text
stringlengths
27.9k
483k
in the currently popular hierarchical picture of structure formation , galaxies are thought to be embedded in massive dark halos . these halos grow from density fluctuations in the early universe and initially contain baryons in a hot gaseous phase . this gas subsequently cools and some fraction eventually condenses into stars . much progress has been made in observationally delineating the global star formation history and the resulting build - up of stellar mass ( e.g. , madau et al . 1998 , brinchmann & ellis 2000 ; dickinson et al . 2003 ; bundy , ellis & conselice 2005 ) . however , many of the physical details , particularly the roles played by feedback and cooling , essential for a full understanding of how galaxies form , remain uncertain . models ( e.g. , van den bosch 2002 ; abadi et al . 2003 ) have great predictive power in this area but only by assuming presently - untested prescriptions for these effects . obtaining further insight into how such processes operate is thus an important next step not only in understanding galaxy evolution , but also in verifying the utility of popular models , as well as the hierarchical concept itself . one approach towards understanding this issue is to trace how the stellar mass in galaxies forms in tandem , or otherwise , with its dark mass . the first step in this direction began with studies of scaling relations between the measurable properties of disk galaxies , specifically the relation between luminosity and maximum rotational velocity ( tully & fisher 1977 ) . studies utilizing roughly a thousand spiral galaxies have revealed a tight correlation between absolute magnitude and the maximum rotational velocity for nearby galaxies ( haynes et al . 1999 ) . the limited data at high redshift suggests the tf relation evolves only modestly , equivalent to at most 0.4 - 1 magnitudes of luminosity evolution to @xmath4 ( vogt et al . 1997 ; ziegler et al . 2002 ; bohm et al . how the tully - fisher relation evolves with redshift is still controversial , although it appears that fainter disks evolve the most ( bohm et al . 2004 ) , and that selection effects are likely dominating the differences found between various studies . furthermore , it has been difficult for modelers to reproduce the tully - fisher relation to within 30% ( e.g. , cole et al . 2000 ) , making it an important constraint on our understanding of the physics behind galaxy formation . unfortunately , any interpretation of the tf relation is complicated by the fact that both luminosity and virial mass might be evolving together . a more physically - motivated comparison would be between stellar and virial mass . not only does this relation break potential degeneracies in the tf technique , but it also samples more fundamental quantities . in this paper we begin this task by investigating the evolution in the fraction of the total mass that is in a stellar form . this can be accomplished with some uncertainty by contrasting the _ stellar mass _ of a galaxy with its _ halo mass_. we selected disk galaxies for our initial study since these two quantities can be effectively probed observationally for these galaxies with various assumptions ( e.g. , van den bosch 2002 ; baugh et al . 2005 ) . this paper presents the first study of the near - ir tf relation , as well as a comparison between stellar and halo masses , for 101 disk galaxies within the redshift range @xmath5 drawn mostly from the deep1 redshift survey ( vogt et al . our goal is to address several questions relating to the mass assembly history of disks . as our sample is not formally complete in any sense , we can not derive general conclusions concerning the history of _ all _ present - day disks . however , we can determine whether the disks selected from the deep1 survey in the sampled redshift range are still accreting matter and converting baryons into stellar disks at a significant rate . we construct stellar mass tully - fisher and stellar mass / halo mass relation for our sample and find that there is little evolution in either from @xmath6 . this suggests that the dark and stellar components of disk galaxies grow together during this time . the paper is organized as follows : 2 contains a description of the sample including the fields used and the different data products and a discussion on quality control . 3 describes how various quantities , such as the halo and stellar masses , are derived from the basic data . 4 presents our results and 5 summarizes our conclusions . we assume the following cosmology throughout this paper : h@xmath7 = 70 km s@xmath8 mpc@xmath8 , @xmath9 = 0.7 and @xmath10 = 0.3 . our sample consists of 101 galaxies , 93 of which are drawn from the deep1 survey ( see vogt et al . 2004 ; simard et al . 2002 ; weiner et al . 2004 ) , and 8 of which were obtained independently . each of these systems has a resolved rotation curve which was obtained with the keck telescope using lris ( oke et al . 1995 ) in the redshift range @xmath11 to @xmath12 . the additional data presented in this paper , which enables stellar masses to be compared with virial masses , consists of near - infrared imaging described in @xmath132.2 . full details of the selection criteria for the galaxies are discussed in deep1 papers ( vogt et al . 1996 , 1997 , 2005 ) to which the interested reader is referred . moreover , the necessary assumptions implicit in the derivation are detailed in those papers , to which we refer the reader . briefly , galaxies were selected morphologically as elongated disks in hubble space telescope ( hst ) f814w ( i@xmath14 ) images with i@xmath15 . the inferred inclination was chosen to be greater than 30to facilitate a measurement of the rotational velocity . the optical images used for both photometric and morphological analyses come from hst wide field planetary camera-2 ( wfpc2 ) observations of the groth strip ( groth et al . 1994 ; vogt et al . 2004 ) , the hubble deep field ( williams et al . 1996 ) , and cfrs fields ( brinchmann et al . 1998 ) . using i@xmath14 images , structural parameters were determined for each galaxy using the gim2d and galfit packages ( simard et al . 2002 ; peng et al . we fit a two - component model to the surface brightness distribution , assuming a de vaucouleurs law for the bulge , and an exponential for the disk component . based on these fits , the disk scale length , r@xmath16 , and bulge - to - disk ratio ( @xmath17 ) were determined . the uncertainties in the r@xmath16 values determined through this method are possibly underestimated , as has been explored through careful 2-d fitting ( e.g. jong et al 1996 ) . to be inclusive of possible effects , we will incorporate an additional 30% uncertainty for our overall error on the measured values of @xmath18 . details of the observations , reductions and extraction of maximal rotational velocities @xmath19 from the lris spectroscopy are presented in vogt et al . ( 1996 , 1997 , 2005 ) . briefly , each disk galaxy was observed along its major axis , as determined from the hst images ( simard et al . the maximum velocity , @xmath19 , is determined by fitting a fixed form for the rotation curve scaled according to the i@xmath14 disk scale length @xmath18 . the assumed rotation curve has a linear form which rises to a maximum at 1.5@xmath20 r@xmath21 , and remains flat at larger radii . our assumption that 1.5@xmath20 r@xmath16 is the radius where the rotation curves for disks reach their maximum is reasonable , based on a similar behavior for local disk systems of similar magnitudes ( persic & salucci 1991 ; sofue & rubin 2001 ) . our rotation curves are also visible out to several scale lengths , or roughly 2 - 5 ( e.g. , vogt et al . 1997 ) , adequate for measuring v@xmath22 . this model form is then convolved with a seeing profile that simulates the conditions under which the observations were taken , and v@xmath23 is determined by iterative fitting . effects from the width of the slit , slit misalignment with the major axis , and varying inclination were taken into account when performing these fits , and calculating the resulting errors . typically , @xmath19 is determined to a precision of 10 - 20% due to these various sources ( e.g. , vogt et al . 1997 ) . the new data we present in this paper consist of deep near - infrared observations of the deep1 extended sample . precision near - infrared photometry is the critical ingredient for determining stellar masses for our sample ( brinchmann & ellis 2000 ) . photometry was acquired with a @xmath24 filter with three different instruments : the keck near infrared camera ( nirc , matthews & soifer 1994 ) , the ukirt fast - track imager ( ufti , roche et al . 2003 ) , and the cooled infrared spectrograph and camera for ohs ( cisco , motohara et al . 2002 ) on the subaru 8.4 meter telescope ( see bundy et al . 2004 for a more detailed description of these data ) . nirc has a field of view of 38 and a pixel scale of 0.15 pixel@xmath8 . the equivalent numbers for ufti are : 96 field of view , with a pixel scale of 0.091 pixel@xmath8 . the cisco camera has a field of view of 108with a pixel scale of 0.105 pixel@xmath8 . the typical depths for these images is @xmath25 = 20.5 - 21 ( vega ) with a typical seeing of @xmath26 0.8 . in each case , the infrared data were taken with a dither pattern whose step size exceeded the typical size of the galaxies of interest . the data were reduced by creating sky and flat - field images from sets of disregistered neighboring science frames . standard stars were observed for calibration purposes during the observations . some images taken in good seeing but through thin cloud were subsequently calibrated in photometric conditions via shallower exposures taken with the wide field infrared camera ( wirc , wilson et al . 2003 ) on the hale 5 meter telescope . we measure our photometry in the @xmath25-band , and hst i@xmath14 and v@xmath27 bands , within a scale - length factor , either 1.5r@xmath16 or 3r@xmath16 , both of which are generally large enough to avoid seeing effects from the @xmath24 imaging . we then extrapolate the total magnitudes , within each band , out to infinite radius by using the fitted parameters for an exponential disk derived in the hst i@xmath14 band . to compare observables over a range in redshift , it is necessary to reduce all measures to a standard rest - frame . galactic extinction corrections were applied using the formalism of schlegel , finkbeiner & davis ( 1998 ) and internal extinction was accounted for according to measured inclinations and luminosities using the precepts of tully et al . it is debatable whether internal extinction corrections derived for nearby spirals are applicable to higher redshift disks . our sample is mostly composed of systems with m@xmath28 where extinction could be a cause for concern . however , direct extinction measurements in moderate redshift disks , determined through overlapping pairs , find a modest overall extinction ( white , keel & conselice 2000 ) . to derive absolute magnitudes and rest - frame colors , we estimated non - evolutionary @xmath29-corrections . the k - band @xmath29-corrections were computed from spectral energy distributions determined alongside the stellar mass fits . we experimented with both the original algorithm developed by brinchmann & ellis ( 2000 ) and also an independent one developed by bundy , ellis & conselice ( 2005 , see @xmath133.2 ) . both methods agree very well . the approach is similar to that used by vogt et al . ( 1996 , 1997 ) where @xmath29-corrections were calculated from model seds from gronwall & koo ( 1995 ) based on various star formation histories in the bruzual & charlot ( 1993 ) code . we also note that our derived k - correction values are very similar to non - evolutionary @xmath29-corrections derived by poggianti ( 1997 ) using seds from poggianti & barbaro ( 1996 ) . the virial mass of a galaxy - that is , the combined total of the dark , stellar and gaseous components - is perhaps its most fundamental property . however , obtaining an accurate estimate from observed quantities is necessarily difficult since there is no _ a priori _ agreed model for the relative distributions of the various components . our approach to this challenge will be to use both analytical techniques and semi - analytical simulations to investigate the relationship between the total halo mass and our dynamical and structural observables . although necessarily approximate and debatable in terms of the assumptions made , we will attempt , where possible , to investigate the uncertainties involved by contrasting the two approaches in the context of our data . for a simple virialized system , such as a circularly rotating disk , we can place constraints on the total mass within a given radius @xmath30 independent of model assumptions . if @xmath31 , where we estimate the maximum rotational velocity v@xmath22 is reached , the mass within @xmath30 is given by , @xmath32 where @xmath31 . this assumes that the rotation curve reaches the maximum velocity by 1.5 @xmath20 r@xmath16 , ( persic & salucci 1991 ) , otherwise v@xmath22 should be replaced by v@xmath33 . there are a few possible approaches for determining total halo masses , some of which require the use of simulations to convert observed dynamical qualities , usually v@xmath22 , into halo masses . we take a basic approach using eq . ( 1 ) to obtain the total mass with a given radius , and take a suitably large total radius of 100 kpc to measure the total halo mass . this is often the extent of disk hi rotation curves , and similar to the sizes of dark matter halos ( e.g. , sofue & rubin 2001 ) . another approach now being used ( e.g. , bohm et al . 2004 ) has been proposed by van den bosch ( 2002 ) . in this case , it is argued from analytic simulations of disk galaxy formation that the quantity @xmath34 gives , on average , the best empirical representation of the virial mass for simulated disks . the zero - point of the relationship between r@xmath16v@xmath35/g and virial mass is claimed to be independent of feedback and independent of the mass of the halo ( van den bosch 2002 ) . semi - analytic models based on @xmath36cdm ( e.g. , cole et al . 2000 ; benson et al . 2002 ; baugh et al . 2005 ) suggest , however , that the ratio between v@xmath37r@xmath16/g and halo mass is not as simple as the above formalism implies . the latest _ galform _ models from baugh et al . ( 2005 ) and lacey ( priv . communication ) show that equation [ 1 ] and van den bosch ( 2002 ) under - predicts the dark halo mass . these models show that the relationship between the virial mass at r@xmath16 and the total halo mass changes as a function of mass in the sense that the ratio is higher for lower mass halos . physically this can be understood if high mass halos have a larger fraction of their baryonic mass in a hot gaseous phase that is not traced by the formed stellar mass . using the _ galform _ models to convert our observables into a total halo mass is however potentially inaccurate as these models can not reproduce the tully - fisher relation to better than 30% . however , there are reasons to believe that the semi - analytic models put the correct amount of halo mass into their modeled galaxies ( e.g. , benson et al . this however does not necessarily imply that the v@xmath22 values in these models are able to accurately match the halo masses . independent determinations of total halo masses are necessary to perform this test . however , the masses from this approach match within 0.5 dex the masses from using eq . ( 1 ) with a suitably large total radius . with these caveats in mind , we have used the _ galform _ model results to fit the relationship between the virial mass at r@xmath16 ( eq . 1 ) , and the total mass of the halo ( m@xmath38 ) , a ratio which we call @xmath39 . we fit @xmath40 as a linear function of @xmath41 , such that @xmath42 . using the _ galform _ results we fit @xmath43 and @xmath44 at redshifts @xmath45 . we find that the functional form of @xmath40 does not change significantly with redshift , with typical values @xmath46 and @xmath47 . the value of the halo mass m@xmath38 is then given by : @xmath48 observational and model uncertainties contribute to errors on these virial and halo mass estimates in two ways . measured scale lengths given by the gim2d and galfit fitting procedure ( 2.1 ) , give an average error of 0.12 kpc , although we add an extra error to this to account for systematics seen when performing one component models to disks / bulge systems ( de jong 1996 ) . at the same time , systematic difficulties in the rotation curve analysis ( 2.1 ) can arise . vogt et al . ( 2005 ) discuss these issues in some detail and conclude the average error is @xmath49 kms@xmath8 ( typically @xmath5010 - 20% ) . following the discussion of rotation curve fitting in @xmath132.1 , it is possible we do not measure the true v@xmath22 required for insertion in equations ( 1 & 2 ) . even a modest underestimate of v@xmath22 would lead to a significant error in @xmath51 . in combination , these measurement uncertainties imply virial and halo masses precise to no better than 30% . the semi - analytic method for computing total halo masses is likely limited , and thus we also investigate the total halo masses found through the van den bosch ( 2002 ) formalism , and through the use of equation ( 1 ) out to 100 kpc . we find that the total halo mass through eq . ( 1 ) , eq . ( 2 ) and van den bosch ( 2002 ) are all fairly similar . we account for these differences in our use of total halo masses in 4.3 . our procedure for deriving stellar masses follows the multi - color method introduced by brinchmann & ellis ( 2000 ) . combining hst v@xmath27 , i@xmath14 , and near - infrared @xmath24 photometry for a galaxy of known redshift , we fit a range of template seds synthesized using software by bruzual & charlot ( 2003 ) . our photometry is done in a matched aperture large enough to avoid seeing problems associated with the @xmath24-band . these fitted seds constrain the @xmath24-band mass / light ratio . our code for computing stellar masses only uses models with ages less than the age of the universe at the redshift observed . a @xmath52 analysis normalized by the near infrared @xmath24-band luminosity yields the stellar mass . the template seds were constructed sampling a range of exponentially - declining star formation rates , metallicities and ages with a salpeter imf . using other stellar mass functions , such as chabrier , kennicutt or kroupa would result in computed stellar masses that are smaller by @xmath53 0.3 dex . we further assumed a simple exponentially declining star formation history with @xmath54 values ranging from 0.001 - 15.0 gyr , and metallicities from z=0.005 - 5 in solar units . typical uncertainties in this method are a factor of three ( brinchmann & ellis 2000 ; papovich , dickinson & ferguson 2001 ; drory et al . 2004 ; bundy , ellis & conselice 2005 ) . the sed is constrained by the observed colors as measured in an aperture of radius 1.5 @xmath20 r@xmath16 in each band which is optimal in terms of signal - to - noise ( s / n ) . we assume that there are no color gradients and use the colors measured within this aperture as the global color . the total @xmath24-band light is measured by extrapolating the 1.5 @xmath20 r@xmath16 flux to infinity , assuming the same exponential fit as measured in the hst i@xmath14 image . errors on the stellar masses arising from photometric uncertainties can be determined in a monte carlo fashion . simulated exponential disks of a known magnitude were inserted into the reduced hst and ground based images and photometrically recovered using the tools that were applied to the sample . the simulated disks were arranged to randomly sample the selected ranges of disk scale - length @xmath18 and inclination . the derived photometric errors were then input into the stellar mass calculations from which a @xmath55 range was calculated . the average error in m@xmath56 is 0.47 dex . the calculation of stellar masses through this technique is limited to a degree by systematics which are difficult to constrain with the current data set , with @xmath24-band magnitude errors only a small source of uncertainty . the models of the spectral evolution of galaxies depend both on the observed stellar libraries and underlying theory , and some fundamental uncertainties remain ( c.f . bruzual & charlot 2003 ) . we expect uncertainties of about 5 - 10% , given the range of possible models ( charlot , worthey & bressan 1996 ) . this is not important for the conclusions in this paper but might be an issue with larger and more accurate datasets in the future . we also can not constrain the amount of recent star formation produced in bursts ( e.g. , kauffmann et al . 2003 ; bell & de jong 2001 ) , resulting in a slight systematic overestimate of the stellar mass . the latter effect only becomes important at large burst fractions and should be @xmath57% for our big spirals ( c.f . drory et al 2003 ) . the final systematic uncertainty is the dust correction adopted . we use the calzetti ( 1997 ) extinction law in our stellar mass calculations , although other extinction laws produce very small differences ( at most 10 - 20% ) in the resulting stellar mass ( papovich et al . 2001 ) . taken together we are likely to have systematic uncertainties in our stellar mass estimates amounting to @xmath58 dex which is lower than our random uncertainty , thus they will not influence our results . the tully - fisher ( tf ) relation has been the traditional method for investigating how dark halos and the stellar components of disk galaxies relate . the optical relation has been studied in disks galaxies out to redshifts @xmath4 by e.g. , vogt et al . ( 1996 , 1997 , 2005 ) , ziegler et al . ( 2002 ) and bohm et al . these investigations have found between 0.4 and @xmath59 magnitudes of rest - frame b - band luminosity evolution in disks between @xmath60 and @xmath61 . this luminosity evolution is derived by assuming that the slope of the tf relation at high redshift is the same as it is at @xmath60 . the question of differential evolution in the relation ( ziegler et al . 2002 ; bohm et al . 2004 ) , for example that the faintest galaxies evolve more rapidly , remains an important unknown . although the purpose of this paper is to move beyond the tf relation , we begin by plotting the k - band tf relation for our sample ( figure 1 ) . one might expect evolution of the tf relation in the k - band to display a clearer signal than the b - band since the effects of dust are mitigated and passive evolution should be more uniform in its effect across the sample . when we assume that the slope of the k - band tully - fisher is the same as the local value from verheijn ( 2001 ) we find no significant evolution , as is also found in the b - band ( vogt et al . we perform these fits using both a downhill simplex amoeba and levenberg - marquardt @xmath62 minimization , both of which give the same results . we find a fading of 0.04@xmath630.24 magnitudes for systems at @xmath64 compared with the @xmath60 relationship and a brightening of 0.37@xmath630.23 magnitudes for systems at @xmath65 , consistent with no evolution . the scatter does not evolve significantly ( 1.13 magnitudes for the @xmath660.7 sample , and 0.72 magnitudes for systems at @xmath64 ) . in all cases , the observed slope and scatter change only slightly when we ignore internal extinction corrections . although our results are broadly consistent with earlier , smaller samples , the interpretation of any evolutionary signal is complicated in two ways . first , only a limited range of luminosity and rotational velocity can be sampled at high redshift , leading to great uncertainties given the intrinsic scatter . second , as luminosity and rotational velocity are indirect measures of the assembly state of the galaxy , both may be evolving in complex ways , that mask actual evolutionary changes . the first step beyond the tf relation is to compare the stellar mass to the measured maximum velocity - a relation we will call the _ stellar mass - tully fisher relation_. the classical b - band tf relation scales such that @xmath67 . this coupling becomes even steeper for the local stellar mass tully - fisher relation in nearby disks , @xmath68 ( bell & de jong 2001 ) . ideally , we seek to measure the all - inclusive baryonic tf relation , but measuring the gas content of high redshift disks is not yet feasible . we can estimate how much cold gas we are missing in our stellar mass inventory by investigating the gas mass fractions for nearby disk galaxies . through examinations of the luminosities and hi masses for nearby disks , mcgaugh & de blok ( 1997 ) conclude that galaxies which are massive , bright , red , or have a high - surface brightness , have very little gas in comparison to bluer , fainter , lower surface brightness systems . systems which are brighter than m@xmath69 have gas mass fractions that are typically 0.1 or lower . since our selection finds the most luminous , high surface brightness systems , which are also red , they are the least likely sub - class of disks to have a high gas content . the stellar mass tf relation is shown in figure 2 where , as before , we divide the sample into two redshift bins , split at @xmath70 . each panel contains a solid line giving the @xmath60 best fit and a dashed line illustrating the @xmath63 3 @xmath71 uncertainty in this fit ( bell & de jong 2001 ) . as was the case for the conventional tf relations , no significant evolution in the zero - point is observed . the bell & de jong ( 2001 ) @xmath72 stellar mass tully - fisher relation can be written as m@xmath73 . by holding the slope of this relationship constant , we find that the zero point is best fit by 0.45@xmath630.12 at @xmath74 and 0.41@xmath630.13 at @xmath64 . neither of these are however significantly different from the @xmath60 relationship , and are very similar to each other . this implies that , if growth continues , the stellar and dark components are growing together . for example , if disk assembly since @xmath75 proceeded only by the addition of stellar mass at a uniform rate of 4 m@xmath76 year@xmath8 , the local zero point would be discrepant at the 4@xmath71 level . this lack of evolution is important for understanding how disk galaxy formation is occurring ( see 4.4 ) . moreover , the scatter in the stellar mass tully - fisher relation is similar to that observed in the k - band tully - fisher after converting the k - band magnitude scatter into a luminosity and assuming an average stellar mass to light ratio . the typical scatter ( in log m@xmath56 units ) in stellar mass for these is 0.65 for disks at @xmath74 and 0.48 for those at @xmath64 . the final step in our analysis will be an attempt to convert our measured quantities into a comparison of the stellar and halo masses as discussed in @xmath133 . recognizing the considerable uncertainties involved , figure 3 shows as a function of redshift the ratio , @xmath77 m@xmath56/m@xmath78 , between the stellar masses and virial masses ( eq . 1 ) within 3r@xmath16 for our sample . as expected , the stellar masses are nearly always less than the independently - derived virial masses , indicating that our methods for computing these values are not dominated by large systematic errors . it can also be seen from this figure that there is wide range of m@xmath56/m@xmath78 values at every redshift from @xmath79 . the open pentagons show the median values of @xmath80 as a function of redshift . the dotted horizontal line shows the global baryonic mass fraction @xmath81 = 0.171 derived from wmap results ( spergel et al . within 3r@xmath16 it appears that many disks have m@xmath56/m@xmath78 values higher than this limit . it also appears that within 3r@xmath16 the stellar mass of the disk dominates the virial mass . this indicates that within the visual parts of some of our sample , the disk component accounts for roughly all the mass , which is consistent with a maximal disk interpretation . however , there are clear examples on figure 3 where either the stellar mass is overestimated and/or the virial mass is underestimated . both of these are possibilities , since our stellar masses are potentially too high from using a salpeter imf , and we might be underestimating the value of v@xmath22 due to seeing , or a lack of depth in the lris observations . there is also a slight bias such that at higher redshifts nearly all galaxies sampled have high v@xmath22 values , which are likely maximal , while at lower redshifts we are sampling systems with lower v@xmath22 values , that have not yet fully formed their stellar masses . this is likely part of the reason that the stellar mass to virial mass ratio increases slight at higher redshifts . however , we are also interested in constraining the relationship between the total halo mass , m@xmath38 , and the total stellar mass m@xmath56 . as we discuss in 3 , it is very difficult to accurately obtain halo masses . we therefore show in figure 4 total halo masses derived through eq . ( 1 ) at 100 kpc and through the semi - analytical approach ( eq . although we can not accurately determine total halo masses for individual systems , our main goal is to compare how the ratio of stellar to halo mass changes with redshift . figure 4 shows this relationship divided into the same redshift bins as in figures 1 & 2 . to first order , the halo masses and stellar masses of disk galaxies should correlate if star formation is regulated in the same manner in halos of different masses ( steinmetz & navarro 1999 ) . although there is no reason to expect any particular functional form between these two quantities , there is a reasonably well - fit linear relationship between them . we find that the zero point and slope of this relation does not change significantly between low and high redshift . there is also no obvious change in the scatter from high to low redshift ( 0.26 cf . the most significant outcome of figures 3 and 4 however is the remarkable similarity in trends found at high and low redshift suggesting some disks had completed the bulk of their stellar assembly by @xmath821 or more likely that the stellar and dark masses of galaxies grow together continuously . figure 5 shows the distribution of @xmath80 = m@xmath56/m@xmath38 for systems more massive than , and less massive than , the average halo mass , m@xmath83 . the solid line is the universal baryonic mass ratio . as mentioned earlier , most disk stellar mass fractions are lower than the cosmic ratio , which seems appropriate given our stellar mass inventory is not intended to account for all associated baryons . although some dispersion is expected , conceivably some fractions are overestimated or underestimated , for example by making incorrect assumptions about the imf , or underestimating v@xmath22 ( and hence m@xmath38 ) by insufficient sampling of the rotation curve . figure 5 also shows a tentative population of disk galaxies with remarkably low stellar fractions , the most extreme cases occurring in objects with halo masses m@xmath84 . these galaxies deviate from the @xmath72 stellar mass tf relation by more than 4 @xmath71 . investigation of the individual systems that lie in this category shows them to be undergoing vigorous star formation as inferred by bluer than average @xmath85 colors . a weak correlation was found between @xmath80 and @xmath85 . because these systems are blue , they may have kinematic asymmetries in their rotation curves which may raise their v@xmath22 values . however , this is unlikely the effect producing this slight correlation as most of the low @xmath80 systems also have low v@xmath22 values . we investigated several other possible correlations involving @xmath80 ( for example with the bulge / disk ratio ) but no significant trends were found . to investigate the implications of our results for the assembly history of stellar mass in disk galaxies since @xmath4 , we return to the behavior of the stellar fraction @xmath80 = m@xmath56(@xmath533r@xmath16)/m@xmath78(@xmath533r@xmath16 ) vs. redshift ( figure 3 ) , and the relationship between total stellar mass ( m@xmath56 ) and halo mass ( m@xmath38 ) at @xmath64 and @xmath74 ( figure 4 ) . trends in these relations may shed light on the physical processes regulating star formation in disk galaxies . this is only the case however if the value of r@xmath16 for disks does not grow with time , which appears to be the case for the largest disks , based on our own limited data , and in statistical studies of @xmath86 disks ( ravindranath et al . 2004 ; conselice et al . 2004 ; cf . ferguson et al . 2004 for higher redshifts ) . first , figure 3 shows that the median stellar vs. virial mass value for our sample does not change significantly with redshift . we can compare two simple extreme models to this result to determine the likely method by which disk galaxies are forming at @xmath3 . in a ` monolithic collapse ' model , where the dark halo and baryons for a galaxy are in place at high redshift , and there is no change in these mass components with time ( e.g. , eggen , lynden - bell & sandage 1962 ) , the average value of m@xmath56/m@xmath78 should increase with time . this assumes that effective radii distribution for disks does not change with time , which is observed ( ravindranath et al . there is evolution in the monolithic model only in the sense that baryons are gradually converted to stars over time . this is the opposite to the observed trend in which the ratio of m@xmath56/m@xmath78 is roughly constant . on the other hand , in a hierarchical picture , the value of m@xmath56/m@xmath78 remains relatively constant with time . shown in figure 3 are model predictions from _ galform _ which show that the value of m@xmath56/m@xmath78 remains relatively constant with redshift . while we do not have a complete sample , we do find that the average and median values of m@xmath56/m@xmath78 for our sample remain constant with redshift within the errors , which is consistent with the hierarchical idea . the higher average m@xmath56/m@xmath78 value at @xmath86 is likely produced by our selection of the brightest disks at these redshifts ( vogt et al . the _ galform _ model however does not predict the full range of mass fractions that we find . at all redshifts there are systems dominated within r@xmath16 by their stellar masses , while some systems have low m@xmath56/m@xmath78 ratios . this is either produced by an observational bias ( 4.3 ) , or the simulations predict too much dark matter in the centers of disks . figure 4 shows the relationship between total stellar and total halo masses ( see 3.1 ) for our sample , divided into our two redshift ranges . we also show on figure 4 the _ galform _ semi - analytic model values for this relationship for disk dominated galaxies with the 80% completeness indicated . as in figure 3 , no strong evolution in m@xmath56/m@xmath38 ratios is predicted . as galaxies in the semi - analytic models grow by accreting smaller systems , or intergalactic gas coupled with dark matter , and then converting the newly - obtained gas into stars quickly , the relationship between m@xmath56 and m@xmath38 remains constant . this is generally what we find , as well as a good agreement with the model predictions . in fact , the agreement between the total stellar and halo masses with the _ galform _ models is slightly better than the comparison between the stellar and virial masses with 3@xmath20r@xmath16 , which further suggests that the luminous components of disk galaxies are dominated by the stellar mass ( 4.3 ) . in summary , the fact that the ratio m@xmath56/m@xmath38 remains relatively constant in our sample with redshift , and that the stellar - mass tf relation does not evolution , are indications that disk galaxies are forming through the accretion of both dark and baryonic mass . disk galaxies are undergoing star formation at @xmath3 at a rate of a few solar masses per year . if disks at @xmath4 contained all the baryonic or halo masses that they have at @xmath60 , we would see an increase in m@xmath56/m@xmath38 with time . because we do not see this trend , it appears that new stars form out of gas accreted from the intergalactic medium , which is coupled with dark matter at a constant ratio . this baryonic to dark matter ratio is such that both m@xmath56 and m@xmath38 grow with time , creating a stellar mass to halo mass ratio that is on average relatively constant . we present the results of a dynamical and structure study of 101 disk galaxies drawn mostly from the deep1 survey with redshifts in the range @xmath820.2 - 1.2 . new infrared imaging observations are presented which enable us to derive reasonably reliable stellar masses and thereby to construct the stellar mass tully - fisher relation and its redshift dependence . using various formalisms drawn from analytic and semi - analytical models , we attempt to convert our dynamical data to make the first comparisons of the relative fractions of stellar and total mass in our sample . notwithstanding the considerable uncertainties and sample incompleteness , the results are encouraging and suggest remarkably little evolution in the mix of baryons and dark matter since @xmath821 . although our sample is not formally complete in luminosity or mass , we explore the degree to which there may have been evolution in the relative distribution of stellar and virial masses and find the following : 1 . massive disk galaxies exist out to @xmath4 with halo masses as large as 10@xmath87 , roughly as large as the most massive disks in the nearby universe . these systems also contain a large amount of stellar mass . at least some disk galaxies are nearly fully mature in their stellar content at @xmath88 . we confirm earlier studies based on smaller samples and find no significant evolution in the zero point or scatter of the rest - frame @xmath89-band tully - fisher relation out to @xmath12 . the stellar mass tully - fisher relation out to @xmath12 is likewise largely consistent with the relation found for nearby disks . we find no significant evolution in our sample after comparing systems at redshifts greater than and less than @xmath90 . 4 . although there are clearly great uncertainties in estimating total halo masses from our dynamical data , we find the distribution of the ratio of stellar and halo masses remains relatively similar from @xmath60 to @xmath12 . the stellar fraction observed can be understood if the bulk of the baryons associated with massive disk galaxies have already formed their stars . a modest number of massive galaxies have very low stellar fractions , consistent with continued star formation as revealed by their blue @xmath91 colors . these results are in relatively good agreement with @xmath36cdm analytical and semi - analytical models ( benson et al . 2002 ; baugh et al . 2005 ) , suggesting that disk galaxy formation is hierarchical in nature . our overarching conclusion from this study is that no significant evolution in the stellar mass fraction can be detected in the population of regular massive disks since @xmath821 . although biases and uncertain assumptions may affect detailed quantities at the @xmath92 dex level , the absence of gross trends is consistent with the conclusion that the bulk of these systems grow at @xmath3 by the accretion of dark and baryonic material . this conclusion is however tempered by the fact that our selection is affected by a bias that likely changes with redshift . for example , because our sample was selected by a @xmath93 limit , we are studying the brightest disks at the highest redshifts . these results , while intriguing , will be examined in much greater detail with the upcoming deep2 redshift survey ( davis et al . 2002 ) , which will combine near - infrared imaging from the hale 5 meter telescope with rotation curves obtained using the deimos spectrograph on the keck ii telescope . there are several issues that can be better addressed with deeper spectroscopy and a more careful selection of high - z disks . the total number of disks in this much larger sample is expected to be on the order of thousands and will solidify and expand on the preliminary results revealed by this study . we thank the deep team for generously enabling us to augment their optical sample and catalogs with near - infrared data in order to make this comparison . we also thank cedric lacey and andrew benson for access to their semi - analytical model simulation results , and xavier hernandez and ken freeman for comments regarding this work . cjc acknowledges support from a national science foundation astronomy and astrophysics postdoctoral fellowship . npv is pleased to acknowledge support from nsf grants nsf-0349155 from the career awards program , nsf-0123690 via the advance it program at nmsu , and ast 95 - 29098 and 00 - 71198 administered at ucsc , and nasa stsci grants go-07883.01 - 96a , ar-05801.01 , ar-06402.01 , and ar-07532.01 . abadi , m.g . , navarro , j.f . , steinmetz , m. , & eke , v.r . 2003 , apj , 591 , 499 baugh , c.m . , lacey , c.g . , frenk , c.s . , granato , g.l . , silva , l. , bressan , a. , benson , a.j . , & cole , s. 2005 , mnras , 356 , 1191 bell , e.f . , & de jong , r.s . 2001 , apj , 550 , 212 benson , a.j . , cole , s. , frenk , c.s . , baugh , c.m . , & lacey , c.g . 2000 , mnras , 311 , 793 benson , a.j . , lacey , c.g . , baugh , c.m . , cole , s. , & frenk , c.s . 2002 , mnras , 333 , 156 bohm , a. , et al . 2004 , a&a , 420 , 97 brinchmann , j. , & ellis , r.s . 2000 , apj , 536 , l77 brinchmann , j. , et al . 1998 , apj , 499 , 112 brinchmann , j. 1999 , phd , cambridge university bruzual , a.g . , & charlot , s. 2003 , mnras , 344 , 1000 bundy , k. , fukugita , m. , ellis , r.s . , kodama , t. , & conselice , c.j . 2004 , apj , 601 , l123 bundy , k. , ellis , r.s . , & conselice , c.j . 2005 , apj , astro - ph/0502204 calzetti , d. 1997 , aj , 113 , 162 cole , s. , lacey , c.g . , baugh , c.m . , & frenk , c.s . 2000 , mnras , 319 , 168 charlot , s. , worthey , g. , & bressan , a. 1996 , apj , 457 , 625 conselice , c.j . 2004 , apj , 600 , l139 davis , m. , et al . 2002 , spie , 4834 , 161 de jong , r.s . 1996 , a&a , 118 , 557 dickinson , m. , papovich , c. , ferguson , h. c. , & budvari , t. 2003 , apj , 587 , 25 drory , n. , bender , r. , feulner , g. , hopp , u. , maraston , c. , snigula , j. , & hill , g.j . 2004 , apj , 608 , 742 eggen , o.j . , lynden - bell , d. , & sandage , a.r . 1962 , apj , 136 , 748 gronwall , c. , & koo , d.c . 1995 , apj , 440 , l1 groth , e.j . , kristian , j.a . , lynds , r. , oneil , e.j . , balsano , r. , rhodes , j. 1994 , bass , 185 , 5309 ferguson , h.c . , 2004 , apj , 600 , l107 haynes , m.p . , et al . 1999 , aj , 117 , 2039 kauffmann , g. , et al . 2003 , mnras , 341 , 33 kinney , a. , calzetti , d. , bohlin , r. , mcquade , k. , storchi - bergmann , t. , & schmitt , h.r . 1996 , apj , 467 , 38 madau , p. , pozzetti , l. , & dickinson , m. 1998 , apj , 498 , 106 mcgaugh , s.s . , & de blok , w.j.g . 1997 , apj , 481 , 689 matthews , k. , & soifer , b.t . 1994 , in `` infrared astronomy with arrays , the next generation '' , editied by ian s. mclean , 190 , 239 motohara , k. , et al . 2002 , pasj , 54 , 315 oke , j.b . 1995 , pasp , 107 , 375 papovich , c. , dickinson , m. , & ferguson , h.c . 2001 , apj , 559 , 620 peng , c.y . , ho , l.c . , impey , c.d . , & rix , h - w . 2002 , aj , 124 , 266 persic , m. , & salucci , p. 1991 , apj , 368 , 60 poggianti , b.m . 1997 , a&as , 122 , 399 poggianti , b.m . , & barbaro , g. , 1996 , a&a , 314 , 379 ravindranath , s. , et al . 2004 , apj , 604 , l9 roche , p.f . 2003 , spie , 4841 , 901 schlegel , d.j . , finkbeiner , d. , & davis , m. 1998 , apj , 500 , 525 simard , l. et al . 2002 , apjs , 142 , 1 sofue , y. , & rubin , v. 2001 , ara&a , 39 , 137 spergel , d.n . 2003 , apjs , 148 , 175 steinmetz , m. , & navarro , j.f . 1999 , apj , 513 , 555 tully , r.b . , & fisher , j.r . 1977 , a&a , 54 , 661 tully , r.b . , pierce , m.j . , huang , j - s . , saunders , w. , verheijen , m.a.w . , & witchalls , p.l . 1998 , aj , 115 , 2264 van den bosch , f.c . 2002 , mnras , 332 , 456 verheijen , m.a.w . 2001 , apj , 563 , 694 vogt , n.p . , forbes , d.a . , phillips , a.c . , gronwall , c. , faber , s.m . , illingworth , g.d . , & koo , d.c . 1996 , apj , 465 , l15 vogt , n.p . , 1997 , apj , 479 , l121 vogt , n.p . 2004 , apj , submitted vogt , n.p . , et al . 2005 , in preparation weiner , b.j . , et al . 2004 , apj , in press williams , r. , et al . 1996 , aj , 112 , 1335 white , r.e . , keel , w.c . , conselice , c.j . 2000 , apj , 542 , 761 wilson , j.c . , 2003 , spie , 4841 , 451 ziegler , b.l . 2002 , apj , 564 , l69
ultracold atomic gases offer various opportunities for the study of interacting many - body quantum systems in a well - controlled environment @xcite . for instance , the bose - hubbard model a simplified description for bosons in a periodic potential can be studied with bose - einstein condensates confined in optical lattices @xcite . quantum gases have also gained much attention lately regarding the emergence of an effective space - time @xcite : their low - energy phase fluctuations obey the same covariant field equations as a scalar quantum field in a certain curved space - time . hence , the study of phonons in this laboratory system might shed some light on aspects of cosmic quantum effects , e.g. , hawking radiation @xcite or the freezing and amplification of quantum fluctuations in expanding spacetimes @xcite . although the fluctuations in bose - einstein condensates are usually small , it has recently become possible in experiments to go beyond the classical order parameter and resolve signatures of the fluctuations @xcite . theoretically , the fluctuations in a bose - einstein condensate are usually treated as small perturbations of the mean field . the solution of the coupled field equations is rather demanding and often requires further approximations , especially for time - dependent condensates . the hartree - fock - bogoliubov method , see , e.g. , @xcite , permits in principle the self - consistent propagation of the mean field and the quantum correlations for arbitrary time - dependences of the trap potential or interaction strength . but the scaling of the numerics with system size often limits the actual calculations to a low number of dimensions , certain symmetries , or a short time interval . thermal condensates might be studied using the projected gross - pitaevskii equation @xcite , where the low - energy part of the fluctuations is expanded into an arbitrary basis and , in view of the large thermal occupation , treated classically ; higher excitations as well as the vacuum effect are omitted . on the other hand , studies in the context of expanding spacetimes , e.g. , @xcite , indeed focus on the quantum fluctuations but often assume a homogeneous background or start with the hydrodynamic action , which is only valid on scales longer than the healing length . in this article , i will discuss the evolution of the quantum fluctuations in trapped time - dependent bose - einstein condensates . the linear field operators will be expanded into their eigenmodes thus permitting the diagonalization of the ( initial ) evolution equations . although basis expansions of the field operator are frequently used ( e.g. , in @xcite ) , these references usually consider the harmonic oscillator eigenfunctions , a large number of which must be used in order to describe the excitations properly . by adopting the eigenmodes , a much smaller part of the basis needs to be considered and many more situations will become numerically feasible . ( note , however , that in order to obtain the fluctuation eigenmodes , a relatively large number of oscillator functions must be employed but they need not be propagated . ) this article is organized as follows . section [ secfieldequations ] reviews the field equations and their linearization . the bogoliubov - de gennes equation for the linear quantum fluctuations can be diagonalized by an eigenmode expansion , which will be performed in sec . [ seceigenmode ] . however , as soon as trap frequency or interaction strength are varied , off - diagonal terms appear . this coupling of different modes is negligible for excitations with energies much smaller than the chemical potential even in time - dependent condensates , as will be shown in sec . [ sectf ] , where the order parameter is treated in the thomas - fermi approximation . it is then also possible to establish the analogy between phase fluctuations and a massless scalar field in a certain curved spacetime . if the excitation energies are of the same order as the chemical potential , the coupling of different modes might lead to a population transfer , which will be illustrated in an example in sec . [ sec1d ] . in dimensionless units , the field operator @xmath0 of a trapped ( quasi)-@xmath1-dimensional bose - einstein condensate obeys the non - linear schrdinger field equation @xcite @xmath2 \hat \psi \ , , \label{fieldequation } \end{aligned}\ ] ] with @xmath1-dimensional coupling strength @xmath3 . by changing the @xmath4-wave scattering length @xmath5 through feshbach resonances @xcite or by varying the trap frequency @xmath6 , an external time - dependence can be prescribed on the condensate . the gas cloud will adapt to these changes and it will either expand or contract and with it the quasi - particle excitations residing upon it . a part of this background motion can be accounted for by transforming to new coordinates @xmath7 with scale factor @xmath8 . the field operator then reads @xcite @xmath9 the phase @xmath10 is chosen such as to generate an isotropic velocity field @xmath11 which ( at least partially ) describes the expansion / contraction of the condensate . if the scale factor @xmath8 obeys @xmath12 with @xmath13 being the initial value of the coupling strength , a scaled field equation follows @xmath14 \hat \psi \ , , \label{scaledheisenberg } \end{aligned}\ ] ] where trapping and interaction terms have acquired the same time - dependent pre - factor @xmath15 and all other coefficients are time - independent . ( the scale factor @xmath8 on the left hand side might be included into a redefined time @xmath16 , see sec . [ secspacetime ] . ) for large particle numbers @xmath17 , one might formally expand the field operator into inverse powers of @xmath18 @xcite @xmath19 here , @xmath20 and @xmath21 are the atomic operators . they commute with the linear @xmath22 and higher - order quantum excitations @xmath23 and thus yield the exact conservation of particle number . the order parameter @xmath24 in the center of the trap but diminishes towards the edge of the condensate . insertion of the expansion into the scaled heisenberg equation yields the gross - pitaevskii equation for the classical background @xmath25 @xcite @xmath26 \psi_0 \ , . \label{gp } \end{aligned}\ ] ] the linear quantum fluctuations @xmath27 obey the bogoliubov - de gennes equation @xcite @xmath28 \hat \chi + f^2 g_0 \psi_0 ^ 2 \hat\chi^\dagger \ , , \label{bdg } \end{aligned}\ ] ] and the residual terms comprise the equation of motion for @xmath23 @xmath29 \hat \zeta + g_0 \psi_0 ^ 2 \hat \zeta^\dagger \nonumber\\ & \quad + g_0 \left ( 2 \psi_0 \hat \chi^\dagger \hat \chi + \psi_0^ * \hat \chi^2 + \hat \chi^\dagger \hat \chi^2 \right ) + { { \cal o } } ( g_0 \hat \zeta ) \ , . \label{zeta } \end{aligned}\ ] ] these higher orders @xmath23 must remain small in order for the mean - field expansion to be valid , i.e. , for the linearized equation to be applicable . this means that the terms involving products of @xmath27 must remain small because they act as source terms for higher orders @xmath23 . from the gross - pitaevskii equation , i can infer when the evolution of the order parameter @xmath25 is solely described by the scale factor @xmath8 : apart from the trivial case @xmath30 , this occurs only when the spatial derivatives can be neglected with respect to the interaction and trapping terms , @xmath31 , i.e. , in the thomas - fermi approximation . density and phase of the order parameter @xmath32 then assume the form @xmath33 where the heaviside step function @xmath34 is @xmath35 for @xmath36 and @xmath37 elsewhere . in this approximation , the motion of the classical background becomes stationary and the scaled coordinates @xmath38 are co - moving with the condensate . the bogoliubov - de gennes equation can be tackled by introducing self - adjoint operators @xmath39 with @xmath40 being the phase of the order parameter . these operators resemble ( relative ) density and phase fluctuations @xmath41 and @xmath42 up to the prefactor @xmath43 . since this prefactor eventually becomes large near ( and beyond ) the surface of the condensate , the smallness of @xmath44 and @xmath45 can not be ensured . therefore , i will stick to @xmath46 in the following , but still refer to them as density and phase fluctuations . they obey @xmath47 \hat \chi_- \ , = \ , - \mathcal k_+ \hat \chi_+ \ , , \nonumber\\ & \frac{1}{2 } \left [ b^2 \frac{\partial}{\partial t } + \bm v_0 \bm \nabla_x + \frac{1}{2 } ( \bm \nabla_x \bm v_0 ) \right ] \hat \chi_+ \ , = \ , \mathcal k_- \hat \chi_- \ , , \label{eomchipm } \end{aligned}\ ] ] where the velocity field @xmath48 results from the residual background phase beyond the thomas - fermi approximation . however , @xmath49 is small for @xmath50 and if @xmath51 does not change too swiftly , because an almost homogeneous phase will develop with only small deviations near the boundary of the condensate . whereas attractive @xmath52 invalidate the thomas - fermi approximation and generally @xmath53 even in the center of the trap . the differential operators on the right - hand sides @xmath54 generally do not commute for inhomogeneous condensates @xmath55 @xmath56 \ , & = \ , f g_0 \left\ { ( { \mbox{\boldmath$\nabla$}}_x^2 \varrho_0 ) + 2 ( { \mbox{\boldmath$\nabla$}}_x\varrho_0 ) { \mbox{\boldmath$\nabla$}}_x \right\ } \ , . \label{commkpm } \end{aligned}\ ] ] in order to define the initial state unambiguously , i will assume that the condensate is at rest before @xmath57 . then @xmath58 , @xmath59 , @xmath60 , and @xmath61 such that the left hand sides of eqs . reduce to partial time derivatives and the initial eigenmode equations for @xmath62 follow @xmath63 because @xmath64 for inhomogeneous condensates , cf . eq . , density and phase fluctuations of each mode must have different space dependences . this leads to the expansions @xcite ( i will adopt the sum convention throughout this article for brevity ; any indices appearing only on one side of the equation are to be summed ) @xmath65 of @xmath62 into different eigenmode bases @xmath66 and @xmath67 , see appendix [ appeigenmodes ] for more details on how to obtain @xmath68 . usually , these two bases are neither orthogonal nor normalized , , but instead can be chosen to be dual to each other @xmath69 note that this condition does not fix the norm of @xmath68 but still permits the multiplication by an arbitrary factor , @xmath70 and @xmath71 . observables must be unaffected by this ambiguity , see app.[secobservables ] . insertion of the eigenmode expansion into the linear field equations yields a set of coupled first - order differential equations @xmath72 with time - dependent coefficients . the symmetric matrices @xmath73 are initially diagonal @xmath74 and @xmath75 . at later times , they acquire off - diagonal elements because of the different - time commutators @xmath76 \neq 0 $ ] and @xmath77 \neq 0 $ ] when @xmath78 . the velocity coupling matrix @xmath79 h_m^- \label{velo } \end{aligned}\ ] ] is not symmetric , but vanishes for homogeneous phases @xmath80 of the order parameter , e.g. , initially or in the thomas - fermi approximation . for repulsive @xmath50 and slow variations of interaction strength @xmath81 and trap frequency @xmath6 , the order parameter phase is homogeneous except for small ripples near the boundary of the condensate such that the matrix @xmath82 is usually negligible . from the evolution equations with the initially diagonal coupling matrices , the introduction of bosonic operators @xmath83 and @xmath84 creating or annihilating an initial quasi - particle is straightforward @xmath85 here , a bar shall denote complex conjugation , e.g. , @xmath86 . the coefficients obey the initial values @xmath87 where the phase has been appropriately chosen and the frequencies @xmath88 . the upper index of the coefficients @xmath89 and @xmath90 labels the mode , while the lower index denotes the components of this particular mode when expanded in a certain basis , e.g. , the initial eigenfunctions @xmath91 . since the coupling matrices become non - diagonal even for slow ( adiabatically ) variation of the trap frequency @xmath6 or coupling strength @xmath81 , the initial bases @xmath91 can not represent the eigenmodes at later times . although @xmath68 might be employed in order to calculate the spatial correlation functions , see appendix [ secobservables ] , the use of these functions might be misleading regarding the correlations between different modes . furthermore , when probing the excitations using , e.g. , the scheme proposed in @xcite , the proper particles defined at the time of measurement will be detected and not the initial ones . the particle definition in time - dependent background is a non - trivial task , see , e.g. , @xcite . nonetheless , it is always possible to expand @xmath62 @xmath92 into bases @xmath93 , which are defined such that the coupling matrices and , cf . , become diagonal at any particular instant @xmath94 @xmath95 of course , the velocity term @xmath96 is then generally non - diagonal and the evolution equations of the different modes will not exactly decouple at this particular instant @xmath94 . but one should bear in mind that measurement occurs usually in an adiabatic region , where the external parameters are only slowly - varying functions of time . then , the background phase is approximately homogeneous and the velocity @xmath97 . hence , quasi - particle creators and annihilators might be introduced analogous to eq . @xmath98 where @xmath99 . in the previous sections , i made no approximations except for the linearization and the assumption of an isotropic trap . the formalism is , in principle , applicable for arbitrary variations of trap frequency @xmath6 and interactions @xmath81 . to this end , it would be necessary to solve the gross - pitaevskii equation and the linear evolution equations simultaneously . the numerical solution is complicated by the fact that the coupling matrices and need to be calculated at each time step . some of the numerical difficulties can be circumvent by adopting the thomas - fermi profile , where density and phase of the background become time - independent ( in the coordinates @xmath38 ) and thus require the calculation of the coupling matrices only once . despite some shortcomings regarding the dynamics of the order parameter , this approximation is usually applicable for repulsive interactions and in the center of the trap , but becomes inaccurate towards the surface of the condensate , where the quantum pressure @xmath100 is relevant . within the thomas - fermi approximation , the coordinate transformation @xmath101 associated with the scaling transformation renders the background density time - independent @xmath102 , while the phase becomes homogeneous @xmath103 and thus @xmath104 . the integrals of the coupling matrices simplify considerably and the evolution equations can be cast into the form @xmath105 i.e. , @xmath106 and @xmath107 can be split into time - independent diagonal parts , cf . @xmath108 and constant , symmetric , non - diagonal coupling matrices @xmath109 with time - dependent prefactors @xmath110 . the external variation of trap frequency @xmath6 and coupling strength @xmath81 is solely encoded in the scale factor @xmath8 and the scalar function @xmath111 . note also that the coefficients and thus also the evolution equations for the fluctuations are independent of the initial coupling strength @xmath13 . the addend @xmath112 appearing in the parentheses of eqs . and can be expressed through the chemical potential @xmath113 , where @xmath114 , cf . the thomas - fermi equation . the mode functions @xmath68 of excitations with low energies , @xmath115 , are localized inside the condensate . hence , changing the bounds of the integrals and from infinity to the thomas - fermi radius @xmath116 will not alter these matrix elements significantly . bearing further in mind that for @xmath117 , it follows @xmath118 and . also because of the restriction to low energies , , and one gets @xmath119 . hence , the evolution equations of different eigenmodes approximately decouple and i obtain second - order equations of motion for phase @xcite @xmath120 \delta\hat \phi_n \ , = \ , 0 \ , , \label{effkfgmode } \end{aligned}\ ] ] and density fluctuations @xmath121 \delta\hat\varrho_n \ , = \ , 0 \ , , \label{effkfgdensitymode } \end{aligned}\ ] ] where i also introduced proper time @xmath16 . equation is the evolution equation of a mode of a minimally - coupled massless scalar field in a friedman - lematre - robertson - walker spacetime @xcite , provided the scale factor @xmath122 of the space - time is identified with @xmath123 , cf . @xcite @xmath124 [ note that the prefactor of the damping term is @xmath125 in eq . , while it is @xmath1 in a @xmath126-dimensional friedman - lematre - robertson - walker spacetime . ] the analogy is not restricted to the evolution equations in mode expansion but applies in the low - energy limit of the field equations as well : within the thomas - fermi approximation , @xmath127 for @xmath117 and @xmath61 , and for low excitations , @xmath128 , the phase fluctuations obey a second - order field equation @xcite @xmath129 \delta\hat\phi \ , = \ , 0 \ , , \label{effkfg } \end{aligned}\ ] ] which is similar to that of a minimally coupled scalar field in a friedman - lematre - robertson - walker spacetime . having established this kinematical analogy , cf . , some of the concepts of general relativity can be applied to time - dependent bose - einstein condensates . sonic analogs of horizons @xcite are of particular interest for the study of non - equilibrium effects , because they give a rough estimate whether and when adiabaticity will be violated and the ( quantum ) fluctuations freeze and get amplified , i.e. , ( quasi-)particle production occurs . an effective particle horizon occurs , if a phonon emitted at a time @xmath130 can only travel a finite ( co - moving ) distance , i.e. , if the integral @xmath131 converges to a finite value @xmath132 for @xmath133 . wavepackets emitted at time @xmath130 at the origin @xmath134 can reach only points within the horizon , @xmath135 , in a finite time . all other points are concealed by the horizon . [ for simplicity , i assumed in eq . an homogeneous sound velocity @xmath136 . ] in order to point out the analogy of phase fluctuations to cosmic quantum fields , i formulated the evolution equations and using proper time @xmath137 . on the other hand , experiments are usually performed in the laboratory and thus the variations of trap frequency @xmath138 and coupling strength @xmath139 are prescribed in laboratory time @xmath140 . since @xmath137 is a complicated function of @xmath140 , it is not quite obvious whether or not the quantum fluctuations will experience non - adiabatic evolution for a given modulation of @xmath6 or @xmath81 . in laboratory time @xmath140 , eq . reads @xmath141 \delta\hat\varrho_n \ , = \ , 0 \ , , \label{effkfgdensitymodet } \end{aligned}\ ] ] which is the evolution equation of a damped harmonic oscillator with time - dependent coefficients @xmath142 and @xmath143 initially , when @xmath59 , the field modes perform free oscillations . upon the gradual increase of the damping term @xmath144 with respect to the oscillation frequencies @xmath145 , the non - adiabatic evolution of the quantum fluctuations slowly sets in , until they finally freeze and get amplified when both terms , @xmath142 and @xmath145 are of the same order @xcite . let me discuss the two extremal ways a time - dependent scale factor @xmath8 can be achieved : firstly , only the trap frequency might be varied , while the interaction strength @xmath146 . with instantaneous frequencies @xmath147 , adiabaticity can be violated for any finite change @xmath148 if @xmath149 becomes sufficiently large . the quantum fluctuations can not adapt to the changing background any more , they freeze and get amplified . and , secondly , for static traps , @xmath150 , where @xmath81 is time - dependent . then , the situations is not so clear because @xmath151 , cf . hence , only the rapid acceleration of the scale factor @xmath152 will lead to notable changes of the excitation frequencies . on the other hand , adiabaticity could be violated by increasing the magnitude of the damping term , @xmath142 . then , however , a continuous acceleration of @xmath149 is required , because otherwise , if @xmath153 was constant , the system would equilibriate . as an example for the absence of particle production inside a static trap , @xmath150 , i will consider an exponential sweep of the coupling coefficient @xmath154 with @xmath155 . in this case , eq . for the scale factor permits an analytic solution @xmath156^{\frac1{2+d } } \exp\left\{\frac{\gamma t}{2+d}\right\ } \ , . \label{bexpsweepg } \end{aligned}\ ] ] the coefficients of eq . become time - independent @xmath157 and the density eigenmodes are just damped harmonic oscillators with solutions @xmath158 where @xmath159 is some residual oscillating function with frequency @xmath145 . hence , the density - density fluctuations diminish @xmath160 , and , consequently , the phase - phase fluctuations increase . but this is just the adiabatic evolution , because @xmath161 and thus @xmath162 , cf . [ secobservables ] . this means that no quasi - particle production occurs for the dynamics in the hydrodynamic regime , i.e. , for low - energy excitations with order parameter treated in the thomas - fermi approximation . these findings can also be inferred from the ( absence of an ) effective particle horizon . for the particular shape of the scale factor @xmath149 follows @xmath163 because @xmath164 . note that the presented solution assumes @xmath165 at all times , especially also for @xmath166 . hence @xmath167 at some time @xmath168 when @xmath169 , cf . eqs . and . on the other hand , @xmath170 and @xmath171 for a condensate at rest , see eq . . since both solutions ( static initial state and exponential sweep ) can not be matched at @xmath172 such that @xmath139 and @xmath149 are both continuous , the switching on of the exponential sweep would excite breathing oscillations . these oscillations , however , generally affect particle production , e.g. , through parametric resonance . the simplest application of the presented formalism consists in a quasi - one - dimensional condensate . in highly anisotropic traps , where the perpendicular trap frequency @xmath173 is much larger than the chemical potential , the motion in the perpendicular directions is restricted to the ground state and might be integrated out . an effectively one - dimensional field equation follows , where the interaction strength @xmath174 can be varied through feshbach resonance or by changing @xmath173 . however , one should bear in mind that the transversal extent of the condensate @xmath175 has to be much larger than the @xmath4-wave scattering length @xmath5 such that the interaction of different atoms can still be described through three - dimensional scattering theory . for simplicity , i will adopt in this section the thomas - fermi approximation for the order parameter but will permit arbitrary energies for the excitations . calculated using 100 harmonic oscillator basis functions . the frequencies equal the chemical potential at the intersection with the black line . the lowest odd mode converges for @xmath176 to the classical thomas - fermi breathing frequency @xmath177 ( green ) , which follows from eq . . , scaledwidth=40.0% ] in figure [ spectrum1d ] , the excitation frequencies @xmath178 of the lowest modes are plotted versus the chemical potential @xmath179 . for @xmath180 , one has the equidistant spectrum of the harmonic oscillator @xmath181 . whereas for high @xmath182 , the frequencies become almost independent of the chemical potential . in particular , the frequency @xmath183 of the lowest excitation with odd parity tends for @xmath176 towards the thomas - fermi breathing frequency @xmath184 obtained from eq . . the discrepancy between these two frequencies @xmath183 and @xmath185 for finite values of the chemical potential @xmath179 hints at shortcomings of the thomas - fermi approximation . in particular , eq . does not describe the breathing motion of the background properly and care must be taken when employing eqs . for the order parameter . nonetheless , eq . still predicts the correct order of magnitude of the characteristic response time of the background , @xmath186 , to variations of trap frequency or interaction strength . hence , it is still possible to discuss several cases , where the difference of @xmath187 and @xmath188 is either small or does not matter : firstly , if the shape of the condensate varies only slowly , i.e. , if @xmath189 and the condensate can adapt to changes of @xmath138 and/or @xmath139 immediately . secondly , if no breathing oscillations are excited , e.g. , because the condensate expands or contracts , see also eqs . and . and , thirdly , for very large chemical potentials , @xmath176 , the thomas - fermi approximation becomes exact . the quantum fluctuations are in the hydrodynamic regime and their effective evolution equations decouple . in order to address cases where breathing of the background occurs , it would be necessary to abandon the thomas - fermi approximation and to solve the gross - pitaevskii equation , which could , e.g. , be done by expanding the order parameter @xmath25 into oscillator functions @xcite . a stationary condensate , i.e. , @xmath58 , can be accomplished through simultaneous variation of trap frequency @xmath138 and interaction strength @xmath139 , cf . @xmath190 though , one should be aware that this only holds within the thomas - fermi approximation : the instantaneous chemical potential must at all times be much larger than the trap frequency @xmath191 if both were of the same order , the kinetic term , @xmath192 , in the gross - pitaevskii equation becomes important ; no stationary background could be realized even for simultaneous variation of @xmath138 and @xmath139 . for an exponential sweep @xmath193 the effective second - order equation for the phase fluctuations in the hydrodynamic limit @xmath194 \delta\hat\phi \ , = \ , 0 \ , . \label{effkfgexp } \end{aligned}\ ] ] is that of a massless scalar field in a de sitter spacetime with exponentially growing scale factor @xmath195 , cf . eq . which is believed to describe the universe during the epoch of cosmic inflation @xcite for the time - dependence , the integral is finite and an effective sonic horizon occurs ; the quantum fluctuations freeze and get amplified . instead of solving eq . for @xmath196 , i will consider the evolution equation of the density fluctuations @xmath197 @xmath198 \hat x_n^+ \ , = \ , 0 \ , , \label{densitysweepeom } \end{aligned}\ ] ] where @xmath199 . obviously , all modes undergo the same evolution just at different times . eq . can be solved analytically in terms of bessel functions @xcite @xmath200 where @xmath201 . the hankel functions @xmath202 have the proper asymptotics for early times @xmath203 such that the operators @xmath84 annihilate the initial vacuum state . the phase fluctuations read @xmath204 from these expressions and , i can infer the correlations of each mode . at late times follows @xmath205 comparison with the adiabatic values @xmath206 , cf.eq . , yields the quasi - particle number at late times @xmath207 the occupation number of all modes grows exponentially though at different times @xmath208 , where the shift @xmath209 is determined by the excitation frequencies @xmath210 . however , one should bear in mind that eq . is only valid for a limited time before leaving the hydrodynamic regime . in order to go beyond the effective space - time description and thus the analytical findings - , i will now consider the full evolution equations . the sweep rate @xmath211 shall be chosen such that all modes evolve adiabatically at first , i.e. , @xmath212 for all @xmath213 . when subsequently reducing trap potential and coupling strength , the excitation frequencies @xmath214 decrease and non - adiabatic evolution sets in at different times for each mode . the fluctuations freeze and get amplified . figure [ sweep50 ] shows the instantaneous particle numbers of the lowest three modes for initial chemical potential @xmath215 and @xmath211 . the lowest excitation , @xmath216 , which becomes non - adiabatic first , acquires the largest particle number . the next two modes , @xmath217 , experience less squeezing , though , remarkably , @xmath218 an unexpected result , which can be explained by the coupling of different modes : the second even mode , @xmath219 , gets populated from the principle excitation , @xmath216 , whereas the coupling matrix elements between @xmath216 and @xmath220 are zero because of different parity . and the sweep rate @xmath211 . at @xmath221 , the chemical potential equals the trap frequency . for the calculation of the eigenmodes @xmath68 and the coupling matrices , i used the harmonic oscillator functions @xmath222 up to @xmath223 , cf . appendix [ seceigenmode ] . the lowest 20 modes @xmath224 were then propagated . the numerical accuracy was set to @xmath225 . , scaledwidth=40.0% ] the main objective of this article was the investigation of quantum fluctuations in time - dependent harmonically - trapped bose - einstein condensates with repulsive interactions . to this end , the linear fluctuations were expanded into their initial eigenmodes and the field equations were diagonalized . this diagonal form , however , persists only as long as the condensate is at rest ; as soon as trap frequency or interaction strength are varied , the coupling of different modes sets in . ( only part of which can be accounted for by transforming to the instantaneous eigenmodes , though the definition of instantaneous eigenmodes is a non - trivial task . ) two regimes were identified : firstly , for energies much smaller than the chemical potential , the coupling of different modes is negligible and an effective space - time metric might be introduced for the phase fluctuations . this , however , necessitates a redefinition of the time coordinate such that the required change of trap frequency and/or interaction strength for a certain dynamics of this effective space - time is not obvious . it turned out that the sole variation of the interaction coefficient @xmath81 in a smooth monotonic way is hardly sufficient to render the evolution of the quantum fluctuations non - adiabatic , since the expansion / contraction of the background might compensate for changes of @xmath139 such that the sound velocity ( in comoving coordinates ) remains constant . breathing oscillations of the background , excited , e.g. , by the sudden change of the interaction coefficient @xmath139 , might still yield a notable amount of quasi - particles . and secondly , if the excitation energy is of the same order as the chemical potential , different quasi - particle modes couple . the amplification and freezing of the fluctuations and also the coupling of different modes was illustrated in an example , where the trap frequency and interaction strength were exponentially ramped down such that the shape of the condensate remains constant . for the considered parameters , a quasi - particle number of 0.5 was obtained in the lowest mode , though higher occupation numbers could be achieved by faster sweep rates or starting with a higher chemical potential . the inversion of the occupation number in the next two modes could be attributed to the inter - mode coupling : although the third excitation , @xmath219 , experiences a much briefer period of non - adiabatic evolution than the second mode , @xmath220 , only the former couples to the lowest mode , @xmath216 , and gets populated from it . i would like to thank craig m. savage and uwe r. fischer for helpful discussions . this research was supported by the australian research council . the aim of this appendix is the derivation of the initial eigenfunctions @xmath68 of density and phase fluctuations . to this end , let me expand the field operator into any orthonormal basis @xmath226 of the underlying hilbert space @xmath227 with some operator - valued coefficients @xmath228 @xmath229 lower - case greek indices ( @xmath230 , @xmath231 , ... ) shall denote the components in this arbitrary basis , while lower - case latin indices ( @xmath232 , @xmath213 , ... ) label the initial eigenmodes . if the condensate is initially at rest , the scale factor and the phase of the order parameter is homogenous @xmath233 such that @xmath61 and thus @xmath234 , cf .. the evolution equations simplify considerably and can be expanded into the basis @xmath235 . it follows , cf . @xmath236 where @xmath237 are defined in eq . . the matrices @xmath238 and @xmath239 are real and symmetric but do not commute due to @xmath240 \neq 0 $ ] , cf . . from eqs . , i obtain second - order evolution equations for phase and density fluctuations @xmath241 which can be diagonalized by a transformation to the eigenvectors . because the matrix @xmath242 is not symmetric , phase and density fluctuations obey different eigenvalue equations @xmath243 where @xmath213 denotes the vectors and @xmath230 merely counts the components in the particular basis representation . for brevity , i will only discuss the eigenvectors of @xmath244 in the following , though the same applies for those of @xmath245 as well . the eigenvectors are generally not orthogonal @xmath246 but @xmath247 usually still forms a basis . ( note , however , that the eigenvectors of a non - symmetric matrix not always span the entire vector space . but if the @xmath248 were no basis , the evolution equation could not be diagonalized . this would mean that there existed some fluctuations , which have constant losses to some eigenmodes rather unphysical in view of the stationary initial state considered here . therefore , i will not discuss this case any further . ) with the @xmath249 forming a basis of @xmath227 , there must exist a dual basis @xmath250 in the space of linear functionals ( the dual ) on @xmath227 , which obeys @xmath251 [ roughly speaking , the elements of @xmath227 become column vectors in the basis expansion , whereas row vectors correspond to the functionals on @xmath227 , i.e. , the elements of the dual . since the eigenvectors @xmath248 form a basis , the matrix with the @xmath248 as columns must be invertible . the rows of the ( left ) inverse matrix then comprise the elements of the dual basis @xmath250 in the particular basis expansion . ] after the multiplication of the first of eqs . with @xmath252 from the left and summation over @xmath230 follows @xmath253 which , because @xmath254 is a basis , implies that the @xmath255 are the left eigenvectors of @xmath256 with the same eigenvalues @xmath257 @xmath258 transposition yields @xmath259 i.e. , the eigenvalue equation of @xmath245 , cf . . hence , @xmath260 and @xmath261 and @xmath262 ( i.e. , density and phase fluctuations ) must have the same spectrum @xmath263 . for simplicity , @xmath264 , which can be achieved by renormalization of @xmath265 , in the following . note that the spectrum of a real non - symmetric matrix might contain pairs of complex conjugate eigenvalues @xmath257 , @xmath266 , which can be seen when taking the complex conjugate of . complex eigenvalues @xmath267 are associated with exponentially growing solutions , i.e. , unstable modes . since i am interested in the quantization of the stationary initial state , i will not discuss this case but instead assume @xmath268 @xmath269 . ( as can be easily verified , the components of the eigenvectors @xmath270 and @xmath271 must be real - valued as well . ) nonetheless , complex eigenvalues might still occur in dynamical situations , e.g. , during the signature change event proposed in @xcite or during ( quantum ) phase transitions @xcite ( see also ref . @xcite for an illustrative example ) . the initial evolution equations can be diagonalized by multiplication with the left eigenvectors @xmath272 which leads to the definition of the density and phase fluctuation eigenmodes @xmath224 through @xmath273 where the spatial mode functions @xmath274 follow from comparison of eqs . with and the duality of the @xmath270 and @xmath271 implies eq . for the @xmath68 . in view of , the initial evolution equations can be transformed to the new basis , cf.eq . @xmath275 where the transformed matrices @xmath276 have become diagonal . note that the transformation is not orthogonal because the matrices comprising of the eigenvectors @xmath270 and @xmath271 are not orthogonal . hence , the commutators are _ not _ preserved , in particular @xmath277 where the latter can be inferred from eq . @xmath278 since @xmath106 and @xmath107 also commute with their product @xmath279 , which is diagonal , they must be diagonal , too . the duality condition does not fix the norm of the eigenvectors but still permits the multiplication with an arbitrary factor @xmath280 @xmath281 this renormalization then leads to a stretching / shrinking of the operator - valued coefficients , cf . @xmath282 since this is merely a basis transformation , the time - evolution of the quantum fluctuations @xmath224 must be unaffected . to see this , recall the definitions and of the coupling matrices : they are the matrix elements of the operators @xmath237 and of @xmath283 with respect to the basis functions @xmath68 and thus acquire additional factors as well . as expected , all of these factors cancel such that the time - evolution remains unchanged . in particular , the eigenfrequencies are invariant @xmath284 . furthermore the observables should not depend on the particular normalization of the basis functions . the relative density - density correlations at time @xmath140 read @xmath285 where i used the instantaneous eigenmode basis @xmath286 , see eq .. obviously , the factors @xmath280 and @xmath287 contributed by @xmath286 and @xmath288 cancel each other and the spatial correlations are independent of the normalization @xmath280 . similarly , the expression for the spatial phase - phase correlations @xmath289 yields the same result regardless of the employed basis . hence , the factors @xmath280 can be chosen at will . there exist , however , several convenient choices for @xmath280 : firstly , the density modes might be normalized to unity , @xmath290 . this is advantageous if left and right eigenvectors are the same , i.e. , if @xmath244 is symmetric . the drawback is that if @xmath244 is not symmetric and therefore @xmath291 , only one of the mode functions , @xmath292 , can be normalized to unity , whereas the norm of the @xmath293 follows from eq . . and , secondly , these factors @xmath280 can be fixed by demanding @xmath294 . in this case , the prefactors of initially obey @xmath295 and @xmath296 and both quadratures @xmath297 . however , one should note that neither of the eigenfunctions @xmath68 is generally normalized to unity , @xmath298 . of course , the coefficients @xmath299 and @xmath300 do depend on the particular choice of the basis functions @xmath301 and thus also on the factors @xmath280 . for instance , the adiabatic density - density and phase - phase correlations of a particular mode read @xmath302 the dependence on the factor @xmath280 becomes important regarding the low excitations in the thomas - fermi approximation , cf . [ sectf ] : the modes decouple and it is not necessary to introduce a new spatial basis @xmath68 at the time of measurement . one has instead @xmath303 and @xmath304 such that @xmath305 while @xmath306 , i.e. , the phase and density fluctuations apparently increase or decrease even for adiabatic evolution . in view of this @xmath280 ambiguity , the ( absolute ) density - density or phase - phase correlations provide no adequate measure for the squeezing ( i.e. , non - adiabaticity ) of a single mode . the fourier transforms of eqs . and on the other hand are independent of @xmath280 but do not represent the excitation eigenmodes . as will be shown in the following , the ( instantaneous ) quasi - particle number measures the relative deviation of density and phase fluctuations from their adiabatic values . in view of the different expansions and of density and phase fluctuations into their initial and adiabatic eigenfunctions , the corresponding creation and annihilation operators @xmath83 , @xmath84 and @xmath307 , @xmath308 , respectively , can be transformed by virtue of a bogoliubov transformation . for the annihilators @xmath308 follows in particular @xmath309 with the bogoliubov coefficients @xmath310 and @xmath311 . the first line follows from inversion of eq . and the second line can be inferred after transforming to the new basis @xmath301 . since @xmath312 for non - adiabatic evolution , the quasi - particle number operator @xmath313 will have a non - zero expectation value as well @xmath314 \right\rangle \end{aligned}\ ] ] where the commutator @xmath315 = -i$ ] . noting that the prefactors @xmath316 and @xmath317 are just the adiabatic density - density and phase - phase correlations , see eq . , the particle number can be rewritten @xmath318 i.e. , the instantaneous particle number gives just the relative deviation of the density and phase correlations from their adiabatic values . note that expression does not contain the correlations between different modes . to this end , it would be necessary to evaluate @xmath319 , which is fourth order in the @xmath308 . this observable can be reduced to expectation values quadratic in in the @xmath308 by virtue of wick s theorem , see , e.g. , @xcite . another interesting aspect regards the scaling of the correlation functions and with the interaction strength : within the thomas - fermi approximation , see sec . [ sectf ] , the evolution equations are independent of @xmath13 . all properties of the linear excitations are determined by the chemical potential @xmath179 and the variations @xmath320 and @xmath321 , in particular the expectation values @xmath322 . only the normalization factor @xmath323[\mu_0 - v(\bm x')]}$ ] in eqs . and depends on @xmath13 . hence , the relative density - density and phase - phase correlations in the thomas - fermi approximation are both proportional to @xmath13 for fixed @xmath179 . a. fisher , p. b. weichman , g. grinstein , and d. s. fisher , phys . b * 40 * , 546 ( 1989 ) ; d. jaksch _ et al . _ , phys . lett . * 81 * , 3108 ( 1998 ) ; d. jaksch and p. zoller , ann . ( n.y . ) * 315 * , 52 ( 2005 ) ; r. schtzhold , m. uhlmann , y. xu , and u. r. fischer , phys . rev . lett . * 97 * , 200601 ( 2006 ) ; c. kollath , a. m. luchli , and e. altman , phys . rev . * 98 * , 180601 ( 2007 ) . c. barcel , s. liberati , and m. visser , phys . a * 68 * , 053613 ( 2003 ) ; c. barcel , s. liberati , and m. visser , int . j. mod . d * 12 * , 1641 ( 2003 ) ; c. barcel , s. liberati , and m. visser , class . quant . grav . * 18 * , 1137 ( 2001 ) ; _ ibid . _ * 18 * , 3595 ( 2001 ) . dimensionless units can be introduced through the initial trap frequency @xmath324 and the atom mass @xmath232 . all energies are then measured in units of @xmath324 , the dimensionless time is defined through @xmath325 , and the characteristic length @xmath326 , where @xmath327 . e. tiesinga , b. j. verhaar , and h. t. c. stoof phys . rev . a * 47 * , 4114 ( 1993 ) ; s. inouye _ et al . _ , nature * 392 * , 151 ( 1998 ) ; ph . courteille , r. s. freeland , and d. j. heinzen , phys . lett . * 81 * , 69 ( 1998 ) . as similar expansion of the linear fluctuations is also considered in @xcite . there , the authors expand the linear field operator @xmath27 of a static / stationary condensate into eigenmodes and not the phase and density fluctuations @xmath62 . this leads to complex mode functions @xmath328 and @xmath329 , while @xmath68 are real . the eigenfunctions of density and phase fluctuations @xmath330 and @xmath45 are defined slightly differently than those of @xmath62 @xmath331 noting that @xmath332 the center of the trap , the eigenfunctions @xmath292 and @xmath333 ( and similarly @xmath293 and @xmath334 ) must be approximately proportional for low excitations . hence , the @xmath335 for @xmath336 indeed represent the proper density and phase eigenmodes @xmath337 and @xmath338 up to a constant prefactor @xmath339 . l. e. sadler _ et al . _ , nature * 443 * , 312 ( 2006 ) ; h. saito and m. ueda , phys . a * 72 * , 023610 ( 2005 ) ; h. saito , y. kawaguchi , and m. ueda , phys . lett . * 96 * , 065302 ( 2006 ) ; a. lamacraft phys . lett . * 98 * , 160404 ( 2007 ) ; m. uhlmann , r. schtzhold , and u. r. fischer , phys . * 99 * , 120407 ( 2007 ) .
the skyrme model is a classical field theory which describes atomic nuclei and their low energy interactions @xcite . it is a non - linear theory admitting topological solitons , so - called skyrmions , labelled by a topological charge @xmath4 which can be identified with the baryon number . when the theory is quantized the skyrmions give a good description of nuclei , @xmath5 resonance @xcite and even bound states of nuclei @xcite . a striking feature is that skyrmions can be quantized as fermions @xcite . when the theory is coupled to a fermion field two different ways of describing fermions appear in the same theory . physically , the fermion field can be thought of as light quarks in the presence of atomic nuclei @xcite . hiller calculated the spectrum of the fermions in the background of a skyrmion @xcite . however , the approximation for the skyrmion configuration was very crude . the spectrum shows a spectral flow of the eigenvalues depending on the value of the coupling constant . in particular one mode crosses zero . this general behaviour has been verified by other authors @xcite . to gain a better understanding we consider a generalization of the model . since static field configurations in the original skyrme model in flat space are topologically equivalent to maps @xmath6 , the model can be generalized to the base space being a sphere of radius @xmath7 . the original model is recovered in the limit @xmath8 . it turns out that for @xmath9 and baryon number @xmath10 the faddeev - bogomolny bound is satisfied and the minimal energy solution can be worked out explicitly @xcite . in this case the map @xmath6 is an isometry . for larger @xmath7 the skyrmions are localized and tend to the usual flat space skyrmions in the limit @xmath8 . monopoles and skyrmions have many properties in common , and it is therefore instructive to recall the results for monopoles . there is an index theorem making use of the bogomolny equations which states that for @xmath11 monopoles there is a @xmath11 dimensional vector space of fermionic zero modes @xcite . furthermore , the monopole and its fermionic zero mode can be calculated explicitly for @xmath12 . we will examine how far we can extend the analogy between monopoles and skyrmions . in this paper we study fermions coupled to skyrmions on @xmath13 . the dirac equation on @xmath13 is derived in sect . [ diracequation ] . we also explain the skyrmion - fermion interaction . in sect . [ diracsolution ] , we derive the ordinary differential equation which describes the solutions with total angular momentum @xmath2 for general shape functions . in sect . [ differentialequations ] , we give a brief discussion of ordinary differential equations and their regular singular points . in sect . [ constantskyrmion ] , the equation of sect . [ diracsolution ] is solved for a constant skyrme configuration , which is equivalent to a mass term for the fermions . we also discuss the boundary conditions which are necessary to derive the spectrum . in sect . [ hedgehogskyrmion ] , we derive the eigenfunctions and the spectrum for the hedgehog configuration with the special shape function @xmath14 . in sect . [ numerics ] the results of sect . [ constantskyrmion ] and [ hedgehogskyrmion ] are verified numerically . the spectrum for more general shape functions is also evaluated numerically and compared with results in the literature . we find an interesting relationship between the topological charge of the skyrmion and certain singular points . we end with a conclusion . in this section we derive the dirac equation on a @xmath0-sphere of radius @xmath9 . first we define an atlas of two charts @xmath15 and @xmath16 which are the stereographic projections from the north pole @xmath17 and the south pole @xmath18 , respectively . next we define vierbeins on both charts which uniquely determine a connection one - form using the metric compatibility condition and the torsion - free condition . the connection one - form induces the spin connection so that we can finally write down the dirac equation for each chart . at the end of this section we discuss the interaction term which couples fermions and skyrmions . in order to construct the stereographic projection let @xmath13 be embedded in euclidean @xmath19 with coordinates @xmath20 . the points of @xmath13 can now be labelled by the projection from the north pole @xmath17 onto the @xmath21 plane through the equator introducing the coordinates @xmath15 . alternatively , one can project from the south pole @xmath18 introducing the coordinates @xmath16 . these two charts are well defined everywhere apart from their projection point @xmath17 and @xmath18 , respectively . in terms of the @xmath19 cartesian coordinates we obtain @xmath22 in the following , we derive the equations of motion for both charts . as we shall see in sect . [ constantskyrmion ] , this enables us to discuss the regularity conditions at the north and south pole in a simple way . we will also use the two charts to discuss a nontrivial symmetry of the energy spectrum in sect . [ generalization ] . the induced metric on @xmath13 can be calculated from the euclidean metric in @xmath19 . taking into account that our base space is a lorentzian space time @xmath23 , the metric can be written as @xmath24 the other coordinate system gives rise to the same metric replacing @xmath15 by @xmath16 . on the overlap the two charts are related via the transition functions @xmath25 therefore , the coordinate vectors @xmath26 and @xmath27 are related by @xmath28 note that this transformation has the determinant @xmath29 . we now choose non - coordinate bases @xmath30 and @xmath31 . it is convenient to choose diagonal vierbeins @xmath32 and @xmath33 such that @xmath34 where all other components vanish . the minus sign occurs because @xmath15 and @xmath16 are related by inversion which changes the orientation . the vierbeins satisfy @xmath35 where @xmath36 is the flat minkowski metric . the same relations are true for the vierbeins @xmath33 . the non - coordinate basis vectors are related by an @xmath37-transformation @xmath38 this is in fact a @xmath39-rotation about the vector @xmath40 . with this choice of vierbeins the connection one - form @xmath41 can be calculated from the metric compatibility condition @xmath42 and the torsion - free condition @xmath43 where @xmath44 is the dual basis of @xmath45 . after a short calculation we obtain @xmath46 and the same relation holds for @xmath47 expressed in @xmath16-coordinates . the spin connection @xmath48 is now given by @xmath49 where @xmath50 $ ] , and the @xmath51-matrices satisfy the standard anticommutation relations @xmath52 . the lagrangian for fermions in curved space time is given by @xmath53 with our choice of coordinates and vierbeins we obtain @xmath54 using the coordinates @xmath15 and @xmath55 for the coordinates @xmath16 . equation ( [ l2 ] ) can be transformed into ( [ l1 ] ) by using @xmath56 and the induced spinor transformation @xmath57 of the fermion field @xmath58 which we calculate in the following . the spinor transformation @xmath57 is defined as @xmath59 the transformation @xmath57 is a lifting of the local lorentz transformation @xmath60 of the vierbeins in ( [ erotation ] ) . note that the transformation @xmath61 can be rewritten as @xmath62 for @xmath63 . here @xmath64 . since we are only concerned with the spatial part we work with the euclidean @xmath65 rather than with @xmath66 infinitesimally , the rotation is given by @xmath67 . this can be lifted to a spinor transformation by @xmath68 . for @xmath69 we obtain @xmath70 note that there is a sign choice . as the spinor transformation @xmath57 is an element of the double cover of the rotation group , the transformation @xmath71 is also a lifting of the rotation @xmath72 . in the following we discuss fermions coupled to skyrmions . the full lagrangian @xmath73 of this model is the sum of the fermion lagrangian @xmath74 , the skyrmion lagrangian @xmath75 and the interaction lagrangian @xmath76 . in this paper we consider fermions in the background of a static skyrme configuration and neglect the backreaction . therefore , we do not discuss the skyrmion lagrangian any further , and the interested reader is referred to @xcite . the interaction lagrangian @xmath76 given by @xmath77 where @xmath78 is a convenient parameterization of the skyrme field . equation ( [ lint ] ) is the standard parity - invariant manner in which fermions couple to a skyrme field and is also called the yukawa interaction term . this lagrangian was first introduced by gell - mann and levy ( without a fourth order skyrme term ) , @xcite . also see @xcite for a discussion of various different chiral models . recently , the lagrangian ( [ lint ] ) has been discussed by balachandran _ _ @xcite . note that @xmath79 is a spin and an isospin spinor . in our notation the pauli matrices @xmath80 act on the isospin indices , whereas the @xmath51-matrices act on the spin indices . it is convenient to split the dirac spinor into two @xmath81 spin isospin matrices @xmath82 and @xmath83 such that @xmath84 in this notation the @xmath51-matrices can be written in terms of quaternionic @xmath81 matrices using @xmath85 times the spin pauli matrices @xmath86 and the two dimensional identity matrix @xmath87 as a basis for the quaternions . since any complex @xmath81 matrix can also be expressed in terms of quaternions with complex coefficients it is convenient to choose the same basis for the quaternions as before . then the spin isospin matrices @xmath88 can be written as @xmath89 . with this notation the spin operators act on @xmath79 by left - multiplication , @xmath90 , whereas the isospin matrices act on @xmath79 by right multiplication @xmath91 where the transpose in ( [ quat1 ] ) guarantees the group structure , and ( [ quat2 ] ) gives a useful trick to evaluate @xmath92 . as mentioned above , we make the approximation that the skyrmion is a fixed static background field , _ i.e. _ we ignore the backreaction of the fermion fields . furthermore , we only consider spherically symmetric skyrmions . for @xmath10 , skyrmions are believed to be spherically symmetric , but for @xmath93 this is no longer true . however , we expect that some generic features of the skyrmion - fermion interaction are not sensitive to the precise shape of the skyrmion . for more details see sect . [ generalization ] . spherically symmetric skyrme fields are best written in polar coordinates @xmath94 where @xmath95 is the `` radial '' shape function and @xmath96 is the normal vector in @xmath97-direction , see ( [ unitvectors ] ) . the shape function @xmath95 can be calculated by solving the euler - lagrange equation of the skyrme lagrangian numerically . however , instead of using a numerical solution it is more convenient in this context to approximate @xmath95 by an explicit function . manton showed in @xcite that a good ansatz for @xmath7 not too large is the conformal ansatz given by the ansatz tends to @xmath98 . qualitatively , this is still a good ansatz . ] @xmath99 here @xmath100 is a constant which has to be determined by minimizing the energy of the skyrmion . this ansatz describes a skyrme configuration which is either localized around one of the poles or uniformly distributed . the value @xmath101 corresponds to the @xmath102-symmetric ansatz of the previous section , whereas for @xmath103 , or @xmath104 , the skyrme field is localized at the south pole , or the north pole , respectively . manton also showed that for @xmath105 the stable solution is @xmath101 . finally , we can write down the dirac equation for fermions coupled to a spherically symmetric background skyrmion . it is given by @xmath106 using the coordinates @xmath15 and @xmath107 using the coordinates @xmath16 . here the skyrmion gives rise to @xmath108 and the same equation holds in the coordinate system @xmath16 . in order to solve the dirac equation we introduce spherical polar coordinates for equation ( [ dirac1 ] ) and ( [ dirac2 ] ) . in terms of the coordinates @xmath109 in @xmath19 , the polar coordinates for the @xmath0-sphere of radius @xmath110 can be written as @xmath111 the angles @xmath112 can now be expressed in terms of @xmath15 as @xmath113 in terms of the @xmath114 , the relations for @xmath115 and @xmath116 are identical to the ones above , and for the @xmath97 coordinate we obtain @xmath117 equation ( [ dirac1 ] ) gives rise to the following dirac equation in polar coordinates : @xmath118 where @xmath119 and the unit vectors @xmath120 are given by @xmath121 @xmath122 can be written as @xmath123 . similarly , equation ( [ dirac2 ] ) give rise to @xmath124 the spin connection term could be removed by setting @xmath125 and @xmath126 , respectively . however , it is easier to discuss the regularity of the solution in terms of the original fields . we adopt the standard representation of the @xmath51-matrices where @xmath127 is diagonal . this is convenient because the dirac equation is invariant under parity . @xmath128 now , the spinor transformation @xmath57 in ( [ rhospin ] ) can be evaluated explicitly : @xmath129 with this choice of @xmath51-matrices , we can derive the differential equation for the stationary solutions of the dirac equation by setting @xmath130 : @xmath131 where @xmath132 the components of the matrix in equation ( [ eq1 ] ) commute with the total angular momentum operator @xmath133 where @xmath134 is the orbital angular momentum operator , the spin operator is @xmath135 and the isospin operator is @xmath136 . in polar coordinates , the orbital angular momentum @xmath134 is given by @xmath137 equation ( [ eq1 ] ) is also invariant under parity @xmath138 with @xmath139 it is worth discussing the parity operator @xmath138 in more detail . @xmath138 acts on the @xmath51-matrices and the partial derivatives as follows : @xmath140 using the spinor transformation @xmath57 in ( [ rhospin ] ) we can show that @xmath141 therefore , the parity operator @xmath142 which is defined for the coordinates @xmath16 differs from @xmath138 by a minus sign . the @xmath143 positive parity spin isospin spinors are given by @xmath144 this gives rise to the following system of first order differential equations : @xmath145 where @xmath146 , @xmath147 and @xmath148 are functions of @xmath97 . note that the @xmath149 spinor @xmath150 gives rise to the same equation with @xmath151 replaced by @xmath152 . in @xmath16-coordinates positive and negative parity are interchanged such that the @xmath153 is given by @xmath154 this gives rise to @xmath155 again the @xmath156 spinor @xmath157 gives rise to the same equation with @xmath151 replaced by @xmath152 . equation ( [ general1 ] ) can be transformed into a second order differential equation for @xmath158 where @xmath159 . note that since @xmath160 $ ] it is consistent to set @xmath161 . @xmath162 @xmath163 and similarly equation ( [ general2 ] ) can be transformed into @xmath164 @xmath165 it is not easy to transform equation ( [ od1 ] ) and ( [ od2 ] ) into each other . in fact , due to the spinor transformation ( [ trafo ] ) @xmath166 and @xmath167 are related via @xmath168 also note that @xmath167 is given by @xmath169 and similarly @xmath170 can be written as @xmath171 in this section we solve the dirac equation which has been derived in the previous section for various shape functions @xmath95 . however , first we review some basic facts about ordinary differential equations and regular singular points , following @xcite . any linear second order differential equation can be brought into the standard form @xmath172 if @xmath173 and @xmath174 are analytic functions in the neighbourhood of @xmath175 then @xmath176 is called a regular point . in this case there are locally two linearly independent solutions @xmath177 and @xmath178 such that any solution of ( [ diffeq ] ) is a linear combination of @xmath177 and @xmath178 . moreover , @xmath179 is an analytic function , _ i.e. _ it can be written as @xmath180 if either @xmath181 or @xmath182 is singular then @xmath175 is called a singular point . in the special case that @xmath183 where @xmath184 and @xmath185 are analytic functions , _ i.e. _ @xmath186 the point @xmath175 is called a _ regular singular point_. equations which only have regular or regular singular points are known as fuchsian differential equations . equation ( [ diffeq ] ) can then be solved by the following ansatz : @xmath187 the exponent @xmath57 can be calculated by solving the so - called _ indicial equation _ @xmath188 which is derived by inserting the ansatz ( [ eqansatz ] ) into ( [ diffeq ] ) . the two solutions @xmath189 of the indicial equation ( [ indicial ] ) are called the _ exponents _ of the regular singular point @xmath190 . our convention will be that @xmath191 . if the difference of the exponents @xmath192 is not a positive integer or zero , then the ansatz ( [ eqansatz ] ) gives rise to two linearly independent solutions of ( [ diffeq ] ) which are uniformly convergent in some neighbourhood of @xmath190 . note , however , that the solution diverges at @xmath190 if its exponent @xmath57 is negative . let @xmath192 be a positive integer or zero . this situation is slightly more subtle because the two solutions can become linearly dependent , and this could lead to further logarithmic terms . the solution @xmath177 with exponent @xmath193 is always uniformly convergent in some neighbourhood of @xmath190 . therefore , we reduce equation ( [ diffeq ] ) to a differential equation of first order by setting @xmath194 . this gives rise to the following equation for @xmath195 : @xmath196 this equation can be solved by integration and separation of variables , and we obtain @xmath197 where @xmath198 is an analytic function in a suitable domain , and @xmath199 and @xmath200 are constants . a short calculation shows that @xmath201 let @xmath202 then for @xmath203 the most general solution is @xmath204 where @xmath199 and @xmath4 are arbitrary constants , and @xmath177 is the solution of the form ( [ eqansatz ] ) . the constant @xmath205 and the function @xmath206 are completely determined by the differential equation ( [ difeqz ] ) . the function @xmath207 is given by @xmath208 the coefficients @xmath209 being constants . in the special case that @xmath210 no logarithms occur , _ i.e. _ the two solutions @xmath177 and @xmath178 are linearly independent . if @xmath211 the corresponding form of the solution is @xmath212 it is also possible to discuss the singularity at infinity . the transformation @xmath213 maps the point at infinity to the origin . if we apply this transformation to equation ( [ diffeq ] ) we obtain a new differential equation in terms of @xmath214 . if the new equation has a regular or a regular singular point at the origin @xmath215 then ( [ diffeq ] ) has a regular or regular singular point at infinity . if a fuchsian differential equation has @xmath11 finite regular singular points @xmath216 with exponents @xmath217 and a regular singular point at infinity with exponents @xmath218 then the following equation holds : @xmath219 this is a useful consistency check for the exponents . we are interested in the behaviour of fermions in the presence of a background skyrmion . the skyrmion introduces a space - dependent complex mass for the fermions . therefore , it is instructive to consider first a massive fermion on @xmath13 . this corresponds to @xmath220 so that the fermion mass is @xmath151 . this example can be solved explicitly in terms of hypergeometric functions . in the following sections we proceed to more complicated configurations . in sect . [ hedgehogskyrmion ] , we consider the spherically symmetric shape function @xmath14 which can also be solved explicitly , but in a less standard way . in sect . [ generalization ] , the conformal shape function ( [ conformal ] ) is discussed . in this case , the differential equation can no longer be solved explicitly . we present numerical algorithms in sect . [ numericalalgorithm ] . setting @xmath221 , equation ( [ od1 ] ) can be written as @xmath222 where @xmath223 . equation ( [ od1f0 ] ) has three regular singular points , one at @xmath224 , one at @xmath225 and one at @xmath226 with exponents @xmath227 , @xmath228 and @xmath229 respectively . note that relation ( [ erelation ] ) between the exponents is satisfied , _ i.e. _ the sum of the exponents is equal to @xmath110 . in this example we will show that the solution is non - singular on @xmath13 if its exponents are @xmath230 and @xmath231 . equation ( [ od1f0 ] ) can be solved explicitly : @xmath232 where @xmath233 is a hypergeometric function , _ e.g. _ see @xcite . the hypergeometric function is non - singular for @xmath234 and is normalized at @xmath235 , _ i.e. _ @xmath236 . it may be singular at @xmath237 , corresponding to @xmath238 , which is the north pole . as @xmath158 has to be non - singular at the south pole , @xmath239 , we have to set @xmath240 . since @xmath158 is not well defined at @xmath224 we have to express @xmath158 in terms of @xmath166 and calculate @xmath241 . using the transformation ( [ trafo ] ) from @xmath242 to @xmath79 we obtain @xmath243 which is the analogue of equation ( [ fgtilde ] ) . setting @xmath244 , in equation ( [ od2 ] ) we can derive a similar equation for @xmath245 which has exponents @xmath246 , @xmath247 and @xmath248 . the solution of this equation can again be expressed in terms of hypergeometric functions : @xmath249 this is only non - singular at @xmath224 if we set @xmath250 . assuming that @xmath251 , @xmath170 can be calculated from equation ( [ tildef ] ) : @xmath252 therefore , @xmath158 can be written as @xmath253 @xmath166 has to be non - singular at @xmath224 . equation ( [ g(1)nonsing ] then implies that @xmath158 is also non - singular at @xmath224 . therefore , the regular solution @xmath158 has the exponents @xmath230 and @xmath231 . note that this argument only relies on the local behaviour and explicit solutions are not needed . as we shall see , in the generic case , the exponents at the singular points do not depend on the chosen shape function , so that our analysis is also valid for general shape functions . problems can occur if additional singular points coalesce with the singular points at @xmath254 because this changes the exponents .. ] in the generic case , the regular solution can be calculated by imposing that the solution can be expanded as a power series at the north and at the south pole . later , this will be the basis for our numerical algorithm . now , we impose this condition in order to calculate the spectrum of equation ( [ od1f0 ] ) . using formula ( 15.3.6 ) in @xcite we can express @xmath158 in terms of hypergeometric functions at the second singular point . ) only holds if @xmath255 . however , this inequality is satisfied because @xmath256 . ] @xmath257 the last term has to vanish , otherwise @xmath170 would be singular at the north pole . therefore , either @xmath258 or @xmath259 has to have a pole . since @xmath260 , we obtain @xmath261 , and the spectrum is therefore @xmath262 with this condition the eigenfunctions @xmath263 reduce to jacobi polynomials ( @xcite , formula 15.4.6 ) . the spectrum is displayed in figure [ fermion1 ] . it is symmetric with respect to @xmath264 . the importance of this result lies in the fact that for @xmath265 the energy @xmath266 and the corresponding solutions @xmath263 are the same for all possible shape functions @xmath95 . therefore , these solutions can be used as starting values for the relaxation method , which will be described in sect . [ numerics ] . as the radius of the @xmath0-sphere tends to infinity the spectrum tends to the usual spectrum of the dirac equation , which is continuous for @xmath267 . also note that if we assume that @xmath158 and its first and second derivative remain finite as @xmath151 tends to infinity then the energy @xmath268 has to be of order @xmath151 in this limit , so @xmath269 is the leading order asymptotic for @xmath268 . therefore , we displayed the lines @xmath270 in figure [ fermion1 ] . as we shall see in sect . [ numerics ] the region @xmath271 also plays a special role for our numerical algorithm . as a function of the coupling constant @xmath151 for @xmath244 . @xmath272 is shown as dashed lines . [ fermion1],height=472 ] another way of deriving the spectrum ( [ specf=0 ] ) is to note that if the exponents at @xmath273 both vanish then @xmath158 is an integral function , that is it has only one singular point at infinity . the only integral functions which do not possess an essential singular point at infinity are the polynomials , _ e.g. _ @xcite . therefore , one of the exponents at infinity is equal to @xmath274 for @xmath275 . since the exponents at infinity are @xmath276 we obtain the same spectrum as before . note however , that since the exponents @xmath277 differ by an integer , we have to check in this approach that the solution is indeed a polynomial , _ i.e. _ @xmath205 in equation ( [ gs ] ) vanishes . this spectrum has also been calculated by other authors , _ e.g. _ @xcite . in this section we construct the solutions of equation ( [ od1 ] ) for @xmath14 . configurations with @xmath278 are called `` hedgehog configurations '' because at each point the vector in isospin space is proportional to the coordinate vector . for @xmath279 the map from the base @xmath13 to the target @xmath13 is even an isometry and possesses @xmath102-symmetry . setting @xmath14 in equation ( [ od1 ] ) we obtain @xmath280 @xmath281 this is a homogeneous differential equation of fuchsian type with four singular points . they are located at @xmath273 , @xmath282 and @xmath283 . assuming @xmath284 the exponents are @xmath285 for @xmath286 the singular points at @xmath287 and @xmath288 coalesce , and therefore , the exponents at @xmath287 become @xmath289 and the exponent at infinity is @xmath290 . the relation for the exponents ( [ erelation ] ) is satisfied . for @xmath291 the singular points at @xmath224 and @xmath292 coalesce . this changes the exponents at @xmath224 to @xmath293 . in this case , there is an essential singular point at infinity and the equation is no longer fuchsian . since we are interested in solutions that can be expanded as a power series around @xmath273 , it is natural to look for polynomial solutions . the singular point at @xmath294 might impose a constraint . in order to verify whether there is a logarithmic term we have to calculate the coefficient @xmath205 in equation ( [ gs ] ) . here after a standard calculation we obtain @xmath296 . therefore , there are two independent power series at the point @xmath297 . this imposes no further restriction , since we do not have to exclude any of these two solutions . this implies that @xmath158 is again an integral function and we can calculate the energy spectrum using the exponents at infinity . again , the exponents @xmath277 differ by an integer , and we have to verify that the solutions are indeed polynomials . it turns out that the only problem occurs for @xmath298 . in the following , we derive the recurrence relation for these polynomials . yet , it is instructive to consider first the solution where @xmath158 is a constant @xmath299 , _ i.e. _ @xmath298 . then we can solve equation ( [ exk=1 ] ) for @xmath268 and obtain @xmath300 and @xmath301 . setting @xmath302 in equation ( [ g(u)k=1 ] ) , we obtain @xmath303 it is easy to see that this is not a solution for @xmath304 . @xmath302 is only a solution for arbitrary @xmath151 if @xmath305 . this solution is very interesting because its spectrum crosses @xmath264 for @xmath306 . now we derive the general solution for equation ( [ g(u)k=1 ] ) . since we are interested in the solutions for which the relevant exponent vanishes for @xmath287 , we expand @xmath158 around the south pole @xmath287 : @xmath307 assuming that @xmath308 , we derive the following recurrence relations . the initial condition is @xmath309 and the recurrence relation is given by @xmath310 since we have an initial condition we expect to be able to reduce this second order recurrence relation to a recurrence of first order . therefore , we make the ansatz @xmath311 substituting this ansatz into the recurrence relation ( [ recurrence ] ) we obtain a non - linear first order recurrence for the @xmath312 . however , this can be solved by setting @xmath313 by construction the series ( [ ansatza ] ) has the right behaviour at the singular point @xmath287 . to find the behaviour at @xmath224 we have to discuss the radius of convergence @xmath314 of the series . the radius @xmath314 can be calculated as the following limit : @xmath315 therefore , the series converges for @xmath316 . for @xmath224 , equation ( [ ansatza ] ) can be written as @xmath317 for large @xmath11 , @xmath312 behaves like @xmath318 . therefore , @xmath319 is proportional to @xmath320 so that the terms in the sum ( [ g(1 ) ] ) are of order @xmath110 and do not vanish . hence , the series is divergent , unless it terminates . but because of the special form of @xmath319 and @xmath312 it is easy to see when the series terminates . in fact , for @xmath321 we can now write @xmath319 as the following product : @xmath322 where @xmath323 is the product of all odd numbers less than or equal to @xmath324 . all @xmath319 vanish for @xmath325 which is the constant solution that we have derived previously . if the term @xmath326 vanishes then the coefficient for @xmath319 vanishes individually . however , if one of the terms in the product vanishes , say for @xmath327 , then not only @xmath328 vanishes but also all @xmath329 for @xmath330 , _ i.e. _ the series terminates . as functions of the coupling constant @xmath151 for @xmath14 @xmath272 is shown as dashed lines.[fermion2],height=472 ] from this condition the complete energy spectrum can be calculated . we obtain the special level @xmath331 which crosses from positive to negative energies , and two series of energy levels @xmath332 and @xmath333 which are given by @xmath334 the spectrum is displayed in figure [ fermion2 ] . if we only consider the modes with energy @xmath335 the spectrum is symmetric under reflection at the axes @xmath336 and @xmath337 . however , the mode @xmath338 with energy @xmath339 is only symmetric under combined reflections @xmath340 and @xmath341 , which corresponds to a @xmath39-rotation around the point @xmath342 . this spectrum agrees with our previous calculation where we used the fact that @xmath158 is an integral function . in this section we first describe numerical algorithms to calculate the spectrum for more general shape functions @xmath95 . then we discuss the spectrum for the conformal ansatz ( [ conformal ] ) . we also discuss the simple ansatz @xmath343 which gives rise to skyrmions with @xmath93 . finally , we compare our results with the literature . the energy spectrum can be calculated for various shape functions using a standard relaxation method @xcite . the differential equation ( [ od1 ] ) is rewritten as a system of first order differential equations . @xmath344 the energy @xmath268 is considered as an additional dependent variable whose derivative vanishes , @xmath345 , so that the complete system becomes @xmath346 for simplicity , let us discuss the numerical algorithm for @xmath102-symmetric shape function @xmath347 . equation ( [ g(u)k=1 ] ) has a regular singular point at @xmath287 which implies that at this point we can not choose value and derivative of the solution independently . however , once the value of @xmath348 is chosen we can calculate @xmath349 . in fact , inserting @xmath287 into equation ( [ g(u)k=1 ] ) we obtain @xmath350 similarly we obtain for @xmath224 , @xmath351 the system ( [ system ] ) is discretized such that the boundary conditions at the singular points @xmath273 are satisfied . starting from an initial configuration the relaxation algorithm calculates an improved configuration by linearizing the equation around the solution and minimizing an error function which indicates how well the discretized equations are satisfied . the secret of a successful relaxation method is a good guess for the starting configuration . fortunately , for @xmath265 the analytic solution of ( [ od1 ] ) is known for all possible shape functions @xmath95 . therefore , we start with this configuration at @xmath265 and increase @xmath151 by a small amount @xmath352 . the relaxed solution is a good initial configuration for @xmath353 , and so on . as a test , the spectrum for @xmath244 is calculated using the relaxation method . this reproduces the analytic result . for @xmath14 the relaxation method converges well as long as @xmath354 is not inside the interval @xmath355 $ ] . however , if the regular singular point @xmath354 is inside the interval , the relaxation algorithm no longer converges and has to be modified . this is one of the main reasons for including the lines @xmath272 in the figures . in this case a specially chosen shooting method proves more suitable . it deals with the additional singularities in @xmath355 $ ] by simply avoiding them . as mentioned before , our differential equation can be defined on the complex plane . due to the boundary condition the solution @xmath158 is an analytic function . the original solution is recovered by restricting @xmath158 to be a real function of a real variable @xmath356 $ ] . in order to avoid further singular points on the real line we can set @xmath357 where @xmath358 $ ] . then equation ( [ od1 ] ) induces a differential equation for @xmath116 , that is the equation is solved on a semi - circular contour . the resulting differential equation can be written as a system of 5 real first order differential equations , one of which corresponds to the energy @xmath268 . now , the spectrum can be obtained by a standard shooting method , @xcite . the boundary conditions for the shooting method are that at the singular points @xmath359 the solution is non - singular and furthermore that the equation is real at @xmath239 . since the equation has real coefficients , the fact that @xmath158 is real for @xmath287 implies that @xmath158 is real for @xmath356 $ ] . a successful shooting method needs a good initial guess . for @xmath360 we can rely on our explicit solution and impose @xmath361 and @xmath362 where equation ( [ initial1a ] ) follows from equation ( [ hypid ] ) and equation ( [ initial1b ] ) follows from the reflection relation for the gamma function ( see ( 6.1.17 ) in @xcite ) . as in the relaxation method the initial guess for the coupling constant @xmath363 is the value of the solution for coupling constant @xmath151 . it is also possible to reconstruct the solution @xmath158 for @xmath356 $ ] using the solution @xmath364 for @xmath365 $ ] . cauchy s formula states that if @xmath366 is an analytic function and @xmath367 is a contour enclosing the point @xmath368 then @xmath369 see for example @xcite . in our case @xmath158 is a real function on the interval @xmath355 $ ] , so @xmath370 where the coefficients @xmath371 are real . it follows that @xmath158 can be written as @xmath372 and a similar formula can be obtained for its derivative @xmath373 . for @xmath374 both the relaxation method and the shooting method give the same result . with the conformal shape function ( [ conformal ] ) , equation ( [ od1 ] ) gives rise to a rather lengthy differential equation with five regular singular points , see @xcite for more details . assuming that the singular points do not coalesce we obtain the following exponents : @xmath375 where @xmath376 and @xmath377 and @xmath378 . the relation ( [ erelation ] ) is again satisfied . moreover , the exponents at infinity are the same as for @xmath244 . in the limit @xmath379 the exponents at infinity change discontinuously ( see ( [ exk=1 ] ) ) . the exponents change if one or more singular points coalesce . however , in general , there are five singular points which all have different characteristics . the regular singular points at @xmath359 provide the boundary conditions for the regular solution , and are not very sensitive to the exact form of the shape function . they imply that the solution of this equation locally still is an analytic function of @xmath380 . one might expect that the regular singular point at infinity could be used to calculate the degree of the polynomial as in the previous two sections . however , for general @xmath100 , the solutions are not regular on the entire complex plane , so that the exponents at infinity do not contain any information about the spectrum . this is because there is a further singular point at @xmath381 with exponents @xmath382 . generically , the solution diverges logarithmically at this singular point such that @xmath383 is a branch point , see equation ( [ log ] ) . the existence of polynomial solutions might be related to the spherical @xmath102 symmetry of the problem , which is broken for @xmath384 . note that the singular point at @xmath385 is independent of @xmath268 , and it can never be inside the interval @xmath386 because @xmath387 for @xmath388 . finally , the singular point at @xmath389 depends on the energy @xmath268 . moreover , for certain values of @xmath268 , @xmath100 and @xmath151 this point can be inside the interval @xmath386 . this singular point could force a potential solution to diverge , such that it would no longer be well - defined . therefore , we are interested in the local behaviour near @xmath390 . after a straightforward but long calculation using maple , we obtain @xmath391 . therefore , locally , both solutions are convergent power series , and no logarithmic divergencies occur . ) for the shape function @xmath392 where @xmath393 . @xmath272 is shown as dashed lines.[fermion3],height=472 ] in the following we calculate spectra as functions of the coupling constant @xmath151 for different values of @xmath100 numerically . for @xmath101 we reproduce the spectrum for @xmath14 of sect . [ hedgehogskyrmion ] . also note that @xmath394 corresponds to the spectrum for @xmath395 which agrees with the spectrum for @xmath244 in sect . [ constantskyrmion ] . as @xmath100 increases the spectrum moves towards the @xmath396 spectrum . at the same time , the energy @xmath339 intersects @xmath264 for increasing values of the coupling constant @xmath151 , such that for @xmath397 the intersection point also appears to move to @xmath398 . in figure [ fermion3 ] we display the spectrum for @xmath393 . in the following we discuss the eigenfunction which crosses zero in more detail . we restrict out attention to the following lorentz scalar @xmath399 which is defined as @xmath400 a short calculation shows that @xmath401 is a function of @xmath97 only . therefore , we can integrate out the angular coordinates @xmath115 and @xmath116 and obtain : @xmath402 the density @xmath403 measures where the fermion field is localised . to obtain equation ( [ qtilde2 ] ) we used equations ( [ 0+spinor ] ) and ( [ fgtilde ] ) . @xmath158 and @xmath166 can be calculated from ( [ od1 ] ) and ( [ od2 ] ) , respectively . recall @xmath404 . expressing @xmath167 in terms of @xmath166 avoids having to worry about the denominator in ( [ f ] ) which is rather difficult to handle numerically . note however , that @xmath167 is non - singular for @xmath405 . this follows from the behaviour at the corresponding regular singular point of equation ( [ od1 ] ) . the function @xmath166 is related to @xmath158 via equations ( [ fgtilde ] ) and ( [ f ] ) . this implies @xmath406 in the following , we calculate @xmath403 for @xmath101 , @xmath407 and @xmath408 and compare it to the baryon density @xmath409 which measures where the skyrme field is localized . is defined with a minus sign , but this is of no relevance here . ] the expression for @xmath79 can be calculated explicitly for @xmath101 . in this case @xmath302 . since we are interested in the zero crossing mode we have to set @xmath410 and obtain @xmath411 surprisingly , this is independent of @xmath151 . also note that in this case @xmath403 is proportional to @xmath412 . in the following , we normalize @xmath403 so that @xmath413 . for @xmath414 , the skyrme configuration is localised near the north pole @xmath415 ( see figure [ fermion4 ] ) . in figure [ fermion4 ] ( a ) and ( b ) we show @xmath412 for @xmath416 and @xmath393 , respectively . in figure [ fermion4 ] ( c ) and ( d ) we show @xmath403 for several values of the coupling constant @xmath151 . for @xmath265 , @xmath403 can also be calculated explicitly and is independent of @xmath100 . for @xmath378 and @xmath417 , the fermion configuration is also localised near the north pole . for @xmath416 , the energy @xmath268 vanishes when @xmath418 . for @xmath393 , the energy @xmath268 vanishes when @xmath419 . in figure [ fermion4 ] ( e ) and ( f ) we display @xmath403 for these values of @xmath151 and compare it to the baryon density @xmath412 . there is quite a good agreement between these two functions . therefore , the @xmath264 states can be interpreted as a fermion - skyrmion bound states . the simplest method for obtaining a skyrmion with baryon number @xmath93 is the ansatz @xmath420 . for @xmath93 this skyrme configuration has very high energy . however , we are mainly interested in the interplay of topology and the spectrum of the dirac equation which should be independent of the precise form of the skyrmion . . @xmath272 is shown as dashed lines . [ fermion5],height=472 ] for @xmath421 the spectrum is displayed in figure [ fermion5 ] . it is symmetric with respect to the @xmath268-axis . note that the symmetry of the spectrum can be derived from equation ( [ od1 ] ) and ( [ od2 ] ) . for even @xmath4 the function @xmath422 is an even function of @xmath380 and @xmath423 is an odd function of @xmath380 . under @xmath424 and @xmath425 equation ( [ od1 ] ) transforms into equation ( [ od2 ] ) . however , the equations are just expressed in different coordinate systems , and hence have the same energy eigenvalues . therefore , we have shown that the spectrum is symmetric under @xmath424 if @xmath4 is even , provided that @xmath148 is an odd function of @xmath380 . for @xmath265 all the eigenmodes have the same value , independent of the background configuration , and parity . as @xmath151 increases , the modes increasingly feel the influence of the background configuration , in particular its topology . it appears that the winding number @xmath421 of the skyrmion forces @xmath407 modes to cross zero , one for @xmath417 and one for @xmath426 . in order to follow the spectrum for @xmath271 the shooting method has to be used . . here we only show the spectrum for @xmath427 , see text . @xmath272 is shown as dashed lines.[fermion6],height=472 ] this pattern can also be verified for @xmath428 , see figure [ fermion6 ] . now there are three modes which cross zero , two of positive parity ( @xmath417 ) and one of negative parity @xmath429 . it seems probable that the number zero - crossing modes of positive parity differs at most by one from the number of zero - crossing modes of negative parity . further calculations for @xmath430 and @xmath408 confirm this pattern . for @xmath430 we find two zero - crossing modes of positive parity and two of negative parity , whereas for @xmath431 there are three positive parity modes and two negative parity modes . this numerical evidence shows that the singular points of equation ( [ od1 ] ) play an important role . in the following we will assume that @xmath95 is an analytic function , and in particular that it can be expanded in a taylor series with respect to the variable @xmath159 , for @xmath432 . it follows that @xmath422 and @xmath423 are also analytic functions of @xmath380 . therefore , the only singular points of ( [ od1 ] ) inside the interval @xmath386 occur for @xmath433 . if such a singular point is non - degenerate its exponents are @xmath338 and @xmath407 . furthermore , it can be shown be shown by expanding @xmath148 as a power series in @xmath380 that the coefficient @xmath434 always vanishes . in our previous examples @xmath95 was always monotonic . let @xmath95 be monotonic with boundary conditions @xmath435 and @xmath436 . if @xmath271 then @xmath433 has @xmath4 nondegenerate solutions for @xmath437 . in this case , these singularities of equation ( [ od1 ] ) all have exponents @xmath438 and the coefficient @xmath434 vanishes , so that the solution of ( [ od1 ] ) is regular at these points . we conjecture that the existence of @xmath4 regular singular points inside the interval @xmath386 indicates that there are @xmath4 modes crossing from one side of the spectrum to the other . this conjecture is equivalent to the statement that for topological charge @xmath4 there are @xmath4 modes that cross zero . if we allow for more general skyrme configurations than the hedgehog , such a statement becomes even more difficult to prove , because now we have to solve partial differential equations . however , both the topological number and the number of singular points are stable under small perturbations . therefore , there seems to be a deep interplay between the topology of the configuration and the singular points of a differential operator . this can be compared to monopoles where we have an index theorem which connects the topology of the monopole configuration with the index of the dirac operator . in the following we compare our results with the literature . kahana and ripka considered the problem of coupling a hedgehog configuration to fermions in flat space using an approximation for the skyrme configuration with a size parameter . varying this parameter they computed the spectral flow of the eigenvalues . they found a @xmath439 zero mode for a special value of the `` size '' of the hedgehog . since the size is given by minimizing the energy density for the skyrmion their result corresponds to varying the coupling constant @xmath151 . in @xcite a linear approximation for the shape function , @xmath440 , was used . here @xmath441 so that the dirac equation for the fermions is independent of @xmath151 . however , we defined the skyrmion as a given background configuration which minimizes the energy . therefore , @xmath442 where @xmath443 is the size of the skyrmion . @xmath443 can be calculated by minimizing the energy @xcite : @xmath444 . kahana and ripka showed by numerical calculation that there is a zero mode for @xmath445 which corresponds to @xmath446 . in @xcite an exponential function was used to approximate the shape function : @xmath447 . in this case minimizing the energy results in @xmath448 and the zero mode occurs for @xmath449 which corresponds to @xmath450 . the fact that there is a mode which crosses zero seems to be independent of the special ansatz for the shape function . this is consistent with our results . in @xcite it has been argued that a skyrmion in flat space is reasonably well approximated by the stereographic projection of a skyrmion on a @xmath0-sphere of optimal radius . the optimal radius for @xmath10 is @xmath9 so we can use our analytic result for the mode @xmath339 in sect . [ hedgehogskyrmion ] . for @xmath306 this is a zero mode , and the value of @xmath151 is in the same order of magnitude as in the literature mentioned above . we have derived the dirac equation for fermions on the @xmath0-sphere which are chirally coupled to a skyrmion . in this paper , we only considered spherically symmetric skyrme configurations and fermionic states with the total angular momentum @xmath451 . we were particularly interested in the shape functions @xmath244 and @xmath14 and the conformal ansatz ( [ conformal ] ) . the shape function @xmath221 leads to a hypergeometric equation and can therefore be solved explicitly in terms of hypergeometric functions . to derive the spectrum we have to impose that the solutions are non - singular at the north and south pole using the correct chart . we derived the spectrum @xmath452 which agrees with the spectrum calculated by sen @xcite using maurer cartan forms . moreover , the eigenfunctions @xmath263 are given by jacobi polynomials . the shape function @xmath14 is more complicated . it leads to a fuchsian equation with four regular singular points , two at the boundary @xmath359 , one at infinity and one depending on @xmath268 and @xmath151 . however , the equation could still be solved in terms of polynomials . we were able to solve the recurrence relation for the coefficients of the polynomial and derived the corresponding spectrum : @xmath453 the eigenfunctions have been evaluated explicitly : @xmath454 where @xmath455 we also considered more general shape functions numerically . for @xmath10 we discussed the conformal ansatz ( [ conformal ] ) and calculated the spectrum . for @xmath101 the spectrum agrees with the analytic solution . in particular there is a mode @xmath338 whose energy @xmath339 crosses zero for a certain value of the coupling @xmath151 . the graph of @xmath339 as a function of @xmath151 changes as the parameter @xmath100 in the ansatz is varied . however , for finite @xmath456 , this mode always stays present . according to our numerical evidence the eigenfunction for @xmath264 can be interpreted as a fermion - skyrmion bound state . for larger @xmath4 we considered the ansatz @xmath420 . this leads to @xmath4 modes crossing zero . we conjectured a connection between zero modes and certain singular points in the equation which in turn are connected to the winding number of the configuration . the author wants to thank n s manton , h pfeiffer and y shnir for many discussions during various stages of this project . the author also would like to thank j m speight for many helpful suggestions and continuous support . t. h. r. skyrme , _ a nonlinear field theory _ , proc . . lond . * a260 * : 127 ( * 1961 * ) , g. s. adkins , c. r. nappi and e. witten , _ static properties of nucleons in the skyrme model _ , nucl . phys . * b228 * : 552 ( * 1983 * ) , r. a. leese , n. s. manton and b. j. schroers , _ attractive channel skyrmions and the deuteron _ , . phys . * b442 * : 228 ( * 1995 * ) , , p. irwin , _ zero mode quantization of multi - skyrmions _ , phys . rev . * d61 * : 114024 ( * 2000 * ) , , s. krusch , _ homotopy of rational maps and the quantization of skyrmions _ ( * 2002 * ) , , d. finkelstein and j. rubinstein , _ connection between spin , statistics , and kinks _ , j. math . phys . * 9 * : 1762 ( * 1968 * ) , e. witten , _ global aspects of current algebra _ , nucl . phys . * b223 * : 422 ( * 1983 * ) , a. p. balachandran and s. vaidya , _ skyrmions , spectral flow and parity doubles _ , int a14 * : 445 ( * 1999 * ) , , j. r. hiller and t. f. jordan , _ solutions of the dirac equation for fermions in skyrme fields _ , phys * d34 * : 1176 ( * 1986 * ) , s. kahana and g. ripka , _ baryon density of quarks coupled to a chiral field _ , nucl . phys . * a429 * : 462 ( * 1984 * ) , g. ripka and s. kahana , _ the stability of a chiral soliton in the fermion one loop approximation _ , phys . lett . * b155 * : 327 ( * 1985 * ) , g. ripka , _ quarks bound by chiral fields _ , oxford science publications ( * 1997 * ) . n. s. manton and p. j. ruback , _ skyrmions in flat space and curved space _ , phys . lett . * b181 * : 137 ( * 1986 * ) , n. s. manton and b. j. schroers , _ bundles over moduli spaces and the quantization of @xmath457 monopoles _ , ann 225 * : 290 ( * 1993 * ) , m. nakahara , _ geometry , topology and physics _ , graduate student series in physics , iop publishing ltd ( * 1990 * ) . t. eguchi , p. b. gilkey and a. j. hanson , _ gravitation , gauge theories and differential geometry _ , . rept . * 66 * : 213 ( * 1980 * ) , s. krusch , _ @xmath458 skyrmions and the rational map ansatz _ , nonlinearity * 13 * : 2163 ( * 2000 * ) , , m. gell - mann and m. levy , _ the axial vector current in beta decay _ , nuovo cim . * 16 * : 705 ( * 1960 * ) , n. s. manton , _ geometry of skyrmions _ , commun . math . phys . * 111 * : 469 ( * 1987 * ) , e. t. whittaker and g. n. watson , _ a course of modern analysis _ , cambridge mathematical library , cambridge university press , fourth edition ( * 1927 * ) . d. sen , _ fermions in the space - time @xmath459 _ , j. math . * 27 * : 472 ( * 1986 * ) , m. carmeli and s. malin , _ field theory on @xmath459 topology . 3 . the dirac equation _ , found . phys . * 15 * : 1019 ( * 1985 * ) , w. h. press , b. p. flannery , s. a. teukolsky and w. t. vetterling , _ numerical recipes _ , cambridge university press ( * 1992 * ) .
according to the most popular progenitor model of long - duration gamma - ray bursts ( grbs ) , the collapsar model @xcite , a grb is the result of ultra - relativistic jets ejected by an accreting black hole formed by the core - collapse of a massive star ( most probably a wolf - rayet star ) . this predicts a physical link between grbs and supernovae ( sne ) that has been spectroscopically confirmed in four cases so far : xrf 020903 @xcite , grb 021211/sn2002lt @xcite , grb 030329/sn2003dh ( e.g. , * ? ? ? * ; * ? ? ? * ) and grb 031203/sn2003lw . further evidence comes from the observation of late - time bumps in grb afterglows that can be modelled very well by an underlying sn component @xcite , and which have led to the conclusion that in fact all long - duration grbs are physically related to sn explosions ( * ? ? ? * paper i ) . furthermore , the collapsar model implies that the progenitors of long - duration grbs are associated with regions of high - mass star formation @xcite , which might reveal themselves by a detectable extinction in the grb host galaxies along the lines of sight towards the burster . this idea is further supported by the so - called dark bursts @xcite , for which no optical afterglow has been discovered despite rapid and deep searches in small error box regions . while most none - discoveries are the result of too shallow searches and too large error boxes ( e.g. , * ? ? ? * ; * ? ? ? * ) , a small percentile remains that require intrinsic extinction to dim the afterglow , e.g. , grb 970828 @xcite , 990506 @xcite and 020819 @xcite , while others might have been intrinsically underluminous ( for a discussion , see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . of particular interest within the dust extinction model is the statistical distribution of the amount of visual extinction in the grb host galaxies , as well as the nature of the corresponding dust . given the fact that grbs and their afterglows can be observed up to high redshift , they offer the possibility to get insight into the nature of cosmic dust when the universe was much younger and galaxies were much less evolved . furthermore , since the optical properties of the dust grains trace the environmental conditions in the interstellar medium @xcite , they are to some degree an indicator for the physical conditions to make a grb progenitor . the present paper is the third in a series of papers where we employ a large database of photometry gathered from the literature to reanalyze all optical afterglow light curves of grbs in the pre-_swift _ era in a consistent manner to derive a homogenous sample , which is then used to study afterglow properties in a statistical way . in paper i and an update @xcite , the properties of the supernovae underlying nearby afterglow light curves were explored . in paper ii @xcite the optical light curve parameters were derived for a large ( and basically complete ) sample of afterglows observed by the end of 2004 , up to the launch of the _ swift _ mission . in this paper , we extend this systematic analysis to the spectral energy distribution ( sed ) of grb afterglows in the optical / near - infrared bands in order to search for signals from extinction by dust in the grb host galaxies . in [ data ] , we present the methods with which we analyzed the afterglow seds . among the 59 grbs studied in paper ii , 30 had data quality sufficiently high to be included in a sample for an investigation of the seds . we then further reduce this sample to a golden sample of 19 grb afterglows with parameters derived from specific dust model fits . in [ rad ] we present the results derived from our analysis and discuss our findings in the context of the standard fireball model . the fitting process allows us to derive the host galaxy extinction @xmath0 along the line of sight and the intrinsic , extinction - corrected spectral slope @xmath1 ( @xmath2 ) of the afterglow light in the optical / near - infrared bands . the @xmath0-@xmath1 sample is then further analyzed statistically to derive conclusions about the environment of grbs and the dust properties of high - redshift galaxies . in paper ii we presented the analysis of a complete list of optical / nir afterglow light curves for which sufficient public data was available , up to grb 041006 . one parameter derived in these light curve fits is the normalization constant @xmath3 , the magnitude of the afterglow at a certain time , being one day after the grb trigger in the case of a light curve fit with a simple power - law ( @xmath4 ) , or the jet break time , @xmath5 , in case of a fit with a smoothly broken power - law with a pre - break decay slope @xmath6 and a post - break decay slope @xmath7 @xcite . all data used to fit these light curves is corrected for galactic extinction with @xmath8 derived from the cobe maps @xcite employing the milky way extinction curve of @xcite . the contemporaneous afterglow brightness in different photometric bands is then transformed into a spectral energy distribution , using zero fluxes and median wavelengths taken from the literature . the sed for grb 030329 is derived in an alternate way , see appendix [ litcomp ] . the fits use a levenberg - marquardt @xmath9 minimization algorithm . unless stated otherwise , all errors in this paper are at the @xmath10 confidence level . initially , we assume no extinction and the seds are fit with a simple power - law , @xmath2 . in the following , we label a slope derived from such a fit @xmath11 . a steep @xmath11 combined with a non - linear ( curved ) @xmath12 is then indicative of extinction in the grb host galaxy . in order to derive the visual extinction @xmath0 in the grb host galaxy along the line of sight , we transform the wavelengths of the sed into the host galaxy frame following @xcite , using redshifts taken from the compilation of @xcite . in two cases redshift estimates are used . for grb 980519 , we use a redshift of @xmath13 , following @xcite who state that @xmath14 due to the lack of a supernova bump . for xrf 030723 , we varied the redshift and fit the sed for each redshift , and find the best results in a redshift range of @xmath15 to 0.4 . thus , we adopt @xmath16 . @xcite find tentative evidence for a low redshift ( although the host galaxy would be very faint ) , and @xcite find a preferred @xmath17 from the amati relation @xcite . in both cases , the sed is fit well by a power - law with an additional small amount of source frame extinction . the influence of intergalactic lyman absorption on the photometry was accounted for by excluding sed data points lying beyond @xmath18 hz in the host frame from the fit ) , the v band also had to be excluded as the host galaxy is a dla with a very wide lyman @xmath19 absorption line @xcite . ] . the observed sed ( corrected for galactic extinction ) is then modeled by the function @xmath20 with @xmath21 here , @xmath1 is the intrinsic power - law slope of the sed , and @xmath22 a normalization constant ( we choose the unextinguished flux density at @xmath23 hz in the host galaxy frame ) . the function @xmath24 is the extinction law assumed for the interstellar medium ( ism ) of the grb host galaxy . we call this extinction source frame extinction . it encompasses local extinction close to the site of the grb and host galaxy extinction further away in case the afterglow passes through a significant part of the host galaxy along the line of sight . the extinction law for the milky way ( mw ) , the large magellanic cloud ( lmc ) , and the small magellanic cloud ( smc ) was taken from @xcite . these three dust types differ strongly in the wavelength region we examine ( host frame uv / optical / nir ) , especially toward shorter wavelengths . the sequence mw , lmc , smc is given by a decreasing strength of the 2175 bump , an increasing fuv extinction , and a decreasing reddening per h atom @xcite . the former implies the absence of very small carbonaceous grains @xcite , while the latter is consistent with the trend observed for the metallicity in hii regions , with mw representing the highest metallicity and smc the lowest . thus , smc dust traces the dust properties in low - metallicity environments . as the wavelengths of the sed are shifted from the uv into the optical / nir , a distinction between the three dust types is easiest for grbs at high redshift , whereas for nearby grbs the three extinction curves are almost identical in the considered wavelength regime . since the fit has three free parameters ( @xmath22 , @xmath1 and @xmath0 ) , the observed sed has to have at least four data points . the one exception is grb 000131 ( appendix [ litcomp ] ) , which we include because of its very high redshift ( @xmath25 , * ? ? ? the fitting process is very sensitive to slight variations in the sed , especially if the @xmath3 data points have small errors , so we did a careful check for each sed and removed outliers . furthermore , we added a systematic error of 0.03 magnitudes to each data point of the sed ( with the exception of those small number of cases where @xmath3 is based on only one data point , as these almost always have much larger errors anyway ) to encompass uncertainties that may derive from the process of data reduction . this results in a strong decrease of @xmath9 for the fits and reduces the importance of this measure for discerning among different models for sed fits . our final sample ( table [ taball ] ) comprises 30 afterglows out of the 59 bursts included in the sample of paper ii , with fits to the three dust models for each afterglow . figure [ seds ] shows all 30 seds and the associated best fits for each of the three dust models . this figure contains all fit results that we obtain , including those with unphysical results , just in order to show the failure of certain extinction laws to reproduce the observational data . table [ taball ] reveals that many of the fits are unsatisfactory . first of all , some fits have very large error bars @xmath26 and @xmath27 . since the seds are fit by allowing all three free parameters ( incl . @xmath22 ) to vary simultaneously , this creates a cross - sectional error , as the errors of @xmath1 and @xmath0 are correlated . the steepest spectral slope ( @xmath28 ) results in the smallest extinction ( @xmath29 ) and vice secondly , many fits , especially when using the mw dust model , find _ negative _ extinction . this would imply bright emission features , for which we have no evidence . thirdly in some cases @xmath30 is found , while the standard fireball model implies @xmath31 in the timespans ( hours to days ) and wavelength bands that afterglows are typically observed in , and thus the timespan where we derive @xmath3 . these afterglow seds ( e.g. grb 971214 ) are categorized by a relatively shallow unextinguished slope @xmath32 to 1 but a strong spectral curvature . correction of this curvature results in @xmath30 . one solution would be dust with an increased fuv extinction , but recent results on high redshift quasars @xcite do not support this idea . we note that in these cases the data base is also sparse in photometric bands other than @xmath33 . to derive a more homogenous sample , which we call the golden sample , we employ the following criteria : 1 . the 1@xmath34 error in @xmath1 ( @xmath27 ) and the 1@xmath34 error in @xmath0 ( @xmath26 ) should both be @xmath35 . 2 . @xmath36 . 3 . we do not consider grbs where all fits ( mw , lmc , and smc ) find @xmath37 , even if the previous criterium is fulfilled . 4 . @xmath38 ( although we do not reject cases with @xmath39 ) . 5 . a known redshift . as there is no case where lmc dust is clearly preferred compared to smc and mw dust ( table [ taball ] ) , in the following we remove all lmc dust cases from further consideration . after applying these criteria , for the remaining 19 grbs , in eight cases only one dust model remains . for grb 000131 no distinction can be made , as it is not fit with extinction . the remaining ten grbs give ambiguous results when it comes to distinguishing between the two dust models . with the exception of grb 020813 and grb 030328 , where visual inspection of the fits shows that the inclusion of smc dust reproduces the observed sed better than mw dust , no true distinction can be made at all . for these grbs , we do not claim any dust model preference , but for the sake of consistency , we choose the dust model result which produces smaller errors in the derived parameters . this results in 19 @xmath1-@xmath0 pairs which we designate our golden sample ( table [ tabgold ] ) . there are some grbs for which high visual source frame extinction has been deduced in the literature but which are not included in table [ taball ] . for grb 980329 , no redshift is known , although it is assumed to be high . @xcite derive a bayesian photometric redshift of @xmath40 . the very steep observed spectral slope ( uncorrected for extinction ) has been noted by several authors . @xcite use their own data combined with data from the literature and derive an @xmath41 for @xmath42 , corresponding to @xmath43 assuming a galactic dust - to - gas ratio . this is concurrent with our result @xmath44 for smc dust and under the assumption of a redshift of @xmath40 . however , this result also has @xmath30 , implying that the true host extinction is lower than @xmath45 . two afterglows which have redshifts but viable data in only two colors , grb 990705 and grb 000418 , could not be fitted with the procedure outlined in [ analysis ] . @xcite find @xmath46 , in agreement with our result . while they do discuss reddening by intrinsic dust , their lack of a redshift measurement does not allow them to find a definite result . the steep initial decay suggests a wind environment with a cooling break blueward of the optical bands . using this case and smc dust , we derive @xmath47 . for grb 000418 , assuming a wind ( ism ) environment with the cooling break blueward of the optical bands and the new derived light curve parameters from paper ii we fix @xmath1 via the @xmath19-@xmath1 relations and find @xmath48 ( @xmath49 ) . the latter is slightly smaller than in @xcite ( who used an ism environment ) , while @xcite find @xmath50 for lmc dust . finally , there are indications of very high source frame extinction from the nir afterglows of grb 030528 @xcite and grb 040827 @xcite , but we were not able to perform any fits . in the selection of our golden sample , there were eight bursts where application of our selection criteria yielded an unambiguous result . smc dust is by far the most preferred model , with seven out of the eight cases ( table [ tabgold ] ) , only grb 970508 has evidence for mw dust . this result is expected from studies of grb host galaxies , which indicate that most hosts are low - metallicity galaxies , as they are lyman @xmath19 emitters , blue , and subluminous ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? thus we deduce that the ism in these galaxies is probably metal - poor . while the preference of smc - like dust in grb environments has already been found by many groups for several bursts , e.g. , grb 000301c @xcite , grb 000926 @xcite , grb 010222 @xcite , grb 020813 @xcite , grb 021004 @xcite , grb 030226 @xcite , grb 030429 @xcite , xrf 030723 @xcite , our study puts this conclusion on a statistical basis . given the sparse knowledge we have so far on the dust properties in cosmologically remote galaxies we note however that this finding does not necessarily imply the requirement of low - metallicity environments for the creation of grb progenitors , as it might be indicated by smc - like dust . the fact that at least some afterglows seds are fitted best assuming mw dust raises doubt on such a general requirement . even though one can not rule out the possibility that in such cases the line of sight passes through foreground material with different grain properties and ism metallicity . in any case , it is clear from this study that grb afterglows can be used as a tool in order to explore the properties of cosmic dust in the cosmologically remote universe , in star - forming regions in particular and in galaxies in general . but high quality photometric data are essential for building larger samples . among the 19 grbs in our golden sample ( table [ tabgold ] ) , we find eleven with @xmath51 . we note that almost all of these cases are also those with the highest quality data . basically , this is not surprising , as more sed data reduces errors . it is more speculative to state that , given high quality sed data , there will always be small amounts of source frame extinction detected in grb afterglows . in eight of these cases , the detection is significant at the 2@xmath34 level or higher , it is significant at more than 3@xmath34 in two cases and at 4@xmath34 in one case . the distribution of derived intrinsic extinctions is given in figure [ a_vhist ] . a strong clustering toward low extinctions ( @xmath52 ) is evident . zooming in with smaller bins shows that values between @xmath53 and @xmath54 are preferred , with a non - error - weighted mean value of @xmath55 . no afterglow is extinguished by more than 0.8 magnitudes . as a comparison , the prime example of a dark burst is plotted , grb 970828 . for this burst , @xcite derive @xmath56 by interpolating between x - ray and radio data and comparing this with deep optical observations that found no afterglow @xcite . the intervening space is free of examples , creating a potential dark burst desert . on the other hand , there are two afterglows in table [ taball ] that have mean @xmath0 values lying in this dark burst desert but are not included in the golden sample due to the large errors of their fits ( criterium ( 1 ) in [ golden ] ) . for grb 010921 , mw dust is preferred , as smc dust finds @xmath57 . we derive @xmath58 , in agreement with @xcite . while this is a positive detection , the error encompasses most other bursts of the golden sample . for grb 980703 , no dust model preference can be derived , but even for the smc result with smaller extinction , we find @xmath59 . how real is the non - existence of highly extinguished afterglows ? a first argument against the reality of the dark burst desert is that our sample is biased toward afterglows with low source frame extinction , as we require viable photometric data in several wavebands and a known redshift . intrinsic extinction complicates the gathering of photometry ( starting with the discovery of the afterglow ) and a spectroscopic redshift . in [ others ] , we mentioned several probably highly extinguished afterglows that are not part of our sample due to the sparsity of their photometric data . highly extinguished afterglows are not observed well enough to determine conclusively that they are highly extinguished and not just intrinsically faint @xcite . figure [ a_vtoz ] shows the derived visual extinctions ( table [ tabgold ] ) plotted as a function of redshift @xmath60 ( values taken from * ? ? ? * ) . with the exception of grb 030429 the amount of extinction drops sharply toward high redshifts . most likely , this can be explained by an observational bias , at least in the pre-_swift _ era where only very few afterglows were already localized within minutes after a burst . in particular , due to the wavelength dependence of the dust opacity , the higher the redshift the more effective dust can dim an afterglow in the optical bands . furthermore , the redshift measurement itself is biased toward bright and unextinguished afterglows , as many redshifts are determined from absorption spectra taken when the afterglow is much brighter than the underlying host galaxy . finally , our data indicate that , on average , low-@xmath60 grbs have brighter afterglows ( cf . figure [ bigpic1 ] ) , and thus have a greater chance of being detected through significant extinction . unfortunately , the chances of detecting a `` rosetta stone dark burst '' at low redshift that is highly obscured by dust but still bright enough to yield viable data and a redshift might be small if the grb frequency is coupled with the star - formation rate ( e.g. , * ? ? ? * ; * ? ? ? * ) and low metallicities @xcite . it is also visible that the the ability to discern between different dust models is coupled to the redshift , as the strongest deviations ( 2175 bump , fuv extinction ) appear in the uv region and the burst has to lie at a certain redshift to shift these features into the optical bands . assuming that our database is not seriously affected by an observational bias ( dark bursts ) , figure [ submm ] shows that the standard grb afterglow is nearly unextinguished in its host galaxy . since any long - lasting dust destruction by the intense fireball radiation should represent itself in a color evolution of the afterglow @xcite , which we do not find in our data , we conclude that on average there is not much evidence for dust in grb environments along the line of sight . furthermore , any dust destruction could only affect the dust in the immediate grb environment , say , within 10 pc around the burster @xcite . dust at larger distances will still produce extinction . additionally , the _ swift _ satellite has made observations of early grb afterglows routine , and a major result is the almost complete absence of reverse shock uv flashes @xcite . it seems that in most cases the absence of large amounts of dust is not due to dust destruction but due to the fact that the standard grb progenitor ( seen pole - on along its rotational axis ) is not enshrouded by dust , globally and locally . thus , if in those cases where significant extinction is found the extinction is not just produced in the immediate grb environment ( local extinction ) but prevails within the host galaxy ( global extinction ) , we would expect that grbs for which we find significant optical extinction are located in very dusty , submm - bright hosts . in order to explore this possibility , we have taken from the literature all submm data for grb hosts with a flux density @xmath61 at the 1@xmath34 level that are also in our main sample of 30 afterglows @xcite and calculated the corresponding star formation rate ( sfr ) via the procedure developed by @xcite ( cf . * ? ? ? * ) to take the different redshifts into account . since the sfr is a direct measure of the far - infrared luminosity of a galaxy @xcite , it traces its total amount of radiating dust . for grb 980703 , we conservatively used the result for smc dust , which has the lowest extinction , and took the sfr from @xcite . we added grb 000418 ( @xmath48 , [ others ] ) , arbitrarily assuming a 1 @xmath34 error of 0.5 mag , and transformed the @xmath0 result for grb 011211 ( smc dust ) into an upper limit , as we find @xmath30 , making @xmath0 unsure . the resulting relation of sfr ( @xmath62 per year ) versus @xmath0 is shown in figure [ submm ] . a trend is visible , a rising sfr is coupled to a rising intrinsic visual extinction along the line of sight , with the exception of grb 010222 . a linear fit to the data ( excepting grb 010222 relation is grb 010222 . a hubble space telescope image of the grb 010222 host galaxy reveals the location of the grb being offset by a small margin from the center of the galaxy @xcite . we find a low extinction value along the line of sight , whereas persistent submillimeter flux indicates that this is a dusty starburst galaxy @xcite . the discrepancy is resolved if the grb happened toward the edge of the galaxy , but our line of sight places it in front of the galaxy . ] ) gives @xmath63 . we conclude that , on average , grb afterglows that show significant extinction along the line of sight in their host are located in galaxies with a substantial star formation rate and , hence , a globally acting extinction by large amounts of dust . if this interpretation is correct , then the trend seen in the data indicates that the line - of - sight can pass through a significant extent of the host galaxy . since the sample is still very small and the error bars on both the sfr and the extinction @xmath0 are very large , it is clear that more and better data are required in order to verify this result . figure [ betahist ] displays the distribution of the intrinsic spectral slopes , @xmath1 , for the bursts from table [ tabgold ] . the distribution is broad , ranging from 0.2 to 1.2 , and features a peak around @xmath64 . we find a non - error - weighted mean value of @xmath65 . this result of a broad @xmath1 distribution of the intrinsic sed of the afterglows is probably robust , since ( 1 ) our light curve fits of the individual afterglows usually include many data points covering several days , so that we are not much sensitive to individual measurement errors and ( 2 ) these fits do not include the very early phase of an afterglow when its spectral properties might develop much faster than at later times . in fact , with the exception of the early afterglows of grb 021004 and grb 030329 , we have never found clear evidence for color variations of the genuine afterglow light in our data . while it is not the goal of the present paper to find reliable explanations for the observed width of the distribution of spectral slopes @xmath1 of the afterglows in the optical / nir bands , we note that the mean of this distribution gives a value for the power - law index of the electron distribution ( @xmath66 ) of @xmath67 , assuming a wind environment and the cooling frequency blueward of the optical bands ( * ? ? ? * and references therin ) and the peak of the histogram ( figure [ betahist ] ) . this is in general agreement with theoretical predictions for ultra - relativistic shocks @xcite . the maximum values we find for @xmath1 ( figure [ betahist ] ) could then be explained by those afterglows which had the cooling frequency redward of the optical bands during the entire time span when they were observed , provided that @xmath68 is a universal number . on the other hand , a universality of @xmath68 seems to be difficult to reconcile with those afterglows which have @xmath69 . assuming the standard afterglow models @xcite , these afterglows require @xmath70 , provided that the basic model assumptions are indeed fulfilled in these cases . note that in paper ii we also found afterglows that require @xmath70 based on their light curve shapes alone . the fact that some afterglows require @xmath70 has already been reported and in detail explored by others ( cf . * ? ? ? * ; * ? ? ? * ; * ? ? ? based on our sed fits three of 19 afterglows have @xmath71 within their 1@xmath34 error bar ( grb 970508 , grb 991216 and grb 030429 ) which implies @xmath70 , while a very shallow afterglow decay slope @xmath7 brings grb 990123 , grb 991216 , grb 010222 , grb 030328 and grb 041006 into this sample ( paper ii ) . obviously , only a minority of afterglows require @xmath70 . it is not obvious why these afterglows are specific in some sense . the afterglow of grb 991216 is one of the brightest afterglow for redshifts @xmath72 . on the other hand , the afterglow of grb 041006 is one of the less luminous afterglows ( [ redshifting ] ) . thus , @xmath70 is not a question of luminosity . note that both afterglows have only small intrinsic extinction ( @xmath73 ) , so these low values for @xmath1 are not just an artifact of the fitting process . for grb 991216 , the unextinguished slope is @xmath74 , for grb 041006 it is @xmath75 . we also note that all these afterglow light curves are sampled fairly ( grb 990123 ) to very well ( grb 010222 ) , implying that the finding of a flat spectral slope is not a question of data quality either . x - ray observations or the modeling of lyman @xmath19 absorption allow the determination of the hydrogen column density @xmath76 along the line of sight to the grb in its host galaxy after correcting for the column density in our galaxy @xcite . in figure [ nhtoa_v ] we present a sample of afterglows for which we have derived @xmath0 and for which @xmath76 values are reported in the literature ( table [ xraydata ] ) . this is an update of the plot first presented in @xcite and expanded in @xcite . , leading to a linear depiction of the dust model curves . ] the results reinforce our findings presented in [ smcchapter ] . with two exceptions , all points in the plot lie on the theoretical prediction for smc dust or even above it . the two exceptions are grb 021004 and 030329 , where only upper limits for @xmath76 are given in the literature , leading to dust - to - gas ratios higher than even for the milky way . on the other hand , both afterglow seds are best fit with smc dust ( although the preference is only weak for grb 030329 ) , implying that the three dust models we use are not applicable in all cases . incidentally , grb 021004 and grb 030329 have the two best observed afterglow light curves with the most pronounced substructure ( paper ii ) . the strongest outlier in figure [ xraydata ] is grb 990123 , for which we find a very low source frame extinction . the extremely bright uv flash of this burst @xcite may have burned significant amounts of dust along the line of sight ( cf . * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , reducing the dust - to - gas ratio . no early multicolor data exists to probe the color variations that are expected in the early light curve . knowledge of the intrinsic spectral slope of the afterglows allows us to determine their luminosity distribution ( appendix [ math ] ) . at first , in figure [ bigpic1 ] , we show the @xmath33-band light curves of all 30 afterglows of the sed sample ( table [ taball ] ) plotted with smoothed splines connecting the data points to guide the eye ( for reasons of clarity , error bars on the photometry have been omitted ) . the data have been corrected for galactic extinction and host contribution . for the light curve of grb 030329 , the supernova contribution has also been subtracted , using the data from @xcite . a very large spread of magnitudes is seen , the range at one day after the trigger is 7.5 magnitudes between the afterglows of grb 021211 and grb 030329 . further bright afterglows are those of grb 991208 and grb 991216 , other faint afterglows are grb 040924 , grb 030227 and grb 971214 . figure [ bigpic2 ] displays the afterglow light curves of the golden sample ( table [ tabgold ] ) after applying the cosmological @xmath77-correction ( the second term of eq . [ delta1 ] ) ( table [ tabzcorr ] ) and after the time shifting to a common redshift of @xmath78 ( the first term of eq . [ delta1 ] ) . in other words , figure [ bigpic2 ] is a measure of the absolute @xmath33 band magnitudes of the afterglows up to a constant . compared to figure [ bigpic1 ] the magnitude range has decreased , now being 5.7 magnitudes at one day after the burst . nine of the 16 afterglows that have data one day after the burst lie in a range only two magnitudes wide , approximately clustered around the afterglow of grb 030329 . in other words , this afterglow is now seen to be quite typical . the two afterglows above this range are those of grb 021004 and grb 991208 ( assuming for the latter burst an extrapolation of the decay with @xmath79 to one day after the burst in the host rest frame , paper ii ) . the six afterglows beneath this range are grb 030226 , grb 020405 , grb 030328 , grb 011121 , grb 041006 and grb 040924 . of these six afterglows , only two ( grb 030226 and grb 030328 ) lie at @xmath80 . grb 021004 is the most luminous afterglow at all times , although it is possible that the afterglow of grb 991208 was brighter at earlier times when it was not yet discovered @xcite . knowledge of the unextinguished light curves allows us to determine the luminosity distribution of our afterglow sample . based on fig . [ bigpic2 ] we derive the @xmath33 band magnitudes at one and four days ( corresponding to half a day and two days after the burst in the host frame at @xmath60=1 ) and transform them into absolute magnitudes @xmath81 . we do not extrapolate the afterglow light curves , except for grb 030429 and grb 020813 , which both have their final data points close to 4 days and are post - break . thus , not all light curves are included ( e.g. , grb 991208 is not yet detected at one day , grb 030328 is not detected any more at four days if at @xmath82 ) . the sample then contains 16 grbs at both one day and at four days , the results are given in table [ tabmb ] . thereby , we computed the luminosity errors in a conservative fashion . given that the errors of @xmath1 and @xmath0 are correlated ( the result @xmath83 is coupled to @xmath84 and the other way around ) , we calculate three values of @xmath81 for the pairs @xmath85 , @xmath86 , and @xmath87 . the pair @xmath88 results in a lower luminosity , and the pair @xmath87 in a higher luminosity . the difference between these luminosities and the luminosity derived from the pair ( @xmath89 ) is a conservative upper limit for the error of @xmath81 . in addition , we impose limits @xmath90 and @xmath91 . if @xmath92 or @xmath93 , we set @xmath94 or @xmath95 , respectively , when computing @xmath81 . the resulting luminosity distributions are given in figures [ mb1 ] and [ mb4 ] . in addition , in order to search for a potential evolutionary effect or an observational bias , we distinguish grbs with @xmath96 and @xmath97 , this being the median of the redshift distribution of our sample . a bimodal distribution of the afterglow luminosities is evident in figures [ mb1z ] and [ mb4z ] . afterglows at @xmath96 tend to be less luminous , which might be explainable by an observational bias : the chances to detect intrinsically faint afterglows are higher for a lower @xmath60 . at one day after the grb at @xmath82 , the complete distribution has a two peaks , the mean lying at @xmath98 , with the most luminous afterglow , grb 021004 , having @xmath99 , and the faintest , grb 040924 , having @xmath100 . the means for the distributions of the @xmath96 and @xmath97 grbs are @xmath101 and @xmath102 , respectively . the difference between the mean values is @xmath103 mag , indicating a bimodality . at four days after the burst at @xmath82 , a single broad peak remains , for the complete distribution , the mean value is @xmath104 . now , the brightest afterglow ( grb 021004 ) has @xmath105 and the faintest afterglow ( grb 041006 ) has @xmath106 . once again , the bimodal distribution for nearby ( @xmath96 ) and distant ( @xmath97 ) grb afterglows is evident . the means for the distributions of the @xmath96 and @xmath97 grbs are @xmath107 and @xmath108 , respectively . the difference between the mean values is @xmath109 mag , the significance of the bimodality has been strongly reduced . @xcite analysed the x - ray afterglow light curves of grbs and shifted them to @xmath82 in an analog process , finding a bimodal flux distribution . there are eight grbs in their sample ( grb 970508 , grb 990123 , grb 991216 , grb 000926 , grb 010222 , grb 011121 , grb 030226 , and grb 030329 ) that are also among the afterglows we have plotted in figure [ bigpic2 ] . grb 990123 , grb 991216 , grb 000926 , and grb 010222 are in _ group i _ of @xcite , while grb 970508 , grb 011121 , grb 030226 , and grb 030329 are in their _ group ii_. we note that for @xmath82 , @xmath110 day lies at the end of the initial plateau phase of the light curve of grb 970508 . the afterglow brightens by 1.5 magnitudes shortly afterward . a comparison reveals that while the two groups mix in the optical ( the afterglow of grb 030329 being more luminous than those of grb 990123 and grb 010222 at one day after the burst at @xmath82 ) , the mean absolute magnitude of _ group i _ afterglows is @xmath111 , while it is @xmath112 for _ group ii _ afterglows , with the difference being @xmath113 mag . while this finding is intriguing , the sample is too small to draw any conclusions . knowledge of the extinction corrected afterglow magnitudes allows us to compare the luminosities of the afterglows at a common point in the evolution of the jet , the jet break time . ignoring any fine structure in the light curves , for all bursts in our golden sample with a jet break ( paper ii ) we computed the apparent @xmath33-band afterglow magnitudes at the time of the jet break from the beuermann equation @xcite . we included the fit around the jet break of grb 030329 but excluded grb 021004 , as the reality of the late break we found in paper ii is unclear due to the many rebrightening episodes . these magnitudes were then converted to luminosities and normalized to the luminosity of the afterglow of grb 990123 at the time of its jet break . we use these luminosity ratios to search for correlations between afterglow luminosity and parameters of the prompt @xmath114 emission of the grbs . we take the isotropic energy release @xmath115 and the source frame peak energy @xmath116 from @xcite , and the collimation corrected energies @xmath117 from paper ii . we do not find any correlations between the afterglow luminosity during the break time and these three parameters , with the absolute value of the correlation coefficient being smaller than 0.25 in all cases . we have presented a sample of 30 grb afterglow spectral energy distributions in the optical / nir bands which have been modeled with various dust extinction curves ( milky way , large magellanic cloud and small magellanic cloud ) to derive the source frame extinction , @xmath0 , intrinsic to the host galaxies and the spectral slope , @xmath1 , of the afterglows unaffected by any dust extinction . as all afterglows have been analyzed in a systematic way , the results are fully comparable , making this sample unique in terms of both size and consistency . for the further statistical study , we selected 19 afterglows , our golden sample , which have physically reasonable results and small error bars . the preferred dust models we find ( [ smcchapter ] ) as well as the deduced source frame dust - to - gas ratios ( [ nh ] ) based on the inclusion of data taken from the literature , both indicate that the majority of grbs we have investigated , covering the redshift range from 0.1 to 4.5 , occur in low - metallicity environments . the @xmath0 distribution that we have derived from these data ( [ avchapter ] ) highlights a sparsity of strongly extinguished afterglows , creating a dark burst desert , even though it is unclear if the preference of low extinctions is more than an observational and sample selection bias . our finding that most afterglows suffer from only low extinction in their hosts could indicate that afterglows are usually not obscured by dust close to the burster . one would then expect that the most extinguished afterglows are in fact located in globally dusty hosts . indeed , we find weak evidence for a correlation between the submm flux of grb host galaxies and the source frame extinction @xmath0 . although the statistical significance is low due to the small sample size and the large errors , this finding calls for a more thorough investigation . knowledge of @xmath1 and @xmath0 allowed us then to correct the afterglow light curves for intrinsic extinction and to derive the true luminosity distribution of our afterglow sample at chosen times in the host galaxy frame ( [ redshifting ] ) . we find that , on average , low-@xmath60 afterglows are less luminous than high-@xmath60 afterglows . the most likely explanation we have at hand for this finding is an observational bias against intrinsically faint afterglows at high redshifts . a bimodal distribution found by @xcite in similarly corrected x - ray afterglows is not clearly seen in the optical although , on average , grbs with fainter x - ray afterglows also have fainter optical afterglows . unfortunately , the available sample size is still too small to reach definite conclusions . a search for correlations between prompt emission parameters and the luminosity of the optical afterglows at jet break time has come up empty . since our sample is exclusively composed of grbs from the pre-_swift _ era , a similar study in a few years time on _ swift_-discovered grb afterglows will shed light on the dark burst desert and the true afterglow luminosity distribution by removing observational bias factors via rapid and highly precise grb localizations . already , _ swift _ has lead to the discovery of very faint afterglows including what may be the `` darkest '' burst ever , grb 050412 @xcite . the recent discovery of the first afterglows of short grbs @xcite opens the possibility of finally making a comparison of the environment of the two different classes of grbs . we thank the anonymous referee for helpful comments that improved this paper . d.a.k . and s.k . acknowledge financial support by dfg grant kl 766/13 - 2 . a.z . and s.k . acknowledge financial support by dfg grant kl 766/11 - 1 . we wish to thank s. covino , j. gorosabel , t. kawabata , b. c. lee , k. lindsay , e. maiorano , n. masetti , r. sato , m. uemura , p. m. vreeswijk and k. wiersema for contributing unpublished or otherwise unavailable data to the database , and s. cortes ( clemson university ) for reducing additional data . d.a.k . wishes to thank n. masetti and d. malesani for enlightening discussions . furthermore , we wish to thank scott barthelmy , nasa for the upkeep of the gcn circulars and jochen greiner , garching , for the `` grb big list '' . in the following , we consider the proper afterglow , i.e. , cleaned from any underlying host component as well as corrected for galactic extinction and extinction in the host galaxy . in cases where the late - time light curve is dominated by a supernova component , the data was removed due to a lack of knowledge concerning the supernova sed . we start with the expression for the flux density per unit frequency , @xmath118 , of a time - dependent source that is shifted to a redshift @xmath119 , which was originally observed at a redshift @xmath120 @xcite : @xmath121 here , @xmath122 and @xmath123 are measured in the observers frame , @xmath124 is the luminosity distance . the exponential function corrects for a postulated extinction in the host galaxy at redshift @xmath120 , where @xmath125 is the optical depth in the host at the observed frequency . when shifting an afterglow to a redshift @xmath119 , its apparent magnitude changes by an amount @xmath126 , with @xmath127 where @xmath128 is the wavelength - dependent filter response function for the given photometric band . for an unabsorbed afterglow @xmath129 we assume @xmath1 = const . since the light curve shape can be affected by a jet break and by rebrightening episodes ( which are achromatic according to our present data base ) , we do not specify @xmath130 . using eq . ( [ ag ] ) , it follows that in eq . ( [ dado ] ) @xmath131 with @xmath132 . any extinction correction is included in @xmath133 in eq . ( [ dado ] ) . after transforming flux density per unit frequency into flux density per unit wavelength and inserting eqs . ( [ dado ] , [ ag ] , [ f1 ] ) into eq . ( [ dm ] ) , we obtain @xmath134^{1-\beta}\ \frac{d_l(z_0)^2}{d_l(z_1)^2}\,\frac{\int_0^\infty s(\lambda ) \ , \lambda^{\beta-2 } \exp\{\tau(\lambda/(1+z_0))\}\ , d\lambda } { \int_0^\infty s(\lambda ) \ , \lambda^{\beta-2 } \,d\lambda}\right ) \label{delta1}\ ] ] in a logarithmic plot ( apparent magnitude vs. log time ) the first term in eq . ( [ delta1 ] ) represents a shift of the observed light curve in time . the second term in eq . ( [ delta1 ] ) shifts the light curve along the magnitude axis ; the light curve shape is not affected . in our calculations we assumed a flat universe with matter density @xmath135 , cosmological constant @xmath136 , and hubble constant @xmath137 km s@xmath138 mpc@xmath138 @xcite . for @xmath128 we used the filter functions of the vlt / fors1 bessel filters ( see http://www.eso.org/instruments/fors1/filters.html ) . we have searched the literature for publications that also derive the source frame extinction toward the gamma - ray - bursts listed in table [ taball ] . in most cases , these results are in good agreement with our findings . _ grb 970508 _ to our knowledge , @xcite was the first who discussed the extinction in a grb host galaxy derived via a fit of the observed sed in the optical bands . analysing collected data from the literature , he finds for grb 970508 @xmath139 ( albeit for a slightly higher @xmath60 , having left the redshift as a free parameter ) , concurrent with our result @xmath140 for the preferred mw dust extinction model . _ grb 971214 _ @xcite find a strong spectral curvature in their data and infer a large @xmath141 for an assumed redshift of @xmath142 . as a fixed value for the spectral slope , they took the @xmath143 color of the afterglow of grb 970508 . this strong spectral curvature is mirrored in our sed fit which finds a large extinction and @xmath30 . fits with a fixed @xmath1 derived from the @xmath19-@xmath1 relations @xcite are strongly rejected . _ grb 980519 _ we have found no references to intrinsic extinction in the literature . this makes our preferred smc dust result of @xmath144 , derived under the assumption of a redshift of @xmath13 , a novel result . _ grb 980703 _ this burst has the highest extinction in our sample ( table [ taball ] ) . this result is not unprecedented , however . while finding different values , several authors reported on a high extinction at different times after the burst : @xcite find @xmath145 0.9 days post - burst , @xcite find @xmath146 at 1.2 days , and @xcite derive @xmath147 at 5.3 days . our achromatic fit gives @xmath148 for the mw dust model , close to the result from @xcite , while smc dust gives @xmath59 . the light curves are not sampled well enough to check if the discrepancies between the various authors are the result of chromatic changes of the sed , a bumpy structure , or a true decrease of line - of - sight extinction with time . _ grb 990123 _ @xcite find a negative extinction for this burst and therefore fix it to zero , consistent with the low value we derive : @xmath149 for the preferred smc dust model . on the other hand , @xcite derive a very high extinction of @xmath150 from metal column abundances . given the high quality of the optical photometry , we can not reproduce such a high extinction . _ grb 990510 _ both , @xcite and @xcite find a slight curvature in the sed . removing the @xmath151-band data point , @xcite find @xmath152 , comparable to our result , @xmath153 for smc dust . neither publication fits the sed with a dust model to derive the host extinction . _ grb 991208 _ almost all data on this afterglow ( in @xmath154 ) is from @xcite . they do not discuss intrinsic extinction in the host galaxy . together with additional @xmath155-band data from @xcite and @xcite , we find a high extinction of @xmath156 for mw dust , the weakly preferred model , and an excellent fit ( @xmath9/dof=0.20 ) . _ grb 991216 _ after correcting for significant galactic foreground extinction , both @xcite and @xcite find a sed well approximated by a power - law , concurrent with no dust reddening , although these authors do not discuss intrinsic reddening . with a larger database and more colors , we find @xmath157 for the weakly preferred mw dust , still a low value . this is consistent with the value @xmath158 that @xcite find from a low - resolution spectrum . they also detect a broad absorption feature centered at 2360 , which can be interpreted as a redshifted 2715 bump , giving credence to our choice of mw dust as the preferred model . _ grb 000131 _ for an smc dust model , @xcite find @xmath159 . we are not able to perform a free fit as the sed has only three colors ( @xmath33 and @xmath160 are affected by lyman dampening ) , but fitting a straight power - law shows the sed is consistent with no extinction ( i.e. , no sed curvature is seen ) . _ grb 000301c _ @xcite find very low host extinction ( @xmath161 ) for this burst using an smc dust curve , and no solution for mw and lmc dust . @xcite derive a best fit result of @xmath162 for smc dust , find a slightly worse fit for lmc dust and a much worse fit for mw dust . we also find negative extinction for mw dust , and also low values for lmc and smc dust ( @xmath163 for smc dust ) , completely in concordance with the results of the other authors . _ grb 000911 _ comparing their host galaxy sed with synthetic spectra of extinguished galaxies , @xcite find @xmath164 for smc dust , consistent with our result of @xmath165 . the lmc result ( @xmath166 compared to @xmath167 ) is also in agreement . _ grb 000926 _ for smc dust , we find @xmath168 , in good agreement with @xcite who find @xmath169 for smc dust . several other analyses @xcite disagree more with our results , but @xcite employ a @xmath170 spectral energy distribution as we do , while the other studies have shorter baselines . based on metal column abundances from high - resolution spectra , @xcite find a large @xmath171 , which we can not confirm via the broad band photometry . _ grb 010222 _ working with five - color sdss photometry , @xcite find low extinction ( @xmath172 for @xmath173 fixed ) for this burst with an smc dust curve , which is less than what we derive ( @xmath174 for @xmath175 and smc dust ) . our result is consistent with that of @xcite , who find @xmath176 . @xcite find a higher extinction of @xmath177 based on metal column abundances . _ grb 010921 _ not finding a supernova bump in this nearby grb , @xcite invoke @xmath41 . while @xcite do find a weak sn bump and we derive the sed from this fit , we also find significant extinction of @xmath58 for mw dust , the preferred model , thus validating the analysis in @xcite . _ grb 011121 _ @xcite deduce a galactic visual extinction of @xmath178 along the line of sight by fitting the uncorrected data points with a mw extinction law at @xmath179 , which translates into a galactic @xmath180 , assuming @xmath1813.1 . they could not constrain the host extinction , however . @xcite find a galactic @xmath182 and a spectral slope of @xmath183 between about 0.4 and 1.5 days after the burst ; a possible extinction in the host galaxy is not considered . @xcite adopt @xmath184 based on the cobe data @xcite , find no evidence for host extinction , and @xmath185 at 2.5 days after the burst . they note , however , that the hi maps of @xcite favor a lower galactic extinction of @xmath171 mag , i.e. @xmath186 ( assuming @xmath181 3.1 ) . we refitted the combined data set taken from @xcite , adding carefully selected ( early and hst data ) data from @xcite and @xcite . in addition , we have now used @xmath186 deduced via the hi data @xcite , in order to overcome the potential mixture of an unknown source frame extinction and an uncertain galactic extinction along the line of sight . we omitted the @xmath187-band data @xcite , however , since they worsen the fit . in doing so , we now derive a moderate host extinction of @xmath188 mag ( smc dust ) . the mw dust model is not preferred , although the low redshift of the grb makes a differentiation among the three dust models hard . _ grb 011211 _ both @xcite and @xcite find very low extinction for this burst . @xcite rule out mw dust and find that smc dust gives the best fit with @xmath189 . our result is that mw dust is ruled out , lmc dust gives a low @xmath0 with a very bad fit , and smc dust gives a moderate @xmath190 but with @xmath191 . thus the results are comparable in principle . _ grb 020124 _ @xcite find @xmath52 by fixing @xmath192 for smc dust , deeming @xmath69 unrealistic . their sed is consistent with ours , showing a strong curvature . they additionally employ a synthetic @xmath151-band point derived from extrapolating the spectrum and find @xmath193 from a free fit of smc dust , still higher than what we derive ( excluding the @xmath151 band , as it is affected by lyman dampening ) , although the results are identical within the large errors . @xcite find @xmath95 to 0.9 . _ grb 020405 _ working with almost the same data set as we do , and including x - ray observations , @xcite find negligible ( though unspecified ) dust extinction in this afterglow . on the other hand , @xcite , comparing the x - ray and the optical brightness , invoke a gray extinction curve with a very high @xmath194 . we find moderate extinction @xmath195 for all three dust models . _ grb 020813 _ @xcite performed a very stringent analysis of an early high - resolution spectrum of this afterglow and , via analysis of metal column densities , derive @xmath196 , @xmath197 and @xmath198 for mw , lmc , and smc dust , respectively . our analysis gives a very good match , with definite results : @xmath199 , @xmath200 and @xmath201 , respectively . furthermore , we also come to the conclusion that smc dust is the preferred model . to make their unextincted spectral slope more compatible with standard models , @xcite invoke @xmath202 ( which converts to @xmath203 ) . @xcite deduce @xmath204 , but for the galactic extinction curve of @xcite . _ grb 021004 _ for an smc dust curve , @xcite derive @xmath205 , coupled with @xmath206 . this work finds a lower @xmath207 coupled with a higher @xmath208 . this is for the free fit , a wind model with the cooling break blueward of the optical gives @xmath209 fixed and @xmath210 , which is comparable . fitting an sed derived on october 11.67 , @xcite find exactly this result ( @xmath210 ) for smc dust , associated with @xmath211 . they find higher extinction for lmc dust and a worse fit , and negative extinction for mw dust , just as we do . _ grb 021211 _ @xcite derive extinction values for this grb by fixing @xmath1 . as our light curve fitting finds a very blue afterglow with @xmath212 ( i.e. , almost flat ) , these values can not be compared adequately . @xcite find up to @xmath213 , while we only derive small negative extinction for all dust models . since these authors have derived their spectral slope @xmath214 from an early @xmath215 color , this may imply color evolution ( cf . * ? ? ? @xcite derive an unextincted slope of @xmath216 at 0.87 days after the burst in the observer frame and an upper limit on the source frame extinction of @xmath196 . _ grb 030226 _ @xcite find negligible extinction for this afterglow based on their multicolor data set , in contrast to @xcite who , on theoretical grounds , argue for a fixed @xmath217 and thus derive small extinction from their @xmath218 . we utilize both data sets , and find a result concurrent with @xcite , mw and lmc dust are ruled out , and the reddening using smc dust ( @xmath219 ) is negligible within errors . _ grb 030227 _ @xcite were not able to conclusively fit their sed . their @xmath220 is in perfect agreement with the result we find , @xmath221 . while the data set is sparse , it is , with the exception of one data point , from @xcite exclusively . the steep @xmath11 supports dust reddening , and the unextinguished @xmath222 , a reasonable value , further supports our findings of @xmath223 , which could explain the faintness of this not too distant ( @xmath224 , * ? ? ? * ) afterglow . the sparse data points of the sed and their large errors lead to very large errors in @xmath1 and @xmath0 , however . _ grb 030323 _ the sed of this burst is unusual . the @xmath151 band can not be included due to lyman dampening , and the spectrum ( a strong dla ) shows that the @xmath160 band is also unreliable . with the exception of the @xmath155 band , the sed shows a strong curvature without being very steep . as there is no reason to remove the @xmath155 band point ( based on three photometric points , and not two each as in @xmath225 and @xmath187 ) , we tried to fit the sed with it ( removing it results in very good fits that have @xmath30 ) . as the data points have small errors , the result is a very bad fit ( @xmath226 for two degrees of freedom ) . @xcite do not fit the light curves , but analyse the sed at different post - burst times from observations at near - identical epochs . they fix @xmath227 , derived from the @xmath19-@xmath1 relations . using this value , they find @xmath228 , @xmath229 and @xmath230 for mw , lmc and smc dust curves , respectively . fixing @xmath1 to this value and fitting our sed yields @xmath231 , @xmath232 and @xmath233 for mw , lmc and smc dust curves , respectively . while these values are higher than those in @xcite , they are comparable if conservative errors are assumed . however , the @xmath155-band data is a strong outlier in these fits . _ grb 030328 _ after paper ii went into press , @xcite presented a large amount of @xmath234 data on this burst that was not available to us before . we refitted the optical light curves and derive the following parameters which supersede those presented in table 1 of paper ii : @xmath235 ( 1.34 in paper ii ) , dof = 81 ( 18 in paper ii ) , @xmath236 mag ( @xmath237 in paper ii ) , @xmath238 ( @xmath239 in paper ii ) , @xmath240 ( @xmath241 in paper ii ) , @xmath242 days after the burst ( @xmath243 in paper ii ) , @xmath244 ( fixed to 10 in paper ii ) . the host galaxy magnitude remains unchanged . ( see paper ii for the definitions of these values . ) these changes do not influence the conclusions of paper ii . we note that the value for the break smoothness parameter @xmath245 is fully in agreement with the possible correlation between @xmath6 and @xmath245 ( figure 8 of paper ii ) , and is almost identical to the value pair derived for grb 010222 . _ grb 030329 _ the sed of this burst was derived in an alternate way compared to all other seds . the rebrightening episodes starting at one day are found to be achromatic over several days . the high data quality makes it possible to shift the light curves to a common magnitude , the amount of shift is used to construct the sed , assuming conservative errors . this method is independent of the light curve fit , which is very complicated in the case of grb 030329 ( cf . paper ii ) . @xcite find @xmath248 , coupled with @xmath249 , for mw dust . our result is : @xmath250 , coupled with @xmath251 , which is comparable within errors . for @xmath252 fixed , @xcite derive @xmath253 , comparable to @xmath254 which we derive when fixing @xmath1 at this value . _ grb 030429 _ @xcite rule out mw dust ( as do we ) and find @xmath255 for smc dust at 0.548 days after the burst , slightly lower than our value of @xmath256 . they also find higher extinction for lmc dust , but the fit is much worse , contrary in part to our result . we find much lower source frame extinction for lmc dust , but here also the quality of the fit is bad . _ grb 040924 _ @xcite use their own data and data from the gcn archives to derive @xmath259 , slightly lower than the value we find ( @xmath260 ) . they do not fit the sed with a dust model . @xcite derive @xmath261 and argue that for the standard blastwave model , @xmath262 and thus @xmath263 mag . this is in agreement with our result @xmath264 mag . _ grb 041006 _ @xcite find @xmath265 and , to keep @xmath266 , assume negligible host galaxy extinction . this is in agreement with our result @xmath267 , but we note that the shape of the sed is indicative of a slight amount of dust . the derived @xmath268 and the shallow post - break light curve decay slope @xmath269 ( paper ii ) are both indicative of a hard electron spectrum with @xmath270 . achterberg , a. , gallant , y. a. , kirk , j. g. , & guthmann , a. w. 2001 , , 328 , 393 akerlof , c. , et al . 1999 , , 398 , 400 amati , l. , et al . 2002 , , 390 , 81 andersen , m. i. , et al . 2000 , , 364 , l54 ballantyne , d. r. , et al . 2002 , , 389 , l74 berger , e. , et al . 2001a , , 556 , 556 berger , e. , et al . 2001b , , 560 , 652 berger , e. , et al . 2002 , , 581 , 981 berger , e. , et al . 2003a , , 588 , 99 berger , e. , et al . 2003b , , 426 , 154 berger , e. , et al . 2005a , , 629 , 328 berger , e. , et al . 2005b , , 438 , 988 beuermann , k. , et al . 1999 , , 352 , l26 bloom , j. s. , et al . 1998 , , 508 , l21 bloom , j. s. , et al . 1999 , , 401 , 453 bloom , j. s. , et al . 1999 , gcn circ . 480 , http://gcn.gsfc.nasa.gov/gcn/gcn3/480.gcn3 bloom , j. s. , et al . 2004 , , 572 , l45 bloom , j. s. , et al . 2004 , , 127 , 252 bohlin , r. c. , et al . 1978 , , 224 , 132 bradley , j. , et al . 2005 , science , 307 , 244 butler , n. r. , et al . 2005a , , 621 , 884 butler , n. r. , et al . 2005b , , 629 , 908 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 castro - tirado , a. j. , et al . 1999 , , 511 , l85 castro - tirado , a. j. , et al . 2001 , , 370 , 398 castro - tirado , a. j. , et al . 2003 , , 411 , l315 chary , r. , et al . 2002 , , 566 , 229 chevalier , r. a. , & li , z .- y . 1999 , , 520 , l29 covino , s. , et al . 2003 , , 404 , l5 covino , s. , et al . 2005 , , in press ( astro - ph/0509144 ) dado , s. , dar , a. , & de rjula , a. 2002 , , 388 , 1079 della valle , m. , et al . 2003 , , 406 , l33 de luca , a. , et al . 2005 , , 440 , 85 dickey & lockman 1990 , , 28 , 215 djorgovski , s. g. , et al . 2001 , , 562 , 654 draine , b. t. 2000 , , 532 , 273 draine , b. t. , & hao , l. 2002 , , 569 , 780 firmani , c. et al . 2004 , , 611 , 1033 fitzpatrick , e. l. , & massa , d. 1986 , , 307 , 286 fox , d. w. , et al . 2003 , , 586 , l5 fox , d. w. , et al . 2005 , , 437 , 845 frail , d. a. , et al . 2002 , , 565 , 829 friedman , a. s. , & bloom , j. s. 2005 , , 627 , 1 fruchter , a. 2001 , gcn circ . 1087 , http://gcn.gsfc.nasa.gov/gcn/gcn3/1087.gcn3 fynbo , j. p. u. , et al . 2001a , , 369 , 373 fynbo , j. p. u. , et al . 2001b , , 373 , 796 fynbo , j. p. u. , et al . 2003 , , 406 , l63 fynbo , j. p. u. , et al . 2004 , , 609 , 962 fynbo , j. p. u. , et al . 2005 , , 633 , 317 galama , t. j. , & wijers , r. a. j. m. 2001 , , 549 , l209 galama , t. j. , et al . 1999 , , 398 , 394 galama , t. j. , et al . 2003 , , 587 , 135 garnavich , p. m. , loeb , a. , & stanek , k. z. 2000 , , 544 , l11 garnavich , p. m. , et al . 2000 , , 543 , 61 garnavich , p. m. , et al . 2003 , , 582 , 924 gendre , b. , & bor , m. 2005 , , 430 , 465 greiner , j. , et al . 2003 , , 599 , 1223 groot , p. j. , et al . 1998 , , 493 , l27 halpern , j. p. , et al . 1998 , , 393 , 41 halpern , j. p. , et al . 2000 , , 543 , 697 hjorth , j. , et al . 2003 , , 423 , 847 hjorth , j. , et al . 2003 , , 597 , 699 hjorth , j. , et al . 2005 , , 437 , 859 holland , s. t. , et al . 2002 , , 124 , 639 holland , s. t. , et al . 2003 , , 125 , 2291 holland , s. t. , et al . 2004 , , 128 , 1955 hopkins , p. f. , et al . 2004 , , 128 , 1112 jakobsson , p. , et al . 2003 , , 408 , 941 jakobsson , p. , et al . 2004a , , 427 , 785 jakobsson , p. , et al . 2004b , , 617 , l21 jakobsson , p. , et al . 2005 , , 629 , 45 jakobsson , p. , et al . 2005 , , 362 , 245 jaunsen , a. o. , et al . 2001 , , 546 , 127 jaunsen , a. o. , et al . 2003 , , 402 , 125 jensen , b. l. , et al . 2001 , , 370 , 909 kennicutt , r. c. , jr . 1998 , , 498 , 541 kirk , j. g. , guthmann , a. w. , gallant , y. a. , & achterberg , a. 2000 , , 542 , 235 klose , s. , et al . 2000 , , 545 , 271 klose , s. , et al . 2003 , , 592 , 1025 klose , s. , et al . 2004 , , 128 , 1942 kosugi , g. , et al . 2005 , gcn circ . 3263 , http://gcn.gsfc.nasa.gov/gcn/gcn3/3263.gcn3 lazzati , d. , covino , s. , & ghisellini , g. 2002 , , 330 , 583 le floch , e. , et al . 2003 , , 400 , 499 lee , b. c. , et al . 2001 , , 561 , 183 li , w. , et al . 2003 , , 115 , 844 liang , e. , & zhang , b. 2005 , , submitted ( astro - ph/0508510 ) lloyd - ronning , n. m. , et al . 2002 , , 574 , 554 macfadyen , a. i. , & woosley , s. e. 1999 , , 524 , 262 malesani , d. , et al . 2004 , , 609 , l5 maiolino , r. , et al . 2004 , , 431 , 533 maiorano , e. , et al . 2006 , , submitted masetti , n. , et al . 2000 , , 354 , 473 masetti , n. , et al . 2003 , , 404 , 465 masetti , n. , et al . 2005 , , 438 , 841 mirabal , n. , et al . 2003 , , 587 , 128 mller , p. , et al . 2002 , , 396 , l21 nardini , m. , et al . 2005 , , submitted ( astro - ph/0508447 ) nysewander , m. c. , et al . 2005 , , submitted ( astro - ph/0505474 ) owens , a. , et al . 1998 , , 339 , l37 paczyski , b. 1998 , , 494 , l45 panaitescu , a. , & kumar , p. 2000 , , 543 , 66 panaitescu , a. , & kumar , p. 2001 , , 556 , 1002 panaitescu , a. 2005 , , 362 , 921 pandey , s. b. , et al . 2004 , , 417 , 919 pei , y. c. 1992 , , 395 , 130 perna , r. , lazzati , d. & fiore , f. 2003 , , 585 , 775 piro , l. , et al . 2001 , , 558 , 442 predehl , p. , & schmitt , j. h. m. m. 1995 , , 293 , 889 price , p. a. , et al . 2001 , , 549 , l7 price , p. a. , et al . 2002 , , 572 , l51 price , p. a. , et al . 2003 , , 584 , 931 rau , a. , et al . 2004 , , 427 , 815 reichart , d. e. 1998 , , 495 , l99 rhoads , j. e. 2005 , gcn circ . 3527 , http://gcn.gsfc.nasa.gov/gcn/gcn3/3527.gcn3 rhoads , j. e. , & fruchter , a. s. 2001 , , 546 , 117 rol , e. , et al . 2005 , , 624 , 868 roming , p. w. a. , et al . 2005 , , submitted ( astro - ph/0509273 ) sagar , r. , et al . 2001 , basi 29 , 1 sari , r. , piran , t. , & halpern , j. p. 1999 , , 519 , l17 sari , r. , piran , t. , & narayan , r. 1998 , , 497 , l17 savaglio , s. , & fall , s. m. 2004 , , 614 , 293 savaglio , s. , et al . 2003 , , 585 , 638 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 schneider , d. p. , et al . 2005 , , 130 , 367 silvey , j. , et al . 2004 , gcn circ . 2833 , http://gcn.gsfc.nasa.gov/gcn/gcn3/2833.gcn3 soderberg , a. m. , et al . 2005 , , 627 , 877 soderberg , a. m. , et al . 2006 , , 636 , 391 soderberg , a. m. , & berger , e. 2005 , gcn circ . 4375 , http://gcn.gsfc.nasa.gov/gcn/gcn3/4375.gcn3 spergel , d. n. , et al . 2003 , , 148 , 175 stanek , k. z. , et al . 1999 , , 522 , l39 stanek , k. z. , et al . 2003 , , 563 , 592 stanek , k. z. , et al . 2003 , , 591 , l17 stratta , g. , et al . 2004 , , 608 , 846 stratta , g. , et al . 2005 , , 441 , 83 tanvir , n. , et al . 2004 , , 352 , 1073 taylor , g. b. , et al . 2000 , , 537 , l17 tiengo , a. , et al . 2004 , , 423 , 861 vreeswijk , p. m. , et al . 1999 , , 523 , 171 vreeswijk , p. m. , et al . 2004 , , 419 , 927 vreeswijk , p. m. , et al . 2005 , , in press ( astro - ph/0510404 ) watson , d. , et al . 2003 , , 595 , l29 waxman , e. , & draine , b. t. 2000 , , 537 , 796 weingartner , j. c. , & draine , b. t. 2001 , , 548 , 296 woosley , s. e. 1993 , , 405 , 273 yost , s. a. , et al . 2002 , , 577 , 155 yun , m. s. , & carilli , c. l. 2002 , , 568 , 88 zeh , a. , klose , s. , & hartmann , d. h. 2004 , , 609 , 952 ( paper i ) zeh , a. , klose , s. , & hartmann , d. h. 2005 , in proceedings of the 22nd texas symposium on relativistic astrophysics , ed . p. chen , e. bloom , g. madejski ( slac ) , & v. patrosian ( stanford university ) , preprint ( astro - ph/0503311 ) zeh , a. , klose , s. , & kann , d. a. 2006 , , in press ( astro - ph/0509299 ) ( paper ii ) zhang , b. , & mszros , p. 2004 , int.j.mod.phys . a19 ( 2004 ) 2385 l r c | c c c | c c c | c c c 970508 & & @xmath271 & 2.73 & @xmath272 & @xmath273 & 2.00 & @xmath274 & @xmath275 & 1.14 & @xmath276 & @xmath277 + 971214 & & @xmath278 & 0.32 & @xmath279 & @xmath280 & 0.14 & @xmath281 & @xmath282 & 0.09 & @xmath283 & @xmath284 + 980519 & 1 & @xmath234 & 0.61 & @xmath285 & @xmath286 & 0.73 & @xmath287 & @xmath288 & 0.01 & @xmath289 & @xmath290 + 980703 & & @xmath291 & 2.28 & @xmath292 & @xmath293 & 2.10 & @xmath294 & @xmath295 & 2.33 & @xmath296 & @xmath297 + 990123 & 2 & @xmath298 & 0.21 & @xmath299 & @xmath300 & 0.35 & @xmath301 & @xmath302 & 0.20 & @xmath303 & @xmath304 + 990510 & 3 & @xmath154 & 0.03 & @xmath305 & @xmath306 & 0.01 & @xmath307 & @xmath308 & 0.58 & @xmath309 & @xmath310 + 991208 & & @xmath311 & 0.20 & @xmath312 & @xmath313 & 0.17 & @xmath314 & @xmath315 & 0.26 & @xmath316 & @xmath317 + 991216 & 4 & @xmath291 & 0.12 & @xmath273 & @xmath318 & 0.16 & @xmath272 & @xmath319 & 0.29 & @xmath320 & @xmath321 + 000131 & & @xmath322 & @xmath3230.01&@xmath324 & 0 & @xmath3230.01 & @xmath324 & 0 & @xmath3230.01 & @xmath324 & 0 + 000301c & 5 & @xmath325 & 2.48 & @xmath326 & @xmath286 & 2.03 & @xmath327 & @xmath328 & 1.14 & @xmath329 & @xmath330 + 000911 & & @xmath331 & 0.26 & @xmath332 & @xmath333 & 0.28 & @xmath334 & @xmath335 & 0.33 & @xmath336 & @xmath337 + 000926 & 6 & @xmath291 & 1.02 & @xmath338 & @xmath339&1.67 & @xmath340 & @xmath341 & 0.37 & @xmath342 & @xmath343 + 010222 & 7 & @xmath344 & 1.35 & @xmath345 & @xmath346 & 1.36 & @xmath347 & @xmath348 & 0.58 & @xmath349 & @xmath350 + 010921 & & @xmath351 & 0.07 & @xmath352 & @xmath353 & 0.06 & @xmath354 & @xmath355 & 0.03 & @xmath356 & @xmath357 + 011121 & 8 & @xmath344 & 0.55 & @xmath275 & @xmath358 & 0.54 & @xmath275 & @xmath359 & 0.49 & @xmath360 & @xmath361 + 011211 & 9 & @xmath325 & 2.03 & @xmath362 & @xmath363 & 7.12 & @xmath364 & @xmath365 & 1.89 & @xmath366 & @xmath367 + 020124 & 10&@xmath368 & 0.14 & @xmath369 & @xmath370 & 0.15 & @xmath371 & @xmath372 & 0.01 & @xmath373 & @xmath374 + 020405 & 11&@xmath375 & 1.82 & @xmath376 & @xmath377 & 1.75 & @xmath378 & @xmath379 & 1.75 & @xmath380 & @xmath381 + 020813 & 12&@xmath170 & 1.62 & @xmath382 & @xmath383 & 1.10 & @xmath384 & @xmath385 & 0.95 & @xmath386 & @xmath387 + 021004 & & @xmath375 & 1.62 & @xmath388 & @xmath389 & 1.57 & @xmath390 & @xmath391 & 0.69 & @xmath392 & @xmath393 + 021211 & & @xmath375 & 1.16 & @xmath394 & @xmath395 & 1.13 & @xmath396 & @xmath397 & 1.05 & @xmath398 & @xmath399 + 030226 & 13&@xmath291 & 0.15 & @xmath400 & @xmath401 & 0.57 & @xmath402 & @xmath403 & 0.32 & @xmath404 & @xmath405 + 030227 & & @xmath406 & 1.41 & @xmath407 & @xmath408 & 0.74 & @xmath409 & @xmath410 & 0.48 & @xmath411 & @xmath412 + 030323 & & @xmath413 & 5.91 & @xmath414 & @xmath415 & 6.35 & @xmath416 & @xmath417 & 6.41 & @xmath418 & @xmath419 + 030328 & 14&@xmath234 & 0.23 & @xmath420 & @xmath421 & 0.22 & @xmath422 & @xmath423 & 0.18 & @xmath424 & @xmath425 + 030329 & & @xmath426 & 0.10 & @xmath427 & @xmath428 & 0.09 & @xmath429 & @xmath301 & 0.06 & @xmath430 & @xmath431 + 030429 & 15&@xmath432 & 2.60 & @xmath433 & @xmath434 & 7.72 & @xmath435 & @xmath436 & 1.79 & @xmath437 & @xmath438 + 030723 & & @xmath439 & 0.16 & @xmath440 & @xmath441 & 0.15 & @xmath442 & @xmath443 & 0.19 & @xmath444 & @xmath445 + 040924 & & @xmath446 & 0.10 & @xmath447 & @xmath448 & 0.10 & @xmath449 & @xmath450 & 0.09 & @xmath451 & @xmath452 + 041006 & 16&@xmath453 & 0.06 & @xmath454 & @xmath455 & 0.05 & @xmath456 & @xmath457 & 0.05 & @xmath458 & @xmath459 + [ taball ] l c c c c 970508 & * _ m _ & 2.73 & @xmath272 & @xmath273 + 990123 & * _ s _ & 0.20 & @xmath303 & @xmath304 + 991208 & m & 0.20 & @xmath312 & @xmath313 + 991216 & m & 0.12 & @xmath273 & @xmath318 + 000131 & @xmath460 & @xmath3230.01 & @xmath324 & 0 + 000301c & * _ s _ & 1.14 & @xmath329 & @xmath330 + 000911 & m & 0.26 & @xmath332 & @xmath333 + 000926 & * _ s _ & 0.37 & @xmath342 & @xmath343 + 010222 & * _ s _ & 0.58 & @xmath349 & @xmath350 + 011121 & s & 0.49 & @xmath360 & @xmath361 + 020405 & s & 1.75 & @xmath380 & @xmath381 + 020813 & s & 0.95 & @xmath386 & @xmath387 + 021004 & * _ s _ & 0.69 & @xmath392 & @xmath393 + 030226 & * _ s _ & 0.32 & @xmath404 & @xmath405 + 030328 & s & 0.18 & @xmath424 & @xmath425 + 030329 & s & 0.06 & @xmath430 & @xmath431 + 030429 & * _ s _ & 1.79 & @xmath437 & @xmath438 + 040924 & s & 0.09 & @xmath451 & @xmath452 + 041006 & m & 0.06 & @xmath454 & @xmath455 + [ tabgold ] * * * * * * * *
the parametric beating of a weak probe field with a prepared raman coherence in a far - off - resonance medium has been extensively studied @xcite . it has been demonstrated that multimode laser radiation @xcite and incoherent fluorescent light @xcite can be replicated into raman sidebands . since a substantial molecular coherence can be produced by the two - color adiabatic raman pumping method @xcite , the quantum conversion efficiency of the parametric beating technique can be maintained high even for weak light with less than one photon per wave packet @xcite . to describe the statistical properties of a weak quantum probe and its first - order stokes and anti - stokes sidebands in the parametric beating process , a simplified quantum treatment has recently been performed @xcite . it has been shown that the statistical properties of the quantum probe can be replicated into the two sidebands nearest to the input line , in agreement with the experimental observations @xcite . however , many experiments have reported the observations of ultrabroad raman spectra with a large number of sidebands @xcite . in the experiments with solid hydrogen @xcite , at least two anti - stokes sidebands and two stokes sidebands have been observed . in the experiment with molecular deuterium @xcite , a large raman coherence @xmath0 and about 20 raman sidebands , covering a wide spectral range from near infrared through vacuum ultraviolet , have been generated . in rare - earth doped dielectrics with low raman frequency and long - lived spin coherence , a substantial raman coherence @xmath1 and an extremely large number of sidebands ( about @xmath2 ) can also be generated @xcite . broad combs of raman sidebands @xcite have been intensively studied because they may synthesize to subfemtosecond @xcite and subcycle @xcite pulses . the generation of broad combs of raman sidebands has always been examined as a semiclassical problem . while classical treatments are sufficient for many purposes , a quantum treatment is required when the statistical properties of the radiation fields are important . on the other hand , broad combs of raman sidebands with similar nonclassical properties and different frequencies may find useful applications for high - performance optical communication . therefore , it is intriguing to examine the quantum aspects of high - order coherent raman processes . in this paper , we extend the treatment of ref . @xcite to study various quantum properties of multiorder sidebands generated by the beating of a quantum probe field with a prepared raman coherence in a far - off - resonance medium . under the conditions of negligible dispersion and limited bandwidth , we derive a bessel - function solution for the sideband field operators . we analytically and numerically calculate various quantum statistical characteristics of the sideband fields generated from a single - mode quantum input . we show that , with increasing the effective medium length or the raman sideband order , the autocorrelation functions , cross - correlation functions , photon - number distributions , and squeezing factors undergo oscillations governed by the bessel functions . meanwhile , the normalized autocorrelation functions and normalized squeezing factors of the single - mode probe field are not altered and can be replicated into a broad comb of generated multiorder raman sidebands . we study the mixing and modulation of photon statistical properties in the case of two - mode input . we show that the prepared raman coherence and the medium length can be used as control parameters to switch a sideband field from one type of photon statistics to another type , or from a non - squeezed state to a squeezed state and vice versa . we also discuss two - photon interference in coherent raman scattering . although the multiorder coherent raman scattering can produce a broad comb of sideband fields with different frequencies , it behaves in many aspects as a beam splitter @xcite . therefore , in this paper , we also make comparison of this conventional device with our system as and when it is possible . before we proceed , we note that , in related problems , the generation of correlated photons using the @xmath3 and @xmath4 parametric processes has been studied @xcite . the correlations between the stokes and anti - stokes sidebands and the possibility of transferring a quantum state of light from one carrier frequency to another carrier frequency ( multiplexing ) have been discussed for resonant systems @xcite . the paper is organized as follows . in sec . [ sec : model ] , we describe the model and present the basic equations . in sec . [ sec : single - mode ] , we study various quantum characteristics of the sideband fields generated from a single - mode quantum input . in sec . [ sec : two - mode ] , we discuss the quantum properties of the sideband fields generated from a two - mode quantum input . finally , we present the conclusions in sec . [ sec : summary ] . we consider a far - off - resonance raman medium shown schematically in fig . [ fig1 ] . level @xmath5 with energy @xmath6 is coupled to level @xmath7 with energy @xmath8 by a raman transition via intermediate levels that are not shown in the figure . we send a pair of long , strong , classical laser fields , with carrier frequencies @xmath9 and @xmath10 , and a short , weak , quantum probe field @xmath11 , with one or several carrier frequencies , through the raman medium , along the @xmath12 direction . the timing and alignment of these fields are such that they substantially overlap with each other during the interaction process . the driving laser fields are tuned close to the raman transition @xmath13 , with a small finite two - photon detuning @xmath14 , but are far detuned from the upper electronic states @xmath15 of the molecules . we assume that all the frequency components of the input probe field are separated by integer multiples of the raman modulation frequency @xmath16 . the driving fields adiabatically produce a raman coherence @xmath17 @xcite . when the probe field propagates through the medium , it beats with the prepared raman coherence . since the probe field is weak and short compared to the driving fields , the medium state and the driving fields do not change substantially during this step . the beating of the probe field with the prepared raman coherence leads to the generation of new sidebands in the total output field @xmath18 . the frequencies of the sideband fields @xmath19 are given by @xmath20 , where @xmath21 is integer and @xmath22 is a carrier frequency of the input probe field . the range of @xmath21 should be appropriate so that @xmath23 is positive . the probe field is taken to be not too short so that the fourier - transformation limited broadening is negligible . we assume that the prepared raman coherence @xmath17 is substantial so that the spontaneous raman process is negligible compared to the stimulated and parametric processes . consequently , the quantum noise can be neglected . unlike ref . @xcite , our model does not require any restriction on the magnitude of the coherence as all raman sidebands are included . when we take the propagation equation for the classical raman sidebands @xcite and replace the field amplitudes by the quantum operators , we obtain @xmath24 here , @xmath25 and @xmath26 are the dispersion and coupling constants , respectively . we have denoted @xmath27 , where @xmath28 is the molecular number density . we take all the sidebands to be sufficiently far from resonance that the dispersion of the medium is negligible . in this case , we have @xmath29 and @xmath30 . we write @xmath31 $ ] , where @xmath32 and @xmath33 , and assume that @xmath34 and @xmath35 are constant in time and space . we change the variables by @xmath36 $ ] . using photon operators , we can write @xmath37 . here , @xmath38 is the quantization length taken to be equal to the medium length , @xmath39 is the quantization transverse area taken to be equal to the beam area , @xmath40 is a bloch wave vector , and @xmath41 and @xmath42 are the annihilation and creation operators for photons in the spectral mode @xmath21 and the spatial mode @xmath40 . for simplicity , we restrict our discussion to the case where each sideband field contains only a single spatial mode ( with , e.g. , @xmath43 ) . then , eq . ( [ 1 ] ) yields @xmath44 where @xmath45 . for the medium length @xmath38 , the evolution time is @xmath46 . it follows from eq . ( [ 4 ] ) that the total photon number is conserved in time . note that eq . ( [ 4 ] ) represents the heisenberg equation for the fields that are coupled to each other by the effective interaction hamiltonian @xmath47 the interaction between the sideband fields via the prepared raman coherence is analogous to the interaction between the transmitted and reflected fields from a conventional beam splitter @xcite . however , the two mechanisms are very different in physical nature . the most important difference between them is that the two fields from the conventional beam splitter have the same frequency while the sideband fields in the raman scheme have different frequencies . in addition , the model hamiltonian ( [ 8b ] ) involves an infinitely large number of raman sidebands , separated by integer multiples of the raman modulation frequency @xmath48 . despite these differences , the model ( [ 8b ] ) can be called the multiorder raman beam splitter . the parameters @xmath49 determine the transmission and scattering coefficients for the fields at the raman beam splitter . we assume that the bandwidth of the generated raman spectrum is small compared to the characteristic probe frequency @xmath22 . in this case , the @xmath21-dependence of the coupling parameters @xmath50 can be neglected , that is , we have @xmath51 . with this assumption , we find the following solution to eq . ( [ 4 ] ) : @xmath52 here , @xmath53 is the @xmath54th - order bessel function . the expression ( [ 8 ] ) for the output field operators is a generalization of the bessel - function solution obtained earlier for the classical fields @xcite . the number of generated raman sidebands is characterized by the effective interaction time @xmath55 or , equivalently , the effective medium length @xmath56 , where @xmath57 the coefficient @xmath58 characterizes the strength of the parametric coupling and is proportional to the prepared raman coherence @xmath34 , that is , to the intensities of the driving laser fields . the bessel functions @xmath59 are the transmission ( @xmath60 ) and scattering ( @xmath61 ) coefficients for the raman sidebands , similar to the transmission and reflection coefficients of a conventional beam splitter . the assumption of limited bandwidth requires @xmath62 , that is , @xmath63 @xcite . in what follows we use the expression ( [ 8 ] ) to calculate various quantum statistical characteristics of the sideband fields , namely , the autocorrelation functions , the two - mode cross - correlation functions , the squeezing factors , and the relation between the @xmath64 representations of the output and input states in this section , we consider the case where the input probe field has a single carrier frequency @xmath22 . in other words , we assume that the sideband @xmath65 is initially prepared in a quantum state @xmath66 and the other sidebands are initially in the vacuum state . the density matrix of the initial state of the fields is given by @xmath67 we study the autocorrelations of photons in the generated raman sidebands . we use eq . ( [ 8 ] ) and apply the initial density matrix ( [ 8a ] ) to calculate the normally ordered moments @xmath68 of the photon - number operators @xmath69 . the result is @xmath70 in particular , the mean photon numbers of the sidebands are given by @xmath71 here , @xmath72 is the photon - number operator for the input field . the @xmath73th - order autocorrelation functions of the sidebands are defined by @xmath74 . from eqs . ( [ 9 ] ) and ( [ 9a ] ) , we find @xmath75 where @xmath76 is the @xmath73th - order autocorrelation function of the input field . ( first row ) and second - order autocorrelation function @xmath77 ( second row ) , both scaled to their initial values for the probe field , as functions of the effective medium length @xmath56 ( first column ) and the sideband order @xmath21 ( second column ) . in ( a ) and ( c ) , the sideband order is @xmath78 ( solid line ) , @xmath79 ( dashed line ) , and @xmath80 ( dotted line ) . in ( b ) and ( d ) , the effective medium length is @xmath81 . in ( c ) , we have amplified @xmath82 ( dashed line ) and @xmath83 ( dotted line ) by 5 and 10 times , respectively . in ( b ) and ( d ) , the negative side of the @xmath21 axis is not shown because the functions plotted are symmetric in @xmath21 . ] equations ( [ 9a ] ) and ( [ 9b ] ) indicate that , when we increase the effective medium length @xmath56 or the sideband order @xmath21 , the mean photon number @xmath84 and the autocorrelation function @xmath85 undergo oscillations as described by even powers of the bessel function @xmath86 . such oscillatory behavior is illustrated in fig . when the sideband order @xmath21 is higher , the onset of @xmath84 occurs later [ see fig . [ fig2](a ) ] and , hence , so does the onset of @xmath85 [ see fig . [ fig2](c ) ] . for a fixed @xmath21 , both @xmath84 and @xmath85 reach their largest values at the same optimal medium length @xmath87 , where @xmath88 is the position of the first peak of @xmath89 . the higher the sideband order @xmath21 , the larger is the optimal length @xmath90 and the smaller are the maximal output values of @xmath84 and @xmath85 [ see figs . [ fig2](a ) and [ fig2](c ) ] . figures [ fig2](b ) and [ fig2](d ) show that @xmath84 and @xmath85 are substantially different from zero only in the region where @xmath91 is not too large compared to @xmath56 . for a given @xmath56 , both @xmath84 and @xmath85 achieve their maximal values at @xmath92 . the normalized @xmath73th - order autocorrelation functions of the sidebands are defined by @xmath93 . these functions characterize the overall statistical properties , such as sub - poissonian , poissonian , or super - poissonian photon statistics , regardless of the mean photon number . unlike the mean photon number @xmath84 and the autocorrelation function @xmath85 , the normalized autocorrelation function @xmath94 does not oscillate when we change the effective medium length @xmath56 or the sideband order @xmath21 . indeed , with the help of eq . ( [ 9 ] ) , we find @xmath95 where @xmath96 . equation ( [ 10 ] ) indicates that the generated sideband fields and the probe field have the same normalized autocorrelation functions , which are independent of the evolution time and are solely determined by the statistical properties of the input field . in other words , the normalized autocorrelation functions of the probe field do not change during the parametric beating process and are precisely replicated into the comb of generated sidebands . such a replication of the normalized autocorrelation characteristics can be called autocorrelation multiplexing . in particular , if the photon statistics of the input field is sub - poissonian , poissonian , or super - poissonian , the photon statistics of each of the sideband fields will also be sub - poissonian , poissonian , or super - poissonian , respectively . this result is in agreement with the experiments on replication of multimode laser radiation @xcite and broadband incoherent light @xcite . the ability of the raman medium to multiplex the autocorrelation characteristics is similar to the ability of a conventional beam splitter @xcite . it is not surprising that the normalized autocorrelation functions of the probe field are replicated into the sidebands in the parametric beating process . such a replication is possible because the medium is far off resonance and the quantum probe field is weak compared to the driving fields . under these two conditions , the photon annihilation operators are linearly transformed as described by the linear differential equation ( [ 4 ] ) . this equation shows that the photon annihilation operators are not mixed up with the creation operators , and therefore the evolution of the annihilation operators is linear with respect to the initial annihilation operators . we should , however , emphasize here that the replication of all of the normalized autocorrelation functions @xmath97 ( for all orders @xmath73 ) of the input field does not mean the replication of the quantum state @xmath66 . in fact , the oscillations of the mean photon number @xmath84 and the autocorrelation function @xmath85 indicate that the photon - number distributions and consequently the quantum states of the sidebands evolve in a rather complicated way and are quite different from those for the input field . a separable state at the input can produce an entangled state @xcite . in addition , cross - correlations between the sidebands can be generated from initially uncorrelated fields , and a fock state at the input does not produce sideband fields in isolated fock states [ see below ] . we study the correlations between the generated raman sidebands . for two different sidebands @xmath54 and @xmath98 ( @xmath99 ) , we have @xmath100 the cross - correlation function for the two sidebands is defined by @xmath101 . using eqs . ( [ 9a ] ) and ( [ 11b ] ) , we find @xmath102 when we extend @xmath103 for @xmath104 , we have @xmath105 . according to eq . ( [ 11a ] ) , the cross - correlation function @xmath103 oscillates when we change the effective medium length @xmath56 or the sideband orders @xmath54 and @xmath98 . such oscillatory behavior is illustrated in fig . [ fig3 ] . , scaled with the second - order autocorrelation function of the input field , as a function of ( a ) the effective medium length @xmath56 and ( b ) the sideband order @xmath98 . here , the sideband order @xmath54 is fixed to @xmath79 . in ( a ) , the sideband order @xmath98 is @xmath78 ( solid line ) , @xmath106 ( dashed line ) , and @xmath80 ( dotted line ) . in ( b ) , the effective medium length is @xmath81 . in ( b ) , the negative side of the @xmath98 axis is not shown because the function plotted is symmetric with respect to the sideband orders . ] the normalized cross - correlation function is defined by @xmath107 . unlike the function @xmath103 , the normalized function @xmath108 does not oscillate . indeed , we find the relation @xmath109 as seen from the above relation , the normalized cross - correlation functions @xmath108 for all possible sideband pairs @xmath110 are equal to each other , to the normalized second - order autocorrelation function @xmath111 for each sideband @xmath21 , and to the normalized second - order autocorrelation function @xmath112 of the input field . when @xmath113 , that is , when the photon statistics of the input field is non - poissonian , we obtain @xmath114 , a signature of cross - correlations between the sidebands . such correlations are generated although the sidebands are initially not correlated . in particular , if the input field has a sub - poissonian photon statistics ( @xmath115 ) , anti - correlations between the sidebands ( @xmath116 ) will be generated . note that the conventional beam splitters also have a similar property @xcite . the generation of cross - correlations between the sidebands indicates that the quantum states of the generated sidebands are different from that of the input field . we emphasize that the anti - correlation generation can not be explained by the classical statistics of the fields with positive @xmath64 functions although the sideband dynamics is linear with respect to the field variables . to get deeper insight into the quantum properties of the generated raman sidebands , we derive the photon - number distributions for the output fields . the joint photon - number distribution for the output fields is defined by @xmath117 , where @xmath118 is the density matrix of the output state . here , @xmath119 is the evolution operator . using eqs . ( [ 8 ] ) and ( [ 8a ] ) , we find @xmath120 where @xmath121 is the photon - number distribution of the input field , and @xmath122 is the total photon number . from eq . ( [ 10a ] ) , the marginal photon - number distribution @xmath123 for the sideband @xmath21 is obtained as @xmath124^k p_{\mathrm{in}}(n+k ) . \label{10b}\ ] ] clearly , @xmath123 is in general different from @xmath125 . thus , despite the replication of the normalized autocorrelation functions , the photon - number distribution of the probe field is not replicated into the sidebands . we examine several particular cases . first , we consider the case where the probe field is initially prepared in a coherent state @xmath126 . this state is characterized by a poisson distribution @xmath127 for the photon number , where @xmath128 . in this case , ( [ 10a ] ) and ( [ 10b ] ) yield @xmath129 and @xmath130 respectively , where @xmath131 . thus , the generated sidebands are not correlated , and the marginal photon - number distributions for the individual sidebands remain poisson distributions during the evolution process . second , we consider the case where the probe field is initially in a fock state @xmath132 . in this case , we have @xmath133 . therefore , eq . ( [ 10b ] ) yields @xmath134^{n - n } \frac{n!}{n!(n - n)!}\ ] ] for @xmath135 , and @xmath136 for @xmath137 . meanwhile , eq . ( [ 10a ] ) yields @xmath138 , that is , the joint photon - number distribution is not a product of the marginal photon - number distributions for the individual sidebands . thus , the sidebands are correlated . they are not generated in isolated fock states . finally , we consider the case where the probe field is initially in a thermal state , which is characterized by a boltzmann photon - number distribution @xmath139 . in this case , eq . ( [ 10b ] ) yields @xmath140 where @xmath131 . thus , the marginal photon - number distributions for the individual sidebands remain boltzmann distributions during the evolution process . however , according to eq . ( [ 10a ] ) , we have @xmath138 , a signature of correlations between the generated sidebands . we examine the squeezing of the field quadratures . a field quadrature of the @xmath21th mode is defined by @xmath141 . we say that the @xmath21th mode is in a squeezed state if there exists such a phase @xmath142 that @xmath143 or , equivalently , @xmath144 , where @xmath145 . the squeezing degree is measured by the quantity @xmath146 . note that the relation between the squeezing factor @xmath147 and the conventional squeezing parameter @xmath148 is @xmath149 . in terms of the photon operators , we have @xmath150 + [ ( \langle \hat b_q^2\rangle-\langle \hat b_q\rangle^2 ) e^{-2i\varphi}+\mathrm{c.c . } ] . \label{12c}\ ] ] using eqs . ( [ 8 ] ) and ( [ 8a ] ) , we find @xmath151 and @xmath152 when we insert eqs . ( [ 9a ] ) , ( [ 12a ] ) , and ( [ 12b ] ) into eq . ( [ 12c ] ) , we obtain @xmath153 here , @xmath154 denotes the squeezing factor for the @xmath142-quadrature of the input field . equation ( [ 12 ] ) shows that , if @xmath155 , then @xmath156 . thus , if the input field is in a squeezed state , then the generated sidebands are also in squeezed states . in other words , the squeezing of the input field is transferred to the comb of generated sidebands during the parametric beating process . the squeezing factor @xmath157 of the sideband @xmath21 is reduced from the input squeezing factor @xmath154 by the factor @xmath158 . unlike the case of linear directional couplers and beam splitters @xcite , the squeezing degree of the probe field can not be completely transferred to the raman sidebands . this difference is due to the fact that the linear directional coupler and the beam splitter involve only two output modes while the multiorder coherent raman process involves many more output modes . note that the phase of the squeezed quadrature of the sideband @xmath21 changes by @xmath159 . this means that the squeezed quadrature of a generated even - order ( odd - order ) raman sideband is parallel ( orthogonal ) to that of the input field . we introduce the normalized squeezing factor @xmath160 . we find the relation @xmath161 where @xmath162 is the normalized squeezing factor for the input field . equation ( [ 12d ] ) implies that , besides a shift of the quadrature phase angle , the normalized squeezing factor for the input field is replicated into the comb of generated sidebands . this result can be used to convert squeezing to a new frequency , i.e. , to perform squeezing multiplexing . the relation @xmath163 indicates that the @xmath56- and @xmath21-dependences of the squeezing factor @xmath157 are similar to those of the mean photon number @xmath164 [ see figs . [ fig2](a ) and [ fig2](b ) ] . note that , when the input field is in a coherent state , the sideband fields have no squeezing . this property is similar to the case of four - wave mixing but is unlike the case of degenerate parametric down - conversion , where perfect squeezing can in principle be obtained . since squeezed states are nonclassical states , the ability of the raman medium to multiplex squeezing from a probe field to its sidebands is a quantum property that can not be described by the classical statistics of the fields with positive @xmath64 functions . we calculate the quantum state of the output fields for several cases . first , we consider the case where the input sideband @xmath78 is initially in a coherent state @xmath126 . the state of the fields at the input is @xmath165 the state of the fields at the output is given by @xmath166 . since @xmath167 , we have @xmath168 . using eq . ( [ 8 ] ) , we find @xmath169 here , @xmath170 is a coherent state of the @xmath21th mode , with the amplitude @xmath171 thus , a probe field in a coherent state can produce sideband fields that are also in coherent states . such a process can be called coherent - state multiplexing . this property of the raman medium is similar to the case of conventional beam splitters @xcite . second , we consider the case where the input sideband @xmath78 is initially prepared in a fock state @xmath132 . the state of the fields at the input is written as @xmath172 the output state of the fields is given by @xmath173 . with the help of eq . ( [ 8 ] ) , we find @xmath174 here , @xmath175 for @xmath176 , and @xmath177 for @xmath178 . when @xmath179 and @xmath180 , the output state ( [ 15a ] ) is , in general , a multipartite inseparable ( entangled ) state . in a particular case where the input state of the probe field is a single - photon state , i.e. , @xmath181 , eqs . ( [ 15a ] ) and ( [ 15b ] ) yield @xmath182 here , @xmath183 is the quantum state of a single photon in the sideband @xmath21 with no photons in the other sidebands . in this case , the entanglement between two different sidebands @xmath54 and @xmath98 at the output can be measured by the bipartite concurrence @xmath184 , see @xcite . finally , we consider the case where the input sideband @xmath78 is initially in an incoherent mixed state @xmath185 with the help of eq . ( [ 15a ] ) , the density matrix of the output state of the fields is found to be @xmath186 to obtain the reduced density matrix @xmath187 for an arbitrary sideband @xmath21 , we trace the total density matrix ( [ 15d ] ) over all sidebands except for the sideband @xmath21 . then , we find @xmath188 where the marginal photon - number distribution @xmath123 is given by eq . ( [ 10b ] ) . as seen , the reduced state of each sideband is also an incoherent superposition of fock states . however , if the input state is different from the vacuum state , then , for @xmath180 , we have @xmath189 , a signature of correlations between the generated sidebands . moreover , the total density matrix ( [ 15d ] ) of the output fields contains nonzero off - diagonal matrix elements in the fock - state basis . in a particular case where the initial state of the probe field is a thermal state , i.e. , @xmath190(|n\rangle\langle n|)_0 $ ] , the reduced state of each generated sideband is also a thermal state , namely , @xmath191 here , @xmath131 is the mean photon number for the sideband @xmath21 . the reduced thermal states of the generated sidebands are , however , not isolated from each other . a far - off - resonance medium with a substantial raman coherence , prepared by two strong driving fields , can efficiently mix and modulate the quantum statistical properties of the sideband fields . to understand this mechanism , we study the case where the input probe field has two carrier frequencies , @xmath22 and @xmath192 , separated by an integer multiple @xmath193 of the raman modulation frequency @xmath48 . we assume that the raman sidebands @xmath78 and @xmath193 are initially in independent quantum states @xmath66 and @xmath194 , respectively , while the other sidebands are initially in the vacuum state . the density matrix of the initial state of the fields is given by @xmath195 here , @xmath196 . we study the mixing and modulation of photon statistics of the sideband fields . when we use eq . ( [ 8 ] ) to calculate the mean photon numbers of the sidebands generated from the initial state ( [ 18 ] ) , we find @xmath197 . \label{19}\end{aligned}\ ] ] furthermore , we find @xmath198 . \qquad \label{20b}\end{aligned}\ ] ] hence , the second - order autocorrelation function @xmath199 of the sideband @xmath21 is found to be @xmath200 , \nonumber\\ \label{20a}\end{aligned}\ ] ] where @xmath201 , \nonumber\\ \delta_3&=&\langle\hat b_0^\dagger(0)\rangle [ \langle\hat b_\nu^\dagger(0)\hat b_\nu^2(0)\rangle -\langle\hat b_\nu^\dagger(0)\hat b_\nu(0)\rangle\langle \hat b_\nu(0)\rangle].\end{aligned}\ ] ] the first two terms on the right - hand sides of eqs . ( [ 19 ] ) , ( [ 20b ] ) , and ( [ 20a ] ) are the contributions of the individual input sidebands @xmath78 and @xmath193 . the other terms result from the interference between the two interaction channels . as a function of ( a ) the effective medium length @xmath56 and ( b ) the sideband order @xmath21 in the case of two - mode input . the input sideband @xmath78 is initially prepared in a fock state , with @xmath80 photons . the input sideband @xmath79 is initially prepared in a thermal state , with 1 photon in average . in ( a ) , the sideband order is @xmath202 ( solid line ) and @xmath106 ( dashed line ) . in ( b ) , the effective medium length is @xmath203 . ] unlike the case of single - mode input , in the case of two - mode input , the normalized second - order autocorrelation function @xmath204 depends , in general , on @xmath56 and @xmath21 . such behavior is illustrated in fig . [ fig4 ] . when @xmath56 is such that @xmath205 or @xmath206 , we have @xmath207 or @xmath208 , respectively . consequently , if the two input sidebands have different normalized autocorrelation functions , i.e. , @xmath209 , then , with increasing @xmath56 or @xmath21 , the normalized autocorrelation function @xmath210 will oscillate between the values @xmath211 and @xmath212 [ see fig . [ fig4 ] ] . in particular , if the photon statistics of one of the input fields , e.g. , the sideband @xmath78 , is sub - poissonian [ @xmath213 and that of the other input field is super - poissonian [ @xmath214 , then each generated sideband @xmath21 will have complex statistical properties and will oscillate between sub - poissonian [ @xmath215 and super - poissonian [ @xmath216 photon statistics [ see fig . [ fig4 ] ] . using the prepared raman coherence @xmath34 or the medium length @xmath38 as a control parameter , we can switch a sideband field from super - poissonian photon statistics to sub - poissonian or vice versa . similar modulation of photon statistics has been demonstrated in a linear directional coupler @xcite . we study the mixing and modulation of the squeezing properties of the sideband fields . when we use eq . ( [ 8 ] ) to calculate the amplitudes @xmath217 and @xmath218 of the sidebands generated from the initial state ( [ 18 ] ) , we find the expressions @xmath219 and @xmath220 we insert eqs . ( [ 19 ] ) , ( [ 20 ] ) , and ( [ 21 ] ) into eq . ( [ 12c ] ) . then , we obtain the squeezing factor @xmath221 where @xmath222 and @xmath223 are the initial squeezing factors of the sidebands @xmath78 and @xmath193 , respectively . as seen , the squeezing factor @xmath147 of the sideband @xmath21 is a superposition of the input squeezing factors @xmath222 and @xmath223 , taken with the quadrature phase shifts @xmath224 and @xmath225 , respectively , and weighted by the factors @xmath158 and @xmath226 , respectively . as a function of ( a ) the effective medium length @xmath56 and ( b ) the sideband order @xmath21 in the case of two - mode input . the input sideband @xmath78 is initially prepared in a squeezed vacuum state , with the squeezing parameter @xmath227 and the phase @xmath228 . the input sideband @xmath79 is initially prepared in a thermal state , with the mean photon number 1 . in ( a ) , the sideband order is @xmath202 ( solid line ) and @xmath106 ( dashed line ) . in ( b ) , the effective medium length is @xmath203 . ] unlike the case of single - mode input , in the case of two - mode input , the normalized squeezing factor @xmath229 varies , in general , with @xmath56 and @xmath21 . such behavior is illustrated in fig . [ fig5 ] . when @xmath56 is such that @xmath205 or @xmath206 , we have @xmath230 or @xmath231 , respectively . consequently , if the normalized squeezing factors @xmath232 and @xmath233 of the two input fields are different , the normalized squeezing factor @xmath234 will oscillate between the values @xmath232 and @xmath233 . in particular , if one of the two input fields , e.g. , the sideband @xmath78 , is squeezed [ @xmath235 and the other input field is not squeezed [ @xmath236 , then , each generated sideband @xmath21 will have complex squeezing properties and will oscillate between a squeezed state [ @xmath237 and a non - squeezed state [ @xmath238 , see fig . using the prepared raman coherence @xmath34 or the medium length @xmath38 as a control parameter , we can switch a sideband field from a non - squeezed state to a squeezed state or vice versa . note that a similar result has been obtained for a linear directional coupler @xcite . and ( b ) mean photon number @xmath239 as functions of the effective medium length @xmath56 in the case where the sideband @xmath78 is initially prepared in a weak squeezed vacuum state and the sideband @xmath79 is initially prepared in a strong coherent state . the parameters for the initial states of the input sidebands are @xmath227 , @xmath228 , and @xmath240 . the curves are calculated for the sidebands @xmath78 ( solid line ) , @xmath79 ( dashed line ) , and @xmath241 ( dotted line ) . ] we analyze a particular case where the sideband @xmath78 is initially in a squeezed vacuum state @xmath242|0\rangle_0 $ ] and the sideband @xmath193 is initially in a coherent state @xmath243 . here , @xmath244 is a complex number , the modulus @xmath245 characterizes the amount of squeezing , and the phase angle @xmath246 characterizes the alignment of the squeezed vacuum state in phase space . for the input sideband @xmath78 , we have @xcite @xmath247 , @xmath248 , and @xmath249 . for the input sideband @xmath193 , we have @xmath250 , @xmath251 , and @xmath252 . then , we find from eq . ( [ 19 ] ) that the mean photon number of an arbitrary sideband @xmath21 is @xmath253 we find from eq . ( [ 22 ] ) that the maximal squeezing of the sideband @xmath21 occurs in the @xmath254-quadrature where @xmath255 . the corresponding value of the squeezing factor is @xmath256 as seen from eq . ( [ 24 ] ) , squeezing can be transferred from the initial squeezed vacuum state of the sideband @xmath78 to the other sidebands . the squeezing factors of the sidebands are independent of the amplitude @xmath257 of the initial coherent state of the sideband @xmath193 . meanwhile , the mean photon number of each sideband is governed not only by the squeezing parameter @xmath258 of the initial state of the sideband @xmath78 but also by the amplitude @xmath257 of the initial state of the sideband @xmath193 . using this fact , we can manipulate to get optimized mean photon numbers and squeezing degrees of the sideband fields at the output as per requirement . in particular , we can convert squeezing from a weak field to a much stronger field . to illustrate this possibility , we plot in fig . [ fig6 ] the squeezing factor @xmath259 and the mean photon number @xmath239 as functions of the effective medium length @xmath56 for the parameters @xmath227 , @xmath228 , @xmath260 , and @xmath261 . in this case , the most negative value of the input squeezing factor @xmath262 is achieved at @xmath263 and is given by @xmath264 , indicating the squeezing degree 86% . the mean photon number of the input squeezed vacuum state is @xmath265 , rather small . the solid lines in fig . [ fig6 ] show that the sideband @xmath78 , initially prepared in a weak squeezed vacuum state , can be significantly enhanced while keeping its squeezing degree substantial . meanwhile , the dashed lines show that , for @xmath266 , the sideband @xmath79 , initially prepared in a strong coherent state , is squeezed by about 29% and has the mean photon number of about 41 . similarly , the dotted lines show that , for @xmath267 , a generated new sideband @xmath241 is squeezed by about 16% and has the mean photon number of about 39 . thus , from a weak squeezed field at the input , we can obtain other output squeezed fields that have smaller but still substantial squeezing degrees , much larger mean photon numbers , and different frequencies . we show the possibility of quantum interference between the probability amplitudes for a pair of photons with different frequencies in the coherent raman process . we assume that the sidebands @xmath78 and @xmath79 are initially prepared in independent single - photon states . this initial condition corresponds to the situation where two photons with different frequencies @xmath22 and @xmath268 are incident into the raman medium . the input state of the fields can be written as @xmath269 the output state of the fields is given by @xmath270 . with the help of eq . ( [ 8 ] ) , we find @xmath271|1_k 1_l\rangle . \label{16a}\end{aligned}\ ] ] here , the fock state @xmath272 is the state of two photons in the sideband @xmath21 with no photons in the other sidebands , and the fock state @xmath273 is the state in which there is one photon in each of the sidebands @xmath54 and @xmath98 but no photons in the other sidebands . it follows from eq . ( [ 16a ] ) that the probability for finding two photons in the sideband @xmath21 is @xmath274 the joint probability for finding one photon in each of the sidebands @xmath54 and @xmath98 ( @xmath99 ) is given by @xmath275 ^ 2 . \label{16d}\ ] ] the probability @xmath276 for having one and only one photon in the sideband @xmath21 is @xmath277 the mean photon number of the sideband @xmath21 is @xmath278 we find the relations @xmath279 , @xmath280 , and @xmath281 , which reflect the symmetry of the generated stokes and anti - stokes sidebands with respect to the two input sidebands 0 and 1 . for finding one photon in each of the sidebands @xmath78 and @xmath79 as a function of the effective medium length @xmath56 . ] when we insert @xmath60 and @xmath282 into eq . ( [ 16d ] ) , we obtain the following expression for the joint probability for finding one photon in each of the sidebands @xmath78 and @xmath79 : @xmath283 ^ 2.\end{aligned}\ ] ] this expression shows that the joint probability @xmath284 may become zero at certain values of @xmath56 [ see fig . [ fig7 ] ] . this is a signature of destructive interference between two channels that form the state @xmath285 . in the first channel , each of the two photons individually transmits through the medium without any changes . the two - photon probability amplitude for this channel is @xmath286 . in the second channel , both the photons are scattered from the prepared raman coherence and exchange their sidebands . the two - photon probability amplitude for this channel is @xmath287 . since raman scattering produces a phase shift of @xmath288 for each photon , the probability amplitudes for the two channels ( the transmission and scattering of both the photons ) are @xmath289 out of phase . the interference between the two channels is therefore destructive , yielding the output state @xmath285 with the joint probability @xmath284 given above . when the medium length @xmath38 is such that @xmath290 , the interference between the two two - photon amplitudes becomes completely destructive , and therefore the state @xmath285 is removed from the output state ( [ 16a ] ) . we denote such a medium length by @xmath291 . the positions of the zeros of @xmath284 depicted in fig . [ fig7 ] indicate that the first three values of @xmath291 are given by @xmath292 , 3.11 , and 4.68 . it is interesting to note that @xmath293 can be determined in an experiment using a single - mode input . indeed , in the case where a single sideband @xmath78 is initially excited , the mean photon numbers of the generated sidebands are given by eq . ( [ 9a ] ) . therefore , the effective medium length @xmath293 corresponds to the situation where the probe sideband @xmath78 and its adjacent sidebands @xmath294 have the same mean photon numbers at the output . there exist literatures on two - photon interference in various systems @xcite . two - photon interference in coherent raman scattering , described above , is an analogy of two - photon interference at a conventional beam splitter @xcite . we emphasize that two - photon interference in coherent raman scattering involves copropagating photons with different frequencies in a collinear scheme . to be more general , we consider the case where an arbitrary number of sidebands is initially excited . we find that an arbitrary multimode coherent state @xmath295 of the input fields produces a coherent state @xmath296 of the output fields . here , the output amplitudes @xmath297 are linearly transformed from the input amplitudes @xmath298 as given by @xmath299 consequently , the diagonal coherent - state representation @xmath300 of an arbitrary input quantum state @xmath301 determines the representation @xmath302 of the output state @xmath303 via the equation @xmath304 here , we have introduced the notation @xmath305 if the input state @xmath301 is a classical state @xcite , @xmath300 must be non - negative and less singular than a @xmath14 function , and consequently so must @xmath302 . in this case , the output state @xmath303 is also a classical state . moreover , since the multimode coherent state @xmath306 is separable and the weight factor @xmath302 is non - negative , the output state @xmath303 is , by definition , separable @xcite . therefore , a necessary condition for the output fields to be in an inseparable ( entangled ) state or , more generally , in a nonclassical state is that the input field state is a nonclassical state . a similar condition has been derived for the beam splitter entangler @xcite . note that , in the case where we use a single - mode input field @xmath65 , prepared in an arbitrary quantum state with the coherent - state representation @xmath307 ( the stokes and anti - stokes sideband fields are initially in the vacuum state ) , eq . ( [ 17 ] ) becomes @xmath308 it has been shown in ref . @xcite that , when the input field is prepared in an even or odd coherent state , a multipartite entangled coherent state can be generated . we have studied the quantum properties of multiorder sidebands generated by the beating of a quantum probe field with a prepared raman coherence in a far - off - resonance medium . under the conditions of negligible dispersion and limited bandwidth , we have derived a bessel - function solution for the sideband field operators . we have analytically and numerically calculated various quantum statistical characteristics of the multiorder sideband fields . we have examined the quantum properties of the sideband fields in the case of single - mode quantum input . we have shown that , when we change the effective medium length or the raman sideband order , the autocorrelation functions , the cross - correlation functions , the photon - number distributions , and the squeezing factors undergo oscillations governed by the bessel functions . when the sideband order is higher , the onset of the sideband generation occurs later and , therefore , so does the onset of the sideband autocorrelation functions . the mean photon number and the autocorrelation functions of each sideband reach their largest values at the same optimal medium length determined by the first peak of the corresponding bessel function . the higher the sideband order , the larger is the optimal length and the smaller is the maximal output values of the mean photon number and sideband autocorrelation functions . meanwhile , the normalized autocorrelation functions and normalized squeezing factors of the probe field are not altered by the parametric beating process . they are replicated into the comb of generated multiorder sidebands . as the result , the single - mode input field and the generated sidebands have identical normalized autocorrelation functions and identical normalized squeezing factors . in other words , they have similar quantum statistical properties the same type of photon statistics and the same type of squeezing . in addition to this resemblance , it has been shown that an input field in a coherent state can produce sideband fields in coherent states . it has also been shown that , when the input field is prepared in a thermal state , the reduced state of each generated sideband is also a thermal state . therefore , the multiorder coherent raman process can be used to multiplex the statistical properties of a quantum probe field into a broad comb of different frequencies . as far as replicating the statistical properties of the input probe into its sidebands is concerned , the raman medium appears to behave as a linear system . however , the replication of the normalized autocorrelation functions and normalized squeezing factors of the probe field does not mean the replication of the quantum state at all . the photon - number distributions and the quantum states of the sidebands evolve in a rather complicated way . cross - correlations between the sidebands can be generated from initially uncorrelated fields . an inseparable state can be generated from a separable nonclassical state . although the dynamics of our model system is linear with respect to the field variables , the possibilities of interesting quantum phenomena such as anti - correlation generation , squeezing multiplexing , and entangled - state generation represent the quantum properties that can not be described by the classical statistics of the fields with positive @xmath64 functions . we have also studied the mixing and modulation of photon statistical properties in the case of two - mode quantum input . we have shown that the prepared raman coherence and the medium length can be used as control parameters to switch a sideband field from one type of photon statistics to another type , or from a non - squeezed state to a squeezed state and vice versa . in addition , we can switch nonclassical properties , such as sub - poissonian photon statistics and squeezing , from one frequency to another frequency . we have also shown an example of quantum interference between the probability amplitudes for a pair of photons with different frequencies . we have made interesting observations that the multiorder coherent raman scattering behaves in many aspects as a conventional beam splitter and hence can be called a multiorder raman beam splitter . the two systems have the same underlying physics : the fields are linearly transformed from the input values . however , the two systems are different in their natures . unlike the conventional beam splitter , the multiorder coherent raman process can efficiently produce a broad comb of sideband fields whose frequencies are different and are separated by integer multiples of the raman modulation frequency . the number of generated raman sidebands increases with the effective medium length , which is proportional to the product of the medium length and the prepared raman coherence . the bessel functions of the effective medium length play a similar role as the transmission and reflection coefficients of a conventional beam splitter . the ability of the far - off - resonance raman medium to generate a broad comb of fields with similar quantum statistical properties and to switch the quantum statistical characteristics of the radiation fields from one type to another type may find useful applications for high - performance optical communication networks . in addition , two - photon interference in coherent raman scattering may find various applications for high - precision measurements and also for quantum computation . finally , we emphasize that the coupling between the raman sidebands can be controlled by the magnitude of the prepared raman coherence , that is , by the intensities of the driving fields . in a realistic far - off - resonance raman medium , such as molecular hydrogen or deuterium vapor @xcite , solid hydrogen @xcite , and rare - earth doped dielectrics @xcite , a large raman coherence and , consequently , a large number of raman sidebands can be generated by the two - color adiabatic pumping technique . in such a system , the generation of a broad comb of high - order raman sidebands with nonclassical statistical properties is , in principle , feasible . therefore , we expect that the coherent raman scattering technique using quantum fields will become a practical and efficient method for a wide range of applications in nonlinear and quantum optics . v. p. kalosha and j. herrmann , phys . lett . * 85 * , 1226 ( 2000 ) ; v. p. kalosha and j. herrmann , opt . lett . * 26 * , 456 ( 2001 ) ; fam le kien , nguyen hong shon , and k. hakuta , phys . a * 64 * , 051803(r ) ( 2001 ) ; fam le kien , k. hakuta , and a. v. sokolov , _ ibid . _ * 66 * , 023813 ( 2002 ) ; v. kalosha , m. spanner , j. herrmann , and m. ivanov , phys . lett . * 88 * , 103901 ( 2002 ) ; r. a. bartels , t. c. weinacht , n. wagner , m. baertschy , chris h. greene , m. m. murnane , and h. c. kapteyn , _ ibid . _ * 88 * , 013903 ( 2002 ) . s. e. harris and a. v. sokolov , phys . a * 55 * , r4019 ( 1997 ) ; a. v. sokolov , d. d. yavuz , and s. e. harris , opt . lett . * 24 * , 557 ( 1999 ) ; a. v. sokolov , d. d. yavuz , d. r. walker , g. y. yin , and s. e. harris , phys . a * 63 * , 051801(r ) ( 2001 ) . h. p. yuen and j. h. shapiro , ieee trans . theory * it-26 * , 78 ( 1980 ) ; m. ley and r. loudon , opt . commun . * 54 * , 317 ( 1985 ) ; b. yurke , s. l. mccall , and j. r. klauder , phys . rev . a * 33 * , 4033 ( 1986 ) ; s. prasad , m. o. scully , and w. martienssen , opt . commun . * 62 * , 139 ( 1987 ) ; z. y. ou , c. k. hong , and l. mandel , opt . commun . * 63 * , 118 ( 1987 ) ; h. fearn and r. loudon , opt . commun . * 64 * , 485 ( 1987 ) ; r. a. campos , b. e. a. saleh , and m. c. teich , phys . rev . a * 40 * , 1371 ( 1989 ) . j. torgerson , d. branning , c. monken , and l. mandel , phys . a * 204 * , 323 ( 1995 ) ; k. mattle , h. weinfurter , p. g. kwiat , and a. zeilinger , phys . * 76 * , 4656 ( 1996 ) ; d. bouwmeester , j. pan , k. mattle , m. eibl , h. weinfurter , and a. zeilinger , nature ( london ) * 390 * , 575 ( 1997 ) ; t. c. ralph , n. k. langford , t. b. bell , and a. g. white , phys . a * 65 * , 062324 ( 2002 ) ; t. b. pittman , b. c. jacobs , and j. d. franson , phys . rev . lett . * 88 * , 257902 ( 2002 ) ; s. p. walborn , a. n. de oliveira , s. padua , and c. h. monken , phys . rev . lett . * 90 * , 143601 ( 2003 ) . m. d. lukin , a. b. matsko , m. fleischhauer , and m. o. scully , phys . 82 * , 1847 ( 1999 ) ; a. s. zibrov , a. b. matsko , o. kocharovskaya , y. v. rostovtsev , g. r. welch , and m. o. scully , _ ibid . _ * 88 * , 103601 ( 2002 ) . z. y. ou and l. mandel , phys . lett . * 61 * , 54 ( 1988 ) ; z. y. ou and l. mandel , _ ibid_. * 62 * , 2941 ( 1989 ) ; h. huang and j. h. eberly , j. mod . optics * 40 * , 915 ( 1993 ) ; m. o. scully , u. w. rathe , c. su , and g. s. agarwal , opt . commun . * 136 * , 39 ( 1997 ) ; a. k. patnaik and g. s. agarwal , j. mod . optics * 45 * , 2131 ( 1998 ) .
detecting a very weak signal which is almost inaccessible within the classical ( i.e. , non - quantum ) regime is one of the most important subjects in quantum information science . a strong motivation to devise such an ultra - precise sensor stems from the field of gravitational wave detection @xcite . in fact , a variety of linear sensors composed of opto - mechanical oscillators have been proposed @xcite , and several experimental implementations of those systems in various scales have been reported @xcite . it is well known that in general a linear sensor is subjected to two types of fundamental noises , i.e. , the _ back - action noise _ and the _ shot noise_. as a consequence , the measurement noise is lower bounded by the _ standard quantum limit ( sql ) _ @xcite , which is mainly due to the presence of back - action noise . hence , high - precision detection of a weak signal requires us to devise a sensor that evades the back - action noise and eventually beats the sql ; i.e. , we need to have a sensor achieving _ back - action evasion ( bae)_. in fact , many bae methods have been developed especially in the field of gravitational wave detection , e.g. , the variational measurement technique @xcite or the quantum locking scheme @xcite . moreover , towards more accurate detection , recently we find some high - level approaches to design a bae sensor , based on those specific bae methods . for instance , ref . @xcite provides a systematic comparison of several bae methods and gives an optimal solution . also systems and control theoretical methods have been developed to synthesize a bae sensor for a specific opto - mechanical system @xcite ; in particular , the synthesis is conducted by connecting an auxiliary system to a given plant system by _ direct - interaction _ @xcite or _ coherent feedback _ @xcite . along this research direction , therefore , in this paper we set the goal to develop a general systems and control theory for engineering a sensor achieving bae , for both the coherent feedback and the direct - interaction configurations . the key tool used here is the _ geometric control theory _ @xcite , which had been developed a long time ago . this is indeed a beautiful theory providing a variety of controller design methods for various purposes such as the non - interacting control and the disturbance decoupling problem , but , to our best knowledge , it has not been applied to problems in quantum physics . actually in this paper we first demonstrate that the general synthesis problem of a bae sensor can be formulated and solved within the framework of geometric control theory , particularly the above - mentioned disturbance decoupling problem . + this paper is organized as follows . section [ sec:2 ] is devoted to some preliminaries including a review of the geometric control theory , the general model of linear quantum systems , and the idea of bae . then , in section [ sec:3 ] , we provide the general theory for designing a coherent feedback controller achieving bae , and demonstrate an example for an opto - mechanical system . in section [ sec:4 ] , we discuss the case of direct interaction scheme , also based on the geometric control theory . finally , in section [ sec:5 ] , for a realistic opto - mechanical system subjected to a thermal environment ( the perfect bae is impossible in this case ) , we provide a convenient method to find an approximated bae controller and show how much the designed controller can suppress the noise . + for a matrix @xmath0 , @xmath1 , @xmath2 , and @xmath3 represent the transpose , hermitian conjugate , and element - wise complex conjugate of @xmath4 , respectively ; i.e. , @xmath5 , @xmath6 , and @xmath7 . @xmath8 and @xmath9 denote the real and imaginary parts of a complex number @xmath10 . @xmath11 and @xmath12 denote the zero matrix and the @xmath13 identity matrix . @xmath14 and @xmath15 denote the kernel and the image of a matrix @xmath4 , i.e. , @xmath16 and @xmath17 . let us consider the following _ classical _ linear time - invariant system : @xmath18 where @xmath19 is a vector of system variables , @xmath20 and @xmath21 are vectors of input and output , respectively . @xmath22 , and @xmath23 are real matrices . in the laplace domain , the input - output relation is represented by @xmath24 where @xmath25 and @xmath26 are the laplace transforms of @xmath27 and @xmath28 , respectively . @xmath29 is called the _ transfer function_. in this subsection , we assume @xmath30 . now we describe the geometric control theory , for the disturbance decoupling problem @xcite . the following _ invariant subspaces _ play a key role in the theory . * definition 1 : * let @xmath31 be a linear map . then , a subspace @xmath32 is said to be @xmath4-_invariant _ , if @xmath33 . * definition 2 : * given a linear map @xmath31 and a subspace im@xmath34 @xmath35 , a subspace @xmath32 is said to be @xmath36-_invariant _ , if @xmath37 . * definition 3 : * given a linear map @xmath31 and a subspace ker@xmath38 @xmath35 , a subspace @xmath39 is said to be @xmath40-_invariant _ , if @xmath41 . * definition 4 : * assume that @xmath42 is @xmath40-invariant , @xmath43 is @xmath36-invariant , and @xmath44 . then , @xmath45 is said to be a @xmath46-pair . from definitions 2 and 3 , we have the following two lemmas . * lemma 1 : * @xmath32 is @xmath36-invariant if and only if there exists a matrix @xmath47 such that @xmath48 * lemma 2 : * @xmath32 is @xmath40-invariant if and only if there exists a matrix @xmath49 such that @xmath50 the disturbance decoupling problem is described as follows . the system of interest is represented , in an extended form of eq . , as @xmath51 where @xmath52 is the disturbance and @xmath53 is the output to be regulated . @xmath54 and @xmath55 are real matrices . the other output @xmath28 may be used for constructing a feedback controller ; see fig . [ fig_feedbackframework ] . the disturbance @xmath52 can degrade the control performance evaluated on @xmath53 . thus it is desirable if we can modify the system structure by some means so that eventually @xmath52 dose not affect at all on @xmath53 . to @xmath56 is zero for all @xmath57 , for the modified system . or equivalently , the controllable subspace with respect to @xmath52 is contained in the unobservable subspace with respect to @xmath53 . ] this control goal is called the disturbance decoupling . here we describe a specific feedback control method to achieve this goal ; note that , as shown later , the direct - interaction method for linear quantum systems can also be described within this framework . the controller configuration is illustrated in fig . [ fig_feedbackframework ] ; that is , the system modification is carried out by combining an auxiliary system ( controller ) with the original system ( plant ) , so that the whole closed - loop system satisfies the disturbance decoupling condition . the controller with variable @xmath58 is assumed to take the following form : @xmath59 where @xmath60 , @xmath61 , @xmath62 , and @xmath63 are real matrices . then , the closed - loop system defined in the augmented space @xmath64 is given by @xmath65 & = \left[\begin{array}{cc } a+bd_{\scriptscriptstyle k}c & bc_{\scriptscriptstyle k}\\ b_{\scriptscriptstyle k}c & a_{\scriptscriptstyle k}\\ \end{array}\right ] \left[\begin{array}{c } x\\ x_{{\scriptscriptstyle k}}\\ \end{array}\right ] + \left[\begin{array}{c } e\\ o\\ \end{array}\right ] d , ~ \nonumber \\ z&=\left[\begin{array}{cc } h & o\\ \end{array}\right ] \left[\begin{array}{c } x\\ x_{{\scriptscriptstyle k}}\\ \end{array}\right].\end{aligned}\ ] ] the control goal is to design @xmath66 so that , in eq . , the disturbance signal @xmath52 dose not appear in the output @xmath53 : see the below footnote . here , let us define @xmath67 , \label{eq_extended_matrixae}\end{aligned}\ ] ] @xmath68 , @xmath69 , @xmath70 , and @xmath71 . then , the following theorem gives the solvability condition for the disturbance decoupling problem . * theorem 1 : * for the closed - loop system , the disturbance decoupling problem via the dynamical feedback controller has a solution if and only if there exists a @xmath46-pair @xmath72 satisfying @xmath73 note that this condition does not depend on the controller matrices to be designed . the following corollary can be used to check if the solvability condition is satisfied . * corollary 1 : * for the closed - loop system , the disturbance decoupling problem via the dynamical feedback controller has a solution if and only if @xmath74 where @xmath75 is the maximum element of @xmath36-invariant subspaces contained in @xmath76 , and @xmath77 is the minimum element of @xmath40-invariant subspaces containing @xmath78 . these subspaces can be computed by the algorithms given in appendix a. once the solvability condition described above is satisfied , then we can explicitly construct the controller matrices @xmath79 . the following intersection and projection subspaces play a key role for this purpose ; that is , for a subspace @xmath80 , let us define @xmath81 \in \mathcal{v}_{{\scriptscriptstyle e } } \right\ } , ~ \\ \mathcal{v}_{{\scriptscriptstyle p}}:&= \left\{x \in \mathcal{x}~\bigg| \left[\begin{array}{c } x\\ x_{{\scriptscriptstyle k}}\\ \end{array}\right ] \in \mathcal{v}_{{\scriptscriptstyle e}},~ \exists { x}_{{\scriptscriptstyle k } } \in \mathcal{x}_{{\scriptscriptstyle k } } \right\}.\end{aligned}\ ] ] then , the following theorem is obtained : * theorem 2 : * suppose that @xmath82 is a @xmath83-pair . then , there exist @xmath84 , @xmath85 , and @xmath63 such that @xmath86 and @xmath87 hold , where @xmath88 . moreover , there exists @xmath89 with @xmath90 , and @xmath91 has an invariant subspace @xmath92 such that @xmath93 and @xmath94 . also , @xmath95 satisfies @xmath96 where @xmath97 is a linear map satisfying @xmath98 . in fact , under the condition given in theorem 2 , let us define the following augmented subspace @xmath99 : @xmath100 \bigg|~ x \in \mathcal{v}_{2 } \right\}.\end{aligned}\ ] ] then , @xmath93 and @xmath94 hold , and we have @xmath101 & = \left[\begin{array}{cc } a+bd_{\scriptscriptstyle k}c & bc_{\scriptscriptstyle k}\\ b_{\scriptscriptstyle k}c & a_{\scriptscriptstyle k}\\ \end{array}\right ] \left[\begin{array}{c } x\\ nx\\ \end{array}\right ] \nonumber\\ & = \left[\begin{array}{c } ( a+bf)x\\ n(a+bf)x\\ \end{array}\right ] \in \mathcal{v}_{{\scriptscriptstyle e}},\end{aligned}\ ] ] implying that @xmath102 is actually @xmath91-invariant . now suppose that theorem 1 holds , and let us take the @xmath46-pair @xmath82 satisfying eq . . then , together with the above result ( @xmath103 ) , we have @xmath104^{\top } \subseteq \mathcal{v}_{{\scriptscriptstyle e } } \subseteq { { \rm ker}\hspace{0.07cm}}[h~o]$ ] . this implies that @xmath52 must be contained in the unobservable subspace with respect to @xmath53 , and thus the disturbance decoupling is realized . here we describe a general linear quantum system composed of @xmath105 bosonic subsystems . the @xmath106-th mode can be modeled as a harmonic oscillator with the canonical conjugate pairs ( or quadratures ) @xmath107 and @xmath108 satisfying the canonical commutation relation ( ccr ) @xmath109 . let us define the vector of quadratures as @xmath110^{\top}$ ] . then , the ccrs are summarized as @xmath111 . \end{aligned}\ ] ] note that @xmath112 is a @xmath113 block diagonal matrix . the linear quantum system is an open system coupled to @xmath114 environment fields via the interaction hamiltonian @xmath115 , where @xmath116 is the field annihilation operator satisfying @xmath117 . also @xmath118 is given by @xmath119 with @xmath120 . in addition , the system is driven by the hamiltonian @xmath121 with @xmath122 . then , the heisenberg equation of @xmath123 is given by @xmath124 where @xmath125 is defined by @xmath126 = \left[\begin{array}{cc } ( \hat{a}_j+\hat{a}_j^*)/\sqrt{2 } \\ ( \hat{a}_j-\hat{a}_j^*)/\sqrt{2 } i \\ \end{array}\right].\ ] ] the matrices are given by @xmath127 and @xmath128 with @xmath129^\top \in \mathbb{r}^{2 \times 2n}$ ] . also , the instantaneous change of the field operator @xmath125 via the system - field coupling is given by @xmath130 summarizing , the linear quantum system is characterized by the dynamics and the output , which are exactly of the same form as those in eq . ( @xmath131 in this case ) . however note that the system matrices have to satisfy the above - described special structure , which is equivalently converted to the following _ physical realizability condition _ @xcite : @xmath132 the opto - mechanical oscillator illustrated in fig . [ fig_opticalsensor ] is a linear quantum system , which serves as a sensor for a very weak signal . let @xmath133 and @xmath134 be the oscillator s position and momentum operators , and @xmath135 represents the annihilation operator of the cavity mode . the system hamiltonian is given by @xmath136 ; that is , the oscillator s free evolution with resonant frequency @xmath137 plus the linearized radiation pressure interaction between the oscillator and the cavity field with coupling strength @xmath138 . the system couples to an external probe field ( thus @xmath139 ) via the coupling operator @xmath140 , with @xmath141 the coupling constant between the cavity and probe fields . the corresponding matrix @xmath142 and vector @xmath143 are then given by @xmath144,~~ c_{1}=\sqrt{\frac{\kappa}{2 } } \left[\begin{array}{c } 0 \\ 0 \\ 1 \\ i \\ \end{array}\right].\end{aligned}\ ] ] the oscillator is driven by an unknown force @xmath145 with coupling constant @xmath146 ; then the vector of system variables @xmath147^\top$ ] satisfies @xmath148 where @xmath149 , \ , b=\sqrt{\gamma}\left[\begin{array}{c } 0 \\ 1 \\ 0 \\ 0 \\ \end{array}\right ] , \nonumber \\ c_1&=-b_1^{\top } = \left[\begin{array}{cccc } 0 & 0 & \sqrt{\kappa } & 0\\ 0 & 0 & 0 & \sqrt{\kappa}\\ \end{array}\right ] , \nonumber \\ \hat{w}_{1 } & = [ \hat{q}_{1 } , \hat{p}_{1}]^{\top } , ~~ \hat{w}_{1}^{\rm out } = [ \hat{q}_{1}^{\rm out } , \hat{p}_{1}^{\rm out}]^{\top}. \label{eq_matrix_originalplant}\end{aligned}\ ] ] note that we are in the rotating frame at the frequency of the probe field . these equations indicate that the information about @xmath150 can be extracted by measuring @xmath151 by a homodyne detector . actually the measurement output in the laplace domain is given by @xmath152 where @xmath153 , @xmath154 , and @xmath155 are transfer functions given by @xmath156 thus , @xmath151 certainly contains @xmath150 . note however that it is subjected to two noises . the first one , @xmath157 , is the back - action noise , which is due to the interaction between the oscillator and the cavity . the second one , @xmath158 , is the shot noise , which inevitably appears . now , the normalized output is given by @xmath159 and the normalized noise power spectral density of @xmath160 in the fourier domain @xmath161 is calculated as follows : @xmath162 the lower bound is called the sql . note that the last inequality is due to the heisenberg uncertainty relation of the normalized noise power , i.e. , @xmath163 . hence , the essential reason why sql appears is that @xmath164 contains both the back - action noise @xmath165 and the shot noise @xmath166 . therefore , toward the high - precision detection of @xmath150 , we need bae ; that is , the system structure should be modified by some means so that the back - action noise is completely evaded in the output signal ( note that the shot noise can never be evaded ) . the condition for bae can be expressed in terms of the transfer function as follows @xcite ; i.e. , for the modified ( controlled ) sensor , the transfer function from the back - action noise to the measurement output must satisfy @xmath167 equivalently , @xmath164 contains only the shot noise @xmath158 ; hence , in this case the signal to noise ratio can be further improved by injecting a @xmath158-squeezed ( meaning @xmath168 ) probe field into the system . there are two schemes for controlling a quantum system via feedback . the first one is the _ measurement - based feedback _ @xcite illustrated in fig . [ fig_mfvscf ] ( a ) . in this scheme , we measure the output fields and feed the measurement results back to control the plant system . on the other hand , in the _ coherent feedback _ scheme @xcite shown in fig . [ fig_mfvscf ] ( b ) , the feedback loop dose not contain any measurement component and the plant system is controlled by another quantum system . recently we find several works comparing the performance of these two schemes @xcite . in particular , it was shown in @xcite that there are some control tasks that can not be achieved by any measurement - based feedback but can be done by a coherent one . more specifically , those tasks are realizing bae measurement , generating a quantum non - demolished variable , and generating a decoherence - free subsystem ; in our case , of course , the first one is crucial . hence , here we aim to develop a theory for designing a coherent feedback controller such that the whole controlled system accomplishes bae . as discussed in section [ sec:2 - 1 ] , the geometric control theory for disturbance decoupling problem is formulated for the controlled system with special structure ; in particular , the coefficient matrix of the disturbance @xmath52 is of the form @xmath169^\top$ ] and that of the state vector in the output @xmath53 is @xmath170 $ ] . here we consider a class of coherent feedback configuration such that the whole closed - loop system dynamics has this structure , in order for the geometric control theory to be directly applicable . first , for the plant system given by eqs . and , we assume that the system couples to all the probe fields in the same way ; i.e. , @xmath171 this immediately leads to @xmath172 . next , as the controller , we take the following special linear quantum system with @xmath173 input - output fields : @xmath174 where the matrices @xmath175 satisfy the physical realizability condition . note that , corresponding to the plant structure , we assumed that the controller couples to all the fields in the same way , specified by @xmath176 . here we emphasize that the number of channels , @xmath114 , should be as small as possible from a viewpoint of implementation ; hence in this paper let us consider the case @xmath177 . now , we consider the coherent feedback connection illustrated in fig . [ fig_3i3o_general ] , i.e. , @xmath178 where @xmath179 and @xmath180 are @xmath181 unitary matrices representing the scattering process of the fields ; recall that the scattering process @xmath182 with @xmath183 the phase shift can be represented in the quadrature form as @xmath184 = s(\theta ) \left[\begin{array}{cc } \hat{q } \\ \hat{p } \\ \end{array}\right ] = \left[\begin{array}{cc } \cos \theta & -\sin \theta \\ \sin \theta & \cos \theta \\ \end{array}\right ] \left[\begin{array}{cc } \hat{q } \\ \hat{p } \\ \end{array}\right ] . \label{eq_phaseshifter}\ ] ] here we apply the coherent feedback scheme elaborated in section [ sec:3 - 2 ] to the opto - mechanical system studied in section [ sec:2 - 3 ] . the goal is , as mentioned before , to determine the controller matrices @xmath95 such that the closed - loop system achieves bae . here , we provide a step - by - step procedure to solve this problem ; the relationships of the class of controllers determined in each step is depicted in fig . [ fig_controllerconstraint ] . -phase shifter corresponding to eq . . , width=264 ] * ( ii ) * in the next step we apply theorem 1 to check if there exists a feedback controller such that the above closed - loop system achieves bae ; recall that the necessary and sufficient condition is eq . , i.e. , @xmath206 , where now @xmath207 \right\ } , \\ \mathcal{h}&={{\rm ker}\hspace{0.07cm}}h={\rm span } \left\ { \left[\begin{array}{c } 1\\ 0\\ 0\\ 0\\ \end{array}\right ] , \left[\begin{array}{c } 0\\ 1\\ 0\\ 0\\ \end{array}\right ] , \left[\begin{array}{c } 0\\ 0\\ 1\\ 0\\ \end{array}\right ] \right\}. \end{aligned}\ ] ] to check if this solvability condition is satisfied , we use corollary 1 ; from @xmath208 and @xmath209 , the algorithms given in appendix a yield @xmath210 implying that the condition in corollary 1 , i.e. , @xmath211 , is satisfied . thus , we now see that the bae problem is solvable , as long as there is no constraint on the controller parameters . next we aim to determine the controller matrices @xmath95 , using theorem 2 . first we set @xmath212 and @xmath213 ; note that @xmath82 is a @xmath83-pair . then , from theorem 2 , there exists a feedback controller with dimension @xmath214 . moreover , noting again that @xmath196 , there exist matrices @xmath84 , @xmath85 , and @xmath215 such that @xmath216 these conditions lead to @xmath217 , ~~ g=\left[\begin{array}{cc } 0&g_{\scriptscriptstyle 12}\\ -\frac{g}{\sqrt{\kappa}}&g_{\scriptscriptstyle 22}\\ g_{\scriptscriptstyle 31}&g_{\scriptscriptstyle 32}\\ 0&\sqrt{\kappa}\\ \end{array}\right ] , \\ n&=\left[\begin{array}{cccc } n_{\scriptscriptstyle 11 } & n_{\scriptscriptstyle 12 } & 0 & n_{\scriptscriptstyle 14 } \\ n_{\scriptscriptstyle 21 } & n_{\scriptscriptstyle 22 } & 0 & n_{\scriptscriptstyle 24}\\ \end{array}\right ] , \end{aligned}\ ] ] where @xmath218 , and @xmath219 are free parameters . then the controller matrices @xmath95 can be identified by eq . with the above matrices @xmath220 ; specifically , by substituting @xmath221 , @xmath222 , and @xmath223 in eq . , we have @xmath224 which yield @xmath225 where @xmath226 is the right inverse to @xmath215 , i.e. , @xmath227 . * ( iii ) * note again that the controller has to satisfy the physical realizability condition , which is now @xmath228 and @xmath229 . these constraints are represented in terms of the parameters as follows : @xmath230 where @xmath231 and @xmath232 . this is one of our main results ; the linear controller achieving bae for the opto - mechanical oscillator can be fully parametrized by eq . satisfying the condition . we emphasize that this full parametrization of the controller can be obtained thanks to the general problem formulation based on the geometric control theory . * ( iv ) * in practice , of course , we need to determine a concrete set of parameters to construct the controller . especially here let us consider a _ passive system _ ; this is a static quantum system such as an empty optical cavity . the main reason for choosing a passive system rather than a non - passive ( or _ active _ ) one such as an optical parametric oscillator is that , due to the external pumping energy , the latter could become fragile and also its physical implementation must be more involved compared to a passive system @xcite . now the condition for the system @xmath233 to be passive is given by @xmath234 and @xmath235 ; the general result of this fact is given in theorem 3 in appendix b. from these conditions , the system parameters are imposed to satisfy , in addition to eq . , the following equalities : @xmath236 there is still some freedom in determining @xmath237 , which however corresponds to simply the phase shift at the input - output ports of the controller , as indicated from eq . . thus , the passive controller achieving bae in this example is unique up to the phase shift . here particularly we chose @xmath238 and @xmath239 . then the controller matrices satisfying eqs . and are determined as @xmath240,~~ c_{\scriptscriptstyle k}=-b_{\scriptscriptstyle k}^{\top } = \left[\begin{array}{cc } \frac{g}{\sqrt{\kappa } } & 0 \\ 0 & \frac{g}{\sqrt{\kappa } } \\ \end{array}\right ] . \ ] ] as illustrated in fig . [ fig_3i3obae ] , the controller specified by these matrices can be realized as a single - mode , 2-inputs and 2-outputs optical cavity with decay rate @xmath241 and detuning @xmath242 . in other words , if we take the cavity with the following hamiltonian and the coupling operator ( @xmath243 is the cavity mode ) @xmath244 then to satisfy the bae condition the controller parameters @xmath245 must satisfy @xmath246 summarizing , the above - designed sensing system composed of the opto - mechanical oscillator ( plant ) and the optical cavity ( controller ) , which are combined via coherent feedback , satisfies the bae condition . hence , it can work as a high - precision detector of the force @xmath150 below the sql , particularly when the @xmath247-squeezed probe input field is used ; this fact will be demonstrated in section [ sec:5 ] . in this section , we study another control scheme for achieving bae . as illustrated in fig . [ fig_mankei_scheme ] ( a ) , the controller in this case is directly connected to the plant , not through a coherent feedback ; hence this scheme is called the _ direct interaction_. the controller is characterized by the following two hamiltonians : @xmath248 where @xmath249^{\top}$ ] is the vector of controller variables with @xmath250 the number of modes of the controller . @xmath251 is the controller s self hamiltonian with @xmath252 . also @xmath253 with @xmath254 , @xmath255 represents the coupling between the plant and the controller . note that , for the hamiltonians @xmath251 and @xmath253 to be hermitian , the matrices must satisfy @xmath256 and @xmath257 ; these are the physical realizability conditions in the scenario of direct interaction . in particular , here we consider a plant system interacting with a single probe field @xmath197 , with coupling matrices @xmath258 and @xmath259 . then , the whole dynamics of the augmented system with variable @xmath260^{\top}$ ] is given by @xmath261 where @xmath262 , ~~ b_{{\scriptscriptstyle e}}=\left[\begin{array}{c } b\\ o \\ \end{array}\right ] , \nonumber \\ c_{{\scriptscriptstyle e}}&=\left[\begin{array}{cc } c & o \\ \end{array}\right ] , ~~ b_{{\scriptscriptstyle e}}=\left[\begin{array}{cc } b^{\top } & o\\ \end{array}\right ] ^{\top}. \end{aligned}\ ] ] note that @xmath263 , @xmath264 , and @xmath265 are the same matrices as those in eq . . also , comparing the matrices and , we have that @xmath266 , which thus leads to @xmath267 and @xmath268 in theorem 2 . now , again for the opto - mechanical system illustrated in fig . [ fig_opticalsensor ] , let us aim to design the direct interaction controller , so that the whole system achieves bae ; that is , the problem is to determine the matrices @xmath269 so that the back - action noise @xmath270 does not appear in the measurement output @xmath271 . for this purpose , we go through the same procedure as that taken in section [ sec:3 - 3 ] . * ( i ) * because of the structure of the matrices @xmath263 and @xmath264 , the system is already of the form , where the geometric control theory is directly applicable . * ( ii ) * because we now focus on the same plant system as that in section [ sec:3 - 3 ] , the same conclusion is obtained ; that is , the bae problem is solvable as long as there is no constraint on the controller matrices @xmath269 . the controller matrices can be determined in a similar way to section [ sec:3 - 3 ] as follows . first , because the @xmath46-pair @xmath82 is the same as before , it follows that @xmath272 , i.e. , @xmath273 . then , from theorem 2 with the fact that @xmath267 and @xmath268 , we find that the direct interaction controller can be parameterized as follows : @xmath274 the matrices @xmath47 , @xmath49 , and @xmath215 satisfy @xmath275 , and @xmath98 , which lead to @xmath276 , ~~ g=\left[\begin{array}{cc } 0&g_{\scriptscriptstyle 12}\\ -\frac{g}{\sqrt{\kappa}}&g_{\scriptscriptstyle 22}\\ g_{\scriptscriptstyle 31}&g_{\scriptscriptstyle 32}\\ 0&0\\ \end{array}\right ] , \nonumber \\ n&=\left[\begin{array}{cccc } n_{\scriptscriptstyle 11}&n_{\scriptscriptstyle 12 } & 0&n_{\scriptscriptstyle 14}\\ n_{\scriptscriptstyle 21}&n_{\scriptscriptstyle 22 } & 0&n_{\scriptscriptstyle 24}\\ \end{array}\right],\end{aligned}\ ] ] where @xmath218 , and @xmath219 are free parameters . * ( iii ) * the controller matrices have to satisfy the physical realizability conditions @xmath256 and @xmath257 ; these constraints impose the parameters to satisfy @xmath277 where @xmath231 and @xmath232 . + equations , , and provide the full parametrization of the direct interaction controller . * ( iv ) * to specify a set of parameters , as in the case of section [ sec:3 - 3 ] , let us aim to design a passive controller . from theorem 4 in appendix b , @xmath278 and @xmath279 satisfy the condition @xmath280 and @xmath281 , which lead to the same equalities given in eq . . then , setting the parameters to be @xmath238 and @xmath239 , we can determine the matrices @xmath278 and @xmath282 as follows : @xmath283 , ~~ r_2=r_1^\top = \left[\begin{array}{cccc } 0&0&g&0 \\ 0&0&0&g \\ \end{array}\right ] . \end{aligned}\ ] ] the controller specified by these matrices can be physically implemented as illustrated in fig . [ fig_mankei_scheme ] ( b ) ; that is , it is a single - mode detuned cavity with hamiltonian @xmath284 , which couples to the plant through a beam - splitter ( bs ) represented by @xmath285 . * remark : * we can employ an active controller , as proposed in @xcite . in this case the interaction hamiltonian is given by @xmath286 , while the system s self - hamiltonian is the same as above ; @xmath284 . that is , the controller couples to the plant through a non - degenerate optical parametric amplification process in addition to the bs interaction . to satisfy the bae condition , the parameters must satisfy @xmath287 . note that this direct interaction controller can be specified , in the full - parameterization , , and , by @xmath288 we have demonstrated in sections [ sec:3 - 3 ] and [ sec:4 ] that the bae condition can be achieved by engineering an appropriate auxiliary system and connecting it to the plant . however , in a practical situation , it can not be expected to realize such perfect bae due to several experimental imperfections . hence , in a realistic setup , we should modify our strategy for engineering a sensor so that it would accomplish _ approximate bae_. then , looking back into section [ sec:2 - 3 ] where the bae condition , @xmath289 , was obtained , we are naturally led to consider the following optimization problem to design an auxiliary system achieving the approximate bae : @xmath290 where @xmath291 denotes a valid norm of a complex function . in particular , in the field of robust control theory , the following @xmath292 norm and the @xmath293 norm are often used @xcite : @xmath294 that is , the @xmath292 or @xmath293 control theory provides a general procedure for synthesizing a feedback controller that minimizes the above norm . in this paper , we take the @xmath292 norm , mainly owing to the broadband noise - reduction nature of the @xmath292 controller . then , rather than pursuing an optimal quantum @xmath292 controller based on the quantum @xmath292 control theory @xcite , here we take the following geometric - control - theoretical approach to solve the problem . that is , first we apply the method developed in section [ sec:3 ] or [ sec:4 ] to the idealized system and obtain the controller achieving bae ; then , in the practical setup containing some unwanted noise , we make a local modification of the controller parameters obtained in the first step , to minimize the cost @xmath295 . as a demonstration , here we consider the coherent feedback control for the opto - mechanical system studied in section [ sec:2 - 3 ] , which is now subjected to the thermal noise @xmath296 . following the above - described policy , we employ the coherent feedback controller constructed for the idealized system that ignores @xmath296 , leading to the controller given by eqs . and , illustrated in fig . [ fig_3i3obae ] . the closed - loop system with variable @xmath297^{\top}$ ] , which now takes into account the realistic imperfections , then obeys the following dynamics : @xmath298 where @xmath299 . \ ] ] @xmath263 , @xmath264 , and @xmath265 are the same matrices given in eq . . @xmath296 is the thermal noise satisfying @xmath300 , where @xmath301 is the mean phonon number at thermal equilibrium @xcite . note that the damping effect appears in the @xmath302 component of @xmath303 due to the stochastic nature of @xmath296 . also , again , @xmath304 and @xmath305 are the decay rate and the detuning of the controller cavity , respectively . in the idealized setting where @xmath296 is negligible , the perfect bae is achieved by choosing the parameters satisfying eq . . the measurement output of this closed - loop system is , in the laplace domain , represented by @xmath306 the normalized noise power spectral density of @xmath307 is calculated as @xmath308 the coefficient of the back - action noise is given by @xmath309 our goal is to find the optimal parameters @xmath310 that minimize the @xmath292 norm of the transfer function , @xmath311 . norm @xmath312 versus the coupling constant @xmath304 and the detuning @xmath305.,width=264 ] the system parameters are taken as follows @xcite : @xmath313 mhz , @xmath314 mhz , @xmath315 khz , @xmath316 mhz , @xmath317 , and the effective mass is @xmath318 kg . we then have fig . [ fig_h2norm ] , showing @xmath319 as a function of @xmath304 and @xmath305 . this figure shows that there exists a unique pair of @xmath320 that minimizes the norm , and they are given by @xmath321 mhz and @xmath322 mhz , which are actually close to the ideal values . [ fig_noisepower ] shows the value of eq . with these optimal parameters @xmath320 , where the noise floor @xmath323 is subtracted . the solid black line represents the sql , which is now given by @xmath324 then the dot - dashed blue and dotted green lines indicate that , in the low frequency range , the coherent feedback controller can suppress the noise below the sql , while , by definition , the noise power of the autonomous ( i.e. , uncontrolled ) plant system is above the sql . moreover , this effect can be enhanced by injecting a @xmath158-squeezed probe field ( meaning @xmath325 and @xmath326 ) into the system . in fact the dashed red line in the figure illustrates the case @xmath327 ( about 9 db squeezing ) , showing the significant reduction of the noise power . the main contribution of this paper is in that it first provides the general theory for constructing a back - action evading sensor for linear quantum systems , based on the well - developed classical geometric control theory . the power of the theory has been demonstrated by showing that , for the typical opto - mechanical oscillator , a full parametrization of the auxiliary coherent - feedback and direct interaction controller achieving bae was derived , which contains the result of @xcite . note that , although we have studied a simple example for the purpose of demonstration , the real advantage of the theory developed in this paper will appear when dealing with more complicated multi - mode systems such as an opto - mechanical system containing a membrane @xcite . another contribution of this paper is to provide a general procedure for designing an approximate bae sensor under realistic imperfections ; that is , an optimal approximate bae system can be obtained by solving the minimization problem of the transfer function from the back - action noise to the measurement output . while in section [ sec:5 ] we have provided a simple approach based on the geometric control theory for solving this problem , the @xmath292 or @xmath293 control theory could be employed for systematic design of an approximate bae controller even for the above - mentioned complicated system . this is also an important future research direction of this work . the set of @xmath36-invariant subspaces has a unique maximum element contained in a given subspace @xmath330 . this space , denoted by @xmath331 , can be computed by the following algorithm : @xmath332 similarly , the set of @xmath40-invariant subspaces has a unique minimum element containing a given subspace @xmath333 , and this space , denoted by @xmath334 , can be computed by the following algorithm : @xmath335 note that a passive system is constituted only with annihilation operator variables ; a typical optical realization of the passive system is an empty optical cavity . moreover , @xmath345 is already satisfied and @xmath344 leads to @xmath346 . this is the reason why it is sufficient to consider the constraints only on @xmath347 and @xmath348 . then the goal here is to represent the conditions @xmath349 in terms of the coefficient matrices of eq . . for this purpose , let us introduce the permutation matrix @xmath350 as follows ; for a column vector @xmath351^{\top}$ ] , @xmath350 is defined through @xmath352^{\top}$ ] . note that @xmath350 satisfies @xmath353 . then , the coefficient matrices of the above two system representations are connected by @xmath354 where @xmath355.\end{aligned}\ ] ] @xmath356 , @xmath357 , and @xmath358 have the same forms as above . then , we have the following theorem , providing the passivity condition in the quadrature form : * theorem 3 : * the system is passive if and only if , in addition to the physical realizability condition , the following equalities hold : @xmath359 _ proof : _ let us first define @xmath360 $ ] , which leads to @xmath361 . then , we can prove @xmath362 the condition in the right hand side is equivalent to @xmath363 and @xmath364 , which thus leads to @xmath343 . also , from a similar calculation we obtain @xmath365 let us next consider the passivity condition of the direct interaction controller discussed in section [ sec:4 ] . the setup is that , for a given linear quantum system , we add an auxiliary component with variable @xmath366 , which is characterized by the hamiltonians . the point is that these hamiltonians have the following equivalent representations in terms of the vector of annihilation operators @xmath367 and @xmath368 : @xmath369 \mathscr{r}_{\scriptscriptstyle k } \left[\begin{array}{c } \hat{a}_{\scriptscriptstyle k } \\ \hat{a}_{\scriptscriptstyle k}^{\sharp } \\ \end{array}\right ] , \nonumber \\ \hat{h}_{\rm int}&=\frac{1}{2}\left ( \left[\begin{array}{cc } \hat{a}^{\dag } & \hat{a}^{\top } \\ \end{array}\right ] \mathscr{r}_1 \left[\begin{array}{c } \hat{a}_{\scriptscriptstyle k } \\ \hat{a}_{\scriptscriptstyle k}^{\sharp } \\ \end{array}\right ] + \left[\begin{array}{cc } \hat{a}_{\scriptscriptstyle k}^{\dag } & \hat{a}_{\scriptscriptstyle k}^{\top } \\ \end{array}\right ] \mathscr{r}_2 \left[\begin{array}{c } \hat{a } \\ \hat{a}^{\sharp } \\ \end{array}\right ] \right).\end{aligned}\ ] ] the matrices @xmath370 , and @xmath371 are of the same forms as those in eq . . note that they have to satisfy the physical realizability conditions @xmath372 and @xmath373 . now we can define the passivity property of the direct interaction controller ; that is , if the hamiltonians does not contain any quadratic term such as @xmath374 and @xmath375 , then the direct interaction controller is passive . the formal definition is given as follows : * definition 6 : * the direct interaction controller constructed by hamiltonians is said to be passive if , in addition to the physical realizability conditions @xmath372 and @xmath373 , the matrices satisfy @xmath376 and @xmath377 . * theorem 4 : * the direct interaction controller constructed by hamiltonians is passive if and only if , in addition to the physical realizability conditions @xmath256 and @xmath257 , the following equalities hold : @xmath378 c. m. caves , k. s. thorne , r. w. p. drever , v. d. sandberg , and m. zimmermann , on the measurement of a weak classical force coupled to a quantum - mechanical oscillator . i. issues of principle , rev . * 52 * , 341 ( 1980 ) c. l. latune , b. m. escher , r. l. de matos filho , and l. davidovich , quantum limit for the measurement of a classical force coupled to a noisy quantum - mechanical oscillator , phys . a * 88 * , 042112 ( 2013 ) . t. corbitt , y. chen , e. innerhofer , h. m@xmath379ller - ebhardt , d. ottaway , h. rehbein , d. sigg , s. whitcomb , c. wipf , and n. mavalvala , an all - optical trap for a gram - scale mirror , phys . lett . * 98 * , 150802 ( 2007 ) . m. underwood , d. mason , d. lee , h. xu , l. jiang , a. b. shkarin , k. brkje , s. m. girvin , and j. g. e. harris , measurement of the motional sidebands of a nanogram - scale oscillator in the quantum regime , phys . a * 92 * , 061801 ( 2015 ) . h. j. kimble , y. levin , a. b. matsko , k. s. thorne , and s. p. vyatchanin , conversion of conventional gravitational - wave interferometers into quantum nondemolition interferometers by modifying their input and/or output optics , phys . d * 65 * , 022002 ( 2001 ) . y. liu , s. shankar , n. ofek , m. hatridge , a. narla , k. m. sliwa , l. frunzio , r. j. schoelkopf , and m. h. devoret , comparing and combining measurement - based and driven - dissipative entanglement stabilization , phys . x * 6 * , 011022 ( 2016 ) .
over the last more than three decades ferroelectric liquid crystals ( flcs ) have attracted considerable attention as promising chiral liquid crystal materials for applications in fast switching display devices ( a detailed description of flcs can be found , e.g. , in monographs @xcite ) . equilibrium orientational structures in flcs are represented by helical twisting patterns where flc molecules align on average along a local unit director @xmath2 where @xmath3 is the smectic tilt angle ; @xmath4 is the twisting axis normal to the smectic layers and @xmath5 is the @xmath6-director . the flc director lies on the smectic cone depicted in fig . [ subfig : director ] with the _ smectic tilt angle _ @xmath3 and rotates in a helical fashion about a uniform twisting axis @xmath4 forming the flc helix with the _ helix pitch _ , @xmath7 . this rotation is described by the azimuthal angle around the cone @xmath8 that specifies orientation of the @xmath6-director in the plane perpendicular to @xmath4 and depends on the dimensionless coordinate along the twisting axis @xmath9 where @xmath10 is the helix twist wave number . the important case of a uniform lying flc helix in the slab geometry with the smectic layers normal to the substrates and @xmath11 where @xmath12 is the electric field applied across the cell , is illustrated in fig . [ fig : geom ] . this is the geometry of surface stabilized flcs ( ssflcs ) pioneered by clark and lagerwall in @xcite . they studied electro - optic response of flc cells confined between two parallel plates subject to homogeneous boundary conditions and made thin enough to suppress the bulk flc helix . it was found that such cells exhibit high - speed , bistable electro - optical switching between orientational states stabilized by surface interactions . the response of flcs to an applied electric field @xmath12 is characterized by fast switching times due to linear coupling between the field and the _ spontaneous ferroelectric polarization _ @xmath13 where @xmath14 is the _ polarization unit vector_. there is also a threshold voltage necessary for switching to occur and the process of bistable switching is typically accompanied by a hysteresis . figure [ subfig : cell ] also describes the geometry of deformed helix flcs ( dhflcs ) as it was introduced in @xcite . this case will be of our primary concern . in dhflc cells , the flc helix is characterized by a short submicron _ helix pitch _ , @xmath15 @xmath16 m , and a relatively large _ tilt angle _ , @xmath17 deg . by contrast to ssflc cells , where the surface induced unwinding of the bulk helix requires the helix pitch of a flc mixture to be greater than the cell thickness , a dhflc helix pitch is 5 - 10 times smaller than the thickness . this allows the helix to be retained within the cell boundaries . electro - optical response of dhflc cells exhibits a number of peculiarities that make them useful for lc devices such as high speed spatial light modulators @xcite , colour - sequential liquid crystal display cells @xcite and optic fiber sensors @xcite . the effects caused by electric - field - induced distortions of the helical structure underline the mode of operation of such cells . in a typical experimental setup , these effects are probed by performing measurements of the transmittance of normally incident linearly polarized light through a cell placed between crossed polarizers . a more general case of oblique incidence has not received a fair amount of attention . theoretically , a powerful tool to deal with this case is the transfer matrix method which has been widely used in studies of both quantum mechanical and optical wave fields @xcite . in this work we apply the method for systematic treatment of the technologically important case of dhflcs with subwavelength pitch also known as the _ short - pitch dhflcs_. recently , the transfer matrix approach to polarization gratings was employed to define the effective dielectric tensor of short - pitch dhflcs @xcite that gives the principal values and orientation of the optical axes as a function of the applied electric field . biaxial anisotropy and rotation of the in - plane optical axes produced by the electric field can be interpreted as the _ orientational kerr effect _ @xcite . it can be expected that the electric field dependence of the effective dielectric tensor will also manifest itself as electric - field - induced transformations of the polarization - resolved angular ( conoscopic ) patterns in the observation plane after the dhflc cells illuminated by convergent light beam . these patterns are represented by the two - dimensional ( 2d ) fields of polarization ellipses describing the polarization structure behind the conoscopic images @xcite . as it was originally recognized by nye @xcite , the key elements characterizing geometry of such stokes parameter fields are the _ polarization singularities _ that play the fundamentally important role of structurally stable topological defects ( a recent review can be found in ref . in particular , the polarization singularities such as the _ c points _ ( the points where the light wave is circular polarized ) and the _ l lines _ ( the curves along which the polarization is linear ) frequently emerge as the characteristic feature of certain polarization state distributions . for nematic and cholescteric ( chiral nematic ) liquid crystals , the singularity structure of the polarization - resolved angular patterns is generally found to be sensitive to both the director configuration and the polarization characteristics of incident light @xcite . in this study , we consider the polarization - resolved angular patterns of dhflc cells as the stokes parameter fields giving detailed information on the incidence angles dependence of the polarization state of light transmitted through the cells . in particular , we explore how the polarization singularities transform under the action of the electric field . our analysis will utilize the transfer matrix approach in combination with the results for the effective dielectric tensor of biaxial flcs evaluated using an improved technique of averaging over distorted helical structures . we also emphasize the role of phase singularities of a different kind and discuss the electro - optic behavior of dhflcs near the exceptional point where the condition of zero - field isotropy is fulfilled . the layout of the paper is as follows . in sec . [ sec : theory ] we introduce our notations and describe the transfer matrix formalism rendered into the @xmath18 matrix form suitable for our purposes . this formalism is employed to deduce a number of the unitarity and symmetry relations with emphasis on the planar anisotropic structures that represent dhflc cells and posses two optical axes lying in the plane of substrates . in sec . [ sec : optics - dhf ] we evaluate the effective dielectric tensor of dhflc cells , discuss the orientational kerr effect and show that electro - optic response of dhflc cells is enhanced near the exceptional point determined by the condition of zero - field isotropy . geometry of the polarization - resolved angular patterns emerging after dhflc cells is considered in sec . [ sec : conoscopy ] . after providing necessary details on our computational approach and the polarization singularities , we present the numerical results describing how the singularity structure of polarization ellipse fields transforms under the action of the electric field . finally , in sec . [ sec : discussion ] we draw the results together and make some concluding remarks . details on some technical results are relegated to appendixes [ sec : op - evol][sec : flc - spiral ] . in order to describe both the electro - optical properties and the polarization - resolved angular patterns of deformed helix ferroelectric liquid crystal layers with subwavelength pitch we adapt a general theoretical approach which can be regarded as a modified version of the well - known transfer matrix method @xcite and was previously applied to study the polarization - resolved conoscopic patterns of nematic liquid crystal cells @xcite . this approach has also been extended to the case of polarization gratings and used to deduce the general expression for the effective dielectric tensor of dhflc cells @xcite . in this section , we present the transfer matrix approach as the starting point of our theoretical considerations , with emphasis on its general structure and the symmetry relations . the analytical results for uniformly anisotropic planar structures representing homogenized dhflc cells are given in appendix [ sec : planar ] . we deal with a harmonic electromagnetic field characterized by the free - space wave number @xmath19 , where @xmath20 is the frequency ( time - dependent factor is @xmath21 ) , and consider the slab geometry shown in fig . [ fig : geom_4rays ] . in this geometry , an optically anisotropic layer of thickness @xmath22 is sandwiched between the bounding surfaces ( substrates ) : @xmath23 and @xmath24 ( the @xmath25 axis is normal to the substrates ) and is characterized by the dielectric tensor @xmath26 and the magnetic permittivity @xmath16 further , we restrict ourselves to the case of stratified media and assume that the electromagnetic fields can be taken in the following factorized form @xmath27 where the vector @xmath28 represents the lateral component of the wave vector . then we write down the representation for the electric and magnetic fields , @xmath12 and @xmath29 , @xmath30 where the components directed along the normal to the bounding surface ( the @xmath25 axis ) are separated from the tangential ( lateral ) ones . in this representation , the vectors @xmath31 and @xmath32 are parallel to the substrates and give the lateral components of the electromagnetic field . substituting the relations into the maxwell equations and eliminating the @xmath25 components of the electric and magnetic fields gives equations for the tangential components of the electromagnetic field that can be written in the following @xmath18 matrix form @xcite : @xmath33 where @xmath34 is the _ differential propagation matrix _ and its @xmath35 block matrices @xmath36 are given by [ eq : mij ] @xmath37 general solution of the system @xmath38 can be conveniently expressed in terms of the _ evolution operator _ which is also known as the _ propagator _ and is defined as the matrix solution of the initial value problem [ eq : evol_problem ] @xmath39 where @xmath40 is the @xmath41 identity matrix . basic properties of the evolution operator are reviewed in appendix [ sec : op - evol ] . in the ambient medium with @xmath42 and @xmath43 , the general solution can be expressed in terms of plane waves propagating along the wave vectors with the tangential component . for such waves , the result is given by @xmath44 where @xmath45 is the eigenvector matrix for the ambient medium given by @xmath46 @xmath47 are the pauli matrices @xmath48 from eq . , the vector amplitudes @xmath49 and @xmath50 correspond to the forward and backward eigenwaves with @xmath51 and @xmath52 , respectively . figure [ fig : geom_4rays ] shows that , in the half space @xmath53 before the entrance face of the layer @xmath23 , these eigenwaves describe the _ incoming and outgoing waves _ @xmath54 whereas , in the half space @xmath55 after the exit face of the layer , these waves are given by @xmath56 in this geometry , there are two plane waves , @xmath57 and @xmath58 , incident on the bounding surfaces of the anisotropic layer , @xmath23 and @xmath24 , respectively . then the standard linear input - output relations @xmath59 linking the vector amplitudes of transmitted and reflected waves , @xmath60 and @xmath61 with the amplitude of the incident wave , @xmath62 through the transmission and reflection matrices , @xmath63 and @xmath64 , assume the following generalized form : @xmath65 where @xmath66 is the matrix the so - called _ scattering matrix _ that relates the outgoing and incoming waves ; @xmath67 ( @xmath68 ) is the transmission ( reflection ) matrix for the case when the incident wave is incoming from the half space @xmath53 bounded by the entrance face , whereas the mirror symmetric case where the incident wave is impinging onto the exit face of the sample is described by the transmission ( reflection ) matrix @xmath69 ( @xmath70 ) . so , we have @xmath71 it is our task now to relate these matrices and the evolution operator given by eq . . to this end , we use the boundary conditions requiring the tangential components of the electric and magnetic fields to be continuous at the boundary surfaces : @xmath72 and @xmath73 , and apply the relation to the anisotropic layer of the thickness @xmath22 to yield the following result @xmath74 on substituting eqs . into eq . we have @xmath75 where the matrix @xmath76 linking the electric field vector amplitudes of the waves in the half spaces @xmath77 and @xmath78 bounded by the faces of the layer will be referred to as the _ transfer ( linking ) matrix_. the expression for the transfer matrix is as follows @xmath79 where @xmath80 is the rotated operator of evolution . this operator is the solution of the initial value problem with @xmath81 replaced with @xmath82 . from eqs . and , the block structure of the transfer matrix can be expressed in terms of the transmission and reflection matrices as follows @xmath83 similarly , for inverse of the transfer matrix , @xmath84 we have @xmath85 in appendix [ sec : op - evol ] , it is shown that , for non - absorbing media with symmetric dielectric tensor , @xmath86 , the operator of evolution satisfies the unitarity relation . by using eq . in combination with the algebraic identity [ eq : vm_identity ] @xmath87 where @xmath88 , for the eigenvector matrix given in eq . , we can deduce the unitarity relation for the transfer matrix @xmath89 the unitarity relation for non - absorbing layers can now be used to derive the energy conservation laws [ eq : energy_laws ] @xmath90}^{\dagger } } { { { \mathbf{t}}}^{t}}_{\pm}+ { { [ { { { \mathbf{r}}}^{t}}_{\mp}]}^{\dagger } } { { { \mathbf{r}}}^{t}}_{\mp } = { \mathbf{i}}_2 , \ ] ] where a dagger and the superscript @xmath91 will denote hermitian conjugation and matrix transposition , respectively , along with the relations for the block matrices [ eq : w_ij_unit ] @xmath92}^{\dagger } } , \\ & \label{eq : w_12_unit } { \mathbf{w}}_{12}=-{\mathbf{t}}_{+}^{-1 } { \mathbf{r}}_{-}= { { [ { \mathbf{t}}_{-}^{-1 } { \mathbf{r}}_{+}]}^{\dagger } } , \\ & \label{eq : w_21_unit } { \mathbf{w}}_{21}={\mathbf{r}}_{+ } { \mathbf{t}}_{+}^{-1}= -{{[{\mathbf{r}}_{- } { \mathbf{t}}_{-}^{-1}]}^{\dagger}}.\end{aligned}\ ] ] note that eqs . and can be conveniently rewritten in the following form [ eq : rel - w_12_21 ] @xmath93}^{\dagger } } { { { \mathbf{r}}}^{\dagger}}_{+},\end{aligned}\ ] ] so that multiplying these identities and using the energy conservation law gives the relations . in the translation invariant case of uniform anisotropy , the matrix @xmath34 is independent of @xmath94 and the operator of evolution is given by @xmath95 then , the unitarity condition @xcite @xmath96 can be combined with eq . to yield the additional symmetry relations for @xmath97 @xmath98 where an asterisk will indicate complex conjugation , that give the following algebraic identities for the transmission and reflection matrices : @xmath99 it can be readily seen that the relation for the transposed matrices can be derived by substituting eq . into the conservation law . for the important special case of uniformly anisotropic planar structures with @xmath100 , the algebraic structure of the transfer matrix is described in appendix [ sec : planar ] . equation shows that the symmetry relations remain valid even if the dielectric constants are complex - valued and the medium is absorbing . since identities are derived from eqs . and without recourse to the unitarity relations , they also hold for lossy materials . similar remark applies to the expression for inverse of the transfer matrix . from eq . , it follows that the relation between the transmission ( reflection ) matrix , @xmath67 ( @xmath68 ) , and its mirror symmetric counterpart @xmath69 ( @xmath70 ) can be further simplified and is given by @xmath101 from eqs . and , we have the relation for the transposed reflection matrices @xmath102 whereas the transmission matrices are symmetric . we now pass on to the electro - optical properties of dhflc cells and extend the results of ref . @xcite to the case of biaxial ferroelectric liquid crystals with subwavelength pitch . in addition , the theoretical treatment will be significantly improved by using an alternative fully consistent procedure to perform averaging over distorted flc helix that goes around the limitations of the first - order approximation . we consider a flc film of thickness @xmath22 with the @xmath25 axis which , as is indicated in fig . [ fig : geom ] , is normal to the bounding surfaces : @xmath23 and @xmath24 , and introduce the effective dielectric tensor , @xmath103 , describing a homogenized dhflc helical structure . for a biaxial flc , the components of the dielectric tensor , @xmath104 , are given by @xmath105 where @xmath106 , @xmath107 is the kronecker symbol ; @xmath108 ( @xmath109 ) is the @xmath110th component of the flc director ( unit polarization vector ) given by eq . ( eq . ) ; @xmath111 are the _ anisotropy parameters _ and @xmath112 ( @xmath113 ) is the _ anisotropy ( biaxiality ) ratio_. note that , in the case of uniaxial anisotropy with @xmath114 , the principal values of the dielectric tensor are : @xmath115 and @xmath116 , where @xmath117 ( @xmath118 ) is the ordinary ( extraordinary ) refractive index and the magnetic tensor of flc is assumed to be isotropic with the magnetic permittivity @xmath16 . as in sec . [ sec : theory ] ( see fig . [ fig : geom_4rays ] ) , the medium surrounding the layer is optically isotropic and is characterized by the dielectric constant @xmath119 , the magnetic permittivity @xmath120 and the refractive index @xmath121 . at @xmath122 , the ideal flc helix @xmath123 where @xmath124 is the free twist wave number and @xmath125 is the equilibrium helical pitch , is defined through the azimuthal angle around the smectic cone @xmath8 ( see fig . [ fig : geom ] and eq . ) and represents the undistorted structure . for sufficiently small electric fields , the standard perturbative technique applied to the euler - lagrange equation gives the first - order expression @xcite for the azimuthal angle of a weakly distorted helical structure @xmath126 where @xmath127 is the electric field parameter linearly proportional to the ratio of the applied and critical electric fields : @xmath128 , and @xmath129 . according to ref . @xcite , normally incident light feels effective in - plane anisotropy described by the averaged tensor , @xmath130 : @xmath131 where @xmath132 , and the effective dielectric tensor @xmath133 can be expressed in terms of the averages @xmath134^{-1}}\rangle } } , \label{eq : eta } \\ & \beta_{z\alpha}= { \ensuremath{\langle{\epsilon_{z\alpha}/\epsilon_{zz}}\rangle } } = \left\langle \frac{u_1 d_z d_\alpha+u_2 p_z p_\alpha}{1+u_1 d_z^2+u_2 p_z^2 } \right\rangle , \label{eq : beta } \end{aligned}\ ] ] as follows @xmath135 general formulas - give the zero - order approximation for homogeneous models describing the optical properties of short pitch dhflcs @xcite . assuming that the pitch - to - wavelength ratio @xmath136 is sufficiently small , these formulas can now be used to derive the effective dielectric tensor of homogenized short - pitch dhflc cell for both vertically and planar aligned flc helix . the results for vertically aligned dhflc cells were recently published in ref . @xcite and we concentrate on the geometry of planar aligned dhflc helix shown in fig . [ fig : geom ] . for this geometry , the parameters needed to compute the averages @xmath130 ( see eq . ) , @xmath137 ( see eq . ) and @xmath138 ( see eq . ) are given by @xmath139 formulas can now be inserted into eqs . to yield the explicit expressions for the elements of the dielectric tensor : [ eq : epsilon - ij - planar ] @xmath140 the simplest averaging procedure previously used in refs . @xcite involves substituting the formula for a weakly distorted flc helix into eqs . and performing integrals over @xmath141 . this procedure thus heavily relies on the first - order approximation where the director distortions are described by the term linearly proportional to the electric field ( the second term on the right hand side of eq . ) . quantitatively , the difficulty with this approach is that the linear approximation may not be suffice for accurate computing of the second - order contributions to the diagonal elements of the dielectric tensor . in this approximation , the second - order corrections describing the helix distortions that involve the change of the helix pitch have been neglected . in order to circumvent the problem , in this paper , we apply an alternative approach that allows to go beyond the first - order approximation without recourse to explicit formulas for the azimuthal angle . this method is detailed in appendix [ sec : flc - spiral ] . the analytical results substituted into eqs . give the effective dielectric tensor in the following form : @xmath142 the _ zero - field dielectric constants _ , @xmath143 and @xmath144 , that enter the tensor are defined in eqs . and , respectively , and can be conveniently rewritten as follows [ eq : epsilon_ph ] @xmath145 a similar result for the _ coupling coefficients _ @xmath146 , @xmath147 and @xmath148 ( see eq . ) reads [ eq : coupling - coeffs ] @xmath149 note that , following ref . @xcite , we have used the relation to introduce the electric field parameter @xmath150 where @xmath151 is the dielectric susceptibility of the goldstone mode @xcite . the above dielectric tensor is characterized by the three generally different principal values ( eigenvalues ) and the corresponding optical axes ( eigenvectors ) as follows @xmath152 ^ 2+[\gamma_{xy } \alpha_{e}]^2}\end{aligned}\ ] ] where @xmath153 the in - plane optical axes are given by @xmath154.\end{aligned}\ ] ] from eq . , it is clear that , similar to the case of uniaxial flcs studied in ref . @xcite , the zero - field dielectric tensor is uniaxially anisotropic with the optical axis directed along the twisting axis @xmath155 . the applied electric field changes the principal values ( see eqs . and ) so that the electric - field - induced anisotropy is generally biaxial . in addition , the in - plane principal optical axes are rotated about the vector of electric field , @xmath156 , by the angle @xmath157 given in eq . . in the low electric field region , the electrically induced part of the principal values is typically dominated by the kerr - like nonlinear terms proportional to @xmath158 , whereas the electric field dependence of the angle @xmath157 is approximately linear : @xmath159 . this effect is caused by the electrically induced distortions of the helical structure and bears some resemblance to the electro - optic kerr effect . following refs . @xcite , it will be referred to as the _ orientational kerr effect_. it should be emphasized that this effect differs from the well - known kerr effect which is a quadratic electro - optic effect related to the electrically induced birefringence in optically isotropic ( and transparent ) materials and which is mainly caused by the electric - field - induced orientation of polar molecules @xcite . by contrast , in our case , similar to polymer stabilized blue phase liquid crystals @xcite , we deal with the effective dielectric tensor of a nanostructured chiral smectic liquid crystal . this tensor is defined through averaging over the flc orientational structure . typically , in experiments dealing with the electro - optic response of dhflc cells , the transmittance of normally incident light passing through crossed polarizers is measured as a function of the applied electric field . for normal incidence , the transmission and reflection matrices can be easily obtained from the results given in appendix [ sec : planar ] by substituting eq . into eqs . - . when the incident wave is linearly polarized along the @xmath160 axis ( the helix axis ) , the transmittance coefficient @xmath161 where @xmath162 is the thickness parameter , describes the intensity of the light passing through crossed polarizers . note that , under certain conditions such as @xmath163 , @xmath164 and the transmittance can be approximated by simpler formula @xmath165 where @xmath166 is the difference in optical path of the ordinary and extraordinary waves known as the _ phase retardation_. in ref . @xcite , the relation was used to fit the experimental data using the theory based on the linear approximation for the helix distortions ( see eq . ) . these results are reproduced in figure [ fig : expt_th ] along with the theoretical curve computed using the modified averaging technique . from fig . [ fig : expt_th ] , it is seen that , in the range of relatively high voltages , the averaging method described in appendix [ sec : flc - spiral ] improves agreement between the theory and the experiment , whereas , at small voltages , the difference between the fitting curves is negligibly small . given the anisotropy and biaxiality ratios , @xmath167 and @xmath168 , the zero - field dielectric constants and the coupling coefficients are determined by the smectic tilt angle , @xmath3 . figure [ fig : gamma_theta ] shows how the coupling coefficients depend on @xmath3 for both uniaxially and biaxially anisotropic flcs . as it can be seen in fig . [ subfig : gamma_uni ] , in the case of conventional flcs with @xmath169 , all the coefficients are positive and the difference of the coupling constants @xmath170 that define the electrically induced part of @xmath171 ( see eq . ) is negative at @xmath172 . from eqs . and , it follows that , at @xmath169 , the principal values of dielectric constants @xmath173 and @xmath174 are decreasing functions of the electric field parameter @xmath175 so that anisotropy of the effective dielectric tensor is weakly biaxial . in addition , for non - negative @xmath175 and @xmath176 , the azimuthal angle of in - plane optical axis , @xmath177 , given in eq . increases with @xmath175 from zero to @xmath178 . figure [ subfig : gamma_bia ] demonstrates that this is no longer the case for biaxial flcs . it is seen that , at @xmath179 , the coupling coefficient @xmath147 and the difference @xmath170 both change in sign when the tilt angle @xmath3 is sufficiently small . at such angles , the dielectric constant @xmath173 increases with @xmath175 and the electric field induced anisotropy of dhflc cell becomes strongly biaxial . when @xmath170 and @xmath175 are positive , electric field dependence of the azimuthal angle @xmath177 is non - monotonic and the angle decays to zero in the range of high voltages where @xmath180 . at @xmath122 , the zero - field anisotropy is uniaxial and is described by the dielectric constants , @xmath143 and @xmath181 , given in eq . . [ fig : eph_theta ] , these constants are plotted against the tilt angle . it is shown that , at small tilt angles , the anisotropy @xmath182 is positive . it decreases with @xmath3 and the zero - field state becomes isotropic when , at certain critical angle @xmath183 , the condition of zero - field isotropy @xmath184 is fulfilled and @xmath185 . so , the angle @xmath186 can be referred to as the _ isotropization angle_. in what follows we discuss peculiarities of the electro - optic response in the vicinity of the _ isotropization point _ where @xmath171 is proportional to @xmath158 ( see eq . ) and the kerr - like regime breaks down . mathematically , the isotropization point represents a square root branch - point singularity of the eigenvalues of the dielectric tensor which is known as the _ exceptional point _ @xcite . in the electric field dependence of the in - plane dielectric constants , @xmath187 and @xmath174 , this singularity reveals itself as a cusp where the derivatives of @xmath188 with respect to @xmath175 are discontinuous . more precisely , we have @xmath189 as is illustrated in fig . [ subfig : epsilon_c_uni ] , the cusp is related to the effect of reconnection of different branches representing solutions of an algebraic equation . since the azimuthal angle @xmath177 is undetermined at @xmath183 @xmath190 the isotropization point also represents a phase singularity . the electric field dependence of @xmath177 is thus discontinuous and the relation @xmath191 describes its jumplike behaviour at @xmath122 . this behaviour is demonstrated in fig . [ fig : psi_alp ] . we can now use eq . and write down the condition of zero - field isotropy in the following explicit form : @xmath192 the case of a uniaxially anisotropic flc with @xmath169 can be treated analytically . in this case , it is not difficult to check that @xmath193 gives the special solution of eq . that does not depend on the tilt angle and corresponds to an isotropic material with @xmath194 . another solution is given by the relation @xmath195 linking the isotropization angle @xmath186 and the anisotropy parameter , @xmath196 . in fig . [ fig : theta_r1 ] , this solution is represented by the solid line curve . the isotropization angle is shown to be a slowly decreasing function of the anisotropy ratio @xmath167 . from eq . , it starts from the maximal value of @xmath186 is @xmath197 and decays approaching @xmath198 . when the biaxiality ratio @xmath168 differs from unity , the solution of the isotropy condition can only be written in the parametrized form as follows @xmath199 where @xmath200 the @xmath186 versus @xmath167 curves computed from the representation are shown in fig . [ fig : theta_r1 ] . it can be seen that , by contrast to the case of uniaxial anisotropy [ @xmath169 ] , for biaxial flcs with @xmath202 , each curve has two branches separated by a gap . the isotropization angle vanishes , @xmath203 , at one of the endpoints of the gap , @xmath204 , whereas the angle @xmath186 equals @xmath178 at the other endpoint . thus , dependence of the isotropization angle on the anisotropy ratio @xmath167 being smooth and continuous for uniaxially anisotropic flcs is found to be splitted into two branches when the flc anisotropy is biaxial . from fig . [ fig : theta_r1 ] , one of the branches with @xmath205 is associated with the endpoint at @xmath204 and lies below of the solid line curve representing conventional flcs . for this branch , the angle @xmath186 decreases with the biaxiality ratio reaching zero at @xmath206 . we conclude this section with the remark on the electro - optic response of dhflc cells in the vicinity of the isotropization point . one of the important factors governing the electric field dependence of the transmittance is the phase retardation @xmath207 ( see eq . ) proportional the effective birefringence @xmath208 . the electrically dependent part of this birefringence is plotted as a function of the electric field parameter in fig . [ fig : dn_alp ] . it shown that , at @xmath183 , the kerr - like regime breaks down and the birefringence is dominated by the terms linearly dependent on the electric field . such a pockels - like behaviour manifests itself in the perfectly harmonic dependence of the transmittance , @xmath209 , on the electric field parameter depicted in fig . [ subfig : transm_c_uni ] . figure [ subfig : transm_c-2_uni ] illustrates the effect of small deviations from the isotropization angle . though the curve presented in fig . [ subfig : transm_c-2_uni ] and the ones for flc-576a ( see fig . [ fig : expt_th ] ) are quite similar in shape , it is clear that sensitivity to the electric field and the magnitude of transmission peaks are both considerably enhanced near the isotropization point . such behaviour comes as no surprise and derives from the above discussed fact that this point plays the role of a singularity ( an exceptional point ) ( see eqs . ) . we can now combine the general relations deduced in sec . [ sec : theory ] ( and in appendix [ sec : planar ] ) using the transfer matrix method with the results of sec . [ sec : optics - dhf ] to study the _ polarization - resolved angular ( conoscopic ) patterns _ describing the polarization structure behind the conoscopic images of short - pitch dhflc cells that are characterized by the effective dielectric tensor . this polarization structure is represented by a two - dimensional ( 2d ) distribution of polarization ellipses and results from the interference of eigenmodes excited in the dhflc cells by the plane waves with varying direction of incidence . geometrically , the important elements of the 2d stokes parameter fields are the polarization singularities such as @xmath1 points ( the points where the light wave is circular polarized ) and @xmath0 lines ( the curves along which the polarization is linear ) . in this section the focus of our attention will be on the singularity structure of the polarization - resolved angular patterns emerging after the dhflc cells . our starting point is the computational method used to evaluate the patterns as the polarization ellipse fields . we shall use the electric field vector amplitudes of incident , reflected and transmitted waves conveniently rewritten in the circular basis @xmath210 and the incidence angles , @xmath211 and @xmath212 , related to the lateral component of the wave vector as follows @xmath213 where @xmath211 ( @xmath212 ) is the _ polar ( azimuthal ) angle of incidence_. dependence of the polarization properties of the waves transmitted through the dhflc cell on the incidence angles , @xmath211 and @xmath212 , will be of our primary concern . the transmission matrix describing conoscopic patterns on the transverse plane of projection is given by @xcite @xmath214 where @xmath215 and @xmath216 are the polar coordinates in the observation plane ( @xmath217 and @xmath218 are the cartesian coordinates ) and @xmath219 is the aperture dependent scale factor . the transmission matrix of dhflc cells , @xmath220 , can be computed from general formulas given in appendix [ sec : planar ] ( see eq . ) . for this matrix , the parameters @xmath221 that enter the expression for the dielectric tensor of planar structures should be replaced with the characteristics @xmath222 of the effective dielectric tensor given in eqs . . in what follows the incident light is assumed to be linearly polarized @xmath223 where @xmath224 is the polarization azimuth of the incident wave , and the state of polarization of the transmitted wave @xmath225 is defined by the polarization ellipse characteristics . the orientation of the polarization ellipse is specified by the azimuthal angle of polarization ( _ polarization azimuth _ ) @xmath226}^{\ast } } e_{-}^{({\mathrm{trm}})},\end{aligned}\ ] ] where @xmath227 is the @xmath110th component of the stokes vector , and its eccentricity is described by the signed ellipticity parameter @xmath228 that will be referred to as the _ ellipticity_. the ellipse is considered to be right handed ( left handed ) if its helicity is positive ( negative ) , so that @xmath229 ( @xmath230 ) . + from eq . , the incidence angles and the points in the observation plane are in one - to - one correspondence . so , computing the polarization azimuth , @xmath231 , the ellipticity , @xmath232 , at each point of the projection plane yields the 2d field of polarization ellipses which is called the polarization - resolved angular ( conoscopic ) pattern . the point where @xmath233 and thus the transmitted wave is circularly polarized with @xmath234 will be referred to as the _ c@xmath235 point_. this is an example of the _ polarization singularity _ that can be viewed as the _ phase singularities _ of the complex scalar field @xmath236 where the phase @xmath237 ( see eq . ) become indeterminate . such singularities are characterized by the _ winding number _ which is the signed number of rotations of the two - component field @xmath238 around the circuit surrounding the singularity @xcite . the winding number also known as the _ signed strength of the dislocation _ is generically @xmath239 . since the polarization azimuth is defined modulo @xmath240 and @xmath241 , the dislocation strength is twice the index of the corresponding @xmath242 point , @xmath243 . for generic @xmath1 points , @xmath244 and the topological index can be computed as the closed - loop contour integral of the phase @xmath245 modulo @xmath246 @xmath247 where @xmath0 is the closed path around the singularity . in addition to the handedness and the index , the @xmath1 points are classified according to the number of streamlines , which are polarization lines whose tangent gives the polarization azimuth , terminating on the singularity . this is the so - called _ line classification _ that was initially studied in the context of umbilic points @xcite . mathematically , the straight streamlines that terminate on the singularity are of particular importance as they play the role of separatrices , separating regions of streamlines with differently signed curvature . as is illustrated in fig . [ fig : singul - types ] , for generic @xmath1 points , the number of the straight lines , @xmath248 , may either be 1 or 3 . this number is @xmath249 provided the index equals @xmath250 , @xmath251 , and such @xmath1 points are called _ stars_. at @xmath252 , there are two characteristic patterns of polarization ellipses around a @xmath1 point : ( a ) _ lemon _ with @xmath253 and ( b ) _ monstar _ with @xmath254 @xcite . different quantitative criteria to distinguish between the @xmath1 points of the lemon and the monstar types were deduced in refs . @xcite . from these criteria it can be inferred that a lemon becomes a monstar as it approaches a star and @xmath1 point annihilation occurs only between stars and monstars @xcite . the case of linearly polarized wave with @xmath255 provides another example of the polarization singularity where the handedness is undefined . the curves along which the polarization is linear are called the _ l lines_. from eq . the @xmath1 points are nodal points of the scalar complex function @xmath256 which can be found as intersection points of the stokes parameter nodal lines @xmath257 and @xmath258 . similarly , equation implies that nodes of the stokes parameter field @xmath259 provide the @xmath0 lines where @xmath260 and @xmath255 . now we present the theoretical results for the polarization - resolved patterns of the dhflc cells . these patterns are computed for the cell of thickness @xmath261 @xmath16 m filled with the flc mixture flc-576a which was studied in ref . @xcite and described at the end of sec . [ subsec : kerr - effect ] . our first remark is that the angular dependence of the elements of the transmission matrix is determined by the angle difference @xmath262 which is the angle between the in - plane optical axis @xmath263 ( see eq . ) and the lateral wave vector @xmath264 ( see eq . ) . then the vector amplitudes [ eq : tilde_e_phi ] @xmath265 where @xmath266 is the angle between the optical axis @xmath263 and the polarization plane of the linearly polarized incident wave ( see eq . ) , describing the incident and transmitted waves with polarization ellipses rotated by the angle @xmath267 are related by the transformed transmission matrix @xmath268\,{\mathbf{t}}_{c}(\rho,-\tilde{\phi})\ , \exp[i\tilde{\phi}{\boldsymbol{\sigma}}_3].\end{aligned}\ ] ] from relation it follows that , given the angle @xmath266 , the sole effect of changing the azimuthal angle of the optical axis : @xmath269 is the rotation of the polarization ellipse field by the angle @xmath270 . in dhflc cells , this effect manifests itself as the electric field induced rotation and can be clearly seen in fig . [ fig : pol - f-20 ] that shows the patterns emerging after the dhflc cell calculated at @xmath271 deg for two values of the electric filed parameter : @xmath272 ( see fig . [ subfig : pol-00 - 20 ] ) and @xmath273 ( see fig . [ subfig : pol-03 - 20 ] ) . figure [ fig : pol - f-20 ] also illustrates the case of angular patterns that do not contain @xmath1 points . the geometry of such patterns is completely characterized by the @xmath0 lines . interestingly , at @xmath274 deg , it turned out that the @xmath259 nodal lines can be evaluated using the simplified equation @xmath275 where @xmath207 is the phase retardation expressed in terms of the eigenvalues of the matrix for uniformly anisotropic planar layers ( see eqs . and ) , that thus gives a sufficiently accurate approximation for @xmath0 lines . the angle @xmath266 can be regarded as the governing parameter whose magnitude determines the formation of @xmath1 points . given the aperture , the latter occurs only if the magnitude of @xmath266 exceeds its critical value . the case where the angle @xmath266 is close to the critical value is illustrated in fig . [ subfig : pol-00 - 39_5 ] . it can be seen that , at @xmath276 deg , the singularity structure of the polarization ellipse fields becomes complicated and is characterized by the presence of symmetrically arranged star - monstar pairs of @xmath1 points . in addition to the above discussed electric - field - induced rotation , the electric field is shown to facilitate the formation of @xmath1 points . clearly , the field induced biaxial anisotropy is responsible for this effect . in fig . [ fig : pol - f-41 ] , we show how the singularity structure of the polarization resolved patterns develops when the angle @xmath266 further increases . this structure can be described as symmetrically arranged chains of star - monstar pairs of @xmath1 points . as is indicated in fig . [ subfig : pol-03 - 41 ] , these tightly packed chains of @xmath1 points are generally located in the vicinity of the lines @xmath277 that give a high accuracy approximation for the @xmath278 nodal lines where @xmath258 and , similar to eq . , are determined by the phase retardation @xmath279 . note that , since @xmath280}^{\ast } } e_{p}^{({\mathrm{trm}})}$ ] , applicability of approximate formulas and implies that the phase difference between the components of the transmitted waves , @xmath281 and @xmath282 , is close to the phase retardation : @xmath283}^{\ast } } e_{p}^{({\mathrm{trm}})}\}\approx \delta$ ] . in sec . [ subsec : kerr - effect ] , a similar approximation has been used to derive the expression for the transmittance given by eq . . as it was mentioned in the previous section , the loci of @xmath1 points on the projection plane are determined by intersections of the @xmath278 and @xmath284 nodal lines . in our case , the star - monstar pairs are produced as a result of small - scale oscillations of the nodal line around the smooth curve described by eq . . experimentally , it is a challenging task to resolve accurately the chains of @xmath1 points resulting from such ripplelike oscillations in polarimetry measurements . the 2d polarization - resolved patterns are centrally symmetric being invariant under inversion through the origin : @xmath285 . the reason is that optical properties of planar anisotropic structures are unchanged under a 180-degree rotation about the normal to the cell ( the @xmath25 axis ) . more specifically , we have the central symmetry relation @xmath286 which is an immediate consequence of the fact that the matrix @xmath34 given in eqs . and remains intact when the azimuthal angle of the in - plane optic axis is changed by @xmath240 . another symmetry relation @xmath287 describes the transformation of the transmission matrix under reflection in the mirror symmetry axis directed along @xmath263 : @xmath288 . this relation immediately follows from eq . deduced in appendix [ sec : planar ] . by using formula , it is not difficult to show that the polarization ellipse field @xmath289 transforms into its mirror symmetric counterpart @xmath290 when the polarization azimuth @xmath266 changes its sign : @xmath291 . in this paper , we have performed transfer matrix analysis of polarization - resolved angular patterns emerging after electrically controlled short - pitch dhflc cells . our formulation of the transfer matrix method , which is a suitably modified version of the approach developed in refs . @xcite , involves the following steps : ( a ) derivation of the system of equations for the tangential components of the wave field in the @xmath18 matrix form ( see eq . ) ; ( b ) introducing the evolution operator ( propagator ) and the scattering matrix ; ( c ) defining the transfer matrix through the propagator and , finally , ( d ) deducing formulas that link the transfer and scattering matrices . description of this method is augmented by discussion of a variety of unitarity and symmetry relations ( see sec . [ subsec : symmetry ] and appendix [ sec : planar ] ) , with an emphasis on the special case of anisotropic planar structures representing homogenized dhflc cells . interestingly , the relations given in eqs . , and are shown to be essentially independent of the assumption of lossless materials and thus can be used when the medium is absorbing . in general , we found that , owing to its mathematical structure , the transfer matrix approach provides the framework particularly useful for in - depth analysis of symmetry related properties ( recent examples of such analysis can be found , e.g. , in refs . @xcite ) . similarly , one of the important results of a rigorous analysis performed within such a framework in ref . @xcite is the expression for the effective dielectric tensor describing the electro - optical properties of uniform lying flc helical structures with subwavelengh pitch . in sec . [ sec : optics - dhf ] we have extended theoretical considerations of ref . @xcite to the case of biaxial flcs and have applied an alternative technique of averaging over distorted helix to evaluate the dielectric tensor . this technique is presented in appendix [ sec : flc - spiral ] and gets around the difficulties of the method that relies on the well - known first - order expression for a weakly distorted helix . the modified averaging procedure allows high - order corrections to the dielectric tensor to be accurately estimated and improves agreement between the theory and the experimental data in the high - field region ( see fig . [ fig : expt_th ] ) . the resulting electric field dependence of the effective dielectric tensor is linear ( quadratic ) for non - diagonal ( diagonal ) elements with the coupling coefficients given by eq . . these coupling coefficients along with the zero - field dielectric constants determine how the applied electric field @xmath292 changes the principal values of the effective dielectric tensor ( see eqs . and ) and the azimuthal angle of optical axis . generally , at @xmath122 , the dhflc cell is uniaxially anisotropic with the dielectric constants , @xmath144 and @xmath143 , and the optical axis directed along the helix axis . then there are two most important effects induced by the electric field : ( a ) producing biaxial anisotropy by changing the eigenvalues of the dielectric tensor ; and ( b ) rotation of in - plane optical axes by the field dependent angle @xmath157 defined in eq . at sufficiently low electric field @xmath292 and non - vanishing zero - field anisotropy , the kerr - like regime takes place so that the principal values depend on the electric field quadratically whereas the optical axis angle @xmath157 is approximately proportional to @xmath292 . this is the orientational kerr effect previously studied in refs . @xcite for different geometries . our results on dependence of the coupling coefficients and the zero - field dielectric constants on the smectic tilt angle @xmath3 described in sec . [ subsec : tilt ] indicate a number of differences between uniaxial and biaxial flcs . what is more important , they show that the zero - field anisotropy may vanish at certain value of @xmath3 which might be called the isotropization angle : @xmath183 ( see fig . [ fig : eph_theta ] ) . in sec . [ subsec : isotropy ] , the isotropization point determined by the condition of zero - field isotropy is found to represent a singularity known as the exceptional point @xcite . for analytic continuation of the dielectric tensor in the complex @xmath175 plane , the exceptional points occur at the zeros of the square root in eq . where @xmath293 ^ 2+[\gamma_{xy}\alpha_e]^2=0 $ ] . in general , there are two pairs of complex conjugate values of electric field parameter representing four exceptional ( branch ) points . when the difference @xmath294 vanishes , the two branch points coalesce on the real axis at the origin . in the case of conventional uniaxial flcs , the analytic solution of the condition of zero - field isotropy can be obtained in the closed form and is given by simple formula where the isotropization angle , @xmath186 , is found to be a decreasing function of the anisotropy parameter @xmath295 . for biaxial flcs with @xmath296 , the solution can only be written in the parametrized form . as it can be seen in fig . [ fig : theta_r1 ] , the corresponding @xmath186-@xmath167 curves are splitted into two branches separated by the gap . these results significantly differ from the relation @xmath297 that can be easily obtained @xcite for the dielectric tensor averaged over the flc helix : @xmath298{\mathbf{i}}_3 + [ u_1\cos^2\theta - ( u_1\sin^2\theta+u_2)/2]{\hat{\mathbf{x}}}\otimes{\hat{\mathbf{x}}}$ ] . the difference stems from the fact that , in our approach , the effective dielectric tensor @xmath299 is defined through the averaged differential propagation matrix @xmath300 and thus is not equal to the averaged dielectric tensor : @xmath301 at the exceptional point , the kerr - like regime breaks down and the electric field dependence of the birefringence becomes linear ( see fig . [ fig : dn_alp ] ) . this might be called the pockels - like regime which is characterized by the harmonic electric field dependence of the transmittance of light passing through crossed polarizers ( see fig . [ subfig : transm_c_uni ] ) . the curve shown in fig . [ subfig : transm_c-2_uni ] illustrates the electro - optical response of a dhflc cell near the exceptional point . it is seen that sensitivity to the electric field and the magnitude of the transmittance at peaks are both considerably enhanced as compared to the case studied in ref . @xcite ( see also fig . [ fig : expt_th ] ) . we now try to put these results in a more general physical context . in quantum physics , the exceptional points are known to produce a variety of interesting phenomena including level repulsion and crossing , bifurcation , chaos and quantum phase transitions @xcite . for optical wave fields , a recent example is unidirectional propagation ( reflection ) of light at the exceptional points in parity - time ( @xmath302 ) symmetric periodic structures and metamaterials that has been the subject of intense studies @xcite . to the best of our knowledge , the role of exceptional points in optics of liquid crystal systems has yet to be recognized . the main problem with conventional uniaxial flcs is that the isotropy condition requires large values of the smectic tilt angle that are typically well above @xmath303 deg . though there are no fundamental limitations preventing preparation of flc mixtures with large tilt angles , this task still remains a challenge to deal with in the future . biaxial flcs , where the isotropization tilt angle can be sufficiently small when @xmath304 is close to @xmath305 , also present a promising alternative approach for future work . in sec . [ sec : conoscopy ] , in order to gain further insight into the electro - optical properties of the dhflc cells , we have combined the transfer matrix approach and the results for the effective dielectric tensor to explore the polarization - resolved angular patterns which are the polarization ellipse fields representing the polarization structure of conoscopic images of dhflc cells . in the observation plane , such 2d patterns encode information on how the polarization state of transmitted light is changed with the incidence angles and exhibit singularities of a different kind , the polarization singularities such as @xmath0 lines ( lines of linear polarization ) and @xmath1 points ( points of circular polarization ) . note that , similar to the above discussed exceptional point at which the angle @xmath157 becomes undetermined , @xmath1 points can be regarded as phase singularities ( optical phase singularities are reviewed in ref . @xcite ) . since the differential propagation matrix of planar structures is invariant under rotation of in - plane optical axes by @xmath240 , the patterns are centrally symmetric [ see figs . [ fig : pol - f-20]- [ fig : pol - f-41 ] ] . it was shown that , at fixed the angle @xmath306 between the optical axis @xmath263 and the polarization plane of incident wave , the sole effect of the electric - field - induced rotation by the angle @xmath157 is rotation of the polarization ellipse field as a whole by the same angle . the symmetry axis of the @xmath259 nodal lines ( @xmath0 lines ) is found to be directed along @xmath263 . when the @xmath266 is not too small , they can be approximated by solving eq . and thus are mainly determined by the phase retardation @xmath207 . similar remark applies to the @xmath278 lines and approximate formula . it turned out that this is the angle @xmath266 that plays the role of the parameter governing formation of @xmath1 points . when the magnitude of @xmath266 exceeds its critical value which , in our case , is close to @xmath307 deg , @xmath1 points emerge as symmetrically arranged and densely packed chains of star - monstar pairs ( see figs . [ fig : pol - f-39_5]-[fig : pol - f-41 ] ) . so , in dhflc cells , rotation of polarization ellipse fields and formation of @xmath1 points are two most important effects describing electrically induced transformations of the polarization - resolved angular patterns . these predictions can be verified experimentally . this work is now in progress . this work is supported by the hkust grant itp/039/12np . we begin with the relation @xmath308 known as the _ composition law_. this result derives from the fact that the operator @xmath309 is the solution of the system that satisfies the initial condition with @xmath310 replaced by @xmath311 . from the composition law it immediately follow that the inverse of the evolution operator is given by @xmath312 and can be found by solving the initial value problem @xmath313 for non - absorbing media with symmetric dielectric tensor , the matrix @xmath34 is real - valued , @xmath314 , and meets the following symmetry identities @xcite : @xmath315 where an asterisk and the superscript @xmath91 indicate complex conjugation and matrix transposition , respectively . in this case , the evolution operator and its inverse are related as follows : @xmath316 where a dagger will denote hermitian conjugation . by using the relations , it is not difficult to verify that the operator on the right hand side of eq . is the solution of the cauchy problem . in this section we present the results for anisotropic planar structures characterized by the dielectric tensor of the following form : @xmath317 where the optical axes @xmath318 lie in the plane of substrates ( the @xmath160-@xmath319 plane ) . the operator of evolution can be expressed in terms of the eigenvalue and eigenvector matrices , @xmath320 and @xmath321 , as follows @xmath322 for the dielectric tensor , @xmath323 and @xmath324 ( see eq . ) . assuming that @xmath325 , we have @xmath326 where @xmath327 and @xmath328 are the principal refractive indices ; @xmath329 is the parameter of in - plane anisotropy . for the case where the diagonal block - matrices , @xmath330 and @xmath331 , vanish , it is not difficult to show that the eigenvector and eigenvalue matrices can be taken in the following form : @xmath332 in addition , the eigenvectors satisfy the orthogonality conditions ( a proof can be found , e.g. , in appendix a of ref . @xcite ) that , for the eigenvector matrix of the form , can be written as follows @xmath333 upon substituting eqs . - into eq . , some rather straightforward algebraic manipulations give the transfer matrix @xmath334 { \mathbf{n}}^{-1 } , \\ & \label{eq : a_pm - pln } { \mathbf{a}}_{\pm}={\mathbf{e}}_{{\mathrm{m } } } { \mathbf{h}}\pm{\mathbf{h}}_{{\mathrm{m } } } { \mathbf{e}},\end{aligned}\ ] ] where @xmath335 . from eq . , the block @xmath35 matrices of @xmath336 are given by [ eq : tw_ij - pln ] @xmath337}^{t}}= \notag \\ & = { \mathbf{a}}_{- } { \mathbf{w}}_{- } { \mathbf{a}}_{+}^{t}-{\mathbf{a}}_{+ } { \mathbf{w}}_{+ } { { { \mathbf{a}}}^{t}}_{-}. \end{aligned}\ ] ] finally , we can combine eq . and eq . with eq . to derive the expressions for the transmission and reflection matrices @xmath338 describing the case where the incident wave is impinging onto the entrance face of the layer , @xmath23 . as it can be seen from formulas , the symmetry relations are satisfied even if the dielectric constants @xmath339 , @xmath340 and @xmath173 are complex - valued . so , the applicability range of identities includes lossy ( absorbing ) anisotropic materials described by the dielectric tensor of the form given in eq . . interestingly , inverse of the transfer matrix , @xmath341 , can be obtained from formula by changing sign of the thickness parameter @xmath342 : @xmath343 . in formulas , this transformation interchanges the matrices @xmath344 and @xmath345 , so that @xmath346 and @xmath347 . so , from eq . , the block matrices of @xmath341 are given by [ eq : w_inv_pln ] @xmath348 from the other hand , eqs . and give the transfer matrix and its inverse , respectively , expressed in terms of the transmission and reflection matrices , @xmath349 and @xmath350 . these expressions can now be substituted into eq . to yield the relations @xmath351 linking the transmission ( reflection ) matrix , @xmath352 ( @xmath353 ) , and its mirror symmetric counterpart @xmath69 ( @xmath70 ) . in conclusion of this section , we consider how the transmission and reflection matrices transform under the reflection in the @xmath354 plane when the azimuthal angle @xmath355 changes its sign : @xmath356 . from eqs . and , we have @xmath357 by using eq . it is not difficult to deduce a similar relation for the transfer matrix @xmath358 that can be combined with eq . to yield the result for the transmission and reflection matrices in the final form : @xmath359 an important point is that , similar to identities , the assumption of lossless ( non - absorbing ) medium is not required to derive the symmetry relations and eq . . for the case of uniaxially anisotropic structure with @xmath360 , it is not difficult to find the expressions for the eigenvalues that enter the eigenvalue matrix @xmath361 where @xmath362 ( @xmath363 ) is the refractive index for ordinary ( extraordinary ) waves and @xmath329 is the anisotropy parameter . similarly , after computing the eigenvectors , we obtain the eigenvector matrix in the following form : @xmath364 & m_yq_o \\ m_y & -m_xq_o \end{pmatrix } , \quad { \mathbf{h}}= \begin{pmatrix } m_x q_e & m_y n_o^2 \\ m_y q_e & -m_x [ n_o^2-q_p^2 ] \end{pmatrix } , \\ & \label{eq : n_oe - pln - uni } n_{e}=\dfrac{2q_e\mu}{n_o^2}(n_o^2-q_p^2m_x^2 ) , \quad n_{o}=2q_o\mu ( n_o^2-q_p^2m_x^2).\end{aligned}\ ] ] equations - can now be substituted into the general expression for the transfer matrix defined by formulas - so as to obtain the transmission and reflection matrices . for the general case of biaxial anisotropy , the expressions for the eigenvalues are more complicated than those for uniaxially anisotropic layer ( see eq . ) . these can be written in the following form : @xmath365 ^ 2 - 4\det{\tilde{{\mathbf{m } } } } } , \end{aligned}\ ] ] where the matrix @xmath366 is given by [ eq : tilde - m - pln ] @xmath367 \cdot{\mathbf{rt}}(\psi)= \begin{pmatrix } \tilde{m}_{11 } & \tilde{m}_{12}\\ \tilde{m}_{21 } & \tilde{m}_{22 } \end{pmatrix } , \\ & \label{eq : tilde - m-11 } \tilde{m}_{11}=n_{e}^2-(u_a+u_z[1+u_a])\,q_p^2 m_x^2 , \quad \tilde{m}_{12}=(u_a+u_z[1+u_a])\ , q_p^2 m_x m_y , \\ & \label{eq : tilde - m-22 } \tilde{m}_{22}=n_{o}^2-u_z q_p^2 m_y^2 , \quad \tilde{m}_{21}=u_z q_p^2 m_x m_y,\quad u_z=(\epsilon_{\perp}-\epsilon_z)/\epsilon_z.\end{aligned}\ ] ] it can be readily checked that the result for uniaxial anisotropy is recovered from eq . as the limiting case where the parameter of out - of - plane anisotropy @xmath368 is negligible and @xmath360 . for the eigenvector and normalization matrices , @xmath321 and @xmath369 , given in eq . and eq . , respectively , the results are @xmath370 these relations along with formulas give the transfer matrix for biaxially anisotropic films with two in - plane optical axes . before closing this section we briefly comment on the important special case of normal incidence that occurs at @xmath371 . in this case , the matrices @xmath372 defined in eq . can be written in the factorized form @xmath373 where @xmath374 is the matrix describing rotation about the @xmath25 axis by the angle @xmath216 . substituting eq . into eq . gives the block matrices @xmath375 expressed as a function of the director azimuthal angle @xmath355 . the result for the transmission and reflection matrices @xmath376 where the diagonal matrices @xmath377 and @xmath378 describe the case in which the director lies in the incidence plane , immediately follows from the relations and . in sec . [ subsec : diel - tensor - dhf ] , the effective dielectric tensor of a deformed helix flc cell is expressed in terms of the averages given in eqs . and . in this apppendix , we describe how to perform averaging over the helix pitch without recourse to explicit formulas for the azimuthal angle @xmath8 the flc director . we assume that the azimuthal angle is a function of @xmath160 , so that the free energy density can be written in the following form : @xmath379 where @xmath124 is the free twist wave number . then the free energy functional per unit volume can be written as the free energy density averaged over the helix pitch @xmath380/v={\ensuremath{\langle{f}\rangle}}_{x}\equiv \frac{1}{p}\int_0^p f { \mathrm{d}}x.\end{aligned}\ ] ] the first integral of the stationary point ( euler - lagrange ) equation @xmath381 is given by @xmath382 ^ 2 + u_{e}\cos\phi = e.\end{aligned}\ ] ] assuming that @xmath383 and @xmath384 is non - negative , equation can be recast into the differential form @xmath385 integrating eq . over the period yields the relation for the helix wave number @xmath386 where @xmath387 . this relation gives the helix pitch , @xmath7 , expressed in terms of the dimensionless parameter @xmath388 . we can now use equation to rewrite eq . in the following form @xmath389^{-1/2}.\end{aligned}\ ] ] an important consequence of this equation is the relation @xmath390 that allows to perform averaging over the helix pitch by computing integrals over the azimuthal angle @xmath8 . in particular , with the help of eqs . and , it is not difficult to deduce the following expression for the free energy : @xmath391)= \notag \\ u_k\left [ 1-m^{-1}\bigl\ { 1 + 2 ( \sqrt{m}-{\ensuremath{\langle{r^{-1}}\rangle}}_{\phi})/{\ensuremath{\langle{r}\rangle}}_{\phi } \bigr\ } \right].\end{aligned}\ ] ] in the low voltage regime , the parameter @xmath392 is small and the left hand side of eq . can be expanded into the power series in @xmath392 . the expansion up to the second order terms is given by @xmath393 and can be used to average the @xmath25 component of the polarization vector @xmath394 defined in eq . . the result reads @xmath395 where @xmath151 is the dielectric susceptibility of the goldstone mode @xcite and @xmath396 . similarly , the averages that enter the formulas for the elements of the effective dielectric tensor can be expressed in terms of the electric field parameter @xmath175 as follows : [ eq : averages - planar ] @xmath397^{-1/2}(1 + 3 v\gamma_v^2\alpha_e^2 ) , \\ & { \ensuremath{\langle{v_{zz}^{-1}\cos^2\phi}\rangle}}_{\phi } \approx \gamma_v(1 + 3 [ 1+v]^{1/2}\gamma_v\alpha_e^2 ) , \\ & { \ensuremath{\langle{v_{zz}^{-1}\cos\phi}\rangle}}_{\phi } \approx 2 \gamma_v\alpha_{e } , \quad \gamma_v= [ \sqrt{1+v}+1]^{-1 } , \\ & { \ensuremath{\langle{v_{zz}^{-1}\sin\phi}\rangle}}_{\phi}= { \ensuremath{\langle{v_{zz}^{-1}\sin\phi\cos\phi}\rangle}}_{\phi}= 0.\end{aligned}\ ] ] substituting the relations into eqs . give the effective dielectric tensor which is expressed in terms of the zero - field dielectric constants @xmath398 , \\ & \label{eq : epsilon_p - v } \epsilon_p = \epsilon_2\sqrt{1+v } , \quad \gamma_v=[\sqrt{1+v}+1]^{-1},\end{aligned}\ ] ] and the coupling coefficients @xmath399 , \\ & \gamma_{yy } =3 \epsilon_2 v \gamma_v^2\sqrt{1+v } , \quad \gamma_{xy } = 2 ( \epsilon_{1}-\epsilon_{\perp})\gamma_v\cos\theta\sin\theta.\end{aligned}\ ] ] 45ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop _ ( , , ) p. @noop _ _ , the liquid crystals book series ( , , ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) p. @noop _ _ , ed . ( , , ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . , , vol . ( , , ) p. @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1007/978 - 94 - 007 - 1530 - 1 [ _ _ ] , ed . ( , , ) p. link:\doibase 10.1109/tap.2012.2220316 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.105.053902 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.064103 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.213901 [ * * , ( ) ] http://dx.doi.org/10.1038/nmat3576 [ * * , ( ) ] link:\doibase 10.1103/physreva.89.065801 [ * * , ( ) ]
the motivation of this paper is the main theorem in pointwise asymptotic formulas on flat surfaces " of a. eskin and h. masur @xcite . they prove that on strata of quadratic differentials @xmath2 , parametrizing half translation surfaces , there exists almost everywhere a limit of the quadratic growth rate and it is constant ( with respect to a natural @xmath3 invariant liouville measure @xmath4 on @xmath2 see veech @xcite ) . the following questions arise : + 1 . does the quadratic growth rate associated to these constants majorize all other quadratic growth rates ? + 2 . how can one compute the constants ? are there ( sub ) spaces of quadratic differentials , where one can prove not only almost everywhere results but pointwise results ? + 3b . and if one has pointwise results , on which sets do the constants take different " values ? + our example the simplest one can find , but it has the advantage that one can compute things elementary . the above stated questions could all be answered in a precise way . + the paper contains results about the growth rates of cylinders of periodic trajectories or saddle connections on the two dimensional torus @xmath5 , @xmath6 with @xmath0 marked points . if the number of cylinders of periodic orbits ( po ) or saddle connections ( sc ) of length less than @xmath7 on @xmath8}$ ] is denoted by @xmath9},t)$ ] then the existence of the limits @xmath10},t)}{t^2}\ ] ] for every constellation of marked points i.e. for all @xmath11\in { { \mathbb{t}}^{2n}}$ ] is proved . , because this differs from the general marking only by a translation . all our counting results are invarinat under translations of the marking . ] remark : in the mathematical sense this is not the true parameterspace , because we would like to have all marked points different , but since we make either pointwise statements with respect to one marking or only measure theoretic statements with respect to the lebesgue measure on this space the difference to a well choosen parameter space plays no role . moreover all statements are true for any conformal class the tori might have . the existence of the limits is established by counting lattice points of certain lattices in the universal cover @xmath12 of @xmath13 . as a biproduct of this method we conclude that there is a subset @xmath14 ( which depends on what one counts : saddle connections or periodic families ) of measure one in the parameter space of all n markings ( contained in the set of all non rational n markings ) where the above limit takes a maximal value @xmath15 . moreover it is easy to see that if one approaches a point of the set @xmath16 the asymptotic constants of the sequence converge to @xmath15 . in other words the map @xmath17 \longmapsto \lim_{t\rightarrow \infty } \frac{n_{po / sc}({\mathbb{t}}^2_{[x_0, ... ,x_{n-1}]},t)}{t^2}\ ] ] is continous at the points of the set @xmath18 . in the case of two marked points and only in this case the set @xmath19 does not depend on what is counted ( po s or sc s ) . without restrictions one can assume that the first marking is @xmath20 $ ] , then the set of points in general position consists exactly of all the non rational points @xmath21 . in this case the quadratic growth rate function for saddle connections , assuming @xmath22 , is given by : @xmath23 & \longmapsto & \frac{6}{\pi}\left ( 1 + \prod_{p| n\ prime}\left(1-\frac{1}{p^2}\right)^{-1 } \sum_{\gcd(i , n)=1}\left(\frac{1}{i^2}-\frac{1}{n^2 } \right)\right ) \quad \mbox{on } { \mathbb{q}}^2 \nonumber\\ \left[x , y \right ] & \longmapsto & \frac{6}{\pi}+\pi \quad \mbox{for } ( x , y ) \notin { \mathbb{q}}^2\end{aligned}\ ] ] the continuity at non rational points implies that the above formula approaches @xmath24 as @xmath25 approximates a non rational number . in contrast to the general for two marked points this continuity is calculated directly by an estimate . furthermore for the two marked torus we write down the veech and affine groups for rational markings . an argument shows that every veech surface coming from a rational marking is a @xmath1 deformation of one of the family @xmath26 } : \ n \in { \mathbb{n}}\}$ ] . therefore all the veech groups of rationally marked tori are isomorphic to one of the veech groups @xmath27}$ ] associated to the family of surfaces @xmath28 . finally the asymptotic formula of veech and the asymptotic counting of lattice points in the plane is used to compute the index @xmath29}]$ ] . with the help of the description of the veech groups for rational 2 markings we rediscover the index of some classical congruence subgroups of @xmath1 . it is now well known @xcite that one can also use ratners theorem to obtain pointwise results in modulispaces of torus coverings branched over n marked points . of course from the point of pure existence of the growth rates this is more general and in fact covers the examples described in the paper . on the other hand the results in this paper are explicit and the extremality , or more restrictive the maximality of the almost everywhere constants are proved for the above cases . in this direction one of the main objects of the paper was to see if there is something in between purely rational markings and markings in general position which has not an extremal growth rate . this question also could be answered with yes " ( see theorem [ tlimes ] and the preparatory lemma [ schnigi ] ) . unfortunately our elementary methods can not directly taken over to prove something on families of branched coverings of tori , but nevertheless with the more sophisticated approach in @xcite one can compute all the siegel veech constants for two marked torus coverings which are not veech . the results in @xcite show , that in this cases the siegel veech constants , as functions of the degree of the covering , behave monotonically , too . in the light of eskins and this results it would be of interest to compute and prove related statements for marked veech surfaces and their branched coverings . acknowledgements : + i have to thank for the hospitality of the cpt in marseille luminy in person of ricardo lima and his group during the jear 99 as well as alex eskin and pascal hubert for fruitful discussions . i owe thanks to eugene gutkin for pointing out an unprecise statement to me and the referee for finding a mistake . this paper is part of my thesis and i like to give many thanks to tyll krger and serge troubetzkoy for their continuous encouragement making this work possible . for @xmath30, ... ,[x_n]\in { \mathbb{r}}^2/ { \mathbb{z}}^2 $ ] let @xmath31 ) } $ ] be the translation torus @xmath32 marked at the points @xmath30,[x_1], ... ,[x_{n-1}]\in { \mathbb{t}}^2 $ ] . if we do not write down the zeroth entry @xmath30 $ ] in the index vector we assume @xmath30=0+{\mathbb{z}}^2 $ ] . our calculations are restricted to the torus which is isometric to @xmath33 with the induced euclidean metric . with respect to this choice the periodic directions ( of the geodesic flow ) are exactly the ones with rational slope . the results are not restricted to these special translation tori , they hold for all marked translation tori . in fact all the arguments are invariant under @xmath3 deformations of the marked tori and any marked translation torus is an @xmath3 deformation of the studied ones . the vector @xmath34 * associated * to a saddle connection @xmath35 ( or a prime periodic orbit ) is the unique vector @xmath34 in @xmath36 that maps isometrically to @xmath35 in @xmath37 under the natural projection map . recall that in the case of a periodic orbit @xmath35 the associated vector @xmath34 characterizes a whole cylinder of closed geodesics . a rational two vector like @xmath38 is always assumed to fulfill the condition @xmath39 . @xmath40 * two points @xmath41,[y]\in { \mathbb{t}}^2 $ ] are said to be * relatively rational * , if @xmath42 . * @xmath43 denotes the set of saddle connections between the points @xmath41\in{\mathbb{t}}^2 $ ] and @xmath44\in{\mathbb{t}}^2 $ ] . * @xmath45 is the length of a @xmath46 and finally if @xmath47,[z]\in s \in sc_{(x , y)}$ ] let @xmath48 be the distance between @xmath47 $ ] and @xmath49 $ ] * measured along @xmath35 . * relatively rational is obviously an equivalence relation . every marking splits up in classes of relatively rational points . each class defines a marking , eventually of lower order . by the results of veech @xcite or gutkin and judge @xcite @xmath8}$ ] is a covering of the one marked torus @xmath50}$ ] if and only if all marked points @xmath51 are pairwise relatively rational . thus @xmath8}$ ] is a veech surface if and only if for all markings are in one and the same marking class and from @xcite it follows : @xmath52}\right)\subset sl_2({\mathbb{q}})$ ] , where @xmath52}\right)$ ] denotes the veech group of @xmath8}$ ] . any affine diffeomorphism @xmath53 of the unmarked torus @xmath54 is given by an element @xmath55)\in sl_2({\mathbb{z}})\oplus { \mathbb{t}}^2 $ ] : @xmath56)=[az+v]$ ] . since every affine diffeomorphism of the marked torus defines one of the unmarked torus by forgetting marked points one obtains : @xmath57}\right)\subset sl_2({\mathbb{z}}).\ ] ] a * lattice * @xmath58 is a subset of @xmath12 of the form @xmath59 a * point distribution * is a finite union of lattices . the set of * visible points * @xmath60 of of a discrete subset @xmath61 consists of all points @xmath62 such there is no @xmath63 and a point @xmath64 with @xmath65 . in other words : there is no other point of @xmath58 on the line between the origin and the points @xmath66 . for the correct counting of periodic orbits and saddle connections we need further : if a point distribution @xmath58 is contained in another point distribution @xmath67 , then the visible points @xmath68 of @xmath67 with respect to @xmath58 are those @xmath69 for which there is a @xmath70 such that @xmath71 . @xmath72 is called @xmath73 * complete * if @xmath74 . + @xmath58 is called * closed * , if @xmath75 for any integer @xmath76 such that there exist two points @xmath77 with @xmath78 . let @xmath41,[y]\in { \mathbb{t}}^2 $ ] , again we put @xmath79 . then the set of saddle connections @xmath43 is described by the visible points of the lattice of associated vectors @xmath80 . + for a discrete subset @xmath61 ( in general point distributions ) we define for @xmath81 @xmath82 here @xmath83 is the ( closed ) disc of radius @xmath7 centered in the origin of @xmath12 . given a nonrational ( marked ) point @xmath44 $ ] on the marked torus @xmath84}^2 $ ] . take a saddle connection @xmath85 from the origin to @xmath44 $ ] . it might be represented in @xmath12 by @xmath86 with two fixed integers @xmath87 and an @xmath88 choosen from the set @xmath89 . assume we take a second point @xmath41\in { \mathbb{t}}_{[0]}^2 $ ] @xmath90 , different from @xmath44 $ ] so , that there is a saddle connection @xmath91 , which fulfills an equation @xmath92 this means after mark the point @xmath41 $ ] the saddle connection @xmath93 destroys @xmath94 and generates two new saddle connections , @xmath93 and one in the set @xmath95 . it is clear by equation ( [ relation ] ) , that if the factor @xmath96 is a rational number , say @xmath97 , all saddle connections represented by @xmath98 , @xmath99 are divided by the marking @xmath100 to give saddle connections in @xmath95 and @xmath101 . now assume @xmath41 $ ] rationally divides two saddle connections from @xmath102 with the equations : @xmath103 with integers @xmath104 , @xmath105 and @xmath106 as above . subtract the two equations to obtain : @xmath107 since @xmath88 is non rational @xmath108 . assuming this we have : @xmath109 this equation has solutions if @xmath110 is ( componentwise ) divisible by @xmath111 . thus the saddle connections in the set @xmath112 dividing the ones from @xmath113 rationally , all have a common stretch factor @xmath114 . hence , if there is more than one rationally dividing saddle connection , then they are all given by the ( lattice ) set of vectors @xmath115 if @xmath44 $ ] , @xmath116 are fixed the different sets of rationally dividing saddle connections for the stretch factor @xmath114 are parameterized by @xmath117 . it is now clear that @xmath95 contains a lattice . = 7truecm to make shure there are not too much parallel saddle connections in @xmath101 and @xmath102 having non rational length ratio , consider the set @xmath118 of saddle connections that are represented by @xmath119 , where @xmath120 is non rational . [ countnrat ] @xmath121 . assume that the growth rate in question is of bigger than linear order , then there must be an integer @xmath122 where one can find two saddle connections @xmath123 represented by the points @xmath124 and @xmath125 , @xmath126 , @xmath127 ( we assume @xmath128 ) . if that would be not the case , for each @xmath122 there could be at most one saddle connection of the form @xmath129 in @xmath130 , therefore @xmath131 contradicting our assumption . now by the strahlensatz of elementary geometry ( see figure 1 ) each intersection point of the saddle connections @xmath132 and @xmath133 as above , must divide the length in rational amounts contradicting the assumption . let @xmath43 and @xmath134 be two sets of saddle connections on the marked tori @xmath135}$ ] , @xmath136}$ ] and suppose @xmath49 $ ] is rationally divided by @xmath44 $ ] . then @xmath137 is the set of vectors associated to saddle connections @xmath95 on @xmath138}$ ] which are parallel to vectors associated to saddle connections in @xmath139 and shorter in length . + to summarize : [ schnigi ] let @xmath41,[y],[z]\in { \mathbb{t}}^2 $ ] be different points and @xmath140 . if @xmath44\in s \in sc_{(x , z)}$ ] , then @xmath141 contains a lattice , if and only if there is a rational number @xmath142 such that @xmath143 , moreover the number @xmath144 is unique in the sense that this equation holds for any saddle connection represented by elements of @xmath141 . these are just the statements above translated to the situaion @xmath41\neq [ 0]$ ] . the lemma is viewed as the generalisation of the following observation on the torus @xmath145 ( with the origin as a marked point ) : a point @xmath146 is rational if and only if it is in the intersection of saddle connections . if this is the case the saddle connections through @xmath146 give rise to a lattice ( by taking representatives in @xmath12 ) . the moral of the lemma is : when counting saddle connections for general markings , all markings which are lying on some saddle connection and dividing the length of it rationally will cause a lattice of intersection points . this is important as we will see later , this decreases the asymptotic constant of that marking . because of this observation we also have to take care about the definition of the set of markings in @xmath147 , where we assume the growth rates to be extremal . thus for saddle connections we define the set of markings ` in general position ' ( shortly : ` gp ' ) in @xmath147 inductively : @xmath148\in { \mathbb{t}}^{4 } : \ x_1-x_0 \notin { \mathbb{q}}^2 \}\\ \mbox { } \\ { \mathbb{t}}^{2(n+1)}_{gp(sc)}=\left\ { \begin{array}{l } [ x_0, ... ,x_n]\in { \mathbb{t}}^{2(n+1)}:[x_0, ... ,x_{n-1}]\in { \mathbb{t}}^{2n}_{gp(sc)},[x_n ] \in { \mathbb{t}}^2 \ \mbox{and } \\ \forall \ i \in \ { 0, ... ,n \ } \ x_i \mbox { does not divide any saddle connection}\\ \mbox{defined by the marking } [ x_0, ... ,x_{i-1},x_{i+1}, ... ,x_{n } ] \mbox { rationally } \end{array } \right\}. \end{array}\ ] ] thus the set of non extremal n markings is a set of full lebesque measure on @xmath147 ( see theorem [ tlimes ] ) . for periodic orbits the set of markings in general position is defined in the following way : @xmath149:\ x_j - x_i \notin { \mathbb{q}}\ \forall \ i , j \in \{0, ... ,n\ } \mbox { and } j \neq i \}.\ ] ] observe that for all @xmath150 @xmath151 . the sets are equal in the cases @xmath152 ( trivial ) and @xmath153 , for more marked points the inclusion is strict . it will turned out the markings in general position are exactely the sets of markings on which the growth rates are maximal . note that in the case of two marked points the set of markings in general position is the same as the set of non rational markings , always assuming one of the markings to be @xmath154 $ ] . it should be noted here , that the pure existence of the growth rate functions for any choice of marking could also be seen using the siegel veech formula together with ratners theorem for ergodic measures on homogeneous spaces ( see @xcite ) . the more elementary argumentation here and in the chapter on general markings shows , that even for complicated markings one has to count only simple lattice point distributions to obtain the quadratic growth constant . to compute quadratic growth rates we recall some preparatory lemmas . the first is quite well known and uses an argument going back to gauss . [ gauss ] . then @xmath156 to begin we treat the case @xmath157 . by drawing horizontal and vertical lines through the lattice points , we got a decomposition of the plane in rectangles with sidelengthes @xmath158 . every lattice point of @xmath58 can be viewed as the upper right vertice of one rectangle . so the number of points of @xmath58 in the concentric circle @xmath83 of radius @xmath7 is determined by the number of rectangles in that circle up to a certain correction term coming from the fact that a rectangle might be only partly inside the ball . but after increasing the radius of the ball by the length of the diagonal of the rectangles @xmath159 the estimate @xmath160 holds . and by decreasing the radius with the same amount we have : @xmath161 putting together these two estimates leds to : @xmath162 in the general case we have the estimate : @xmath163 the statement follows easily . before stating the next proposition let us make a general remark on our strategy of counting . if our interest is for example counting prime closed geodesics , we associate a certain point distribution in @xmath164 to all geodesics in question . this point distribution @xmath58 does not reflect a priory the prime geodesics , but it will be choosen in a way that its quadratic asymptotics as in the last lemma can be calculated easily . the points in @xmath58 which are associated to prime geodesics will be the visible points of @xmath58 with respect to @xmath165 . the price we have to pay is that we can not relate the distribution of visible points and all points of ( the completed and closed ) @xmath58 just by the factor @xmath166 . we have to choose our multiplicities with respect to @xmath58 : [ vielf ] let @xmath167 be a @xmath165 complete and closed lattice . let @xmath168 be the set of numbers @xmath169 such that @xmath170 . then the relation @xmath171 holds between the numbers of points of @xmath58 and the number of visible points @xmath60 of @xmath58 . for the lattice @xmath172 this implies : @xmath173 first let us check that there is no @xmath174 that is not an integer multiple of some @xmath175 . assume there is , then there exists a @xmath176 and a rational number @xmath177 with @xmath178 . so @xmath179 divides @xmath180 , a contradiction to the @xmath165 completeness of @xmath58 . since @xmath58 is closed and every element in @xmath58 is some @xmath181-multiple of an element in @xmath60 , thus : @xmath182 the lattice @xmath172 is obviously @xmath165 complete , closed and @xmath183 , thus @xmath184 counting only a certain percentage @xmath146 of the length of each line in a point distribution @xmath58 is equivalent to counting the length of vectors in the point distribution + @xmath185 . thus there is a relation : @xmath186 remark : as is seen above the relating factor between cylinders of geodesics and prime geodesics will be @xmath187 instead of @xmath166 because we do nt distinguish between the two directions a prime geodesic might have . our decision to do that is based on the point of view : count only ( the length of ) the geometric object . define the set of lattices parametrized by @xmath188 , with @xmath189 as : @xmath190 further for @xmath150 and a divisor @xmath191 of @xmath0 define @xmath192 to be the set of numbers which are relatively prime to @xmath193 @xmath194 ( if @xmath195 we write shortly @xmath196 instead of @xmath192 ) and the point distributions @xmath197 with this conventions [ pvert ] the point distributions @xmath198 are for all @xmath150 complete with respect to @xmath165 and closed . we have : @xmath199.\ ] ] as a consequence , for all prime numbers @xmath146 @xmath200 first one has to find all integers which map @xmath198 under elementwise multiplication to @xmath198 itself . for @xmath201 and @xmath202 we have @xmath203 , if and only if @xmath204 therefore @xmath169 has to be in @xmath205 . obviously the multiplication with a number from this set maps points of @xmath198 again to @xmath198 . so if the point distributions are complete with respect to @xmath165 they are also closed . to see the last statement take @xmath206 and choose @xmath207 with @xmath208 . this means @xmath209 with @xmath210 and @xmath211 . the equation in the first coordinate says that @xmath169 and @xmath212 are relatively prime to @xmath0 and from this it follows that @xmath213 has to be a multiple of @xmath0 . thus @xmath214 is already contained in @xmath198 and the completeness follows . using equation ( [ factor ] ) from proposition [ vielf ] gives the first part of the lemma . for the second part we simply observe that because @xmath146 is prime @xmath215 : @xmath216 remark : obviously the lattices @xmath217 are for @xmath218 never @xmath165 complete . before we compute the relating factor of equation ( [ vf ] ) in general , an observation is helpful : let @xmath219 be the principal character modulo @xmath0 , that is @xmath220 if @xmath221 and is equal to @xmath222 if not , then obviously : @xmath223 where @xmath224 is the @xmath225 series with respect to @xmath226 evaluated at @xmath227 . the following is well known ( see for example exercise ( 3.d ) page 166 in @xcite or serre @xcite ) [ sumform ] if @xmath35 is a complex number with real part bigger than @xmath228 then : @xmath229 the condition on the real part of @xmath35 guaranties absolute convergence of the series so that the following manipulations are justified . we start with the right hand side : @xmath230 we have used the well known mbius @xmath4 function , the formula @xmath231 and summation over @xmath232 to reorder the sum in a useful way . evaluating the above @xmath225 series for @xmath233 we find @xmath234 the left hand side of this equation is easy to compute with the help of lemma [ gauss ] and the definition of @xmath198 @xmath235 use eulers @xmath236 function to identify @xmath237 for later use we show [ completion ] + @xmath198 containes the @xmath165 completion of the lattice @xmath238@xmath239 . it is enough to show that a multiple of each point @xmath240 is in @xmath241 . that is we have to solve @xmath242 . but this is always possible since @xmath96 and @xmath169 are relatively prime with respect to @xmath0 . we now describe the affine group @xmath245 for a rational point @xmath246 . [ afftor ] assume @xmath247 . then @xmath248 if @xmath249 we have @xmath250 @xmath251 is a consequence of the discussion around equation ( [ sl2 ] ) in the preparation section . the only thing which is left to prove is the statement on the translation vector @xmath252 . it is defined only modulo @xmath165 , thus one can assume @xmath253 . the subgroup @xmath254 of pure translations is trivial if @xmath255 : + assume the lattice @xmath256 generated is by use of @xmath252 translated to @xmath165 and by the same translation @xmath165 is moved to @xmath257 , then @xmath258 must hold . thus either @xmath252 is trivial , or @xmath100 is one of the exceptions . in general the same problem comes up if the map should interchange the lattices @xmath165 and @xmath257 . the only possibility is : @xmath259 has to map @xmath257 to @xmath260 and @xmath252 translates this into @xmath165 . thus we obtain the conditions : @xmath261 for the three exceptions there is in fact only one condition , because in these cases @xmath262 , thus @xmath263 is automatically contained in these affine groups . moreover the translation @xmath264 is contained in the groups . [ kong]the veech group @xmath265 where @xmath266 $ ] * for @xmath267 is @xmath268 * for @xmath269 ( or @xmath270 analogously ) is @xmath271 * for @xmath272 is @xmath273 in the case @xmath274 that is @xmath275 we have to add the rotation : @xmath276 * for @xmath277 ( or equivalently @xmath278 ) is @xmath279 by the last proposition we have to evaluate the condition @xmath280 on the coefficents @xmath281 . if for example @xmath282 then we have @xmath283 it follows directly that @xmath284 , @xmath285 and @xmath286 is arbitrary . since the determinant of the matrix is always @xmath228 we have the condition @xmath287 . therefore @xmath288 , if @xmath289 . the other three cases are equally simple , we omit them . we prove now that all 2 marked tori are affine equivalent to one from a special family , if the marking is rational . [ reduction ] if @xmath290 $ ] ( @xmath22 ) , @xmath291 is affine isomorphic to @xmath292}$ ] . more precisely : there exists an affine diffeomorphism of @xmath13 represented by an element of @xmath1 which maps @xmath293 $ ] to @xmath294 $ ] ( and preserves @xmath20 $ ] ) . the veech group @xmath295 of the one marked torus is represented by @xmath1 . thus it is enough to show : given a point @xmath296\in{\mathbb{t}}^2 $ ] with @xmath39 , there is a linear map in @xmath1 that maps it to the point @xmath297\in{\mathbb{t}}^2 $ ] . let @xmath298 then the equation @xmath299 holds . we want the vector @xmath300 to become the second row of a matrix @xmath301 , thus the vector @xmath302 representing the first row has to fulfill @xmath303 . this equation has solutions @xmath304 because @xmath305 and @xmath306 are relatively prime . taking solutions @xmath307 defines an @xmath301 with @xmath308 with @xmath309 relatively prime to @xmath0 . hence the congruence @xmath310 has a solution @xmath311 and one can use the linear dehn twists @xmath312 and @xmath313 to obtain the result up to a rotation of 90 degrees . [ isom ] the veech groups @xmath314 for points @xmath315 $ ] with @xmath39 are all isomorphic . the isomorphism is given by conjugation with an element of @xmath1 . moreover the asymptiotic growth rates are the same , if the surfaces are affine isomorphic . the first two statements follow from the above proposition . veechs asymptotic formula depends only on the conjugacy class of the veech group in @xmath3 and on the areas of the cylinders of closed geodesics ( see for example vorobetz paper @xcite ) , which are invariant under affine isomorphisms . in other words the asymptotic constants depend only on the @xmath3 orbit of the given veech surface in the moduli space . but proposition ( [ reduction ] ) shows the marked surfaces in question are all on one @xmath3 orbit . since every rational number @xmath100 has a unique representation as above , one can restrict the considerations to families of markings defined by @xmath316\quad \mbox{or } \quad x_{(n , n)}:=\left[\frac{1}{n},\frac{1}{n}\right]\quad \mbox{where } \ 1\neq n \in { \mathbb{n}}\ ] ] and to the corresponding tori . we choose the family @xmath317 . to compute the index @xmath318})]$ ] , the first observation is that the only affine map @xmath319})$ ] ( @xmath39 ) which interchanges @xmath20 $ ] and @xmath320 $ ] is given by rotation of 180 degrees and a translation by the vector @xmath38 ( for @xmath321 ) . since @xmath53 maps @xmath320 $ ] to @xmath222 it is of no interest for the following discussion up to the subgroup of generated by the idempotent element @xmath53 , @xmath322})$ ] can be viewed as the isotropy subgroup @xmath323}({\mathbb{t}}^2)\subset a\mathit{ff}({\mathbb{t}}^2 ) $ ] . for a translation structure @xmath324 and @xmath325 one defines the isotropy group @xmath326 of @xmath100 and its associated veech group in the usual way : @xmath327 these are subgroups of @xmath328 , @xmath329 respectively . as a direct consequence from the definitions and elementary group theory one obtains : [ shindex ] let @xmath330 be a translation structure and @xmath331 the structure @xmath330 where the point @xmath100 is marked . then @xmath332= \operatorname{ord}(a\mathit{ff}(u)/ a\mathit{ff}_x(u))= [ v(u):v_x(u)]=\operatorname{ord}(v(u)/v_x(u))\ ] ] where @xmath333 is the orbit of @xmath100 under @xmath328 and @xmath334 denotes the set of right cosets @xmath335 of the subgroup @xmath336 . see for example chapter i 5 in s. langs book on algebra @xcite . lemma [ reduction ] above implies together with proposition [ shindex ] for @xmath321 @xmath337 the last equation on the cardinality of the set of pairs @xmath302 is well known . @xmath338 is normal in @xmath1 and therefore also in @xmath339})$ ] , thus @xmath340}({\mathbb{t}}^2)\cong v({\mathbb{t}}^2_{[\frac{p}{n},\frac{q}{n}]})/(-id).\ ] ] since @xmath263 generates a subgroup of order @xmath227 and the subgroup @xmath341 } ) \subset a\mathit{ff}({\mathbb{t}}^2_{[\frac{p}{n},\frac{q}{n}]})$ ] of pure translations is trivial if @xmath342 we have @xmath343})]=\frac{n^2}{2}\prod_{p|n \ prime } \left(1-\frac{1}{p^2}\right)\ ] ] for @xmath153 the index is @xmath344 because @xmath263 is already an element in @xmath345}({\mathbb{t}}^2)$ ] . in contrast to this calculation there is a second way to compute the index of the veech groups using the asymptotic formula of veech and the results from the last section . to begin , the veech groups are presented as subgroups of @xmath1 , but since we do not want to count orbits in both directions we take the quotient modulo @xmath346 : @xmath347 . since both modulo @xmath263 equivalent elements are in @xmath348 we can do the same with the veech groups and loose no information when taking the preimage in @xmath3 . thus , denote by @xmath349 the image of @xmath348 in @xmath350 . the orbits @xmath351 ( @xmath352 is the number of different orbits ) of the vectors @xmath353 associated to a maximal cylinder @xmath354 of closed geodesics which fills the torus under @xmath355 give always the same asymptotic constant ( for the formula see proposition 6.3 in @xcite : @xmath356}{area({\mathcal{c}}_i ) } \ ] ] since each cylinder @xmath354 fills the torus we have @xmath357 and @xmath358=1 $ ] . here @xmath359 is the group generated by linear dehn twists around @xmath354 . thus we can write : @xmath360 \ vol\left({\mathbb{h}}/psl_2({\mathbb{z}})\right)= \nonumber \\ & = & \frac{\pi}{3}[sl_2({\mathbb{z}}):v({\mathbb{t}}^2_{x_{n}})].\end{aligned}\ ] ] the other side of veech s asymptotic formula is the quadratic growth rate of the lengths of the maximal cylinders that completely fill the torus . their directions are exactly the directions of lines in @xmath12 which begin at the origin and cross some point of the lattice @xmath361 . to get a set of integer points which represent ( eventually multiples of the length of ) the closed cylinders in question we simply multiply with @xmath0 and get the lattice @xmath362 studied earlier . since we are on the torus @xmath363 to count the length spectrum of the cylinders correctly one has to look for the visible points of the completion of @xmath364 with respect to @xmath165 . this completion , by proposition [ completion ] , is the point distribution @xmath198 , thus finally : @xmath365^{-1}\ ] ] what is left to calculate is @xmath352 : [ invariante ] @xmath366 observe that the ratio of the length of the two saddle connections connecting @xmath20 $ ] and @xmath367 $ ] and bounding the same cylinder of periodic orbits is an invariant under the operation of the veech group . furthermore modulo @xmath368 we can represent every periodic orbit or saddle connection by a point in @xmath369 . thus the directions with only one periodic family are represented by : @xmath370 the line from @xmath371 to @xmath372 intersects exactly one point of the set @xmath361 characterized by @xmath373 if @xmath374 runs through the numbers relatively prime to @xmath0 , then @xmath100 does this as well . thus the above regarded length ratio is in any case one of the numbers @xmath375 we see , exactly two relatively prime numbers modulo @xmath0 , namely @xmath100 and @xmath376 having the same invariant . if this invariant is complete , which will be shown in the next lemma , then the number of different one cylinder orbits under the operation of the veech group is given by @xmath377 the exceptional case @xmath153 is trivial , because there exists only one direction of one cylinder in @xmath378 . let @xmath291 be a rational marked torus . two directions @xmath379 with only one periodic cylinder are in the same orbit under the operation of the veech group , if they have the same ratio @xmath380 here @xmath381 are the two saddle connections which are on the boundary of the cylinder in direction @xmath382 . obviously the condition is necessary . all that is left is to show that it is sufficent . that is we have to find an element of the veech group that is a map between the two directions with the same invariant . forgetting for a moment the marked point @xmath100 we can map any periodic direction to any other by elements of @xmath1 . especially the saddle connections in question are mapped on one another . because the invariant is the same for both possible orientations of the saddle connections and @xmath383 we have found a map , which also transports one marked point to the other . now we can compute the index with the help of the counting formula again : let @xmath384 $ ] , @xmath321 , @xmath385 then @xmath386=\frac{n^2}{2}\prod_{p|n \ prime}\left(1-\frac{1}{p^2}\right)\ ] ] holds . in the case @xmath153 the index is @xmath344 . from equality ( [ kxn ] ) together with @xmath387 for @xmath321 it follows @xmath388=\frac{3}{2\pi } \varphi(n)\left[\lim_{t\rightarrow \infty } \frac{n(g^{v}_{i^n } , t)}{t^2 } \right]^{-1}.\ ] ] then put in the right hand side of equation ( [ vffinal ] ) to get the result . if @xmath153 then @xmath389 and we obtain @xmath344 . the two ways of computing the index of the veech groups of two marked tori might be used to obtain the asymptotic formula ( [ vffinal ] ) , counting the lattice points , without using the second section . one can take the first index computation and the veech formula together with the knowledge about the number of cusps @xmath390 ( see proposition [ invariante ] ) to compute @xmath391 backwards . nevertheless the first given way to compute the index of the veech groups seems to be a simple method to calculate for the index of the well known groups ( for example compare s.lang @xcite @xmath392 in @xmath1 . either the @xmath393 are subgroups of index two in @xmath348 if @xmath321 , or in the case @xmath153 @xmath394 and @xmath395 is isomorphic to @xmath396 . thus : @xmath397= \left\ { \begin{array}{cl } n^2\prod_{p|n \ prime}\left(1-\frac{1}{p^2}\right ) & \textrm { if } \ n>2 \\ 3 & \textrm { if } \ n=2 \end{array } \right.\ ] ] we begin with a general observation concerning the directions of periodic families on the torus . these directions are exactly all the rational " directions and this fact will not change how many points we will mark and wherever we will place them on the torus . the only thing that might change is the number of periodic families in a given rational direction , but in this case they are all of equal length . for the two marked torus @xmath291 there are always two families in any rational direction if @xmath100 is non rational . there are saddle connections connecting @xmath20 $ ] und @xmath41 $ ] but which bound no periodic family because they are not in a rational direction . for rational markings @xmath100 this can never happen . collecting things for rational @xmath100 : either there is a rational direction with two periodic families and the bounding saddle connections connecting the same marked point , or there is only one family but then the two bounding saddle connections will connect the marked points @xmath20 $ ] and @xmath41 $ ] . therefore counting the length @xmath398 of cylinders of periodic trajectories for extremal markings @xmath100 on @xmath291 is easy : @xmath399 to treat the rational case we have to subtract the directions where only one family is from the doubled number of periodic orbits of the one marked torus . let us denote the number of single cylinders on @xmath291 with length smaller than @xmath7 by @xmath400}({\mathbb{t}}^2_x , t)$ ] . thus we write : @xmath401}({\mathbb{t}}^2_x , t)+o(t^2)\ ] ] with the help of the discussion at the end of the last paragraph one computes for a rational point @xmath402 : @xmath403 to count saddle connections is less easy , but again using the last paragraph ( and the discussion before equation ( [ kxn ] ) we can write down the associated growth rates as follows : @xmath404 here ( [ vffinal ] ) and proposition [ invariante ] is used . the last term needs some explanation . we have seen that @xmath405 already counts the length of the simple periodic cylinders . it counts as well the length of their boundaries that is the sum of the length of the two bounding saddle connections . by proposition [ invariante ] there are @xmath406 different families of orbits of saddle connections labeled by their length ratios . equation ( [ kxn ] ) shows that these orbits under the veech group all have the same growth rate , namely @xmath407 . the relation of the length of a saddle connection bounding a family to the length of the family is more or less the invariant @xmath408 . anyway it is one of the numbers @xmath409 or @xmath410 where @xmath411 . with the help of equation ( [ scaling ] ) one finds the above expression . continuing the evaluation : @xmath412 the last equality follows with the help of equation ( [ in ] ) . for non rational markings one obtains : @xmath413 summarizing : the limits @xmath414 exist for all @xmath100 , moreover there are estimates @xmath415 and @xmath416 for @xmath88 non rational . the inequalities are strict if @xmath100 is rational . moreover for each sequence of rational points @xmath417 converging to a non rational point @xmath100 : @xmath418 this is the continuity of the quadratic constants at non rational points . the existence of the limits is proved in the equations ( [ ponr ] ) , ( [ po ] ) , ( [ rat ] ) and ( [ snrat ] ) , even more the computations give explicitly the values of the limits . in the case of saddle connections it is indeed not easy to see the continuity of the growth rate function from the expression ( [ rat ] ) directly ( in fact before the author found the esimate below he does not believe in the stated formula ) . by comparing equation ( [ snrat ] ) ( the limiting constant ) to the first line of ( [ rat ] ) one has to show : @xmath419 by using the equation ( [ sumform ] ) , which states @xmath420 and the estimates @xmath421 we have : @xmath422 clearly if @xmath423 the right and left hand side of the inequalities converge to @xmath424 . the continuity in the case of periodic orbits follows immediately from equation ( [ po ] ) . * let @xmath8}$ ] be a n marked translation torus . then the limits @xmath425},t)}{t^2}\ ] ] exist . * the set @xmath16 has full lebesque measure on @xmath426 and the inequalities @xmath427},t)}{t^2 } \leq \lim_{t \rightarrow \infty}\frac{n_{sc}({\mathbb{t}}^2_{[y_0, ... ,y_{n-1}]},t)}{t^2}= \frac{n(n+1)}{2}\pi+\frac{3n}{\pi}\ ] ] and @xmath428},t)}{t^2 } \leq \lim_{t \rightarrow \infty}\frac{n_{po}({\mathbb{t}}^2_{[y_0, ... ,y_{n-1}]},t)}{t^2}= \frac{3n}{\pi}\ ] ] hold for all @xmath429 . in both cases ( @xmath430 s and @xmath431 s ) the inequality is strict , if @xmath432 * + if @xmath433 is a sequence with + @xmath434 @xmath435 then @xmath436},t)}{t^2}\right)= \lim_{t \rightarrow \infty}\frac{n_{sc / po}({\mathbb{t}}^2_{[y_0, ... ,y_{n}]},t)}{t^2}\ ] ] we prove the statements by induction over the number @xmath0 of marked points . the idea is to write down , modulo lower order terms , the numbers of saddle connections ( or po s ) for a given marking as sums over the numbers for an marking with one marked point less and the terms occuring when one marks the forgotten point . the results from the first section will imply the statements . to start for one point there is nothing to show , moreover we have already seen the proof for @xmath153 . so we assume the statements are true for all n markings @xmath437 of the torus . we add another marking @xmath438 . if the resulting marking is in general position , @xmath439 ( for saddle connections ) we obtain for saddle connections @xmath440},t)=n_{sc}({\mathbb{t}}^2_{[x_0, ... ,x_{n-1}]},t)+ \sum^{n}_{i=0}n(sc_{(x_i , x_n)},t)\ ] ] furthermore by the definition of general position markings @xmath441 for all @xmath442 . for the case @xmath443 : saddle connections starting and ending in @xmath438 are bounding cylinders of periodic trajectories . we have to take care of multiplicities when counting these @xmath444 thus @xmath445},t)}{t^2}&= & \lim_{t \rightarrow \infty}\frac{n_{sc}({\mathbb{t}}^2_{[x_0, ... ,x_{n-1}]},t)}{t^2}+ n\pi+\frac{3}{\pi}=\nonumber \\ & = & \frac{n(n+1)}{2}\pi+\frac{3n}{\pi}. \nonumber\end{aligned}\ ] ] we assume now that the marking is in general position with respect to periodic orbits . since a saddle connection starting and ending in @xmath438 bounds a cylinder of periodic trajectories , one has to add in each rational direction a new periodic family . thus the periodic orbit growth rates for markings in general position are : @xmath446},t)}{t^2}= \lim_{t \rightarrow \infty}\frac{n_{po}({\mathbb{t}}^2_{[x_0, ... ,x_{n-1}]},t)}{t^2}+ \frac{3}{\pi}=\frac{3n}{\pi}.\ ] ] if the marking @xmath447 is not in general position , it could be divided in classes of relatively rational points ( this is of course meaningful in any case , po s or sc s ) . the set of these classes is denoted by @xmath448}$ ] . for each class the veech theory holds ( see @xcite ) and guaranties that all the limits exist . for the inequalities between the limits one observes : if the marked points are rational there exists always a direction in which there are strictely less than the maximal number of @xmath0 cylinders of closed trajectories . the orbit of this direction under the operation of the veech group leads to a positive asymptotic growth constant ( proposition 6.1 in @xcite ) . but a lower ( than maximal possible ) number of closed cylinders in a given direction causes multiplicities " ( on has to count only visible points ) . hence the limit growth rates are in both cases ( po s and sc s ) smaller than the ones for markings in general position . this shows : if ther is one class in @xmath448}$ ] containing more than one point , then the growth rates have to be smaller than for markings in general position . to compute the growth rates in the cases of periodic orbits , we observe that each class @xmath449}$ ] defines a marked veech torus @xmath450 , by marking @xmath13 only with the points of that class . thus : @xmath451},t))=\sum_{m \in \mathcal{r}_{[x_0, ... ,x_n ] } } n_{po}({\mathbb{t}}^2_{m},t ) + o(t^2).\ ] ] here is used that by definition between points of different classes there are never saddle connections with rational slope ( or equivalently : with bound a closed trajectory ) . all remains is to count the saddle connections between points of different classes . to do this let @xmath51 be a marked point not rationally equivalent to @xmath438 and @xmath452 the set of saddle connections connecting @xmath51 and @xmath453 on the n marked torus @xmath454}$ ] . what we have to count are sets of saddle connections from a given set of saddle connections between two marked points ( for example the set @xmath452 ) which are parallel and shorter to another set of saddle connections ( for example @xmath455 ) . however by lemma ( [ schnigi ] ) and proposition ( [ countnrat ] ) sets of this kind are always lattices modulo a mistake of order @xmath456 , which does not count in the quadratic limit . + there are two different cases : first , there is a nonempty subset in @xmath455 consisting of saddle connections parallel to saddle connections in the set @xmath452 and shorter in length . the set of all indices @xmath457 of such @xmath453 are denoted with @xmath408 . second , a nonempty subset @xmath455 is parallel to one from @xmath452 , but the first are longer . as above denote the set of all indices of such @xmath453 with @xmath181 . moreover the sets of saddle are ( modulo sets of lower order growth rates ) given by some lattices @xmath458 or @xmath459 respectively . we can write : @xmath460 here @xmath461 denotes the lattices of saddle connections between @xmath51 and @xmath438 on the two marked tori @xmath462}$ ] . for all @xmath463 @xmath464 modulo terms of strictly lower order size the above expressions are all lattices , hence they have quadratic growth rate limits . after these steps for all markings @xmath465 $ ] the induction is complete . from the two equalities it follows immediately : + if intersections of the above kind have a nonzero quadratic growth rate ( so the marking is rational in the sense of our definition ) than the associated growth rates are smaller than for a non rational marking . the continuity follows from the fact that by approximating a marking in general position all the denominators of all the length ratios are growing over all bounds . because of the simplicity of our methods we are not able to conclude anything about ( families of ) branched coverings of tori , without using the existence of the growth rate limits by the method explained in @xcite . even the sets where the growth rate is the biggest is not seen as just the preimage of the non rational markings on the torus . for closed cylinders of geodesics at least it has to be contained in this set . this is because the geometry and combinatorics of the covering might cause new sets where the growth rates are different . geometrically the reason for this is that the length spectrum of the cylinders of periodic trajectories is connected in a way to the saddle connections bounding them , which depends on the covering and where the singular points are . if one makes the simplifying assumption that all inverse images of the marked points are itself marked ( such coverings are called ` balanced coverings ' by some people ) , then from theorem [ tlimes ] one can conclude : let @xmath466}$ ] a covering branched only over @xmath467 with the induced translation structure @xmath468 . further denote by @xmath469 the translation structure where each inverse image of a marked point is itself marked . then , after rescaling the volume of @xmath470 ( with respect to the induced flat metric ) to one : @xmath471},t)}{t^2}\ ] ] ( @xmath472 is the degree of @xmath473 ) . moreover , if some of the markings are artificial ( i.e. not singular points of the induced metric ) then the growth rates after unmark " any subset of these points will be smaller than the above one . thus @xmath474 remark : similarly one can write down the corresponding estimate for maximal periodic cylinders . + if one includes ratners theorem as done in @xcite then the @xmath475 in equation ( [ verh ] ) can be replaced by a limit . + the statement is a direct consequence of the fact that each saddle connection on the torus has exactly @xmath472 preimages on @xmath476 and theorem [ tlimes ] . the second part is clear , because by removing nonsingular markings , the directions in which there are saddle connections are not changed . but in any such direction the number of saddle connections could decrease and their length increase . remark : with the results of veech @xcite every translation surface can be approximated by torus coverings in spaces of abelian differentials @xmath477 . moreover torus coverings are equally distributed with respect to the liouville measure @xmath4 mentioned in the introduction . since periodic cylinders and saddle connections are locally stable with respect to deformations in this spaces we have some sort of control on their numbers by the corollary . but because of the increasing number of artifical markings while approximating a general translation surface or equivalently the increasing degree of the covering , the estimate above is to weak to predict for example quadratic bounds of the growth rates of po s or sc s in general . on the other hand the equidistribution of torus coverings in @xmath477 is used to compute the @xmath4 volume of these spaces of abelian differentials see @xcite . the knowledge of these volumes in turn is the main step to evaluate siegel veech formulas , compare @xcite . eskin , alex , counting problems and semisimple groups , doc . j. dmv , extra volume icm ii ( 1998 ) 539 - 552 . eskin , alex ; masur , howard . pointwise asymptotic formulas on flat surfaces , ergodic theory and dynamical systems ( 2001 ) 37 pages . eskin , alex ; masur , howard ; schmoll , martin . billiards with barriers , preprint ( 2001 ) 37 pages . eskin , alex ; masur , howard ; zorich , anton . the siegel veech constants , in preparation . eskin , alex ; okounkov , andrei . asymptotics of numbers of branched coverings of a torus and volumes of moduli spaces of holomorphic differentials . preprint ( 2000 ) 42 pages . to appear in inventiones math . . gutkin , e. ; judge , c. ; the geometry and arithmetic of translation surfaces with applications to polygonal billiards , math . lett.3 ( 1996 ) , 391 - 403 . gutkin , e. , judge , c. ; affine mappings of translation surfaces , geometry and arithmetic , duke journal ( 1999 ) . lang , serge ; algebra . second edition . addison - wesley publishing co. , reading , mass . , xv+714 pp . isbn : 0 - 201 - 05487 - 6 lang , serge ; introduction to modular forms , grundlehren der mathematischen wissenschaften 222 , springer verlag 1976 . masur , h. ; closed trajectories for quadratic differentials with an application to billiards , duke math . j. 53 ( 1986 ) 307 - 314 . masur , h. ; lower bounds for the number of saddle connections and closed trajectories of a quadratic differential , in holomorphic functions and moduli 1 , 215 - 228 , msri publ . 10 , springer verlag 1988 . masur , h. ; the growth rate of trajectories of a quadratic differential , ergodic theory and dynamical systems 10 ( 1990 ) , 151 - 176 . serre , j.p . ; cours darithmtique , le mathmaticien , presses universitaires de france,(1970 ) . tenenbaum , g. ; introduction la thorie analytique et probabiliste des nombres , cours spcialiss , socit mathmatique de france,(1995 ) . veech , w. ; the teichmller geodesic flow , annals of mathematics , 64 ( 1986),441 - 530 . veech , w. ; teichmller curves in moduli space , eisenstein series , and an application to triangular billiards , invent . 97 ( 1989 ) , 553 - 583 . veech , w. ; moduli spaces of quadratic differentials , journal danalyse mathmatique vol.55 ( 1990),341 - 379 . veech , w. ; the billiard in the regular polygon , geom . and funct . analysis 2 ( 1992),341 - 379 . veech , w. ; siegel measures . ann . of math . ( 2 ) 148 ( 1998 ) , no . 3 , 895 - 944 . vorobets , y. ; planar structures and billiards in rational polygons , russ . surveys 51 ( 1996 ) , 177 - 178 . vorobets , y. ; planar structures and billiards in rational polygons : the veech alternative , russ . surveys 51 ( 1996 ) , 779 - 817 . zemlyakov , a.n . , katok , a.b . ; topological transitivity of billiards in polygons . zametki 18(2)(1975 ) pp . 291 - 300 . english translation in math . notes 18(2 ) ( 1976 ) pp 760 - 764 .
in a previous paper @xcite , we presented a method for reducing a general quadratic theory of gravitation to a canonical second - order form . the quadratic action provides an example of a higher - derivative theory , where the gravitational equations of motion are higher than second order . this means that the theory contains more degrees of freedom than just the simple massless graviton . by rewriting the action , we showed that quadratic gravitation is classically equivalent to einstein s gravity coupled to a massive real scalar field and a massive symmetric tensor field describing a spin - two field , with a specific lagrangian . we found that , even for simple quadratic gravity , the reduced action gave highly non - trivial potential energy and kinetic coupling terms . this suggests that higher - derivative theories have a great deal of structure , not immediately apparent from a simple linear analysis where one expands around flat space . one particular feature , which is the main thesis of this paper , is that the transformed theory may have a complicated vacuum structure . here , by a vacuum solution , we mean a stable solution of the second - order theory . in this paper , we will only consider vacua in which the auxiliary fields are covariantly constant . it is important to note that only once the vacuum of the transformed theory has been identified can the nature of the elementary field excitations be discussed . thus , for instance , the exact mass and couplings of the excitations , as well as which excitations are potentially ghost - like , may be very different around different vacuum states . in general , they will bear little relation to the excitations of the linearized analysis . the vacuum structure of the quadratic theories remained comparatively simple . the only stable vacuum was flat space with zero vacuum expectation value for both of the auxiliary fields . the elementary excitations were a massive scalar field and a massive ghost - like spin - two field . we would like to investigate how other higher - derivative theories can introduce a more interesting vacuum structure . a natural generalization is to consider actions which are not quadratic but general functions of the curvature tensors . since these tensors involve only second derivatives of the metric , the corresponding field equations can be at most fourth - order ; that is , the same order as the quadratic actions . we can start by asking how many new degrees of freedom we expect in such theories . to give a rough count , we turn to the cauchy problem . first recall that even higher - derivative gravity theories remain diffeomorphism invariant . thus we are always able to transform away four components of the symmetric metric , leaving six free components . since the field equations are fourth - order , if the cauchy problem can be solved we expect to be required to give four initial conditions for each independent component of the metric , namely the component itself and the first three time derivatives . this would imply that we have at most twelve degrees of freedom . rewriting the theory in a second - order form , since the auxiliary fields are set equal to terms involving second derivatives of the metric , we expect that giving the initial conditions of the auxiliary fields is equivalent to fixing second and third derivatives of the metric , a total of twelve conditions . thus we find that the auxiliary fields should carry six degrees of freedom , just as in the general quadratic theory . this leaves a possible six further degrees of freedom in the metric . however , since the metric in the second - order theory obeys einstein gravity , the number of degrees of freedom is reduced to the two helicity states of the usual massless graviton . by this rough argument , we expect an action in the form of any general function of the curvature tensors to describe , at most , the propagation of a massless graviton plus an additional six degrees of freedom ; that is , the same as for the special case of the general quadratic theory discussed in our previous paper @xcite . be this as it may , the structure of such theories is potentially much richer than in the quadratic case . as we will see , the vacuum structure is now non - trivial . a second generalization is to consider gravity actions which have higher derivatives acting on the curvature tensors . since these theories involve more than two derivatives on the metric , the corresponding field equations will be generically higher than fourth - order . consequently , we now expect new degrees of freedom in addition to those of the general fourth - order theory . again , we will show that the vacuum structure of such theories is , in general , non - trivial . some discussion of reducing both types of generalized theories to a second - order form already exists in the literature . that the equations of motion following from actions given by a general function @xmath1 are equivalent to the equations of motion of a scalar field dilatonically coupled to gravity , was first shown by teyssandier and tourrenc @xcite . the equivalence at the level of the action was given by magnano _ et al . _ @xcite , who also rewrote the reduce theory in canonical form . this latter group and jakubiec and kijowski @xcite also rewrote actions given by a general function @xmath2 in second - order form , though without the canonical separation of the new degrees of freedom . actions including derivatives of the scalar curvature were considered in the context of inflation by gottlber _ et al . _ @xcite , who again showed , in some special cases , the equivalence of the field equations to those describing scalar fields coupled to gravity . still at the level of the field equations , this work was extended to a general class of actions by schmidt @xcite and wands @xcite . possible vacua of the @xmath1 theory are discussed , in terms of the original higher - derivative field equations , by barrow and ottewill @xcite , and later in terms of the second - order field equations by barrow and cotsakis @xcite . this paper is organized as follows . in the next three section we will consider actions which are general functions first of the curvature scalar only , then of the ricci tensor , and finally of the full riemann tensor . rewriting these theories in a canonical second - order form , we will find a rich vacuum structure . this will require a careful discussion of the different `` branches '' of the theory , a concept we define below . we will show that the vacuum states are in a one - to - one correspondence with the stable , constant curvature , desitter or anti - desitter solutions of the higher - derivative theory . furthermore , we will show that if the vacuum of the transformed theory has zero cosmological constant , then the corresponding spacetime for the original higher - derivative theory is also flat . we will then discuss the elementary excitations around a given vacuum and argue that they remain a graviton , a scalar field and a ghost - like spin - two field , with masses and couplings depending on the particular vacuum and the functional form of the original action . in section [ fboxr ] we will consider a class of gravity actions that are general functions of the curvature scalar @xmath3 and the derivatives @xmath0 , with @xmath4 a positive integer . we show how to rewrite such theories in second - order form and show that they have , in general , a non - trivial vacuum structure . in particular , we present an example with two new scalar degrees of freedom , neither of which is ghost - like , coupled to einstein gravity with a stable anti - desitter space as its vacuum . we briefly present our conclusions in section [ conclusion ] . throughout the paper our conventions are to use a metric of signature @xmath5 and define the ricci tensor as @xmath6 . as in the case of quadratic gravity , the higher - derivative theory described by actions which are general functions of the curvature scalar is classically completely equivalent to the canonical second - order theory of a scalar field coupled to gravity . this equivalence was first shown at the level of the action by magnano _ et al . _ we shall derive this equivalent theory in a slightly different form , by first introducing an auxiliary field to reduce the action to second order and then making a suitable conformal transformation . we will point out the need to consider different branches of the theory when making the reduction . we introduce the auxiliary field in two steps . first we write @xmath7 } , \label{f(r)s}\end{aligned}\ ] ] where @xmath8 . the auxiliary field @xmath9 has the equation of motion @xmath10 provided we have @xmath11 , this gives @xmath12 . substituting back into the action we return to the original higher - derivative form . thus , the reduced action is equivalent to the original theory , but only away from the critical points defined by @xmath13 . a continuous region of @xmath9 between critical points , where @xmath14 never vanishes , will be called a branch . typically , there will be several branches in the theory . since @xmath9 gets set equal to @xmath3 , in terms of the original theory , the condition for a critical point can be written @xmath15 . thus , branches in @xmath9 correspond to branches in @xmath3 in the original higher - derivative theory . as a concrete example , let us assume that @xmath16 where @xmath17 . solving @xmath13 implies that there is a single critical point at @xmath18 . therefore , the second - order formalism in terms of the auxiliary field @xmath9 is valid in each of the two branches @xmath19 and @xmath20 , but breaks down at the critical point @xmath18 . since @xmath9 gets set equal to @xmath3 , these regions correspond to branches in the space of curvature given by @xmath21 and @xmath22 with a critical point at @xmath23 . the second step , which will then allow us to rewrite the action in canonical form , is to change variables to a new field @xmath24 . in terms of this new variable , the action becomes @xmath25 } .\ ] ] clearly this action is well defined only for those regions where the @xmath9 action is valid ; that is , away from the critical points . therefore , the @xmath26 formulation is only defined over ranges of @xmath9 for which @xmath11 . note that to define the action we must be able to invert the expression @xmath27 to give @xmath28 . locally this requires the same non - degenerate condition @xmath11 . globally , there may still be many different roots when we solve for @xmath9 in terms of @xmath26 . however , in any given branch of @xmath9 there is only a single root . thus there is a valid formulation of the theory in terms of @xmath26 for each branch of @xmath9 . it is important to note that , in each branch , the inverted function @xmath29 which enters is different . as a concrete example , consider once again @xmath1 defined in . it follows that @xmath30 note that @xmath31 . this expression can be inverted to give @xmath32 clearly one must take the @xmath33 root in the @xmath20 branch and the @xmath34 root in the @xmath19 branch . this situation is quite generic , as we will see below . for a particular reduced theory , corresponding to a given branch of the original higher - derivative theory , we can define @xmath35 and perform the conformal transformation @xmath36 . action then becomes , @xmath37 } .\ ] ] as promised , we find a theory of einstein gravity coupled to a scalar field with a particular potential dependent on the choice of the original function @xmath1 . as in the quadratic case there is one subtlety in defining @xmath38 . to keep @xmath38 real , we must have @xmath39 . when @xmath40 we must introduce , instead , the field @xmath41 , the effect of which is to change the sign of the overall normalization of the action as compared with the form given in above . further , at the special point @xmath42 the action can not be put in canonical form . in this paper we will restrict our attention to the @xmath43 case only . as a concrete example , consider @xmath1 defined in . as pointed out above , in this case @xmath31 in either branch of @xmath9 . therefore , @xmath35 is always well defined . we point out , however , that the range of @xmath38 is restricted to @xmath44 . note that the scalar field kinetic energy has the usual sign and , hence , @xmath38 is not a ghost . we can now look for the vacua in any given branch of the theory . the vacua of the theory described by action are defined to be stable solutions of the @xmath45 and @xmath38 equations of motion given by @xmath46 - \tfrac32{\overline g}_{\mu\nu}v(\chi ) \\ \label{canr2chieom } { \overline \nabla}^2\chi = \frac{dv}{d\chi } \end{gathered}\ ] ] respectively , where the potential is given by @xmath47 where we recall that the form of @xmath48 depends on the branch in question . as stated previously , we will only consider vacua satisfying the covariant constant condition @xmath49 it follows that @xmath38 is a constant and , from , that it must extremize the potential . furthermore , it follows from that the vacuum is a space of constant curvature with @xmath50 since @xmath51 and @xmath38 is constant , these vacua also correspond to spaces of constant curvature with respect to the original metric , with @xmath52 . since the vacuum field @xmath38 must be an extremum of the potential , we have the condition @xmath53 we started with a completely general function @xmath54 , so this condition may generically have multiple solutions , including solutions away from @xmath55 . further , given a particular solution for @xmath38 , the value of the potential at this point need not be zero ; that is , the vacuum generally has non - zero cosmological constant . in this case , the curvature scalars @xmath3 and @xmath56 will be non - vanishing . as a concrete example of all this , we return to the example specified in . as discussed previously , there are two branches where the second order theory is defined ; branch a where @xmath20 and branch b where @xmath19 . the field @xmath57 , which is given as a function of @xmath9 in , satisfies @xmath31 everywhere and , hence , @xmath58 is well defined in both branches , although its range in restricted to @xmath44 . expression was inverted to give @xmath9 as a function of @xmath57 in . this expression is branch dependent , being given by @xmath59 with the positive root chosen in branch a and the negative root chosen in branch b. it follows that the form of the potential energy defined in also depends on the branch chosen . it is given by @xmath60 for branch a and b respectively . the two potentials are plotted in figures [ fig : r3vp ] and [ fig : r3vm ] . we see that each has a stationary point . however , since vacua must be minima of the potential , it follows that only branch b has a stable vacuum . this is located at @xmath61 , which is non - zero . note that @xmath62 , so that the vacuum state has a non - vanishing negative cosmological constant . it follows that @xmath63 and @xmath64 , which implies that the vacuum state is an anti - de sitter space both in terms of the metric @xmath45 and the metric @xmath65 . thus we have given an example of a higher - derivative theory which has a new vacuum state away from the flat - space solution of ordinary einstein gravity . it is important to note that none of this structure would have been evident if we had made a simple expansion of the original action around flat space in terms of @xmath66 . in fact , keeping only the first non - trivial , quadratic terms in @xmath67 , we would not even have been aware that there was an additional scalar degree of freedom in the theory . the lowest order term in @xmath68 is cubic in @xmath67 , and so in a quadratic expansion we would only see the usual einstein term . the conclusion is that generic higher - derivative corrections to pure einstein gravity introduce completely new vacuum states into the theory . flat space is no longer a unique point in field space , so that if , for instance , we wish to investigate the fundamental excitations in the theory , we must start by specifying which vacuum state we are considering . at this point , we would like to present another specific example that further illustrates the preceding discussion . consider the cubic function @xmath69 with @xmath70 . we now have @xmath71 so that the theory has a single critical point at @xmath72 . it follows that the second - order theory is defined on two branches ; branch a where @xmath73 and branch b where @xmath74 . on either branch we can define a new field @xmath24 , which in this case is given by @xmath75 note that , since @xmath76 , @xmath57 is always a positive real number in the range @xmath77 for both branches . expression can be inverted to give @xmath78 clearly the @xmath79 solution is correct on branch a whereas the @xmath80 solution is to be used in branch b. since @xmath57 is always positive , the definition of @xmath81 is valid in both branches . note that @xmath38 is then restricted to lie in the range @xmath82 . since the expression for @xmath9 in terms of @xmath57 , and hence @xmath38 , given in is different in each branch , it follows from that the potential energy function also is different in each branch . we have @xmath83 where @xmath84 and the @xmath85 expression is valid in branch a and the @xmath86 expression valid in branch b. the two potentials are plotted in figures [ fig : r3genvp ] and [ fig : r3genvm ] . unlike the previous example , in this case each branch contains a single , stable minimum . in branch a , this minimum is located at @xmath55 and has vanishing cosmological constant @xmath87 . it follows that , for this vacuum , @xmath88 and spacetime is flat with respect to both the @xmath45 and @xmath65 metrics . in branch b , the minimum is located at @xmath89 with a non - vanishing negative cosmological constant given by @xmath90 . for this vacuum @xmath91 and , hence , @xmath64 . it follows that the spacetime is an anti de - sitter space with respect to both metrics @xmath45 and @xmath65 . thus we have found an example which , aside from a conventional minimum at @xmath23 with zero cosmological constant , has an additional minimum with negative cosmological constant . furthermore , there is an unstable maximum with positive cosmological constant . the mass of the scalar field is different at each of the different minima . here a quadratic expansion of the action around flat - space would have identified a new scalar degree of freedom , but would never have revealed the presence of a second stable vacuum state . it will not have escaped the readers notice that finding the vacua in the second - order formalism requires a careful discussion of the branches in the @xmath9 variable . the branching structure was not too difficult in the preceding examples , but as we will show below , it can be , and generally is , extremely complicated . however , the following remarks will allow us to define a simpler procedure for determining the vacua . recall that , for a covariantly constant scalar @xmath38 satisfying @xmath92 , the corresponding spacetime structure in terms of the metric @xmath65 is that of a space of constant curvature with @xmath93 . provided we are not at a degenerate point , spaces of constant curvature @xmath3 and vacuum solutions are in one - to - one correspondence , since the relation @xmath93 not only means that constant @xmath38 implies constant @xmath3 , but also the reverse . note that , as a corollary , any vacuum solution with zero cosmological constant , namely @xmath94 , must correspond to flat space in the original higher - derivative theory . an important consequence of this result is that , since flat space is a single point in the field space of the original theory , there can be only one vacuum state with zero cosmological constant in the second - order theory . we conclude that all vacuum solutions of the above type can be found as constant curvature solutions of the equations of motion for the higher - derivative @xmath65 theory . all such vacua can be found directly , without reference to any branch structure . having found a vacuum of interest , one can then proceed in reverse , introducing the second - order theory in the appropriate branch containing that vacuum . this is often an easier procedure . the @xmath65 equations derived from the higher - derivative action are @xmath95 if we look for solutions of constant curvature , then @xmath96 where @xmath3 is a constant . it follows that the derivative terms in the equations of motion drop out and we are left with the simple condition , first derived by barrow and ottewill @xcite , @xmath97 we see that this is exactly the condition we obtained for stationary points of the potential aside from a factor of @xmath98 . in the latter case , however , the expression was taken to be a function of the scalar field @xmath38 expressed in terms of @xmath9 , and as such only valid in a branch - by - branch sense , whereas here the expression is valid for all curvature @xmath3 . this now provides us with a procedure for finding all the vacua of the theory as well as the nature of the excitations around a given vacuum . we start by looking for constant curvature solutions of the original higher - derivative equations of motion , solving the equation , which is valid globally . having identified these points we can then , locally around each solution , make a transformation to the reduced theory , choosing , if necessary , the appropriate branch of @xmath48 . in the transformed frame , the new scalar degree of freedom is made explicit . we can then address questions of local stability and the mass of the scalar field . the only breakdown of this procedure occurs when the minimum is at a degenerate point of the transformation ; that is , a point where @xmath15 . in this case , any discussion of the particle spectrum and local stability must be in terms of the original theory . in general , we shall not consider such degenerate points . as an example of this procedure , consider the function @xmath99.\ ] ] where @xmath100 is a constant . the condition for a degenerate point is then @xmath101 this equation has solutions @xmath102 where @xmath103 is any integer . the theory thus has an infinite number of branches given by @xmath104 . rather than treating each branch separately , we follow our procedure for finding the vacuum states of the theory by solving the condition for the constant curvature solutions of the original theory . the condition reads @xmath105 = 0.\ ] ] the left - hand side of this expression is plotted in figure [ fig : constrex ] . we have the solutions @xmath106 where @xmath103 is an integer and @xmath107 are the solutions of @xmath108 explicitly , the first few solutions of near @xmath109 are @xmath110 furthermore @xmath111 and as @xmath112 becomes large the solutions approach @xmath113 . the first observation to make is that the constant curvature solutions at @xmath114 , including the point @xmath23 , all correspond to degenerate points . as such , a second - order form of the theory does not exist at these points . thus , although these are constant curvature solutions , we can not identify them as vacuum solutions of a canonical second - order theory . however the solutions at @xmath115 are not at degenerate points . from our previous discussion this implies that they must correspond to vacuum solutions of the canonical second - order theory . nonetheless , we find that each vacuum solution lies in a different branch of the theory . let us consider two particular solutions , @xmath116 and @xmath117 , in the second - order formulation . for the first solution we must take the branch @xmath118 . following our usual procedure for writing the theory in a second - order form , we first introduce the auxiliary field @xmath9 and then define @xmath119.\ ] ] however , as usual , we are required to invert this expression to give @xmath9 as a function of @xmath26 . it follows that @xmath120 we note that @xmath26 is only defined in the range @xmath121 . to put the action in canonical second - order form we define @xmath81 . the potential for @xmath38 , is then given by @xmath122\ ] ] where , @xmath123 we now have that @xmath38 is restricted to the range @xmath124 . the potential is plotted in figure [ fig : sinrvp ] . we find that there is a single maximum of the potential at @xmath125 . the value of the potential at this point is @xmath126 , so that @xmath127 and @xmath128 . thus we see that the maximum corresponds to the flat space solution @xmath129 as expected , and we can conclude that this vacuum is unstable . for the solution at @xmath130 , we must take the branch @xmath131 . the expression for the potential is the same as the expression above , except that we must now define @xmath132 the resulting potential is plotted in figure [ fig : sinrvm ] . again @xmath38 is restricted to lie in the range @xmath124 . we now find a stable minimum at @xmath125 . this gives @xmath133 , @xmath134 and @xmath135 , showing that the minimum of the potential does indeed correspond to the constant curvature solution @xmath130 . we can concluded that this is a stable vacuum state . in fact , further analysis shows that all the vacua @xmath115 with @xmath136 are unstable maxima , while those with @xmath137 are stable minima . this final example demonstrates that the vacuum structure of higher - derivative theories can be complicated . we find an infinite number of stable and unstable vacua , each in a different branch of the theory . in summary , we have shown that theories which are general functions of the scalar curvature @xmath3 , can be reduced to a canonical second - order form describing einstein gravity coupled to a scalar field with a potential which may have a vacuum structure of arbitrary complexity . further these vacua all correspond to constant curvature solutions of the original theory . the advantage of discussing the theory in the second - order formalism is that it is in a canonical form where we can easily identify the excitations around the vacuum . the disadvantage is that the problem of inverting @xmath138 means that we are forced to break the theory into regions separated by degenerate points . for this reason it is often easier to identify vacua as constant curvature solutions of the higher - derivative theory and then make a local transformation in the region of each solution to investigate the excitations around that vacuum . all of the non - trivial vacuum structure arises from non - linear terms in the field equations and so is missed in a quadratic expansion of the higher - derivative theory in terms of the fluctuation of the metric around flat space . further , any analysis of the particle content of the theory , in this case the presence of a new scalar degree of freedom and identification of its mass , can only be made once the vacuum is identified . in this and the following section we shall extend our discussion to include actions which are general functions of first the ricci tensor and then the riemann tensor . we shall use the same techniques as in the previous section , namely introducing auxiliary fields , showing that vacuum states correspond to constant curvature solutions of the higher derivative theory , and then investigating the excitations around a given vacuum . the reduction to a second - order form was first discussed by magnano _ et al . _ @xcite and jakubiec and kijowski @xcite . here , we shall use a slightly different analysis , again pointing out the need for identifying different branches of the theory and also giving a canonical separation of the new variables about any given vacuum . we shall only give the general formulation without specific examples . again we will find a rich structure of vacua which would be completely missed in a linearized analysis . this will re - emphasize the need to identify the vacuum in question before investigating the masses and couplings of the elementary excitations of the theory . starting with an action @xmath2 we can introduce an auxiliary field and put the theory in a second - order form @xmath139 } \notag \\ & = \frac{1}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ \pi^{\mu\nu } \left(r_{\mu\nu } - x_{\mu\nu}(\pi_{{\rho}{\sigma}})\right ) + f(x_{\mu\nu}(\pi_{{\rho}{\sigma } } ) ) \right ] } .\end{aligned}\ ] ] again we introduce the auxiliary field in two steps , first writing the action in terms of the field @xmath140 which gets set to @xmath141 on solving its equation of motion . we then define @xmath142 and invert the expression to give @xmath140 as a function of @xmath143 . note again that the introduction of both @xmath140 and @xmath143 requires the non - degeneracy condition @xmath144 if the auxiliary field is to be properly eliminated to return to the original action . furthermore , in inverting , it may be necessary to divide the theory into branches corresponding to different possible roots for @xmath140 . thus for the variable @xmath143 it may be necessary to introduce a collection of auxiliary variable theories each valid for a different branch of @xmath141 . we can ask how many degrees of freedom are there in the auxiliary field ? we have argued that we expect six . if we consider the @xmath140 formulation for now , we see first that we can again derive a spin - two divergence condition . since the auxiliary field equation of motion gives @xmath145 , we clearly can show , using the bianchi identity @xmath146 , that @xmath147 . constraining the components of @xmath140 , these four conditions imply that we do indeed have six new propagating degrees of freedom . we would like to be able to separate these into scalar and spin - two degrees of freedom as was done in the quadratic case in @xcite . for a general function @xmath54 , this is not possible since we are not able to obtain the necessary trace condition from the @xmath65 equation of motion . however , as we will show below , having identified a vacuum state , we can always make the separation locally in an expansion around the vacuum solution . what then are the vacua of the @xmath2 theory ? since we are unable to separate the spin - two and scalar degrees of freedom in the auxiliary field , we can not put the theory in a canonical form and look to minimize the potential as we did for the @xmath1 theory . however , we recall that the vacuum states in question are none other than states where the auxiliary field is covariantly constant . this means , since the @xmath140 equation of motion gives @xmath145 that they are states of constant tensor curvature . we can also impose the condition that the states should have no preferred direction , so that @xmath140 is proportional to the metric @xmath65 . thus vacuum states are states of constant scalar curvature @xmath148 with @xmath3 constant . consequently , one approach to finding vacuum solutions is simply to look for constant curvature solutions of the original theory . the equations of motion of the original higher derivative theory are given by @xmath149 to incorporate the presence of the metric in @xmath2 in deriving the equations of motion , we consider @xmath54 as a function of the mixed index object @xmath150 , with all contractions made between raised and lowered indices so that the metric does not enter explicitly . then in the above expression @xmath151 , so that , for instance , @xmath152 . looking for constant curvature solutions of the form @xmath96 , we obtain the condition @xmath153 where @xmath154 and @xmath155 , and we have used the fact that @xmath156 evaluated at the constant curvature solution must be proportional to @xmath157 . note that this condition has exactly the same form as the condition we obtained for actions which were general functions of the scalar curvature . given a particular vacuum solution satisfying the condition , we would like to investigate the excitations around the vacuum state . the natural way to do this is to consider an expansion of the action around a given vacuum @xmath158 . expanding @xmath2 and keeping terms to quadratic order in the curvature only , we have @xmath159 } \notag \\ & = \frac{\beta}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ - \tfrac12 r_0 + r + \frac{1}{6{m_0}^2}r^2 - \frac{1}{{m_2}^2}\left(r_{\mu\nu}r^{\mu\nu } -\tfrac13r^2\right ) \right ] } , \label{expandf}\end{aligned}\ ] ] here we have evaluated @xmath54 and its derivative at the vacuum @xmath160 and defined @xmath161 using the general symmetric decomposition of the second two expressions . the variables @xmath162 , @xmath163 and @xmath164 are then given by @xmath165 where in writing the final line of we have used the constant curvature condition @xmath166 . we see that we have succeeded in putting the action in a quadratic form , though with the addition of a cosmological constant . the final line of is identical to the general quadratic form we discussed in our previous paper @xcite , except for the addition of a cosmological constant term and a renormalization of the gravitational coupling constant by a factor @xmath162 . we can thus introduce auxiliary fields @xmath38 and @xmath167 exactly as we did in @xcite to give @xmath168 . \end{gathered}\ ] ] the only effect of the cosmological term is to modify the potential for the scalar field , adding a term @xmath169 . we find that the extremum of the potential is now at @xmath170 , which , relating the auxiliary field @xmath38 back to the original curvature , gives @xmath158 , as required for a consistent expansion . following exactly the analysis of our previous paper , the auxiliary field @xmath167 satisfies generalized divergence and trace conditions , so it does indeed describe the degrees of freedom of a spin - two field . by making a final field redefinition , we can write the action in canonical form with explicit kinetic energy terms for the spin - two field . again following the discussion in our previous paper , the spin - two field will have the correct pauli - fierz limit , but will unfortunately be ghost - like . thus we have shown that locally , around any vacuum of the higher - derivative theory ( that is a solution of constant curvature ) , we can expand the theory to identify a new scalar and a new spin - two degree of freedom ( provided we are not at a degenerate point ) . the spin - two field satisfies divergence and trace conditions as before but importantly , we see that it remains ghost - like . thus generalizing to @xmath2 actions fails to remove the problem of the ghost spin - two degree of freedom . the masses of the degrees of freedom are fixed by the form of the function @xmath54 around the constant curvature solution . it is worth noting that we could equally well have done this analysis in the second - order form , looking for solutions with constant @xmath140 proportional to @xmath65 . then expanding in small @xmath140 about such solutions , gives a linear coupling between @xmath140 and @xmath141 and quadratic ` mass ' terms for @xmath140 . from this form we could then extract the scalar and spin - two parts of @xmath140 . in this sense , the auxiliary field @xmath140 always carries six degrees of freedom , which about any given vacuum can be decomposed into a scalar field and a ghost spin - two field , with the form of the decomposition changing as we go from vacuum to vacuum . our last generalization is to consider actions which are general functions of the riemann tensor @xmath171 . as mentioned earlier , the suggestion is that such theories have an additional six degrees of freedom , which we know in the quadratic case can be decomposed into scalar and spin - two fields . we start the discussion by demonstrating that , as in all previous cases , we can introduce an auxiliary field to write the action in a second order form . @xmath172 } \notag \\ & = \frac{1}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ \rho^{{\lambda}\mu\nu{\rho } } \left(r_{{\lambda}\mu\nu{\rho } } - x_{{\lambda}\mu\nu{\rho}}(\rho_{\alpha\beta\gamma\delta})\right ) + f(x_{{\lambda}\mu\nu{\rho}}(\rho_{\alpha\beta\gamma\delta } ) ) \right ] } .\end{aligned}\ ] ] again we introduce the auxiliary field in two stages ; first introducing @xmath173 , which gets set equal to the riemann tensor on solving its equation of motion , and then defining @xmath174 in both cases we require the non - degeneracy condition @xmath175 in order to be able to eliminate the auxiliary field and return to the original action . when using the variable @xmath176 we may be required to break the theory into branches , introducing a collection of second - order theories , each taking a different branch when inverting to find @xmath177 in terms of @xmath176 . turning to the vacuum states , we again find that states with constant @xmath177 have constant riemann curvature , since @xmath177 is set equal to @xmath178 by its equation of motion . furthermore , imposing the condition that there is no preferred direction in spacetime we require that @xmath177 is proportional to @xmath65 . therefore , given the symmetries of @xmath177 , we have @xmath179 and , hence , we are considering solutions of constant ricci scalar curvature . as before the easiest way to obtain such solutions is from the equations of motion of the original higher - derivative action . again , to circumvent the problem , when deriving the equation of motion , of the metric explicitly entering the function @xmath54 , we consider @xmath54 as a function of @xmath180 with all contractions made between raised and lowered indices . we then derive the equations of motion @xmath181 where @xmath182 . restricting to constant curvature solutions of the form , we get the familiar condition @xmath183 where now @xmath184 and @xmath185 , and we have used the fact that , by symmetry , @xmath186 is proportional to @xmath187 . to investigate the excitations around a given vacuum we expanding the action about the constant curvature solution . we have , keeping terms up to quadratic order in the curvature only , @xmath188 \right ) \left ( r^{{\lambda}\mu\nu{\rho } } - \tfrac{1}{12}r_0 \left[g^{{\lambda}\nu}g^{\mu{\rho}}-g^{{\lambda}{\rho}}g^{\mu\nu}\right ] \right ) \right ] \notag \\ & = \frac{\beta}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ - \tfrac12 r_0 + r + \frac{1}{6{m_0}^2}r^2 - \frac{1}{{m_2}^2}\left(r_{\mu\nu}r^{\mu\nu } -\tfrac13r^2\right ) \right . } \notag \\ & \qquad\qquad \left . + \gamma\left(r_{{\lambda}\mu\nu{\rho}}r^{{\lambda}\mu\nu{\rho } } -4r_{\mu\nu}r^{\mu\nu}+r^2\right ) \right ] . \label{riemannexpand}\end{aligned}\ ] ] here we evaluate @xmath54 and its derivatives at the constant curvature solution , defining , for the contraction of the derivatives with a tensor @xmath189 which has the symmetries of the riemann tensor but in otherwise arbitrary , @xmath190 with @xmath191 and @xmath192 . we have also introduced the parameters @xmath193 and have used the constant curvature condition @xmath166 in the final line of . the final expression in is a quadratic action with a gauss - bonnet term , a weyl - squared term and a ricci - scalar - squared term , together with a cosmological constant . the gauss - bonnet term can be dropped classically as a total divergence , leaving the action in the same quadratic form as discussed in our previous paper @xcite . reducing the action to a second - order form then follows exactly as in the case for @xmath2 actions , so that we obtain the same transformed action . again we can verify that the transformed action has a vacuum solution at @xmath158 as is required for the consistency of our expansion . we also note that here too the expansion could have been made in terms of the variable @xmath173 , which could then be decomposed into its scalar and spin - two parts , the form of the decomposition depending on which vacuum is being considered . in conclusion , general actions of the form @xmath194 may have a variety of vacuum solutions , generically not apparent in a linear analysis . around any vacuum the new degrees of freedom in the theory , aside from the massless graviton , can always be separated into a scalar field and a spin - two field . unfortunately , the spin - two field is always ghost - like . in this section , we shall briefly discuss how to extend our analysis to actions with higher - order equations of motion . we have seen that an action involving any function of the curvature tensors gives at most fourth - order equations of motion . for higher - order equations we need to consider actions which include some derivatives of the curvature . here , we shall be concerned with only the simplest form , @xmath195 where we require that the function @xmath54 has been reduced , by integration by parts and dropping total derivatives , so as to minimize @xmath4 . the equations of motion following from similar actions have been considered by schmidt @xcite and wands @xcite , who showed them to be equivalent to those for a set of scalar fields coupled to gravity . here will shall show the full equivalence in a new way , working at the level of the action . it will be important to distinguish between possible forms of the function @xmath54 . if we write @xmath196 , ( where in the action we have @xmath197 , @xmath198 and so on ) , we find there are two possible cases : either @xmath199 is a function of @xmath200 ( case one ) or it is not ( case two ) . if it is not then it must be a function of @xmath201 , since otherwise the original form was reducible ; that is , the action was not written in a form minimized with respect to @xmath4 . thus , in case two , we can always decompose @xmath54 as @xmath202 further , we see that for case one the equations of motion are @xmath203th - order , while for case two they are @xmath204th - order . we would like to reduce the action to a canonical second - order form , by introducing auxiliary fields . given the order of the higher - derivative equations of motion we see that in case one we must introduce @xmath205 new fields , while in case two we need only @xmath206 new fields . the procedure we shall use is essentially a generalization of ostrogradski s method for reducing a higher - order action to a first - order form @xcite , only that here we shall be reducing to a second - order form . ( the ostrogradski result is usual given as a hamiltonian , but this can always be rewritten as a helmhotz lagrangian , the analog of the form we shall use . ) we start by introducing a set of lagrange multipliers , so @xmath207.\end{gathered}\ ] ] clearly eliminating fields via the lagrange multiplier equations of motion , starting with @xmath208 and working down in @xmath4 returns one to the original action . however , we have introduced at least one too many new fields . it is clear that the @xmath200 equation of motion is purely algebraic and eliminating it will not introduce higher - derivatives ; that is , the action will remain second order . the @xmath209 equation of motion reads @xmath210 where we have distinguished between the two cases discussed above , and substituted the special form of @xmath54 in case two . in case one , to form the analog of ostrogradski s lagrangian , we solve the equation to give @xmath200 as a function of @xmath211 and @xmath208 , writing @xmath212 . it should be noted that in general the solution is not unique , and we must divide the original theory into pieces corresponding to different branches of the solution , just as in the case of actions of the form @xmath1 discussed in the previous section . in case two we simply substitute for @xmath208 and the special form of @xmath54 . we get in case one @xmath213 , \end{gathered}\ ] ] the exact analog of the ostrogradski - helmhotz lagrangian , while for case two we have a slightly different form , @xmath214 . \end{gathered}\ ] ] we see that , as expected , in case one we have a total of @xmath205 auxiliary fields , while in case two we have only @xmath206 new fields . all that remains is to transform the action into a canonical form , growing canonical kinetic energy terms for all the new auxiliary fields . let us concentrate on actions of case one . terms of the form @xmath215 are easy to deal with . simply introducing a pair of new fields , @xmath216 and @xmath217 by @xmath218 we have , in the action , @xmath219 } \notag \\ & = \int { d^4x\sqrt{-g } \left [ - \left(\nabla\chi_i\right)^2 + \left(\nabla\psi_i\right)^2 \right ] } . \end{aligned}\ ] ] we see that the @xmath216 field has a canonical kinetic term , but the @xmath217 field has the wrong sign ; it is a ghost . this is characteristic of higher- order theories . in reducing the theory to second - order the new fields always enter as a pair of a ghost - like field with a ordinary field . if we define a potential function , @xmath220 where it is understood that the right - hand side is evaluated at @xmath221 , @xmath222 , the action in case one becomes @xmath223 } .\ ] ] to complete the transformation to canonical form all that is left is to make a conformal rescaling of the metric to remove the @xmath224 coupling . as usual we define @xmath225 and rescale to a metric @xmath36 , giving @xmath226 .\end{gathered}\ ] ] we conclude that the original case one higher - derivative gravity theory is equivalent to canonical einstein gravity coupled to @xmath205 scalar fields , @xmath227 of which , @xmath38 and @xmath216 , propagate physically and @xmath4 of which , @xmath217 , are ghost - like . to put the case two action in canonical form is more complicated because of the @xmath228 term . however , in principle , it is always possible to introduce a set of new fields @xmath229 which simultaneously diagonalize the kinetic terms for the @xmath230 and @xmath231 , though now the form of the transformation will depend on the function @xmath232 . at least @xmath233 of the new fields will be ghost - like . we can then make a conformal rescaling as in case one to put the action in the same canonical form , though the potential function will have a different form . the conclusion is that there is a procedure for rewriting the higher - order action in a canonical second - order form . however , @xmath4 of the new fields in case one and at least @xmath233 of the fields in case two will be ghost - like . in each case we obtain a specific potential , and so we can again look for vacuum states as stationary points of the potential . generically , as in the case of @xmath1 actions discussed in the previous section , the non - trivial vacua do not correspond to the flat - space solution of the original higher - derivative theory . as an simple example of this procedure consider the function @xmath234 . this example is of the first case since writing @xmath235 with @xmath197 and @xmath198 , we have @xmath236 , which is not independent of @xmath237 . following our general procedure , the first step is to introduce a set of lagrange multipliers , @xmath238 } \notag \\ & = \frac{1}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ { \lambda}_1 + \alpha{{\lambda}_2}^2 + \mu \left(r-{\lambda}_1\right ) + \mu_1 \left(\nabla^2{\lambda}_1-{\lambda}_2\right ) \right ] } . \end{aligned}\ ] ] as discussed above we have introduced one too many auxiliary fields . we can eliminate @xmath237 by solving its equation of motion , which reads @xmath239 implying @xmath240 . substituting back into the action gives @xmath241 } .\ ] ] next , to put the kinetic energy for @xmath242 and @xmath243 in canonical form , we define @xmath244 and @xmath245 , so that @xmath246 } .\ ] ] finally we make the conformal rescaling @xmath36 with @xmath225 to put the action in canonical form @xmath247 . \end{gathered}\ ] ] thus we see that the original higher - derivative gravity theory is equivalent to canonical einstein gravity coupled to three scalar fields . one field is ghost - like , and the potential has a single , unstable stationary point at @xmath248 . as a second example consider @xmath249 writing @xmath197 and @xmath198 , we now have @xmath250 , independent of @xmath237 , so this example clearly falls under case two . repeating our procedure , we introduce lagrange multipliers to give @xmath251 } \notag \\ & = \frac{1}{2\kappa^2 } \int { d^4x \sqrt{-g } \left [ \alpha + \beta{\lambda}_1 + \gamma{{\lambda}_1}^2 + \epsilon{\lambda}_1{\lambda}_2 + \mu \left(r-{\lambda}_1\right ) + \mu_1 \left(\nabla^2{\lambda}_1-{\lambda}_2\right ) \right ] } . \end{aligned}\ ] ] again we have too many new fields . the @xmath237 equation of motion now reads @xmath252 which can not be solved for @xmath237 since the action is case two . however substituting this solution into the action eliminates both @xmath237 and @xmath243 , leaving the second - order form , @xmath253 } .\ ] ] the kinetic energy for @xmath242 is already in canonical form , but we must make a final conformal rescaling by @xmath254 to put the action in the completely canonical form @xmath255 } , \ ] ] where we have the potential , @xmath256 again the original higher - derivative gravity is shown to be equivalent to ordinary einstein gravity though now coupled to only two scalar fields . by choosing @xmath17 we can ensure that neither field is ghost - like . note that we argued above that case two theories must have at least @xmath233 ghost - like fields . it is thus only in this special case of @xmath257 that we are able to have all the scalar fields non - ghost - like . as before we obtain a specific potential for the fields . we find that @xmath258 has a single stationary point at @xmath259 , @xmath260 , provided @xmath261 . expanding around this point we find that , if @xmath262 and @xmath263 , we have a stable minimum . the value of the potential at the minimum is @xmath264 which is negative . we conclude that , for the given range of @xmath265 , @xmath162 and @xmath266 , the theory has a single stable vacuum state with negative cosmological constant . from einstein s equation we find that this state is an anti - desitter space with @xmath267 . using the fact that @xmath52 for covariantly constant @xmath38 , we find that @xmath268 . thus the vacuum state also corresponds to an anti - desitter space in the original higher - derivative theory . the most important conclusion of this paper is that higher - derivative theories of gravitation generically have multiple stable vacua . one of these may be trivial , corresponding to flat spacetime , but all the other vacua are non - trivial with the associated manifold being either desitter or anti - desitter spacetime with non - vanishing cosmological constant . while of interest from various points of view , such non - trivial vacua can not represent the universe as it is now since the radius of curvature of these solutions is of order the inverse planck mass . thus , one might conclude that non - trivial gravitational vacua are irrelevant for particle physics theories . however , this is not the case . we have recently shown that if we extend the methods and results of this paper to the realm of @xmath269 supergravity , then non - trivial vacua can exist with vanishing cosmological constant @xcite . furthermore , we find that supersymmetry is generically spontaneously broken in these vacuum states . it follows that higher - derivative @xmath269 supergravitation could play a pivotal role in high energy physics . this possibility is being pursued elsewhere @xcite . this work was supported in part by doe grant no . de - fg02 - 95er40893 and nato grand no . crg-940784 . a. hindawi , b. a. ovrut , and d. waldram , _ soft supersymmetry breaking induced by higher - derivative supergravitation in the electroweak standard model _ , 381b * ( 1996 ) , 154162 , hep - th/9602075 .
superclusters are the largest structures in the universe . their sizes span the range from a few times the radius of a typical galaxy cluster ( @xmath3 3 - 4 @xmath4 ) to the lengths of the `` great walls '' of galaxies observed by the cfa redshift survey @xcite and the sloan digital sky survey @xcite . these latter sizes of @xmath5 represent a significant fraction ( @xmath6% ) of the horizon scale . because they are so large relative to a typical survey volume , superclusters have proven difficult to study , especially in a statistical manner . observationally , attempts to characterize the large - scale structure using distributions of galaxies date back to @xcite , but these methods did not become powerful until the advent of galaxy cluster and galaxy redshift surveys . @xcite used the palomar sky survey to construct a catalog of rich clusters , with which he was able to find evidence for what he called `` second - order clusters , '' i.e. clusters of clusters , which are indeed superclusters . subsequent work included @xcite s study of the supercluster containing the rich coma cluster , and @xcite s work on the perseus supercluster . other observational approaches to supercluster identification have been based , like that of @xcite , on cluster catalogs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) or on analysis of smoothed density fields @xcite . these have been important in characterizing the low - redshift universe , but they have not yet allowed the study of supercluster evolution from early cosmic times ( i.e. , from redshifts of order unity ) . how do superclusters form , and how do their shapes , multiplicities , sizes , and structures evolve with time ? the answers to these fundamental questions are not yet known . recently , there has been exciting progress on both the observational and computational fronts . the sloan digital sky survey ( sdss , * ? ? ? * ; * ? ? ? * ) is collecting deep , broadband imaging and spectroscopy of a large , contiguous portion of the sky . sdss has created the largest galaxy redshift survey catalog to date , and much work is also underway to improve photometric redshift techniques ( e.g. * ? ? ? * ; * ? ? ? * ) , which would permit measurements of 3-d galaxy positions for the much larger sdss imaging catalog . on the theoretical side , a concordance cosmological model has gained widespread acceptance @xcite . applying this model to cosmological dynamics codes such as that described below , we can now reliably predict the properties and evolution of superclusters . the time is therefore ideal for a computational analysis of large - scale structure using new , powerful simulations , the results of which can be used to determine the properties and evolution of superclusters . in addition , these results can be compared with observations , thus providing a new test of the current cosmological model complementary to that provided by cmb , large - scale power spectra , and other studies as well as shed light on structure formation and evolution . we begin by describing , in 2 , the process by which clusters and in turn superclusters were identified in the simulation output . we provide quantitative descriptions of the multiplicity and size distributions of the identified superclusters in [ sec : mult ] and [ sec : size ] . we examine the radial structure of our superclusters in [ sec : radial ] and the dimensionality of superclusters in [ sec : shapes ] . the evolution of the above properties with redshift is presented in [ sec : evo ] . results for superclusters selected from only higher - mass clusters are presented in [ sec : him ] . our conclusions are summarized in [ sec : conc ] . we use the efficient , parallel dark matter simulation code of @xcite . a large - scale , high - resolution simulation was evolved with a particle - mesh method to compute long - range gravitational forces , and a tree code to treat high - density regions . we use the cosmological parameters of the concordance model : @xmath7 and @xmath8 ( cf . @xcite ; see also @xcite ) . a total of @xmath9 particles , each assigned the mass @xmath10 , is evolved in a periodic box @xmath11 on a side , and positions are saved at the appropriate times in order to construct a light cone with vertex at one corner of the box . clusters are then identified in two different , partially overlapping volumes : a `` low - redshift '' sphere of radius @xmath11 , corresponding to a maximum redshift of @xmath12 , and a `` high - redshift cone , '' which includes the low-@xmath13 sphere s positive octant ( @xmath14 ) and extends this octant out to @xmath15 , a distance of @xmath16 . using the cluster selection scheme described below , a total of 1,442,616 clusters with @xmath17 are found in the low-@xmath13 sphere , and 720,550 in the high-@xmath13 cone . further details on this light cone simulation are given by @xcite , hereafter hbb . clusters are identified in the simulation using the friends - of - friends ( fof ) algorithm ( see hbb ; * ? ? ? the fof results depend on the chosen linking length @xmath18 . a common parameterization uses the linkage parameter @xmath19 , where @xmath20 is the mean interparticle separation . this is useful in applications where the mean density of objects , and in turn their mean separation , changes over time . holding @xmath21 constant imposes the same minimum overdensity for clusters at all redshifts . we use the value @xmath22 ( see also hbb ) , corresponding to an overdensity of @xmath23 times the mean density . the center of each fof cluster is taken to be the position of the most bound particle . this center is used to compute the virial radius @xmath24 , defined as the radius enclosing the virial overdensity expected from spherical top - hat collapse ( which evolves with redshift in the @xmath0cdm cosmology ) . in this analysis , we consider only clusters with virial mass threshold ( i.e. , the total mass within the virial radius ) @xmath17 . this mass threshold is typical for a poor cluster of galaxies . the mean mass of all clusters in the low-@xmath13 sphere is @xmath25 . in [ sec : him ] we discuss superclusters selected from clusters with a higher mass threshold . the ellipticities of the clusters identified at low and high @xmath13 are studied in detail by hbb . relevant to our present work on superclusters , they find alignments between nearby cluster pairs out to separations of @xmath26 ; the strength of alignment increases with redshift and with decreasing cluster separation . hbb explain these alignments as being due to clusters forming and gathering material along `` filamentary superstructures '' ; they provide evidence for overdensities along the separation vectors between pairs of aligned clusters . as demonstrated by @xcite , @xcite and references therein , superclusters can be identified by their overdensity of clusters . we use the fof method to identify superclusters from the cluster sample defined above ( 2.1 - 2.2 ) . the mean density of clusters in the low-@xmath13 sphere ( we shall not discuss the high-@xmath13 cone until [ sec : evo ] ) is @xmath27 , yielding a mean intercluster separation of @xmath28 . we therefore select linking lengths of @xmath29 and @xmath30 , corresponding to @xmath31 and @xmath32 , respectively . all scales in this paper are in comoving coordinates . the different linking lengths sample different portions of the supercluster geometry . small linking lengths identify only the dense cores of superclusters , whereas larger linking lengths allow the fof algorithm to percolate to lower - density filaments of clusters . large @xmath18-values therefore reach to the outer parts of superclusters , as well as identify looser superclusters , and often connect structures that appear discrete with smaller @xmath18 . we use monte carlo simulations to compare the significance of the selected superclusters with that of random associations . clusters are placed at randomly chosen positions within a sphere such that the total number density is the same as that for the simulated clusters , and the fof method is applied , as before , with @xmath29 and @xmath30 . we find significantly fewer clusters per supercluster , and significantly fewer superclusters down to multiplicity as small as 3 , for all four linking lengths tested . this result implies that most of the superclusters are real , even for @xmath33 . the only exception is the case of binary clusters ( superclusters with only 2 members ) selected with @xmath33 , of which we find only @xmath34% more using the simulated clusters than we do with the monte carlo sample . the random rate of superclusters is typically @xmath35% for @xmath36 superclusters , decreasing to @xmath37% for @xmath38 superclusters . the monte carlo results are compared with the @xmath0cdm multiplicity functions below . to help visualize the typical geometries of superclusters identified by the fof method , we present supercluster maps of a typical region ( a cube of side length @xmath39 , concentric with the low-@xmath13 sphere ) for @xmath40 ( figs@xmath41[fig1]-[fig2 ] ) . each point is a cluster , and a given symbol represents membership in the same supercluster . the maps present a projection of the 3-d distribution onto the @xmath42-plane in the first plot , and onto the @xmath43-plane in the second plot . viewing the plots in turn corresponds to a @xmath44 rotation of the viewing angle about the @xmath45-axis . for clarity , we show only superclusters with at least 5 member clusters . members of a supercluster that extend outside the limits of our cubical region are not shown , so some superclusters appear to have fewer members than the imposed minimum . comparing these maps to analogous figures generated with @xmath46 and @xmath30 reveals , as expected , that increasing the linking length often causes neighboring superclusters to be connected into one . this is especially true when @xmath18 increases from 7 to @xmath30 , resulting in many large , complicated structures with central cores joining multiple lower - density filaments . the _ multiplicity _ of a supercluster is the number of clusters it contains , @xmath47 . in fig@xmath41[fig3 ] we present , for various linking lengths , the integrated supercluster multiplicity function the number density ( per cubic comoving @xmath4 ) of superclusters with @xmath47 or more members , as a function of @xmath47 . also shown for comparison are the corresponding results from random catalogs , as described in the previous section . longer linking lengths yield more superclusters of all multiplicities , as expected . shorter linking lengths yield curves that rise more steeply toward low multiplicities , implying that they preferentially find small superclusters with low multiplicity . conversely , longer linking lengths find a higher percentage of larger , richer superclusters . the shorter linking lengths naturally trace high - density regions ( e.g. , dense cores of superclusters ) , while larger linking lengths characterize lower - density regions , such as the outer parts of superclusters or diffuse superclusters . small @xmath18 can also divide structures that are identified with larger linking lengths . the fact that we see superclusters with @xmath48 or more clusters may seem surprising , but the total number density of such structures is only @xmath49 per @xmath50 , and it is based on a very low mass threshold of cluster members , @xmath17 . such clusters are numerous , and any comparison with observations should of course account for this threshold . for superclusters selected using higher - mass clusters , see [ sec : mevo ] . we repeated the above analysis after selecting superclusters in redshift - space coordinates ; the resulting multiplicity functions are nearly identical to the real - space functions at low multiplicities , and are shifted to slightly higher @xmath47 ( by up to 20% ) at the highest multiplicities . the supercluster size function i.e. , the number of superclusters above a given size is presented in fig@xmath41[fig4 ] for different linking lengths . the size of a supercluster is defined as the maximum distance of any member cluster from the supercluster center of mass . as with the multiplicity function , the size function rises more steeply for shorter linking lengths . this implies that the longer linking lengths find a greater ratio of large superclusters to small superclusters , analogous to the result for the multiplicity function . similar results are found using redshift - space coordinates ; for the largest linking length used , the sizes are slightly ( @xmath51 ) larger . we find that the largest supercluster radii are @xmath52 , corresponding to diameters of @xmath53 . this is consistent with the largest structures seen in observations ( e.g. * ? ? ? * ; * ? ? ? * ) , and it is also the largest distance at which hbb detect cluster alignment in the simulation . the average multiplicity of superclusters is plotted as a function of supercluster size for different linking lengths in fig@xmath41[fig5 ] . for @xmath54 or @xmath55 objects more aptly called binary or triple clusters than superclusters the maximum radius is set mainly by the separation(s ) of the 2 or 3 clusters ; thus the average multiplicity changes slowly over this range of separations . beyond this regime , the multiplicity and size of superclusters appear to be related nearly linearly . this result suggests that the spherically averaged density profile in superclusters is roughly _ isothermal _ ; i.e. , @xmath56 , where the cluster number density @xmath57 satisfies : @xmath58 this profile yields multiplicity @xmath59 for @xmath56 . we can investigate the spherically averaged density profile more directly by examining the distribution of clusters within individual superclusters and determining the average number density profile for the entire sample . to do so , we superpose the centers - of - mass ( calculated using only the mass in clusters ) of all superclusters , and scale all superclusters to the same size by dividing all cluster coordinates ( relative to the center - of - mass ) for a given supercluster by its @xmath60 . we add up the total number of clusters inside radial bins , divided by the total number of clusters in superclusters for the given linking length . we have included only superclusters with 5 or more members ; because the multiplicity function drops off rapidly , most of the superclusters included have only 5 or 6 members . this fact affects the combined profile in two ways . first , only in few cases will one of the five members happen to lie very near the supercluster s center - of - mass ; thus it is only at @xmath61 that we begin to find a significant fraction of the clusters . second , one out of roughly five clusters in each supercluster lies at @xmath62 . thus , for the smaller linking lengths , the integrated number fraction approaches a limit of @xmath63 ; for the larger linking lengths , which find richer superclusters , the limiting fraction is slightly higher . the fraction of clusters within @xmath64 increases linearly with radius for the range @xmath65 . this linear relationship indeed implies @xmath66 for the supercluster density profile in this range . the profile steepens on larger scales . the average mass of clusters in superclusters , @xmath67 for the @xmath40 superclusters , is somewhat larger than the average mass of clusters in the total sample , @xmath68 , indicating that superclusters contain slightly more massive clusters , on average , than the mean . the @xmath40 superclusters with @xmath69 members have an even higher average cluster mass of @xmath70 , implying that richer superclusters contain more massive clusters than poorer superclusters . we measure the shape , and in particular the dimensionality , of a given supercluster by fitting a 3-d ellipsoid to the distribution of its member clusters . we use the code developed by hbb to fit ellipsoids to the dark matter particles comprising their clusters . the code constructs , for a given supercluster , the @xmath71 matrix of second moments of cluster positions relative to the supercluster center - of - mass , i.e. @xmath72 where @xmath73 and @xmath74 are two of the position coordinates for a given cluster , and @xmath75 is the mass of the cluster ; the sum is over all member clusters . the eigenvalues of this matrix are simply the three axis lengths of the best - fit ellipsoid for the supercluster , times a known constant . following the notation used by hbb , we shall denote these lengths @xmath76 for the primary , secondary , tertiary axes , respectively ; i.e. , @xmath77 . we restrict our analysis to superclusters with @xmath69 ; for poorer superclusters , the `` best - fit ellipsoid '' would not be very meaningful . we have already examined size distributions for superclusters ( [ sec : size ] ) , so we now focus on the _ axis ratios _ , which allow us to probe the shape and dimensionality of superclusters . fig@xmath41[fig6 ] displays the distribution of _ primary axis ratios _ @xmath78 for superclusters selected with @xmath79 . fig@xmath41[fig7 ] shows the distribution of _ secondary axis ratios _ @xmath80 , and fig@xmath41[fig8 ] shows @xmath81 . the mean and peak axis ratios from each curve in these figures are listed in table 1 . the secondary axis ratios are generally smaller than the primary axis ratios . we find that a primary axis ratio of @xmath82 is typical for superclusters selected using a broad range of linking lengths ; this corresponds to a primary ellipticity of @xmath83 . the secondary axis ratio is more strongly dependent on the linking length used to find the superclusters . for @xmath84 , we find a peak @xmath80 of @xmath85 , corresponding to a secondary ellipticity of @xmath86 ; for @xmath33 , we find a peak of @xmath87 , or @xmath88 . the peak @xmath81 ratios range from @xmath89 for @xmath33 to @xmath90 for @xmath84 . these measurements allow us to construct a picture of the typical shape and dimensionality of superclusters . the interpretation of different ellipticities can be described as follows : 1 . a combination of high primary and low secondary axis ratios is typical of two - dimensional pancake - like structures ( two dimensions are large , and one is small ) . a combination of low primary and high secondary axis ratios is typical of one - dimensional filamentary structures ( one dimension is large , and two are small ) . a combination of primary and secondary axis ratios that are both significantly less than unity is typical of structures most appropriately described as triaxial ( each dimension is a different size ) . 4 . if both axis ratios are close to unity , then the distribution is nearly spherical . the results we find suggest that superclusters are typically triaxial structures nearing 2-d , pancake - like structures at @xmath91 . they are neither spherical nor purely filamentary in nature . as the linking length is increased , the resulting superclusters become more two - dimensional ; the cores of superclusters are thus more triaxial , spreading out to more pancake - like structures on larger scales . we plot the bivariate distribution of @xmath78 and @xmath80 for @xmath40 in fig@xmath41[fig9 ] ( the distributions for @xmath92 and @xmath93 are similar ) . applying the interpretation rules listed above , 1-d structures reside in the upper left corner of fig@xmath41[fig9 ] , 2-d in the lower right corner , and 3-d ( spherical ) in the upper right . points in the middle or toward the lower left corner represent triaxial structures . fig@xmath41[fig9 ] shows that a large majority of the superclusters are more nearly two - dimensional than one - dimensional . no nearly spherical superclusters are found with this linking length , but there is a smaller population of filamentary structures . similar distributions are found for @xmath92 and @xmath93 , though showing more flattened structures ( 2d ) as @xmath18 increases , and more triaxial structures as @xmath18 decreases . this is the same trend depicted in fig@xmath41[fig7 ] as a shift toward lower secondary axis ratio with increasing linking length . the shorter linking lengths also yield significantly higher fraction of filamentary structures i.e. , superclusters with low primary axis ratio and a secondary axis ratio approaching unity . this trend suggests that most superclusters at @xmath94 are triaxial structures with one dimension considerably smaller than the other two . this geometry is apparent in fig@xmath41[fig9 ] , and becomes more significant when a larger fof linking length is used . since the superclusters are not completely flat , smaller linking lengths may probe only the dense core regions , which are smaller in size than the supercluster thickness . thus , while the thickness sets an upper bound to the tertiary axis length identified with any linking length , the higher @xmath18-values find longer primary and secondary axes by probing farther out into the pancake . this causes larger linking lengths to find a smaller mean secondary axis ratio , and reveal the underlying 2-d nature of the large - scale superclusters . similarly , for a given @xmath18 , the largest , highest - multiplicity superclusters ( @xmath95 ) are on average flatter ( @xmath96 for @xmath97 ) than those with @xmath98 5 - 10 ( @xmath99 ) . this flatness is still apparent in projection : for superclusters selected ( from @xmath17 clusters ) in redshift - space , the projected axis ratios have a broad distribution with a mean @xmath100 and a peak @xmath101 , for all linking lengths @xmath1025 , 7 , and 10@xmath4 . the triaxiality of superclusters can also be quantified using the measurement commonly employed for geometrical descriptions of elliptical galaxies , first introduced explicitly by @xcite : @xmath103 with @xmath77 , @xmath104 will take on values between 0 and 1 ( unless @xmath105 , in which case @xmath104 is undefined ) . two - dimensional pancake structures approach @xmath106 , while filaments approach @xmath107 ( note , however , that the increase is not linear with axis ratio ) . intermediate values of @xmath104 represent triaxial systems , with triaxiality increasing with @xmath104 . by this measure our low-@xmath13 superclusters are triaxial , ranging from a mean of @xmath108 for @xmath33 to @xmath109 for @xmath84 . for all linking lengths we see a tail in the distribution that describes structures which are very nearly one - dimensional . thus , some mass concentrations are filamentary at @xmath91 , especially in the relatively dense regions probed by @xmath84 . indeed , we would expect structures that have collapsed along two dimensions to be of higher average density than those that have collapsed along only one dimension . this result agrees with hbb s identification of filamentary density enhancements in the large - scale structure ; here we add the prediction of more diffuse flattened structures surrounding and possibly connecting the filaments . we examine the time evolution of the supercluster properties studied above . we study superclusters in two redshift slices at @xmath110 and @xmath111 , denoted @xmath112 and @xmath113 , respectively . superclusters are identified by the fof method as before . however , since the cluster abundance decreases at higher redshift , and their mean separation thus increases , we compare superclusters found at different redshifts _ not _ with the same linking length @xmath18 , but instead with the same linkage parameter @xmath19 , where @xmath20 is the mean comoving intercluster separation . as discussed in 2.2 , this approach sets a minimum overdensity for detection of superclusters , which is constant with redshift . this use of a constant linkage parameter accounts for the reduced density of clusters at high redshift . however , because clusters are clustered , the density of superclusters depends on the relative rates of cluster formation in regions of high density vs. low density . hierarchical supercluster formation , in which clusters gravitate toward each other over time , also contributes to a reduced supercluster density at early times . we indeed find a lower density of superclusters in the high-@xmath13 cone . the total number of superclusters decreases by a factor of a few from @xmath114 to @xmath112 , and by over an order of magnitude from @xmath112 to @xmath113 . in order to construct a sufficiently large sample , we consider broad ranges in redshift , especially for the @xmath113 slice . in addition , we omit the @xmath21-value corresponding to the @xmath115 ( @xmath114 ) case from our subsequent analysis , as it results in too few superclusters at high redshift . as stated in 2.3 , our @xmath114 linking lengths of @xmath79 correspond to @xmath116 , respectively . we use these latter linkage parameters @xmath21 to refer to the linking lengths used in our subsequent analysis . a given value of the @xmath21-parameter can be converted to an actual comoving linking length at a given redshift simply by multiplying by the mean intercluster separation at that redshift . the mean cluster densities in our @xmath112 and @xmath113 shells are @xmath117 and @xmath118 , corresponding to mean cluster separations of @xmath119 and @xmath120 , respectively . therefore , @xmath121 0.23 , 0.33 , and 0.47 correspond to @xmath102 7 , 10 , and 14.3@xmath4 at @xmath122 , and 16.8 , 24 , and 34.3@xmath4 at @xmath123 . the evolution of the multiplicity function is presented in fig@xmath41[fig10 ] . we present the integrated multiplicity function for all three redshifts , using two different values of @xmath21 . the number density of superclusters drops noticeably with increasing redshift . monte carlo simulations analogous to those described in 2.3 were also run at @xmath112 and @xmath124 , with results similar to those obtained at low @xmath13 : the multiplicity functions are physically significant for all multiplicities and linking lengths , except for binary clusters in the @xmath125 case , where only @xmath126 more binary clusters are found over the monte carlo results . the redshift evolution of the multiplicity function depends on the linking length of the superclusters . the vertical distances between the @xmath127 curves in fig@xmath41[fig10 ] increase with increasing multiplicity , implying that the abundance of the richest ( highest-@xmath47 ) superclusters found with this linking length drops off especially swiftly with increasing @xmath13 . however , for @xmath128 the vertical distances _ decrease _ with increasing @xmath47 , to the extent that the abundance of the richest superclusters found with this linking length is in fact roughly the same at @xmath112 as it is today . we interpret these trends as further evidence that small linking lengths ( e.g. @xmath128 ) select the densest regions of superclusters . whereas the larger linking lengths do not find many rich superclusters at high redshift because of the lack of clusters , the smaller linking lengths find the same dense supercluster cores even at much earlier times . these dense cores are likely the sites of the largest overdensities present at early times , and the first places where observable structures form through gravitational collapse . thus smaller linking lengths relative to the mean intercluster separation are needed to locate the highest overdensities in the observable universe . the lack of rich superclusters at high @xmath13 is also evidence that these structures form largely through gravitational accretion of clusters that are initially too distant to be identified as supercluster members . supercluster size remains highly correlated with multiplicity at high @xmath13 , but with a lower @xmath129 normalization than at low @xmath13 . the size function does not evolve significantly with redshift for the larger superclusters ( see fig@xmath41[fig11 ] ) , while , as expected , the number of cluster members decreases . using the method described in [ sec : shapes ] , we fit ellipsoids to the superclusters identified at @xmath112 and @xmath113 . because we only perform these fits on superclusters with @xmath69 , which are rare at high redshift , the statistical uncertainties are larger . for example , only 68 superclusters are identified at @xmath113 for @xmath128 . we use the axis ratios of the best - fit ellipsoids to characterize the evolution of supercluster dimensionality . tables 2 and 3 present the mean primary and secondary axis ratios , respectively , for @xmath116 at @xmath130 . we also construct the axis ratio distributions ( similar to figs@xmath41[fig6]-[fig7 ] ) as a function of @xmath13 for @xmath131 and @xmath32 ( there are too few @xmath128 superclusters at high redshift to create meaningful distributions ) . the results for @xmath127 are shown in figs@xmath41[fig12]-[fig13 ] . the number of high - redshift superclusters is too low to construct bivariate distributions as in fig@xmath41[fig9 ] , but we can consider how the changes in _ mean _ ratios correspond to translations of these distributions . the strongest evolution observed is the increase in the secondary axis ratio with increasing redshift , for @xmath125 ( fig@xmath41[fig13 ] ) . most of the change occurs between @xmath114 and @xmath112 , with only a slight additional increase between @xmath112 and @xmath113 ( see table 3 ) . typical secondary axis ratios increase from @xmath132 at low @xmath13 to @xmath133 at high @xmath13 . thus the flattened structures observed with @xmath125 at @xmath114 ( see [ sec : shapes ] ) appear less flattened at high redshifts , indicating their gravitational collapse over time . very little evolution is evident in the secondary axis ratio for superclusters identified with @xmath131 , but the population of high - density filaments with @xmath134 evolves between @xmath112 and @xmath114 , as there are almost none of these filamentary structures at high redshift . the @xmath135 distributions for @xmath136 are shown in figs@xmath41[fig6]-[fig7 ] ; at higher redshifts the distributions are similar to the corresponding curves in figs@xmath41[fig12]-[fig13 ] . the primary axis ratios are slightly smaller at higher redshifts for both @xmath131 and @xmath32 . typical values decrease from @xmath82 at @xmath114 to @xmath137 at @xmath138 1 - 2 . in a plot similar to fig@xmath41[fig9 ] , this would correspond to a leftward shift in the distribution , i.e. a shift toward greater triaxiality at high redshift , and a shift to a 2-d geometry as the systems approach @xmath114 . quantitatively , our high-@xmath13 superclusters have mean triaxiality @xmath139 for both linking lengths , significantly higher than the mean @xmath104 for any linking length at low @xmath13 . we conclude that superclusters had highly triaxial structure at early times , and that between a cosmological look - back time of @xmath140 gyr ( corresponding to @xmath112 for the cosmology described in 2.1 ) and the present day , many collapsed along their smallest dimensions to form the more nearly pancake - like structures observed in the simulation at @xmath114 . in the same span of time , some structures in regions of highest density also collapsed along a second dimension to form nearly one - dimensional filaments . how do the supercluster properties change if superclusters are selected using higher - mass clusters as their seeds ? while the abundance of high - mass clusters is significantly lower , and the statistical uncertainties larger , the massive clusters are more easily detected in observational surveys . we repeat the analysis described above for superclusters selected using the rarer , higher - mass clusters with @xmath141 . this threshold corresponds to typical `` rich clusters '' ( e.g. * ? ? ? * ; * ? ? ? there are @xmath142 clusters above this mass threshold at @xmath91 , and @xmath143 clusters at @xmath122 . the cluster abundance is @xmath144 at @xmath91 , decreasing to @xmath145 at @xmath122 . the intercluster mean separation is @xmath146 at @xmath91 , increasing to @xmath147 at @xmath122 . superclusters are selected from these clusters at @xmath91 and @xmath122 using the linkage parameters @xmath136 and @xmath32 ; these correspond to @xmath148 and @xmath149 respectively at @xmath91 , and @xmath150 and @xmath151 at @xmath122 . the large linking lengths @xmath18 reflect the large mean intercluster separation of these massive clusters . we also include , for comparison , the smaller linking length of @xmath97 at @xmath91 . there are fewer superclusters identified , and a smaller number of cluster members in each , as compared to those selected using the low - mass clusters . the overall structure of the superclusters is not significantly changed , as described below . the multiplicity function of the superclusters is presented in fig@xmath41[fig14 ] for both @xmath91 and @xmath152 . the results show , as expected , a similar shape to the multiplicity function of superclusters selected using the more numerous low - mass clusters ( figs@xmath41[fig3 ] , [ fig10 ] , [ fig11 ] ) , but with a greatly reduced amplitude : the number of superclusters at a given multiplicity is lower , and equivalently the supercluster multiplicity ( richness ) is smaller for a given supercluster abundance . this is partly due to the much smaller number of massive clusters . the evolution of the multiplicity function is stronger than that of superclusters selected with low - mass clusters ( cf . fig@xmath41[fig10 ] ) . superclusters with linkage parameters of @xmath153 contain up to @xmath154 rich clusters at @xmath91 . the supercluster multiplicity decreases sharply at higher redshifts . the supercluster size function is presented in fig@xmath41[fig15 ] . the maximum size of @xmath91 superclusters ranges from @xmath155 ( i.e. , twice the plotted @xmath60 radius ) for @xmath97 , to @xmath156 for @xmath157 ( @xmath136 ) , to @xmath158 for @xmath159 ( @xmath127 ) . for a given linkage parameter @xmath21 , the maximum supercluster size is somewhat larger than that of superclusters selected with low - mass clusters , since the linking length values @xmath18 are larger . for a given value of @xmath21 , no significant evolution is observed in the size of the largest superclusters , reflecting the existence of these extended structures at early times . the same result is observed for superclusters selected with the lower mass threshold clusters . the number of cluster members within these large structures decreases with redshift , as seen above , due to the lowered cluster abundance at high @xmath13 . the supercluster multiplicity versus size relation is similar but reduced in amplitude , as expected to the superclusters selected with the lower mass threshold ( fig@xmath41[fig5 ] ) . a nearly linear relation between multiplicity and size is observed for all but the binary clusters . the ellipticity distribution shows that superclusters exhibit a triaxial geometry on average , but with significantly larger scatter than the more numerous superclusters selected with lower - mass clusters ( figs@xmath41[fig12]-[fig13 ] ) . the mean primary and secondary axis ratios are each @xmath160 to @xmath161 ; for example , for @xmath136 at @xmath91 , the mean values are @xmath162 and @xmath163 . these ratios do not change significantly with either linking length or with redshift to @xmath122 . the distribution of ellipticities around this mean is , however , quite broad . the supercluster shapes are not significantly different from those selected using the lower - mass clusters , except for the stronger flattening observed in the latter at @xmath91 , where the mean triaxial shape has evolved slightly towards a flattened two - dimensional shape ( with @xmath164 and @xmath165 ) . the superclusters selected with high - mass clusters are triaxial , with somewhat higher triaxiality at @xmath91 than those selected using lower - mass clusters i.e . , the high - mass superclusters are approaching a slightly more filamentary nature , with mean axis ratios of @xmath160 in both @xmath78 and @xmath80 ( cf . fig@xmath41[fig9 ] ) , and a mean triaxiality of @xmath166 . we have investigated the multiplicities , sizes , radial structures , and dimensionalities of superclusters from @xmath114 to @xmath113 in @xmath0cdm cosmology . the results improve our understanding of the properties ( especially shapes ) and evolution of superclusters , and can also be used as predictions for testing the cosmology with future observations . at low redshift , we predict a total number density of @xmath167 , and @xmath168 superclusters with at least five member clusters with mass @xmath17 for superclusters selected with linking length @xmath169 , and @xmath93 , respectively . the maximum size of these superclusters ranges from @xmath170 for @xmath97 to @xmath171 for @xmath172 . the abundance of superclusters decreases rapidly with increasing redshift . there are many more superclusters at low @xmath13 than at high @xmath13 largely because there are many more clusters that have formed by the present time and that have gravitated to form superclusters . the richest superclusters are almost entirely missing at high redshifts ; these structures form late in cosmic time . however , the densest `` core '' regions of superclusters , traced well by fof linking lengths @xmath173 of the mean intercluster separation , are present at @xmath122 almost as abundantly as they are today . these are likely the sites of the highest overdensities in the early universe . the spherically averaged density profiles of superclusters are well fit by an isothermal profile , @xmath174 , over a broad range in radius . the profile is shallower at small radii and is steeper at large radii . a nearly linear relation exists between supercluster size and multiplicity , arising from this density profile . we find that superclusters are triaxial in shape , especially at early times , where the mean triaxiality of our sample is @xmath139 . over time , many collapse along a single dimension to approach two - dimensional shapes on the largest scales ( i.e. , those sampled by the fof linking length @xmath33 at @xmath114 , using low - mass clusters with @xmath17 ) ; their cores remain triaxial . quantitatively , the mean triaxiality for long linking lengths decreases to @xmath108 at @xmath114 , whereas for smaller @xmath18 ( i.e. , in the cores ) it decreases to @xmath109 . the mean primary axis ratio of superclusters at @xmath114 is @xmath175 , corresponding to a primary ellipticity @xmath176 . typical secondary axis ratios at @xmath114 are @xmath177 and @xmath178 for @xmath84 , and @xmath179 and @xmath180 for @xmath33 . gravitational collapse increases ellipticity , as the potential gradient is larger along the minor rather than the major axis @xcite ; collapse thus leads to lower dimensionality ( cf . * ? ? ? * for a more recent treatment ) . a nonnegligible population of dense , filamentary superstructures is also present at @xmath114 , though not at higher redshift . superclusters selected using higher - mass clusters , @xmath181 , show consistent results , with a reduced multiplicity function , as expected . these superclusters are triaxial in shape , with somewhat greater mean triaxiality at @xmath91 ( @xmath182 0.8 - 0.85 ) than superclusters with lower - mass clusters . these properties , derived from a large - scale , high - resolution cosmological simulation , provide direct information on the size , content , shape , and evolution of superclusters , as well as testable predictions of the currently favored @xmath0cdm model with which future and currently accumulating observations can be compared . once sdss is complete , the best statistical analyses to date of large - scale structure using observations will be possible , especially if accurate and well - behaved photo-@xmath13 algorithms are developed . there are other exciting observing programs on the horizon . the pan - starrs project @xcite is expected to see first light in 2006 , and will carry out multi - band imaging surveys to identify galaxy clusters , as well as gravitational weak lensing observations to map the matter distribution in the universe . the large synoptic survey telescope @xcite , which is expected to see first light in 2012 , will provide improved weak lensing maps that will probe the dark matter distribution to the largest scales . the analysis presented here will aid in the comparison of these major observational data with the predictions of the concordance cosmological model . this research was supported in part by nsf grant ast-0407305 . computations were performed on the national science foundation terascale computing system at the pittsburgh supercomputing center ; additional computational facilities at princeton were provided by nsf grant ast-0216105 . abell , g. o. 1958 , apjs , 3 , 211 bahcall , n. a. 1988 , ara&a , 26 , 631 bahcall , n. a. & soneira , r. m. 1984 , apj , 277 , 27 bahcall , n. a. , ostriker , j. p. , perlmutter , s. , & steinhardt , p. 1999 , science , 284 , 1481 bennett , c. l. , et al . 2003 , apjs , 148 , 1 bode , p. , & ostriker , j. p. 2003 , apj , 145 , 1 budavri , t. , szalay , a. s. , connolly , a. j. , csabai , i. , & dickinson , m. e. 2000 , aj , 120 , 1588 csabai , i. , connolly , a. j. , szalay , a. s. , & budavri , t. 2000 , aj , 119 , 69 davis , m. , efstathiou , g. , frenk , c. s. , & white , s. d. m. 1985 , apj , 292 , 371 einasto , m. , einasto , j. , tago , e. , dalton , g. , & andernach , h. 1994 , mnras , 269 , 301 einasto , j. , et al . 2003a , a&a , 405 , 425 einasto , j. , et al . 2003b , a&a , 410 , 425 einasto , j. , et al . 2006 , a&a , submitted ( astro - ph/0603764 ) erdogdu , p. , et al . 2004 , , 352 , 939 geller , m. j. , & huchra , j. p. 1989 , science , 246 , 897 gott , j. r. , et al . 2005 , apj , 624 , 463 gregory , s. a. , & thompson , l. a. 1978 , apj , 222 , 784 gregory , s. a. , thompson , l. a. , & tifft , w. g. 1981 , apj , 243 , 411 hopkins , p. f. , bahcall , n. , & bode , p. 2005 , apj , 618 , 1 hubble , e. 1936 , the realm of the nebulae ( new haven : yale university press ) kaiser , n. , et al . 2002 , proc . spie , 4836 , 154 kolokotronis , v. , basilakos , s. , & plionis , m. 2002 , mnras , 331 , 1020 lin , c. c. , mestel , l. , & shu , f. h. 1965 , apj , 142 , 1431 navarro , j. f. , frenk , c. s. , & white , s. d. 1996 , apj , 462 , 563 navarro , j. f. , frenk , c. s. , & white , s. d. 1997 , apj , 490 , 493 oort , j. h. 1983 , ara&a , 21 , 373 ostriker , j. p. , & steinhardt , p. j. 1995 , nature , 377 , 600 peacock , j. a. 1999 , cosmological physics ( cambridge , uk : university press ) spergel , d. n. , et al . 2003 , apjs , 148 , 175 spergel , d. n. , et al . 2006 , astro - ph/0603449 statler , t. s. 1994 , aj , 108 , 111 stoughton , c. , et al . 2002 , aj , 124 , 1810 tyson , j. a. 2002 , proc . spie , 4836 , 10 york , d. g. , et al . 2000 , aj , 120 , 1579 yoshisato , a. , morikawa , m. , gouda , n. , & mouri , h. 2006 , apj , 637 , 555 rrrrrrr @xmath18 & mean @xmath78 & peak @xmath78 & mean @xmath80 & peak @xmath80 & mean @xmath81 & peak @xmath81 + 5 & 0.60 @xmath183 .009 & 0.64 & 0.50 @xmath183 .007 & 0.41 & 0.29 @xmath183 .004 & 0.23 + 7 & 0.57 @xmath183 .004 & 0.62 & 0.43 @xmath183 .003 & 0.31 & 0.23 @xmath183 .001 & 0.19 + 10 & 0.54 @xmath183 .002 & 0.64 & 0.33 @xmath183 .001 & 0.20 & 0.16 @xmath183 .001 & 0.13 + rrrrrrr @xmath13 & @xmath184 & @xmath136 & @xmath127 + 0 & 0.60 @xmath183 .009 & 0.57 @xmath183 .004 & 0.54 @xmath183 .002 + 1 & 0.44 @xmath183 .03 & 0.47 @xmath183 .01 & 0.48 @xmath183 .01 + 2 & 0.52 @xmath183 .06 & 0.50 @xmath183 .03 & 0.50 @xmath183 .02 + rrrrrrr @xmath13 & @xmath184 & @xmath136 & @xmath127 + 0 & 0.50 @xmath183 .007 & 0.43 @xmath183 .003 & 0.33 @xmath183 .001 + 1 & 0.36 @xmath183 .02 & 0.40 @xmath183 .01 & 0.44 @xmath183 .01 + 2 & 0.41 @xmath183 .05 & 0.41 @xmath183 .03 & 0.46 @xmath183 .01 +
in the early universe , a scalar field often oscillates coherently after inflation @xcite . coherent oscillation of a scalar field may have various cosmological effects such as generation of curvature perturbation , dark matter , baryon asymmetry and unwanted relics . in a simple model of a real scalar field @xmath4 with a canonical kinetic term and a scalar potential of @xmath5 , it is known that the scalar field stops at the initial position @xmath6 until the hubble parameter becomes equal to the effective mass of the scalar . after that , @xmath4 begins a coherent oscillation for @xmath7 with @xmath8 being the critical exponent , which depends on the background equation of state ( eq . ( [ cond1 ] ) ) . in this case , the mass scale of @xmath4 becomes larger and larger compared with the hubble scale and such a coherent scalar dissipates its energy due to the hubble expansion . the amplitude of the coherent oscillation @xmath9 decreases as @xcite @xmath10 with @xmath11 being the scale factor of the universe . such a scalar field can act as a curvaton @xcite , which generates the primordial density perturbation of the universe @xcite . in particular , curvaton models with @xmath12 were studied in some literature @xcite . on the other hand , for @xmath13 , the scalar field never oscillates and it just follows a so - called scaling solution @xcite ) scaling solution . we do not regard an oscillating solution as a scaling solution even if its amplitude satisfies a simple power law like ( [ scale1 ] ) . ] @xmath14 this is an attractor and it was pointed out that such a scalar field can not generate curvature perturbation since the initial fluctuation of @xmath6 is erased during the scaling regime @xcite . in this paper , we revisit the scalar dynamics for @xmath15 . in ref . @xcite it is stated that in this critical case , the scalar field is not attracted to a scaling solution but it is something like between a scaling and oscillation . nevertheless , they say that for @xmath15 both the oscillating solution ( [ scale1 ] ) and scaling solution ( [ scale2 ] ) show the same behavior with the cosmic expansion and hence the curvature perturbation also vanishes . however , we will see that the actual dynamics of the scalar field for @xmath15 is far from both the oscillating solution ( [ scale1 ] ) and scaling solution ( [ scale2 ] ) . even averaged features of the scalar dynamics can not be described by these solutions . instead , a peculiar feature of the scalar dynamics shows up , which we call a `` pseudo scaling solution '' . we will show it both in a numerical and analytical way and see that the curvature perturbation does not vanish in this case . the pseudo scaling solution can have a variety of physical implications . as we noted above , the curvature perturbation can be generated by a curvaton even if it follows the pseudo scaling solution , and hence it may offer the seed of the large scale structure of our universe . another example is a symmetry breaking field such as the peccei - quinn field @xcite . in the supersymmetric ( susy ) axion model , it is possible that the peccei - quinn field follows the pseudo scaling solution in the early universe . as we will see later , we can naturally solve the isocurvature perturbation as well as the domain wall problem in such a case . the pseudo scaling behavior also has some implications on the baryon number asymmetry produced by the affleck - dine mechanism @xcite . therefore , it is worth studying detailed features of the pseudo scaling solution . the organization of this paper is as follows . in sec . [ sec : mono_pot ] , we discuss the dynamics of the scalar field with a monomial potential in a general background equation of state . here the pseudo scaling " scalar dynamics is extensively studied . also , some comments on the hubble mass term are given there . cosmological implications of the pseudo scaling solution are explained in sec . [ sec : cos ] . we discuss the curvature perturbation and the domain wall problem there . the last section is devoted to conclusions and discussion . [ sec : mono_pot ] let us consider the lagrangian @xmath17 where the scalar potential is given by the monomial one @xmath18 where @xmath12 is an integer . , @xmath4 should be interpreted as @xmath19 . ] the equation of motion of @xmath4 is given by @xmath20 with @xmath21 being the hubble parameter . we take @xmath22 and assume @xmath23 . in the matter ( radiation)-dominated ( md ( rd ) ) universe , @xmath24 . a more general equation of state may be obtained in the inflaton oscillation dominated universe ( see app . [ sec : app ] ) . let us define @xmath25 with @xmath11 being the scale factor of the universe and @xmath26 the initial time . we have taken @xmath27 . then the equation of motion becomes @xcite @xmath28 where the prime denotes derivative with respect to @xmath29 and @xmath30 the equation ( [ eqofm1 ] ) represents the motion of the rescaled field @xmath31 in the effective potential @xmath32 in these variables , @xmath31 has a minimum at @xmath33 where @xmath34 thus if @xmath35 the sign of the friction term is positive , and hence the oscillation of @xmath31 damps and it relaxes to the minimum of the potential @xmath33 . let us suppose the condition ( [ cond1 ] ) is satisfied so that there is a nontrivial minimum of the effective potential and @xmath31 relaxes there . , and @xmath36 for @xmath37 . thus , @xmath38 is ensured and there is a nontrivial minimum of the potential if @xmath39 . it does not always hold if there is a hubble mass term in the potential ( see eq . ) . ] this represents the scaling solution : in terms of the original field @xmath4 , @xmath40 in this case , independently of the initial condition , @xmath4 finally follows the scaling solution without oscillation . note also that the scaling solution implies that the effective mass scale of the scalar decreases in proportion to the hubble parameter : @xmath41 . on the other hand , if the inequality ( [ cond1 ] ) is inverted , the friction term of ( [ eqofm1 ] ) has a negative sign , and hence @xmath31 never relaxes to its potential minimum . this corresponds to an oscillating solution . this behavior is also understood as follows . assuming an oscillating solution for @xmath4 with potential @xmath42 , we obtain @xmath43 by making use of the virial theorem with @xmath9 being the oscillation amplitude of @xmath4 . thus the effective mass of @xmath4 scales as @xmath44 . if this can continue to exceed the hubble parameter @xmath45 , the oscillation survives . this condition is written as @xmath46 this is just an inversion of the condition ( [ cond1 ] ) . we have seen that a scalar field approaches to a scaling solution for @xmath47 and oscillating solution for @xmath48 independently of the initial condition . however , the case of @xmath15 is more nontrivial . in this case , the friction term of ( [ eqofm1 ] ) vanishes : @xmath49 in terms of @xmath31 , it does not universally approach to some fixed solutions and continues to oscillate without damping . the resulting dynamics significantly depends on the initial condition of @xmath4 . if the initial condition is chosen so that @xmath31 oscillates around @xmath33 without crossing @xmath50 , @xmath4 also does not cross @xmath51 . the resulting dynamics of @xmath4 looks like neither scaling nor oscillating solution ; we call this a `` pseudo scaling solution '' . hereafter we consider the case of @xmath15 . before solving the equation of motion , we must specify the initial condition . the natural initial condition is set just after inflation as @xmath52 and @xmath53 , where @xmath54 and @xmath55 are the hubble parameter and the cosmic time at the end of the inflation , respectively . we assume @xmath56 without loss of generality . here @xmath6 should satisfy @xmath57 since otherwise @xmath4 rolls down the potential during inflation . this condition roughly means @xmath58 . therefore , in terms of @xmath31 , it is initially located close to the origin and rolls down the effective potential ( [ veff ] ) toward the minimum . in particular , we consider the case @xmath59 by noting that the initial velocity of @xmath31 is given by @xmath60 we obtain the conserved energy density of @xmath31 : @xmath61 here we have used the approximation @xmath62 which is valid as long as @xmath63 . since the energy density is negative , @xmath31 does not overshoot the origin @xmath50 , which means that @xmath4 does not oscillate around @xmath51 . the rescaled field @xmath31 just oscillates around @xmath33 without damping .. although it holds during inflation , the relation is modified after inflation depending on the effective equation of state of the inflaton oscillation dominated universe as @xmath64 for @xmath65 . the approximation @xmath66 as an initial condition is valid for inflation models in which the inflaton mass scale is much larger than the hubble scale , such as new inflation . we have numerically checked that @xmath67 is indeed negative for such a case . on the other hand , for chaotic inflation , our approximation is not justified and @xmath67 may be positive . however , the dynamics is almost identical to the case of negative @xmath67 after @xmath4 is replaced with @xmath19 even if @xmath4 once overshoots @xmath51 as long as @xmath68 is satisfied . ] for @xmath15 for the potential @xmath0 for @xmath69 in the left panel and @xmath70 in the right panel . we have normalized @xmath4 by the planck mass @xmath71 . the time is normalized by @xmath72 , where @xmath73 is a natural mass scale of this system . the red solid line represents a numerical solution , which corresponds to a `` pseudo scaling '' solution , while the green dashed line shows a hypothetical scaling solution for representative purpose . , title="fig : " ] for @xmath15 for the potential @xmath0 for @xmath69 in the left panel and @xmath70 in the right panel . we have normalized @xmath4 by the planck mass @xmath71 . the time is normalized by @xmath72 , where @xmath73 is a natural mass scale of this system . the red solid line represents a numerical solution , which corresponds to a `` pseudo scaling '' solution , while the green dashed line shows a hypothetical scaling solution for representative purpose . , title="fig : " ] the dynamics looks so simple in terms of the rescaled field @xmath31 . this `` pseudo scaling '' solution is more nontrivial in terms of the original field @xmath4 . let us see fig.[fig : ps ] where we have numerically solved the equation of motion of @xmath4 with initial conditions @xmath74 for @xmath69 in the left panel and @xmath70 in the right panel . we have normalized @xmath4 by the planck mass @xmath71 . the time is normalized by @xmath72 , where @xmath73 is a natural mass scale of this system . the red solid line corresponds to a numerical solution , which corresponds to a pseudo scaling solution , while the green dashed line shows a scaling solution @xmath75 for representative purpose . for comparison , we also show time evolution of @xmath76 for cases with @xmath77 ( scaling ) , @xmath78 ( pseudo scaling ) and @xmath79 ( oscillating ) for @xmath80 in fig . [ fig : ps2 ] . it is clear from this figure that the pseudo scaling solution resembles neither the scaling solution nor the oscillating solution even qualitatively . to understand this behavior , let us solve the equation ( [ eqofm2 ] ) more carefully . our goal is to understand main features of the pseudo scaling solution and estimate how the periodicity depends on the initial condition @xmath81 . ) , but the following discussion is enough for our purpose . ] here it is important to note that the dynamics of @xmath31 is dominated by the region @xmath82 since @xmath83 in the case of our interest . is satisfied . ] it means that we can almost neglect the second term of eq . in order to understand main features of the dynamics . in the region @xmath82 , the equation of motion is approximated as @xmath84 this has a general solution consisting of the growing and decaying mode , @xmath85 where @xmath86 and @xmath87 are determined by the initial conditions . in terms of @xmath4 , this solution corresponds to @xmath88 let us divide the time into the first phase @xmath89 , the second phase @xmath90 and the third phase @xmath91 . here , we define @xmath92 as the time at which the motion of @xmath4 starts . the time when the motion of @xmath4 stops is denoted as @xmath93 . some time after @xmath94 the field @xmath4 again starts to move , and we define that time as @xmath95 ( see fig.[fig : ps ] ) . the @xmath96 term in the effective potential becomes important only for some short periods at around @xmath97 and @xmath98 . [ [ first - phase ] ] first phase + + + + + + + + + + + in the time interval @xmath99 , @xmath86 and @xmath87 are determined by the initial conditions : @xmath100 this solution breaks down at @xmath101 where @xmath31 becomes close to @xmath33 . after that @xmath31 passes through @xmath33 and goes back to @xmath102 within order one unit time in terms of @xmath29 . we can estimate @xmath103 and @xmath92 as @xmath104 thus the time @xmath92 is roughly the epoch at which the hubble parameter becomes equal to the effective mass of @xmath4 . it is obvious that @xmath4 remains almost constant during this period . [ [ second - phase ] ] second phase + + + + + + + + + + + + next , let us consider the interval @xmath90 . just what we have to do is to determine the constants @xmath86 and @xmath87 in ( [ sol ] ) with appropriate initial conditions . we set the initial condition at @xmath105 ( and @xmath106 ) , with @xmath107 taken somewhere between @xmath108 and @xmath33 , say , @xmath109 . the precise position of @xmath107 is not important since the dynamics is dominated by the region @xmath82 as we noted before . since @xmath67 is conserved during the whole process , we can express @xmath110 as @xmath111 where @xmath112 the sign of @xmath113 is flipped because @xmath31 changes the direction of motion due to the higher order term in @xmath114 at @xmath115 . then , defining @xmath116 and @xmath117 as the constants @xmath86 and @xmath87 in eq . at @xmath118 , we obtain @xmath119 hence @xmath120 . then the end of the second period , @xmath121 , is the time at which the decaying mode @xmath122 becomes equal to the growing mode @xmath123 . thus we obtain @xmath124 as @xmath125 where in the last equality we have approximated @xmath126 and @xmath127 . here we have included an @xmath128 constant @xmath129 which comes from the uncertainty of @xmath107 as well as effects of the @xmath96 term . we have numerically checked that the @xmath108 dependence of @xmath93 is precisely given by eq . for @xmath63 , or @xmath130 . it should be noticed that @xmath93 is orders of magnitude longer than the inverse mass scale of the scalar field . from ( [ solphi ] ) , we can see that @xmath4 scales as @xmath131 during this time interval , so we have @xmath132 and eventually at @xmath94 it becomes @xmath133 note that during this regime , @xmath4 satisfies @xmath134 , so the potential term becomes irrelevant and the effective mass scale @xmath135 becomes smaller and smaller than the hubble parameter . the energy density of @xmath4 is dominated by the kinetic term and scales as @xmath136 during this regime . [ [ third - phase ] ] third phase + + + + + + + + + + + lastly , let us consider the third period @xmath91 . this is just a continuation of the second period , where the growing mode dominates over the decaying mode . thus the solution is @xmath137 this is valid until it becomes close to @xmath33 . thus the end of the third period , @xmath138 is estimated as here is exactly the same as that in eq . since the dynamics of @xmath31 is periodic . ] @xmath139 therefore it is seen that the relative time interval @xmath140 is equal to @xmath141 . during this time interval , @xmath142 remains almost constant as is clear from ( [ solphi ] ) . [ [ section ] ] after @xmath143 , we have exactly the same dynamics as the second phase and the following third phase since there is no damping in the equation of motion of @xmath31 . this one `` pseudo - oscillation '' consisting of one set of the second and third period takes much longer time than the inverse mass scale of the scalar field and even than the hubble time scale . in terms of @xmath4 , the field value reduces by an amount of @xmath144 during this one pseudo oscillation . the scaling of @xmath4 consists of @xmath145 regime and @xmath146 regime , both are completely different from the scaling solution @xmath147 . the scaling solution is obtained only after averaging the pseudo oscillation , but the period of the pseudo oscillation is much longer than the hubble time scale , hence using the scaling solution would lead to qualitatively wrong conclusions on the physical implications such as the curvature perturbation . finally , we make some comments on qualitative differences between the scaling solution for @xmath148 and the pseudo scaling solution for @xmath149 . in the case of the scaling solution , the kinetic term , the hubble friction term and the potential term in the equation of motion are all comparable . therefore , if the potential is not monomial , the motion of @xmath4 soon deviates from the scaling solution once other terms than the original one dominate the potential . on the other hand , in the case of the pseudo scaling solution , the kinetic term and the hubble friction term dominate over the potential term . this can be seen by the fact that @xmath150 and @xmath151 are the solutions of the equation @xmath152 for @xmath153 . , it means that the quadratic term dominates the effective potential . ] thus , even after the potential is dominated by other terms than the one with @xmath149 , the motion of @xmath4 still follows the pseudo scaling solution for some period until the potential term becomes comparable to the kinetic and friction terms . changes , e.g. due to the inflaton decay , the pseudo scaling behavior terminates soon.]we will see this in more detail in sec . [ sec : cos ] . here we briefly comment on the case where the hubble mass term exists in the potential : @xmath154 the equation of motion of @xmath4 is given by @xmath155 the equation of motion of the rescaled field @xmath31 is the same as ( [ eqofm1 ] ) except that @xmath156 is now replaced with @xmath157 in particular , for @xmath15 , the equation of motion becomes @xmath158 as long as @xmath159 , the existence of the hubble mass term does not qualitatively affect the pseudo scaling dynamics studied in the previous subsection . if @xmath160 , the crucial difference from the case of @xmath159 is that the initial position of @xmath4 can not be taken freely . in particular , for positive @xmath161 , we have @xmath162 and there is no dynamics . let us consider the case of negative hubble mass : @xmath163 . in this case , @xmath4 initially sits at the minimum of the potential @xmath164 where @xmath165 this is the same order of @xmath33 during inflation , and hence @xmath31 begins oscillation around @xmath33 with an oscillation time scale of order one in terms of @xmath166 . note that the temporal minimum of the potential @xmath167 exhibits the same scaling as the scaling solution @xmath75 . thus @xmath76 follows the temporal minimum @xmath168 with oscillation around it . in this case , this oscillation time scale is close to the hubble time scale . this is qualitatively different from the case of @xmath159 in the previous subsection . let us see the dynamics more closely . in order to evaluate the initial amplitude @xmath169 around the temporal minimum @xmath167 , we must specify the origin of the hubble mass term of @xmath4 . it arises from the planck - suppressed interaction as @xmath170 where @xmath171 and @xmath172 are constants of order unity , @xmath173 denotes the inflaton filed and @xmath174 is the inflaton potential . during a slow - roll inflation , the potential energy of the inflaton dominates , hence we have @xmath175 . after inflation , the inflaton begins a coherent oscillation and the kinetic energy and potential energy are equally distributed , roughly speaking . in the case of harmonic oscillation of the inflaton , we have @xmath176 . therefore , unless @xmath177 , the temporal minimum @xmath167 just after inflation is different from that during inflation . this shift of the potential minimum occurs in the time scale of inflaton mass , rather than the hubble scale , so @xmath4 can not track the temporal minimum @xcite . thus the @xmath4 oscillation around the temporal minimum @xmath167 is induced just after inflation , and the amplitude is of the order of @xmath178 . , @xmath4 can not track the temporal minimum @xcite . ] another term is coupling to the ricci scalar @xmath179 : @xmath180 if this is a dominant origin of the hubble mass term , @xmath181 during inflation and @xmath182 after inflation for the harmonic inflaton oscillation . oscillates between positive and negative values . ] anyway , the oscillation amplitude around the temporal minimum @xmath169 decreases as the universe expands according to @xmath183 because of the number density conservation . by using this , we obtain the following scaling : @xmath184 the oscillation amplitude relatively decreases compared with the position of the temporal minimum for @xmath13 , while the ratio is constant for @xmath15 . thus @xmath4 safely relaxes to its temporal minimum for @xmath13 . for @xmath15 , @xmath4 oscillates around @xmath167 and never cross @xmath51 thereafter if @xmath185 initially . this fact is important to study the formation of topological defects , such as domain walls in the axion model@xcite . so far we have considered the dynamics of scalar field with @xmath15 having only the non - renormalizable potential @xmath0 and the hubble mass term . more realistic models may have other terms in the potential . in this section we consider cosmological implications of the scalar dynamics with @xmath15 based on the potential @xmath186 the presence of the hubble mass term like @xmath187 does not change the result much as long as @xmath188 . in this type of the potential , the dynamics of @xmath4 is eventually dominated by the mass term at @xmath189 , even if @xmath4 obeys a pseudo scaling law at first . this is easily seen by rewriting the equation of motion as @xmath190 since @xmath191 , the effect of the mass term becomes important at @xmath189 . note that even if the potential is dominated by the mass term , the scalar field does not always feel the mass term if @xmath4 obeys a pseudo scaling law as we explained at the end of sec . [ sec : ps ] . , where @xmath192 is defined below eq . ] fig . [ fig : mass ] shows numerical results for the time evolution of @xmath76 for @xmath78 for @xmath193 with a mass term . in the left ( right ) panel we have assumed positive ( negative ) mass squared . the red solid ( green dashed ) line corresponds to @xmath194 @xmath195 . other parameters and initial conditions are the same as those in fig . [ fig : ps ] . these results clearly show that the pseudo scaling law is maintained even after the mass term dominates the potential and the oscillation by the mass term starts at at @xmath196 . first consider the abundance and the curvature perturbation of @xmath4 . since @xmath4 is light during inflation , it obtains a quantum fluctuation of @xmath197 and @xmath4 may act as a curvaton - like field . however , it is stated in ref . @xcite that the curvature perturbation vanishes for both @xmath148 and @xmath15 because of the scaling behavior of the curvaton field . if @xmath4 obeys a scaling solution ( [ scaling ] ) , the field value of @xmath76 at some uniform density slice does not depend on the initial position @xmath6 : this is easily checked as @xmath198 hence the @xmath6 dependence vanishes . here @xmath199 is the time at which the effective mass becomes equal to the hubble scale ( see eq . ( [ t1 ] ) ) . thus curvature perturbation is not generated by @xmath4 with a scaling law . this is true for @xmath13 in which the scaling solution is an attractor , but for @xmath15 we have seen that the actual dynamics is far from the scaling law . for @xmath200 , @xmath4 scales as @xmath201 the initial @xmath6 dependence remains and hence the perturbation @xmath202 is not always erased . for concreteness , let us suppose that @xmath4 decays into radiation at @xmath203 with a constant decay rate @xmath204 after the dynamics is dominated by the quadratic mass term for @xmath200 and oscillates around the origin . we also assume that the inflaton energy density scales as @xmath205 until the curvaton enters in the quadratic oscillation regime , and decays into radiation before the curvaton decays , i.e. , we assume @xmath206 with @xmath207 being the inflaton decay rate . in this subsection we mainly consider the case of positive mass squared @xmath208 and give only results for the case of negative mass squared @xmath209 at the end . for details on the case of negative mass squared , we refer readers to app . [ sec : hill ] . the energy density of the curvaton at the onset of its oscillation , @xmath210 , is evaluated as @xmath211 where we have defined @xmath212 . it represents the boundary at which the mass term begins to dominate the potential . therefore , the fraction of the @xmath4 energy density at @xmath213 is given by @xmath214 this expression assumes that @xmath4 is subdominant at the decay . otherwise , it is equal to @xmath215 . now , let us evaluate the curvature perturbation generated in this scenario . from the expression ( [ rhom ] ) , we can see that the initial @xmath6 dependence explicitly remains . thus the curvature perturbation of the order of @xmath216 is generated if @xmath4 eventually dominates the universe . in the case of positive mass squared @xmath240 , to the second order in @xmath241 , we obtain the curvature perturbation as @xmath242\left(\frac{\delta\phi_i(\vec x)}{\phi_i } \right)^2,\end{aligned}\ ] ] where we have neglected the inflaton fluctuation . the non - linearity parameter of the local type is given by , @xmath243 \frac{n_\phi^4}{(n_i^2+n_\phi^2)^2}.\ ] ] here we also show the results in the case of negative mass squared @xmath209 . if the mass term dominates the dynamics during the regime @xmath244 , the curvature perturbation is given by @xmath245,\end{aligned}\ ] ] where @xmath246 is a numerical factor . the non - linearity parameter is given by @xmath247 for more details , see app . [ sec : hill ] . this is one example of the property of the curvature perturbation from the pseudo scaling curvaton . although these expressions depend on how the background evolves and when the curvaton decays , it generally holds that the curvature perturbation is generated for @xmath15 in which @xmath4 obeys a pseudo scaling law . next , we consider implications on the domain wall formation . consider the case of negative mass squared : @xmath248 . the potential has a @xmath249 symmetry @xmath250 if @xmath1 is even . if @xmath4 obtains a large initial value during inflation and @xmath251 , it oscillates around the symmetric point @xmath252 after inflation and large fluctuations of @xmath4 particles ( and other particles which interact with @xmath4 ) are induced due to the efficient particle production . such large fluctuations may cause the non - thermal restoration of the symmetry and @xmath4 may be trapped at the origin ( @xmath252 ) @xcite . the trapping effect becomes diluted as the universe expands , and @xmath4 finally rolls down to the minimum of the potential . eventually , the domain wall is formed and the universe is expected to be overclosed . this is the rough picture of the domain wall problem we consider here . a physically well - motivated example is the peccei - quinn ( pq ) field , which often leads to the axionic domain wall problem @xcite due to the nonthermal restoration of the pq symmetry @xcite . on the other hand , for @xmath253 , the scalar field @xmath4 obeys the pseudo scaling law . as we saw earlier , the energy density of @xmath31 is negative for a reasonable initial condition ( see eq . ) . hence @xmath4 never crosses the origin during the pseudo scaling and no particle production occurs . , no efficient particle production is expected to occur since the period of the oscillation is much longer than the cosmic time in the case of the pseudo scaling solution . ] moreover , even after the dynamics is dominated by the mass term , @xmath4 never crosses the origin because the mass term also prevents @xmath31 from crossing the origin ( see eq . ) . the field @xmath4 just oscillates around @xmath254 and the domain wall problem does not occur . therefore , the pseudo scaling solution provides a natural solution to the domain wall problem . here we emphasize that the pseudo scaling solution for @xmath253 is more powerful in solving the domain wall problem than the scaling solution for @xmath255 . to see this point , let us consider the following potential : @xmath256 where we assume @xmath257 and @xmath258 . first consider the scaling solution @xmath255 . in this case , if the initial value of @xmath4 is large enough , the field @xmath4 obeys the scaling solution after inflation . if @xmath259 , the dynamics is dominated by the @xmath260 term and @xmath4 starts to oscillate around the origin when @xmath261 . thus , for sufficiently large @xmath262 , the domain wall problem occurs even if @xmath4 obeys the scaling solution at first . on the other hand , in the case of the pseudo scaling solution @xmath253 , the dynamics becomes dominated by the @xmath260 term when @xmath263 where @xmath264 is defined as , and hence the dynamics is dominated by the @xmath260 term before the mass term becomes effective for @xmath265 . ] @xmath266 we can deduce this as follows . in terms of @xmath31 , the effective potential is given by @xmath267 the amplitude of @xmath31 is of the order of @xmath33 , and hence the @xmath268 term becomes effective once the condition @xmath269 is satisfied . this condition is equivalent to eq . . therefore , for example , if the @xmath260 term dominates the dynamics for the regime @xmath270 , the field value of @xmath4 at @xmath271 is given by @xmath272 if the initial amplitude satisfies @xmath273 , @xmath4 starts to oscillate with the amplitude much smaller than @xmath274 . thus , the domain wall problem is less likely to occur even in this case compared to the scaling solution . so far we have focused only on the domain wall problem , but similar implications can be obtained for other types of topological defects . we have shown that a scalar field with potential @xmath0 with @xmath15 exhibits a peculiar behavior which we call a pseudo scaling . it resembles neither oscillating nor scaling solution . we have also shown that the primordial curvature perturbation is generated by this pseudo - scaling curvaton . the critical exponent @xmath8 depends on the background expansion law : @xmath275 @xmath276 in the md ( rd ) universe . even a renormalizable term can lead to a pseudo - scaling dynamics , i.e. , @xmath277 is obtained for @xmath278 , where the hubble parameter is given by @xmath279 . this observation has implications on some realistic models and cosmological scenarios . in supersymmetric ( susy ) theories , for example , it is often the case that a scalar field has a non - renormalizable potential . an example is the affleck - dine ( ad ) field @xcite which consists of the flat direction in the mssm and have the superpotential of the form @xcite @xmath280 where @xmath4 denotes the ad field and the scalar potential is then given by @xmath281 where @xmath282 and @xmath283 represent the effect of susy breaking . if there is no hubble mass term or its coefficient is small , and the last term dominates the potential , the ad field may enter the pseudo - scaling regime . the @xmath284 flat direction , which is lifted up by the @xmath285 superpotential , matches the condition for the pseudo scaling if the background is matter - dominated as in the inflaton oscillation dominated era . in such a case , the lepton or baryon asymmetry generated by the ad mechanism will be significantly suppressed compared with the naive expectation , e.g. , the case of negative hubble mass term . another example is the susy axion model , in which the superpotential is given by @xcite @xmath286 where @xmath4 and @xmath287 are the pq field having pq charge @xmath288 and @xmath289 respectively . the scalar potential is @xmath290 where we have taken @xmath291 for simplicity . if @xmath292 , @xmath4 can have a large vev which may be suitable as a solution to the strong cp problem @xcite . in the case where the pq symmetry is broken before inflation , the axion field usually obtains a large fluctuation and may cause the isocurvature perturbation problem . one way to avoid the isocurvature problem is to consider the situation where the pq field takes a very large value during inflation @xcite . in such a case , the isocurvature perturbation gets suppressed due to the large expectation value of the pq field . however , since the initial field value is large , particles are expected to be effectively produced due to the oscillation of the pq field after inflation . as a result , the pq symmetry may be non - thermally restored and hence the domain wall problem may occur after the qcd phase transition @xcite . in this paper , we have seen that if the pq scalar obeys the pseudo scaling law after inflation , no such a disastrous situation is realized . thus , for example , if @xmath293 and the pq scalar starts to move during the inflaton oscillation dominated era ( with the mass term ) , the domain wall problem as well as the isocurvature perturbation problem can be naturally solved . this work was supported by the grant - in - aid for scientific research on scientific research a ( no.26247042 [ kn ] ) , young scientists b ( no.26800121 [ kn ] ) and innovative areas ( no.26104009 [ kn ] , no.15h05888 [ kn ] ) . this work was supported by world premier international research center initiative ( wpi initiative ) , mext , japan . the work of y.e . and m.t . was supported in part by jsps research fellowships for young scientists and the program for leading graduate schools , mext , japan . in this appendix we derive the expansion law @xmath294 or @xmath22 in the inflaton oscillation dominated universe . we assume that the inflaton potential around the potential minimum is approximated as @xmath295 where @xmath173 denotes the inflaton field measured from its potential minimum . the equation of motion of the inflaton field is given by @xmath296 from this we obtain @xmath297 where @xmath298 is the inflaton energy density . on the other hand , we have @xmath299 taking time average over the inflaton oscillation time period and assuming that the hubble parameter is much smaller than the inflaton mass scale , we obtain the virial theorem @xmath300 where @xmath301 indicates time average over the inflaton oscillation period . by using this relation , we obtain @xmath302 thus the energy density scales as @xmath303 on the other hand , the freedman equation @xmath304 implies @xmath305 . therefore , we obtain the expansion law @xmath306 for @xmath307 @xmath308 , the usual md ( rd)-like universe @xmath193 @xmath309 is realized . more unconventional expansion laws are also obtained , for example , @xmath310 for @xmath311 and @xmath312 for @xmath313 . in terms of @xmath314 , the critical exponent @xmath315 is given by @xmath316 in this appendix , we give a formal solution of the equation of motion . note that the energy density of @xmath31 @xmath317 is conserved . thus , the kinetic term of @xmath31 is given by @xmath318 by integrating this , we obtain the formal solution of @xmath31 as @xmath319 for @xmath320 region and @xmath321 for @xmath322 region . here @xmath323 and @xmath324 are some initial values of @xmath29 and @xmath31 , respectively . in particular , the period of the oscillation of @xmath31 is given by @xmath325 where @xmath326 are the solutions of the equation @xmath327 with @xmath328 .. for @xmath329 , we must replace @xmath330 by zero and also double the right hand side of the equation when @xmath1 is even . ] note that @xmath326 as well as @xmath331 depends on the initial condition . however , it is difficult to solve this equation analytically . moreover , the approximation given in sec . [ sec : ps ] is sufficient for our purpose in this paper . in this section we briefly review the curvature perturbation generated by the curvaton with a hilltop potential of the form @xmath332 as shown in sec . [ sec : cos ] , if the initial position @xmath6 is set @xmath333 , where @xmath334 is the minimum position of the potential and @xmath335 , @xmath4 obeys a pseudo scaling law until @xmath21 becomes comparable to @xmath282 . for concreteness , let us suppose that this breaking of pseudo scaling law happens at @xmath91 . the field value at this stage is given by ( [ phit2 ] ) : @xmath336 and this can be much smaller than @xmath192 . the scalar dynamics thereafter is the same as that with a scalar potential ( [ tachyon ] ) with initial position @xmath337 close to the origin . in other words , the pseudo scaling scalar dynamics naturally provides us with a hilltop initial condition . in the following , we treat @xmath338 as if it is the `` initial '' condition of the dynamics , although the true initial position is @xmath6 . without loss of generality . ] as in sec . [ sec : abu ] , let us assume @xmath339 . the abundance of the curvaton at @xmath213 is evaluated as @xmath340 this expression assumes that @xmath4 is subdominant at the decay . otherwise , it is equal to 1 . here @xmath341 and @xmath342 denote the hubble parameter and the curvaton field value at the onset of its oscillation , respectively . what is nontrivial in the hilltop scenario is that @xmath341 strongly depends on the initial condition @xmath337 in the hilltop limit @xmath343@xcite . let us express @xmath341 and @xmath342 in terms of @xmath337 . before the onset of oscillation , we can neglect the second term of ( [ tachyon ] ) . the formal solution of the equation of motion with a tachyonic mass term @xmath344 and @xmath22 is written as @xmath345 where @xmath346 is the modified bessel function of the first kind . in general , the solution is a superposition of the modified bessel function of the first and second kind . however , without loss of generality we can consider the region @xmath347 and @xmath348 , and hence the bessel function of the first kind is sufficient to describe the dynamics . the solution is conveniently approximated as @xmath349 & ~~~{\rm for}~~~mt \ll 1,\\ \displaystyle \phi _ * \frac{2^\nu\gamma(\nu+1)}{\sqrt{2\pi } } \frac{e^{mt}}{(mt)^{3p/2}}&~~~{\rm for}~~~mt \gg 1 . \end{cases } \label{phi_sol}\ ] ] first note that the velocity of the scalar field is given by , we have a solution for a general potential : @xmath350 . ] @xmath351 since the oscillation epoch is significantly delayed in the hilltop limit @xmath343 , i.e. , @xmath352 , the condition of the start of the oscillation is given by @xmath353 on the other hand , from eq . ( [ phi_sol ] ) , we obtain @xmath354 for reasonable choice of @xmath355 , we have @xmath356 . this justifies our assumption @xmath357 . then following relations are derived , @xmath358 using these equations and noting that @xmath359 , we obtain @xmath360 where we have taken only leading terms in @xmath361 . hence the curvature perturbation ( [ zeta ] ) is given by @xmath362,\end{aligned}\ ] ] to the leading order in @xmath361 . if @xmath337 would be a true initial condition set during inflation , then @xmath363 and this expression gives a correct estimate for the curvature perturbation from the hilltop curvaton . in this case , the non - linearity parameter of @xmath364 for @xmath365 is a generic prediction@xcite . in our pseudo - scaling curvaton case , the true initial position is @xmath6 and it is @xmath202 that obtains a quantum fluctuation of @xmath366 . from the relation ( [ phi_ini ] ) , we obtain @xmath245 . \end{aligned}\ ] ] the non - linearity parameter is given by @xmath247 therefore , if the inflaton contribution to the curvature perturbation is negligible , we have @xmath367 for @xmath365 . thus in this case the non - linearity parameter may become sizable . 99 a. h. guth , phys . d * 23 * , 347 - 356 ( 1981 ) ; a. a. starobinsky , phys . b * 91 * ( 1980 ) 99 ; k. sato , mon . not . soc . * 195 * , 467 - 479 ( 1981 ) ; a. d. linde , phys . b * 108 * ( 1982 ) 389 ; a. albrecht and p. j. steinhardt , phys . * 48 * , 1220 ( 1982 ) . m. s. turner , phys . d * 28 * , 1243 ( 1983 ) . s. mollerach , phys . d * 42 * , 313 ( 1990 ) ; a. d. linde and v. f. mukhanov , phys . d * 56 * , 535 ( 1997 ) [ astro - ph/9610219 ] ; k. enqvist and m. s. sloth , nucl . b * 626 * , 395 ( 2002 ) [ hep - ph/0109214 ] ; d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) [ hep - ph/0110002 ] ; t. moroi and t. takahashi , phys . b * 522 * , 215 ( 2001 ) [ erratum - ibid . b * 539 * , 303 ( 2002 ) ] [ hep - ph/0110096 ] . k. dimopoulos , g. lazarides , d. lyth and r. ruiz de austri , phys . rev . d * 68 * , 123515 ( 2003 ) [ hep - ph/0308015 ] . k. enqvist , s. nurmi , g. rigopoulos , o. taanila and t. takahashi , jcap * 0911 * , 003 ( 2009 ) [ arxiv:0906.3126 [ astro-ph.co ] ] ; k. enqvist , s. nurmi , o. taanila and t. takahashi , jcap * 1004 * , 009 ( 2010 ) [ arxiv:0912.4657 [ astro-ph.co ] ] . c. t. byrnes , k. enqvist , s. nurmi and t. takahashi , jcap * 1111 * , 011 ( 2011 ) [ arxiv:1108.2708 [ astro-ph.co ] ] . k. mukaida , k. nakayama and m. takimoto , phys . d * 89 * , no . 12 , 123515 ( 2014 ) [ arxiv:1402.1856 [ astro-ph.co ] ] . a. r. liddle and r. j. scherrer , phys . d * 59 * , 023509 ( 1999 ) [ astro - ph/9809272 ] . r. d. peccei and h. r. quinn , phys . lett . * 38 * , 1440 ( 1977 ) . i. affleck and m. dine , nucl . b * 249 * , 361 ( 1985 ) . m. dine , l. randall and s. d. thomas , nucl . b * 458 * , 291 ( 1996 ) [ hep - ph/9507453 ] . k. harigaya , m. ibe , m. kawasaki and t. t. yanagida , arxiv:1507.00119 [ hep - ph ] . k. nakayama , f. takahashi and t. t. yanagida , phys . d * 84 * , 123523 ( 2011 ) [ arxiv:1109.2073 [ hep - ph ] ] . a. a. starobinsky , jetp lett . * 42 * , 152 ( 1985 ) [ pisma zh . . fiz . * 42 * , 124 ( 1985 ) ] ; m. sasaki and e. d. stewart , prog . phys . * 95 * , 71 ( 1996 ) [ astro - ph/9507001 ] ; m. sasaki and t. tanaka , prog . phys . * 99 * , 763 ( 1998 ) [ gr - qc/9801017 ] ; d. h. lyth , k. a. malik and m. sasaki , jcap * 0505 * , 004 ( 2005 ) [ astro - ph/0411220 ] ; m. sasaki , j. valiviita and d. wands , phys . d * 74 * , 103003 ( 2006 ) [ astro - ph/0607627 ] . l. kofman , a. d. linde and a. a. starobinsky , phys . lett . * 76 * , 1011 ( 1996 ) [ hep - th/9510119 ] . i. i. tkachev , phys . b * 376 * , 35 ( 1996 ) [ hep - th/9510146 ] . for a review , see m. kawasaki and k. nakayama , ann . nucl . part . sci . * 63 * , 69 ( 2013 ) [ arxiv:1301.1123 [ hep - ph ] ] . s. kasuya , m. kawasaki and t. yanagida , phys . b * 409 * , 94 ( 1997 ) [ hep - ph/9608405 ] ; phys . b * 415 * , 117 ( 1997 ) [ hep - ph/9709202 ] ; m. kawasaki , t. t. yanagida and k. yoshino , jcap * 1311 * , 030 ( 2013 ) [ arxiv:1305.5338 [ hep - ph ] ] . t. moroi , k. mukaida , k. nakayama and m. takimoto , jhep * 1306 * , 040 ( 2013 ) [ arxiv:1304.6597 [ hep - ph ] ] ; jhep * 1411 * , 151 ( 2014 ) [ arxiv:1407.7465 [ hep - ph ] ] . t. gherghetta , c. f. kolda and s. p. martin , nucl . b * 468 * , 37 ( 1996 ) [ hep - ph/9510370 ] . h. murayama , h. suzuki and t. yanagida , phys . b * 291 * , 418 ( 1992 ) . for a review , see j. e. kim , phys . rept . * 150 * , 1 ( 1987 ) ; j. e. kim and g. carosi , rev . * 82 * , 557 ( 2010 ) [ arxiv:0807.3125 [ hep - ph ] ] . a. d. linde and d. h. lyth , phys . b * 246 * , 353 ( 1990 ) ; a. d. linde , phys . b * 259 * , 38 ( 1991 ) . m. kawasaki , k. nakayama and f. takahashi , jcap * 0901 * , 026 ( 2009 ) [ arxiv:0810.1585 [ hep - ph ] ] . m. kawasaki , t. kobayashi and f. takahashi , phys . d * 84 * , 123506 ( 2011 ) [ arxiv:1107.6011 [ astro-ph.co ] ] .
two - band superconductivity was first proposed in the late 1950 s and further studied in the 1960 s and 1970 s as a possible explanation for understanding superconductivity in s- and d- band metals . however , it was not until the discovery of superconductivity at relatively high temperature in mgb@xmath5 in 2001 , that the most promising example of two - band superconductivity was found . after nearly four years of effort and the development of an extensive literature based on this material , it is now evident that this compound is an electron - phonon mediated superconductor , but with two distinct bands and hence two energy gaps . indeed , this is firmly established from detailed bandstructure calculations that provide the electron - phonon spectral densities which , when in turn are used in a two - band eliashberg formalism , give rise to predictions which are in accord with experiment . in fact , the picture that emerges is one of excellent agreement with the data , with the match being as good as is found between theory and experiment in conventional one - band superconductors@xcite . a detailed review and summary of the level of agreement found for the thermodynamic properties , bcs ratios , penetration depth and temperature - dependent energy gaps is found in ref . with such an extensive comparison between rigorous eliashberg theory and experiment completed , it is now appropriate to move on to the transport properties where less theoretical work has been done . in this paper , we consider the critical current which is of fundamental interest in the discussion of transport . while most technological applications focus on the bulk critical current density , which requires understanding of flux lattice pinning , etc . , here we are presenting the critical current in thin films . this geometry is more relevant to understanding issues associated with the fundamental superconducting state , rather than focussing on material details that lead to vortex pinning , etc . , and hence can be expected to provide further insight into the nature of multiband superconductivity in these materials . we note in passing , at this point , that some authors use the term depairing current for the critical current although , technically , the depairing current is the current at which the first cooper pair is broken and the gap decreases , and the critical current is the maximum current possible , which is greater than the depairing current . in simple one - band s - wave bcs theory the values are almost the same and hence there is the tendency to use the terms interchangeably . here , these two quantities can be quite different and so we will use the term critical current to refer to the global maximum obtainable in the current density . the full temperature - dependence of the critical current @xmath2 was first studied in conventional one - band s - wave bcs superconductors by rogers@xcite and bardeen@xcite . further work following on this was done by parameter@xcite for the clean limit , and by maki@xcite for the dirty limit , where the latter author introduced a description of the superfluid current density using the green s function formalism . a study of the effect of of a current on the quasiparticle states density of states of a superconductor was done by fulde@xcite . the most general work was that of kupriyanov and lukichev@xcite , who calculated the full temperature dependence of the critical current in bcs theory for arbitrary impurity scattering , which returned the usual ginzburg - landau results for the critical current near @xmath6@xcite . further considerations led to an initial attempt to include strong inelastic scattering@xcite followed by a full eliashberg calculation and study of strong coupling effects for conventional electron - phonon superconductors done by ourselves@xcite . thus , the formalism and basic one - band results have been well - established in both the weak - coupling bcs limit and strong - coupling eliashberg regime , at all temperatures and for arbitrary impurity scattering . more recently there has been a renewed theoretical effort on the critical current as applied to exotic gaps , which are anisotropic , such as d - wave@xcite and f - wave@xcite . the two - band case is , in some sense , a highly anisotropic system and hence the study of the critical current in this instance also complements these recent works . with regard to mgb@xmath5 , specifically , there has been some recent experimental work by kunchur@xcite , where the critical current has been measured in these systems . on the theoretical side , koshelev and golubov@xcite have gone beyond ginzburg - landau to use the usadel equations to study this system and find interesting effects such as a kink in the temperature - dependence of the c - axis critical current that reflects the underlying two - band nature . this type of feature is observed in other properties such as the penetration depth which also shows an inflection point in the temperature dependence , however , these effects may disappear with increased interband coupling . finally , a more recent work has been done on the topic of the non - linear term in the superfluid current in mgb@xmath5 and its effect on the nonlinear microwave response , which is relevant for device applications@xcite . in our work here , we use the general green s function formulation presented by maki@xcite where the cooper pairs are given a finite center of mass momentum @xmath1 . this @xmath1 leads to a boost in the quasiparticle energy ( sometimes called a doppler shift ) . the superfluid velocity @xmath7 is taken as uniform in the case of discussing currents in thin films or wires , unlike the doppler shift effect in the context of superfluid currents in the vortex state , where @xmath8 varies with spatial position . this latter topic has been the subject of a large number of papers in the recent literature relating to d - wave superconductors@xcite . a very fine physical description of the effect of this doppler shift on the quasiparticle density of states and the resulting effect on the superfluid density is presented in a paper by xu et al.@xcite , where they consider the non - linear meissner effect in d - wave superconductors from a starting point that discusses the basics of the superfluid current density . thus , with a body of literature developing on the topic of critical currents and superfluid currents , it is of interest to elucidate the features in the critical current due to two - band superconductivity , where unusual effects can occur due to a transfer from one band to the other of the dominant contribution to superfluid current and hence the critical current . our paper is structured in the following manner . in section ii , we introduce the basic theoretical equations which we have evaluated . we then examine , in section iii , results for renormalized bcs theory near @xmath6 and at @xmath9 and apply these results to mgb@xmath5 . in section iv , we present our eliashberg results for finite temperature and with impurities , evaluating the critical current for mgb@xmath5 and other model parameters . we also compare our results to the data for mgb@xmath5 . finally , we form our conclusions in section v. while various approaches have been used in the past for calculating the superfluid current , we will use the green s function method as presented by maki@xcite . thus the superfluid current density @xmath10 is given in terms of finite temperature green s functions as : @xmath11\quad , \label{eq : js1}\ ] ] where @xmath12 is the electron charge , @xmath13 , the electron mass , @xmath14 is the electron momentum , and @xmath15 is the temperature . the green s function is given in terms of the matsubara representation by @xmath16^{-1 } , \label{eq : greensfn}\ ] ] where @xmath17 , @xmath18 is the chemical potential , and the @xmath19 and @xmath20 are the pauli matrices for the particle - hole and spin spaces , respectively . the @xmath21 is the applied superfluid momentum and it results in a shift of the quasiparticle energies , given to first order in @xmath1 by @xmath22 . here , @xmath23 and @xmath24 are the imaginary axis renormalized superconducting order parameter and renormalized matsubara frequencies , respectively , and are defined below . the integral over energy @xmath25 for the combined eqns . ( [ eq : js1 ] ) and ( [ eq : greensfn ] ) can be done and the result , generalized to two bands , is given in terms of a sum of two partial currents @xmath26 and @xmath27 , for the first and second band , respectively : @xmath28 with @xmath29 where @xmath30 , @xmath31 is the electron density and @xmath32 is the fermi velocity of the @xmath33th band ( @xmath34,2 ) . the nonlinear eliashberg equations for @xmath35 and @xmath36 have been generalized to two bands and with the inclusion of the effect of the @xmath1 , they are given as@xcite : @xmath37\nonumber\\ & \times&\int_{-1}^{1}\frac{dz}{2 } \frac{\tilde\delta_j(m)}{\sqrt{(\tilde\omega_j(m)-is_jz)^2+\tilde\delta_j^2(m)}}\nonumber\\ & + & \pi\sum_j(t^+_{ij}-t^-_{ij})\int_{-1}^{1}\frac{dz}{2 } \frac{\tilde\delta_j(n ) } { \sqrt{(\tilde\omega_j(n)-is_jz)^2+\tilde\delta_j^2(n)}}\nonumber\\ \label{eq : del}\end{aligned}\ ] ] and @xmath38 where @xmath39 and @xmath40 are the ordinary and paramagnetic impurity scattering rates , respectively , and the electron - phonon kernels @xmath41 with phonon energy @xmath42 enter through @xmath43 here , the @xmath44 indexes the @xmath44th matsubara frequency @xmath45 , with @xmath46 , where @xmath47 . the @xmath48 are coulomb repulsions , with a high energy cutoff @xmath49 usually taken to be about six to ten times the maximum phonon frequency . as written , eqns . ( [ eq : js2])-([eq : z ] ) are for three dimensions with isotropic fermi surface . for general anisotropic fermi surface , a fermi surface integral would remain , which would need to be done numerically@xcite . this goes beyond the scope of this paper , however , such details are not expected to change the qualitative features of our results but rather lead to quantitative changes only . in the following sections , we present both eliashberg results and renormalized bcs ( rbcs ) results , the latter of which can be quite successful at capturing the essential features of the eliashberg calculations without the full numerical complications of the more sophisticated calculations even for two - band models@xcite . to develop the rbcs results we use the two - square - well approximation or @xmath50 model written for two - bands ( for details see ref . @xcite ) , where now @xmath51\delta_j(t)}{\sqrt{(\omega_m - i\bar s_jz)^2+\delta_j^2}}. \label{eq : rbcsdel}\ ] ] in this expression , @xmath52 , @xmath53 and @xmath54 with @xmath55 taken to represent either the debye frequency or some other characteristic energy scale representing the phonons in the system . likewise , the current from eq . ( [ eq : js2 ] ) can be reduced to @xmath56 equations ( [ eq : rbcsdel ] ) and ( [ eq : rbcsjs ] ) will be used in the next section to develop analytic results for comparison with full numerical calculation . in order to evaluate the critical current near @xmath6 , we need to know the effect of the current on the gap near @xmath6 . in this limit , both @xmath57 and @xmath1 will be small . after long but straightforward algebra , to first order in @xmath58 where @xmath59 is the reduced temperature , the gap equation of eq . ( [ eq : rbcsdel ] ) reduces to the form @xmath60 with a second equation obtained from eq . ( [ eq : bcs1 ] ) by switching indices @xmath61 . here @xmath62 and @xmath63 where @xmath64 and @xmath65 . for @xmath66 as @xmath67 and with no superfluid momentum @xmath68 , we recover the equation for the critical temperature ( symmetric in @xmath61)@xcite : @xmath69 with @xmath70 . throughout this paper , we will find it instructive to consider several simplifying limiting cases . the completely decoupled case corresponds to @xmath71 and the assumption that @xmath72 . in this case , the first channel determines the @xmath6 , which is given by @xmath73 . as an aside , we note that the small separable @xmath74 anisotropy model studied extensively in the older literature@xcite maps on to this model if @xmath75 , @xmath76 and @xmath77 , with the anisotropy @xmath74 assumed small . in this case , eq . ( [ eq : bcs3 ] ) simplifies greatly since @xmath78 so that @xmath79 = ( 1+\lambda)/[\lambda(1+a^2)]$ ] which gives @xmath80)$ ] . this is the well - known result that anisotropy increases the value of the critical temperature over its isotropic value ( @xmath81 ) . returning to eq . ( [ eq : bcs1 ] ) which gives the gap just below @xmath6 , we need to specify @xmath82 . it is given ( after more lengthy algebra ) as @xmath83 and an equivalent expression for @xmath84 is obtained from eq.([eq : bcs4 ] ) by switching indices @xmath61 . solving for @xmath85 , we find @xmath86 with @xmath87 \label{eq : bcs6}\ ] ] and @xmath88 . \label{eq : bcs7}\ ] ] likewise , the gap @xmath89 , is formed from the above three equations by exchanging @xmath61 . when @xmath90 , these equations properly reduce to those obtained previously by nicol and carbotte@xcite . in the limit of decoupled bands , @xmath91 and @xmath92 . some care is required when treating the equivalent equations for @xmath93 and @xmath94 because the combination @xmath95 is replaced by @xmath96 , which is zero . nevertheless , we can show @xmath97 , which is expected on physical grounds as the second band does not contribute near @xmath6 , where @xmath98 for @xmath99 , and @xmath100 $ ] , a well - known result . we can also recover other well - known results for the small separable anisotropy model defined previously . note first that when the fermi velocities are equal , @xmath101 in eq . ( [ eq : bcs7 ] ) reduces to @xmath102 , with @xmath103 given by eq . ( [ eq : bcs6 ] ) . further , @xmath104 and @xmath105 which leads to ( eq . ( [ eq : bcs5 ] ) ) @xmath106 \label{eq : bcs8}\ ] ] and an identical equation for the second gap with the first factor of @xmath107 replaced by @xmath108 . for @xmath90 , eq . ( [ eq : bcs8 ] ) reduces to the known result that the average gap @xmath109 with @xmath110 and @xmath111 . for the isotropic case ( equivalent to one band ) @xmath81 and eq . ( [ eq : bcs8 ] ) reduces to @xmath112 as is well known . the expression for the contribution to the current from the @xmath33th band is given in renormalized bcs by eq . ( [ eq : rbcsjs ] ) . expanding near @xmath113 for small @xmath1 and @xmath57 gives @xmath114 the total current is the sum over both bands and can be written as @xmath115 \frac{7\zeta(3)}{4(\pi t_c)^2 } eq_s , \label{eq : bcs13}\ ] ] where @xmath116 . substituting into eq . ( [ eq : bcs13 ] ) the new expression from eq . ( [ eq : bcs5 ] ) ( and equivalent for @xmath61 ) for @xmath85 , one obtains two terms . the first proportional to @xmath1 and the second to @xmath117 . this gives the total current @xmath0 as a function of @xmath1 . the critical current @xmath118 is obtained as the maximum of @xmath119 as a function of @xmath1 . after finding the extremum value of @xmath1 , we obtain for @xmath118 @xmath120^{3/2 } \biggl [ \frac{n_1}{m_1^*}\frac{1}{\chi_1^\prime}+\frac{n_2}{m_2^ * } \frac{1}{\chi_2^\prime}\biggr]^{-1/2}. \label{eq : bcs14}\end{aligned}\ ] ] we first note that the standard ginzburg - landau @xmath121 temperature dependence of the critical current remains unmodified in the two - band model . as a first check on our expression in eq . ( [ eq : bcs14 ] ) , we can recover the known one - band result when both bands are assumed to be identical by taking @xmath122 , @xmath123 , and the value of @xmath124 . here @xmath125 is the total electron density per unit volume equal to the sum @xmath126 . we obtain the standard ginzburg - landau result@xcite : @xmath127 note that all effective mass renormalization @xmath128 factors have cancelled in eq . ( [ eq : bcs15 ] ) . applying the separable gap anisotropy model to eq . ( [ eq : bcs14 ] ) leads to a modification of eq . ( [ eq : bcs15 ] ) by a multiplicative factor of @xmath129 . this gap anisotropy will decrease @xmath118 over its value for @xmath130 . decoupled bands give the usual expression of eq . ( [ eq : bcs15 ] ) since the second band makes no contribution near @xmath113 , but now @xmath44 , @xmath13 , and @xmath131 are those of the first band only . in the general case , we need to return to eq . ( [ eq : bcs14 ] ) which tells us how material parameters affect the slope of @xmath132 which varies linearly with @xmath58 near @xmath6 . if we normalize the current , as we will in the next section , to a slope of negative one near @xmath6 , this will mean that the current at @xmath9 will vary with material parameters in contrast to the one - band bcs curve for which it is a universal number for the clean limit case that we are discussing here . a well known result of ginzburg - landau theory is that near @xmath6 , @xmath133 , where @xmath134 is the thermodynamic critical field and @xmath135 is the london penetration depth , both for the case of zero current . in a previous paper@xcite , we have given expressions for @xmath134 and @xmath136 as @xmath99 . using these with eq . ( [ eq : bcs14 ] ) , we arrive at the expression relating the two - band @xmath4 , @xmath3 and @xmath2 : @xmath137 where @xmath138 is the velocity of light and @xmath139 is a complicated material - dependent parameter . it is given by @xmath140^{-1 } \label{eq : bcs18}\ ] ] where we have found it convenient to change from electron density @xmath31 , to density of states at the fermi surface @xmath141 . this modifies the usual ginzburg - landau relation by a constant factor . for the decoupled case where @xmath142 , @xmath143 , and @xmath144 , @xmath139 becomes equal to 1 . there is no change in the ginzburg - landau relation and this makes sense since near @xmath145 we are dealing with the single band ( the first one ) . for the isotropic gap case , where @xmath122 for any @xmath146 , from eqs . ( [ eq : bcs6 ] ) and ( [ eq : bcs7 ] ) we obtain @xmath147 and @xmath148 , the average of the squares of the fermi velocities in the two bands , which gives @xmath149 for identical bands , this expression gives 1 , as it must . this also holds if either the two fermi velocities are the same or the two density of states match , but not if both parameters for one band differ from those of the other band . we see that , in general , @xmath139 can be either greater or less than one and the ginzburg - landau relation ceases to hold . for identical bands but with gap anisotropy in the separable small anisotropy @xmath74 model , @xmath150 and @xmath151 , with @xmath104 and @xmath105 which gives @xmath152 to order @xmath74 . this result is expected since , as is well known , the @xmath74 anisotropy correction to @xmath134 and @xmath153 are both of the form @xmath154 which gives a factor of @xmath129 in the ratio @xmath155 . this agrees perfectly with the anisotropy factor we derived earlier for @xmath118 . thus , in this case , the ginzburg - landau relation will hold . we wish to comment on the use of the ginzburg - landau relation @xmath133 for comparison with data at all temperatures , which is done by substituting two - fluid model forms for @xmath134 and @xmath135 . the ginzburg - landau result should only hold near @xmath6 and indeed we find that when we form the ratio of @xmath155 using the @xmath134 and @xmath135 from the full eliashberg procedure , the resulting @xmath118 curve rapidly deviates from the eliashberg calculated @xmath118 . the results of using such a phenomenological procedure can give curves which are lower or higher than the true result , and will not even reproduce the qualitative feature of a kink in the @xmath118 that we have found in the two - band calculation . we conclude that the ginzburg - landau procedure should only be used near @xmath6 as is appropriate to its derivation and no reliable use can be made of it phenomenologically for lower temperatures . we now turn to low temperature and examine results of rbcs at @xmath9 . at zero temperature , the renormalized bcs gap equation ( [ eq : rbcsdel ] ) takes on the form @xmath156 where the barred quantities are renormalized by the factor of @xmath157 of eq . ( [ eq : rbcsz ] ) and @xmath158 ( 250,200 ) these equations need to be solved numerically . results are shown in the four lefthand panels of fig . the top frame is for @xmath159 , @xmath160 , @xmath161 ( nearly decoupled bands ) and @xmath123 . because @xmath162 is so small , the larger gap @xmath163 exhibits the known behavior as a function of @xmath164 of a one - band s - wave superconductor@xcite with a small modification . in one - band , the gap would be unchanged until @xmath165 at which point it would begin a sharp drop toward zero which is reached at @xmath166 . here , the same behavior occurs but there is a very slight drop from the @xmath90 value of @xmath57 at @xmath167 due to the coupling to the second band . the lower gap behaves differently . it retains its @xmath90 value until @xmath167 at which point it also begins a rapid drop but retains a small finite value until the point when @xmath168 . this small tail is due to the coupling @xmath169 which guarantees that the second gap is nonzero , although it can be very small , as long as the first gap is finite . in the second frame @xmath169 has been increased to 0.1 . this results in a considerable integration of the two gaps . at @xmath90 the lower gap ratio @xmath170 has increased from the 0.4 value of the top frame to 0.57 . more significantly for the present paper , as @xmath1 increases both gaps retain their @xmath90 value until @xmath171 , which is the point at which the second gap starts to decrease , but this drop is not as fast as in the top frame because of the large interband coupling . at the same point as the second gap starts to decrease , the first is also reduced but by much less . beyond @xmath165 both gaps drop sharply as expected and reach zero together at @xmath172 , a value which is smaller than the one band value of 1.35 . this shift in zero along the horizontal axis is also accompanied by a shift to the left , relative to the decoupled case , of the inflection points just described , but these shifts are small . these are signatures of the two - band nature of our system . in the third frame of fig . [ fig1 ] ( lefthand side ) , we have retained @xmath173 but now have increased the ratio @xmath174 to 2.0 instead of 1.0 . this has the effect of reducing the region of @xmath1 in which the second gap remains constant at its @xmath90 value and is due to the increase in @xmath175 . for the last frame , we have kept the parameters as in the first frame but have increased @xmath176 to a value of 0.7 . this has the effect of increasing the @xmath90 value of the second gap and also increasing the range over which it stays constant . additionally , the tail beyond @xmath167 is not as small as it is in the top frame due to less gap anisotropy . ( 250,200 ) in renormalized bcs , eq . ( [ eq : rbcsjs ] ) for the partial current can be evaluated analytically at zero temperature to obtain @xmath177^{3/2}\biggr],\nonumber\\ & & \qquad\qquad\qquad \qquad { \rm for } \frac{\bar s_i}{\delta_{i0}}>1\end{aligned}\ ] ] results for the current are presented in the righthand side frames of fig . [ fig1 ] for the four sets of parameters considered on the lefthand side . in all cases , the dotted lines give the results for the partial contribution to the current coming from the second band and the dashed for those from the first . the sum of the two partial contributions is the solid curve . a first feature to note is that for the three frames for which @xmath173 ( nearly decoupled ) , the dashed curves are essentially identical . the straight line segment has slope 0.5 in our units and the zero is very nearly at @xmath178 as in the one - band case . note that on the vertical axis , we have used the total electronic density of states @xmath179 with @xmath180 . if we had used only @xmath181 then the slope of the straight line segment would have been 1 which corresponds to what is expected in the one - band case . a second point to note is that increasing @xmath169 to 0.1 as in the second frame from the top has not had a large effect on the partial current from the first band except to reduce somewhat the value of @xmath1 at which the current goes to zero . this is accompanied by a small shift to the left in the position of the peak and a small reduction in its height . by contrast , the effect of the parameter changes are much more significant on the dotted curve ( the current in the second band ) . in the top frame , the slope of the straight line portion of the dotted curve is a little larger than 0.5 because of the renormalization factors on @xmath175 relative to @xmath182 . more significantly the partial current starts to deviate from linearity at @xmath183 at which point it drops rapidly towards zero but remains finite ( although small ) until the current in the first band has reached zero . this also applies to the third frame although in that case the current in the second band has been increased in value because of the increase in @xmath184 by a factor of two . in all cases the total current ( solid curve ) exhibits two peaks . the peak arising at lower values of @xmath1 has a significant contribution from both bands while the other peak at higher values of @xmath1 is due almost entirely to the partial current of the first band when @xmath169 is very small . on the other hand , when there is more integration of the two systems ( second frame from top ) both bands can make a significant contribution . in the top three frames , the first peak is lower than the second while in the fourth frame , the opposite holds . the height of the first peak can be made higher by increasing the value of @xmath184 as in the third frame compared with the first ( or equivalently , increasing @xmath185 as the density of states enters in combination with @xmath131 and so will have a similar effect , which is redundant to discuss here ) or by increasing @xmath176 as in the fourth , again compared with the first . it can also be increased by increasing the offdiagonal coupling as in the second frame . when this is done , however , the contribution of the second band to the total current in the second peak is also increased . finally we note that the peak heights in the two middle frames are only very slightly different and the lower @xmath1 peak could be made to be the highest through a small tuning of the parameters . as the highest peak determines the value of @xmath2 , this transfer from one peak to the other by varying parameters is very interesting and will produce nonstandard effects at finite temperature . before leaving our discussion of rbcs at @xmath9 , it is of interest to consider the specific case of mgb@xmath5 now well - established to be a two - band electron - phonon superconductor . microscopic parameters have been calculated from extensions of bandstructure calculations to obtain the electron - phonon spectral densities . the coulomb repulsion parameters that enter the gap equations ( [ eq : del ] ) and ( [ eq : z ] ) are also known . the parameters are @xmath186 , @xmath187 , @xmath188 , @xmath189 , @xmath190 , @xmath191 , @xmath192 , @xmath193 , with @xmath194 mev . the ratio of the two density of states @xmath195 and of the fermi velocities @xmath196 . to switch to our notation , @xmath197 . we start with rbcs results . in the top frame of fig . [ fig2 ] , we show the dependence of @xmath163 and @xmath198 normalized to @xmath199 as a function of @xmath164 . on comparison with some of the results of fig . [ fig1 ] , we note that mgb@xmath5 shows considerable integration of the two bands and also both gaps become zero at @xmath200 which is considerably reduced from the one - band bcs value . an interesting quantity that could be measured is the quasiparticle density of states @xmath201 in the presence of a current , which was first examined for one - band s - wave superconductors by fulde@xcite . such measurements have recently been reported for pb in ref . @xcite . for each band separately @xmath202\end{aligned}\ ] ] with @xmath203 . in the bottom frame of fig . [ fig2 ] , we give results for @xmath204 as a function of @xmath205 for various values of @xmath1 as indicated in the figure . the solid curve is essentially the zero current limit and shows the expected superposition of two bcs quasiparticle density of states with square root singularities at the two gaps . as the current is increased in the linear region only ( cross - referencing with the upper frame of fig . [ fig3 ] which shows @xmath0 for mgb@xmath5 ) , the singular structure is smeared and the gap region fills as does the region between the two gaps . gapless behavior appears at a value of @xmath1 near the first peak in the total current due to the density of states in the small band going gapless . this is in contrast to what is found in a one - band s - wave superconductor@xcite where the gapless behavior occurs at the critical current . ( 250,200 ) the superfluid current @xmath0 versus @xmath1 for the rbcs calculation with mgb@xmath5 parameters is plotted in the upper frame of fig . the line types are defined in the same way as for the @xmath0 curves in fig . we note that this graph is very similar to that shown in the second frame ( righthand side ) of fig . [ fig1 ] , where it has been discussed in relation to other cases . as in that graph , the peak in the total current seen at low @xmath1 is lower than the one at higher @xmath1 , so that this second peak determines the critical current . it is of considerable interest to compare these renormalized bcs results with those from full eliashberg calculations based on the @xmath206 spectra given in ref . very similar values of the parameters can also be found in ref . such results are presented in the lower frame of fig . there are several differences that are worthy of note . the full eliashberg results are at @xmath207 and the rbcs are for @xmath9 . this difference in temperature is small and is not expected to lead to any significant differences . examination of the partial currents shows that the dashed curves are almost the same in both plots except that strong coupling effects have shifted the point of zero current to a slightly smaller value of @xmath1 . this is accompanied by a shift to the left of the position of the peak , a reduction in its height and an increase in the rounding , all small effects . these differences are seen to be slightly more prominent in the dotted curve . this shows that rbcs can be used with confidence as a first approximation . however , from this point on , we will consider only eliashberg results as we have available to us the electron - phonon spectral functions and we wish to consider the full effects of temperature and impurities . ( 250,200 ) at the end of the last section , we showed a comparison between rbcs and eliashberg theory ( fig . [ fig3 ] ) . given that the electron - phonon spectral functions for mgb@xmath5 are available , it is appropriate to work with the full theory for @xmath2 as a function of temperature and impurity scattering . we begin with a consideration of temperature effects . in the upper frame of fig . [ fig4 ] , we show the current @xmath208 normalized by @xmath209 as a function of @xmath164 for nine values of temperature from @xmath207 to 0.9 in steps of 0.1 . the parameters used are those appropriate to mgb@xmath5 with the upper curve reproduced from the bottom frame of fig . [ fig3 ] . as the temperature is increased , the peak with larger value of @xmath1 remains the highest and therefore determines the critical current given as the solid curve in the lower frame . here , @xmath2 has been normalized by @xmath210 so that the slope at @xmath145 is negative one by arrangement for the typical critical current plot of @xmath211 , which is designed to bring out the ginzburg - landau behavior near @xmath6 . we see a smooth ( concave downward ) increase in @xmath118 as @xmath212 is reduced . the value of the critical current at @xmath213 is @xmath214 . this is lower than the one - band bcs value of 0.72 . this reduction could reflect the two - band nature of mgb@xmath5 as well as strong coupling corrections@xcite due to our use of the full eliashberg equations ( [ eq : del ] ) and ( [ eq : z ] ) to calculate the current given by eq . ( [ eq : js2 ] ) . however , reference to the lower frame of fig . [ fig3 ] shows that the second band contributes only about 10% to the total critical current at @xmath9 and this remains true at @xmath6 . consequently , this curve is very nearly the single band result and deviates from the classical bcs curve almost entirely because of retardation effects . also shown is the data of kunchur@xcite , plotted as solid dots . there is good agreement between the data and the solid curve which is quite remarkable as the calculation has no free parameters , in principle , as they are all given in the literature . including impurity scattering has an effect on the temperature dependence of @xmath118 , as we will discuss later in relation to fig . unfortunately , information on the individual intraband scattering rates in these films for @xmath215 and @xmath216 bands , separately , is not available to us . including some impurity scattering in the @xmath215 band would futher improve the agreement with experiment . the other critical current curves in fig . [ fig4 ] are for different ratios of @xmath174 as indicated in the figure caption . in addition to allowing for some uncertainty in the reported @xmath131 values in the literature , this also mimics the possibility of currents associated with the c - axis where the @xmath216-band is known to have a much smaller @xmath131 than the @xmath215-band . likewise , we have not included the details of the full fermi surface average in the expression for the current , however , it is known that the @xmath216-band is quasi - two - dimensional as opposed to the more three - dimensional @xmath215-band . dahm and scalapino@xcite find significant geometrical corrections associated with such averaging which could also accentuate the difference between the two bands . this may also be approximately captured by our use of different @xmath131 ratios . all the curves are normalized to have a slope of negative one at @xmath145 . as @xmath184 is increased , the normalized critical current at zero temperature increases and can become considerably larger than its one - band bcs value . there is also a change in the temperature variation of the curves , the dashed curve shows a clear kink around @xmath217 while the long - dashed curve shows concave upward behavior at intermediate temperatures . this can be understood better when the pattern of behavior shown in fig . [ fig5 ] is considered . ( 250,200 ) the top frame gives the temperature evolution for the case @xmath218 and the bottom is for a ratio of 4 . the notation is as for the top frame of fig . [ fig4 ] and for the same nine reduced temperatures . in the top frame , it is the peak at lower @xmath1 which determines the critical current at @xmath213 while at higher @xmath212 , it is the peak with higher @xmath1 . this crossover occurs between @xmath219 and @xmath220 and leads to the kink seen in the dashed curve of fig . [ fig4 ] ( lower frame ) . if we look at the individual contributions from each band ( not shown here ) to the total current , we find that at @xmath9 it is the second band which makes the dominant contribution because of its large value of @xmath131 but near @xmath6 this contribution has decayed sufficiently that it only makes a 30% contribution to the total . in the lower frame of fig . [ fig5 ] , the peak at smaller value of @xmath1 has greatly increased in magnitude over its value in the top frame . it has also moved to a lower value of @xmath1 as expected from fig . 1 . however , now there is no crossover from smaller to larger @xmath1 peak and hence no kink . nevertheless , the dotted curve is very different from a one - band bcs variation . we can understand better how this can arise by once more looking at the partial contribution to the critical current from each band . at @xmath9 it is the second band which dominates while at @xmath6 it accounts only for 2/3 of the total with the remaining 1/3 contribution from band one . this difference in admixture of the two bands with temperature is seen to be enough to cause an upward curvature in the dotted curve at intermediate temperatures as well as to increase the value of the normalized critical current at @xmath9 to a value considerably larger than the bcs value of 0.72 . increasing @xmath184 further as in the long - dashed curve of the lower frame of fig . [ fig4 ] increases the @xmath9 value even further as the peak at lower @xmath1 value becomes even more prominent . details of the temperature variation of the critical current clearly depend on details of the microscopic parameters of the two bands involved , such as values of @xmath221 . for example , it is possible to shift the kink in @xmath118 to higher values of reduced temperature and to make it sharper by considering other parameter sets . ( 250,200 ) to illustrate this , in fig . [ fig6 ] we show results for a lorentzian @xmath222 spectrum defined in detail in our previous paper@xcite . in this model , @xmath159 and we take @xmath223 to decrease the gap anisotropy as compared with mgb@xmath5 . we have also taken @xmath173 ( nearly decoupled bands ) . the top frame gives the current @xmath224 for the case of @xmath218 , for the nine values of reduced temperature previously chosen . with this @xmath131 ratio , the peak with lower value of @xmath1 remains highest at small @xmath212 and the crossover of the maximum to the other peak occurs at a reduced temperature between 0.6 and 0.7 . this manifests itself as a sharp kink in the solid curve for @xmath2 ( bottom frame ) at this same temperature . also shown in this same frame are results for the critical current when @xmath225 ( dotted ) and @xmath226 ( dot - dashed ) . for the larger value of @xmath184 , the normalized current at @xmath9 is increased over the @xmath218 case as expected and for @xmath227 it is reduced but remains above bcs . in this latter case , the two peaks in the current versus @xmath1 are closer together than they are in the top frame of fig . [ fig6 ] and seem to almost merge before the crossover from lower to higher @xmath1 peak occurs , which means that the kink in the dotted curve can hardly be seen . it is clear that a rich pattern of behavior exists for the current versus @xmath1 at different reduced @xmath212 in a two - band model . this translates into quite distinct temperature dependence for @xmath118 as compared with the one - band bcs canonical behavior . some addition insight into this complicated behavior can be obtained from a consideration of simplified but analytic bcs results in the limiting case of two decoupled , well - separated bands , i.e. , for small values of the gap anisotropy @xmath228 . under this assumption , @xmath144 , @xmath229 , @xmath230 , and @xmath231 so that the critical current near @xmath6 given by formula ( [ eq : bcs14 ] ) reduces to @xmath232 and is determined entirely by band one , as previously commented upon . as the temperature is lowered , however , two possible circumstances can arise . the critical current can remain determined by the same band ( number one ) in which case we can write approximately @xmath233 a second possibility is that a crossover to the peak at lower @xmath1 value occurs and we have approximately @xmath234 for the first case , @xmath235 which is the one - band bcs result where we have normalized the critical current in such a way that @xmath211 has a slope of negative one at @xmath145 . this conforms with what we have done in the lower frames of figs . [ fig4 ] and [ fig6 ] . for the second case , by assumption , there is a crossover and @xmath236^{2/3}. \label{eq : bcse}\ ] ] this simplified approximate formula shows clearly that the anisotropy @xmath237 on its own reduces the value of @xmath238 below its universal bcs value of 0.72 while a large value of @xmath239 relative to @xmath240 or @xmath184 relative to @xmath241 increases it and that this reduced dimensionless quantity is no longer universal . it can depend on the band structure parameters as well as on the gap anisotropy . next we consider the effect of impurities on the current . it is instructive to begin by expanding eq . ( [ eq : js2 ] ) for @xmath224 to lowest order in @xmath1 . doing so leads to the well - known and physically expected result that for small superfluid velocity @xmath242 where @xmath243 is the superfluid density of the @xmath33th band given for a strong coupling superconductor by @xmath244^{3/2}}\biggr ] \label{eq : bcsg}\ ] ] in bcs at @xmath9 , this becomes @xmath245 and @xmath246 for band 1 and band 2 , respectively . the slope of @xmath247 as a function of @xmath8 gives @xmath248 with eq . ( [ eq : bcsg ] ) valid for any temperature and any impurity content . the impurities enter only through the eliashberg gap equations ( [ eq : del ] ) and ( [ eq : z ] ) . as the temperature is increased , the superfluid density is reduced and so the slope of @xmath224 seen in figs . [ fig4 ] to [ fig6 ] is correspondingly reduced . impurities also reduce the superfluid density . in bcs , the known result for each of the two bands separately , to lowest order , is @xmath249 with @xmath250 and @xmath251 , the coherence length and impurity mean free path in the @xmath33th band , respectively . this applies to intraband scattering which reduces the initial slope of the partial current coming from each band . ( 250,200 ) results are presented in fig . [ fig7 ] , where we also consider the interband case @xmath252 . in the top frame , we show @xmath0 versus @xmath1 for the reduced temperature @xmath253 in the lorentzian model used in fig . the solid curve applies to the pure case and is for comparison . we show only three cases , one where @xmath254 ( dotted curve ) , @xmath255 ( dashed curve ) , and @xmath256 , where @xmath257 is the @xmath6 without impurities ( in the case of @xmath258 and @xmath259 , @xmath6 will be unaffected , but not so for interband impurities ) . intraband impurities reduce only the partial contribution to the current coming from that band . it reduces its value on the vertical axis and extends the x - axis to higher values of @xmath1 leaving the second band largely unaffected . of course , since both bands contribute significantly to the total current in the region of the peak corresponding to the smaller value of @xmath1 , its magnitude is reduced even in the dotted curve . on the other hand , for the dashed curve , there is little change in the region of the second peak which is dominated by the first band . by contrast interband impurities modify both bands in a similar fashion . the corresponding critical current as a function of @xmath260 is shown in the lower frame . the solid line is the pure case of fig . as expected , a finite value of @xmath258 has reduced the critical current below its value in the pure case at all temperatures , however , when normalized to a slope of magnitude one near @xmath6 , the effect pushes the curve up at low @xmath15 . a kink remains and the crossover temperature is only slightly shifted toward higher values . for the dashed curve , the significant modifications arise only below the kink temperature . by contrast the main effect of the interband impurities is to wash out the kink structure leaving the rest of the curve largely unaffected ( of course the @xmath6 would show a small shift downwards relative to @xmath257 , not shown here ) . in conventional one - band s - wave superconductors , the energy gap @xmath57 remains unaffected by the flow of a supercurrent up to a critical value of the superfluid velocity @xmath7 ( ie . , @xmath261 ) at which point pairbreaking becomes possible . in this region , the current @xmath0 is linear in @xmath8 with proportionality constant equal to the superfluid density . beyond this critical value of @xmath8 , the rapid reduction in the gap in addition to a quasiparticle backflow current leads to @xmath262 at @xmath263 . for two nearly decoupled bands , we would expect that the total current to exhibit a two - peak structure as a function of @xmath8 . a first one near the critical value of the superfluid velocity for band two and the other near the larger critical value of @xmath8 corresponding to the first band alone . the first peak has in general a significant contribution from both bands ( unless @xmath264 ) while the second peak could be due mainly to the first band . the relative admixture depends on microscopic parameters such as density of states @xmath265 and fermi velocities @xmath32 , as well as gap anisotropy . when @xmath184 is made large as compared to @xmath241 , the first peak moves to lower values of @xmath8 and can also become larger than the second at low temperature . as the temperature is increased the superfluid density in each band is reduced but that in the second band is more strongly affected because of its smaller gap value . consequently , since the critical current is determined by the height of the highest peak , one can have a crossover from the first to second peak as @xmath15 is raised from @xmath266 to @xmath6 and this will reflect itself as a kink in the temperature dependence of the critical current . such a feature is also found in the dirty limit calculations presented in ref . @xcite for the critical current along the c - axis . when there is a significant interband coupling or the two gaps are close in value ( isotropic case ) the two peaks in the total current cease to be independent and a rich pattern of behavior emerges for the temperature and impurity dependence of the two peaks . in particular , as the temperature is raised , the first peak can remain the determining peak for @xmath2 up to @xmath6 or it can become the smaller of the two at a crossover temperature which can be increased or decreased through judicious choices of parameters . another possibility is that the two peaks merge into one or that the second peak is the dominant one at all temperatures . application to the specific case mgb@xmath5 , which involves no adjustable parameters , gives good agreement with the in - plane data of kunchur@xcite . if the resulting critical current @xmath211 is normalized to have slope of negative one as a function of @xmath58 near @xmath145 , the corresponding value of the critical current at @xmath9 will differ from its classic one - band bcs value of 0.72 . strong coupling effects resulting from our use of the full eliashberg equations , with appropriate electron - phonon spectral densities describing the electron - phonon interaction , are known@xcite to always reduce the normalized value of @xmath267 while two - band effects can reduce or increase it depending on the detailed values of the microscopic parameters involved . in particular , a large value of @xmath184 relative to @xmath241 can lead to values that can be even larger than one . this is expected for the case of current orientated along the c - axis for the specific parameters representative of mgb@xmath5 . in this case , because the @xmath216 band ( band one in our notation ) is nearly two - dimensional , the fermi velocity of the @xmath215 band in the c - direction will be much greater than that for the @xmath216 band . the superfluid density in each subband , which determines the initial slope of the linear in @xmath8 regime for the supercurrent , is reduced with the increase in intraband impurity scattering . this fact can be used to manipulate the position along the @xmath8 axis and size of the two peaks in the total current and consequently the temperature dependence of the resulting critical current . introducing impurities in the first band will reduce the critical current at all temperatures since this band always makes a contribution to the total current . on the other hand , if they are introduced only in the second band , this will leave the critical current largely unaltered around @xmath113 . interband impurity scattering affects both bands although not necessarily in exactly the same way and leads to the smoothing out of the kink in @xmath268 versus @xmath212 when it exists . in conclusion , we predict a complex pattern of behavior for the total current as a function of @xmath8 , temperature and impurity content which can be used to restrict further the values of the microscopic parameters involved in two - band superconductors , such as mgb@xmath5 . ejn acknowledges funding from nserc , the government of ontario ( prea ) , and the university of guelph . jpc acknowledges support from nserc and the ciar . this research was supported in part by the national science foundation under grant no . phy99 - 07949 .
many quantum algorithms have been developed under the presumption that the hardware upon which they will run is effectively error - free : the error rate is so low the entire algorithm can be executed successfully without a single error . it is now known that such hardware can , in principle , be created using components that have far higher error rates . _ fault - tolerant _ quantum computing can be achieved by encoding qubits in non - abelian anyons in topological materials @xcite or using the quantum error correction codes @xcite . while the former is still in the early stages of its development , for the latter approach sub - threshold quantum operations have been demonstrated in ion - trap and superconducting systems @xcite . however , quantum error correction involves a substantial multiplication of resources ; the number of physical qubits required may be orders of magnitude greater than the number error - free logical qubits seen by the algorithm . a recent study audited the cost of implementing shor s algorithm to solve a classically - infeasible task , and found that even with state - of - the - art techniques for magic state distillation the machine would need over six million of today s highest quality qubits @xcite . the need for millions of qubits contrasts starkly with the fact that only fifty qubits are needed to achieve so - called ` quantum supremacy ' , i.e. to create a quantum processor that is so complex that conventional supercomputers @xcite can not predict its behaviour . machines involving this many qubits , under good but imperfect control , are expected to emerge in the next few years . the challenge for researchers is to identify useful functions for such devices , in order to motivate further investment and continue the evolution toward the longer - term goal of fully fault tolerant systems . recently some hybrid quantum / classical algorithms have been developed which are promising for near future quantum applications @xcite . a common feature of these algorithms is that the quantum computer is only in charge of carrying out a subroutine , acting as a ` coprocessor ' while the larger scale algorithm is governed by a classical computer . the task of the quantum computer is thus simplified and may be accomplished with relatively few quantum operations . a higher error rate per operation is therefore tolerable ; in a fault - tolerant machine this would imply a more modest resource overhead for the code , but it may even be possible to implement such quantum algorithms without quantum error correction . hybrid approaches are very relevant to quantum simulation , i.e. feynman s vision @xcite of using a controlled quantum processor to model another quantum system . such a technology would be highly advantageous for the investigation of various large quantum systems , e.g. simulating quantum chemistry systems @xcite , or novel materials and other condensed - matter systems @xcite . a powerful tool that has been exploited in several hybrid protocols is the variational method @xcite . typically the state of the target system is to be found by writing a trial quantum state with a large but tractable number of real parameters , and then discovering the optimal value of these parameters . implicitly this requires the scientist to use their understanding of the target system ( the novel molecule , or material ) to select a set of parameters that , while large , is far smaller than the total number of parameters needed to specify an arbitrary quantum state . the latter is of course exponential in the number of particles composing the target system . . variational parameters are determined by the classical computer according to the schrdinger equation . the quantum computer is used to implement a subroutine : inputs are parameters @xmath0 , and outputs are values of certain derivatives required by the main program in the classical computer . ( b ) variational parameters are determined iteratively given their initial values . parameters at the time @xmath1 are sent to the quantum computer , which evaluates the quantities required by the classical computer . then , the classical computer can work out parameters at the time @xmath2 , where @xmath3 is a short time . note that the curves here represent the actual evolution of parameters in the example described in appendix [ sec : model ] . ] in this paper , we propose a hybrid quantum algorithm for simulating the dynamics of a quantum system . the conventional approach for simulating quantum dynamics employs trotterisation @xcite , which usually requires many quantum operations therefore seems likely to necessitate the full machinery of fault - tolerant quantum computing @xcite . our approach is based on the variational method and our hope is that it could be implemented using small - size quantum circuits , i.e. quantum circuits with a small number of quantum operations and suffering significant noise compared with fault - tolerant quantum computers . a novel feature of our algorithm is that it compensates for errors through classical inference and without encoding : if the noise in the quantum computer mainly results in stochastic errors and the rate of errors can be amplified in a controllable way , then we find that errors can be approximately corrected . the condition that noise is stochastic can be met by engineering for many systems : if , for example , single qubit gates have relatively high fidelity @xcite and their noise is statistical , then arbitrary two - qubit gate errors can be made statistical though a twirling - like technique which we presently discuss . moreover , the severity of the such errors can be deliberately _ increased _ artificially , allowing one to create curves which the classical algorithm can extrapolate to estimate the zero - error limit . we performed numerical emulations of the process on small systems , finding that this technique does indeed lead to robustness : the impact of physical errors on the simulator s performance far lower than in an ( optimised ) trotterisation protocol , and moreover this impact does not worsen with the duration of the simulation . the remainder of this paper is organised as the following . in sec . [ sec : review ] , we review the trotterisation algorithm and variational methods . in sec . [ sec : algorithm ] , our hybrid algorithm is introduced . in sec . [ sec : principle ] , the variational theory is discussed . in sec . [ sec : hybrid ] , the task for the quantum computer and the overall program are described in detail . errors in our algorithm are analysed in sec . [ sec : analysis ] . the method for reducing errors is given in sec . [ sec : noise ] , in which we also discuss how to convert non - stochastic errors into stochastic errors and how to tune the rate of errors . numerical results are presented in sec . [ sec : numerics ] . a summary is given in sec . [ sec : summary ] . the trotterisation approach to simulation , which we will use as a basis for comparison with our protocol , exploits the fact that time evolution under a general hamiltonian @xmath4 can be approximated according to the trotter - suzuki decomposition @xcite @xmath5 here , each term @xmath6 corresponds to the evolution driven by the term @xmath7 for a short time @xmath8 , which can be realised by a quantum gate or a combination of quantum gates . usually , the short time is taken uniformly as @xmath9 , where @xmath10 is the time of the simulated evolution . when @xmath11 is larger , the approximation is better , and errors in the approximation scales with the simulated time and the number of quantum gates as @xmath12 @xcite . our approach is based on a variational approach . variational methods have numerous applications in the numerical study of many - body quantum systems , for examples , density functional theory @xcite , the matrix product state method @xcite , and simulating molecular dynamics using the variational principle @xcite . in these methods , typically a trial sate is used to approximate the true state of the system . the trial state must of course be specified by some tractable number of parameters . but since existing realisations are entirely classical , there is a stronger condition on the trial function : it must be possible to efficiently evaluate its fit to the true quantum state using only a classical algorithm . this requirement limits the application of variational methods . sometimes it may be impossible to evaluate a trial state that provides a good approximation to the true state in a classical computer . in such a case , a quantum computer could be helpful , because we may be able to complete tasks difficult for a classical computer using a quantum computer . an example is the unitary coupled cluster method @xcite , in which the energy of the trial state can be evaluated using a quantum computer when it is hard for a classical computer . the protocol we describe here is another example . the purpose of our hybrid algorithm is to solve the schrdinger equation @xmath13 ( @xmath14 ) , assuming that the state @xmath15 can be approximated by a trial state @xmath16 , where @xmath0 are variational parameters . as shown in fig . [ fig : hybrid](a ) , the hybrid algorithm is implemented on both of a quantum computer and a classical computer . the task of the classical computer is to determine variational parameters according to the schrdinger equation , and this proceed requires as input certain derivatives associated with the state @xmath17 which the quantum computer provides . the hybrid algorithm works out variational parameters iteratively as shown in fig . [ fig : hybrid](b ) . parameters at the time @xmath1 ( @xmath0 ) are sent to the quantum computer , with which the quantum computer finds the values required by the classical computer . based on results from the quantum computer , the classical computer can determine parameters at the time @xmath2 ( @xmath18 ) , where @xmath3 is a short time . then , these new parameters are sent back to the quantum computer . in this way , given parameters of the initial state ( @xmath19 ) , parameters at the time @xmath10 ( @xmath20 ) are systematically inferred by iterating the process carried out by two computers . the simulation is successful if the state @xmath21 is a good approximation of the state @xmath22 . using the variational method , the degrees of freedom provided by variational parameters allow us to use quantum circuits with a size much smaller than the circuit of the trotterisation algorithm to simulate the time evolution of a quantum system . note that this is an ` apples to oranges ' comparison because our algorithm only simulates the time evolution of a given initial state while the trotterisation algorithm simulates the time evolution of arbitrary initial states , i.e. the time evolution operator . thus our algorithm aims at an easier problem than the trotterisation algorithm . tracking the evolution from a specific initial state is the main goal in many simulations , and other more general tasks can also be reexpressed this way . for example the determination of the fixed ground state of a quantum system can made via an adiabatic time evolution @xcite , thus our algorithm is relevant to that task . in other hybrid algorithms for determining the ground state of a quantum system @xcite , one may need to find the global minimum of the energy in the parameter space to maximise the fidelity . however , finding the global minimum in a high - dimensional parameter space is usually a non - trivial computing task . in our algorithm , parameters are worked out iteratively , therefore the global minimisation is not required . we remark that trotterisation is used in some hybrid algorithms @xcite . in principle our algorithm can be used to replace the trotterisation method in these instances , to further simplify the task of the quantum computer . the time - dependent variational principle corresponding to the schrdinger equation reads @xmath23 , where the lagrangian @xcite @xmath24 assuming that the state @xmath17 is determined by a set of real parameters @xmath25 , i.e. @xmath16 , the lagrangian can be rewritten as @xmath26 which is a function of @xmath27 and their time derivatives @xmath28 . according to @xmath29 , the euler - lagrange equation describing the evolution of parameters ( hence the state @xmath30 ) is @xmath31 where @xmath32 here , @xmath33 , both @xmath34 and @xmath35 are real , and @xmath34 is antisymmetric . there are other variational principles for the quantum time evolution @xcite . for example , mclachlan s variational principle reads @xmath36 @xcite , which leads to the same equation as eq . ( [ eq : difeq ] ) but @xmath37 . here , the norm @xmath38 . in the following , we will focus on the time - dependent variational principle , but the hybrid algorithm can be adapted to mclachlan s variational principle . recall that we can always express a state as @xmath39 , where @xmath40 and @xmath41 are real , and @xmath42 are orthonormal basis states . taking parameters @xmath43 , eq . ( [ eq : difeq ] ) leads to the schrdinger equation ; but of course we will require a parameterisation such that the number of parameters remains tractable for the sizes of target system that we are interested in . thus our variational approach , in common with others , is relevant when the scientist can make an educated guess as to the general form of the quantum state there can be a large number of free parameters @xmath44 , but typically far fewer than would be needed to specify an arbitrary state . we consider trial states that can be directly prepared in the quantum computer , i.e. states can be expressed as @xmath45 , where @xmath46 is an initial state of the quantum computer , and @xmath47 is a sequence of quantum gates determined by parameters @xmath27 , i.e. @xmath48 here , @xmath49 is a unitary operator describing a quantum gate , and the total number of gates ( i.e. parameters ) is @xmath50 . if @xmath50 is smaller than the dimension of the hilbert space ( @xmath51 to be exact ) , @xmath30 only spans a subset of the hilbert space . in this restricted trial - state space , ( [ eq : difeq ] ) approximates the exact evolution if the exact state is close to the trial - state space . in the following analysis , we describe each @xmath49 gate as dependent on only one parameter @xmath52 . however it is worth noting that the trial state can be generalised to the case where each gate @xmath49 depends on multiple parameters , including parameters that vary in a pre - defined way with time , and that both the trotter - suzuki decomposition @xcite and the unitary coupled cluster ansatz @xcite can be expressed in this form . indeed , one can even vary the number of gates @xmath50 depending on the simulated time , providing that we understand how to re - express the trial state using the new gates ( adding gates to the set is of course trivially possible ) . our point is that the set of gates with which we create our trial state , can itself evolve over the simulated time . we are interested in the case that evaluating coefficients @xmath34 and @xmath35 in eq . ( [ eq : difeq ] ) is intractable in a classical computer , therefore these coefficients are obtained using the quantum computer . each parameter is determined in turn , by appropriately configuring a quantum circuit involving @xmath53 gates and a single measurement outcome ; this fixed circuit is run repeatedly until the expected measurement outcome is known to a given precision . note that this implies the overall protocol is trivially parallelisable over a large number of quantum processors with no quantum link between them . we express the hamiltonian in the form @xmath54 where @xmath55 are unitary operators . in many quantum systems , the number of terms in this expression scales with the size of the system polynomially . similarly , we write @xmath56 where @xmath57 are also unitary operators . for many frequently used single - qubit gates and two - qubit gates ( @xmath58 ) , there is only one term in this expression , and @xmath57 is also a one - qubit or two - qubit gate . because any operator can be expressed using pauli operators , we can choose unitary operators @xmath55 and @xmath57 as ( single - qubit and multi - qubit ) pauli operators . using the expression ( [ eq : drk ] ) , we rewrite the derivative of the state as @xmath59 where @xmath60 then , differential equation coefficients can be expressed as @xmath61 and @xmath62 where we have used the expression ( [ eq : ham ] ) . in eqs . ( [ eq : m ] ) and ( [ eq : v ] ) , each term is in the from @xmath63 where the amplitude @xmath64 and phase @xmath65 are determined by either @xmath66 or @xmath67 @xcite , and @xmath68 is a unitary operator either @xmath69 or @xmath70 . such a term can be evaluated using the quantum circuit shown in fig . [ fig : circuit ] . this circuit needs an ancillary qubit initialised in the state @xmath71 and a register initialised in the state @xmath46 . the ancillary qubit is measured in the @xmath72 basis after a sequence of quantum gates on the register and two controlled gates , in which the ancillary qubit is the control qubit . the value of each term is given by @xmath73 , where @xmath74 is the final state of the quantum computer , and @xmath75 is the x - direction pauli operator of the ancillary qubit . the overall flow of the algorithm is as follows : firstly , we select initial parameters @xmath76 . secondly , we solve the differential equation ( [ eq : difeq ] ) numerically using the classical computer , in which process the matrix @xmath34 and the vector @xmath35 in the equation are evaluated using the quantum coprocessor . the solution permits us to project our parameters forward by a small time increment , and repeat the second step . eventually we reach the parameters @xmath77 which allow us to prepare the final state in the quantum computer . there are many different numerical methods for solving a differential equation , and the choice of specific numerical method determines the details of the information exchange loop between quantum and classical processors . in the following , we take the euler method as an example , but the algorithm can be adapted to other numerical methods , e.g. runge - kutta methods @xcite . time is discretised as @xmath78 , where @xmath79 is the initial time , and @xmath80 is the simulated evolution time . firstly , @xmath34 and @xmath35 corresponding to parameters @xmath81 are evaluated using the quantum computer . then , the following process is repeated . given @xmath34 and @xmath35 corresponding to the time @xmath82 , eq . ( [ eq : difeq ] ) is solved numerically on the classical computer to obtain values of @xmath83 . as parameters @xmath84 have been obtained from previous calculations , one can approximately calculate parameters of the time @xmath85 using @xmath86 . repeating the process until @xmath87 , we can work out parameters @xmath77 of the final state . , respectively . @xmath88 denotes the exact time evolution during the time from @xmath89 to @xmath82 . gray circles denote trial wavefunctions , and each empty circle with a dashed edge is the wavefunction at the time @xmath90 given by the exact time evolution @xmath88 and taking the trial wavefunction at the time @xmath89 as the initial state . distances are marked by red double lines . @xmath91 and @xmath92 . the distance between @xmath93 and @xmath94 is not larger than the sum of all these red double lines in the main figure . ] there are four types of errors that can result in infidelity in the variational quantum simulation ; they are i ) errors due to limited generality of the trial wavefunction , which may be only able to describe the simulated system approximately ; ii ) errors in the numerical integration obtained by solving equation ( [ eq : difeq ] ) , which is always approximate due to the finite discretisation of time ; iii ) shot noise in measuring equation coefficients @xmath34 and @xmath35 ; and iv ) errors due to noise in the quantum machine , e.g. decoherence , quantum gate infidelity . a good trial wavefunction allows us to not only reduce trial - wavefunction errors but also minimise the difficulty of implementing the algorithm , e.g. only use small - size quantum circuits . whether a good trial wavefunction can be found depends on the simulated system and our understanding of the physics in that system . however , a trial wavefunction that contains a polynomial number of parameters and is a high - fidelity approximation to the true wavefunction always exists . one can set the trial wavefunction as a wavefunction in the trotter - suzuki form , i.e. take the sequence of gate operations @xmath47 in the form of eq . ( [ eq : decomposition ] ) , and choose the evolution time of each term as a parameter ( @xmath95 ) @xcite . using the trotter - suzuki - form trial wavefunction , we know that the probability of trial - wavefunction errors decreases with the number of trotterisation slices @xmath11 as ` @xmath96 ' @xcite in the worst case . integration errors depend on the numerical method for solving the differential equation ( [ eq : difeq ] ) . we take the euler method as an example . in the euler method , the probability of error is proportional to the size of each step @xmath3 @xcite . therefore , by choosing a small step size , integration errors can be suppressed . shot noise and machine noise occur in the implementation of the algorithm while trial - wavefunction errors and integration errors are due to the imperfection of the algorithm itself . the effect of implementation errors in the integration process is that coefficients @xmath34 and @xmath35 evaluated using the quantum computer are inaccurate , i.e. their values are different from their true values @xmath97 and @xmath98 given by eqs . ( [ eq : m ] ) and ( [ eq : v ] ) ( that is , the values given by a quantum computer without shot noise and machine noise ) . if coefficients @xmath34 and @xmath35 are inaccurate , time derivatives @xmath99 obtained from @xmath34 and @xmath35 [ see eq . ( [ eq : difeq ] ) ] are different from their true values @xmath100 obtained from @xmath97 and @xmath98 . using @xmath101 , @xmath102 and @xmath103 to denote deviations from true values and using @xmath104 to denote the solution of eq . ( [ eq : difeq ] ) , we have @xmath105 . when the matrix is invertible in the vicinity of @xmath97 , @xmath106 . to analyse errors in the hybrid algorithm , we use the trace distance @xmath107 @xcite as the measure of error severity . the degree of error in the overall process is given by @xmath108 . here , @xmath109 denotes the true wavefunction , @xmath110 denotes the trial wavefunction , and @xmath94 is the state prepared in the quantum computer according to the state @xmath111 . these two states @xmath94 and @xmath111 are different because of the machine noise . the triangle inequality holds for the trace distance , i.e. @xmath112 . therefore , an upper bound of @xmath113 is given by ( see fig . [ fig : distance ] ) @xmath114 here , @xmath88 is the exact evolution during the time from @xmath89 to @xmath82 ( @xmath115 if the hamiltonian is time - independent ) , therefore @xmath116 . to obtain the upper bound , we have used @xmath117 . to distinguish algorithm errors and implementation errors , we use the inequality @xmath118 . here , the state @xmath119 is a trial state corresponding to parameters @xmath120 , i.e. assuming that the quantum computer reports accurate values of @xmath34 and @xmath35 at the time @xmath89 . then , the upper bound can be rewritten as @xmath121 where @xmath122 here , @xmath123 corresponds to algorithm errors , and @xmath124 corresponds to implementation errors . algorithm errors in each time step can be expressed as @xmath125 where @xmath126 @xmath127 and @xmath128 here , we have used taylor expansions of @xmath129 and @xmath119 , i.e. @xmath130 and @xmath131 , where @xmath132 . the matrix norm is induced by the vector norm , therefore @xmath133 . implementation errors in each time step are due to the difference between time derivatives obtained from the real quantum computer and their true values , i.e. @xmath134 , which can be expressed as @xmath135 where @xmath136 is a positive semi - definite matrix and @xmath137 here , we have used the taylor expansion of @xmath138 , which is the same as @xmath119 but @xmath100 are replaced by @xmath99 . using @xmath139 to denote the maximum value of the quantity @xmath140 for all @xmath82 , we have @xmath141 the first term of @xmath123 and the second term of @xmath124 are due to imperfections in approximating the initial state using the trial wavefunction and preparing the final state in the quantum computer with machine noise , respectively . note that these two terms are not @xmath10 dependent ; for simulations over a substantial time @xmath10 we may expect them to make relatively small contributions . in the following , we analyse the other three terms one by one . the second term of @xmath123 is the accumulation of trial - wavefunction imperfections in each time step . using the circuit in fig . [ fig : circuit ] , every term in @xmath142 can be measured by a method analogous to that used for obtaining @xmath34 and @xmath35 . the accuracy is again limited by the shot noise and machine noise . therefore , algorithm errors due to the trial wavefunction can be continually estimated during the execution of the the hybrid quantum computation . the third term of @xmath123 is caused by the finite integration step size , which can be reduced by decreasing @xmath3 . in order to limit this term to be @xmath143 , we need to choose a step size @xmath144 , i.e. the number of time steps @xmath145 . when the trial wavefunction is a good approximation to the exact state , we can expect that @xmath146 , which implies @xmath147 in the subspace of the problem , and in this case @xmath148 . the first term of @xmath124 is due to the difference between @xmath99 and their true values @xmath100 . the difference @xmath149 . @xmath34 and @xmath35 are evaluated as required by the algorithm . similar to @xmath142 , each element of @xmath136 can also be measured using the circuit in fig . [ fig : circuit ] . therefore , the susceptibility to shot noise and machine noise in the integration process can be estimated during the execution of the hybrid quantum computation . the algorithm is susceptible to implementation errors when @xmath97 is singular , which should be avoided when choosing the trial wavefunction . as a worst case , implementation errors accumulate linearly with the simulated time . however , it can be far less severe : in sec . [ sec : numerics ] , we will explore an example in which the accumulation of errors due to the machine noise is almost negligible compared with errors in a quantum simulation based on the conventional trotter - suzuki decomposition . shot noise can be suppressed by repeating quantum circuits for measuring @xmath34 and @xmath35 for many times . to measure the quantity @xmath150 , the deviation due to the shot noise decreases with the the number of repetitions @xmath151 as @xmath152 . therefore , if there is only short noise , @xmath153 , where @xmath154 , where @xmath155^{1/2}$ ] and @xmath156^{1/2}$ ] . in order to limit the overall effect of shot noise to be @xmath157 , we need to choose @xmath158 . the number of distinct circuits @xmath159 required for finding the @xmath34 and @xmath35 parameters depends on the number @xmath160 of terms in the hamiltonian @xmath161 [ see eq . ( [ eq : ham ] ) ] , the number @xmath162 of terms in each time derivative of @xmath49 [ see eq . ( [ eq : drk ] ) ] , and @xmath50 . @xmath163 , where the first term corresponds to @xmath34 , and the second term corresponds to @xmath35 . if each @xmath49 is realised by @xmath164 gates , and each term @xmath165 or @xmath57 in the hamilton or the time derivative of @xmath49 is a pauli operator of less than @xmath166 qubits , each circuit includes at most @xmath167 gates . here , we have assumed that each controlled-@xmath68 gate in the circuit ( fig . [ fig : circuit ] ) is realised by @xmath166 two - qubit controlled-@xmath168 gates , where @xmath168 is a single - qubit pauli operator . to complete the circuit , other operations include preparing initial states of the ancillary qubit and the register and measuring the ancillary qubit . the overall computation includes @xmath169 times steps , in each time step , @xmath159 circuits are implemented , each circuit contains @xmath170 gates and is repeated for @xmath151 times . therefore , the overall number of gates is @xmath171.\end{aligned}\ ] ] from this expression , we see that the cost is a polynomial function with respect to the integration error @xmath143 , the shot - noise error @xmath157 and the simulated time @xmath10 . factors @xmath172 , @xmath173 and @xmath174 depend on the form of the trial wavefunction . however , during the actual execution of the hybrid computation , it will be possible to estimate @xmath172 and @xmath173 . moreover @xmath148 when the trial wavefunction is a good approximation to the true state . it is important to remember that while the overall gate count will be a large ( albeit polynomially - scaling ) total , the complete calculation is formed of many small quantum calculations of depth @xmath170 . each small computation is isolated from the others , i.e. there is no shared or persistent quantum resource , and indeed @xmath175 such circuits could be performed in parallel using that many separate small quantum computers . is inferred by measuring @xmath150 for a set of error factors @xmath176 and fitting data using the function @xmath177 . ( b ) error twirling and simulation . non - stochastic errors in a controlled - phase gate can be converted into stochastic errors by performing pauli gates before and after the gate , and error probabilities can be tuned by applying pauli gates after the gate . ] of the implementation errors , machine noise is the more problematic . shot noise can be suppressed merely by repeating each quantum circuit for many times , and the repeating number scales polynomially with the accuracy . machine noise is less easily dismissed , and we focus on it in this section . machine noise need not necessarily result in computing errors . the task of the quantum computer is to work out coefficients @xmath34 and @xmath35 , so that the classical computer can work out time derivatives @xmath99 according to eq . ( [ eq : difeq ] ) . machine noise ( as well as shot noise ) can potentially cause computing errors by changing these time derivatives from their true values @xmath100 . however , sometimes machine noise does not change values of @xmath99 . for example , consider the case that the effect of machine noise is to depolarise the ancillary qubit at a fixed level , i.e. the output of the quantum computer becomes @xmath178 , where @xmath179 is a constant independent of the quantum circuit . in this case , all equation coefficients are scaled as @xmath180 and @xmath181 . as long as @xmath179 is non - zero , the solution @xmath99 of eq . ( [ eq : difeq ] ) is the same for any value of @xmath179 . therefore , only an inhomogeneous scaling of quantum outputs @xmath150 results in computing errors . an example of the homogeneous scaling is the case of balanced measurement errors . errors in the measurement on the ancillary qubit ( see fig . [ fig : circuit ] ) can be modelled as the following : if the state of the qubit is @xmath182 ( @xmath183 ) , the measurement outcome is correct , i.e. @xmath184 ( @xmath185 ) , with the probability @xmath186 ( @xmath187 ) , and the outcome is incorrect , i.e. @xmath185 ( @xmath184 ) , with the probability @xmath188 ( @xmath189 ) . if there are no other implementation errors , quantum outputs are changed from @xmath190 to @xmath191 . if measurement errors are balanced , i.e. @xmath192 , the effect of measurement errors is a fixed scaling factor @xmath193 , which does not result in computing errors . therefore , our hybrid algorithm is inherently insensitive to measurement errors on the ancillary qubit if these errors are balanced . we remark that if single - qubit gates are reliable , one can flip the qubit before the measurement , so that measurement errors are effectively balanced . measurement errors can be corrected even if they are not balanced . if @xmath194 and @xmath195 can be evaluated by benchmarking measurement operations , one can easily work out the true value @xmath190 using the value obtained from the real machine : @xmath196 / ( 1-p_0-p_1)$ ] . next , we will show that a similar procedure can be applied to any machine noise if errors due to the machine noise are stochastic with tunable probabilities . we will also show how to convert errors in two - qubit entangling gates , which are expected to be the main sources of errors , into stochastic errors if they are not stochastic , and how to simulate stochastic errors to tune error probabilities . errors in an operation are stochastic if the operation is described by a superoperator @xmath197 and @xmath198 has the form @xmath199 . here , @xmath200 is the ideal operation without errors , @xmath198 is the superoperator describing the effect of the noise , @xmath201 is an identity operation , and errors @xmath202 occurs with the probability @xmath203 . @xmath202 is a valid quantum operation , i.e. trace - preserving completely positive map . given an initial state @xmath46 , after a sequence of operations , the final state of the quantum computer is @xmath204 where @xmath205 denotes the @xmath206th operation . taking account that errors are stochastic , the quantum outcome can be rewritten in the form @xmath207 here , @xmath208 , @xmath209 is the value without errors , @xmath210 @xmath211 , and @xmath212 the @xmath213 term of @xmath214 corresponds to the case that only the @xmath213 operation causes errors and all other operations are ideal . note that in these equations we have replaced error probabilities @xmath215 with @xmath216 , where @xmath176 is a convenient scale factor allowing us to write @xmath217 as a function of @xmath176 , and @xmath218 . if probabilities of errors are tunable , we can infer the value of @xmath219 by measuring values of @xmath217 of a set of different factors @xmath176 [ see fig . [ fig : error](a ) ] . clearly @xmath176 can never be zero for our machine , as this would imply that we have the ability to fully switch off the machine noise , making it perfect . if @xmath220 are the minimum error probabilities allowed by the machine , the minimum value of @xmath176 achievable by the machine is @xmath185 . to infer the value of @xmath219 , firstly , we take @xmath221 values of @xmath176 as @xmath222 and measure @xmath223 using the quantum computer , where we can take @xmath224 ; secondly , we fit quantum outputs using the function @xmath225 , where @xmath226 . in this way , the first - order contribution of machine noise can be corrected . similarly , considering second - order terms in the expansion ( [ eq : expansion ] ) , one can fit data using a function with second - order terms ( i.e. @xmath227 terms ) to correct the second - order contribution of machine noise . the error reduction protocol only works for small - size circuits , which are used in the hybrid algorithm while the trotterisation algorithm usually needs large - size circuits . the true value @xmath190 can be inferred because the contribution of high - order terms is much smaller than the contribution of lower - order terms , i.e. @xmath228 . the total rate of errors in the quantum circuit with @xmath170 gates is @xmath229 . the first term in the expansion corresponds to the first order contribution @xmath230 , and so on . therefore , high - order terms can not be neglected if @xmath231 . when there are too many gates in the circuit or the error rate is too high , the quantum state will be populated with errors and one can not retrieve the true value @xmath190 even if we consider high - order terms in the interpolation . the best experiments to date @xcite have reduced two - qubit gate infidelity to the range @xmath232 to @xmath233 . with hardware of that kind , our protocol could support hundreds of gates . thus a simulation of a quantum system with a @xmath234 dimensional hilbert space , using a trial wavefunction with hundred(s ) of parameters , may be feasible . implementing the error reduction protocol requires knowledge of the inherent machine noise . therefore , quantum operations need to be benchmarked , e.g. using quantum process tomography @xcite , before the quantum coprocessor is used , and the nature of the machine noise should not vary significantly during the simulation . alternatively , one can monitor the machine noise in the process by stopping the protocol and benchmarking operations , because in the hybrid algorithm the quantum computer only performs small - size circuits and can be stopped at any stage . non - stochastic errors can be converted into stochastic pauli errors using some redundant pauli gates @xcite . it has been seen as a common feature of quantum computing systems that two - qubit gates are the main sources of errors , i.e. probabilities of errors in two - qubit gates are much higher than probabilities of errors in single - qubit gates @xcite . in this case , arbitrary errors in two - qubit gates can be converted into stochastic errors though the use of redundant pauli gates , without introducing significant additional noise . we consider the controlled - phase gate @xmath235 as an example , and it is similar for other two - qubit gates , e.g. the controlled - not gate . here c and t denote the control qubit and the target qubit , respectively . the controlled - phase gate with noise is @xmath236 , where @xmath237 . the noise may not be in the stochastic - error form , so that we express the noise operation in the kraus form , i.e. @xmath238 , where @xmath239 , and each two - qubit matrix can be written as @xmath240 @xcite . here @xmath241 correspond to pauli operators @xmath242 , @xmath243 , @xmath244 and @xmath245 , respectively . in order to convert errors , pauli gates @xmath246 are randomly chosen and applied on each qubit before and after the controlled - phase gate . if the gate before the controlled - phase gate is @xmath68 , the gate after the controlled - phase gate is restricted to be @xmath247 to let random pauli gates cancel with each other . here , @xmath68 and @xmath247 are both two - qubit pauli gates . the circuit is shown in fig . [ fig : error](b ) , in which we take @xmath248 and @xmath249 , so that @xmath250 up to a phase factor . as a result , the overall operation is @xmath251 , where @xmath252 is the superoperator describing the effective noise after applying random pauli gates . the effective noise superoperator is in the stochastic form @xcite ( see appendix [ sec : twirling ] ) , i.e. @xmath253 + \sum _ { ( a , b ) \neq ( 0,0 ) } \epsilon_{a , b } [ { \sigma^{a}_{\rm c } } { \sigma^{b}_{\rm t } } ] , \label{eq : effective}\end{aligned}\ ] ] where the fidelity is @xmath254 , and error probabilities are @xmath255 . @xmath256 $ ] is a superoperator , and @xmath256\rho = u\rho u^\dag$ ] . we have assumed that pauli gates are ideal for simplification . in the case that errors in pauli gates are much less than errors in the controlled - phase gate , noise of pauli gates will be a perturbation to the effective noise of the controlled - phase gate . error severity in the controlled - phase gate will effectively increase by four units of the single - qubit pauli gate error rate , and the effective noise of the controlled - phase gate may not be fully stochastic if pauli - gate errors are non - stochastic . to tune probabilities of errors , we can randomly perform pauli gates after the controlled - phase gate according to desired error probabilities [ see fig . [ fig : error](b ) ] . assuming that we want to tune error probabilities from @xmath257 to @xmath258 , we can perform the pauli gate @xmath259 [ @xmath260 with the probability @xmath261 . because we are only interested in the case that @xmath262 , overall error probabilities are approximately @xmath263 where the first term is due to the raw controlled - phase gate ( with noise ) , and the second term is due to simulated errors using single - qubit pauli gates . to demonstrate the hybrid algorithm , we numerically simulate a small quantum computer using classical computers . we take the quantum ising model @xcite as an example . the hamiltonian of the model reads @xmath264 , where @xmath265 and @xmath266 . here , @xmath267 is the number of spins , and @xmath268 . in our numerical simulations , we take @xmath269 and @xmath270 , therefore we need at least four qubits in the quantum computer to implement the hybrid algorithm . the trial state is chosen to be @xmath271 where the initial state @xmath272 is a one - dimensional cluster state . in the cluster state , qubits are in the eigenstate state of stabilisers @xmath273 ( @xmath274 ) with the same eigenvalue @xmath275 , which is prepared by performing the controlled - phase gate on each pair of nearest neighbouring qubits initialised in the state @xmath276 @xcite . the evolution of the true state is @xmath277 . in this example , the trial state is capable of exactly matching the true state given the correct values of the parameters . we consider a quantum computer that can initialise qubits in the state @xmath182 , measure a qubit in the @xmath278 basis , and perform single - qubit and two - qubit quantum gates . quantum gates include the hadamard gate , pauli gates , phase gates @xmath279 , flip gates @xmath280 , two - qubit gates @xmath281 , the controlled - phase gate and the controlled - not gate . with one of three types of two - qubit gates , the other two can be efficient realised , e.g. the gate @xmath282 can be realised using two controlled - phase gates and three single - qubit gates . here , @xmath283 is the hadamard gate on qubit-1 . we assume all three types of two - qubit gates can be directly implemented for simplification . the state @xmath276 is prepared by initialising the qubit in the state @xmath182 and performing a hadamard gate ; the measurement in the @xmath72 basis is done by performing a hadamard gate before measuring the qubit in the @xmath278 basis . we model the machine noise in the quantum computer as depolarising errors . a qubit may be initialised in the incorrect state ( @xmath183 ) with the probability @xmath284 . the measurement outcome is incorrect with the probability @xmath285 . for single - qubit gates , the noise superoperator is @xmath286 + \frac{\epsilon^{(1)}}{3}\sum_{a=0}^{3}[{\sigma^{(a)}_{}}].\ ] ] for two - qubit gates , the noise superoperator is @xmath287 + \frac{\epsilon^{(2)}}{15}\sum_{a=0}^{3}\sum_{b=0}^{3}[{\sigma^{(a)}_{1}}{\sigma^{(b)}_{2}}].\ ] ] here , @xmath288 and @xmath289 are rates of errors per gate . we assume that error rates of all single - qubit gates are the same , error rates of all two - qubit gates are the same , and error rates of single - qubit operations are only one tenth of error rates of two - qubit gates , i.e. @xmath290 . because the size of quantum circuits for implementing the hybrid algorithm is small , we neglect memory errors . in this model of the machine noise , errors in quantum operations are all stochastic , and error rates can be tuned by simulating errors using single - qubit pauli gates . numerical simulations are performed to find the trace distance @xmath291 between the true state @xmath15 and the state @xmath292 prepared in the quantum computer according to the trial state @xmath17 [ see fig . [ fig : plot](a ) ] . because of the machine noise , @xmath292 is different from @xmath17 . the average values of stabilisers can be used to describe the quality of a cluster state . the performance of quantum algorithms in estimating the average values of stabilisers is shown in fig . [ fig : plot](b ) . in our numerical simulations , we have taken @xmath293 , which is the state - of - the - art error rate @xcite . see appendix [ sec : detnum ] for some details of our numerical simulations . we observe that in this simulation the hybrid algorithm proves to be far more reliable than the trotterisation algorithm . the distance in the hybrid algorithm is about ten times lower than the distance in the trotterisation algorithm , and moreover the increase of the distance as a function of time can be largely suppressed in the hybrid algorithm by using the error reduction scheme [ fig . [ fig : plot](a ) ] . as a result , the hybrid algorithm can provide a much better estimation of the average values of stabilisers [ fig . [ fig : plot](b ) ] . we would like to stress that in order to make the comparison fair , the time interval selected for the trotterisation algorithm has been optimised to minimise errors ( in practice this would be possibly only if the distance from the true state can be measured in the quantum computer , an ability that would probably negate the need for simulation ! ) . in the hybrid algorithm , we have used the fourth - order runge - kutta method and taken @xmath294 to eliminate errors due to the numerical integration ; and we have neglected the effect of shot noise , therefore all errors are due to machine noise . given a finite time cost of implementing the hybrid algorithm , we need to consider the effect of errors in the numerical integration and shot noise . taking into account the effect of these two types of imperfections , we find that the distance in the hybrid algorithm may be increased but still much lower than the distance in the trotterisation algorithm [ blue and red bands in fig . [ fig : plot](a ) ] : we take @xmath295 and assume that each circuit is repeated for @xmath296 ( @xmath297 ) times to measure @xmath150 in the hybrid algorithm ( with the error reduction ) . the increasing of the distance only slightly changes the estimation of average values of stabilisers [ fig . [ fig : plot](b ) ] . we have proposed a quantum algorithm for simulating the time evolution of a quantum system . in this algorithm , both a classical processor and a quantum coprocessor are tightly integrated . because of the assistance of the classical computer , the algorithm can be implemented with quantum circuits of much less depth ( i.e. fewer quantum operations ) compared with the canonical trotterisation algorithm . we discussed the robustness of the algorithm to noise , and we found that the algorithm can automatically correct some errors induced by the noise in the quantum computer . moreover , by deliberately amplifying stochastic noise in a controllable way , the zero - error limit can be estimated ; consequently , the effect of errors can be significantly suppressed without the need for code - based quantum error correction and its concomitant resource overheads . this quantum algorithm can also be parallelised easily . the task of the quantum coprocessor is to repeatedly implement a set of small - size quantum circuits , therefore the computing speed can be accelerated by using a cluster of quantum coprocessors , in which each works independently and there are not quantum channels linking them . in view of these various merits , we believe that our algorithm provides a promising candidate application of early - stage non - fault - tolerant quantum computers . this work was supported by the epsrc national quantum technology hub in networked quantum information technology . the authors would like to acknowledge the use of the university of oxford advanced research computing ( arc ) facility in carrying out this work . http://dx.doi.org/10.5281/zenodo.22558 . * note added for second arxiv submission : * shortly after we posted the first version of this paper , a highly relevant paper was also posted by authors temme , bravyi , and gambetta ( arxiv:1612.02058 ) , which we commend to the reader s attention . in fig . [ fig : hybrid](b ) , the time evolution of a qubit is simulated . the time evolution is driven by the hamiltonian @xmath298 . the trial state is in the form @xmath299 . the initial state is given by @xmath300 and @xmath301 , and @xmath302 . demonstrating this example needs a quantum computer of two qubits . the twirled controlled - phase gate on a state @xmath74 reads @xmath303 \mathcal{n}_\lambda \mathcal{u}_\lambda [ { \sigma^{a}_{\rm c } } { \sigma^{b}_{\rm t } } ] \rho = \bar{\mathcal{n}}_\lambda \mathcal{u}_\lambda \rho\end{aligned}\ ] ] the effective noise superoperator is @xmath304 \mathcal{n}_\lambda [ { \sigma^{a}_{\rm c } } { \sigma^{b}_{\rm t}}].\end{aligned}\ ] ] using @xmath305{\sigma^{b}_{}}$ ] , we can find that the effective noise is in the stochastic form of eq . ( [ eq : effective ] ) . in the hybrid algorithm with the error reduction , we take @xmath224 and @xmath306 and fit the function @xmath177 to infer the value of @xmath190 . to simulate shot noise , we use the normal distribution to approximate the binomial distribution , i.e. the value of @xmath150 taking into account the shot noise is given by @xmath307 , where @xmath308 is given by a normal distribution with the mean @xmath309 and the standard deviation @xmath310 . in the trotterisation algorithm , the time evolution is simulated using eq . ( [ eq : decomposition ] ) . for the quantum ising model , the hamiltonian is decomposed into two terms @xmath311 and @xmath312 . in our numerical simulations , we take @xmath313 , @xmath314 if @xmath315 , and @xmath316 . such a method of determining trotterisation parameters coincides with the optimal choice of the number @xmath11 to minimise errors @xcite . the state @xmath292 is given by replacing @xmath10 with @xmath1 and determining @xmath11 and @xmath8 using the method we just described . to find the optimal @xmath3 , we consider the average trace distance @xmath317 , where @xmath318 is the true state , and @xmath74 is the state prepared in the quantum computer according to the trotterisation algorithm . the average trace distance is plotted in fig . [ fig : plot](a ) , taking @xmath319 in the average . to obtain other results in fig . [ fig : plot ] , we take @xmath320 in the trotterisation algorithm , which minimises the average distance . r. barends , j. kelly , a. megrant , a. veitia , d. sank , e. jeffrey , t. c. white , j. mutus , a. g. fowler , b. campbell , y. chen , z. chen , b. chiaro , a. dunsworth , c. neill , p. omalley , p. roushan , a. vainsencher , j. wenner , a. n. korotkov , a. n. cleland , and j. m. martinis , _ superconducting quantum circuits at the surface code threshold for fault tolerance _ , nature * 508 * , 500 ( 2014 ) . t. p. harty , d. t. c. allcock , c. j. ballance , l. guidoni , h. a. janacek , n. m. linke , d. n. stacey , and d. m. lucas , _ high - fidelity preparation , gates , memory and readout of a trapped - ion quantum bit _ lett . * 113 * , 220501 ( 2014 ) . j. p. gaebler , t. r. tan , y. lin , y. wan , r. bowler , a. c. keith , s. glancy , k. coakley , e. knill , d. leibfried , and d. j. wineland , _ high - fidelity universal gate set for @xmath321be@xmath322 ion qubits _ , phys . lett . * 117 * , 060505 ( 2016 ) . a. peruzzo , j. mcclean , p. shadbolt , m .- h . yung , x .- q . zhou , p. j. love , a. aspuru - guzik , and j. l. obrien , _ a variational eigenvalue solver on a photonic quantum processor _ , nat . commun . * 5 * , 4213 ( 2014 ) . j. m. kreula , l. garca - lvarez , l. lamata , s. r. clark , e. solano , and d. jaksch , _ few - qubit quantum - classical simulation of strongly correlated lattice fermions _ , epj quantum technology * 3 * , 11 ( 2016 ) . p. j. j. omalley , r. babbush , i. d. kivlichan , j. romero , j. r. mcclean , r. barends , j. kelly , p. roushan , a. tranter , n. ding , b. campbell , y. chen , z. chen , b. chiaro , a. dunsworth , a. g. fowler , e. jeffrey , e. lucero , a. megrant , j. y. mutus , m. neeley , c. neill , c. quintana , d. sank , a. vainsencher , j. wenner , t. c. white , p. v. coveney , p. j. love , h. neven , a. aspuru - guzik , and j. m. martinis , _ scalable quantum simulation of molecular energies _ , phys . x * 6 * , 031007 ( 2016 ) . b. p. lanyon , c. hempel , d. nigg , m. mller , r. gerritsma , f. zhringer , p. schindler , j. t. barreiro , m. rambach , g. kirchmair , m. hennrich , p. zoller , r. blatt , and c. f. roos , _ universal digital quantum simulation with trapped ions _ , science * 334 * , 57 ( 2011 ) . r. barends , l. lamata , j. kelly , l. garca - lvarez , a. g. fowler , a. megrant , e. jeffrey , t. c. white , d. sank , j. y. mutus , b. campbell , y. chen , z. chen , b. chiaro , a. dunsworth , i .- c . hoi , c. neill , p. j. j. omalley , c. quintana , p. roushan , a. vainsencher , j. wenner , e. solano , and j. m. martinis , _ digital quantum simulation of fermionic models with a superconducting circuit _ , nat * 6 * , 7654 ( 2015 ) . r. c. bialczak , m. ansmann , m. hofheinz , e. lucero , m. neeley , a. d. oconnell , d. sank , h. wang , j. wenner , m. steffen , a. n. cleland , and j. m. martinis , _ quantum process tomography of a universal entangling gate implemented with josephson phase qubits _ , nat . phys . * 6 * , 409 ( 2010 ) .
the study of the infrared problem in quantum gravity goes mainly back to the work of weinberg @xcite who has investigated the infrared behaviour of virtual and real soft graviton emission processes . the transition amplitude for a process in which an emission of a single graviton occurs is logarithmically divergent . in the sum over an infinite number of soft gravitons emitted he claims that infrared divergencies in the transition amplitude cancel to every order of perturbation theory . in his treatment he considers virtual and real bremsstrahlung seperately . for the _ virtual _ ones he obtained in the sum over an infinte number of bremsstrahlungs processes a transition amplitude which is proportional to some power of the lower cutoff parameter @xmath0 . so this rate will vanish in the limit @xmath1 . for an infinite number of _ real _ processes he obtained a transition amplitude which is proportional to the same negative power in @xmath0 . so the cutoff disappears if virtual and real processes are combined . from our point of view we think that the problem can not be treated in this rather simple way for the following reasons . the structure of the problem is the same as in quantum electrodynamics ( qed ) . here it is well known that the infrared divergencies cancel in lowest order but in higher orders one has to deal with subdivergences in an arbitrary feynman diagram . the question of wheather or not these divergencies cancel to every order of perturbation theory is a topic which had long been investigated , see @xcite , but is still under discussion @xcite . as far as we know the cancellation of these divergencies is not at all clear . so we think that one has to investigate these processes carefully order by order . in this article we will reinvestigate the infrared problem in quantum gravity in the framework of causal perturbation theory . especially we consider the emission of a real soft graviton in two particle scattering . we use the method of adiabatic switching with a testfunction from the schwartz space which cuts off the interaction at large distances in spacetime . this makes all expressions well defined during the calculation and it turns out to be a natural infrared cutoff . we therefore avoid the introduction of a graviton mass . the latter , although widely used in the literature , is less satisfactory on physical grounds because it modifies the interaction at short distances , too . it is not at all clear that the different infrared regularizations give the same result for observable quantities . to study this question we want to calculate the differential cross section for bremsstrahlung from first principles . causal perturbation theory goes back to ideas of bogoliubov @xcite and was carefully developed by epstein and glaser @xcite in the seventies . later scharf @xcite applied it successfully to qed . the idea is to fix the first order interaction and then to construct higher orders of the scattering - matrix @xmath2 by induction using free fields only . since the free quantum fields as basic objects are operator valued distributions the scattering - matrix will be constructed as an operator valued functional on some testfunction space . the most important ingredient for the inductive construction is the causality requirement , which roughly states that the scattering - matrix of a sum of two testfunctions factorizes if the supports of the testfunctions can be separated in time . as testfunction space one considers for convenience the schwartz space of functions of rapid decrease , since this space is invariant under fourier transformation . then all expressions are well defined tempered distributions . in doing this we cut off the interaction at large but finite distances which is unphysical in most cases and has to be removed at the end . this is the so called adiabatic limit where we take the limit that the testfunction goes to a constant . in this way we investigate the long range behaviour of the theory . we consider a symmetric second rank tensor field @xmath3 , which is a solution to the wave equation , i.e. @xmath4 . all tensor indices are raised and lowered with the minkowski metric @xmath5 with components @xmath6 . we denote by @xmath7 the minkowski scalar product of two four vectors . throughout the paper equal indizes in one expression are always be properly contracted with the minkowski metric irrespective of their position . we quantize the graviton field @xmath3 by imposing the lorentz covariant commutation relations @xmath8=-ib^{{\alpha}{\beta}\mu\nu}d_0(x - y ) \label{hh}\ ] ] where the tensor @xmath9 is defined by @xmath10 and @xmath11 is the massless pauli - jordan distribution . we can write down the fourier - representation of the field @xmath12 . it is given by @xmath13 here @xmath14 and @xmath15 , @xmath16 are annihilation and creation operators on a bosonic fock - space . from ( [ hh ] ) we find that they have the following commutation relations @xmath17=b^{{\alpha}{\beta}\mu\nu}\delta^{(3)}(\vec{k}-\vec{k}^{\prime } ) \label{h - com}\ ] ] it should be emphasized that this quantization procedure is only formal in the sense that we do nt describe physical gravitons , since our @xmath12 contains too many degrees of freedom for a real gravitational field . there must be imposed further conditions like a specific gauge condition , etc . in order to obtain physical gravitons . this was done in @xcite . the massive scalar field @xmath18 obeys the klein - gordan equation @xmath19 . it has the following representation in terms of annihilation- and creation - operators : @xmath20 the energy @xmath21 is as usual given by @xmath22 . the operator @xmath23 annihilates a particle with momentum @xmath24 whereas @xmath25 creates an antiparticle with momentum @xmath24 . these operators obey the following commutation relations @xmath26 & = \delta^{(3)}(\vec{p}-\vec{p_{}}^{\prime } ) \\ \bigl[{\tilde{a}}(\vec{p}),{{\tilde{a}}(\vec{p_{}}^{\prime})}^{{\dagger}}\bigr ] & = \delta^{(3)}(\vec{p}-\vec{p_{}}^{\prime } ) \\ \end{split } \label{phi - com}\ ] ] and all the other commutators vanish . the energy - momentum tensor of @xmath18 is given by @xmath27 the interaction with the gravitational field will be described through the following first order coupling @xcite @xmath28 where @xmath29 with newton s constant @xmath30 . this leads to the following explicit form of the first order interaction @xmath31 \label{t1}\ ] ] where we have introduced the trace of the graviton field @xmath32 . for the calculation of the bremsstrahlung process we need to know the explicit form of the @xmath2-matrix up to third order in the coupling constant @xmath33 . this means that we have to construct the time - ordered product @xmath34 . according to the inductive construction of epstein and glaser we proceed by first calculating @xmath35 which will be done in the first subsection and after that we calculate the relevant terms , i.e. terms with four external scalar field operators and one graviton field operator , in the next subsection . for the construction of the time - ordered product @xmath35 we start with the first order term([t1 ] ) and apply the inductive construction due to epstein - glaser as in scharf s book @xcite . first of all we have to build the two distributions @xmath36 and @xmath37 since @xmath38 and @xmath39 are products of normally ordered field operators we apply wick s theorem to obtain a normally ordered expression . we are interested in scattering processes so we have to concentrate on terms with zero or one contraction only . the contraction of the free field operators are defined by @xmath40{.3pt}{.3em}\hrulefill \rule[-.27em]{.3pt}{.3em}\!\ , } } { \rule{0pt}{0.75em}h^{{\alpha}{\beta}}(x_1)h^{\mu\nu}}}(x_2 ) : = & -ib^{{\alpha}{\beta}\mu\nu}d^{(+)}_0(x_1-x_2)\\ { \overset{\makebox { \ , \rule[-.27em]{.3pt}{.3em}\hrulefill \rule[-.27em]{.3pt}{.3em}\!\ , } } { \rule{0pt}{0.75em}{\varphi}(x_1){\varphi}}}(x_2 ) : = & -id^{(+)}_m(x_1-x_2)\\ \end{split}\ ] ] where @xmath41 is the positive frequency part of the massive pauli - jordan distribution . for the splitting operation we need the difference @xmath42 this distribution has causal support and it is given by @xmath43 \\ \end{split}\ ] ] to obtain the time - ordered product @xmath44 we have to split the numerical distribution in @xmath45 according to it s singular order ( see the appendix for the definition of the singular order ) . the singular order of the pauli - jordan distribution is @xmath46 , so the splitting is trivial and we can use the formula @xmath47 the retarded distribution @xmath48 is then given by @xmath49 if we replace all pauli - jordan distributions by the retarded ones @xmath50 . finally the time - ordered distribution @xmath35 is given by @xmath51 for later reference we give here the explicit expression @xmath52 \\ \end{split } \label{t2}\ ] ] this result will be needed in the next subsection for the calculation of @xmath53 . in the previous subsection we have calculated the distribution @xmath44([t2 ] ) from the given first order coupling @xmath54([t1 ] ) . now to obtain the time - ordered product to third order we proceed in much the same way as in the calculation of @xmath44 . we calculate the distributions @xmath55 and @xmath56 which are given by @xmath57 and @xmath58 where @xmath59 and @xmath60 are defined by @xmath61 and @xmath62 it should be noticed that we only collect those terms which have two contractions involving all the spacetime points @xmath63 and @xmath64 , i.e. there is no product of contraction functions with the same argument . after we have built the causal difference @xmath65 we split the numerical part into @xmath66 with support in the backward light cone and @xmath67 with support in the forward light cone . the time - ordered distribution @xmath53 is then given by @xmath68 because the calculation is staightforward and not very enlightning we give only the result @xmath69\\ & \ \times b^{{\alpha}{\beta}{\gamma}{\varepsilon}}d^f_0(x_{\pi(2)}-x_{\pi(3)})\partial^{x_{\pi(3)}}_{{\beta}}\partial^{x_{\pi(1)}}_{{\rho}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 4m^2\bigl[:h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})_{{{^{\scriptstyle{,}}\!\alpha}}}:\\ & \ + : h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\alpha}}}:\bigr]\\ & \ \times d^f_0(x_{\pi(2)}-x_{\pi(3)})\partial^{x_{\pi(3)}}_{{\alpha}}\partial^{x_{\pi(1)}}_{{\sigma}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 4m^2\bigl[:h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(3)}):\\ & \ + : h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(3)})^{{\dagger}}:\bigr]\\ & \ \times d^f_0(x_{\pi(3)}-x_{\pi(2)})\partial^{x_{\pi(1)}}_{{\sigma}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 4m^2\bigl[:h(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}{\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(3)})_{{{^{\scriptstyle{,}}\!\beta}}}:\\ & \ + : h(x_{\pi(1)}){\varphi}(x_{\pi(1)}){\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\varepsilon}}}{\varphi}(x_{\pi(3)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\beta}}}:\bigr]\\ & \ \times b^{{\alpha}{\beta}{\gamma}{\varepsilon}}d^f_0(x_{\pi(3)}-x_{\pi(2)})\partial^{x_{\pi(3)}}_{{\alpha}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 2m^4\bigl[:h(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}{\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(3)}):\\ & \ + : h(x_{\pi(1)}){\varphi}(x_{\pi(1)}){\varphi}(x_{\pi(2)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(2)})_{{{^{\scriptstyle{,}}\!\gamma}}}{\varphi}(x_{\pi(3)})^{{\dagger}}:\bigr]\\ & \ \times d^f_0(x_{\pi(3)}-x_{\pi(2)})d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 2m^4\bigl[:h(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})_{{{^{\scriptstyle{,}}\!\beta}}}:\\ & \ + : h(x_{\pi(1)}){\varphi}(x_{\pi(1)}){\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\beta}}}:\bigr]\\ & \ \times d^f_0(x_{\pi(3)}-x_{\pi(2)})\partial^{x_{\pi(3)}}_{{\beta}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 8m^4\bigl[:h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)}):\\ & \ + : h^{{\rho}{\sigma}}(x_{\pi(1)}){\varphi}(x_{\pi(1)})_{{{^{\scriptstyle{,}}\!\rho}}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})^{{\dagger}}:\bigr]\\ & \ \times d^f_0(x_{\pi(3)}-x_{\pi(2)})\partial^{x_{\pi(1)}}_{{\sigma}}d^f_m(x_{\pi(1)}-x_{\pi(3)})\\ & \ + 4m^6\bigl[:h(x_{\pi(1)}){\varphi}(x_{\pi(1)})^{{\dagger}}{\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)}):\\ & \ + : h(x_{\pi(1)}){\varphi}(x_{\pi(1)}){\varphi}(x_{\pi(2)})^{{\dagger}}{\varphi}(x_{\pi(2)}){\varphi}(x_{\pi(3)})^{{\dagger}}:\bigr]\\ & \ \times d^f_0(x_{\pi(3)}-x_{\pi(2)})d^f_m(x_{\pi(1)}-x_{\pi(3)})\biggr]\\ \end{split } \label{t3}\ ] ] due to the sum over all permutations of the indices this expression for @xmath53 is obviously symmetric in it s arguments as it is required by the definition of the time - ordered product . the third order @xmath2-matrix for the bremsstrahlungs processes is then given by @xmath70 in the preceeding section we have shown explicitly the construction of the time - ordered product to second and to third order . the @xmath2-matrix to third order was given by ( [ s - matrix ] ) . now we consider the scattering process of two massive scalar particles in which a bremsstrahlungs graviton is emited in the final state . that is we want to calculate the expectation value of @xmath71 between the following initial and final states @xmath72 these are wave packets in momentum space where @xmath73 are one particle wave functions from @xmath74 , @xmath75 is a two particle wave function from @xmath76 and @xmath77 is a tensor - valued square integrable wave function on @xmath78 . the vector @xmath79 is the vacuum vector in the asymptotic fock - hilbert space . we want to calculate the following @xmath2-matrix element : @xmath80 to show the details of the calculation we restrict ourself to the following term from the above computed @xmath53 ( [ t3 ] ) @xmath81 where the @xmath82 refers to this particular term of @xmath53 . the @xmath2-matrix element corresponding to this contribution is then given by @xmath83 then we have to compute the vacuum expectation value @xmath84 this can easily be evaluated by inserting the fourier representations of the free field operators ( [ h - field ] ) and ( [ phi - field ] ) and the use of the commutation relations in momentum space ( [ h - com]),([phi - com ] ) . the result is given by @xmath85\,p_1^{{\beta}}\,q_2^{{\rho}}\,p_2^{{\gamma}}\,q_1^{{\varepsilon } } \\ & + \exp\bigl[+i(kx_1-q_1x_3+q_2x_1+p_2x_2-p_1x_2)\bigr]\,q_1^{{\beta}}\,q_2^{{\rho}}\,p_2^{{\gamma}}\,p_1^{{\varepsilon}}\\ & + \exp\bigl[+i(kx_1-p_1x_3+p_2x_1+q_2x_2-q_1x_2)\bigr]\,p_1^{{\beta}}\,p_2^{{\rho}}\,q_2^{{\gamma}}\,q_1^{{\varepsilon}}\\ & + \exp\bigl[+i(kx_1-q_1x_3+p_2x_1+q_2x_2-p_1x_2)\bigr]\,q_1^{{\beta}}\,p_2^{{\rho}}\,q_2^{{\gamma}}\,p_1^{{\varepsilon}}\bigr]\\ \end{split}\ ] ] the @xmath2-matrix element then reads @xmath86\,p_1^{{\beta}}\,q_2^{{\rho}}\,p_2^{{\gamma}}\,q_1^{{\varepsilon } } \\ & + \exp\bigl[+i(kx_1-q_1x_3+q_2x_1+p_2x_2-p_1x_2)\bigr]\,q_1^{{\beta}}\,q_2^{{\rho}}\,p_2^{{\gamma}}\,p_1^{{\varepsilon}}\\ & + \exp\bigl[+i(kx_1-p_1x_3+p_2x_1+q_2x_2-q_1x_2)\bigr]\,p_1^{{\beta}}\,p_2^{{\rho}}\,q_2^{{\gamma}}\,q_1^{{\varepsilon}}\\ & + \exp\bigl[+i(kx_1-q_1x_3+p_2x_1+q_2x_2-p_1x_2)\bigr]\,q_1^{{\beta}}\,p_2^{{\rho}}\,q_2^{{\gamma}}\,p_1^{{\varepsilon}}\bigr]\\ \end{split } \label{sfi3}\ ] ] we introduce the abbreviations @xmath87 for the four parts of @xmath88 . then we consider the spatial integrations in @xmath89 . they can be carried out by inserting the fourier transforms of the testfunctions as well as the feynman propagators : @xmath90\\ = & \ ( 2\pi)^{-10}\int d^4x_1\ldots d^4x_3d^4l_1\ldots d^4l_5\hat{g}(l_1)\ldots\hat{g}(l_3)\exp\bigl[-i(l_1x_1+l_2x_2+l_3x_3)\bigr ] \\ & \ \hat{d}^f_0(l_4)\exp\bigl[-il_4(x_2-x_3)\bigr]\partial^{x_3}_{{\alpha}}\partial^{x_1}_{{\sigma}}\hat{d}^f_m(l_5)\exp\bigl[-il_5(x_1-x_3)\bigr ] \\ & \ \exp\bigl[+i(kx_1-p_1x_3+q_2x_1+p_2x_2-q_1x_2)\bigr]\\ = & \ ( 2\pi)^{-10}\int d^4x_1\ldots d^4x_3d^4l_1\ldots d^4l_5\hat{g}(l_1)\ldots\hat{g}(l_3)\exp\bigl[-i(l_1x_1+l_2x_2+l_3x_3)\bigr ] \\ & \ \exp\bigl[-il_4(x_2-x_3)\bigr]l_5^{{\alpha}}l_5^{{\sigma}}\hat{d}^f_m(l_5)\exp\bigl[-il_5(x_1-x_3)\bigr ] \\ & \ \exp\bigl[+i(kx_1-p_1x_3+q_2x_1+p_2x_2-q_1x_2)\bigr]\\ = & \ ( 2\pi)^2\int d^4l_1\ldots d^4l_5\hat{g}(l_1)\ldots\hat{g}(l_3)\hat{d}^f_0(l_4)l_5^{{\alpha}}l_5^{{\sigma}}\hat{d}^f_m(l_5)\delta^{(4)}(l_1+l_5-k - q_2 ) \\ & \ \delta^{(4)}(l_2+l_4-p_2+q_1)\delta^{(4)}(l_3-l_4-l_5+p_1)\\ \end{split } \label{i1}\ ] ] we can do the integrations w.r.t . @xmath91 and @xmath92 immediately and get @xmath93[-l_1^{{\sigma}}-p_1^{{\sigma}}+k^{{\sigma}}+q_2^{{\sigma}}]\hat{d}^f_m(-l_1+k+q_2)\\ & \ \delta^{(4)}(l_1+l_2+l_3+p_1+q_1-q_2-p_2-k)\\ \end{split}\ ] ] then the first term of @xmath94 becomes @xmath95[-l_1^{{\sigma}}-p_1^{{\sigma}}+k^{{\sigma}}+q_2^{{\sigma}}]\hat{d}^f_m(-l_1+k+q_2)\\ & \ b^{\mu\nu{\rho}{\sigma}}p_1^{{\beta}}q_2^{{\rho}}p_2^{{\gamma}}q_1^{{\varepsilon}}\bigl(32\omega(\vec{k})\omega(\vec{p}_1)\omega(\vec{p}_2)\omega(\vec{q}_1 ) \omega(\vec{q_2})\bigr)^{-1/2}\\ & \ \delta^{(4)}(l_1+l_2+l_3+p_1+q_1-q_2-p_2-k)\\ = & \ ( 2\pi)^{-15/2}i\kappa^3\int d^3\vec{p}_1d^3\vec{q_1}d^3\vec{k}d^3\vec{p}_2d^3\vec{q}_2\hat{\psi}_1(\vec{p}_1)\hat{\psi}_2(\vec{q}_1 ) \hat{\psi}(\vec{p}_2,\vec{q}_2)\hat{\phi}^{\mu\nu}(\vec{k})\\ & \ \int d^4l_1\ldots d^4l_3\hat{g}(l_1)\ldots\hat{g}(l_3)b^{{\alpha}{\beta}{\gamma}{\varepsilon}}\frac{-l_1^{{\alpha}}-p_1^{{\alpha}}+k^{{\alpha}}+q_2^{{\alpha}}}{(l_1+l_3+p_1-k - q_2)^2+i0 } b^{\mu\nu{\rho}{\sigma}}\\ & \ \frac{-l_1^{{\sigma}}-p_1^{{\sigma}}+k^{{\sigma}}+q_2^{{\sigma}}}{m^2-(-l_1+k+q_2)^2-i0}p_1^{{\beta}}q_2^{{\rho}}p_2^{{\gamma}}q_1^{{\varepsilon}}\bigl(32\omega(\vec{k } ) \omega(\vec{p}_1)\omega(\vec{p}_2)\omega(\vec{q}_1)\omega(\vec{q_2})\bigr)^{-1/2}\\ & \ \delta^{(4)}(l_1+l_2+l_3+p_1+q_1-q_2-p_2-k)\\ \end{split } \label{sfi4}\ ] ] where we have inserted the propagators explicitely . in this expression we can carry out the adiabatic limit in the variable @xmath96 since it does nt appear in the argument of the propagators . we do this in the following way . first of all we choose a fixed testfunction @xmath97 with the property @xmath98 . then a new testfunction is defined by @xmath99 the adiabatic limit @xmath100 is then equivalent to the limit @xmath101 . the fourier transform of @xmath102 is given by @xmath103 then we have @xmath104 using this in ( [ sfi4 ] ) the integration w.r.t . @xmath96 gives just the factor @xmath105 . we are interested in the @xmath2-matrix element in the limit where only gravitons at low momenta ( soft gravitons ) are emited . so we omit all small quantities in the numerators , and we take only the leading singularities in the denominators . this is known as the eikonal approximation in the literature , see e.g. @xcite . with this approximation we obtain for the argument of the massless propagator @xmath106 where @xmath107 is the scaling parameter from the adiabatic limit . the argument of the massive propagator becomes @xmath108 since the momentum @xmath109 is on the mass shell . then the @xmath2-matrix element becomes @xmath110 now the adiabatic limit in @xmath111 can be done as above and we arrive at the result for the @xmath2-matrix element @xmath112 there remains to calculate the other three parts @xmath113 . the calculations are similiar to the one above so we only present the results @xmath114 now we are ready to consider the differential cross section . this will be done in detail in the next subsection . we have to calculate the absolute square of the @xmath2-matrix element @xmath94 . we do this again separately for the four parts @xmath116 . for the first part we start with ( [ sfi6 ] ) and write @xmath117 where a @xmath118 indicates complex conjugation . we use the completeness relations of the final states , i.e. @xmath119 where the sum is taken over all final states . then we have @xmath120 now we assume that the wavefunctions @xmath121 and @xmath122 are sharply concentrated around the initial momenta @xmath123 and @xmath124 , so that we can replace the momenta in the propagators by these initial momenta . then the expression can be simplified to @xmath125^{-1}\\ & \ \int d^3\vec{p}_1d^3\vec{q}_1d^3\vec{p_1}^{\prime}d^3\vec{q_1}^{\prime}\hat{\psi}_1(\vec{p}_1)\hat{\psi}_2(\vec{q_1 } ) \hat{\psi}_1(\vec{p_1}^{\prime})^{\ast}\hat{\psi}_2(\vec{q_1}^{\prime})^{\ast}\\ & \ \delta^{(4)}(l_1+p_1+q_1-q_2-p_2-k)\delta^{(4)}(l_2+p_1^{\prime}+q_1^{\prime}-q_2-p_2-k)\\ \end{split } \label{sigma3}\ ] ] we observe that the last integral herein depends on the initial state only . we denote it by @xmath126 , where we have introduced @xmath127 . we have @xmath128\exp\bigl[+i(l_2+p_1^{\prime}+q_1^{\prime}-p)y_2\bigr]\\ = & \ \int d^4y_1d^4y_2\psi_1(y_1)\psi_2(y_1)\psi_1(y_2)^{\ast}\psi_2(y_2)^{\ast}\exp[-il_1y_1]\exp[+il_2y_2]\\ & \ \exp\bigl[-i(y_1-y_2)p\bigr]\\ \end{split}\ ] ] the function @xmath129 is normalized according to @xmath130\ ] ] and it is concentrated around @xmath131 . in the limit of infintely sharp wave packets we may represent it by @xmath132\ ] ] this will be inserted in ( [ sigma3 ] ) and we obtain @xmath133^{-1}\\ & \ \delta^{(4)}(p_2+q_2+k - p_i - q_i)\int d^4y|\psi_1(y)|^2|\psi_2(y)|^2\exp\bigl[-i(l_1-l_2)y\bigr]\\ \end{split } \label{sigma4}\ ] ] for the definition of the cross section we follow sect . 3.4 of @xcite . we can set @xmath134 in the last integral because we neglect all contributions @xmath135 in the numerator , see the remarks after ( [ sfi4 ] ) . with the notation as in @xcite we define the cross section as @xmath136 where @xmath137 is the radius of the beam of incoming particles . then we restrict ourself again to ( [ t3term1 ] ) of @xmath53 and furthermore to the first integral @xmath89 ( [ i1 ] ) of @xmath138 . if we denote the integration variables @xmath139 and @xmath109 by @xmath140 and @xmath141 , respectively , then we obtain for the cross section @xmath142^{-1}\frac{\omega(\vec{p_i})\omega(\vec{q_i})}{\sqrt{\bigl((p_iq_i)^2-m^4\bigr)}}\int d^4l_1d^4l_2\hat{g}(l_1)\hat{g}(l_2)^{\ast}\\ & \ \int\frac{d^3\vec{k}}{2\omega(\vec{k})}\frac{d^3\vec{q_f}}{2\omega(\vec{q_f})}\frac{d^3\vec{p_f}}{2\omega(\vec{p_f})}b^{{\alpha}{\beta}{\gamma}{\varepsilon}}\frac{(q_f^{{\alpha}}- p_i^{{\alpha}})}{(q_f - p_i)^2}\frac{(q_f^{{\rho}}-p_i^{{\rho}})}{2q_f(l_1-k)}p_i^{{\beta}}q_f^{{\rho}}p_f^{{\gamma}}q_i^{{\varepsilon}}b^{{\alpha}^{\prime}{\beta}^{\prime}{\gamma}^{\prime}{\varepsilon}^{\prime}}\\ & \ \frac{(q_f^{{\alpha}^{\prime}}-p_i^{{\alpha}^{\prime}})}{(q_f - p_i)^2}\frac{(q_f^{{\rho}^{\prime}}-p_i^{{\rho}^{\prime}})}{2q_f(l_2-k)}p_i^{{\beta}^{\prime}}q_f^{{\rho}^{\prime } } p_f^{{\gamma}^{\prime}}q_i^{{\varepsilon}^{\prime}}\delta^{(3)}(\vec{p_f}+\vec{q_f}+\vec{k}-\vec{p_i}-\vec{q_i})\\ & \ \delta(\omega(\vec{p_f})+\omega(\vec{q_f})+\omega(\vec{k})-\omega(\vec{p_i})-\omega(\vec{q_i}))\\ \end{split}\ ] ] to carry out the remainig integrations we have to know the integrand explicitely . so at this point of the calculation the tensor structure of the integrand becomes important and we have to write down the various contractions between the four - vectors @xmath143 . with the tensor @xmath144 ( [ b - tensor ] ) we get for the cross section @xmath145\bigl({\omega}(\vec{q_f})^2 -\vec{q_f}^2\bigr)\\ & -\bigl[\bigl({\omega}(\vec{q_f}){\omega}(\vec{p_f})-\vec{q_f}\vec{p_f}\bigr)(p_iq_i)-\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_i})-\vec{q_f}\vec{q_i}\bigr)\bigl({\omega}(\vec{p_i } ) { \omega}(\vec{p_f})-\vec{p_i}\vec{p_f}\bigr)\\ & + \bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\bigl({\omega}(\vec{p_f}){\omega}(\vec{q_i})-\vec{p_f}\vec{q_i}\bigr)+\bigl({\omega}(\vec{p_i}){\omega}(\vec{p_f})- \vec{p_i}\vec{p_f}\bigr)(p_iq_i)\\ & + ( p_iq_i)\bigl({\omega}(\vec{p_i}){\omega}(\vec{p_f})-\vec{p_i}\vec{p_f}\bigr)-p_i^2\bigl({\omega}(\vec{p_f}){\omega}(\vec{q_i})-\vec{p_f}\vec{q_i}\bigr)\bigr]\bigl({\omega}(\vec{q_f } ) { \omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\biggr]^2\\ & \ \frac{\delta^{(3)}(\vec{p_f}+\vec{q_f}+\vec{k}-\vec{p_i}-\vec{q_i})\delta(\omega(\vec{p_f})+\omega(\vec{q_f})+\omega(\vec{k})-\omega(\vec{p_i } ) -\omega(\vec{q_i}))}{(q_f - p_i)^22q_f(l_1-k)(q_f - p_i)^22q_f(l_2-k)}\\ \end{split}\ ] ] it is convenient to choose the center - of - mass system , defined by @xmath146 . we can easily do the integration w.r.t . @xmath147 : @xmath148\bigl({\omega}(\vec{q_f})^2-\vec{q_f}^2\bigr ) -\bigl[\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_f}+\vec{k})\\ & + \vec{q_f}(\vec{q_f}+\vec{k})\bigr)(p_iq_i)-\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_i})-\vec{q_f}\vec{q_i}\bigr)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k } ) + \vec{p_i}(\vec{q_f}+\vec{k})\bigr)\\ & + \bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\bigl({\omega}(\vec{q_f}+\vec{k}){\omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)+\bigl({\omega}(\vec{p_i } ) { \omega}(\vec{q_f}+\vec{k})\\ & + \vec{p_i}(\vec{q_f}+\vec{k})\bigr)(p_iq_i)+(p_iq_i)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k})+\vec{p_i}(\vec{q_f}+\vec{k})\bigr)\\ & -p_i^2\bigl({\omega}(\vec{q_f}+\vec{k}){\omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)\bigr]\bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\biggr]^2\\ & \ \frac{\delta({\omega}(\vec{q_f})+{\omega}(\vec{q_f}+\vec{k})+{\omega}(\vec{k})-2{\omega}(\vec{p_i}))}{(q_f - p_i)^22q_f(l_1-k)(q_f - p_i)^22q_f(l_2-k)}\\ \end{split}\ ] ] now we are left with the integration w.r.t . @xmath149 . in order to be able to do this we have to rewrite the argument of the delta distribution showing the dependence on @xmath150 explicitely . we want to use the formula @xmath151 where the @xmath152 are simple zeros of the function @xmath153 . for that purpose we define @xmath154 where @xmath155 and we have used the notation @xmath156 . in order to simplify the notation in the following formulas we write @xmath157 . the zeros of the function @xmath153 are given by @xmath158 where we have introduced @xmath159 clearly these solutions are itself functions of the momentum @xmath160 and the angle @xmath161 . if we consider the limit of these functions ( [ zeros ] ) as @xmath160 goes to zero we obtain @xmath162 where @xmath163 . the derivative of @xmath153 with respect to @xmath164 is given by @xmath165 by inserting the two zeros of @xmath153 into ( [ fprime ] ) we observe that , by taking the limit @xmath166 , we get a finite result , namely @xmath167 now we can rewrite the cross section as @xmath168\bigl({\omega}(\vec{q_f})^2-\vec{q_f}^2\bigr ) -\bigl[\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_f}+\vec{k})\\ & + \vec{q_f}(\vec{q_f}+\vec{k})\bigr)(p_iq_i)-\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_i})-\vec{q_f}\vec{q_i}\bigr)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k } ) + \vec{p_i}(\vec{q_f}+\vec{k})\bigr)\\ & + \bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\bigl({\omega}(\vec{q_f}+\vec{k}){\omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)+\bigl({\omega}(\vec{p_i } ) { \omega}(\vec{q_f}+\vec{k})\\ & + \vec{p_i}(\vec{q_f}+\vec{k})\bigr)(p_iq_i)+(p_iq_i)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k})+\vec{p_i}(\vec{q_f}+\vec{k})\bigr)\\ & -p_i^2\bigl({\omega}(\vec{q_f}+\vec{k}){\omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)\bigr]\bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\biggr]^2\\ & \ \frac{1}{(q_f - p_i)^22q_f(l_1-k)(q_f - p_i)^22q_f(l_2-k)}\sum_{j=1}^{2}\frac{1}{\lvert f^{\prime}({\mathsf{q_f}})\rvert_{{\mathsf{q_f}}={\mathsf{q_f}}^{(j)}}\rvert } \delta({\mathsf{q_f}}-{\mathsf{q_f}}^{(j)})\\ \end{split}\ ] ] before we do the integration w.r.t . @xmath164 let us denote by @xmath169 the expression which comes from the tensor part in the numerator , i.e. @xmath170\\ & \ \bigl({\omega}(\vec{q_f})^2-\vec{q_f}^2\bigr)-\bigl[\bigl({\omega}(\vec{q_f}){\omega}(\vec{q_f}+\vec{k})+\vec{q_f}(\vec{q_f}+\vec{k})\bigr ) \ ( p_iq_i)-\bigl({\omega}(\vec{q_f})\\ & \ { \omega}(\vec{q_i})-\vec{q_f}\vec{q_i}\bigr)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k})+\vec{p_i}(\vec{q_f}+\vec{k})\bigr ) + \bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\\ & \ \bigl({\omega}(\vec{q_f}+\vec{k}){\omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)+\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k } ) + \vec{p_i}(\vec{q_f}+\vec{k})\bigr)\\ & \ ( p_iq_i)+(p_iq_i)\bigl({\omega}(\vec{p_i}){\omega}(\vec{q_f}+\vec{k})+\vec{p_i}(\vec{q_f}+\vec{k})\bigr)-p_i^2\bigl({\omega}(\vec{q_f } + \vec{k})\\ & \ { \omega}(\vec{q_i})+(\vec{q_f}+\vec{k})\vec{q_i}\bigr)\bigr]\bigl({\omega}(\vec{q_f}){\omega}(\vec{p_i})-\vec{q_f}\vec{p_i}\bigr)\biggr]^2\\ \end{split}\ ] ] then the expression for the cross section is given by @xmath171 from this expression we can identify the differential cross section @xmath172 . we do the @xmath164-integration and obtain @xmath173}\\ & \ \frac{1}{2\bigl[\sqrt{({\mathsf{q_f}}^{(j)})^2+m^2}l_2 ^ 0-{\mathsf{q_f}}^{(j)}{\mathsf{l_2}}\cos{\gamma}-\sqrt{({\mathsf{q_f}}^{(j)})^2+m^2}{\mathsf{k}}+{\mathsf{q_f}}^{(j ) } { \mathsf{k}}\cos\theta\bigr]}\\ \end{split}\ ] ] where we have introduced @xmath174 in the next step we omit all small quantities in the numerator , and we set @xmath175 in the denominators if there occur no singularities . then the expression can be simplified to @xmath176}\\ & \ \frac{1}{2\bigl[{\omega}(\vec{p_i})l_2 ^ 0-{\mathsf{q_f}}^{(j)}|_{{\mathsf{k}}=0}{\mathsf{l_2}}\cos{\gamma}-{\omega}(\vec{p_i}){\mathsf{k}}+{\mathsf{q_f}}^{(j)}|_{{\mathsf{k}}=0}{\mathsf{k } } \cos\theta\bigr]}\\ \end{split}\ ] ] now one observes that the function @xmath177 is independent of @xmath178 and depends only on the initial momenta @xmath123 and @xmath124 so we can take it outside the integral . then we get @xmath179}\\ & \ \frac{1}{2\bigl[{\omega}(\vec{p_i})(l_2 ^ 0-{\mathsf{k}})-{\mathsf{q_f}}^{(j)}|_{{\mathsf{k}}=0}({\mathsf{l_2}}\cos{\gamma}-{\mathsf{k } } \cos\theta)\bigr]}\\ \end{split } \label{dsdo}\ ] ] where the function @xmath180 is given by @xmath181 if we use ( [ limitqf ] ) we can write the differential cross section in the form @xmath182\bigl[{\omega}(\vec{p_i})(l_2 ^ 0 -{\mathsf{k}})-{\mathsf{p_i}}({\mathsf{l_2}}\cos{\gamma}-{\mathsf{k}}\cos\theta)\bigr]}\\ & + f_2(-{\mathsf{p_i}})\int\frac{d^3\vec{k}}{2{\omega}(\vec{k})}\\ & \ \frac{1}{\bigl[{\omega}(\vec{p_i})(l_1 ^ 0-{\mathsf{k}})+{\mathsf{p_i}}({\mathsf{l_1}}\cos{\beta}-{\mathsf{k}}\cos\theta)\bigr]\bigl[{\omega}(\vec{p_i})(l_2 ^ 0 -{\mathsf{k}})+{\mathsf{p_i}}({\mathsf{l_2}}\cos{\gamma}-{\mathsf{k}}\cos\theta)\bigr]}\biggr]\\ \end{split } \label{diffcrosssec}\ ] ] in order to do the integration w.r.t . @xmath183 we use an identity due to feynman @xmath184 ^ 2},\ a , b\in \mathbb{c}\ ] ] we consider the first term in ( [ diffcrosssec ] ) . then we define the functions @xmath185 and @xmath186 by @xmath187 \\ \bigl[{\omega}(\vec{p_i})(l_2 ^ 0-{\mathsf{k}})-{\mathsf{p_i}}({\mathsf{l_2}}\cos{\gamma}-{\mathsf{k}}\cos\theta)\bigr ] \\ \end{split } \label{ab}\ ] ] the integral becomes @xmath188 ^ 2}\\ = & \ \frac{2\pi}{2}\int{\mathsf{k}}d{\mathsf{k}}\int_0 ^ 1dx\int_{-1}^{+1}\frac{d\cos\theta}{\bigl[ax+b(1-x)\bigr]^2}\\ \end{split}\ ] ] we insert @xmath185 and @xmath186 and obtain @xmath189 ^ 2}\\ \end{split } \label{theta - int}\ ] ] we set @xmath190 and we observe that the integral w.r.t . @xmath191 is of the form @xmath192 see @xcite , where we ve set @xmath193 then we obtain for ( [ theta - int ] ) @xmath194 where @xmath195 and @xmath196 are given by @xmath197\\ b(l_1,l_2,x ) = & \ \bigl[\bigl({\omega}(\vec{p_i})l_1 ^ 0-{\mathsf{p_i}}{\mathsf{l_1}}\cos{\beta}\bigr)x+\bigl({\omega}(\vec{p_i})l_2 ^ 0-{\mathsf{p_i}}{\mathsf{l_2}}\cos{\gamma}\bigr)(1-x)\bigr]^2\\ \end{split}\ ] ] then this integral w.r.t . @xmath160 is of the form @xmath198 where the constants @xmath199 are given by @xmath200 in our case the discriminant @xmath201 is given by @xmath202 ^ 2\\ & -4{\omega}(\vec{p_i})^2\bigl[\bigl({\omega}(\vec{p_i})l_1 ^ 0-{\mathsf{p_i}}{\mathsf{l_1}}\cos{\beta}\bigr)x+\bigl({\omega}(\vec{p_i})l_2 ^ 0-{\mathsf{p_i}}{\mathsf{l_2}}\cos{\gamma}\bigr)(1-x)\bigr]^2\\ = & \ -4{\mathsf{p_i}}^2\bigl[\bigl({\omega}(\vec{p_i})l_1 ^ 0-{\mathsf{p_i}}{\mathsf{l_1}}\cos{\beta}\bigr)x+\bigl({\omega}(\vec{p_i})l_2 ^ 0-{\mathsf{p_i}}{\mathsf{l_2}}\cos{\gamma}\bigr)(1-x)\bigr]^2\\ \end{split}\ ] ] so we see that @xmath203 . then the integral ( [ bs44 ] ) is given by @xmath204\ ] ] here we consider gravitons up to an energy @xmath205 which is much smaller than the energy of the incident particles . since every real detector has a finite energy resolution below which he can not detect any particles we have to integrate over all these contributions up to the value @xmath205 . then the final result of the @xmath160-integration is given by the following expression @xmath206\\ \end{split}\ ] ] the second term in ( [ diffcrosssec ] ) can be treated in the same way . the only difference is the sign of @xmath207 in the definition of the functions @xmath185 and @xmath186 , see ( [ ab ] ) , which has no consequences for the calculation of the integrals ( [ bs1 ] ) and ( [ bs44 ] ) . in the preceeding subsection we ve found the differential cross section for bremsstrahlung in a scattering process of two massive scalar particles . now we want to discuss the adiabatic limit . for the differential cross section we ve found @xmath208\\ \end{split } \label{diffcrosssec1}\ ] ] where @xmath209 and @xmath210 . this expression , as it stands , is well defined and finite due to the presence of the testfunctions . to obtain a physically relevant result we have to remove this cutoff and therefore test the infrared behaviour of ( [ diffcrosssec1 ] ) . we will show below that part of this expression becomes singular in the adiabatic limit . in the adiabatic limit in momentum space we let the testfunctions tend to a delta distribution in it s argument . this will be done in the same way as in subsection 5.1 , see ( [ adia ] ) . we scale the argument of the testfunctions with a parameter @xmath107 in the integrals of ( [ diffcrosssec1 ] ) and consider the limit @xmath101 . the functions @xmath211 and @xmath212 are homogeneous of degree 2 in the variables @xmath213 and @xmath96 , i.e. @xmath214 while the function @xmath215 is homogeneous of degree 1 , i.e. @xmath216 . with this in mind let us now look at the various logarithms in ( [ diffcrosssec1 ] ) . the argument of the first logarithm contains the constant @xmath217 so it stays finite if @xmath107 goes to zero . the constant in front of the second and the fourth logarithm has equal powers of @xmath107 in the numerator as well as in the denominator so it tends to a constant . the argument of the second logarithm goes to one , so the logarithm itself goes to zero and the fourth goes to another constant . it remains to consider the third logarithm . this one becomes singular in the adiabatic limit . this can be seen as follows : @xmath218 ^ 2\biggr)\\ & + \text{finite terms}\\ \end{split}\ ] ] if we concentrate on the singular part only we obtain by inserting the scaled testfunctions @xmath219 where we ve introduced @xmath220 . so we get a logarithmic divergence if @xmath107 tends to zero . the origin of this singularity is the massless graviton propagator as can be clearly seen from the above calculation . since all the other terms in ( [ t3 ] ) have the same structure of propagators we would get the same logarihmic divergence as in our example . furthermore there can be no cancellation between these terms since they all have a positive sign . so we can omit the discussion of the other terms . this result is similar to the case of quantum electrodynamics @xcite where a logarithmic divergence in the differential cross section was obtained as well in the bremsstrahlung contribution to the scattering of an electron due to an external classical source . as is well known from qed , the infrared divergence in the bremsstrahlung must cancel against contributions from radiative corrections , otherwise scattering theory would break down . in this appendix we give the definition of the singular order @xmath221 of a numerical distribution @xmath222 . the singular order measures the singularity of the distribution @xmath223 at the origin @xcite . let @xmath222 be given . then the distribution @xmath223 has scaling degree @xmath224 at @xmath225 , if @xmath226 the singular order is then defined by @xmath227 - 4 $ ] , where @xmath228 $ ] is the greatest integer less or equal to @xmath224 . it s an easy excercise to see that differentiation increases the singular order of a distribution , i.e. @xmath229 where @xmath230 for every multi - index @xmath231 with @xmath232 .
quantum chromodynamics ( qcd ) and the gluon field were first proposed by nambu in 1966 @xcite just after the introduction of color degrees of freedom @xcite , and qcd has been established as the fundamental gauge theory of the strong interaction , through the explanation of asymptotic freedom @xcite , the success of perturbative qcd for high - energy phenomena based on the parton model @xcite , and lattice qcd calculations @xcite . in spite of many successes of qcd , there are still unsolved problems in the low - energy region of qcd , owing to its strong - coupling nature . indeed , the mechanism of color confinement is regarded as one of the most difficult important problems in theoretical physics @xcite , and spontaneous chiral - symmetry breaking @xcite is also a difficult issue in describing it quantitatively directly from qcd @xcite . one of the difficulties of qcd lies on the large gap between the fundamental fields ( quarks and gluons ) and the observable particles ( hadrons ) . in fact , quarks and gluons , which are the building blocks of qcd , have no physical asymptotic states , as a result of color confinement or the gauge invariance . this is in contrast to the ordinary perturbation theory , where all the observable phenomena can be directly described with the fundamental fields appearing in the lagrangian . furthermore , the field - theoretical description of the confined particles is an interesting but unsolved difficult subject . in general , the green function is one of the most basic quantities to describe the motions and the interactions of particles @xcite , but most green s functions of quarks and gluons are still unknown in the nonperturbative description . as for the essence of nonperturbative qcd , the central issue is gluon dynamics rather than quarks @xcite . in fact , the strong gluon interaction makes the qcd vacuum highly nontrivial , and color confinement and chiral symmetry breaking are realized even at the quenched level @xcite . then , the analysis of gluon properties is the key point to clarify the nonperturbative aspects of qcd . in particular , the gluon propagator , _ i.e. _ , the two - point green function is one of the most basic quantities in qcd , and has been investigated with much interest in various gauges , such as the landau gauge @xcite , the coulomb gauge @xcite , and the maximally abelian ( ma ) gauge @xcite , in the context of various aspects of qcd . for example , an infrared - vanishing gluon propagator @xcite is proposed from the mathematical analysis of the gribov horizon @xcite , and the infrared singularity of the gluon propagator has been investigated in terms of color confinement @xcite , and also from the viewpoint of renormalon cancellation @xcite . dynamical gluon - mass generation @xcite is also an important subject related to the infrared gluon propagation . in spite of massless perturbative gluons , the gluon field is conjectured to acquire a large effective mass as the self - energy through the self - interaction of gluons in a nonperturbative manner . for , the glueball states , color - singlet bound states of gluons , are considered to be fairly massive _ e.g. _ , about 1.5gev for the lowest @xmath19 and about 2gev for the lowest @xmath20 , as indicated in lattice qcd calculations @xcite . so far , a large effective gluon mass estimated about @xmath21gev has been proposed in various context of qcd physics : analytical framework based on qcd @xcite , lattice qcd calculations @xcite , pomeron physics @xcite , glueball phenomenology @xcite , and heavy - quark phenomenology @xcite . since the color @xmath22 symmetry is unbroken in qcd , the effective - mass generation of gluons is quite different from the higgs mechanism , which is a standard mass generation in quantum field theories . in the electro - weak unified theory , the weak bosons , @xmath23 and @xmath24 , obtain quite a large mass of about 100gev , as a result of the higgs mechanism , _ i.e. _ , spontaneous breaking of the @xmath25 gauge symmetry . however , the gluon mass generation is not a result of spontaneous gauge - symmetry breaking , but stems from much more complicated nonperturbative dynamics of gluons , and generally depends on the gauge choice . the nonperturbative effects originate from the strong - coupling infrared region of qcd . actually , a recent lattice qcd study clarifies that the relevant energy scale for confinement is the infrared gluon component below 1.5gev @xcite : the string tension is almost unchanged even by cutting off high - momentum gluons above 1.5gev . reflecting the asymptotic freedom or the running coupling @xmath26 , qcd exhibits various different features according to the energy scale here , we roughly classify three scale regions of qcd as ultraviolet ( uv ) " , infrared ( ir)/intermediate ( i m ) " , and deep - infrared ( deep - ir ) " regions , in terms of the length @xmath27 . * we define the uv region as @xmath28 0.1fm , which corresponds to the high - energy region more than a few gev s . in this region , perturbative qcd is approximately applicable to the reaction process , and the inter - quark potential is almost coulomb - like @xcite . * we define the ir / im region as 0.1fm @xmath29 1fm , which ranges from a few hundred mev to a few gev in energy . in this region , the system is described by quark - gluon degrees of freedom in a nonperturbative way , which is usually substituted with some effective models @xcite . * we define the deep - ir region as @xmath30 1fm , which is low - energy below @xmath31 0.2gev . in this region , the perturbative running coupling @xmath26 diverges , and the confinement effect is extremely large , so that quark - gluon degrees freedom are hidden and the system is described by hadrons . so far , the gluon propagator has been studied mainly in the landau gauge both in analytic framework @xcite and in lattice qcd @xcite . in the uv region , the gluon propagator takes a massless perturbative form , @xmath32 , apart from the tensor factor . in the ir / im and the deep - ir regions , the perturbative approach breaks down , and we need a nonperturbative approach such as lattice qcd calculations @xcite . basic important ideas were proposed and investigated in the pioneering early - time lattice studies @xcite , and recent lattice studies @xcite followed them and presented high - precision data with the main interests in the deep - ir behavior by using huge - volume lattices @xcite . nevertheless , the functional form of the gluon propagator is unclear still now . in this paper , we study the functional form of the gluon propagator in the landau gauge in su(3 ) lattice qcd monte carlo calculations , especially for _ the ir / im region _ of @xmath33fm , which is considered to be relevant for the quark - hadron physics @xcite , and also aim to describe nonperturbative gluon properties , based on the obtained function form of the gluon propagator . the organization of this paper is as follows . in sec.ii , we give the formalism of the landau gauge fixing and the gluon propagator both in continuum and in lattice qcd . in sec.iii , we show the lattice qcd result of the gluon propagator in the landau gauge both in the coordinate space and in the momentum space . in sec.iv , we estimate the effective gluon mass from the gluon propagator and the effective - mass plot in lattice qcd . in sec.v , we investigate the functional form of the gluon propagator in the landau gauge , by analyzing the lattice qcd data . we show that the landau - gauge gluon propagator is well described with the yukawa - type function in the ir / im region . in sec.vi , as the applications of the yukawa - type gluon propagator , we derive analytic expressions for the zero - spatial - momentum propagator , the effective mass , and the spectral function of the gluon field . section vii is devoted to summary and discussions . in this section , we review the formalism of the landau gauge fixing and the gluon propagator in the euclidean space - time . the landau gauge is one of the most popular gauges in qcd . as a remarkable feature , the landau gauge keeps lorentz covariance and global @xmath22 symmetry . owing to these symmetries and the transverse property , the color and lorentz structure of the gluon propagator is uniquely determined . in the euclidean space - time formalism such as lattice qcd , the landau gauge is usually defined so as to minimize the gauge - field fluctuation . we then expect that only the minimal fluctuation of the gluon field survives in the landau gauge , and the physical essence of gluon properties can be investigated without suffering from large stochastic fluctuations of gauge degrees of freedom . to begin with , let us consider the landau gauge fixing and its physical meaning in euclidean qcd . in the su@xmath34 continuum qcd , the gluon field is expressed as @xmath35 with the generator @xmath36 and @xmath37 . the landau gauge is usually defined by the local condition on the gauge field as @xmath38 in euclidean qcd , the landau gauge has a global definition to minimize the global quantity , @xmath39 by the gauge transformation . this global definition is more strict , and the local condition ( [ eq : locallandau ] ) is derived from the minimization of @xmath40 . the global quantity @xmath40 can be regarded as the total amount of the gauge - field fluctuation in the euclidean space - time . in the global definition , the landau gauge has a clear physical interpretation that it maximally suppresses the artificial gauge - field fluctuation relating to the gauge degrees of freedom . in lattice qcd , the theory is formulated on the discretized space - time @xcite . the qcd action is constructed from the link - variable @xmath41 , instead of the gauge field @xmath42 . the link - variable is defined as @xmath43 , with the lattice spacing @xmath44 and the gauge coupling constant @xmath45 . the gauge transformation of the link - variable is given by @xmath46 with the gauge function @xmath47 . in lattice qcd , the landau gauge fixing is also expressed in terms of the link - variable : the landau gauge is defined by the maximization of @xmath48 by the gauge transformation of the link - variable . for small @xmath44 , using the expansion @xmath49 in terms of lattice spacing @xmath44 , @xmath50 is expressed as @xmath51 therefore , the maximization of @xmath50 corresponds to the minimization of the gauge - field fluctuation as well as the continuum theory . this minimization of the gluon - field fluctuation in turn justifies the expansion in eq.([eq : linkvariableexpansion ] ) . in this subsection , we formulate the gluon propagator in the landau gauge . to begin with , we extract the gluon field from the gauge - fixed link - variable , which is obtained by the gauge transformation of the link - variable to maximize @xmath50 . using the expansion ( [ eq : linkvariableexpansion ] ) , we define the bare gluon field @xmath52 as @xmath53 - \frac{1}{2iagn_c } { \rm tr } \left [ u_\mu(x ) - u_\mu^\dagger(x ) \right ] , \label{eq : gluonfield}\ ] ] where the second term is added to make @xmath54 traceless . the renormalized gluon field @xmath55 is obtained by multiplying a real renormalization factor @xmath56 as @xmath57 we abbreviate @xmath55 as @xmath58 hereafter . the gluon field @xmath58 defined above is traceless and hermite , and is expressed as @xmath35 with @xmath37 . in the landau gauge which maximizes @xmath50 , the gluon field @xmath58 satisfies the local condition , @xmath59 with the forward or backward derivative @xmath60 on the lattice . the gluon propagator @xmath0 is defined by the two - point function as @xmath61 note that the time - ordered product is unnecessary in the euclidean metric , and the translational invariance of the vacuum leads to the @xmath62-dependence . from eq.([eq : locallandaul ] ) , @xmath63 satisfies the transverse property , @xmath64 next , we consider the gluon propagator @xmath65 in the momentum space , which is defined by the fourier transformation of the coordinate - space propagator as @xmath66 on the @xmath67 lattice , this fourier transformation is discretized , and the momentum - space gluon propagator is expressed as @xmath68 here , the discretized momentum @xmath69 and the continuum momentum @xmath70 are defined as @xcite @xmath71 with @xmath72 . in the landau gauge , the color and tensor structure of the propagator is uniquely determined as @xmath73 from the @xmath22 global symmetry , the lorentz symmetry , and the transverse property , @xmath74 therefore , we only have to consider the scalar factor of the gluon propagator , @xmath75 , which is a function of the continuum - momentum squared @xmath76 . in the coordinate space , we investigate the scalar combination of the gluon propagator @xmath77 as a function of the four - dimensional euclidean distance , @xmath78 here , the denominator factor @xmath79 in eq.([eq : defpropagator ] ) has been introduced considering the tensor factor @xmath80 . the scalar factor @xmath81 in eq.([eq : gpmom ] ) is expressed by the fourier transformation of @xmath82 as @xmath83 from which one can prove that @xmath82 depends only on @xmath84 near the continuum limit . in this paper , we call @xmath82 and @xmath81 scalar - type propagator " . we perform @xmath85 lattice qcd monte carlo calculations at the quenched level using the standard plaquette action . here , we adopt three different lattices with the lattice parameter @xmath86=5.7 , 5.8 , and 6.0 . the used lattice size is @xmath87 , @xmath88 , and @xmath89 at @xmath1 = 5.7 , 5.8 , and 6.0 , respectively . in this lattice calculation , we mainly use @xmath90 . the lattice spacing @xmath44 is found to be @xmath91 , and @xmath92fm , at @xmath1 = 5.7 , 5.8 , and 6.0 , respectively , when the scale is determined so as to reproduce the string tension as @xmath93mev from the static q@xmath94 potential @xcite . the number of used gauge configurations is 50 , 40 and 30 for @xmath1 = 5.7 , 5.8 , and 6.0 , respectively . the gauge configurations are picked up every 1,000 sweeps after a thermalization of 20,000 sweeps . we summarize the parameter and calculation conditions in table [ tab : latparam ] . .[tab : latparam ] the lattice parameter @xmath1 , lattice size , and the gauge - configuration number @xmath95 . the corresponding lattice spacing @xmath44 and the lattice volume in the physical unit are added . the lattice spacing @xmath44 is determined so as to reproduce the string tension @xmath93mev . [ cols="^,^,^,^,^",options="header " , ] as a remarkable fact , the yukawa - type function @xmath96 well reproduces the lattice qcd data in the whole region of @xmath7fm . on the other hand , the dipole - type function @xmath97 fails to reproduce the whole region of the lattice data , since it gives too strong reduction at the short distance . the massive propagator @xmath98 also fails to reproduce the whole region of the lattice data , as was already shown . to see the difference of the three fit results clearer , we show in fig.[fig : threefittingratio ] the ratio of the lattice qcd data @xmath99 to the three fit - functions on the scalar - type gluon propagator , _ i.e. _ , @xmath100 , @xmath101 , and @xmath102 . one finds @xmath103 in the whole region of @xmath104fm , while @xmath100 and @xmath102 differ from the unity for small @xmath84 . the ratio of the lattice qcd data @xmath99 at @xmath1 = 6.0 to the fit - functions @xmath98 , @xmath96 and @xmath97 on the scalar - type gluon propagator , _ i.e. _ , @xmath100 , @xmath101 , and @xmath102 . ] we thus find an appropriate functional form of the gluon propagator in the landau gauge . in fact , the scalar - type gluon propagator @xmath82 in the coordinate space is well described by the yukawa - type function @xmath105 with @xmath106gev and @xmath107 in the whole range of @xmath33fm . as a summary figure , we show in fig.[fig : yukawasummary ] the comparison between the obtained yukawa - type function @xmath96 and all the lattice qcd data of the scalar - type gluon propagator @xmath82 at @xmath1=5.7 , 5.8 , and 6.0 in the range of @xmath33fm . for clearer presentation of the yukawa - functional behavior of the lattice gluon propagator @xmath82 , we also show the logarithmic plot of @xmath108 in fig.[fig : yukawasummary](b ) . all the lattice qcd data at @xmath1=5.7 , 5.8 , and 6.0 are found to be well reproduced with the yukawa - type function in the range of @xmath33fm . note also that all the lattice data of @xmath108 locate around a straight line in the logarithmic plot of fig.[fig : yukawasummary](b ) . the yukawa - type function @xmath105 ( solid line ) with @xmath109gev and @xmath110 obtained by the fit analysis at @xmath90 , and the lattice qcd data of the scalar - type gluon propagator @xmath82 in the landau gauge at @xmath1=5.7 , 5.8 , and 6.0 in the range of @xmath33fm . the lower figure is the logarithmic plot of @xmath108 and @xmath111 . , title="fig : " ] the yukawa - type function @xmath105 ( solid line ) with @xmath109gev and @xmath110 obtained by the fit analysis at @xmath90 , and the lattice qcd data of the scalar - type gluon propagator @xmath82 in the landau gauge at @xmath1=5.7 , 5.8 , and 6.0 in the range of @xmath33fm . the lower figure is the logarithmic plot of @xmath108 and @xmath111 . , title="fig : " ] next , we investigate the gluon propagator @xmath75 in the momentum space in terms of the yukawa - type function . in fig.[fig : yukawamom ] , we show the scalar - type gluon propagator @xmath75 in the landau gauge obtained in lattice qcd at @xmath112 , and @xmath113 , which is the fourier transformation of the yukawa - type function @xmath96 . the horizontal axis is @xmath114 . here , we use the same parameters @xmath115gev and @xmath116 as those used in the best - fit analysis for the coordinate - space gluon propagator . from figs . [ fig : latresmom ] and [ fig : yukawamom ] , the lattice qcd data of @xmath81 at @xmath1=5.7 , 5.8 , and 6.0 are found to be approximated with @xmath117 in the range of @xmath118 . we also perform the best - fit analysis of the lattice data of the gluon propagator @xmath81 in the momentum space with @xmath113 . for the lattice data at @xmath1=6.0 in the fit range of @xmath119 , the best - fit parameters are found to be @xmath120gev and @xmath121 , which are close to the values obtained from the fit - analysis of the coordinate - space gluon propagator . the yukawa - type propagator in the momentum space , _ i.e. _ , @xmath113 ( solid line ) with @xmath115gev and @xmath116 , the same values used in fig.[fig : yukawasummary ] . the horizontal axis is @xmath114 . the symbols denote the lattice qcd data of the scalar - type gluon propagator @xmath81 in the landau gauge at @xmath122 , where the momentum is defined as @xmath123 . ] as a caution , in the uv region of @xmath124 , the gluon propagator @xmath75 gradually deviates from the yukawa form @xmath125 and gradually approaches the perturbative form @xmath32 . however , the coordinate - space gluon propagator @xmath82 is found to be almost unchanged for @xmath33fm by correcting the uv deviation . also in the deep - ir region of @xmath126 , as is briefly summarized in appendix b , there appears some deviation between @xmath75 and @xmath127 . in fact , such a deep - ir deviation is indicated by recent huge - volume lattice qcd studies @xcite . in the momentum space , the true gluon propagator @xmath75 turns out to take a saturated value smaller than the yukawa - type propagator @xmath127 in the deep - ir region of @xmath128gev . in other words , @xmath129gev is the lower bound on the applicability of the yukawa - type propagator @xmath127 to the gluon propagator . based on the huge - volume lattice data @xcite , we investigate the deep - ir - corrected gluon propagator in the coordinate space in appendix b. as the conclusion , even after the correction in the deep - ir region , the yukawa - type function is found to work well for @xmath7fm . ( see fig.[fig : hugelatticecorr ] in appendix b. ) as the main conclusion of this paper , we summarize the functional form of the gluon propagator in the landau gauge obtained in su(3 ) lattice qcd . 1 . the coordinate - space gluon propagator @xmath82 in the landau gauge is well described by the four - dimensional yukawa - type function as @xmath130 with @xmath131 600mev and @xmath132 0.16 , for the whole region of @xmath133fm . ( this is valid even after the possible correction in the uv and deep - ir regions . ) the gluon propagator @xmath81 in the momentum space is also well described by the corresponding new - type propagator ( four - dimensional fourier transformed yukawa - type function ) as @xmath134 with @xmath131 600mev and @xmath132 0.16 ( same values ) , in the momentum region of @xmath135 . note here that all the component of the gluon propagator @xmath136 and @xmath65 in the landau gauge can be analytically expressed , starting from the yukawa - type function . ( see sec.ii . ) quantitatively , the yukawa - type propagator @xmath82 exhibits a slower decreasing feature , and @xmath75 exhibits faster decreasing , in comparison with the ordinary massive propagator . it seems suggestive to rewrite @xmath75 as @xmath137 where @xmath138 corresponds to the wave - function renormalization of the gluon field , in a similar manner to the schwinger - dyson formalism . near the on - shell - like condition of @xmath139 , @xmath138 tends to diverge as @xmath140 , which leads to anomalous gluon propagation and may mimic the gluon confinement , _ i.e. _ , the absence of on - shell gluon states . to see the physical meaning of the massive parameter @xmath14 in the yukawa - type function , we compare the yukawa - type propagator form @xmath5 with the massive propagator form @xmath141 in the coordinate space . in spite of a significant difference at the short distance , their difference is just the prefactor at the large distance , where @xmath142 and the main reduction factor is @xmath143 . then , the mass parameter @xmath131 600mev in the yukawa - type gluon propagator directly corresponds to the effective gluons mass in the infrared region . ( see sec.vi c. ) here , we briefly comment on the other functional forms for the gluon propagator . up to now , many functional forms of the gluon propagator @xmath75 have been considered and compared with the lattice qcd result @xcite . some functional forms well describe the gluon propagator better than the yukawa - type function , but they need four or more fit parameters and take highly non - analytical complicated form . on the other hand , the yukawa - type function @xmath96 has _ only two parameters _ @xmath144 and takes an analytical form , which are its advantages . for example , owing to the analyticity , the yukawa - type gluon propagator @xmath96 leads to analytical applications of gluonic nonperturbative quantities , as will be demonstrated in the next section . in this section , as the applications of the yukawa - type gluon propagator , we derive analytical expressions for the zero - spatial - momentum propagator @xmath11 , the effective mass @xmath12 , and the spectral function @xmath13 of the gluon field . all the derivations can be analytically performed , starting from the yukawa - type gluon propagator @xmath96 . second , we consider the zero - spatial - momentum propagator @xmath11 , associated with the yukawa - type propagator @xmath96 . @xmath11 was introduced in sec.iv-b in the context of the effective mass in lattice qcd . here , we mainly deal with the continuum formalism with infinite spatial volume . for the simple argument , we first neglect the temporal periodicity , which is justified for large temporal lattice size . we start from the yukawa - type gluon propagator , @xmath145 with @xmath146 . like eq.([eq : zmpcorrl ] ) , the zero - spatial - momentum propagator is given by @xmath147 in the continuum formalism . using the three - dimensional polar coordinate of @xmath148 , we calculate this integral as follows : @xmath149 with @xmath150 . here , we have used eq.([eq : integformmodbessel ] ) on the modified bessel function . thus , we derive an analytical expression for the zero - spatial - momentum propagator , @xmath151 for the actual comparison with the lattice qcd data , we take account of the temporal periodicity , which is used in lattice calculations . in this case , @xmath11 is given as @xmath152 . \label{eq : corranalytical}\ ] ] in fig.[fig : corryukawa ] , we show the lattice qcd result of @xmath11 in the landau gauge , and the theoretical curve of eq.([eq : corranalytical ] ) with @xmath14=0.624gev and @xmath153=0.162 , the same values used in the previous section . the lattice qcd data of @xmath11 are found to be well described by the theoretical curve , associated with the yukawa - type gluon propagator . the zero - spatial - momentum propagator @xmath11 of gluons in the landau gauge . the symbols are the lattice qcd data at @xmath112 , and the solid line denotes the theoretical curve of eq.([eq : corranalytical ] ) , derived from the yukawa - type propagator with @xmath14=0.624gev and @xmath153=0.162 , the same values in fig.[fig : yukawasummary ] . ] we also consider @xmath11 in the lattice formalism , for more direct comparison with the lattice qcd data . here , we use the expression of @xmath11 with @xmath127 , @xmath154 on the lattice , @xmath155 is given as @xmath156 with @xmath157 ( @xmath158 ) in the lattice unit . then , we obtain an analytical expression for the zero - spatial - momentum propagator , @xmath159 no significant numerical difference turns out to be found between the continuum and the lattice formulae , under the condition of our lattice qcd calculation . this means that the integral quantity @xmath11 is not so sensitive to the details of the uv - behavior of the propagator @xmath96 . in the uv region , the yukawa - type propagator @xmath160 deviates from the correct behavior of the perturbative propagator @xmath32 . for the quantitative estimate of the influence from the deviation in the uv region , we calculate the zero - spatial - momentum propagator @xmath11 using the uv - corrected yukawa propagator @xmath161 , which is @xmath127 for @xmath162gev and @xmath32 in the uv region of @xmath163gev . the difference of @xmath11 is found to be at most 1% between the cases with @xmath127 and @xmath161 . thus , the integral quantity @xmath11 is insensitive to the uv behavior of the gluon propagator , so that it is also expected to be insensitive to the discritization error @xcite in the uv region in the lattice calculation . second , we investigate the effective mass @xmath12 , as the consequence of the yukawa - type propagator @xmath96 . the effective - mass plot is a general useful technique for the mass estimation in lattice qcd , and was actually examined for gluons in sec.iv-b . for simplicity , we here treat the three - dimensional space as a continuous infinite - volume space , while the temporal variable @xmath164 is discrete and periodic . for the simple notation , we here use the lattice unit for @xmath164 . when the temporal periodicity can be neglected , the zero - spatial - momentum propagator @xmath11 is expressed by eq.([eq : d0yukawa ] ) , and we obtain an analytical expression of the effective mass , @xmath165 from the asymptotic form of @xmath166 , the effective mass of gluons is approximated as @xmath167 for large @xmath164 . this functional form indicates that @xmath12 is an increasing function and approaches @xmath14 from below , as @xmath164 increases . note that the mass parameter @xmath14 in the yukawa - type gluon propagator directly corresponds to the effective mass @xmath12 of gluons for large @xmath164 . in fact , @xmath131 600mev has a definite physical meaning of the effective gluon mass in the infrared region . note also that the simple analytical expression of eq.([eq : emgyukawa ] ) or ( [ eq : meffasymptotic ] ) reproduces the anomalous increasing behavior of the effective mass @xmath12 of gluons , as was observed in fig.[fig : effmass ] . as for the increasing behavior , there is a general argument : this can occur when the spectral function is not positive definite @xcite , although the concrete form of the gluon spectral function is not yet known . instead , this framework with the yukawa - type gluon propagator gives an analytical and quantitative method , and is found to well reproduce the lattice result . ( the actual comparison is demonstrated with eq.([eq : meffanaytical ] ) . ) next , we take account of the temporal periodicity , which is used in lattice qcd calculations . in this case , the effective mass @xmath12 is defined by cosh - type " as eq.([eq : emcosh ] ) , and the zero - spatial - momentum propagator @xmath11 is given by eq.([eq : corranalytical ] ) . then , the effective mass @xmath12 of gluons is expressed as @xmath168 } { \cosh\left [ m_{\rm eff}(t ) ( n_t/2-t)\right]}= \\ & & \frac{(t+1)k_1(m(t+1 ) ) + ( n_t -(t+1))k_1(m(n_t-(t+1)))}{tk_1(mt)+(n_t - t)k_1(m(n_t - t))}. \nonumber\end{aligned}\ ] ] in fig.[fig : effmassyukawa ] , we show the theoretical curve obtained by eq.([eq : meffanaytical ] ) together with the lattice result of @xmath12 . here , we take @xmath14=0.624gev , the same value used in the previous section . the lattice qcd data of @xmath12 are found to be well described by the theoretical curve , associated with the yukawa - type gluon propagator . the effective mass @xmath12 of gluons in the landau gauge . the symbols denote the lattice qcd data at @xmath112 , and the solid line denotes the theoretical curve of eq.([eq : meffanaytical ] ) derived from the yukawa - type propagator with @xmath14=0.624gev , the same value used in fig.[fig : yukawasummary ] . ] from the analytical expression of the zero - spatial - momentum propagator @xmath169 in eq.([eq : d0yukawa ] ) , we can derive the spectral function @xmath13 of the gluon field , associated with the yukawa - type gluon propagator . for simplicity , we take continuum formalism with infinite space - time . the relation between the spectral function @xmath13 and the zero - spatial - momentum propagator @xmath11 is given by the laplace transformation , @xmath170 when the spectral function is given by a @xmath17-function such as @xmath171 , which corresponds to a single mass spectrum , one finds a familiar relation of @xmath172 . for the physical state , the spectral function @xmath13 gives a probability factor , and is non - negative definite in the whole region of @xmath173 . this property is related to the unitarity of the @xmath174-matrix . in general , the laplace transformation is defined by @xmath175 and the inverse laplace transformation is expressed as @xmath176 then , from eq.([eq : corrvssf ] ) , the spectral function @xmath13 is expressed as @xmath177 with @xmath178 , @xmath179 and @xmath180 . performing the partial integration in eq.([eq : integformmodbessel ] ) on the modified bessel function , we obtain a formula of the laplace transformation , @xmath181 which leads to the inverse laplace transformation , @xmath182 by differentiating this formula by @xmath173 , we find @xmath183 where the second term includes an infinite factor besides the @xmath17-function . then , we apply this formula to eq.([eq : sfint ] ) , and obtain the spectral function @xmath13 as @xmath184 for more rigorous derivation , we avoid the singularity at @xmath18 by regularizing eq.([eq : ltk1 ] ) as @xmath185 with a positive infinitesimal @xmath186 , and find the formula of @xmath187 which leads to the regularized spectral function , @xmath188 in the calculation of the laplace transformation from @xmath13 to @xmath11 , we can avoid the divergence of the integral at @xmath18 by using the regularized spectral function @xmath189 . then , we can perform the integration , and properly obtain @xmath190 , by taking the limit of @xmath191 after the integration . we have numerically confirmed that the spectral function @xmath192 in eq.([eq : sfyukawar ] ) reproduces @xmath190 by the laplace transformation in the limit of @xmath191 . in this way , we derive the spectral function @xmath13 of the gluon field , associated with the yukawa - type propagator : @xmath193 which is regularized as @xmath189 in more rigorous derivation . here , @xmath131 600mev is the mass parameter in the yukawa - type function for the landau - gauge gluon propagator . the first term expresses a negative continuum spectrum , and the second term a @xmath17-functional peak with the residue including an infinite factor , which is positive as @xmath194 at @xmath195 . we show in fig.[fig : spectral ] the spectral function @xmath13 of the gluon field . although the appearance of the negative - value region in the gluon spectral function is expected , @xmath13 exhibits two anomalous behaviors : it has a positive @xmath17-functional peak with the residue of @xmath140 at @xmath196 , and it takes negative values for all the region of @xmath197 . this negative contribution of the spectral function gives unusual behavior of the effective mass @xmath12 of gluons , _ i.e. _ , its increasing behavior on @xmath164 . the spectral function @xmath13 of the gluon field , associated with the yukawa - type propagator . the unit is normalized by the mass parameter @xmath131 600mev . as eq.([eq : sfyukawa ] ) indicates , @xmath13 shows anomalous behaviors : it has a positive @xmath17-functional peak with the residue of @xmath140 at @xmath196 , and takes negative values for all the region of @xmath197 . ] as was discussed in sec.iv , if the spectral function is non - negative definite , like the ordinary hadronic correlator , the effective mass must be a monotonously decreasing function , due to the mathematical nature of the summation of @xmath198 with non - negative coefficients , as can be proven with eq.([eq : dmdt ] ) . physically , this is due to a larger positive contribution of the excited states to the effective mass at a shorter distance . however , if the spectral function includes the negative - value region , such a definite tendency is lost @xcite . this is the case of the gluon field in the landau gauge . more precisely , from eq.([eq : dmdt ] ) , we find that the increasing property of @xmath12 can be realized , only when there is some suitable coexistence of positive- and negative - value regions in the spectral function @xmath13 . as a remarkable fact , the obtained gluon spectral function @xmath13 is negative - definite for all the region of @xmath197 , except for the positive @xmath17-functional peak at @xmath18 . the negative property of the spectral function in coexistence with the positive peak leads to the anomalous increasing behavior of the effective mass @xmath12 . in the uv region , the true gluon propagator deviates from the yukawa - type function . then , we estimate the contribution from the uv part of this spectral function @xmath13 to the zero - spatial - momentum propagator @xmath11 . introducing the uv cutoff @xmath199 , we define the integration @xmath200 which reduces to the laplace transformation in the @xmath201 limit . for @xmath202 0.1fm , the @xmath199-dependence of the integral quantity @xmath203 is found to be negligible for @xmath204 gev , _ i.e. _ , @xmath205 . in other words , due to this uv - insensitivity , the obtained information on the uv part of the spectral function suffers from a large uncertainty , as is also seen in the maximum entropy method ( mem ) analysis . on the other hand , the ir / im part of the spectral function @xmath13 is relatively stable and reliable . we note that the gluon spectral function @xmath13 is divergent at @xmath195 , and the divergence structure is complicated and consists of two ingredients : a @xmath17-functional peak with a positive infinite residue and a negative wider power - damping peak . on the finite - volume lattice , these singularities are to be smeared , and @xmath13 is expected to take a finite value everywhere on @xmath173 . on the lattice , we conjecture that the spectral function @xmath13 includes a narrow positive peak stemming from the @xmath17-function in the vicinity of @xmath206 and a wider negative peak near @xmath207 in the region of @xmath197 . in this way , the yukawa - type gluon propagator suggests an extremely anomalous spectral function of the gluon field in the landau gauge . note that this framework gives an analytical and concrete expression for the gluon spectral function @xmath13 at the quantitative level . actually , the resulting effective mass @xmath12 well describes the lattice result , as shown in fig.[fig : effmass ] . the obtained gluon spectral function @xmath13 is negative almost everywhere , and includes a complicated divergence structure near the anomalous threshold " , @xmath208 . these anomalous features of the gluon spectral function may have some relation to the various nonperturbative qcd phenomena , such as the gluon confinement and the gluonic instability of the qcd vacuum , _ e.g. _ , gluon condensation , the savvidy vacuum @xcite , and the copenhagen vacuum @xcite . in any case , the yukawa - type gluon propagator @xmath96 , which models the landau - gauge gluon propagator , would be useful for the analytical and quantitative investigation of nonperturbative qcd . we have studied the gluon propagator @xmath0 in the landau gauge in the infrared / intermediate region of @xmath209 1.0fm , which is relevant to the quark - hadron physics , in su(3 ) lattice qcd at @xmath1 = 5.7 , 5.8 , and 6.0 at the quenched level . from the gluon propagator analysis and the effective - mass plot , we have estimated the effective gluon mass of @xmath2mev in the infrared region of @xmath210fm . the effective gluon mass exhibits a significant @xmath84-dependence in this region : it takes a smaller value in the smaller @xmath84-region . we have also studied the functional form of the gluon propagator @xmath0 in lattice qcd . as a remarkable fact , the lattice qcd result of the landau - gauge gluon propagator @xmath4 is fairly well described by the yukawa - type form @xmath211 with the mass parameter @xmath6mev in the whole region of @xmath7fm in four - dimensional euclidean space - time . this yukawa - type propagator corresponds to the new - type propagator @xmath160 in the momentum space , through the fourier transformation , and this also well describes the lattice qcd result of the gluon propagator @xmath212 in the momentum space . as the application of the yukawa - type gluon propagator , we have derived the analytical expression of the zero - spatial - momentum propagator @xmath11 , and the effective mass @xmath12 . the obtained analytical functions for @xmath11 and @xmath12 well reproduce the lattice qcd results , in particular the anomalous increasing behavior of @xmath12 . we have found that the mass parameter @xmath14 of the yukawa - type gluon propagator directly corresponds to the effective gluon mass in the infrared region of @xmath151fm . we have also derived the analytical expression of the spectral function @xmath13 of the gluon field , associated with the yukawa - type gluon propagator , using the inverse laplace transformation of the temporal propagator @xmath11 . as a remarkable fact , the obtained spectral function @xmath13 is negative - definite almost everywhere for @xmath197 , except for a positive @xmath17-functional peak with the residue of @xmath140 at @xmath213 . the coexistence of negative- and positive - value regions of @xmath13 lead to the anomalous increasing behavior of the effective mass @xmath12 of gluons . thus , the theoretical analysis with the yukawa - type gluon propagator gives a new analytical and quantitative method for the nonperturbative gluonic phenomena . the yukawa function @xmath211 in the coordinate space is highly analytic , and this analyticity plays an important role in deriving the analytical expressions for @xmath11 , @xmath12 , and @xmath13 . on the other hand , the momentum - space propagator @xmath160 includes a singular cut stemming from the square root . if this singularity is taken seriously , there would arise a problem , because analyticity of the green function is important in quantum field theories , _ e.g. _ , in the wick rotation converting between the euclidean space and the minkowski space . of course , this yukawa - type function is an approximate form for the landau - gauge gluon propagator in the region of @xmath214fm . there would be more regular and more thorough expression for the gluon propagator . in this paper , we have mainly considered the gluon propagator in the coordinate space instead of the momentum space , since the coordinate - space variable is more directly obtained in lattice qcd . for the confined particles , however , the reason to use the momentum representation would be less clear , compared with ordinary particles . in the ordinary particles , the momentum representation of the green function is clearly useful to express the pole structure , to distinguish the on - shell and off - shell states , and so on . however , for the confined field , there is no on - shell state , _ i.e. _ , no physical asymptotic state , so that there is no definite reason to use the momentum representation , besides the total momentum conservation . indeed , it is difficult to image the non - zero momentum - space propagator with no pole , and the coordinate - space representation may be more convenient for some description of the confined field , similar to potential problems in quantum mechanics , where the coordinate - space wave - function is convenient . the yukawa - type gluon propagator @xmath96 includes @xmath143 as the main reduction factor in the infrared region , and @xmath215gev can be regarded as the infrared effective gluon mass . in terms of the infrared reduction , a simple constituent gluon picture " may be approximately obtained as @xmath216 for the glueball mass @xmath217 . in general , the glueball mass @xmath217 can be estimated from the infrared reduction of the glueball correlator @xmath218 with the glueball operator , _ e.g. _ , @xmath219 for the scalar glueball , with the field strength tensor @xmath220 @xcite . by the wick contraction , @xmath221 can be expressed by some combination of the gluon green functions . in the framework with the yukawa - type gluon propagator , the leading reduction term of @xmath221 can be expressed with some derivative of @xmath222 for large @xmath84 . then , @xmath221 includes @xmath223 as one of the main infrared reduction factors , and the lowest - glueball mass is roughly estimated as @xmath224 1.3gev , by neglecting all the prefactor and higher - order terms , which express the interactions between gluons . in spite of the crude estimate , this value gives the same order of the lowest - glueball mass of about 1.5gev obtained in lattice qcd @xcite . in this subject , there remain difficult problems related to the confinement mechanism , the infinite - volume limit , the gribov copies , and the gauge dependence . for example , it is a highly difficult mathematical problem to find out the precise description of the physical hadronic states like glueballs in terms of the confined particles , quarks and gluons . as other example , zwanziger s theorem @xcite is derived from the argument of the gribov horizon : all connected gluon correlation functions including the gluon propagator must vanish at zero momentum in the infinite - volume limit . however , our lattice qcd results and the yukawa - type propagator @xmath75 indicate an infrared non - vanishing property of the gluon propagator as @xmath225 . this may be due to the absence of an infinite - volume effect in the deep - ir region , as is conjectured by analytical studies @xcite . for this problem , it is desired to clarify the deep - ir behavior of the gluon propagator , and recent huge - volume lattice studies @xcite and a recent analytical study based on the schwinge - dyson equation @xcite also indicate the infrared non - vanishing property of the gluon propagator . as for the gribov - copy problem , it is reported that the gribov - copy effect is quantitatively rather small in the actual lattice qcd calculation for the landau - gauge gluon propagator @xcite , while the ghost propagator slightly suffers from it . however , this is a fundamental problem in qcd , and it would be serious in the argument of the large - volume limit , so that it is also desired to remove the gribov copies and to extract the fundamental moduli region . finally , we discuss the yukawa - type gluon propagation and a possible dimensional reduction due to the stochastic behavior of the gluon field in the infrared region . as shown in this paper , the landau - gauge gluon propagator is well described by the yukawa function in _ four_-dimensional euclidean space - time . however , the yukawa function @xmath5 is a natural form in _ three_-dimensional euclidean space - time , since it is obtained by the three - dimensional fourier transformation of the ordinary massive propagator @xmath226 . in fact , the yukawa - type propagator has a three - dimensional " property . in this sense , as an interesting possibility , we propose to interpret this yukawa - type behavior of the gluon propagation as an effective reduction of the space - time dimension " . such a dimensional reduction " sometimes occurs in stochastic systems , as parisi and sourlas pointed out for the spin system in a random magnetic field @xcite . on the infrared dominant diagrams , the @xmath227-dimensional system coupled to the gaussian - random external field is equivalent to the @xmath228-dimensional system without the external field . in fact , the space - time dimension of the theory is apparently reduced by two . for the system coupled to the gaussian - random external source , the dimensional reduction is universal and is associated with a hidden supersymmetry : in the superspace formalism with @xmath229 , the integration over two grassmann - variables ( @xmath230 ) reduces the space - time coordinates @xmath231 by two @xcite . we note that the gluon propagation in the qcd vacuum resembles the situation of the system coupled to the stochastic external field . in fact , as is indicated by a large positive value of the gluon condensate @xmath232 in the minkowski space , the qcd vacuum is filled with a strong color - magnetic field @xcite , which can contribute spontaneous chiral - symmetry breaking @xcite , and the color - magnetic field is considered to be highly random at an infrared scale @xcite . since gluons interact each other , the propagating gluon is violently scattered by the other gluon fields which are randomly condensed in the qcd vacuum at the infrared scale , as schematically shown in fig.[figgluonprogagation ] . a schematic figure for a propagating gluon in the qcd vacuum . the qcd vacuum is filled with color - magnetic fields which are stochastic at an infrared scale , and the gluon propagates in the random color - magnetic fields . ] actually in the infrared region , the gluon field shows a strong randomness due to the strong interaction , and this infrared strong randomness is considered to be responsible for color confinement , as is indicated in strong - coupling lattice qcd @xcite . in the chiral random matrix theory for qcd @xcite , the infrared randomness of gluons is taken account as an essence of qcd in a simplified manner , and the gluon field is replaced by a gaussian - random external field coupled to the quark field . notice that there are two kinds of randomness in the gluon field : one is a completely - random gauge degree of freedom , which is fake , and the other is a net physical gluonic degree of freedom , which is not completely but highly random at the infrared scale . for the argument of physical randomness , these two concepts should be clearly divided , since the gauge degree of freedom is just a fake . even after the removal of the fake gauge degree of freedom by gauge fixing , the gluon field exhibits a strong randomness @xcite accompanying a quite large fluctuation at the infrared scale . as a generalization of the parisi - sourlas mechanism , we conjecture that the infrared structure of a theory in the presence of the quasi - random external field in higher - dimensional space - time has a similarity to the theory without the external field in lower - dimensional space - time . from this point of view , the yukawa - type behavior of gluon propagation may indicate an effective reduction of the space - time dimension " by one , due to the stochastic interaction between the propagating gluon and the other gluon fields in the qcd vacuum , of which net physical fluctuation is highly random at the infrared scale . in this paper , the yukawa - type gluon propagator is obtained phenomenologically from lattice qcd results , but we expect some deeper theoretical reasons for the yukawa - type propagation , which may be an effective reduction of the space - time dimension , due to the stochastic behavior of the infrared gluon field . in any case , the yukawa - type gluon propagator would provide a new analytical framework for the study of nonperturbative qcd . h. s. is grateful to professor j.m . cornwall for his useful suggestions on the dynamical gluon mass . h. s. is supported in part by the grant for scientific research [ ( c ) no . 19540287 ] from the ministry of education , culture , science and technology ( mext ) of japan . this work is supported by the global coe program , the next generation of physics , spun from universality and emergence " at kyoto university . the lattice qcd calculations have been done on nec - sx8 at osaka university . in this appendix , we derive several fourier transformations used in sec.[sec : functionalforms ] . we here calculate the fourier transformation of the massive propagator @xmath226 . first , we consider the fourier integral of @xmath226 in the four - dimensional euclidean space - time , @xmath243 by rotating the coordinate , we set @xmath244 without loss of generality . then , the integral is expressed as @xmath245 with @xmath246 . with the three - dimensional polar coordinate of @xmath247 , @xmath248 is written as @xmath249 with @xmath250 and @xmath251 . using the integral representation of the modified bessel function , @xmath252 we obtain the fourier transformation formula , @xmath253 next , we consider the fourier integral of @xmath226 in the two - dimensional euclidean space - time , @xmath254 by rotating the coordinate , we set @xmath255 without loss of generality , and integrate @xmath256 as @xmath257 using mehler s integral representation of the modified bessel function , @xmath258 we obtain the fourier transformation formula , @xmath259 in the deep - ir region , there has been reported to appear some deviation on the momentum - space gluon propagator @xmath75 between small - size lattice data and huge - volume lattice data @xcite . in this appendix , we demonstrate that the deviation in the deep - ir region does not affect the yukawa - type behavior of the coordinate - space gluon propagator @xmath82 in the ir / im region of @xmath33fm . figure [ fig : hugelattice ] shows the scalar - type gluon propagator @xmath81 in the recent lattice - qcd calculation with a huge volume @xmath265 at @xmath266 , taken from ref.@xcite . here , a renormalization constant is multiplied for the huge - volume lattice data so as to adjust them to the renormalization condition ( [ renormcond ] ) at @xmath267=4gev . in the momentum space , the true gluon propagator @xmath75 turns out to take a saturated value smaller than the yukawa - type propagator @xmath127 in the deep - ir region of @xmath128gev . in other words , @xmath129gev is the lower bound on the applicability of the yukawa - type propagator @xmath127 to the gluon propagator . the infrared behavior of the gluon propagator @xmath75 . the triangle symbols denote recent huge - volume lattice data taken from ref.@xcite . the solid line denotes the yukawa - type propagator @xmath117 , and the dashed line the deep - ir - corrected propagator @xmath268 with @xmath269gev . ] taking account of the deviation in the deep - ir region , we define the deep - ir - corrected momentum - space propagator as @xmath270 with the ir - saturation momentum of @xmath271 . this value of @xmath272 is determined so as to consist with the huge - volume lattice result in the deep - ir region . using this deep - ir - corrected propagator @xmath273 , we calculate the deep - ir - corrected coordinate - space propagator by the fourier transformation as @xmath274 in fig.[fig : hugelatticecorr ] , we show the yukawa - type propagator @xmath96 and this deep - ir - corrected propagator @xmath275 , together with the lattice qcd data . the difference between @xmath96 and @xmath275 is fairly small for @xmath7 fm . the yukawa - type propagator @xmath96 ( solid line ) , and deep - ir - corrected propagator @xmath275 ( dashed - line ) , together with the lattice data . the difference between them is fairly small in the ir / im region of @xmath33fm . ] 99 y. nambu , proc . of _ preludes theoretical physics , in honor of v.f . weisskopf _ ( north - holland , 1966 ) han and y. nambu , phys . rev . * 139 * , b1006 ( 1965 ) . gross and f. wilczek , phys . lett . * 30 * , 1343 ( 1973 ) ; h.d . politzer , phys . lett . * 30 * , 1346 ( 1973 ) . feynman , proc . of _ high energy collision of hadrons _ ( stony brook , n.y . , 1969 ) ; bjorken and e.a . paschos , phys . rev . * 185 * , 1975 ( 1969 ) . w. greiner , s. schramm , and e. stein , _ quantum chromodynamics _ , ( springer , 2007 ) , and its related references . wilson , phys . d*10 * , 2445 ( 1974 ) ; j.b . kogut and l. susskind , phys . rev . d*11 * , 395 ( 1975 ) . m. creutz , phys . rev . lett . * 43 * , 553 ( 1979 ) ; phys . d * 21 * , 2308 ( 1980 ) . as a recent textbook , _ e.g. _ , h.j . rothe , _ lattice gauge theories : an introduction _ , 3rd edition , ( world scientific , 2005 ) , and its related references . for example , yang - mills and mass gap " is nominated in millennium problems by clay mathematics institute . see the website : http://www.claymath.org/millennium/ y. nambu , g. jona - lasinio , phys . rev . * 122 * , 345 ( 1961 ) ; phys . rev . * 124 * , 246 ( 1961 ) . miransky , _ dynamical symmetry breaking in quantum field theories _ ( world scientific , 1993 ) ; + k. higashijima , prog . . suppl . * 104 * , 1 ( 1991 ) . as an instructive review , e.g. _ , j.e . mandula , phys . rept . * 315 * , 273 ( 1999 ) , and its references . j.m . cornwall , phys . rev . d*26 * , 1453 ( 1982 ) ; phys . rev . d*76 * , 025012 ( 2007 ) , and references therein . gribov , nucl . phys . * b139 * , 1 ( 1978 ) . d. zwanziger , nucl . phys . * b364 * , 127 ( 1991 ) ; phys . b*257 * , 168 ( 1991 ) ; nucl . phys . * b378 * , 525 ( 1992 ) ; phys . rev . d*65 * , 094039 ( 2002 ) ; phys . d*69 * , 016002 ( 2004 ) . m. stingl , phys . rev . d*34 * , 3863 ( 1986 ) ; _ erratum - ibid . _ d*36 * , 652 ( 1987 ) ; u. hbel , r. konning , h.g . reusch , m. stingl , and s. wigard , z. phys . a * 336 * , 423 ( 1990 ) . l. von smekal , a. hauck , and r. alkofer , phys . * 79 * , 3591 ( 1997 ) ; r. alkofer and l. von smekal , phys . rept . * 353 * , 281 ( 2001 ) , and references therein . aguilar , d. binosi , and j. papavassiliou , phys . rev . d*78 * , 025010 ( 2008 ) . kondo , phys . b*678 * , 322 ( 2009 ) . mandula and m. ogilvie , phys . b*185 * , 127 ( 1987 ) . r. gupta , g. guralnik , g. kilcup , a. patel , s.r . sharpe , and t. warnock , phys . d * 36 * , 2813 ( 1987 ) . bernard , c. parrinello , and a. soni , phys . * , 1585 ( 1994 ) . p. marenzoni , g. martinelli , n. stella , and m. testa , phys . b*318 * , 511 ( 1993 ) ; p. marenzoni , g. martinelli , and n. stella , nucl . phys . * b455 * , 339 ( 1995 ) . a. cucchieri , nucl . * b508 * , 353 ( 1997 ) ; nucl . phys . * b521 * , 365 ( 1998 ) . d.b . leinweber , j.i . skullerud , a.g . williams , and c. parrinello , phys . rev . d*58 * , 031501 ( 1998 ) ; phys . rev . d*60 * , 094507 ( 1999 ) , _ erratum - ibid . _ d*61 * , 079901 ( 2000 ) . d. becirevic , ph . boucaud , j.p . leroy , j. micheli , o. pne , j. rodriguez - quintero , c. roiesnel , phys . d*60 * , 094509 ( 1999 ) ; phys . rev . d*61 * , 114508 ( 2000 ) . bonnet , p.o . bowman , d.b . leinweber , and a.g . williams , phys . rev . d*62 * , 051501 ( 2000 ) ; f.d.r . bonnet , p.o . bowman , d.b . leinweber , a.g . williams , and j.m . zonotti , phys . rev . d*64 * , 034501 ( 2001 ) . k. langfeld , h. reinhardt , and j. gattnar , nucl . b621 * , 131 ( 2002 ) . s. furui and h. nakajima , phys . rev . d*69 * , 074505 ( 2004 ) . bowman , u.m . heller , d.b . leinweber , m.b . parappilly , and a.g . williams , phys . rev . d*70 * , 034509 ( 2004 ) ; phys . rev . d*76 * , 094505 ( 2007 ) . a. sternbeck , e .- ilgenfritz , m. mueller - preussker , and a. schiller , phys . d*72 * , 014507 ( 2005 ) . bogolubsky , e .- ilgenfritz , m. mller - preussker , and a. sternbeck , phys . b*676 * , 69 ( 2009 ) ; pos(lat2007 ) , 290 ( 2007 ) . p.j . silva and o. oliveira , phys . rev . d*74 * , 034513 ( 2006 ) ; a. cucchieri , t. mendes , o. oliveira , and p.j . silva , phys . d*76 * , 114507 ( 2007 ) . a. cucchieri and t. mendes , pos(lat2007 ) , 297 ( 2007 ) ; phys . rev . lett . * 100 * 241601 ( 2008 ) . j. greensite and s. olejnik , phys . rev . d*67 * , 094503 ( 2003 ) ; j. greensite , s. olejnik , and d. zwanziger , phys . rev . d*69 * , 074506 ( 2004 ) , and their references . a. cucchieri and d. zwanziger , phys . b*524 * , 123 ( 2002 ) , and references therein . k. amemiya and h. suganuma , phys . rev . d*60 * , 114509 ( 1999 ) ; h. suganuma , k. amemiya , a. tanaka , and h. ichie , nucl . * a670 * , 40 ( 2000 ) ; h. suganuma , k. amemiya , h. ichie , n. ishii , h. matsufuru , t.t . * b * ( proc . suppl . ) * 106 - 107 * , 679 ( 2002 ) . kondo , phys . rev . d*57 * , 7467 ( 1998 ) ; phys . rev . d*58 * , 105016 ( 1998 ) ; phys . rev . d*58 * , 105019 ( 1998 ) . s. mandelstam , phys . rev . d*20 * , 3223 ( 1979 ) . m. baker , j. s. ball , and f. zachariasen , nucl . b186 * , 531 , 560 ( 1981 ) ; phys . rept . * 209 * , 73 ( 1991 ) . h. suganuma , s. sasaki and h. toki , nucl . phys . * b435 * , 207 ( 1995 ) ; prog . . phys . * 94 * , 373 ( 1995 ) . m. beneke and v.m . braun , nucl . phys . * b426 * , 301 ( 1994 ) ; m. beneke , phys . rept . * 317 * , 1 ( 1999 ) . n. brambilla , y. sumino , and a. vairo , phys . b*513 * , 381 ( 2001 ) ; y. sumino , phys . b*571 * , 173 ( 2003 ) . n. ishii , h. suganuma , and h. matsufuru , phys . rev . d*66 * , 094506 ( 2002 ) ; phys . rev . d*66 * , 014507 ( 2002 ) . bernard , phys . b*108 * , 431 ( 1982 ) ; nucl . phys . * b219 * , 341 ( 1983 ) . f. halzen , g. krein , and a.a . natale , phys . rev . d*47 * , 295 ( 1993 ) . cornwall and a. soni , phys . b*120 * , 431 ( 1983 ) . hou , c.s luo , and g.g . wong , phys . * , 014028 ( 2001 ) . g. parisi and r. petronzio , phys . b*94 * , 51 ( 1980 ) . a. yamamoto and h. suganuma , phys . * 101 * , 241601 ( 2008 ) ; phys . rev . d*79 * , 054504 ( 2009 ) . h. suganuma , t.t . takahashi , and h. ichie , _ color confinement and hadrons in quantum chromodynamics _ , ( world scientific , 2004 ) 249 ; t.t . takahashi , h. suganuma , y. nemoto , and h. matsufuru , phys . rev . d*65 * , 114509 ( 2002 ) ; t.t . takahashi , h.matsufuru , y.nemoto , and h. suganuma , phys . lett . * 86 * , 18 ( 2001 ) . h. iida , m. oka , and h. suganuma , eur . j. * a23 * , 305 ( 2005 ) . c. itzykson and j. zuber , _ quantum field theory _ ( mcgraw - hill inc . , 1980 ) . f. de soto and c. roiesnel , jhep * 09 * , 007 ( 2007 ) . savvidy , phys . b*71 * , 133 ( 1977 ) . nielsen and p. olesen , nucl . b144 * , 376 ( 1978 ) ; j. ambjorn and p. olesen , nucl . phys . * b170 * , 60 ( 1980 ) . g. parisi and n. sourlas , phys . lett . * 43 * , 744 ( 1979 ) . h. suganuma and t. tatsumi , ann . phys . * 208 * , 470 ( 1991 ) ; prog . theor . phys . * 90 * , 379 ( 1993 ) . verbaarschot and t. wettig , ann . nucl . part . sci . * 50 * , 343 ( 2000 ) , and its references . h. ichie and h. suganuma , nucl . phys . * b548 * , 365 ( 1999 ) ; nucl . phys . * b574 * , 70 ( 2000 ) .
cataclysmic variables ( cvs ) are a diverse group of short - period binaries in which a late - type , roche - lobe filling secondary star transfers matter through an accretion disk onto a white dwarf primary . the standard paradigm for the formation of cvs postulates that they start out as wide binaries ( on the order of 1 au ) , of moderate orbital period ( @xmath5 1 yr ) . as the more massive component , the white dwarf progenitor , evolves off of the main sequence , the secondary star suddenly finds itself orbiting within the red giant photosphere . during this `` common envelope '' ( ce ) phase , much of the angular momentum of the binary is believed to be shed due to interactions of the secondary star with the atmosphere of the red giant . during this process , the binary period gets shortened , and the red giant envelope gets ejected . depending on the input parameters used in standard ce models , between 50 and 90% of all of these close binary stars merge ( c.f . , politano & weiler 2007 ) . after emerging from the ce phase , an epoch of angular momentum loss via magnetic braking ( verbunt & zwaan 1981 ) , lasting @xmath6 10@xmath7 yr ( warner 1995 ) , is then required that leads to the formation of a semi - detached binary , and the mass transfer phase that signals the birth of a cv . in this scenario , the vast majority of cv secondary stars should not show any signs of evolution , as there just has not been sufficient time for the low mass secondary stars ( m@xmath8 @xmath9 1 m@xmath10 ) in these systems to have begun to evolve off of the main sequence . thus , the composition of the secondary stars should be relatively normal unless they have accreted matter with an unusual composition during the ce phase , or perhaps during classical novae eruptions . in an infrared spectroscopic survey of cvs , harrison et al . ( 2004 , 2005 ) found that the co features of the secondary stars in a number of cvs were much weaker than expected for their spectral types . for u gem , the co features were extremely weak , yet the water vapor feature was normal for a spectral type of m4v . this suggested a deficit of carbon . hamilton et al . ( 2011 ) have tabulated the list of cvs with weaker than expected co features , and show that , when such data exists , analysis of uv spectra _ also _ find evidence for carbon deficiencies . one issue with the infrared spectroscopy surveys was that they were qualitative , and were derived from lower resolution ( r @xmath0 2000 ) spectroscopy . the line features in those spectra _ might _ be compromised due to veiling by a continuum source that arises from the accretion disk . we have obtained moderate , and lower resolution , @xmath1-band spectroscopy of three of the brightest cvs : ss cyg , ru peg , and gk per . previous near - infrared spectroscopy with spex ( harrison et al . 2004 , 2007 ) revealed that the secondary stars in these three systems had weaker than expected co features . in the following , we use a modified version of the moog spectral synthesis program to model moderate resolution @xmath1-band spectra to examine the atomic and molecular abundances in their secondary stars . as shown in hamilton et al . ( 2011 ) , higher resolution spectra dramatically reduce the effects of accretion disk veiling , allowing us to see secondary stars even in cv systems where the accretion luminosity totally dominates that of the secondary star . in the systems discussed here , however , the secondary star dominates the accretion disk flux in the @xmath1-band , and any veiling has to be quite small . we then compare the best fit model spectra derived from the higher resolution data to lower resolution data . in the next section we discuss the data to be modeled , in section 3 we discuss the spectral synthesis procedure , in section 4 we present the modeling of the data , discuss the results in section 5 and draw our conclusions in section 6 . ss cyg , ru peg , and gk per were observed using nirspec in high resolution mode on keck ii on 2004 august 28 . the 0.432 x 12 slit was used providing a resolving power of r @xmath0 25000 , or a velocity resolution of @xmath11 @xmath0 12 km s@xmath12 . we employed the two - nod script , and we used five minute exposure times for the three program cvs . to completely cover the @xmath1-band using nirspec requires two grating settings . due to time limitations , however , we only obtained data at a single grating position chosen to insure coverage of the first overtone bandhead of @xmath13co@xmath14 at 2.294 @xmath15 m . this grating setting also covers the @xmath13co@xmath16 bandhead at 2.383 @xmath15 m . in addition to the three program cvs , we also observed two late - type stars , @xmath17 eri ( k2v ) and @xmath18 psc ( g9iii ) , to provide templates for our synthetic spectral program given that the properties of those two objects are well known . to correct for telluric absorption , we observed bright a0v stars located close to the program objects so as to minimize their relative differences in airmass . these data were reduced using the idl routine `` @xmath19 '' , specially developed for nirspec package can be found here : http://www2.keck.hawaii.edu/inst/nirspec/redspec/index.html ] . in the @xmath1-band , the spectra of a0v stars are nearly featureless , except for the prominent h i brackett @xmath18 feature at 2.16 @xmath15 m . the @xmath19 package does not attempt to correct for this feature , but can interpolate across such lines to reduce their impact upon division into the program star spectrum . note that there is a weak telluric feature located very close to the brackett @xmath18 line , and thus the h i line profiles in spectra produced by the division of a `` patched '' a - star spectrum are slightly compromised . just prior to our observing run on keck , we observed the three program cvs using spex at the irtf on 2004 august 15 . spex was used in cross - dispersed mode generating spectra spanning the range from 0.8 to 2.45 @xmath15 m . in the @xmath1-band , the dispersion is 5.4 / pix . the raw data were reduced using spextool ( cushing et al . 2004 ) . a log of all of our observations is presented in table 1 , and includes the orbital phase at the mid - point of the observational sequence for each of the three cvs . the aavso light curve data base shows that ss cyg was in quiescence at the time of both the spex and nirspec observations , having reached visual maximum on 17 july . ru peg was in outburst on 23 july , but was in quiescence for both observing runs . at the time of the infrared observations , gk per had @xmath20 = 13.1 , its normal minimum light value . to calculate a synthetic spectrum , three basic building blocks are required : a stellar atmosphere model , a list of atomic and molecular lines , and a tool to calculate the corresponding flux at the stellar surface as a function of wavelength . a full description of the modified version of moog and the grid generation and spectral matching python code ( called `` _ kmoog _ '' ) used below is described in hamilton ( 2013 ) , here we highlight the most important aspects . there are several groups that provide precomputed atmosphere models , each with their own slightly different assumptions , simplifications , and basic input physics . most of the models in general circulation , however , have the same underlying assumptions : the atmosphere is either modeled as a series of plane - parallel slabs ( 1d ) , or as a series of concentric shells ( 2d ) . the 1d and 2d atmosphere models are calculated assuming local thermodynamic equilibrium ( lte ) conditions , in which one temperature can describe the particle velocity distribution as well as the population of atomic levels and ionization states around a given point in the stellar atmosphere . it is also assumed that the atmosphere is time invariant , with the effects of dynamical processes such as mixing and convection parameterized . bonifacio et al . ( 2012 ) provide an overview of the more detailed assumptions and differences between the available atmosphere computation methods and programs . full 3d ( radiation - hydrodynamic ) stellar atmosphere models have been calculated across parameter grids ( e.g. , ludwig et al . 2009 , hauschildt and baron 2010 , magic et al . 2013a , 2013b ) . for this program , however , we have only used the 1d and 2d models . the most commonly used atmosphere model codes are phoenix ( hauschildt et al . ( 1999 ) , allard et al . 2011 ) , marcs ( gustafsson et al . 1975 , 2008 ) , and atlas ( sbordone et al . 2004 ) . phoenix is perhaps the most computationally sophisticated , sampling 42 million atomic and 550 million molecular lines when calculating opacity . marcs model atmospheres are consistently updated through a web interface providing both 1d and 2d models . ultimately for this project , we chose to use marcs atmospheres . an important consideration for this choice was the existence of the interpolation code to create subgrid models , making it much simpler to construct a large grid of models quickly . very recently , mszros et al . ( 2012 ) presented a new grid of atmospheres based off of atlas and marcs codes with an updated h@xmath21o line list as well as finer grid spacings in temperature , gravity , metallicity , and c and @xmath22 abundances . given their recent availability , however , all analyses and spectra computed here use solely the original marcs grid . the next crucial step to constructing a synthetic spectrum is to assemble a list of atomic and molecular transitions for which you wish to calculate an opacity at the stellar surface . for our project we used the atlas line list with the addition of the co line list by goorvitch ( 1994 ) . each line list was converted into a common format before combining , and , in the case of the co line list , only lines with @xmath23 and 2.28 @xmath24 2.405 @xmath15 m were kept . the final list contains a total of 11,681 lines . for the purposes of this work , only computations involving diatomic molecules were considered . this eliminates lines of the important molecular species for the cooler m dwarfs commonly found in cvs , such as h@xmath21o . as described below , synthetic spectra for the sun and arcturus were generated and compared to observations to provide the initial validation of this line list . there are a number of codes that can readily make use of marcs and kurucz atmospheres . moogchris / moog.html ] ( sneden 1973 ) has one of the longest track records , and given its ubiquity , we chose it to compute the synthetic spectra for this work . when creating a spectrum , moog is internally limited to considering only 2500 lines in the line list . given the size of our line list , we modified moog ( see hamilton 2013 ) to allow for lists of 250,000 lines . two additional changes were made to the moog source code . first , a version was created to produce absolutely no output to the screen . this prevents unnecessary lag and slowdowns , as well as greatly enhances usability when running a large number of models simultaneously . second , and most importantly , a new binary output file type of moog was created . significant savings in storage size were realized by creating a true fits output type , reducing the file size of a successfully created model by a factor of five or more . the fits files created also have the benefit of being fully compliant and compatible with the standard cfitsio libraries , unlike the fits - like output already in the code . the fits output created also has the benefit of storing both a full resolution spectrum as well as one convolved to lower resolutions , useful for quickly comparing against observations from different telescopes and instruments . to validate our line list , we constructed synthetic spectra for the sun ( using the solar marcs model ) and arcturus ( using the parameters listed in table 2 , but with log@xmath25 = 3.0 ) to compare them to high resolution @xmath1-band spectra published by wallace et al . ( 1996 ) , and wallace & hinkle ( 1996 ) . we concentrated on the reproduction of only the strongest spectral features observed in these two objects . note that the arcturus comparison was only used to insure that we were not missing lines that become important at lower temperatures . we then altered the oscillator strengths of the most highly discrepant lines to insure that they would both fit the observations better , and not bias the automated fitting process . for example , there were a large number of sc i transitions that had oscillator strengths that were much too large . unfortunately , oscillator strengths for many of the transitions in the @xmath1-band do not appear in the nist atomic spectra database , and thus we used trial and error adjustment of the tabulated oscillator strength in the line list to better reproduce the observed spectral feature ( for a primer on the identification of the strongest spectral features found in the @xmath1-band for k - type stars , see fig . 5 in harrison et al . 2004 ) . to test our ability to reproduce observations , we have constructed synthetic spectra to compare to the nirspec observations of @xmath17 eri and @xmath18 psc , and to a sample lower resolution spectra of k dwarfs contained in the irtf spectral library ( cushing et al . we summarize the spectroscopically derived parameters for these objects in table 2 , averaging the data compiled in the pastel catalog ( soubiran et al . 2010 ) for each object , constrained to measures made after 1970 . the final column of table 2 denotes how many unique measurements went into the determination of each of the average properties for each star ( note that hd45977 only has a single measurement of its characteristics by sousa et al . as we will see , this sample more than spans the range in metallicity observed for our program cvs . note that the solar abundance pattern derived by grevesse et al . ( 2007 ) was used for all of the synthetic spectra , and only the abundance of carbon was altered for the investigation of the cvs . in preparation for comparison to the models , the spectra were continuum normalized using a third order polynomial fit in iraf . for the high resolution spectra , the process was quite trivial . for the full @xmath1-band spectra , however , we used an identical set of wavelength bins for identifying the continuum to be fit . the continua were then fitted over the range 2.08 to 2.29 @xmath15 m , with bins that avoided all of the strongest spectral features . the nirspec spectrum of @xmath17 eri is presented in fig . [ eps ] , where data for five of the seven spectral orders are shown ( the first order , spanning @xmath262.43 to 2.47 @xmath15 m , was too far red to obtain appreciable signal in our cv spectra , while the seventh order spectrum is dominated by the strong telluric feature at 2.05 @xmath15 m ) . as shown in table 2 , @xmath17 eri has a temperature near 5100 k , and a near - solar metallicity of [ fe / h ] = @xmath20.1 . in fig . [ eps ] , we overplot a synthetic spectrum with t = 5100 k , and [ fe / h ] = @xmath20.125 ( interpolated between the model atmospheres with [ fe / h ] = @xmath20.25 , and [ fe / h ] = 0.0 ) . the model fit to the strongest lines is excellent . note that we have inserted an h i br@xmath18 line into our line list . we do this to primarily show where this line is located . the profiles of this feature as found in the infrared spectra for each of the objects are unreliable due to the reduction technique . in addition , the profile of such a strong line can not be easily reproduced using moog . @xmath18 psc allows us to test our ability to reproduce a much lower gravity object with a very low metallicity . we used a marcs spherical , alpha - enhanced model atmosphere with log@xmath25 = 2.5 , and [ fe / h ] = @xmath20.5 to generate a synthetic spectrum for @xmath18 psc . as noted in alves - brito et al . ( 2010 ) , @xmath18 psc has a modest alpha - enhancement of @xmath0 @xmath270.2 , and thus many of the metal lines are stronger in @xmath18 psc than they would be in a model that has a solar abundance pattern . since @xmath28 does not yet have the capabilities to adjust the abundances of any other species besides carbon , we ran the moog binary used by @xmath28 in a stand - alone mode , increasing the abundances of o , al and mg as detailed in alves - brito et al . we overplot this model on the observed data in fig . [ gamma ] . the resulting synthetic spectrum is a reasonable match to the data . given that the majority of cvs are too faint for moderate resolution infrared spectroscopy , most of the existing @xmath1-band data have been obtained at a lower resolution . to examine how well we can model such data sets we have downloaded spex data for dwarf stars spanning the spectral types k0v to k7v . the spectral type range k0v to k3v is shown in fig . [ k0tok3 ] , while the range k4v to k7v is shown in fig . [ k4tok7 ] . we generated synthetic spectra for each of these stars using the parameters listed in table 2 . the published metallicities produced excellent fits over most of the spectral range for all of the stars except hd145675 ( k0v ) . we found that the best fitting model for hd145675 over the entire @xmath1-band required a lower metallicity , [ fe / h ] = @xmath270.25 , than tabulated for this object ( though within the observed error bars ) . close inspection of fig . [ k0tok3 ] shows that the synthetic spectra do not do a very good job of matching the strongest spectral features in the 2.10 to 2.15 @xmath15 m bandpass . the origin of this discrepancy is not obvious . the fit of the synthetic spectra over the rest of the @xmath1-band suggests the tabulated ( or derived ) metallicities are correct . this might be a problem with continuum subtraction . to investigate this , we reduced and plot the spex data for the g9v star hd161198 observed during the same run as our program cvs [ note that hd161198 is a spectroscopic binary with a cool secondary star ( duquennoy et al . 1996 ) , and it is not a useful spectral template ] . the model for hd145675 does a good job at fitting the spectrum of hd161198 in this range , indicating that continuum subtraction is not the issue . this suggests that the problem could be due to either the data reduction process , or some issue with the telluric standard and/or object spectra . the maximum amount of observed signal occurs in this bandpass , and the k0v , k1v , and k3v have @xmath1 @xmath9 4.7 ( hd3765 has @xmath1 = 5.16 , while hd145675 has @xmath1 = 5.55 ) , so perhaps it is a linearity issue . the discrepancy between the models and observed spectra in the 2.10 to 2.15 @xmath15 m bandpass is not as severe for the k4v to k7v templates ( even though the k7v template is 61 cyg b , with @xmath1 = 2.32 ) . clearly , however , our synthetic spectra do an excellent job at matching objects with known temperatures and metallicities . especially important for the current program is the match of the model spectra to the observed co features over a wide range of temperature and metallicity . in the preceding we have modeled the spectra of objects with known parameters . for modeling the cvs , where the temperatures and metallicities are poorly known , requires the generation of a large number of models , and a robust method to match the synthetic spectra to the observations . with the python wrapper we have constructed to the modified version of the moog program described above , we are able to generate a large grid of models covering a user - defined range of the following parameters : t@xmath29 , log @xmath25 , [ fe / h ] , and [ c / fe ] . microturbulence was fixed at 2 km s@xmath30 , and macroturbulence was not included since it is a small effect at the resolution of our observational data . in _ , if the synthetic spectra are designated to be stored as fits files , each model actually contains two synthetic spectra : one unsmoothed , full resolution model , and one model convolved with a rotationally broadened spectrum appropriate for the target at the observed spectral resolution ( which must be specified before the grid creation ) . a full grid covering the entire parameter space ( 3800 @xmath9 t@xmath29 @xmath9 5200 k , 3.0 @xmath9 log @xmath25 @xmath9 5.0 , @xmath31 @xmath9 [ fe / h ] @xmath9 @xmath271 , and @xmath31 @xmath9 [ c / fe ] @xmath9 @xmath271 ) requires about one day to create on a 32 core machine and occupies @xmath0 86 gbyte in fits output format . obviously , smaller grids can be generated much more rapidly if the input parameter space can be narrowed . with such large grids , it is easy to be overwhelmed by the quantity of data . given a suitable subset of parameters , a direct grid search is possible ; gather a list of valid models , and step through them one at a time comparing the model to observations . given the sometimes low s / n of the observational data , however , a @xmath32 statistic is not calculated across the entire observational range , but restricted to the areas of greatest impact . for example , including na i , ca i , mg i , al i , fe i lines , and the co bands . this restricted region has the advantage of avoiding areas prone to increased levels of noise , and generally improves the quality of the fits to the observations . another method of stepping through the parameter space involves the use of an evolutionary algorithm to find the best fit to the data . evolutionary algorithms are based off the ideas of biological evolution , the simple premise that in some given population only the `` fittest individuals '' ( combination of all parameters ) will `` breed '' and influence the next generation of that population . if we imagine representing our parameters as 1d binary strings , then the behavior and the manipulation of the binary strings over the course of the fit is not unlike that of dna during reproduction . sets of parameters can be mutated , introducing diversity into the population which helps to avoid getting stuck in a local maximum or minimum in the `` @xmath33 landscape '' . mutations take the form of bit flips or swaps , analogous to transcription errors in dna replication . we used a genetic algorithm framework called @xmath34 . the algorithm proceeds as follows : 1 . make an initial first guess at the parameters based on spectral shape 2 . choose an appropriate maximum number of generations to evolve through as well as the initial population size 3 . `` breed '' and `` mutate '' the `` fittest '' members in the population to create a new generation 4 . repeat # 3 until the population converges to the desired level in this manner , all parameters are varied continually as the population converges towards some minimum @xmath32 , at least a factor of four faster than the plain grid searching technique . using both grid search and genetic algorithm techniques , the grid created with moog can be readily compared with the appropriate subset of cv spectra to attempt to determine as many stellar parameters as possible . both methods are available in _ kmoog _ , though we find the brute force is sufficiently rapid and as reliable as the genetic algorithm technique . heat maps showing the best fitting models for any two parameters are then generated to allow the determination of the best fitting model(s ) . as we will find below , determining the exact value t@xmath29 or [ fe / h ] for an unconstrained object from a @xmath1-band spectrum is difficult . many of the relevant oscillator strengths for transitions in this bandpass are unknown , and we have had to estimate a number of them by comparison of models to observations . in addition , our line list is incomplete , and thus weaker spectra features can contribute in a negative fashion to the assessment of the fit of a model to the data . the issues of telluric correction and continuum subtraction , as noted above , can introduce additional uncertainties into the goal of finding the best fitting solution . thus , the @xmath32 analysis just described must be done carefully , focussing on the strongest spectral features , insuring that there is not a spectroscopic anomaly in that feature which can be attributed to random noise variation , or to the telluric correction process . it is extremely difficult to produce a fully automated best fit regimen that can blindly solve for the parameters of a previously unconstrained object in the @xmath1-band , even when focused on specific spectral lines . currently , we manually examine the @xmath32 analysis for each line / region to insure that it is reliable , and that no unexpected events have occurred that affect the line / region under analysis . one such anomaly will be encountered shortly when we discuss the al i line ( at 2.116 @xmath15 m ) in ss cyg . it has an unusual strength that defies obvious explanation . @xmath35 analysis of this line results in a super - solar abundance . we continue to work on the matching aspects of _ kmoog _ , but in its current state , a more interactive ( `` hands - on '' ) approach is required to insure the most robust results . given that we have some information on the spectral types of the secondary stars in our cvs , we decided to use smaller grid than can be generated with the default settings of _ kmoog_. for example , tests in hamilton ( 2013 ) show that even the moderate resolution @xmath1-band spectra are inadequate for assessing gravity . given the large rotational broadening of our targets , it will be impossible to constrain the surface gravity for the program cvs . thus we fixed log @xmath25 = 4.5 for modeling ss cyg and ru peg , and log @xmath25 = 4.0 for gk per ( see below ) , to reduce parameter space , and increase the efficiency of the grid searching process . we postpone the discussion of error bar estimation to section 4.4 . ss cyg is the prototype long period ( p@xmath36 = 6.603 hr ) dwarf nova with eruptions that recur on a monthly basis . harrison et al . ( 2004 ) estimate the spectral type of the secondary as k4/k5 , consistent with other published values . they noted , however , that the co features were very weak for this spectral classification . bitner et al . ( 2007 ) have used moderate resolution visual band spectroscopy to examine the masses of both components , and found that the secondary star is 10 to 50% larger than a star of its mass and temperature should be . they suggest that the secondary star in ss cyg is evolved , and may have a core that is depleted in hydrogen . they also measured a rotational velocity of @xmath37sin@xmath38 = 89 km s@xmath12 for the secondary star . the nirspec spectrum of ss cyg is presented in fig . [ sscyg ] . the large rotational broadening of the spectral features is apparent . our observations spanned @xmath39 = 0.05 . at the observed phase ( using the k@xmath8 = 162.5 km s@xmath12 value derived by bitner et al . ) , the total change in radial velocity over our set of observations would be @xmath11 = 11 km s@xmath12 , resulting in minimal `` orbital smearing . '' concentrating on the sixth order bandpass ( 2.095 to 2.128 @xmath15 m ) , we convolved solar metallicity models having t = 4700 k with rotational velocities ranging from 70 to 110 km s@xmath12 . the lowest @xmath33 values were for @xmath37sin@xmath38 = 90 @xmath40 10 km s@xmath12 . for the modeling below , we adopt the bitner et al . value of @xmath37sin@xmath38 = 89 km s@xmath12 for ss cyg . to attempt to reproduce the nirspec data , we generated a grid of models spanning 4200 k @xmath9 t@xmath29 @xmath9 4900 k , @xmath41 @xmath9 [ fe / h ] @xmath9 @xmath270.50 , and @xmath42 @xmath9 [ c / fe ] @xmath9 @xmath270.25 . the resulting heat map for the sixth order , based on @xmath33 fits to _ only _ the first two of the three strong lines in the wavelength interval 2.104 and 2.118 @xmath15 m is shown in fig . [ sshm02 ] . the best fit occurs for t = 4700 k , and [ fe / h ] = @xmath20.25 . the fourth order contains a number of metal lines from ti i and fe i , though this spectral region is noisier than the data at shorter wavelengths . the resulting heat map , fig . [ sshm04 ] , is consistent with the result for the sixth order , though the value of [ fe / h ] is less constrained . the results for the third order containing the @xmath13co@xmath14 bandhead is shown in fig . [ sshm05 ] , and suggests [ c / fe ] @xmath9 @xmath20.5 . the best fit synthetic spectrum is overplotted on the data in fig . [ sscyg ] has these values , and fits the data set quite well . there is only one apparent anomaly in the nirspec spectrum of ss cyg , and that is the al i line at 2.116 @xmath15 m . we overplot the derived best fit model spectrum on the irtf spex observation in fig . [ ssirtf ] . the fit is quite good , given the s / n of those data . the main differences between the model and the low resolution spectrum is again the al i line 2.116 @xmath15 m , and an mg i feature at 2.282 @xmath15 m . the al i line was perfectly reproduced for the solar spectrum and for @xmath17 eri , but does appear to be slightly under - predicted for the irtf spectral sequence ( though note our earlier discussion of the anomalous continuum in this bandpass ) . the line at 2.109 @xmath15 m is also due to al i , thus the abundance of aluminum is not unusual . it is odd that this line is perfectly reproduced for our other cvs . there is an he i line at this position that is often seen weakly in emission in the @xmath1-band spectra of other cvs with more prominent accretion disks , such as tt ari , see fig . [ ttari ] . this emission line might be the cause of the peaks in the continuum on either side of this line in fig . [ ssirtf ] , but the effect of any such emission line would be to lessen the depth of the feature , not increase its strength . the mg i line at 2.282 @xmath15 m is well reproduced by _ kmoog _ for all of the template star spectra . this line is very weak in the spex spectrum of ss cyg . since the mg i line at 2.106 @xmath15 m is at the appropriate strength in both the moderate and lower resolution data sets , we conclude that this is simply a s / n or telluric correction issue . the nirspec and spex data were obtained at quite different orbital phases , 0.20 and 0.40 , respectively . the @xmath1-band light curve of ss cyg presented in harrison et al . ( 2007a ) shows no evidence for irradiation at @xmath43 = 0.5 , and the @xmath1 magnitudes at these two phases in this light curve are nearly identical . thus , the fact that the same synthetic spectrum works for both epochs is not surprising . this does prove , however , that irradiation of the secondary star is not the source that weakens the co absorption features , as the amount of the irradiated hemisphere seen at the dramatically different orbital phases of our spectra has no influence on the strength of the co features . ru peg is a u gem type dwarf nova of very long period ( p@xmath36 = 8.99 hr ) , whose binary parameters were first examined by kraft ( 1962 ) , observing doubled ca ii lines at some points in the orbital phase and a spectrum broadly consistent with a g8 iv to k0 v. friend et al . ( 1990 ) observed the @xmath268190 na i doublet to measure radial velocity motion of ru peg , finding that the secondary star is cooler and less massive than expected , with the spectrum being consistent with a k3 dwarf . they also note that the secondary star must be somewhat evolved , as the temperature , and masses derived are inconsistent with a main sequence dwarf . harrison et al . ( 2004 ) presented irtf observations of ru peg , showing that the nir spectrum was consistent with a spectral type of k2 , but that ru peg must contain a subgiant since it would be over - luminous for a dwarf at the distance given by its parallax . dunford et al . ( 2012 ) recently presented spectra of ru peg analyzed with roche tomograms , deriving a much cooler spectral type of k5 . along with the estimation of a cooler spectral type than found previously , dunford et al . used moderate resolution visual band spectra to estimate @xmath37sin@xmath38 = 89 km s@xmath12 . as for ss cyg , we convolved a solar abundance model with t@xmath29 = 4700 k to make our own estimate of the rotation velocity and found a slightly higher value of @xmath37sin@xmath38 = 95 @xmath40 10 km s@xmath12 . we have used this value in generating the synthetic spectra that follow . using the same size grid as that for ss cyg , the @xmath44 analysis for the sixth order , shown in fig . [ ruhm04 ] , suggests solar metallicities , and 4600 @xmath9 t@xmath29 @xmath9 4700 k. the results for the fourth order ( fig . [ ruhm02 ] ) indicate slightly subsolar metallicities , and 4500 @xmath9 t@xmath29 @xmath9 4900 k. we conclude that the secondary star has t@xmath29 = 4700 k , and [ fe / h ] = 0.0 . analysis of the co features in the third spectral order ( fig . [ ruhm05 ] ) finds [ c / fe ] = @xmath20.75 . a synthetic spectrum with these values is overplotted on the nirspec data in fig . [ rupeg ] , and provides an excellent fit to the data in all of the spectral orders . the h i emission from the disk is lower than seen in ss cyg , and the underlying h i br@xmath18 absorption from the secondary star leaves its imprint on the emission line profile . note that the long period and orbital phase at which the observations occurred make for minimal orbital smearing for the duration of our data taking : @xmath11 = 20 km s@xmath12 . the final model derived for the nirspec spectrum is overplotted on the spex data in fig . [ rupegirtf ] . this model fits the lower resolution data very well , including all of the other strong @xmath13co bandheads in the red part of the @xmath1-band . sion et al . ( 2004 ) found that the carbon abundance in the photosphere of the white dwarf was 10% of solar . thus , the carbon deficit observed in the uv can be directly traced to the secondary star . sion et al . also found that silicon suffered a similar deficit as that of carbon . the most prominent si i lines in the wavelength range plotted in fig . [ rupegirtf ] are at 2.136 and 2.189 @xmath15 m , and while these lines are weak in our data , and the s / n not very high , they seem to be well represented by our solar abundance model . gk per is an old classical nova , having experienced an extremely luminous nova eruption in 1901 ( see harrison et al . it has since exhibited periodic dwarf nova outbursts every 2 - 3 years since 1966 , brightening by @xmath0 3 mag during these events . kraft ( 1964 ) detected the mass donor star in the optical , adopting a spectral type of k2 iv and noting that the spectrum was highly variable with an equivalent width of sr ii 4077 consistent with a giant at one point and a dwarf the next . more recently , morales - reuda et al . ( 2012 ) analyzed spectra of gk per , finding that the donor star appeared as k1 iv . their analysis consisted of an iterative template subtraction method , however , not direct detection of the secondary star . harrison et al . ( 2007b ) presented a @xmath1-band spectrum of the mass donor star in gk per , and found that it has extremely weak co features . the radial velocity study conducted by morales - reuda et al . found @xmath37sin@xmath38 = 61.5 @xmath40 11.8 km s@xmath12 , k@xmath8 = 120.5 @xmath40 0.7 km s@xmath12 , and a mass ratio of @xmath45 = 0.55 @xmath40 0.21 . given the luminosity of its classical nova eruption suggests a massive white dwarf . if this is true , the radial velocity solution implies a surface gravity for the secondary star of log @xmath25 @xmath9 4 . as noted above , it is not possible to constrain the surface gravity of the program objects given the limitations of our data set . thus , in contrast to ss cyg and ru peg , we set the surface gravity to log @xmath25 = 4.0 for our modeling of gk per ( though this produced identical results to models with log @xmath25 = 4.5 ) . as for the preceding two objects , we analyzed the rotational velocity apparent in our nirspec data and find @xmath37sin@xmath38 = 55 @xmath40 10 km s@xmath12 , slightly lower than that derived by morales - rueda et al . , but agreeing within the error bars of the two measurements . we use this value in all of our modeling . the nirspec data is plotted in fig . [ gkper ] . @xmath35 analysis of the model grid found these best fit values : t@xmath29 = 5100 k , [ fe / h ] = @xmath20.125 , and [ c / fe ] = @xmath21.0 . gk per is fainter than our other two targets so the co region is very noisy , and thus the measurement of [ c / fe ] for gk per is slightly less certain than found for the other two targets . it is interesting to note that the h i absorption of the underlying secondary star is sufficient to create the appearance of no h i br@xmath18 emission at the center of this feature . the spex spectrum of gk per , fig . [ gkperirtf ] , is even noisier than the nirspec data . clearly , however , the same model does an excellent job at fitting the observations . as demonstrated above , we are easily able to reproduce the spectra of objects with known parameters . our examination of k stars in the irtf spectral library shows that if we know the value of [ fe / h ] , we can derive values of t@xmath29 , or vice versa , that are correct to within the published error bars for that object . to examine how well we can determine t@xmath29 or [ fe / h ] if we do not know either value , we present a @xmath35 analysis for the irtf template hd36003 . the heat map for the spectral region 2.18 @xmath9 @xmath26 @xmath9 2.37 @xmath15 m of this star is shown in fig . [ k4vhm ] . while the models with parameters similar to the published values for both t@xmath29 ( 4615 @xmath40 29 k ) and [ fe / h ] ( @xmath20.14 @xmath40 0.08 ) do have low values of @xmath35 , there is a range of solutions with quantitatively similar fits to the data . as demonstrated in the plots shown in figs . [ k0tok3 ] and [ k4tok7 ] , for k dwarfs , the strong na i doublet ( at 2.209 @xmath15 m ) , the ca i triplet ( at 2.26 @xmath15 m ) , and the co features all slowly get stronger as we decrease temperature . thus , models can mimic the observed spectrum by either an enhanced abundance of [ fe / h ] and a hotter temperature , or a reduced value of [ fe / h ] and a lower temperature . our @xmath35 analysis over such a large span in wavelength is certainly also affected by missing transitions for weak lines , the s / n of the data , and for this lower resolution spectrum , the continuum subtraction process . the red end of the @xmath1-band is filled with strong telluric features , and even for high s / n data , small discrepancies can appear due to a slight mis - match of the telluric standard with the program object . this is especially clear for the co features , as some bandheads in a particular spectrum are well fitted by the models , while others are not . given these results , if we are able to constrain the limits on the temperature of the object , such as `` the secondary star in ss cyg has a spectral type near k5 '' , and not expect super - solar metallicities , we find that temperatures derived from these data are good to @xmath40 250 k , and [ fe / h ] is good to @xmath40 0.25 . moderate resolution data that covered either the na i doublet or ca i triplet would almost certainly lead to more precise values of both quantities . putting error bars on our values for [ c / fe ] is more difficult in that none of our templates are known to have unusual values for this parameter . it is clear that our models reproduce the co features in the template spectra over a large range in metallicity . thus , altering just the abundance of carbon should not introduce any significant issues , as it simply acts to reduce the strength of the co features to mimic those seen in lower metallicity stars . how a reduced carbon abundance affects the input stellar atmosphere is difficult to assess given the unknown evolutionary state of the cv secondary stars . the presence of more than one co bandhead in all of the spectra also allows us to have more confidence in a particular value of [ c / fe ] . for the higher s / n spectra , we expect that the precision of the values for [ c / fe ] are similar in scale to [ fe / h ] : @xmath40 0.25 . for gk per the data are quite poor compared to the other two cvs , and the derived value for [ c / fe ] is less certain . it is obvious , however , that the value of [ c / fe ] is lower in gk per than in the other two sources . this result will be amplified if the gravity of gk per is significantly lower than our input value of log @xmath25 = 4.0 . we have obtained moderate resolution @xmath1-band spectroscopy of three cvs that were previously noted to have weaker than expected co features : ss cyg , ru peg , and gk per . because those earlier data were obtained at lower resolution , r @xmath0 2,000 , questions were raised about the possibility of `` veiling '' of the co features to make them appear weaker . while we show below that this argument has no validity if realistic veiling scenarios are examined , it is the increased dispersion of the spectra obtained with nirspec that easily argues against such a scenario . the nirspec data have a dispersion of 0.32 / pix , nearly 1/17 that of the spex data . thus , the flux per pixel from any veiling component would be diminished by this factor . model spectra fitted to the two different sets of data would give different values for the strength of the co features if there was some type of veiling confined to the co region . to provide quantitative estimates of the relative carbon abundances for cvs we have altered moog to allow us to model the entire @xmath1-band , and developed a software package that constructs large grids of synthetic spectra , and then searches these grids to find the best fitting solution to the observed data . we specifically allowed for the abundance of carbon to be a separate input under the assumption that the @xmath13c abundance has been changed in the initial branches of the cno cycle , as argued in harrison et al . obviously , weak co features could be explained by a deficit of oxygen , but a process that could cause such a deficit is not obvious . with this software we derived values of t@xmath29 , [ fe / h ] , and [ c / fe ] from the nirspec spectra for the three program objects . we list these results in table 3 . except for the deficits of carbon , no other abundance anomalies are apparent , and the secondary stars of all three objects have solar , or slightly sub - solar abundances . the total @xmath13c abundance for ss cyg and ru peg are similar , @xmath0 25% of solar , while for gk per it is @xmath0 10% of solar . if these peculiar values are the result of the cno cycle ( c.f . , marks & sarna 1998 ) , than the abundance of @xmath46c should be enhanced , allowing for the detection of the bandheads of @xmath46co . the two strongest such bandheads with @xmath47 2.4 @xmath15 m are @xmath46co@xmath14 at 2.345 @xmath15 m , and @xmath46co@xmath48 at 2.374 @xmath15 m . the first of these is not covered by our nirspec data , but dhillon et al . ( 2002 ) present a @xmath1-band spectrum of ss cyg which appears to show a strong @xmath46co@xmath14 absorption feature . they attribute this feature to metal line absorption in star spots as indicated by anomalously strong na i absorption features ( at 2.3355 and 2.3386 @xmath15 m ) in their spectra . these two na i lines are not anomalous in our spex data , nor is there a @xmath46co@xmath14 feature of the strength observed by dhillon et al . the second of the @xmath46co features is covered by our nirspec observations , but there is a deep telluric feature at this wavelength that makes good correction quite difficult , and results in a noisy portion of the second order spectrum . even with this issue , there is certainly no sign of a strong absorption feature at this location in either ss cyg or ru peg . the spex data covers both features , and while there are hints of both of these bandheads in the data for ss cyg and ru peg , the spectra are simply too noisy to draw any firm conclusions . new , higher s / n moderate resolution spectroscopy will be required to properly investigate the @xmath13c/@xmath46c ratio . harrison et al . ( 2004 ) found that the @xmath1-band continua of many long period cvs were flatter than expected if the secondary star was the sole source of emission in that bandpass . to more closely examine the slope of the continuum of ss cyg , we compare the full @xmath1-band spectrum of ss cyg with that of the irtf template star that has the most similar temperature : hd45977 . in fig . [ sscygcomp ] , we subtract the spectrum of hd45977 from that of ss cyg , both spectra having been normalized at 2.2 @xmath15 m . the result is a residual spectrum that is fairly flat , with f@xmath49 @xmath50 @xmath51 . given the uncertainties in the temperatures of the two objects , and the differing sources of the data , such a flat spectrum is consistent with free - free emission . the continuum subtraction process step required to allow us to directly compare the spex data to the synthetic spectra easily removes a weak continuum source such as this , and this is why the results for the nirspec and spex are simultaneously consistent with the same model . we can also attempt to model the contribution of line emission sources that might preferentially veil the co features . there are only two possible sources that we can envision that might do this : the h i pfund continuum , and co emission . as shown in fig . [ sscygcomp ] , the h i br@xmath18 emission line in ss cyg is quite strong , so the pfund series of hydrogen will also be present . the pfund series limit is at @xmath262.279 @xmath15 m , and the transitions pf23 and higher , will fall in our modeling bandpass . the emissivities of these lines are very small ; hummer & storey ( 1987 ) list them as being @xmath9 1.7% that of br@xmath18 ( for `` case b '' conditions ) . fitting the br@xmath18 line with a gaussian profile gives fwhm = 73 . we will assume all of the h i lines have this profile . in fig . [ sscygpfcomp ] , we present spectra where we added a rotationally broadened spectrum of the pfund series ( up to pf100 ) to the k4v spectral template , and compare it to the spectrum of ss cyg . given their low emissivities , the effect of the h i lines is not detectable . we also artificially multiplied the emissivities of the pfund lines by factors 10 and 100 and added them to the k4v template . only at very large multiplicative factors can their presence be detected , and even then , they add a smooth continuum to the co region , leaving the absorption features relatively undisturbed , but producing a highly distorted continuum . even with this simplistic model , it is clear that h i pfund emission can not provide the veiling necessary to explain weak co features . the detection of co emission in wz sge ( howell et al . 2004 ) has led to the suggestion that co emission is the source of the veiling needed to produce weaker than expected co features seen in the program cvs . since the co emission must come from the accretion disk , a similar result as that for the pfund continuum emission attains . as discussed in scoville et al . ( 1980 ) , such emission is expected to be confined to regions where t @xmath52 3000 k , and where the densities are high , @xmath53 @xmath54 10@xmath55 @xmath56 . since ss cyg was in a minimum during our observations , presumably these conditions exist somewhere within its accretion disk . as shown in bik et al . ( 2006 ) , optically thin co emission features are essentially the inverse of the co absorption features seen in late type stars . this similarity allows us to construct crude models to investigate what co emission might look like if its origin was in a cv accretion disk . if we assign the co emission to the disk , then the broadening of any spectral feature located there will be much greater than that of the slower rotational broadening of the secondary star ( 89 km s@xmath12 in the case of ss cyg ) . for example , if the accretion disk extends to 75% of r@xmath57 ( the roche lobe radius of the white dwarf ) , using the stellar parameters for ss cyg in bitner et al . ( 2007 ) , the keplerian velocity at this radius will be @xmath0 400 km s@xmath12 . in fig . [ sscygcocomp ] , we add a rotationally broadened co spectrum to the best fitting synthetic spectrum for ss cyg using the value for the h i br@xmath18 line ( fwhm = 73 ) , and a co emission spectrum broadened by half that value . as could be expected , the result is a dramatic change in the profiles of the co features , with an emission peak preceding each bandhead , a subtle redward shift for the location of the strongest absorption , and a dramatic narrowing of the co bandhead profiles . obviously , the data for none of our objects resembles this result . more complicated scenarios for the emission profile could be constructed , but to truly diminish the co features without distorting the resulting spectrum _ requires _ the addition of co emission with a broadening profile that is very similar to that of the secondary star . there is no other location in a cv system besides the secondary star photosphere where such low velocities occur . we have modified the spectral synthesis program moog and have demonstrated that we are able to model moderate and lower resolution @xmath1-band spectra for stars hotter than t@xmath29 @xmath54 4000 k. we then used this program to investigate such spectra for three cvs : ss cyg , ru peg , and gk per . it is clear that all three cvs in this sample have subsolar carbon abundances . that synthetic spectra with identical parameters simultaneously model data with dramatically different dispersions for each of the objects , and realistic co veiling scenarios fail to reproduce the observed co features , proves that the carbon deficits are real . it is important to note that these results represent the first direct measurements of abundances in the photosphere of the mass donor stars in cvs . while our determinations of t@xmath58 , [ fe / h ] , and [ c / fe ] for these three objects are not overly precise , they have no parallel . there have not been many predictions for the behavior of the abundances in the atmospheres of cv donor stars as a function of age / evolutionary state . marks & sarna ( 1998 ) explored how the photospheric abundances of cno elements in cv secondaries might evolve with time , including the possibilities of pollution by the common envelope phase , or through the accretion of classical novae ejecta . they found that the latter two scenarios were inadequate in producing observable changes , and concluded that only pre - contact evolution could significantly alter the cno abundances and isotopic ratios in cv secondary star photospheres . schenker et al . ( 2002 ) examined how cvs might evolve from supersoft binaries and , if so , could show dramatic changes in cno species . in light of our results , it would be useful to revisit this research with newer generations of binary star evolution codes , such as starevol ( stancliffe & eldridge 2009 ) . it is also obvious that moderate resolution spectroscopy of the brightest cvs covering the _ entire _ @xmath1-band are required to achieve more robust measures of [ fe / h ] , as well as allow for the proper examination of the co bands . if the deficits of carbon arise from the cno cycle , than the isotopic abundances of carbon will be altered . moderate resolution @xmath1-band spectroscopy covering the main @xmath46co features will allow for the confirmation of this inference . both teh and rth were partially supported by a grant from the nsf ( ast-1209451 ) . we acknowledge with thanks the variable star observations from the aavso international database contributed by observers worldwide and used in this research . allard , f. , homeier , d. , & freytag , b. 2011 , aspc , 448 , 91 + alves - brito , a. , melendez , j. , asplund , m. , ramirez , i. , et al . 2010 , a&a , 513 , 35 + baron , e. , chen , b. , & hauschildt , p. h. 2009 , in aip conf . , vol . 1171 , american institute of physics conference series , ed . i. hubeny , j. m. stone , k. macgregor , & k. werner , 148160 + bitner , m. a. , robinson , e. l. , & behr , b. b. 2007 , apj , 662 , 564 + bonifacio , p. , caffau , e. , ludwig , h .- , & steffen , m. 2012 , iaus 282 , 213 + cushing , m. c. , rayner , j. t. , & vacca , w. d. 2005 , apj , 623 , 1115 + cushing , m. , vacca , w. d. , & rayner , j. t. 2004 , pasp , 116 , 362 + dhillon , v. s. , littlefair , s. p. , marsh , t. r. , sarna , m. j. , & boakes , e. h. 2002 , a&a , 393 , 611 + dunford , a. , watson , c. a. , & smith , r. c. 2012 , mnras , 422 , 3444 + duquennoy , a. , tokovinin , a. a. , leinert , ch . , glindemann , et al . 1996 , a&a , 314 , 846 + friend , m. t. , martin , j. s. , connon - smith , r. , & jones , d. h. p. 1990 , mnras , 246 , 654 + goorvitch , d. 1994 , apjs , 95 , 535 + gray , d. f. , the observation and analysis of stellar photospheres ( cambridge : cambridge univ . press ) , 374 + gray , r. o. , & corbally , c. j. 1994 , aj , 107 , 742 + grevesse , n. , asplund , m. , & sauval , a. j. 2007 , space sci . , 130 , 105 + gustafsson , b. , bell , r. a. , eriksson , k. , & nordlund , a. 1975 , a&a , 42 , 407 + gustafsson , b. , edvardsson , b. , eriksson , k. , j@xmath59rgensen , u. g. , nordlund , a. , & plez , b. 2008 , a&a , 486 , 951 + hamilton , r. t. 2013 , phd thesis , new mexico state university + hamilton , r. t. , harrison , t. e. tappert , c. , & howell , s. b. 2011 , apj , 728 , 16 + harrison , t. e. , bornak , j. , mcarthur , b. e. , & benedict , g. f. 2013 , apj , 767 , 7 + harrison , t. e. , howell , s. b. , szkody , p. , & cordova , f. a. 2007a , aj , 133 , 162 + harrison , t. e. , campbell , r. k. , howell , s. b. , cordova , f. a. , & schwope , a. d. 2007b , apj , 656 , 444 + harrison , t. e. , osborne , h. l. , & howell , s. b. 2005 , aj , 129 , 2400 + harrison , t. e. , osborne , h. l. , & howell , s. b. 2004 , aj , 127 , 3493 + hauschildt , p. h. , & baron , e. 2010 , a&a , 509 , 36 + hauschildt , p. h. , allard , f. , ferguson , j. , baron , e. , & alexander , d. r. 1999 , apj , 525 , 871 + howell , s. b. , harrison , t. e. , & szkody , p. 2004 , apj , 602 , l49 + hummer , d. g. , & storey , p. j. 1987 , mnras , 224 , 801 + kraft , r. r. 1964 , apj , 139 , 457 + kraft , r. p. 1962 , apj , 135 , 408 + ludwig , h. -g . , caffau , e. , steffen , m. , freytag , b. , et al . 2009 , mmsai , 80 , 711 + magic , z. , collett , r. , hayek , w. , & asplund , m. 2013 , a&a , 560 , 8 + magic , z. , collett , r. , asplund , m. , trampedach , r. , et al . 2013 , a&a , 557 , 26 + marks , p. b. , & sarna , m. j. 1998 , mnras , 301 , 699 + martnez - arniz , r. , maldonado , j. , montes , d. , eiroa , c. , & montesinos , b. 2010 , a&a , 520 , 79 + mszros , sz . , allende prieto , c. , edvardsson , b. , castelli , f. , et al . 2012 , aj , 144 , 120 + morales - reuda , l. , still , m. d. , roche , p. , wood , j. h. , & lockley , j. j. 2002 , mnras , 329 , 597 + politano , m. & weiler , k. p. 2007 , apj , 665 , 663 + reddy , r. r. , & viswanath , r. 1990 , journal of astrophysics and astronomy , 11 , 67 + sbordone , l. , bonifacio , p. , castelli , f. , & kurucz , r. l. 2004 , memorie della societa astronomica italiana supplementi , 5 , 93 + scoville , n. z. , krotkov , r. , & wang , d. 1980 , apj , 240 , 929 + schenker , k. , king , a. r. , kolb , u. , wynn , g. a. , & zhang , z. 2002 , mnras , 337 , 1105 + sion , e. m. , cheng , f. , godon , p. , urban , j. a. , & szkody , p. 2004a , aj , 128 , 1834 + sneden , c. a. 1973 , phd thesis , university of texas + soubiran , c. , le campion , j. -f . , cayrel de strobel , g. , & caillo , a. 2010 , a&a , 515 , 111 + sousa , s. g. , santos , n. c. , israelian , g. , mayor , m. , et al . 2011 , a&a , 533 , 141 + stancliffe , r. j. , & eldridge , j. j. 2009 , mnras , 396 , 1699 + verbunt , f. & zwaan , c. 1981 , a&a , 100 , 7 + wallace , k. , & hinkle , k. 1996 , apjs , 107 , 312 + wallace , k. , livingston , w. , hinkle , k. , & bernath , p. 1996 , apjs , 106 , 165 + warner , b. 1995 , cataclysmic variable stars ( cambridge : cambridge univ . press , 443 + lccccc ss cyg & nirspec & 2004 aug . 28 & 12:31:51 & 4 @xmath60 300 & 0.20 + ru peg & nirspec & `` & 13:17:12 & 4 @xmath60 300 & 0.87 + gk per & nirspec & '' & 14:20:24 & 6 @xmath60 300 & 0.46 + @xmath18 psc & nirspec & `` & 13:45:17 & 8 @xmath60 1 & + @xmath17 eri & nirspec & '' & 15:06:18 & 24 @xmath60 1 & + ss cyg & spex & 2004 aug . 15 & 08:55:17 & 8 @xmath60 180 & 0.40 + ru peg & spex & `` & 10:25:06 & 8 @xmath60 240 & 0.85 + gk per & spex & '' & 13:53:54 & 6 @xmath60 240 & 0.94 + lccccc sun & g2v & 5777 & 4.44 & 0.0 & + arcturus & k0iii & 4313 @xmath40 84 & 1.65 @xmath40 0.30 & @xmath20.54 @xmath40 0.10 & 34/31/31 + @xmath17 eri & k2v & 5088 @xmath40 72 & 4.54 @xmath40 0.16 & @xmath20.09 @xmath40 0.09 & 26/24/24 + @xmath18 psc & g9iii & 4853 @xmath40 45 & 2.59 @xmath40 0.26 & @xmath20.46 @xmath40 0.12 & 11/8/8 + hd145675 & k0v & 5320 @xmath40 114 & 4.45 @xmath40 0.07 & @xmath270.41 @xmath40 0.12 & 18/15/15 + hd10476 & k1v & 5189 @xmath40 53 & 4.46 @xmath40 0.12 & @xmath20.06 @xmath40 0.06 & 16/10/10 + hd3765 & k2v & 5023 @xmath40 66 & 4.53 @xmath40 0.16 & @xmath270.05 @xmath40 0.10 & 7/4/4 + hd219134 & k3v & 4837 @xmath40 138 & 4.51 @xmath40 0.13 & @xmath270.05 @xmath40 0.10 & 15/10/10 + hd45977 & k4v & 4689 @xmath40 174 & 4.30 @xmath40 0.39 & @xmath270.03 @xmath40 0.18 & 1/1/1 + hd36003 & k5v & 4615 @xmath40 29 & 4.35 @xmath40 0.06 & @xmath20.14 @xmath40 0.08 & 3/2/2 + hd201092 & k7v & 3964 @xmath40 175 & 4.50 @xmath40 0.18 & @xmath20.26 @xmath40 0.27 & 10/5/5 + lcccc ss cyg & 4700 @xmath40 250 & 4.50 & @xmath20.25 @xmath61 0.25 & @xmath20.50 @xmath40 0.25 + ru peg & 4700 @xmath40 250 & 4.50 & @xmath270.00 @xmath40 0.25 & @xmath20.75 @xmath61 0.25 + gk per & 5100 @xmath40 250 & 4.00 & @xmath20.125 @xmath40 0.25 & @xmath21.00 @xmath40 0.50 + eri ( black solid line ) . the synthetic model spectrum ( green , dashed line ) generated for this object is overplotted . for the purposes of this paper , the bottom panel will be defined as the second order , while the top panel will be labeled the sixth order . the synthetic spectrum has been rotationally broaded by 4 km s@xmath12 as measured by martinez - arnaiz et al . the s / n ratio ranged from 214 in the sixth order , to 95 in the second order.,title="fig:",width=566 ] m are not especially well matched by the models . the source of this discrepancy is hard to identify . in the top panel we also plot in red the spectrum of a g9v star ( hd161198 ) that we observed with spex during the 2004 august observing run , and the model spectrum generated for the k0v template fits the spectrum of hd161198 quite well.,title="fig:",width=566 ] heat map for the order 6 spectral region for the nirspec spectrum of ss cyg . in this , and all subsequent heat maps , blue denotes @xmath62 @xmath63 0.7 , green is 0.7 @xmath9 @xmath62 @xmath63 0.9 , yellow is 0.9 @xmath9 @xmath62 @xmath63 1.5 , magenta is 1.5 @xmath9 @xmath62 @xmath63 2.5 , and red is @xmath62 @xmath54 2.5 . , title="fig:",width=566 ] heat map for the order 3 spectral region for the nirspec spectrum of ss cyg . for this analysis , a grid of models with the metallicity fixed to [ fe / h ] = @xmath20.25 was constructed , while the carbon abundance was altered over the range @xmath21.0 @xmath9 [ c / fe ] @xmath9 @xmath270.25 . , title="fig:",width=566 ] heat map for the model grid fits to the third order spectrum of ru peg . like that for ss cyg , a grid of single metallicity models ( here with [ fe / h ] = 0.0 ) were generated with only the carbon abundance altered for this analysis.,title="fig:",width=566 ] analysis of the irtf spectral template hd36003 for a large grid of model spectra . the mean of the published values for this star , see table 2 , are t@xmath29 = 4550 k , and [ fe / h ] = @xmath20.17 . the grid has temperature intervals of 100 k , and [ fe / h ] stepped by @xmath400.125.,title="fig:",width=491 ] line of ss cyg . the spectrum below that has had the emissivities of the pfund lines increased by a factor of 10 . the bottom spectrum has the pfund line emissivity increased by a factor of 100.,title="fig:",width=491 ] line , added to the best fitting ( solar abundance ) synthetic spectrum . the bottom spectrum is the result of the identical process , except the velocity broadening was one half that of br @xmath18 profile . the co emission component was normalized so that when it was added to the model spectrum , the depth of the minima in the co features of ss cyg were reproduced . dashed vertical lines indicate the wavelengths of the @xmath13co@xmath65 ( left ) and @xmath13co@xmath66 ( right ) bandheads.,title="fig:",width=642 ]
the interaction between high - energy jets from active galactic nuclei ( agn ) and their surroundings has long been a topic of great astrophysical interest . it is well known that agn feedback can control the size of a galaxy by influencing star formation , but the mechanism behind this is not well understood . several recent observations @xcite , as well as numerical studies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , have demonstrated that agn feedback can be `` negative '' or `` positive . '' the exact astrophysical conditions in the jet and cloud are important in determining what direction feedback takes . jets can be roughly divided according to their fanaroff - riley classification @xcite . fast , energetic frii jets seem more likely to result in negative feedback . negative feedback curbs or even halts star formation , and is thought to result from the extreme radiative and kinetic energies of the jet , which heat and disperse the star - forming gas . additionally , the kinetic energy of a jet creates turbulence that can prevent ambient gas from cooling and subsequently coalescing ( e.g. * ? ? ? for example , a study of the system 3c 326 by @xcite found that , despite the strong h@xmath1 line emission and an inferred molecular gas mass of 2 @xmath2 m@xmath3 , the star formation is 20 times lower than predicted by the kennicutt - schmidt law . they infer that turbulent heating from the jet is inhibiting star formation . other studies , though , suggest increased star formation may be seen in the cocoon of such jets ( e.g. * ? ? ? in contrast , fri jets propagate through the ism / igm with energies high enough to create compression in the surrounding gas , but low enough to reduce the chance of significant turbulent heating . these jets are observed in positive feedback cases , wherein the effect of the jet serves to enhance star formation , including centaurus a ( * ? ? ? * and references therein ) , 4c 41.17 @xcite , and minkowski s object ( hereafter m.o . ) * hereafter c06 ) . we give more details on both positive and negative feedback in section [ sec : jetcloud ] . in this paper , we focus on the case of m.o . , a peculiar star forming object located at a redshift of @xmath4 ( c06 ) that is currently being bombarded by a fr i radio jet from the nearby galaxy ngc 541 . system is of particular interest due to the lack of evidence for an especially dense ism or igm . there is also not much evidence for cold gas outside of the jet interaction site as is the case in centaurus a. as a result , it is unlikely that significant star formation would proceed in m.o . without the interaction of the jet . a strong argument in favor of jet - induced star formation in m.o . is the morphology of the jet - cloud interaction site . outside of the jet interaction , the gas in m.o . is warm ( @xmath5 k ) and clumpy . near the jet interaction site there is a double structure of gas wrapped around the jet and numerous h@xmath1 regions ( c06 ) . c06 thus determined that it is likely that the jet interaction in m.o . caused the warm gas to cool into the h@xmath0 regions straddling the jet , in contrast to the pre - existing cold gas regions in centaurus a. also , the star forming regions in m.o . correlate with the jet - cloud morphology ; the region where the star formation is the highest is the center of the jet - cloud interaction , and the star formation rate ( sfr ) decreases laterally from this point ( c06 ) . unless m.o . happens to be a very unique system , then it may be that jet - induced star formation is a more general phenomenon than previously thought . the present work can be viewed as an extension of our earlier study of the interactions of radiative shocks with clouds @xcite . that work focused mostly on the effects of planar shocks overtaking individual ( or small collections of ) warm clumps on the scale of @xmath6 pc . in the current work , we explore the much richer problem of a full jet intersecting an inhomogeneous intergalactic cloud on the scale of tens of kpc . the paper proceeds as follows : section 2 covers the theory behind jet - cloud interactions , section 3 describes the numerical models used to capture the m.o . system , section 4 details the simulation results , and section 5 concludes the findings . the basic idea of jet - induced star formation ( i.e. positive feedback ) is that the collision of the jet with the cloud will trigger a series of shocks within the cloud . the immediate effect of these shocks will be to compress and heat the gas . depending on how the radiative processes scale with density and temperature , the net result can be to dramatically increase the radiative efficiency within the cloud . if the temperature dependence is shallower than the density dependence , the cloud can enter a phase of runaway cooling . this process occurs most quickly in relatively over - dense regions of the original cloud . these over - dense regions then proceed to collapse at an accelerating pace . provided some of these clumps start sufficiently close to the jeans limit , this collapse will push them beyond this limit , such that gravitational collapse can take over and the clump will proceed to form stars . the properties of the jet and cloud are key to controlling this process . for positive feedback to be important , the initial cloud must be dense enough for some parts to be reasonably close to the jeans limit . the temperature must also be such that any increase in temperature is met with a dramatic increase in cooling ( the hydrogen cooling edge at @xmath5 k is a good example ) . it generally helps for the jet to be significantly less dense than the cloud . finally , the jet velocity needs to be fast enough to trigger shocks in the cloud , yet not so fast that the cloud is disrupted before cooling can have much of an effect . our numerical simulations are performed using the well - tested _ cosmos++ _ computational astrophysics code @xcite , specifically its newtonian hydrodynamics solvers . _ cosmos++ _ carries over many of the multi - physics capabilities found in its predecessor code _ cosmos _ @xcite . the newtonian solvers have previously been utilized to study the bar mode instability in magnetized , rotating neutron stars @xcite and the galactic center g2 event @xcite . the current work uses the high resolution shock capturing ( hrsc ) scheme , which was described in its relativistic form in @xcite . as there are few differences between the relativistic and newtonian forms , we do not give a full presentation here , focusing instead on the packages that are most important to this paper : chemistry , cooling , and star formation . note that , although magnetic fields can play an important role in shock - induced star formation ( cf . * ) , they are not considered in this work . most prior numerical studies of stimulated star formation from jet - generated shocks have been hampered by the resolution of the computational mesh ( e.g. * ? ? ? * ) . in the present work , we significantly improve on previous resolution limitations by employing the adaptive mesh refinement ( amr ) capabilities of _ cosmos++_. _ cosmos++ _ employs a _ amr scheme , in which refinement and de - refinement decisions are made on a cell - by - cell basis , using an oct - tree network to traverse the grid hierarchy . each level of refinement doubles the spatial resolution in each dimension within a given parent cell . this style of local amr scheme ensures that the refinement and de - refinement conform as closely as possible to the shape of the region of interest , in this case the shocks and unstable cooling fronts triggered inside the cloud by the jet . in this work , we present an idealized case of a direct , axially symmetric collision between a jet and a pre - existing spherical cloud . as such , it is a simple , and well controlled , test simulation , though its correspondence with m.o . is imperfect . in the case of m.o . , it is thought that the collision was between a jet and a stellar bridge connecting the elliptical galaxy , ngc 541 , with the interacting galaxies , ngc 545/547 ( c06 ) . there is also evidence that the jet is slowly sweeping across m.o . ( c06 ) . despite these differences , the correspondence in parameters between our simulation and m.o . ensure that our results are applicable and the simulation can be used to better understand the dynamics of this particular object and of jet - induced star formation more generally . although we performed some two - dimensional simulations to test different code options and explore our parameter space , we mostly focus on reporting the results of our 3d simulations . the 3d simulations have a base resolution of either @xmath7 or @xmath8 zones to cover a domain that is approximately @xmath9 with reflection boundaries applied in the @xmath10- and @xmath11-directions , so that we only simulate one quadrant of the full problem . the finest spatial resolution achieved is 19.5 pc per zone , reached by including 2 levels of refinement on top of the @xmath8 base mesh , equivalent to a uniform mesh of @xmath12 zones . the criterion used for refinement is that any zone with @xmath13 @xmath14 is kept at the maximum refinement , while zones that fall below @xmath15 @xmath14 are allowed to de - refine , provided neighboring zones never differ by more than one level of refinement and no zone is allowed to drop below the base resolution . zones are checked against the refinement and de - refinement criteria once every ten evolution steps in the numerical code . we find that a minimum resolution close to our base value is required even in the background in order to get reasonable convergence in the star formation rate ( as will be discussed in section [ sec : sfr ] ) . the cloud , which represents the parent object of m.o . , is initialized with a radius of @xmath16 pc . it is , therefore , somewhat smaller , in terms of projected area , than the real value of 275 kpc@xmath17 ( c06 ) . the gas within the cloud is distributed using a random , log - normal density , @xmath18 @xmath14 , with mean and median values of @xmath19 @xmath14 and @xmath20 @xmath14 , respectively , where @xmath21 is a randomly drawn variable with a mean of 0 and variance of 1 . the gas has a mean molecular weight of @xmath22 , appropriate for approximately solar metallicity gas . the total mass of our cloud ends up being @xmath23 , about a factor of four more than the estimated mass of m.o . ( @xmath24 ; c06 ) . the cloud is immersed in a background gas with @xmath25 @xmath14 and @xmath26 k. the cloud and the background are initially assumed to be in pressure equilibrium , which is used to set the initial temperature of the cloud . this gives a value around @xmath27 k. each simulation is run for five sound - crossing - times , @xmath28 , of the cloud , which corresponds to about 73 myr total . to avoid numerical problems caused by the background gas density dropping too far below its initial value , we set a density floor of @xmath29 @xmath14 . a jet is introduced into the simulation domain through one end . for this work , we choose @xmath30 @xmath14 ( with @xmath31 the same as the cloud ) , a jet diameter of @xmath32 pc , and a jet velocity of 10% of the speed of light . for a cylindrical jet , this yields a kinetic power , @xmath33 erg s@xmath34 , where @xmath35 is the cross - sectional area of the jet . our dependence on these model parameters is explored in section [ sec : jet_power ] . in our simulations , we follow the abundances of 9 atomic and molecular species : , , , , , @xmath36 , h@xmath37 , h@xmath0 , and h@xmath38 . the evolution of each species is governed by equations of the form @xmath39}}{\partial t } + \nabla \cdot ( \rho^{[m ] } \mathbf{v } ) = { \sum_{i=1}^{n_{s}}}{\sum_{j=1}^{n_{s } } } { { k_{ij}(t)}{\rho^{[i]}}{\rho^{[j ] } } } + \sum_{i=1}^{n_{s } } { i_i ( \nu ) \rho^{[i]}}. \label{eqn : dens_m}\ ] ] these rate equations are solved using a stable , semi - implicit , backward difference scheme that we developed in @xcite , and which has since become a standard method adopted by the general community ( e.g. * ? ? ? * ) due to its combination of robustness , efficiency , and accuracy . a total of 27 gas - phase chemical reactions are included in the full network ( @xmath40 ) , including 19 collisional and 8 photoionization / photodissociation processes . the exact reaction chains are spelled out in @xcite . the photoionization field is @xmath41 s@xmath34 , appropriate for cosmic uv background radiation at low redshift @xcite . the photodissociation rate is @xmath42 s@xmath34 , appropriate for the local interstellar medium @xcite . the energy equation @xmath43 = -\lambda(t , n^{[m]})~ , \label{eqn : dual_en}\ ] ] where @xmath44 is the total energy density , including the internal and kinetic contributions , accounts for the cooling and heating of the gas via a total of seven different mechanisms : collisional - excitation , collisional - ionization , recombination , bremsstrahlung , metal - line cooling ( dominantly carbon , oxygen , neon , and iron ) , molecular - hydrogen cooling , and photoionization heating . the cumulative cooling function is @xmath45 } ) = \sum_{i=1}^{n_{s } } \sum_{j=1}^{n_{s } } \dot{e}_{ij}(t ) n^{[i ] } n^{[j ] } - \sum_{i=1}^{n_{s } } j_i n^{[i ] } + \dot{e}_m(t ) n^2 ~ , \label{eqn : coolingchem}\ ] ] where @xmath46 are the cooling rates from 2-body interactions between species @xmath47 and @xmath48 and @xmath49 represents the frequency - integrated photoionization and photodissociation heating rates . @xmath50 is the temperature - dependent cooling rate for metals ( including carbon , oxygen , neon , and iron lines , assuming solar metallicity ) , taken from @xcite . a cooling floor is set at @xmath51 k , below which only adiabatic cooling is possible . in practice , we rarely see temperatures below @xmath52 k. in this study , we are particularly interested in tracking the formation of stars within the cloud . we follow the approach of @xcite in defining a density threshold , @xmath53 , above which star formation is triggered at a rate given by @xmath54 where @xmath55 is the local free - fall time and @xmath56 controls the star formation efficiency . we use values @xmath57 @xmath14 and @xmath58 for most of this work , though in section [ sec : sfr ] we explore how these choices affect our results . this star formation appears as a sink term in the continuity equation @xmath59 by tracking how much mass is lost to star formation during each compute cycle , we are able to continuously track the star formation rate throughout the simulation . we track the most massive ( @xmath60 ) `` stars '' created in this way using tracer particles . each star particle is given an initial velocity equal to the velocity of the gas in the zone in which the star is created . each star particle s velocity remains constant for the remainder of the simulation , as these particles are not acted on by any force . this is not entirely realistic as some momentum may get redistributed during the free - fall process that leads to star formation . on longer timescales , gravitational forces from the rest of the cloud will also act on the stars , eventually altering their velocities . these effects are not considered in this work . at the temperatures typical of star formation in our simulations ( @xmath61 k ) , the jeans length is @xmath62 which is considerably less than the radius of the cloud , yet well above the resolution limits of our simulations . however , since self - gravity is not included in our current simulations , the clumps can not truly become jeans unstable . table [ tab : params ] summarizes the 15 simulations presented in this work . the naming convention refers to the base resolution and how many total grid resolution levels there are . figure [ fig : volume ] presents a volume visualization from the end time of our highest resolution 3d simulation ( 384x128x128_3level ) . a number of general features are apparent . first , the cloud is dense enough and has enough inertia to dramatically slow the propagation of the jet ( represented by its temperature in red ) . at the speed the jet is traveling , if not for the cloud ( and , to a lesser extent , the background gas ) impeding its progress , it should have traversed 74.6 box lengths ( 298.4 cloud radii ) over the duration of the simulation . instead , most of the jet material is deflected to the sides of the cloud , though a significant fraction of its energy is deposited within the cloud gas . the green material represents neutral hydrogen ( ) and shows that the core of the original cloud remains relatively intact . some cooling ( represented by the blue , h@xmath0 , gas ) is happening on the fringes of the cloud ( at a radius of about @xmath63 in figure [ fig : volume ] ) , where weak shocks are being driven into the cloud by what is effectively now an over - pressure cocoon around the cloud . more substantial cooling is happening toward the front of the cloud in the post - shock gas ( dense blue filament just downstream of where the jet terminates ) . cccccccc [ tab : params ] 288x96_2level & @xmath64 & 2 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96_3level & @xmath64 & 3 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96_4level & @xmath64 & 4 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96_5level & @xmath64 & 5 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96_3level_n1_e.05 & @xmath64 & 3 & 1 & 0.05 & 1.5 & 0.1 & 5 + 288x96_3level_n4_e.05 & @xmath64 & 3 & 4 & 0.05 & 1.5 & 0.1 & 5 + 288x96_3level_n7_e.05 & @xmath64 & 3 & 7 & 0.05 & 1.5 & 0.1 & 5 + 288x96_3level_rhoj10 & @xmath64 & 3 & 4 & 0.1 & 1.5 & 0.1 & 50 + 288x96_3level_dj2 & @xmath64 & 3 & 4 & 0.1 & 3.0 & 0.1 & 5 + 288x96_3level_vj5 & @xmath64 & 3 & 4 & 0.1 & 1.5 & 0.5 & 5 + 384x128_3level & @xmath65 & 3 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96x96_2level & @xmath66 & 2 & 4 & 0.1 & 1.5 & 0.1 & 5 + 288x96x96_3level & @xmath66 & 3 & 4 & 0.1 & 1.5 & 0.1 & 5 + 384x128x128_2level & @xmath8 & 2 & 4 & 0.1 & 1.5 & 0.1 & 5 + 384x128x128_3level & @xmath8 & 3 & 4 & 0.1 & 1.5 & 0.1 & 5 + myr ) of our highest resolution 3d simulation ( 384x128x128_3level ) . red represents hot , @xmath67 k , material , mostly the jet , although possibly with some cloud gas mixed in ; green represents regions of the cloud with a neutral hydrogen density , @xmath68 @xmath14 ; and blue represents regions of the cloud with a molecular hydrogen ( our tracer for cold gas ) density , @xmath69 @xmath14 . data have been reflected across the @xmath70 and @xmath71 planes to create this image . axes are marked in units of cloud radii , @xmath72 . [ fig : volume ] ] figure [ fig:2d ] shows what the evolution looks like in one of our 2d simulations ( 384x128_3level ) . this evolution shows broad similarities to what is seen in the 3d simulations , although there are notable differences . for one , the jet has penetrated much further into the cloud in 2d by the end of the simulation . we will also discuss in section [ sec : cooling_front ] how the cooling and star formation in 2d take place in a `` wall '' proceeding forward through the cloud , while in 3d , they take place in a ring , centered about the jet axis . in section [ sec : star_particles ] , we also find that the star particles have a broader velocity distribution in 2d than 3d . such differences illustrate the necessity of 3d simulations . , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] + , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] , i.e. red represents @xmath67 k , green represents @xmath73 @xmath14 , and blue represents @xmath74 @xmath14 . the panels show the evolution , spaced at intervals of one cloud - crossing time , i.e. @xmath75 . the dramatic change in the size of the cloud between the first and second panels is due to photoionization . the overall size and density of the cloud does not change much , but the ionization levels do . [ fig:2d],title="fig : " ] figure [ fig : shocks ] is a volume visualization showing regions of significant velocity convergence ( @xmath76 ) , as an indicator of where shocks are located . from it , we see that , even after 73 myr , the jet - driven shock still has not reached the center of the cloud . we can compare this plot with the estimated shock velocity in the cloud . for our chosen jet velocity , @xmath77 km s@xmath34 , we can use the shock jump conditions , @xmath78 where @xmath79 and @xmath80 are measured in the rest frame of the pre - shock gas , to estimate the speed of the shocks in the background gas , @xmath81 km s@xmath34 . if the shock in the background is strong , then the postshock pressure is approximately @xmath82 . if we assume the shocks are also strong inside the cloud and that the postshock background and cloud gas reach pressure equilibrium , then we can estimate the speed of the shock in the cloud @xmath83 for our median cloud density , @xmath84 @xmath14 , this gives us a shock speed in the cloud of @xmath85 km s@xmath34 . thus , the shock should have traveled approximately 22% of the way through the cloud by the end of our simulation , roughly consistent with figure [ fig : shocks ] . as expected , material that is ahead of the shock ( downstream ) remains relatively undisturbed , except on the fringes of the cloud where the over - pressure cocoon is driving weaker shocks into the cloud ( blue outline of the cloud in figure [ fig : shocks ] ) . ) , indicative of shocks , from the final time dump ( @xmath86 myr ) of our highest resolution 3d simulation ( 384x128x128_3level ) . yellow , orange , and red shades indicate the strong shocks associated with the jet , while blue indicates much weaker shocks . data have been reflected across the @xmath70 and @xmath71 planes to create this image . axes are marked in units of cloud radii , @xmath72 . [ fig : shocks ] ] as mentioned in section [ sec : jetcloud ] , for the jet feedback to be positive , it is critical for the cooling timescale to be short compared to the shock - crossing time of the cloud . following @xcite , we estimate the cooling time to be @xmath87 this is _ much _ shorter than any other relevant timescale in the problem . interestingly , the greatest amount of cooling ( and associated star formation ) is seen in a ring radiating away from the initial impact point at the head of the cloud . in figure [ fig : volume ] , you can see evidence of this in the light green ring of just to the left of the cloud s center . a smaller , denser ring of h@xmath0 follows behind the ring . figure [ fig : formation ] shows how this cooling front and associated star formation moves over the course of the simulation . this distribution is significantly different than what one sees in 2d simulations ( figure [ fig:2d ] ) , where cooling and star formation occurs more uniformly across the cloud diameter ( perpendicular to the jet propagation ) . ) at @xmath88 ( orange ) , 44 ( green ) , 58 ( blue ) , and 73 myr ( purple ) from our highest resolution 3d simulation ( 384x128x128_3level ) . the left panel shows a side view , while the right panel shows a head - on view . data have again been reflected across the @xmath70 and @xmath71 planes . [ fig : formation],title="fig : " ] ) at @xmath88 ( orange ) , 44 ( green ) , 58 ( blue ) , and 73 myr ( purple ) from our highest resolution 3d simulation ( 384x128x128_3level ) . the left panel shows a side view , while the right panel shows a head - on view . data have again been reflected across the @xmath70 and @xmath71 planes . [ fig : formation],title="fig : " ] figure [ fig : slices ] helps clarify the nature of these rings . note first that , as expected , there is a compression front moving forward through the body of the cloud ( approximately vertical feature extending from @xmath89 to @xmath90 in both panels ) . there is also a compression front moving radially into the cloud from all sides ( arced feature extending from @xmath91 to @xmath92 in both panels , although it extends around the entire cloud ) , driven by the over - pressured cocoon . cooling seems to be operating effectively in both of these regions , as the temperature remains @xmath93 k ( blue regions in the right panel ) . however , in most parts of these fronts , the density does not quite reach the threshold for star formation ( @xmath94 ) . this does _ not _ appear to be a resolution effect as the fronts are well resolved in most places and our overall s.f.r . seems to have converged ( see figure [ fig : sfr ] ) . instead , the ring of star formation seen in figures [ fig : formation ] and [ fig : stars ] seems to correspond to regions where the forward propagating compression wave intersects the radial compression front ( most strongly around @xmath95 ) . the combined effect of the two shocks is enough to push the density over the star formation threshold . . the right panel shows temperature in k. note how the densest , coldest gas is found around @xmath96 , where the nearly planar front moving forward through the cloud intersects the nearly spherical front being driven into the cloud . [ fig : slices],title="fig : " ] . the right panel shows temperature in k. note how the densest , coldest gas is found around @xmath96 , where the nearly planar front moving forward through the cloud intersects the nearly spherical front being driven into the cloud . [ fig : slices],title="fig : " ] as a reminder , whenever more than @xmath97 of gas is converted into stars within a given zone within a single cycle , then a star particle is created to track the properties , such as position , age , and velocity , of that `` star . '' following this prescription , we created over @xmath98 star particles in our highest resolution simulation ( 384x128x128_3level ) . the star particles span an age range from 0 72 myr , although the star particles associated with the jet interaction are all younger than about 50 myr ; the oldest stars formed spontaneously at the start of the simulation inside zones whose random density exceeded the star formation threshold . even the 50 myr age is considerably older than the stellar age range of 7.0 - 8.5 myr estimated for m.o . much of this discrepancy can be explained by the delay associated with the free - fall time , which is not accounted for in the `` ages '' of our star particles . for our threshold density , @xmath53 , the associated free - fall time is @xmath99 myr , bringing the ages more in line with m.o . furthermore , the stellar ages in c06 may actually be telling us more about the age of the jet . it may be that m.o . has not yet evolved for as long as we ran our simulations . an interesting point about the stellar ages in our simulations is that they show a negative curvature along the direction of jet propagation , that is , the youngest ( currently forming ) stars are found between two populations of slightly older stars , one in the upstream direction and one downstream . this can be seen for one of our 2d simulations in the right panel of figure [ fig : stars_2d ] . the upstream population are stars that formed recently , only slightly before the current star formation . the downstream population are some of the first stars to form from the jet interaction , but they are now actually located ahead of the current star formation front because they received a velocity kick larger than the current shock speed . since the star particles are not coupled to the gas , they can actually pass ahead of the shock as it slows down , giving an apparently older population ahead of the current star formation front . myr ) of our 384x128_3level 2d simulation . in the left panel , the color of each particle indicates its velocity magnitude ( in km s@xmath34 ) , while in the right panel , the color indicates age ( in myr ) . [ fig : stars_2d],title="fig : " ] myr ) of our 384x128_3level 2d simulation . in the left panel , the color of each particle indicates its velocity magnitude ( in km s@xmath34 ) , while in the right panel , the color indicates age ( in myr ) . [ fig : stars_2d],title="fig : " ] by using our results from sec . [ sec : shock ] and inverting the shock jump condition , we predict a post - shock velocity in the cloud of @xmath100 km s@xmath34 , consistent with the observed 40 km s@xmath34 shear velocity in m.o . ( c06 ) . since our star particles are assigned their velocity based upon the velocity of the gas from which they form , we expect the simulated star particle velocities to be similar . figure [ fig : histogram ] shows a histogram of the velocity distribution of all of the star particles . the distribution peaks at around 50 km s@xmath34 , which is fairly consistent with our crude predictions and the observations of m.o . figures [ fig : stars_2d ] ( left panel ) and [ fig : stars ] show the spatial distributions of the star particles , colored by the magnitude of their velocities , in 2d and 3d , respectively . note the significant differences in the distributions , although the magnitudes of the velocities ( and ages of stars ) are comparable . as mentioned above , the fastest moving star particles are on the downstream edges of the distributions . they are also found further out , i.e. at larger radii , on average in the 3d simulation . bins are filled with stars that formed in the initial cloud , prior to its interaction with the jet . the peak of the 3d distribution is roughly consistent with observations of m.o . [ fig : histogram],title="fig : " ] bins are filled with stars that formed in the initial cloud , prior to its interaction with the jet . the peak of the 3d distribution is roughly consistent with observations of m.o . [ fig : histogram],title="fig : " ] myr ) of our highest resolution 3d simulation ( 384x128x128_3level ) . the color of each particle indicates its velocity magnitude ( in km s@xmath34 ) . the left panel shows a side view , while the right panel shows a head - on view . data have again been reflected across the @xmath70 and @xmath71 planes . [ fig : stars],title="fig : " ] myr ) of our highest resolution 3d simulation ( 384x128x128_3level ) . the color of each particle indicates its velocity magnitude ( in km s@xmath34 ) . the left panel shows a side view , while the right panel shows a head - on view . data have again been reflected across the @xmath70 and @xmath71 planes . [ fig : stars],title="fig : " ] figure [ fig : sfr ] shows the measured star formation rate ( s.f.r . ) as a function of time for four different 3d simulations done at different effective resolutions from 57.1 pc per zone ( 288x96x96_2level ) to 19.5 pc per zone ( 384x128x128_3level ) . as might be expected , the simulated s.f.r . is sensitive to resolution . at higher resolutions , more gas is going to be able to reach the density threshold , @xmath53 , of our star formation model . however , at some point a limit should be reached where the high density filaments are well enough resolved that their size and peak density are no longer functions of resolution . it appears we may have reached this point in our highest resolution simulation . note that the peak star formation rate actually goes down compared to the next lower resolution . also , the time at which star formation begins remains consistent at about 25 myr for the two highest resolutions . a convergence at a resolution of about 25 pc per zone also seems to be supported by our 2d tests ( right panel of figure [ fig : sfr ] ) , where we were able to consider even higher resolutions . . together , the figures suggest that our results are reasonably well converged at resolutions above about 25 pc per zone . note that the highest resolution 3d simulation produces a steady s.f.r . of @xmath101 yr@xmath34 ( corresponding to an h@xmath102 luminosity of @xmath103 erg s@xmath34 ) . this falls short of the observed rate in m.o . ( grey , dashed line ) by about a factor of 6 . the units on the s.f.r . axis in 2d are arbitrary because there is no meaningful way to assign a volume to the cloud . [ fig : sfr],title="fig : " ] . together , the figures suggest that our results are reasonably well converged at resolutions above about 25 pc per zone . note that the highest resolution 3d simulation produces a steady s.f.r . of @xmath101 yr@xmath34 ( corresponding to an h@xmath102 luminosity of @xmath103 erg s@xmath34 ) . this falls short of the observed rate in m.o . ( grey , dashed line ) by about a factor of 6 . the units on the s.f.r . axis in 2d are arbitrary because there is no meaningful way to assign a volume to the cloud . [ fig : sfr],title="fig : " ] the s.f.r . , or correspondingly the h@xmath102 luminosity , since @xcite @xmath104 in figure [ fig : sfr ] is about a factor of 6 less than the observed value in m.o . ( @xmath105 yr@xmath34 and @xmath106 erg s@xmath34 ; c06 ) . this most likely means that one or more of our simulation parameters does not match the situation in m.o . to increase the s.f.r . in the simulation , we could : decrease the star formation density threshold , @xmath53 ; increase the star formation efficiency , @xmath56 ; increase the jet power ; or increase the initial cloud density . we use a set of 2d simulations to test the impact of changing @xmath53 , @xmath56 , and the jet power . in the rest of this section , we focus on @xmath53 and @xmath56 . in the next section , we consider varying the parameters that affect the jet power . we do not , however , explore the effect of further increasing the cloud density , as it is already high enough to trigger spontaneous star formation . we do not want such spontaneous star formation to overwhelm the jet - induced star formation that we are studying . figure [ fig : sfrmodel ] shows the measured s.f.r . for four different 2d simulations using different values for @xmath53 and @xmath56 . not surprisingly , if we lower the star formation threshold density ( black , dashed curve ) , then star formation occurs much more easily , so much so that the rate of spontaneous star formation ( that not associated with the jet interaction ) is within a factor of 10 of the jet - induced rate . interestingly , cutting the star formation efficiency in half ( green , dot - dashed curve vs. blue , dotted curve ) does not simply cut the s.f.r . in half . this illustrates the non - linearity of this star formation process and the need for high fidelity simulations . and @xmath56 , for 2d simulations with a base resolution of @xmath64 and 3 total resolution levels . the dotted , blue curve shows results with our default parameters . the units on the s.f.r . axis are arbitrary because in 2d , there is no meaningful way to assign a volume to the cloud . [ fig : sfrmodel ] ] ultimately , we find that adjusting the star formation parameters does not really aid in raising our measured s.f.r . our standard parameters , @xmath57 @xmath14 and @xmath58 , already produce the highest induced s.f.r . of any of these simulations . increasing the star formation efficiency further might yield some further increase in the s.f.r . , but we are already at what we consider a reasonable upper limit . using the known correlation between jet power and radio luminosity , @xmath107 @xcite , and the radio luminosity of ngc 541 , @xmath108 erg s@xmath34 @xcite , we estimate the kinetic power of the jet impacting m.o . to be @xmath109 erg s@xmath34 . by contrast , our default jet parameters give a power that is more than two orders of magnitude lower . in figure [ fig : sfrjet ] , we explore how increasing the jet power would impact our results . increasing the jet density ( black , dashed curve ) increases the s.f.r . and also causes star formation to start earlier . this is because a denser jet more easily penetrates the background gas to reach the cloud . increasing the jet diameter ( green , dot - dashed curve ) also increases the s.f.r . , although this happens gradually , not even peaking by the end of the simulation . finally , increasing the jet velocity ( red , double - dot - dashed curve ) again causes star formation to start earlier , but also makes the shock crossing time comparable to the cooling timescale , such that the cloud is disrupted ( and ultimately blown off the simulation domain ) before a significant number of stars can form ( s.f.r . goes to zero around @xmath110 myr ) . , @xmath111 , and @xmath112 , for 2d simulations with a base resolution of @xmath64 and 3 total resolution levels . the dotted , blue curve shows results with our default parameters . the units on the s.f.r . axis are arbitrary because in 2d , there is no meaningful way to assign a volume to the cloud . [ fig : sfrjet ] ] from these tests , it appears that the best option for matching the s.f.r . in m.o . would be a simulation with a denser jet . extrapolating from our 2d results , and assuming the same increase is seen in 3d , it appears that a factor of 25 increase in jet density might be enough to have the peak s.f.r . in our simulation match the currently observed value in m.o . this would also bring the kinetic power of our jet more in line . figure [ fig : history ] tracks the total mass of , h@xmath0 , and stars over the course of our highest resolution simulation . while the simulation produces roughly the correct amount of [ @xmath113 in the simulation vs. @xmath24 for m.o . ( c06 ) ] , it produces far less h@xmath0 [ @xmath114 vs. @xmath115 in m.o . ( lacy et al . , in prep . ) ] , and about a factor of 10 fewer stars [ @xmath116 vs. @xmath117 in m.o . ( c06 ) ] . we find an electron number density in our cloud of @xmath118 @xmath14 , also somewhat lower than the range of 1 - 10 @xmath14 obtained for m.o . since the amount of gas is comparable , this suggests that the cooling and star formation in the simulations are not nearly as efficient as they are in m.o . the sizable discrepancy in h@xmath0 could be reduced if we included the effects of self - shielding , which would reduce the amount of photodissociation and increase the yield of h@xmath0 . however , even in simulations where we completely ignore photodissociation , we still find the h@xmath0 mass to be at least two orders of magnitude too low . another important point about our h@xmath0 values is that we are only forming h@xmath0 through gas phase reactions . in m.o . , it may be that substantial h@xmath0 formation takes place on dust grains . however , simulations that include precipitation of h@xmath0 on dust , with standard grain parameters , show only a very modest increase in @xmath119 . likely , further work will be needed to better understand the h@xmath0 levels in m.o . , @xmath119 , @xmath120 , and @xmath121 over time for our highest resolution 3d simulation ( 384x128x128_3level ) . [ fig : history ] ] figure [ fig : history ] also tracks the `` star formation efficiency , '' @xmath121 , over time . by this measure , our simulation achieves a peak star formation efficiency of 0.6% . this , too , is lower than the value of 4% measured in m.o . all of these discrepancies can be traced to the low s.f.r . seen in our simulations . as mentioned previously , the _ peak _ s.f.r . in our simulations ( see figure [ fig : sfr ] ) is about a factor of 6 less than the _ current _ s.f.r in m.o . , though we showed in the previous section that increasing the jet power in the simulations might be able to remove this discrepancy . in this work , we use 2d and 3d simulations to model the evolution of a radio jet impacting a single , dwarf - galaxy - scale cloud in a direct , axially - symmetric collision . our intention is to use these simulations to better understand observations of m.o . , the peculiar starburst galaxy located at the termination point of the radio jet from ngc 541 . our first , main conclusion is that jet - induced star formation ( i.e. positive feedback ) is possible under this scenario . figures [ fig : sfr][fig : history ] all show dramatic increases in star formation attributable to the jet interaction . interestingly , we find that most of the star formation in our 3d simulations takes place in a propagating ring centered about the jet axis ( figure [ fig : formation ] ) . the ring lies at the intersection of a forward propagating shock driven into the cloud by the jet and a spherical compression shock driven by the over - pressure cocoon ( figure [ fig : slices ] ) . these rings may potentially explain the torus of star formation observed in m.o . importantly , this star formation is occurring upstream of an ring ( compare figures [ fig : volume ] and [ fig : formation ] ) , also consistent with observations ( c06 ) . additionally , the velocity histogram of our star particles peaks at @xmath122 km s@xmath34 ( left panel of figure [ fig : histogram ] ) , which is again consistent with observations ( c06 ) . the spatial distribution of the star particle velocities is also interesting . the fastest moving star particles are found furthest downstream ( figures [ fig : stars_2d ] and [ fig : stars ] ) . these are also some of the first star particles to form ( i.e. they are the oldest ) , which is consistent with a slowing of the propagation speed of the star formation front . despite these successes , the correlation of our results with m.o . is imperfect . the main discrepancy is that our measured star formation rate is about a factor of 6 lower than the observed value . c06 also suggested that the star formation rate in m.o . may have been even higher in the past . along with not producing the expected mass of stars , our simulations also fall short on the amount of h@xmath0 they produce . in the paper , we have tried to explore possible avenues to rectify these discrepancies . we first considered our star formation model ; there we found that our results are not strongly sensitive to its parameters . further , we are already close to what we consider a reasonable upper limit for the star - formation efficiency , so there is little chance to make up the discrepancy that way . the sensitivity of our results to the jet parameters was another matter . first , we found that the jet power may have been lower than what is inferred in the case of m.o . by as much as two orders of magnitude . in our 2d simulations , though , we found a complicated dependence of the s.f.r . on jet power ( figure [ fig : sfrjet ] ) . while increasing the jet diameter by a factor of two or the jet density by a factor of ten roughly doubled the s.f.r . , increasing the jet velocity by a factor of 5 actually lowered the s.f.r . and destroyed the cloud on a relatively short timescale . a similar strong sensitivity to the shock velocity was seen in our earlier work @xcite . going the other direction and significantly lowering the jet velocity would introduce a different problem star formation would likely proceed too slowly . once the first generation of massive stars form , there is only a limited amount of time ( of the order a few myr ) for star formation to continue before negative feedback , in the form of heating from the first massive stars , would effectively shut it off @xcite . this effect might explain why the s.f.r . in m.o . appears to be lower now that it was in the past negative feedback may already be kicking in . as future observations continue to constrain the star formation history of m.o . , we plan to continue to refine our simulations . future modifications to our setup may include : simulating non - axially - symmetric interactions between the jet and cloud ; having the jet sweep across the cloud ; or adding more feedback mechanisms , such as heating from young stars and supernovae . this work used the extreme science and engineering discovery environment ( xsede ) , which is supported by national science foundation grant number aci-1053575 . pcf acknowledges support from national science foundation grants ast-1211230 and ast-1616185 . jwlw acknowledges support from nrao student observing support grant sospa3 - 020 . work by pa was performed in part under the auspices of the u.s . department of energy by lawrence livermore national laboratory under contract de - ac52 - 07na27344 .
inspired by wheeler s quest for the existence of geons @xcite , boson stars were introduced by feinblum and mckinley @xcite , kaup @xcite , and bonazzola and ruffini @xcite . in these boson stars the electromagnetic vector field was replaced by some tentative scalar field . with the discovery of a higgs - like boson last year at the lhc @xcite the first fundamental scalar field has been found . but numerous scalar fields have been predicted to exist in high energy physics and cosmology . boson stars arise as stationary localized solutions of the coupled einstein - klein - gordon equations @xcite . the physical properties of boson stars depend strongly on the type of scalar field potential employed ( see e.g. the review articles @xcite ) . mini boson stars arise , when only a mass term is present but no self - interaction . they can reach only relatively small masses . larger boson stars are obtained , when a repulsive quartic self - interaction is included @xcite . for these two types of boson stars gravity is a necessary ingredient . in contrast , in the presence of a sextic potential one obtains solitonic boson stars @xcite , which possess a flat space - time limit , where they correspond to non - topological solitons @xcite ( or @xmath0-balls @xcite ) . moreover , these solitonic boson stars can reach even higher masses @xcite . here we consider boson stars , which are compact in the sense , that the scalar field of these spherically symmetric configurations is finite inside a ball of radius @xmath3 , but vanishes identically outside this radius . in this respect the compact @xmath0-balls resemble stars @xcite . obtained from a v - shaped self - interaction potential , these compact boson stars also possess a flat space - time limit , compact @xmath0-balls @xcite . these represent solutions of the signum - gordon equation . the study of scalar fields with a v - shaped self - interaction potential has revealed interesting physical phenomena . when coupled to electromagnetism , the balance of forces allows for shell - like configurations @xcite . in these @xmath0-shells the scalar field vanishes identically both inside a certain radius @xmath4 and outside a certain radius @xmath3 . the scalar field thus forms a shell of charged matter , @xmath5 . when coupling these shells to gravity the resulting boson shells possess an empty minkowski space interior @xmath6 . however , the shells need not be empty in their interior , they can harbour a black hole in there @xcite . thus one finds that uniqueness and no - hair theorems for black holes can be avoided in the presence of boson shells @xcite . whereas these previous studies considered only asymptotically flat solutions , we here include a cosmological constant . on the one hand , a positive cosmological constant is relevant from an observational point of view , since it can model the dark energy of the universe . since boson stars are very compact objects that can possess very high densities , they have been suggested as alternatives to supermassive black holes , e.g. in the center of galaxies @xcite . even if that would be excluded by observations ( for a discussion see e.g. @xcite ) boson stars could still act as toy models for very compact objects , e.g. neutron stars . such a model of a compact star in a space - time with positive cosmological constant should be a more realistic description of compact stars in the universe , since all observations seem to indicate the existence of a form of dark energy . a negative cosmological constant , on the other hand , leads to solutions that can be interpreted within the ads / cft correspondence @xcite . recently , the study of boson stars in ads space - time received increasing attention @xcite . this is related to the fact that within the context of a holographic description of superconductors and superfluids @xcite ( for reviews see @xcite ) the formation of scalar hair on charged solitons in asymptotically ads has been interpreted as an insulator / superconductor phase transition @xcite . the limit of setting the electric charge of the scalar field @xmath7 to infinity , which due to the scaling symmetries corresponds to setting newton s constant @xmath8 to zero , is called the `` the probe limit '' in this context . we adapt this nomenclature here and refer to the case , where the matter field equation is solved in a fixed background , as `` the probe limit '' . in the opposite limit @xmath9 , the gauge symmetry becomes global and the resulting solutions are uncharged solitons in ads . these are essentially uncharged boson stars and have been suggested to be the holographic description of glueball condensates @xcite . boson stars in asymptotic ads are also of interest from another point of view . it has been suggested that the dynamical formation of a black hole in ads is the dual description of thermalization in a strongly coupled quantum field theory . as such the stability of ads space - time was studied with respect to perturbations , and it was conjectured that ads is unstable under arbitrarily small scalar perturbations and that eventually a black hole would form due to the reflection of the perturbations on the ads boundary @xcite . however , in @xcite it was shown that boson stars appear to be nonlinearly stable . if that were true , thermalization in the dual field theory would not occur . hence , boson stars in ads play an important rle in the context of the nonlinear ( in)stability of ads space - time , and thus of its dual description . here we first consider the set of boson star solutions for various space - time dimensions @xmath10 in the probe limit . interestingly , the presence of a positive cosmological constant allows for the existence of boson shells without an electromagnetic field . we note , that all objects constructed here are electrically neutral . by solving the coupled set of the einstein - signum - gordon equations , we subsequently determine the domain of existence of the compact ( a)ds boson stars and shells in @xmath10 space - time dimensions . we analyze their physical properties and briefly address the stability of the boson stars from a catastrophe theory point of view @xcite . we also address astrophysical aspects of boson stars and boson shells in four dimensions for the physical value of the cosmological constant . the paper is organized as follows . in section 2 we present the action , the ansatz , the equations of motion together with the scaling property , the boundary conditions and the global charges . we present the solutions in the probe limit in section 3 . the boson stars and shells obtained with the back reaction taken into account are discussed in section 4 . we end with our conclusions and an outlook in section 5 . we consider the action of a self - interacting complex scalar field @xmath11 coupled to einstein gravity in @xmath1 dimensions @xmath12 \sqrt{-g } d^dx , \label{action}\ ] ] with curvature scalar @xmath13 , cosmological constant @xmath14 , newton s constant @xmath8 , and the asterisk denotes complex conjugation . the scalar potential @xmath15 is chosen as @xmath16 variation of the action with respect to the metric and the matter fields leads , respectively , to the einstein equations @xmath17 with stress - energy tensor @xmath18 and the matter field equation , @xmath19 where @xmath20 denotes the covariant derivative . invariance of the action under the global phase transformation @xmath21 leads to the conserved current @xmath22 and the associated conserved charge @xmath0 . to construct spherically symmetric boson star solutions we employ schwarzschild - like coordinates and adopt the spherically symmetric metric @xmath23 where @xmath24 is the metric on the @xmath25 dimensional unit sphere . the associated ansatz for the boson field takes the form @xmath26 with the frequency @xmath27 . the conserved scalar charge @xmath0 @xmath28 is then proportional to @xmath27 . we next introduce dimensionless quantities by @xmath29 thus the coupling strength of gravity is expressed in terms of the coupling constant @xmath30 . this yields the set of equations @xmath31 let us now specify the boundary conditions for the metric and the boson field . for the metric function @xmath32 we adopt @xmath33 where @xmath34 is the outer radius of the boson star . since it retains this value to infinity , this fixes the time coordinate . for the metric function @xmath35 we require at the origin the regularity condition for globally regular ball - like boson star solutions @xmath36 and for globally regular shell - like solutions @xmath37 where @xmath38 is the inner radius of the shell . for boson stars we require for the boson field function one condition at the origin and two conditions at the outer radius @xmath34 @xmath39 since this is one condition too many , we introduce another auxiliary differential equation , @xmath40 , by treating @xmath34 as a function . thus @xmath41 is constant , but the value of the constant is adjusted in the numerical scheme such that the boundary conditions eqs . ( [ bc ] ) are satisfied . this determines the outer radius of the star . for boson shells , on the other hand , we require at the inner radius @xmath42 and at the outer radius @xmath43 the conditions @xmath44 we now also make the ratio of inner and outer radius @xmath45 an auxiliary ( constant ) variable . for the numerical computation we introduce the scaled coordinate @xmath46 , such that the inner radius is at @xmath47 and outer radius is at @xmath48 . we refer to the solution in the exterior region @xmath49 as the outer solution of the boson stars and boson shells . in the asymptotically flat case , the outer solution is given by the schwarzschild solution . in the presence of a cosmological constant the schwarzschild - de sitter and schwarzschild - anti - de sitter solutions @xmath50 are exact solutions of the odes in the exterior region . hence the mass parameter of the solutions is given by @xmath51 analogously , we refer to the solution in the interior region @xmath52 as the inner solution of the boson shells . in the asymptotically flat case , the regular inner solution corresponds to flat minkowski space , whereas in the presence of a cosmological constant the regular inner solutions correspond to either de sitter or anti - de sitter space . note however , that in general @xmath53 . thus a rescaling of the time coordinate , @xmath54 , is required to obtain the de sitter or anti - de sitter line element in the standard form . in principle , we could replace these regular inner solutions by the appropriate black hole solutions , analogously to the asymptotically flat case @xcite . then these inner solutions would correspond to schwarzschild - de sitter or schwarzschild - anti - de sitter solutions . here we present the families of solutions in the so called probe limit . thus we obtain the solutions for vanishing coupling to gravity , i.e. , @xmath55 , in the respective background . for vanishing cosmological constant the @xmath0-ball solutions can be found analytically and expressed in terms of bessel functions . the @xmath56 solution was given in @xcite @xmath57 for @xmath10 dimensions this generalizes according to @xmath58 where @xmath59 . the constant @xmath60 and the outer radius @xmath41 are determined by the conditions @xmath61 and @xmath62 . the latter yields @xmath63 . hence @xmath34 is the smallest non - vanishing zero of @xmath64 . from the first condition it then follows that @xmath65 . thus we find @xmath66 the properties of the unique solution in @xmath56 dimensions was discussed in @xcite . when going to higher dimensions , the properties of the respective solutions vary only slowly with @xmath1 , and likewise when going to @xmath67 . this is seen in fig . [ fig1 ] , when restricting to a vanishing cosmological constant , @xmath68 . let us now consider the @xmath0-ball solutions in an ( anti-)de sitter background in @xmath1 dimensions . to obtain these solutions , we have solved the scalar field equation in the respective background numerically , employing a newton - raphson scheme . when the scaled cosmological constant @xmath69 is varied , the solutions change smoothly from the minkowski background solutions . in fig . [ fig1 ] we exhibit the dependence of the solutions on @xmath69 . here we show the value of the scalar field at the origin @xmath70 together with the value of the outer radius @xmath41 of the solutions for @xmath67 , 4 , 5 and 10 dimensions . as the scaled cosmological constant @xmath69 increases from zero , the value of the scalar field at the origin @xmath70 decreases along with the outer radius @xmath41 . interestingly , there is a maximal value @xmath71 , which increases with the dimension @xmath1 , for which these solutions exist . at @xmath71 a second branch of solutions is encountered . moving backwards along this second branch the value of the scalar field at the origin @xmath70 continues to decrease , until it reaches zero at a critical value @xmath2 . at @xmath2 the solutions change character and @xmath0-shells arise . as @xmath69 decreases further , the inner radius @xmath72 increases along with the outer radius @xmath41 . in the limit @xmath73 the size of the shells diverges while the ratio @xmath74 tends to one . thus there are no @xmath0-shells in a minkowski or anti - de sitter background . likewise , when the @xmath0-ball solutions are continued to negative values of the cosmological constant , they change smoothly from the minkowski background solutions , as seen in fig . as the scaled cosmological constant @xmath69 decreases from zero , the value of the scalar field at the origin @xmath70 increases along with the outer radius @xmath41 until a minimal value @xmath75 encountered , beyond which no such solutions exist . our data indicate , that @xmath76 . for a derivation of this limit in @xmath56 dimensions see appendix a. to study the back reaction of the @xmath0-balls and @xmath0-shells on the space - time , we have solved the coupled system of equations for the metric and the scalar field numerically . in the following we first discuss the case of @xmath56 dimensions , and then turn to other dimensions . let us start by briefly recalling the properties of the single family of asymptotically flat compact boson star solutions found in @xcite . at @xmath55 this family starts from the @xmath0-ball solution in the minkowski background . as seen in fig . [ fig2 ] , with increasing @xmath30 the value of the scalar field at the origin @xmath70 decreases , until it reaches a finite minimum , and then increases strongly , while @xmath30 undergoes damped oscillations . in other physical quantities these damped oscillations with respect to @xmath30 lead to a spiral - like pattern , as seen for the outer radius @xmath41 or the mass parameter @xmath77 . such a behaviour is typical for boson stars and neutron stars . here we have extended this family of compact boson star solutions to negative values of the coupling constant @xmath30 , as seen in fig . the physical interpretation of the solutions with negative @xmath30 is that they represent compact solutions made from phantom scalar fields . thus the negative sign of @xmath30 can be absorbed by the negative lagrangian of the phantom field . ever since the relevance of dark energy for cosmology became apparent , such phantom fields are found ubiquitously in the literature . moreover , phantom fields also allow for the formation of various types of wormholes . let us now turn to compact de sitter boson stars by increasing the value of @xmath69 from zero . we demonstrate the effect of a positive cosmological constant on the compact solutions in fig . [ fig2 ] by exhibiting the physical properties of such families of solutions for several values of @xmath69 . we see that for finite @xmath69 the minimum of @xmath70 reaches zero . this signals the occurrence of boson shells . in particular , for a given value of @xmath69 , @xmath70 reaches zero at a critical value @xmath78 . then for @xmath79 boson shells exist . @xmath78 is exhibited in fig . for very small positive @xmath69 , the critical value @xmath78 is negative . thus there exist phantom boson shells in this case , which turn into ordinary boson shells , as @xmath30 increases beyond zero . extrapolation to @xmath80 indicates , that indeed no shells exist in this limit . for all values of @xmath69 the compact boson stars exhibit the characteristic spirals . in fig . [ fig2 ] these are seen for the value of the outer radius @xmath41 , the value of the metric function at the outer radius @xmath81 and the value of the scaled mass @xmath77 . interestingly , phantom type boson star solutions exist only for small values of @xmath69 . since the mass @xmath77 has the same sign as @xmath30 , branches with negative mass exist only for small values of @xmath69 . whereas there is an upper bound @xmath82 for the compact boson stars , there is no such bound for the boson shells . however , with increasing @xmath30 their outer radius decreases , and tends to a finite limiting value . at the same time , the ratio of the inner and outer radii @xmath74 increases , and tends to one . thus the shells become smaller and thinner , while their scaled mass @xmath77 grows . on the other hand , as @xmath30 is kept fixed while @xmath69 is decreased , the shells grow in size , while the ratio @xmath74 tends to one . here in the limit @xmath83 the shell size diverges . in the above subsection we have constructed the domain of existence of compact boson stars and boson shells in terms of dimensionless quantities . we can obtain physical solutions with dimensionful quantities by scaling these dimensionless solutions appropriately . in the case of vanishing cosmological constant we have considered compact stars @xcite . in particular , we have shown , that when the mass of these boson stars is on the order of the solar mass then their radius is on the order of ten(s ) of kilometers , thus they can correspond in mass and size to neutron stars . moreover , spirals are also encountered for neutron stars , when they approach the black hole limit . concerning their stability , we employ arguments from catastrophe theory @xcite . according to catastrophe theory , the stability changes only at turning points . thus when starting from a stable configuration , the stability should change at the maximum of the mass . therefore solutions inside the spiral should be unstable . for neutron stars or ordinary boson stars this has been confirmed by a mode analysis . since the value of the cosmological constant as obtained from cosmology is very small , @xmath84 m@xmath85 in metric units , its presence hardly affects the properties of boson stars that have masses on the order of the mass of the sun . thus the results for boson stars correspond to those obtained before @xcite . however , the presence of a positive cosmological constant , no matter how small , does allow for boson shells . but those boson shells possess cosmological mass and length scales . it would be interesting to see whether such thin boson shells can be associated with voids , i.e. with vast regions of empty space surrounded by a shell of matter . to see the effect of @xmath14 on the compact boson stars let us now consider very big scales . in fig . [ fig4 ] we show our scaled results for the mass @xmath86 in units of @xmath87 for boson stars and boson shells versus the outer radius @xmath41 in units of gigaparsec . keeping @xmath14 at its physical value , we have translated it into a timescale @xmath88 via the relation ( [ scaling ] ) , employing units of gigayears . shown in addition are the black hole horizon and the cosmological horizon of the corresponding schwarzschild - de sitter space - times . note , that these two horizons coincide for the extremal configuration with the maximum value of the mass . these schwarzschild - de sitter values form the boundary , within which all extended objects must remain . as seen in fig . [ fig4 ] , the boson stars reside well within these bounds . note , that since we consider only positive values of the gravitational coupling , i.e. , no phantom fields , the boson star curves associated with the smaller values of the parameter @xmath88 consist of two disconnected parts . the boson shells , in contrast , exist until they reach this cosmological bound . in particular , all boson shell curves extend precisely to the extremal value of the schwarzschild - de sitter curve , where the two horizons coincide . when approaching this limiting configuration , the inner radius of the shells approaches the outer radius . thus the ratio @xmath74 tends to the value one in this limit . more massive shells can not exist . converting the values in fig . [ fig4 ] into numerical values we find that with @xmath89 the mass of the boson stars is on the order of @xmath90 kg and their radius is on the order of gpc . these are sizes that are beyond those of galaxies and galaxy clusters . if there were dark matter distributions on such large scales , our solutions would be able to model those . choosing somewhat smaller values of @xmath88 , on the other hand , we would find sizes relevant for galaxies or galaxy clusters , so that these solutions could be considered to model the dark matter halo of galaxies or the dark matter in galaxy clusters , respectively . note that in the limit of vanishing mass the radius @xmath91 becomes spurious since the non - scaled boson field vanishes identically . as expected , compact boson stars exist also for negative values of the cosmological constant . indeed , the asymptotically minkowski solutions can be smoothly extended to negative values of @xmath69 , thus yielding asymptotically ads boson stars . in fig . [ fig5 ] we exhibit some of their physical properties versus the coupling constant @xmath30 , for several values of @xmath69 . we note , that the domain of existence of these compact ads boson stars decreases with decreasing @xmath69 . this suggests that there is a limiting minimal value for @xmath69 for compact boson stars . analogously to the ds case , some of the physical properties of ads boson stars exhibit damped oscillations with respect to @xmath30 whereas other properties exhibit spirals . moreover , as in the ds case , there are phantom boson stars , associated with negative values of @xmath30 . however , we do not find ads boson shells . in the ads case , the scalar field @xmath92 never reaches the value zero at the origin , necessary for boson shells to arise . we conclude , that the extra attraction associated with negative @xmath14 inhibits the formation of ads shells even stronger than in the asymptotically flat case , @xmath93 . the existence of shells needs repulsion , that can be provided either by a positive cosmological constant , as seen in the previous subsection , or by the presence of electric charge @xcite . here we consider the domain of solutions and their properties for various space - time dimensions . we have made a complete study for dimensions @xmath67 , 5 and 10 . since the dependence on @xmath1 is mostly rather smooth , we exhibit only a number of selected cases . as an example of a solution , we show an asymptotically de sitter boson shell solution in fig . [ fig6 ] . we start our discussion by considering boson stars and boson shells in @xmath95 dimensions . some of their properties are exhibited in fig . [ fig7 ] , and can be compared to those of fig . [ fig2 ] . a surprising feature is that in @xmath95 boson shells seem to exist in the asymptotically flat case , @xmath96 . this follows from the additional ( almost vertical ) line present in fig . [ fig7 ] for @xmath70 . however , since this line is reaching zero for a very small negative value of @xmath30 , those shells are phantom shells . the small branch of phantom shells is seen more clearly in the plot of @xmath97 in the inset . as @xmath83 , the size of these phantom shells diverges . for finite values of @xmath69 , however , we obtain also ordinary ds boson shells . the critical value @xmath78 of the transition between the boson stars and shells is seen in fig . in particular , the resulting phantom ds boson shells continue to exist beyond @xmath55 , where they smoothly turn into ordinary boson shells . as in @xmath56 , at fixed @xmath69 with increasing @xmath30 the outer radius @xmath41 of these boson shells decreases , tending to a finite limiting value , while their ratio @xmath74 of inner and outer radius increases towards one . the compact ds boson stars , on the other hand , exist only below a maximal value of @xmath30 , which is either given by the onset of the spiral or by the transition to ds boson shells . for a given @xmath69 , the domain of existence of ds boson stars in @xmath95 with respect to @xmath30 is smaller than in @xmath56 . when going to higher dimensions , this trend continues . in contrast , the domain of existence of phantom ds boson stars increases with increasing @xmath1 . in @xmath67 dimensions gravity is non - dynamic . therefore we may expect that boson stars may exhibit a different behavior in @xmath67 dimensions than in higher dimensions . let us first inspect the dependence of some of the properties of the compact 3-dimensional objects on @xmath30 for several values of @xmath69 . [ fig8 ] exhibits the scalar field @xmath92 at the origin , @xmath70 , the outer radius @xmath41 , and the scaled mass @xmath77 . here we observe , that indeed some properties of compact boson stars are very different in @xmath67 dimensions . first of all , we notice a maximal value of @xmath70 , that decreases with increasing @xmath69 . this maximal value is encoutered for negative values of @xmath30 , thus these configurations correspond to phantom boson stars . as this maximal value of @xmath70 two branches of solutions merge . along one of these branches @xmath70 reaches zero . thus boson shells emerge at the corresponding critical value @xmath98 . with increasing @xmath69 the critical value @xmath98 increases ( at least for positive @xmath98 ) , analogously to other dimensions . the second branch of boson star solutions , however , exhibits a different behaviour from the one observed before . clearly , the damped oscillations of @xmath70 with @xmath30 are not present . instead a monotonic decrease of @xmath70 with increasing @xmath30 is observed , and no maximal value of @xmath30 is encountered . we further observe that for small @xmath69 , @xmath70 and the outer radius @xmath41 depend only weakly on @xmath69 . the absence of a maximal value of @xmath30 suggests to consider the dependence of the compact boson stars on @xmath69 , choosing fixed large values of @xmath30 . as can be seen from figs . [ fig8]e and f , for a given @xmath30 boson star solutions exist only up to a maximal value of @xmath69 , that increases with @xmath30 . finally , we note that the solutions with @xmath68 are not asymptotically flat , since their mass parameter @xmath77 is finite . this holds for boson stars and boson shells , alike . let us turn finally to ads boson stars in @xmath1 dimensions . some properties of asymptotically ads boson stars in @xmath95 are exhibited in fig . [ fig9 ] . as in four dimensions , there are no ordinary ads boson shells in other than four dimensions . however , there is a very small region of phantom ads boson shells . this is seen by inspecting the critical curve @xmath78 for negative values of @xmath30 close to zero . here the critical curve passes negative values of @xmath69 as well . like the phantom shells for @xmath99 these phantom ads shells diverge in size as @xmath83 . otherwise all the basic properties of these ads solutions in @xmath95 are similar to those of the @xmath56 ads solutions discussed above . moreover , we observe only gradual changes with increasing @xmath1 . in particular , the domain of existence of ads boson stars with respect to @xmath30 decreases with @xmath1 . only the domain of phantom ads shells increases slightly , as seen in fig . [ fig3 ] . the lower dimensional case @xmath67 is special again , as can be seen in figs . [ fig8]e and f , where the dependence of @xmath70 and @xmath77 on @xmath69 is shown for several values of @xmath30 . interestingly , for a given @xmath30 there is no lower bound of @xmath69 encountered . moreover , the mass becomes practically independent of @xmath30 as @xmath69 becomes sufficiently small . when @xmath69 is fixed instead , while @xmath30 is varied , no spirals are encountered for the compact 3-dimensional boson stars , whereas spirals are present in all higher dimensions . this was observed before for ordinary boson stars @xcite . restricting to positive @xmath30 , there is a maximal value of @xmath70 for a given @xmath69 . however , when allowing for phantom fields , a minimal value of @xmath30 is encountered , beyond which @xmath70 increases without bound , while the mass decreases . we have studied compact boson stars and shells obtained with a v - shaped interaction potential in @xmath10 dimensions . the v - shaped potential confines the scalar field to a finite region , which can be ball - like or shell - like . in the probe limit , we have given the general analytical solution for @xmath0-balls in a minkowski background . here no @xmath0-shells exist . @xmath0-shells arise only beyond a critical value @xmath100 of the cosmological constant , which increases with the number of dimensions @xmath1 . likewise , @xmath0-balls exist only above a minimal value of the cosmological constant , which seems to correspond to @xmath76 . subsequently , we have taken the backreaction into account . the resulting configurations correspond to compact boson stars and boson shells . by solving the coupled set of einstein - scalar field equations , we have obtained the full set of solutions , subject to minkowski , de sitter and anti - de sitter asymptotics for a number of space - time dimensions , ranging from 3 to 10 . for any dimension @xmath101 there are compact boson stars with all three types of asymptotics . but concerning their properties , we see a distinct behaviour in three dimensions , that is different from the common behaviour encountered in all higher dimensions . in four and higher dimensions , these boson stars exist in a finite intervall @xmath102 . in constrast , in three dimensions there is no upper bound on the value of @xmath30 . also , all boson stars in four and higher dimensions exhibit a spiral - like dependence of the outer radius and the mass on the coupling constant @xmath30 . at the same time , the scalar field value @xmath103 exhibits damped oscillations . in constrast , in three dimensions the respective boson star properties do not exhibit such a spiral - like dependence or damped oscillations . by exploring the parameter space , we also find boson stars for negative values of @xmath30 . these boson stars with negative @xmath30 correspond to phantom boson stars , since the negative sign of @xmath30 can be reinterpreted as a negative sign associated with the scalar field in the lagrangian , and thus with a phantom field . ample motivation for the consideration of phantom fields is nowadays provided by cosmology . in constrast to boson stars , boson shells do not exist for minkowski asymptotics , if there is no additional force present , balancing the gravitational attraction . consequently , there are no ordinary ads boson shells . however , ds boson shells do exist . here the positive cosmological constant provides the necessary repulsion . on the other hand , there exist small regions of phantom ads shells in more than four dimensions . in four dimensions we have also considered astrophysical aspects of the compact boson stars and boson shells . while we can always adjust the parameters of the solutions to describe compact astrophysical objects with masses and sizes of neutron stars as discussed in @xcite , the influence of the cosmological constant on these objects is negligible , when the physical value of @xmath14 is taken . the new feature is , however , that in addition to boson stars there exist also boson shells for positive values of @xmath14 . we have then addressed the question , what the properties of such compact objects would be , if we set the scale by the physical value of @xmath14 . interestingly , in this case the resulting sets of boson stars reach huge masses and sizes , that are more akin to structures on the largest scales of the universe . the boson shells , on the other hand , can grow in mass until they reach the limit , set by the extremal schwarzschild - de sitter solution , for which the event horizon and the cosmological horizon merge . for negative @xmath14 one might be tempted to consider the ads / cft correspondence and try to interpret the solutions within this framework . however , all of our solutions are compact . the outer solutions are all given in terms of the schwarzschild - ads solutions . thus the ads boundary does not feel anything of the solutions except for their mass . consequently , for such compact solutions the concept of holography does not work . indeed , lots of different compact objects may sit in the bulk , and if they have the same mass , the boundary does not notice a difference . as our next step we plan to include rotation @xcite . rotating boson stars are known for non - compact configurations @xcite . interestingly , their angular momentum @xmath104 is quantized in terms of their particle number @xmath0 , @xmath105 , where @xmath106 is an integer . we expect , that the rotation of compact boson stars will lead to interesting new features . moreover , there may be rotating boson shells in the presence of a cosmological constant . it should also be interesting to construct interacting compact @xmath0-balls and @xmath0-shells for finite cosmological constant . these should arise in the presence of several complex scalar fields @xcite . we would like to thank eugen radu for helpful discussions . we gratefully acknowledge support by the dfg , in particular , also within the dfg research training group 1620 models of gravity . here we consider the ads solutions in the probe limit . we introduce the scaled coordiante @xmath107 , where @xmath108 , and the scaled scalar field @xmath109 . this yields for the ode of the function @xmath110 @xmath111 where @xmath112 , @xmath113 , and prime denotes the derivative with respect to @xmath114 . now the general solution of the inhomogeneous ode can be written as @xmath119 where @xmath120 and @xmath121 are constants and @xmath122 \label{inhsol}\ ] ] is a special solution of the inhomogeneous ode . let us assume that @xmath116 is regular at @xmath123 and @xmath124 . as a consequence the integral in eq . ( [ y2 ] ) diverges and @xmath125 is singular at @xmath123 . however , the special solution of the inhomogeneous ode is regular and vanishes at @xmath123 . thus to obtain the regular solution with @xmath126 we set @xmath127 and @xmath128 . next we consider the conditions for compact solutions , i.e. @xmath129 and @xmath130 at some @xmath131 . @xmath132 = 0 \ , \label{bc1}\\ \psi'(\eta_o ) & = & y_1'(\eta_o ) + \frac{\ell^2}{\hat{\phi}(0 ) } \left[y_2'(\eta_o)\int_0^{\eta_o}\tau^m y_1(\tau)d\tau - y_1'(\eta_0)\int_0^{\eta_o}\tau^m y_2(\tau)d\tau\right ] = 0 \ . \label{bc2}\end{aligned}\ ] ] the linear superpositions @xmath133 and @xmath134 yield @xmath135\int_0^{\eta_o}\tau^m y_1(\tau)d\tau \ , \label{bc1a}\\ 0 & = & \left[(y_1'(\eta_o)y_2(\eta_0)-y_2'(\eta_o)y_1(\eta_o)\right ] \left(1-\frac{\ell^2}{\hat{\phi}(0)}\int_0^{\eta_o}\tau^m y_2(\tau)d\tau\right ) \ , \label{bc2a}\end{aligned}\ ] ] respectively , where we set @xmath136 . since @xmath116 and @xmath125 are solutions of the homogeneous ode it follows that @xmath137 for some constant @xmath138 . as a consequence @xmath139 \neq 0 $ ] and the conditions ( [ bc1a ] ) and ( [ bc2a ] ) reduce to @xmath140 the first equation determines the point @xmath131 and the second the value of @xmath70 . now we are left with the problem to determine the range of @xmath141 for which solutions of eq . ( [ bc1b ] ) exist . although the solutions of the homogeneous ode can be expressed in terms of hypergeometric functions , we did not succeed to determine the minimal values of @xmath141 in the general case , i.e. in all dimensions . however , in @xmath56 dimensions the solution @xmath116 can be expressed in terms of trigonometric functions , @xmath142 \ , \label{homsol}\ ] ] which simplifies the problem considerably . here we consider only @xmath143 . the case @xmath144 needs special treatment . we will show that eq . ( [ bc1b ] ) has only a solution @xmath131 if @xmath145 . clearly , the integral in eq . ( [ bc1b ] ) can only vanish if the integrand possesses ( at least ) one zero . to analyze this condition we introduce a new coordinate @xmath146 , with @xmath147 , and consider the function @xmath148 since @xmath149 does not change sign on the interval @xmath147 it is sufficient to consider the function @xmath150 now we are left with the question for which values of @xmath141 the function @xmath151 does not possess a zero . let us first restrict to @xmath152 . expanding the function @xmath151 for small values of @xmath153 we find @xmath154 . on the other hand evaluating @xmath151 at @xmath155 yields @xmath156 for @xmath157 . thus , @xmath158 possesses at least one zero for @xmath157 . consequently , we can restrict to @xmath159 . we rewrite eq . ( [ eqf1 ] ) as @xmath160 \label{eqf2 } \\ & = & -\frac{\ell+1}{2}\left[b\cos a \sin a+ \sin(ba)\right ] \ , \label{eqf3}\end{aligned}\ ] ] where @xmath161 , @xmath162 are restricted to @xmath163 and @xmath164 . it can be seen from fig . [ fig11 ] that @xmath165 indeed does not a possess a zero in the region @xmath166 . consequently there are no compact solutions with @xmath167 . now we turn to the case @xmath168 . with @xmath169 and @xmath170 the function @xmath151 reads @xmath171 \ . \label{eqf4 } \\ & = & \frac{1}{2\ell'}\left[(\ell'-1)\sin(z'(\ell'+1))+(\ell'+1 ) \sin(z'(\ell'-1))\right ] \label{eqf5}\end{aligned}\ ] ] note that now @xmath172 . therefore , @xmath173 and @xmath174 imply @xmath175 and @xmath176 . this shows that the function @xmath151 does not have a zero for @xmath168 . to conclude , we have found a lower bound for @xmath141 , i.e. @xmath177 , below which no compact solution can exist . however , this does not prove that solutions exist for @xmath178 , since the condition eq . ( [ bc1b ] ) is more restrictive than @xmath179 for some @xmath114 . indeed , for @xmath180 eq . ( [ bc1b ] ) yields an equation of the form @xmath181 , which has no solution for @xmath182 . however , consider the case when the function @xmath158 possesses only one zero on the interval @xmath183 for some @xmath145 . denote the zero by @xmath184 . the condition ( [ bc1b ] ) can then be written as @xmath185 the first integral yields a finite negative contribution , since @xmath158 is bounded and negative on @xmath186 . the second integral on the other hand assumes any positive value between zero and infinity , as @xmath187 ranges between @xmath184 and @xmath188 , since @xmath158 is bounded and positive on @xmath189 . consequently , there exists a @xmath187 on @xmath189 for which both integrals in eq . ( [ intf ] ) cancel , implying that compact solutions exist for some @xmath190 . as an example we computed the exact solution for @xmath191 . we found @xmath192\right.\right . \nonumber\\ & & \left.\left . \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ -15(\eta^4 + 10\eta^2 - 15)\eta\log(1+\eta^2)\right)\right\ } \ .\end{aligned}\ ] ] for this solution the values @xmath193 and @xmath194 have been computed numerically . remarkably , they coincide with the corresponding values of our numerically computed solution of the ode up to eight digits . 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 ] ] . e. w. mielke and r. scherzer , phys . d * 24 * ( 1981 ) 2111 . r. friedberg , t. d. lee and a. sirlin , phys . d * 13 * , 2739 ( 1976 ) . s. r. coleman , nucl . b * 262 * , 263 ( 1985 ) [ erratum - ibid . b * 269 * , 744 ( 1986 ) ] . b. hartmann , b. kleihaus , j. kunz and i. schaffer , phys . b * 714 * , 120 ( 2012 ) [ arxiv:1205.0899 [ gr - qc ] ] . h. arodz and j. lis , phys . d * 77 * , 107702 ( 2008 ) . [ arxiv:0803.1566 [ hep - th ] ] . h. arodz and j. lis , phys . d * 79 * , 045002 ( 2009 ) . [ arxiv:0812.3284 [ hep - th ] ] . h. arodz , j. karkowski and z. swierczynski , acta phys . b * 43 * ( 2012 ) 79 [ arxiv:1201.2279 [ hep - th ] ] . b. kleihaus , j. kunz , c. lammerzahl and m. list , phys . b * 675 * , 102 ( 2009 ) . [ arxiv:0902.4799 [ gr - qc ] ] . b. kleihaus , j. kunz , c. lammerzahl and m. list , phys . rev . d * 82 * , 104050 ( 2010 ) . [ arxiv:1007.1630 [ gr - qc ] ] . f. e. schunck and a. r. liddle , lect . notes phys . * 514 * , 285 ( 1998 ) [ arxiv:0811.3764 [ astro - ph ] ] . j. m. maldacena , adv . * 2 * , 231 ( 1998 ) [ hep - th/9711200 ] . e. witten , adv . * 2 * , 253 ( 1998 ) [ hep - th/9802150 ] . k. sakamoto and k. shiraishi , jhep * 9807 * , 015 ( 1998 ) [ gr - qc/9804067 ] .
chern - simons quantum field theory in 3-dimensions ( perturbative , or non - perturbative ) produces a plethora of numerical invariants of knotted 3-dimensional objects . we introduce a packaging of these invariants into two power series : one that encodes non - perturbative invariants ( model np ) , and one that encodes perturbative invariants ( model p ) . the paper is concerned with the analytic continuation , the asymptotic behavior and and arithmetic properties of those power series . let us begin with a conjecture concerning the structure of nonperturbative quantum invariants . consider the generating series @xmath0 of the _ witten - reshetikhin - turaev _ invariants @xmath1 ( see section [ sec.cs ] ) of a closed , oriented , connected 3-manifold @xmath2 , using a compact lie group @xmath3 and a level @xmath4 . the power series is known to be convergent inside the unit disk @xmath5 since unitarity implies that @xmath1 grows at most polynomially with respect to @xmath6 ; see @xcite . ( analytic continuation ) for every pair @xmath7 as above , the series @xmath8 has analytic continuation as a multivalued function on @xmath9 , where @xmath10 is a finite set that contains zero and the exponentials of the negative of the critical values of the complexified chern - simons action . a key observation is that @xmath11 may contain elements inside the unit disk @xmath12 despite the fact that the power series @xmath8 is analytic for @xmath13 such that @xmath12 . one may compare this behavior with the power series @xmath14 that define the classical dilogarithm , whose analytic continuation is a multivalued analytic function in @xmath15 . schematically , the analytic continuation of @xmath8 may be depicted as follows : @xmath16 the above conjecture has the following features : * it can be formulated for pairs @xmath17 where @xmath18 denotes a knotted object , i.e. , a knot @xmath19 in 3-space or a closed 3-manifold @xmath2 and @xmath3 denotes a compact lie group . * it implies via elementary complex analysis two well - known asymptotic conjectures in quantum topology ; namely the volume conjecture ( in the case of knots ) , and the witten conjecture ( in the case of 3-manifolds ) . the complex analysis argument uses the cauchy formula to write @xmath1 as a contour integral of @xmath20 and then deform the contour around the singularities of the integrand nearest to the origin . for a detailed discussion , see ( * ? ? ? * sec.7 ) . * it states a precise relation between exact chern - simons theory and its perturbation expansion around a trivial ( or not ) flat connection . namely , perturbation theory is simply the expansion of the multivalued function @xmath8 around one of its singularities . * it explains the effect of complexifying a compact gauge group and to the partition function of the corresponding gauge theory . indeed , analytic continuation captures the critical values of the complexified action ; compare also with @xcite . * the conjecture can be extended to state - sum invariants of sum - product type that generalize @xmath1 and do not necessarily come from topology . * the conjecture has been proven for power series of 1-dimensional sum - product type , which includes the case of the @xmath21 and @xmath22 knots ; see @xcite and @xcite . * the conjecture can , and has been , numerically tested . see section [ sec.evidence ] . our next conjecture is a symmetry conjecture . recall that @xmath2 denotes an oriented 3-manifold ; we let @xmath23 denote the _ orientation reversed _ manifold . ( symmetry ) for every pair @xmath7 with @xmath2 an integer homology sphere , we have : @xmath24 has singularities at @xmath25 . let us make some comments regarding the above _ paradoxical statement _ : * the left ( resp . right ) hand side is given by a convergent power series for @xmath5 ( resp . @xmath26 ) . thus , the power series never make sense simultaneously , but their analytic continuations do . * zagier calls a similar statement in ( * ? ? ? * eqn.7 ) a _ strange identity _ since the the two sides never make sense simultaneously . our symmetry conjecture is closely related to a _ modular property _ , at least for the series studied by kontsevich - zagier ; see ( * ? ? ? * sec.6 ) . * in physics , equation is usually called a _ duality_. * in algebraic geometry and number theory , one may compare with the following symmetry for the polylogarithm : @xmath27 where @xmath28 is the @xmath29-th polylogarithm and @xmath30 is the @xmath29-th bernoulli polynomial ; see ( * ? ? ? * sec.1.3 ) . * in analysis one may use the above symmetry to deduce the asymptotic behavior ( and even more , the asymptotic expansion ) of @xmath8 for large @xmath31 . in particular , if @xmath32 , it follows that @xmath33 for large @xmath31 . * the above symmetry may be explained by the fact that cs changes sign under orientation reversal . since the level is nonnegative , the path integral formula for @xmath8 formally implies the above symmetry . * if we use the normalized invariants @xmath34 and if @xmath2 is an integer homology sphere , then it is possible that @xmath35 * when @xmath32 , it follows that the asymptotic expansion of @xmath36 around its singularities _ uniquely determines _ @xmath36 . indeed , the difference between two determinations is an entire function which is bounded by a constant by the symmetry conjecture . thus , the difference is identically zero . * if @xmath37 is _ amphicheiral _ ( for example , @xmath2 is given by a connected sum @xmath38 where @xmath39 is the orientation reversing involution ) , then equation predicts that @xmath40 . so far , we considered nonperturbative quantum invariants . let us now consider perturbative quantum invariants of pairs @xmath17 . they can be packaged into a power series @xmath41 which is convergent at @xmath42 . for a detailed definition , see section [ sec.cs ] . our next conjecture describes an explicit relation between the perturbative and nonperturbative quantum invariants . ( exact implies perturbative ) for every integer homology sphere @xmath2 we have : @xmath43 where @xmath44 is a holomorphic function of @xmath13 at @xmath42 and @xmath45 denotes the analytic continuation at @xmath46 along _ any _ path that avoids the singularities . as before , we can extend conjecture [ conj.2 ] to pairs @xmath47 , where @xmath19 is a knot in 3-space . in calle s terminology ( see @xcite and also ( * ? ? ? * sec.2.3 ) ) , if @xmath48 denotes the _ alien derivative _ in the direction @xmath13 , conjecture [ conj.2 ] states that : @xmath49 equation is reminiscent of the condition @xmath50 that defines the _ bloch group _ ; see for example @xcite . the series @xmath51 also satisfies a symmetry property : @xmath52 unlike the case of conjecture [ conj.symmetry ] , this is an easy corollary of its very definition . chern - simons theory with complex gauge group was studied extensively by gukov in @xcite . it is an interesting problem to compare forthcoming work of gukov - zagier on modularity properties of the quantum invariants with our conjectures . based on some partial results of @xcite and @xcite and stimulating conversations with o. costin , j. calle and d. zagier , more is actually expected to be true . namely , we expect arithmetic restrictions on the singularities of the series @xmath8 and of its monodromy , local and global . these restrictions lead us naturally to the notion of arithmetic resurgence , and the gevrey series of mixed type . in the rest of the paper , we will formulate these expected algebraic / arithmetic aspects of quantum invariants in a precise way and to expose the reader to the wonderful world of resurgence , introduced by calle in the eighties for unrelated reasons ; @xcite . the logical dependence of the sections is the following : @xmath53{\text{section \ref{sec.cs } } } \arrow{sw}\arrow{se } \node[2]{\text{section \ref{sec.ar } } } \arrow{sw}\arrow{se } \\ \node{\text{section \ref{sec.habiro } } } \arrow{see } \node[2]{\text{section \ref{sec.conj } } } \arrow{s } \node[2]{\text{section \ref{sec.sp } } } \arrow{sww } \\ \node[3]{\text{section \ref{sec.evidence } } } \end{diagram}\ ] ] in this section @xmath2 will denote a closed 3-manifold and @xmath3 will denote a simple , compact , simply connected group @xmath3 . for example , @xmath54 . the witten - reshetikhin - turaev invariant is a map : @xmath55 for a definition of @xmath1 see @xcite . formally , for @xmath4 , @xmath1 is the expectation value of a _ path integral _ with a topological _ chern - simons lagrangian _ at level @xmath6 ; see @xcite . since the chern - simons lagrangian takes values in @xmath56 , it follows that the level @xmath6 has to be an integer number , which without loss we take it to be nonnegative . we can convert the sequence @xmath57 into a generating series as follows : for every @xmath2 and @xmath3 as above , we define : @xmath58 unitarity of the chern - simons theory implies that for every @xmath59 the sequence @xmath57 grows polynomially with respect to @xmath6 . in other words , it was shown in @xcite that there exists positive constant @xmath60 and @xmath61 ( that depend on @xmath2 and @xmath3 ) so that @xmath62 for all @xmath4 . thus , @xmath8 is analytic inside the unit disk @xmath63 . the path integral interpretation of @xmath1 formally leads to a perturbation theory along a distinguished critical point of the chern - simons action , namely the trivial flat connection . this gives rise to a graph - valued power series invariant , which has been defined by le - murakami - ohtsuki in @xcite . additional definitions of this powerful invariant were given by kuperberg - thurston ; see @xcite . more precisely , lmo define a graph - valued invariant @xmath64 where @xmath65 is a completed vector space of jacobi diagrams . a _ jacobi _ diagram of degree @xmath6 is a trivalent graph with @xmath66 oriented vertices , considered modulo the as and ihx relations ; see @xcite . jacobi diagrams are diagrammatic analogues of tensors on a lie algebra with an invariant inner product . indeed , given a simple lie algebra @xmath67 , there is a _ weight system map _ that replaces a jacobi diagram of degree @xmath6 by a rational number times @xmath68 : @xmath69\ ] ] see @xcite . recall the _ borel transform _ : @xmath70 \longto \bq[[z ] ] , \qquad \calb\left(\sum_{n=0}^\infty \frac{a_n}{x^n } \right ) = \sum_{n=0}^\infty \frac{a_{n+1}}{n ! } z^n\ ] ] let @xmath71 denote the borel transform of @xmath72 . in @xcite it was proven that if @xmath2 is a homology sphere and @xmath67 is a simple lie algebra , then the formal power series @xmath72 is gevrey-1 . in other words , @xmath71 is an analytic function in a neighborhood of @xmath42 . our main resurgence conjecture [ conj.1 ] formulated in section [ sec.conj ] below links the singularities of the analytic continuation of the series @xmath73 and @xmath51 to some classical geometric invariants of 3-manifolds , namely the critical values of the complexified chern - simons function . let us recall those briefly , and refer the reader to @xcite for a more detailed discussion . let us fix a closed 3-manifold @xmath2 , and simple , compact simply connected group @xmath3 , and a trivial bundle @xmath74 with the trivial connection @xmath75 . let @xmath76 denote the set of @xmath3-connections on @xmath74 . there is a chern - simons map : @xmath77 where , as common in algebraic geometry , we denote @xmath78 even though @xmath76 is an affine infinite dimensional vector space acted on by an infinite dimensional gauge group , the set @xmath79 of gauge equivalence classes of the critical points of @xmath80 is a compact semialgebraic set that consists of flat @xmath3-connections . up to gauge equivalence , the latter are determined by their monodromy . in other words , we may identify : @xmath81 this gives rise to a map : @xmath82 stokes s theorem implies that @xmath80 is a locally constant map . since @xmath79 is a compact set , @xmath80 takes finitely many values in @xmath83 . let us now _ complexify _ the action ; see also @xcite . this means that we replace the compact lie group @xmath3 by its complexification @xmath84 , the moduli space @xmath79 by @xmath85 , and the chern - simons action @xmath80 by @xmath86 : @xmath87 @xmath86 is again a locally constant map , and takes finitely many values in @xmath88 . thus , we may define the following geometric invariants of 3-manifolds . for @xmath2 and @xmath3 as above , we define @xmath89 under complex conjugation ( but keeping the orientation of the ambient manifold fixed ) , we have @xmath90 . it follows that @xmath91 ( resp.@xmath92 ) is invariant under @xmath93 ( resp . @xmath94 ) . the involution @xmath94 preserves the set of rays through zero . on the other hand , under orientation reversal , we have @xmath95 and @xmath96 , where @xmath97 for @xmath98 and @xmath99 . we thank c. zickert for help in identifying those involutions . thus , a typical picture for @xmath91 and @xmath100 is the following : @xmath101 where the horizontal spacing between two dots in any horizontal line is @xmath102 . complexification is a key idea , theoretically , as well as computationally . for example , @xmath85 is an algebraic variety whereas its real part @xmath103 is only a compact set with little structure . the only systematic way ( known to us ) to give exact formulas for the critical values of @xmath80 is to actually compute the critical values of @xmath86 and then decide which of these are critical values of @xmath80 . for @xmath54 , there are exact and numerical computer implementations for the critical values of @xmath86 : see snap @xcite and @xcite . complexification also reveals the arithmetic structure of @xmath104 : its elements are _ periods _ of weight @xmath105 ( in the sense of kontsevich - zagier @xcite ) , of a rather special kind . namely , the critical values of @xmath86 are @xmath106-linear combinations of the _ rogers dilogarithm _ function evaluated at algebraic numbers . the latter is defined by : @xmath107 for @xmath108 and analytically continued as a multivalued analytic function in @xmath15 . here , @xmath109 is the classical dilogarithm function . for @xmath54 , our identification of the complexified chern - simons action with @xcite is as follows : @xmath110 for higher rank groups , exact formulas for the critical values of @xmath86 may also be given in terms of the rogers dilogarithm function at algebraic numbers . this will be explained in detail in a separate publication . as an illustration of the above discussion let us give an example . if @xmath2 is obtained by @xmath111 surgery on the @xmath22 knot , then @xmath112 consists of @xmath113 points plotted as follows : @xmath114 in this picture , a higher resolution reveals that the points nearest to the vertical axis consist of _ two _ distinct but close pairs . we thank c. zickert for providing us with an exact and numerical computation of the critical values of the complexified chern - simons map . let us end this section with a problem : give an direct relation between the cubic polynomial action and the rogers dilogarithm . a transcendental relation between the chern - simons action and the rogers dilogarithm was given in ( * ? ? ? * sec.6.2 ) , using the third algebraic @xmath19-theory group @xmath115 . so far , the discussion involved closed 3-manifolds . let us now consider knots @xmath19 in the 3-sphere . for simplicity , we will assume that @xmath54 ( so @xmath116 ) in this section . let us fix a knot @xmath19 in 3-sphere and a nonnegative integer @xmath6 . let us denote by @xmath117 the _ kashaev invariant _ of @xmath19 ( see @xcite ) , which is also identified by @xcite with the value of the jones polynomial colored by the @xmath6-th dimensional irreducible representation of @xmath118 evaluated at @xmath119 ( and normalized to be @xmath120 at the unknot ) . thus , we may define : @xmath121 we define the perturbative invariant @xmath122 as follows : * take the sequence @xmath123 $ ] of the jones polynomials of @xmath6 , colored by the @xmath6-dimensional irreducible representation of @xmath118 , and normalized by @xmath124 . see for example , @xcite . * it turns out that there exists a power series @xmath125 $ ] so that @xmath126 $ ] for all @xmath4 ; see for example @xcite . * consider the power series @xmath127 $ ] . * define @xmath128 $ ] . in @xcite ( resp . @xcite ) it was shown that @xmath129 ( resp . @xmath122 ) is analytic for @xmath13 in a neighborhood of @xmath130 . regarding the critical values of the chern - simons action , we will consider only _ parabolic _ @xmath131 representations ; i.e. , those representations so that the trace of every peripheral element is @xmath132 . as in the case of closed 3-manifolds , we may identify the moduli space of _ parabolic _ flat @xmath131-connections on the knot complement with @xmath133 : @xmath134 in addition , we have a map , described in detail in @xcite : @xmath135 @xmath86 is again a locally constant map , and takes finitely many values in @xmath88 , and allows us to define the sets @xmath136 and @xmath137 . let us end this section with an example of the simplest knot @xmath21 and the simplest hyperbolic knot @xmath22 . if @xmath138 is the right hand trefoil knot @xmath21 , then @xmath139 can be plotted as follows : @xmath140 if @xmath141 is the simplest hyperbolic knot , then @xmath142 can be plotted as follows : @xmath143 where @xmath144 is the _ volume _ of @xmath22 ; see @xcite , and numerically , @xmath145 the arithmetic nature of @xmath104 is only the beginning . it turns out that * the _ ray - singer torsion invariants _ associated to a @xmath84-representation of @xmath146 are also algebraic numbers ( this is proven and discussed in detail in @xcite ) , * in case @xmath2 is hyperbolic and @xmath54 the geometric representation is defined over a number field ; see @xcite , * the perturbative expansions of the quantum invariants @xmath1 around an @xmath84-representation of @xmath146 are conjectured to be algebraic numbers ; see for example @xcite and @xcite . the need to formulate these algebricity properties in a uniform way , as well as some results in some key cases , lead us to the notion of an arithmetic resurgent series , which is the focus of this section . along the way , we will also discuss the auxiliary notion of a gevrey series of mixed type , perhaps of interest on its own . _ resurgence _ was coined by calle in his study of analytic continuation of formal and actual solutions of differential equations , linear or not ; see @xcite . an earlier term used by calle was the notion of _ endless analytic continuation_. the concept of resurgence has influenced our thinking deeply . unfortunately , it is not easy to find an accepted definition of resurgence , or a reference for it in the literature . on the other hand , there are several expositions of instances of resurgence , covering special cases of this rather general notion . the curious reader may consult @xcite for a detailed discussion in addition to the original work @xcite . given the gap in the literature , we will do our best to give a working ( and exact ) definition of resurgence , which features some properties which are arithmetically important , and analytically rare . let us begin by recalling the monodromy of multivalued germs of interest in this paper . we refer the reader to @xcite for further details . the next definition is motivated by the types of singularities that appear in algebraic geometry ; see @xcite . a _ multivalued analytic germ _ @xmath147 at @xmath42 is called _ quasi - unipotent _ if its monodromy @xmath148 around @xmath130 satisfies the condition : @xmath149 for some nonzero natural numbers @xmath150 and @xmath151 . it is easy to see that a quasi - unipotent germ @xmath147 can be written as a finite sum of germs of the form : @xmath152 where @xmath153 , @xmath154 , and @xmath155 , where @xmath156 is the ring of power series convergent at @xmath42 ( identified with the ring of germs of functions analytic at @xmath42 ) . see for example , @xcite . series of the form are often known in the literature as series of the _ nilsson class _ ; see @xcite . the rationality of the exponents @xmath157 above is an important feature that always appears in algebraic geometry and arithmetic and rarely appears in analysis . for a further discussion ; see @xcite . we say that @xmath158 is an _ resurgent series _ ( and write @xmath159 ) if * @xmath160 is convergent at @xmath42 . * @xmath160 has analytic continuation as a multivalued function in @xmath161 , where @xmath162 is a discrete subset of @xmath163 . * the local monodromy is quasi - unipotent . in what follows , we will make little distinction between a germ , its analytic continuation , and the corresponding function . so , we will speak about the algebra of resurgent functions . the following definition is motivated by the properties of some power series that are associated to knotted 3-dimensional objects . we say that series @xmath158 is a _ gevrey series of mixed type _ @xmath164 if * @xmath165 and the coefficients @xmath166 lie in a number field @xmath19 , and * there exists a constant @xmath167 so that for every @xmath4 the absolute value of every galois conjugate of @xmath166 is less than or equal to @xmath168 , and * the common denominator of @xmath169 is less than or equal to @xmath170 . if @xmath160 is gevrey of mixed type @xmath164 and @xmath171 , @xmath172 , then @xmath160 is also gevrey of mixed type @xmath173 . let @xmath174 ( resp . @xmath175 ) denote the @xmath176$]-algebra of gevrey series of mixed type ( resp . mixed type @xmath164 ) . the gevrey series of mixed type @xmath177 are precisely the important class of _ arithmetic gevrey series of type _ @xmath150 , introduced and studied by andr ; see @xcite . a @xmath3-_function _ @xmath160 in the sense of siegel is a gevrey series of mixed type @xmath178 which is _ holonomic _ , i.e. , it satisfies a linear ode with coefficients in @xmath176 $ ] ; see @xcite . restricting the functions @xmath179 in to be gevrey series of type @xmath180 , we arrive at the notion of an arithmetic quasi - unipotent germ . we say that a multivalued analytic germ @xmath147 at @xmath42 is _ arithmetic quasi - unipotent _ if it can be written as a finite sum of germs of the form where @xmath181 . combining this definition with the notion of a resurgent function , we arrive at the notion of an arithmetic resurgent function . let @xmath182 denote the countable set of _ periods _ in the sense of kontsevich - zagier ; see @xcite . we say that @xmath158 is an _ arithmetic resurgent series _ ( and write @xmath183 ) if * @xmath160 is a resurgent series . * the singularities @xmath184 of @xmath160 is a discrete subset of @xmath185 , where @xmath185 denotes the set of _ periods _ as defined by kontsevich - zagier ; see @xcite . * the local monodromy is arithmetic quasi - unipotent . * the global monodromy is defined over @xmath186 . in a separate publication we will give the proof of the following proposition which shows that the coefficients of the taylor series at the origin of an arithmetic resurgent function have asymptotic expansions themselves . the transseries conclusion of the proposition below ( without any claims on the mixed gevrey type ) follows from the resurgence hypothesis on @xmath160 alone , and are studied systematically in the upcoming book of costin @xcite , as well as in @xcite . if @xmath158 is arithmetic resurgent , then @xmath187 where the sum is over the finite set of singularities of @xmath160 nearest to the origin , and @xmath188 is a finite sum of series of the form where @xmath189 are gevrey of mixed type @xmath190 ( for some @xmath151 ) . obvious arithmetic invariants of an arithmetic resurgent function @xmath160 are : * the set of singularities @xmath191 . * the local quasi - unipotent monodromy , and its field of definition . * the global monodromy , defined over @xmath186 . this section is logically independent of the rest of the paper and can be skipped at first reading , although it provides some useful examples of arithmetic resurgent functions . a main example of arithmetic resurgent functions comes from the following theorem of andr . @xmath3-functions are arithmetic resurgent with singularities a finite set of algebraic numbers . @xmath3-functions arise naturally in three contexts : * from geometry , related to the regularity of the gauss - manin connection . * from arithmetic . * from enumerative combinatorics . for a geometric construction of resurgent functions , let us recall the following result from @xcite ; see also @xcite . let @xmath192 be a projective non - singular connected curve and @xmath193 the complement of a finite set of points . suppose that @xmath194 is a proper and smooth morphism . for every @xmath195 , the algebraic de rham cohomology @xmath196 is an algebraic differential equation on @xmath197 , and the local system @xmath198 ( for @xmath199 ) is the local system of germs of solutions of that equation . let @xmath148 denote the local monodromy around a point @xmath200 . then , we have the following theorem . @xcite the algebraic differential equation is regular singular and the local monodromy @xmath148 is quasi - unipotent . the @xmath3-functions obtained by theorems [ thm.katz ] and [ thm.andre ] are closely related . the main conjecture is that all @xmath3-functions come from geometry . for a discussion of this topic , and for a precise formulation of the bombieri - dwork conjecture , see the survey papers of @xcite and also @xcite . let us discuss a third source of resurgent functions , which was discovered recently by the author in @xcite . a _ hypergeometric term _ @xmath201 ( in short , _ _ ) in variables @xmath202 where @xmath203 is an expression of the form : @xmath204 where @xmath205 for @xmath206 , @xmath207 for @xmath208 , and @xmath209 are integral linear forms in the variables @xmath202 . we assume that for every @xmath4 , the set @xmath210 is finite . we will call a _ balanced _ if in addition it satisfies the _ balance condition _ : @xmath211 given a @xmath212 , consider the corresponding sequence @xmath213 defined by @xmath214 where the summation index lies in the finite set , and the corresponding generating series : @xmath215.\ ] ] we will call sequences of the form @xmath213 _ balanced multisum sequences_. @xcite for every @xmath212 , the generating series @xmath216 is a @xmath3-function . let us point out that the proof of theorem [ thm.ga ] in general offers no help of locating the singularities of the function @xmath216 . to fill this gap , the author developed an efficient ansatz for the location of the singularities of @xmath216 ; see @xcite . a _ knotted object _ @xmath18 denotes either a closed 3-manifold @xmath2 or a knot @xmath19 in 3-space . a _ pair _ @xmath17 denotes either a closed 3-manifold @xmath2 and a compact lie group @xmath3 , or a knot @xmath217 in 3-space and @xmath54 . if @xmath54 and @xmath2 or @xmath19 is hyperbolic , let @xmath218 denote the critical values of the galois conjugates @xmath219 of the geometric @xmath131-representation . we now have all the ingredients to formulate our arithmetic resurgence conjecture , which is a refinement of the analytic continuation conjecture [ conj.0 ] . ( arithmetic resurgence ) for every pair @xmath17 , @xmath220 and @xmath51 are arithmetic resurgent with possible singularities in the set @xmath221 . if @xmath18 is hyperbolic and @xmath54 , then the singularities of the above functions include @xmath222 . conjecture [ conj.1 ] implies the following corollaries . for every pair @xmath17 , the power series @xmath73 has analytic continuation as a multivalued function on the complement @xmath223 of the finite set @xmath224 . if @xmath2 is a closed hyperbolic 3-manifold , then the witten - reshetikhin - turaev invariants determine the volume of @xmath2 by : r @xmath225 this follows from the fact that @xmath226 has a singularity at @xmath227 and @xmath228 where @xmath229 is a discrete faithful representation . witten s conjecture ( formulated in @xcite ) regarding the asymptotic expansion of the witten - reshetikhin - turaev invariants holds . for every hyperbolic knot @xmath19 in 3-space , the kashaev invariants determine the volume of @xmath19 by : @xmath230 moreover , there is an asymptotic expansion of the kashaev invariants in powers of @xmath231 using proposition [ prop.expar ] . in this section we we describe an arithmetic relation , due to habiro , between the perturbative @xmath51 and the non - perturbative @xmath73 invariants of knotted objects . this section is independent of our conjecture . however , habiro s results * are a good complement of our conjecture , * are important and interesting on their own right , * point to a different arithmetic origin for the invariants of knotted objects . this point of view has been studied by gukov - zagier @xcite . in this section , a knotted object @xmath18 denotes either a homology sphere @xmath2 or a knot @xmath19 . for simplicity , we will assume that @xmath54 . a theorem of habiro implies that @xmath232 determines @xmath233 and vice - versa ; @xcite . let us explain more about habiro s key results . in @xcite habiro introduces the ring @xmath234}=\lim_{\leftarrow n } \bz[q^{\pm}]/((q)_n)\ ] ] where @xmath235 is the quantum @xmath6-factorial defined by : @xmath236 with @xmath237 . in a sense , one may think of elements of the habiro ring as complex - valued _ analytic functions _ with domain @xmath238 , the set of complex roots of unity . this way of thinking is motivated by the following features of the habiro ring , shown in @xcite : * it is easy to see that every element @xmath239}$ ] can be written ( nonuniquely ) in the form : @xmath240 , \quad \text{for } \quad n \in \bn.\ ] ] note however that the above form is not unique , since for example : @xmath241 nevertheless the form can be used to generate easily elements of the habiro ring . * elements of the habiro ring can be evaluated at complex roots of unity . in other words , there is a map : @xmath242 } \longto \bc^{\om } , \qquad f(q ) \longto \left ( f : \omega \longto \bc , \qquad \w \mapsto f(\w)\right).\ ] ] in particular , we can associate a map : @xmath243 } \longto \bc[[z ] ] , \qquad f(q ) \longto \lnp_f(z)=1+\sum_{n=1}^\infty f(e^{2 \pi i / n } ) z^n.\ ] ] * elements of the habiro ring have taylor series expansions around @xmath244 ( and also around every complex root of unity ) . in other words , we can define a map : @xmath245 } \longto \bq[[z ] ] , \qquad f(q ) \longto \lp_f(z)=\calb ( f(e^{1/x})).\ ] ] * as in the case of analytic functions , the maps and are 1 - 1 . thus , @xmath246 determines @xmath247 and vice - versa . however , we need all the coefficients of the power series @xmath246 to determine a single ( eg . the third ) coefficient of @xmath247 . * given a homology sphere @xmath2 , there exists an element @xmath248}$ ] such that its image under the maps and coincide with the non - perturbative and perturbative invariants of @xmath2 discussed in section [ sub.npinvariants ] and [ sub.pinvariants ] . this was a main motivation for habiro , and was extended to knots in 3-space by huynh - le in @xcite . one may ask for an extension of conjecture [ conj.1 ] for the series @xmath246 and @xmath247 that come from the habiro ring . unfortunately , the habiro ring is uncountable ( whereas all quantum invariants of knotted objects lie in a countable subring ) and it has little structure as such . thus , it is unlikely that the series @xmath246 associated to a random sequence of laurent polynomials @xmath249 ( as in ) will be resurgent . concretely , we can pose the following problem with overwhelming numerical evidence : show that @xmath246 is not resurgent when @xmath250 in @xcite , le and the author introduced a countable subring @xmath251}^{\mathrm{hol}}$ ] that consists of elements of the form : @xmath252 , \quad ( f_n(q ) ) \quad \text{is $ q$-holonomic}\ ] ] where @xmath253-holonomic means that @xmath249 satisfies a linear @xmath253-difference equation with coefficients in @xmath254 $ ] ; see @xcite . in @xcite it was shown that the elements @xmath255 and @xmath256 of the habiro ring actually lie in the countable subring @xmath251}^{\mathrm{hol}}$ ] . show that for every @xmath257}^{\mathrm{hol}}$ ] , the series @xmath246 and @xmath246 are arithmetic resurgent . in the next section , we will discuss formulate a resurgence conjecture for some special elements of the habiro ring that do not always come from topology . conjecture [ conj.1 ] and problem [ prob.hab ] ask for proving that certain power series are arithmetic resurgent . however , they do not explain the source of resurgence . usually , resurgence is associated with a differential equation , linear or not ; see for example @xcite and also @xcite . in this section we will give another construction of powers series @xmath258 and @xmath259 which aims to explain the origin of arithmetic resurgence . this section was motivated by conversations with o. costin and j. calle whom we thank for their generous sharing of their ideas . let us first introduce the notion of _ series of sum - product type_. consider function @xmath260 analytic in @xmath261 $ ] with @xmath262 and the corresponding sequence and series of sum - product type : @xmath263 and the corresponding series @xmath264.\ ] ] since @xmath262 , it follows that the formal power series @xmath265\ ] ] is also well - defined . let @xmath259 denote the borel transform : @xmath266.\ ] ] let us now give a flavor of some results from @xcite and @xcite . in the rest of the section , let us consider @xmath260 of the following trigonometric type : @xmath267 where @xmath268 where @xmath269 , @xmath270 , @xmath271 for all @xmath150 , and @xmath272 for all but finitely many @xmath150 . in @xcite we construct elements of the extended bloch group , given by solutions of the algebraic equations : @xmath273 the values of these elements under the rogers dilogarithm defines a set @xmath274 , and its exponentiated cousin @xmath275 . @xcite @xmath258 and @xmath259 are arithmetic resurgent with singularities included in @xmath276 . notice that @xmath277 and @xmath278 where @xmath279 is an element of the countable habiro subring @xmath251}^{\mathrm{hol}}$ ] , where @xmath280 thus , theorem [ thm.cg ] is a special case of problem [ prob.habiro ] . equations appear in the _ dilogarithm ladders _ of lewin and others , whose aim is to produce interesting elements of algebraic @xmath19-theory . for a detailed discussion , see @xcite and @xcite . for the simplest knot @xmath21 and the simplest hyperbolic knot @xmath22 , we have : @xmath281 ( and likewise , equality for the @xmath282-series ) , where @xmath283 are both covered by theorem [ thm.cg ] . the resurgence conclusion of theorem [ thm.cg ] is valid for very general entire functions @xmath260 , with some mild hypothesis . for a detailed discussion , see @xcite and @xcite let us summarize what is known about conjecture [ conj.1 ] . conjecture [ conj.1 ] is known * for all 3-manifolds @xmath2 of the form @xmath284 where @xmath285 is a closed surface and all compact groups @xmath3 . indeed , this follows from the fact that @xmath1 is a polynomial in @xmath6 . thus , @xmath8 is a rational function of @xmath13 with denominator a power of @xmath286 . on the other hand , @xmath287 . * for @xmath288 where @xmath2 is the _ poincare homology sphere _ , or small seifert fibered 3-manifolds , see @xcite . in this lucky case , one uses explicit formulas for the coefficients of @xmath289 given zagier ( see @xcite ) which allow one show resurgence relatively easily . * for the simplest knot @xmath21 ( and also for @xmath290 torus knots ) ; and for the simplest hyperbolic knot @xmath22 ; see @xcite and @xcite . see remark [ rem.sp3 ] . our sample calculations below show the importance of the fractional polylogarithms and their analytic continuation , studied in detail in @xcite . let us confirm conjecture [ conj.1 ] for @xmath291 . for simplicity , we will choose @xmath54 . the case of other lie groups is similar . the witten - reshetikhin - turaev invariant is given by ( * ? ? ? * eqn.2.26 ) : @xmath292 expanding the above as a convergent power series in @xmath231 : @xmath293 and using the _ fractional polylogarithm _ @xmath294 defined for @xmath295 and @xmath5 by the convergent series : @xmath296 it follows that @xmath297 since @xmath298 has analytic continuation as a multivalued function in @xmath15 ( see @xcite ) , it follows that @xmath299 has analytic continuation on @xmath15 . we can further compute the monodromy around @xmath42 and @xmath300 using @xcite . let us confirm conjecture [ conj.1 ] for 3-manifolds of the form @xmath302 . for simplicity , we will choose @xmath54 . the case of other lie groups is similar . the witten - reshetikhin - turaev invariant is given by the famous _ verlinde formula _ @xcite : @xmath303 altough not a priori obvious , it it true that the right hand side of the above expression is a polynomial in @xmath6 of degree @xmath304 . in fact , we have ( * ? ? ? * sec.3 ) : @xmath305 when @xmath306 . for example , we have , @xmath307 it follows that @xmath308 is a rational function with denominator a power of @xmath309 . for example , we have : @xmath310 since @xmath311 is connected for all @xmath312 and @xmath3 , it follows that @xmath313 confirming conjecture [ conj.1 ] . let us give some details about how the work of @xcite and @xcite verify conjecture [ conj.1 ] for the simplest @xmath21 knot . an independent verification of the conjecture , valid for series of sum - product type , can be obtained by @xcite . equation ( 36 ) of @xcite and @xcite imply that we can write : @xmath314 where @xmath315 and @xmath160 is a multivalued analytic function ( analytic at @xmath42 ) : @xmath316 where @xmath317 denotes the _ unique primitive character of conductor _ @xmath318 : @xmath319 together with proposition [ prop.expar ] , this implies that the singularities of @xmath320 are @xmath321 . moreover , the local expansion of @xmath320 around @xmath322 is given by : @xmath323 where @xmath44 is a function holomorphic at @xmath42 . since @xmath324 is multivalued analytic at @xmath15 ( see @xcite ) , this confirms conjecture [ conj.1 ] for @xmath21 . using a mittag - leffler type decomposition for the fractional polylogarithm from ( * ? ? ? * eqn.13 ) , we can also verify the symmetry conjecture for @xmath21 . conjecture [ conj.1 ] gives an exact formula for the singularity of @xmath122 which is nearest to the origin . notice that this singularity does not in general coincide with the critical value of @xmath86 corresponding to the discrete faithful representation . this was indeed observed numerically for several _ twist knots_. let @xmath325 denote the twist knot with negative clasp and @xmath326 full twists , where @xmath327 : @xmath328 in particular , we have : @xmath329 the invariant trace field of @xmath325 is of type @xmath330 $ ] for @xmath331 and @xmath332 $ ] for @xmath333 . it follows that @xmath334 is a subset of a union of @xmath335 ( resp . @xmath336 ) horizontal lines , symmetric with respect to the @xmath13-axis , and a superposition of @xmath105 ( resp . @xmath120 ) copies of the @xmath13-axis for @xmath331 ( resp . @xmath333 ) . this set can be computed exactly and numerically using the methods of @xcite . the corresponding element of the habiro ring is given by : @xmath337 where @xmath338 $ ] denote the @xmath6-cyclotomic polynomial of @xmath325 . the latter may be computed inductively with respect to @xmath6 for each fixed @xmath326 , see @xcite . using this formula , one can compute 500 coefficients of @xmath339 , and then numerically compute the singularity of the series @xmath340 nearest to the origin . the numerical method used was the following : * fix a truncated power series @xmath341 where @xmath342 is a sufficiently large integer ( eg @xmath343 ) . * using the root test , one can compute approximately the radius of convergence @xmath344 of the series @xmath345 * plot @xmath346 for @xmath150 near the inverse of the radius of convergence . the plot reveals a blow - up at certain values @xmath347 of @xmath348 . * this suggests singularities at @xmath349 , and an asymptotic expansion of @xmath166 with a term of the form : @xmath350 in general , we have a finite sum ( over @xmath347 ) of terms of the above form . * one can numerically compute the constants @xmath351 and @xmath352 by fitting data . a fitting method ( also used by zagier in @xcite ) can improve the rate of convergence to @xmath353 for any @xmath75 . we used @xmath354 . this was done for the twist knots of equation . if @xmath355 denote the inverse of the radius of convergence of the series @xmath340 then we obtain numerically that : @xmath356 these numbers agree with the absolute value of @xmath357 for @xmath219 some galois conjugate of the discrete faithful representation . we thank n. dunfield , s. shumakovitch and c. zickert for their help in the numerical computations . additional numerical evidence for @xmath358 and @xmath359 ( and for many series of 1-dimensional sum - product type ) was obtained by @xcite . an early version of these conjectures was delivered by lectures in columbia university , cuny , universit paris vii , and orsay in the fall of 2006 and in aarhus , baton - rouge , hanoi and strasbourg in the summer of 2007 . the author wishes to thank the organizers of the conferences for their hospitality , and also n. a campo , y. andr , n. dunfield , m. kontsevich , w. neumann , t. pantev , d. sullivan , d. thurston , d. zeilberger and c. zickert . above all , the author expresses his gratitude to o. costin , j. calle and d. zagier for many enlightening conversations . in this section we will give a formal proof of the relation among @xmath258 and @xmath259 for series of sum - product type . we will use the notation from section [ sec.sp ] . suppose that @xmath360 for all @xmath361 $ ] . then , it is easy to see that @xmath258 is convergent inside the unit disk @xmath5 . the next theorem describes an explicit relation between @xmath258 and @xmath259 . we have : @xmath362 where @xmath44 is analytic at @xmath42 . we will give only the formal calculation , leaving the analytic details to the reader . below , @xmath44 will denote a germ of an analytic function at @xmath42 . for @xmath4 , let @xmath363 denote the coefficient of @xmath364 in @xmath365 given in . let us fix @xmath366 and consider @xmath6 large enough . then , we have : @xmath367 thus , @xmath368 ignore the @xmath369 terms , and interchange summation and integration . we obtain that @xmath370 where @xmath371 is the @xmath29-polylogarithm . the latter is a multivalued analytic function on @xmath15 with an asymptotic expansion at @xmath300 of the form : @xmath372 where @xmath44 is an analytic function at @xmath42 ; see for example , ( * ? ? ? * eqn.6 ) . thus , @xmath373 letting @xmath374 , and replacing @xmath13 by @xmath375 it follows that : @xmath376 this concludes the formal calculation . in this section we will give a path integral formula for @xmath8 using as input the famous chern - simons path integral studied in the seminal paper of witten ; see @xcite . with the notation of @xcite and with our normalization we have : @xmath377 where @xmath76 is the affine space @xmath76 of @xmath3-connections on the trivial bundle @xmath74 over @xmath2 . since @xmath80 takes values in @xmath88 , we require that the level @xmath6 ( which plays the role of the inverse planck s constant ) be integer . without loss of generality , we assume that @xmath4 . formally separating the @xmath378 contribution in and interchanging summation and integration in , it follows that the above formula is an infinite dimensional analogue of a _ riemann - hilbert problem _ , and was obtained during a conversation with kontsevich in the fall of 2006 . for a detailed discussion on the riemann - hilbert problem , see @xcite . finite dimensional analogues of the riemann - hilbert problem are discussed in @xcite .
self - propelled particles convert chemical @xcite , magnetic @xcite , or radiation energy @xcite into directed motion . in this way , kinetic energy is intrinsically introduced into the corresponding system on the single particle level and length scale , in contrast to a process of macroscopic stirring or shaking from outside . far - reaching consequences on the single particle behavior as well as on their collective properties emerge . single artificially generated janus particles can start to self - propel mostly due to phoretic effects when one of the two sides is selectively heated by laser light @xcite or catalyzes a chemical reaction @xcite . in a liquid environment , such single colloidal particles obey rotational as well as translational diffusion . on time scales that are short when compared to the characteristic rotational diffusion time , self - propulsion shows up as directed motion . on significantly longer time scales , the overall diffusion coefficient is considerably enhanced @xcite . also the consequences of a possible additional particle deformability were studied extensively @xcite as it may occur , for example , for self - propelled droplets on surfaces @xcite or in a bulk fluid @xcite . another example of artificial self - propelled particles are granular hoppers that transform vertical vibration energy into directed horizontal motion @xcite . on the biological side , the conversion of chemical energy into directed motion allows bacteria to individually search for food @xcite or to swim towards or away from illuminating light @xcite . likewise , chemical energy is used by bacteria , amoebae , or tissue cells to crawl on substrates @xcite . propulsion mechanisms are manifold and reach from the beating of a single flagellum @xcite via twisting deformations of whole filamental bacterial bodies @xcite to the synchronization of the motion of thousands of cilia @xcite . when many self - propelled particles act together , collective modes of migration emerge , like swarms that move as single entities @xcite , traveling density bands @xcite , or lanes of counterpropagating particles @xcite . even turbulent states arise that are based on the energy conversion on the single particle length scale @xcite . already without adhesive interactions and in the absence of alignment mechanisms , clusters of self - propelled particles can form in high - density systems @xcite . these clusters appear when there is no interaction that aligns the migration directions of different particles so that they can mutually block their motion . for monodisperse two - dimensional systems , hexagonal packing can arise @xcite . confining walls can support a local clustering or trapping of self - propelled particles at the boundaries @xcite . if the particle interaction is strong enough compared to their self - propulsion , then a formation of active crystals composed of self - propelled particles is to be expected . for this scenario , particle simulations demonstrated the emergence of crystal - like objects that collectively migrate as a single structure @xcite . in contrast , the approach that we recently introduced @xcite and analyze here in breadth is a complementary continuum one . it combines the description of crystalline materials by the phase field crystal model @xcite with ideas from the toner - tu model @xcite to characterize active crystalline systems of self - propelling objects @xcite . to our knowledge , there is no field - theoretical approach preceding ref . @xcite that includes the translational ordering in active crystals of self - propelled particles . the phase field crystal model is a microscopic field approach introduced by elder et al . to efficiently model processes in crystalline materials with particle resolution , but on diffusive time scales @xcite . in this way it can reproduce results from molecular dynamics simulations @xcite , yet in a computationally much more efficient way . motivated by dynamical density functional theory @xcite , the phase field crystal model can in principle be viewed as a microscopic theory describing a manifold of solidification and crystallization phenomena . examples are defect motion in strained crystals and multidomain structures @xcite , epitaxial growth @xcite , crystal pinning @xcite , crystal growth of binary alloys @xcite , branching in crystal growth @xcite , or even crack propagation @xcite . quite contrarily , the toner - tu model is a macroscopic hydrodynamic - like continuum description of systems of self - propelling objects @xcite . local spatial i.e. translational order of the single particles is not resolved or explicitly taken into account . the focus in this model is on the orientational order of the migration directions and thus on the macroscopic emergence of collective motion . it was argued that , if locally the migration directions spontaneously order , long - ranged orientational order of the migration directions and therefore global collective motion into one single collective direction emerges even in two spatial dimensions @xcite . our present work starts from an `` active phase field crystal model '' that we recently introduced in a previous study @xcite . it combines ideas from the two outlined theories , the equilibrium phase field crystal model and the toner - tu approach namely . we extend our previous investigation by providing for example more details about inverse active lattice structures and the migration speeds of the collectively migrating textures . our main focus here is on the stability of traveling active crystals . we therefore perform a linear stability analysis of these structures and demonstrate linear stability even on large length scales that are not accessible by direct numerical simulations . furthermore , for particles self - propelling on a substrate and surrounded by a thin fluid film , we include the influence of hydrodynamic interactions : they can disturb or destroy the crystalline and orientational order . as a first step we derive in this section our active phase field crystal model on the basis of dynamical density functional theory for brownian systems @xcite . we consider a system of self - propelled particles in two spatial dimensions . examples would be amoebae or tissue cells crawling on a substrate @xcite or granular hoppers that move horizontally on a vibrating plate and can not jump over each other @xcite . such systems can directly exchange momentum with the ground . the same should also be true for low reynolds number swimmers in a thin film of fluid on a substrate @xcite . we assume that each particle migrates with an effective active drive of magnitude @xmath0 into an individual direction that is characterized by a unit vector @xmath1 . the one - particle density distribution function is therefore given by @xmath2 , with @xmath3 the spatial coordinate and @xmath4 the time . in two spatial dimensions , the orientation of @xmath1 can be parameterized by a single angle @xmath5 . thus the dynamical equation for the one - particle density follows as @xcite @xmath6 in this equation @xmath7 sets the inverse thermal energy , @xmath8 is the translational diffusion tensor , @xmath9 denotes the rotational diffusion constant , and @xmath10 is an equilibrium energy functional that will be specified below . the effect of self - propulsion enters through the last term on the right - hand side via the magnitude of the active drive @xmath0 . our next step is to derive dynamical equations for the particle density and for the polar orientational order of the self - propulsion directions . we denote the average value of the one - particle density distribution function @xmath2 as @xmath11 , i.e. @xmath12 which is a conserved quantity in a closed system . here , @xmath13 is the spatial volume of the system and @xmath14 is the surface area of the unit sphere . the modulation of the particle density @xmath15 around @xmath11 is obtained by integrating out the orientational degrees of freedom from @xmath2 and subtracting @xmath11 , @xmath16 furthermore , we take into account the first orientational moment of @xmath2 , which we denote as @xmath17 with @xmath18 the dimensionality of the system . from this definition it becomes evident that the polarization field @xmath19 describes the local orientational order of the active driving directions @xmath1 of the self - propelled particles . similarly to the procedure in refs . @xcite , we expand the one - particle density with respect to its orientational dependence as @xmath20 to proceed we insert this expansion into eq . ( [ eqrho ] ) . from the resulting equation , the different fourier modes in terms of the angle @xmath21 are calculated . in this way we obtain dynamical equations for @xmath22 and @xmath23 similar to the ones in refs . @xcite , now , however , including the active drive . applying the approximation of constant mobility and assuming an isotropic translational diffusion @xmath24 , the resulting expressions are considerably simplified . we finally find @xmath25 \partial_{\tau}(\bar{\rho}\tilde{\mathbf{p } } ) & = & \frac{\beta \tilde{d}\bar{\rho}}{\pi}\nabla^2\frac{\delta\tilde{\mathcal{f}}}{\delta(\bar{\rho}\tilde{\mathbf{p } } ) } -\frac{\beta \tilde{d}_r \bar{\rho}}{\pi}\frac{\delta\tilde{\mathcal{f}}}{\delta(\bar{\rho}\tilde{\mathbf{p } } ) } \nonumber\\[.05 cm ] & & { } -v\nabla(\bar{\rho}\tilde{\phi}_1 ) . \label{eqpdft}\end{aligned}\ ] ] in the following we specify the free energy functional @xmath10 . more precisely , we introduce @xmath26 as a sum of a translational part @xmath27 and an orientational part @xmath28 . on the one hand , @xmath29\phi + \frac{u}{4}\phi^4 \right\},\ ] ] is the free energy introduced as the phase field crystal model @xcite . here , the order parameter @xmath30 is related to the field @xmath31 in eq . ( [ rhoexpand ] ) via @xmath32 and @xmath33 . the parameters @xmath34 , @xmath35 , and @xmath36 determine the magnitude of the energetic contributions , whereas @xmath37 is a measure for the temperature . minimizing this functional for an equilibrium system leads to fluid - like states at higher temperatures or high values of @xmath38 , hexagonally crystalline states at lower temperatures and intermediate values of @xmath38 , and lamellar states at lower temperatures and low values of @xmath38 @xcite . the characteristic length scale of the hexagonally crystalline and lamellar structures is determined by the magnitude of @xmath39 in eq . ( [ fpfc_phi ] ) . on the other hand , the orientational part is given by @xmath40 ^ 2 \right\}.\ ] ] the concept behind this free energy is similar to the original approach by toner and tu @xcite . we have @xmath41 due to thermodynamic stability . if @xmath42 and @xmath43 , the system can reduce its free energy by an orientational ordering of the self - propulsion directions , given by a nonzero solution of @xmath23 . this corresponds to an intrinsic tendency towards ordered collective motion . if @xmath44 , orientational ordering of the self - propulsion directions always costs orientational free energy . consequently there is a tendency to avoid collectively ordered motion . this case shows a richer behavior than the other one and will mainly be considered below . predominantly using the ramakrishnan - yussouff expansion @xcite , the phase field crystal model could be connected to density functional theory @xcite ; for a review see ref . expressions for the phenomenological parameters in terms of microscopic correlation functions could be derived . afterwards the procedure was extended to the case of an additional orientational order @xcite . likewise microscopic expressions for the corresponding phenomenological parameters were listed in these references , and we do not repeat them here . to finally obtain our dynamical equations , we introduce eqs . ( [ fsum])([ftildep ] ) into eqs . ( [ eqpsi1dft ] ) and ( [ eqpdft ] ) . moreover we apply the rescaling rules @xcite @xmath45 this rescales @xmath32 to @xmath46 . all lengths from now on are measured in units of the characteristic length scale @xmath39 of the phase field crystal structures . an additional simplification of the notation follows from the further rescaling @xmath47 \tilde{c}_4 & = & \frac{u}{4}\,c_4 , \\[.1 cm ] \tilde{d}_r & = & \tilde{d}q_0 ^ 2\,d_r.\end{aligned}\ ] ] at the end of this procedure we obtain our dynamical order parameter equations in the form @xcite @xmath48\psi_1 % \nonumber\\ % & & \qquad { } + 3\bar{\psi}\psi_1 ^ 2+\psi_1 ^ 3\,\big\ } \nonumber\\ & & { } - v_0\,\nabla\cdot\mathbf{p } , \label{eqpsi1}\\[.1 cm ] \partial_{t}\mathbf{p } & = & \nabla^2\left ( c_1\mathbf{p}+c_4\mathbf{p}^2\mathbf{p } \right ) - d_r\left ( c_1\mathbf{p}+c_4\mathbf{p}^2\mathbf{p } \right)\nonumber\\[.05 cm ] & & { } - v_0\nabla\psi_1 . \label{eqp}\end{aligned}\ ] ] the field @xmath49 is a measure for the local particle density , whereas the field @xmath50 measures the local orientational order of the active driving directions of the self - propelled particles . @xmath51 controls the temperature , @xmath52 the mean density , and @xmath53 the strength of the active drive of the individual particles . the ordering behavior of the active driving directions of the self - propelled particles is determined by @xmath54 and @xmath55 , whereas @xmath56 describes the orientational diffusion of the active drive directions . as noted above , the equilibrium phase field crystal model based on eq . ( [ fpfc_phi ] ) shows hexagonally crystalline structures in a certain range of temperature @xmath51 and mean density @xmath52 . changing the temperature or mean density , hexagonal textures can be transformed into lamellar ones . in the non - equilibrium case considered here , such a transition can also be achieved by increasing the active drive @xmath53 of the individual self - propelled particles @xcite . this scenario is depicted in the upper row of fig . [ snapshots ] . the snapshots shown in the figure were obtained by numerically solving eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) , starting from random initial conditions . we start with a low value of the active drive @xmath53 and a hexagonal structure in panel ( a ) . for the parameter values listed in the figure caption , density peaks form that are symmetrically located within `` + 1''-defects of the polarization field . the overall structure remains at rest . this changes with increasing @xmath53 , when the symmetry of the density peak positions with respect to the `` + 1''-defects of the polarization field is spontaneously broken . this spontaneous shift of the density peaks out of the centers of the `` + 1''-defects of the polarization field represents the symmetry breaking necessary for a net active propulsion to emerge . as a whole , the structure starts to migrate ( travel ) collectively ( b ) . further increasing @xmath53 deforms the traveling hexagonal crystal to a traveling rhombic and then to a traveling quadratic crystal ( c ) . finally it leads to a transition to traveling lamellae ( d ) . typically , when thinking of crystalline textures in this way , we would identify each density peak with one particle . however , also the situation of cluster crystals is included by our approach . in that case several particles occupy one lattice site @xcite . such active cluster crystalline structures were observed before in a system of deformable self - propelled particles @xcite and in a modified vicsek model confined between two parallel walls @xcite . furthermore , traveling crystalline textures have been observed previously in three - component reaction - diffusion systems @xcite . when we thus identify each density peak in the top row of fig . [ snapshots ] with a cloud of self - propelled particles , we can similarly interpret the corresponding active inverted textures in the bottom row . in each column of fig . [ snapshots ] the two snapshots can be mapped onto each other by noting a special symmetry relation of eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) . the equations remain invariant under the simultaneous transformations @xmath57 , @xmath58 , and @xmath59 . consequently the series of patterns in the bottom row of fig . [ snapshots ] follows directly as an inversion of the top row patterns . we will concentrate on the characterization of hexagonal , rhombic , and quadratic crystalline textures in the following . nevertheless the results apply equally for the inverted structures due to the symmetry property of the dynamical equations . active honeycomb textures are formed for example by flagellated marine bacteria @xcite . to quantify the collective motion of the active crystals , we tracked the migration of each individual density peak in the sample . for this purpose , the center of each density peak was determined at fixed time intervals . on the one hand , this time interval was short enough so that identical density peaks could be identified before and after each migration step . on the other hand , it should be long enough so that the migration step is larger than the mesh size of the calculation grid . it was shown previously that at least for the finite system sizes investigated a collectively traveling single crystal develops from the random initial conditions via an intermediate multidomain structure @xcite . at the end of this coarsening process towards the active single crystal , all density peaks migrate collectively into the same direction . the migration speed @xmath60 of the resulting active single crystal follows from a sample average of the peak velocity magnitudes of all density peaks , @xmath61 . here @xmath62 denotes the total number of all density peaks in the sample . the velocities @xmath63 of the individual density peaks , with @xmath64 , follow from the tracking procedure described above . corresponding results are depicted in fig . [ vmvsv0 ] for characteristic values of the parameters @xmath54 and @xmath55 . as mentioned above , these parameters control the spontaneous orientational ordering of the migration directions of the individual self - propelled particles . of the single crystals as a function of the active drive @xmath53 of the individual particles at @xmath65 . for @xmath66 and @xmath67 ( squares ) the directions of active drive of the individual particles order spontaneously and the structures travel collectively with @xmath68 . a nonzero threshold appears for @xmath69 and @xmath67 ( rhombi ) . at @xmath70 and @xmath71 ( circles ) the directions of active drive of the individual particles do not order spontaneously . then the structures are at rest ( @xmath72 ) below a finite threshold value @xmath73 ; they start to travel collectively with nonzero @xmath74 above the threshold.,width=321 ] we plot in fig . [ vmvsv0 ] the collective migration speed @xmath60 of the single crystal as a function of the active drive @xmath53 of the individual self - propelled particles . for @xmath70 the active driving directions of the individual particles do not order spontaneously . at low values of the active drive @xmath53 , the overall hexagonally crystalline structure remains at rest , i.e. @xmath72 . this corresponds to the situation in fig . [ snapshots](a ) . only at a nonzero threshold value @xmath73 do the crystals start to travel collectively , see also fig . [ snapshots](b ) . here , the nonzero value of @xmath67 only plays a minor quantitative role , and a similar curve is obtained for @xmath75 . in contrast , there is no threshold when the active driving directions of the individual particles order spontaneously , i.e. @xmath66 and @xmath67 . the whole crystalline structure migrates at each nonzero value of the active drive @xmath53 of the individual particles . even more , the whole crystal propels collectively at the speed of the individual particles , i.e. @xmath68 . in addition to this curve of negative @xmath54 , we show in fig . [ vmvsv0 ] an intermediate case of @xmath69 and @xmath67 . there , the nonlinear @xmath55-term leads to a small residual threshold behavior at low values of @xmath53 . during the remaining part of this paper we will be concerned with the case of @xmath70 and @xmath75 . it shows the richer behavior due to the emergence of resting as well as traveling structures that are separated by the finite threshold value @xmath73 . introducing a hydrodynamic - like continuum theory , toner and tu studied the collective motion of flocks of self - propelling objects @xcite . their focus was mainly on the orientational ordering of the self - propulsion directions , and not on a translational , possibly crystalline order of the individual objects . the theory contains nonlinear convective terms due to self - propulsion . analyzing their model , toner and tu found that long - ranged orientational order of the self - propulsion directions can arise in two spatial dimensions @xcite , in contrast to what is known for orientational ordering in equilibrium systems @xcite . such nonlinear convective terms that could lead to long - ranged orientational order of the self - propulsion directions are absent in our approach . therefore the question arises whether a very large crystal will still migrate as a single structure into one direction , or whether it will break up into domains of different migration directions . instead of the nonlinear convective terms , however , our approach contains a kind of elastic interaction . the phase field crystal free energy functional @xmath76 of eq . ( [ fpfc_phi ] ) acts in the form of a chemical potential in the dynamical equation ( [ eqpsi1 ] ) for the density . it enforces a locally spatially periodic arrangement of the density peaks with a characteristic wave number . deviations from this order are energetically penalized . it has been demonstrated in particle simulations @xcite that solely elastic interactions between self - propelled particles , without any explicit alignment interaction @xcite , can induce global collective motion into one common migration direction . there the elastic interactions guided the particles to find a global mode of migration , even under the influence of moderate stochastic noise . in this way , elasticity enforced global collective motion . due to the finiteness of the system sizes it is difficult , however , to make a rigid statement about what happens on very large length scales on the basis of the particle simulations . our continuum field approach offers an advantage in this respect . the explicit dynamical equations ( [ eqpsi1 ] ) and ( [ eqp ] ) contain all length scales . in principle , their investigation is not limited to a certain system size . an arbitrary large system can be analyzed . in reality , however , to get an explicit solution , we have to solve the equations on a finite numerical lattice . nevertheless , we can extract from our equations the information of whether an arbitrarily large single crystal with a single migration direction is linearly stable . our procedure will be described in the following . as mentioned before , we confine ourselves for this purpose to the case in which the self - propulsion directions do not spontaneously order , i.e. @xmath70 ( @xmath75 ) . as a first step we linearize the dynamical equations ( [ eqpsi1 ] ) and ( [ eqp ] ) with respect to a small perturbation @xmath77 and @xmath78 of the order parameter fields , @xmath79 here , @xmath80 and @xmath81 correspond to the solutions obtained from directly solving numerically the dynamical equations ( [ eqpsi1 ] ) and ( [ eqp ] ) . these solutions describe , on the finite size of the numerical calculation grid , a single crystalline structure as illustrated for example in fig . [ snapshots ] . in our case , the converged steady - state solutions @xmath80 and @xmath81 , around which we perturb the system , are spatially inhomogeneous . this significantly complicates the analysis as will be shown below . from the linearization we obtain @xmath82\delta\psi_1 + 6\bar{\psi}\psi_{10}\delta\psi_1 \nonumber\\ & & { } \qquad+3\psi_{10}^2\delta\psi_1 \big\ } - v_0\,\nabla\cdot\delta\mathbf{p } , \label{eqdeltapsi1}\\[.1 cm ] \partial_{t}\delta\mathbf{p } & = & \nabla^2 c_1\delta\mathbf{p } - d_r c_1\delta\mathbf{p } - v_0\nabla\delta\psi_1 . \label{eqdeltap}\end{aligned}\ ] ] since the coefficients of the perturbations in the second equation are constant , the system of equations can be significantly simplified in fourier space . we apply the transformations in the form @xmath83 eqs . ( [ eqdeltapsi1 ] ) and ( [ eqdeltap ] ) are transformed accordingly . then the second equation can be solved for @xmath84 , which is inserted into the first equation . in this way , the system of eqs . ( [ eqdeltapsi1 ] ) and ( [ eqdeltap ] ) is reduced to a single equation on the scalar order parameter field @xmath85 , which reads @xmath86\,\delta\psi_1(\mathbf{q},\omega ) \nonumber\\ & & { } + \frac{6\bar{\psi}}{(2\pi)^3}\int\!d^2q'\int\!d\omega'\ ; \psi_{10}(\mathbf{q}-\mathbf{q}',\omega-\omega')\delta\psi_1(\mathbf{q}',\omega ' ) \nonumber\\ & & { } + \frac{3}{(2\pi)^3}\int\!d^2q'\int\!d\omega'\ ; \psi_{10}^2(\mathbf{q}-\mathbf{q}',\omega-\omega')\delta\psi_1(\mathbf{q}',\omega ' ) \,\big\ } \nonumber\\ & & { } -\frac{q^2v_0 ^ 2}{c_1(d_r+q^2)+i\omega}\,\delta\psi_1(\mathbf{q},\omega ) . \label{stabaneq}\end{aligned}\ ] ] as we can see , the analysis becomes nonlocal in fourier space . the reason is the spatial modulation of the traveling or resting unperturbed structures , as they are for example depicted in fig . [ snapshots ] . in principle , eq . ( [ stabaneq ] ) represents an infinite system of equations in which all modes @xmath85 of different @xmath87 and @xmath88 are coupled . a similar situation occurs in the study of microphase - separated diblock copolymer systems ref . there , however , an equilibrium situation is investigated , which allows a different treatment of the problem . fortunately the situation can be simplified by exploiting the regular nature of the unperturbed states @xmath80 . first , the structures form a steady state in the sense that they migrate as a single object with a constant migration velocity @xmath89 . in the co - moving frame , the single crystals are at rest . second , the spatial modulation of the unperturbed structures is periodic . using these observations , eq . ( [ stabaneq ] ) can be split into discrete sets of equations in fourier space , each of them still being infinite , however . a numerical evaluation is required to handle the problem . since the unperturbed structures @xmath80 form a steady state , we can write them in the form @xmath90 this relation translates itself into fourier space in the following way : @xmath91 introducing these expressions into eq . ( [ stabaneq ] ) decouples the modes in the frequencies @xmath88 , which significantly simplifies the problem . for the second simplification , we use the fact that the unperturbed structures are spatially periodic . they are resting or traveling crystals . in fourier space they are therefore characterized by a reciprocal lattice of a discrete set of wave vectors . translating in reciprocal space the whole structure by a reciprocal lattice vector maps the positions of the reciprocal lattice points onto themselves . in this sense , the reciprocal lattice is closed by itself under such translations . the subtractions @xmath92 in eq . ( [ stabaneq ] ) correspond to these translations . as a consequence , the set of equations ( [ stabaneq ] ) splits into different decoupled discrete subsets of equations . within one such set , all modes are connected via addition or subtraction of the reciprocal lattice vectors of the unperturbed structure . still each set generally contains infinitely many modes , but the discretization already implies a major reduction . in the following we investigate three different structures : a traveling hexagonal crystal , a traveling rhombic crystal , and a traveling quadratic crystal . as mentioned above , the textures are obtained in this sequence when increasing the active drive @xmath53 . on the left - hand side , [ ftstructure ] shows snapshots of the steady - state structures in real space , i.e. @xmath93 . on the right - hand side , the corresponding power spectra @xmath94 in two - dimensional fourier space are included . and ( a ) @xmath95 for a traveling hexagonal crystal , ( b ) @xmath96 for a traveling rhombic crystal , and ( c ) @xmath97 for a traveling quadratic crystal . only a fraction of the numerical calculation box is shown . right column : corresponding power spectra in two - dimensional fourier space . a distorted hexagonal lattice structure is obtained in fourier space in panel ( d ) from the traveling hexagonal case . panel ( e ) gives the power spectrum for the traveling rhombic crystal . finally , the square pattern in real space transforms itself into a square pattern of the two - dimensional power spectrum ( f ) . the intensity of the peaks in the power spectrum is represented by their brightness and was partially rescaled for better visualization . circles mark the directions of the reciprocal lattice vectors of smallest magnitude @xmath98 . the examples of linear stability analysis against linear dynamical instabilities that are shown in fig . [ logdeterminantdynamic ] were performed in these directions.,width=321 ] the different textures hexagonal , rhombic , and quadratic can be identified by eye from the real - space plots . also the spatial arrangements of the inner intensity peaks on the right - hand side of fig . [ ftstructure ] reflect these symmetries . however , when we take into account the magnitude of the intensities , we see that the collective migration breaks the symmetry . most obviously , there is no four - fold symmetry in the intensity magnitudes of fig . [ ftstructure ] ( f ) for the quadratic texture . likewise , the single real - space density peaks in fig . [ ftstructure ] ( c ) appear a little distorted . the direction of collective migration marks the orientation of the remaining symmetry axis . we can see that the magnitude of the peaks in the power spectrum , coded by their brightness in fig . [ ftstructure ] , significantly decays with increasing wave number . as a consequence we introduce a positive threshold value @xmath99 . @xmath93 is approximated by only those modes that satisfy @xmath100 . the same procedure is applied to @xmath101 , i.e. the fourier transform of @xmath102 . for each investigated structure , the above procedure must be performed only once in practice . we numerically iterate eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) forward in time starting from random initial conditions . when the system has reached the steady state , @xmath103 is used as an input for the unperturbed states @xmath93 and @xmath102 . after performing the fourier transform , the values @xmath104 and @xmath101 that satisfy the above threshold criterion together with their locations @xmath87 in the reciprocal space are stored in an ordered list structure for the further analysis . we also numerically measure @xmath89 during this initial numerical iteration and use this value in our further evaluation . ( it is a challenging task to calculate the collective migration velocity @xmath89 analytically . a corresponding attempt in one spatial dimension that is restricted to the threshold vicinity is included in the appendix . ) our intention is to perform a linear stability analysis of the traveling single crystal at long wave lengths , i.e. at small wave numbers @xmath105 . direct numerical solution of eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) showed that the structures are stable on the length scale of the size of the numerical box . but the question of what happens on larger length scales has so far remained unanswered . in other words , we are interested in the question whether the single crystal will break up into a multidomain texture for large system sizes . below , we denote the reciprocal lattice vectors of the unperturbed structure as @xmath106 , with @xmath107 a labeling index . they localize the positions of the spectral points in the right column of fig . [ ftstructure ] . the reciprocal lattice vector of smallest magnitude is called @xmath98 . we probe a square region in reciprocal space bordered at two opposing edges by @xmath108 and @xmath109 . this region covers the first brillouin zone . all wave vectors @xmath87 lying in the first brillouin zone belong to a different discrete coupled subset of equations spanned by the nonlocal terms in eq . ( [ stabaneq ] ) . we can not connect two different wave vectors in this region by adding or subtracting reciprocal lattice vectors . still , these subsets of equations are infinite . we have seen , however , that the coupling strengths @xmath110 and @xmath111 that are responsible for the nonlocal coupling in eq . ( [ stabaneq ] ) decay with increasing distance @xmath112 . therefore we introduce a cutoff distance @xmath113 and only include modes that satisfy @xmath114 in the subset of equations for each wave number @xmath87 . in this way , the discrete subsets of equations become finite and can be handled numerically . in practice , we sample the square region between @xmath108 and @xmath109 in small wave vector steps . at each @xmath87 in this region , we build up the discrete set of equations described above . for this purpose , we numerically search for all modes @xmath115 with wave vectors @xmath116 that can be obtained by iteratively adding or subtracting the reciprocal lattice vectors that predominantly contribute to @xmath117 and @xmath118 ( those were previously stored in a list when calculating the spectra , see above ) . only modes @xmath115 that satisfy the condition @xmath114 are included . ( [ stabaneq ] ) leads for each wave vector @xmath87 and frequency @xmath88 to a linear system of equations for all modes @xmath115 that are coupled to @xmath85 . in practice , our cut - off values are chosen such that the coupling to several hundred modes is included for each wave vector @xmath87 . the coefficients of this system of equations are calculated and stored in matrix form . for practical reasons , to obtain a matrix of real coefficients , we consider the real and imaginary parts of the perturbations @xmath119 as independent variables . each equation in the system of equations ( [ stabaneq ] ) is split into its real and imaginary part , which doubles the number of equations . finally , the determinant @xmath120 of the resulting matrix is calculated by standard numerical procedures @xcite . at the end of this procedure , a vanishing determinant @xmath120 signals a linear instability of the single crystal with respect to the perturbation @xmath85 . before continuing with the results , we add a technical remark to explain the appearance of the plots included below . due to the many modes that are coupled in each case , the number of multiplications during the process of obtaining the determinant @xmath120 is very large . in fact , @xmath120 often exceeds the size that can be processed on the computer . to circumvent this problem , we rescale the factors in the final multiplication by an adjusted constant positive number . since we are only interested in the question whether @xmath120 vanishes ( or changes sign ) this is a legal procedure . nevertheless , the absolute values of @xmath120 should not be interpreted any more . furthermore , we calculate and plot the logarithm of the determinant @xmath121 $ ] . in a first step , we consider static instabilities in the co - moving frame . that is , we set @xmath122 ( and likewise @xmath123 ) in eq . ( [ stabaneq ] ) . thus @xmath88 does not describe an additional degree of freedom but is determined by the wave vector @xmath87 . the results corresponding to the three textures in fig . [ ftstructure ] are depicted in fig . [ logdeterminantco ] . against instabilities of small wave - numbers in the co - moving frame . the analyzed regions in @xmath87-space contain the first brillouin zones of the textures in fig . [ ftstructure ] , respectively . surface plots show the logarithm of the determinant @xmath124 of the stability matrix in the analyzed @xmath87-regions . it has been rescaled for technical reasons in each panel , so that the absolute values should not be interpreted . below the surfaces , the shadow plots represent colored projections of the surfaces into the plane . we define @xmath125 , @xmath126 the component of @xmath87 parallel to @xmath98 , and @xmath127 its component in the perpendicular direction , where @xmath98 is the reciprocal lattice vector of smallest magnitude . a linear instability would be indicated by a vanishing @xmath124 , however , we observe nonzero values in all probed instances ( and no sign changes ) . we thus conclude that the structures are linearly stable against static instabilities in the co - moving frame.,width=321 ] the determinant naturally vanishes at @xmath128 and at @xmath129 , which can not be shown in the logarithmic plot and leads to the white wholes in the projected shadow plots below each surface plot . we do not find any other position in the first brillouin zone where the determinant vanishes . the small dips visible in the surfaces of @xmath130 $ ] were found to decrease with increasing size of the numerical calculation box that was used to obtain @xmath117 . they were not observed to indicate a linear instability . we thus conclude that the investigated structures are linearly stable against static instabilities in the co - moving frame . after excluding static linear instabilities in the co - moving frame , we ask for dynamical linear instabilities . for example , on larger length scales , there might appear domains migrating in different directions . this question adds a further degree of freedom to our analysis , namely the frequency @xmath88 . previously , it was fixed by the relation @xmath122 in the co - moving frame . now we can in principle choose any value independently of the selected wave vector @xmath87 . fortunately , for each perturbation @xmath85 , the coupled system of equations has the same size as above in the analysis within the co - moving frame . the additional degree of freedom @xmath88 does not lead to further couplings . we can understand this result from eqs . ( [ stabaneq ] ) , ( [ psi10co ] ) , and ( [ psi10sqco ] ) . on the one hand , eq . ( [ stabaneq ] ) implies that the modes @xmath131 are coupled via the unperturbed steady - state fields @xmath132 and @xmath133 . on the other hand , due to the steady - state form of @xmath80 and @xmath134 , eqs . ( [ psi10co ] ) and ( [ psi10sqco ] ) contain a rigid relation between @xmath87 and @xmath88 in the @xmath135-functions . in this way , each coupling wave vector @xmath136 in eq . ( [ stabaneq ] ) uniquely fixes one coupling frequency @xmath137 . the complete @xmath87-@xmath88 space is too large to be systematically tested . we therefore particularly focused on the representative directions given by the reciprocal wave vectors of each structure , and the directions perpendicular to these . in neither case did we find a vanishing determinant that would indicate a dynamical linear instability . this provides evidence that the investigated traveling active single crystals are also dynamically linearly stable . a few examples of our results are depicted in fig . [ logdeterminantdynamic ] . against linear dynamical instabilities . the panels show stability probes performed in @xmath87-space along the direction of the reciprocal lattice vector of smallest magnitude @xmath98 . in addition , the frequency @xmath88 was varied in an interval centered around @xmath138 at each wave vector @xmath87 , with @xmath89 the collective migration velocity . we define @xmath139 , @xmath125 , and @xmath140 . surface plots show the logarithm of the determinant @xmath141 of the stability matrix in the analyzed @xmath87-@xmath142 regions . it has been rescaled for technical reasons in each panel , so that the absolute values should not be interpreted . below the surfaces , the shadow plots represent colored projections of the surfaces into the plane . our examples correspond to the directions indicated by the circles in the power spectra of fig . [ ftstructure ] for ( a ) @xmath95 , ( b ) @xmath96 , and ( c ) @xmath97 . a linear instability would be indicated by a vanishing @xmath141 ( or a sign change ) . apparently , the structures are linearly stable against linear dynamical instabilities.,width=321 ] one might argue that there are sources of uncertainty in our analysis . for example , we obtain our initial reciprocal lattice vectors from a numerical calculation on a grid of finite size . however , if this approximation were problematic , it would rather show up as an instability of the unperturbed state . in contrast to that , we here find that the system is linearly stable despite such approximations . we are therefore confident that our analysis reflects the real behavior of the structures , and that the active single crystals are linearly stable against perturbations . in the previous section , we have seen that the active single crystals are linearly stable even at very large system sizes . the underlying stabilization mechanism is provided by the effective elastic interactions between the density peaks . it is introduced by the phase field crystal free energy functional that acts like a chemical potential for the density peaks in our dynamical equations ( [ eqpsi1 ] ) and ( [ eqp ] ) . this picture is further supported by an analysis of related particle simulations @xcite . on the contrary the question arises , whether in real experimental systems there might be a counteracting mechanism that could destroy this long - ranged order . a natural candidate is given by hydrodynamic interactions . it turns out that hydrodynamic interactions can destabilize the system already at relatively small system sizes . this can be studied by direct numerical simulation of our dynamical equations , so that a complicated analysis as in the previous section is not necessary . in the situation that we investigate , the self - propelled particles can still directly exchange momentum with the ground . this requirement is satisfied by particles that directly migrate on a substrate , or approximately by swimmers that propel in a low reynolds number environment close to a surface with no - slip boundary conditions . however , we assume that the density peaks and lamellae can additionally interact with each other through a surrounding background fluid film . our goal in the following is a simple qualitative estimate of the effect of hydrodynamic interactions . we assume that the particles and the fluid have the same mass density , and that the system is incompressible . in this way the dynamical equation of mass continuity is trivially satisfied and does not need to be considered explicitly in the following . the location of the particles is described by the concentration or density field @xmath143 . we introduce a velocity field @xmath144 to parameterize the fluid flow that is induced by the particle motion . on the one hand , additional convective terms appear in the dynamical equations ( [ eqpsi1 ] ) and ( [ eqp ] ) that now read @xmath145 \hspace{-.7cm}\partial_{t}\mathbf{p } & = & \nabla^2\frac{\delta\mathcal{f}}{\delta\mathbf{p } } - d_r\frac{\delta\mathcal{f}}{\delta\mathbf{p } } - v_0\nabla\psi_1 -\mathbf{u}\cdot\nabla\mathbf{p } + \mathbf{\omega}\cdot\mathbf{p}. \label{eqphydro}\end{aligned}\ ] ] here the tensor @xmath146 with components @xmath147 describes rotations due to convection . we do not consider flow alignment of the polarization field in this qualitative picture @xcite . on the other hand , the additional dynamical equation for the velocity field @xmath144 in the rescaled form reads @xmath148 it is supplemented by the incompressibility condition @xmath149 , so that an explicit pressure term is not necessary . we can neglect the convective term on the right - hand side in the low reynolds number regime . when the self - propelled particles migrate on the substrate , they push the surrounding fluid and set it into motion . this effect is included by the force term @xmath150 that we introduce as @xmath151 with a sufficiently large coefficient @xmath152 . again @xmath62 is the number of density peaks in the sample . @xmath153 and @xmath63 denote , respectively , the current positions and velocities of the density peaks for @xmath154 . in practice we track the positions and motion of the centers of the density peaks as described above . at each peak center @xmath153 , the fluid flow velocity @xmath155 is then adjusted to the peak velocity @xmath63 . to mimic the presence of the particles in the fluid , the viscosity @xmath156 varies with the density @xmath49 . we set @xmath157 at positions of lowest density @xmath49 and let it linearly increase by a factor of @xmath158@xmath159 when moving to the regions of high density @xmath49 . due to the smooth density profiles in our numerical samples the effect of the contribution @xmath160 is mostly negligible . overall , the volume elements of high density @xmath49 , which represent the presence of the self - propelled particles , are considered as part of the fluid flow . this approach is similar to the `` fluid particle dynamics '' introduced by tanaka and araki @xcite . the last term @xmath161 in eq . ( [ equ ] ) represents a simple way to model the friction of the thin fluid film with its environment , with @xmath162 the friction constant . this term was used previously to investigate the complex flow behavior in quasi - two - dimensional systems of fluid films @xcite . two limits for the magnitude of @xmath163 are obvious . on the one hand , if @xmath163 is small , the friction between the fluid film and its surfaces is low and the migrating particles can easily set the surrounding fluid into motion . on the other hand , if @xmath163 is very large , we see from eq . ( [ equ ] ) that the fluid remains practically at rest with @xmath164 . then the convective terms in eqs . ( [ eqpsi1hydro ] ) and ( [ eqphydro ] ) are negligible and we reobtain the previous eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) . thus hydrodynamic interactions are not important in the regime of large @xmath163 . we illustrate the flow behavior at different values of the friction parameter @xmath163 in fig . [ fluidflow ] . for high @xmath163 , corresponding to strong fluid friction , most parts of the fluid film can not be set into motion as illustrated in panels ( a ) and ( d ) . the fluid flow remains very localized around the density peaks and does not extend between different peaks . consequently the hydrodynamic interactions between the density peaks are low . around the density peaks some `` backflow '' of the fluid oppositely to the peak migration directions occurs . this is stressed in a rescaled version of panel ( d ) , included as panel ( e ) ; a `` zoom '' into the region around one single density peak is provided for illustration . with decreasing fluid friction @xmath163 , the `` backflow '' vanishes and the fluid flow more and more extends between the density peaks , as depicted in panels ( b ) and ( f ) . at very low friction parameters @xmath163 , the whole fluid can be set into motion nearly homogeneously , as shown in panels ( c ) and ( g ) . since the fluid flow here extends between the density peaks , hydrodynamic interactions are strong in this case . as for fig . [ vmvsv0 ] , we measured again the collective migration speed @xmath60 in the form of the sample - averaged peak velocity magnitude , @xmath61 , with @xmath62 the number of the density peaks and @xmath63 the velocity of the single density peaks , @xmath154 . a result of @xmath60 as a function of the active drive @xmath53 of the individual self - propelled particles is plotted in fig . [ alphadep ] for two different of the single crystals as a function of the active drive @xmath53 of the individual particles in the presence of hydrodynamic interactions at @xmath165 . for low hydrodynamic interactions , corresponding to a large fluid friction parameter @xmath163 , the collective migration speed is lower . for strong hydrodynamic interactions , corresponding to a low fluid friction parameter @xmath163 , the collective migration speed is enhanced . in the latter case the particles can push each other by setting the fluid between them into motion . the critical value of the self - propulsion velocity @xmath73 remains unchanged by the presence of the hydrodynamic interactions . , width=321 ] values of the fluid friction parameter @xmath163 . in this case @xmath70 , i.e. there is no alignment mechanism that would lead to a spontaneous orientational ordering of the self - propulsion directions . as a first result , we found that , within our numerical resolution , the critical self - propulsion velocity @xmath73 , at which collective motion of the active crystal sets in , remains unchanged by the hydrodynamic interactions . second , the hydrodynamic interactions speed up the collective motion of the active single crystal . we can see this from the two data curves plotted in fig . [ alphadep ] . at high values of @xmath163 the fluid friction is large , and the self - propelling particles can not set the surrounding fluid into motion . however , at low fluid friction for low values of @xmath163 , the particles can set the whole surrounding fluid film into motion , as was also demonstrated in fig . [ fluidflow ] . in this way , one density peak can push the preceding peak via the fluid between them . the density peaks support each other in migration via hydrodynamic interactions . at fixed active drive @xmath53 of the individual self - propelled particles , the collective migration speed @xmath60 increases with decreasing fluid friction @xmath163 and increasing hydrodynamic interactions . however , the hydrodynamic interactions disturb the order of the active single crystals . this is illustrated in fig . [ destabilizedcrystal ] . . ( a ) negligible hydrodynamic interactions at strong fluid friction ( @xmath166 ) allow the emergence of a perfect traveling active single crystal . ( b ) strong hydrodynamic interactions at low fluid friction ( @xmath167 ) destabilize this texture.,width=321 ] the figure includes two identical systems that only differ by the fluid friction parameter @xmath163 . at high @xmath163 and therefore little hydrodynamic interaction ( a ) , the sample shows a perfect active single crystal that migrates collectively into one direction . on the contrary , the crystal is broken up into different domains at low @xmath163 corresponding to strong hydrodynamic interactions ( b ) . also a partial transition to lamellar textures is induced by the hydrodynamic interactions . in the case without hydrodynamic interactions this transition would typically occur only at higher magnitudes of the active drive @xmath53 @xcite . in summary , we can say that at least within the qualitative picture presented hydrodynamic interactions have a destabilizing effect on the active crystalline structures considered here . in this paper we focused on the stability of traveling active single crystals . an active phase field crystal model was introduced for this purpose . we outlined its derivation via dynamical density functional theory from a microscopic consideration of self - propelled particles that feature an active drive . besides resting hexagonal crystals , traveling hexagonal , rhombic , and quadratic crystals , as well as traveling lamellar textures were obtained from this description . also the inverted structures like resting and traveling honeycomb textures , as well as inverted traveling rhombic and quadratic lattices were observed . a linear stability analysis did not indicate a linear instability of the collectively migrating single crystals . in particular , long - wave - length instabilities beyond the size of the numerical calculation grids did not show up , indicating linear stability also for large system sizes . we found , however , that hydrodynamic interactions can destabilize the single - crystalline textures . the linear stability analysis was significantly complicated by the spatial modulation of the unperturbed steady state . as a consequence of this spatial modulation , our equations became nonlocal in fourier space . we could reduce the problem to manageable size by exploiting the periodicity of the unperturbed state and its steady - state migration . this made the linear stability analysis possible . the problem of a spatial modulation of the unperturbed state occurred already in a previous equilibrium stability study of micro - phase separated diblock copolymer melts @xcite . there , however , the equilibrium conditions allowed a different strategy to perform the analysis . our results on the stability of the traveling crystals are important when we think of the design of new active materials . it is expected that , for example , features of traveling crystals like phononic properties are different from their passive counterparts . we have seen that the structure of the migrating crystals , i.e. their hexagonal , rhombic , or quadratic form , can be tuned by the activity of the constituent self - propelled particles . it will be interesting to test whether other features of active crystals , such as their phononic behavior , can be similarly tuned by the properties of the individual active components . we shall investigate this interesting question in the future . the authors thank the deutsche forschungsgemeinschaft for support of this work through the german - japanese project lo 418/15 . in this appendix we briefly analytically address the onset of collective migration . our goal is to derive an expression for the collective migration speed @xmath60 . this is a challenging task . we confine ourselves to one spatial dimension that we denote as the @xmath168-direction . furthermore , we consider the case of @xmath70 and @xmath75 . still our results remain restricted to the very close vicinity of the threshold for the onset of collective motion . under these circumstances , and using the fact that the structures collectively migrate in a steady - state solution with speed @xmath60 , see eq . ( [ steadystate ] ) , we can rewrite eqs . ( [ eqpsi1 ] ) and ( [ eqp ] ) as @xmath169\psi_1 \label{apppsi1 } \nonumber\\ & & { } \qquad+ 3\bar{\psi}\psi_1 ^ 2+\psi_1 ^ 3\,\big\ } - v_0\,\partial_z{p } , \\[.1 cm ] -v_m\partial_z{p } & = & c_1\!\left ( \partial_z^2-d_r \right)\!p - v_0\partial_z\psi_1 . \label{appp}\end{aligned}\ ] ] introducing the abbreviations @xmath170 and @xmath171 , we can write the formal solution of eq . ( [ appp ] ) as @xmath172 \partial_z\psi_1 \nonumber\\ & = & { } -\bar{v}_0\partial_z\left [ \gamma+\bar{v}_m\gamma^2 \partial_z + \bar{v}_m^2\gamma^3\partial_z^2+\bar{v}_m^3\gamma^4\partial_z^3+ ... \right]\psi_1 , \nonumber\\&&\label{solp}\end{aligned}\ ] ] where we have defined @xmath173 via @xmath174 with the abbreviation @xmath175 when the operator @xmath173 is applied to an arbitrary function @xmath176 . we have further used the relation @xmath177\psi(z ' ) \nonumber\\ & = & \partial_z \int dz'\,\gamma(z - z')\psi(z')\nonumber\\ & & \end{aligned}\ ] ] in the last step of eq . ( [ solp ] ) . ( [ solp ] ) is now inserted into eq . ( [ apppsi1 ] ) to obtain a closed equation for @xmath49 : @xmath169\psi_1 \nonumber\\ & & { } \qquad+ 3\bar{\psi}\psi_1 ^ 2+\psi_1 ^ 3\,\big\ } \nonumber\\ & & { } + v_0\bar{v}_0\partial_z^2\big [ \gamma+\bar{v}_m\gamma^2 \partial_z \nonumber\\ & & { } \qquad+\bar{v}_m^2\gamma^3\partial_z^2 + \bar{v}_m^3\gamma^4\partial_z^3+ ... \big]\psi_1 . \nonumber\\\end{aligned}\ ] ] we rewrite this expression in the form @xmath178\psi_1 \nonumber\\ & & { } \qquad+ 3\bar{\psi}\psi_1 ^ 2+\psi_1 ^ 3\,\big\ } \nonumber\\ & & { } + v_0\bar{v}_0\big [ \gamma+\bar{v}_m\gamma^2 \partial_z \nonumber\\ & & { } \qquad+\bar{v}_m^2\gamma^3\partial_z^2 + \bar{v}_m^3\gamma^4\partial_z^3+ ... \big]\psi_1 , \nonumber\\ \label{gintroduced}\end{aligned}\ ] ] where we have defined @xmath179 with an abbreviation analogous to eq . ( [ abbrgamma ] ) . next , we multiply eq . ( [ gintroduced ] ) by @xmath180 and carry out the integral over @xmath168 : @xmath181\psi_1.\qquad \nonumber\\ & & \label{inteq}\end{aligned}\ ] ] here the integrals are taken over the whole domain of size @xmath182 . the periodic boundary conditions at the domain boundaries imply @xmath183 for any integrable functional @xmath184 . our last step consists of expanding the operator @xmath173 on the right - hand side of eq . ( [ inteq ] ) into powers of @xmath185 . on the one hand , we have @xmath186 and on the other hand , @xmath187 with integers @xmath188 and @xmath189 . using these relations , we finally obtain from eq . ( [ inteq ] ) an expression for the collective migration speed at the onset of collective motion : @xmath190 where @xmath191g(z - z')\partial_{z'}\psi_1(z ' ) , \nonumber\\ & & \label{b0 } \\ b_1 & = & \frac{v_0 ^ 2}{c_1 ^ 2}\int dz\,\int dz'\,\int dz''\,\left[\partial_z\psi_1(z)\right ] \nonumber\\ & & { } \qquad\times\gamma(z - z')\gamma(z'-z'')\partial_{z''}\psi_1(z '' ) , \nonumber\\ & & \label{b1 } \\ b_3 & = & \frac{v_0 ^ 2}{c_1 ^ 4}\int dz\,\int dz'\,\int dz''\,\int dz'''\,\int dz''''\ , \nonumber\\ & & { } \qquad\times\left[\partial_z^2\psi_1(z)\right]\gamma(z - z')\gamma(z'-z '' ) \nonumber\\ & & { } \qquad\times\gamma(z''-z''')\gamma(z'''-z'''')\partial_{z''''}^2\psi_1(z '' '' ) . \nonumber\\ & & \label{b3}\end{aligned}\ ] ] from the symmetry of the integrands in these expressions , and since @xmath192 and @xmath173 are positive operators , we expect all coefficients @xmath193 , @xmath194 , and @xmath195 to be positive . this implies a supercritical bifurcation of the collective migration speed at the onset of collective motion , which is in line with our simulations in two spatial dimensions . however , since @xmath196 generally depends on @xmath60 , the coefficient @xmath195 might be renormalized from the @xmath60-dependence of @xmath193 and @xmath194 . directly at the threshold for the onset of collective motion , we may assume a harmonic functional dependence of the form @xmath197 . when we calculate the coefficients @xmath193 , @xmath194 , and @xmath195 from eqs . ( [ b0])([b3 ] ) , we obtain from eq . ( [ eqvm ] ) for the collective migration speed directly above onset : @xmath198^{\frac{1}{2}}.\ ] ] however , higher harmonics in the expression for @xmath196 and its @xmath60-dependence will affect this result . 97ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )
the presence of a tight correlation between the rest - frame optical color and luminosity of early - type galaxies , the so - called color - magnitude relation ( cmr ) or red - sequence , is well established in the literature . the relation implies that the properties of the stellar populations in early - type galaxies are strongly dependent on the total stellar mass of the system . the tight correlation observed along this relationship ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) requires a strong coherence between both the age and metallicity of stellar populations present in early - type galaxies . measurements of the slope , zero - point , and dispersion around the color - magnitude relation probe the star formation histories of early - type galaxies and can provide physical insight into galaxy formation and evolution . one possible explanation for the slope of the cmr is that massive , and thus more luminous , galaxies retain more metals than less massive ones . the deeper potential wells of massive galaxies may allow fewer metals to be expelled in supernova driven winds @xcite . alternatively , @xcite have shown that the color - magnitude relation can be reproduced in hierarchical merging models if the epoch of mergers occurs at very early times . since the age of a stellar population is degenerate with its metallicity on the cmr , however , metallicity may not be the sole parameter behind the relationship @xcite . while both age and metallicity effects may create the cmr , studies of the evolution of the relationship toward high - redshift argue that the relation is driven primarily by metallicity and can not be generated by age effects alone @xcite . the color - magnitude relation of galaxies in clusters has been studied extensively . using high precision @xmath2 and @xmath3 photometry , @xcite determined the dispersion around the cmr for early - type galaxies in coma and virgo to be small , @xmath4 milli - mag ( mmag ) . this small scatter implies early - type galaxies in clusters must be uniformly old with formation epochs earlier than a @xmath5 . @xcite found @xmath6 mmag for clusters at @xmath7 ; this scatter , combined with the early age of the universe at the observed epoch , indicates that elliptical galaxies must have stopped forming stars in these clusters by a @xmath8 for a simple approximation to the color evolution of stellar populations . the cluster ms 1054 - 03 at a @xmath9 has a 29 mmag dispersion in the @xmath10 color @xcite , while @xcite and @xcite showed that the scatter in single clusters is only 40 and 30 mmag at @xmath11 . these small dispersions , seemingly independent of redshift , indicate that the coordination of star formation between galaxies in clusters is quite strong . using a sample of 158 clusters with @xmath12 drawn from the early data release of the sloan digital sky survey ( sdss ) , @xcite found that the cmr between clusters is very homogeneous . locally , the cmr is universal between clusters of various masses as well @xcite . while the mean color of red sequence galaxies in over - dense regions is only slightly redder than for similar galaxies in less dense regions and the slope of the cmr is independent of environment @xcite , several studies show that the scatter around the cmr is dependent on galaxy environment . a re - analysis of the data presented in @xcite indicates the dispersion around the cmr in clusters is 2-@xmath13 smaller than that for galaxies in groups or in the field @xcite . the scatter in galaxy colors in the core of the cluster cl 1358 + 62 at @xmath14 have @xmath15 mmag while early - types at large radii ( @xmath16 mpc ) have nearly double that value , indicating that the dispersion of galaxy colors in the field may be larger than that measured in clusters @xcite . early - type galaxies in the hubble deep field have a scatter of 120 @xmath17 60 mmag in rest - frame @xmath18 with an average redshift of 0.9 @xcite . while this measurement is rather uncertain , it is larger than that measured in clusters , further indicating that field galaxies may have less coordinated star formation than galaxies in clusters . a majority of the work on the cmr has focused on early - type galaxies in clustered environments with luminosities near @xmath19 . this is natural , as statistically significant samples are more easily selected in these over - dense regions and @xmath19 galaxies are more common than their more massive analogs . in this work , we extend the analysis of the color - magnitude relation to a sample of nearly 20,000 very luminous red field galaxies from the sloan digital sky survey . these galaxies are not selected to reside in clustered environments and are chosen to be quite luminous with @xmath20 . we create two galaxy samples from the sdss spectroscopic data . at low - redshifts ( @xmath21 ) we use galaxies from the entire survey region while at higher redshifts ( @xmath22 ) , where the galaxies are apparently fainter , we select galaxies from a 270 deg@xmath23 region which has been imaged an average of 10 times and up to 29 times . the repeated imaging of our moderate - redshift galaxies makes comparisons between the galaxy samples possible with similar fidelity . our sample allows us to probe a new area in parameter space luminous galaxies in the field and compare our results to those found for galaxies in clustered environments . the paper is organized as follows : 2 describes the sloan digital sky survey and the selection for the galaxies used here . in 3 , we present our measurements of the slope and scatter of the color - magnitude relations . we discuss simple star formation history models in 4 and use our measurements of the scatter around the cmr as a constraint on the evolutionary history of massive early - type galaxies in the field . in 5 , we discuss the role environment plays on the cmr of galaxies . we construct the composite spectra of galaxies across the red sequence in 6 before closing in 7 . throughout this work , we use @xmath24 = @xmath25 km s@xmath26 mpc@xmath26 and ( @xmath27 , @xmath28 ) = ( 0.3,0.7 ) to calculate luminosities and distances . when calculating ages of stellar populations , we set @xmath29 . all magnitudes used here have been corrected for reddening using the @xcite dust maps . the sloan digital sky survey @xcite is imaging @xmath30 steradians of the northern sky through 5 passbands @xcite . the imaging is conducted with a ccd mosaic in drift - scanning mode @xcite on a dedicated 2.5 m telescope located at apache point observatory in new mexico . after the images are processed @xcite and calibrated @xcite , targets are selected for spectroscopy with two double - spectrographs mounted on the same telescope using an automated spectroscopic fiber assignment algorithm @xcite . two spectroscopic galaxy samples are created using the sdss imaging . the main galaxy sample @xcite is a complete , flux limited , sample of galaxies with @xmath31 . this cut is nearly five magnitudes brighter than the sdss detection limit of @xmath32 22.5 , and thus the photometric quantities for these galaxies are measured with signal - to - noise of a few hundred . the luminous red galaxy ( lrg ) sample @xcite selects luminous early - type galaxies out to @xmath33 with @xmath34 19.5 using several color - magnitude cuts in @xmath35 , @xmath36 , and @xmath37 . the combination of these two samples allows us to study the broadband colors of luminous field galaxies at moderate - redshifts with statistically significant samples . while the focal point for the sdss is a contiguous survey of the northern galactic cap , the sdss also conducts a deep imaging survey , the sdss southern survey , by repeatedly imaging an area on the celestial equator in the southern galactic cap . currently , objects in the 270 deg@xmath23 region have been imaged an average of 10 times and up to 29 times , resulting in improved photometric precision for faint galaxies . objects detected in each observational epoch were matched using a tolerance of 0.5 arcseconds to create the final coadded catalog used here . the photometric measurements from each epoch were combined by converting the reported asinh magnitudes @xcite into flux and then calculating the mean value . errors on each parameter are simply the standard deviation of the flux measurements . the southern survey is an ideal region to compare the properties of faint galaxies at moderate - redshift with brighter low - redshift analogs drawn from the entire sdss survey area with photometry of similar fidelity . several methods are used to measure galaxy fluxes in sdss . below , we briefly describe the two flux measurements used throughout this paper . the petrosian ratio , @xmath38 , the ratio of the local surface brightness at radius @xmath39 to the average surface brightness at that radius , is given by @xmath40}{\left . \int_0^\theta dr ' 2\pi \theta ' i(\theta ' ) \right/ [ \pi \theta^2]},\ ] ] where @xmath41 is the azimuthally averaged surface brightness profile of a galaxy and @xmath42 and @xmath43 are chosen to be 0.85 and 1.25 for sdss . the petrosian flux is given by the flux within a circular aperture of @xmath44 , where @xmath45 is the radius at which @xmath38 falls below 0.2 . in sdss , @xmath45 is determined in the @xmath36 band then subsequently used in each of the other bands . this flux measurement contains a constant fraction of a galaxy s light , independent of its size or distance , in the absence of seeing . more details of sdss petrosian magnitudes can be found in @xcite , and @xcite . for each galaxy imaged by sdss , two seeing - convolved models , a pure @xcite profile and a pure exponential profile , are fit to the galaxy image . the best - fit model in the @xmath36 band is used to measure the flux of a galaxy in each of the other bands . these model magnitudes are unbiased in the absence of color gradients and provide higher signal - to - noise ratio colors than petrosian colors . a more complete description of model magnitudes is given in @xcite . throughout this paper , we use petrosian magnitudes when calculating galaxy luminosities while model magnitudes are used to determine galaxy colors . we calculate photometric @xmath46-corrections using the method of @xcite ( ` kcorrect v3_2 ` ) . this program constructs a linear combination of carefully chosen spectral templates in order to best match the observed photometry at the measured redshift of the galaxy . the rest - frame colors of these best - fit spectra are used to derive corrections to the observed galaxy colors . with this method , it becomes simple to transform between bandpasses and , if necessary , photometric systems given a detailed understanding of the filter characteristics . after calculating @xmath46-corrections with ` kcorrect ` , we remove any mean @xmath46-corrected color versus redshift trends with a low order polynomial to account for the passive evolution of stellar populations between observed epochs . this evolutionary correction is normalized to a redshift of 0.3 , near the median of the lrg redshift distribution . the color evolution correction is small ( less than 2% for all bands in both samples ) and does not affect any of the results presented here . we further assume a galaxy dims by one magnitude per @xmath47 change in redshift in the @xmath35-band due to the passive evolution of its stellar populations . again , this correction is normalized to @xmath48 . in the remainder of this paper , all of the color and luminosity measurements refer to quantities which have been adjusted for the @xmath46-correction and evolution . cc @xmath49 & 4026 + @xmath50 & 5307 + @xmath51 & 6441 + @xmath52 & 7687 + @xmath53 & 3408 + @xmath54 & 4494 + @xmath55 & 5454 + @xmath56 & 6509 + throughout this paper , we work in the @xmath49 , @xmath50 , @xmath51 , @xmath52 and @xmath53 , @xmath54 , @xmath55 , @xmath56 systems for the low- and moderate - redshift galaxy samples , respectively . here , the notation @xmath57 indicates the reported ab magnitudes are derived through a standard @xmath35 filter that has been blueshifted by @xmath58 . for a filter system which has been shifted by @xmath58 , the @xmath46-correction for galaxies at that redshift is trivial , @xmath59 , and is independent of the galaxy spectral energy distribution . in choosing shifts which nearly match the median redshift of each sample , we minimize the uncertainty introduced through the @xmath46-corrections @xcite . table 1 lists the effective wavelengths for each of the bandpasses used here . in order to investigate the color - magnitude relation of massive galaxies in the range @xmath60 , we create two samples from the sdss spectroscopic database to be analyzed separately ; figure [ sample ] shows the selection cuts used to define each of these samples . from the main sample of galaxies , we extracted 16622 galaxies in the redshift range @xmath21 satisfying the rest - frame color cut @xmath61 which approximates a cut of constant stellar mass @xcite . this selection criterion results in a cut diagonally across the red sequence , which could bias our measurements of the scatter in red sequence colors and will greatly affect any slope measurement of the color - magnitude relation . in order to avoid biased slope measurements and test the consistency of our scatter measurements , we also construct a sample of 5100 galaxies in the luminosity range @xmath62 with @xmath63 for comparison ( shown by the dashed region in figure [ sample ] ) . we further select 2782 lrg galaxies from the sdss southern survey in the redshift range @xmath22 which satisfy @xmath64 and have been imaged 6 or more times . equation 3 is the same rest - frame color cut applied to the low - redshift galaxies ( equation 2 ) shifted to the @xmath65 color the effective wavelengths of the filters that define each of these colors are similar , and thus these colors probe similar regions in the rest - frame galaxy spectrum ( see table 1 ) . of these 2782 sample galaxies , 62% were imaged more than 9 times while 20% of the galaxies were imaged more than 12 times in the sdss southern survey . as in the low - redshift sample , we adopt a simple constant color cut , @xmath66 and @xmath62 , to define a subsample of 1380 high - redshift galaxies to test the consistency of our scatter measurements and allow us to determine the slope of the color - magnitude relation without a strong cut across the red sequence . we find no differences within our quoted error between the dispersion around the cmr measured from galaxies defined using the stellar mass cut and the constant color cut ; throughout the remainder of this paper , scatter measurements will be based upon galaxies drawn in the first manner while all slope measurements are referenced to samples created using the latter method . figure [ zhist ] shows the redshift distribution of the galaxies used to measure the scatter around the color - magnitude relation in this paper . in detail , the red sequence is populated by elliptical and s0 galaxies as well as late - type galaxies reddened by dust . the arrow in the lower right of figure [ sample ] shows the reddening vector given by the @xcite extinction curve . in order for a dusty late - type galaxy to scatter into our sample , it must have a large unobscured luminosity as we only consider the most luminous red - sequence galaxies in this work . there is a clear paucity of very luminous blue galaxies in figure [ sample ] and thus the contamination from dusty spiral galaxies is likely small in our sample . furthermore , @xcite constructed the average spectrum of massive galaxies , such as those used in our analysis , and found that only 5% of the galaxy spectra in the @xmath67 luminosity range were contaminated by emission lines uncharacteristic of early - type galaxies . in the remainder of this paper , we will use the red - sequence and early - type classifications synonymously . throughout this paper , we will refer to the galaxies considered here as field galaxies as they are not specifically selected to reside in clustered environments . it is important to note , however , that galaxies which are of similar color and luminosity as those in our sample tend to lie in over - dense environments @xcite . @xcite found the clustering strengths and mean separations of lrgs are comparable to those of poor clusters or rich groups . for comparison , of the lrgs in the volume surveyed by @xcite to find galaxy clusters with 13 or more detected cluster members , 16% are projected less than 500 kpc from the inferred location of a cluster and within 0.05 of the photometric redshift for that cluster . a detailed examination of the density dependence of the scatter around the cmr is conducted in 5 . when determining the slope of the color - magnitude relation , we only consider the galaxies selected with our simple luminosity and constant color cuts as any slope measurements based on samples with a strong cut across the red sequence will be quite biased . the galaxy samples considered in this work only span a small range of luminosities and thus are not ideally suited for measurements of the slope of the cmr we simply use our measurement of this quantity to remove the mean relationship before determining the scatter around it . the slopes and zero - points for the cmr for several colors are listed in table 2 . figure [ fig : slopeplot ] shows the measured slopes of the cmr as a function of the effective wavelengths of the two bandpasses used to define the color . the slope in bluer colors is more pronounced than that for redder bands which tend to show small or negligible slopes . this trend is consistent with a metallicity sequence as the primary driver along the red - sequence . as noted by @xcite using models from @xcite , the flattening of the color - magnitude relation toward redder colors can also be reproduced by an age trend along the red sequence but only if a very specific relationship between galaxy mass and star formation time holds with very little scatter . in the bluest colors , which have been used historically in the literature , the slopes measured here are in general agreement with past work ; for comparison , @xcite report a slope of @xmath68 mag mag@xmath26 in the @xmath69 color for a large sample of galaxies at @xmath70 . rcrrcrrcr[hb ] @xmath71 & 1.162 & @xmath72 & & 1.162 & @xmath73 & & 1.164 & @xmath74 + @xmath75 & 0.449 & @xmath76 & & 0.449 & @xmath77 & & 0.451 & @xmath78 + @xmath79 & 0.329 & @xmath80 & & 0.329 & @xmath81 & & 0.329 & @xmath82 + @xmath83 & 1.611 & @xmath84 & & 1.610 & @xmath85 & & 1.612 & @xmath86 + @xmath87 & 1.750 & @xmath88 & & 1.750 & @xmath89 & & 1.754 & @xmath90 + @xmath91 & 0.613 & @xmath92 & & 0.612 & @xmath93 & & 0.614 & @xmath94 + @xmath95 & 0.379 & @xmath96 & & 0.377 & @xmath97 & & 0.381 & @xmath98 + @xmath65 & 0.991 & @xmath99 & & 0.980 & @xmath100 & & 0.983 & @xmath101 + figure [ fig : sigmaresid ] illustrates the typical scatter about the mean color - magnitude relation for the @xmath75 and @xmath91 colors ; the measured dispersion about this relationship is quite small . the second column of table 3 lists the measured dispersion around the cmr for each of the colors used in this study . for each color studied , we fit a gaussian to the observed color distribution using poisson errors on each color bin and define the dispersion of the best - fit gaussian to be the scatter in galaxy colors on the red sequence . by fitting the histograms rather than calculating the standard deviation directly , we place more weight on the core of the distribution compared to the wings . thus , our scatter measurements will not be heavily affected by any non - red - sequence interlopers in the wings of the color distributions . as this method could potentially depend on the binning used , we are careful to vary the bin size for each fit and ensure the measured scatter is not dependent on our choice of bin size . it should be noted that past work has used the standard deviation to quantify the scatter in galaxy colors . while the standard deviation and gaussian widths may give systematically different dispersions in the case of non - gaussian distributions , figure [ fig : sigmaresid ] shows the color distribution on the red sequence is well described by a gaussian and thus the two measurements are comparable . bootstrap resampling of the data ( e.g. * ? ? ? * ) was employed to determine the uncertainty on our scatter measurements . it is important to note that the scatter measurements in the second column of table 3 represent upper limits to the intrinsic scatter in galaxy colors as these values are not corrected for systematic or experimental effects , which could be significant . in order to investigate the level of skewness of the observed color distribution , we define @xmath102 to be the absolute value of the color difference between the @xmath103 quantile and the median of the distribution the level of asymmetry in the distribution is then estimated using @xmath104 for this statistic , a maximally skewed distribution would have @xmath105 such that a tail of galaxies to the red would result in @xmath106 . a symmetric distribution would have @xmath107 . we adopt this approach rather than a traditional third moment calculation as the calculation of higher moments of a distribution is strongly dependent on the wings of the distribution where contamination may be important . for the histograms shown in figure [ fig : sigmaresid ] , we find that @xmath108 for the @xmath75 color and @xmath109 for the @xmath91 color . both of these measurements reflect the slight over - abundance of red galaxies compared to a gaussian in both panels of figure [ fig : sigmaresid ] . this red tail is due to the diagonal cut across the red sequence imposed by our stellar mass selction criterion ; red galaxies are selected to lower luminosities than blue galaxies . since the luminosity function of galaxies is steeply rising toward lower luminosities in this range , more red galaxies are selected than blue galaxies . for comparison , we measure @xmath110 for the @xmath75 color and @xmath111 for the @xmath91 color when the simple luminosity cuts are used to select our sample galaxies . since our intermediate redshift sample is drawn from regions that have been repeatedly imaged by the sloan digital sky survey , we can assess the level to which the dispersion we observe in the color - magnitude relation is dependent on measurement noise . if the scatter is contaminated by measurement noise , we would expect the squared scatter to decrease inversely with the number of observations : @xmath112 figure [ fig : richardson ] shows the results of this test ; the methods used for the ensemble sample were used in the same manner to calculate the scatter and error for each bin . in each bin of @xmath113-measurements , we include galaxies which were imaged a total of @xmath113 times , and thus each bin contains a unique subsample of galaxies . for the moderate - redshift sample , the third and fourth columns in table 3 lists the instrumental contribution and intrinsic galaxy scatter predicted by extrapolating the fits to equation 2 to infinite observations for each of the three colors . cccccc[hb ] @xmath71 & @xmath114 & @xmath115 & @xmath116 & @xmath117 & @xmath118 + @xmath75 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 + @xmath79 & @xmath124 & @xmath125 & @xmath125 & @xmath126 & @xmath127 + @xmath83 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 + @xmath87 & @xmath133 & @xmath134 & @xmath135 & @xmath136 & @xmath137 + @xmath91 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 + @xmath95 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath147 + @xmath65 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 + while the intrinsic dispersion around the color - magnitude relation at moderate redshift can be extracted from the measured scatter based on subsamples of galaxies with different number of measurements , the low - redshift galaxies do not allow for this test as most galaxies in this sample have been imaged only once . in order to estimate the intrinsic scatter in galaxy colors for this sample , we identified 5839 of our low - redshift galaxies that have been imaged more than once under photometric conditions . we adopt the average root - mean squared variation of each photometric measurement as the instrumental error for each color of interest . we find the mean measurement uncertainty to be 17.6 , 15.1 , and 20.9 mmag in the @xmath71 , @xmath75 , and @xmath79 bands , respectively . the internal dispersion in each color is the difference , in quadrature , between the measured scatter and inferred mean uncertainty in our single pass photometry . for the low - redshift galaxies , the third and fourth columns of table 3 lists the observational error and intrinsic scatter determined by this correction . while our estimate of the instrumental contribution to the observed color dispersion corrects for uncorrelated errors on each independent measurement , we do not correct for systematic errors common to all observations which could cause an overestimation of the intrinsic dispersion in galaxy colors . figure [ fig : sigmaplot ] shows our scatter measurements as a function of the effective wavelength of the filters used to construct each color . the scatter about the cmr is clearly wavelength dependent ; the bluest colors , which are most affected by small changes in the age , metallicity , or dust attenuation of the system , have systematically larger dispersions than the reddest colors where changes in metallicity and age have less impact . here , we consider three simple star formation histories and derive the range of permitted parameters for these histories following a method similar to that of @xcite . in these toy models , we will assume that only age variations between red - sequence galaxies creates the scatter in early - type galaxy colors . in reality , the scatter across the cmr is likely driven by some combination of age , metallicity , dust , and possibly @xmath153-enhancement variations across the red - sequence . these models will thus illustrate the broadest range of ages allowed by our measurements ; the inclusion of any variation in the other galaxy properties will only constrain the age more strongly . in this section , we will primarily consider the high - redshift sample , as these are the intrinsically youngest galaxies in our sample and hence should provide the tightest constraints on the past star formation history of massive early - type galaxies . an identical analysis of the low - redshift sample provides results consistent with those presented below . in all three model cases , the low - redshift galaxies provide looser constraints than the high - redshift sample ; star formation is allowed to continue to slightly later times . first , we consider a star formation history composed of a single delta - function burst . the probability of a burst occurring is distributed uniformly between the earliest allowed time for galaxy formation , @xmath154 , set to @xmath155 , and @xmath156 , the time at which the youngest early - type galaxies form . using a grid of monte - carlo simulations with 5000 galaxies per realization , we create samples of galaxy spectra synthesized using the method of @xcite with solar metallicity , a @xcite initial mass function , and varying values of @xmath156 . we find that realizations allowing galaxies younger than 8.8 gyr ( @xmath157 ) generate color distributions with standard deviations larger than that measured . we perform a similar test using a scheme in which 80% of the stars in a galaxy are formed in a single burst at @xmath155 and 20% are formed in a second burst which occurs at a random time between @xmath155 and a minimum age , as done by @xcite . for this scenario , the secondary bursts must occur between @xmath158 to reconstruct the observed scatter in the color - magnitude relation . for bursts that generate less than 5% of a galaxy s stars , the time scale for the secondary burst is unconstrained - small amounts of star formation can occur at late epochs without violating the constraints placed by our measurements . finally , we create a grid of models in which star formation begins at @xmath155 and continues at a constant rate until it is truncated at @xmath159 . for each realization , we allow @xmath159 to range from @xmath155 to a critical cutoff time at which all galaxies have stopped forming stars . star formation in this manner must have ended by @xmath160 in order to reproduce the observed scatter in the color - magnitude relation at @xmath161 . the allowed range of parameters and the distribution of galaxy colors found in each model are shown graphically in figure [ fig : models ] . to allow comparison with figure [ fig : sigmaresid ] , typical observational errors have been included in the color distributions in the lower panels of the figure . interestingly , all of these models predict similar scatter trends to those seen in figure [ fig : sigmaplot ] . in all three cases , the scatter in the @xmath87 color is nearly double that predicted for the @xmath91 and @xmath95 colors . while each of the three models succeeds in reproducing the observed scatter in early - type galaxy colors , none of the models adequately recreates the color distribution measured from real galaxies , a sign that the models chosen do not perfectly track the true evolution of the star formation history of early - type galaxies . the @xmath162 parameter , as defined in equation 4 , for each of the model distributions is shown in figure [ fig : models ] . the three models produce color distributions with @xmath163 compared to the value of @xmath164 calculated for the observed sample of galaxies at the same redshift . the difference between the observed skewness and that determined from our simple models reiterates the mismatch between the observed and predicted color distributions . in our simulations , we assume all galaxies have the same metallicity and no dust attenuation . also , the onset of star formation is uniform in the second two models . in reality , these parameters are not constant , and thus the sharp truncations seen on the red edges of model color distributions will be blurred by the addition of other complications to the models . however , any of these corrections will increase the observed scatter in galaxy colors thus driving the last epoch of star formation needed to satisfy our constraints to earlier times . a majority of the work on the color - magnitude relation of early - type galaxies has focused on cluster galaxies at various redshifts . while some work ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) tackles the problem in the field , these studies suffer from small numbers of galaxies . the measurement of the scatter around the cmr presented here , on the other hand , focuses on the most massive field galaxies at moderate redshifts using a statistically significant sample . we can now , for the first time , statistically compare the dispersion around the cmr in both clustered and field environments . in order to explore the environmental dependence of the color - magnitude relation within our data set , we use sdss imaging of 5305 deg@xmath23 to identify a comparison sample of 16 million normal galaxies down to @xmath165 . at the location of each lrg galaxy in our sample , we count the number of neighboring galaxies within a @xmath166 mpc ( proper ) aperture which have colors similar to those expected for an @xmath19 galaxy on the red sequence , and luminosities in the @xmath167 to @xmath168 range ; the details of the luminosity and color cuts are given in @xcite . using the number of neighbor galaxies as a proxy for the environment of each lrg in our sample , we divide each of our galaxy sets into a clustered subsample and a low density subsample . we define lrg galaxies with more than 5 ( 7 ) neighboring galaxies in the low- ( moderate- ) redshift samples to be in clustered environments while a field subsample is comprised of galaxies with fewer neighbors . this threshold was chosen such that the clustered sample in both redshift bins composes 30% of the total galaxy sample at that redshift . to place these values in physical context , at @xmath169 ( @xmath161 ) there are 25.3 ( 170 ) tracer galaxies per square degree . given the angular diameter distance of 398 ( 712 ) @xmath170 mpc to that redshift , this yields a total of 1.65 ( 3.46 ) background galaxies per aperture on average . the average number of neighbors in the field sample is 2.3 ( 3.9 ) galaxies while the clustered galaxies have a mean of 9.1 ( 11.4 ) neighbors . thus our clustered samples are over - dense by a factor of 4 ( 3 ) compared to the field galaxies . @xcite found that the slope and zeropoint of the color - magnitude relation do not depend strongly on the local galaxy environment . we find the same result ; table 2 lists the measurements of the slope and zeropoint of the relationship for the ensemble sample as well as for the field and clustered subsamples . the variation between the two samples is small . in the following calculations , we use the ensemble slope and zeropoint measurements as these are measured with the highest signal - to - noise . the results are unaffected if the individual measurements of the slope and zeropoint are used for each subsample . the fifth and sixth columns of table 3 list the measured scatter in galaxy colors for the field and clustered galaxies respectively . in all of the bands , the scatter for the field subsample is larger than that in clustered environments . if we assume the scatter introduced by instrumental effects is the same for both the field and clustered subsamples , the same correction determined for the ensemble galaxy sample can be applied to these subsamples , as well . with this correction , we find that the scatter in galaxy colors in dense environments is @xmath171 smaller than for that of isolated systems in both redshift intervals studied here . our clustered galaxies reside in a range of densities , not simply rich clusters , as have often been used for similar studies , and thus we likely underestimate the significance of the difference in scatter between the field and strongly clustered environments . furthermore , our field sample is not completely composed of isolated galaxies as the average number of neighbor galaxies in this sample is larger than the average density of background galaxies . again , this leads to an underestimation of the true difference in the dispersion around the color - magnitude relation for field and clustered galaxies . the larger scatter in field galaxy colors supports previous claims of an environmental dependence for the scatter around the color - magnitude relation for early - type galaxies ( e.g. * ? ? ? * ; * ? ? ? * ) , though the effect measured here is weak . comparisons of the scatter in galaxy colors for clustered galaxies often draw on galaxies from a single cluster whereas our clustered sample is created by combining galaxies from a number of clustered environments . it is possible that the star formation histories of galaxies in a single cluster are more highly coordinated than the star formation histories between different clusters , thus the scatter measured using galaxies from a single cluster would be smaller than when several clusters are combined . as the fraction of blue galaxies is larger in the field compared to clustered environments , the field subsample may have a larger contamination fraction compared the the clustered subsample . this would also cause an increased scatter in field galaxies colors compared to the galaxies in more dense environments . having measured the scatter in galaxy colors , we next investigate the spectral differences between galaxies by examining the average spectrum of massive early - type galaxies across the red sequence . the composite early - type galaxy spectrum has been investigated as a function of environment , luminosity , redshift , size , and velocity dispersions @xcite ; here , we construct the composite spectrum for red sequence galaxies of different mean colors . since the galaxies in our low - redshift sample are brighter , and thus have higher quality spectra , and are more numerous than the higher redshift galaxies , we only consider these galaxies in our analysis . to construct the average spectrum of the galaxies in each color bin , we first shift the spectra according to the observed redshift and then normalize the observed sdss spectrum by the flux at rest - frame 5000 @xmath172 . we augment the sdss pixel mask to include all pixels within 10 @xmath172 of the bright sky lines at 5577 , 5890 , 5896 , 6300 , 6364 , 7245 , and 7283 @xmath172 . since the galaxies of interest are in the redshift range @xmath21 , the masked pixels occur at different rest - frame wavelengths for each galaxy , leaving the mean spectrum unaffected . we then find the number - weighted mean of the spectra , ignoring all masked pixels in the mean . galaxies are divided into two distinct color bins red galaxies having @xmath71 colors between 1-@xmath13 and 3-@xmath13 redward of the mean and a blue sample in the same range blueward of the mean . the red galaxy bin contains 1321 galaxies with a mean @xmath71 color of 1.20 and @xmath173 while the blue sample contains 1161 galaxies with mean color and luminosity of 1.09 and @xmath174 . figure [ fig : coadd](a ) shows the coadded spectra of the red and blue galaxies . the gross properties of each composite spectrum are similar . figure [ fig : coadd](b ) shows the difference between the blue and red spectra . this difference spectrum shows clear signs of enhanced balmer absorption in the blue galaxies compared to the red composite , indicating that , on average , the blue galaxies contain more young stars , in agreement with past studies ( e.g. * ? ? ? * ; * ? ? ? * ) . while the red composite shows stronger [ n @xmath175ii ] emission lines relative to the blue composite , no other prominent emission lines , such as [ o @xmath175ii ] , [ o @xmath175iii ] , and h@xmath153 are present ; this emission line difference is likely due to a larger fraction of agn in the red galaxies compared to the blue we consider two possible evolutionary scenarios to explain the spectral differences between the blue and red galaxies on the red sequence : pure age ( i.e. passive ) evolution of a single age stellar population and a recent epoch of star formation mixed with an existing old population . in the first scenario , we fit each composite spectrum with a solar metallicity , single age , stellar synthesis model created using the method of @xcite . the red and blue spectra are best fit by 6.3 gyr and 3.7 gyr populations , respectively . in the second scenario , we find the blue composite spectrum is well fit by a 6.3 gyr old stellar population mixed with a 1 gyr population containing 3.5% of the stellar mass and we model the red composite with a simple 6.3 gyr population . the difference spectra for both scenarios are shown in figure [ fig : coadd](b ) . while each of the two model hypotheses would suggest quite different evolutionary histories for massive early - type galaxies , the predicted difference spectra are remarkably similar . the model consisting of a recent burst of star formation slightly over - estimates the continuum in the difference spectrum near 3900 @xmath172 and thus provides a marginally poorer fit to the data than the passive evolution model . the residuals between the data and the model difference spectra are shown in figure [ fig : coadd](c ) . both of the models fail to match the strengths of the @xmath176 and mg @xmath177 features near 4140 and 5160 @xmath172 . @xcite interpret these differences between observations and synthetic spectra as a enhanced @xmath153-to - iron ratio present in the galaxies but not included in the synthetic models . also , several weak iron features are evident in the two model spectra near 4300 and 5300 @xmath172 which are missing in the observed difference spectrum . in a pure - age model for the evolution across the red sequence , the enhanced blue continuum provided by the hotter stars in the young population dilutes the metal lines seen in the older population . this dilution leads to differences in the strengths of the metal lines of the two population models , thus causing the residual iron features in both model difference spectra . the absence of these residuals in our observed difference spectrum indicates that a pure - age model for the spectral differences between red and blue galaxies on the red sequence is an incomplete model . in order to match the variations of the spectral features across the red sequence , a more detailed model is required . this model should include both age and metallicity variation and , more importantly , non - solar abundance ratios . until spectral synthesis models with realistic @xmath153-enhancement prescriptions are available , more quantitative analysis of our observed difference spectrum is difficult as the interplay between age , metallicity , and @xmath153-to - iron ratio variations must be included for a full understanding of the physical differences between these galaxies . another clear difference between the models and the data is the strong na d feature near 5892 @xmath172 . here , the red composite spectrum has considerably more na d absorption than predicted by the stellar synthesis models . as the model spectra do not contain absorption due to the interstellar medium , this discrepancy may indicate that the red galaxies contain more neutral gas than their blue counterparts . we present new multi - color observations of a large sample of galaxies in the range @xmath1 gathered from the sloan digital sky survey . these galaxies represent the most massive galaxies in the universe and are not selected to be in clustered environments . past studies of the color - magnitude relationship have focused on @xmath178 galaxies in clusters , thus our work focuses on an area of parameter space previously unexplored . using the robust photometry of bright galaxies provided by the sdss as well as multiple observations of moderate - redshift galaxies , we are able to construct the color - magnitude relations for massive field galaxies in several colors . while our data only span one magnitude in luminosity , and thus are not ideal for in - depth analysis of the slope and zeropoint of the cmr , our large sample of galaxies allows us to measure the dispersion about this relationship quite well . the scatter around the red sequence is 20 mmag in the reddest bands while the bluest colors have nearly double this value . we construct several toy model star formation histories in order to place constraints on the evolutionary processes at work in massive galaxies . we find that the majority of star formation in these early - type galaxies must have occurred before @xmath179 , in agreement with other authors . for secondary bursts of star formation with strengths less than 5% of the total stellar mass of the system , our data are consistent with bursts of star formation at any epoch . however , one should note that none of the toy models reproduce the observed distribution of galaxy colors . furthermore , we make comparisons between our new measurement of the scatter around the color - magnitude relation for massive field galaxies and early - type galaxies in clusters . an internal comparison shows that the scatter around the cmr in dense environments is @xmath180 smaller than that measured using more isolated galaxies . the composite spectra of lrg galaxies redward and blueward of the mean galaxy color are physically distinct . the blue galaxies show stronger balmer absorption than the red galaxies , consistent with the presence of a younger stellar population . simple models slightly favor passive evolution of stellar populations as the cause of the observed differences in spectra across the red sequence compared to a more recent injection of young stars onto an old stellar population , though the difference between the predicted spectra from these models is small . we would like to thank michael blanton , scott burles , douglas finkbeiner , david hogg , david schlegel , john moustakas , amelia stutz , and jane rigby for software and helpful comments during the preparation of this paper . we further thank the anonymous referee for a thorough and critical reading of our paper which lead to substantial improvements to the text . this research made use of the nasa astrophysics data system . rjc is funded through a national science foundation graduate research fellowship . dje is supported by nsf grant ast 339020 and an alfred p. sloan research fellowship . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the korean scientist group , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . abazajian , k. , et al . 2003 , , 126 , 2081 abazajian , k. , et al . 2004 , , 128 , 502 abazajian , k. , et al . 2005 , , 129 , 1755 andreon , s. 2003 , , 409 , 37 arimoto , n. & yoshii , y. 1987 , , 173 , 23 bahcall , n. a. , et al . 2003 , , 148 , 243 baldry , i. k. , glazebrook , k. , brinkmann , j. , ivezi ' c , . , lupton , r. h. , nichol , r. c. , & szalay , a. s. 2004 , , 600 , 681 bell , e. f. , & de jong , r. s. 2000 , , 312 , 497 bernardi , m. , et al . 2003 , , 125 , 1882 bernardi , m. , sheth , r . k. , nichol , r. c. , schneider , d. p. , & brinkmann , j. 2005 , , 129 , 61 blakeslee , j. p. , et al . 2003 , , 596 , l143 blanton , m. r. , et al . 2001 , , 121 , 2358 blanton , m. r. , et al . 2003 , , 594 , 186 blanton , m. r. , lin , h. , lupton , r. h. , maley , f. m. , young , n. , zehavi , i. , & loveday , j. 2003 , , 125 , 2276 blanton , m. r. , et al . 2003 , , 125 , 2348 bower , r. g. , lucey j. r. , & ellis , r. s. 1992 , , 254 , 601 bressan , a. , chiosi , c. , & tantalo , r. 1996 , , 311 , 425 bruzual , g. & charlot , s. 2003 , , 344 , 1000 de lucia , g. , springel , v. , white , s. d. m. , croton , d. , & kauffmann , g. 2005 , arxiv astrophysics e - prints , arxiv : astro - ph/0509725 de vaucouleurs , g. 1948 , annales dastrophysique , 11 , 247 eisenstein , d. j. , blanton , m. , zehavi , i. , bahcall , n. , brinkmann , j. , loveday , j. , meiksin , a. , & schneider , d. 2005 , , 619 , 178 eisenstein , d. j. , et al . 2003 , , 585 , 694 eisenstein , d. j. , et al . 2001 , , 122 , 2267 ellis , r. s. , smail , i. , dressler , a. , couch , w. j. , oemler , a. j. , butcher , h. , & sharples , r. m. 1997 , , 483 , 582 faber , s. m. 1972 , , 20 , 361 faber , s. m. 1973 , , 179 , 731 fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p. 1996 , , 111 , 1748 gladders , m. d. , lopez - cruz , o. , yee , h. k. c. , & kodama , t. 1998 , , 501 , 571 gunn , j. e. , et al . 1998 , , 116 , 3040 hogg , d. w. , finkbeiner , d. p. , schlegel , d. j. , & gunn , j. e. 2001 , , 122 , 2129 hogg , d. w. , et al . 2003 , , 585 , l5 hogg , d. w. , et al . 2004 , , 601 , l29 ivezi ' c , . , et al . 2004 , astronomische nachrichten , 325 , 583 kauffmann , g. & charlot , s. 1998 , , 294 , 705 kodama , t. 1997 , ph.d . thesis , univ . tokyo kodama , t. & arimoto , n. 1997 , , 320 , 41 kodama , t. , bower , r. g. , & bell , e. f. 1999 , , 306 , 561 larson , r. b. 1974 , , 166 , 585 larson , r . b. , tinsley , b. m. , & caldwell , c. n. 1980 , , 237 , 692 lugger , p. m. 1984 , , 278 , 51 l ' opez - cruz , o. , barkhouse , w. a. , & yee , h. k. c. 2004 , , 614 , 679 lupton , r. h. , gunn , j. e. , ivezi ' c , z. , knapp , g. r. , kent , s. , & yasuda , n. 2001 , asp conf . 238:astronomical data analysis software and systems x , 10 , 269 lupton , r. h. , gunn , j. e. , & szalay , a. s. 1999 , , 118 , 1406 matteucci , f. & tornambe , a. 1987 , , 185 , 51 mcintosh , d. h. , zabludoff , a. i. , rix , h. , & caldwell , n. 2005 , , 619 , 193 oconnell , r. w. 1980 , , 236 , 430 odonnell , j. e. 1994 , , 422 , 158 pier , j. r. , munn , j. a. , hindsley , r. b. , hennessy , g. s. , kent , s. m. , lupton , r. h. , & ivezi ' c , . 2003 , , 125 , 1559 press , w. h. 2002 , numerical recipes in c++ : the art of scientific computing by william h. press . xxviii , 1,002 p. : ill . ; isbn : 0521750334 , rose , j. a. 1985 , , 90 , 1927 salpeter , e. e. 1955 , , 121 , 161 sandage , a. & visvanathan , n. 1978 , , 225 , 742 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 smith , j. a. , et al . 2002 , , 123 , 2121 stoughton , c. , et al . 2002 , , 123 , 485 stoughton , c. , et al . 2002 , , 4836 , 339 strauss , m. a. , et al . 2002 , , 124 , 1810 terlevich , a. i. , kuntschner , h. , bower , r. g. , caldwell , n. , & sharples , r. m. 1999 , , 310 , 445 terlevich , a. i. , caldwell , n. , & bower , r. g. 2001 , , 326 , 1547 trager , s. c. , faber , s . m. , worthey , g. , & gonz ' alez , j. j. 2000 , , 119 , 1645 van dokkum , p. g.,franx , m. , kelson , d. d. , illingworth , g. d. , fisher , d. , & fabricant , d. 1998 , , 500 , 714 van dokkum , p. g. , franx , m. , fabricant , d. , illingworth , g. d. , & kelson , d. d. 2000 , , 541 , 95 van dokkum , p. g. , stanford , s. a. , holden , b. p. , eisenhardt , p. r. , dickinson , m. , & elston , r. 2001 , , 552 , l101 visvanathan , n. & sandage , a. 1977 , , 216 , 214 worthey , g. , trager , s. c. , & faber , s. m. 1995 , asp conf . ser . 86 : fresh views of elliptical galaxies , 86 , 203 york , d. g. , et al . 2000 , , 120 , 1579 zehavi , i. , et al . 2005 , , 621 , 22 zepf , s. e. , whitmore , b. c. , & levison , h. f. 1991 , , 383 , 524
discovery of the higgs boson as predicted by the standard model ( sm ) by the atlas and the cms experiments became the moment of triumph for particle physics @xcite . such a historic discovery together with decades of eletroweak precision data have well established the validity of sm up to accessable energies . however , there is no verified explanation of the origin of the small neutrino masses and no viable candidate for the dark matter in the sm . due to these unwavering issues , various extensions of sm have been proposed . the secret of neutrino masses may lie in some form of seesaw mechanism , where a sm singlet right - handed neutrinos with large majorana masses cause the light neutrino masses ( type - i seesaw ) @xcite or a sm weak scalar triplet with a tiny induced vacuum expectation value ( vev ) generates the small neutrino masses ( type - ii seesaw ) @xcite . if neutrino masses are generated as loop corrections , the masses will naturally be suppressed and such extensions of sm are both theoretically well motivated and phenomenologically viable @xcite . searches for a stable dark matter candidate have also been motivation for various extensions of the sm . some form of symmetry usually stabilizes the dark matter . simple discrete symmetries such as r - parity in supersymmetric models @xcite can perform an excellent job of preventing the particle from decaying . kaluza - klein parity @xcite in universal extra dimension models and t - parity in the littlest higgs models @xcite can stabilize the lightest particle ; turning them into promising dark matter candidates . a similar role is played by a @xmath5 symmetry for the case of inert doublet models @xcite or scotogenic models @xcite . sm extended by a scalar singlet carrying a discrete @xmath5 parity is yet another example for a simple dark matter model . instead of being an adhoc symmetry , this @xmath5 symmetry can be a remnant of the @xmath6 generator of @xmath7 grand unified theories ( guts ) @xcite . @xmath7 guts provide one of the most lucrative frameworks , where one can incorporate many of the aforementioned extensions of sm along with a beautiful unified picture of sm gauge couplings . among the classes of @xmath7 guts , supersymmetric versions have multiple features such as successful unification of gauge couplings and natural dark matter candidate owing to an automatic r - parity , while it solves the gauge hierarchy problem based on symmetry principle . in addition , supersymmetric models offer a mechanism for triggering electroweak symmetry breaking ( ewsb ) via radiative effects @xcite . in this scenario , the positive mass parameter of the higgs boson at high energy becomes negative at low energy due to the renormalization group flow which dictates how the parameters evolve with scale . the purpose of this paper is to explore the possibility of radiative ewsb breaking in non - supersymmetric ( non - susy ) extensions of the sm . since this is an attractive mechanism to trigger ewsb , checking its viability in non - susy models is of great interest . as we argue below , such a radiative ewsb may be necessary in certain unified theories which have two stages of symmetry breaking . in a general context a positive mass parameter for the higgs boson turning negative also enhances the available parameter space . in some extensions of sm , the higgs boson is a part of the larger multiplet , which breaks some higher symmetry . this occurs in trinification models based on @xmath8 gauge symmetry @xcite broken by two @xmath9 higgs multiplets . these multiplets contain sm singlet components which acquire large vevs breaking the gauge symmetry down to sm . the same @xmath9 multiplets also contain the higgs boson of the sm which should develop a negative squared mass to trigger ewsb . consistency of the high scale symmetry breaking , however , would demand that all physical higgs bosons , including the sm higgs , have positive squared masses at a high energy . one could introduce new higgs fields to break the electroweak symmetry in which case the model looses its minimality and predictivity . for such class of models one might employ radiative loop corrections to turn higgs mass parameter negative at low energy from positive value it obtained at high energy and thus cause eletroweak symmetry breaking . a second example is provided by a class of @xmath7 models with the symmetry breaking sectors containing @xmath10 along with either a @xmath11 or a @xmath12 @xcite where flavor mixing is induced by vector - like fermions in the @xmath13 representation . in such models , the sm singlet from @xmath10 acquires a gut scale vev , breaking @xmath7 down to @xmath14 . the @xmath10 also contains a sm higgs doublet which must have positive squared mass at the gut scale . this positive mass term can turn negative at low energy due to renormalization group flow . similar arguments can be applied for the case where a sm singlet of @xmath15-representation breaks @xmath7 down to sm @xcite . the higgs doublet is also part of @xmath15 , which should have a positive squared mass at the gut scale . in this paper we explore the possibility of radiative electroweak symmetry breaking in several popular extensions of the sm . the mechanism fails in the sm , as reviewed in sec . [ sec : sm and type - i ] . in type - i seesaw models which includes right - handed neutrinos to the spectrum of sm , radiative ewsb is not achieved - in fact the effect of @xmath16 fields is to provide _ positive _ corrections to the higgs mass parameter in evolving from high to low energies . the situation is different in type - ii seesaw models which contain a weak scalar triplet , if the mass of the triplet is around the tev scale . the key difference is the cross coupling between the sm higgs boson and an additional scalar field in the higgs potential . the need for this additional scalar field to be at the tev scale arises from the needed magnitude of the @xmath17 parameter : @xmath18 . as the correction to @xmath17 from the scalar cross coupling grows as @xmath19 of the new scalar , this scalar should not be much heavier than a tev , assuming that the quartic cross couplings are not extremely weak . radiative mass generation is a popular mechanism for neutrino masses where one assumes new scalars at the tev scale for lepton number violation . such models are amenable to radiative ewsb . dark matter models employing a scalar singlet or an inert doublet also exhibit radiative ewsb . finally , we propose and analyze a universal seesaw model wherein a new @xmath0 symmetry is broken at tev scale , which also shows radiative ewsb . in non - supersymmtric extensions of the sm such as the ones studied here , the gauge hierarchy problem has to be somehow solved . here we simply assume that this is done by fine - tuning . the dimensionful parameters of the sm extensions are given by @xmath20 where the @xmath21 donate mass parameters for additional scalar fields that may be present and @xmath22 is the cosmological constant . these dimensionful parameters may take , for reasons not understood , special values , rather than their natural values " which are of order the planck scale . once the scalar masses are set at these special ( or fine - tuned ) values , we assume that the corrections to @xmath17 arising from other particles present in the model do not exceed the physical mass of @xmath23 . a positive mass parameter turning negative via rge flow leads to dimensional transmutation as can be seen in a coleman - weinberg @xcite analysis of the effective potential . the rge evolution that we employ is in one to one correspondence with the effective potential , where the minimization is performed at a momentum scale close to the mass of the higgs scalar . the outline of the paper is as follows . in sec . [ sec : sm and type - i ] , we discuss the absence of such radiative ewsb in sm and type - i seesaw model . even though such a mechanism fails for type - i seesaw models , in sec . [ sec : type - ii seesaw ] we show that the presence of a tev scale weak triplet makes radiative ewsb a success for the case of type - ii seesaw models . in sec . [ sec : two - loop neutrino mass model ] , we show that for a two - loop neutrino mass model positive higgs mass parameter at a high energy scale turns negative in the renormalization group flow to low energy . in sec . [ sec : idm ] and sec . [ sec : scalar singlet dm model ] , we show that simple dark matter models such as inert doublet model and scalar singlet dark matter model also exhibit radiative ewsb . when the models have tev scale scalar(s ) coupled to the sm higgs boson . in sec . [ sec : universal seesaw ] , we study the radiative ewsb for a universal seesaw model . finally in sec . [ sec : conclusion ] we conclude . the mechanism of radiative ewsb occurs when the renormalization group flow of higgs mass parameter @xmath24 receives enough negative contribution from various parameters of the model turning the positive quantity into a negative one while evolving from high to low energies . unfortunately such can not be the case in the sm . the higgs potential of the sm is given by : @xmath25 and the renormalization group equation ( rge ) of mass parameter @xmath24 is given by @xcite : @xmath26 the evolution of the higgs mass parameter @xmath24 is dominated by the gauge couplings and the top quark yukawa couplings . however , these corrections are proportional to @xmath17 itself , which implies that a positive @xmath17 can not turn into negative @xmath17 in rge evolution making radiative ewsb an impossibility within sm . for type - i seesaw models , the additional lagrangian is given by : @xmath27 this part of the lagrangian manages to contribute in the renormalization group flow as it adds a new term to the rge of higgs mass parameter : @xmath28 unfortunately , the contributions coming from the @xmath16 fields make the situation worse as they only strengthen the positivity of the mass parameter as it evolves from high to low energies . one should also notice that if we want to use the criterion of naturalness , ie . the correction to the higgs mass parameter @xmath29 the scale of the majorana mass of the right - handed neutrino should not exceed @xmath30 @xcite . in supersymmetry , the radiative ewsb becomes successful @xcite as the scalar partners of the fermions contribute significantly in the renormalization group flow of the higgs mass parameter in the right direction , making the positive term negative as it evolves from high to low energies . similar incidents occur in other extensions of sm such as type - ii seesaw models where tev scale particle(s ) manages to dominate the renormalization group flow and turn the positive value of the higgs mass parameter into negative value triggering radiative ewsb . a common feature of these extensions is the presence of new scalar(s ) at the tev scale as we show in the next five sections . while type - i seesaw needs right - handed ( rh ) neutrinos which are neutral under the standard model gauge group with large majorana masses , the minimal type - ii seesaw mechanism requires the existence of a weak scalar triplet . the most natural source for such triplets is provided by the left - right symmetric theories which can be realized either at low energy or can be embedded in grand unified theories ( guts ) such as @xmath7 or @xmath31 . here we study the potential radiative ewsb scenario of the type - ii seesaw extension of the sm . we consider the possibility that the weak scalar triplet @xmath32 is the only low - energy remnant of the new physics beyond the sm and the neutral component @xmath33 acquires a very small induced vev at low energy . the sm electroweak doublet @xmath34 and the electroweak triplet @xmath35 are denoted by : @xmath36 where @xmath37 s are the pauli matrices . the most general renormalizable tree - level scalar potential is @xmath38\\ & + \lambda_4 \ ; \phi^\dagger \phi \ ; \tr ( \delta^\dagger \delta)+ \lambda_5 \ ; \phi^\dagger [ \delta^\dagger , \delta ] \phi + \ { \dfrac{\mu}{\sqrt{2}}\ ; \phi^t\ ; i \sigma_2 \ ; \delta^\dagger \phi + h.c.\}. \end{split}\ ] ] the weak triplet also generates a majorana mass term for the neutrinos through the yukawa lagrangian : @xmath39 with the vev of the electroweak doublet @xmath40 , an effective dimension 5 operator generates the neutrino masses through the small but non - zero induced vev , @xmath41 and /or @xmath42 is so small that @xmath43 as the electroweak triplet decouples . the neutrino mass matrix is given by : @xmath44 here one assumes @xmath45 so that @xmath46 is induced only after @xmath47 is generated . one also needs to realize the fact that integrating out the heavy scalar triplet in the tree level approximation will also have an effect on the sm higgs quartic coupling . the effective quartic coupling below the scale @xmath48 is given by : @xmath49 this is the connecting formula for the standard model quartic coupling @xmath50 at the scale @xmath48 . one needs to be careful while studying the solution to the set of rges of the parameters of a model . the parameters in the scalar potential have to satisfy certain conditions at all energy scales which ensures that the potential is bounded from below . for that purpose one must identify the necessary and sufficient conditions for boundedness of the potential . for type - ii seesaw models , the stability conditions for the potential to be bounded from below can be derived to be : 1 . 2 . 3 . @xmath51 this is compact set of constraints that is necessary and sufficient as we show in the appendix [ appendix : stability cond ] . for a less compact set of constraints the couplings of the lagrangian have to maintain these stability conditions upto the energy scale of new physics such as guts . using vertex corrections and the wave function renormalization factors , we can calculate the complete set of @xmath52-functions and renormalization group equations ( rges ) . we have also determined the rges for the mass parameters of the model which are related to the anomalous dimensions @xmath53 of the scalar masses by @xmath54 where @xmath55 and @xmath56 is the running scale . the set of rges for the mass parameters is given by : and @xmath57 in the rge of the mass parameter @xmath42 given in eq . ( 17 ) of ref . we also disagree with the coefficient in front of the mass parameter @xmath58 in the rge of @xmath59 given in eq . ( 18 ) of ref ] @xmath60 \mu^2_\phi + 6 \lambda_4 \mu_\delta^2 + 6|\mu|^2;\\ 16\pi^2 \dfrac{d \mu_\delta^2 } { d t } = & \left[\l(-\frac{18}{5}g_1 ^ 2 - 12g_2 ^ 2\r)+8\lambda_2 + 2\lambda_3 + 2\tr\left(\mathbf{y_\delta^\dagger y_\delta}\right)\right]\mu_\delta^2 + 4\lambda_4 \mu_\phi^2 + 2|\mu|^2;\\ 16\pi^2 \dfrac{d \mu } { d t } = & \left[\lambda_1 + 4\lambda_4 - 8\lambda_5-\frac{27}{10}g_1 ^ 2-\frac{21}{2}g_2 ^ 2 + 2 t + \tr\left(\mathbf{y_\delta^\dagger y_\delta}\right)\right]\mu;\\ \end{split}\ ] ] where @xmath61 .\ ] ] a complete set of rges for all the couplings of the lagrangian is given in the appendix [ appendix : rges for type - ii ] . with this set of rges we proceed towards its numerical solution . we can already see that the contribution from the cubic coupling @xmath42 and quartic couplings such as @xmath62 has the potential to drive the positive mass parameter @xmath17 at high energy scale towards a negative value at lower energy scale to trigger radiative ewsb . for a tev scale scalar triplet mass one realizes that to get the correct order of neutrino mass the cubic parameter @xmath42 needs to be very small ( @xmath63 gev ) , which makes the contribution of @xmath42 term in the rges of mass parameters irrelevant . this choice is natural , since quantum corrections to @xmath42 are proportional to @xmath42 itself - see eq . ( [ mass rges type - ii ] ) . to analyze the evolution of the mass parameters , one needs to solve the set of rges which in turn requires one to define the relevant couplings at some energy scale @xcite . in this case , all the sm gauge couplings , yukawa couplings and higgs quartic couplings were evaluated at two - loop level upto the energy scale corresponding to the scalar triplet mass . at energies above the triplet mass the gauge couplings were evolved continuously but with the updated rges given in the set of rges in eq . ( [ rges type - ii gauge ] ) . the quartic coupling of the sm electroweak doublet higgs has a discontinuity at the triplet mass scale due to the matching condition of the parameter given in eq . ( [ quartic matching condition ] ) . above the energy scale @xmath64 , the full set of rges was used to evolve all the parameters of the model . .quartic couplings and mass parameter values for the sample point used for the type - ii seesaw model in fig . [ fig : sample rge type - ii seesaw ] . [ cols="^,^,^,^",options="header " , ] .5 . in fig . [ fig : sub1vf ] the evolution of the absolute value of the sm higgs mass parameter @xmath65 and @xmath66 along with the masses parameters @xmath67 and @xmath42 have been plotted . the point at which @xmath68 touches the horizontal line corresponding to mass = @xmath69 , is the energy scale where radiative ewsb is triggered as the sign of the sm higgs mass - squared parameter @xmath70 switches from positive to negative while evolving from high to low energies . for the sample point , this happens around tev energy scale . note that the mass parameter of the scalar @xmath71 also turns negative around the renormalization energy scale @xmath72 indicating that the @xmath73 symmetry also gets broken radiatively . in fig . [ fig : sub2vf ] shows the evolution of all the quartic couplings of the model from planck scale down to weak scale . note that the model remains perturbative all the way for the selected sample point.,title="fig:",width=302 ] .5 . in fig . [ fig : sub1vf ] the evolution of the absolute value of the sm higgs mass parameter @xmath65 and @xmath66 along with the masses parameters @xmath67 and @xmath42 have been plotted . the point at which @xmath68 touches the horizontal line corresponding to mass = @xmath69 , is the energy scale where radiative ewsb is triggered as the sign of the sm higgs mass - squared parameter @xmath70 switches from positive to negative while evolving from high to low energies . for the sample point , this happens around tev energy scale . note that the mass parameter of the scalar @xmath71 also turns negative around the renormalization energy scale @xmath72 indicating that the @xmath73 symmetry also gets broken radiatively . in fig . [ fig : sub2vf ] shows the evolution of all the quartic couplings of the model from planck scale down to weak scale . note that the model remains perturbative all the way for the selected sample point.,title="fig:",width=321 ] to find the solution of the set of rges , we took a more simplified case where we kept all the yukawa coupling @xmath74 to be small and negligible and yukawa coupling @xmath75 . the numerical solution was hunted for the case where one of the eigenvalue of the scalar mass matrix @xmath76 corresponds to the sm higgs boson and another one corresponds to scalar boson of mass @xmath77 . here , we used @xmath78 gev just as an example as similar universal seesaw model @xcite was used to explain apparent diphoton excess @xcite which eventually became statistically insignificant @xcite . the vector - like fermion mass was kept around tev scale where the mass is approximated by @xmath79 . one such sample point is given by table [ table : sample point vector - like fermion model ] . as the first new particle in this model is at @xmath78 gev , the sm rges were evolved at two - loop level upto the scale @xmath80 and then the new set of rges was deployed to do the evolution of the couplings and mass parameters . [ fig : sample rge vector - like model ] shows that the both the vevs ( electroweak vev and vev for the single @xmath71 ) can be generated by radiative correction . we have presented in this paper various extensions of the sm where electroweak symmetry breaking is triggered by the renormalization group flow . even though such symmetry breaking fails to occur in sm , the scalar extensions are able to incorporate this attractive mechanism . extensions like type - ii seesaw models , loop induced neutrino mass models and scalar dark matter models all have this built - in feature . a common shared feature of all models where radiative ewsb is realized is the presence of new scalars at the tev scale . these tev scale scalars may be detected in the large hadron collider ( lhc ) in the near future . ksb would like to thank bartol research institute for hospitality during a visit where the work was initiated . ksb and sk would like to thank the organizers of cetup * 2016 for hospitality and the participants for helpful discussions and comments . the work of ksb and sk is supported in part by the u. s. department of energy grant no . de - sc0016013 and the work of ig is supported in part by bartol research institute . the gauge transformation of the fields defined in eqs . ( [ eq : type - ii scalar potential])-([eq : type ii yukawa ] ) can be written as : @xmath81 where @xmath82 is a unitary matrix . under this gauge transformation the yukawa term remains invariant : @xmath83 let us define @xmath84 : @xmath85 the gauge transformation of the field @xmath86 can be written as @xmath87 now , one can diagonalize @xmath86 by gauge transformation and as @xmath88 , we can write @xmath86 in such a basis as @xmath89 let us define the higgs doublet as @xmath90 in terms of these real fields the quartic part of the scalar potential ( given in eq . ( [ eq : type - ii scalar potential]))becomes @xmath91 where @xmath92 the quartic couplings of the potential form a vector space spanned by the real valued vector , @xmath93 and the quartic couplings of the scalar potential can be written as @xmath94 where @xmath95 we use known results from the copositivity conditions of real symmetric matrices @xcite to determine the boundedness conditions as 1 . 2 . 3 . where @xmath96 as this set of boundedness conditions need to be satisfied for all values of @xmath52 , this set of conditions reduces to the set of inequalities ( [ cond1 ] , [ cond2 ] and [ cond3 ] ) note that , all the conditions mentioned in inequalities ( [ cond1 ] ) and ( [ cond2 ] ) and at least one of the conditions in inequality ( [ cond3 ] ) need to be satisfied for the potential to be bounded from below . the rges for the yukawa couplings are given by @xcite : @xmath97 + \mathbf{y_d}\left [ t- \frac{1}{4 } g_1 ^ 2 - \frac{9}{4 } g_2 ^ 2 - 8\,g_3 ^ 2\right ] ; \\ 16\pi^2 \dfrac{d\mathbf{y}_u}{d t } = & \mathbf{y_u } \left [ \frac{3}{2 } \mathbf{y_u^\dagger y_u } - \frac{3}{2}\ , \mathbf{y_d^\dagger y_d}\right]+\mathbf{y_u}\left [ t-\frac{17}{20 } g_1 ^ 2 - \frac{9}{4 } g_2 ^ 2 - 8\,g_3 ^ 2\right ] ; \\ 16\pi^2 \dfrac{d \mathbf{y}_e}{d t}=&\mathbf{y_e}\left[\frac{3}{2}\mathbf{y_e^\dagger y_e}+\frac{3}{2}\mathbf{y_\delta^\dagger y_\delta}\right ] + \mathbf{y_e}\left[t-\frac{9}{4}g_1 ^ 2-\frac{9}{4}g_2 ^ 2\right];\\ 16\pi^2 \dfrac{d \mathbf{y}_\delta}{d t}=&\left[\frac{1}{2}\mathbf{y_e^\dagger y_e}+\frac{3}{2}\mathbf{y_\delta^\dagger y_\delta}\right]^t \mathbf{y_\delta } + \mathbf{y_\delta}\left[\frac{1}{2}\mathbf{y_e^\dagger y_e}+\frac{3}{2}\mathbf{y_\delta^\dagger y_\delta}\right ] \\&+\left[-\frac{3}{2}\left(\frac{3}{5}g_1 ^ 2 + 3g_2 ^ 2\right)+\tr\left(\mathbf{y_\delta^\dagger y_\delta}\right)\right]\mathbf{y_\delta}.\\ \end{split}\ ] ] the rges for the quartic couplings of the lagrangians are given by @xcite : @xmath98\lambda_4\nonumber + 8\lambda_5 ^ 2;\\ 16\pi^2 \dfrac{d \lambda_5}{d t}=&-\frac{9}{2}g_1 ^ 2\lambda_5-\frac{33}{2}g_2 ^ 2\lambda_5-\frac{18}{5}g_1 ^ 2g_2 ^ 2+\[2\lambda_2 - 2\lambda_3 + 2\lambda_1 + 8\lambda_4 + 2 t + 2 \tr\left(\mathbf{y_\delta^\dagger y_\delta}\right)\]\lambda_5.\nonumber\\ \end{split}\ ] ] @xmath99 ; \\ h & = \tr \[\mathbf{y_e^\dagger y_e y_e^\dagger y_e } + 3 \mathbf{y_d^\dagger y_d y_d^\dagger y_d } + 3 \mathbf{y_u^\dagger y_u y_u^\dagger y_u } \ ] . \end{split}\ ] ] the rges for the mass parameters are given by eq . ( [ mass rges type - ii ] ) . + beyond the energy scale corresponding to the mass of the triplet @xmath100 the sm gauge coupling evolution also needs to be recalculated due to the triplet s contribution . while the weak triplet does not effect the evolution of the @xmath101 gauge coupling evolution , it does change the rges of the other gauge couplings . the rges for the gauge couplings are given by : @xmath102 where @xmath103 are the three gauge couplings with the one loop @xmath52-function coefficient @xmath104 . for the two - loop neutrino mass model , among the gauge couplings only the hypercharge gauge coupling is modified due the additional scalar particles . so , the rges for the sm gauge couplings are given by : @xmath105 where @xmath103 are the three gauge couplings with the one loop @xmath52-function coefficient @xmath106 + the rges for the yukawa couplings are given by @xcite : @xmath107 the rges for the quartic scalar couplings are given by @xcite : @xmath108 - 8 \tr \l(\mathbf{f^\dagger f y^\dagger_\ell y_\ell}\r ) ; \end{split}\ ] ] @xmath109 - 8 \tr \l(\mathbf{y^\dagger_\ell y_\ell h^\dagger h } \r ) ; \\ 16 \pi^2 \dfrac{d\lambda_6}{d t}=&\ ; 4 \lambda_2 \lambda_6 + 4 \lambda_3 \lambda_6 + 4 \lambda_4 \lambda_5 + 4 \lambda_6 ^ 2-\dfrac{90}{5 } \lambda_6 g_1 ^ 2 + \dfrac{432}{25 } g_1 ^ 4 + 8 \lambda_6 \ [ \tr ( \mathbf{f^\dagger f})+\tr ( \mathbf{h^\dagger h})\].\\ \end{split}\ ] ] the rges for the mass parameters are given by eq . ( [ mass rges two - loop neutrino ] ) . the one loop rges for the inert doublet model have already been computed . the sm gauge coupling rges are given by : @xmath110 where @xmath111 are the @xmath52-coefficients of the sm gauge couplings updated with the added particles . + the quark sector of the model remains unchanged , while the leptonic sector needs to be revisited . the rges for the leptonic yukawa couplings are @xcite : @xmath112 for the quartic scalar coupling we find the following set of rges @xcite : @xmath113 @xmath114;\\ \end{split}\ ] ] @xmath115 ; \hspace{.5 cm } t_{4\nu } \equiv \tr \ [ \mathbf{h^\dagger h h^\dagger h}\];\hspace{.5 cm } t_{\nu e } \equiv \tr \ [ \mathbf{h^\dagger h y_e^\dagger y_e}\].\\ \end{split}\ ] ] the rges for the mass parameters are given by eq . ( [ mass rges for idm ] ) . one notices from the set of rges that the evolution of majorana mass @xmath116 , the new yukawa coupling @xmath117 and the scalar quartic coupling @xmath57 are proportional to the respective quantities themselves . the upshot of this setting is that these parameters remain small if they are small at any energy scale . this feature of the model becomes self - explanatory upon realization that if any of these parameters becomes zero , the neutrino becomes massless and global @xmath0 symmetry conserving the lepton number is restored . while the rges for the mass parameters are given by eq . ( [ mass rges sm + real singlet scalar ] ) , the rges for the quartic couplings are given by : @xmath118 rges for the yukawa couplings and the gauge couplings remain the same as sm . the rges for the mass parameters are given by eq . ( [ mass rges for vector - like fermion model ] ) . the rges for the gauge couplings are given by : @xmath119 where @xmath120 . the set of rges for all the yukawa couplings is given by : @xmath121;\\ 16 \pi^2 \dfrac{d\mathbf{{y}_d}}{d t}=&\ ; \mathbf{y_d } \ [ \dfrac{3}{2 } \l ( \mathbf{y_d^\dagger y_d } - \mathbf{y_u^\dagger y_u } \r)+\dfrac{1}{2}\mathbf{g_d^\dagger g_d}+ t- \dfrac{1}{4}g_1 ^ 2-\dfrac{9}{4}g_2 ^ 2 - 8g_3 ^ 2 - 3g_4 ^ 2\];\\ 16 \pi^2 \dfrac{d\mathbf{{y}_e}}{d t}=&\ ; \mathbf{y_e } \ [ \dfrac{3}{2 } \mathbf{y_e^\dagger y_e } + \dfrac{1}{2}\mathbf{g_e^\dagger g_e}+ t- \dfrac{9}{4}g_1 ^ 2-\dfrac{9}{4}g_2 ^ 2 - 3g_4 ^ 2\];\\ 16 \pi^2\dfrac{d\mathbf{{f}_u}}{d t}= & \ ; \mathbf{f_u } \[\mathbf{f_u^\dagger f_u}+t_f-\dfrac{8}{5}g_1 ^ 2 - 8g_3 ^ 2 - 15 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{g_u g_u^\dagger f_u};\\ 16 \pi^2 \dfrac{d\mathbf{{f}_d}}{d t}= & \ ; \mathbf{f_d } \[\mathbf{f_d^\dagger f_d}+t_f-\dfrac{2}{5}g_1 ^ 2 - 8g_3 ^ 2 - 15 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{g_d g_d^\dagger f_d};\\ 16 \pi^2 \dfrac{d\mathbf{{f}_e}}{d t}= & \ ; \mathbf{f_e } \[\mathbf{f_e^\dagger f_e}+t_f-\dfrac{18}{5}g_1 ^ 2 - 15 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{g_e g_e^\dagger f_e};\\ 16 \pi^2 \dfrac{d\mathbf{{g}_u}}{d t}= & \ ; \mathbf{g_u } \[\mathbf{g_u^\dagger g_u}+\mathbf{y_u^\dagger y_u}+t_g-\dfrac{8}{5}g_1 ^ 2 - 8 g_3 ^ 2 - 6 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{f_u f_u^\dagger g_u};\\ 16 \pi^2 \dfrac{d\mathbf{{g}_d}}{d t}= & \ ; \mathbf{g_d } \[\mathbf{g_d^\dagger g_d}+\mathbf{y_d^\dagger y_d}+t_g-\dfrac{2}{5}g_1 ^ 2 - 8 g_3 ^ 2 - 6 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{f_d f_d^\dagger g_d};\\ 16 \pi^2\dfrac{d\mathbf{{g}_e}}{d t}= & \ ; \mathbf{g_e } \[\mathbf{g_e^\dagger g_e}+\mathbf{y_e^\dagger y_e}+t_g-\dfrac{18}{5}g_1 ^ 2 - 6 g_4 ^ 2 \]+\dfrac{1}{2}\mathbf{f_e f_e^\dagger g_e}.\\ \end{split}\ ] ] the rges for scalar quartic couplings are given by : @xmath122 @xmath123 where @xmath124 ; \\ t_f & = \tr \[\mathbf{f_e^\dagger f_e } + 3 \mathbf{f_d^\dagger f_d } + 3 \mathbf{f_u^\dagger f_u } \ ] ; \\ t_g & = \tr \[\mathbf{g_e^\dagger g_e } + 3 \mathbf{g_d^\dagger g_d } + 3 \mathbf{g_u^\dagger g_u } \ ] ; \\ h & = \tr \[\mathbf{y_e^\dagger y_e y_e^\dagger y_e } + 3 \mathbf{y_d^\dagger y_d y_d^\dagger y_d } + 3 \mathbf{y_u^\dagger y_u y_u^\dagger y_u } \];\\ h_f & = \tr \[\mathbf{f_e^\dagger f_e f_e^\dagger f_e } + 3 \mathbf{f_d^\dagger f_d f_d^\dagger f_d } + 3 \mathbf{f_u^\dagger f_u f_u^\dagger f_u } \];\\ h_g & = \tr \[\mathbf{g_e^\dagger g_e g_e^\dagger g_e } + 3 \mathbf{g_d^\dagger g_d g_d^\dagger g_d } + 3 \mathbf{g_u^\dagger g_u g_u^\dagger g_u } \];\\ h_{yg } & = \tr \[\mathbf{y_e^\dagger y_e g_e^\dagger g_e } + 3 \mathbf{y_d^\dagger y_d g_d^\dagger g_d } + 3 \mathbf{y_u^\dagger y_u g_u^\dagger g_u } \];\\ h_{fg } & = \tr \[\mathbf{f_e^\dagger f_e g_e^\dagger g_e } + 3 \mathbf{f_d^\dagger f_d g_d^\dagger g_d } + 3 \mathbf{f_u^\dagger f_u g_u^\dagger g_u } \ ] . \end{split}\ ] ] 99 s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * , 30 ( 2012 ) [ arxiv:1207.7235 [ hep - ex ] ] ; g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * , 1 ( 2012 ) [ arxiv:1207.7214 [ hep - ex ] ] . g. aad _ et al . _ [ atlas and cms collaborations ] , phys . lett . * 114 * , 191803 ( 2015 ) [ arxiv:1503.07589 [ hep - ex ] ] . p. minkowski , phys . b * 67 * , 421 ( 1977 ) ; t. yanagida , proceedings of the _ workshop on unified theories and baryon number in the universe _ , tsukuba , 1979 , eds . a. sawada , a. sugamoto ; s. glashow , in cargese 1979 , _ proceedings , quarks and leptons _ ( 1979 ) ; m. gell - mann , p. ramond and r. slansky , conf . c * 790927 * , 315 ( 1979 ) [ arxiv:1306.4669 [ hep - th ] ] ; r. n. mohapatra and g. senjanovic , phys . lett . * 44 * , 912 ( 1980 ) . j. schechter and j. w. f. valle , phys . d * 22 * , 2227 ( 1980 ) ; g. lazarides , q. shafi and c. wetterich , nucl . phys . b * 181 * , 287 ( 1981 ) ; r. n. mohapatra and g. senjanovic , phys . d * 23 * , 165 ( 1981 ) ; c. wetterich , nucl . b * 187 * , 343 ( 1981 ) ; j. schechter and j. w. f. valle , phys . d * 25 * , 774 ( 1982 ) . a. zee , phys . b * 93 * , 389 ( 1980 ) erratum : [ phys . lett . b * 95 * , 461 ( 1980 ) ] . k. s. babu , phys . b * 203 * ( 1988 ) 132 . g. r. farrar and p. fayet , phys . b * 76 * , 575 ( 1978 ) . h. c. cheng , k. t. matchev and m. schmaltz , phys . d * 66 * , 036005 ( 2002 ) [ hep - ph/0204342 ] . h. c. cheng and i. low , jhep * 0309 * , 051 ( 2003 ) [ hep - ph/0308199 ] ; h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ hep - ph/0405243 ] ; i. low , jhep * 0410 * , 067 ( 2004 ) [ hep - ph/0409025 ] . n. g. deshpande and e. ma , phys . rev . d * 18 * , 2574 ( 1978 ) . e. ma , phys . d * 73 * , 077301 ( 2006 ) [ hep - ph/0601225 ] . t. w. b. kibble , g. lazarides and q. shafi , phys . b * 113 * , 237 ( 1982 ) ; r. n. mohapatra , phys . rev . d * 34 * , 3457 ( 1986 ) ; a. font , l. e. ibanez and f. quevedo , phys . b * 228 * , 79 ( 1989 ) ; l. m. krauss and f. wilczek , phys . . lett . * 62 * , 1221 ( 1989 ) ; l. e. ibanez and g. g. ross , phys . b * 260 * , 291 ( 1991 ) ; l. e. ibanez and g. g. ross , nucl . b * 368 * , 3 ( 1992 ) ; s. p. martin , phys . d * 46 * , 2769 ( 1992 ) [ hep - ph/9207218 ] . l. e. ibanez and g. g. ross , phys . b * 110 * , 215 ( 1982 ) ; k. inoue , a. kakuto , h. komatsu and s. takeshita , prog . phys . * 68 * , 927 ( 1982 ) erratum : [ prog . phys . * 70 * , 330 ( 1983 ) ] ; l. e. ibanez , phys . b * 118 * , 73 ( 1982 ) ; j. r. ellis , d. v. nanopoulos and k. tamvakis , phys . b * 121 * , 123 ( 1983 ) ; j. r. ellis , j. s. hagelin , d. v. nanopoulos and k. tamvakis , phys . lett . b * 125 * , 275 ( 1983 ) ; l. alvarez - gaume , j. polchinski and m. b. wise , nucl . b * 221 * , 495 ( 1983 ) . a. de rujula , h. georgi , and s. l. glashow , in _ fifth workshop on grand unification _ , edited by k. kang , h. fried , and p. frampton ( world scientific , singapore , 1984 ) , p. 88 . for an earlier discussion , see y. achiman and b. stech , in _ new phenomena in lepton - hadron physics _ , edited by d. e. c. fries and j. wess ( plenum , new york , 1979 ) , p. 303 . k. s. babu , x. g. he and s. pakvasa , phys . d * 33 * , 763 ( 1986 ) ; g. m. pelaggi , a. strumia and s. vignali , jhep * 1508 * , 130 ( 2015 ) [ arxiv:1507.06848 [ hep - ph ] ] . k. s. babu , b. bajc and s. saad , arxiv:1605.05116 [ hep - ph ] . k. s. babu , i. gogoladze , p. nath and r. m. syed , phys . d * 72 * , 095011 ( 2005 ) [ hep - ph/0506312 ] . s. r. coleman and e. j. weinberg , phys . d * 7 * , 1888 ( 1973 ) . h. arason , d. j. castano , b. keszthelyi , s. mikaelian , e. j. piard , p. ramond and b. d. wright , phys . d * 46 * , 3945 ( 1992 ) ; m. x. luo and y. xiao , phys . lett . * 90 * , 011601 ( 2003 ) [ hep - ph/0207271 ] . f. vissani , phys . d * 57 * , 7027 ( 1998 ) [ hep - ph/9709409 ] . a. arhrib , r. benbrik , m. chabab , g. moultaka , m. c. peyranere , l. rahili and j. ramadan , phys . d * 84 * , 095005 ( 2011 ) [ arxiv:1105.1925 [ hep - ph ] ] . m. a. schmidt , phys . d * 76 * , 073010 ( 2007 ) erratum : [ phys . d * 85 * , 099903 ( 2012 ) ] [ arxiv:0705.3841 [ hep - ph ] ] . i. gogoladze , n. okada and q. shafi , phys . d * 78 * , 085005 ( 2008 ) [ arxiv:0802.3257 [ hep - ph ] ] ; a. kobakhidze and a. spencer - smith , jhep * 1308 * , 036 ( 2013 ) [ arxiv:1305.7283 [ hep - ph ] ] . k. s. babu and j. julio , aip conf . proc . * 1604 * , 134 ( 2014 ) . k. s. babu and c. macesanu , phys . d * 67 * , 073010 ( 2003 ) [ hep - ph/0212058 ] . m. nebot , j. f. oliver , d. palao and a. santamaria , phys . d * 77 * , 093013 ( 2008 ) [ arxiv:0711.0483 [ hep - ph ] ] ; j. herrero - garcia , m. nebot , n. rius and a. santamaria , nucl . b * 885 * , 542 ( 2014 ) [ arxiv:1402.4491 [ hep - ph ] ] ; d. schmidt , t. schwetz and h. zhang , nucl . b * 885 * , 524 ( 2014 ) [ arxiv:1402.2251 [ hep - ph ] ] ; j. herrero - garcia , m. nebot , n. rius and a. santamaria , arxiv:1410.2299 [ hep - ph ] . r. barbieri , l. j. hall and v. s. rychkov , phys . d * 74 * , 015007 ( 2006 ) [ hep - ph/0603188 ] ; j. a. casas , j. r. espinosa and i. hidalgo , nucl . b * 777 * , 226 ( 2007 ) [ hep - ph/0607279 ] . e. ma , mod . a * 21 * , 1777 ( 2006 ) [ hep - ph/0605180 ] . t. hambye and m. h. g. tytgat , phys . b * 659 * , 651 ( 2008 ) [ arxiv:0707.0633 [ hep - ph ] ] . a. merle and m. platscher , phys . d * 92 * , no . 9 , 095002 ( 2015 ) [ arxiv:1502.03098 [ hep - ph ] ] . v. silveira and a. zee , phys . * 161b * , 136 ( 1985 ) . j. mcdonald , phys . rev . d * 50 * , 3637 ( 1994 ) [ hep - ph/0702143 [ hep - ph ] ] . m. kadastik , k. kannike and m. raidal , phys . d * 80 * , 085020 ( 2009 ) erratum : [ phys . d * 81 * , 029903 ( 2010 ) ] [ arxiv:0907.1894 [ hep - ph ] ] ; y. mambrini , n. nagata , k. a. olive and j. zheng , phys . d * 93 * , no . 11 , 111703 ( 2016 ) [ arxiv:1602.05583 [ hep - ph ] ] . s. m. barr , phys . lett . * 53 * , 329 ( 1984 ) ; a. e. nelson , phys . * 136b * , 387 ( 1984 ) . k. s. babu and r. n. mohapatra , phys . lett . * 62 * , 1079 ( 1989 ) . k. s. babu and r. n. mohapatra , phys . d * 41 * , 1286 ( 1990 ) . r. n. mohapatra and y. zhang , jhep * 1406 * , 072 ( 2014 ) [ arxiv:1401.6701 [ hep - ph ] ] . s. b. dev , r. n. mohapatra and y. zhang , jhep * 1602 * , 186 ( 2016 ) [ arxiv:1512.08507 [ hep - ph ] ] . cms collaboration [ cms collaboration ] , collisions at 13tev , cms - pas - exo-15 - 004 ; the atlas collaboration , atlas - conf-2015 - 081 ; m. aaboud _ et al . _ [ atlas collaboration ] , arxiv:1606.03833 [ hep - ex ] ; v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1606.04093 [ hep - ex ] . cms collaboration [ cms collaboration ] , cms - pas - exo-16 - 027 ; the atlas collaboration [ atlas collaboration ] , atlas - conf-2016 - 059 . k. p. hadeler , linear algebra appl . , * 49 * , 79 ( 1983 ) . k. g. klimenko , theor . phys . , * 62 * , 58 ( 1985 ) .
einstein s theory of general relativity ( gr ) predicts that space - time is curved and warped by the mass and the angular momentum of a gravitating body . the kinematics of objects in the vicinity of such a body are expected to differ from those based on newton s theory of gravity . the nasa / stanford spaceborne mission gravity probe b ( ) was designed to measure these differences with four gyroscopes essentially freely falling around earth , in a polar orbit at an altitude of about 640 km . according to gr , the gyroscopes spin axes should precess by @xmath46 yr@xmath5 in a north - south direction due to the curving of space - time by the earth s mass , and by @xmath439 in an eastward direction due to the warping of space - time by the earth s angular momentum . these precessions are dubbed the geodetic " effect and the frame - dragging , " ( `` lense - thirring , '' or `` gravitomagnetic '' ) effect , respectively , and can be considered as rotations of any near - earth inertial frame relative to the distant universe . the gyroscopes were set spinning within a quartz block with respect to which , through superconducting quantum interference devices ( squids ) , each of the spin axes precessions could be measured with an accuracy of 100 within @xmath440 days . the quartz block was bonded to a @xmath415 cm diameter quartz optical telescope . all of these devices the quartz block , the squids , and the quartz telescope ( together termed the probe ) were placed in a dewar filled with liquid helium and cooled to a temperature of 1.8 k. ideally , the orientation of the probe could be best provided by pointing the telescope to one suitable quasar , one of the most distant compact types of single object in the universe . however , quasars were too dim for the telescope to track . only stars in our galaxy were bright enough . the challenge was therefore to find a guide star , " bright enough for the on - board telescope to track , sufficiently isolated to limit light contamination , and suitably located to not unnecessarily decrease the accuracy of the measurement of the precessions of the gyroscopes . then the guide star s motion on the sky needed to be measured with respect to quasars or distant galaxies so that the precession of the gyroscopes could be determined with respect to the distant universe . we described the astronomical effort undertaken with the guide star for the support of the mission , vlbi for gravity probe b , " in seven papers : overview " ( * ? ? ? * paper i ) ; monitoring of the structure of the reference sources 3c 454.3 , b2250 + 194 , and b2252 + 172 " ( * ? ? ? * paper ii ) ; a limit on the proper motion of the ` core ' of the quasar 3c 454.3 " ( * ? ? ? * paper iii ) ; a new astrometric analysis technique and a comparison with results from other techniques " ( * ? ? ? * paper iv ) ; proper motion and parallax of the guide star , i m pegasi " ( * ? ? ? * paper v ) ; the orbit of i m pegasi and the location of the source of radio emission " ( * ? ? ? * paper vi ) ; and the evolution of the radio structure of i m pegasi " ( * ? ? ? * paper vii ) . here we give a synopsis of these papers . a suitable guide star for had to simultaneously meet a number of requirements : * the motion of the star needed to be known or measurable with an uncertainty small enough not to significantly increase the prelaunch anticipated standard error of of @xmath60.5 in each of the two precessions . the specific requirement was that the standard error be @xmath70.14 for measurements of the star s motion in each coordinate . in 1990 , when we entered the first stage of our search for a suitable guide star , no star s ( or any other celestial object s ) motion on the sky was known with such accuracy in the optical . such accuracy had previously been achieved only in the radio . pioneered by @xcite , astrometric measurements with the radio technique of very long baseline interferometry ( vlbi ) of a quasar relative to another quasar nearby on the sky yielded a proper - motion estimate with a standard error less than half of the requirement for @xcite . therefore vlbi appeared to be at that time the only option to determine the motion of a suitable guide star with the desired accuracy . * the star needed to be sufficiently bright in the radio , preferably with a flux density at least of order 1 mjy , for vlbi astrometry measurements to be successful . * the star needed to be sufficiently bright in the optical , at least 7th magnitude , for the on - board telescope to detect . * the star s angular distance , @xmath8 , from the ecliptic needed to be between 20@xmath9 and 40@xmath9 . the lower limit was set by the requirement that , with a suitable sun shield at the telescope , sunlight not enter the dewar and boil off the liquid helium . the upper limit was set by the power requirements of the spacecraft , since the solar panels were fixed to the spacecraft , which would continually point to the guide star , making the power received by the solar panels dependent on @xmath8 . * the star needed to be located as close as possible to the celestial equator , so as to maximize the sensitivity of the measurement of the frame - dragging precession , which essentially decreases with the cosine of the star s declination . * the star needed to be sufficiently isolated on the sky . neighbouring stars or reflection nebulae could possibly intolerably decrease the accuracy of the pointing of the telescope to the star s center of brightness . to find the most suitable guide star for , we conducted a vla survey at 8.4 ghz from 1990 to 1992 of @xmath41200 stars with v magnitude @xmath76.0 and a declination between @xmath10 and @xmath11 . no single stars were found in the survey with a flux density above @xmath41 mjy . in fact , we detected only previously known radio stars , all of them binaries of the rs canum venaticorum type ( see also , for a later survey , * ? ? ? we identified four potentially suitable guide stars : @xmath12 andromeda ( hr 8961 , @xmath13 declination ) , hr 1099 ( @xmath14 ) , hr 5110 ( @xmath15 ) , and ( hr 8703 , @xmath16 ) . after investigating each of the four candidates with vlbi , studying the sky fields around them in the optical , and checking on possible vlbi reference sources for each of them , we selected as the most suitable for . is a binary with a giant primary and sun - type secondary . we summarize its optical characteristics and previously known properties in table [ topt ] . the radius of the primary is @xmath413.3 @xmath3 0.6 , which translates at a distance of of @xmath4100 pc , to an angular radius of @xmath40.64 @xmath3 0.03 mas . this radius is comparable to the full - width at half - maximum ( fwhm ) of the synthesized beam of a global vlbi array operating at 8.4 ghz , and therefore in principle allows for an accurate determination of the location of the radio emission with respect to the center of the primary . a prerequisite for successful guide star tracking was that the onboard telescope be able to lock sufficiently accurately on the center of the optical disk of the star ( or a point with an offset from this center constant during the mission ) . one concern here was the appearance of variable dark spots on the primary , shown by doppler imaging and photometry to cover @xmath415% of the star s surface . if the variations in the size or location of these spots resulted in a linear trend over the course of the mission in the centroid of optical brightness , relative to the center of the disk of the i m peg primary , then that trend would have the same effect on the data reduction as an equal error in our vlbi measurement of the proper motion of i m peg . such a trend is not necessarily implausible , given the year - to - year variations in i m peg s optical brightness and the astrophysical plausibility of variations in typical spot latitudes analogous to the systematic variations of sunspot locations on the sun . however , @xcite showed , through extensive spot mapping spanning the duration of the mission , that any error due to variability in the spot pattern would yield no more than a 0.04 drift in the optical center . a second concern was the surroundings of i m peg on the sky . any star or nebula within the field of view of the telescope when locked on , possibly combined with s photometric variability , could cause systematic astrometric errors . however , for , no other star within , e.g. @xmath17 , is brighter than v magnitude 10 , which can be compared with the corresponding relative large brightness of , that varied during the mission only between v magnitude 5.7 and 6.0 . all in all , through extensive optical and even millimetric observations of the co(j=1@xmath180 ) line ( the latter to search for any molecular cloud that could be associated with a reflection nebula ) , we constrained any astrometric systematic error to a negligible value : the combined error from dark spots , neighbouring stars , and a nebula , combined with the photometric variability of , we estimated to be smaller than 0.05 . l c c@ c c c _ hipparcos _ , @xmath19 , epoch 1991.25 & & & & 1 + _ hipparcos _ , @xmath20 , epoch 1991.25 & & & & 1 + _ hipparcos _ , parallax ( mas ) & & & & 1 + _ hipparcos _ , distance ( pc ) & & & & + & & & & & + + & & & & & + v magnitude range during mission & & 5.7 to 6.0 & & & 2 + mass ( @xmath21 ) & & @xmath22 & @xmath23 & & 3,3 + spectral type & & k2 iii & g v & & 4 + @xmath24 ( k ) & & @xmath25 & @xmath26 & & 4,3 + radius ( @xmath27 ) & & @xmath28 & @xmath29 & & 4,3 + radius ( mas ) & & @xmath30 & @xmath31 & & 4,3 + & & & & & + + & & & & & + @xmath32 ( @xmath27 ) & & @xmath33 & @xmath34 & & 3,3 + @xmath32 ( mas ) & & @xmath35 & @xmath36 & & + @xmath37 ( days ) & & & & 3 + @xmath38 ( @xmath9 ) & & & & 4,5 + @xmath39 & & & & 4 + @xmath40 ( hjd ) & & & & 3 + [ topt ] l l l l l l l 3c 454.3 & quasar & & & 7 10 & 0.859 & 1610 + b2250 + 194 & galaxy & @xmath41 & 3.6 & 0.35 0.45 & 0.28 & 880 + b2252 + 172 & unidentified & 0.4 & 1.4 & 0.017 & & + + i m peg & rs cvn & @xmath42 & 0.7 & 0.0002 0.08 & 0.0 & 0.0 + [ t2refsources ] our goal was to determine the motion on the sky of relative to the distant universe , with a standard error not exceeding 0.14 in either coordinate . to achieve this goal , we selected radio reference sources with respect to which the motion of could be measured . these sources needed to be at cosmological distances , be sufficiently strong and compact for vlbi observations , and located nearby to on the sky so as to minimize systematic astrometric errors . we selected three reference sources , with the strong quasar , , as the main one . the other two were the active nucleus of the radio galaxy , , and the unidentified source ( most likely also a quasar or a radio galaxy ) , . we display their positions on the sky , together with that of , in figure [ f1skypos ] , and list their characteristics in table [ t2refsources ] . ( east - west ) and @xmath43 ( north - south ) directions on the plot are on the same scale . figure taken from . ] in figure [ f2images ] we present typical vlbi images at 8.4 ghz of and each of the three references sources . contour ( fwhm ) in the inset . north is up and east to the left . for more information , see and . figure taken from , but rearranged.,title="fig : " ] contour ( fwhm ) in the inset . north is up and east to the left . for more information , see and . figure taken from , but rearranged.,title="fig : " ] contour ( fwhm ) in the inset . north is up and east to the left . for more information , see and . figure taken from , but rearranged.,title="fig : " ] contour ( fwhm ) in the inset . north is up and east to the left . for more information , see and . figure taken from , but rearranged.,title="fig : " ] all the sources are approximately located along the same axis ( north - south ) , allowing perhaps for a more accurate relative correction of tropospheric and ionospheric effects . however , each of the three sources has advantages and disadvantages as reference sources for . the quasar 3c 454.3 is both the strongest and the closest on the sky to . it was used as a reference source for from 1991 to 1994 @xcite , and could therefore extend the time baseline of our vlbi observations . the disadvantage is the complexity of 3c 454.3 s structure and its superluminal nature . in figure [ f2images ] we identified six components of which the easternmost component , c1 , is still compact at the highest angular resolution , has a flat or inverted high - frequency spectrum @xcite , and is located almost exactly at the same position ( ) as the core of the quasar seen at 43 ghz @xcite . the component c1 was therefore considered to be the core at 8.4 ghz , the part most closely related to the supermassive black hole and to the center of mass of the quasar ; it was thus the reference point in the quasar s structure for our astrometric measurements . the other components are moving away from c1 , with apparent speeds of up to @xmath45 c ( ) . the remaining sources were both rather compact , so that the brightness peak could simply be taken as the reference point in the structure ; their disadvantages were that b2250 + 194 was located relatively far away from and b2252 + 172 was rather faint ( with an unknown redshift ) . in addition , we used geodetic and astrometric vlbi observations regularly conducted over many years to determine the limit on the proper motion of two of the three reference sources , 3c 454.3 and b2250 + 194 . these determinations were made without consideration of a particular reference point in the structure of the two sources but were made with respect to a celestial reference frame , crf . this frame is defined by a large number of extragalactic radio sources distributed over the sky and is for our purposes essentially the same as the international celestial reference frame , icrf2 @xcite . all in all , we took c1 in 3c 454.3 as the primary reference point with respect to which the motion of was measured . the other measurements served to determine the stability of c1 on the sky so that finally the motion of could be determined with respect to the distant universe . to determine the motion of relative to c1 and test c1 s stability relative to the brightness peaks of the other two reference sources , we took into account : * is catalogued as a binary . however , we could not rule out that it is perhaps part of an extended triple or even larger multiple system . therefore , to ensure our obtaining a sufficiently accurate value of the proper motion of for the @xmath415 months data - collection interval in 2004 and 2005 , we spread a sufficiently large number of vlbi measurements over eight and a half years to be able to determine any nearly constant change of the proper motion with time ( proper acceleration " ) , or place a limit on it . * for the best determination of s parallax , we spread the vlbi measurement epochs appropriately to cover the different seasons each year . similarly , to allow us to model any orbital component of the motion of the radio emission , we carefully distributed the epochs over the phase of the binary orbit , which could be computed from earlier spectroscopic results . * to measure the time - variable brightness distributions of and the three reference sources , and in particular to identify c1 in 3c 454.3 , we selected each vlbi observation session to be long enough to ensure that the coverage was sufficiently dense for us to obtain excellent imaging of the sources . * since we expected to fluctuate strongly in radio brightness , and since we knew that radio emission from b2252 + 174 was relatively weak , we needed our vlbi array to be sufficiently sensitive to nearly guarantee detection of the sources in each observing session . * lastly , we needed an observing frequency for which receivers were available at each of the telescopes of the vlbi array , and which guaranteed sufficiently high angular resolution and low system temperatures . all said , 35 sessions of astrometric vlbi observations were made between 1997 january 16 and 2005 july 16 with about four sessions every year , each about 11 to 15 h long . the observations were made with a global array of 12 to 16 radio telescopes comprised of most or all of : mpifr s 100 m telescope at effelsberg , germany ; nasa / caltech / jpl s 70 m dsn telescopes at robledo , spain , goldstone , ca , and tidbinbilla , australia ; nrao s ten 25 m telescopes of the vlba , across the u.s.a ; nrao s phased vla , equivalent to a 130 m telescope , near socorro , nm ; and , at early times , nrcan s 46 m algonquin radio telescope near pembroke , ont , canada , and nrao s 43 m telescope in green bank , wv . the observing frequency was 8.4 ghz . all of these vlbi observations were made to extract interferometric phase delays related to the difference of arrival times of a radio wave from a celestial source at each pair of antennas of a vlbi array . such observations allow the most accurate astrometric measurements @xcite and can yield relative positions and proper motions referenced to particular points in the structure of the sources with uncertainties as low as @xmath410 and @xmath410 , respectively ( for the earliest such measurements , see , e.g. , marcaide & shapiro 1983 ; bartel et al . 1986 ; and for a recent review , see reid & honma 2014 ) . the phase delay is given by @xmath44 where @xmath45 is the interferometric phase or phase " at observing angular frequency , @xmath46 , and time , @xmath47 . the phase can be expressed as @xmath48 + 2\pi n(\omega , t ) , \label{phi}\ ] ] where @xmath49 is the `` geometric delay , '' the difference in the arrival times of the radio wave in vacuum at the two antennas ; @xmath50 describes the difference in the instrumental delays ( including clock behavior ) at the two antenna sites ; @xmath51 gives the difference in radio wave propagation times to the two antennas due to all tropospheric and ionospheric effects ; @xmath52 is the delay contribution from source structure to account for any non - pointlike brightness distribution of the celestial source ; @xmath53 is the ( thermal ) noise contribution to the phase measurement ; and @xmath54 describes the integer number of 2@xmath55 ambiguities , or `` phase wraps , '' in the measurement . the astrometric information of a vlbi measurement is given by the geometric delay which , apart from relativistic contributions and those due to earth s motion relative to the solar system barycenter , is given by @xmath56 , \label{geom}\ ] ] where @xmath57 is the speed of light in vacuum , @xmath58 is the 3-dimensional vector between two antennas of a baseline of a vlbi array , and @xmath59 is the unit vector in the direction of the observed source . for phase - delay vlbi observations to succeed , to be able to remove the 2@xmath55 phase ambiguities , and to reduce many sources of astrometric errors , we needed to switch the antennas rapidly between sources . for the first 23 sessions , we used a typical sequence of 3c 454.3 ( 80 s ) - i m peg ( 170 s ) - b2250 + 194 ( 80 s ) . for the remaining 12 sessions , b2252 + 172 was included in the observations and the sequence was altered to 3c 454.3 ( 80 s ) - i m peg ( 125 s ) - b2250 + 194 ( 80 s ) - 3c 454.3 ( 80 s ) - i m peg ( 125 s ) - b2250 + 194 ( 80 s ) - b2252 + 172 ( 90 s ) . in addition to the astrometric 8.4 ghz observations , we also observed once at 5.0 ghz and once at 15.0 ghz , to allow investigation of the compactness and spectral properties of the components of the sources . almost all observations were recorded in both right and left circular polarization , and all were processed with the vlba correlator at socorro , nm . we used a custom software package ( ) that allowed us to analyze the data efficiently . first , we connected the phases and removed the 2@xmath55 phase ambiguities to generate the correct phase delays . then we corrected the phase delays for the effects of the troposphere and the ionosphere with _ a priori _ estimates . we further processed them with a kalman filter to model the residuals of the troposphere and the ionosphere , and the clock offset at each telescope from the one chosen as the reference clock for the vlbi array . then , for the more extended source 3c 454.3 , the structure effects were removed from the phase delays based on the reference point , c1 , defined with images made with nrao s aips software package . the other two reference sources were sufficiently compact that no correction for their source structure was needed . the relative weakness of s radio emission , and sometimes also that of b2252 + 172 , made it necessary for us to develop a special technique for the analysis of vlbi data . it combines the advantages of parametric model - fitting via weighted least - squares applicable for the relatively strong sources 3c 454.3 and b2250 + 194 with the sensitivity of phase - referenced mapping , the latter required for and b2252 + 172 . the merged analysis technique yielded superior results to the phase - referenced mapping technique ( ) . the radio emission from varied from session to session , reflecting the complex astrophysical nature of the environment of the star . in most cases the images of showed only one clearly defined component . in these cases the coordinates were determined for the maximum of a two - dimensional gaussian fit to that component . in the remaining cases , there was more than one local maximum of approximately equal brightness in the emission region : in eight cases there were two such maxima , and in one case there were three , all of them significant . in such cases , we took the position of i m peg to be the mean position of the different maxima . the final result of the analysis of the vlbi observations was a set of coordinates with statistical standard errors for each source and each observing epoch . these sets were the basis for the study of the motion , or upper bounds on it , of the sources with respect to each other . in addition , we analyzed geodetic astrometric vlbi observations and made use of interferometric group delays . group delays are given by @xmath60 these are observations of thousands of sources from all over the sky . these observations are done on a routine basis , independent of the program . for 3c 454.3 we used all available data from a total of 1,119 observing sessions , from 1980 to 2008 . in support of , the second reference source , b2250 + 194 , was included in 38 sessions of routine geodetic astrometric group - delay observations between 1997 and 2008 . group - delay observations have the advantage that 2@xmath55 ambiguities are not inherent in the data and that the positions are determined relative to a celestial reference frame ( crf ) . the disadvantage , however , is that the parameter , @xmath52 , was not determined and that therefore the exact reference point in the brightness distribution of the sources for the position measurement remained undefined . all available geodetic astrometric data with a total of 6.5 million group - delay determinations were processed via least - squares and yielded estimates of coordinates with statistical standard errors for 3c 454.3 and b2250 + 194 for each observing session , keeping the coordinates of the other sources constant . this set of solutions forms our crf , which for s purposes is virtually identical ( see , e.g. , * ? ? ? * ) to the icrf2 @xcite , the most fundamental such frame presently in use . the phase - delay determinations of the position of c1 , the core of 3c 454.3 , relative to those of the brightness peaks of b2250 + 194 and b2252 + 172 , are plotted in figure [ f3posref ] . the group - delay determinations of the position of b2250 + 194 relative to the crf are also plotted in figure [ f3posref ] . see for a similar plot of the group - delay determinations of the position of 3c 454.3 relative to the crf . are also given . bottom panel : the relative coordinates of b2250 + 194 as determined from routine geodetic group - delay vlbi observations of up to @xmath44000 extragalactic sources scattered over the sky . figure taken from . , title="fig:",scaledwidth=80.0% ] are also given . bottom panel : the relative coordinates of b2250 + 194 as determined from routine geodetic group - delay vlbi observations of up to @xmath44000 extragalactic sources scattered over the sky . figure taken from . , title="fig:",scaledwidth=72.0% ] weighted least - squares straight line fits to each set of position determinations yielded the position at epoch and proper motion for each coordinate separately , along with their statistical standard errors and correlations . in each case the fits to the data gave @xmath61 ( @xmath62 per degree of freedom ) larger than unity , indicating systematic errors in the position determinations . these errors can for instance be due to uncorrected effects from source structure and/or atmospheric and ionospheric variations , and are difficult to quantify . therefore the total standard errors of the position determinations were determined partly empirically by including a particular constant for each fit . this constant was added in quadrature to the statistical standard errors for each set of position determinations and for each coordinate separately so as to obtain @xmath63 for each fit . the resultant proper - motion values with their standard errors are listed for each set of position determinations in the panels of figure [ f3posref ] . the upper panel of figure [ f3posref ] indicates that there appears to be some motion of c1 relative to b2250 + 194 at the 3.5@xmath64 significance level . the source 3c 454.3 has a long jet , indicating strong activity . some motion of a component like c1 even if located close to the black hole could therefore be expected . no significant motion is discernible for b2250 + 194 relative to b2252 + 172 . no matter whether there is some motion due to activity near the black hole or not , the 1@xmath64 upper limits of the proper - motion values for c1 relative to b2250 + 194 and b2250 + 194 relative to the crf are each smaller than 40 . combining the phase - delay proper - motion determination of c1 relative to b2250 + 194 with the group - delay proper - motion determination of b2250 + 194 relative to the crf gives for the time period from 1998 to 2005 a 1@xmath64 standard error for the mean position of c1 in the crf of 45 and 68 . the proper motion of c1 relative to the crf for that same period is @xmath65 and @xmath66 in and , respectively , which we do not consider significant . we therefore give a 1@xmath64 upper limit on the magnitude of the proper - motion components of 46 and 56 , respectively . the latter pair of determinations is our limit on the level of stationarity of the reference point c1 in 3c 454.3 relative to the distant universe ( ) . with the stationarity of the reference point in 3c 454.3 determined , the motion of relative to the distant universe could be analyzed . five kinds of motion needed to be considered : * proper motion mainly due to the motion of relative to the motion of the sun in the galaxy , * annual parallax due to the orbital motion of earth around the sun , together with the much smaller contribution of the motion of the sun with respect to the solar system barycenter , * change of proper motion over time due to a hypothetical third member of the binary system , * orbital motion around the common center of mass of the binary , and * erratic motion of the center of radio emission within the binary system . to determine the proper motion , parallax , proper acceleration , and orbital motion , we used weighted least - squares to fit a linearized model to the 35 determinations of the positions of the reference point in relative to the position of c1 in 3c 454.3 . the uncertainties of these position determinations are hard to compute theoretically . we therefore took them from an empirical analysis based on the computation of the root - mean - square ( rms ) scatter separately in @xmath67 and @xmath43 , of the postfit residuals obtained for the 12 positions of relative to c1 and the 35 positions of also relative to c1 . the resulting uncertainty was 0.06 mas in each coordinate . in addition there is the previously discussed uncertainty of the position of c1 and the upper limit of the proper motion of c1 in the crf that we considered for the standard errors of the position determinations of relative to the crf . the parameters of the model fit were the position of at epoch , proper motion , parallax , and four scalar parameters for the description of the projection on the sky of an orbit with zero - eccentricity @xcite and a period known from optical spectroscopic observations @xcite . the fit resulted in rms values of the residuals somewhat different for each coordinate but much larger than the standard errors of the position determinations of . the source of this large scatter is the astrophysical nature of the radio emission , characterized by motion related to radio brightness changes @xcite and plausibly in part to stellar magnetic field changes implied by spot maps from optical spectroscopy ( e.g. , * ? ? ? for the fit we used uniform weighting independently for and , and allowed their errors to have non - zero correlation . after iteration , we obtained the final set of parameter estimates with statistical standard errors and @xmath63 . this set is presented in table [ tfinal ] . for the determination of the proper acceleration , we enlarged the set of positions of 35 epochs by four additional positions of obtained at epochs between 1991 and 1994 , also at 8.4 ghz , by @xcite in support of the _ hipparcos _ mission . these observations considerably extended our time baseline ; however , they were only of limited use for the estimate of parameters other than the proper acceleration because of inferior coverage , lower angular resolution , and the position determinations not being referred to c1 . we found no significant proper acceleration ; i.e. , none larger than one statistical standard error ; and so we include no such proper acceleration in the fit used to obtain our final results in table [ tfinal ] . figure [ f4proppar ] displays the 39 position determinations and the fit to the 35 position determinations , with the four earliest positions not used in the fit . the effect of the proper motion and the parallax can be clearly seen in the data as well as in the fit . figure [ f5paraellipse ] displays the effect of the parallax still more clearly . it shows the 35 position determinations with the estimated position at epoch , proper motion , and the primary s orbital motion subtracted . it also shows the estimated parallax ellipse . figure [ f6morbit ] displays the orbit clearly . again we plot the 35 position determinations with the estimated position at epoch , proper motion , but now with parallax rather than orbital motion subtracted . also shown are the estimated orbit and the corresponding positions predicted from the orbit model . figure [ f7resvt ] shows the 39 position determinations and the fit to the above 35 positions , but now with all model contributions , namely the estimated position at epoch , proper motion , parallax , and the primary s orbital motion subtracted . we compare them to the corresponding flux densities of . the scatter of the position residuals appears to be random with certainly no correlation to the measured flux densities . the rms of the scatter is @xmath40.4 mas in each coordinate , much larger than the standard errors of the position determinations and most likely dominated by the fluctuation of the positions of the stellar radio emission relative to the center of the primary of . figure [ f8imschemradio ] gives an artist s three - dimensional rendition of with the primary as a giant with dark spots and the secondary as a sun - like star , each in its estimated orbit . errors ( here and hereafter in the figure captions in the sense of the full length of @xmath68 error bars ) are much smaller than the symbols . the curve shows the fit of the solution , with orbital terms omitted for clarity . in addition , the four positions from 1991 to 1994 are shown ( upper left ) ; however , they were not used in the fit . the arrow indicates the direction and its length the magnitude of the proper motion in one year . figure taken from . ] errors ( of position determinations relative to c1 ) are about a third of the size of the circles . the relatively large scatter is most likely dominated by the fluctuations of the positions of the stellar radio emission relative to the center of the disk of the primary . the model positions ( not shown ) are almost exactly on the plotted parallax ellipse and close to the observed positions . the parallax ellipse , however , is slightly time - dependent due to the influence of planets . over the course of our observations of several years , the parallax ellipse in the figure shifts by about the width of the line . , scaledwidth=70.0% ] errors ( for the position determinations relative to c1 ) are about twice the size of the circles . a solid line connects each observed position with the corresponding fitted position indicated by a dot on the estimated orbit . the large scatter of the observed positions about the model orbit ellipse is again most likely dominated by the fluctuations of the positions of the stellar radio emission relative to the center of the disk of the primary . figure adapted from . , scaledwidth=70.0% ] . the secondary is the smaller yellow star . the projected orbit of the primary is the same as the inferred radio source orbit shown in figure [ f6morbit ] . the projected orbit of the secondary and the diameters of the two stars correspond to the nominal values given in table [ topt ] . the system is shown with the primary at its ascending node . note that the primary s visible pole is near its southeast side . the size of the figure is approximately 3.8 mas by 3.4 mas . , scaledwidth=95.0% ] a detailed analysis of systematic errors affecting each of the nine fit parameters of s motion relative to the distant universe led us to the following conclusions . for the position at the mean epoch of the vlbi observations of 2001.29 , the systematic error in each coordinate was computed from 0.5 times the angular radius of the primary along the sky projection of the normal to the binary orbital plane ( p.a . = 130.5 @xmath3 8.6 ) . for the later epoch of 2005.08 in table [ tfinal ] , the value for each component was added in quadrature with a possible maximum rms drift during that time interval of the radio brightness reference point of . the rms drift rate was assumed to be not larger than one stellar radius over the vlbi observation time span of 8.5 years or 0.06 in @xmath67 and 0.05 in @xmath43 . the systematic error of the proper motion estimate was found to be mostly comprised of the upper limit of the proper motion of c1 in the crf and of the aforementioned possible maximum rms drift of the radio brightness reference point of . the systematic errors of the parallax and orbit estimates were found to be most clearly defined by the position values of b2250 + 194 relative to those of c1 in 3c 454.3 , since the true parallax and orbit parameter values should be zero . the same parameter fit used for resulted in a parallax estimate for b2250 + 194 relative to c1 of @xmath69 @xmath3 0.074 mas , with the statistical standard error decreasing to 0.026 mas for @xmath63 . for the five - fold smaller separation of from 3c 454.3 the systematic uncertainty was estimated to be @xmath700.015 mas in each coordinate . the same parameter fit also resulted in orbit parameter estimates for b2250 + 194 relative to c1 much smaller than the estimates for itself . for each of the parameters in table [ tfinal ] the total error is also listed . in general it is computed as the root - sum - square of the statistical standard error and the estimated systematic error . for the position and proper motion values , however , the statistical errors are doubled before computing the root - sum - square to allow for correlated noise in the vlbi positions . for the parallax and orbit parameter estimates , the systematic errors were too small to contribute appreciably to the total standard errors . l c c c c @xmath67 at epoch 2005.08 ( errors in mas ) & & 0.12 & 0.33 & 0.40 + @xmath43 at epoch 2005.08 ( errors in mas ) & & 0.13 & 0.29 & 0.39 + @xmath71 ( ) & @xmath72 & 0.026 & 0.073 & 0.090 + @xmath73 ( ) & @xmath74 & 0.030 & 0.074 & 0.095 + parallax ( mas ) & @xmath75 & 0.074 & @xmath700.015 & 0.074 + ( mas ) & @xmath76 & 0.10 & @xmath770.1 & 0.10 + ( mas ) & @xmath78 & 0.11 & @xmath770.1 & 0.11 + ( mas ) & @xmath79 & 0.09 & @xmath770.1 & 0.09 + ( mas ) & @xmath80 & 0.11 & @xmath770.1 & 0.11 + semimajor axis ( mas ) & 0.89 & 0.09 & @xmath770.1 & 0.09 + axial ratio & 0.30 & 0.13 & @xmath770.1 & 0.13 + p.a . of ascending node ( deg ) & 40.5 & 8.6 & @xmath778 & 8.6 + @xmath40 ( heliocentric jd ) & 2450342.56 & 0.44 & @xmath770.4 & 0.44 + comparing our results with the most precise previous measurements of proper motion and parallax , we find that our estimates agree with those in the _ hipparcos _ catalogue @xcite and those of @xcite within their respective larger standard errors . however , we found that our estimates slightly disagree with the values in the _ hipparcos _ re - reduction @xcite by 1.6 and 2.4 times the combined standard error in and in parallax , respectively . this discrepancy in is , however , more than 10 times smaller than the standard error of either the geodetic or the frame - dragging effect @xcite , and therefore of no consequence for the results . comparing the orbital parameters , we find that our estimate of the time of conjunction is consistent with that of @xcite and the combination of our estimates of orbital inclination and semimajor axis are consistent with their value of @xmath32 . this agreement , in turn , is consistent with the reasonable expectation that the radio emission is on average centered on the primary of the system , and that it orbits with the same inclination . any offset in phase of the radio orbit from that of the primary corresponds to less than one - forth of the radius of the primary . with position at epoch , proper motion , parallax , orbital motion , and average location relative to the center of the primary determined , how can the radio emission be described and where does it appear relative to the primary of from epoch to epoch ? figure [ f9image ] displays a set of vlbi images of relative to the center of the primary in its orbit about the binary s barycenter . the rms background ; and contours at 50% and above are drawn in black . the center of each panel ( red dot ) is the fit position of the star s center , as derived from the astrometric results in . in other words , the coordinate origin in our radio image should approximately represent the center of the disk of the primary star . the red circle indicates the angular size of the primary star ( radius of @xmath28 ; * ? ? ? the cyan dotted ellipse shows the orbit of the primary as in and . figure taken from , where also the complete set of our vlbi images is given . , scaledwidth=80.0% ] the analysis of the images showed that the radio emission is highly variable and frequently partly circularly polarized . the morphology is also variable with the average size of the radio emission slightly larger than the disk of the primary . the positions of the peaks of the emission regions are scattered over an area on the sky slightly larger than the disk of the primary . the scatter has a tendency to be distributed along the orbit normal and expected primary s spin axis . comparison with simulations suggest that the brightness peaks preferentially occur near the polar regions similar to the dark spots in the optical @xcite . the height of the emission is likely close to the surface of the primary with 2/3 of the emission peaks located within 0.25 times the stellar radius above the surface . the radio emission may be due to flares linked to a possibly dipolar magnetic field , whose axis is normal to the orbit plane , as could be expected giving the tidally locked rotation of the primary . for a movie of the star based on these vlbi images , see the website , www.yorku.ca/bartel/impeg.mpg , or the online version of . our series of 35 vlbi sessions between 1997 and 2005 resulted in the most comprehensive and detailed radio investigations ever made of a star . * the proper motion of relative to the distant universe is @xmath0 @xmath3 0.09 in and @xmath2 @xmath3 0.09 in , where the errors are intended to represent one - standard deviation errors and include estimates of systematic errors . * these results met the pre - launch requirements of the mission of standard errors no larger than 0.14 in each coordinate so as not to discernibly degrade the estimates of the geodetic and frame - dragging effects . * the parallax of is 10.37 @xmath3 0.07 mas , corresponding to a distance of 96.4 @xmath3 0.7 pc . * the proper - motion and parallax estimates agree with previous estimates , including those in the _ hipparcos _ catalogue , within the combined standard errors , with the exception of the revised _ hipparcos _ results of @xcite where the agreement is within 1.6 and 2.4 times the respective combined standard errors . any discrepancy at these levels is however of no consequence for the results . * the radio emission of is highly variable . * the vlbi images of show emission regions slightly larger than the disk of the primary and their peaks scattered about a similarly sized area , consistent with being located preferentially near the polar regions and within @xmath40.25 stellar radii above the surface . * the vlbi images of were assembled for a movie of the star ( see the website of the first author , www.yorku.ca/bartel/impeg.mpg , or the online version of ) . acknowledgements . we thank each of the following people , listed alphabetically , for their contributions of observations and data reductions to our investigations of and its sky neighborhood , for providing us with unpublished updated catalog data , and/or for enlightening discussions on various aspects of this research program : s. berdyugina , a. buffington , d. buzasi , n. caldwell , r. campbell , j. chandler , t. dame . r. donahue , g. fazio , d. fischer , o. franz , g. gatewood , d. gordon , e. guinan , p. hemenway , a. henden , m. holman , the late j. huchra , p. kalas , g. keiser , j. kolodziejczak , r. kurucz , b. lange , d. latham , j. lederman , j .- f . lestrade , j. lipa , p. luca , s. marsden , h. mcalister , d. monet , n. nunes , l. petrov , r. schild , g. torres , w. traub , d. trilling , n. turner , the late h. wendker , and the late c. worley , this research was primarily supported by nasa , through a contract from stanford university to sao , with a major subcontract from sao to york university . we obtained observations made with the 100-m telescope of the max - planck - institut fr radioastronomie at effelsberg , with antennas of the national radio astronomy observatory ( nrao ) , which is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . , with antennas of the dsn , operated by jpl / caltech , under contract with nasa , and with the nasa / esa hubble space telescope , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . we made use of the sao / nasa astrophysics data system abstract service , initiated , developed , and maintained at sao . , a. l. , gordon , d. , & jacobs , c. s. , ( eds . ) 2009 , iers technical note no . 35 , http://www.iers.org/iers/en/publications/technicalnotes/tn35.html ( frankfurt : verlag des bundesamts fr kartographie und geodsie ) , a. , krichbaum , t. p. , graham , d. a. , et al . 2004 , in european vlbi network on new developments in vlbi science and technology ( trieste : proceedings of science ) , ed . r. bachiller , f. colomer , j .- f . desmurs , & p. de vicente , 710
recently , @xcite reported on the use of minor planets to assess gravity in the outer solar system . that paper was mostly devoted to the use of astrometry of asteroids to investigate gravity in the that region and the potential use of such observations to ascertain the reality of the pioneer effect , an unexplained acceleration towards the sun that perturbs the motion of pioneer 10 and 11 beyond a distance of 20 au @xcite however , some statements were made that applied to comets and the outer planets . in particular , with respect to the outer planets , @xcite commented that the ephemerides of the outer planets are almost entirely based on optical observations and are much less accurate than those for the inner planets . this , coupled with the pioneer effect not being observed in the inner solar system argues that the major planets are not good candidates for investigating gravity in the outer solar system . @xcite , carried out detailed calculations on the impact that the pioneer effect would have on orbital motions of the outer planets . they found a pioneer - like acceleration would produce significant secular and periodic effects and conclude that the absence of such evidence indicates that the pioneer effect is ruled out as a phenomenon effecting the outer planets . the approach of these authors involves analytically and numerically investigating the gaussian rate equations for rates of change in orbital elements as a function of time and a perturbing force . this approach will be discussed more fully later in section [ results ] . additionally , @xcite observes that @xcite offered no calculations to substantiate its conclusions about the outer planets and indicates that other authors reach different conclusions about the use of the outer planets in this role . @xcite also sought to illuminate the issues involved by using a simplified four parameter model of orbital motion . referencing several papers @xcite , @xcite calls the dispute about the observability of pioneer effect - like perturbations in the motion of the outer planets indicative of a lack of consensus in discussions of the planet s motion . both @xcite and @xcite refer to @xcite to buttress their contention that a pioneer - like acceleration should be observable in the motion of the outer planets . in discussing options for a non - dedicated spacecraft investigation of the pioneer effect , @xcite briefly considers whether the effect would be detectable in the motion of the outer planets and conclude that the pioneer effect would have an impact on outer planet motion that would be at variance to that observed . this conclusion is largely based on the change in the @xmath0 product that would be required to result in the pioneer effect at the distance of neptune and the comparison of that value with the otherwise known uncertainty of the @xmath0 value . the basis of their conclusion seems to be similar to that expressed in @xcite , where modifications to kepler s law are used to assess the possibility of several modifications to gravity . @xcite restricts its attention to the case of the inner planets , however , where ranging data is available from both earth - based radar and spacecraft . however , @xcite s conclusion that the pioneer effect would have been observed in the motion of the outer planets is too all - encompassing . for example , some alternative theories of gravity provide for a variation in the force that a spacecraft would experience , as would distributed mass densities concentrated in the outer solar system . the issue here is whether such accelerations are really observationally detectable independent of their source . this paper s purpose is twofold . first , we wish to deal with the problem of astrometric determination of the orbital elements of pluto , in particular by assessing the ability of astrometric observations to detect perturbations of the sort expected from the pioneer effect . secondly , we wish to show that while using simplified models of the dynamics can lead to some insights , one must be careful to not over - simplify the issues involved lest one be misled by the analysis onto false paths . specifically , this paper shows that , contrary to recent assertions in the literature , the current ephemeris for pluto does not preclude the existence of the pioneer effect . we show that the orbit of pluto is currently not well enough characterized to make such an assertion . thus we address the views alluded to above that have now propagated through a number of papers ( ) . we believe that the conclusions presented in those papers are not established unambiguously and that caution needs to be used in drawing further inferences about the dynamics of the outer solar system . notwithstanding our comments about the dangers of using simplified methods , we note that our approach itself is a simplification . the outer planets must be dynamically assessed as a system . in order to rigorously determine whether a small perturbation like the pioneer effect is detectable by astrometry , we should simultaneously include the changes it produces in the orbits of uranus , neptune , and pluto . only in this way can all the second order perturbations to the system be taken into account . however , in order to illustrate the ideas concerned and the weaknesses of the approaches outlined above , we restrict our attention here to manipulating the orbit of pluto . this approach is further discussed in section [ disc ] . the remainder of this paper is divided into four further sections . section [ m&m ] describes several `` back of the envelope '' approaches to understanding the nature of the orbital changes that a pioneer effect would cause , and a detailed discussion of the analysis methodology we have chosen to make our case . section [ results ] addresses our results in terms of the relationships between orbital parameters and observation arc length on the observability of small perturbations . finally , section [ disc ] provides a discussion of our results and section [ concl ] presents conclusions . to begin , we must describe the working definition of the pioneer effect used in this paper . following @xcite , we take the pioneer effect to be manifested by a radial acceleration , directed sunward , of magnitude @xmath1 m s@xmath2 . since the primary purpose of this paper is to investigate the dynamical consequences of the pioneer effect , which apparently begin at about 20 au from the sun , and because there are no data indicating a more gradual onset of the pioneer effect , we will assume that the anomalous acceleration begins abruptly at a heliocentric distance of 20 au . we recognize that this is a simplistic model of the pioneer effect . alternative mechanisms exist that cause the acceleration to vary with object mass , orbital eccentricity , radial distance , and other parameters of the motion . as further observations are made of the pioneer effect , they can be used to investigate different force models to explore various alternatives until the pioneer effect is either ruled out or its origin is found . however , the current status of knowledge of the pioneer effect argues that this simple model should be investigated first . additionally , a perturbation beginning somewhat closer to the sun at a more gradual pace would be more easily seen in the motion of the planets . as shown recently by @xcite , if the pioneer effect occurred at shorter distances from the sun , its effect on planetary ephemerides would have already been detected . thus , our assumptions about the pioneer effect represent the minimum plausible perturbation given the available data . notwithstanding our concentration on the pioneer effect , the analysis presented here should be valid for any small constant perturbation to gravity in the outer solar system . there are several ways to approach assessing whether small perturbations like the pioneer effect can be observed in the motions of the outer planets . for clarity , however , let us make clear that we are talking of perturbations to the motion of the outer planets that are spherically symmetric and directed towards the sun . perturbations due to localized mass concentrations ( i.e. , planet x ) are specifically excluded . however , the latter can be instructive . some years ago , there were allegations that the motion of the outer planets contained anomalies that indicated the presence of a large mass concentration in the outer solar system . @xcite showed how these presumed anomalies in uranus motion vanish when the orbital elements are adjusted while using correct values of neptune s mass as determined by spacecraft . the important point here is that it is not enough to merely compare projected positions ; rather , one must adjust the orbital elements , and even other parameters defining the problem , to best fit observational data . there are a number of approaches to addressing this problem . one approach was taken by @xcite in which a variation in the value of the astronomical unit is investigated for the inner planets . their analysis assumes a small eccentricity and makes use of ranging data available from radar and spacecraft observations . this shows that quite stringent constraints are placed on the nature of gravity at scales approximating that of the inner solar system . in particular , any pioneer - like acceleration at those scales would long since have been detected . another approach is to consider the relative magnitudes of the acceleration due to the sun and that due to the pioneer effect . this is the approach taken in @xcite , although it was not documented in that paper . certainly the pioneer effect should not be expected to have a dominating impact on the motion of the outer planets . the ratio of the pioneer acceleration to that produced by the sun at a distance equal to the semimajor axis of the outer planets is 0.005 , 0.013 , and 0.023 percent for uranus , neptune , and pluto , respectively . if we integrate an orbit without any perturbations to newtonian gravity other than the pioneer effect , and compare with the keplerian case , we find the orbital periods of these objects are systematically shorter . uranus period shortens by 5.8 days and neptune s by 24.1 days , while pluto s period drops by 79.7 days . these intervals correspond to 0.02 , 0.04 , and 0.09 percent of the periods of uranus , neptune , and pluto , respectively . differentiating kepler s third law implicitly , we obtain a relationship between a small change in an orbiting object s period and a corresponding change in the orbit s semimajor axis . in particular , @xmath3 where @xmath4 is the semimajor axis ( in au ) and @xmath5 is the orbital period ( in years ) . for uranus , the fractional period shortening due to the pioneer effect is equivalent to a fractional change in semimajor axis of approximately 0.013 percent . similarly , the orbital period shortening is equivalent to a reduction in semimajor axis of 0.027 and 0.059 percent for neptune and pluto , respectively . these simple calculations imply an equivalent change in aphelion distance of @xmath6 , @xmath7 , and @xmath8 km for uranus , neptune , and pluto . in the cases of uranus and neptune , this is less than the approximate maximum errors in range of @xmath9 km @xcite ) . in the case of pluto , the change in semimajor axis that would correspond to the shortening of its orbital period with the pioneer effect is about twice the radial distance uncertainty . however , pluto has completed less than one - third of an orbit since its discovery and its orbital elements are even less well - determined than the other outer planets . thus , small changes in other orbital elements could easily obscure any orbital changes due to the pioneer effect . the values of orbital elements are not directly observable and to reduce observational data to orbital elements it is necessary to proceed by using numerical approximations . the normal manner in which one proceeds is to first determine a `` preliminary orbit '' from a small number of observations and then to refine it by successive approximations into a `` definitive orbit '' as many more observations are added . the method used to improve the orbit and obtain the definitive orbit through additional observations is called `` differential correction '' and is well described in standard celestial mechanics texts ( e.g. , @xcite , @xcite , or @xcite ) . differential correction uses a least squares approach to iteratively refine the estimates of the elements as more observations become available . additionally , statistical information on the errors of the elements naturally results from the differential correction process . the process results in a set of orbital elements , along with error estimates for the elements and covariance parameters showing the degree to which the elements are correlated . it is worth remembering that the equations governing celestial mechanics are nonlinear and normally a linearized version of the problem is used to determine the covariances and elements . however , the nonlinearity manifests itself in occasional difficulties in determining elements and some of these difficulties will be discussed later in section [ sect3-b ] . the whole chain of analysis outlined above for orbital element determination is associated with a number of errors that must be understood and characterized . some are associated with the description of the solar system dynamics and calculation of the orbits , while some originate in the observations themselves . however , in determining elements from observations , there is no prior warning about the pathological situations that can arise . for example , if the inclination is small , the error in the longitude of the ascending node will be large . similarly , if the eccentricity is small , the error in the argument of perihelion will be large and the time of perihelion passage or mean anomaly will be poorly defined . such problems can be avoided through the use of alternative orbital elements ( e.g. , equinoctial elements ) . thus , the orbital characterization process outlined above must be carried out to enable predictions of the future positions of objects on the sky . the errors associated with the predictions must be compared statistically to determine whether a truly observable difference can be asserted between the alternative models of gravity . importantly , since orbital elements are not directly observable , it is not enough to simply integrate an object s equations of motion forward in time from a set of initial conditions , with and without the perturbation ( as was done in @xcite ) , and then compare the differences . one must first determine element values by adjusting them to a set of observations and a gravity model to obtain a new set of elements with associated error metrics . this can be understood physically by noting that the orbital elements describe the conserved mechanical energy of the body and if the potential field changes due to the existence or absence of a perturbation the elements need to be redetermined . in order to address these issues , we use the orbfit software package @xcite . this program uses observational data and data on the dynamics of the solar system to determine orbital elements and predict ephemerides for minor bodies . since orbfit is available in source code , we added a simple option to include forces arising from the pioneer effect . by comparing orbits resulting from synthetic observations with and without the pioneer effect , we explore the expected effects of such a perturbation on asteroid orbits and examine when this effect can be detected astrometrically . orbfit is a complex piece of software that performs a number of tasks that relate to this analysis . the first task of interest here is that of orbit determination . this process begins with astrometric observations of the body of interest . an initial orbit is determined from a few of the available observations and a process of differential correction is applied to correct the initial elements and take into account any additional observations . the second task of interest performed by orbfit is orbital position prediction . to predict the position of bodies of interest , orbfit takes the position and velocity of the body of interest , and integrates the equations of motion of that body with many additional perturbations due to the other bodies in the solar system , solar oblateness , etc . additionally , the process used by orbfit naturally provides measures of the errors in elements that result from the unavoidable errors in position measurement and the associated element covariances . these data can be mapped on to positions in the sky and the error associated with the prediction , which permit our analysis to take place . by using orbfit for our calculations , we are able to include planetary and other perturbations and factors that affect the determination and prediction of orbits . these factors are dealt with in orbfit to create orbital predictions and also conduct orbital analyses in a manner that very closely mimics the process by which data from observations are reduced to orbital predictions and elements in the real world . in particular , orbfit is able to deal with various errors and noise that are both unavoidable and necessary for calculations at a level of accuracy sufficient for distinguishing between orbits affected by the pioneer effect and those subject to gravity without the additional perturbation . the calculations outlined above represent standard techniques of celestial mechanics and can be performed by any number of systems . orbfit was chosen because of its availability in source code . this permitted addition of a simple option to include the force arising from the pioneer effect , which were simply added to the forces exerted on the bodies of interest by the major planets . rather than integrating the orbits of the main bodies in the solar system , their dynamics are introduced via the jpl de405 ephemeris . in this context the planetary ephemeris is a sort of lookup table containing positions of the planets so that their effect on the motion of a body of interest might be determined . in this paper , we will parametrically vary the eccentricity of hypothetical test bodies , generate synthetic observations for them with and without a pioneer - like acceleration , and determine the conditions under which the pioneer effect can be observed . with the exception of eccentricity , the orbital elements of our test bodies are those of pluto @xcite . the eccentricity of the bodies are 0.001 , 0.005 , 0.01 , 0.05 , 0.1 , 0.2 , and 0.3 . these values are chosen to bracket the actual eccentricity of pluto while extending to nearly circular orbits similar to those found in the other outer planets . the nominal elements at the epoch j2000 ( jd 2,451,545.0 ) , are referred to the mean ecliptic and equinox of jd 2000.0 and are given in table [ elements ] . ll semimajor axis & 39.48168677 au + eccentricity & see text + inclination & 17.14175 degrees + longitude of ascending node & 110.30347 degrees + argument of pericenter & 113.76329 degrees + mean anomaly & 14.86205 degrees + the observation cadence we have chosen to use roughly reflects that which exists for pluto . if one reviews the observations for pluto that were used to construct the de405 ephemeris , we find approximate observation frequencies as indicated in table [ obsfreqs ] . with the exception of the 1914 - 1929 period , we have adopted these cadences and generate synthetic observations accordingly . the pre - discovery observations occuring from 1914 - 1929 are aggregated and an overall rate of one observation per year is used for this time period . lrrl 1914 - 1919 & 11 & 6 & 2 per year + 1920 - 1929 & 6 & 10 & 1 per 2 years + 1930 - 1939 & 431 & 10 & 1 per week + 1940 - 1949 & 233 & 10 & 2 per month + 1950 - 1959 & 113 & 10 & 1 per month + 1960 - 1969 & 113 & 10 & 1 per month + 1970 - 1979 & 364 & 10 & 3 per month + 1980 - 1989 & 361 & 10 & 3 per month + 1990 - 1999 & 1,125 & 10 & 1 per 3 days + 2000 - 2006 & 962 & 7 & 1 per 3 days + 2007 - 2011 & est . & 5 & 1 per 3 days + 2012 - 2163 & est . & 152 & 1 per day + in the analysis , we use observation arc lengths running up to 250 years that are evaluated at 50 year intervals . since the orbital period of these objects is approximately 250 years , our synthetic observations span a complete orbit . the total number of synthetic observations occurring over the observation arc lengths are shown in table [ arclength ] . lrr 1914 - 1963 & 952 & 952 + 1964 - 2013 & 4,212 & 5,164 + 2014 - 2063 & 18,262 & 23,426 + 2064 - 2113 & 18,262 & 41,688 + 2114 - 2163 & 18,262 & 59,950 + the general approach is to use orbfit with a set of elements defined as described in table [ elements ] to generate ephemerides separately in both the perturbed ( that is , with the pioneer perturbation present ) , and unperturbed ( without the pioneer effect ) cases . the predicted positions of the test body on the sky represent `` perfect '' observations with no uncertainties in either observations or elements . in each case , we can take these predicted ephemeris positions and add gaussian observational error . these randomly altered positions then represent the results of synthetic astrometric observations . the random observational error applied is equal to 0.3 seconds of arc , representing the results of good quality ccd astrometry reduced against modern star catalogs . this value of error is assumed to be isotropic on the sky ; thus , we generate and apply synthetic errors in right ascension and declination equal to this value divided by the square root of two with appropriate adjustment to the right ascension for the cosine of the declination . the resulting different sets of synthetic observations and alternative gravity models can then be used with orbfit to determine elements and errors associated with the elements as a function of the eccentricity and the observation arc length . it is noted that this is the procedure used to investigate the motion of minor planets in @xcite and @xcite . we recognize that this assumed error represents a level of accuracy much better than that found in observations taken until relatively recently . indeed , there are many potential sources of position error that might be modeled , some of a known nature and some unknown . the data were taken at different observatories and different techniques may have been used to reduce the observations , leading to systematic errors in the derived positions . the positions have been reduced with various catalogs and transformed , sometimes repeatedly , to different reference frames . even for modern observations , there may be errors in the adopted precession values for the j2000 epoch . additional error sources include errors due to uncertainties of the initial conditions of the other planets ( e.g. , ephemeris errors ) , errors due to corrections to earth rotation , and unmodeled instrumental corrections . since the primary impact of increasing the a priori positional error is to de - emphasize those observations with larger errors , we chose to equally weight all our synthetic observations . thus , we assume much more optimistic error estimates for early observations than are representative in the actual observation archive and results in equal weighting for all the synthetic observations . this should result in our calculations providing an optimistic estimate of the detectability of the pioneer perturbation . an additional reason for electing to model a uniform observational accuracy across the span of the synthetic observations in order to minimize variation in our results due to another source of observational noise . for parts of the synthetic observational arc prior to the advent of modern catalogs and ccds , this error will be considerably less than that found in the actual observational archive . interestingly , the orbfit program is capable of varying a priori observational error across time or across different observatories , but we did not make use of this capability in this paper . given these assumptions , we have four cases , which we illustrate in table [ cases ] . the first pair of cases ( the `` gravity with pe '' column in table [ cases ] ) , and the focus of our analysis , is for observations that are generated with a pioneer - effect perturbation present meaning that the pioneer effect exists in nature and determines an object s motion . these same observations are analyzed with two gravity models . the first model is that of newtonian gravity with the addition of a constant radial acceleration with a value commensurate with that associated with the pioneer effect ; the second is standard newtonian gravity . the comparison of the orbital solutions for these two cases correspond to a situation in which the universe is one where the pioneer effect actually exists and we analyze it to determine which of our two gravity models is correct . this is the situation portrayed in the first column of table [ cases ] . the second pair of cases are comparable except that they are associated with a universe where the pioneer effect does not exist . again , we compare two theories of gravity to see if we can observationally distinguish between the two . this situation is portrayed in the second column of table [ cases ] . this paper investigates the first case only ; as shown in @xcite , the second case is nearly symmetric with the first , and produces similar conclusions . lcc gravity with pe & `` matched '' & `` mismatched '' + only gravity & `` mismatched '' & `` matched '' + as outlined above , our plan is to take ephemerides created with various known orbital elements and predict positions on the sky when the motion is governed by newtonian gravity with and without a pioneer - like perturbing acceleration . this process involves integrating the equations of motion of a body subjected to two different force laws . the results of this calculation provide sky positions as a function of time . since the angular differences between the predicted positions are small , we separately consider the difference in right ascension and declination resulting from the two situations . [ fig1 ] shows the resulting position difference . since the epoch of the elements is jd 2,451,545.0 , the angular differences are zero at that date and diverge as one moves forward and backwards in time from the element epoch . it is worth emphasizing that the two ephemerides being compared are generated with two separate gravity models with identical and exactly specified elements . figure [ fig1 ] can be compared with figs . 9 and 10 in @xcite . they drew the conclusion that the divergence in position , since not observed over the past 90 years , indicates the pioneer effect does not exist . however , such a conclusion is based on the idea that the elements ( i.e. , the initial conditions ) are known precisely . the reality is that these position predictions are those related to the elements derived from the assumption of newtonian gravity coupled with observed positions . if there is a `` mismatch '' between the gravity model used to derive orbital elements and the reality that determines the actual motion of the orbital bodies , these will not be the correct elements . even if the elements are derived from a `` matching '' reality and gravity model , it is remarkable how rapidly orbital predictions can degrade outside the observation arc , especially when that arc is short relative to the orbital period . these aspects will be discussed in detail in the next subsection . if we take the ephemeris positions associated with one `` reality '' ( e.g. , the one containing the pioneer perturbation see table [ cases ] ) consonant with the observation cadence shown in table [ obsfreqs ] , and add an isotropic , normally distributed random position error of 0.3 arc seconds to each of the positions , the result is a set of synthetic positions that are illustrative and representative of those that might have been obtained as observations of the bodies . this set of observations can be divided into five different arcs covering 250 years ( approximately one orbital period ) ; thus , each arc covers an additional 50 years from the start of the observations and represents an incremental one - fifth of an orbital revolution . each synthetic observation arc can be analyzed when a pioneer - like perturbation is included and when it is not , corresponding to the two rows in the first column of table [ cases ] . the observation fitting process serves to determine the elements of the orbit and these are the elements that can then be used to predict sky position . how do the observed positions differ when this procedure is followed ? [ fig2 ] through [ fig6 ] shows the difference in sky position as a function of time for the orbital fits associated with the two gravity models . in all cases , the left hand graph shows the right ascension difference while the right hand graph shows the declination difference . comparing these figures with fig . 1 above and figs . 9 and 10 in @xcite shows that for shorter arcs ( those ranging up to 100 years long ) the difference in predicted motion for the two gravity models is quite small for the duration of the observation arc and only begins to substantially diverge when the position is being extrapolated to times beyond those for which observations exist . for a 150 year arc , small irregularities in the position difference begin to be noticeable during the observational arc . as the arcs grow still longer ( to 200 years and beyond ) , more substantial irregularities are seen . also , it is generally true that positional differences are greater the greater the eccentricity , all other things being equal . recall what the differences plotted in figs . [ fig2 ] through [ fig6 ] represent . synthetic observations for a universe with a pioneer effect were generated and orbital elements were determined by fitting the same observations to two alternative gravity models . one of these models is correctly `` matched '' to the observations ; that is , the gravity model used for orbit fitting is the same one used to generate the observations . the other model does not correspond to the one used to generate the observations . thus , we would a priori expect that the former would fit the data better than the latter . what we see , however , is surprising . for observation arcs as long as a century , which at best is the situation obtaining for pluto at the current time , there are only relatively small differences in the positions predicted by the correct gravity model and those predicted by the mismatched model . in all cases , however , the predicted positions outside the range of observations rapidly diverge from one another . this behavior is a manifestation of two inter - related factors . first , the orbital fitting problem is inherently nonlinear and is normally solved in the linear approximation . even if not mathematically chaotic , the system of equations is certainly sensitively dependent upon initial conditions ; small changes in elements can result in large changes in predicted position outside the range of observations . secondly , this sensitivity is exacerbated by the problem of a short observation arc . the length of the entire observational archive for pluto is no more than about one - third of its orbital period . even in as simple a case as linear least squares curve fitting , a limited amount of independent variable data ( corresponding to a short observation arc ) will lead to relatively large errors in the fitting parameters ( corresponding to the orbital elements ) . together , these factors conspire to potentially generate large position errors outside the observation arc , while increasing the length of the observation arc can markedly reduce error over the whole of the arc and even beyond it . some of these issues can sometimes be alleviated by a change in variables . use of other than keplerian orbital elements ( e.g. , equinoctial elements ) can allow one to avoid some problems with determining an orbital solution , in particular in cases with small eccentricities or small inclinations . similarly , a linear combination of keplerian orbital elements can be used to improve the accuracy of some elements ( but see the footnotes on pages and ) . however , those types of variable change do nothing for the issues associated with the ill - conditioned nature of the linear matrix and a short observational arc . in any case we see that we must fit observations to particular gravity models and adjust orbital elements before predicted positions on the sky can be compared . noting that deriving orbital elements and predicting an orbit beyond a short observation arc provides little information about the actual motion that will be seen once new observations are made , we conclude that drawing conclusions from such extrapolations is totally unwarranted . thus , we can conclude nothing regarding the pioneer effect and the motion of pluto as it is currently known , at least subject to the limitations of our methodology . now let us examine more closely the relatively small position differences at the beginning of the observation arcs in figs . [ fig2 ] through [ fig6 ] . [ fig12 ] is similar to figs . [ fig2 ] through [ fig6 ] , but shows an expanded view of the difference in position of a body with an eccentricity of 0.3 ( e.g. , similar to pluto ) and a 100 year observation arc . at the beginning of the arc , the difference in right ascension for the two gravity models is nearly six seconds of arc and the difference in declination is about 1.5 arcsec . one would certainly think that this level of difference would be observable ; however , we argue below that it is not a distinguishable difference given our current knowledge of pluto s orbit . figure [ fig13 ] shows a subset of the residuals in both right ascension and declination of actual observations with respect to a recent jpl ephemeris , de414 . the full set of residuals show a number of clear outliers and we have removed from the set any residuals of magnitude greater than ten seconds of arc . prior to about 1960 there is clearly a greater spread in the residuals than at later times . additionally , prediscovery images taken prior to 1930 and dating to as early as 1914 have been found in the archive . these observations seem to contain serious irregularities manifested by large and systematically biased residuals . supporting this assertion , @xcite find that systematic differences between the observed and calculated orbit are a continuing issue . additionally , they find that only about half of pluto s motion since its discovery has been observed in a systematic and organized manner . they also comment that it is impossible to extrapolate pluto s position more than a few years into the future . this is indicative of a poorly characterized orbit . the residuals of the synthetic observations with respect to a `` mismatched '' gravity model are similar in character and magnitude to the differences between predictions for different gravity models that are shown in fig . 7 . as observed previously , there seems to be a significant difference between the two sets of predictions . a standard comparison of these models @xcite would calculate and compare @xmath10 values for the two models . if the resulting value is about unity , the fit is good . the overall issues associated with model goodness - of - fit are complex and will be discussed later in section [ quality ] ; however , anticipating this complexity let us look more deeply at the data . if we compare our `` mismatched '' orbit fit with actual observational data and their associated fit , we will see that there are comparable systematic errors in both the synthetic and the real datasets . systematic trends in residuals can be visually discerned more readily by examination of normal points rather than individual data points . in a normal point plot , means and standard deviations of the residuals are collected and plotted for each planetary opposition . fig . [ fig14 ] shows the normal points for the de414 residuals in the `` high - low - close '' portion of the plot , where the `` closing '' value indicated by the small horizontal mark is the average residual for each opposition and the length of the vertical line indicates a one standard deviation variation above and below the mean residuals . the comments above about the systematic bias of the prediscovery observations and the larger dispersion of residuals before 1960 are borne out in the normal plot . now , let us compare this data with normal points for the synthetic observations of pluto . in developing the observation cadence for the synthetic observations , we did not consider the relative positions of the earth , the sun , and pluto and thus did not concentrate synthetic observations around particular yearly oppositions . since our primary purpose was to illustrate broader issues with the data , this was done to avoid introducing additional noise through a varying observational cadence . thus , for the post - discovery residuals associated with our synthetic observations we found the average and standard deviations for each calendar year . for prediscovery images , our assumed observation cadence of one observation per year clearly made the dispersion calculation impossible . in these cases , we accumulated data by decade ; thus , we calculated the mean and standard deviation for the 1914 - 1919 period and the 1920 - 1929 period and plotted the resulting values accordingly . the normal points associated with the synthetic observations are also shown in fig . [ fig14 ] . the average is shown by the solid line and the plus and minus one standard deviation limits are shown as dotted lines . several comments are in order about these graphs . first , the width of the standard deviation spread on the synthetic observations is different from that of the de414 residuals . this is due to our assumption of a constant 0.3 arc second astrometric accuracy of the synthetic observations . the actual observations that contribute to the dispersion of the de414 residuals have varying accuracies , ranging from large values for the early observations to small values for current ccd observations . as pointed out earlier , the level of astrometric accuracy used in the synthetic observations is much better than that found in the actual observation archive until relatively recently . we chose to model a uniform observational accuracy across the span of our synthetic observation arc in order to make an optimistic estimate of the detectability of the pioneer perturbation as well as to minimize variation in our results due to another source of noise ( e.g. , the observation error ) . thus , rather than the uniform difference between the upper and lower limits , the normal points of the actual de414 residuals vary in width . if we had modeled a varying error , such a variation in width would be expected for the synthetic observations as well . another point to be made about fig . [ fig14 ] is that while the prediscovery residuals for the synthetic observations and those with respect to de414 are of opposite signs , their magnitudes are of the same order . in the former case , the large residuals are due to fitting a `` mismatched '' gravity model consisting of only nominal gravity to observations that result from gravity plus an anomalous pioneer acceleration ; in the latter case , unknown observational errors have resulted in large and biased residuals . what are we to make of the difference in sign ? we would argue that since there are errors of an unknown nature present in these early observations , we have absolutely no basis is even assuming the sign of the residual is indicative of reality . although the prediscovery residuals have characteristics that indicate that they might represent a constant offset in right ascension , an effect that can be caused by many errors in the analysis of the observations , the limits on the residual s variation can also accomodate a linear trend as is seen in the corresponding synthetic observations . we can only observe that the approximate magnitude of the residuals at early times are roughly the same for both real observations relative to the de414 ephemeris and the synthetic observations with respect to their ephemeris . thus , we can not draw any conclusions about the existence or nonexistence of the pioneer effect from our current knowledge of the orbit of pluto ; its observation arc is too short for such a determination . a third point with respect to fig . [ fig14 ] is that post - discovery residuals are similar for both the synthetic observations and those of de414 and very close for post-1960 observations . overall , the only place where the two sets of normal points diverge is in the prediscovery era . thus , neither set seems to indicate the existence or nonexistence of an additional gravitational anomaly . both sets of residuals possess patterned irregularities , but it is impossible to say that one is superior to the other . as a final comment on the comparison of the two sets of residuals , fig . [ fig15 ] shows the total rms residual by epoch for the two cases . this quantity is the square root of the sum of the squares of the residuals in right ascension and declination with an appropriate adjustment for the cosine of the declination . as above , de414 s residuals are accumulated by opposition , while the residuals of the synthetic observations relative to their ephemeris are accumulated by decade for prediscovery observations and by year thereafter . the quality of the residuals in the two cases is quite high over most of the observation arc . even in the prediscovery era , the trends of the residuals are similar . as before , there is a linear trend in the synthetic residuals , but the entire prediscovery regime only includes two points . the trend of those two points is not dissimilar to that of the de414 residuals . even if the prediscovery observations possessed a systematic error of as large as two seconds of arc , the trend in rms residual is not substantially different from those of the synthetic observations . from the discovery of pluto to about 1960 , the de414 total rms residuals are greater than those found for the synthetic observations . this difference is expected due to the conservative assumptions of astrometric accuracy that were made in generating the synthetic observations . however , the trend of the residuals in both cases is similar , indicating a comparable fit to the observations . from about 1960 onwards , both sets of residuals fit quite well . given the likely , but uncharacterized , errors in pre - discovery observations , we are led to the observation that real observations are explained about as well by de414 as the synthetic observations are explained by their ephemeris . the context of this comment must be kept in mind . the de414 residuals relative to its ephemeris represent a fit to the actual observation archive by a normal gravity model without a pioneer anomaly . the synthetic observations are created with a pioneer anomaly present and thus their underlying motion is not in keeping with normal gravity . when the synthetic observations are fit to a gravity model without a pioneer anomaly included in the dynamics , the fit is almost indistinguishable from that of the de414 ephemeris . especially given the uncharacterized errors apparently present in the prediscovery observation archive , we can not distinguish between the two gravity models on the basis of the current observational record . however , we note that this is at least as much due to the shortness of the observation arc for pluto as due to any difference in gravity models . what adds special interest to the pluto case , however , is that due to its relatively large eccentricity pluto shows potentially detectable differences between the two gravity models if the entire one hundred year observation arc were uniformly reliable . thus , we are led again to the desirability of continuing to gather astrometric data on pluto and expect that an astrometric difference in position might soon be able to falsify the existence of the pioneer anomaly . nevertheless , for the length of the currently available observation arc and because of the uneven quality of the observational archive , we can not currently distinguish between the two gravity models on the basis of orbital position . there are six orbital elements corresponding to the six degrees of freedom of the classical dynamical problem in three dimensions . given that elements are determined from observations by a model fitting process , we must ask whether this large number of degrees of freedom is somehow redundant and whether a simpler model of the dynamics can be used to illuminate the problem . this is the approach taken by @xcite , where synthetic observations are generated for a simplified dynamical model with four degrees of freedom . conclusions are then drawn about the degree to which a pioneer effect perturbation can be fit to this simplified model and thus whether the pioneer effect exists . the problem with this approach is that the elements that are ignored or suppressed in @xcite are precisely those with the greatest uncertainty resulting from the orbital fitting process . further , as will be discussed below , care must be taken when using rms residuals as a primary measure of goodness of fit as far as model selection is concerned . the underlying cause of the problems outlined above was discussed earlier . briefly , this behavior is due to two inter - related factors : first , the orbital fitting problem is inherently nonlinear and is normally solved in the linear approximation ; second , the sensitivity of the orbital solution to initial conditions is exacerbated , especially in the case of pluto , by the problem of a short observation arc . these issues can sometimes be alleviated by a change in variables . use of other than keplerian orbital elements can result in an improved determination of orbital elements @xcite . however , this type of variable change does nothing for the issues associated with a short observational arc . the uncertainties arising from these sources make the approach in @xcite problematic . in simplifying the problem by assuming away two or more degrees of freedom , the full parameter space , which has ample room to conceal the differences in predicted position , is not available for that purpose . this provides the false impression that perturbation effects are observable when , in fact , they are not . one might think that the pioneer acceleration , assumed to be radial , would only effect the motion of an orbiting body in its orbital plane . in a theoretical sense , this is true . the pioneer effect exerts no torque on the orbiting body , and the plane of the orbit remains unchanged as time progresses . an analytical investigation of such an orbit can be considered as two dimensional and the number of degrees of freedom of the motion can be reduced accordingly . this is the essence of the approach of @xcite . however , in an observational context the motion is not restricted to a plane . there are at least two factors that force orbital motion to be considered to occur in three dimensional space with the associated degrees of freedom . the first is that unless the observer s location ( e.g. , earth ) remains in the plane of the orbit of the body , there will be parallax introduced which forces the motion into three dimensions . similarly , observational errors will unavoidably make the object s position vary from its theoretical planar motion . since both of these are a priori unknown and are only determined in the context of a least squares orbital solution for the observed object , the object s motion must necessarily be considered in its complete dimensionality . of course , this requires all six degrees of freedom of the classical dynamical problem . there can be large errors associated with orbital elements and the extra degrees of freedom can provide ample room to `` hide '' orbital variations . this will be discussed in detail below . as shown in any numerical analysis text ( for example , @xcite ) , an upper limit on the relative error of a fitted model parameter is related to the relative error of the independent variables by a condition number . the condition number can be estimated by the square root of the ratio of the largest to the smallest eigenvalues of the associated normal system of equations @xcitethe differential correction problem can be stated as a linear matrix equation @xmath11 , where @xmath12 is an appropriate jacobian matrix , @xmath13 is a vector of elements , and @xmath14 is a vector of observations . the normal equations are formulated by multiplying this equation by the transpose of @xmath12 ( denoted by @xmath15 ) , giving @xmath16 . this system of equations is solved formally by inverting the @xmath17 product and multiplying again , giving @xmath18 . in general , the @xmath12 matrix is not square . it has a column for each orbital element ( e.g. , six columns ) and a row for each measured sky coordinate ( e.g. , twice the number of observations ) . however , the @xmath17 matrix is square and possesses a corresponding set of eigenvalues . the ratio of the largest to the smallest eigenvalues of the @xmath17 matrix provides the condition number of the normal equations . however , the condition number of the normal equations is the square of the condition number of the @xmath12 matrix . thus , the original equation @xmath11 has a condition number equal to the square root of the condition number of the normal equations . ] . it might be thought that the ill - posed nature of the orbital fitting problem could be obviated through a coordinate transformation . indeed , there are alternative orbital elements that alleviate some difficulties with orbits of too small eccentricity or inclination , for example , equinoctial elements @xcite . alternatively , new variables that are linear combinations of the elements considered here can be used to alter the mix of errors among them . thus , rather than specifying the time of perihelion passage , we can specify some instant of time and then specify either the mean anomaly , the mean longitude ( the sum of the longitude of the ascending node , the argument of perihelion , and the mean anomaly ) , or the true longitude ( the sum of the longitude of the ascending node , the argument of perihelion , and the true anomaly ) at that moment . however , it can be shown that the condition number of a problem like this is invariant under linear coordinate transformations @xcitematrices @xmath12 and @xmath19 are said to be similar when @xmath20 , where @xmath21 is nonsingular . such transformations arise from changes of variables . now , suppose we have an eigenvalue problem , @xmath22 and let us perform this similarity transformation . then @xmath22 implies @xmath23 , which gives @xmath24 which can be stated as @xmath25 so that @xmath12 and @xmath19 have the same eigenvalues and the eigenvectors of @xmath12 and @xmath19 are related by @xmath26 . thus , similarity transformations preserve eigenvalues but not eigenvectors . since the condition number of a problem is the ratio of the largest to smallest eigenvalues , we see that condition number is preserved under linear coordinate transformations . ] although nonlinear coordinate transformations might alleviate this problem , there are none known to the authors that are used in practical astrometric work using orbfit or other standard software packages . fig . [ fig7 ] shows the condition number of our orbital fitting problem as a function of orbital eccentricity and observation arc length . for low eccentricities , the condition number can become quite large . even for eccentricities similar to that of pluto , the condition number for a short arc problem implies large errors in orbital elements . ordinarily , the condition number is important in the context of numerical roundoff with finite precision arithmetic in a computer . this is not the situation here ; however , our precision constraint is the accuracy of the position observations that drive the orbital fitting process . thus an angular precision of 0.3 seconds of arc in right ascension and declination represents a relative error on the order of @xmath27 percent or less . with a condition number of 100 , this translates into an error of about 30 seconds of arc in orbital parameters like mean anomaly and argument of pericenter . the corresponding results when we fit our synthetic observations bear out these theoretical considerations . [ fig8 ] shows the relative error ( one sigma estimated error in the linear approximation ) in orbital elements obtained with different observation arc lengths . the four panels show a representative set of orbital eccentricities . what is generally true for these figures is that for short arcs , the relative error in the elements is much greater than it is for the longer arcs . similarly , the uncertainty in the location of the perihelion ( as measured by the argument of the pericenter and the mean anomaly ) is quite high and remains so for relatively long observation arcs . additionally , since these errors are derived in the linear approximation , the real errors are likely to be greater , perhaps much greater . + this implies a large error in mean anomaly or argument of pericenter , both associated with the spatial and temporal location of perihelion . this , coupled with the observational nature of the astrometric problem necessitating a full three dimensional treatment , forces us to conclude that a substantial amount of error can be absorbed into a multidimensional parameter space and , since it is unclear how the error can be allocated across the parameter space dimensions , the full parameter space must be included to properly reflect the uncertainties of the motion . thus in order to drive the error in orbital elements down to levels allowing us to determine whether the pioneer effect exists , we need a full dynamical model in three - dimensional space and either a relatively high eccentricity object or a long observation arc or both . as observed before , our approach is itself an approximation . a `` full dynamical model '' should include not only adjustment of the orbital elements of pluto , but also simultaneous adjustment to those of uranus and neptune as well . it is only in this way that all the second order perturbations to the system can be taken into account . however , as observed above , in order to illustrate the ideas concerned and the weaknesses of the approaches outlined above , we restrict our attention here to manipulating the orbit of pluto . this approach will be justified further in section [ disc ] . the problem of assessing model validity and comparing alternative models is huge and any kind of complete treatment is well beyond the scope of this paper . however , given the comments above , it is appropriate to offer some observations on this large and interesting topic . the issue of goodness - of - fit for models is not a simple one . often , a @xmath28 is calculated and compared to determine which of competing models is the preferred . however , there are complexities involved . for example , @xcite discusses goodness - of - fit and points out that while use of the @xmath10 statistic as a goodness - of - fit measure for binned data is justifiable and is often done , it has flaws . he further observes that for unbinned data ( such as we have here ) , there is no equivalent popular method for measuring goodness - of - fit . indeed , @xcite gives several examples of problematic goodness - of - fit cases . similarly , @xcite advocates investigating the structure of residuals to find patterns , biases , and sytematic differences between the model and the data . we can divide metrics for model quality into two broad categories that , although useful , are certainly not mutually exclusive . first , we can consider what might be termed point estimates of fit quality . often , a measure of merit is used that is related to the rms residuals between the model and the data . [ fig9 ] compares our rms residuals for various eccentricities . one panel of this figure shows the results of fitting the `` mismatched '' case of the pioneer - perturbed observations being fit to a strictly newtonian gravity model ; the other panel shows the corresponding `` matched '' case where the gravity model includes a pioneer perturbation . in both cases the underlying synthetic observations contain a pioneer perturbation . the most striking feature of this figure is how slowly the residual degrades as the arc length increases , especially for low eccentricity objects . indeed , until the observation arc is over 100 years long , there are only small differences in the quality of the fit as measured by rms residuals for any of the eccentricities evaluated . even in the `` mismatched '' case , the rms residual is still less than half an arc second when the arc is 150 years long . in keeping with the discussion above about the ill - conditioned nature of the orbital fitting problem , we see that strictly relying on rms residuals as a measure of merit for model selection can be problematic . another measure of merit that is sometimes invoked is the uncertainties in the orbital elements themselves . the orbital fitting process automatically provides mathematically well - justified estimates of the elements , although these are in a linear approximation that has its own difficulties since the orbital fitting problem is intrinsically nonlinear . this approximation provides error estimates similar to those shown in fig . [ fig8 ] . across the eccentricities shown , there is substantial error spread throughout the parameter space . in particular , it might be instructive to compare panels a and d in fig . [ fig8 ] . in panel a , a very low eccentricity case , the errors in semimajor axis , inclination , and longitude of the ascending node drop to a level allowing seven or eight significant digits to be present in the corresponding element values . at the same time , however , the uncertainties in eccentricity , argument of the pericenter , and mean anomaly remain relatively large . in the case with the largest eccentricity evaluated , panel d shows the uncertainty to be more uniformly spread across elements . trading off the uncertainties of one element against those of another to compare two models is therefore at best arbitrary and could be misleading . position error in the sky plane is yet another point metric for fit quality . in this case , uncertainties in the orbital elements are mapped directly onto the sky . here , all the comments above about the uncertainty of the elements are valid , plus the observation that the mapping from the six - dimensional space of orbital elements to the two - dimensional sky plane is highly degenerate ; many sets of elements can map to the same region of sky . thus , the sky plane position , by itself , can have problems as a point measure of model fit quality . in particular , comparing the nominal solution for two models can result in a large error volume about the calculated positions of objects . all three approaches to providing point estimates of model fit quality have one common weakness . they are all based on a linearized form of the orbital fitting problem and make the assumptions associated with least squares fitting like normally distributed errors , independence , and no bias or systematic errors . possibly the biggest manifestation of this is that we should not expect to be able to accurately extrapolate very far beyond the available observation arc unless there is at least a full orbital revolution s volume of data and even then , extrapolation is dangerous . a possibly better approach to comparatively assessing model alternatives is to use some type of global comparison . of broadest applicability in this context is the idea of testing for the normality of the residuals between the model and the data . the overall basis for these tests is to look at the structure of the residuals and to perform statistical tests for lack of fit ( see , for example , ( * ? ? ? * section 4.4.4.6 ) ) . the basis for these tests is to search for patterns , biases , and systematic differences between the model and the data . another approach that could be used to address the global quality of fit is to use bootstrap techniques . this approach could involve , as done here , synthesizing observations whose error characteristics reflect those found actual observations and then using the synthetic observations to assess actual variations in the fit quantities . another approach of this sort would be to remove some of the data from the fit and see how the fitted solution extrapolates to the times that were removed . a similar approach would be simply to remove data randomly from the observation set and analyze the resulting variations in fitting parameters . one form that this approach could take would be to conduct a full bayesian analysis of competing models . the distributions of orbital elements resulting from synthetic observations that arise from a monte carlo process could be used to compare the probability of obtaining the observations in light of the competing models ( see , for example , @xcite ) . although beyond the scope of this paper , the findings discussed here show that there are not currently enough data to warrant this type of analysis ; however , the advent of pan - starrs and lsst will change that situation in the near future . in the analysis described above we first showed that one can not simply take orbital elements resulting from the fit of observations to a particular force model and use them to predict positions resulting from motion under the influence of another force . rather , we must refit the orbits to the observations under the new force model . in the case of pluto , this produces a well - fitting orbit that is indistinguishable in a practical sense , at least as long as the observation arc is short enough , from the unperturbed motion . thus , without redetermining the elements we can not make sweeping generalizations about whether or not the outer planets orbits show that the pioneer effect does or does not exist . similarly , although making simplifying assumptions about a physical situation in order to draw conclusions is a time - honored theoretical mode of attack , if the physical model is oversimplified we can be misled into erroneous conclusions . as seen above , in an observational context , both observer position and observational errors lead to the necessity of introducing the third spatial dimension with its associated degrees of freedom . if we do not keep an appropriate number of degrees of freedom , the problem can be oversimplified too much and mislead us into unwarranted conclusions . in particular , we must take care in using such a simplified model to conclude that the pioneer effect does not exist . once again , it should be emphasized that our approach is itself an approximation . to properly conduct an analysis of the sort outlined here , the orbital parameters of the entire system of outer planets should be included . this would bring into the calculation all second order perturbations . however , we have taken a simplified approach and believe it accurately and fairly addresses the uncertainties in the orbit of pluto . one reason we believe our simplified approach is valid is illustrated in fig . [ fig16 ] . [ fig16 ] shows the force per unit mass ( in units of the pioneer acceleration ) exerted by uranus and neptune at pluto s position over the period of time we have considered . perhaps surprisingly , due to the relative positions of the planets in their orbits , neptune exerts less force on pluto than uranus over most of this period . the magnitude of the force is of the order of the pioneer acceleration for both planets . if the pioneer effect exists , we would expect the orbital elements of uranus and neptune to change , but their positions would change by very little . the magnitude of the forces they exert on pluto would therefore change by an amount much less than the current magnitude of those forces . thus , we argue that the approach used in this paper , while not as accurate as a full second order calculation involving all the outer planets , is accurate enough for our purposes . given the comments above , how then are we to compare alternative models ? there are a number of methods that we can use to compare and assess models and their results . in the orbit fitting context , these have been discussed above in terms of point estimates and global estimates of goodness of fit . the simplest point estimate is to simply calculate the rms residual of the fit compared with the input observations . however , as we ve seen , the ill - conditioned nature of the orbital fitting problem can sometimes make the residual a poor candidate for this role . errors , in the linear approximation , to the orbital elements can be assessed to determine the quality of the fit . further physical insight can be gathered from inspecting the sky position errors that result from element errors . global assessments of model fit primarily revolve around the normality of residuals . there are lack of fit tests that can be used to test the residuals and determine if there are any indications of a deficient model . generally , a lack of fit is manifested as patterns , biases , and systematic variations in the residuals , which would indicate a poorly fitting model . which of these metrics is best ? the reality is that determining the quality of a model and comparing model effectiveness is a complex problem . we need all these measures of model fit . statistical tests should be performed comparing predicted positions , taking proper account of the associated errors , to test the hypothesis that the positions predicted under the two dynamical models are different . only in this way can the existence of a perturbation like the pioneer effect be falsified through astrometric methods . as discussed earlier , pluto is very interesting in that orbital fits to the two gravity models show deviations in predicted position that are probably detectable if the entire observation arc consisted of reliable measurements . however , prior to about 1960 , there is a significantly greater dispersion in residuals that at later times . only about half of pluto s motion since its discovery has been the result of systematic and organized observing campaigns @xcite . what adds special interest , however , is that because of its relatively large eccentricity pluto is likely to show differences in position predictions for the two gravity models in the relatively near future as more observations are accumulated . the analysis described above shows two major things . first , we must fit observations to a particular dynamical model and adjust orbital elements before predicted positions on the sky can be compared . orbital parameters are derived from observations which have associated an unavoidable error . the determination of orbits is a model fitting process which has its own associated error sources . extrapolating sky positions very far past the end of an observation arc can result in predicted observations becoming inaccurate so rapidly as to be worthless . the implication of these findings is that `` matched '' and `` mismatched '' gravity models can not be distinguished on the basis of observable sky positions for observation arc lengths similar to those currently obtaining for pluto . similarly , in order to draw conclusions about differences in position in the sky , we must be careful not to oversimplify the dynamical model used to draw the conclusions . suppressing degrees of freedom in the dynamics simplifies the orbital determination problem to just such a degree . the orbital determination problem is nonlinear and the customary solution methods are approximations . thus , any missing or ignored degrees of freedom can , if present , conceal dynamical effects associated with differing gravity models ; we are forced to make use of the full dimensionality of the dynamical problem . in particular , a substantial amount of variation can be absorbed into a multidimensional parameter space and the full parameter space must be considered to properly reflect differences in motion of the outer planets due to the pioneer effect . the problem with the simplified approach is a two - headed one . first , the orbital fitting problem is inherently nonlinear and is normally solved in the linear approximation . even if not mathematically chaotic , the system of equations is sensitively dependent upon initial conditions . thus , small changes in elements can result in large changes in predicted position outside the range of observations . this sensitivity is exacerbated by the problem of a short observation arc . the length of the entire observational archive for pluto is less than about one - third of a complete revolution . together , these factors conspire to potentially generate large errors outside the observation arc , while increasing the length of the observation arc can markedly reduce error over the whole of the arc and even beyond it . we must conclude that we do not know the orbit of pluto as well as we might have thought . we must continue to perform astrometry on it in order to be able to comment on the accuracy with which we know its orbit . using current data , we can not assert that the motion of pluto demonstrates that the pioneer effect does not exist . that jury is still out . of course , this does not mean that the pioneer effect exists . it does mean that we can not deny the existence of the pioneer effect on the basis of motions of the pluto as currently known . further observations are required before such an assertion can be made with confidence . it should again be emphasized that our approach is itself an approximation . the dynamical system that should be analyzed to provide a comprehensive answer to the question of the detectabilty of the pioneer effect should include not only adjustment of the orbital elements of pluto , but also simultaneous adjustment to those of uranus and neptune as well . it is only in this way that all the second order perturbations to the system can be taken into account . our approach here , however serves to illustrate the ideas concerned and the weaknesses of the approaches outlined above . finally , it should be pointed out that , in addition to the observations of individual objects discussed in this paper , there are other related approaches to assessing gravity in the outer solar system . for example , recently @xcite have investigated the use of ensembles of trans - neptunian objects ( tnos ) to ascertain whether their motion reflects unknown additional perturbations and showed that the pioneer effect was not consistent with the motion of tnos . on the other hand , in the area of the observation of individual objects as discussed in this paper , the advent of pan - starrs and lsst in the next several years should provide sufficient data to determine whether the motion of outer solar system bodies reflect the action of unknown forces . this determination should occur over time frames discussed in this paper . however , a combination of the techniques of @xcite and the considerations presented here should provide definitive answers more quickly . the authors wish to thank dr . myles standish for providing the residuals of jpl s de414 ephemeris . we also wish to thank the anonymous referee for a close reading of the manuscript and comments that resulted in significant improvements to the paper . , p. r. , & robinson , d. k. 2003 , data reduction and error analysis for the physical sciences ( data reduction and error analysis for the physical sciences , 3rd ed . , by philip r. bevington , and keith d. robinson . boston , ma : mcgraw - hill , isbn 0 - 07 - 247227 - 8 , 2003 . ) , p. k. 1992 , explanatory supplement to the astronomical almanac ( explanatory supplement to the astronomical almanac , by p. kenneth seidelmann . published by university science books , isbn 0 - 935702 - 68 - 7 , 752pp , 1992 . )
since its discovery fifty years ago @xcite anderson localization has remained a very active field of research ( see for instance the reviews @xcite ) . according to the scaling theory @xcite , there is no delocalized phase in dimensions @xmath20 , whereas there exists a localization / delocalization at finite disorder in dimension @xmath21 . to get some insight into this type of transition , it is natural to consider anderson localization on the cayley tree ( or bethe lattice ) which is expected to represent some mean - field limit . the tight - binding anderson model on the cayley tree has been thus studied by various authors . in @xcite , the recursion equation for the self - energy has been studied to establish the mobility edge as the limit of stability of localized states . in @xcite , the exponent @xmath19 governing the divergence of the localization length was shown to be @xmath22 . in @xcite , the supersymmetric formalism has been used to predict the critical behavior of some disorder - averaged observables ( see also @xcite where similar results have been obtained for the case where anderson tight - binding model is replaced by a non - linear @xmath23-model ) . in @xcite , the problem was reformulated in terms of recursions on riccati variables to obtain weak - disorder expansions . other studies have focused on random - scattering models on the cayley tree @xcite . more recently , the interest in anderson localization on the cayley tree has been revived by the question of many - body localization @xcite , because the geometry of the fock space of many - body states was argued to be similar to a cayley tree @xcite . in the present paper , we consider the anderson tight binding model on the cayley tree already studied in @xcite and we study numerically the statistical properties of landauer transmission and eigenstates as a function of the disorder strength @xmath0 and of the number @xmath1 of generations . we find that several probability distributions propagate as traveling waves with a fixed shape , as the number @xmath1 of tree generations grows . the fact that traveling waves appear in disordered models defined on trees has been discovered by derrida and spohn @xcite on the specific example of the directed polymer in a random medium and was then found in various models @xcite . for the case of anderson localization on trees , the analysis of @xcite concerning the distribution of the self - energy in the localized phase is actually a traveling wave analysis ( see appendix a for more details ) , although it is not explicitly mentioned in these terms in @xcite . the fact that traveling waves occur has been explicitly seen in the numerical study of the transmission distribution for a random - scattering model on the cayley tree ( see fig . 8b of @xcite ) and has been found within the supersymmetric formalism @xcite . in the field of traveling waves and front propagation ( see the reviews @xcite ) , there exists an essential separation between two classes : in pulled fronts , the velocity is determined by the form of the tail of the front and thus by the appropriate linearized equation in the tail region , whereas in pushed fronts , the velocity is determined by the bulk properties and thus by the non - linear dynamics in the bulk region . it turns out that for disordered systems defined on trees , the traveling waves that appear usually correspond to cases where it is the tail of the probability distribution that determines the velocity @xcite . in particular for anderson localization on the cayley tree , the traveling wave propagation is also determined by the tails @xcite . from this traveling wave point of view , the localization / delocalization anderson transition thus represents a traveling / non - traveling critical point . such a traveling / non - traveling phase transition for a branching random walk in the presence of a moving absorbing wall has been studied recently in @xcite and we find here very similar behaviors in the critical region . the paper is organized as follows . in section [ trans ] , we describe the statistical properties of landauer transmission . in section [ eigen ] , we discuss the statistical properties of eigenstates , as measured by the entropy and inverse participation ratios ( i.p.r . ) . we summarize our conclusions in section [ conclusion ] . in appendix a , we translate the analysis of ref . @xcite concerning the distribution of the self - energy in the localized phase into a traveling - wave tail analysis for the landauer transmission discussed in section [ trans ] . in appendix b , we explain how such a similar tail analysis can be performed in the delocalized phase for auxiliary variables that are closely related to inverse participation ratios . finally , in appendix c , we recall the results of ref . @xcite concerning the traveling / non - traveling phase transition for a branching random walk in the presence of a moving absorbing wall , since these results are used as a comparison in the text to understand the finite - size scaling properties in the critical region . in quantum coherent problems , the most appropriate characterisation of transport properties consists in defining a scattering problem where the disordered sample is linked to incoming wires and outgoing wires and in studying the reflexion and transmission coefficients . this scattering theory definition of transport , first introduced by landauer @xcite , has been much used for one - dimensional systems @xcite and has been generalized to higher dimensionalities and multi - probes measurements ( see the review @xcite ) . in dimension @xmath24 , the transfer matrix formulation of the schrdinger equation yields that the probability distribution of the landauer transmission becomes asymptotically log - normal @xcite , i.e. one has @xmath25 where @xmath26 represents the localization length and where @xmath27 is a sample - dependent random variable of order @xmath28 distributed with a gaussian law . although it is often assumed that this log - normal distribution persists in the localized phase in dimension @xmath29 , recent numerical studies @xcite are in favor of the following scaling form for the logarithm of the conductance @xmath30 with exponents of order @xmath31 and @xmath32 @xcite , whereas @xmath33 from eq . [ trans1d ] . for the cayley tree that we consider in this paper , we will find below that the fluctuation exponent vanishes @xmath34 , and we will discuss the probability distribution of the variable @xmath27 . but let us first recall the appropriate scattering framework for the cayley tree @xcite . we consider the anderson tight - binding model @xmath35 where the hopping between nearest neighbors @xmath36 is a constant @xmath37 and where the on - site energies @xmath38 are independent random variables drawn from the flat distribution @xmath39 the parameter @xmath0 thus represents the disorder strength . starting at generation @xmath40 and ending at generation @xmath41 ( on the figure @xmath42 ) is attached to one incoming wire and to @xmath43 outgoing wires . in section [ trans ] , we discuss the properties of the total transmission @xmath44 where @xmath45 is the reflexion amplitude of the incoming wire , and @xmath46 the transmission amplitudes of the outgoing wires.,height=226 ] we consider the scattering geometry introduced in @xcite and shown on fig . [ figscattering ] : the finite tree of branching ratio @xmath47 is attached to one incoming wire at its root ( generation @xmath40 ) and to @xmath43 outgoing wires at generation @xmath41 . one is interested into the eigenstate @xmath48 that satisfies the schrdinger equation @xmath49 inside the disorder sample and in the wires where one requires the plane - wave forms @xmath50 these boundary conditions define the reflexion amplitude @xmath45 of the incoming wire and the transmission amplitudes @xmath46 of the @xmath51 outgoing wires . to satisfy the schrdinger equation of eq . [ schodinger ] within the wires with the forms of eq . [ psiqoutside ] , one has the following relation between the energy @xmath52 and the wave vector @xmath53 @xmath54 to simplify the discussion , we will focus in this paper on the case of zero - energy @xmath55 and wave - vector @xmath56 @xmath57 because the zero - energy @xmath55 corresponds to the center of the band where the delocalizion first appears when the strength @xmath0 of the disorder is decreased from the strong disorder localized phase . inside the cayley tree @xmath58 , the schrdinger equation of eq . [ schodinger ] involves one ancestor denoted by @xmath59 and @xmath47 descendants denoted by @xmath60 @xmath61 whereas for the last generation @xmath41 there is only one ancestor and one descendant ( outgoing wire ) @xmath62 as explained in @xcite , it is convenient to introduce the riccati variables @xmath63 that represent the ratio of the wave function of two neighboring sites . on the outgoing wires , these riccati variables are fixed by eq . [ psiqoutside ] to be @xmath64 the schrdinger equation of eq . [ schodingerbord ] gives the first recursion @xmath65 whereas the schrdinger equation of eq . [ schodingerinside ] gives the recursion inside the tree for @xmath58 @xmath66 on the other hand , the value of the riccati variable for the origin of the tree @xmath40 is fixed by the incoming wire of eq . [ psiqoutside ] @xmath67 and thus the reflexion coefficient @xmath45 can be obtained via @xmath68 from the @xmath69 obtained via the recursion of eq . [ riccatiinside ] . from the conservation of energy , the total transmission @xmath70 is related to the reflexion coefficient @xmath71 @xmath72 as explained in @xcite , the criterion for the localization / delocalization phases is then the following : \(a ) if the riccati variable @xmath69 at the root of the tree converges towards a real random variable as @xmath10 , the reflexion is total @xmath73 and the transmission vanishes @xmath74 . \(b ) if the riccati variable @xmath69 keeps a finite negative imaginary part as @xmath10 , the reflexion is only partial @xmath75 and the transmission remains finite @xmath76 we refer to @xcite for the results of a weak disorder expansion within this framework , and for a numerical monte - carlo approach to determine the mobility edge in the plane @xmath77 . here we will instead study the statistical properties of the transmission @xmath70 at zero energy @xmath55 as a function of the disorder strength @xmath0 ( eq [ flat ] ) and of the number @xmath1 of generations . but before discussing the disordered case , let us first describe the finite - size properties of the pure case . in the pure case where all on - site energies vanish @xmath78 , all branches are equivalent and there is no dependence on @xmath79 . the recursions for riccati variables of eqs [ riccatioutgoing ] , [ riccatifirst ] and [ riccatiinside ] give @xmath80 i.e. one obtains the simple alternation between odd and even generations @xmath81 the reflexion coefficient @xmath45 thus vanishes exactly for any even size @xmath82 @xmath83 and the transmission coefficient in each branch @xmath84 reads @xmath85 ( note that for pure trees of uneven size , the reflexion would not vanish . this is why in the disordered case , we will only consider trees of even sizes @xmath41 where in the corresponding pure case , the reflexion vanish exactly for any finite tree ) . it is instructive to write now the corresponding wave function as a function of the generation n , @xmath86 this exponential decay of the wavefunction in this delocalized case is very peculiar to the tree geometry : it is imposed by the energy conservation and by the exponential growth of the number of sites with the generation @xmath87 . if one wishes to study numerically real trees , one is limited to rather small number of generations of order @xmath88 ( see for instance the study @xcite based on exact diagonalization ) because the number of sites and thus of random energies grows exponentially in @xmath1 . from the point of view of convergence towards stable probability distributions via recursion relations , it is thus better to use the so - called pool method that will allow us to study much larger number of generations ( see below eq . [ pool5 ] ) . the idea of the pool method is the following : at each generation , one keeps the same number @xmath89 of random variables to represent probability distributions . within our present framework , the probability distribution @xmath90 of the riccati variables @xmath91 at generation @xmath87 will be represented by a pool of @xmath89 complex values @xmath92 . it is convenient from now on to change the notation @xmath93 with respect to fig . [ figscattering ] so that @xmath40 now corresponds to the contacts with the outgoing wires ( see fig . [ figscattering ] ) . the numerical results presented below have been obtained with the following procedure : \(i ) the random variables of the initial pool are given by ( eq . [ riccatifirst ] ) @xmath94 where @xmath95 are independent random variables drawn with eq . [ flat ] . \(ii ) from the random variables @xmath96 of the pool at generation @xmath97 , the pool at generation @xmath87 is constructed as follows . for each @xmath98 , one generates a new random energy @xmath99 with the law of eq . [ flat ] and one draws @xmath47 random indices @xmath100 among the pool of generation @xmath101 to construct the variable @xmath102 as follows(eq . [ riccatiinside ] ) @xmath103 \(iii ) from these pools of riccati variables , one may compute via eqs [ reflexioncoef ] and [ deftotaltrans ] a pool of total transmission @xmath104 for trees of @xmath87 generations using @xmath105 for the anderson model on the bethe lattice , the pool method has been already used , in particular in @xcite with a pool @xmath106 with a number @xmath107 of generations , in @xcite with pools up to @xmath108 , and in @xcite with pools up to @xmath109 with a number @xmath110 of generations . the pool method is also very much used for disordered systems on hierarchical lattices ( see for instance@xcite ) , in the remaining of this section , we present the numerical results obtained with a pool of size @xmath111 we have also results for a pool of size @xmath112 with a number of generations @xmath113 to see how the results change with the pool size . however , the number of generations @xmath114 for this bigger pool @xmath115 has turned out to be less precise in the critical region . all figures shown below thus corresponds to data obtained with the pool of size @xmath116 . as is usual with the pool method @xcite , the location of the critical point depends on the pool , i.e. on the discrete sampling with @xmath89 values of probability distributions . it is expected to converge towards the thermodynamic critical point only in the limit @xmath117 ( see the discussion of section [ fspool ] in appendix b ) . nevertheless , for each given pool , the critical behaviors with respect to this pool - dependent critical point usually allows a good measure of critical exponents @xcite . for instance , for the pool of size @xmath116 , the critical value @xmath118 of the disorder strength where the localization - delocalization occurs for the landauer transmission is of order @xmath119 for the pool of size @xmath115 , we find that it is higher and of order @xmath120 . this rather important shift of the pseudo - critical point with the pool size which has already been seen in @xcite , seems to be due to the very slow logarithmic convergence of traveling wave velocity in the presence of cut - off ( see the discussion of section [ fspool ] in appendix b ) . so we stress that here , in contrast to @xcite , our goal is not to determine the true thermodynamic mobility edge @xmath121 , but instead to understand the critical behaviors of the finite - pool results with respect to the pool - dependent critical point of eq . [ critipoole5 ] . we will first discuss the behavior of the typical transmission @xmath122 defined by @xmath123 as a function of the number @xmath1 of generations and disorder strength @xmath0 , before we turn to the distribution around this typical value . in the localized phase @xmath125 : ( a ) linear decay of @xmath126 as a function of the number @xmath1 of generations ( see eq . [ transloc ] ) . ( b ) behavior of the slope @xmath127 as a function of the disorder strength @xmath0 : it vanishes linearly @xmath128 with @xmath129 ( see eq . [ critipoole5 ] ) and @xmath6 ( see eq . [ xiloc ] ) , title="fig:",height=226 ] in the localized phase @xmath125 : ( a ) linear decay of @xmath126 as a function of the number @xmath1 of generations ( see eq . [ transloc ] ) . ( b ) behavior of the slope @xmath127 as a function of the disorder strength @xmath0 : it vanishes linearly @xmath128 with @xmath129 ( see eq . [ critipoole5 ] ) and @xmath6 ( see eq . [ xiloc ] ) , title="fig:",height=226 ] in the localized phase , one expects that the typical transmission @xmath122 defined by eq . [ defttyp ] decays exponentially with the number @xmath1 of generations @xmath130 where @xmath131 represents the localization length that diverges at the delocalization transition @xmath132 . we show on fig . [ figtransloc ] our numerical results for the pool of size @xmath116 ( eq . [ pool5 ] ) : the exponential decay with @xmath1 of eq . [ transloc ] is shown on fig . [ figtransloc ] ( a ) for various disorder strength @xmath0 . the corresponding slope @xmath127 is shown as a function of @xmath0 on fig . [ figtransloc ] ( b ) : we find that this slope vanishes linearly in @xmath133 , in agreement with the exact result @xcite @xmath134 and in agreement with figure 7 of ref . @xcite concerning a random - scattering model on the cayley tree . of the infinite tree in the delocalized phase : ( a ) @xmath137 as a function of the disorder strength @xmath0 ( b ) same data in a log - log plot to measure the exponent of the essential singularity of eq . [ essentialsingularitytransdeloc ] : @xmath138 as a function of @xmath139 : the slope is of order @xmath140 , title="fig:",height=226 ] of the infinite tree in the delocalized phase : ( a ) @xmath137 as a function of the disorder strength @xmath0 ( b ) same data in a log - log plot to measure the exponent of the essential singularity of eq . [ essentialsingularitytransdeloc ] : @xmath138 as a function of @xmath139 : the slope is of order @xmath140 , title="fig:",height=226 ] in the delocalized phase , the typical transmission remains finite in the limit where the number of generations @xmath1 diverges @xmath141 as shown on fig . [ figtransdeloc ] , we measure the following essential singularity behavior of the typical transmission @xmath142 the presence of essential singularities in transport properties near the localization transition on the bethe lattice has been found in @xcite via the supersymmetric formalism ( see also @xcite where similar results have been obtained for the case where anderson tight - binding model is replaced by a non - linear @xmath23-model ) . in particular , eq . 71 of ref . @xcite ) states that the leading critical behavior of the diffusion constant is given by : @xmath143 ( eq . 71 of ref . we note that the exponent in this essential singularity is @xmath144 instead of the exponent of order @xmath145 that we measure . the reason for this difference could be that the results of ref . @xcite are based on the computation of the _ disorder - averaged two - point density - density correlation function _ ( see eq . 2 of ref . @xcite ) , whereas our numerical results concern the _ typical value of landauer transmission _ , and not the disorder - averaged transmission which is expected to be governed by rare events ( see below the section [ distritrans ] ) . also it is not clear to us what exactly represents the diffusion constant computed in @xcite , because it is known that on the bethe lattice the dynamics is not diffusive but ballistic in the delocalized phase @xcite , and that more generally the random walk on the bethe lattice is not diffusive because the tree - geometry induces an effective bias away from the origin ( see for instance @xcite and references therein ) . as a consequence , some implicit reinterpretation of the bethe lattice seems to underlie the statements of ref . @xcite that makes difficult a precise comparison with our present numerical results . if there exists some finite - size scaling in the critical region for the typical landauer transmission of the form @xmath146 the matching of our results in the localized phase ( see eq . [ transloc ] ) and in the delocalized phase ( eq . [ essentialsingularitytransdeloc ] ) requires a finite - size correlation length exponent @xmath147 of order @xmath148 in another traveling / non - traveling phase transition studied in ref @xcite ( see the summary in appendix c ) , it has been obtained that the finite - size scaling exponent @xmath149 is determined by the relaxation rate towards the finite value in the non - traveling phase . we have thus studied the relaxation length towards the finite value in the delocalized phase . we find that our data for @xmath150 are compatible with the form @xmath151 where @xmath152 is a random stationary process as a function of @xmath1 . we find that its autocorrelation function is exponential @xmath153 and we measure that the relaxation length @xmath154 diverges with an exponent @xmath155 of the order of the exponent @xmath147 of eq . [ nufstrans ] . we thus obtain that the critical properties are qualitatively similar to the critical properties described in ref @xcite ( see the summary in appendix c ) : the traveling phase is characterized by a velocity that vanishes linearly , but the finite size scaling is governed by the relaxation length towards the asymptotic finite value of the non - traveling phase . exactly at criticality , we thus expects the following stretched exponential decay of the typical transmission @xmath156 where the exponent @xmath157 is related to the other exponents by ( see the scaling relations of eqs [ kapparelation ] and [ nutyprelation ] in appendix c ) @xmath158 from our previous estimate of the exponent @xmath159 , this would correspond to the numerical value @xmath160 we have not been able to measure this stretched exponential behavior exactly at criticality from our data , because a precise measure of the exponent @xmath157 would require to be exactly at the critical point . of the probability distribution @xmath161 of the logarithm of landauer transmission : ( a ) in the delocalized phase ( here @xmath162 ) , the probability distribution @xmath161 does not move with @xmath1 , and vanishes with a discontinuity at the boundary @xmath163 . ( b ) in the localized phase ( here @xmath164 ) , the probability distribution @xmath161 moves with @xmath1 as a traveling wave of fixed shape . , title="fig:",height=226 ] of the probability distribution @xmath161 of the logarithm of landauer transmission : ( a ) in the delocalized phase ( here @xmath162 ) , the probability distribution @xmath161 does not move with @xmath1 , and vanishes with a discontinuity at the boundary @xmath163 . ( b ) in the localized phase ( here @xmath164 ) , the probability distribution @xmath161 moves with @xmath1 as a traveling wave of fixed shape . , title="fig:",height=226 ] up to now , we have only discussed the behavior of the typical transmission of eq . [ defttyp ] as a function of @xmath1 and @xmath0 . we now turn to the probability distribution of the logarithm of the transmission around its averaged value , i.e. we consider the distribution of the relative variable @xmath165 we find that as @xmath10 , this variable remains finite not only in the delocalized phase where @xmath166 is finite , but also in the localized phase where @xmath167 decays linearly in @xmath1 ( eq . [ transloc ] ) . this means that in the localized phase , the probability distributions @xmath168 actually propagates as a traveling wave with a fixed shape around its moving center @xmath167 as shown on fig . [ fighistotrans ] ( b ) . this phenomenon has already been seen for random scattering models on the bethe lattice ( see fig . 8b of ref . this is in contrast to the broadening with @xmath169 observed in low dimensions @xmath170 ( see eq . [ transd ] ) : the cayley tree thus corresponds to @xmath34 in eq . [ transd ] . of @xmath171 as @xmath0 varies in the localized phase for various disorder strengths @xmath172 : the exponential decay as @xmath173 corresponds to slopes of order @xmath174 , @xmath175 , @xmath176 , @xmath177 . ( b ) width @xmath178 of the probability distribution of @xmath179 as a function of @xmath0 : it remains finite both in the localized phase and in the delocalized phase , but it presents a cusp singularity at criticality . , title="fig:",height=226 ] of @xmath171 as @xmath0 varies in the localized phase for various disorder strengths @xmath172 : the exponential decay as @xmath173 corresponds to slopes of order @xmath174 , @xmath175 , @xmath176 , @xmath177 . ( b ) width @xmath178 of the probability distribution of @xmath179 as a function of @xmath0 : it remains finite both in the localized phase and in the delocalized phase , but it presents a cusp singularity at criticality . , title="fig:",height=226 ] the fact that the shape is fixed can be used numerically to measure more precisely the tails of this probability distribution by accumulating data over iterations : we show on fig . [ figtranswidth ] ( a ) the histograms of @xmath27 obtained by this procedure for various disorder strength @xmath0 . an essential property of this distribution is the right exponential tail @xmath180 in terms of the rescaled transmission @xmath181 ( see eq . [ variableu ] ) , this corresponds to the power - law decay @xmath182 we find that the selected exponent @xmath183 slightly grows as the disorder strength @xmath0 decreases , from a value of order @xmath177 for the strong disorder @xmath184 towards a value of order @xmath174 near criticality ( see figure [ figtranswidth ] a ) . the velocity of the traveling wave propagation of the whole distribution is actually determined by this power - law tail , as explained in detail in appendix a. in particular , the critical value @xmath185 has been predicted in @xcite ( see eq . [ betacdemi ] in appendix a ) . an important consequence of the power - law tail of eq . [ tailpowert ] with @xmath186 is that all integer moments @xmath187 of the landauer transmission will be governed by rare events . finally , we show on fig . [ figtranswidth ] ( b ) the width of the probability distribution of @xmath188 as a function of @xmath0 : it remains finite both in the localized phase and in the delocalized phase , but it presents a cusp singularity at criticality because in the delocalized phase , the distribution presents a singularity at finite distance from the typical value , as is clearly visible on fig . [ fighistotrans ] ( a ) : the transmission is bounded by @xmath189 , i.e. the histogram of @xmath188 presents a discontinuity at @xmath163 . the landauer transmission studied in the previous section is of course the most appropriate observable to characterize the transport properties and to find the transition between the conducting / non - conducting phases . however , one expects that these transmission properties that emerge when the disordered sample is linked to incoming and outgoing wires are related to the nature of eigenstates of the disordered sample in the absence of these external wires ( see fig . [ figtree ] ) to determine whether a normalized eigenstate @xmath190 is localized or delocalized , the usual parameters are the inverse participation ratios ( i.p.r . ) @xmath191 of arbitrary power @xmath192 . another important quantity to characterize the spatial extent of an eigenstate is its entropy @xmath193 the normalization condition yields the idendity @xmath194 , and the entropy corresponds to @xmath195 on an hypercubic lattice of size @xmath169 in dimension @xmath196 , containing @xmath197 sites , the delocalized phase is characterized by the following typical behaviors @xmath198 and @xmath199 these scalings are the same for pure homogeneous eigenfunctions with @xmath200 for all sites @xmath201 . at criticality , the scalings of the i.p.r . involve a whole series of non - trivial exponents and the wave function is said to be multifractal ( see the review @xcite ) . finally in the localized phase , the i.p.r . and the entropy are finite . in the present section , we discuss the statistical properties of these i.p.r . and entropy for zero - energy eigenstates for the cayley tree of fig . [ figtree ] as a function of the number @xmath1 of tree generations and of the disorder strength @xmath0 . where each interior site has @xmath202 neighbors . on the figure the tree ends at generation @xmath42 . , height=226 ] we refer to @xcite where it is explained how eigenstates of finite trees can be constructed and how the density of states can be obtained . note that on a tree , boundary sites dominate so that one needs a substraction procedure to obtain the appropriate bulk density of states . in the following , we are not interested into the density of states , which does not contain any information on the localized / delocalized nature of the spectrum . we wish to study instead the spatial properties of eigenstates of zero - energy @xmath55 ( center of the band ) . the schrdinger equation [ schodingerinside ] yields as before the recursion of eq . [ riccatiinside ] for the riccati variables defined in eq . [ riccati ] . the difference with the scattering case is now in the boundary conditions @xmath203 that replace eq . [ riccatifirst ] . as a consequence , the riccati variables are now real ( and not complex ) . the energy @xmath55 will indeed be an eigenstate only if the schrdinger equation is also satisfied at the center of the finite tree at the center site @xmath204 since the on - site energy @xmath205 is a random variable drawn with some distribution , we may consider that we choose @xmath205 to satisfy eq . [ riccaticenter ] to obtain a typical eigenstate of zero energy and to study its spatial properties . let us first describe zero - energy eigenstate in the pure case to stress the peculiarities of the tree geometry . in the pure case where all on - site energies vanish @xmath78 , the zero - energy eigenstate with radial symmetry reads @xmath206 where the root amplitude @xmath207 is determined by the normalization condition @xmath208 \label{normapur}\end{aligned}\ ] ] this state is the analog of the scattering state described in section [ puretrans ] . in particular , one sees again an exponential decay of the wavefunction in this delocalized case which is very peculiar to the tree geometry : in the normalization condition of eq . [ normapur ] , this exponential decay is exactly compensated by the exponential growth of the number of sites at generation @xmath209 , so that all generations @xmath210 carry exactly the same weight in the normalization condition , and the root weight @xmath211 vanishes as @xmath212 in the thermodynamic limit , where @xmath82 is the number of generations of the tree . for this state , the usual inverse participation ratio of eq . [ defiq ] for @xmath213 reads @xmath214 \label{ipppur}\end{aligned}\ ] ] more generally for @xmath215 , one obtains the following decay with the number of generations @xmath216 in terms of the total number of sites @xmath217 this corresponds to @xmath218 this behavior is thus very anomalous with respect to the corresponding decay as @xmath219 for the i.p.p . of pure states in finite dimension @xmath196 ( see eq . [ delociqdimd ] ) . moreover , since the eigenfunction normalization yields the idendity @xmath194 , one sees that there exists some discontinuity as @xmath220 . in particular , the entropy of eq . [ defs ] which can usually be computed as the derivative of eq . [ sderiiq ] should be computed directly here , and one obtains @xmath221 \label{entropiecaspurcalcul}\end{aligned}\ ] ] using eq [ normapur ] , one obtains the following behavior for large @xmath1 @xmath222 in terms of the total number of sites of eq . [ calncaspur ] , the entropy is proportional to the logarithm of the number of sites @xmath223 that should be compared with the growth as @xmath224 for pure eigenstates in dimension @xmath196 ( see eq . [ delocsdimd ] ) . in conclusion , the tree geometry induces very anomalous scalings in @xmath225 for the logarithm of i.p.r . of pure states [ ipppurcaln ] ) with respect to the case of finite dimension @xmath196 , whereas the entropy is well behaved in @xmath226 ( see eq . [ entropiecaspurfin ] ) . in the disordered case , we expect to observe similar behaviors , whereas the i.p.r . and the entropy will remain finite in the localized phase . to compute the inverse participation ratio @xmath227 ( eq . [ defiq ] ) via recursion , one needs to introduce besides the riccati variable of eq . [ riccati ] the auxiliary variables defined by @xmath228 where the sum is over all sites @xmath229 that are descendants of the site @xmath230 of the tree . these variables satisfy the following recurrence inside the tree @xmath231 @xmath232 \label{reccq}\end{aligned}\ ] ] and the initial conditions at the boundaries @xmath233 at the central root , on needs to impose the normalization @xmath234 \label{normapsi2}\end{aligned}\ ] ] that determines the weight @xmath235 of the root in terms of the variables @xmath236 of the branches . to compute the inverse participation ratio of parameter @xmath192 of eq . [ defiq ] , one needs the variables @xmath237 together with the variables @xmath236 @xmath238 = \frac{\left [ 1 + \displaystyle \sum_{m=1}^{k+1 } c^{(q ) } ( 1,m)\right]}{\left [ 1 + \displaystyle \sum_{m=1}^{k+1 } c^{(1 ) } ( 1,m)\right]^q } \label{iqtree}\end{aligned}\ ] ] to compute recursively the entropy of eq . [ defs ] , one needs similarly to introduce the auxiliary variable @xmath239 where the sum over @xmath229 denotes the sum over the site @xmath230 and all its descendants : @xmath240 thus represents the entropy for the branch containing @xmath230 and its descendants . the initial conditions at the boundaries are simply @xmath241 and the recursion inside the tree @xmath231 can be written as @xmath242 where the first term represents the weighted contribution of the branches entropies and where the second term represents the mixing entropy @xmath243 with the weights @xmath244 normalized to @xmath245 . at the central site of the tree , one uses the same formula but with @xmath246 branches instead of @xmath47 branches . to study the statistical properties of eigenstates , we have used again the pool method explained in [ poolmethod ] . the only difference is that for the transmission , we have followed the recursions for the complex riccati variables , whereas here we follow the recursions for the real riccati variable @xmath91 and for the auxiliary variables @xmath236 , @xmath247 and @xmath23 described above . of a normalized zero - energy eigenstate : ( a ) evolution of the typical entropy @xmath248 with the number @xmath1 of generations : it grows linearly in @xmath1 in the delocalized phase , whereas it remains finite in the localized phase . ( b ) asymptotic width @xmath249 of the relative variable @xmath250 as a function of the disorder strength @xmath0 : the width converges to @xmath251 ( as @xmath212 ) in the delocalized phase , whereas it is finite in the localized phase , title="fig:",height=226 ] of a normalized zero - energy eigenstate : ( a ) evolution of the typical entropy @xmath248 with the number @xmath1 of generations : it grows linearly in @xmath1 in the delocalized phase , whereas it remains finite in the localized phase . ( b ) asymptotic width @xmath249 of the relative variable @xmath250 as a function of the disorder strength @xmath0 : the width converges to @xmath251 ( as @xmath212 ) in the delocalized phase , whereas it is finite in the localized phase , title="fig:",height=257 ] we first consider how the eigenstate entropy defined in eq . [ defs ] evolves with the number @xmath1 of generations . as shown on fig . [ figentropy ] ( a ) , the typical value grows linearly in @xmath1 in the delocalized phase @xmath252 where the factor @xmath253 varies smoothly with @xmath0 and does not vanish continuously near the critical point . in this delocalized phase , we moreover find that the width of the relative variable @xmath254 decays to zero as @xmath10 @xmath255^{1/2}_{w < w_c } \opsimeq_{n \to \infty } \frac{1}{n } \label{entropywidthdeloc}\end{aligned}\ ] ] in the localized phase on the contrary , the typical value and the width of the relative variable @xmath254 remain finite as @xmath256 ( see fig . [ figentropy ] ( a ) and ( b ) ) . in the localized phase ( a ) @xmath257 as a function of the disorder strength @xmath0 near the critical point . ( b ) same data as a function of @xmath258 to measure the exponent of the power law of eq . [ critientropy64loc ] : the slope is of order @xmath259.,title="fig:",height=226 ] in the localized phase ( a ) @xmath257 as a function of the disorder strength @xmath0 near the critical point . ( b ) same data as a function of @xmath258 to measure the exponent of the power law of eq . [ critientropy64loc ] : the slope is of order @xmath259.,title="fig:",height=226 ] as shown on fig . [ fig64loc ] , we find that the typical entropy of an infinite tree of the localized phase diverges with the following power law near criticality @xmath260 the entropy measures the size of the region where the weight @xmath261 is concentrated , whereas the localization length @xmath262 describes the far exponential decay of the transmission @xmath70 ( eq . [ transloc ] ) . we thus conclude that in the localized phase , besides the localization length @xmath262 known since @xcite , there exists a larger diverging length scale @xmath263 that characterizes the size where the weight @xmath261 is concentrated . as explained above , the root weight @xmath264 and of the i.p.r . @xmath265 are determined by the auxiliary variables @xmath237 introduced in eq . [ defcq ] . we thus expect that the localized and delocalized phases correspond to the following behaviors for the auxiliary variables @xmath237 \(i ) in the localized phase , the auxiliary variables @xmath237 will remain finite random variables as the number of generations diverge @xmath10 . then the root weight @xmath266 and the i.p.r . @xmath227 remain finite as @xmath10 . \(ii ) in the delocalized phase , the auxiliary variables @xmath237 will instead grow exponentially with the number @xmath1 of generations with some lyapunov exponents @xmath267 defined by @xmath268 the corresponding typical behaviors of the root weight and of the i.p.r . then reads @xmath269 and @xmath270 we find numerically that the probability distributions of auxiliary variables @xmath271 move as traveling waves of velocity @xmath272 ( see eq . [ lyapunovcq ] ) with fixed shape ( see figure [ fighistoc1c2 ] of appendix b where the fixed shape is shown for @xmath273 and @xmath213 ) . we explain in appendix b how the lyapunov exponents @xmath272 can be determined via a tail analysis that yields the identity ( see eq . [ selectionqidentity ] ) @xmath274 so that the logarithm of the i.p.r . @xmath227 decays slower than linearly in @xmath1 , because the coefficient in eq . [ lyapunoviq ] exactly vanishes . physically , one could expect the decay to be of order @xmath275 as in eq . [ ipppurcaln ] concerning pure states on the cayley tree , but we have not been able to measure the behavior of the i.p.r . , because within the pool method that we use , it turns out that the pool - dependent critical point is not the same for the variables @xmath236 and for @xmath247 ( see more details in section [ fspool ] ) , so that it does not seem easy with the pool method to extract reliable results concerning @xmath265 . in the remaining of this section , we thus focus on the statistical properties of the root weight @xmath235 that is determined by the auxiliary variable @xmath236 via eq . [ normapsi2 ] . the fact that the pool - dependent critical point for @xmath236 is the same as the critical point found for the landauer transmission can be understood from the tail analysis of the appendices a and b that involve exactly the same integral kernel . : the figure shows that the slope @xmath276 vanishes linearly @xmath277 ( see eq . [ xideloc ] with @xmath6 ) . ( b ) width @xmath278 of the relative variable @xmath279 in the limit @xmath256 as a function of the disorder strength @xmath0 : it remains finite both in the localized and delocalized phase , but it presents a cusp at the critical point @xmath118 . , title="fig:",height=188 ] : the figure shows that the slope @xmath276 vanishes linearly @xmath277 ( see eq . [ xideloc ] with @xmath6 ) . ( b ) width @xmath278 of the relative variable @xmath279 in the limit @xmath256 as a function of the disorder strength @xmath0 : it remains finite both in the localized and delocalized phase , but it presents a cusp at the critical point @xmath118 . , title="fig:",height=188 ] in the delocalized phase , the root weight @xmath266 presents the typical decay of eq . [ lyapunovpsi0 ] . we show on fig . [ fig44deloc ] ( a ) that the corresponding lyapunov exponent @xmath276 vanishes linearly in the critical region @xmath280 i.e. we find the same exponent @xmath6 as for the divergence of the correlation length of the localized phase ( see eq . [ nueq1 ] ) . this can be understood from the tail analysis presented in the appendices a and b. we find that the width of the relative variable @xmath279 remains finite in the limit @xmath256 both in the localized phase and in the delocalized phase : its behavior as a function of the disorder strength @xmath0 presents a cusp at @xmath118 as shown on fig . [ fig44deloc ] ( b ) . of an infinite tree in the localized phase : ( a ) @xmath281 as a function of the disorder strength @xmath0 ( b ) @xmath282 as a function of @xmath258 to measure the exponent of the essential singularity of eq . [ essentialsingularity44loc ] : we measure asymptotically a slope of order @xmath283.,title="fig:",height=226 ] of an infinite tree in the localized phase : ( a ) @xmath281 as a function of the disorder strength @xmath0 ( b ) @xmath282 as a function of @xmath258 to measure the exponent of the essential singularity of eq . [ essentialsingularity44loc ] : we measure asymptotically a slope of order @xmath283.,title="fig:",height=226 ] in the localized phase , the root weight remains finite as @xmath10 . as show on fig . [ fig44loc ] , we measure the following essential singularity @xmath284 as in section [ fsstrans ] , we now discuss the finite - size scaling in the critical region . if there exists some finite - size scaling in the critical region for the typical root weight , the matching of our results in the delocalized phase ( see eq . [ lyapunovpsi0 ] and [ xideloc ] ) and in the localized phase ( eq . [ essentialsingularity44loc ] ) requires a finite - size correlation length exponent @xmath285 of order @xmath286 we have performed the analysis described in section [ fsstrans ] and we measure that the relaxation length @xmath154 towards the finite value of eq . [ essentialsingularity44loc ] diverges with an exponent @xmath287 of the order of the exponent @xmath285 of eq . [ nufszero ] . we thus obtain that the critical properties are again qualitatively similar to the critical properties described in ref @xcite ( see the summary in appendix c ) : the traveling phase is characterized by a velocity that vanishes linearly , but the finite size scaling is governed by the relaxation length towards the asymptotic finite value of the non - traveling phase . exactly at criticality , we thus expects the following stretched exponential decay of the typical transmission @xmath288 where the exponent @xmath289 is related to the other exponents by ( see the scaling relations of eqs [ kapparelation ] and [ nutyprelation ] in appendix c ) @xmath290 from our previous estimate of the exponent @xmath291 , this corresponds to the numerical value @xmath292 again , as explained after eq . [ transrhonume ] , we have not been able to measure this stretched exponential behavior exactly at criticality from our data , because a precise measure of the exponent @xmath293 would require to be exactly at the critical point . in summary , for the anderson model on the bethe lattice , we have studied numerically the statistics of the landauer transmission @xmath2 and the statistics of eigenstates at the center of the band @xmath55 , as a function of the disorder strength @xmath0 and the number @xmath1 of generations . we have shown that both the localized phase and the delocalized phase are characterized by the traveling wave propagation of some probability distributions . in the text , we have presented detailed numerical results , and in appendices a and b , we have explained how the velocities of the traveling waves are determined by the tails , via the properties of the integral kernel introduced in @xcite . the anderson transition then corresponds to a traveling / non - traveling critical point for these traveling waves , and the critical properties obtained are very similar to the traveling - wave phase transition studied in @xcite : ( i ) the finite value of the non - traveling phase presents an essential singularity ( ii ) the relaxation length towards this essential singularity determines the finite - size scaling in the critical region ( iii ) the finite - size correlation length exponent @xmath149 is different from the value @xmath6 that govern the vanishing of the velocity in the traveling - wave phase . we thus hope that in the future , these properties for anderson localization on the cayley tree will be better understood by extending the methods that have been developed recently for the class of the fisher - kpp traveling waves ( see @xcite and references therein ) . how results obtained on the bethe lattice are related to properties of anderson localization in finite dimension @xmath196 is of course a difficult question . during the history of localization , many values for the upper critical dimension have been proposed such as @xmath294 . even within the supersymmetric community , there seems to be different interpretations . mirlin and fyodorov @xcite consider that the essential singularities that appear on the bethe lattice , are directly related to the exponential growth of the number of sites with the distance , and will become conventional power - law behaviors as soon as @xmath295 , so that the upper critical dimension is actually @xmath296 . on the contrary , efetov @xcite argues that the results obtained for the bethe lattice are relevant to anderson transition in high dimension @xmath196 when reinterpreted within the so - called effective medium approximation. for the landauer transmission , the upper critical dimension @xmath297 can be defined as the dimension where the exponent @xmath298 concerning the width of the sample - to - sample distribution in the localized phase ( see eq.[transd ] ) vanishes @xmath299 . this means that for @xmath300 , the probability distribution would travel as a traveling wave in the whole localized phase . to determine whether @xmath296 or @xmath301 , it would be thus interesting to understand whether the traveling wave propagation with a fixed shape is possible only on trees or whether it can also occur in sufficiently high dimension @xmath196 . in low dimensions , the probability distribution of the logarithm of the transmission is known to broaden with some exponent @xmath302 ( see eq . [ transd ] ) in the localized phase , whereas the probability distribution of the logarithm of i.p.p . is known to shrink with the system size in the delocalized phase @xcite ( see also fig . ( 4a ) of @xcite where the same phenomenon occurs for the directed polymer in @xmath303 dimensions ) . however traveling wave propagations of the probability distribution @xmath304 of inverse participation ratios @xmath227 have actually been found in finite dimensions but only _ exactly at criticality_. this phenomenon has been first obtained for the power - law random band matrix model @xcite and has been then observed for the anderson model in dimensions @xmath305 @xcite : the motions of the typical values @xmath306 with @xmath307 determine the multifractal spectrum , whereas the reduced variables @xmath308 keep fixed distributions presenting power - law tails with @xmath192-dependent exponents . we have actually observed the same behavior for the directed polymer in @xmath303 dimensions exactly at the localization / delocalization transition ( see fig . ( 3a ) and ( 3b ) of ref . @xcite ) . in all these cases , it would be thus very nice to better understand the relations between the tails of the distributions and the motions of the typical values , and to identify the wave equation that underlies these traveling waves . finally , our work raises once again the question whether anderson transitions are characterized by several length scales that diverge with various @xmath19 exponents . here for the cayley tree , besides the exponent @xmath6 that has been predicted for a long time @xcite , we have found other diverging lengths with the following physical meanings . ( i ) in the localized phase , besides the exponent @xmath6 associated to the divergence of localization length @xmath262 that describes the far exponential decay of the transmission @xmath70 ( eq . [ transloc ] ) , we have found that the critical behavior of the entropy is governed by another diverging length scale with @xmath309 ( eq . [ xisloc ] ) that characterizes the size where the weight @xmath261 is concentrated . ( ii ) in the delocalized phase , we have found that the landauer transmission reaches its asymptotic value after a length diverging as @xmath310 ( eq . [ nufstrans ] ) that governs the finite - size scaling of the transmission in the critical region . even if the precision of numerically measured critical exponents can be always discussed , we feel nevertheless that our numerical results are not compatible with the existence of the single exponent @xmath6 . note that for the directed polymer on the cayley tree , we have also found previously that the critical properties involve two exponents @xmath311 and @xmath312 @xcite . whether there exists various exponents @xmath19 for anderson model in finite dimension @xmath196 has been debated for a long time . the majority of papers on anderson localization seem in favor of a single @xmath19 ( see the reviews @xcite ) , but it seems to us that numerical papers actually study always the same type of observables . in particular , we have not been able to find numerical studies concerning the entropy of eigenstates . on the theoretical side , we are not aware of many theories in favor of various @xmath19 exponents , except ( i ) the supersymmetric studies of refs @xcite that involve two different diverging length scales , called respectively the localization length and the phase coherence length . ( ii ) the pseudo - delocalization transition of the random hopping model in dimension @xmath24 , that is characterized by the two exactly known correlation lengths exponents @xmath22 and @xmath311 ( see @xcite and references therein ) , as in the other models described by the same strong disorder fixed point ( see the review @xcite ) . we feel that if localization models exhibit several @xmath19 exponents both on the cayley tree which represents some mean - field @xmath313 limit and in some models in dimension @xmath24 , the possibility of various @xmath19 for anderson transition in dimension @xmath314 should be reconsidered by studying in detail the statistics of various observables . in this appendix , we translate the analysis of ref . @xcite concerning the distribution of the self - energy in the localized phase for the traveling - wave propagation of the landauer transmission discussed in section [ trans ] . the recursion of eq . [ riccatiinside ] for the complex riccati variable @xmath315 reads more explicitly in terms of its real and imaginary parts @xmath316 with @xmath317-\infty,+\infty[$ ] and @xmath318 . @xmath319 where the random on - site energies @xmath320 are drawn from the flat distribution of eq . the landauer transmission of eq . [ transresume ] then reads @xmath321 in the delocalized phase , this recursion of eq . [ ricattiinsidexy ] is difficult to analyze because one has to find the joint distribution of the two finite variables @xmath322 that remain stable upon iteration . in the localized phase however , the problem is simpler @xcite as we now recall . in the localized phase , if the imaginary part @xmath323 converge towards zero exponentially in @xmath87 as @xmath324 where @xmath325 and where @xmath326 remains a finite random variable upon iteration , one has then to study the simpler recurrence for large @xmath87 @xcite @xmath327 note that the form of eq . [ ynscaling ] corresponds to a traveling wave of velocity @xmath328 for the variable @xmath329 @xmath330 the real part @xmath331 satisfies now a closed recurrence independent of the @xmath323 , and its stable distribution @xmath332 satisfies the closed equation @xmath333 \label{petoilex}\end{aligned}\ ] ] an important property of this distribution is that it presents the power - law tail @xmath334 because whenever one of the @xmath47 variables @xmath335 on the right handside of eq . [ petoilex ] is close to @xmath251 , the variable @xmath336 of the left - handside is large with @xmath337 . the stable joint distribution @xmath338 satisfies @xmath339 \delta \left [ y - \sum_{m=1}^k \frac { y_m } { e^{-v } x_m^2 } \right ] \label{ricattistab}\end{aligned}\ ] ] the idea of @xcite is to look for the power - law tail in @xmath340 that is compatible with the recursion equation , i.e. one assumes @xmath341 with some exponent @xmath342 . note that , in the traveling wave language of eq . [ lnynscaling ] , this is equivalent to look for solutions with an exponential tail @xmath343 for the variable @xmath344 . so this corresponds to the usual exponential tail analysis of fronts @xcite , except for the following difference : in usual studies of propagation into unstable phases , it is the forward tail that has an exponential decay and that determines the velocity , whereas in our present case , it is the backward tail that determines the propagation . in laplace transform with respect to @xmath340 , the power - law decay of eq . [ tail ] corresponds to the following singular expansion near the origin @xmath345 equation [ ricattistab ] becomes for the laplace transform @xmath346 @xmath347 \label{ricattistablaplace}\end{aligned}\ ] ] using the expansion of eq . [ devlaplace ] , one obtains that the function @xmath348 has to satisfy the eigenvalue equation @xmath349 where the function @xmath350 represents the stationary distribution of the variable @xmath351 @xmath352 \label{fonctionq}\end{aligned}\ ] ] ( see eqs 6.5 and 6.6 in @xcite ) . the tail analysis thus leads to the eigenvalue problem the integral kernel appearing in eq . [ eqpropreax ] @xmath353 for each @xmath354 and @xmath0 , the integral kernel is positive , and thus one expects some continuous analog of the perron - froebenius theorem : the iteration of the integral kernel will converge towards a positive eigenvector @xmath355 that is associated to the maximal eigenvalue @xmath356 of the kernel . for instance , for the special case @xmath357 , the solution is simply @xmath358 ( see eqs [ petoilex ] and [ fonctionq ] ) and @xmath359 except for this case @xmath357 , we are not aware of any explicit solution for the eigenvalue @xmath356 ( even for the simpler case where the distribution of the on - site energies @xmath360 is a cauchy law and where the distributions @xmath361 and @xmath350 are also cauchy laws @xcite ) . with eqs [ eqpropreax ] and [ kernel ] , one concludes that at fixed @xmath0 , each mode @xmath354 is associated to the velocity @xmath362 ( eq . [ ynscaling ] ) @xmath363 the selection of the tail exponent @xmath354 of eq . [ tail ] and of the corresponding velocity @xmath364 of eq [ ynscaling ] usually depend on the form of the initial condition @xcite . in our present case , the initial condition is completely localized ( see eq . [ riccatifirst ] ) @xmath365 in this case , one expects that the solution that will be dynamically selected @xcite corresponds to the tail exponent @xmath183 and to the velocity @xmath366 determined by the following extremization @xmath367_{\beta=\beta_{selec}(w ) } = \left [ \frac{1}{\beta^2 } \ln \lambda_0(\beta , w ) - \frac{1}{\beta } \frac { \partial_{\beta}\lambda_0(\beta , w ) } { \lambda_0(\beta , w ) } \right]_{\beta=\beta_{selec}(w ) } \label{selection}\end{aligned}\ ] ] the critical point is then determined by the two conditions @xmath368 since the general discussion is rather obscured by the absence of an explicit expression for the eigenvalue @xmath356 of the kernel of eq . [ kernel ] , it is useful to consider the following strong disorder approximation ( called upper limit condition in @xcite ) , where the recursion for the @xmath331 in eq . [ ricattilinearized ] is simply replaced by @xcite @xmath369 the argument is that in the limit of very large @xmath0 , the distribution of @xmath336 has also a width of order @xmath0 , so that the neglected terms in eq . [ ricattilinearized ] are of order @xmath370 . of course the approximation of eq . [ strongapprox ] is not very well controlled , because it concerns random variables , and it suppresses important correlations between the variables @xmath322 of eq . [ ricattilinearized ] . nevertheless , it is useful to consider it before returning to the true recursions of eq . [ ricattilinearized ] , because with eq . [ strongapprox ] , the eigenvalue @xmath356 is replaced by simple expression @xcite ( see also section 7.2 of @xcite where the approximation of independence between @xmath322 is considered for the cauchy case ) @xmath371 the velocity then reads @xmath372 for the flat distribution @xmath360 of random site energies ( eq . [ flat ] ) , one thus obtains @xmath373 the function @xmath374 is defined on the interval @xmath375 : it flows towards @xmath376 in the limits @xmath377 and @xmath378 with the following behaviors @xmath379 and @xmath380 it has a single maximum at @xmath381 given by the extremum condition of eq . [ selection ] , so that @xmath381 is actually independent of @xmath0 . for @xmath382 , the selected exponent is of order @xmath383 for any @xmath0 . we note that is value is close to the value @xmath177 that we measure for the large disorder @xmath184 ( see figure [ figtranswidth ] a ) . in conclusion , this strong disorder approximation allows to see explicitly how things work on a simple example , and seems to give a reasonable value of @xmath384 for very large @xmath0 . we now return to the analysis of the full problem of eq . [ ricattilinearized ] . here we should say that we do not agree with the discussion of @xcite concerning the selection of the tail exponent @xmath354 : \(i ) in @xcite , the authors conclude that within the localized phase , @xmath183 decreases from unity and reaches @xmath386 exactly at criticality ( see the text between eqs . 6.8 and 6.9 in @xcite ) . \(ii ) in our numerical results of section [ trans ] we have found instead that the selected exponent @xmath183 is always smaller than @xmath144 : it slightly grows as the disorder strength @xmath0 decreases and has a value close to @xmath387 near criticality ( see figure [ figtranswidth ] a ) . as a function of the tail exponent @xmath354 for various disorder strength @xmath0 : ( i ) in the localized phase @xmath125 , the velocity is extremum at a @xmath0-dependent value @xmath388 and it is positive @xmath389 ( ii ) at the critical point @xmath390 , the velocity is extremum at @xmath386 where it vanishes @xmath391 ( iii ) in the delocalized phase , there exists another traveling - wave moving in the other direction of velocity @xmath392 : this velocity is extremum at @xmath393 for any @xmath136 ( see appendix b for more details ) , height=226 ] we propose the following argument to justify our finding @xmath385 . we think that the tail analysis is actually well defined only on the interval @xmath375 for the following reasons . the eigenvalue eq . [ kernel ] relates the behavior of @xmath348 at @xmath394 to the behavior of @xmath348 near the origin @xmath395 . for @xmath396 , the function @xmath397 will be finite in the region where @xmath398 is finite , i.e. the integration is dominated by the region @xmath399 and we obtain the power - law decay @xmath400 however the function @xmath348 has to be integrable at @xmath401 to obtain from eq . [ tail ] the probability of the variable @xmath340 alone at large @xmath340 @xmath402 according to eq . [ axinfinity ] , the function @xmath348 is integrable only for @xmath403 i.e. we obtain that the tail analysis has a meaning only for @xmath404 . in the limit @xmath377 , we expect from eq . [ lambda0beta0 ] the same behavior as in eq . [ vapproxflatzero ] @xmath405 exactly at @xmath393 , the authors of @xcite have argued that @xmath406_{\beta=1/2 } = 0 \label{partialdemi}\end{aligned}\ ] ] for any @xmath0 , as a consequence of the symmetry @xmath407 coming from the consideration of the adjoint kernel ( see more details around eqs . 6.7 and 6.8 in @xcite ) . here in contrast to @xcite , we think that the region @xmath408 is not physical because of eq . [ axinfinity ] , but the condition of eq . [ partialdemi ] is useful to understand why the critical point determined by eqs [ criticonditions ] corresponds to the tail exponent @xcite @xmath409 to summarize the selection mechanism , the shape of the velocity @xmath362 as a function of the tail exponent @xmath354 in shown on fig . [ figvelocity ] for various disorder strength @xmath0 : \(i ) in the localized phase @xmath125 , the velocity derivative is negative at @xmath393 : @xmath410_{\beta=1/2 } = \left [ \frac{1}{\beta^2 } \ln \lambda_0(\beta , w > w_c ) \right]_{\beta=1/2 } = - \left [ \frac{1}{\beta } v(\beta , w > w_c ) \right]_{\beta=1/2 } < 0 \label{derineg}\end{aligned}\ ] ] the velocity is extremum at a @xmath0-dependent value @xmath388 and the corresponding selected velocity is positive @xmath389 . \(ii ) at the critical point @xmath390 , the velocity is extremum at @xmath386 where it vanishes @xmath391 . \(iii ) in the delocalized phase , the velocity derivative is positive at @xmath393 : @xmath411_{\beta=1/2 } = \left [ \frac{1}{\beta^2 } \ln \lambda_0(\beta , w < w_c ) \right]_{\beta=1/2 } = - \left [ \frac{1}{\beta } v(\beta , w < w_c ) \right]_{\beta=1/2 } > 0 \label{deripos}\end{aligned}\ ] ] we explain in appendix b that in this delocalized phase , there exists another traveling - wave moving in the other direction with the velocity @xmath392 : this velocity @xmath412 is then extremum at @xmath393 for any @xmath136 ( see appendix b for more details ) . as a final remark , we believe that the solution exactly at @xmath118 is not valid anymore because of eq . [ axinfinity ] , and that the appropriate treatment exactly at criticality should replace the finite velocity motion in @xmath413 assumed in eq . [ lnynscaling ] by the form @xmath414 where the anomalous exponent @xmath415 has been discussed in the text ( see eqs [ transfss ] and [ transrhonume ] ) . in conclusion , the analysis of @xcite is thus very close to the traveling wave analysis of the directed polymer model @xcite : the difference is that in the directed polymer , the variables @xmath331 are random variables independent of the @xmath323 and the selected @xmath384 can be simply obtained from the non - integer moments of @xmath336 ( see the section 7.2 in @xcite ) , whereas in the present localization model , the variables @xmath331 and @xmath323 are correlated and one has thus to solve the eigenvalue problem for the integral kernel of eq . [ kernel ] . in the following appendix , we explain how the same ideas can be used in the delocalized phase . as explained in section [ eigen ] , in the delocalized phase , we have considered the recurrence for the real riccati variable @xmath331 @xmath416 together with the recurrences of eqs [ reccq ] for the auxiliary variables @xmath237 introduced in eq . [ defcq ] . to make more visible the similarities with the previous appendix , it is convenient to perform the change of variables @xmath417 to obtain the following recursions for these variables @xmath418 the recursion of eq . [ ricattix ] for @xmath331 alone will as before converge to some distribution @xmath332 satisfying eq . [ petoilex ] . in the delocalized phase , one expects that the variables @xmath237 will grow exponentially ( see eqs [ lyapunovcq ] ) . we thus set @xmath419 where @xmath267 governs the exponential growth and where @xmath420 remains a finite random variable upon the iteration @xmath421 the similarity with the discussions of appendix a is now obvious ( see eqs [ ynscaling ] and [ ricattilinearized ] ) . assuming the power - law decay @xmath422 with @xmath423 , one finds that the function @xmath424 has to satisfy the eigenvalue problem @xmath425 i.e. it is the same integral kernel as in eq . [ eqpropreax ] with the correspondences @xmath426 and @xmath427 , but now we are interested in the phase @xmath267 . we have argued above ( see section [ beta0demi ] ) that the problem for the function @xmath424 is well defined only for @xmath428 , i.e. the exponent @xmath429 that governs the power - law of eq . [ dtail ] is restricted to the interval @xmath430 of the variable @xmath431 of eq . [ scalingdn ] for various disorder strength @xmath432 : the exponent @xmath433 of eq . [ dtail ] is of order @xmath434 in the whole delocalized phase . ( b ) for @xmath435 , comparison of the probability distributions @xmath436 and @xmath437 of the variable @xmath438 for @xmath273 and @xmath213 : the exponents of eq . [ dtail ] are respectively @xmath434 and @xmath439 in agreement with eq . [ muselec].,title="fig:",height=226 ] of the variable @xmath431 of eq . [ scalingdn ] for various disorder strength @xmath432 : the exponent @xmath433 of eq . [ dtail ] is of order @xmath434 in the whole delocalized phase . ( b ) for @xmath435 , comparison of the probability distributions @xmath436 and @xmath437 of the variable @xmath438 for @xmath273 and @xmath213 : the exponents of eq . [ dtail ] are respectively @xmath434 and @xmath439 in agreement with eq . [ muselec].,title="fig:",height=226 ] denoting as before @xmath356 the maximal eigenvalue of the kernel of eq . [ kernel ] , each mode @xmath440 is associated to the lyapunov exponent ( see eq . [ ynscaling ] ) @xmath441 from eq . [ lambda0beta0 ] , we have the following behavior for @xmath442 @xmath443 near the other boundary @xmath444 , the property of eq . [ partialdemi ] , yields that the partial derivative with respect to @xmath440 is still negative at @xmath444 @xmath445_{\mu \to 1/(2q ) } = \left [ -\frac{1}{\mu^2 } \ln \lambda_0 ( 1/2,w ) \right]_{\mu \to 1/(2q ) } < 0 \label{derimu}\end{aligned}\ ] ] for the special case @xmath273 , the curve @xmath446 is directly related via @xmath447 to the curve @xmath364 discussed in appendix a and shown on fig . [ figvelocity ] ( see the delocalized case @xmath136 ) . since the selected tail exponent @xmath448 has to extremize @xmath449 , we conclude that the selected tail exponents are given by the simple values @xmath450 in the whole delocalized phase . this is in agreement with our numerical results . we show on fig . [ fighistoc1c2 ] ( a ) the probability distribution of @xmath451 of the variable @xmath452 of eq . [ scalingdn ] for various disorder strength @xmath432 : the tail exponent @xmath433 of eq . [ dtail ] remains of order @xmath434 . for @xmath213 , we find similarly that the selected tail exponent remains the same within the whole delocalized phase and that it is of order @xmath439 . on fig . [ fighistoc1c2 ] ( b ) , we compare for @xmath435 the probability distributions @xmath451 and @xmath453 of the variables @xmath431 : the exponents of eq . [ dtail ] are respectively @xmath434 and @xmath439 . from eq . [ muselec ] , we obtains that the lyapunov exponents satisfy the identities @xmath454 as explained in the text around eq . [ lyapunoviqidentity ] , this identity is important to understand the decay of i.p.r . with the number @xmath1 of generations in the delocalized phase . as explained in section [ poolmethod ] , the pool method consists in representing a probability distribution @xmath455 by a large number @xmath89 of random variables @xmath456 . in particular this introduces a cut - off in the tail of @xmath455 around @xmath457 with @xmath458 . in the field of traveling waves , the presence of such a cut - off has been much studied ( see @xcite and references therein ) : the leading correction to the selected velocity is usually logarithmic in @xmath89 @xmath459 for our present problem , where the critical point corresponds to a vanishing velocity @xmath460 , we thus expect that the true critical point @xmath461 corresponding to @xmath462 will be shifted towards a pool - dependent pseudo - critical point @xmath463 corresponding to @xmath464 . as explained in the text , all the traveling waves encountered in the present paper vanish linearly in @xmath133 . we thus expect that the difference @xmath465 $ ] will also decay only logarithmically as @xmath466 . from the analysis presented in appendix a and b , it is obvious that the pool dependent critical point for the traveling wave of the landauer transmission ( see appendix a ) and for the auxiliary variable @xmath236 or @xmath467 ( see appendix b ) are the same because they are defined by the same condition in terms of the eigenvalue @xmath468 of the integral kernel . however , the other variables @xmath237 or @xmath469 that have the same true critical point @xmath461 , will not have the same pool - dependent critical point @xmath463 , because the constant @xmath470 in eq . [ vpool ] will depend on @xmath192 . this is indeed what we observe with our numerical computations with the pool @xmath116 : the pseudo - critical point @xmath463 for the variable @xmath247 is below the pseudo - critical point @xmath463 for the variable @xmath236 . we observe that the difference between the two becomes smaller for the bigger pool @xmath115 . in conclusion , numerical studies based on the pool method are valid for observables that are related to a single traveling wave , but one can not study the critical properties of observables that depend on two distinct traveling waves that have different pseudo - critical points . this is why in the text , we have not been able to present reliable numerical results for the statistics of i.p.r . @xmath265 that depends on both @xmath236 and @xmath247 . in this appendix , we briefly recall the finite - size scaling properties of the traveling / non - traveling phase transition studied in ref . @xcite , because these properties are useful to interpret our numerical results described in the text . for a one - dimensional branching random walk in the presence of an absorbing wall moving at a constant velocity @xmath328 , the survival probability @xmath471 presents a phase transition at @xmath472 with the following critical behaviors ( see more details in @xcite ) : \(i ) for @xmath473 , it converges exponentially in time towards a finite limit @xmath474 @xmath475 where the limit @xmath476 presents an essential singularity @xmath477 and where the relaxation time @xmath478 diverges as @xmath479 \(iv ) the critical region is described by some finite - size scaling form governed by the relaxation time @xmath478 that appears in eq . [ qbelow ] ( and not by the relaxation time @xmath482 appearing in eq . [ qabove ] ) @xmath485 where the scaling function @xmath486 has the following asymptotic behaviors . for @xmath487 , has the one has the power - law @xmath488 to recover the finite limit of eq . [ qessential ] and one has the scaling relation @xmath489 for @xmath490 , one has the power - law @xmath491 to recover the exponential decay of eq . [ qabove ] , and one has the scaling relation @xmath492 for the problem considered in @xcite all the exponents are exactly known @xmath493 , @xmath494 , @xmath495 and @xmath496 . in our numerical results presented in the text concerning the traveling / non - traveling phase transitions that occur in anderson localization on the bethe lattice , we find very similar critical behaviors : the velocity of the traveling wave vanishes linearly with @xmath493 , and the finite - size scaling in the critical region is governed by the other exponent @xmath19 that appears in the relaxation towards the finite value of the non - traveling phase . majumdar and p.l . krapivsky , phys . e 62 , 7735 ( 2000 ) ; s.n . majumdar and p.l . krapivsky , phys . e 65 , 036127 ( 2002 ) s.n . majumdar and p.l . krapivsky , physica a 318 , 161 ( 2003 ) ; s.n . majumdar , phys . e 68 , 026103 ( 2003 ) . j. prior , a.m. somoza and m. ortuno , phys . rev . b 72 , 024206 ( 2005 ) ; a.m. somoza , j. prior and m. ortuno , phys . rev . b 73 , 184201 ( 2006 ) ; a.m. somoza , m. ortuno and j. prior , phys . 99 , 116602 ( 2007 ) . e. brunet and b. derrida , phys . e 56 , 2597 ( 1997 ) ; e. brunet , b. derrida , j. stat . 103 , 269 ( 2001 ) ; e. brunet , b. derrida , phys . e 70 , 016106 ( 2004 ) ; e. brunet , b. derrida , a. h. mueller , s. munier , europhys . lett . , 76 , 1 ( 2006 ) ; e. brunet , b. derrida , a. h. mueller , and s. munier ; phys . e 73 , 056126 ( 2006 ) ; e. brunet , b. derrida , a. h. mueller , and s. munier phys . e 76 041104 ( 2007 ) .
fault - tolerant quantum computing requires that the fidelities of elementary quantum logic gates exceed certain thresholds @xcite . these fidelity thresholds depend on the architecture . however , even for the most forgiving cases the threshold is high ; therefore , reaching and exceeding it is an extraordinarily difficult task . the implementation of high - fidelity quantum operations is also relevant , although to a lesser extent , in other quantum technologies , including quantum communication and sensing . qualitatively speaking , the fidelity of a gate depends on the ratio between the quantum coherence time and the gate time , implying two approaches to increase gate fidelities : increasing coherence times and reducing gate times . in most qubit architectures , single - qubit gates are implemented by applying an oscillating field that is resonant with the qubit transition frequency . most of the experiments and theoretical studies on such gates were performed with a driving strength that is significantly smaller than the qubit transition frequency . in this weak - driving regime , the qubit undergoes rabi oscillations between the two energy eigenstates at a rate that is proportional to the driving strength . this situation is described rather accurately using the rotating wave approximation @xcite ( rwa ) , which predicts sinusoidal oscillations between the energy eigenstates . with strong driving , the rwa breaks down and the dynamics becomes very complex . despite the increased complexity in controlling dynamics , the strong driving regime is interesting because it brings the opportunity to implement advanced quantum control @xcite and achieve faster quantum operations than in the weak driving regime . a two - level quantum system ( or a qubit ) driven by strong resonant pulses has been studied experimentally using nv centers in diamond @xcite , semiconductor quantum dots @xcite and superconducting circuits @xcite . in a few experiments @xcite , the qubit population exhibits complex dynamics containing a few frequency components , as the driving strength approaches or exceeds the qubit transition frequency , a signature of the breakdown of the rwa . therefore , for the proper design of quantum gates in this regime , different theoretical methods are required . floquet theory @xcite , which provides a general framework for treating periodically driven quantum systems with any driving strength , is the natural method for analyzing strong - driving dynamics . in ref . , the quasienergies and quasienergy states ( hereafter referred to as floquet states ) , as predicted by floquet theory , were observed in a strongly driven superconducting qubit . moreover , the observed dynamics pointed to the important role that the pulse shape plays in the qubit evolution , as determined by adiabaticity conditions in the floquet picture . in this paper , we perform a comprehensive theoretical analysis of qubit dynamics under strong resonant pulses using floquet theory . we derive approximate analytical expressions for the quasienergies and floquet states as functions of the driving amplitude for the practically important case where the qubit is biased at its symmetry point . we then analyze the qubit dynamics induced by a driving pulse , obtained by modulating the amplitude of a periodic signal . the quantum state of the qubit is naturally expressed as a superposition of floquet states . the occupation probabilities of the floquet states remain fixed as long as the driving strength is fixed , but they can change when the pulse envelope varies in time . we analyze the qubit dynamics using a representation of its state in the floquet picture @xcite . when the change in the driving strength is slow , the occupation probabilities of the floquet states remain almost constant , although the floquet states themselves change in accordance with the instantaneous driving strength . the quantum superposition of the floquet states acquires a phase that depends on the evolution of the quasienergies over time . this acquired phase corresponds to a qubit state rotation , generalizing the notion of rabi oscillations to the case of large driving amplitudes . however , a pulse with slowly varying amplitude corresponds to relatively slow quantum gates . we therefore analyze the dynamics with short pulses , when in general nonadiabatic transitions between floquet states can not be neglected . we find that with suitable pulse parameters the nonadiabatic transitions can be largely suppressed . this effect provides the basis of floquet interference efficient suppression of transitions in the adiabatic basis ( fiesta ) , a method to optimize quantum gates with strong driving . finally , we show that fiesta can be used to implement high - fidelity single - qubit operations in very short times , significantly alleviating the effect of decoherence . the remainder of this paper is organized as follows : in sec . [ sec : floquet ] , we derive expressions for the quasienergies and floquet states of a qubit biased at the symmetry point under harmonic driving with arbitrary strength . in sec . [ sec : adiabatic_theory ] , we describe the adiabatic theory in the floquet picture and derive the adiabatic condition . in sec . [ sec : adiabatic ] , we use the adiabatic theory in the floquet picture to describe the quantum state evolution effected by pulses with slowly varying amplitude . in sec . [ sec : nonadiabatic ] , we present a quantitative analysis of the nonadiabatic transitions between the floquet states when the amplitude of the pulse changes rapidly . in sec . [ sec : gates ] , we calculate the fidelities of single - qubit gates optimized so as to suppress nonadiabatic transitions . in sec . [ sec : oqc ] , we discuss the connections between fiesta and other optimal quantum control methods . section [ sec : conclusion ] contains concluding remarks . in this section we discuss the quasienergies and floquet states of a qubit driven with a single - frequency tone . some of the results in this section have been presented in our previous paper @xcite and are included here for completeness . we start with a general model in which a qubit is driven by a harmonic field . the hamiltonian is given by @xmath0 with @xmath1 ( @xmath2 ) the pauli matrices . the ground ( @xmath3 ) and excited ( @xmath4 ) states are ordered such that @xmath5 and @xmath6 . the hamiltonian in eq . ( [ eq : hamstart ] ) is written under the standard convention where in the absence of driving the energy eigenstates are the ground and excited states @xmath3 and @xmath4 . in a frame rotating about the @xmath7 axis with frequency @xmath8 , the hamiltonian is : [ eq : hamrf ] h _ = & -_z + & + a ( t + ) ((t)_x + ( t)_y ) ) . in the weak driving limit ( @xmath9 ) , the terms in eq . oscillating at frequencies @xmath10 can be ignored under rwa . with this approximation the hamiltonian is given by [ eq : hrwa ] h _ = -_z + ( _ x + _ y ) . this is a well - known time - independent hamiltonian representing a spin-1/2 particle precessing in a magnetic field . in this paper , we will analyze the dynamics under the more general hamiltonian , eq . , for large driving amplitude . we will refer to eq . while connecting the general driven dynamics to the special cases in the well - known weak driving regime . for simplicity , we will treat eq . under @xmath11 in the remainder of this section . our derivation can be easily generalized to arbitrary @xmath12 . another alternative form of eq . , which is related to it by a @xmath13 rotation about the @xmath14 axis , is @xmath15 this form is convenient for our analytical derivations for the quasienergies and floquet states ( in which we will closely follow ref . ) , and we shall therefore use this form of the hamiltonian in the remainder of this section . according to floquet theory , for a periodic hamiltonian there exist solutions to the schrdinger equation of the form @xmath16 where @xmath17 are the quasienergies and @xmath18 , to which we shall refer as floquet modes , are periodic with the periodicity of the hamiltonian . because of the periodicity , we can write @xmath19 with @xmath20 , where @xmath21 is the period of the hamiltonian , and the state fourier components @xmath22 are vectors in the hilbert space of the system . each fourier coefficient @xmath22 contains two complex numbers corresponding to the states @xmath3 and @xmath4 : @xmath23 . the index @xmath24 in eq . ( [ eq : generalfloquet ] ) takes a number of values equal to the dimension of the hilbert space . importantly , the floquet states form a complete basis , such that any quantum state can be expressed as a superposition of floquet states . with a fixed driving strength , the solution of the schrdinger equation is a superposition of floquet states with fixed coefficients . substituting eqs . and in the time - dependent schrdinger equation for a qubit with the hamiltonian given by eq . ( [ eq : hamrotated ] ) , we find the relation _ j |u_j , n = ( - _ x + n ) |u_j , n - _ z ( |u_j , n-1 + |u_j , n+1 ) . the above set of equations can be expressed as a single equation : @xmath25 where @xmath26 is the vector @xmath27 , and @xmath28 we now perform a basis transformation where the basis states after the transformation are related to those before the transformation by the formula @xmath29 in other words , when expressed in the original , time - domain representation , |_j , n,0 & = _ m=-^ e^imt j_m - n ( ) |0 , + by the fact that it diagonalizes the hamiltonian @xmath30 when @xmath31 . when considering strong driving , one can use the point @xmath31 as a starting point and treat @xmath32 as a small parameter . the hamiltonian in the new basis reads @xmath33 a sufficiently large truncated version of the matrix @xmath30 yields eigenvectors and eigenvalues that converge well . a size of @xmath34 for example is easily sufficient for reliable numerical calculations . as explained in ref . , truncating this hamiltonian to a @xmath35 matrix gives good results in the strong - driving limit , but it fails in the weak driving limit . as also explained in ref . , a truncation that allows an analytical solution while giving a good approximation for the quasienergies for both weak and strong driving is obtained by using the hamiltonian @xmath36 inspired by the symmetry in the above matrix , which is most easily seen in the limiting case @xmath37 , we now perform a basis transformation @xmath38 , with @xmath39 we obtain @xmath40 this matrix can be split into two @xmath35 decoupled blocks . the outer block gives quasienergies that are close to ( or even degenerate with ) quasienergies that were ignored when we truncated the infinite - dimensional matrix @xmath30 to a @xmath41 matrix . we can therefore expect that the quasienergies obtained from the outer block will not be accurate , and indeed they are not . we therefore focus on the inner block : @xmath42 the eigenvalues of this matrix ( i.e. the quasienergies ) are given by : _ 0 & = ( - - ) , + _ 1 & = ( - + ) . [ eq : quasienergiesanalytical ] the floquet modes are given by the corresponding eigenvectors of the floquet hamiltonian . in the basis of eq . , these eigenvectors are given by @xmath43 where @xmath44 we now note that we have made two transformations to go from the floquet hamiltonian @xmath30 in eq . ( [ eq : floquethamiltonian ] ) to @xmath45 , and we need to transform the eigenvectors back to the original basis . in the basis of eq . , the floquet modes are given by @xmath46 transforming further to the basis of eq . , we find the expressions c + + ) , + c + + ) . [ eq : periodicpartsoffloquetstatesoriginalbasis ] the rabi frequency is given by the difference between the two quasienergies ( see eq . ): @xmath47 ^ 2 + \delta^2 j_1 ^ 2\left(\frac{2a}{\omega}\right ) } . \label{eq : rabifrequencyanalyticalformula}\ ] ] in the case of exact resonance ( i.e. @xmath48 ) , we obtain the expression @xmath49 ^ 2 + j_1 ^ 2\left(\frac{2a}{\omega}\right ) } . \label{eq : rabifrequencyanalyticalformularesonantcase}\ ] ] this expression reduces to the well - known expressions in the weak- and strong - driving limits : when @xmath50 we obtain @xmath51 , and when @xmath52 we obtain @xmath53 , which upon shifting by @xmath8 gives @xmath54 ( and the minus sign here is physically insignificant because it is the absolute value of this expression that gives the meaningful frequency ) . it is worth noting that ref . has an expression for the rabi frequency that resembles eq . ( [ eq : rabifrequencyanalyticalformularesonantcase ] ) . the approach used there , however , is designed to extend the regime of validity from the weak - coupling limit to somewhat large driving strengths , and it breaks down in the strong - driving limit . in contrast , our results are most accurate in the weak- and strong - driving limits , with some deviation from the exact results at intermediate values of @xmath55 . furthermore , the approach of ref . relies on a numerical evaluation of one of the parameters in the argument of the bessel functions , while our approach is purely analytical . , as well as two additional ( and redundant ) quasienergies , as functions of driving amplitude @xmath55 . the solid lines are obtained from numerically solving the schrdinger equation and is essentially exact . the dashed lines are the approximate analytical expressions given by eq . . ( b ) the fidelity of the analytical expression for the floquet modes given by eq . with respect to the exact floquet states , which are evaluated by a numerical integration of the schrdinger equation , as a function of @xmath55 . away from the first two zeros of @xmath56 , and in particular for both the weak and the strong driving limits , the analytical formulae give good approximations for the quasienergies and the floquet modes . the vertical dotted line in ( b ) marks the point of maximum deviation between the analytical and the exact floquet modes.,width=302 ] in fig . [ fig : comparisonbetweenanalyticalformulaandexactresults ] , we plot the quasienergies obtained from eq . and compare them with those obtained from diagonalizing a @xmath57 truncated version of the hamiltonian in eq . . we also plot the fidelity @xmath58 of the floquet modes obtained from eq . with the exact floquet modes : @xmath59 where we have used the definition @xmath60 the analytical formula for the quasienergies agrees very well with the numerical ( and essentially exact ) results , except for small errors around the locations of the floquet - state degeneracy points . the floquet modes agree reasonably well with the exact results for most values of the driving strength . it should be noted here that a fidelity of 0.93 is not very high when dealing with qubit states : in the time - independent case , this value of the fidelity corresponds to an angle of 30 degrees in the bloch sphere representation . this difference in the quality of the approximation for the quasienergies and floquet modes can be understood by considering ( non - degenerate ) perturbation theory . perturbations that appear as off - diagonal matrix elements in the hamiltonian modify the quantum states at the first order but modify the energies only at the second order . in our case these perturbations are the matrix elements that we ignored when we truncated the floquet hamiltonian to a @xmath61 matrix . for example , taking into consideration the fact that @xmath62 has maxima of magnitude @xmath63 , we can estimate that an off - diagonal perturbation of magnitude @xmath64 coupling a quantum state to another one that is @xmath65 in energy away should give a fidelity between the perturbed and unperturbed states of about @xmath66 where @xmath67 , which agrees with the values plotted in fig . [ fig : comparisonbetweenanalyticalformulaandexactresults ] . as one can see in fig . [ fig : comparisonbetweenanalyticalformulaandexactresults ] , there are points where the quasienergies are degenerate , including the point @xmath37 . any superposition of two floquet states that have the same quasienergy is also a floquet state , meaning that at these degeneracy points there is ambiguity ( and freedom ) in how to define the floquet modes . since in this paper we are interested in the dynamics under the influence of short pulses with a varying amplitude @xmath55 , the natural definition of the floquet modes is obtained by considering the path that the pulse takes in parameter space approaching or moving away from any of the degeneracy points . in particular , for a resonant pulse ( i.e. with @xmath48 ) starting from @xmath37 the floquet states at the initial time should be defined by taking the limit @xmath68 . the degeneracy of floquet states at finite values of @xmath55 is a result of the symmetry @xmath69 in the sinusoidal driving waveform that we consider here @xcite . if we modify the waveform and the symmetry is lifted , the quasienergy crossings will turn into avoided crossings . before concluding this section , we address a rather surprising aspect in our analysis . obtaining the expression for the rabi frequency that is valid in the weak - driving limit was more difficult than obtaining the expression that is valid in the strong - driving limit , although the former limit is generally considered to be the simpler of the two limits . the reason is that we followed a derivation that is well suited for the strong - driving limit and modified it in order to extend its validity to the weak - driving limit . if we were interested in the weak - driving limit only , we could have started the derivation differently and written @xmath30 in eq . in the basis @xmath70 with @xmath71 and @xmath72 : @xmath73 near resonance ( i.e. when @xmath74 with @xmath75 ) one can obtain a good approximation of the quasienergies and floquet modes by truncating @xmath30 to the @xmath76 matrix @xmath77 which gives the well known expression for the rabi frequency @xmath78 . this derivation is indeed simpler than the one that we have followed above . it is not obvious , however , how one could modify this derivation and extend its validity to the strong - driving limit . we start this section by giving a derivation of the schrdinger equation in the floquet basis . before any manipulation , the equation reads @xmath79 inspired by the fact that when @xmath55 is independent of time the probability amplitudes of the floquet states are also time independent , we express the quantum state @xmath80 as a superposition in the instantaneous floquet basis : @xmath81 where @xmath82 is the floquet state with index @xmath24 taken at time @xmath83 and @xmath55 is assumed to implicitly depend on time . differentiating this expression with respect to time , we find an alternative expression for the derivative on the left - hand side of eq . ( [ eq : schroedingerequation ] ) : @xmath84 we now note that the partial derivative in the last term corresponds to the full derivative of the floquet states with respect to time assuming that @xmath55 remains constant in time . with this point in mind , we know that the floquet states are solutions of the schrdinger equation : @xmath85 the last term in eq . ( [ eq : expansionofpsidot ] ) therefore cancels the right - hand side in eq . ( [ eq : schroedingerequation ] ) , and the schrdinger equation in the floquet basis reduces to @xmath86 multiplying this equation on the left by @xmath87 , we find that @xmath88 which can alternatively be expressed as @xmath89 the two terms in the above equation have clear meanings . the first term describes the fact that as @xmath55 changes ( and @xmath90 changes with it ) the coefficients @xmath91 acquire phase shifts in order to correct for the difference between the actually accumulated phase ( from the initial time until time @xmath83 ) and the phase that would have accumulated assuming that the instantaneous quasienergy had been in effect from @xmath92 until time @xmath83 . the second term on the right - hand side of eq . ( [ eq : schroedingerequationinfloquetbasis ] ) describes geometric phase accumulation , as well as nonadiabatic transitions between floquet states . the first term in eq . has the undesirable property that its coefficient contains the factor @xmath83 , which grows without bounds , a feature that looks rather unnatural and could complicate calculations for long pulses . this term can be eliminated by defining the coefficients @xmath93 } c_k(t ) . \label{eq : c2ctilde}\ ] ] in terms of the coefficients @xmath94 , the quantum state in eq . takes the form |(t ) = _ j c_j(t ) a clear analogy with quantum state evolution in standard adiabatic theory : the floquet modes @xmath95 are the eigenstates of the instantaneous floquet hamiltonian , the dynamical phases @xmath96 are the time integrals of the quasienergies , and the evolution of the coefficients @xmath94 incorporates geometric phases and nonadiabatic transitions . with the definition of @xmath94 , eq . is transformed into = - _ j e^-i_0^t[_j(t)-_k(t)]dt | |u_j(a , t ) _ [ eq : schroedingerequationinfloquetbasiswithtildes ] note that although we have eliminated the indefinitely growing coefficient with this last transformation we now have a new undesirable feature in the equation of motion , namely the fact that the time derivative of the state now depends on the entire history of the system , as opposed to just the system parameters and state at time @xmath83 . is the schrdinger equation that describes the transition dynamics between the floquet modes . if the driving strength @xmath55 changes very slowly over time @xmath83 , we are in the adiabatic limit of the floquet picture . using eq . together with @xmath97 , we conclude that all the coefficient @xmath98 are constants . in this case , there is no population transition between the floquet modes . from eq . , we find that the state dynamics are fully described by applying a time - dependent dynamical phase @xmath99 to each of the two floquet modes @xmath95 . the constant coefficients @xmath100 can be obtained by decomposing the initial state @xmath101 in the floquet state basis at the initial time , e.g. the basis states @xmath102 if we are considering a pulse whose amplitude @xmath55 starts from zero at the initial time @xmath92 ( and we remind the reader here that setting @xmath37 in @xmath102 should be understood as taking the limit @xmath68 ) . ( upper panel ) and the quasienergy spectrum ( lower panel ) for the case of a qubit under resonant driving , @xmath103 . the quasienergies and floquet modes are calculated by numerically diagonalizing the floquet hamiltonian , eq . . the positions of the avoided crossings ( labeled with dashed lines in the lower panel ) align well with the local maxima of @xmath104.,width=302 ] if the driving strength @xmath55 varies fast over time , nonadiabatic transitions between the floquet modes can be excited . according to eq . , the nonadiabatic transition rate , given by the change of the coefficient @xmath105 , depends on the inner product @xmath106 . in the upper panel of fig . [ fig : innerproduct ] , we plot the amplitude of the inner product @xmath104 for the case of a qubit under resonant driving , @xmath103 . this quantity exhibits multiple local maxima at driving amplitudes which correspond to the positions of the avoided crossings of the quasienergies spectrum ( labeled with dashed lines in the lower panel of fig . [ fig : innerproduct ] ) . in a pulse where the driving strength changes and the system traverses such an avoided crossing , landau - zener transitions can occur between floquet states following the usual formula for landau - zener transitions , just modified from the language of energies to that of quasienergies . the time dependence of the floquet modes @xmath107 appears explicitly in eq . . it is possible to rewrite the equation using the fourier decomposition of these modes , similarly to what is done for example when transforming the time - dependent floquet hamiltonian into its time - independent infinite matrix form . as we shall see shortly , this change makes it possible to evaluate the necessary inner products directly from the eigenstates of the floquet hamiltonian in its time - independent large - matrix form . in order to perform this transformation , we take the inner product on the right - hand side of eq . and express it as a fourier series : @xmath108 where the double - bracket notation is defined in eq . , and we have also used the definition @xmath109 note that @xmath110 is different from @xmath111 , which was introduced in sec . [ sec : floquet ] and is the @xmath112-th fourier coefficient of @xmath95 . apart from the differentiation with respect to @xmath55 , the inner products in eq . can be evaluated by taking the simple inner products between two vectors that are obtained by diagonalizing the hamiltonian @xmath30 . because of the ambiguity in defining @xmath17 and @xmath18 ( more specifically the fact that one can freely move integer multiples of @xmath8 between the two ) , the above definition of @xmath113 naturally suggests the accompanying definition @xmath114 which leads to the relation @xmath115 for all values of @xmath112 . in other words , the different values of @xmath112 give identical copies of the floquet states , although the quasienergies and floquet modes for the different values of @xmath112 are different by integer multiples of @xmath8 and a factor of @xmath116 . equation now becomes @xmath117dt ' } e^{in\omega t } \langle\langle u_{k}(a , t ) | \frac{\partial}{\partial a } | u_{j}^{(n)}(a , t ) \rangle\rangle \tilde{c}_j(t ) \nonumber \\ & = & - \frac{da}{dt } \sum_j \sum_{n=-\infty}^{\infty } e^{-i\int_0^t[\epsilon_{j}^{(n)}(t')-\epsilon_k(t')]dt ' } \langle\langle u_{k}(a , t ) | \frac{\partial}{\partial a } | u_{j}^{(n)}(a , t ) \rangle\rangle \tilde{c}_j(t).\end{aligned}\ ] ] this equation is asymmetric in that it uses the quasienergies and floquet modes both with and without the additional index @xmath112 . it can be made symmetric by first rewriting it in the form @xmath118dt ' } \langle\langle u_{k}^{(0)}(a , t ) | \frac{\partial}{\partial a } | u_{j}^{(n)}(a , t ) \rangle\rangle \tilde{c}_{j}^{(0)}(t ) \label{eq : schroedingerequationinfloquetbasiswithtildesandspaceexpansionunsymmetrized}\end{aligned}\ ] ] and observing that the equation would still be valid if we replaced the index 0 by @xmath119 throughout the equation . we can therefore replace the single equation by an infinite number of equivalent and independent equations . because these equations are linear in the coefficients @xmath120 , one can distribute the two probability amplitudes @xmath121 at the initial time ( with complete freedom ) among the different identical copies of the floquet states ( and hence use the coefficients @xmath122 with @xmath123 ) and solve all the equations in order to find the probability amplitudes at the final time , keeping in mind that after obtaining the probability amplitudes at the final time all the coefficients with the same value of @xmath24 must be summed before calculating the occupation probabilities for the two floquet states , i.e. @xmath124 . noting once more the linearity in the schrdinger equations and the equivalence between the different copies of floquet state , we can rearrange the terms on the right - hand side among the different equations . one possible rearrangement gives the set of equations @xmath125dt ' } \langle\langle u_{k}^{(m)}(a , t ) | \frac{\partial}{\partial a } | u_{j}^{(n)}(a , t ) \rangle\rangle \tilde{c}_{j}^{(n)}(t ) . \label{eq : schroedingerequationinfloquetbasiswithtildesandspaceexpansion}\end{aligned}\ ] ] we remark that the above equation can also be derived using the ( @xmath126 ) formalism @xcite as described in ref . . in applications to quantum control of a single qubit , one often focuses on the qubit s final state at the end of a control pulse or pulse sequence , which is then followed by state readout . in this section , we discuss the final state of the qubit after a resonant pulse whose amplitude varies slowly . in this case , the adiabatic limit in the floquet picture applies . the pulse amplitude @xmath37 at the beginning and the end of the pulse . as a result , we would like to express the initial and final states of the qubit using floquet modes @xmath127 . from eqs . , and , we obtain the floquet modes expressions |u_0(0,t ) & = ( c 1 + e^it + ) , + c 1 + -e^it + ) in the original qubit basis using the bessel functions at value 0 , @xmath128 , and @xmath129 for @xmath103 and @xmath68 . we notice from the above equations that the floquet modes @xmath127 are the time - independent states @xmath130 and @xmath131 , if we define @xmath132 and @xmath133 as the qubit eigenstates in a rotating frame which rotates at an angular frequency @xmath8 . in this rotating frame , the floquet modes at the beginning and at the end of the pulse correspond to the same set of states . the state evolution from the beginning ( @xmath92 ) to the end ( @xmath134 ) of the pulse is simply described by the accumulation of a dynamical phase @xmath135 , where @xmath136 . in the bloch sphere , the above evolution can be seen as a rotation of the state vector by an angle @xmath137 around the x axis , which points towards @xmath138 . the rotation angle @xmath137 , which is the time integral of the quasienergy difference , does not depend on any other details of the pulse shape . considering the case where the rise and fall parts of the pulse are slow and we vary only the duration of the middle part of the pulse with fixed amplitude @xmath139 , the qubit state undergoes rabi oscillations as @xmath137 changes at a constant rate @xmath140 , i.e. the rabi frequency as discussed in sec . [ sec : floquet ] . in the equatorial plane . the direction of the fictitious field in the equatorial plane is determined by the phase @xmath12 of the driving and the amplitude of the fictitious field has an amplitude given by the quasienergy difference @xmath141.,width=302 ] it is worth noting here that although we have only discussed the case where the driving term of the qubit is @xmath142 above , we can easily generalize our results to the driving waveform @xmath143 with an arbitrary constant phase @xmath12 . in this more general case , we obtain the same form of the quasienergies of the system and floquet modes with a phase shift compared to the @xmath144 case . in particular , the floquet modes at @xmath68 are given by @xmath145 and @xmath146 . by choosing a specific value of @xmath12 in the driving signal , one can control the rotation axis of the rabi oscillation in the equatorial plane of the bloch sphere ( see figure [ fig : adaibaticpulseblochrepresentation ] ) . this point appears naturally in the rwa for weak driving , and our analysis shows that it holds for pulses with arbitrary driving amplitude , as long as the pulse amplitude varies slowly . in sec . [ sec : nonadiabatic ] , we will examine more closely the breakdown of the adiabatic approximation . when the driving strength changes rapidly , e.g. in a short pulse , nonadiabatic transitions between floquet states will be reflected in the evolution of the coefficients @xmath94 in eq . . this evolution can be calculated using eq . or equivalently eq . . these two equations are exact . to proceed further analytically , we will employ adiabatic perturbation theory ( apt ) @xcite and then we will compare the solutions given by apt with the exact numerical solutions . in the first order of apt @xcite , we solve eq . iteratively keeping terms up to first order in @xmath147 . this is equivalent to replacing @xmath148 on the right - hand side of eq . . this way we obtain the approximate expression _ k^(m)(t ) = _ k^(m)(0 ) - _ 0^t dt _ j _ n=-^ e^-i_0^t[_j^(n)(t)-_k^(m)(t)]dt u_k^(m)(a , t ) | | u_j^(n)(a , t ) _ j^(n)(0 ) . [ eq : schrodingerequation1storderapt ] setting @xmath149 at the initial time and summing the equation over @xmath119 , we obtain _ k(t ) = _ k(0 ) - _ 0^t dt _ j _ m=-^ e^-i_0^t[_j^(0)(t)-_k^(m)(t)]dt u_k^(m)(a , t ) | | u_j^(0)(a , t ) _ [ eq:1storderapt ] if we quantify the nonadiabatic transition as the change of each coefficient @xmath150 , we can see from the above equation that this quantity is proportional to n_jk(t ) = _ 0^t dt _ m=-^ e^-i_0^t[_j^(0)(t)-_k^(m)(t)]dt u_k^(m)(a , t ) @xmath151 , denotes the nonadiabatic transition matrix element from the @xmath24-th to the @xmath152-th floquet mode . it can be defined as the complex coefficient @xmath153 of the @xmath152-th floquet mode after a driven evolution with the initial state prepared in the @xmath24-th floquet mode , i.e. @xmath154 . @xmath155 is the sum of complex , oscillating terms that will occasionally give a very small value for the sum , as we shall see below . we also note that the coefficient @xmath153 can be calculated essentially exactly by numerically solving the time - dependent schrdinger equation . next , we calculate the nonadiabatic transition matrix elements for a representative pulse shape that has the simple form : a(t ) = ( 1-(t / t_r ) ) & t t_r + a_m & t_r < t t_r + t_p + ( 1 + ( ) ) & t_r + t_p < t t_r+t_p+t_f , [ eq : pulseshape ] where @xmath156 is the duration of the middle part of the pulse , @xmath157 ( @xmath158 ) is the rise ( fall ) time of the pulse , and @xmath159 is the maximum amplitude of the pulse . we choose to analyze pulses with the rising and falling edge being cosine functions because it is a pulse shape that requires a small bandwidth , minimizing required experimental resources . nonadiabatic transitions occur in both the rising and falling edges of the pulse . we will deal with these two cases separately in the remainder of this section . we will consider the case of resonant driving ( @xmath48 ) and @xmath144 here but our treatment can be generalized to arbitrary values of @xmath8 and @xmath12 . for the pulse shape given by eq . , the rising edge is in the time interval between 0 and @xmath160 . for performing the exact numerics to simulate the nonadiabatic transitions , we assume the initial state of the system to be @xmath161 which coincides with the floquet mode @xmath162 . thus @xmath163 and @xmath164 . we then use a numerical solver to determine the system evolution under the time - dependent hamiltonian in eq . with the specified pulse shape . the final state @xmath165 is written as a superposition of the floquet modes @xmath166 , yielding the floquet mode coefficients @xmath167 and @xmath168 . in this case , the quantity @xmath168 gives an measure of the nonadiabatic transitions during the rising edge of the pulse . in fig . [ fig : nonadiabatic](a ) , we show @xmath168 as a function of the maximum amplitude @xmath159 and the rise time @xmath157 . for comparison we also evaluate the nonadiabatic transition amplitudes using apt . for this purpose , we first numerically calculate all the instantaneous quasienergies and floquet modes for values of @xmath55 from @xmath92 to @xmath169 in small steps . we then numerically evaluate @xmath170 according to eq . , and we plot its absolute value in fig . [ fig : nonadiabatic](b ) . in fig . [ fig : nonadiabatic](c ) , we show the nonadiabatic transition amplitude calculated both by exact numerics and by apt as a function of @xmath157 for a few values of the strong - driving pulse amplitude @xmath159 . we find good agreement between the apt solution and the exact results . considering the physics of landau - zener transitions , and noting that the minimum gap in the first avoided crossing of quasienergies is @xmath65 , we find that for @xmath171 adiabaticity is determined by the product @xmath172 . as expected , for long rise times @xmath173 where the evolution should be adiabatic , we have @xmath174 . for short rise times @xmath175 , the nonadiabatic transition amplitude oscillates as a function of @xmath157 . this oscillatory behavior is a result of interference between different paths that correspond to the nonadiabatic transitions during the rise time . we notice that for the particular pulse shape analyzed here , @xmath176 has a local minimum around @xmath177 ( as well as at other higher points ) , which only weakly depends on the value of @xmath159 provided that @xmath178 . at this point , the net probability for a nonadiabatic transition to have occurred during the amplitude ramp is remarkably small . this suppression of nonadiabatic transitions can be understood as destructive interference of the transitions between the floquet states at different times during the amplitude ramp . the exact values of @xmath160 at which the minima occur depend on @xmath12 . for @xmath179 for example , the first minimum in the nonadiabatic transition probability occurs around @xmath180 . the nonadiabatic transitions in the falling edge of the pulse can be calculated in a similar way as for the rising edge . to simplify notations , we shift the times so that the beginning and end times are @xmath181 and @xmath158 ( e.g. by taking @xmath182 in eq . ) . we have = - ( (-1 ) ) . to exactly simulate the nonadiabatic transitions , we start with the initial state @xmath183 and express the final state @xmath184 in the basis of floquet modes @xmath185 . in fig . [ fig : nonadiabatic](d - f ) we show the amplitude of nonadiabatic transitions occurring during the falling edge of the pulse as a function of the maximum pulse amplitude @xmath159 and the fall time @xmath186 . in the adiabatic limit @xmath187 , we have @xmath174 as expected . we also notice a coherent suppression of nonadiabatic transitions at @xmath188 with only a weak dependence on the value of @xmath139 as long as @xmath178 . our results for the rising and falling edges agree in predicting that when @xmath189 , nonadiabatic transitions are negligible , which corresponds to the adiabatic limit in this driven system . the small differences between the two cases in the exact shapes of the oscillatory behavior of @xmath174 as a function of @xmath157 is not very surprising , given that there is an asymmetry between the two cases , both in the pulse shape and in the fact that in one case the pulse amplitude is minimum at the initial time while in the other the pulse amplitude is maximum at the initial time . remarkably , we find that the transitions between the floquet states during the rise and fall edges of the control pulses are reduced for specific values of the rise and fall times , respectively . this effect provides the basis for a quantum control protocol , which we name floquet interference efficient suppression of transitions in the adiabatic basis ( fiesta ) . in the next section we discuss the application of fiesta to optimization of single qubit gates . as we have discussed in secs . [ sec : adiabatic ] and [ sec : nonadiabatic ] , the absence of nonadiabatic transitions between floquet modes results in smooth and simple rabi oscillations between energy eigenstates of the qubit . this is a desirable property for implementing high - fidelity single - qubit gates . nonadiabatic transitions can be suppressed either by turning the pulse on and off adiabatically or by using the destructive interference of the nonadiabatic transitions discussed in sec . [ sec : nonadiabatic ] . in any realistic setup , the gate fidelity is further reduced by the decoherence that the qubit experiences during the implementation of the gate , and the reduction increases with increasing gate time . for an adiabatic pulse in the strong driving regime , we require @xmath190 , which means that the total pulse duration must satisfy @xmath191 . this condition is the same one obtained for the implementation of a typical single - qubit gate in the weak driving regime , because the pulse duration in that case is given by @xmath192 and @xmath193 . therefore , adiabatic pulses in the strong driving regime do not necessarily yield higher operation speeds than weak - driving pulses . however , with the coherent suppression of the nonadiabatic transitions , which requires careful control of the pulse shapes , one could realize high - fidelity single - qubit gates with gate times on the order of @xmath194 . in this section , we discuss the optimization of the fidelity of single - qubit operations by zeroing the nonadiabatic transitions between floquet modes in fast pulses . universal single - qubit operations require unitary operations corresponding to rotations around two axes in the bloch sphere with arbitrary angles . here , we consider implementing operations which are rotations around axes in the equatorial plane of the bloch sphere since these can be realized with the hamiltonian in eq . in the absence of nonadiabatic transitions between floquet modes . the rotation axis is controlled by the phase of the pulse @xmath12 and the rotation angle @xmath195 is determined by the time integral of the quasienergy difference , i.e. @xmath196 . following the discussions in the previous section , we consider the pulse shape defined in eq . . to achieve the above rotations , we choose the optimal rise time @xmath160 and fall time @xmath158 on the rising and falling edges respectively for the coherent suppression of nonadiabatic transitions and we choose the duration @xmath197 of the middle part of the pulse correspondingly according to the target rotation angle @xmath195 . for finding the optimal rise time @xmath160 for a given operation , we calculate and minimize the nonadiabatic transitions on the rising edge in its time interval between 0 and @xmath160 for pulses with a initial phase @xmath12 which sets the desirable rotation axis . this is essentially the same calculation we did in sec . [ sec : nonadiabatic ] . for operations @xmath198 and @xmath199 which are rotations around the @xmath200 and @xmath14 axes of the bloch sphere , the optimal rise times @xmath160 are approximately @xmath201 and @xmath202 respectively . for finding the optimal fall time @xmath158 , the procedure is more complicated as the nonadiabatic transitions on the falling edge depend on @xmath158 as well as the phase of the pulse at the beginning of the falling edge @xmath203 . this phase depends on the initial phase of the pulse as well as the pulse duration by @xmath204 . in fig . [ fig : pulsefidelity](a ) and ( b ) , we plot the nonadiabatic transition amplitude @xmath176 on the falling edge versus the fall time @xmath158 and the phase @xmath203 with the maximum amplitude @xmath205 . we find that the @xmath158 minimizing @xmath176 changes with @xmath203 periodically . next , we perform simulation on the qubit dynamics under pulses with different parameters and compute the their fidelities with respect to the ideal operations @xmath198 and @xmath199 . with the shape of the pulse in eq . , the pulse is fully determined by the parameters @xmath159 , @xmath12 , @xmath157 , @xmath186 , and @xmath156 . with energy relaxation time @xmath206 taken into account , we can express the resulting operation as a quantum process matrix @xmath207 characterized by quantum process tomography ( qpt ) @xcite . to obtain the process matrix corresponding to a given pulse , we simulate the final state after the pulse with respect to initial states @xmath3 , @xmath4 , @xmath208 , and @xmath209 . the simulations are done by numerically solving the master equation in the lindblad form @xcite : ( t ) = -i[h(t ) , ( t)]+_n ( 2_n_n^- _ n^_n-_n^_n ) , [ eq : gates : mastereq ] where @xmath210 and @xmath211 are the lindblad operator and the decoherence rate for a certain decoherence source respectively . here we consider only one decoherence mechanism , energy relaxation thus we only have one term in eq . , with @xmath212 and @xmath213 . the actual values of the qubit parameters are chosen to be the same as those characterized in the experiments discussed in ref . , where @xmath214 ghz and @xmath215 @xmath216s . we note that the value of @xmath206 here is shorter than those obtained in other experiments with flux qubits , and therefore the fidelities we obtain are conservative estimates . once we have the final states with respect to the four initial states , we can reconstruct the process matrix @xmath217 using standard qpt . from the process matrix @xmath217 obtained by qpt and its ideal counterpart @xmath218 , we can directly calculate the process fidelity , defined as @xmath219 $ ] . the gate fidelity @xmath220 , defined as the state fidelity of the process output state with respect to the ideal output state averaged over all possible input states , can be related to the process fidelity @xmath221 by @xmath222 with @xmath223 the dimension of the system @xcite . we simulate the gate fidelity of operations @xmath198 and @xmath199 implemented by pulses with @xmath11 and @xmath224 at @xmath205 . in these simulations , we choose the rise times of the pulses @xmath225 and @xmath202 for @xmath198 and @xmath199 respectively for the maximum suppression of the nonadiabatic transitions on the rising edge . in fig . [ fig : pulsefidelity](c ) and ( d ) , we plot the gate fidelities versus the maximum amplitude duration @xmath197 with different fall times . for short fall times ( @xmath226 ) , the gate fidelity oscillates as a function of @xmath197 . this is a signature of coherent nonadiabatic transitions between the floquet modes . the highest - fidelities are attained when the combinations of @xmath197 and @xmath158 lead to small nonadiabatic transitions ( minima in fig . [ fig : pulsefidelity](a ) and ( b ) ) . for longer fall times ( @xmath227 ) , the ramping of the falling edge is approximately adiabatic thus the gate fidelity has suppressed oscillations and is limited by @xmath206 . versus the maximum amplitude duration @xmath197 ( a ) or the rotation angle @xmath195 given by @xmath196 ( b ) . the pulses are defined with @xmath11 , @xmath225 and @xmath228 and various values of @xmath197 and @xmath139 . while the optimal @xmath197 for the coherent suppression of nonadiabatic transitions is independent of @xmath139 , different pulses with the same @xmath160 , @xmath158 , and @xmath197 can have different rotation angles @xmath195 depending on @xmath139.,width=302 ] so far , we have shown that high - fidelity operations @xmath198 and @xmath199 can be achieved for some specific rotation angle @xmath195 which depends on the optimal combination of @xmath160 , @xmath197 , and @xmath158 . performing univeral quantum computation , however , requires engineering pulses that correspond to arbitrary rotation angles . this goal is easy to achieve in our optimization method . as discussed in sec . [ sec : nonadiabatic ] , the optimal @xmath160 and @xmath158 are almost independent of the maximum driving amplitude @xmath139 as long as @xmath229 . in fig . [ fig : fidelity_vs_a](a ) , we plot the gate fidelity of @xmath198 versus @xmath197 and @xmath139 with the optimal @xmath225 and a short @xmath228 . we show that the optimal @xmath197 , which minimizes nonadiabatic transitions and leads to high - fidelity operations , is also independent of @xmath139 . in fig . [ fig : fidelity_vs_a](b ) , we plot the gate fidelity of @xmath198 versus @xmath195 at several values of @xmath139 . we show that by fixing @xmath160 , @xmath197 , and @xmath158 to their the optimal values and sweeping @xmath139 , high - fidelity rotations for any rotation angles @xmath195 can be achieved . therefore , to fully optimize a pulse for a given rotation operation , one can first determine @xmath160 , @xmath197 , and @xmath158 based on the rotation axis and a rough range of @xmath139 and then fine - tune @xmath139 by targeting a precise rotation angle given by @xmath196 . as an example , we optimize pulses for the operations @xmath230 and @xmath231 in the single - qubit clifford group @xcite using qubit parameters detailed in ref . . for @xmath230 , the optimized pulse with @xmath225 , @xmath228 , @xmath232 , and @xmath233 has a gate fidelity of @xmath234 . for @xmath231 , the optimized pulse has a fidelity of @xmath235 with @xmath236 , @xmath228 , @xmath237 , and @xmath238 . the above fidelities are limited by @xmath206 , where we assumed @xmath239 ghz and @xmath215 @xmath240 . next , we consider optimization with an experimental limitation , as introduced by the time resolution and bandwidth of the ultrafast arbitrary waveform generator used in ref . . we perform pulse optimization while restricting the time resolution of the waveform to 40 ps and limiting the shortest @xmath157 and @xmath186 to 80 ps . we obtain a fidelity of @xmath241 for @xmath230 with the pulse parameters : @xmath242 ps , @xmath243 ps , @xmath244 ps , and @xmath245 ghz . for @xmath231 , we obtain a fidelity of @xmath246 with @xmath247 ps , @xmath243 ps , @xmath248 ps , and @xmath249 ghz . now the gate fidelities are additionally limited by the unsuppressed nonadiabatic transitions between floquet modes . these nonadiabatic transitions arise from waveform definition errors due to insufficient time - resolution . we note that , thanks to the cosine - function shape of the pulse envelope during the rising and falling edges , this pulse scheme requires a bandwidth of only @xmath10 . for superconducting qubits , which usually have a transition frequency of a few gigahertz , such pulses can be readily generated by a high - bandwidth arbitrary waveform generator . in this section , we discuss the connection between fiesta and other methods in optimal quantum control ( oqc ) . for an in depth discussion of quantum control , we refer the reader to the review articles @xcite . the objective in oqc is maximizing the fidelity of a quantum state or of a quantum operation , given a hamiltonian with controllable parameters . constraints on the parameters can be applied , including amplitude constraints and time - domain constraints . time - domain constraints are of two types : frequency bandwidth constraints or parametrization in terms of analytical functions . a research topic that is closely related to oqc is the quantum speed limit , addressing operations implemented with high fidelity in the minimum possible amount of time @xcite . gradient ascent pulse engineering ( grape ) @xcite and krotov @xcite optimal quantum control rely on assuming a piecewise - constant form of the control parameters . in these methods , an iterative algorithm is used to steadily increase the fidelity and eventually converge to optimal values of the control parameters in all the time intervals . with sufficient discretization , continuously varying control parameters can be approximated with good accuracy . other control methods , including chopped random - basis quantum optimization ( crab ) @xcite and gradient optimization of analytic controls ( goat ) @xcite assume the hamiltonian to be expressed in terms of analytic functions , with a set of parameters entering these functions to be optimized . derivative removal by adiabatic gate ( drag ) @xcite is concerned with the reduction of state leakage in a weakly anharmonic system , providing an approximate analytical solution to this problem . a problem discussed extensively in literature is the control of a qubit with a transverse driving term limited in absolute value . it was shown that the optimal solutions for single - qubit rotations consist of sudden switching of the control between its extreme values , and the optimal time is related to the inverse qubit transition frequency ( see ref . and references therein ) . in contrast to these approaches , closed - loop optimal control is a method in which control is optimized in a system with imperfect knowledge of the system hamiltonian or control transfer functions , based on feedback from a fidelity measurement performed on the physical system @xcite . fiesta is used to perform optimization of quantum gates with resonant driving and a simple pulse shape characterized by three times ( pulse rise , top , and fall ) and an amplitude ( the maximum pulse amplitude ) . this particular optimization problem has elements related to amplitude constraints and bandwidth constraints explored with other methods . one important characteristic of this pulse shape is that it is motivated by experimental implementation constraints , where using a frequency band around the qubit transition frequency can be conveniently implemented using analog quadrature modulation . optimization with amplitude constraints can be done using grape or analytically in some cases . bandwidth constraints can be implemented in goat @xcite or approximately , using filtered optimal pulses @xcite . our approach is physically motivated by the observation that even in the strong driving regime the dynamics can be simply described in terms of three unitaries , for the rise , central , and fall parts of the pulses . we showed that optimizing only the four involved parameters ( @xmath157 , @xmath156 , @xmath186 , and @xmath159 ) extremely high fidelities can be attained , with a total pulse duration of the order of @xmath250 , thus approaching the quantum speed limit @xcite . we expect that the optimization of these parameters is also suitable for a closed - loop approach , and that high fidelities can be attained with moderate levels of errors arising from the transfer function frequency dependence and qubit detuning . we note that bartels and mintert @xcite discussed a numerical optimization method for control with short pulses based on floquet theory . the discussion in ref . focuses on the implementation of single - qubit gates , the method is applicable straightforwardly to optimizing single - qubit gates . nevertheless , we expect in general that optimized pulses obtained using their method will have rather complicated envelopes , possibly complicating experimental implementations . we find that in the strong driving regime the optimal times for @xmath200 and @xmath14 rotations obtained with fiesta are different from each other , and these optimal times depend on the maximum amplitude @xmath159 . these results bear an interesting connection with the discussion presented in ref . ( see also references therein ) of optimized pulses with amplitude constraints and no bandwidth constraints . we finally comment on the relevance of our work for physical implementations of optimal quantum control . superconducting flux qubits , for which we have recently observed floquet dynamics @xcite , are an ideal two - level system to apply these results , because higher energy levels are well separated from the lowest two levels , even when driving field amplitudes are larger than @xmath32 . nv centers have also been used in strong driving time domain experiments by fuchs et al . @xcite . in a recent experiment scheuer et al . @xcite demonstrated the implementation of optimal control in the strong driving regime using optimization based on crab . we also expect our theoretical framework to be extendible to optimization of strong driving control for multi - level systems , with superconducting qubits being a case of particular interest . we have analyzed the dynamics of a two - level system under strong resonant pulses from the perspective of ( non)adiabatic evolutions in the floquet picture . we presented derivations of the approximate analytical expressions for the quasienergies and floquet states of the system under consideration . we analyzed the effects of pulse shaping using the adiabatic theory in the floquet picture . we shown that when the the driving amplitude of a pulse varies slowly , the system remains in its initial superposition of the floquet states and the dynamics are governed by a dynamic phase that depends on the evolution of the quasienergies over time . this phase corresponds to a qubit state rotation , generalizing the notion of rabi oscillations to the case of large driving amplitudes . we have also analyzed and quantified the nonadiabatic transitions between the floquet states by employing the adiabatic perturbation theory as well as exact numerical simulations . we found that the first order adiabatic perturbation theory agrees very well with the exact numerics when the driving amplitude is comparable with the qubit frequency and breaks down at very high driving amplitude . in addition , we found that with suitable pulse shaping the nonadiabatic transitions can be coherently suppressed . finally , we presented fiesta , an optimization scheme for the pulse shapes that minimize the nonadiabatic transitions . we shown that high - fidelity single - qubit operations can be achieved in very short times , significantly alleviating the effect of decoherence . furthermore , our optimized control scheme requires pulse specifications that can be achieved with presently available arbitrary - waveform generators and is therefore ready to be implemented in experiments using superconducting qubits . our method is an addition to the toolset of quantum control , providing a simple recipe for high fidelity control of a single qubit in the strong driving regime and a framework that can be generalized to other strongly driven systems . we thank frank wilhelm for useful discussions . we acknowledge support from nserc , canada foundation for innovation , ontario ministry of research and innovation , industry canada . during this work , al was supported by an early research award . 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 @noop _ ( , , ) link:\doibase 10.1103/physrev.49.324 [ * * , ( ) ] link:\doibase 10.1126/science.259.5101.1581 [ * * , ( ) ] link:\doibase 10.1109/9.928587 [ * * , ( ) ] link:\doibase 10.1103/physreva.63.032308 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1103/revmodphys.76.1037 [ * * , ( ) ] link:\doibase 10.1126/science.1181193 [ * * , ( ) ] http://stacks.iop.org/1367-2630/16/i=9/a=093022 [ * * , ( ) ] link:\doibase 10.1126/science.1183628 [ * * , ( ) ] link:\doibase 10.1038/nphys2149 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.121303 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.246601 [ * * , ( ) ] link:\doibase 10.1038/nature02831 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.93.037001 [ * * , ( ) ] link:\doibase 10.1126/science.1119678 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.187002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.107001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.257003 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.017003 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3093823 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.257003 [ * * , ( ) ] http://stacks.iop.org/1367-2630/17/i=4/a=043058 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.133601 [ * * , ( ) ] link:\doibase 10.1140/epjb / e2003 - 00343 - 8 [ * * , ( ) ] link:\doibase 10.1103/physreva.75.063414 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.physrep.2010.03.002 [ * * , ( ) ] link:\doibase 10.1103/physrev.138.b979 [ * * , ( ) ] link:\doibase 10.1007/s100530050236 [ * * , ( ) ] `` , '' in link:\doibase 10.1002/0471428027.ch3 [ _ _ ] ( , ) pp . link:\doibase 10.1103/physreva.79.032301 [ * * , ( ) ] link:\doibase 10.1103/physreva.86.023831 [ * * , ( ) ] link:\doibase 10.1103/physrevb.67.165301 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.466058 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.052508 [ * * , ( ) ] link:\doibase 10.1080/09500349708231894 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.78.390 [ * * , ( ) ] @noop _ _ , vol . ( , ) \doibase http://dx.doi.org/10.1016/s0375-9601(02)01272-0 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.78.405 [ * * , ( ) ] http://stacks.iop.org/1367-2630/12/i=7/a=075008 [ * * , ( ) ] link:\doibase 10.1140/epjd / e2015 - 60464 - 1 [ * * , ( ) ] http://stacks.iop.org/0953-8984/28/i=21/a=213001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.240501 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.052327 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.260501 [ * * , ( ) ] link:\doibase 10.1016/j.jmr.2004.11.004 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0301-0104(93)80108-l [ * * , ( ) ] link:\doibase 10.1103/physreva.84.022326 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physrevlett.103.110501 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physrevlett.112.240503 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.052315 [ * * , ( ) ]
spheroidal stellar populations , such as elliptical galaxies and the halos and bulges of spiral galaxies , contain a significant fraction of all stars in the local universe . the ages and metallicities of halo stars in these galaxies provide some of the most important observational clues to the formation and earliest evolution of galaxies , whether through studies of globular star clusters ( gcs ) or , in more nearby galaxies , the individual halo stars . below , we broadly summarize what is currently known about the ages and metallicities of stars in the spheroids of nearby galaxies . the spheroid of the milky way has two components , a centrally concentrated bulge composed of more metal - rich stars , and an extended , lower density halo consisting of more metal - poor stars . the halo of the galaxy shows substructure , with ( at least ) two chemically distinct components . stars of the ` inner halo ' have a mean [ fe / h]@xmath2 @xcite , while stars in the ` outer halo ' ( r@xmath315 kpc ) are more metal - poor , with a mean [ fe / h]@xmath4 @xcite . the halo also contains many stellar streams and tidal debris that are probably the remnants of previously accreted satellite galaxies ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . the globular clusters in the galaxy have a weak metallicity gradient inside @xmath510 kpc , with no such trend apparent beyond 10 kpc out to the most distant cluster located at @xmath6120 kpc @xcite . the stars and gcs in the halo of the milky way appear to be mostly old , with ages between 11 and 13 gyr @xcite , indicating that the dominant era of accretion in the milky way took place soon after formation @xcite . detailed studies of the stars in m31 suggest that this galaxy has had a more active , recent accretion history when compared with the milky way . m31 has a markedly metal - rich inner spheroid component @xcite that is dominated by substructure @xcite . m31 accreted a relatively massive satellite galaxy only @xmath7 gyr ago @xcite , and has had other many more minor interactions @xcite as well . deep images taken with the _ hst _ , which reach below the main sequence turnoff region for ancient stars in m31 , indicate the presence of metal - rich , intermediate age stars ( @xmath8 gyr ) in fields located at deprojected distances of 11 , 20 , and 35 kpc , with a larger contribution from older , more metal - poor stars increasing at larger distances @xcite . beyond @xmath9 kpc and extending out to at least @xmath10 kpc , the halo of m31 is dominated by metal - poor stars with [ fe / h]@xmath11 @xcite , and this component too has copious substructure @xcite . little is known about the ages of these stars except for the recent work of @xcite who studied the m31 gc mgc1 , which is located at a projected galactocentric distance of @xmath12120 kpc . the cmd of this cluster reveals a metal abundance of [ m / h]@xmath122.3 and a horizontal branch morphology that is consistent with those of milky way gcs at this metal abundance , suggesting a similar age . studies of stars in the spheroids of several more distant elliptical galaxies suggest that metal - rich stars dominate out quite far , probably to at least 10 - 15@xmath13 . ngc 5128 has a largely metal - rich population at radii of 8 - 33 kpc @xcite . even further out , an analysis of the colors of rgb stars , which are sensitive to metallicity , and the colors and luminosities of the agb bump and the rc , which are sensitive to both metallicity and age , suggests that stars in ngc 5128 at @xmath14 kpc have largely intermediate to old ages , with @xmath15 gyr @xcite . the halos of ngc 3377 and ngc 3379 have a wide range of metallicity , with a significant fraction of metal - rich stars , although metal - poor stars begin to dominate the spheroid of ngc 3379 beyond @xmath16 . the lack of stars brighter than the rgb suggests that ngc 3377 and ngc 3379 contain few young , bright agb stars @xcite . like ngc 3379 , a dominant population of metal - poor stars have been found in the spheroid of the edge - on spiral galaxy ngc 891 @xcite . m81 is an earlier - type spiral ( sab ) than either the milky way or m31 . however , studies of the resolved stars in the outskirts of m81 have already shown that it too is a complex mix of stellar populations . @xcite used 9 archival wfpc2 images of m81 to show that the number density of rgb and agb stars decreases with galactocentric radius , eventually flattening out at a ( deprojected distance ) of @xmath17 kpc . they attribute the stars interior to this as part of the thick disk , and stars beyond this as part of the halo . cmds of fields interior to 25 kpc reveal relatively young stars @xcite , and older rgb stars with a relatively large range in metallicity ( @xmath18 < 0 $ ] ; mouhcine 2005 , williams 2009 ) . these stars are likely located in the disk , either thin or thick , of m81 . more recently , @xcite found a flattending of the stellar density profile beyond @xmath19 kpc ( deprojected ) from m81 , with more metal - poor stars ( [ m / h]@xmath20 ) than in the interior fields , again indicating that the halo dominates at these distances . the only constraint on the _ ages _ of stars in the halo of m81 comes from the measurement of absorption line strengths in globular clusters , which suggest ages that are indistinguishable from the gcs in the galaxy , within the significant uncertainties ( @xmath21 gyr ) @xcite . in this work , we study the metallicities and ages of stars in a field in m81 that lies at a projected distance of 19 kpc , using observations taken with the advanced camera for surveys on @xmath0 . these data are among the deepest available for a portion of m81 . the rest of this paper is organized as follows : section 2 presents the data and reduction ; section 3 presents the cmd for our field and discusses general features ; section 4 presents a new determination of the distance to m81 from the tip of the rgb , and averages this with previous distance determinations to give our adopted value ; section 5 discusses the red clump , and section 6 presents an estimate of the metallicities and ages of stars in our m81 field . section 7 presents some new results on a small population of blue stars discovered in our field , and section 8 discusses the results in terms of the formation of m81 , as well as our main conclusions . observations of m81 were obtained with the wide field channel ( wfc ) of the advanced camera for surveys ( acs ) on - board the @xmath0 , for program go-10604 ( pi : sarajedini ) . a single acs / wfc pointing covering @xmath22 was centered at @xmath23 , @xmath24 ( j2000.0 ) , at a projected distance of 18 from the center of m81 , or 19 kpc at the adopted distance of 3.7 mpc ( see section 4 ) . the specific pointing was chosen to lie along the southwestern semi - minor axis , beyond the radius suggested as the edge of m81 s thick disk @xcite , where the surface brightness of m81 is still high enough that we can determine a meaningful metallicity distribution from rgb stars . figure 1a shows the location of our acs pointing . it also shows the wfpc2 field studied by @xcite , which falls on the edge of the suggested thick disk in m81 , at a projected distance of 12.8 ( 14 kpc ) from the center of m81 . figure 1b shows our pointing superposed on hi contours from @xcite ; while our field is outside of m81 s purported thick disk ( at a deprojected distance of 30 - 34 kpc ) , it still lies along the line - of - sight with some of the many hi filaments that exist outside m81 s optical disk @xcite . while some of this hi has been tidally stripped from m81 by encounters with m82 and ngc 3077 , probably @xmath25 myr ago @xcite , velocities of the hi towards our field are consistent with rotation of the m81 disk ( m.yun , private communication ) . our field was observed on 2005 sept . 11 - 13 for 5 orbits ( 2 images per orbit ) in the f606w filter and 9 orbits ( 2 images per orbit ) in the f814w filter , using a standard 4-point dither pattern . each individual exposure had an exposure time of 1247s , leading to a total exposure time of 12470s in the f606w filter , and 22446s in the f814w filter . we retrieved the pipeline flatfielded images from the @xmath0 archive as well as the pipeline - processed multidrizzled combined images from subsets of the data . photometry was performed with the acs module of the dolphot software package @xcite , which is specifically designed for point - source photometry of objects in the individual flt images , which were taken directly from the @xmath0 archive . object detection and photometry was performed on all exposures simultaneously , using one of our deep @xmath26 drizzled images ( derived from a subset of 8 images ) as the reference frame . the dolphot parameters were set to the recommended values given in the dolphot user s guide , which includes psf - fitting using pre - derived psfs , as well as the calculation and application of cte correction @xcite and aperture corrections . the instrumental magnitudes were converted to the vegamag @xmath0 photometric system by adopting the revised zeropoints ( for data taken before july 4 , 2006 ) of 26.420 for f606w and 25.536 for f814w , which differ by @xmath27 mag from the values originally published by @xcite . we refer to magnitudes in this system by their filter name ( @xmath28 and @xmath29 ) , and in cases when we need to convert to the johnson - cousins systems refer to the filters as @xmath30 and @xmath31 . we determined the photometric completeness limits and associated errors in our point - source photometry by adding and remeasuring 60,000 artificial stars in the acs images . the artificial stars have a range of magnitude ( 24 @xmath32 31 ) and color ( @xmath33 ) . the pritchet interpolation function @xcite @xmath34 was used to fit the binned completeness fractions @xmath35 ( @xmath36 number of stars detected / number of stars added ) as a function of magnitude . here , @xmath37 is a parameter that measures the rate of decline of @xmath38 at @xmath39 , and @xmath39 is defined as the 50% completeness limit . to account for completeness variations in both color and magnitude , we used only the bluest stars ( @xmath40 ) to determine the f814w completeness fraction @xmath41 , and redder stars ( @xmath42 ) to derive @xmath43 . the limiting magnitude ( @xmath44 ; * ? ? ? * ) for each filter is @xmath45 and @xmath46 . we selected point sources in our field as follows . first , we restricted our sample to objects with a dolphot category consistent with stars ( obtype=1 ) . of the sources that met this criterion , only those that were measured on at least 8 of the 10 individual f606w images , and on at least 15 of the 18 individual f814w images were retained , yielding a list of 13,858 sources over the entire acs field . to remove remaining , slightly resolved objects , we applied the following cuts : @xmath47 ( from the psf fitting ) , and a crowding parameter crowd @xmath48 mag , where the latter rejected objects whose magnitudes are likely to be affected by neighboring sources . finally , we kept only those objects with sharpness values within an empirically - derived hyperbolic function : @xmath49 . all of these criteria were chosen based on visual inspection of sources on the acs images . the @xmath28@xmath29 versus @xmath29 color - magnitude diagram ( cmd ) for the 9081 remaining objects ( representing our ` cleaned ' sample of stellar objects in the m81 field ) is shown in the left panel of figure 2 , with the cmd of all objects _ rejected _ by the chi / sharp / crowd cuts plotted in the right hand panel . the photometric errors derived from the artificial star experiments , as well as the 50@xmath50 photometric completeness levels , are also plotted in the left - hand panel . comparison of the two plots shows that the vast majority of rejected objects do not correspond to the sequences visible in the ` clean ' cmd , lending credence to our choice of classification criteria . we note that some true stellar sources will be removed by our somewhat stringent selection criteria , but since we are most interested in those objects with the best quality photometry , these are acceptable ` losses ' . the cleaned cmd extends from above the tip of the first ascent red giant branch ( trgb ) to one magnitude below the horizontal branch ( hb ) . stars on the red giant branch are older than a few gyr . the hb has a morphology which is predominantly redward of the rr lyrae instability strip , although we can not unambiguously detect an extended blue horizontal branch due to the presence of a small population of blue stars ( @xmath28@xmath510 ) , which will be discussed in section 7 . in fact , the red hb resembles a red clump ( rc ) most commonly seen in relatively young ( down to @xmath52 gyr ) and/or metal - rich stellar populations . an analysis of the rgb and rc stars is presented in section 5 . there is a small but noticeable population of stars above the trgb , likely asymptotic giant branch ( agb ) stars of intermediate age ( @xmath125 to 10 gyr ) . from the besanon galaxy model of @xcite , we estimate that there should be @xmath53 ( foreground ) stars from the milky way with @xmath54 and @xmath55@xmath56 in our field of view , so the vast majority of stars in our m81 field in this color and magnitude range belong to m81 . we will explore these populations in more detail in the upcoming sections . we determine the luminosity function of the rgb stars in our field by defining an ` rgb zone ' and thereby excluding stars far from the rgb sequence . we use the z=0.0001 and z=0.004 models of @xcite to provide a general guide to the shape of the rgb in both color and luminosity , after first making the models redder and bluer by 0.05 mag to accomodate photometric uncertainties ; see figure 3 . however , we explicitly include stars brighter than the tip of the rgb , likely to be agb stars , by allowing a larger range of magnitudes than the model predictions . in figure 4 , we show the @xmath29 luminosity function of the rgb in our field for those objects in the extraction region with @xmath57 , with a bin width of 0.08 , and uncertainties given by @xmath58 . we do not make a correction for foreground stars , because we found in 3 that these are very sparse in this direction . the red clump is clearly present at @xmath59 ; we will return to this feature in section 5 . in order to interpret our cmd further , we need to know the distance to m81 . here , we determine the distance from the tip of the rgb ( trgb ) in the @xmath31-band , which has proven to be an excellent distance indicator for metal - poor , old stellar populations @xcite . visual inspection of our cmd shows a transition in the number counts at @xmath60 , albeit with a small number of stars . while a small number of stars in the upper part of the rgb ( there are @xmath61 in our cmd ) can lead to a possible systematic bias in the determination of @xmath62 , @xcite have shown that with at least @xmath63 stars in the top magnitude of the rgb ( as is the case for our luminosity function ) , such biases are expected to be small . to derive a quantitative estimate of the location of the rgb tip , we have employed the sobel edge - detection filter @xcite in order to find the strongest discontinuity in the bright part of our lf in figure 4 ( for the relatively small number of stars in the upper rgb , this simple edge - detection method should suffice for our purposes ) . each object in the lf is modelled as a gaussian distribution with a dispersion equal to the expected photometric error . we use the dispersion in the mean of the individual dolphot magnitudes as our photometric error . the resulting luminosity function @xmath64 is plotted at the top of figure 5 . to determine the location of the rgb tip , we applied to this lf the edge - detection algorithm @xmath65 where @xmath66 is the mean photometric error defined from our artificial star experiments @xcite . the resultant response function is plotted in the bottom of figure 5 , and shows a peak at @xmath67 , where the small formal uncertainty reflects only the width of the sobel function peak . converting to the @xmath31-band via the synthetic transformations of @xcite ( and using the revised zeropoints , as noted above ) , we find @xmath68 , where we have adopted an uncertainty of 0.03 mag in the transformation from @xmath29 to @xmath31 . the true uncertainty in @xmath62 is almost certainly larger than that given above , since they do not include any uncertainties in the photometry itself . the distance to m81 is determined by comparing the measured magnitude for the tip of the rgb found above with the predicted intrinsic magnitude , @xmath69 ( for [ fe / h]@xmath70 ) from @xcite . this predicted value is very similar to @xmath71 ( random error ) derived by @xcite , and to @xmath72 ( for [ fe / h]@xmath73 ) from @xcite . a comparison between observed and predicted @xmath74 magnitudes gives an observed distance modulus of @xmath75 for m81 , or an intrinsic value of @xmath76 ( @xmath77 mpc ) , where we have assumed a foreground extinction of @xmath78 ( using @xmath79 ; * ? ? ? * ) and adopted a 0.02 uncertainty in the foreground reddening to m81 . this is consistent with determinations from independent trgb estimates and other methods , as discussed below . several different techniques have been used previously to estimate the distance to m81 , leading to distance modulus estimates that differ by approximately 0.1 mag . the period - luminosity relationship for cepheid variables is a robust way to determine the distance to nearby galaxies . @xcite determined a value of @xmath80 for m81 , based on 30 cepheids discovered in m81 by @xcite , using @xmath0 observations . more recently , @xcite determined a distance modulus of @xmath81 based on 11 cepheids , in good agreement with the distance determined by ferrarese et al . here , we adopt the mccommas value as the cepheid based distance to m81 . estimates of the distance to m81 based on the tip of the rgb method tend to give a range of values . we found @xmath82 above , and @xcite derived a mean value of @xmath83 , based on fields which are interior to that used here . however , @xcite derive a much shorter distance modulus of @xmath84 from their detection of the trgb in an m81 disk field , and @xcite derived a value of @xmath85 from deep photometry of an acs field in m81 s disk . for this work , we adopt a distance of @xmath86@xmath87@xmath88@xmath89 ( @xmath90 mpc ) to m81 , which is the mean value of the five distance determinations described above ( 4 trgb values and the cepheid value ) . the quoted error is the dispersion of the input values ( each of the individual distance determinations have similar uncertainties ) . inspection of figure 2 shows a well - populated horizontal branch , indicating a population of relatively old stars . while our data do not allow for a clean detection of a blue horizontal branch , the magnitude and color of the red clump ( rc ) allows us to estimate the age and metallicity of the dominant stellar population in our field , as shown in the next section . to derive the photometric properties of the rc , we use only those stars with colors in the range @xmath91 ; a close - up of this part of the cmd is shown in figure 6 , along with the luminosity function of stars in this color range , where the rc is clearly visible . to derive the mean magnitude of this feature , we fit the ` background ' lf immediately brighter and fainter than the rc as a linear function , and fit a single gaussian to the lf once the linear background has been subtracted off . we note that the choice of bin size and the exact background fit have negligible effect on the mean magnitude determined for the rc clump . the resulting fit ( and 1@xmath92 uncertainties ) yields @xmath93 . the gaussian width of @xmath94 , while consistent with the expected photometric errors from the artificial star experiments , does allow for some intrinsic spread in age and/or metallicity . given our adopted distance modulus of @xmath86@xmath87@xmath88@xmath95 , the i - band absolute magnitude in this region of m81 is @xmath96 = 0.21 @xmath1 0.10 . the location of the rc in our dataset is also consistent with the value derived by @xcite for their m81 outer disk field while we get @xmath97 . the williams field exhibits a mean age of 2 - 3 gyr and a metallicity of [ m / h]@xmath98 . in our field ( see next section ) we find a mean age of 9 gyr and a [ m / h]@xmath99 . our m81 field is @xmath100 gyr older and @xmath101 dex more metal - poor than the williams field . a change of @xmath102 dex makes the red clump 0.15 mag brighter and a change of + 6 gyr makes the red clump 0.17 mag fainter ( based on fig . 8 of * ? ? ? * ) , so they offset each other resulting in a rc absolute magnitude that is essentially unchanged . ] . we note the presence of a small positive feature at @xmath103 ( adopted error based on the bin size used ) in the lf in figure 6 . this feature is very likely the faint signal from the rgb bump mag @xcite . we do not see strong evidence for an agb bump here because of the relatively low number of stars in our field . @xcite do see an agb bump , because their field contains many more stars , but do not observe an rgb bump , likely because their significantly larger range of metallicities and ages washes out this feature . ] , the location on the rgb where the hydrogen - burning shell crosses the discontinuity in chemical abundances remaining from the stars convective envelope @xcite . the luminosity of the rgb bump depends on both the metallicity and the age of the stars , and will be used as a consistency check on our results in the next section . the absolute magnitude of the red clump , as well as the shape ( slope ) and location ( color ) of the rgb are both sensitive to the age and metallicity of the stellar population . however , the rgb is more sensitive to metallicity than to age , and the magnitude of the rc is more sensitive to age than to metallicity , so it is possible to use both features together to constrain the age and metallicity of the dominant population . we compare the observations with the theoretical isochrones of @xcite , assuming scaled - solar abundance . cmd as this makes use of the well - determined photometric zeropoints for the acs / wfc instrument . however , the theoretical dependence of the red clump magnitude on metallicity and age is performed in the ground - based system as it requires knowledge of the median i - band rc magnitude for a population of stars , and these have only been constructed for the ground - based filters . this is unlikely to introduce a significant systematic effect because the acs / wfc f814w passband closely matches its ground - based @xmath31-band counterpart . we also note that the rgbs of the @xcite isochrones are identical to those of @xcite . ] we first correct the colors and magnitudes using the values for reddening and distance noted above . next , we assume an age and use the colors of rgb stars with @xmath104@xmath1052.0 to determine metallicity based on an interpolation within the @xcite grid of theoretical isochrones with 2.3@xmath106[m / h]@xmath1060.0 . figure 7 shows these theoretical rgbs overplotted on our m81 cmd for an age of 9 gyr . the @xmath104@xmath1052.0 magnitude range is chosen to minimize the effect of asymptotic giant branch stars on the blue side of the rgb @xcite . the resultant mean metallicity and the i - band absolute magnitude of the rc derived above , @xmath96 = 0.21 @xmath1 0.10 , are used in conjunction with the @xcite models for the median magnitude of the red clump stars to infer the age of the population . this age estimate @xmath107 is then used to redetermine the mean metallicity from the rgb stars and the process is repeated . this iterative procedure converges quickly and yields a mean metallicity of [ m / h ] @xmath108 ( random ) , @xmath10.11 ( systematic ) , where the latter value includes the influence of errors in the distance modulus and reddening . the mean age of the dominant population is then @xmath86@xmath107@xmath88 = 9 @xmath1 2 gyr , where the quoted uncertainty reflects the error in the red clump magnitude , which is the dominant source of uncertainty . the resultant metallicity distribution function ( mdf ) is shown in fig . 8 wherein the fitted gaussian profile gives a 1-@xmath92 width of 0.39 dex in [ m / h ] . the dominant signal in the mdf is due to stars around @xmath109 to @xmath110 . the artificial star tests we have performed suggest that the mean color error at @xmath111 is 0.032 mag , which corresponds to @xmath112 } \sim$ ] 0.12 dex based on the @xcite isochrones in the range 1.3@xmath106[m / h]@xmath1060.7 . this suggests a significant fraction of the @xmath112}$ ] exhibited by the mdf is due to an intrinsic range in metallicity among the m81 stars in our field . in particular , subtracting the dispersion due to errors from the measured dispersion in quadrature yields an intrinsic abundance spread of @xmath112}$]= 0.37 dex . it is also important to keep in mind that we have used scaled - solar isochrones for the determination of the mdf shown in fig . the degree to which field stars in the m81 halo are enhanced in the @xmath37-capture elements is currently not known . we can estimate the effect of such an enhancement on our mdf using the formalism of @xcite : @xmath113 = [ fe / h ] + log(0.638f_{\alpha } + 0.362)\ ] ] where @xmath114 represents the factor by which the @xmath37-elements are enhanced . thus , if @xmath115$]=@xmath116 , then @xmath114=2 and our computed scaled - solar [ m / h ] values are larger than [ fe / h ] by @xmath117 dex . we conclude this section by providing a series of consistency checks on our principle result , which is that the mean metallicity and age of stars in our m81 field are [ m / h]@xmath118 and @xmath119 gyr , respectively . first , the location of the weak rgb bump ( described in the previous section ) at @xmath120 can be used to provide an independent estimate of the age / metallicity of the m81 field stars . indeed , the rgb bump lies above the hb / rc for primarily old , more metal - poor populations @xcite . from our adopted distance modulus , an assumed mean color @xmath121 and @xmath122 , we find @xmath123 . from the relations in @xcite ( and adopting an additional 0.1 dex uncertainty in those relations ) , this translates to a metallicity [ fe / h]@xmath124 for an old , 12 gyr population , or a metallicity [ fe / h]@xmath125 for a 9 gyr population . while the uncertainties in these results are large , and the @xmath37 enhancement unknown , the age and metallicity derived from the rgb bump are broadly consistent with what we find from the location of the rc . a further comparison of our derived value of @xmath126 with predicted values presented by @xcite ( their fig . 22 ; based on models by @xcite ) are also consistent with a metal - poor population with an age of @xmath127 gyr . as an additional check , we compare our cmd with that for the galactic globular cluster ngc 362 , studied as part of the @xmath0 treasury project go-10775 @xcite . this cluster has a metallicity of [ fe / h]@xmath128 ( on the @xcite scale ) , and an age of @xmath1210.5 gyr @xcite , and therefore provides a good reference for our field . figure 9 shows our m81 photometry compared with the ngc 362 fiducial sequence derived from the hst treasury project data . the latter has been shifted by @xmath129=13.12 and @xmath130=0.05 . these are based on adopted values of @xmath131 and @xmath132 for ngc 362 @xcite and @xmath133 and @xmath134 for m81 . we see good agreement between the location and shape of the ngc 362 fiducial and the locus of m81 stars along the red clump and the rgb . the color distribution of m81 rgb stars is slightly bluer than the ngc 362 rgb fiducial , and is most evident within one magnitude of the m81 rgb tip . this could be due to ngc 362 being slightly more metal - rich ( @xmath120.05 dex ) and somewhat older ( 1 to 2 gyr ) than the mean properties of the m81 field . taken together , these could account for a color difference of @xmath120.03 mag in @xmath135 on the rgb . our cmd of the m81 halo field shows a ( small ) population of blue stars with @xmath136 , properties expected for young ( @xmath137 yr ) stars , but not for the stars in the halo of a spiral galaxy . we have checked that these are genuine sources with blue colors . they are clearly point - like and not obvious background galaxies , and are scattered throughout our acs field . we conclude that these are likely individual , young stars associated with m81 . furthermore , the rather sudden appearance of these objects at @xmath138 blueward of the dominant rgb suggests that these are not rgb or rc stars in m81 scattered due to large photometric errors . an overlay of metal - rich ( z=0.008 ; [ m / h]@xmath139 ) isochrones from @xcite onto our cmd is shown in fig . 10 , and suggests that these blue stars have ages between @xmath140-@xmath141 myr . what is the origin of these stars ? they have colors and magnitudes similar to a stellar population seen in the cmd of an outer disk field in m81 ( figure 3 in * ? ? ? potentially then , there is a sparse outer disk in m81 , which contributes a very small fraction of the stars observed in our field . if these stars are an extension of the true outer disk of m81 , then the disk plane extends out @xmath634 - 40 kpc from m81 . a second possibility is that these stars formed in the disk of m81 or in one if its companions , and then were carried to this location along with the hi tidal debris that is coincident with our field , likely as a result of the most recent galactic interactions between m81 , m82 and ngc 3077 , which occurred @xmath142 myr ago ( * ? ? ? * see also weisz 2008 ) . the observed hi gas along the line - of - sight to our acs field has a column density of only @xmath143@xmath144 @xcite . recent ( @xmath145 myr ) star formation has been found not only in the outskirts of m81 itself @xcite but also in nearby tidal dwarf galaxies such as holmberg ix @xcite , bk3n @xcite , and in other nearby tidal debris @xcite . however , in all of these cases , the recent star formation has occurred in areas with higher hi column densities ( @xmath146@xmath144 ) than that observed in our field , and is thus consistent with the lack of _ recent _ ( i.e. @xmath147 yr ) star formation in our field . this is also consistent with the findings of @xcite , who find recent low - level star formation at levels @xmath148@xmath144 ) . we note that while recent episodes of star formation in the outermost regions of late types galaxies have been observed at column densities similar to that in our field ( @xmath149@xmath144 ; e.g. * ? ? ? * ; * ? ? ? * ) , we do not see evidence of this in our ( small ) field . we conclude that while we do see evidence for blue stars in our field , these make only a small contribution to the stellar population in our m81 field . the main result from our analysis of a field at a ( projected ) distance of @xmath150 ( 19 kpc ) from the center of m81 , just beyond the suggested edge of the thick disk , is that the dominant population has a mean age of 9@xmath12 gyr and a peak metallicity of [ m / h]@xmath151 0.11 . below , we discuss these results in the context of the formation history of m81 , and compare with results from the spheroids of other nearby galaxies . the color distribution of stars on the rgb in our m81 field is relatively blue and rather narrow . we derived the metallicity distribution of these rgb stars by comparing with theoretical isochrones , and find a mean [ m / h ] of 1.15 , with an intrinsic dispersion of 0.37 dex as shown in figure 8 . our field is dominated by stars that are more metal - poor than stars located closer to the center of m81 ; indeed , our acs field contained the most metal - poor stars in any m81 field observed thus far . in figure 11 , we compare our mdf ( for an assumed age of 9 gyr ) with that determined by @xcite ( for globular cluster - like ages ; @xmath152 gyr for a field that is also along the southwest minor axis , but somewhat closer to the center of m81 with a projected distance of @xmath1214 kpc ( see figure 1 ) . while the general shape for the mdf is similar in both fields , the mouhcine field is clearly dominated by stars which are more metal - rich by @xmath153 dex than the stars studied here ( where we have already accounted for the differences in the assumed ages of the mdfs adopting a slightly older ( by @xmath154 gyr ) age would effectively _ decrease _ the mean metallicity of our mdf by @xmath155 dex ; making the difference between the mdfs larger ) . this metallicity difference is not likely to be due to uncertainties in the colors or in the adopted reddening values ; to match a 0.4 dex difference in [ m / h ] , our mean rgb would need to be redder by @xmath155 in ( @xmath156 . similarly , the mdfs of @xcite are more metal - poor than the outer ( thick ? ) disk field presented by @xcite , a field dominated by relatively old ( 2/3 of stars formed more than @xmath12 8 gyr ago ) but relatively metal - rich stars . this sequence of declining metallicity with location indicates that either there is a strong metallicity gradient in the disk of m81 , or that our field ( as well as the mouhcine field ) samples a different structural component of m81 than @xcite . the markedly lower metallicity between our field and that of the outer disk of m81 , combined with the relatively old ages inferred for these stars ( discussed below ) and the lack of ( much ) recent star formation , provide tentative evidence that our field is largely dominated by stars in the _ halo _ of m81 , rather than in a thick disk . in a recent wide - field study of bright stars in m81 s outer regions @xcite suggest the presence of a faint , structurally distinct component surrounding m81 with a mean metallicity [ m / h]@xmath157 ; the large uncertainty is due to the rather limited part of the rgb observable in their ground - based data . this metallicity is slightly higher than ( but consistent with ) the value found here . dex ] our field is not located within the region surveyed by @xcite , but lies at a deprojected distance of 34 - 40 kpc from m81 , comparable to the distance of the fields used for their metallicity estimate . this extended component has properties similar ( but not identical ) to either the milky way thick disk or the halo , making the exact nature of this structure somewhat uncertain . future deep observations of m81 s outer regions will be needed to help establish the true nature of this feature . we can compare the mdf that we derive for our 20 kpc field in m81 with the metallicities found for the spheroids / halos of other nearby galaxies . at galactocentric radii of @xmath158 kpc , similar to the location of our m81 pointing , the spheroid of m31 is dominated by a largely metal - rich ( [ fe / h]@xmath159 ) population @xcite . the stars in the halo of the milky way are metal - poor , with typical [ fe / h ] values of @xmath160 to @xmath161 for the inner halo ( out to @xmath162 kpc ) @xcite and @xmath163 beyond this ( * ? ? ? * and references within ) . the halo of the e / s0 galaxy ngc 3379 ( at a galactocentric distance of 33 kpc ) shows a broad metallicity distribution with a significant number of metal - poor ( [ m / h]@xmath164 ) stars @xcite . in contrast , the large elliptical galaxy ngc 5128 has a metal - rich halo with [ m / h]@xmath165 to @xmath166 at _ all _ locations studied thus far , from 8 - 38 kpc @xcite . similarly , the elliptical galaxy ngc 3377 also shows a largely metal - rich population at small galactocentric radii @xcite . taken together , large , nearby galaxies exhibit a wide diversity of halo metallicities , showing no clear trends with parent galaxy mass or morphological type ( although there are suggestions that halo metallicity increases with galaxy mass , * ? ? ? * ) . indeed , it appears that variations in the metallicity of halo stars _ even at different locations within a given ` halo ' _ could make any trends with parent galaxy properties difficult to quantify . for example , the metallicity of stars in our single m81 field are similar to that of the ` halo ' of ngc 3379 , the ` outer halo ' of m31 , and the ` inner halo ' of the milky way ( assuming @xmath115\sim + 0.3 - 0.4 $ ] ) . if we have indeed sampled the halo of m81 , an open question is whether an even more metal - poor population exists at larger galactocentric radii ; again , future studies of more distant m81 field would be worthwhile . while our cmd does not reach the main sequence turnoff region , the most reliable age - sensitive feature in the cmd , it still holds clues to the dominant age of the stars in this region of m81 . formal fits to the rc , rgb , and rgb bump give consistent estimates for the mean age of the stars of @xmath167 gyr . the fact that we do not observe a strong population of blue horizontal branch stars is also consistent with these results . we can compare the ages estimated for stars in the spheroids of different galaxies , although it should be noted that these estimates come from different techniques . the dominant population of our m81 field is slightly younger but consistent with the mean ages ( 9.7 , 11.0 , and 10.5 gyr , for fields at distances of 11 , 21 , and 35 kpc from m31 ) of the m31 halo fields studied by ( * ? ? ? * and references within ) . the halo of m31 also harbors globular clusters as old as those in the milky way @xcite ; however , the presence or absence of any radial age trends among this population is still largely an open question . in contrast , the stars in our m81 field appear to be slightly younger ( by @xmath168 gyr , but just within the age uncertainties ) , than the oldest stars in the milky way halo @xcite , as our comparison with the globular cluster ngc 362 illustrated above . while there is an age spread of a few gyr among the milky way gcs with the inner halo clusters being older in the mean than those of the outer halo @xcite , the ages of globular clusters in m81 ( estimated from integrated spectra ) are indistinguishable from galactic globular clusters ( within the sizeable uncertainties * ? ? ? * ) . it should be noted that these clusters are all projected at smaller distances in m81 than the location of our field . we note that our data do _ not _ rule out the presence of a minority population of older stars , as ancient as the majority galactic globular cluster system , in this region of m81 . in the case of ngc 5128 , @xcite investigated the ages of field stars at a distance of 38 kpc using the magnitude of the rc , and found that it is consistent with an old population with a mean age of @xmath169 gyr , very similar to the ages that we observe in the m81 halo . our age and metallicity results for halo stars in m81 are in broad agreement with predictions from some @xmath170cdm models @xcite , where the outer ( @xmath171 kpc ) halos of @xmath172 galaxies are expected to be formed from the accreted ( over a relatively long timescale ) debris of somewhat lower - mass progenitor galaxies than the more metal - rich systems that ( for instance ) resulted in the formation of the more metal - rich giant southern stream in m31 @xcite . in this scenario , the outer halos of luminous galaxies like m81 will have more metal - poor and somewhat younger combined stellar populations than that of the ( slightly ) older , more metal - rich inner halo / spheroid . our results are broadly consistent with these predictions . we have presented deep photometry of a field located 20 kpc from the center of m81 , believed to be dominated by halo stars . the resulting cmd , based on hst / acs imaging , reveals a relatively blue rgb and a hb populated predominantly redward of the instability strip ( i.e. , a red clump ) . from the shape of the rgb , the magnitude of the red clump , and the location of the rgb bump , we derive a mean metallicity of [ m / h ] = 1.15 @xmath1 0.11 and an age of 9 @xmath1 2 gyr for the dominant population in this region of m81 s halo . this is the lowest metallicity found for stars in any portion of m81 , almost certainly because our field is more distant than any other portion of m81 studied to date . our age and metallicity results are broadly consistent with the predictions from @xmath170cdm models , which predict lower metallicities and younger mean ages for outer halos of spiral galaxies than for inner halos . future studies that probe the metallicities and ages of stars even further out in the halo of m81 will allow for better constraints on the early formation of this galaxy . a.s . and p.r.d . were supported through funds provided by nasa through grants ( go-10604.01-a and go-10604.02.a , respectively ) from the space telescope science institute , which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . the authors would like to thank andy dolphin and ben williams for their assistance with the photometric reductions with dolphot , and the anonymous referee for suggesting changes that improved the paper . we would also like to thank tom brown and ben williams for providing comments on earlier versions of the manuscript , as well as michael siegel , min yun , katie chynoweth and mustapha mouhcine for helpful discussions . abadi , m. g. , navarro , j. f. , & steinmetz , m. 2006 , , 365 , 747 alves , d. r. , & sarajedini , a. 1999 , , 511 , 225 alves - brito , a. , forbes , d. a. , mendel , j. t. , hau , g. k. t. , & murphy , m. t. 2009 , , 395 , l34 allende prieto , c. , beers , t. c. , wilhelm , r. , newberg , h. j. , rockosi , c. m. , yanny , b. , & lee , y. s. 2006 , , 636 , 804 barker , m.k . , ferguson , a.m.n . , irwin , m. , arimoto , n. , & jablonka , p. 2009 , , in press bell , e.f . , 2008 , , 680 , 295 bellazzini , m. , ferraro , f.r . , & pancino , e. 2001 , , 556 , 635 brown , t.m . , ferguson , h.c . , smith , e. , kimble , r.a . , sweigart , a.v . , renzini , a. , rich , r.m . & vandenberg , d.a . 2003 , , 592 , l17 brown , t.m . , ferguson , h.c . , smith , e. , guhathakurta , p. , kimble , r.a . , sweigart , a.v . , renzini , a. , rich , r.m . , & vandenberg , d.a . 2005 , , 130 , 1693 brown , t.m . , smith , e. , ferguson , h.c . , rich , r.m . , guhathakurta , p. , renzini , a. , sweigart , a.v . & kimble , r.a . 2006 , , 652 , 323 brown , t. m. , 2007 , , 658 , l95 brown , t. m. , 2008 , , 685 , l121 bullock , j.s . , & johnston , k.v . 2005 , , 635 , 931 carollo , d. , 2007 , , 450 , 1020 carretta , e. , & gratton , r. g. 1997 , , 121 , 95 chapman , s. c. , ibata , r. , lewis , g. f. , ferguson , a. m. n. , irwin , m. , mcconnachie , a. , & tanvir , n. 2006 , , 653 , 255 chiboucas , k. , karachentsev , i. d. , & tully , r. b. 2009 , , 137 , 3009 dalcanton , j. j. , et al . 2009 , , 183 , 67 davidge , t.j . 2008 , , 120 , 1145 de blok , w. j. g. , & walter , f. 2003 , , 341 , l39 decesar , m. e. , durrell , p. r. , ciardullo , r. , hurley - keller , d. , & feldmeier , j.j . 2004 , baas , 35 , 1396 de mello , d.f . , smith , l.j . , sabbi , e. , gallagher , j.s . , mountain , m. , & harbeck , d.r . 2008 , , 135 , 548 dolphin , a.e . 2000 , , 112 , 1383 dotter , a. et al . , submitted durrell , p.r . , harris , w.e . , & pritchet , c.j . 2001 , , 121 , 2557 durrell , p.r . , decesar , m.e . , ciardullo , r. , & feldmeier , j.j . 2004 , iau symposium 217 : recycling intergalactic and interstellar matter , eds . duc , j. braine , e. brinks , 90 fardal , m. a. , babul , a. , geehan , j. j. , & guhathakurta , p. 2006 , , 366 , 1012 fardal , m. a. , guhathakurta , p. , babul , a. , & mcconnachie , a. w. 2007 , , 380 , 15 fardal , m. a. , babul , a. , guhathakurta , p. , gilbert , k. m. , & dodge , c. 2008 , , 682 , l33 ferguson , a.m.n . , irwin , m.j . , ibata , r.a . , lewis , g.f . & tanvir , n.r . 2002 , , 124 , 1452 ferrarese , l. , 2000 , , 529 , 745 ferraro , f. r. , messineo , m. , fusi pecci , f. , de palo , m. a. , straniero , o. , chieffi , a. , & limongi , m. 1999 , , 118 , 1738 ferraro , f. r. , montegriffo , p. , origlia , l. , & fusi pecci , f. 2000 , , 119 , 1282 fleming , d.e.b . , harris , w.e . , pritchet , c.j . , & hanes , d.a . 1995 , , 109 , 1044 font , a. s. , johnston , k. v. , guhathakurta , p. , majewski , s. r. , & rich , r. m. 2006 , , 131 , 1436 font , a. s. , johnston , k. v. , ferguson , a. m. n. , bullock , j. s. , robertson , b. e. , tumlinson , j. , & guhathakurta , p. 2008 , , 673 , 215 freedman , w.l . , 1994 , , 427 , 628 gilbert , k. m. , et al . 2007 , , 668 , 245 girardi , l. , bressan , a. , bertelli , g. , & chiosi , c. 2000 , , 141 , 371 girardi , l. , bertelli , g. , bressan , a. , chiosi , c. , groenewegen , m.a.t . , marigo , p. , salasnich , b. , & weiss , a. 2002 , , 391 , 195 guhathakurta , p. , 2006 , preprint ( astro - ph/0502366 ) hammer , f. , puech , m. , chemin , l. , flores , h. , & lehnert , m. d. 2007 , , 662 , 322 harris , w. e. 2001 , saas - fee advanced course 28 : star clusters , 223 harris , g. l. h. , harris , w. e. , & poole , g. b. 1999 , , 117 , 855 harris , g. l. h. , & harris , w. e. 2000 , , 120 , 2423 harris , w. e. , & harris , g. l. h. 2002 , , 123 , 3108 harris , w.e . 1990 , , 102 , 949 harris , w.e . , harris , g.l.h . , layden , a.c . , & stetson , p.b . 2007 , , 134 , 43 harris , w.e . , harris , g.l.h . , layden , a.c . , & wehner , e.m.h . 2007 , , 666 , 903 2007 , , 134 , 43 ibata , r. , irwin , m. , lewis , g. , ferguson , a.m.n . & tanvir , n. 2001 , , 412 , 49 ibata , r. , chapman , s. , ferguson , a. m. n. , lewis , g. , irwin , m. , & tanvir , n. 2005 , , 634 , 287 ibata , r. , martin , n.f . , irwin , m. , chapman , s. , ferguson , a.m.n . , lewis , g.f . & mcconnachie , a.w . 2007 , , 671 , 1591 irwin , m. j. , ferguson , a. m. n. , ibata , r. a. , lewis , g. f. , & tanvir , n. r. 2005 , , 628 , l105 ivezi , . , et al . 2008 , , 684 , 287 juri , m. , et al . 2008 , , 673 , 864 kalirai , j.s . 2006 , , 648 , 389 karachentsev , i.d . , karachentseva , v.e . , & borngen , f. 1985 , , 217 , 731 karachentsev , i.d . , 2002 , , 383 , 125 karachentseva , v.e . , karachentsev , i.d . , & borngen , f. 1985 , , 60 , 213 klement , r. , 2009 , , 698 , 865 lee , m.g . , freedman , w.l . , & madore , b.f . 1993 , , 417 , 553 ma , j. , fan , z. , de grijs , r. , wu , z.,zhou , x. , wu , j. , jiang , z. , & chen , j. 2009 , , 137 , 4884 mackey , a. d. , 2009 , , in press ( arxiv:0909.1456 ) madore , b.f . , & freedman , w.l . 1995 , , 109 , 1645 makarova , l.n . , 2002 , , 396 , 473 marin - franch , a. , 2009 , , 694 , 1498 marigo , p. , girardi , l. , bressan , a. , groenewegen , m.a.t . , silva , l. , & granato , g. l. 2008 , , 482 , 883 maybhate , a. , masiero , j. , hibbard , j. e. , charlton , j. c. , palma , c. , knierman , k. a. , & english , j. 2007 , , 381 , 59 mccommas , l.p . , yoachim , p. , williams , b.f . , dalcanton , j.j . , davis , m.r . , & dolphin , a.e . 2009 , , 137 , 4707 mcconnachie , a.w . , irwin , m.j . , ferguson , a.m.n . , ibata , r.a . , lewis , g.f . , & tanvir , n. 2004 , , 350 , 243 mcconnachie , a.w . 2009 , , 461 , 66 mouhcine , m. , & ibata , r. 2010 , , in press mouhcine , m. 2006 , , 652 , 277 mouchine , m. , ferguson , h.c . , rich , r.m . , brown , t.m . , & smith , t.e . 2005 , , 633 , 810 mouhcine , m. , rejkuba , m. , & ibata , r. 2007 , , 381 , 873 mould , j. , & kristian , j. 1986 , , 305 , 591 newberg , h. j. , et al . 2002 , , 569 , 245 newberg , h. j. , yanny , b. , & willett , b. a. 2009 , , 700 , l61 pietrinferni , a. , cassisi , s. , salaris , m. , & castelli , f. 2004 , , 612 , 168 rejkuba , m. , greggio , l. , harris , w.e . , harris , g.l.h . , & peng , e.w . 2005 , , 631 , 262 rejkuba , m. , mouhcine , m. , ibata , r. 2009 , , in press richardson , j. c. , et al . 2008 , , 135 , 1998 riess , a. 2003 , acs isr 2003 - 009 ( baltimore , stsci ) rizzi , l. , tully , r.b . , makarov , d. , makarova , l. , dolphin , a.e . , sakai , s. , shaya , e.j . 2007 , , 661 , 815 robin , a. c. , reyl , c. , derrire , s. , & picaud , s. 2003 , , 409 , 523 ryan , s. , & norris , j. 1991 , , 101 , 1865 sabbi , e. , gallagher , j.s . , smith , lj . , de mello , d.f . , & mountain , m. 2008 , , 676 , l113 sakai , s. , madore , b.f . , & freedman , w.l . 1996 , , 461 , 713 sakai , s. , & madore , b.f . 2001 , , 555 , 280 sarajedini , a. 2009 , in the ages of stars , iau symposium , vol . 258 , p. 221 sarajedini , a. , & jablonka , p. 2005 , , 130 , 1627 sarajedini , a. , 2007 , , 133 , 1658 sirianni , m. , 2005 , , 117 , 1049 schlegel , d.j . , finkbeiner , d.p . , & davis , m. 1998 , , 500 , 525 schroder , l. l. , brodie , j. p. , kissler - patig , m. , huchra , j. p. , & phillips , a. c. 2002 , , 123 , 2473 smith , m. c. , et al . 2009 , , 399 , 1223 stetson , p.b . 1987 , , 99 , 191 stetson , p.b . , davis , l.e . , & crabtree , d.r . 1990 , in asp conf . ser . 8 , ccds in astronomy , ed . jacoby ( san francisco : asp ) , 289 stetson , p.b . 1992 , in asp conf . 25 , astronomical data analysis software and systems i , eds . d.m . worral , c. biemesderfer , and j. barnes ( san francisco : asp ) , 297 thilker , d. a. , et al . 2007 , , 173 , 538 tikhonov , n. a. , galazutdinova , o.a . , & drozdovsky , i.o . 2005 , , 431 , 12 weisz , d. r. , skillman , e. d. , cannon , j. m. , dolphin , a. e. , kennicutt , r. c. , jr . , lee , j. , & walter , f. 2008 , , 689 , 160 williams , b.f . , 2009 , , 137 , 419 yanny , b. , et al . 2003 , , 588 , 824 yun , m.s . , ho , p.t.p . , & lo , k.y . 1994 , , 372 , 530 yun , m.s . 1999 , galaxy interactions at low and high redshift , iau symp . d.sanders , 81 zinn , r. j. 1985 , , 293 , 424
since their discovery , josephson effects play a central role in the investigation of the superconducting state because of their fundamental as well as technical relevance . josephson effects occur whenever two extended superconducting banks are weakly coupled and can be observed , among many other systems , across a geometric constriction of a superconductor with lateral dimensions much smaller than the coherence length . this kind of josephson junction , usually referred to as scs junction , has promising properties for highly sensitive magnetometers @xcite . recent progress in microfabrication technique allows to pattern high quality niobium thin films with structures small enough for constriction - type josephson junctions @xcite . at the same time , a scs josephson junction is a useful basis for theoretical considerations because only the properties of the superconductor and its geometry are relevant . superconducting point contacts serve as a simplified model for scs junctions and have been investigated by many authors . an overview of many useful results for the ac and dc josephson effect can be found in the reviews by likharev @xcite and golubov et al . the most general result was derived by zaitsev @xcite and includes a large number of conduction channels as well as asymmetry and non - bcs structure of the electrodes . + microscopic theory for the dc josephson effect in a model mesoscopic constriction has been presented in @xcite with the restriction to a small number of conducting channels in the constriction . this model clarified the relevance of andreev bound states for the supercurrents in the junction . analytic solutions of microscopic theory for the dc josephson effect in a ballistic superconducting microconstriction have been presented in @xcite . in this approach , the influence of the constriction length on the critical current has been calculated self - consistently for temperatures close to @xmath2 . as already stated by kulik @xcite and ishii @xcite , a more detailed physical picture of cooper pair tunneling in josephson junctions involves andreev reflection @xcite at the interfaces between the weak link and the electrodes . due to this picture , andreev bound states in josephson junctions represent electron and hole waves that bounce back and forth between the two electrodes constructively , transferring one cooper pair in every cycle . thus , cooper pair transfer across josephson junctions is mediated by andreev bound states . with this view in mind , detailed knowledge of the local density of states ( ldos ) in the vicinity of the weak link is essential . this fact has led to the consideration of the ldos in different types of josephson junctions : among others , see @xcite for scs , @xcite for sns , @xcite for sns and sis and @xcite for sns and sfs with s- and d - wave pairing symmetry ( n = normal conductor , i = isolator , f = ferromagnet ) . in the present paper , we examine a two - dimensional model of a scs josephson junction in order to investigate the dependence of the dc josephson effect on temperature , on geometric parameters as well as the influence of magnetic fields , external ones as well as those generated by the supercurrents flowing in the junction . therefore , we will numerically solve microscopic eilenberger theory self - consistently for a conventional ( s - wave ) superconducting junction in the clean limit . in order to achieve electromagnetic self - consistency as well , we employ a convenient method to calculate the magnetic field evoked by the supercurrents . this will allow us to derive transport properties for a realistic geometry at arbitrary temperature and arbitrary external magnetic field . furthermore , the method used to solve microscopic eilenberger theory self - consistently including magnetic fields presented in this paper can be applied to more complex geometries and thus might be of more general interest . the paper is organized as follows . in the next section , we introduce our geometric model . in section [ sec : theory ] , we present the theoretical background for our calculations and explain how we solve the relevant equations . in section [ sec : comp ] , we compare our results to previous ones in order to test our methods . in section [ sec : jeffectinscs ] , we present representative results and explain how the josephson effect emerges in scs junctions . in the following , we consider the influence of the temperature and geometric parameters ( section [ sec : geometry ] ) , of @xmath0 ( [ sec : kappa ] ) and of external magnetic fields ( [ sec : bext ] ) . finally , in section [ sec : concl ] , we close with some concluding remarks . for the calculations presented in this paper , we use a two - dimensional model of a scs josephson junction as depicted in fig.[cap : geometry ] . the geometry consists of a long superconducting channel which is narrowed down by a constriction . we will calculate all relevant quantities on a section of the channel enclosing the constriction with length @xmath3 ( @xmath4 is the coherence length ) . on the left and right end of this section , the phase of the order parameter will be set to @xmath5 . the width of the channel will be set to @xmath6 throughout the paper whilst the constriction is defined by the parameters @xmath7 and @xmath8 as can be seen in the figure . the external magnetic field @xmath9 is oriented perpendicular to the two - dimensional structure . since we assume the fermi surface to be cylindrical with the symmetry axis parallel to the @xmath10 direction , we can restrict our calculations to the @xmath11 plane . the results are applicable to films of arbitrary thickness with an appropriate value for the london penetration depth @xcite . our results correspond to the symmetry plane at half the extension of the sample in the @xmath10 direction and , accordingly , magnetic fields are oriented parallel to @xmath12 . two - dimensional model for a scs josephson junction consisting of a long channel ( width @xmath6 throughout the paper ) with a constriction . the constriction is defined by the parameters @xmath7 and @xmath8 . our calculations have been carried out for a section of the channel with length @xmath3 ( dark grey area ) and fixed phase difference of the order parameter @xmath13 . the magnetic field is oriented perpendicular to the image plane ( along the @xmath10 axis ) . ] our calculations are based on quasiclassical eilenberger theory @xcite which we solve self - consistently including the magnetic vector potential . to obtain josephson weak link behaviour , it is essential to calculate the order parameter self - consistently . a self - consistent solution contains local modulations of the order parameter amplitude and can thus yield local suppression of superconductivity . local suppression of superconductivity in turn constitutes a josephson weak link . + to solve quasiclassical eilenberger theory self - consistently , we have to find solutions for the order parameter @xmath14 and the supercurrent @xmath15 which satisfy the two self - consistency conditions : @xmath16 these two equations , the gap equation and the current equation , contain the components of the quasiclassical propagator @xmath17 and @xmath18 which , in turn , can be expressed in terms of two complex quantities @xmath19 and @xmath20 called the riccati amplitudes @xcite : @xmath21 the riccati amplitudes @xmath19 and @xmath20 are solutions to the riccati differential equations : [ eq : riccatipde ] @xmath22\ , a(x)-\delta(\vec{r}(x ) ) & = & 0\nonumber \\ \label{eq : riccatipdea } \\ \hbar v_{f}\partial_{x}\ , b(x)-[2\tilde{\varepsilon}_{n}+\delta(\vec{r}(x))\ , b(x)]\ , b(x)+\delta^{\dagger}(\vec{r}(x ) ) & = & 0\nonumber \\ \label{eq : riccatipdeb}\end{aligned}\ ] ] the riccati differential equations have to be solved along real space trajectories @xmath23 pointing in the direction of the fermi wave vector @xmath24 using the modified matsubara frequencies @xmath25 : @xmath26 the function @xmath27 in the riccati equations is the complex order parameter . the gradient of the order parameter phase @xmath28 together with the magnetic vector potential @xmath29 consitute the gauge - invariant superfluid velocity @xmath30 : @xmath31 in the following subsection , we will describe the procedure how we calculate the magnetic vector potential @xmath29 . the distribution of the order parameter phase @xmath32 follows from the solution of the gap equation which , together with the current equation ( [ eq : curreq ] ) , minimizes free energy . thus , the way we calculate the magnetic vector potential makes use of the freedom of gauge , but after choosing a certain way to calculate @xmath29 , the order parameter phase is uniquely defined by the self - consistent solution of the gap and the current equation . to obtain stable solutions , the riccati equation for @xmath19 has to be integrated along @xmath33 whereas the riccati equation for @xmath20 in the opposite direction . appropriate initial values for the integration of the riccati differential equations can be constructed by dropping the gradient terms . for @xmath34 , this yields [ eq : abasymp]@xmath35 for the evaluation of the right hand side of the gap and the current equation , the components of the quasiclassical propagator @xmath36 and @xmath18 constructed with the riccati amplitudes according to equations ( [ eq : riccatiparamf ] ) and ( [ eq : riccatiparamg ] ) have to be averaged over the fermi surface of the superconducting material ( denoted with @xmath37 ) . as stated above , we assume the fermi surface to be cylindrical with the symmetry axis along the @xmath10 direction . finally , these averaged quantities have to be summed over all matsubara frequencies @xmath38 smaller than the cutoff @xmath39 . for our calculations , we use a cutoff of @xmath40 which is sufficiently large for the results to be independent of @xmath39 . apart from the left and right end of the channel section , all other boundaries of the geometry are assumed to be impenetrable . thus , we can apply the general boundary conditions for the riccati amplitudes given by shelankov and ozana @xcite in a very simple form . vanishing transmissibility leads to specular reflection of the trajectories ( outgoing angle is equal to incident angle ) and to the conservation of the riccati amplitudes . + the optical behaviour of the trajectories makes it necessary to implement a two - dimensional raytracing procedure . depending on position and angle , this leads to multiple reflections of the trajectories within the geometry . a similar raytracing procedure for the riccati equations has been used in @xcite to calculate the local density of states at polygonal boundaries of d - wave superconductors . in our case , the trajectories can be terminated either if the left or right channel end have been reached or if the length of the trajectory is sufficient for the results of the integration to be independent of the initial values . to solve the above equations , we discretize the area of the channel section and solve the equations at the grid nodes . numerical costs can be reduced significantly by a local variation of the grid width . in the direct vicinity of the constriction , a rather fine grid is needed whereas less nodes are sufficient in the channel . to account for magnetic fields , the two self - consistency equations ( [ eq : gapeq ] ) and ( [ eq : curreq ] ) have to be completed by the maxwell equation @xmath41 and the definition of the magnetic vector potential @xmath42 the total magnetic vector potential has two contributions . first , there is the contribution @xmath43 which represents the externally applied magnetic field . second , the supercurrents flowing in the junction cause a magnetic vector potential @xmath44 . combining both yields the total magnetic vector potential : @xmath45 choosing coulomb gauge @xmath46 , it follows from ( [ eq : maxwell ] ) and ( [ eq : magvecpotdef ] ) for the contribution evoked by the supercurrents that @xmath47 in order to solve this poisson equation for the magnetic vector potential @xmath44 , we use the two - dimensional green s function of the laplacian . this leads to @xmath48 free space boundary conditions for @xmath49 are contained in the green s function whereas the behaviour in the vicinity of the sample is determined by the self - consistency conditions ( [ eq : gapeq ] ) and ( [ eq : curreq ] ) , see subsection [ sub : iterations ] . in a two - dimensional discretization of the geometry , this integral can easily be solved analytically for each grid cell and then summed over all cells . in order to have consistent behaviour of the magnetic vector potential at the connection to the long channel , we assume the currents to be constant away from the constriction ( for @xmath50 ) and also integrate over these in equation ( [ eq : afromcurrs ] ) . + after calculating the magnetic vector potential @xmath51 from the supercurrents via eq.([eq : afromcurrs ] ) , we make use of the freedom of gauge . the phase gradients of the order parameter @xmath28 together with the total magnetic vector potential @xmath29 lead to the final form of the supercurrents ( see eq.([eq : dopplervs ] ) ) . this implies that the resulting phase gradients depend on the form of the magnetic vector potential and vice versa . to minimize both the phase gradients and the magnetic vector potential , we shift the vector potential @xmath51 to minimal absolute values . this can be done via a simple homogeneous offset which neither changes the corresponding magnetic field @xmath52 nor conflicts with the poisson equation([eq : apoisson ] ) . minimizing the absolute values of the phase gradient and the magnetic vector potential leads to more robust convergence of the iterative procedure which we use to solve the gap and the current equation . this iterative procedure will be described in detail in subsection [ sub : iterations ] . to complete the total magnetic vector potential , we include external magnetic fields @xmath9 via @xmath53 which complies with the coulomb gauge chosen for the magnetic vector potential @xmath54 and which minimizes absolute values . in the problem considered in the present work , two factors leading to pair breaking are relevant . first , a phase difference is applied which leads to a transport current and thus suppresses superconductivity . second , the external magnetic field leads to screening currents and pair breaking . in order to apply correct boundary conditions at the connection to the channel , these two effects have to be clearly separated . since we study josephson junction behaviour , the maximum transport current will be given by the josephson critical current which is much lower than the critical current of a massive superconducting lead . thus , the transport current will not significantly lower the gap amplitude in the channel , far away from the constriction . for vanishing external magnetic field , it is therefore a very good approximation to use the temperature dependent bulk gap amplitude @xmath55 as boundary condition at the left and right end of the section of the channel enclosing the constriction . furthermore , for vanishing external magnetic field , the phase is homogeneous over the cross section of the channel and a phase difference can easily be applied by setting the phase boundary conditions to @xmath56 . the situation changes in the presence of an external magnetic field . the magnetic field leads to screening currents and lowers the gap amplitude in the channel . to gain correct behaviour for higher magnetic fields , it is thus essential to include the lowering of the gap amplitude in the channel . with an external magnetic field , we thus proceed in two steps . first , we self - consistently calculate the amplitude and phase of the order parameter for zero transport current ( corresponding to zero phase difference ) without any boundary conditions at the left and right section ends . this will give us the correct self - consistent solution with screening but without transport currents and allows for a lowering of the gap amplitude as well as for variations of the phase over the cross section of the channel . in the second step , we use the amplitude and phase of the order parameter found at the left and right section end in the first step as boundary conditions for finite transport current . therefore , the phase has to be shifted by @xmath57 at the left section end and by @xmath58 at the right section end whereas the amplitude has to be retained . herewith , the pair - breaking influence of the external magnetic field and the screening currents is included for all amplitudes of the transport current . this two - step procedure has to be repeated for every value of the magnetic field . the gradients of the order parameter phase @xmath28 together with the magnetic vector potential @xmath29 constitute the gauge - invariant superfluid velocity @xmath30 given in eq.([eq : dopplervs ] ) . the current - phase relation thus has to be expressed in terms of the gauge - invariant phase difference @xmath59 with @xmath60 where @xmath61 is the flux quantum and @xmath62 indicate the left and right section end . let us now comment on how we gain the self - consistent solutions for the order parameter @xmath14 , the supercurrents @xmath15 and the magnetic vector potential @xmath29 . the self - consistency equations ( [ eq : gapeq ] ) and ( [ eq : curreq ] ) serve as iteration rules with which we improve an initial guess on a discrete grid . in each iterative step , we first have to solve the riccati equations ( [ eq : riccatipde ] ) for all matsubara frequencies @xmath63 and all fermi velocities @xmath64 at every grid node . second , we construct @xmath17 and @xmath18 from the solutions for @xmath19 and @xmath20 via ( [ eq : riccatiparamf ] ) and ( [ eq : riccatiparamg ] ) and plug these into the gap ( [ eq : gapeq ] ) and current equation ( [ eq : curreq ] ) . third , we calculate the magnetic vector potential @xmath51 caused by the supercurrents via ( [ eq : afromcurrs ] ) and construct the total magnetic vector potential @xmath54 according to ( [ eq : atot ] ) . with the new configurations for @xmath14 , @xmath15 and @xmath29 , we can attempt the next iterative cycle . this iterative procedure has to be repeated until convergence . two criteria can be utilized to determine the level of self - consistency of the solution . it is straightforward to observe the change of the functions @xmath14 , @xmath15 and @xmath29 in every iterative step and to continue until it is sufficiently small . since the rate of convergence in a single iteration can be very small , a more significant approach is to verify current conservation which provides a stringent test of self - consistency @xcite . in the formulation of the microscopic eilenberger equations used in the present work , the order parameter in the gap equation retains a complex quantity . thus , iterating the gap equation yields both the amplitude and the phase distribution of the order parameter self - consistently . the magnetic vector potential is calculated from the currents and the external magnetic field in every iteration and the system of equations is thus solved entirely self - consistent including the magnetic vector potential . at first sight , the free space green s function ( [ eq : afromcurrs ] ) for the magnetic vector potential does not provide boundary conditions for the surface of the geometry . since the green s function ( [ eq : afromcurrs ] ) is embedded in a self - consistent calculation , the compliance with the boundary conditions is ensured by the gap ( [ eq : gapeq ] ) and the current equation ( [ eq : curreq ] ) . a self - consistent solution of the eilenberger equations automatically leads to @xmath65 and respect of the boundary conditions at the surface of the superconductor ( @xmath66 with the surface normal @xmath67 ) . since , in the self - consistent calculation , the magnetic vector potential depends on the currents and vice versa , the correct behaviour of the magnetic vector potential including @xmath68 follows . embedding the green s function ( [ eq : afromcurrs ] ) in the self - consistent calculation thus liberates us from the explicit formulation of boundary conditions for the magnetic vector potential . it should be noted that an iterative algorithm usually produces stable solutions . as discussed in @xcite , unstable solutions corresponding to saddle points of the free energy can only be reached with more elaborate methods . however , since we are only interested in stable solutions , an iterative approach is controllable and straightforward . once a self - consistent configuration of the order parameter , the supercurrents and the magnetic vector potential has been found , the normalized local density of states follows via @xmath69 \big>_{fs}\end{aligned}\ ] ] with the definitions of @xmath70 and @xmath71 introduced above @xcite . all quantities with the dimension of an energy will be normalized to @xmath72 . the characteristic lengthscale we use to normalize all lengths is @xmath73 . total currents ( integrated over the cross - section of the channel ) will be given in terms of @xmath74 with @xmath75 . + with @xmath0 and the london penetration depth @xmath76 , we use @xmath77 and @xmath78 . + applying these normalizations , equation ( [ eq : apoisson ] ) becomes @xmath79 thus , it is obvious that for reasonably large values of @xmath1 , the magnetic vector potential evoked by the currents can be neglected and electromagnetic self - consistency becomes unimportant . nevertheless , for smaller values of @xmath1 , magnetic fields become increasingly important and have to be included in the calculations . as a first test of our approach and our numerical code , we compare our results with those of ginzburg - landau theory presented in chapter 4.4 of tinkham s textbook @xcite . there , one can find results for the critical current of a thin wire or film . in order to compare our results with these calculations , we remove the constriction ( @xmath80 ) and consider temperatures close to @xmath2 . additionally , we have to change our boundary conditions slightly in order to allow for suppression of superconductivity in the channel at high transport current densities . this can easily be done via periodic boundary conditions at the left and right end of the channel section . after these minor changes , our program is ready to consider a long thin wire or film and our results perfectly match those from ginzburg - landau theory . the same results for the critical current can also be achieved by a simple doppler shift calculation @xcite , i.e. solving the bulk gap and current equation with a homogeneous phase gradient @xmath81 this consideration provides a second verification . as a third test , we compare our results with those for a ballistic superconducting microconstriction obtained by zareyan et al . @xcite . in their work , zareyan et al . analytically solve quasiclassical eilenberger theory for a ballistic narrow superconducting channel connecting two bulk electrodes for temperatures close to @xmath2 . if we remove the constriction from our geometry ( @xmath80 ) and use our standard boundary conditions for the left and right section end described in section [ sec : theory ] , we can numerically reproduce the results given by zareyan et al . in this section , we show results without magnetic fields in order to explain in detail how the josephson effect in a scs junction emerges . as mentioned above , neglecting the magnetic fields caused by the supercurrents corresponds to @xmath82 . in order to explain the josephson effect in a scs junction in detail , we fix the geometric parameters ( @xmath83 ) as well as the temperature ( @xmath84 ) . in fig.[cap : jjindetail ] , we show the amplitude ( a ) and the phase ( b ) of the order parameter for different values of the phase difference as well as the current - phase relation ( c ) . additionally , we show the ldos in the center of the constriction in ( d ) . ( a ) josephson effect in a scs junction with @xmath83 and @xmath82 at a temperature of @xmath84 . in ( a ) and ( b ) , the order parameter amplitude and phase are plotted along the @xmath85 axis for several values of the phase difference across the junction . the constriction length is indicated with the vertical dashed lines . in ( c ) , we show the corresponding current - phase relation with the josephson critical current indicated by the dashed lines . the ldos in the center of the constriction ( @xmath86 ) is plotted in ( d ) . ] ( b ) josephson effect in a scs junction with @xmath83 and @xmath82 at a temperature of @xmath84 . in ( a ) and ( b ) , the order parameter amplitude and phase are plotted along the @xmath85 axis for several values of the phase difference across the junction . the constriction length is indicated with the vertical dashed lines . in ( c ) , we show the corresponding current - phase relation with the josephson critical current indicated by the dashed lines . the ldos in the center of the constriction ( @xmath86 ) is plotted in ( d ) . ] ( c ) josephson effect in a scs junction with @xmath83 and @xmath82 at a temperature of @xmath84 . in ( a ) and ( b ) , the order parameter amplitude and phase are plotted along the @xmath85 axis for several values of the phase difference across the junction . the constriction length is indicated with the vertical dashed lines . in ( c ) , we show the corresponding current - phase relation with the josephson critical current indicated by the dashed lines . the ldos in the center of the constriction ( @xmath86 ) is plotted in ( d ) . ] ( d ) josephson effect in a scs junction with @xmath83 and @xmath82 at a temperature of @xmath84 . in ( a ) and ( b ) , the order parameter amplitude and phase are plotted along the @xmath85 axis for several values of the phase difference across the junction . the constriction length is indicated with the vertical dashed lines . in ( c ) , we show the corresponding current - phase relation with the josephson critical current indicated by the dashed lines . the ldos in the center of the constriction ( @xmath86 ) is plotted in ( d ) . ] with increasing phase difference , increasing supercurrents flow across the junction . the current concentration in the constriction leads to high local current densities and causes a suppression of the order parameter amplitude . this suppression of the order parameter amplitude in turn limits the current that can flow across the junction . at the same time , the suppression of the order parameter amplitude leads to weak coupling between the left and right electrode . the suppression of the order parameter amplitude caused by the current concentration in the constriction thus evokes the josephson effect in scs junctions . as long as the phase difference and the current density are small , the suppression of the order parameter amplitude is negligible and the self - consistent phase distribution resembles the solution of the laplace equation from london theory , i.e. from local electrodynamics . the stronger the currents , the stronger is the suppression of the order parameter amplitude in the constriction and the stronger is the deviation of the phase distribution from the solution of the laplace equation . at phase differences close to @xmath87 , the whole phase drop occurs in the constriction and only very small currents flow . + for the parameters chosen here , we find that the current - phase relation strongly deviates from the standard sinusoidal form . as long as the suppression of the order parameter in the constriction is small , the current - phase relation is basically linear . however , as the suppression of the order parameter becomes stronger , the current reaches the josephson critical current and vanishes at @xmath87 . in fig.[cap : currfieldnoemsc ] , we show the current flow pattern in the vicinity of the constriction at the critical current ( @xmath88 ) . current flow pattern in the vicinity of the constriction for @xmath83 , @xmath82 and @xmath84 at the critical current ( @xmath88 ) . for better visibility , we plot only every fourth point out of the grid used in our calculations . ] in fig.[cap : jjindetail ] ( d ) , we show the ldos in the center of the constriction . with increasing phase difference , pronounced andreev bound states occur at energies below the gap ( @xmath89 ) . at intermediate values of the phase difference , additional structures in the ldos appear close to the main peak of the bound state which stem from multiple reflections in the constriction . these substructures will be discussed in detail in the next section . at the maximum phase difference of @xmath87 , the andreev bound states occur at an energy of @xmath90 and exhibit the highest peak value . by considering the amplitude and the phase of the order parameter , the currents and the local density of states , we gain a very complete picture of the josephson effect in scs junctions . we can state that the supercurrent across the constriction - type weak link is indeed mediated by the andreev bound states . cooper pairs incident on one side of the junction are transformed into quasiparticles that travel across the junction phase - coherently and recompose to cooper pairs on the other side . in order to explain the relevance of self - consistency , we compare the self - consistent results with those of a s - n - s model for the order parameter : @xmath91 with this model , we calculate the supercurrents for phase differences @xmath92 and extract the current - phase relation and calculate the ldos in the center of the constriction ( see fig.[cap : sns ] ) . the parameters used here are the same as for the self - consistent solutions in fig.[cap : jjindetail ] . + when using the s - n - s model , one implicitly assumes that the junction exhibits weak link behaviour . this justifies the suppression of the order parameter and allows for a rapid variation of the phase . the result is a more sinusoidal current - phase relation because the full phase drop occurs within the constriction ( see fig.[cap : sns ] ( a ) ) . in the self - consistent solution however , part of the phase variation occurs in the channel and the effective phase difference in the direct vicinity of the constriction is smaller , at least for small total phase differences ( cf . fig.[cap : jjindetail ] ( b ) ) . + the position of the bound states for the s - n - s model reflects the fact that the current - phase relation is more sinusoidal ( see fig.[cap : sns ] ( b ) ) . additionally , it is remarkable that the bound states in the s - n - s model do not exhibit additional substructure . ( a ) current - phase relation ( a ) and ldos in the center of the constriction ( b ) for the non - self - consistent s - n - s model given in eq.([eq : sns ] ) . the self - consistent current - phase relation from fig.[cap : jjindetail ] ( c ) is also plotted in ( a ) for comparison . for this figure , we use the same parameters as for fig.[cap : jjindetail ] . ] ( b ) current - phase relation ( a ) and ldos in the center of the constriction ( b ) for the non - self - consistent s - n - s model given in eq.([eq : sns ] ) . the self - consistent current - phase relation from fig.[cap : jjindetail ] ( c ) is also plotted in ( a ) for comparison . for this figure , we use the same parameters as for fig.[cap : jjindetail ] . ] in the first part of this section , we analyze the temperature dependence of the josephson effect in scs junctions . in the second part , we will focus on the influence of the geometric parameters that define the constriction . in order to investigate the temperature dependence of the josephson effect in scs junctions , we keep the geometric parameters fixed ( @xmath83 ) and neither consider external magnetic fields nor those caused by the supercurrents ( @xmath93 , @xmath82 ) . + to study the influence of the temperature , we calculate the order parameter @xmath14 , the currents @xmath94 and the local density of states @xmath95 for different values of the phase difference @xmath59 at temperatures from @xmath96 to @xmath97 . from the resulting current configurations , we extract the current - phase relations which we show in fig.[cap : cphirgapt ] ( a ) . in fig.[cap : cphirgapt ] ( b ) , we plot the amplitude of the order parameter in the center of the constriction for the same values of @xmath98 and @xmath59 as for the current - phase relations . for the two highest temperatures , we increase the cutoff @xmath39 used in the gap equation and the current equation in order to include a sufficient number of matsubara frequencies ( @xmath99 for @xmath100 and @xmath101 for @xmath102 respectively ) . ( a ) current - phase relations for @xmath83 and @xmath82 at different temperatures are shown in ( a ) and the corresponding amplitude of the order parameter in the center of the constriction in ( b ) . points with solid guidelines are the calculated stable branches whereas the dashed lines are a rough sketch of the unstable ones . inset in ( a ) : temperature - dependence of the critical current ( points with solid guideline ) and the kulik - omelyanchuk ( dashed ) and the ambegaokar - baratoff results ( dash - dotted ) for comparison . ] ( b ) current - phase relations for @xmath83 and @xmath82 at different temperatures are shown in ( a ) and the corresponding amplitude of the order parameter in the center of the constriction in ( b ) . points with solid guidelines are the calculated stable branches whereas the dashed lines are a rough sketch of the unstable ones . inset in ( a ) : temperature - dependence of the critical current ( points with solid guideline ) and the kulik - omelyanchuk ( dashed ) and the ambegaokar - baratoff results ( dash - dotted ) for comparison . ] at temperatures close to @xmath2 , the current - phase relation is highly sinusoidal . the lower the temperature , the stronger are the deviations of the current - phase relation from @xmath103 . for the geometry used here , the current - phase relation becomes multivalued at temperatures @xmath104 . as stated above , our iterative approach yields only the stable branch of the solution . the unstable branch of the current - phase relation is roughly sketched in the diagram ( dashed lines , see e.g. @xcite ) . the multivalued character of the current - phase relation implies that the current increases even for phase differences @xmath105 and marks the transition from josephson junction behaviour to bulk current flow . + in the inset of fig.[cap : cphirgapt ] ( a ) , we plot the dependence of the critical current on temperature ( points with solid guideline ) . for comparison , we plot the kulik - omelyanchuk result from @xcite for a point contact in the clean limit ( dashed line ) as well as the ambegaokar - baratoff result for a tunnel junction ( dash - dotted ) from @xcite . close to @xmath2 , the critical current exhibits a linear decrease with increasing temperature and the self - consistent results of our calculations coincide with those for the point contact . at lower temperatures however , the self - consistent treatment leads to a stronger increase of the critical current than predicted for the point contact . in the kulik - omelyanchuk results for the point contact , the current vanishes at @xmath87 for all temperatures whereas , in the self - consistent treatment , the currents increase for @xmath105 @xcite . at low temperatures , this leads to higher values of the critical current in the self - consistent calculation . in calculations for a more point contact - like constriction ( @xmath106 ) , we find that this correction due to self - consistency remains . local density of states in the center of the constriction at three different temperatures ( @xmath96 , @xmath107 and @xmath97 from top to bottom ) for different values of the phase difference @xmath59 as indicated in the diagrams . inset : position of the andreev bound states at @xmath102 as a function of the phase difference @xmath59 ( dots ) and @xmath108 for comparison . for this figure , we use the same parameters as for fig.[cap : cphirgapt ] . ] ( a ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the constriction length @xmath7 as indicated in the diagrams . for this figure , we use @xmath84 , @xmath82 and @xmath109 and do not consider magnetic fields . lines are guides for the eye . ] ( b ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the constriction length @xmath7 as indicated in the diagrams . for this figure , we use @xmath84 , @xmath82 and @xmath109 and do not consider magnetic fields . lines are guides for the eye . ] in fig.[cap : ldost ] , we show the local density of states in the center of the constriction for three different temperatures . with increasing phase difference and thus increasing transport current , andreev bound states occur in the constriction . the larger the phase difference , the lower is the energy of the bound states . at @xmath96 , the andreev bound state for @xmath87 still has finite energy . at higher temperatures , when the current - phase relation is not multivalued , the bound states have zero energy for the maximum phase difference @xmath87 . + for @xmath102 , we additionally plot the position of the andreev bound states as a function of the phase difference @xmath59 ( inset in lowermost panel of fig.[cap : ldost ] ) . at temperatures close to the critical temperature @xmath2 , the position of the bound states is given by @xmath110 which is plotted in the inset for reference . at lower temperatures , deviations occur as can be seen in the upper two panels of fig.[cap : ldost ] . + primarily at low temperatures , the andreev bound states do not only have one peak but exhibit additional local maxima at energies above the main peak . this substructure is due to multiple reflections of quasiparticles within the constriction . if we consider the center of the constriction and measure the angle of the trajectories @xmath111 with respect to the @xmath85 axis , an additional reflection within the constriction sets in at angles @xmath112 with @xmath113 the intervals between the angles @xmath112 each correspond to a local maximum of the bound state . the effective distance over which the phase drop occurs along the trajectory increases with every additional reflection and the effective phase gradient thus decreases . + in the case of the s - n - s model , the substructure of the bound states does not occur because the phase drop occurs in the direct vicinity of the constriction . reflections of the trajectories do not change the effective distance over which the phase drop occurs along the trajectories . ( a ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the constriction width @xmath8 as indicated . the dashed lines correspond to a channel without constriction . for this figure , we use @xmath84 , @xmath82 and @xmath114 and do not consider magnetic fields . lines are guides for the eye . ] ( b ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the constriction width @xmath8 as indicated . the dashed lines correspond to a channel without constriction . for this figure , we use @xmath84 , @xmath82 and @xmath114 and do not consider magnetic fields . lines are guides for the eye . ] now let us address the influence of the geometric parameters on the josephson effect . + in fig.[cap : cphirgapll ] , we plot the current - phase relation and the order parameter amplitude in the center of the constriction for different values of the constriction length @xmath7 . the temperature is set to @xmath84 , the width of the constriction to @xmath109 and we do not consider magnetic fields . if the constriction is shorter , more current flows and the suppression of the order parameter amplitude is stronger . thus , for shorter constrictions , the current - phase relation is more sinusoidal and the josephson critical current is higher than for longer ones . the decrease of the critical current with increasing length of the constriction is consistent with the results found by zareyan et al . @xcite for temperatures close to @xmath2 . in fig.[cap : cphirgapbb ] , we show the current - phase relation and the order parameter amplitude in the center of the constriction for different values of the constriction width @xmath8 . as before , the temperature is set to @xmath84 , the length of the constriction to @xmath114 and we do not consider magnetic fields . for reference , we show the current - phase relation without a constriction but with periodic boundary conditions as described in section [ sec : comp ] ( dashed lines in both fig.[cap : cphirgapbb ] ( a ) and ( b ) ) . the dashed lines thus correspond to a long superconducting lead with homogeneous current flow or equally a homogenous bulk current . as expected , the width of the constriction strongly influences the josephson critical current . the narrower the constriction , the stronger is the current concentration in the constriction and the stronger is the suppression of the order parameter amplitude . hence , less current flows at the same phase difference for a narrower constriction . from this section on , we will include the effects of magnetic fields in our considerations . in this section , we will focus on the influence of the parameter @xmath0 whereas in the next section , we will examine the effect of an external magnetic field . so far , we neither considered external magnetic fields nor those caused by the supercurrents . neglecting the latter corresponds to large values of @xmath1 which can easily be seen from equation ( [ eq : apoissonnorm ] ) . in our case , the lateral extension of the geometry is @xmath115 and thus not much happens for @xmath116 . nevertheless , for smaller values of @xmath1 , the distribution of the currents and the magnetic fields changes drastically . ( a ) current density at the critical current for @xmath117 ( a ) and @xmath118 ( b ) . dark shading represents low current density where s bright areas exhibit high current density . arrows indicate the direction of the current flow . for small values of @xmath1 , the current is strongly concentrated on the surface of the sample . ] ( b ) current density at the critical current for @xmath117 ( a ) and @xmath118 ( b ) . dark shading represents low current density where s bright areas exhibit high current density . arrows indicate the direction of the current flow . for small values of @xmath1 , the current is strongly concentrated on the surface of the sample . ] in fig.[cap : currdensityglk ] , we show the current density at the josephson critical current for @xmath117 ( a ) and @xmath118 ( b ) . to calculate these configurations , we use @xmath84 and @xmath83 . for @xmath117 , the current density is homogeneous over the cross section of the channel . nevertheless , in the vicinity of the constriction , the current is strongly concentrated . for @xmath118 , the current is concentrated on the surface of the sample . in the channel , the current distribution thus changes drastically . however , since the constriction size is equal to the london penetration depth @xmath119 with @xmath118 used here , the current density is homogeneous in the constriction . in fig.[cap : magfieldglk ] , we plot the magnetic fields corresponding to the current configurations in fig.[cap : currdensityglk ] . the strong concentration of the current on the surface of the sample leads to the disappearance of the magnetic field in the interior of the channel . nevertheless , the magnetic field in the vicinity of the constriction does not depend much on the value of @xmath1 as long as the dimensions of the constriction are smaller than the london penetration depth ( @xmath120 ) . accordingly , the current - phase relation and the dependence of the order parameter amplitude in the center of the constriction on the phase difference remain unchanged for @xmath121 . thus , we can conclude that the josephson effect is independent of @xmath1 as long as @xmath120 . ( a ) magnetic field distribution @xmath122 for @xmath117 ( a ) and @xmath118 ( b ) . we multiply the magnetic field @xmath123 with @xmath124 in order to have equal scaling of the vertical axis . the geometry of the sample is indicated with the black lines . for smaller values of @xmath1 , the current is concentrated on the surface of the sample and the magnetic field is suppressed in the interior . ] ( b ) magnetic field distribution @xmath122 for @xmath117 ( a ) and @xmath118 ( b ) . we multiply the magnetic field @xmath123 with @xmath124 in order to have equal scaling of the vertical axis . the geometry of the sample is indicated with the black lines . for smaller values of @xmath1 , the current is concentrated on the surface of the sample and the magnetic field is suppressed in the interior . ] in fig.[cap : ldosglk ] , we show the local density of states in the center of the constriction for @xmath84 and @xmath83 . the inhomogeneous configuration for small values of @xmath1 leads to a splitting of the bound states . for the case of @xmath118 and @xmath125 , we discuss the origin of the peaks labeled 1,2,3 in the upper panel of fig.[cap : ldosglk ] in detail . + peak 1 originates from trajectories which are oriented parallel to the @xmath85 axis and thus pass through regions with small current density . peaks 2 and 3 stem from trajectories which are reflected into the lower ( @xmath126 ) and upper ( @xmath127 ) part of the channel , i.e. in regions with high current density . thus , the concentration of the supercurrents on the surface of the sample leads to differing contributions of the trajectories for different angles and yields a splitting of the bound state . local density of states in the center of the constriction for two different values of @xmath1 ( @xmath118 in the upper panel , @xmath117 in the lower panel ) . for this figure , we use @xmath84 and @xmath83 . please note the different scaling of the vertical axis for the two panels . ] in this section , we will finally examine the influence of an external magnetic field on the josephson effect in a scs junction . therefore , we set the temperature to @xmath84 , the geometric parameters to @xmath83 , the parameter @xmath117 and apply an external magnetic field @xmath128 . at the end of this section , we will discuss the temperature dependence of the influence of the external magnetic field . before turning to the results of our calculations , let us briefly discuss the normalization we use for the magnetic field in terms of more familiar quantities . following the procedure presented in @xcite , we calculate the bulk upper critical field @xmath129 using eilenberger theory @xcite . for s - wave superconductors with cylindrical fermi surface and with the magnetic field oriented parallel to the cylinder axis , we find @xmath130 at @xmath84 , the normalized upper critical field thus is @xmath131 , or @xmath132 respectively . an external magnetic field leads to screening currents which flow on the surface of the superconductor . thus , in our case , the currents are a mixture of transport currents across the josephson junction and screening currents . in fig.[cap : currdensityb0 ] , we show the current density in the section enclosing the constriction for an external magnetic field of @xmath133 without transport currents ( @xmath134 , ( a ) ) and with critical transport currents ( @xmath135 , ( b ) ) . in fig.[cap : magfieldb0 ] , we plot the corresponding magnetic field distributions @xmath136 . ( a ) current density with applied external magnetic field ( @xmath117 , @xmath133 ) without transport current ( @xmath134 , ( a ) ) and with critical transport current ( @xmath135 , ( b ) ) . again , dark shading signifies low current density whereas bright shading signifies high current density . arrows indicate the direction of the current flow . without transport currents , a small circulating current appears in the constriction which leads to local antiscreening ( white circular arrow in the constriction in ( a ) ) . with an applied phase difference , the total current distribution becomes asymmetric and consists of both screening and transport currents . ] ( b ) current density with applied external magnetic field ( @xmath117 , @xmath133 ) without transport current ( @xmath134 , ( a ) ) and with critical transport current ( @xmath135 , ( b ) ) . again , dark shading signifies low current density whereas bright shading signifies high current density . arrows indicate the direction of the current flow . without transport currents , a small circulating current appears in the constriction which leads to local antiscreening ( white circular arrow in the constriction in ( a ) ) . with an applied phase difference , the total current distribution becomes asymmetric and consists of both screening and transport currents . ] ( a ) magnetic field distribution @xmath122 for applied external magnetic field ( @xmath117 , @xmath133 ) without transport current ( @xmath134 , ( a ) ) and with critical transport current ( @xmath135 , ( b ) ) . the geometry of the sample is indicated with the black lines . screening currents lead to a reduction of the magnetic field in the interior of the sample . additional transport currents increase the magnetic field on one side of the sample and decrease it on the other side . ] ( b ) magnetic field distribution @xmath122 for applied external magnetic field ( @xmath117 , @xmath133 ) without transport current ( @xmath134 , ( a ) ) and with critical transport current ( @xmath135 , ( b ) ) . the geometry of the sample is indicated with the black lines . screening currents lead to a reduction of the magnetic field in the interior of the sample . additional transport currents increase the magnetic field on one side of the sample and decrease it on the other side . ] without transport currents , we find screening currents which flow on the surface of the channel and which lower the magnetic field inside the sample . since the width of the channel is equal to the london penetration depth ( @xmath119 with @xmath117 used here ) , the magnetic field is not completely suppressed . in the constriction , a small circulating current flows which is directed opposite to the screening currents and which leads to a local enhancement or antiscreening of the magnetic field . + with a finite phase difference applied between the ends of the section , a transport current flows . this leads to a breaking of the fourfold symmetry visible in fig.[cap : currdensityb0 ] ( a ) and to a combined current configuration consisting of screening as well as transport currents . consequently , the magnetic field distribution shows an asymmetry . the total magnetic field is decreased for @xmath126 and increased for @xmath127 . far away from the sample , the external magnetic field is recovered , but this happens on a much larger scale and can not be seen in fig.[cap : magfieldb0 ] . in fig.[cap : cphirgapb0 ] , we show the current - phase relation and the order parameter amplitude in the center of the constriction for different values of the external magnetic field . for small values of the external magnetic field , the strength of the currents is gradually reduced as is the order parameter amplitude in the constriction , but the functional form of the current - phase relation and the order parameter amplitude is basically unchanged . at larger values of the external magnetic field however , the transport currents become very small and the order parameter amplitude in the constriction remains unaffected . at the same time , the josephson effect disappears . the reason for the disappearance of the josephson effect is that with increasing external magnetic field , superconductivity is suppressed in the channel . since the constriction is much smaller than the london penetration depth , the screening currents do not flow across the constriction ( see fig.[cap : currdensityb0 ] ( a ) ) . this leads to pair breaking in the channel and lowers the amplitude of the order parameter . the screening currents mainly flow in the channel and superconductivity first breaks down here . however , if superconductivity is already suppressed in the channel , no current can flow across the constriction and the josephson effect does not occur . ( a ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the external magnetic field @xmath137 . inset in ( a ) : dependence of the josephson critical current on @xmath137 . inset in ( b ) : amplitude of the order parameter in the center of the constriction ( solid line ) and in the center of a long channel without constriction ( dashed line ) for @xmath134 as a function of the external magnetic field . for the whole figure , we use @xmath84 , @xmath83 and @xmath117 . lines are guides for the eye . ] ( b ) current - phase relation ( a ) and order parameter amplitude in the center of the constriction ( b ) for different values of the external magnetic field @xmath137 . inset in ( a ) : dependence of the josephson critical current on @xmath137 . inset in ( b ) : amplitude of the order parameter in the center of the constriction ( solid line ) and in the center of a long channel without constriction ( dashed line ) for @xmath134 as a function of the external magnetic field . for the whole figure , we use @xmath84 , @xmath83 and @xmath117 . lines are guides for the eye . ] the external magnetic field which leads to suppression of the josephson effect is about twice the approximate upper critical field @xmath138 derived above . to verify the validity of this unusually large value of the magnetic field leading to the suppression of superconductivity in the channel , we carried out calculations for a longer channel with the same width ( @xmath139 , @xmath115 ) , but without a constriction . the amplitude of the order parameter in the center of the constriction as well as in the center of the long channel for @xmath134 is plotted in the inset in fig.[cap : cphirgapb0 ] ( b ) . in the channel , superconductivity is completely suppressed at somewhat lower values of the external magnetic field , namely at about @xmath140 . local density of states in the center of the constriction for three different values of the external magnetic field ( @xmath93 , @xmath141 and @xmath142 from bottom to top ) . for this figure , parameters are set corresponding to fig.[cap : cphirgapb0 ] . please note the different scaling of the vertical axis for the three panels . ] temperature dependence of the critical current without magnetic fields ( @xmath82 , @xmath93 , solid guideline ) and with external magnetic field ( @xmath117 , @xmath137 as indicated ) for @xmath83 . for comparison , we plot the kulik - omelyanchuk result ( @xcite , dashed line ) and the ambegaokar - baratoff result ( @xcite , dash - dotted ) . ] since our two - dimensional geometry implies translational invariance in the @xmath10 direction , the high critical field found in our calculations can be understood in terms of the parallel critical field @xmath143 @xcite . because of the infinite extension of the geometry in the @xmath10 direction , the external magnetic field is oriented parallel to the surfaces . up to an external magnetic field @xmath144 oriented parallel to the surface of a superconductor , there is a superconducting surface sheath of thickness @xmath145 while @xmath146 in the interior . if the geometry consists of two coplanar surfaces with spacing @xmath147 , nucleation is possible at even higher values of the external magnetic field , growing with @xmath148 ( see @xcite and references therein ) . since the channel width is @xmath115 , the value of the external magnetic field at which superconductivity is suppressed is in good agreement with @xmath143 . in the center of the constriction , the spacing between opposite surfaces is even smaller and nucleation occurs at even higher values of the external magnetic field . from the inset in fig.[cap : cphirgapb0 ] ( b ) , we conclude that the current across the constriction should be suppressed at about @xmath140 since this value of the external magnetic field leads to suppression of superconductivity in the channel . the section used for the calculations with a constriction ( @xmath149 ) is too short to reach the undisturbed channel , far enough from the weak link . however , using a longer section would lead to a badly defined phase difference across the junction and to excessive numerical costs . our main conclusion , the fact that the external magnetic field primarily affects the channel and not the constriction , remains unchanged , however . + in a thin film with small extension in the @xmath10 direction , the relevant critical field for the suppression of superconductivity is the upper critical field @xmath150 . the occurrence of the parallel critical field @xmath143 in our results stems from the assumption of translational invariance in the @xmath10 direction . in fig.[cap : ldosb0 ] , we plot the local density of states in the center of the constriction for three different values of the magnetic field . at intermediate strength of the external magnetic field , we find a splitting of the andreev bound states , but the main features of the ldos without a magnetic field persist . we still find the energy gap and pronounced bound states if a phase difference @xmath151 is applied . for @xmath87 , the bound states have zero energy which indicates josephson junction behaviour . for very strong external magnetic fields , the local density of states is strongly disturbed . even without a phase difference , the energy gap is partly filled . a phase difference still leads to bound states , but their spectral weight is strongly reduced . for @xmath87 , the bound states have finite energy and the josephson effect is absent . in fig.[cap : jcbt ] , we show the temperature dependence of the critical current for @xmath83 without ( @xmath82 , @xmath93 ) and with external magnetic field ( @xmath117 , @xmath137 as indicated in the figure ) . for comparison , we plot the kulik - omelyanchuk result for a point contact in the clean limit @xcite and the amgebaokar - baratoff result for a tunnel junction @xcite . we find that the universality of the inclination of @xmath152 close to @xmath2 persists even in an external magnetic field . we present self - consistent solutions of microscopic eilenberger theory for a two - dimensional model of a scs josephson junction . magnetic fields , external ones as well as those evoked by the supercurrents , have been included and the relevant equations have been solved numerically without further assumptions . in the self - consistent calculation , the josephson effect appears without further input . unlike in a non - self - consistent calculation , we do not have to make assumptions about the existence and the behaviour of the weak link . by considering the self - consistent results , we explain in detail how the josephson effect in a scs junction emerges . taking into account the order parameter amplitude and phase , the currents and the local density of states , we point out that the josephson effect is a result of phase - coherent quasiparticle transport across the junction . localized quasiparticle states ( andreev bound states ) appear in the junction if a supercurrent flows . these bound states , together with the corresponding current - phase relations , have been calculated and discussed in detail for different geometric parameters , different temperatures , different values of @xmath0 and different values of the external magnetic field . we show that the josephson effect in scs junctions is independent of @xmath1 as long as the london penetration depth is larger than or comparable to the size of the constriction . we also show that the josephson effect in scs junctions is very insensitive to external magnetic fields . external magnetic fields mainly influence the channel leading to the weak link and suppress superconductivity there . the josephson effect persists up to the upper critical field of the channel . in contrast to the non - self - consistent results , the self - consistently calculated local density of states exhibits andreev bound states with peculiar substructure due to the geometry and magnetic fields . the local density of states should be experimentally accessible via stm / sts . present address : institute for solid state physics , university of tokyo , japan . v. i. shnyrkov and s. i. melnik , fiz . nizk . temp . * 33 * , 22 ( 2007 ) [ low temp . phys . * 33 * , 15 ( 2007 ) ] . s. k. h. lam and d. l. tilbrook , appl . . lett . * 82 * , 1078 ( 2003 ) . v. bouchiat , m. faucher , c. thirion , w. wernsdorfer , t. fournier and b. pannetier , appl . . lett . * 79 * , 123 ( 2001 ) . k. k. likharev , rev . phys . * 51 * , 101 ( 1979 ) . a. a. golubov , m. yu . kupriyanov and e. ilichev , rev . * 76 * , 411 ( 2004 ) . a. v. zaitsev , jetp * 86 * , 1742 ( 1984 ) [ sov . jetp * 59 * , 1015 ( 1984 ) ] . a. martn - rodero , f. j. garca - vidal and a. levy yeyati , phys . * 72 * , 554 ( 1994 ) . a. levy yeyati , a. martn - rodero and f. j. garca - vidal , phys . b * 51 * , 3743 ( 1995 ) . in the present work , we use @xmath153 throughout . m. zareyan , yu . a. kolesnichenko and a. n. omelyanchouk , low . temp . * 25 * , 864 ( 1999 ) . i. o. kulik , jetp * 57 * , 1745 ( 1969 ) [ sov . jetp * 30 * , 944 ( 1970 ) ] . c. ishii , progr . . phys . * 44 * , 1525 ( 1970 ) . c. ishii , progr . . phys . * 47 * , 1464 ( 1972 ) . a. f. andreev , jetp * 46 * , 1823 ( 1964 ) [ sov . jetp * 19 * , 1228 ( 1964 ) ] . a. f. andreev , jetp * 49 * , 655 ( 1965 ) [ sov . jetp * 22 * , 455 ( 1966 ) ] . a. furusaki and m. tsukada , physica b * 165&166 * , 967 ( 1990 ) . a. furusaki and m. tsukada , phys . b * 43 * , 10164 ( 1991 ) . a. furusaki , superlatt . and microstruct . * 25 * , 809 ( 1999 ) . p. miller and j. k. freericks , j. phys . : condens . matter * 13 * , 3187 ( 2001 ) . r. ziki , l. dobrosavljevi - gruji and z. radovi , phys . b * 59 * , 14644 ( 1998 ) . j. pearl , appl . phys . lett * 5 * , 65 ( 1964 ) . g. eilenberger , z. physik * 214 * , 195 ( 1968 ) . n. schopohl and k. maki , phys . b * 52 * , 490 ( 1995 ) . n. schopohl , in _ quasiclassical methods in the theory of superconductivity and superfluidity _ , bayreuth , germany , 1998 , edited by d. rainer and j. a. sauls , cond - mat/9804064 ( unpublished ) . a. shelankov and m. ozana , phys . b * 61 * , 7077 ( 2000 ) . c. iniotakis , s. graser , t. dahm and n. schopohl , phys . b * 71 * , 214508 ( 2005 ) . a. furusaki and m. tsukada , solid state commun . * 78 * , 299 ( 1991 ) . f. sols and j. ferrer , phys . b * 49 * , 15913 ( 1994 ) . m. tinkham , _ introduction to superconductivity _ , mcgraw - hill ( new york , 2 . ed . t. dahm , s. graser , c. iniotakis and n. schopohl , phys . rev . b * 66 * , 144515 ( 2002 ) . i. o. kulik and a. n. omelyanchouk , fiz . nizk . temp . * 3 * , 945 ( 1977 ) [ sov . j. low temp . * 3 * , 459 ( 1977 ) ] . v. ambegaokar and a. baratoff , phys . * 10 * , 486 ( 1963 ) . t. dahm and n. schopohl , phys . * 91 * , 017001 ( 2003 ) . g. eilenberger , z. physik * 190 * , 142 ( 1966 ) . d. saint - james and p. g. de gennes , phys . * 7 * , 306 ( 1963 ) . p. scotto and w. pesch , j. low temp . phys . * 84 * , 301 ( 1991 ) . a. p. van gelder , phys . lett . * 20 * , 1435 ( 1968 ) . h. j. fink , phys . rev . * 177 * , 1017 ( 1969 ) .
narrow line seyfert 1 galaxies ( osterbrock & pogge 1985 ) form a special class of active galactic nuclei . narrowness of the optical lines frequently correlates with the steepness of the soft x - ray spectra ( boller , brandt & fink 1996 ) hinting for a connection of this class with soft state galactic sources . the overall spectra are sometimes well described by a disk emission superimposed on a broad ir - x - ray power law component ( puchnarewicz et al . 2001 ) but in most cases the soft x - ray slope seems to be too shallow for an extension of a disk black body emission . such a steep soft x - ray spectrum of a roughly power law shape may form either as a result of comptonization ( e.g. czerny & elvis 1987 ) by a moderately hot material ( in comparison with the very hot plasma responsible for the hard x - ray component in normal seyfert 1 galaxies ) or as a result of reflection by a strongly ionized material ( e.g. czerny & . zycki 1994 , czerny & dumont 1998 ) . the established presence of warm absorber seen in many of those objects ( collinge et al . 2001 , kastra et al . 2002 , kaspi et al . 2002 , kinkhabwala et al . 2002 , leighly et al . 2002 , yaqoob et al . 2002 ) complicates the analysis . however , with the extremely good spectral resolution of _ chandra _ and _ xmm _ the issue may be addressed through the careful analysis of the existing spectral features . ton s180 ( @xmath0 , corresponding to 286 mpc for hubble constant 65 km s@xmath1 mpc@xmath1 ) is an extreme case of a nls1 with its low value of the fwhm of @xmath2 line ( @xmath3 km s@xmath1 ; wisotzki et al . 1995 ) and steep soft x - ray spectrum of photon index @xmath4 ( comastri et al . 1998 ) the galactic absorption toward the object is low ( @xmath5 @xmath6 ; stark et al . the source was extensively observed in all energy bands . the broad band spectral energy distribution ( sed ) determined by turner et al . ( 2002 ) indicates that most of the energy is emitted in the band @xmath7 ev . furthermore , _ hst _ data show typical emission lines as expected for a seyfert 1 galaxy , and the lack of absorption features . high resolution spectra obtained by _ fuse _ reveal uv absorption by ovi and the absence of neutral hydrogen absorption , indicating high ionization state for the absorbing gas . high resolution x - ray spectra of this object were recently obtained with _ xmm - newton _ and _ chandra _ satellites . _ xmm - newton _ spectrum was steep and featureless ( vaughan et al . _ chandra _ data also appeared predominantly featureless and well represented by a single power law with some soft excess ( turner et al . 2001 , turner et al . the shape of the soft excess was fitted by a diskbb model , however plotted residua hinted for the possible presence of sharp spectral features ( fig . 1 turner et al . 2001 ) . therefore , in the present paper we reanalyze _ chandra _ data in some detail ( sec . [ sec : cha ] and [ sec : spe ] ) , using a new response matrix acisleg1d1999-07-22rmfn0002.fits and including any effects related to the contamination layer on acis - s . we note that the observation of ton s180 was performed on day 143 of the chandra mission when the effects of the contaminant were much smaller than currently observed . we detected several weak narrow absorption lines due to the material located at the line of sight to the source . also , several individual disk lines are fitted with equivalent widths in rough agreement with the predictions of the model of illuminated accretion disk spectrum presented by raska et al . ( 2002 a , b ) . in sec . [ sec : apl ] we discuss existing models ( kaastra et al . 2002 , kinkhabwala et al . 2002 and others ) for fitting absorption by warm , partially ionized gas present in many agn . we point out , after krolik ( 2001 ) , that the warm absorber may not be the slab of constant density matter , but it is rather the distribution of regions with different densities and temperatures . such a multi - phase mixture of gas can exist under constant pressure , as a result of thermal instabilities ( krolik , mckee & tarter 1981 ) . till now , the multi - phase models of warm absorber were not constructed , because popular photoionization codes ( xstar , cloudy ) , compute radiative transfer through the constant density matter . in sec . [ sec : mod ] , we present new set of models computed by the code titan ( dumont , abrassart & collin 2000 ) , which solves radiative transfer through the matter in non - lte approach . introducing assumption of constant pressure ( not density ) of warm absorber , we obtain spectra with many absorption lines from the gas with strong temperature and density gradient . our code is suitable for any optical thickness of the absorber , nevertheless , we consider regions with total column density of order of @xmath8 , as suggested by krolik ( 2001 ) to be a typical property of warm absorber . in sec . [ sec : bbc ] we construct two global disk corona models trying to predict the slope of observed continuum . the model with higher black hole mass ( @xmath9 ) is more promising in the sense that emission lines from the disk are expected in the spectra , since hot skin does not cover disk completely . sec . [ sec : dis ] is devoted to the discussion , while conclusions are listed in sec . [ sec : conc ] . ton s180 was observed with the _ chandra _ satellite in december 14 , 1999 , for about 80 ksec in low - energy transmission grating in acis - s / letg configuration , as described by turner et al . ( 2001 ) and turner et al . the resolution of letg corresponds to an @xmath10 km s@xmath1 for the range of energies from 0.1 up to 6 kev . the source luminosity at this epoch ( 210 kev unabsorbed flux equal @xmath11 erg s@xmath1 @xmath6 and 0.310 kev unabsorbed flux equal @xmath12 erg s@xmath1 @xmath6 ) was comparable to that during the _ beppo - sax _ observation ( 210 kev : @xmath13 erg s@xmath1 @xmath6 ; comastri et al . 1998 ) and _ xmm - newton _ observation ( 0.3 - 10 kev : @xmath14 erg s@xmath1 @xmath6 ; vaughan et al . 2002 ) and somewhat fainter than during the 12 days _ asca _ monitoring ( 210 kev : @xmath15 erg s@xmath1 @xmath6 ; romano et al . 2002 ) . we obtained the data from the chandra archive . the data were processed with the standard cxc data processing pipeline ascdsver 6.7.0 . we analyzed the data using ciao version 2.2 . and follow grating analysis thread ( cxc web page ) to extract the spectrum and create appropriate effective area files ( arf files ) . we combined both sides of the first order spectrum . we used calibration data base version caldb v.2.14 , in particular we applied acisd1997-04-17qen0003.fits , quantum efficiency file and acisleg1d1999-07-22rmfn0002.fits as response matrix . this formally allows us to study the entire energy range between 0.12 till 12.2 kev , although calibration uncertainties are higher below 0.3 kev , while the effective area drops significantly above @xmath16 kev . after removing bad pixels and rebinning the data to achieve at least 20 counts per channel we examined the energy between 0.18 and 6 kev . all spectral fits have been performed with the xspec 9.0 fitting package . since the begining of the _ chandra _ mission acis - s detector has been accumulating a layer of a contaminant which affects any measurements at low energies ( below @xmath17 kev ) . detailed effects of the contaminant are still being investigated by the chandra x - ray center , however , it is clear that the overall degradation of acis quantum efficiency and presence of additional absorption features is time dependent . ton s180 was observed at the day 143 of the _ chandra _ mission and based on the currently available calibration data we expect that the instrument shows maximum of @xmath1840 - 15 % degradation within 0.28 - 0.6 kev range ( i.e. maximum @xmath19 % in 0.28 kev and @xmath20 % in 0.6 kev ) . in the following we identify the possible absorption features due to the absorber in ton s180 using the entire available energy range . we note here that some of the features at energies close to the edges of the contaminant ( c , n , o , f k - edge ) may be affected . we use the available model , acisabs , to include the contamination effects in the analysis . in the first step , we fit the @xmath21 value of galactic absorption together with a broken power law continuum . our best fit value of @xmath22 @xmath6 is slightly larger than galactic absorption @xmath23 @xmath6 determined by dickey & lockman ( 1990 ) and used by comastri et al . ( 1998 ) or than @xmath24 @xmath6 determined by stark et al . ( 1992 ) and used by turner et al . ( 1998 , 2001 ) . all fitted parameters for `` plain continuum '' are listed in table [ tab0 ] . in the next step we fit broken power law with value of galactic absorption fixed at @xmath25 @xmath6 ( see table [ tab0 ] second line ) . it results in slightly higher @xmath26 , but the power law index below 2 kev is @xmath27 , and agrees with the best fit slope @xmath28 presented by comastri et al . 1998 for beppo sax data . lower value obtained by turner et al . ( 2001 and 2002 ) , equal to @xmath29 resulted from the use of additional component to model the soft x - ray excess . the energy break obtained by us in both models ( about @xmath30 kev ) is somewhat lower than the value determined by comastri et al . 1998 ( @xmath31 ) , but the spectral slope above this energy break , @xmath32 agrees with results comastri et al . 1998 . since both broken power - law indices are very similar to those obtained from other satellites we adopt second model ( fixed galactic absorption ) for further analysis , leaving @xmath33 per 1682 d.o.f . ( reduced @xmath34 ) . llllll plain cont . & @xmath35 & @xmath36 & @xmath37 & @xmath38 & 1185/1681 + fixed @xmath39 & 1.52 & @xmath40 & @xmath41 & @xmath42 & 1197/1682 + with lines & @xmath43 & @xmath44 & @xmath45 & @xmath46 & 932/1624 + after fitting continuum , there is still some structure in the residuals of the soft x - ray spectrum , suggesting the presence of absorbing matter along the line of sight toward the source . therefore , we have tried to determine column density of this gas in the simplest way , by fitting ovii , oviii , and cvi edges often seen in seyfert galaxies ( mrk 509 yaqoob et al . 2002 , ngc 3783 kaspi et al . surprisingly , such edges appeared to be very weak in the data of ton s180 . we were able only to give upper limits for @xmath47 @xmath48 , and @xmath49 of order of @xmath50 , which practically indicates the lack of those edges . x - ray absorption lines appear to be common feature in many agn . the widely acceptable explanation is the existence of warm , ionized gas which is illuminated by the energetic radiation from nucleus . trying to find properties of warm absorber we need proper tools to fit those data . till now , models of absorption by the partially ionized plasma available in xspec ( absori , photo , ) do assume slab with constant density . in absori , the radiative transfer of lines is not computed at all , only photoionization cross sections are determined and absorption `` on the spot '' is computed @xmath51 ( done et al . constructed by kinkhabawa et al . ( 2002 ) local photo model ( http://xmm.astro.columbia.edu/research.html ) neglects photo - electric absorption , but allows for individual line absorption , treating each ionic column density separately with an initially unabsorbed power law . this method produces consistent results only when absorbing matter is optically thin , and transmitted spectra do not display photo - electric edges . furthermore , they have assumed that plasma is optically thin to reemitted photons . in the aim to find evidences for warm absorber in ton s180 , we have applied absori to our data . allowing for three parameters to be fitted in this model , @xmath26 decreases by 38 ( now reduced @xmath52 ) . the fit indicates that the medium has column density @xmath53 @xmath6 and ionization parameter ( @xmath54 ) for the fixed default temperature @xmath55 k. this model , however , leaves considerable residual trends and does not allow for a deeper insight into the ionization state of the absorbing material . it is not surprising since the model is based on simple computations of photoionization cross sections without accurate transfer through the medium . we have applied photo to our analysis . the model again indicates very low column densities of each ion , about @xmath56 @xmath6 for cvi , ovii , and oviii . we conclude that most probably , the warm absorber in ton s180 is more complicated than optically thin constant density slab . it may be the stratified density gas , where accurate radiative transfer , including all photoionization processes , should be calculate to model transmitted spectrum . such models are not available for fitting the data yet , so two or three separate absorbers with different constant densities and ionization states are used together to explain spectral features in some agn ( kaastra et al . 2002 , kaspi et al . this motivates us to present in sec . [ sec : mod ] the model of the multi - phase gas with stratified density and temperature , where the radiative transfer of incident power law is carefully computed . the model is not ready for fitting yet , but gives interesting results based on equivalent widths of various lines . observed line properties in the ton s180 data are determined in the section below . more advanced modeling was done with the code xstar ( developed by kallman & krolik 1999 ) which allows to estimate the total column densities of absorbed gas . models are computed assuming constant density slab . such models are very useful for reconstruction of the flat part of curve of growth ( yaqoob et al . 2002 ) , and for subsequent estimation of turbulent velocities of the absorber . nevertheless , at the last stage of analysis separate gaussian components are fitted to the data . we have successfully fitted several absorption lines using standard gaussian model for each line separately ( see figs . [ fig : cont ] , [ fig2:cont ] , [ fig3:cont ] , [ fig4:cont ] , [ fig5:cont ] , [ fig6:cont ] , and [ fig7:cont ] . * gaussian fit requires three free parameters , energy of the line centroid , gaussian width and normalizations , which has to be negative in case of absorption lines . * we include in the fit only those features which are significant . all absorption lines are presented in table [ tab1 ] . for each line we list its central energy in kevs , equivalent width ( ew ) in evs , the decrease of @xmath26 , best identification with observed and rest frame energy , and with possible blends . some of those lines , especially the softest ones , may be caused by dust presented on chandra mirrors . this should show up a common feature in all objects even those with different redshifts . to check whether lines are real or instrumental , we compare observations of ton s180 with _ chandra _ data of quasar 3c273 observed two weeks latter with the same configuration as ton s180 . exemplary residuals ( for soft range between 0.2 - 0.24 kev ) for both objects are shown in fig . [ por:22 ] . in both cases local power - law was fitted with fixed value of galactic absorption appropriate for each object . the same features around energy 0.22 kev are present as well in 3c273 with redshift 0.156 as in ton s180 . we repeated such analysis in the whole energy range . all possible instrumental lines are labeled in table [ tab1 ] . some of the detected features are seen only in ton s180 spectrum and therefore seem to be real , intrinsic to the source . for example , in fig [ por:44 ] we show that line cvi with observed energy 0.4428 kev which is detected in ton s180 data is not seen in 3c273 . lines are very faint . they usually have equivalent width less or about 2 evs , and several of them are just marginal detections , as clearly seen from the line ew error , given at 90% confidence level ( * each error is computed for the one parameter of interest * ) . only two of them are stronger , but from theoretical predictions by analysing our modeled spectra from sec [ sec : mod ] , they are not blends . all possible blends according of our modeled specta obtained with spectral resolution 100 are listed in table [ tab1 ] . there is therefore no major discrepancy between the _ chandra _ results and the _ xmm _ observations ( vaughan et al . 2002 ) where only upper limits of order of 1 ev at energies 0.35 - 0.7 kev were provided . most lines originate from hydrogen or helium like ions indicating that matter is highly ionized . there is an absence of lines from mildly ionized spices . interestingly , we found two different transitions from the same element at the given ionization level , ( cvi@xmath57 and cvi@xmath58 where i j denotes number of levels from and to which the transition proceeds ) . this is helpful in the aim to estimate the optical thickness of cvi ions and then total optical thickness of the warm absorber . if the medium is optically thin in those lines ( i.e. weak line case ) , we are on the linear part of the curve of growth , where equivalent width is proportional to the column density @xmath59 of the absorbed ion : @xmath60 where @xmath61 is the oscillator strength transition and stimulated emission has been set equal to unity . thus , for the given population of ions at the same ionization level , the ratio of equivalent widths is given by : @xmath62 since the oscillator strengths are strictly atomic features independent from the radiation field and are calculated theoretically we use them to find the ratio @xmath63}{ew[{\rm c vi}_{(1 - 5 ) } ] } = 29.86 . \label{equ : tra}\ ] ] on the other hand , the same ratio determined for ton s180 ( see table [ tab1 ] ) equals @xmath64 . error is computed assuming most pessimistic case , when numerator has its maximal value @xmath65 , and denominator has its minimal value @xmath66 . although the error is huge , the value of the ratio determined from observation is still far from theoretical predictions for optically thin medium ( eq . [ equ : tra ] ) . our theoretical spectra ( see sec . [ sec : mod ] ) suggest that c@xmath57 line is possibly a blend with mgxii , sixiv and most strongly with cv . we have computed that real ew of the c@xmath57 line can be even up to four times lower than the ew of the whole blend . but this fact even lowers the observed ratio of ews of two carbon lines . the straight conclusion is that the emitting region is optically thick for those lines and the transfer in lines should be taken into account for proper modeling of the warm absorber . it is also possible that the identification of the lines is not correct or the line is a blend of the carbon line with another transition . nevertheless , we further explore the possibility that the carbon line ratio is determined correctly . therefore , the equation [ equ : column ] gives only a lower limit for the ion column density , equal to @xmath67 @xmath6 from the cvi@xmath57 line and @xmath68 @xmath6 from cvi@xmath58 , which under assumption of all carbon being in the form of cvi ions and with solar abundances translates into the lower limit for the warm absorber column equal @xmath69 @xmath6 and @xmath70 @xmath6 , correspondingly . exact determination of the column density is complicated . at the flat curve of growth we need additional information about velocity dispersion @xmath71 to indicate column densities . in the case of considered data this is very difficult , because @xmath72 error of full width of half maximum ( fwhm ) is too high to estimate @xmath71 parameter . for instance for sxii line , presented in fig . [ fig3:cont ] , @xmath73 km / s . this fact also precludes us to integrate over fitted line profile in the aim of determining exact flux emitted in this line . so , we are unable to compute exact optical depth of the line and therefore the column density of absorbing ion . the number of lines is to small to fit full model ( for instance xstar ) . another lower limit on column density of cvi ion can be derived knowing that the optical thickness of the cvi lines is higher than 1 . in such a case we can write : @xmath74 where @xmath75 is the temperature of the medium and @xmath76 is atomic mass of the ion . for the cvi@xmath58 line we obtain : @xmath77 for the temperature approximately equal to @xmath78 k this value is twice higher than the one determined by mason et al.(2002 ) in case of mrk 766 . this limit is over an order of magnitude weaker than the limit obtained directly from equation [ equ : column ] . fwhm estimated from the @xmath79 k is quite narrow and the line width is rather determined by much larger turbulent motion . the value of reduced @xmath26 was equal 0.6363 after fitting all possible absorption lines . after absorption lines have been identified , we were able to fit a few emission lines from an accretion disk . we applied the standard diskline model with fixed power law index in the radial dependence of the irradiating flux @xmath80 . we assumed @xmath81 and @xmath82 ( where @xmath83 ) . we have found the best value of inclination angle @xmath84 for the line cv , and we adopted it ( keeping frozen ) for the other lines . results are listed in table [ tab2 ] , and shown in figs . [ fig : cont ] , and [ fig8:cont ] . the last column of table [ tab2 ] presents theoretical values of ews of the same emission lines computed for accretion disk illuminated by hard x - ray flux equaled to the local disk flux ( full model presented in raska et al . 2002 a , b ) . the model takes into account harder incident radiation ( @xmath85 ) , than it is seen in case of ton s180 , and solar metal abundances . for moderate illumination in the model we have found for instance the equivalent width of nex equal 11 ev , oviii equal 18.2 , mgxii equal 10.5 , and fexvii about 8 ev , or many others at the same order of magnitude in respect to the reflected continuum ( factor of 2 - 3 smaller in respect to the total continuum ) . models do not predict lines with ews higher than @xmath86 evs , when the disk is in the hydrostatic equilibrium ( raska et al . 2002 a , b ; see also figures in nayakshin , kazanas & kallman 2000 , ballantyne , ross & fabian 2001 ) . higher values can be only obtained from constant density medium , as discussed by ballantyne , ross & fabian ( 2002 ) , particularly when adopting enhanced metalicity . emission lines also become by a factor of a few stronger when the dynamical flare model is adopted , i.e. the time - scale of the flare is shorter than the time - scale for a disk surface layer to achieve a hydrostatic equilibrium ( collin et al . 2002 ) . however , in that case lines should be narrow and not represented well by a diskline shape . soft disk line detection is therefore a promising tool to constrain the conditions within the cold reprocessing medium . detection of standard broad disk lines presented in this paper , if real , suggest no strong departure from hydrostatic equilibrium and the extension of the cold disk roughly down to the marginally stable orbit . after all lines have been fitted , we fit again broken power - law with galactic absorption . results presented in table [ tab0 ] last verse indicate that lines do not affect continuum very strongly . the final value of reduced @xmath26 appeared to be 0.5531 . llcc @xmath87 & @xmath88 & 7.42 & _ instumental _ + @xmath89 & @xmath90 & 6.6 & _ instumental _ + @xmath91 & @xmath92 & 22.29 & _ instumental _ + @xmath93 & @xmath94 & 6.34 & _ instumental _ + @xmath95 & @xmath96 & 15.35 & _ instumental _ + @xmath97 & @xmath98 & 6.92 & _ instumental _ + @xmath99 & @xmath100 & 14 . & mgxi ( 0.2971 ; 0.3155 ) + @xmath101 & @xmath102 & 5.18 & sixiii ( 0.30878 ; 0.3279 ) + @xmath103 & @xmath104 & 7.55 & sxii ( 0.31899 ; 0.3387 ) + @xmath105 & @xmath106 & 3.47 & cv ( 0.33386 ; 0.3545 ) + @xmath107 & @xmath108 & 4.45 & cvi ( 0.3459 ; 0.3673 ) + & & & blend with mgxii ( 0.368 ) , + & & & sixiv ( 0.3703 ) , cv ( 0.3709 ) + @xmath109 & @xmath110 & 5.16 & cvi ( 0.4428 ; 0.4702 ) + @xmath111 & @xmath112 & 3.47 & fexvii ( 0.6868 ; 0.72932 ) + & & & blend with sxvi ( 0.731 ) + @xmath113 & @xmath114 & 3.29 & nex ( 1.1387 ; 1.209 ) + & & & blend with fexxv ( 1.206 ) , + & & & fexxv ( 1.208 ) , and fexxv ( 1.213 ) + @xmath115 & @xmath116 & 2.95 & mgxii ( 1.3861 ; 1.472 ) + @xmath117 & @xmath118 & 3.7 & mgxii ( 1.7327 ; 1.840 ) + & & & blend with sixiii ( 1.853 ) + @xmath119 & @xmath120 & 6.09 & sixiv ( 1.8831 ; 2.000 ) + lllcc @xmath121 & @xmath122 & 72.73 & cv ( 0.2899 ; 0.3079 ) & 8.69 + @xmath123 & @xmath124 & 21.26 & oviii ( 0.6149 ; 0.653 ) & 19.0 + @xmath125 & @xmath126 & 46.26 & nex ( 0.961 ; 1.020 ) & 11.29 + in this section , we present models of warm absorber , allowing density , temperature and so consequently ionization state to be stratified ( krolik 2002 ) . the only assumption which we made is the constant pressure . such an approach is more realistic than the constant density model , since thermal instabilities can produce gas with steep ionization structure keeping all zones in pressure equilibrium . we have used code titan ( dumont , abrassart & collin 2000 ) to compute temperature and density structure of the warm absorber together with radiative transfer . the code is based on two - stream approximation and works in plane - parallel geometry . radiative transfer is computed both in lines and in continuum so the code can be used for very inhomogeneous even optically thick media . thermal , ionization and statistical equilibrium of ions are computed in complete non - lte for 10 elements and all corresponding ions . we consider following elemental abundances : h : @xmath127 , he : @xmath128 , c : @xmath129 , n : @xmath130 , o : @xmath131 , ne : @xmath132 , mg : @xmath133 , si : @xmath134 , s : @xmath135 , fe : @xmath136 , . we transfer totally 721 lines of different ions , taking their oscillator strengths and energies from national institute of standards and technology ( nist , http://physics.nist.gov ) . photo - electric absorption is fully taken into account . in our models , we have assumed the shape of incident radiation consistent with this fitted to the data of ton s180 . we have adopted power law photon index @xmath137 ( sec [ sec : cont ] ) , and spectral limits from @xmath138 up to @xmath139 kev as we have concluded from broad band spectral energy distribution presented by turner et al . ( 2002 ) . the assumption of constant pressure refers to the total : radiation and gas pressure , and its value is self consistently determined by the code . the ionization parameter on the top of the absorber @xmath140 and the total column density are free parameters of our models . we have computed models for @xmath141 and @xmath142 named a1 , @xmath143 and @xmath144 named a2 , and @xmath145 and @xmath142 named a3 . such values of column densities for given @xmath140 are upper limits above which optical depth of cloud is large enough that strong absorption edges become present in outgoing spectra . even for high @xmath140 , the maximum compton temperature is low , @xmath146 k , because the incident spectrum is soft ( @xmath147 ) . in fig . [ fig : mod ] we show the temperature structure for a1 ( dashed line ) ( higher @xmath140 does not change this structure ) compared to a2 ( solid line ) and in fig . [ fig : mod1 ] we show the corresponding spectra . in case of higher @xmath140 ( model a1 ) the temperature range within the medium is much broader and the effective spectrum from such region ( presented in fig . [ fig : mod1 ] dashed line ) displays slightly stronger lines than for a2 ( solid line spectrum ) . we have enlarged range of frequencies to illustrate region observed in ton s180 data . we have also performed comparison of a1 model to the warm absorber with constant density ( fig . [ fig : mod ] dotted - dashed line ) assuming that both clouds have the same total column densities and they are illuminated by the equivalent fluxes . thus , choosing the volume density for constant density cloud , the ionization parameter is determined by the value of flux the same as for stratified cloud . we have chosen volume density equal @xmath148 , which implies @xmath149 . spectrum for constant density cloud presented in fig . [ fig : mod1 ] by dotted - dashed line displays strong edges and can not be applicable to explain our observations . decreasing volume density we increase ionization parameter , and it may remove edges . we plan to make proper calculations in the future work to show how to distinguish between low density constant cloud and stratified cloud . we have also calculated theoretical values of the equivalent widths of transfered lines , the same which are found in ton s180 data . results are presented in table [ tab3 ] . ews for high surface ionization ( a1 ) are almost a factor of two higher that for a2 model . in our data we see slightly larger ews and we can not model them by increasing @xmath140 , since model a3 shows the same ews as a1 . therefore , we have to introduce model labeled a4 in table [ tab3 ] , which has exactly the same parameters as a2 i.e. @xmath143 and @xmath144 and additionally we assume turbulent velocity @xmath150 . turbulent velocity does not change the total temperature structure , it only affects lines , making their ews higher , as shown in table [ tab3 ] . for model a4 several lines , for instance mgxi , sixiii , sxii cvi , and sixiv ) have ews in quite good agreement with those achieved from observations ( see table [ tab1 ] ) . in all presented models the ratio of @xmath151[cvi@xmath152 $ ] / @xmath151[cvi@xmath153 $ ] equals 0.7 , in very good agreement with observed value of 0.6 . llllll sviii & 2s@xmath154 2p@xmath155 @xmath154p@xmath156 - 2s@xmath154 2p@xmath155 3d @xmath154l & 0.522 & 0.293 & 0.506 & 1.431 + fexv & 3s@xmath154 @xmath157s - 3s4p @xmath157p@xmath156 & 0.406 & 0.191 & 0.392 & 1.164 + six & 2s@xmath154 2p @xmath154p@xmath156 - 2s@xmath154 3d @xmath154d & 0.584 & 0.286 & 0.568 & 1.193 + sx & 2s@xmath154 2p@xmath158 @xmath159s@xmath156 -2s@xmath154 2p@xmath154(@xmath158p ) 3l @xmath159p & 0.585 & 0.305 & 0.567 & 1.549 + cv & 1s@xmath154 @xmath157s - 2p @xmath157p@xmath156 & 1.112 & 0.587 & 1.079 & 1.849 + sxi & 2s@xmath154 2p@xmath154 @xmath158p - 2s@xmath154 2p 3d @xmath158l@xmath156 & 0.680 & 0.333 & 0.659 & 1.500 + mgxi & 2p @xmath158p@xmath156 - 4l @xmath158l & 0.834 & 0.468 & 0.809 & 1.995 + sixiii & 2s @xmath157s - 3p @xmath157p@xmath156 & 0.783 & 0.439 & 0.759 & 2.019 + sxii & 2s@xmath154 2p @xmath154p@xmath156 - 2s@xmath154 3d @xmath154d & 0.719 & 0.348 & 0.697 & 1.580 + cv & 1s@xmath154 @xmath157s - 3p @xmath157p@xmath156 & 1.287 & 0.708 & 1.249 & 2.151 + cvi & 1s @xmath154s - 2l @xmath154l & 1.186 & 0.599 & 1.136 & 1.772 + cvi & 1s @xmath154s - 5l @xmath154l & 1.650 & 0.881 & 1.585 & 2.585 + fexvii & 2s@xmath154 2p@xmath160 @xmath157s - 2s@xmath154 2p@xmath1613s @xmath157p@xmath156 & 1.216 & 0.621 & 1.159 & 4.008 + nex & 1s @xmath154s - 3l @xmath154l & 3.151 & 1.856 & 3.063 & 7.485 + mgxii & 1s @xmath154s - 2l @xmath154l & 3.561 & 2.101 & 3.561 & 9.107 + mgxii & 1s @xmath154s - 4l @xmath154l & 5.224 & 2.947 & 5.330 & 12.26 + sixiv & 1s @xmath154s - 2l @xmath154l & 4.594 & 2.426 & 4.598 & 11.56 + in order to draw attention to the possible constraints resulting from detection of the broad disk lines we model the broad band spectrum of ton s180 assuming three basic spectral components : ( a ) an illuminated accretion disk ( b ) an optically thick comptonizing plasma ( c ) an optically thin ( non - thermal ) comptonizing plasma . for low or moderate accretion rate we take the simplest keplerian disk model , emitting locally as a black body and parameterized fully by the black hole mass , @xmath162 , and accretion rate , @xmath163 , given in eddington units ( taking efficiency 1/12 appropriate for newtonian model ) . we consider also highly supercritical accretion disk applying the model described by kawaguchi ( 2002 ) . this model includes the effect of strong radial advection and the departure from the local black body emission due to electron scattering which are essential at very high accretion rates . we assume that the energy generated in the inner disk illuminates outer disk , with the irradiating flux @xmath164 , with @xmath165 . this irradiation effect may come from the scattering of the disk radiation back toward the disk by the warm absorber as well as from the direct interception of the radiation by the flaring outer disk . we adopted a simple parameterization and did not model this irradiation since the process is quite complex . it would require assumption about the radial and angular distribution of the density of the scattering medium ( see e.g. kurpiewski et al . 1997 ) and the computations of the viscosity dependent disk vertical structure , as in raska et al . we did not compute the disk vertical structure in order to verify whether the outer disk is flaring strongly enough to allow for such a strong effect . we assume that the inner disk for ( @xmath166 ) is fully covered by the optically thick plasma . thermal plasma parameters are : the optical depth , @xmath167 , and the temperature , @xmath75 . only soft photons from the inner disk come through this plasma , and we compute the spectrum using the simple code based on sunyaev & titarchuk ( 1980 ) description ( used for example by collin - souffrin et al . we adopt the plane parallel geometry and the generation of the photons at the bottom of the zone . this component forms a soft x - ray component of the spectrum . we next added a contribution from much hotter , or non - thermal , plasma component in order to provide the appropriate hard x - ray tail but the parameters of this component are poorly constraint and we do not address this issue in detail . our model provides a one - parameter family of solutions for the broad band spectrum , i.e. we can find a good model for any mass of the black hole between @xmath168 and @xmath169 , but for a fixed mass all other parameters ( @xmath163 , @xmath170 , @xmath167 , and @xmath75 ) are uniquely determined . the appropriate accretion rate is needed to reproduce the normalization of the spectrum in the optical band , where it is mostly given by the product of @xmath171 ( for illumination case ) . the value of @xmath170 must be chosen in order not to overproduce / underproduce far - uv flux . the slope of the soft x - ray spectrum gives one constraint on the combination of @xmath167 and @xmath75 , but for low values of @xmath75 which are always derived the normalization of the soft comptonized spectrum also depends significantly on the plasma parameters ( for a number of soft photons already specified by @xmath170 ) . examples of the solution are therefore as follows : for @xmath172 : we have @xmath173 , @xmath174 , @xmath175 , @xmath176 kev , and for high accretion rate case : @xmath177 : we have @xmath178 , @xmath179 , @xmath180 , @xmath181 kev ; the broad band data fit for the largest mass is shown in fig . [ fig : bbwidmo_a ] and for the smallest mass in fig . [ fig : toshi ] . those two extreme models differ considerably with the extension of the disk part covered with the hot skin . in the case of small mass , the disk model would overpredict the far - uv data points considerably if the large portion of the inner disk is not fully covered with the thermal plasma . plasma parameters are such that they make formation of any emission lines rather difficult . in the case of high mass the inner disk does not really have to be covered , as indicated by much smaller value of @xmath170 . clumpy corona model would be easily applicable to this case , or even the x - ray emission could come from some shocks in outflowing material at the rotation axis . the fraction of the disk covered may be always further reduced assuming that synchrotron emission of the plasma itself provides additional soft photons . in such a case the lines from the irradiated disk are easily expected . reanalyzing _ chandra _ data of narrow line seyfert 1 galaxy - ton s180 we were able to identify several narrow absorption lines due to the warm absorber . such lines were neither discussed by turner et al . ( 2001 ) and turner et al . ( 2002 ) where they presented the data for the first time nor seen in _ xmm - newton _ data ( vaughna et al . 2002 ) because the lines are indeed very faint . we detected lines of highly ionized carbon and other elements , but we do not seem to detect any oxygen lines while in the fuse data ( turner et al . 2002 ) ovi narrow absorption line components were seen . five of those absorption lines detected in ton s180 ( cvi(@xmath182 kev ) , nex , mgxii(@xmath183 kev ) , mgxii(@xmath184 kev ) , and sixiv ) are also seen in another seyfert 1 galaxy - ngc 3783 analyzed by kaspi et al . ( 2002 ) , with similar values of ews ( 0.46 , 2.3 , 4.3 , 1.6 and 6.6 ev respectively ) . also kaastra et al . ( 2002 ) in ngc 5548 have found the same cv , cvi(@xmath182 kev ) , nex , mgxii(@xmath183 kev ) , and sixiv , with following ews - 0.26 , 0.99 , 0.47 , 3.09 , and 3.88 ev . we found a lower limit for the warm absorber column of @xmath70 @xmath6 from the equivalent with of the one of c vi lines ( see sect . [ sec : abs ] ) and an upper limit of @xmath185 @xmath6 from our model of the warm absorber ( see sect . [ sec : mod ] ) . the upper limit is not firm since it is sensitive to the adopted turbulent velocity as well as abundances . the _ chandra _ data show that the effect of the warm absorber does not modify significantly the x - ray spectrum in the case of ton s180 . therefore our analysis do not change the overall picture of the ton s180 continuum , as shown by turner et al . ( 2002 ) and discussed by vaughan et al . in particular , the huge soft x - ray excess is reasonably explained only within the frame of the comptonization model . variability arguments presented by vaughan et al . ( 2002 ) in favor of hybrid plasma model ( thermal plasma being responsible for the soft x - ray slope and non - thermal plasma component leading to hard x - ray power law ) are convincing . however , the data do not allow to determine all the hybrid plasma parameters uniquely , as explained by vaughan et al . ( 2002 ) , so in the present paper we have used much simpler model in order to draw attention to the possible constraints resulting from detection of the broad disk lines . those emission lines can be present in the broad band spectrum rather for higher black hole mass @xmath186 and accretion rate @xmath187 , since with those assumptions hot skin does not cover disk completely . this conclusion is opposite to the black hole mass estimations done for ton s180 by wang & lu ( 2001 ) , where @xmath188 . but our result is consistent with studies done by siemiginowska ( 1997 ) , where simple disk corona model was fitted to the _ beppo sax _ data ( those presented by comastri et al . we argue , contrary to the previous papers by turner et al . ( 2001 ) and vaughan et al . ( 2002 ) that spectral features in absorption and even in emission are present in the _ chandra _ data of ton s180 . the quality of the data does not allow us to determine column densities of individual metals , we can only put lower limit on column density of cvi . a few emission lines fitted to the ton s180 _ chandra _ data have lower equivalent widths by factor of 6 than those reported by mason et al . ( 2002 ) for seyfert i galaxy mrk 766 , but of similar order as disk lines found by kaastra et al . ( 2002 ) in ngc 5548 . estimation of the mass of the black hole in ton s180 based on fwhm of the @xmath2 line and the formula of kaspi et al . ( 2000 ) for the relation between the size of the broad line region and the object luminosity at 5100 rather suggests small black hole mass , @xmath190 ( wang & lu 2001 ) , in contradiction with our conclusion based on the detection of the soft x - ray lines . _ acknowledgments . _ we are grateful to suzy collin for discussion regarding spectral models of warm absorber computed by titan . we thank liz galle for help in processing the archival data and antonella fruscione and fabrizio nicastro for discussion . this work was supported in part by grant 2p03d00322 of the polish state committee for scientific research and by jumelage / cnrs no . 16 `` astronomie france / pologne '' . as acknowledge support through nasa contract nas8 - 39073(cxc ) ballantyne , d.r . , ross , r.r . , & fabian , a.c . , 2001 , , 327 , 10 ballantyne , d.r . , ross , r.r . , & fabian , a.c . , 2002 , 336 , 867 boller , t. , brandt , w.n . , fink , h. , 1996 , , 305 , 53 collin - souffrin , s. , czerny , b. , dumont , a .- m . , zycki , p. t. , 1996 , , 314 , 393 collin s. , coup , s. , dumont , a .- m . , petrucci , p .- o . , raska , a. , 2002 , ( accepted to ) collinge , m.j . , brandt , w.n . , kaspi , s. , crenshaw , d.m . , elvis , m. , kraemer , s.b . , reynolds , c.s . , sambruna , r.m . , wills , b.j . , 2001 , , 557 , 2 comastri a. , fiore , f. , guainazzi , m. , matt , g. , stirpe , g. m. , zamorani , g. , brandt , w. n. , leighly , k. m. , piro , l. , molendi , s. , parmar , a. n. , siemiginowska , a. , & puchnarewicz , e. m. , 1998 , 333 , 31 czerny , b. & dumont , a .- m . , 1998 , , 338 , 386 czerny , b. & elvis , m. , 1987 , , 321 , 305 czerny , b. & . zycki , p.t . , 1994 , , 431 , l5 dickey , j.m . & lockman , f.j . , 1990 , , 28 , 215 done , c. , mulchaey , j.s . , mushotzky , r.f . , arnaud , k.a . , 1992 , , 395 , 275 dumont , a .- m . , abrassart a. , & collin s. 2000 357 , 823 kaastra , j.s . , steenbrugge , k.c . , raassen , a.j.j . , van der meer , r.l.j . , brinkman , a.c . , liedahl , d.a . , behar , e. , de rosa , a. , 2002 , , 386 , 427 kaspi , s. , et al . 2000 , , 533 , 631 kaspi , s. , et al . , 2002 , , 574 , 643 kawagushi , t. , 2002 , ( submitted to ) kinkhabwala , a. , sako , m. , behar , e. , kahn , s. m. , paerels , f. , brinkman , a.c . , kaastra , j.s . , gu , m.f . , liedahl , d.a . , 2002 , , 575 , 732 krolik , j.h . , 2002 proceedings of the workshop : `` x - ray spectroscopy of agn with chandra and xmm - newton '' , eds . th . boller , s. komossa , s. kahn , h. kunieda krolik , j.h . , kriss , g.a . , 2001 , , 561 , 684 krolik , j.h . , mckee , c.f . , & tarter , c.b . , 1981 , 249 , 422 kurpiewski a. , kuraszkiewicz j. , czerny b. , 1997 , , 285 , 725 leighly , k.m . , zdziarski , a.a . , kawaguchi , t. , matsumoto , c. , 2002 proceedings of the workshop : `` x - ray spectroscopy of agn with chandra and xmm - newton '' , eds . boller , s. komossa , s. kahn , h. kunieda mihalas , d. 1978 , stellar atmospheres , w.h . freeman & co. , san francisco nayakshin , s. , kazanas , d. , & kalman , t. r. , 2000 , , 537 , 833 osterbrock , d.e . , pogge , r.w . , 1985 , , 297 , 166o puchnarewicz , e.m . , mason k.o . , siemiginowska a. , fruscione a. , comastri a. , fiore f. & cagoni i. , 2001 , , 550 , 644 romano p. , turner t.j . , mathur s. , & george i.m . , 2002 , , 564 , 162 raska a. , czerny b. , ycki p.t . , pojmanski g. , 1999 , , 305 , 481 raska , a. , dumont , a .- m . , czerny , b. , & collin , s. , 2002a , , 332 , 799 raska , a. , czerny , b. , dumont , a .- m . , collin , s. , siemiginowska , a. `` the radiation spectra of illuminated accretion disks in agn '' , active galactic nuclei : from central engine to host galaxy , 2002b , eds . : s. collin , f. combes , i. shlosman , asp conference series ( in press ) stark , a.a . , gammie , c.f . , wilson , r.w . , bally , j. , linke , r.a . , heiles , c . , & hurwitz , m. 1992 , , 79 , 77 sunyaev r.a . , titarchuk l.g . , 1980 , , 86 , 121 turner , t. j. , george , i. m. , & nandra , k , 1998 , , 508 , 648 turner , t. j. , george , i. m. , yaqoob , t. , kriss , g. , crenshaw , d. m. , kraemer , s. , zheng , w. , wang , j. , & nandra , k. , 2001 , , 548 , l13 turner , t. j. , romano , p. , kraemer , s. b. , george , i. m. , yaqoob , t. , crenshaw , d. m. , storm , j. , alloin , d. , lazzaro , d. , da silva , l. , pritchard , j. d. , kriss , g. , zheng , w. , mathur , s. , wang , j. , dobbie , p. , & collins , n. r. , 2002 , , 568 , 120 vaughan , s. , boller , th . , fabian , a.c . , ballantyne , d.r . , brandt , w.n . , tr.umper , j. , ( astro - ph/0207650 ) wang & lu 2001 , aa , 377 , 52 . wisotzki , l. , dreizler , s. , engels , d. , fink , h .- h . , & heber , u , 1995 , , 297 , l55 yaqoob , t. 2002 , ( submitted to )
by now , jump - diffusion processes and more general lvy processes or lvy driven processes are widely used in models of the stock dynamics and term structure models , under both historic and risk neutral measure . however , only discrete time series of empirical data are available , therefore the problem of identification of the jump and diffusion components is a non - trivial task . the problem of the identification of the diffusion component was considered in detail in ait - sahalia ( 2002 , 2003 ) ; in the present paper , the focus is on the jump component . clearly , the presence of jumps is most prominent in prices of contingent claims near expiry , especially for out - of - the money options , and therefore it is natural to extract information about the jump part of the process by using empirical data for out - of - the - money european options . carr and wu ( 2003 ) showed that in the presence of jumps , the prices of out - of - the - money european options near expiry are of order @xmath3 , where @xmath3 is the time to expiry , and use this observation to test for the presence of jumps . however , the results in carr and wu ( 2003 ) are qualitative rather than quantitative . the principal aim of the present paper is to derive simple analytical formulas which can be used for the identification of the jump part of a lvy process ( under a risk - neutral measure chosen by the market ) ; the straightforward generalizations to mean - reverting processes driven by lvy processes ( the ornstein - uhlenbeck type processes ) , stochastic volatility models with jumps and lvy - driven term structure models will be published elsewhere . the first group of results of the present paper are formulas for @xmath0 , or time decay , of out - of - the money european call and put options on a stock , at expiry . contrary to the gaussian case , @xmath0 for the former ( resp . , the latter ) is shown to be negative if there are positive ( resp . , negative ) jumps . in other words , we prove that near expiry , these prices are linear functions of time to expiry ( up to relatively small errors ) , and calculate the coefficients . the results are explicit analytical expressions in terms of the parameters which characterize the process . by using these formulas and empirical data for options near expiry as in carr and wu ( 2003 ) , it is possible to identify the jump part of a particular lvy model . after that , by using the data for options far from expiry , one can identify the first instantaneous moment of the process , and calculate the diffusion coefficient by subtracting the second instantaneous moment of the jump part of the process . this gives a new identification method for the diffusion component . by using the formulas for the @xmath0 s for the european put and call , we obtain approximations for the american put price and early exercise boundary near expiry . the pricing of american options in the gaussian case is well understood see e.g. musiela and rutkowski ( 1997 ) , karatzas and shreve ( 1998 ) , carr and faguet ( 1994 ) , carr ( 1998 ) , longstaff and schwartz ( 2001 ) , clment et al . ( 2002 ) , and the bibliography therein . in the infinite horizon case , explicit analytical formulas are obtained , and in the finite horizon case , a number of efficient numerical methods are developed . although explicit formulas are not available in the finite horizon case , several general results are proved for this case as well . consider the american put with the strike price @xmath4 and maturity @xmath5 , on a non - dividend paying stock ; the riskless rate @xmath6 is constant . let @xmath7 be the optimal exercise price of the american put . if the stock log - price @xmath8 follows the brownian motion , then the asymptotic behavior of @xmath9 is known both near expiry : @xmath10 as @xmath11 ( see , e.g. , barles _ ( 1995 ) and lamberton ( 1995 ) ; an asymptotic formula for the put price is also available ) , and for small values of @xmath12 : for any @xmath13 , @xmath14 ( see , e.g. musiela and rutkowski ( 1997 ) ; ( [ asymprgauss ] ) is formulated as follows : if the riskless interest rate is zero , then the early exercise of the american put is not optimal ) . notice that ( [ asbgauss ] ) implies that @xmath15 and that ( [ exlim ] ) and ( [ asymprgauss ] ) do not agree well when both the riskless interest rate and time to expiry vanish . in levendorski ( 2003 ) , a variant of carr s randomization method was developed for wide families of lvy processes , and as a by - product , it was shown that for many families of non - gaussian processes used in empirical studies of financial markets , the analogs of ( [ exlim ] ) and ( [ asymprgauss ] ) agree much better . namely , @xmath16 where @xmath17 depends on some of parameters which characterize the price process , and is _ smaller _ than the strike price , @xmath4 . moreover , it was proved that @xmath17 vanishes with @xmath12 , and therefore , @xmath18 notice that ( [ asymprlevy ] ) looks more natural than ( [ asymprgauss ] ) , and may be of some importance in today s almost deflationary world . further , if the difference @xmath19 is sizable , then any numerical procedure or parameters fitting procedure which does not take this fact into account , may produce significant errors near expiry and strike . hence , the non - standard behavior of the early exercise boundary has practical implications for the parameters fitting and development of accurate numerical methods . however , the proof in levendorski ( 2003 ) has three disadvantages , which stem from the use of the wiener - hopf method . first , the non - standard behavior of the early exercise boundary results from long calculations without clear intuition . next , the approximate formulas for the put price are expressed in terms of the factors in the wiener - hopf factorization formula , hence , they are complicated for practical use , and finally , the method in levendorski ( 2003 ) is not directly applicable to a popular class of variance gamma processes , which was used in carr et al ( 2002 ) , madan and hirsa ( 2003 ) , and carr and hirsa ( 2003 ) for pricing of american put . the generalization of the method for more general classes of processes seems impossible . here we prove ( [ exlimlevy ] ) and ( [ asymprlevy ] ) by using a much simpler argument . we obtain an efficient approximation for the put price near expiry , and formulate the result about the non - standard behavior of the early exercise boundary in a more meaningful form . also , we extend the result in levendorski ( 2003 ) to variance gamma processes , and in addition , we find a condition on a process which ensures that the behavior of the boundary is even more regular than in the gaussian case : there exists a negative constant @xmath20 s.t . @xmath21 we prove that the limit of the boundary at expiry is located where @xmath22 of the european call is equal to the riskless rate . ( if the @xmath22 of the european call at expiry is less than the riskless rate for all spot prices less than the strike , then the limit is the strike . ) when the corresponding equation is solved , one can use the formulas for @xmath0 s of the european call and put and obtain an efficient approximation to the put price near expiry . the formulas are fairly simple and can be used for parameters fitting purposes near expiry , both for the european options and american put . they also can be used to improve performance of numerical methods : when using the backward induction , one can start not at expiry , where the value function is most irregular but close to it , and use the analytical approximation as the basis for the backward induction . we demonstrate how large the deviations from the standard behavior of the early exercise boundary and put price can be by using the empirical data for the american put documented in carr et al . we found that the non - standard behavior should be observed for reasonable values of the riskless rate : typically , the margin between the strike and boundary is 10% or more , and for the spot price in the 5%interval around the strike , the jump premium " in the put price over the pay - off is @xmath23 or more , where @xmath3 is the time to expiry ( with the normalization @xmath24 for a business year ) . the last aim of the paper is to describe the restrictions on the log - price process under an emm , which should be satisfied if the non - standard behavior of the early exercise boundary is observed . as ( [ exlim ] ) , ( [ exlimlevy ] ) and ( [ superregular ] ) imply , various types of the behavior of @xmath9 near expiry are possible however , for many families of processes used in empirical studies of financial markets , namely , hyperbolic processes , normal inverse gaussian processes ( nig ) , and truncated lvy processes of the extended koponen s family ( a.k.a . kobol processes , a.k.a . cgmy model ) of order @xmath25 , the behavior of the early exercise boundary is non - standard for all parameters values . this means that if in the empirical studies , the early exercise boundary for put options on a particular stock exhibits the standard behavior near expiry , then the use of any of the processes above to model the log - price dynamics under a risk - neutral measure chosen by the market is inconsistent with the data . normal inverse gaussian processes , hyperbolic processes and kobol processes of order greater than 1 are processes with sample paths of infinite variation . notice that the non - standard behavior of the early exercise boundary is proved for much wider classes of infinite variation lvy processes as well . for processes with finite variation jump component , the prominent examples being variance gamma processes and kobol processes of order less than 1 , both the non - standard behavior ( [ exlimlevy ] ) and super - regular " behavior ( [ superregular ] ) of the early exercise boundary are possible , depending on parameters of the process ; there are also borderline cases , when we are able only to prove that ( [ exlim ] ) holds . according to a recent empirical study in carr et al . ( 2002 ) , finite variation processes fit empirical data better . even for a finite variation process , if we fix all the parameters of a process except for the drift @xmath26 , let @xmath27 and change @xmath26 accordingly so that the stock remains a martingale , then below a certain critical value of @xmath12 ( depending on the parameters of the process ) , the non - standard behavior is observed , and as @xmath12 decreases further , the optimal exercise price goes to zero uniformly over the interval @xmath2 . it is natural to ask how robust the result about the non - standard behavior of the early exercise boundary is . in levendorski ( 2003 ) and the present paper , it is assumed that dynamics of log - prices @xmath8 follows a non - gaussian lvy process under a risk - neutral measure chosen by the market . this simplest choice of a non - gaussian risk - neutral measure for pricing of the american put was employed in carr et al ( 2002 ) , madan and hirsa ( 2003 ) , carr and hirsa ( 2003 ) , where parameters fitting and computational issues where addressed . better fit can be achieved with stochastic volatility models with jumps but the method in levendorski ( 2003 ) can not be generalized for stochastic volatility models . the method of the present paper can be applied to stochastic volatility models ( and term structure models ) , and under certain parameter restrictions , the margin between the early exercise boundary and strike can be derived . further , in levendorski ( 2003 ) and the present paper , the formulas for the margin are derived in two different approximate models , and in cases which were considered in the both papers , the formulas for the margin are the same . the identical results in two approximations allows one to hope that the margin in the continuous time model is the same . as a proven theorem , we can only claim that the margin in the approximate model is not less than in the continuous time model . the rest of the paper is organized as follows . in section [ levy ] , we recall the definition of lvy processes , and give examples of several families of lvy processes used in empirical studies of financial markets . in section [ europe ] , @xmath0 s of out - of - the - money european put and call options are calculated , and in section [ amerput ] , the approximate formulas for the american put price and early exercise boundary near expiry are derived . in section [ implications ] , we consider several families of lvy processes used ( and suggested for use ) in empirical studies of financial markets , and discuss the implications of non - standard behavior of prices and early exercise boundary for the choice of a family of processes , and parameters fitting in more detail . in section [ empirics ] , we derive properties of the early exercise boundary for kobol processes , for parameters values documented in carr et al . ( 2002 ) , and show that the margin between the strike price and early exercise boundary can be significant indeed : more than 10% . in the appendix , the technical proofs are given . recall that a lvy process is a process with stationary independent increments ( for general definitions , see e.g. sato ( 1999 ) ) . a lvy process may have a gaussian component and/or pure jump component . the latter is characterized by the density of jumps , which is called the lvy density . we denote it by @xmath28 . also , a lvy process can be completely specified by its characteristic exponent , @xmath29 , definable from the equality @xmath30=e^{-t\psi(\xi)}$ ] ( we confine ourselves to the one - dimensional case ) . if @xmath31 is a lvy process with finite variation jump component , then the characteristic exponent is given by @xmath32 where @xmath33 and @xmath26 are the variance and drift coefficient of the gaussian component , and @xmath34 satisfies @xmath35 equation ( [ psi1 ] ) is a special case of the lvy - khintchine formula ; for the general case , see e.g. sato ( 1999 ) . wide families of jump - diffusion processes used in the theoretical and empirical studies of financial markets are lvy processes with finite variation jump component . 0.1cm_example 2.1 . _ let @xmath36 be a lvy process with the lvy density @xmath37 where @xmath38 , and @xmath39 . then @xmath40 where @xmath41 and @xmath42 are the variance and drift of the gaussian component . the @xmath43 is holomorphic in the strip @xmath44 . in order that @xmath45 $ ] be finite , we need to impose an additional condition @xmath46 . for the sake of brevity , we will consider below the exponential jump - diffusions with the characteristic exponent of the form ( [ jumpdif ] ) . the generalization to the case of lvy densities given by general exponential polynomials on positive and negative half - axis is straightforward . 0.1cm_example 2.2 . _ variance gamma processes ( vgp ) have been developed and used by madan and co - authors in a series of papers during 90th ( see madan et al . ( 1998 ) and the bibliography there ) . the characteristic exponent of a vgp without the diffusion component can be represented in the form @xmath47,\ ] ] where @xmath48 and @xmath49 . the condition @xmath45<+\infty$ ] imposes an additional restriction @xmath46 . 0.1cm_example 2.3 . _ truncated lvy processes ( tlp ) constructed by koponen ( 1995 ) were used for modeling in real financial markets in bouchaud and potters ( 1997 ) , cont et al . ( 1997 ) and matacz ( 2000 ) ; a generalization of this family was constructed in boyarchenko and levendorski ( 1999 , 2000 ) and called the extended koponen family of tlp processes . later , this generalization was used in carr et al . ( 2002 ) under the name cgmy - model . as a.n . shiryaev remarked , the name tlp was misleading , and so starting with boyarchenko and levendorski ( 2002a-2002c ) , we use the name kobol processes . the characteristic exponent of a kobol process is of the form @xmath50,\ ] ] where @xmath51 , and @xmath42 . the condition @xmath45<+\infty$ ] imposes an additional restriction @xmath52 . ( in boyarchenko and levendorski ( 1999 , 2000 , 2002b ) , the reader can also find a formula for the kobol process of order @xmath53 ) . for @xmath54 , the equation ( [ kbl1d ] ) is obtained from ( [ psi1 ] ) with the lvy density given by @xmath55 and @xmath56 ( that is , there is no gaussian component ) . in the case @xmath57 , instead of ( [ psi1 ] ) , the general form of the lvy - khintchine formula is needed ( see sato ( 1999 ) and boyarchenko and levendorski ( 2002b ) ) . kobol processes of order @xmath58 are finite - variation processes , and the ones of order @xmath59 are infinite - variation processes . other examples of infinite variation processes are hyperbolic processes ( hp ) , normal inverse gaussian processes ( nig ) and normal tempered stable ( nts ) lvy processes of order @xmath60 ; in the general classification of boyarchenko and levendorski ( 1999 , 2000 , 2002a-2002c ) , hp and nig are processes of order 1 . hyperbolic processes were constructed and used by eberlein and co - authors ( see e.g. eberlein and keller ( 1995 ) , eberlein et al . ( 1998 ) , eberlein and raible ( 1999 ) ) ; hyperbolic distributions were constructed in barndorff - nielsen ( 1977 ) . normal inverse gaussian processes ( nig ) were introduced in barndorff - nielsen ( 1998 ) and used to model german stocks in barndorff - nielsen and jiang ( 1998 ) ; generalization of the class nig , namely , the class of normal tempered stable ( nts ) lvy processes , was constructed in barndorff - nielsen and levendorski ( 2001 ) and barndorff - nielsen and shephard ( 2001 ) . as kobol processes , ntslp can be of any order between 0 and 2 . 0.1 cm _ example 2.4 . _ the characteristic exponent of an nts lvy process is of the form @xmath61,\ ] ] where @xmath62 @xmath63 , and @xmath64 . the condition @xmath45<+\infty$ ] imposes an additional restriction @xmath65 . with @xmath53 , we obtain the characteristic exponent of nig processes . 0.1cm_remark 2.1 . _ for the explicit calculations in the next two sections , it is important that the characteristic exponent of any of lvy processes considered above admit the analytic continuation into the complex plane with cuts @xmath66 , [ i{{\lambda_+ } } , + i\infty)$ ] , where @xmath52 and @xmath67 , and the analytic continuation is defined by the same formula ( in example 2.4 , @xmath68 and @xmath69 , and in example 2.1 , the characteristic exponent is analytic in the complex plane with two poles at @xmath70 and @xmath71 ) . this property is unnecessary if we want to obtain only qualitative results . for instance , the existence of a non - vanishing margin between the strike and early exercise boundary will be proved for more general class of lvy processes with the jump part of infinite variation . to simplify the presentation , we normalize the strike price to 1 . let @xmath31 be a lvy process of any class considered above . let @xmath72 and @xmath73 be the price of the european call and put option , respectively , at time @xmath74 , and the spot price @xmath75 . denote by @xmath76 , and @xmath77 , the opposite to the @xmath78 of the out - of - the - money european call and put options , respectively , at expiry ( the option s @xmath0 is the derivative of the option price with respect to time ) : for @xmath79 , @xmath80 and for @xmath81 , @xmath82 if the process is gaussian , then the prices of out - of - the - money european call , @xmath83 , and put , @xmath84 , vanish faster than any power of @xmath3 , as @xmath85 , hence the limits ( [ cc ] ) and ( [ cp ] ) equal 0 . the next theorem and lemma show that in the case of non - gaussian lvy processes which we consider , both @xmath86 and @xmath77 , exist and do not vanish . the intuition is that when the jumps are present , the value of waiting of a positive ( respectively , negative ) movement in the price is larger than in the gaussian case . [ margin1 ] let @xmath31 be any of lvy processes considered above . then 0.cm a ) for @xmath79 , the limit ( [ cc ] ) exists , and it is positive ; 0.cmb ) for @xmath81 , the limit ( [ cp ] ) exists , and it is positive ; 0.cmc ) the time decay at expiry , of out - of - the money european call , @xmath87 @xmath88 and put , @xmath77 , depends only on the positive and negative jump parts of the process , respectively , but not on the drift and gaussian component . notice that the intuition behind the independence of the time decay at expiry of the gaussian component is simple : the movements caused by the latter are of order @xmath89 , on average , and therefore the main contribution to the price of an out - of - the - money option , ( almost ) at expiry , can come from the jumps in the corresponding direction only ( for a rigorous proof , see the appendix ) . we start with the simplest case of a pure jump process ( example 2.1 with @xmath56 ) . let @xmath90 be the infinitesimal generator of @xmath36 . then the price @xmath91 of a contingent claim with the terminal pay - off @xmath92 can be represented in the form @xmath93g(x+\mu \tau).\ ] ] for the pure jump process , @xmath90 acts as follows : @xmath94 and since @xmath95 for processes in example 1.1 , we conclude that @xmath90 is a bounded operator in @xmath96 . hence , if @xmath97 is bounded ( the case of the put option : @xmath98 ) , we can write ( [ europclaim ] ) in the form @xmath99 if the put option is out - of - the - money , that is , @xmath81 , then for small @xmath3 , @xmath100 , hence the zero - order term in ( [ europclaim2 ] ) is zero , and @xmath101 by dividing by @xmath102 and passing to the limit , we obtain @xmath103 it remains to calculate @xmath104 at @xmath81 . we have @xmath105 and by calculating the integral , we obtain @xmath106 the case of the call option requires essentially the same calculations but @xmath90 should be regarded as a bounded operator in a space of continuous functions , which grow as @xmath107 as @xmath108 ( for details , see boyarchenko and levendorski ( 2002b ) ) , and the result is @xmath109 similarly , @xmath110 and @xmath111 for more general exponential jump - diffusions can be calculated . moreover , the proof above gives @xmath112 and @xmath113 provided @xmath114 and @xmath115 ( the latter condition is necessary for the stock to be priced ) . we conjecture that ( [ ccgen ] ) and ( [ cpgen ] ) are valid for any lvy process satisfying ( [ denspos1 ] ) . however , we were unable to prove ( [ cc])([cp ] ) in the full generality , and so in the proof of theorem [ margin1 ] for the other families of lvy processes listed above , we calculate the limits @xmath110 and @xmath111 directly , by using the formulas for the characteristic exponents . in the case of nts lvy processes and nig , set @xmath116 . now , for any @xmath31 of kobol , nts , and vgp classes , and for @xmath117 and @xmath118 , define @xmath119.\ ] ] [ gen ] let @xmath31 be a vgp , kobol , or nts lvy process . then for @xmath79 , @xmath120 and for @xmath81 , @xmath121 in the appendix . in the gaussian case , @xmath122 , and since @xmath123 depends linearly on @xmath29 , we conclude that @xmath123 depends on the jump component of the process only . hence , @xmath76 , and @xmath77 , depend on the jump component only . the fact that the former depends on the positive jumps only , and the latter on the negative ones , follows by inspection of the lvy - khintchine formula : for @xmath118 ( resp . , @xmath117 ) , @xmath124 is independent of the density of negative ( resp . , positive ) jumps . in the appendix , we will calculate @xmath124 for each family of processes , and derive from ( [ genc])([genp ] ) the following formulas : \1 ) for kobol processes of order @xmath125 : @xmath126 the integrals converge since @xmath127 and @xmath128 ; \2 ) for vgp : @xmath129 the integrals converge since @xmath130 ; \3 ) for nts lvy processes @xmath131^{\nu/2}}{z(z+1)}dz,\quad x<0;\\ \label{cpnts } { { \mathcal p}}(x)&=&\frac{{\delta}}{\pi}\sin\frac{\pi\nu}{2}\int^{+\infty}_{{\alpha}+{\beta } } \frac{e^{-xz}[(z-{\beta})^2-{\alpha}^2]^{\nu/2}}{z(z+1)}dz,\quad x>0;\end{aligned}\ ] ] with @xmath53 , formulas for nig processes obtain . the integrals converge since @xmath132 , @xmath133 , and @xmath128 . by direct inspection of formulas ( [ cckbl])([cpnts ] ) , we conclude that with @xmath134 , the integrands decay as @xmath135 as @xmath136 for processes of order @xmath137 ; if @xmath31 is a vgp or exponential jump - diffusion , then the integrands decay faster than @xmath138 , for any @xmath139 . therefore , with @xmath134 , the integrals diverge if and only if the jump component of @xmath31 is a process of order @xmath140 . further , the integrands are positive monotone continuous functions of @xmath141 . by using the monotone convergence theorem , we deduce [ margin2 ] a ) for processes of order @xmath58 , vgp , and exponential jump - diffusion processes in example 2.1 , there exist finite limits @xmath142 \b ) for nig , and kobol and nts lvy processes of order @xmath143 , the limits @xmath144 and @xmath145 exist but are infinite . \c ) @xmath146 is an increasing continuous function , which maps @xmath147 onto @xmath148 , and @xmath149 is a decreasing continuous function , which maps @xmath150 onto @xmath151 . _ remark 3.1 . _ a ) it is possible to derive formulas for @xmath86 and @xmath77 , for the case of hyperbolic processes , and that part b ) of theorem [ margin2 ] holds in this case . b ) can be proved for more general classes of lvy processes . for instance , @xmath152 can be proved if in a neighborhood of 0 , the density of positive jumps , @xmath153 , admits a lower bound via @xmath154 , where @xmath155 . then there exists a density @xmath156 such that @xmath157 and @xmath158 where @xmath159 . clearly , @xmath110 can not increase if we replace @xmath153 with @xmath160 . the argument in the proof of theorem [ margin1 ] for the jump - diffusion case shows that the contribution of @xmath156 to @xmath110 is bounded uniformly in @xmath79 , and the proof for kobol processes shows that @xmath161 gives a contribution which is unbounded as @xmath162 . 0.1 cm to finish this section , we list approximate formulas for european call and put options near expiry , which follow from theorem [ margin1 ] and the put - call pairity . of course , these approximations are too inaccurate in a very small neighborhood of the strike if the process is of order @xmath163 . for @xmath81 : @xmath164 for @xmath79 : @xmath165 assume that @xmath144 and @xmath145 exist . they are measures of the influence of positive and negative jumps , respectively , and the emm condition is naturally formulated in terms of @xmath166 and @xmath144 and @xmath145 . as it was shown above , the limits ( finite or infinite ) exist for the families of processes in examples 2.12.4 . [ emm ] let @xmath31 be a lvy process with finite variation jump component . then @xmath167 is a local martingale if and only if @xmath168 since the gaussian component @xmath169 and jump component @xmath170 of a lvy process are independent , @xmath167 is a local martingale if and only if for any @xmath171 and @xmath172 , @xmath173e\left[e^{y_t}\right]\\ & = & e^{(\mu - r+{\sigma}^2/2)t+x_s}e\left[e^{y_t}\right ] , \end{aligned}\ ] ] whereupon @xmath174,\ ] ] and @xmath175 - 1,\ ] ] which is @xmath176 ( here @xmath177 denotes the transition kernel of the semigroup @xmath178 , @xmath179=\int_{-\infty}^{+\infty}p_t(z)f(y+z)dz.\right)\ ] ] divide by @xmath180 : @xmath181 and pass to the limit @xmath182 ; the result is ( [ emm1 ] ) . we rewrite ( [ emm1 ] ) as @xmath183 for processes without the gaussian component , ( [ emm2 ] ) is simpler : @xmath184 the first theorem below shows that if negative jumps are present but their influence is not very large , and compensated by a positive drift ( which clearly decreases the value of waiting ) , then the early exercise boundary converges to the strike price . moreover , the convergence is more regular than in the gaussian case in the sense that ( [ superregular ] ) holds . let @xmath185 be the early exercise boundary . the continuous time model can be approximated by a discrete time model with small time steps @xmath186 . in the discrete time approximation , the option can be exercised at @xmath187 only . denote by @xmath188 the optimal exercise boundary at the last moment before the expiry , and set @xmath189 . one should expect that the behavior of @xmath190 for small @xmath191 is a good proxy for the behavior of the optimal exercise boundary near expiry . in particular , @xmath192 if the limit in the lhs exists . the proof of a simplified version @xmath193 is straightforward : as we make the class of admissible stopping times smaller , the price of the american put does not increase , therefore the early exercise boundary does not go down . we were unable to prove the equality ( [ just ] ) but the inequality ( [ just2 ] ) suffices to make the conclusion that if the early exercise boundary in the discrete time approximation exhibits the non - standard behavior in the limit @xmath194 , then the behavior of the early exercise boundary in the continuous time model is also non - standard . [ extrastan ] let @xmath31 be a vgp or kobol process of finite variation , without gaussian component , or an exponential jump process . let @xmath195 then @xmath196 in other words , _ if @xmath22 of the out - of - the - money european put , at the expiry and strike , is less than the drift , then the slope of the boundary near expiry equals the @xmath22 . _ at @xmath197 , a put owner must be indifferent between exercising and holding the put , therefore @xmath190 is the solution to the equation @xmath198 ( the lhs is the value of the put if it is exercised , and the rhs is the expected present value of the put if it is kept alive ) , equivalently , @xmath199 assume that for small @xmath200 , @xmath201 , then @xmath202 , hence @xmath203 . by using the taylor formula and dividing by @xmath191 , we obtain @xmath204 then we pass to the limit @xmath205 @xmath206 condition @xmath207 is necessary lest the assumption @xmath208 lead to a contradiction ; and the condition ( [ cneg ] ) suffices for the inequality @xmath208 to hold for small @xmath191 . now assume that ( [ cneg ] ) fails . this happens when either the drift is negative or the influence of negative jumps is too large . hence , the value of waiting should increase , and the optimal exercise boundary should be lower . as we will see , apart from the borderline case @xmath209 , the value of waiting becomes so large that @xmath190 remains bounded away from 0 , as @xmath205 . [ thm : finvar ] let @xmath31 be a vgp or kobol process of finite variation , without gaussian component , or an exponential jump process . then a ) if @xmath209 , then the limit @xmath210 exists , and it is equal to 0 ; \b ) if @xmath211 then the limit ( [ barh ] ) exists . it is negative , and can be found from the equation @xmath212 which has a unique solution . in other words , _ if @xmath22 of the out - of - the - money european put , at the expiry and strike , is bigger than the drift , then the limit of the early exercise boundary at expiry is less than the strike . it is located where @xmath22 of the out - of - the - money call , at expiry , is equal to the riskless rate . _ we start with part b ) . first of all , we notice that due to ( [ emm3 ] ) , ( [ cneg2 ] ) is equivalent to @xmath213 therefore on the strength of theorem [ margin2 ] , c ) , the solution @xmath214 to ( [ formpos1 ] ) exists and it is unique . secondly , the reader may be surprised that the answer is formulated in terms of the density of positive jumps , but the proof shows that this is a consequence of the put - call pairity ( for european options ) . since @xmath215 is a ( local ) martingale under the risk - neutral measure , we have @xmath216 hence , we can rewrite the rhs in ( [ exb1 ] ) as @xmath217 ( essentially , this is the put - call pairity ) , and therefore , ( [ exb1 ] ) can be written as @xmath218 we see that if @xmath219 is negative for small @xmath191 , then @xmath220 depends only on the upper tail of the probability density . by dividing by @xmath191 and passing to the limit , we obtain ( [ formpos1 ] ) . thus , part b ) has been proved . in part a ) , @xmath209 , hence from ( [ emm1 ] ) , @xmath221 , and if we assume that for some sequence @xmath222 , the sequence @xmath223 remains bounded away from 0 , then by dividing by @xmath224 in ( [ exb2 ] ) and passing to the limit , we obtain an inequality , contradiction . in the following theorem , we allow for a gaussian component . [ gauss ] let @xmath31 be a vgp or kobol of order @xmath58 , or exponential jump - diffusion process , with a gaussian component . then a ) if @xmath225 , then the limit @xmath214 exists . it is negative , and can be found from ( [ formpos1 ] ) ; \b ) if @xmath226 , then the limit @xmath214 exists , and it is equal to 0 . by repeating the proof of theorem [ thm : finvar ] , we obtain a ) , and similar arguments show that under condition @xmath226 , an assumption that the behavior of the boundary is non - standard , leads to contradiction . this gives b ) . we see that in the case @xmath226 , the result is weaker than in the pure jump case , where we can single out the case when the boundary is more regular than in the gaussian case . according to part b ) of theorem [ margin2 ] , now @xmath227 as @xmath228 . by using this observation , and repeating the proof of part a ) of theorem [ thm : finvar ] , we obtain [ thm : infvar ] let @xmath31 be a nig or kobol or ntsl process of infinite variation . then the limit @xmath229 exists . it is negative , and can be found from the equation ( [ formpos1 ] ) . _ remark 4.1 . _ if the density of positive jumps satisfies the conditions in remark 3.1 b ) , then @xmath152 , and repeating the proof of theorem [ thm : finvar ] , we obtain that @xmath214 is negative . let @xmath31 be a process from any of families in examples 2.12.4 . let @xmath230 but the lvy density and the diffusion coefficient of the gaussian component remain fixed . then the drift of the process changes with @xmath12 and converges to a finite value . by the direct inspection of ( [ formpos1 ] ) , we conclude that if the early exercise boundary exhibits the non - gaussian behavior ( [ exlimlevy ] ) near expiry , then the early exercise price near expiry vanishes with @xmath12 as well . the same argument as in the gaussian case shows that for any @xmath102 , @xmath231 as @xmath230 ( see ( [ asymprgauss ] ) ) , therefore we conclude that the optimal exercise price tends to zero uniformly on the interval @xmath2 . notice that if @xmath31 is a vgp or a kobol process of order @xmath232 , or exponential jump - diffusion ( as in example 2.1 ) , then for large @xmath12 , ( [ formpos1 ] ) has no solution , but for sufficiently small @xmath12 , equation ( [ formpos1 ] ) has a unique solution , and this solution tends to 0 together with @xmath12 . this means that for sufficiently low levels of the riskless rate , all the processes considered above should lead to the non - standard behavior of the early exercise boundary . by using the approximate formulas for the european put and call near expiry at the end of section 3 , we obtain approximate formulas for the american put price near expiry : @xmath233 notice that this approximation is too inaccurate very close to the strike , especially for infinite variation processes . the behavior of the early exercise boundary near expiry can be used to determine the type of a lvy process , which can be used to model the log - price dynamics under the risk - neutral measure . this choice of a risk - neutral measure was used in carr et al . ( 2002 ) , hirsa and madan ( 2003 ) and carr and hirsa ( 2003 ) . empirical studies of financial markets clearly indicate the presence of both downward and upward jumps ( under historic measure ) , therefore if the boundary exhibits the standard behavior , then one may not use simultaneously a process of order @xmath234 under the historic measure and lvy process under a risk - neutral measure ( chosen by the market ) . indeed , as it is shown in carr et al . ( 2002 ) , p.312 , the difference between the historic and risk - neutral lvy densities must be of finite variation , therefore a process of infinite ( finite ) variation under the historic measure must remain a process of infinite ( finite ) variation under an emm . processes of order @xmath235 can be used for modelling of log - price dynamics both under historic and risk - neutral measure but additional restrictions on the density of the upward jumps must be imposed . another possibility is to use one - sided kobol processes with densities of positive jumps of an order different from the one for positive jumps ( see boyarchenko and levendorski ( 1999 , 2000 , 2002b ) ) . the characteristic exponents of these strongly asymmetric kobol processes are of the form @xmath236\\ & & \hskip1 cm + c_-\gamma(-\nu_-)[(-{{\lambda_-}})^{\nu_-}-(-{{\lambda_-}}-i\xi)^{\nu_-}],\end{aligned}\ ] ] where @xmath237 , @xmath238 , and @xmath239 ( for the formulas in the case @xmath240 , see boyarchenko and levendorski ( 1999 , 2000 , 2002b ) ) . the analogs of ( [ cckbl ] ) and ( [ cpkbl ] ) are @xmath241 if a finite variation process is chosen for modelling of the american put on a particular stock , then the choice of parameters of the process is constrained by the observed behavior of the early exercise boundary : a. if the non - standard behavior ( [ exlimlevy ] ) is observed , then the constraint is @xmath242 ; b. if the super - regular " behavior ( [ superregular ] ) is observed , then the constraint is @xmath243 ( and probably , there is no gaussian component ) ; however , as empirical examples in the next section show , this possibility is unlikely to realize ; c. if @xmath244 but the super - regular " behavior ( [ superregular ] ) is not documented , then the constraint is @xmath245 , and the gaussian component is admissible . still another possibility is to assume that under an emm chosen by the market , the rate of decay of the density of positive jumps is very large ; then @xmath244 for processes of order @xmath235 , and @xmath214 is negative but very close to 0 for processes of order @xmath246 . for kobol , this means that the steepness parameter @xmath247 must be large in modulus . if under the historic measure , the @xmath247 is not large , this interpretation presumes that the agents in the market are very risk averse . the formulas for @xmath0 of out - of - the money european call and put options , at expiry , can also be used to infer the type of an appropriate family of processes , and parameters of the process . in particular , if @xmath0 at expiry quickly grows as the spot price approaches the strike price , then processes with infinite variation jump component must be used , otherwise the processes with finite variation jump component . the aim of this section is to demonstrate how large the margin between the strike and early exercise boundary and the jump premium " in the american option price over the payoff @xmath248 can be for realistic values of parameters . we use risk - neutral parameters values of the kobol processes with a diffusion component , which were obtained for several stocks in carr et al . ( 2002 ) ( table 3 on p.327 in op . cit . ) . 0.1 cm _ example 5.1 . _ only in one case : ibm1111 , the process is of order @xmath249 : the risk - neutral parameters are @xmath250 , and @xmath251 ( in carr et al . , these parameters are denoted @xmath252 , and @xmath253 ) . since @xmath249 , the non - standard behavior of the early exercise boundary is guaranteed but the gap between the strike price and boundary is small unless the interest rate is very small indeed . it is seen from figure 1 , where we plot the graph of @xmath214 as a function of @xmath12 , that for @xmath254 , @xmath214 is about @xmath255 , and for @xmath256 , @xmath214 is close to 0.01 , so the gap is clearly seen . in figure 2 , we plot the graph of @xmath0 at expiry , for out - of - the - money european call and put ( graph of @xmath110 , for small @xmath79 near zero , and put , @xmath111 , for @xmath81 ) . both are unbounded in a neighborhood of @xmath257 . since the density of positive jumps decays very fast ( the steepness parameter @xmath258 is very large ) , the time decay of out - of - the - money puts is very small unless the spot price is very close to the strike . the density of positive jumps decays faster than the one of negative jumps ( @xmath259 ) , therefore the @xmath0 for the out - of - the - money call is smaller than the one for out - of - the - money put , for the same distance from the strike . 0.1 cm _ example 5.2 . _ for amzn1014 , the parameters are @xmath260 the fairly large @xmath261 indicates that the intensity of jumps is large , and relatively small values of @xmath262 and @xmath247 imply that there are many large jumps , especially negative ones ( recall that @xmath262 characterizes the rate of decay of the density of negative jumps ) . therefore , even not very close to expiry , the owner of an out - of - the - money european option can expect that with a non - negligible probability the option will become in - the - money . hence , the price of these out - of - the - money options is sizable , especially the one of the put . figure 3 demonstrates this effect . we plot graphs of @xmath146 and @xmath149 in a neighborhood of zero . further , @xmath263 , therefore for reasonable @xmath12 , we have @xmath242 , and the behavior of the early exercise boundary is non - standard . since the intensity of large negative jumps is significant , the value of waiting is very large , and the gap between the strike and optimal exercise boundary is also large . from ( [ amputapprox ] ) and figure 3 , we can conclude that even at 5% below the strike , the option value of waiting ( the difference between the american put price and the pay - off @xmath248 ) is of order @xmath23 ; @xmath24 is one year . in the next two figures 4 and 5 , we plot the graph of @xmath146 farther from the strike , and the graph of @xmath214 as the function of @xmath12 . in figure 6 , we plot both the early exercise boundary , at expiry , @xmath264 , as a function of @xmath12 , and that of the early exercise boundary far from expiry , @xmath265 ( the early exercise boundary in the infinite time horizon case ) . similar results are valid for the other stocks documented in carr et al . for several families of lvy processes used in empirical studies of financial markets , we have proved that the time decay of out - of - the - money european options at expiry is not negligible , and derived explicit formulas . these formulas can be used to make numerical calculations near expiry more accurate both for european options , and american ones . by using the same formulas , we proved that the optimal exercise boundary for the american put is separated from the strike price by a non - vanishing margin on the interval @xmath2 , and that as the riskless rate vanishes , the optimal exercise price goes to zero uniformly over the interval @xmath2 . the last result may be especially interesting since the current levels of interest rates are rather low . furthermore , we discussed the restrictions on the process under the risk - neutral measure which the non - gaussian behavior of the early exercise price near expiry imposes , and suggested a natural asymmetric extension of the kobol family ( a.k.a . cgmy model ) , which can be used for any type of the behavior of the boundary . we showed that for the risk - neutral parameters obtained for several stocks in carr et al . ( 2002 ) , the behavior should be non - standard , and the margin between the strike and the early exercise boundary be quite sizable : more than 10% , up to expiry . the jump premium over the payoff is shown to be sizable as well . further , we assume that @xmath46 ; the results for the case @xmath266 obtain by passing to the limit @xmath267 in the formulas for the case @xmath46 . for any @xmath268 , the price of the european call option is given by @xmath269 } { -i\xi(1-i\xi)}d\xi\ ] ] ( see e.g. boyarchenko and levendorski ( 1999 , 2002b ) ) . clearly , the limit @xmath110 is independent of @xmath12 , therefore we may set @xmath270 . the characteristic exponent of a process without the gaussian component is of the form @xmath271 where @xmath272 grows as @xmath273 in the complex plane with the cuts @xmath66 $ ] and @xmath274 , for processes of order @xmath137 . in the complex plane outside two poles @xmath275 . for vgp , @xmath272 grows as @xmath276 , and for exponential jump processes , @xmath272 decays as @xmath277 . set @xmath278 , and write ( [ callprice ] ) as @xmath279 } { -i\xi(1-i\xi)}d\xi.\ ] ] let @xmath31 be a kobol or nts lvy process of order @xmath280 . then we can find @xmath281 s.t . since @xmath283 is bounded away from zero on the line @xmath284 , the integrand in ( [ callprice2 ] ) admits an upper bound via @xmath285 it follows that the integral over @xmath286 makes a contribution of order @xmath287 since @xmath288 , we have @xmath289 on the set @xmath290 the @xmath79 is fixed , therefore for sufficiently small @xmath102 , @xmath291 is negative and bounded away from zero , and therefore , @xmath292 is negative and bounded away from zero , uniformly in @xmath293 and @xmath294 in the half - plane @xmath295 . this means that we can use the cauchy theorem and push the line of the integration in ( [ au1 ] ) down : @xmath296 , and obtain that the integral in ( [ au1 ] ) is zero . hence , @xmath297 by using essentially the same argument as in boyarchenko and levendorski ( 2002b ) , where a similar integral ( 5.33 ) was transformed to the integral over the banks of the cut @xmath298 $ ] in the complex plane , and after the change of variables @xmath299 and @xmath300 on the left and right banks , respectively , to an integral over @xmath301 ( equation ( 5.37 ) in boyarchenko and levendorski ( 2002b ) ) , we obtain , for @xmath302 , @xmath303 as @xmath304 , @xmath305 , therefore by passing to the limit in ( [ genc2 ] ) , we obtain ( [ genc ] ) . recall that the calculations above were made for a kobol or nts lvy process of order @xmath306 . if @xmath31 is a vgp , then we make essentially the same argument by using any @xmath307 . if @xmath31 is a hyperbolic process , nig , or kobol or nts lvy process of order @xmath59 , then we start with the integration by part in ( [ callprice2 ] ) , by using @xmath308 after that we obtain an absolutely converging integral , and can repeat the constructions above with straightforward changes . in the end , we obtain @xmath309dz+o(1).\ ] ] by integrating by part back , and passing to the limit @xmath293 , we obtain ( [ genc ] ) . to finish the proof of ( [ genc ] ) , it remains to show that the result does not change when we add a gaussian component . let @xmath31 be any of the processes above , without a gaussian component , and let @xmath310 be a lvy process with the same drift " term and jump density as @xmath31 , and with the diffusion coefficient @xmath33 . let @xmath311 be the probability density of @xmath310 , and @xmath312 be the probability density of the brownian motion with the volatility @xmath33 . the @xmath312 is the convolution of @xmath177 and @xmath312 , therefore for @xmath81 , @xmath313 take @xmath314 . in the region @xmath315 , the gaussian density is @xmath316 , for any @xmath317 , therefore we may replace the outer integral with the integral over @xmath318 , and add @xmath319 . but then on the support of the integrand , for a fixed @xmath79 , @xmath320 is bounded away from zero , uniformly in @xmath3 . define @xmath321 as @xmath110 but assuming that the log - price follows the jump part of the process . then we can represent the inner integral in the form @xmath322 and therefore , @xmath323 now , define @xmath324 to be @xmath325 for all @xmath326 . then for small @xmath3 , the integral above does not change but now @xmath327 is a continuous bounded function on @xmath328 . by using the super - exponential decay of the gaussian density , we can restore the @xmath329 , and after that pass to the limit and obtain @xmath330 . this finishes the proof of ( [ genc ] ) . [ [ proof - of - cckblcpnts - and - cpjumpccjump - for - jump - processes - with - non - zero - drift ] ] proof of ( [ cckbl])([cpnts ] ) , and ( [ cpjump])([ccjump ] ) for jump processes with non - zero drift ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ let @xmath31 be a vgp . we have @xmath333.\ ] ] if @xmath118 , then the terms with @xmath247 cancel out , and @xmath334 hence , @xmath335 . if @xmath117 , then the terms with @xmath262 cancel out , and @xmath336 thus , @xmath337 . \2 . let @xmath31 be a kobol process . in the case @xmath118 , @xmath338\\ & = & ic\gamma(-\nu)[({{\lambda_+}}-z - i0)^\nu-({{\lambda_+}}-z+i0)^\nu]\\ & = & ic\gamma(-\nu)(z-{{\lambda_+}})^\nu i(e^{-i\pi\nu}-e^{i\pi\nu})\\ & = & 2c\gamma(-\nu)\sin(\pi\nu)(z-{{\lambda_+}})^\nu.\end{aligned}\ ] ] similarly , in the case @xmath117 , @xmath339 \3 . let @xmath31 be an nts lvy process . if @xmath340 , then @xmath341\\ & = & i{\delta}[({\alpha}^2-({\beta}-z+i0)^2)^{\nu/2}-({\alpha}^2-({\beta}-z - i0)^2)^{\nu/2}]\\ & = & i{\delta}({\alpha}-{\beta}+z)^{\nu/2}[({\alpha}+{\beta}-z+i0)^{\nu/2}-({\alpha}+{\beta}-z - i0)^{\nu/2}]\\ & = & i{\delta}({\alpha}-{\beta}+z)^{\nu/2}(z-{\alpha}-{\beta})^{\nu/2}(e^{i\pi\nu/2}-e^{-i\pi\nu/2})\\ & = & -2\sin\frac{\pi\nu}{2}((z-{\beta})^2-{\alpha}^2)^{\nu/2}.\end{aligned}\ ] ] if @xmath342 , then similarly , we obtain the same expression but without the minus sign . finally , if @xmath31 is an exponential jump - diffusion process , then the same proof as in lemma [ gen ] shows that for the call , it suffices to consider the pure jump process , and that in this case , ( [ callprice3 ] ) holds . since @xmath343 the integrand in ( [ callprice3 ] ) has the only pole @xmath344 in the plane @xmath345 . we push down the line of integration , and on crossing the pole , apply the residue theorem . the result is @xmath346 and by passing to the limit , we obtain ( [ ccjump ] ) . ( [ cpjump ] ) is proved similarly . cont , r. , potters , m. , and bouchaud , j .- p . : scaling in stock market data : stable laws and beyond " , in dubrulle , b. , f. graner , and d. sornette ( eds . ) . scale invariance and beyond ( proceedings of the cnrs workshop on scale invariance , les houches , march 1997 ) . berlin : springer 1997 \28 . lamberton , d. : critical price for an american option near maturity . in bolthausen , e. , m. dozzi , f. russo ( eds . ) seminar on stochastic analysis , random fields and applications . boston basel berlin : birkhaser 1995
in this paper , we present a multirate method that is suited for integrating systems of ordinary differential equations ( odes ) which arise from step models describing nanostructure evolution . the strength of our method , in comparison to other existing multirate schemes , is its order of accuracy : our method is fourth order provided solutions are sufficiently smooth . a multirate method is basically one that takes different step sizes for different components of the solution @xcite . when might such a need for different time steps arise ? one situation where multirate methods may be more efficient than single rate ones is when a few of the components contain time singularities or are _ locally stiff_. in this case , ( explicit ) single rate methods are likely to use small time steps for all the components , whereas a multirate one uses small time steps just for singular / locally stiff ones . we believe that this strategy significantly improves the efficiency of solution . integration by our multirate method occurs in two stages . we first use an `` outer '' integrator to handle the non - singular/ non - stiff components and then an `` inner '' integrator to handle the singular / stiff ones . to perform the second inner integration , a small number of components from the outer solution must be interpolated . the high order of accuracy of the multirate scheme relies on the interpolation being of a sufficiently high order . one of our key results in this paper is that for a multirate method to be @xmath0 order , the interpolation must be @xmath1 order . we demonstrate our method by using a fourth order runge - kutta method ( with error control ) for the inner and outer schemes and coupling them together with cubic interpolants . multirate schemes were first studied by gear and wells @xcite . further treatments can be found in @xcite , and the references therein . however , considering the wide variety of methods which researchers have used to improve the performance and accuracy of integration codes , it is surprising to learn that multirate schemes have received only modest attention . this is even more surprising , given that algorithms using somewhat similar concepts are well developed for the numerical solution of pdes e.g. adaptive mesh refinement ( amr ) in the field of hyperbolic pdes @xcite . the applications of multirate methods seem mostly confined to @xmath2-body problems @xcite , and equations arising from electrical networks @xcite . the work presented here is , as far as we are aware , the first instance of a multirate scheme applied to a problem in surface evolution . our method is fairly similar to the one described in @xcite , which is second order . for example : we first advance and interpolate the slow components , and then integrate the fast components this is the `` slowest first '' paradigm described in @xcite . we also automatically detect the fast components by looking at the error estimates produced using an embedded formula . the main differences are that : ( i ) our method is fourth order in time for sufficiently smooth solutions and ( ii ) our method is most effective when applied to the types of odes that commonly arise in models of step surface evolution . we will discuss ( ii ) in more detail later on in the paper , but for now we make the comment that a necessary but not sufficient condition for our method to work is that the ode system be locally coupled . the strengths of our method are that automatic step size selection is simple to implement , and that there is a lower overhead cost because we do not have to interpolate all the slow components of the solution this is one of the benefits of specializing to locally coupled systems . the outline of this paper is as follows : in section [ sec : physical_motivation ] , we discuss why step models are studied and explain the physics behind the equations . in section [ sec02:governing ] we present the step equations , and in section [ sec03:properties ] we describe features of the equations that require special attention . in section [ sec04:code ] we give the details of the numerical method and in particular , we discuss the important issue of interpolation in section [ interpolation ] . we validate our code and present our results in section [ sec05:implementation ] and summarize our findings with a conclusion in section [ sec61:discussion ] . the surface dynamics of crystal structures has received much recent attention @xcite because of its relevance to the fabrication of nano - scale electronic devices , such as quantum dots @xcite . of interest to us here is the behavior of _ vicinal _ surfaces below the roughening temperature @xmath3 . for any given material , a surface forming a small angle with a high symmetry plane of the crystal is called vicinal . the roughening temperature is the critical temperature below which steps become thermodynamically stable . thus , a microscope image of a surface below @xmath4 , with a slight miscut angle , appears as made from a series of terraces separated by steps of atomic height see figure [ fig : silicon ] for an example in silicon . as the surface evolves , the steps move and change their shape , but the steps are well defined and have a lifetime that is long enough to be directly observable . when the temperature is increased above @xmath4 , a _ kosterlitz - thouless _ phase transition occurs @xcite , and the surface becomes statistically rough as characterized by the divergence of the height height correlation function @xcite . for many physical applications ( such as epitaxy ) the operating temperatures are below @xmath3 , and this mesoscale description of a surface in terms of steps and their evolution is very useful . a step model can account for finite size effects occurring at the atomic scale , while remaining computationally simple simulations with step models can be done over much longer time periods than , for example , with atomistic models of the surfaces . the step s shape can , in general , be very complicated . thus , quantitative descriptions for how steps interact with one another can be very difficult to derive , and a complete description of an arbitrarily shaped nanostructure in terms of its steps is currently not available . as a result , theoretical studies have been restricted to simple nanostructure geometries and step shapes . the bcf model , proposed by burton , cabrera and frank @xcite in 1951 , deals with a monotonic step train consisting of an infinite number of parallel steps see figure [ fig : geombcf ] . in this model , the steps edges are separated by atomically smooth terraces , and each step position is uniquely described by a single scalar quantity @xmath5 where the index @xmath6 labels the step . using this model , burton , cabrera and frank were able to describe the evolution of a ( 1 dimensional ) stepped surface , under non - equilibrium conditions , in terms of its steps . in 1988 , rettori and villain @xcite considered a 2d array of circular mounds , and incorporated the effects of step line tension into the bcf model . the nanostructures that they studied consist of a finite number of concentric circular layers , in a `` wedding cake '' configuration see figure [ fig : wedcake ] . this step system can also be used to describe the `` healing '' of small circular pits @xcite produced by scanning tunneling microscopes . the radius of each layer @xmath7 is assumed to be a continuous function of time . physical considerations then lead to a set of locally coupled odes for the radii . similar sets of equations can be found in @xcite . there are two main competing physical processes that take place on a stepped surface , in the absence of evaporation and desorption . the first one is the diffusion of adsorbed atoms ( `` adatoms '' ) across terraces , which is characterized by a diffusivity @xmath8 . the second one is the attachment - detachment of adatoms at step edges , which is characterized by the kinetic coefficients @xmath9 and @xmath10 see figure [ fig : tranprostepsurf ] . experimental evidence @xcite suggests that , for some materials , attachment from the terrace above requires overcoming a higher activation energy barrier than attachment from the terrace below , so that @xmath11 . however , in this paper we consider only @xmath12 , and disregard this ( possible ) asymmetry in the step attachment - detachment known in the literature as an ehrlich - schwoebel ( es ) barrier @xcite . furthermore , we neglect the diffusion of adsorbed vacancies and the diffusion of adatoms along step edges @xcite . in this section , we present the ( non - dimensional ) evolution equations for a finite axisymmetric nanostructure with @xmath13 steps , relaxing in the absence of deposition and evaporation see figure [ fig : wedcake ] . derivations of these equations can be found in @xcite . every step in the structure is subject to two physical effects that drive its motion . the first is a _ step - line tension _ , arising from a gibbs - thomson mechanism @xcite . an isolated , circular step of radius @xmath14 , on top of an infinite substrate , initially devoid of adatoms , reduces its perimeter ( and hence its radius ) by emitting adatoms at a rate proportional to its curvature @xcite i.e. @xmath15 . the second effect is a _ repulsive interaction with neighboring steps _ , characterized by a potential function that is inversely proportional to the square of the distance between the steps @xcite . steps in the bulk of the structure ( with a smaller curvature ) tend to be less affected by the step - line tension compared to steps near the top . let @xmath16 @xmath17 be the positions of the steps numbered starting from the top of the nanostructure . thus @xmath18 , with @xmath19 the radius of the innermost step and @xmath20 the radius of the outermost step . define @xmath21 , @xmath22 , and @xmath23 , by : @xmath24{4.6in}{{${\displaystyle \left . \begin{array}{lclcl } \lambda_1 & = & \lambda(\rho_{1 } , \rho_{2})\/,\\ \lambda_n & = & \lambda(\rho_{n-1 } , \rho_{n } ) + \lambda(\rho_{n } , \rho_{n+1})\/ , \quad \mbox{for } \;\ ; 2 \leq n \leq n-1\/,\\ \lambda_n & = & \lambda(\rho_{n-1 } , \rho_{n})\/ , \end{array } \ ; \right\}\/$}}}\ ] ] @xmath25{4.6in}{{${\displaystyle \hspace*{0.50 in } \mbox{where } \quad \lambda(\rho_i , \rho_j ) = \frac{2\,\rho_i}{\rho_i+\rho_j}\ , \frac{1}{(\rho_i-\rho_j)^3 } + \frac{1}{\rho_j}\ , \left(\frac{\rho_i}{\rho_i^2-\rho_j^2}\right)^2\,,}\/$}}\ ] ] @xmath26 @xmath27 where @xmath28 , @xmath29 , and @xmath30 ( as well as @xmath31 below ) are non - dimensional constants . then the step - flow equations are @xmath32 the non - dimensional constants are as follows : * the parameter @xmath33 measures the strength of the step - step interactions relative to the strength of the step line tension . it is given by @xmath34 where @xmath35 is the step - step interaction coefficient @xcite , @xmath36 is the step stiffness @xcite , @xmath37 is the height of a single step , and @xmath38 is a typical value for the radii for example , it could be the initial radius of the final step in the structure . _ we note that , in many experimental situations , @xmath39_. * the parameters @xmath40 are given by @xmath41 where @xmath42 is the adatom attachment - detachment coefficient at a step , and @xmath43 is the adatom terrace diffusivity . the ratio @xmath44 where @xmath45 and @xmath46 is a typical terrace width , measures the competition between diffusion and attachment - detachment see equation ( [ eqn : sec20:deltan ] ) . * the dimensionless parameter @xmath31 is given by @xmath47 where @xmath48 is the atomic area , @xmath49 is the boltzmann constant , @xmath50 is the absolute temperature , @xmath51 is the equilibrium density of adatoms at a straight , isolated , step and @xmath52 is a typical bulk step velocity . we note that , of the physical parameters involved in the definitions of the non - dimensional constants above , some such as the terrace diffusivity @xmath43 , have been extensively tabulated @xcite , while others such as @xmath51 and @xmath52 , can be inferred from experiments @xcite . however , for the purposes of simulation , we can take @xmath53 without loss of generality , by an appropriate rescaling of the step radii and time . equations ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) constitute a pentadiagonal system , of the form @xmath54 where all the @xmath55 have the same functional form , with the exception of @xmath56 , @xmath57 , @xmath58 , and @xmath59 which govern the behavior of the first and final two steps . notice that if @xmath60 denotes the common rate function for the bulk of the steps ( @xmath61 ) , then @xmath62 in this paper , we present a multirate method for integrating the equations ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) , when @xmath63 . however , we believe that the method presented here should be applicable to _ general sets _ of locally coupled odes , which are _ locally stiff _ ( see section [ local ] ) . in fact , our multirate method was designed to specifically tackle this problem . let us now turn our attention to the difficulties that arise when solving equations ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) numerically . the axisymmetric step - flow equations possess a number of peculiar properties which pose problems for standard integrators hence the need for a multirate algorithm . for example , the singular collapse of the innermost step and stiffness , localized to only a few of the components , are two ( different ) situations under which a standard integrator is forced to use small time steps . in these cases , a single rate method uses small time steps for all components . in contrast , a multirate method will use small time steps only when it has to , so that most of the components are integrated with a large time step . this strategy improves the efficiency of the integration . the singular collapse of the innermost step causes a loss in accuracy for most high order integrators near the point of collapse . hence , we implement a low - order simple euler routine for the innermost step and its neighbors . away from the singular step , we implement a fourth order multirate method with error control , which is able to efficiently integrate locally stiff components . equations ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) have the property that @xmath64 in a finite time . the top step always undergoes a monotonic collapse because its radius always decreases under the effect of step - line tension . as the top step shrinks , it emits adatoms , causing the radii of the second and subsequent steps to grow as these are absorbed . when the top step completely disappears , the number of layers in the structure is reduced by one . as a result of the sequential collapse of top steps , a macroscopically flat region called a facet forms and grows on the top of the structure . provided that the collapse of the top steps is tracked accurately , and the topmost @xmath65 is removed at each collapse , the growth of the facet is automatically accounted for . when the first collapse occurs , @xmath66 , @xmath67 , @xmath68 and @xmath69 replace @xmath70 , @xmath68 , @xmath71 and @xmath72 in ( [ eqn : sec20:eqnrho1 ] ) , and ( [ eqn : sec20:eqnrhon ] ) applies when @xmath73 . a similar replacement occurs for the second and subsequent collapses . in this way , a given index tracks always the same step throughout the integration . let @xmath74 be the collapse time for @xmath19 . then it can be shown @xcite that as @xmath75 , @xmath76 for some constants @xmath77 and @xmath78 . the square root behavior in ( [ square_root ] ) comes from the fact that the leading order behavior for @xmath79 in equation ( [ eqn : sec20:eqnrho1 ] ) stems from a line tension : @xmath80 . thus the derivatives of @xmath19 are divergent at the time of collapse . since ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) is a locally coupled set of equations , we also expect @xmath81 , for @xmath82 , to be singular but for the solutions to become more regular near @xmath83 as @xmath84 becomes larger . since the accuracy of high order integrators usually relies on the solution having enough bounded derivatives , this means that standard high order solvers will lose accuracy near the time of collapse . for example , consider a method with truncation error @xmath85 for smooth solutions @xmath86 , and time step @xmath87 . let @xmath88 be a time @xmath89 steps away from @xmath83 , so that @xmath90 . then , given the square root singularity in ( [ square_root ] ) , the error near @xmath83 will be increased to @xmath91 furthermore , consider the issue of automatic time step selection in an adaptive integration code . this is usually done by estimating the local truncation error , and updating the time step size with a formula for the truncation error that assumes a smooth solution . for example , consider a runge - kutta scheme using an embedded higher order formula to estimate the local truncation error . such an algorithm updates the time step size using a formula like @xcite @xmath92 where @xmath93 is the order of the integrator . equation ( [ stepsizecontrol ] ) is invalid near a time singularity , because from ( [ eq100 ] ) , the error near @xmath83 does not scale as @xmath94 . the resulting behavior is somewhat unpredictable : an adaptive integrator may take a very large number of tiny steps rendering it very inefficient or it may simply abort , stating that the specified error tolerance is not achievable . the aim of this subsection is to attempt to quantify the classes of systems for which the approach in this paper is effective . before explaining what we mean by local stiffness , we first introduce some notation . for @xmath95 , let @xmath96 be the solution of the ode system for some initial condition @xmath97 . for some integer @xmath98 , let @xmath99 be the solution with initial condition @xmath100 , where @xmath101 is the kronecker delta , and @xmath102 is small . finally , let @xmath103 and @xmath104 . we say that the @xmath105 component of the ode system is _ strongly local _ if ( i ) the system is locally coupled and ( ii ) given any @xmath106 , for all @xmath107 , there exists an integer @xmath108 independent of @xmath107 such that for @xmath109 , @xmath110 . if every component is strongly local , we say that the ode _ system _ is strongly local . therefore , a system is strongly local if a small perturbation to any one of its components remains localized in component number . now we explain local stiffness . recall that an ode is stiff when the ratio of the slowest and fastest time scales is much greater than one . the simplest example of this is a situation where the solution of interest is strongly stable , so that small perturbations decay very rapidly , relative to the principal time scale of evolution . now consider again the perturbation described in the previous paragraph . we say that the @xmath105 component of an ode system is _ locally stiff _ if ( i ) it is strongly local and ( ii ) @xmath111 rapidly in time , relative to the principal time scale of evolution . hence , the @xmath105 component of the solution is locally stiff if a perturbation to it remains localized in component number _ and _ decays rapidly in time . if every component is locally stiff , the the solution is globally stiff . once strong locality has been established in an ode system , individual solution _ components _ can be designated as either being ( locally ) stiff or non - stiff . for the rest of this paper , when we refer to `` stiff components '' of the solution , we mean that the components are locally stiff . a `` non - stiff component '' is one that evolves on a time scale comparable to the principal time scale . with this in mind , we can design multirate strategies that handle stiff and non - stiff components separately . for example , we expect to be able to integrate all non - stiff components with large time steps using explicit solvers . if the number of stiff components is relatively small , we can use the same explicit method on the stiff components also , but with much smaller time steps because of stability constraints . if on the other hand the number of stiff components is fairly large , we should resort to a fully implicit stiff solver . at this point it is worth comparing our approach with other work in the literature dealing with problems involving disparate time scales . gear and kevrikidis @xcite propose their `` projective integration '' method to deal with situations where there is a _ gap _ in the spectrum of time scales : the main evolution of a ( stable ) solution occurs slowly , with perturbations decaying much more quickly . for a linear problem , this corresponds to a situation where the eigenvalues can be separated into two groups : one set of moderate sized eigenvalues , and another set with large negative real parts . projective integration requires two ode solvers : an `` inner '' and an `` outer '' integrator . the idea is to take many small steps using the inner solver so that the fast modes are damped out , followed by a large _ projective _ step with the outer integrator . the process is then repeated . this method ( which is not multirate ) is well - suited to handling problems where many , or all , of the solution components are rapidly attracted to a slowly varying manifold . note that there is no notion of `` locality '' in this approach : the fast modes can potentially be coupled with all the slow ones . our multirate method also involves an `` inner '' and an `` outer '' integrator . as discussed above , the property of local stiffness means that a certain subset of the solution components have much stronger stability than the others . these stiff components are handled by an `` inner '' integrator while the non - stiff ones are taken care of using an `` outer '' solver . however , in contrast to the work of gear and kevrikidis , our method is more suited to systems where a small fraction of the components is stiff at any time during the integration . in fact , in terms of the ode s evolution in time , one can think of projective integration as using the inner / outer integrators in `` series '' , whereas our multirate method uses them in `` parallel '' . we will now show that @xmath112 , the radius of a bulk step , in ( [ eqn : sec20:eqnrhon ] ) is strongly local provided @xmath113 and hence , away from the facet and the substrate , any stiffness that arises is localized . for a bulk step , the physical origin of the rapid decay comes from the nature of step interactions in equation ( [ eqn : sec20:lambda ] ) . steps strongly repel each other when they get too close together . consider a configuration where some of the steps in the bulk are tightly bunched together , and most of the other steps are widely spaced apart . in this case , a step strictly ( two steps away , at least , from the edge ) inside a bunch is strongly stable , and hence stiff , because small perturbations in its trajectory are opposed by strong interactions from the neighboring steps . on the other hand , widely spaced steps do not experience such large forces , and respond to perturbations on much slower time scales . it turns out that these `` step bunching '' configurations are quite common in practice and are produced by the natural time evolution of the system . in fact , the step bunching instability @xcite is a well - studied phenomenon in epitaxial growth , with applications in quantum dot technology @xcite and nanolithography @xcite . to analyze the decay of solutions , the direct approach would be to compute the jacobian matrix and analyze its eigenspectrum . unfortunately , while the jacobian for the system in ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) can be computed analytically by linearizing at any fixed set of radii @xmath114 , the expressions involved are very complicated , and do not give much insight as to why the equations should be stiff . instead , we present below a less rigorous calculation , which allows us to relate the degree of local stiffness to the step spacing . our approach is based on the fact that the number of equations , @xmath2 , is generally rather large , and that the solutions of interest have a step spacing that is , piecewise , nearly constant . by this we mean that the step spacing @xmath115 changes slowly with @xmath116 , except for a few places where it may change abruptly the effect of these changes is much harder to analyze , and our method of attack ignores them since it is only valid far away from these rapid transition regions . however , the results of our numerical calculations indicate that their presence does not invalidate our analysis . we begin by considering a configuration of steps which has a nearly constant step spacing , and expand the solution in the form @xmath117 + @xmath118 , where @xmath119 is a constant , @xmath120 is the ( constant ) leading order step spacing and @xmath121 . substituting this expression into the step flow equations , and ignoring the equations for the boundary steps ( corresponding to @xmath122 and @xmath123 ) , results in the following leading order equation for the perturbation @xmath124 @xmath125 where @xmath126 is the discrete laplacian : @xmath127 . to show that equation ( [ conststepspacingpert ] ) has the property of strong locality , we consider the solution to the problem with the initial condition @xmath128 , given by @xmath129 with @xmath130 . when @xmath131 , @xmath132 exponentially because the integrand in ( [ conststepspacingiv ] ) is @xmath133-periodic . hence the delta function initial condition remains localized for all @xmath107 . also , note that equation ( [ conststepspacingpert ] ) has free normal modes given by @xmath134 and @xmath135 is the wave - number . it follows that the time scales behave like the fourth power of the step spacing . hence , widely spaced steps evolve on a slow time scale whereas step bunches , which consist of sets of tightly packed steps , give rise to fast time scales and local stiffness . the goal of our method is to efficiently solve a system of locally coupled odes where only a few of the components are stiff . a standard explicit integrator would take small time steps for all components of the solution . in contrast , our multirate method takes large steps for the non - stiff components , and small steps for the stiff ones . the algorithm starts by taking an explicit , global time step , say : from @xmath136 to @xmath137 . an embedded formula is then used to obtain an estimate of the local truncation error ( lte ) for each component of the solution . in general , some of the ltes will be unacceptably large ( because the associated solution components are stiff ) , while others will have acceptable sizes . the algorithm checks if the components with acceptable ltes satisfy the preset tolerance levels . if they do not , the step size is reduced and another global time step is attempted . if they do , a second round of integration is performed to correct the components with large ltes . hence , the algorithm is as follows : * take a step from @xmath138 to @xmath139 . * let @xmath140 be the lte for the @xmath141-th component , let @xmath142 be the @xmath143 percentile of all the ltes , and let @xmath144 be the required error tolerance for @xmath84-th component . for example , if @xmath145 , then @xmath146 of the errors are larger than @xmath147 . if @xmath148 , @xmath147 is the median . * for some real number @xmath42 , flag all the components whose ltes are greater than @xmath149 ( our code uses @xmath150 ) as being possibly stiff . * check if the unflagged solution components satisfy the tolerance requirements , i.e. @xmath151 where @xmath152 is taken over all unflagged components only . * * * if they do , the step is successful . the step size is increased using the formula ( [ stepsizecontrol ] ) with @xmath153 as the ratio of errors . * * perform a second integration to correct the flagged components ( see figure [ fig4 ] ) . * * increase the step size according to formula ( [ stepsizecontrol ] ) . increase @xmath154 and go back to step 1 . * * * if they do not , the step is not successful . reduce the step size according to ( [ stepsizecontrol ] ) with @xmath155 as the ratio of errors . * * do not increase @xmath154 and go back to step 1 . [ algorithm2 ] the second integration is basically done only for the stiff components , and it involves many small sub - steps within the interval @xmath156}\/$ ] to ensure stability . although this second round of integration takes a large number of steps , it only needs to be done for a small subset of the solution components . to perform the second integration , values for the non - stiff components at all times @xmath157 along the boundaries of any stiff set of steps are needed see figure [ fig4 ] . for example , in the case of a pentadiagonal system , the values of two non - stiff components are needed on each side of a stiff region . one way to generate dense output from the non - stiff components , between @xmath158 and @xmath159 , is through interpolation . in this paper we use cubic interpolatants which are generated by using the intermediate stage function evaluations in a runge - kutta method . in section [ interpolation ] , we give more details on the construction of these interpolants and demonstrate that cubic interpolation is consistent with a multirate method that is globally fourth order . once solution components have been flagged as requiring re - integration , the local coupling means that some of the non - stiff components may also have to be re - integrated . because ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) is a pentadiagonal system of equations , if only @xmath160 and @xmath161 are stiff components with large ltes , then all three of the components @xmath162 and @xmath161 must be re - integrated as a set , using the dense output from @xmath163 , @xmath164 , @xmath165 and @xmath166 as ` boundary conditions ' . hence , the algorithm is slightly wasteful in that although @xmath167 was deemed accurate enough , it still had to be integrated for a second time . note that our algorithm uses the lte in a different way from conventional embedded rk methods : instead of immediately scaling the time step if the smallest lte is greater than the tolerance level , we make a note of _ which _ components had the largest ltes by analyzing their distribution : it might not be efficient to retake the time step for every component , if only a few of them are inaccurate . the largest ltes ( in the sense of being larger than @xmath168 in algorithm [ algorithm2 ] ) are discarded , and then the time step scaled according to the largest of the remaining errors . hence , we get a larger time step for a majority of the solution components , and the way that this time step is adjusted throughout the course of the integration is not affected by the presence of a few either rapidly varying or stiff components . for the rest of this paper , we will call the first time stepper @xmath169 ( used to generate the ltes in the first place ) , and the second time stepper @xmath170 ( used to re - integate stiff components with large ltes ) . in general , @xmath169 and @xmath170 do not have to be the same method , or of the same order , but @xmath169 has to be able to generate estimates of the local truncation error . in our code , @xmath169 is a cash - karp runge - kutta formula @xcite and @xmath170 depends on the solution component : if the re - integration involves the innermost step ( see section [ treatment ] ) , we take @xmath170 to be a simple - euler routine which adjusts its step size by step doubling ; otherwise @xmath171 . in other words , there are two possibilities which can arise when performing the re - integration with @xmath170 : 1 . the re - integration involves solution components which include the innermost step . in the following analysis , we will assume this is @xmath70 . [ case1 ] 2 . the re - integration does not involve the innermost step . [ case2 ] the reason to distinguish between these two cases is that ( [ case1 ] ) will involve integration of singular trajectories ( see equation ( [ square_root ] ) ) , but in general , ( [ case2 ] ) will not . using step doubling in ( [ case1 ] ) is a fairly crude way of adjusting the time step . however , resorting to an embedded formula is not possible when ( [ stepsizecontrol ] ) breaks down , so using step doubling to monitor the quality of the solution is reasonable in this case . from equation ( [ square_root ] ) , we have seen that @xmath172 in a singular fashion , causing problems for standard high order integrators . our treatment uses a simple euler method whenever the re - integration of @xmath70 and neighboring singular components is involved : this is `` optimal '' in the sense that simple euler produces results that have the same accuracy as higher order methods ( due to the singular nature of @xmath70 ) but is computationally cheaper . furthermore , we are able to extract the time of collapse , @xmath173 , using linear interpolation , which is consistent with simple euler s order of accuracy . our method involves solving for @xmath174 instead of @xmath175 . note that from ( [ square_root ] ) , @xmath176 as @xmath75 ( the collapse time ) for some constants @xmath177 and @xmath178 , which means that @xmath179 has exactly one derivative at @xmath83 . our main reason for solving for @xmath179 , instead of @xmath70 , is not to improve accuracy , but rather to enable the algorithm to ` step through ' the singularity at @xmath180 , and use linear interpolation to obtain @xmath83 , the time of collapse of the innermost step . taking square roots to recover @xmath70 will will result in a drastic loss in accuracy near @xmath83 . at time @xmath107 close to @xmath83 , consider taking a time step of size @xmath181 with component @xmath179 using simple euler . let @xmath182 be the result of taking this time step using a ` perfect ' integrator , producing the exact solution at @xmath183 , given @xmath184 . then , since the truncation error in simple euler is @xmath185 from ( [ rho_sq ] ) , we have @xmath186 therefore , if @xmath187 ( @xmath107 sufficiently far away from the singularity ) then the lte for @xmath70 , @xmath188 , is @xmath189 . however , if @xmath190 ( @xmath107 is very close to @xmath173 ) , then the lte for @xmath70 is @xmath191 , which is not a big improvement over ( [ eq100 ] ) . note that these estimates for the lte are independent of the order of @xmath170 . when @xmath170 has ` overstepped ' @xmath83 resulting in @xmath192 and @xmath193 for times @xmath194 ( @xmath195 ) , we set @xmath196 as an approximation to the collapse time . once @xmath70 has collapsed at @xmath83 , it is removed from the system ( [ eqn : sec20:eqnrho1 ] ) ( [ eqn : sec20:eqnrhon ] ) , the number of equations drops by one , and @xmath197 replaces @xmath184 as the new top step . the key to making our multirate method high order lies in the ability to generate dense output from the non - stiff components with high accuracy . one way to generate dense output is to use interpolation and @xmath139 , the function that we derive does not actually pass through @xmath139 . hence strictly speaking , it is not an interpolant , though we will continue to refer to these approximating functions as `` interpolants '' for convenience . ] . for some integrators , such as backward differentiation formulae ( bdf ) , is it obvious how to derive an interpolant that is consistent in order with the underlying integrator bdf use extrapolation to advance the solution in time . for other integrators , such as runge - kutta schemes , constructing the interpolant is less obvious and this is the focus of the section . note that we need to generate interpolants _ during run time _ using only the function evaluations that have already been computed by the integrator within each time step . the extra constraint of generating the interpolants during run time adds a non - trivial complication to the `` traditional '' interpolation problem , which has been studied extensively @xcite . having successfully integrated the non - stiff components from @xmath138 to @xmath139 , we have the point values @xmath198 and the derivative @xmath199 at our disposal to construct the interpolant between @xmath138 and @xmath139 . we do _ not _ have information about the derivative @xmath200 . however , because we are constructing these interpolants during run time , we are at liberty to use the intermediate function evalulations inherent in the application of runge - kutta : this is valuable information that is not usually available in the traditional interpolation problem . we have seen that our method performs integration in two phases : first we integrate a large number of non - stiff components , then we integrate a small number of stiff ones . ideally , we would like the two integrations to have the same order . this is only possible if the interpolation of the non - stiff components is of a sufficiently high order otherwise large interpolation errors will contaminate the accuracy in the stiff components . in the following paragraphs , when we generate dense output between points @xmath107 and @xmath183 , we define an interpolant to be order @xmath201 when the interpolation error is @xmath202 . let us assume that our inner and outer solvers are both @xmath0 order . first of all , let us calculate @xmath201 in terms of @xmath154 if we want our method to be globally @xmath0 order . for simplicity , we assume in this calculation that the effects of round - off error are negligible . consider the ode system @xmath203 let us assume that we have taken a macro step of size @xmath181 and advanced the non - stiff components @xmath204 successfully from time @xmath107 to @xmath183 . also assume that we have taken @xmath2 micro steps of size @xmath205 , @xmath95 , for the stiff components @xmath206 so that @xmath207 . after taking these @xmath2 steps , the total error in @xmath206 will be @xmath208 the first term is the sum of the local truncation errors caused by taking @xmath2 steps each of size @xmath205 . the second term is the sum of the interpolation errors : note that to advance @xmath209 , an evaluation of @xmath210 in between @xmath107 and @xmath183 , in general , is required for the non - stiff neighbours of @xmath209 and this incurs an interpolation error of size @xmath202 . therefore the error in @xmath211 is also @xmath202 and the error in @xmath209 is @xmath212 . it is clear that ( [ toterr ] ) simplifies to @xmath213 and so for our multirate method to be globally @xmath0 order , we require @xmath214 that is , we can afford for the order of the interpolation to be one less than the order of the integrator . for example , if our integrator is fourth order ( @xmath215 ) , we need to be able to construct cubic interpolants during run time . if our integrator is second order , then linear interpolation should be sufficient as observed in @xcite . we will now illustrate how these interpolants are constructed by taking the classical 4th order ( non - adaptive ) runge - kutta formula as an example and applying it to the autonomous ode system @xmath216 : .coefficients in classical rk4 [ cols= " < , < , < , < , < " , ] we used our multirate code to solve the simple wave equation @xmath217 . by discretizing using the method of lines and one - sided ( `` upwind '' ) differences in space , we obtained @xmath2 coupled ordinary differential equations which were solved using the initial condition @xmath218 and periodic boundary conditions . unlike the axisymmetric step equations , this discretization of the wave equation does not have any time singularities and we expect fourth order convergence for every component . this is confirmed in figure [ fig : convergence ] . although @xmath181 and @xmath219 constantly change because our algorithm uses adaptive step size control , we take @xmath220 as a measure of the average step size where @xmath221 is the final integration time . for the single rate method , @xmath2 is the number of steps and for the multirate method , @xmath2 is the total number of micro - steps . in the multirate code , a maximum macro - stepsize @xmath222 was imposed , and all but the first and final macro - steps had size 1 . the exact solution comes from solving the linear system of odes exactly using an eigenfunction expansion . as the integration progresses , the fraction of components that has to be reintegrated increases gradually as the solution broadens and its amplitude decreases : see figure [ fig : wave ] . a matlab multirate code that produces the results in figure [ fig : wave ] is given in the appendix . figure [ fig5](a ) shows the results of an integration with @xmath223 . note that only those steps which are near the facet tend to pack closely together , but steps which are far away move relatively slowly and do not deviate significantly from their initial uniform configuration . this expanding front of closely packed steps represents the @xmath224 expansion of the facet radius @xcite . figure [ fig5](b ) illustrates the separation in time scales of the solution components and shows which components of the solution are integrated for a second time . as expected , our algorithm takes large time steps for components which are far away from the facet . near the facet and the collapsing top step , many relatively small steps are taken . for the rapidly varying components in this figure , only a representative sample of the meshpoints @xmath138 from the @xmath170 integration are shown . in contrast , when @xmath225 , steps can be closely packed even away from the facet . the plots in figure [ fig6 ] show that a step bunching instability arises when @xmath226 is sufficiently small and are qualitatively very different to those in figure [ fig5 ] . the instability originates from steps with smaller radii and gradually spreads outwards so that more and more steps bunch up . our multirate scheme performs a second integration when bunching and local stiffness arise : therefore , our algorithm gradually becomes less efficient over time . however , as long as the fraction of bunched steps is not too large , our algorithm remains competitive compared to a standard adaptive 4th/5th order runge - kutta code . when the fraction of bunched steps becomes close to unity , the optimal strategy is to have the algorithm detect this automatically , and then switch to a fully implicit , single - rate stiff solver . we leave this as future work , noting that inversion of the pentadiagonal jacobian only costs @xmath227 operations ( where @xmath2 is the total number of existing steps ) . in this paper , we present a multirate integration scheme that is designed to efficiently solve the systems of odes that arise in the relaxation of crystal mounds . these odes have two properties that call for a multirate strategy : the singular collapse of the innermost step and local stiffness . our method automatically detects singular / stiff components in the solution and disregards them when computing the size of the bulk ( macro ) time step . the result is that the bulk timestep can be much larger than in a single rate method . the trade - off is a re - integration of the stiff components which usually consists of a small fraction of the total number of equations in the ode system . our method is globally fourth order when applied to odes which have sufficiently smooth solutions for example , the step equations studied in @xcite and the wave equation discretized through the method of lines in section [ subsec : error - spread ] . however , the time singularities present in the axisymmetric step - flow equations mean that near times of collapse , the integration of steps near the facet suffers a loss in accuracy . to specifically deal with the singular inner trajectories , our method couples a simple euler routine to the bulk solver . given that the truncation error reduces to @xmath228 near the collapse time , independent of the method order , simple euler is the preferred method because it is computationally cheaper . furthermore , the use of linear interpolation to extract collapse times is consistent with the method s order . the high order accuracy of our algorithm ( when applied to bulk steps ) relies on the ability to generate high order interpolants during the run - time of a one - step integration method . our algorithm generates these interpolants by using the intermediate stage function evaluations of an embedded runge - kutta ( rk ) formula . specifically , our method computes @xmath229 order interpolants that are consistent with the @xmath230 order accuracy of the integrator . however , for general @xmath0 order rk formulae , we do not know if it is always possible to construct interpolants that have order @xmath231 . we see four main possible extensions to this work . the first is to generalize our multirate paradigm so that it can be used for ( i ) higher order runge - kutta formulae and ( ii ) multistep methods ( e.g. bdf , adams etc . ) we believe that it should be possible to make _ any _ method multirate the main obstacle in doing this is to derive interpolants of a suitably high order . the second is to explore in more detail the types of pdes that our multirate method can apply to . typically , large systems of odes result from discretizations of pdes and it is for large ode systems that our method becomes competitive with single rate methods . we think that a basic requirement of the discretization is that it should be strongly local . however , we have not fully explored which discretizations are strongly local and which are not . for example , a one - sided , upwind discretization @xmath232 of the advection equation @xmath233 is strongly local only when @xmath234 . for @xmath235 , a kronecker delta initial condition is unstable and does _ not _ remain localized . a discretization using centered differences @xmath236 yields a system of odes that is never strongly local for any @xmath237 . for nonlinear equations , our method seems to be efficient for step - flow like odes with repulsive dipolar step - step interactions . we were able to show that the linearized step flow odes are strongly local ; however , we do not know if linearizing an ode system is always sufficient to show strong locality . the third is to explore how the choice of parameters @xmath238 and @xmath239 in algorithm [ algorithm2 ] affect the efficiency of the integration and if there are optimal values of @xmath238 and @xmath239 . a `` good '' choice for @xmath238 and @xmath239 will result in small number of re - integrated components and a large macro - time step . if @xmath238 is too large and @xmath239 too small , the algorithm behaves like a single rate method . on the other hand , if @xmath238 is too small and @xmath239 is too large , many non - stiff components will be re - integrated along with the stiff ones , rendering the method inefficient . furthermore , our choice of @xmath147 as the @xmath143 percentile of the ltes is somewhat arbitrary ( but seems to generate reasonable results ) . another possibility is to take the mean this amounts to increasing the sensitivity of the bulk step size to the presence of one or two extremely stiff components . clearly , the performance of our multirate method is tied to the distribution of ltes , its moments , and identification of the `` largest errors '' . quantification of the `` largest errors '' and deciding which moments to use is work in progress . in summary , this work contributes to the currently growing body of research in multirate methods . we hope that the strategies adopted in this paper can be carried over to other physical problems and used to improve the efficiency and accuracy of future multirate algorithms . * acknowledgements * + we thank dionisios margetis for many helpful discussions and meetings . rrr was partially supported by nsf grant dms-0813648 . here we give the details of a matlab multirate code to solve a wave equation . % parameters for integrator desired_error = 1e-6 ; % desired error per step mr = 1 ; % set mr=1 for multirate mode , mr=0 for single rate t= 20 ; t = 0 ; dt = 0.1 ; dt_max = 1.0 ; k = -3 ; p = 30 ; % approx percentage of components to reintegrate safety = 0.95 ; while t < t [ unew , error , k ] = rk_onestep(t , u , dt,[1 n ] ) ; % % % % % start of multirate modification % % % if mr = = 1 % flag large errors flags = error > 10^k*percentile(error , p ) ; for i=1:length(flags ) if flags(i ) = = 1 r = i ; break ; end end for j = length(flags):-1:1 if flags(j ) = = 1 s = j ; break ; end end % just to be safe r = r - w ; s = s+w ; [ unew , num_micro_steps ] = micro_integrate(t , t+dt , u , unew , ... [ r s],k , desired_error*1e-3 ) ; end % % % end of multirate modification % % % max_error = max ( max(error(1:r-1 ) ) , max(error(s+1:end ) ) ) ; r = ( desired_error / max_error ) ^(1/5 ) ; if r<1 % step failed dt = dt*safety*max(0.1,r ) ; trynum = trynum+1 ; if trynum>10 sprintf('10 failed attempts ! ' ) return end elseif r>1 % step succeeded [ r s ( s - r ) ( s - r)/n ] t = t+dt ; u = unew ; trynum = 0 ; numsteps = numsteps+1 ; dt = dt*safety*min(5,r ) ; if t+dt > t dt = ( t - t ) ; end if dt > dt_max dt = dt_max ; end end if t > t1 & & plot_number = = 1 subplot(4,2,1 ) plot(x , unew,'linewidth',2 ) ; xlabel('x ' ) ; ylabel('u(x , t ) ' ) ; tit = sprintf('t=%f',t1 ) ; title(tit ) ; axis([-l/2 l/2 0 1 ] ) ; subplot(4,2,3 ) spy(flags,20 ) title('flagged components ' ) ; plot_number = 2 ; elseif t > t2 & & plot_number = = 2 subplot(4,2,2 ) plot(x , unew,'linewidth',2 ) ; xlabel('x ' ) ; ylabel('u(x , t ) ' ) ; tit = sprintf('t=%f',t2 ) ; title(tit ) ; axis([-l/2 l/2 0 1 ] ) ; subplot(4,2,4 ) spy(flags,20 ) title('flagged components ' ) ; plot_number=3 ; elseif t > t3 & & plot_number = = 3 subplot(4,2,5 ) plot(x , unew,'linewidth',2 ) ; xlabel('x ' ) ; ylabel('u(x , t ) ' ) ; tit = sprintf('t=%f',t3 ) ; title(tit ) ; axis([-l/2 l/2 0 1 ] ) ; subplot(4,2,7 ) spy(flags,20 ) title('flagged components ' ) ; plot_number=4 ; elseif t > t4 & & plot_number = = 4 subplot(4,2,6 ) plot(x , unew,'linewidth',2 ) ; xlabel('x ' ) ; ylabel('u(x , t ) ' ) ; tit = sprintf('t=%f',t4 ) ; title(tit ) ; axis([-l/2 l/2 0 1 ] ) ; subplot(4,2,8 ) spy(flags,20 ) title('flagged components ' ) end end y_current = y1 ; t = t1 ; trynum = 0 ; numsteps = 0 ; while t < t2 [ y_new , error ] = rk_onestep(t , y_current , dt,[r-2,s+2 ] ) ; r = ( desired_error / max(error(r : s ) ) ) ^(1/5 ) ; if r<1 % step failed dt = dt*max(0.1,r ) ; trynum = trynum+1 ; if trynum>10 sprintf('10 failed attempts ! ' ) return end elseif r>1 % step succeeded t = t+dt ; numsteps = numsteps+1 ; trynum = 0 ; y_new(r-2 ) = interpolate(t , y1(r-2),y2(r-2),t1,t2,k(r-2 , : ) ) ; y_new(r-1 ) = interpolate(t , y1(r-1),y2(r-1),t1,t2,k(r-1 , : ) ) ; y_new(s+1 ) = interpolate(t , y1(s+1),y2(s+1),t1,t2,k(s+1 , : ) ) ; y_new(s+2 ) = interpolate(t , y1(s+2),y2(s+2),t1,t2,k(s+2 , : ) ) ; y_current = y_new ; dt = dt*min(5,r ) ; if t+dt > t2 dt = ( t2-t ) ; end end end y2 = y_current ; a2 = 1/5 ; a3 = 3/10 ; a4 = 3/5 ; a5 = 1 ; a6 = 7/8 ; b21 = 1/5 ; b31 = 3/40 ; b32 = 9/40 ; b41 = 3/10 ; b42 = -9/10 ; b43 = 6/5 ; b51 = -11/54 ; b52 = 5/2 ; b53 = -70/27 ; b54 = 35/27 ; b61 = 1631/55296 ; b62 = 175/512 ; b63 = 575/13824 ; b64 = 44275/110592 ; b65 = 253/4096 ; k1 = dt*f(t , y , cpt ) ; k2 = dt*f(t+a2*dt , y+b21*k1,cpt ) ; k3 = dt*f(t+a3*dt , y+b31*k1+b32*k2,cpt ) ; k4 = dt*f(t+a4*dt , y+b41*k1+b42*k2+b43*k3,cpt ) ; k5 = dt*f(t+a5*dt , y+b51*k1+b52*k2+b53*k3+b54*k4,cpt ) ; k6 = dt*f(t+a6*dt , y+b61*k1+b62*k2+b63*k3+b64*k4+b65*k5,cpt ) ; k = [ k1 ' k2 ' k3 ' k4 ' k5 ' k6 ' ] ; m. kitamura , m. nishioka , j. oshinowo , y. arakawa , in - situ fabrication of self - aligned ingaas quantum dots on gaas multiatomic steps by metalorganic chemical vapor deposition , appl . 66 ( 26 ) ( 1995 ) 36633665 . j. g. mclean , b. krishnamachari , d. r. peale , e. chason , j. p. sethna , b. h. cooper , decay of isolated features driven by the gibbs - thomson effect in an analytic model and a simulation , phys . rev . b 55 ( 3 ) ( 1997 ) 1811 1823 . a. ichimiya , y. tanaka , k. ishiyama , quantitative measurements of thermal relaxation of isolated silicon hillocks and craters on the si(111)-(7x7 ) surface by scanning tunneling microscopy , phys . 76 ( 25 ) ( 1996 ) 4721 4724 .
in the modern mathematical literature , catalan numbers are wonderfully ubiquitous . although they appear in a variety of disguises , we are so used to having them around , it is perhaps hard to imagine a time when they were either unknown or known but obscure and underappreciated . it may then come as a surprise that catalan numbers have a rich history of multiple rediscoveries until relatively recently . below we review over 200 years of history , from their first discovery to modern times . we break the history into short intervals of mathematical activity , each covered in a different section . we spend most of our effort on the early history , but do bring it to relatively recent developments . we should warn the reader that although this work is in the history of mathematics , we are not a mathematical historian . rather , this work is more of a historical survey with some added speculations based on our extensive reading of the even more extensive literature . due to the space limitations , this survey is very much incomplete , as we tend to emphasize first discoveries and papers of influence rather than describe subsequent developments . this paper in part is based on our earlier investigation reported in @xcite . many primary sources are assembled on the _ catalan numbers website _ @xcite , including scans of the original works and their english translations . ming antu , sometimes written mingantu ( c.1692c.1763 ) was a chinese scientist and mathematician , native of inner mongolia . in 1730s , he wrote a book _ quick methods for accurate values of circle segments _ which included a number of trigonometric identities and power series , some involving catalan numbers : @xmath0 the integrality of catalan numbers played no role in this work . ming antu s book was published only in 1839 , and the connection to catalan numbers was observed by luo jianjin in 1988 . we refer to @xcite for more on this work and further references . in 1751 , leonhard euler ( 17071783 ) introduced and found a closed formula for what we now call the _ catalan numbers_. the proof of this result had eluded him , until he was assisted by christian goldbach ( 16901764 ) , and more substantially by johann segner . by 1759 , a complete proof was obtained . this and the next section tell the story of how this happened . on september 4 , 1751 , euler wrote a letter to goldbach which among other things included his discovery of catalan numbers . euler was in berlin ( prussia ) at that time , while his friend and former mentor goldbach was in st . petersburg ( imperial russia ) . they first met when euler arrived to st . petersburg back in 1727 as a young man , and started a lifelong friendship , with 196 letters between them @xcite . euler defines catalan numbers @xmath1 as the number of triangulations of @xmath2-gon , and gives the values of @xmath1 for @xmath3 ( evidently , computed by hand ) . all these values , including @xmath4 are correct . euler then observes that successive ratios have a pattern and guesses the following formula for catalan numbers : @xmath5 for example , @xmath6 . he concludes with the formula for the catalan numbers g.f . : @xmath7 in his reply to euler dated october 16 , 1751 , goldbach notes that the g.f . @xmath8 satisfies quadratic equation @xmath9 he then suggests that this equation can be used to derive catalan numbers via an infinite family of equations on its coefficients . euler writes back to goldbach on december 4 , 1751 . there , he explains how one can obtain the product formula from the binomial formula : @xmath10 from the context of the letter , it seems that euler knew the exact form of before his investigation of the catalan numbers . thus , once he found the product formula , he was able to derive rather than simply guess it . we believe that euler did not include his derivation in the first letter as the g.f . formula appears at the very end of it , but once goldbach became interested he patiently explained all the steps , as well as other similar formulas . johann andreas von segner ( 17041777 ) was another frequent correspondent of euler . although segner was older , euler rose to prominence faster , and in 1755 crucially helped him to obtain a position at university of halle , see @xcite . it seems , there was a bit of competitive tension between them , which adversely affected the story . in the late 1750s , euler suggested to segner the problem of counting the number of triangulations of an @xmath11-gon . we speculate based on segner s later work , that euler told him only values up to @xmath12 , and neither the product formula nor the g.f . . segner accepted the challenge and in 1758 wrote a paper @xcite , whose main result is a recurrence relation which he finds and proves combinatorially : @xmath13 he then uses the formula to compute the values of @xmath1 , @xmath14 , but makes an arithmetic mistake in computing @xmath15 , which then invalidates all larger values . euler must have realized that equation is the last missing piece necessary to prove . he arranged for segner s paper to be published in the journal of st . petersburg academy of sciences , but with his own _ summary _ @xcite . in it , he states , gives segner a lavish compliment , then points out his numerical mistake , and correctly computes all @xmath1 for @xmath16 . it seems unlikely that given the simple product formula , segner would have computed @xmath17 incorrectly , so we assume that euler shared it only with his close friend goldbach , and kept segner in the dark until after the publication . in summary , a combination of results of euler and segner , combined with goldbach s observation gives a complete proof of the product formula . unfortunately , it took about 80 years until the first complete proof was published . semn kirillovich kotelnikow ( 17231806 ) was a russian mathematician of humble origin who lived in st . petersburg all his life . in 1766 , soon after the raid on berlin , euler returned to st . same year , kotelnikow writes a paper @xcite elaborating on catalan numbers . although he claimed to have another way to verify , larcombe notes that he `` does little more than play around with the formula '' @xcite . nicolas fuss ( 17551826 ) was a swiss born mathematician who moved to st . petersburg to become euler s assistant in 1773 . he married euler s granddaughter , became a well known mathematician in his own right , and remained in russia until death . in his 1795 paper @xcite , in response to pfaff s question on the number of subdivisions of an @xmath11-gon into @xmath18-gons , fuss introduced what is now know as the _ fuss catalan numbers _ ( see exercise a14 ) , and gave a generalization of segner s formula , but not the product formula . in 1836 , a young french mathematician joseph liouville ( 18091882 ) founded the _ journal de mathmatiques pures et appliques_. he was in the center of mathematical life in paris and maintained a large mailing list , which proved critical in this story . in 1838 , a jewish french mathematician and mathematical historian olry terquem ( 17821862 ) asked liouville if he knows a simple way to derive euler s formula from segner s recurrence . liouville in turn communicated this problem to `` various geometers '' . what followed is a remarkable sequence of papers giving foundation to `` catalan studies '' . first , gabriel lam ( 17951870 ) wrote a letter to liouville outlining the solution , a letter liouville promptly published in the _ journal _ and further popularized @xcite . lam s solution was to use an elegant double counting argument . let s count the number @xmath19 of triangulations of an @xmath2-gon with one of its @xmath20 diagonals oriented . on the one hand , @xmath21 . on the other hand , by summing over all possible directed diagonals we have @xmath22 combining these two formulas with easily implies . in 1738 , belgium born mathematician eugne charles catalan ( 18141894 ) was a student of liouville at cole polytechnique . inspired by the work of lam , he became interested in the problem . he was the first to obtain what are now standard formulas @xmath23 he then studied the problem of computing the number of different ( non - associative ) products of @xmath11 variables , equivalent to counting the number of bracket sequences @xcite . olinde rodrigues ( 17951851 ) was a descendant of a large sephardic jewish family from bordeaux . he received his doctorate in mathematics and had a career as a banker in paris , but continued his mathematical interests . in the same volume of the _ journal _ , he published two back - to - back short notes giving a more direct double counting proof of . the first note @xcite gives a variation on lam s argument , a beautiful idea often regarded a folklore . roughly , he counts in two ways the number @xmath24 of triangulations of @xmath2-gon where either an edge or a diagonal is oriented . on the one hand , @xmath25 . on the other hand , @xmath26 , since triangulations with an oriented diagonal are in bijection with triangulations of @xmath27-gon obtained by inserting a triangle in place of a diagonal , and such edge can be any edge except the first one . we omit the details . in @xcite , rodrigues gives a related , but even simpler argument for counting the bracket sequences . denote by @xmath28 the number of bracket sequences of labeled terms @xmath29 , e.g. @xmath30 . then , on the one hand @xmath31 . on the other hand , @xmath32 since variable @xmath33 can be inserted into every bracket sequence in exactly @xmath34 ways . to see this , place a bracket around every variable and the whole product , e.g. @xmath35 . now observe that a new variable is inserted immediately to the left of any of the @xmath36 left brackets , or immediately to the right of any of the @xmath36 right brackets , e.g. @xmath37 in 1839 , clearly unaware of euler s letters , a senior french mathematician jacques binet ( 17861856 ) wrote a paper @xcite with a complete g.f . proof of . across the border in germany , johann august grunert ( 17971872 ) became interested in the work of the frenchmen on the one hand , and of fuss on the other . a former student of pfaff , in a 1841 paper @xcite he found a product formula for the fuss catalan numbers . he employed g.f.s to reduce the problem to @xmath38 but seemed unable to finish the proof . the complete proof was given in 1843 by liouville @xcite using the _ lagrange inversion _ ( see @xcite ) . after a few more mostly analytic papers inspired by the problem , the attention of the french school turned elsewhere . catalan however returned to the problem on several occasions throughout his career . even fifty years later , in 1878 , he published _ sur les nombres de segner _ @xcite on divisibility of the catalan numbers . thomas kirkman ( 18061895 ) was a british ordained minister who had a strong interest in mathematics . in 1857 , unaware of the previous work but with cayley s support , he published a lengthy treatise @xcite . there , he introduced the _ kirkman cayley numbers _ , defined as the number of ways to divide an @xmath11-gon with @xmath18 non - intersecting diagonals ( see a41 ) , and states a general product formula , which he proves in a few special cases . in 1857 , arthur cayley ( 18211895 ) was a lawyer in london and extremely prolific mathematically . he was interested in a related counting of _ plane trees _ , and in 1859 published a short note @xcite , where ( among other things ) he gave a conventional g.f . proof of that the number of plane trees is the catalan number . like kirkman , he was evidently unaware of the previous work . his final formula is @xmath39 which he called `` a remarkably simple form '' . curiously , in the same paper he also discovered and computed the g.f . for the _ ordered bell numbers_. in 1860 , cayley won a professorship in cambridge and soon became a central figure in british mathematics . few years later , henry martyn taylor ( 18421927 ) became a student in cambridge , where he remained much of his life , working in geometry , mathematical education and politics . in 1882 , he and r. c. rowe published a paper @xcite where they carefully examined the literature of the french school , but gave an erroneous description of the euler segner story . they evidently missed later papers by grunert and liouville , and computed the fuss catalan numbers along similar lines , using the lagrange inversion , see @xcite . cayley continued exploring g.f . methods for a variety of enumerative problems , among over 900 papers he wrote . in 1891 , cayley used g.f . tour de force to completely resolve kirkman s problem @xcite . the _ ballot numbers _ were first defined by catalan in 1839 , disguised as the number of certain triangulations @xcite . we believe this contribution was largely forgotten since catalan gave a formula for the ballot numbers in terms of the catalan numbers , but neither gave a closed formula nor even a table of the first few values . the ballot sequences were first introduced in 1878 by william allen whitworth ( 18401905 ) , a british mathematician , a priest and a fellow at cambridge . he resolved the problem completely by an elegant counting argument and found interesting combinatorial applications @xcite . despite both geographical and mathematical proximity to cayley , he did not notice that the numbers @xmath40 he computed are the catalan numbers . in modern terminology , the _ ballot problem _ was introduced in 1887 by joseph bertrand ( 18221900 ) . in a half a page note @xcite , he defined the probability @xmath41 that in an election with @xmath42 voters , a candidate who gathered @xmath43 votes is always leading the candidate with @xmath11 votes . bertrand announced that one can use induction to show that @xmath44 he famously concludes by saying : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ il semble vraisemblable quun rsultat aussi simple pourrait se dmontrer dune manire plus directe . ( it seems probable that such a simple result could be shown in a more direct way . ) _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ within months , bertrand s protg joseph - mile barbier ( 18391889 ) announced a generalization of @xmath41 to larger proportions of leading votes @xcite , a probabilistic version of the formula for the fuss catalan numbers . dsir andr ( 18401918 ) was a well known french combinatorialist , a former student of bertrand . he published an elegant proof of bertrand s theorem @xcite in the same 1887 volume of _ comptes rendu _ as bertrand and barbier . although he is often credited with the _ reflection method _ , this attribution is incorrect , as andre s proof was essentially equivalent to that by whitworth ( cf . the reflection principle is in fact due to dmitry semionovitch mirimanoff ( 18611945 ) , a russian - swiss mathematician who discovered it in 1923 , in a short note @xcite . andre s original proof was clarified and extended to barbier s proposed generalization in @xcite . we should mention here takcs s thorough treatment of the history of the ballot problem in probabilistic context @xcite , humphreys s historical survey on general lattice paths and the reflection principle @xcite , and bru s historical explanation on how bertrand learned it from ampre @xcite . despite a large literature , for decades the catalan numbers remained largely unknown and unnamed in contrast with other celebrated sequences , such as the fibonacci and bernoulli numbers . yet the number of publications on catalan numbers was growing rapidly , so much that we can not attempt to cover even a fraction of them . here are few major appearances . the first monograph citing the catalan numbers is the _ thorie des nombres _ by douard lucas ( 18421891 ) , published in paris in 1891 . despite the title , the book had a large combinatorial content , including rodrigues s proof of catalan s combinatorial interpretation in terms of brackets and products @xcite . the next notable monograph is the _ lehrbuch der combinatorik _ by eugen netto ( 18481919 ) , published in leipzig in 1901 . this is one of the first combinatorics monographs ; it includes a lot of material on permutations and combinations , and it devotes several sections to the papers by catalan , rodrigues and some related work by schrder @xcite . we should mention that these monographs were more the exception than the rule . many classical combinatorial books do not mention catalan numbers at all , most notably percy macmahon , _ combinatorial analysis _ ( 1915/6 ) , john riordan , _ an introduction to combinatorial analysis _ ( 1958 ) , and h. j. ryser , _ combinatorial mathematics _ ( 1963 ) . a crucial contribution was made by william g. brown in 1965 , when he recognized and collected a large number of references on the `` euler segner problem '' and the `` pfaff fuss problem '' , as he called them @xcite . to denote the catalan numbers . ] from this point on , hundreds of papers involving catalan numbers have been published and all standard textbooks started to include it ( see e.g. ( * ? ? ? * @xmath453.2 ) and ( * ? ? ? * @xmath453 ) for an early adoption ) . some 465 references were assembled by henry w. gould in a remarkable bibliography @xcite , which first appeared in 1971 and revised in 2007 . at about that time , various lists of combinatorial interpretations of catalan numbers started to appear , see e.g. @xcite , appendix 1 and a very different kind of list in @xcite . because of their chaotic history , the catalan numbers have received this name relatively recently . in the old literature , they were sometimes called the _ segner numbers _ or the _ euler segner numbers _ , which is historically accurate as their articles were the first published work on the subject . perhaps surprisingly , we are able to tell exactly who named them _ catalan numbers _ and when . our investigation of this _ eponymy _ is informal , but we hope convincing ( cf . @xcite for an example of a proper historical investigation ) . first , let us discard two popular theories : that the name was introduced by netto in @xcite or by bell in @xcite . upon careful study of the text , it is clear that netto did single out catalan s work and in general was not particularly careful with the references , but never specifically mentioned `` catalan zahlen '' . similarly , eric temple bell ( 18831960 ) was a well known mathematical historian , and referred to `` catalan s numbers '' only in the context of catalan s work . in a footnote , he in fact referred to them as the `` euler segner sequence '' and clarifies the history of the problem . our investigation shows that the credit for naming catalan numbers is due to an american combinatorialist john riordan ( 19031988 ) . he tried this three times . the first two : _ math reviews _ mr0024411 ( 1948 ) and mr0164902 ( 1964 ) went unnoticed . even marshall hall s influential 1967 monograph @xcite does not have the name . but in 1968 , when riordan used `` catalan numbers '' in the monograph @xcite , he clearly struck a chord . although riordan s book is now viewed as somewhat disorganized and unnecessarily simplistic , back in the day it was quite popular . it was lauded as `` excellent and stimulating '' in p.r . stein s review , which continued to say `` combinatorial identities is , in fact , a book that must be read , from cover to cover , and several times . '' we are guessing it had a great influence on the field and cemented the terminology and some notation . we have further evidence of riordan s authorship of a different nature : the _ ngram chart _ for the words `` catalan numbers '' , which searches google books content for this sentence over the years . the search clearly shows that the name spread after 1968 and that riordan s monograph was the first book which contained it . there were three more events on the way to the wide adoption of the name `` catalan numbers '' . in 1971 , henry gould used it in the early version of @xcite , a bibliography of catalan numbers with 243 entries . then , in 1973 , neil sloane gave this name to the sequence entry in @xcite , with riordan s monograph and gould s bibliography being two of the only five references . but the real popularity ( as supported by the ngram chart ) , was achieved after the martin gardner s _ scientific american _ column in june 1976 , popularizing the subject : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ [ catalan numbers ] have [ .. ] delightful propensity for popping up unexpectedly , particularly in combinatorial problems . indeed , the catalan sequence is probably the most frequently encountered sequence that is still obscure enough to cause mathematicians lacking access to n.j.a . sloane s _ handbook of integer sequences _ to expend inordinate amounts of energy re - discovering formulas that were worked out long ago . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ now that we have covered over 200 years of the history of catalan numbers , it is worth pondering if the subject matter is worth the effort . here we defer to others , and include the following helpful quotes , in no particular order . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ it is not exaggerated to say that the catalan numbers are the most prominent sequence in combinatorics . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the catalan numbers are one of the most important sequences of combinatorial numbers , with a large range of occurrences in apparently different counting problems . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the catalan numbers are , next to the binomial coefficients , the best studied of all combinatorial counting numbers . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ catalan numbers are probably the most frequently occurring combinatorial numbers after the binomial coefficients . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ mathematicians are often amazed that certain mathematical objects ( numbers , sequences , etc . ) show up so often . for example , in enumerative combinatorics we encounter the fibonacci and catalan sequences in many problems that seem to have nothing to do with each other . [ .. ] the answer , once again , is our human predilection for triviality . [ .. ] the catalan sequence is the simplest sequence whose generating function is a ( genuine ) algebraic formal power series . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the catalan numbers are a favorite pastime of many amateur ( and professional ) mathematicians . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ catalan numbers are the subject of [ much ] interest ( sometimes known as _ catalan disease _ ) . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ catalan numbers are even more fascinating [ than the _ fibonacci numbers _ ] . like the north star in the evening sky , they are a beautiful and bright light in the mathematical heavens . they continue to provide a fertile ground for number theorists , especially , catalan enthusiasts and computer scientists . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ i d have to say my favorite number sequence is the catalan numbers . [ .. ] catalan numbers just come up so many times . it was well - known before me that they had many different combinatorial interpretations . [ .. ] when i started teaching enumerative combinatorics , of course i did the catalan numbers . when i started doing these very basic interpretations any enumerative course would have some of this i just liked collecting more and more of them and i decided to be systematic . before , it was just a typed up list . when i wrote the book , i threw everything i knew in the book . then i continued from there with a website , adding more and more problems . @xcite _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ * acknowledgments : * . we are very grateful to richard stanley for the suggestion to write this note . we would also like to thank maria rybakova for her help with translation of @xcite , to xavier viennot for showing us the scans of euler s original letter to goldbach , and to peter larcombe for sending me his paper and help with other references . the author was partially supported by the nsf .
planets that transit their stars offer us the opportunity to study the physics of planetary atmospheres and interiors , which may help constrain theories of planet formation . from the photometric light curve , we can measure the planetary radius and also the orbital inclination , which when combined with radial velocity ( rv ) observations , allows us to measure the mass and density of the planet . infrared observations of planets during secondary eclipse can be used to measure the planet s thermal spectrum @xcite , and spectroscopic observations around the time of transit can constrain the composition of the planet s atmosphere through transmission spectroscopy @xcite . additionally , optical observations of the secondary eclipse can probe a planet s albedo @xcite . all of the known transiting planets are hot jupiters or hot neptunes , which orbit so close to their parent stars that the stellar flux plays a major role in heating these planets . in contrast , the detection of transiting planets with longer periods ( @xmath1 ) and consequently lower equilibrium temperatures , would allow us to probe a completely different regime of stellar insolation , one more like that of jupiter and saturn whose energy budgets are dominated by their internal heat . currently , it is not feasible to find long - period , transiting planets from the ground . while rv surveys have found @xmath2 planets with periods @xmath3 , it is unlikely that any of these transit their parent star , given that for a solar - type host , the transit probability is very small , @xmath4 . thus the sample of long - period planets detected by rv will need to at least double before a transiting planet is expected . given the long lead time necessary to detect long - period planets with rv observations , this is unlikely to happen in the near future . furthermore , the expected transit times for long - period systems are relatively uncertain , which makes the coordination and execution of photometric follow - up observations difficult . putting all this together , radial velocity surveys are clearly an inefficient means to search for long - period , transiting planets . ground - based photometric transit surveys are similarly problematic . a long - period system must be monitored for a long time because typically , at least two transits must be observed in order to measure the period of the system . furthermore , if there are a couple of days of bad weather at the wrong time , the transit event will be missed entirely , rendering years of data worthless . thus transit surveys for long - period planets require many years of nearly uninterrupted observations . additionally , many thousands of suitable stars must be monitored to find a single favorably - inclined system . because of its long mission lifetime ( @xmath5 ) , nearly continuous observations , and large number of target stars ( @xmath6 ) , the _ kepler _ satellite @xcite has a unique opportunity to discover long - period transiting systems . not only will _ kepler _ observe multiple transits of planets with periods up to the mission lifetime , but it is also likely to observe single transits of planets with periods longer than the mission lifetime . as the period of a system increases beyond @xmath7 , the probability of observing more than one transit decreases until , for periods longer @xmath8 , only one transit will ever be observed . for periods longer than @xmath8 , the probability of seeing a single transit diminishes as @xmath9 . even with only a single transit observation from _ kepler _ , these long - period planets are invaluable . we show that planets with periods longer than the mission lifetime will likely be detected and can be characterized using the _ kepler _ photometry and precise radial velocity observations . furthermore , this technique can be applied to planets that will transit more than once during the _ kepler _ mission , so that targeted increases in the time - sampling rate can be made at the times of subsequent transits . in @xmath10[sec : np ] we calculate the number of one- and two - transit systems _ kepler _ is expected to observe . we give a general overview of how the _ kepler _ photometry and radial velocity follow - up observations can be combined to characterize a planet in @xmath10[sec : basic ] . we discuss in @xmath10[sec : lc ] the expected uncertainties in the light curve observables , and we relate these to uncertainties in physical quantities , such as the period , that can be derived from the transit light curve . @xmath10[sec : rv ] describes the expected uncertainties associated with the rv curve and how they influence the uncertainty in the mass of the planet . @xmath10[sec : e ] discusses the potential impact of eccentricity on the ability to characterize long - period planets . we summarize our conclusions in @xmath10[sec : conclude ] . details of the derivations are reserved for an appendix ( @xmath10[sec : append ] ) . the number of single - transit events _ kepler _ can expect to find is the integral of the probability that a given system will be favorably inclined to produce a transit , multiplied by the probability that a transit will occur during the mission , convolved with a distribution of semimajor axes , and normalized by the expected frequency of planets and the number of stars that _ kepler _ will observe , @xmath11 here @xmath12 is the total number of transiting systems expected , @xmath13 is the number of stars being monitored , @xmath14 is the probability that the system is favorably inclined to produce a transit ( the transit probability ) , @xmath15 is the probability that a transit will occur during the mission , and @xmath16 is some distribution of the planet semimajor axis @xmath17 , normalized to the expected frequency of planets . assuming a circular orbit , the transit probability is simply , @xmath18 where @xmath19 and @xmath20 are the radius and mass of the parent star , and @xmath21 is the period of the system . we consider both the probability of observing exactly one transit , @xmath22 , and the probability of observing exactly two transits , @xmath23 . these probabilities are @xmath24 @xmath25 thus there will be a range in periods from @xmath7 to @xmath8 where it is possible to get either one or two transits during the mission . if we assume a mission lifetime of @xmath26 years and a solar - type star ( @xmath27 ) , we can find the total probability ( @xmath28 ) of observing a planet that transits _ and _ exhibits exactly one or exactly two transits as a function of period or semimajor axis . these probability distributions as functions of @xmath17 are shown in the middle panel of fig . [ fig : np ] . for @xmath29 , the total probability , @xmath30 , of observing a single transit is @xmath31 thus the probability that the planet will transit _ and _ that a single transit will occur during the mission is generally small for periods at the mission lifetime , and beyond this is a strongly decreasing function of the period , @xmath32 or @xmath33 . the observed distribution of semimajor axes for planets with minimum masses @xmath34 is shown in the top panel of figure [ fig : np ] , where @xmath35 is the mass of the planet , @xmath36 is the orbital inclination , and the data are taken from the `` catalog of nearby exoplanets '' of @xcite . note that we have included multiple - planet systems . we normalized the distribution by assuming that 8.5% of all stars have planets with @xmath34 and @xmath37 @xcite . we can extrapolate this distribution to larger semimajor axes by observing that it is approximately constant as a function of @xmath38 and adopting the value found by @xcite , @xmath39 . using this extrapolation , we predict @xmath40 and @xmath41 of stars have a @xmath34 planet within @xmath42 and @xmath43 , respectively , in agreement with @xcite . we find the number of planets expected for _ kepler _ by convolving this normalized distribution with the probability distribution and multiplying by @xmath44 . we convolve the distribution of semimajor axes with the total transit probability distribution by weighting the probability of observing a transit for each member of the bin to determine an overall probability for the bin . the bottom panel of fig . [ fig : np ] shows the resulting distribution of one- and two - transit systems . integrating this distribution over all semimajor axes out to the limit of current rv observations predicts that _ kepler _ will observe a total of 4.0 single - transit systems and 5.6 two - transit systems during the mission lifetime of @xmath45 . if we include the extrapolation of the observed distribution to larger semimajor axes , the integrated number of single - transit systems increases to 5.7 ; the effect is modest because the probability of observing a transit declines rapidly with increasing semimajor axis ( @xmath46 ) . thus , _ kepler _ is likely to detect at least handful of single - transit events . it is important to bear in mind that this is a lower limit . first , our estimate does not include planets with @xmath47 , primarily because rv surveys are substantially incomplete for low - mass , long - period planets . as we will show , _ kepler _ will be able to constrain the periods of planets with radii as small as that of neptune with the detection of a single transit . indeed , microlensing surveys indicate that cool , neptune - mass planets are common @xcite . second , @xcite and @xcite showed that a distribution in eccentricity increases the probability that a planet will transit its parent star . finally , if the _ kepler _ survey lifetime is extended , the number of detections of long - period transiting planets will also increase . one might assume that a single - transit event must simply be discarded from the sample because the event is not confirmed by a second transit , and the period can not be constrained by the time between successive transits . although we do not expect to see many of them , long - period transiting planets are precious , as the information we could potentially gain by observing these systems when they transit in the future would allow us to greatly enhance our understanding of the physical properties of outer giant planets , and in particular would allow us to compare them directly to our own solar system giants . thus , these single transit events should be saved , if at all possible . . the shaded region indicates @xmath48 , where we expect the sample to suffer from incompleteness . the dashed line indicates our extrapolation of the distribution , found by assuming the constant distribution in @xmath38 from @xcite . ( _ middle _ ) the total transit probability . the solid line shows the probability that a planet transits _ and _ exhibits a single transit during the _ kepler _ mission lifetime . the dotted line shows the same probability for two transits . ( _ bottom _ ) the expected number of one and two - transit events . the probability distribution has been convolved with the observed fraction of planets and multiplied by the number of stars kepler will observe ( 100,000 ) to give the number of single transit systems ( solid line ) and two transit systems ( dotted line ) that can be expected . the dashed line shows the convolution of the single transit probability distribution with the extrapolation to larger semimajor axes . [ fig : np ] ] with a few basic assumptions , _ kepler _ photometry of a transiting planet can provide a period for the system that can be combined with precise radial velocity measurements to estimate a mass for the planet . here we briefly sketch the basics of how these properties can be estimated and what their uncertainties are , and we provide details in the following sections . we assume the planet is much less massive than the star , @xmath49 , and circular orbits . we consider the impact of eccentricity on our conclusions in [ sec : e ] . assuming no limb - darkening and assuming the out - of - transit flux is known perfectly , a transit can be characterized by four observables , namely the fractional depth of the transit @xmath50 , the full - width half - maximum of the transit @xmath51 , the ingress / egress duration @xmath52 , and the time of the center of transit @xmath53 . these can be combined to estimate the instantaneous velocity of the planet at the time of transit , @xmath54 for a circular orbit , @xmath55 can be used to estimate the period of the planet @xmath56 where we have employed kepler s third law assuming that the planet s mass is much smaller than the star s mass . thus , if the stellar density @xmath57 is known , then the planet s period can be estimated from photometry of a single transit @xcite . note that there is a degeneracy between @xmath57 and @xmath21 , so with a single transit @xmath57 must be determined by other means in order to derive an estimate of @xmath21 . @xmath57 can be estimated via spectroscopy combined with theoretical isochrones , or asteroseismology . the total uncertainty in the period is the quadrature sum of the contribution from the estimate of @xmath57 and the contribution from the _ kepler _ photometry ( i.e. , the uncertainty at fixed @xmath57 ) , @xmath58 the contribution from _ kepler _ is dominated by the uncertainty in the ingress / egress time @xmath52 , @xmath59 where we have assumed @xmath60 , and that the number of points taken out of transit is much larger than the number taken during the transit ( and thus the out - of - transit flux is known essentially perfectly ) . here @xmath61 is approximately equal to the total signal - to - noise ratio of the transit , @xmath62 where @xmath63 is the photon collection rate . for _ kepler _ we assume , @xmath64 and thus _ kepler _ will detect a single transit with a signal - to - noise ratio of @xmath65 where @xmath66 is the radius of the planet . for a neptune - size planet with these fiducial parameters , @xmath67 . _ kepler s _ contribution to the uncertainty in the period is , @xmath68 where we have assumed the impact parameter @xmath69 . due to the strong scaling with @xmath66 and relatively weak scaling with @xmath21 , there are essentially two regimes . for @xmath70 , the uncertainty in @xmath21 is dominated by uncertainties in the estimate of @xmath57 , which is expected to be @xmath71 , whereas for @xmath72 , it is very difficult to estimate the period due to the uncertainty in @xmath52 from the _ kepler _ photometry . once the period is known from the photometry , the mass of the planet , @xmath35 , can be measured with radial velocity observations . for a circular orbit , and for observations spread out over a time that is short compared to @xmath21 , the radial velocity of the star can be expanded about the time of transit , and so approximated by the velocity at the time of transit @xmath73 , plus a constant acceleration @xmath74 , @xmath75 where @xmath76 , and @xmath77 is the stellar radial velocity semi - amplitude . thus , @xmath78 where @xmath79 is the surface gravity of the star , and we have assumed @xmath80 . the uncertainty in the planet mass has contributions from three distinct sources : the uncertainty in @xmath79 , which can be estimated from spectroscopy , the uncertainty in @xmath74 , which is derived from rv observations after the transit , and the uncertainties in @xmath81 and @xmath50 , which are derived from _ kepler _ photometry . in fact , the uncertainty in @xmath52 dominates over the uncertainties in @xmath51 and @xmath50 . therefore , we may write , @xmath82 for reasonable assumptions and long - period planets ( @xmath83 ) , we find that the uncertainty in @xmath74 dominates over the uncertainty in @xmath52 . assuming that @xmath84 equally - spaced rv measurements with precision @xmath85 are taken over a time period @xmath86 after the transit , the uncertainty in @xmath74 is , @xmath87 where we have assumed @xmath88 . thus , radial velocity observations combined with _ kepler _ photometry can confirm the planetary nature of a jupiter - sized planet in a relatively short time span . an accurate ( @xmath89 ) measurement of the mass for a jupiter - mass planet , or even a rough characterization of the mass for a neptune - mass planet , will require either more measurements , or measurements with substantially higher rv precision . of course , additional radial velocity observations over a time span comparable to @xmath21 will further constrain the mass and period . in @xmath10[sec : lc ] and @xmath10[sec : rv ] , we derive the above expressions for the uncertainty in the mass and the period of the planet using a fisher information analysis . in the absence of significant limb - darkening , a transit light curve can be approximated by a trapezoid that is described by the five parameters @xmath53 , @xmath51 , @xmath52 , @xmath50 , and @xmath90 , where @xmath90 is the out - of - transit flux . figure [ fig : lc ] shows this simple trapezoidal model and labels the relevant parameters . mathematically , the flux @xmath91 as a function of time @xmath92 is given by , @xmath93)/\tau , & t_1 \le t \le t_2 , \\ f_0-\delta , & t_2 < t < t_3 , \\ f_0-\delta(1-[(t -[t_c - t/2-\tau/2])/\tau ] ) , & t_3 \le t \le t_4 , \\ f_0 , & t_4 < t , \end{array}\right.\ ] ] where @xmath94 @xmath95 are the points of contact . we also define @xmath96 to be the total duration of the observations . this model is fully differentiable and can be used with the fisher matrix formalism to derive exact expressions for the uncertainties in the parameters @xmath53 , @xmath51 , @xmath52 , @xmath50 , and @xmath90 . we note that the definition for @xmath50 we use here differs slightly from the definition we adopted in [ sec : basic ] . here @xmath50 is the depth of the transit ( e.g. , in units of flux ) , rather than the fractional depth . these two definitions differ by a factor of @xmath90 such that @xmath97 . however , when @xmath98 ( as will be the case for _ kepler _ ) , the uncertainty in @xmath90 is negligible , and so one may define @xmath99 , thus making these two parametrizations equivalent . figure [ fig : models ] compares this simple model with the exact model computed using the formalism of @xcite , in this case for a transit of the sun by neptune with a 3.5-year period and an impact parameter of @xmath100 . we see that the trapezoidal model provides an excellent approximation to the exact light curve , and as we will show the analytic parameter uncertainties are quite accurate . we also show the case of significant limb - darkening as expected for a sun - like star observed in the _ r_-band ( similar to the _ kepler _ bandpass ) . in this case , the match is considerably poorer , but nevertheless we will show the analytic parameter uncertainties we derive assuming the simple trapezoidal model still provide useful estimates ( see @xcite for a more thorough discussion ) . we derive exact expressions for the uncertainties in @xmath53 , @xmath51 , @xmath52 , @xmath50 , and @xmath90 by applying the fisher matrix formalism to the simple light curve model ( a detailed explanation of the fisher matrix is given by @xcite ) . the full details of the derivation are given in the appendix ( @xmath10[sec : append ] ) . in the case of the long - period transiting planets that will be observed by _ kepler _ , we can simplify the full expression by making a couple of assumptions . we assume that the flux out of transit , @xmath90 , is known to infinite precision and that @xmath101 . the uncertainties in the remaining parameters are , @xmath102 we tested the accuracy of these expressions using monte carlo simulations . we generate 1000 light curves for a given set of orbital parameters . we assume that the planet orbits a solar - type star in a circular orbit with an impact parameter of 0.2 . we repeat the analysis for four of the solar - system planets : jupiter , saturn , neptune , and earth . we assume the expected photometric precision for _ kepler _ for a stellar apparent magnitude of @xmath103 ( a representative magnitude for _ kepler _ s stellar sample ) , a sampling rate of one per @xmath104 , and a total duration for the observations of @xmath105 hours . then we fit for the parameters @xmath53 , @xmath51 , @xmath52 , @xmath50 , and @xmath90 using a down - hill simplex method . the uncertainty in each parameter is taken to be the standard deviation in the distribution of the fits for that parameter . a comparison of the analytic expressions and the monte carlo simulations is shown in fig . [ fig : sigobs ] , where we have plotted the uncertainties as a function of period for a planet with radius @xmath106 crossing a star with radius @xmath107 at an impact parameter @xmath100 . we find that the fisher matrix approximation breaks down as the number of points during ingress / egress becomes small . we also considered expected uncertainties for the exact uniform source and limb - darkened models for the transit light curve . @xcite give the full solution for a transit involving two spherical bodies and provide code for calculating the transit light curve . they also include a limb - darkened solution . these models may be parametrized by the same observables described in our simplified model . the full and limb - darkened light curves are plotted in fig . [ fig : models ] as the dotted and dashed curves , respectively . we use the limb - darkening coefficients from @xcite closest to observations of the sun ( @xmath108 , @xmath109 , @xmath110 = 0.0 $ ] ) in the r - band . we apply the fisher matrix method to numerical derivatives of these model light curves to compare the uncertainties for the full and limb - darkened solutions with the analytic uncertainties for the trapezoidal model . a comparison of the uncertainties from the different models is shown in fig . [ fig : sigobs ] . we find that the simplified model is a good approximation for the exact , uniform - source transit model . while the comparison is less favorable with the limb - darkening model , the uncertainties do not differ by more than a factor of a few . we also compared how the uncertainties in the transit observables varied with impact parameter @xmath111 for the three models , because varying @xmath111 will affect the relative sizes of @xmath51 and @xmath52 . these comparisons are shown in fig . [ fig : impact ] for neptune in a circular orbit around the sun with a 3.5-year period . as can be seen from the figure , variations in @xmath111 have little impact in the uncertainties in the observables for @xmath112 . ) , the transit depth ( @xmath50 ) , the ingress / egress time ( @xmath52 ) , the fwhm duration of the transit ( @xmath51 ) , the total duration of observations ( @xmath96 ) , and the time of the transit center ( @xmath53 ) . [ fig : lc],width=576 ] and an impact parameter @xmath100 . typical error bars for kepler assuming @xmath113 and 30-min sampling ( solid ) or 20-min sampling ( dotted ) are indicated . [ fig : models],width=576 ] , @xmath51 , @xmath52 , and @xmath50 are plotted versus period for neptune orbiting the sun with an impact parameter of 0.2 . the different line styles indicate the trapezoidal transit model ( solid line ) , the full solution ( dotted line ) , and a model including the effects of limb - darkening ( dashed line ) . monte - carlo simulations of the trapezoidal model using 30-minute sampling are shown as crosses . the gray crosses show simulations for 20-minute sampling . the dotted line in the bottom panel is not visible because it is nearly identical to the solid line . the vertical dotted lines mark where the ingress / egress duration @xmath52 is equal to the sampling rate @xmath114 , as well as where @xmath115 for 30-minute sampling.[fig : sigobs],width=528 ] , @xmath51 , @xmath52 , and @xmath50 are plotted versus impact parameter for neptune orbiting the sun with a period of 3.5 years . the line styles are the same as for fig . [ fig : sigobs ] . as in fig . [ fig : sigobs ] , the dotted line in the bottom panel is not visible because it is nearly identical to the solid line . [ fig : impact],width=528 ] assuming the planet is in a circular orbit , and assuming the stellar density , @xmath57 , is estimated from independent information , @xcite demonstrated that the planet period , @xmath21 , and impact parameter , @xmath111 , can derived from a single observed transit , with sufficiently precise photometry . in fact , provided independent estimates of @xmath20 and @xmath19 are available , then it is also possible to derive the planet radius , @xmath116 , semimajor axis , @xmath17 , and instantaneous velocity at the time of transit , @xmath55 . the uncertainties in these physical parameters can be calculated through standard error propagation . equations for these quantities and their uncertainties are given in eqs . [ eqn : rp][eqn : p ] ( note that we calculate @xmath117 instead of @xmath111 because numerical methods have difficulty when @xmath118 ) . in each case below , the second approximate uncertainty equation is calculated by substituting the variances and covariances of the observables @xmath119 , and @xmath50 , and simplifying under the assumption that @xmath101 . @xmath120\nonumber\\ & \simeq & t\delta q^{-2}\left(\frac{6t^2}{\tau^3}\right)\nonumber\\ & & \nonumber\\ a & = & \frac{g_{\star}}{4}\left(\frac{t\tau}{\sqrt{\delta}}\right)\\ \sigma_a^2 & = & a^2\left[\frac{1}{g_{\star}^2}\sigma_{g_{\star}}^2 + \frac{1}{t^2}\sigma_t^2 + \frac{1}{\tau^2}\sigma_{\tau}^2 + \frac{1}{4\delta^2}\sigma_{\delta}^2+\frac{2}{t\tau}\sigma_{t\tau}^2 -\frac{1}{t\delta}\sigma_{t\delta}^2 - \frac{1}{\tau\delta}\sigma_{\tau\delta}^2 \right]\nonumber\\ & \simeq & a^2\left[\frac{1}{g_{\star}^2}\sigma_{g_{\star}}^2+q^{-2}\left(\frac{6t}{\tau}\right)\right]\nonumber\\ & & \nonumber\\ v_{\rm tr , p } & = & 2r_{\star}\left(\frac{\sqrt{\delta}}{t\tau}\right)^{1/2}\\ \sigma_{v_{\rm tr , p}}^2 & = & v_{\rm tr , p}^2\left[\frac{1}{r_{\star}^2}\sigma_{r_{\star}}^2+\frac{1}{4t^2}\sigma_t^2+\frac{1}{4\tau^2}\sigma_{\tau}^2+\frac{1}{16\delta^2}\sigma_{\delta}^2+\frac{1}{2t\tau}\sigma_{t\tau}^2-\frac{1}{4t\delta}\sigma_{t\delta}^2-\frac{1}{4\tau\delta}\sigma_{\tau\delta}^2\right]\nonumber\\ & \simeq & v_{\rm tr , p}^2\left[\frac{1}{r_{\star}^2}\sigma_{r_{\star}}^2 + q^{-2}\left(\frac{3t}{2\tau}\right)\right]\nonumber\\ & & \nonumber\\ \label{eqn : p } p & = & \frac{\pi g}{4}\left(\frac{4}{3}\pi\rho_{\star}\right)\left(\frac{t\tau}{\sqrt{\delta}}\right)^{\frac{3}{2}}\\ \sigma_p^2 & = & p^2\left [ \frac{1}{\rho^2_{\star}}\sigma_{\rho_{\star}}^2 + \frac{9}{4t^2}\sigma_t^2 + \frac{9}{4\tau^2}\sigma_{\tau}^2 + \frac{9}{16\delta^2}\sigma_{\delta}^2+\frac{9}{2t\tau}\sigma_{t\tau}^2 -\frac{9}{4t\delta}\sigma_{t\delta}^2 - \frac{9}{4\tau\delta}\sigma_{\tau\delta}^2\right]\nonumber\\ & \simeq & p^2\left[\frac{1}{\rho^2_{\star}}\sigma_{\rho_{\star}}^2 + q^{-2}\left(\frac{27t}{2\tau}\right ) \right]\nonumber\end{aligned}\ ] ] of course , the quantities @xmath19 , @xmath79 , and @xmath57 and their uncertainties must be estimated from some other external source of information , such as spectroscopy or theoretical isochrones . we can generally expect that the fractional uncertainties on these quantities will be of order @xmath121 . for long - period ( @xmath122 ) , jupiter - sized planets , @xmath123 , and thus the variances and covariances of the transit observables will be small in comparison to the expected uncertainties on @xmath124 , and @xmath79 . the fractional uncertainties in these derived parameters are plotted in fig . [ fig : deriv ] . we compared the analytic uncertainties given in eqs . [ eqn : rp][eqn : p ] to the monte carlo simulations described in sec . [ sec : models ] . the fractional uncertainty in each parameter from the monte carlo simulations is one - half the range of the middle 68% of the data divided by the median . notice that as @xmath52 decreases from @xmath125 to @xmath114 , the simulations become increasingly disparate from the theoretical expectations . @xmath126 does not show this behavior because it does not depend on @xmath52 . in this regime , as @xmath52 decreases , it becomes increasingly probable that only one point will be taken during ingress leaving @xmath52 relatively unconstrained and increasing its uncertainty and the uncertainty of quantities that depend on it . we now consider in detail the uncertainty in the estimated period . given that @xmath127 , we have that @xmath128 . since @xmath129 and @xmath130 , the contribution to the uncertainty in the period due to the _ kepler _ photometry is a much stronger function of the planet s radius than its period , @xmath131 . as a result , for a fixed stellar radius @xmath19 , the ability to accurately estimate @xmath21 depends almost entirely on @xmath66 . furthermore , because the scaling with @xmath66 is so strong , there are essentially two distinct regimes : for large @xmath66 , the uncertainty in @xmath21 is dominated by the uncertainty in @xmath57 , whereas for small @xmath66 , the period can not be estimated . the boundary between these two regimes is where the contribution to the uncertainty in the period from @xmath57 is equal to the contribution from the _ kepler _ photometry , i.e. where the two terms in the brackets in eq . [ eqn : p ] are equal . this occurs at a radius of , @xmath132^{1/5},\ ] ] assuming @xmath69 . note that @xmath133 is a weak or extremely weak function of all of the parameters except @xmath19 . [ fig : contour ] illustrates this point by showing contours of constant uncertainty in the period as a function of the radius and period of the planet . the calculations are for systems with @xmath134 , @xmath100 , and @xmath113 . this figure shows that for planets larger than the radius of neptune , the uncertainty in the period will be dominated by the uncertainty in the inferred stellar density , whereas for planets much smaller than neptune , an accurate estimate of the period from the _ kepler _ light curve will be impossible . as shown above , the uncertainty in @xmath21 is dominated by the uncertainty in @xmath57 and @xmath52 . our ability to determine @xmath52 , and its uncertainty , depends both on its length and how many points we have during the ingress or egress . the length of @xmath52 depends on @xmath55 ( which is a proxy for @xmath21 in the circular case ) , @xmath111 , and @xmath66 . these properties are intrinsic to the system , but may be derived from the observables . we can explore how the contours shown in fig . [ fig : contour ] vary with impact parameter . figure [ fig : impact ] shows that a system with a larger impact parameter will have a smaller fractional uncertainty in @xmath52 . thus , a system with a larger impact parameter would have a smaller fractional uncertainty in @xmath21 at fixed period and planet radius . the other effect that influences the contours is the sampling of @xmath52 . as the number of points taken during the ingress / egress becomes small , the fractional uncertainty in @xmath52 , and hence in @xmath21 , increases . in particular , if only one point is taken during the ingress , then the duration is relatively unconstrained . the probability of taking only one point during the ingress increases linearly from 0 to 1 as @xmath52 decreases from @xmath135 to @xmath114 . in fig . [ fig : deriv ] , the monte carlo simulations for the fractional uncertainty in @xmath21 diverge from the simple model over this range . we indicate this region by the shaded portion of fig . [ fig : contour ] . where @xmath136 , the fractional uncertainty in @xmath21 can be several times that predicted by the theoretical calculations , but these uncertainties converge as the shading gets lighter towards @xmath115 . below the line where @xmath136 , the uncertainty in @xmath52 increases rapidly , but fortunately , in much of this regime , _ kepler _ will observe multiple transits , and this analysis will be unnecessary . as shown in the figure , a faster sampling rate , such as 1 per 20 min , significantly expands the parameter space over which our theoretical uncertainties are valid . , @xmath17 , @xmath21 , @xmath55 , and @xmath117 are plotted versus period for neptune orbiting the sun with an impact parameter of 0.2 . the line styles are the same as for fig . [ fig : sigobs ] . the vertical dotted lines mark where the ingress / egress duration @xmath52 is equal to the sampling rate @xmath114 , as well as where @xmath115 for 30-minute sampling . [ fig : deriv],width=528 ] changes with @xmath137 magnitude ; it shows the 0.20 contour for @xmath138 and @xmath100 . the diagonal black dotted line represents the boundary below which the assumptions of fisher matrix break down ( the ingress / egress time @xmath52 is roughly equal to the sampling rate of 1/30 min ) . thus , the contours below this boundary are shown as dash - dotted lines . the region between @xmath115 and @xmath136 is shaded to indicate the increasingly probability of obtaining only one point during ingress or egress leading to an uncertainty in @xmath21 that is larger than theoretical expectations . the diagonal gray dotted line indicates the @xmath136 boundary for 20-min sampling . the vertical dotted line shows the mission lifetime of _ kepler _ ( @xmath139 ) . solar system planets are indicated . [ fig : contour],width=576 ] the mass of the planet comes from sampling the stellar radial velocity curve soon after the transit is observed . near the time of transit we can expand the stellar radial velocity , @xmath140\approx v_0 - a_{\star}(t - t_c ) , \ ] ] where @xmath73 is the systemic velocity , @xmath141 is the stellar radial velocity semi - amplitude , and @xmath142 is the stellar acceleration . because the planet is known to transit and has a long period , we assume @xmath143 . a fisher matrix analysis of the linear form of eq . [ eqn : rv ] gives the estimated uncertainty in @xmath74 to be @xmath144 where @xmath85 is the radial velocity precision , @xmath145 is the total time span of the radial velocity observations , and @xmath84 is the number of observations , and we have assumed @xmath146 and that the observations are evenly spaced in time . the details of this derivation are given in the appendix ( @xmath10[sec : append ] ) . equations for the mass of the planet and the uncertainty are @xmath147\nonumber\\ & \simeq & m_{\rm p}^2\left[\frac{4}{g_{\star}}\sigma_{g_{\star}}^2+\frac{1}{a_{\star}^2}\sigma_{a_{\star}}^2+q^{-2}\left(\frac{24t}{\tau}\right)\right ] . \nonumber\end{aligned}\ ] ] the approximate expression for the uncertainty in the mass is taken in the limit @xmath101 . furthermore , the @xmath148 term can be neglected in most cases because , as we showed for the corresponding term in uncertainty in the period , it will be very small for planets with radii larger than that of neptune . the scaling for the @xmath149 term is given in eq . [ eqn : sigbscale ] . contours of constant fractional uncertainty in the mass of the planet are shown in fig . [ fig : mpplot ] as a function of @xmath21 and @xmath35 , assuming @xmath150 radial velocity measurements are taken over @xmath151 mos with a precision @xmath152 . it shows that the planet mass can be estimated to within a factor of two over this time period , thus establishing the planetary nature of the transiting object . by doubling the length of observations , one can put stronger constraints on the mass of the planet . there are two points to bear in mind when applying this estimate for the uncertainty in the mass of the planet . first , for @xmath153 , the @xmath154 term is no longer small . the second consideration is that as time progresses away from the time of transit , the straight line approximation to the radial velocity curve will break down . in that case , the period will begin to be constrained by the radial velocity curve itself , and the uncertainties should be calculated from a fisher matrix analysis of the full expression for the radial velocity curve with three parameters : @xmath73 , @xmath141 , and @xmath21 . as a function of @xmath35 and @xmath21 . the solid lines show the result for 20 radial velocity measurements with precision of @xmath155 taken over a period of 3 months after the transit . the dashed line shows the contour for @xmath156 for 40 observations taken over 6 months . the dotted line indicates the mission lifetime of 3.5 years . the positions of jupiter and saturn are indicated . [ fig : mpplot],width=576 ] the results presented in the previous sections assumed circular orbits . given that the average eccentricity @xmath157 of planets with @xmath158 is @xmath159 , this is not necessarily a good assumption . in this section , we assess the effects of non - zero eccentricities on the ability to characterize single - transit events detected by _ kepler_. we note that our discussion has some commonality with the study of @xcite , who discuss the possibility of characterizing the orbital eccentricities of transiting planets with photometric observations . however , our study addresses this topic from a very different perspective . in [ sec : basic ] we demonstrated that , under the assumption that @xmath160 , the planet period @xmath21 and mass @xmath35 can be inferred from two observables : the velocity of the planet at the time of transit @xmath161 , and the projected acceleration of the star @xmath74 . in the case of a non - zero eccentricity , there are two additional unknown parameters : the eccentricity @xmath157 and the argument of periastron of the planet @xmath162 . thus , with only two observables ( @xmath161 and @xmath74 ) and four unknowns ( @xmath163 ) , it is not possible to obtain a unique solution . although a unique solution to the planet parameters does not exist using the observable parameters alone , it may nevertheless be possible to obtain an interesting constraint on @xmath21 and/or @xmath35 by adopting reasonable priors on @xmath157 and @xmath162 . from the transit observables , one can estimate the velocity of the planet at the time of transit ( eq . [ eqn : vtr ] ) . in the case of @xmath164 , this is given by , @xmath165 where here and throughout this section we assume @xmath166 . solving for @xmath21 , @xmath167 ^ 3.\ ] ] for a fixed value of @xmath157 , the inferred period relative to the assumption of a circular orbit has extremes due to the unknown value of @xmath162 of @xmath168 taking a typical value for the eccentricity of @xmath169 , this gives a range of inferred periods of @xmath170 relative to the assumption of a circular orbit . in fact , because the probability that a planet with a given @xmath17 transits its parent star depends on @xmath157 and @xmath162 @xcite , a proper bayesian estimate , accounting for these selection effects , might reduce the range of inferred values for @xmath21 significantly . what can be learned from rv observations immediately after the transit ? expanding the stellar projected velocity around the time of transit , we can write , @xmath171 the projected acceleration of the star at the time of transit is , @xmath172 this can be combined with @xmath55 to derive the mass of the planet , @xmath173 thus , for a fixed eccentricity , the inferred mass relative to the assumption of a circular orbit has extremes of , @xmath174 which for @xmath169 yields a range of @xmath175 , which is generally _ smaller _ than the contribution to the uncertainty in @xmath35 due to the measurement uncertainty in @xmath74 for our fiducial case of @xmath176 and @xmath177 ( see eq . [ eqn : sigbscale ] ) . we may also consider what can be learned if it is possible measure the curvature of the stellar radial velocity variations immediately after transit . the projected stellar jerk is given by , @xmath178 this can be combined with @xmath161 and @xmath74 to provide an independent constraint on a combination of the eccentricity and argument of periastron , @xmath179 unfortunately , it will be quite difficult to measure @xmath180 . using the same fisher formalism as we used to estimate the uncertainty in @xmath74 ( [ sec : append ] ) , assuming @xmath84 evenly - spaced rv measurements with precision @xmath85 , taken over time span @xmath86 after the transit , the uncertainty in @xmath180 is , @xmath181 where we have assumed @xmath88 . for our fiducial parameters , the fractional uncertainty in @xmath180 is , @xmath182 where we have approximated @xmath183 . we conclude that , in order to obtain interesting constraints on the eccentricity of long - period planets detected by _ kepler _ , higher - precision rv measurements taken over a baseline comparable to @xmath184 will be needed . the discovery of long - period transiting planets along with subsequent follow - up observations would greatly enhance our understanding of the physics of planetary atmospheres and interiors . such planets would allow us to gather constraints in a regime of parameter space currently only occupied by our own giant planets , namely planets whose energy budgets are dominated by their residual internal heat , rather than by stellar insolation . these constraints , in turn , might provide new insights into planet formation . in this paper , we demonstrated that it will be possible to detect and characterize such long - period planets using observations of single transits by the _ kepler _ satellite , combined with precise radial velocity measurements taken immediately after the transit . indeed , these results can be generalized to any transiting planet survey using the scaling relations we provide , and it may be particularly interesting to apply them to the _ convection , rotation & planetary transits ( corot ) _ mission @xcite . we calculated that _ kepler _ will see a few long - period , single transit events and showed that , for circular orbits , the period of the system can be derived from the _ kepler _ light curve of a single event . we derived an expression for the uncertainty in this period and showed that it is dominated by the uncertainty in the stellar density derived from spectroscopy ( which we assume is @xmath185 ) for planets with radii larger than the radius of neptune , rather than being dependent on the properties of the transit itself . this method can also be applied to planets that _ kepler _ will observe more than once , so that the second time a planet is expected to transit _ kepler _ can make a selective improvement to its time sampling to better characterize the transit . we have also shown how the mass of the planet can be constrained by acquiring precise ( @xmath186 ) radial velocity measurements beginning shortly after the transit occurs . we have shown that 20 measurements over 3 months can measure the mass of a jupiter - sized object to within a factor of a few and that extending those observations to 40 measurements over 6 months significantly reduces the uncertainty in the mass . knowing the mass to within a factor of a few in such a short time can distinguish between brown dwarfs and planets rapidly and allow us to maximize our use of radial velocity resources . we explored the effect of eccentricity on the ability to estimate the planet mass and period . allowing for a non - zero eccentricity adds two additional parameters , and as a result it is not possible to obtain a unique solution for the planet mass , period , eccentricity , and argument of periastron . however , by adopting a reasonable prior on the eccentricity of the planet , the period and mass of the planet can still be estimated to within a factor of a few . detailed characterization of the planet properties will require precision rv measurements obtained over a duration comparable to the period of the planet . thus , in the interest of `` getting the most for your money , '' we have shown that the sensitivity of _ kepler _ extends to planets with periods beyond its nominal mission lifetime . with the launch of the _ kepler _ satellite , we are poised to discover and characterize several long - period transiting systems , provided that we are prepared to look for them . we are grateful to josh carter , jason eastman , eric ford , andy gould , matt holman , yoram lithwick , josh winn , and andrew youdin for helpful discussions . we thank the referee ron gilliland for constructive comments . jcy is supported by a dean s graduate enrichment fellowship from the ohio state university . 19 natexlab#1#1 , a. 2003 , advances in space research , 31 , 345 , j. w. 2007 , , 119 , 986 , g. , borucki , w. j. , & koch , d. 2005 , new astronomy review , 49 , 478 , j .- p . , bennett , d. p. , fouqu , p. , williams , a. , dominik , m. , jrgensen , u. g. , kubas , d. , cassan , a. , coutures , c. , greenhill , j. , hill , k. , menzies , j. , sackett , p. d. , albrow , m. , brillant , s. , caldwell , j. a. r. , calitz , j. j. , cook , k. h. , corrales , e. , desort , m. , dieters , s. , dominis , d. , donatowicz , j. , hoffman , m. , kane , s. , marquette , j .- b . , martin , r. , meintjes , p. , pollard , k. , sahu , k. , vinter , c. , wambsganss , j. , woller , k. , horne , k. , steele , i. , bramich , d. m. , burgdorf , m. , snodgrass , c. , bode , m. , udalski , a. , szymaski , m. k. , kubiak , m. , wic eckowski , t. , pietrzyski , g. , soszyski , i. , szewczyk , o. , wyrzykowski , . , paczyski , b. , abe , f. , bond , i. a. , britton , t. r. , gilmore , a. c. , hearnshaw , j. b. , itow , y. , kamiya , k. , kilmartin , p. m. , korpela , a. v. , masuda , k. , matsubara , y. , motomura , m. , muraki , y. , nakamura , s. , okada , c. , ohnishi , k. , rattenbury , n. j. , sako , t. , sato , s. , sasaki , m. , sekiguchi , t. , sullivan , d. j. , tristram , p. j. , yock , p. c. m. , & yoshioka , t. 2006 , , 439 , 437 , w. , koch , d. , boss , a. , dunham , e. , dupree , a. , geary , j. , gilliland , r. , howell , s. , jenkins , j. , kondo , y. , latham , d. , lissauer , j. , & reitsema , h. 2004 , in esa special publication , vol . 538 , stellar structure and habitable planet finding , ed . f. favata , s. aigrain , & a. wilson , 177182 , c. j. 2008 , arxiv:0801.2579 , 801 , r. p. , wright , j. t. , marcy , g. w. , fischer , d. a. , vogt , s. s. , tinney , c. g. , jones , h. r. a. , carter , b. d. , johnson , j. a. , mccarthy , c. , & penny , a. j. 2006 , , 646 , 505 , j. a. , yee , j. c. , eastman , j. , gaudi , b. s. , & winn , j. n. 2008 , arxiv:0805.0238 , 805 , d. , allen , l. e. , megeath , s. t. , torres , g. , alonso , r. , brown , t. m. , gilliland , r. l. , latham , d. w. , mandushev , g. , odonovan , f. t. , & sozzetti , a. 2005 , , 626 , 523 , d. , brown , t. m. , noyes , r. w. , & gilliland , r. l. 2002 , , 568 , 377 , a. 2000 , , 363 , 1081 , a. , butler , r. p. , marcy , g. w. , vogt , s. s. , wright , j. t. , & fischer , d. a. 2008 , arxiv:0803.3357 , 803 , d. , seager , s. , richardson , l. j. , & harrington , j. 2005 , , 434 , 740 , e. b. , quinn , s. n. , & veras , d. 2008 , arxiv:0801.2591 , 801 , a. 2003 , arxiv : astro - ph/0310577 , a. , udalski , a. , an , d. , bennett , d. p. , zhou , a .- y . , dong , s. , rattenbury , n. j. , gaudi , b. s. , yock , p. c. m. , bond , i. a. , christie , g. w. , horne , k. , anderson , j. , stanek , k. z. , depoy , d. l. , han , c. , mccormick , j. , park , b .- , pogge , r. w. , poindexter , s. d. , soszyski , i. , szymaski , m. k. , kubiak , m. , pietrzyski , g. , szewczyk , o. , wyrzykowski , . , ulaczyk , k. , paczyski , b. , bramich , d. m. , snodgrass , c. , steele , i. a. , burgdorf , m. j. , bode , m. f. , botzler , c. s. , mao , s. , & swaving , s. c. 2006 , , 644 , l37 , k. & agol , e. 2002 , , 580 , l171 , j. f. , matthews , j. m. , seager , s. , miller - ricci , e. , sasselov , d. , kuschnig , r. , guenther , d. b. , moffat , a. f. j. , rucinski , s. m. , walker , g. a. h. , & weiss , w. w. 2007 , arxiv:0711.4111 , 711 , s. & malln - ornelas , g. 2003 , , 585 , 1038 the fisher matrix formalism is a simple way to estimate the uncertainties in the parameters , @xmath187 , of a model @xmath188 , that is being fit to a series of measurements @xmath189 with measurement uncertainties @xmath190 . the covariance of @xmath191 with @xmath192 is given by the element @xmath193 of the covariance matrix @xmath194 , where @xmath195 and the entries of @xmath111 are given by , @xmath196 where @xmath197 is the number of data points . in the limit of infinite sampling ( @xmath198 ) and fixed precision , @xmath199 , @xmath200 where the interval of interest is given by @xmath201 $ ] , and in this case @xmath96 is the total duration of observations . thus , if the partial derivatives of a model with respect to its parameters are known , then the uncertainties in those parameters can be estimated @xcite . for the simplified transit model described in @xmath10[sec : models ] , the observable parameters are @xmath202 $ ] . for a sampling rate @xmath203 , the @xmath111 matrix is , @xmath204 the covariance matrix is , @xmath205 where @xmath206 is the duration of the flat part of the eclipse and @xmath207 is the time spent out of eclipse . thus , the uncertainty on the ingress / egress time , @xmath208 , is given by @xmath209 define @xmath210 . then , in the limit @xmath60 , the uncertainties in the observable parameters are given by , @xmath211 note that @xmath61 is approximately the total signal - to - noise ratio of the transit . assuming that the photometric uncertainties are limited by photon noise , we have that @xmath212 , where @xmath63 is the photon collection rate . this recovers the expression for @xmath61 in [ sec : basic ] . consider @xmath84 measurements of @xmath216 , each with precision @xmath85 , taken at times @xmath217 . fitting these measurements to the linear model in eq . [ eqn : vexpand ] , we can estimate the uncertainties in the parameters @xmath73 and @xmath74 using the fisher matrix formalism , @xmath218 .\ ] ] the covariance matrix is the inverse of @xmath111 , @xmath219-\left[\displaystyle\sum_{k=1}^{n}(t_k - t_c)\right]^2}\left[\begin{array}{cc } \displaystyle\sum_{k=1}^{n}(t_k - t_c)^2 & \displaystyle\sum_{k=1}^{n}(t_k - t_c)\\ \displaystyle\sum_{k=1}^{n}(t_k - t_c ) & n \end{array}\right ] .\ ] ] if the points are evenly spaced by @xmath220 @xmath221 and thus , @xmath222 .\end{aligned}\ ] ] the uncertainty in the projected stellar acceleration is , @xmath223 in the limit as @xmath224 , the covariance matrix reduces to , @xmath225 .\end{aligned}\ ] ] defining the total length of observations as @xmath226 , we can also write , @xmath227,\end{aligned}\ ] ] and so the uncertainty @xmath74 in the limit of @xmath224 becomes , @xmath228
measurements of the masses and radii of neutron stars provide unique information on the equation of state of matter at densities above the nuclear saturation density . extremely accurate mass measurements in binary systems can be obtained through dynamical methods ( see , e.g. , stairs 2006 ; demorest et al . however , without independent determinations of the radii of these neutron stars , these measurements provide only limits for the equation of state . in contrast , an accurate measurement of the radius of even a single neutron star can provide strong constrains on the equation of state of neutron star matter ( lattimer & prakash 2001 ; zel & psaltis 2009 ; read et al . 2009 ) . thermonuclear x - ray bursts observed from low mass x - ray binaries have proven to be one of the most suitable events that can be used to measure both the radii and the masses of neutron stars ( see , e.g. , van paradijs 1978 , 1979 ; damen et al . 1990 ; lewin , van paradijs , & taam 1993 ; zel 2006 ) . using high signal - to - noise x - ray data and time - resolved spectroscopy allows a measurements of the apparent radius during the cooling segments of the bursts . in addition , time - resolved spectroscopy during a photospheric radius expansion burst enables a measurement of the eddington luminosity for each source . combining these two measurements leads to an uncorrelated determination of the neutron star mass and radius . thanks to the rossi x - ray timing explorer ( rxte ) , which provided more than 15 years of observations of bursts ( galloway et al . 2008 ) , it has now become possible to not only make neutron star mass and radius measurements using observations of specific systems ( see , e.g. , zel , gver , & psaltis 2009 ) but also to quantify the systematic uncertainties in these measurements . in paper i of these series ( gver , psaltis , & zel 2012a ) , we analyzed more than 13,000 x - ray spectra from 12 different sources and showed that the errors in the spectroscopic determination of neutron - star radii due to systematic effects in the cooling tails of x - ray bursts are in the range of @xmath23%@xmath08% . in paper ii ( gver , zel , & psaltis 2012b ) , we used all the archival photospheric radius expansion bursts and found that for sources from which a large number of bursts are observed , systematic uncertainties in the determination of the eddington flux are at the 5%@xmath010% level . in this paper , we focus on a third possible source of systematic uncertainty , which is the absolute flux calibration of the proportional counter array ( pca ) on board rxte . absolute flux uncertainties affect the normalization of the spectra obtained during bursts and therefore the measurement of both the apparent radii and of the eddington flux . determining the absolute flux calibration of an x - ray detector requires a major effort from any instrument team and all x - ray observatories put large amount of resources into in - flight calibrations of their detectors . however , these efforts are hindered by the fact that almost all x - ray sources in the sky that may be used as standard candles often show unpredicted variability over both long and short time scales ( see , e.g. , wilson - hodge et al . 2011 , and references therein for the recent activity of the crab nebula ) . additionally , complications due to significantly different responses of different detectors , both in terms of effective areas and the energy ranges that the detectors are sensitive to , become a major hurdle and restrict the inter - calibration efforts . furthermore , the often unknown intrinsic spectral shape of the astronomical objects being used for the calibration makes the problem even more complex . simultaneous observations of sources with well known intrinsic spectra by two or more x - ray detectors that have different energy responses allow us to better understand and quantify the systematic uncertainties in the absolute flux calibrations of individual detectors . in the early days of high energy astrophysics , the crab nebula was seen as the ideal stellar source for the calibration of various instruments . compiling a number of rocket , balloon , and satellite observations performed between 1964 and 1972 , toor & seward ( 1974 ) concluded that the diffuse x - ray emission should be steady at least for the time period they analyzed , since any variation in the pulsar emission larger than 10% at a few kev would have been detected . with these limits , they assumed that the crab is a steady source and provided a `` universal '' power - law amplitude and photon - index with their relevant uncertainties . in the following years , these values were often cited and used as a reference point to calibrate the performance of various instruments . norgaard - nielsen et al . ( 1994 ) even proposed a specific mission called _ excalibur _ to continuously monitor the crab nebula using two pinhole cameras one with a ccd and another with a proportional counter to measure the spectrum of the nebula with an accuracy @xmath32% over the energy range 0.8@xmath020 kev . such a mission could , in principle , be a reliable inter - calibration reference for subsequent detectors . using the crab as a standard source , kirsch et al . ( 2005 ) compared the xmm - newton / epic - pn measurements of the spectrum of crab with those from all the other modern x - ray satellites . in particular , they found that although the photon index inferred from xmm - newton / epic agrees within errors with the toor & seward ( 1974 ) value , the normalization derived from epic - pn is significantly lower than that from the rest of the instruments . they also noted that , with the new generation of instruments , crab is becoming a source that is too bright and too extended to be used as a standard calibration source . in a similar , later study , weisskopf et al . ( 2010 ) investigated possible departures from a single power - law of the intrinsic spectrum of crab for rosat / pspc , rxte / pca , and the xmm - newton / epic - pn , which would affect the instrument calibrations . using the high statistical quality of the rxte / pca data , these authors concluded that they would have been able to detect significant deviations from a simple power - law spectrum but they did not find any evidence for such a deviation . with the advancement of imaging detectors and considering the brightness of the crab for the onboard ccds , searches for new calibration sources and techniques started . in order to coordinate and improve the efforts on inter - calibration , a group of instrument teams formed the international astronomical consortium for high - energy calibration ( iachec , sembay et al . 2010 ) . one of the recent results of this group demonstrates that in the 2.0@xmath07.0 kev range , epic - pn measures systematically lower fluxes than the chandra acis , by 11.0@xmath10.5% ( nevalainen et al . 2010 ) , using a sample of 11 nearby clusters of galaxies . the difference is smaller yet still significant when acis is compared with epic mos1 and mos2 , with differences at 3.0@xmath10.5% and 6.0@xmath10.5% , respectively ( nevalainen et al . 2010 ) . the spectroscopic measurements of neutron star masses and radii have been predominantly performed with rxte . launched in 1995 and operated until 2012 , rxte / pca has always been calibrated against the crab nebula ( jahoda et al . 2006 ; shaposnikov et al . 2012 ) together with an onboard calibration source used mainly for channel - to - energy conversion . throughout the mission , any possible variations in the crab , as the one reported by wilson - hodge et al . ( 2011 ) , were assumed to be averaged out ( shaposnikov et al . 2012 ) ; this issue will be discussed further in section [ variable_source ] . because the pca was intrinsically calibrated against the crab and because it is a non - imaging instrument , inter - calibration of the pca has always been a challenge , especially as more and more sources such as clusters of galaxies or similarly soft and relatively dim objects ( i.e. , rx j1856.5@xmath03754 ) started to be used as standard candles . time resolved x - ray spectra extracted during thermonuclear x - ray bursts observed from low mass x - ray binaries offer a different opportunity in determining the systematic uncertainties caused by calibrations of different detectors . since the early studies of x - ray bursts , it has been shown that the burst x - ray spectra can statistically be modeled with blackbody functions ( see , e.g. , swank et al . 1977 ; lewin , van paradijs , & taam 1993 ; galloway et al . 2008 ) . in paper i , we analyzed more than 13,000 x - ray spectra extracted from the cooling tails of x - ray bursts observed from 12 different sources and showed that a planckian function can statistically fit the vast majority of the observations . the high count rates of rxte and the large number of bursts allowed us to show that any deviation from a planckian spectrum is limited to @xmath4 in the 2.5@xmath025.0 kev range . this almost unique characteristic of burst spectra make them ideal for testing our ability to convert countrate spectra to bolometric fluxes . furthermore , because the obtained spectral parameters are directly related to neutron star mass - radius measurements , results obtained from an analysis of x - ray burst data can be directly converted to uncertainties in the measurements of neutron star masses and radii . using simultaneous rxte , chandra x - ray observatory ( chandra ) , and xmm-_newton _ observations of x - ray bursts from gs 1826@xmath0238 , we quantify the differences in the measurements of the apparent surface area of the thermal emission when using different detectors with independent calibrations . we describe our method in section 2 , which is based on predicting the observed chandra and xmm-_newton _ countrates using time resolved spectral parameters obtained from rxte / pca . section 3 details the analysis of the data from different x - ray satellites . finally , in sections 4 and 5 we present our results and discuss their implications . the rapid evolution of the effective temperature during the cooling tails of thermonuclear x - ray bursts demands very short exposure time observations for direct comparison of spectral data that are obtained simultaneously . the short exposures require very high countrates and , hence , detectors with large photon collecting areas . this has been a unique property of rxte / pca but can not be achieved by the other instruments that we are going to compare its results against . therefore , we formulated an approach that does not rely on directly comparing the extracted spectra of x - ray bursts simultaneously observed with two ( or more ) x - ray satellites . instead , we first analyze the rxte data by performing time resolved x - ray spectroscopy . we extract spectra for short intervals ( 0.25@xmath02.0 s depending on the countrate ) and fit these data with blackbody functions . based on the obtained @xmath5 values , we then calculate the likelihood for a wide range of spectral parameters that cover the @xmath15@xmath6 range of the best fit values . we use this range of individual parameter pairs ( blackbody temperature and normalization ) to create simulations of chandra or xmm-_newton _ x - ray spectra and then to calculate the total rxte - predicted countrate in a given energy range . we calculate the uncertainty in these countrates by taking into account the posterior likelihoods of the rxte / pca parameters for each simulation . finally , we compare this predicted countrate with the countrate actually observed by chandra or xmm-_newton _ in the same energy range , taking into account all the associated uncertainties . assuming that there are no calibration uncertainties , the chandra and/or xmm-_newton _ count rates that were predicted by analysing the rxte observation should be in perfect agreement with the actual count rates observed with these satellites . a systematic offset on the other hand , would imply a discrepancy in the overall calibration of either or both detectors . it is evident that this method allows for the detection of a systematic uncertainty caused by the overall flux calibration of detectors and specifically in the determination of the normalization in a particular energy range . however , the method does not allow us to constrain any calibrational uncertainties in the determination of the spectral shape . we searched the archival rxte observations to find the best source that can be used for this purpose . an ideal source should be a frequent burster with a number of simultaneous observations and predictable burst properties . two low mass x - ray binaries in the archive conform to these conditions : the so - called clockwork burster gs 1826@xmath0238 and 4u 1636@xmath0536 . the problem with the latter source is that its bursts do not last long enough ( only about 20 s on average ) to obtain sufficient amount of data to infer any systematic trends during the cooling tails of x - ray bursts . on the other hand , with bursts that last on average 130 s , gs 1826@xmath0238 provides a significant amount of counts to be used for this purpose . gs 1826@xmath0238 was first discovered with the ginga satellite ( makino et al . 1988 ; tanaka 1989 ) . first conclusive detections of thermonuclear x - ray bursts from this source came from bepposax observations ( ubertini et al . 1997 ) , although x - ray bursts may have been observed from this source earlier by oso 7 ( markert et al . 1977 ) and oso 8 ( becker et al . 1976 ) . using 2.5 years of monitoring observations by bepposax , during which 70 x - ray bursts were detected , ubertini et al . ( 1999 ) reported that the x - ray bursts show a quasi - periodicity of 5.76 hr , a remarkable characteristic of this source . later , using 260 x - ray bursts observed with the wide field camera ( wfc ) on board bepposax over 6 years , cornelisse et al . ( 2003 ) showed that the burst recurrence time is indeed stable but decreases as the persistent flux increases . a total of 40 x - ray bursts were observed with the rxte between 1997 - 2003 ( galloway et al . 2008 ) . during this time interval , the persistent flux steadily increased as the burst recurrence time decreased ( galloway et al . another remarkable feature of the x - ray bursts observed from this source is their similarity . although the initial decay time - scale and the burst fluence show slight variations with persistent flux , the observed x - ray burst light curves are nearly identical and reproducible ( galloway et al . 2004 ) . the nearly identical light curves of individual bursts , predictable recurrence times , and , finally , already existing simultaneous observations with other x - ray satellites ( chandra and xmm- _ newton _ ) make gs 1826@xmath0238 an ideal target to understand and quantify the systematic uncertainties in the measurement of neutron - star masses and radii introduced by the uncertainties in the absolute flux calibration of rxte / pca . in this study , we use data from three different x - ray satellites . gs 1826@xmath0238 was observed simultaneously by chandra and rxte for the first time in 2002 , where three x - ray bursts were recorded . we present the analysis of these bursts below . furthermore , rxte observed the system in 2003 simultaneously with xmm-_newton_. during this observation , 7 x - ray bursts were detected with rxte . we analyze here 5 of these events because , during one of the remaining bursts , xmm-_newton _ was affected by high energy particle background ( burst # 28 ) while the other burst was only partially observed and did not have high time resolution data with rxte / pca ( burst # 29 ) . we present a log of these bursts in table [ bursts ] . cccc 22 & 52484.41775 & 70044 - 01 - 01 - 000 & chandra + 23 & 52484.56684 & 70044 - 01 - 01 - 00 & chandra + 24 & 52485.00672 & 70044 - 01 - 01 - 02 & chandra + 26 & 52736.53279 & 80048 - 01 - 01 - 00 & xmm-_newton _ + 27 & 52736.66488 & 80048 - 01 - 01 - 010 & xmm-_newton _ + 28 & 52736.80048 & 80048 - 01 - 01 - 010 & xmm-_newton _ + 29 & 52738.34118 & 80048 - 01 - 01 - 04 & xmm-_newton _ + 30 & 52738.47686 & 80048 - 01 - 01 - 04 & xmm-_newton _ + 31 & 52738.61013 & 80048 - 01 - 01 - 07 & xmm-_newton _ + 32 & 52738.74636 & 80048 - 01 - 01 - 07 & xmm-_newton _ [ bursts ] the rxte observations were analyzed following the method detailed in paper i. time resolved , 2.5@xmath025.0 kev x - ray spectra were extracted from science event mode data ( 125 @xmath7s time resolution and 64 energy channels ) using the ftool _ seextrct _ from all of the rxte / pca layers . for each burst , spectra were integrated over 0.25 , 0.5 , 1 or 2 s time intervals so that the total number of counts in each spectrum remain roughly the same while the countrate decreases during a burst . a 16 s x - ray spectrum was extracted from a time interval just prior to each burst , as the spectrum of the persistent emission , and subtracted from the x - ray burst spectra as background . we took the pca deadtime correction into account following the procedure described in paper i. response matrix files were created for each burst separately using the pcarsp version 11.7 , heasoft 6.7 and heasarc s remote calibration database . x - ray spectral analysis was performed using the interactive spectral interpretation system ( isis ) , version 1.4.9 - 55 ( houck & denicola 2000 ) and , for each fit , a systematic uncertainty of 0.5% was included as suggested by the rxte calibration team . we note that we also investigated the possibility of photon pile - up in the rxte / pca data following tomsick & kaaret ( 1998 ) and jahoda et al . ( 2006 ) . during the decay of the x - ray bursts investigated here , the observed count rates are @xmath81600 cts / s / pcu , which is far below the threshold of 10000 cts / s / pcu , where pile - up can become important according to those studies . we , therefore , conclude that photon pile - up in the pca does not affect our results . we fit each spectrum with a blackbody function using the _ bbodyrad _ model ( as defined in xspec , arnaud 1996 ) . pinto et al . ( 2010 ) reported an analysis of high resolution x - ray spectroscopy of the interstellar medium toward gs 1826@xmath0238 and found an equivalent hydrogen column density of n@xmath9 . in order to take into account the effect of the interstellar extinction in the x - ray spectra , we used this value with the _ phabs _ model in xspec . we set the elemental abundances to the protosolar abundances of lodders ( 2003 ) , following pinto et al . the resulting histogram of @xmath5/dof values obtained from the cooling tails of all the bursts and the spectral evolution during one of the bursts ( burst # 23 ) are given in figures [ chi_hist ] and [ rxte_burst23 ] . similar to the results we obtained for a number of sources in paper i , the blackbody model could adequately fit the data with a systematic uncertainty @xmath10 of only @xmath11 . chandra observed gs 1826@xmath0238 in 2002 for approximately 70 ks ( obsid 2739 ) . we analyzed the data using the _ chandra_repro _ tool with ciao version 4.3 and caldb 4.4.5 . chandra_repro _ script automates the data processing following the standard pipeline steps but applies the latest available calibrations . using the _ dmextract _ tool we created a 2.5@xmath08.0 kev range background - subtracted source lightcurve using only the data from the grating arms ( + /- 1st orders ) . the zeroth order data for this observation is saturated due to the high countrate of the source and the nominal frame time of 1.7 s. even for the grating arms , pile - up was significant during the peak of the bursts . in order to correctly carry out the comparison between the rxte and chandra observations , we opt to correct for pile - up the _ predicted _ chandra countrates based on the rxte spectra , as opposed to correcting the observed chandra countrates . because rxte observations have better statistics , this forward folding of the correction does not unnecessarily inflate the uncertainties . we applied a pile - up correction procedure that is described in the appendix , which was developed based on the understanding of the detector properties and calibrated against independent data . the evolution of the applied correction is shown in figure [ ch_pileup ] for burst # 23 . at the burst peak , photon pile - up causes a net loss of 25% of the flux , which can be recovered by the pile - up correction model that was developed independently of the present data ( see the appendix ) . nevertheless , the rise and the peak of each burst , which approximately correspond to the first 30 s , are most heavily affected by pile - up . furthermore , the rise and the peak are also the periods where the intrinsic spectrum may show some deviation from a planckian shape ; this is expected from theoretical model spectra and was discussed in paper i using rxte observations . because of these reasons , we exclude the first 30 s of all chandra bursts from our analysis such that in the time interval that we use , the pile - up fraction decreases to 2% . we used the _ mkgrmf _ and _ fullgarf _ tools to create observation specific response and auxiliary response files for the first orders of the high energy transmission grating ( hetg ) and used these when creating the simulations based on the spectral parameters determined from time resolved x - ray spectroscopy . xmm-_newton _ observed gs 1826@xmath0238 in 2003 for approximately 200 ks with two different back - to - back exposures that lasted 107,850 and 91,916 s ( obsids : 0150390101 , 0150390301 ) . the observation was performed with epic - pn in timing mode and the epic - mos detectors were turned off . we analyzed the epic - pn data using the sas version 11.0.0 ( xmmsas_20110223_1801 - 11.0.0 ) and the calibration files as of june 2011 . we calibrated the observation with the _ epproc _ tool . due to the high energy particle background we could only use the first 67 ks of the first observation and @xmath880 ks of the second . similar to the chandra observation , we extracted background subtracted source light curves in the 2.5 @xmath0 10.0 kev range using the sas tool _ epiclccorr_. the observatory guide gives a limit of 450 ct / s for telemetry of the epic - pn timing mode , which can be increased if some of the other instruments are turned off as in this case . in the burst lightcurves that were observed with xmm-_newton _ , two gaps that last approximately 10 s ( see figures [ comp_xmm1 ] and [ comp_xmm2 ] ) are seen . these are most likely due to this limit . while creating the simulations we calculated observation specific effective area and redistribution matrix files using the sas tools _ rmfgen _ and _ arfgen_. in order to explore the possibility of pile - up in the epic - pn data , we created two event lists : one including the full psf for the source and another where the central two pixels , which should be the most affected from pile - up , are excised . if pile - up were significant , the ratio of the countrates in the two event lists would evolve as the count rate changes during the burst because the two event lists would be subject to different amounts of pile - up . the ratio of the two lightcurves is shown in figure [ xmm_pileup ] for bursts # 30 , 31 , and 32 . as expected , a large fraction of counts are collected in the central two pixels , making this ratio approximately equal to two . however , as the burst eveolves , the ratio remains constant at the few percent level , implying that the pile - up effect will be at most of the same order . we consider this as strong evidence for the lack of any pile - up effect on the measured fluxes . nevertheless , in order to avoid any residual effects of photon pile - up and any possible deviations from a planckian shape , as with the chandra bursts , we exclude the first 30 s of each burst when determining the systematic offsets between the instruments . we show these time intervals in the plots for comparison and completeness . we followed the method outlined in section [ method ] for comparing all of the x - ray bursts observed simultaneously with rxte and chandra or with rxte and xmm-_newton_. briefly , our analysis steps are as follows : 1 . we first obtained the best fit values of the parameters for each burst spectrum observed with rxte / pca . we then calculated the two - dimensional confidence contours for the resulting color temperature and the blackbody normalization parameters covering a @xmath12 range from the best fit values . an example confidence contour is shown in figure [ example_conf ] for the spectrum obtained approximately 9.7 s after burst # 26 started . note that we show 68% and 95% confidence contours in the figure but use the entire likelihood in the calculations . 2 . for each pair of spectral parameters , we calculated the predicted countrates for the chandra gratings ( only the @xmath13 1st orders for heg and meg ) in the 2.5@xmath08.0 kev range and for xmm- _ newton_/epic - pn countrates in the 2.5@xmath010.0 kev range ( i.e. , the energy range that matches the rxte / pca sensitivity ) . for the case of the chandra simulations , we applied the pile - up correction , as described in the previous section . 3 . in order to take into account the uncertainties in the rxte / pca modeling , we used the posterior likelihoods over the parameters for each spectral fit to calculate the uncertainty in the total predicted countrates . 4 . finally , we compared the observed chandra hetg / acis - s or xmm-_newton_/epic - pn countrates to the countrates predicted using the rxte / pca spectral fits . for this comparison , we binned the countrates to match the time resolution of each data set and subtracted the persistent flux level of the source from the burst emission . during the chandra observation , the nominal frame time was 1.7 s. the time resolution of the epic - pn observation was much higher . we , therefore , binned the epic - pn data to match the varying exposure time of rxte spectra . in figures [ comp_ch ] , [ comp_xmm1 ] , and [ comp_xmm2 ] , we show the time evolution of the source countrate during each x - ray burst simultaneously observed with rxte and chandra in the 2.5@xmath08.0 kev range or with rxte and xmm-_newton _ in the 2.5@xmath010.0 kev range . in principle , all of the calculated ratios should be equal to unity within their formal uncertainties , if there are no systematic calibration differences . it is already evident from this figure that this is indeed the case between rxte and chandra but that there is a systematic calibration offset between rxte and xmm-_newton_. to explore in more detail the calibration differences between the three instruments and whether such differences evolve with countrate , we performed two analyses . first , for the xmm-_newton_/epic - pn data , for which we infer the largest systematic calibration offset , we created histograms of the ratios of the predicted and observed countrates with varying start times compared to the burst start times ( see figure [ phtest2 ] ) . if there were a significant countrate dependence of the calibration offset , we would have seen a significant change in the distribution of calculated ratios at later stages of these bursts . however , it is evident from figure [ phtest2 ] that there is no such time dependent change in the ratios . second , in order to quantify the level of offset between the three instruments , we followed the bayesian procedure described in papers i and ii and outlined below for completeness . we calculated the ratio @xmath14 between the observed and predicted countrates for each time bin in the 30@xmath0120 s interval after the start of every burst together with its formal uncertainty . we assume that the uncertainty in each measurement is described by a gaussian with a centroid @xmath15 and a dispersion @xmath16 i.e. , @xmath17\;,\ ] ] where @xmath18 is a set of normalization constants . we will also assume that the ensemble of these ratios for each burst is drawn from a gaussian distribution with a centroid at @xmath19 and a dispersion @xmath20 , i.e. , @xmath21\;,\ ] ] where @xmath22 is a proper normalization constant . this distribution is our model for the underlying systematic flux calibration offset between the instruments . if there are no calibration differences , the centroid should be equal to one , to within @xmath23 , where @xmath24 is the number of measurements , and the dispersion should tend to zero . a significant offset in the flux calibration between two instruments will move @xmath25 away from unity , but keep the dispersion consistent with zero . on the other hand , a large dispersion can not be produced by an overall flux calibration offset . instead , it likely points to effects such as an inadequate modeling of the pile - up correction , of the energy response of the detector(s ) , or of the astrophysical properties of the source . our goal is to calculate the posterior likelihood for the parameters @xmath19 and @xmath20 , @xmath26data@xmath27 , given an ensemble of @xmath24 measurements . using bayes theorem , we can write @xmath28 here @xmath29 is a normalization constant , and @xmath30 and @xmath31 are the priors over the centroid and dispersion of the underlying gaussian distribution . we take these prior distributions to be flat in the range of parameters shown in figures [ res_ch ] , [ res_xmm1 ] , and [ res_xmm2 ] , which is large enough to incorporate any range of interest . the quantity @xmath32 measures the likelihood that we will make a particular set of measurements given the values of the parameters of the underlying gaussian distribution . because these measurements are uncorrelated , we write @xmath33 inserting equation ( [ eq : data ] ) into equation ( [ eq : bayes ] ) we obtain the posterior likelihood @xmath34 where @xmath29 is an overall normalization constant . we show two - dimensional contours of these likelihoods for each burst in figures [ res_ch ] , [ res_xmm1 ] , and [ res_xmm2 ] and display the most likely values of the parameters in table [ res_all ] . all figures show that the underlying distributions of offsets are extremely narrow and that the most likely values of their dispersions are consistent with being zero . this indicates that there are no systematic problems in the modeling of the energy response of the detectors and that the corrections described earlier adequately account for any pile - up effects . the most likely values of the centroids are consistent with unity for the comparison between rxte and chandra but as low as 0.86 for the comparison between rxte and xmm-_newton_. because the offsets are stable both throughout the cooling tails of individual bursts and from burst to burst , we combined all the measurements from all the bursts observed by each x - ray satellite to quantify the overall offset and its uncertainty . we computed the posterior likelihood for the offset @xmath25 given the two dimensional posterior likelihood of the centroid value @xmath19 and dispersion @xmath20 of the underlying gaussian distribution as @xmath35 using 5 bursts and 447 individual measurements and 3 three bursts and 170 individual measurements for comparison with xmm-_newton _ and chandra , respectively , we found that the ratio of the observed count - rate with the chandra and xmm-_newton _ compared to the predictions based on the rxte spectral analysis in the 2.5 @xmath0 8.0 kev and 2.5 @xmath0 10 kev range is 1.012@xmath36 and 0.857@xmath10.003 , respectively . in other words , the fluxes measured with chandra hetg / acis are in statistical agreement with the rxte / pca while the epic - pn measures one that is 14% less . in figure [ final_overplot ] , we show the combined likelihoods of the resulting ratios of countrates between the different instruments . lcc 22 & 1.01 & hetg - acis / pca + 23 & 1.01 & hetg - acis / pca + 24 & 0.97 & hetg - acis / pca + 26 & 0.86 & epic - pn / pca + 27 & 0.85 & epic - pn / pca + 30 & 0.86 & epic - pn / pca + 31 & 0.86 & epic - pn / pca + 32 & 0.86 & epic - pn / pca [ res_all ] in the previous section , we addressed the cross calibration between rxte and _ chandra _ as well as between rxte and xmm- _ newton _ , using observations of the burster gs 1826@xmath0238 . we obtained the most likely fractional offsets between the fluxes measured using the various instruments on simultaneous observations and showed that they are consistent with being constant throughout individual bursts as well as across different bursts . however , the offsets we inferred correspond to a single point in time for each comparison . if the efficiencies of the detectors were constant in time , the results of this comparison would be applicable to all the bursts observed thoughout the durations of the missions . however , the responses of the detectors evolve with time for a number of reasons ( e.g. , gas leakage , charge transfer inefficiency ) . this evolution is typically accounted for using smooth , time - dependent models of the response parameters of the detectors . these parameters are determined by fitting the models to regular observations of the cosmic sources with known brightness and spectral shape or using internal calibrations sources ( e.g. epic cameras ) . as an example , rxte / pca had been calibrated against the crab nebula ( shaposnikov et al . 2013 ) throughout the mission . the smooth models of the evolution of the detector response clearly can not reliably capture the true evolution , which is expected to have a stochastic component . this will , therefore , introduce an uncertainty in the measurements that will change in time . [ line - through]*t*his complication is exacerbated if the calibration source shows intrinsic variability , as is the case with the crab nebula , which showed variability at a level of @xmath8 7% ( wilson - hodge et al . it is , therefore , possible that were we to repeat our cross calibration procedure between rxte and chandra at different epochs , we would obtain different offsets . even though randomly catching an epoch of identical flux measurements between rxte and chandra might seem unlikely , we can not base our conclusions on such a posterior likelihood . for this reason , we now explore the possibility that there is a time - dependent systematic uncertainty in the calibration of the instruments and how this uncertainty would affect the scatter in the measured burst fluxes . if there were single - epoch measurements of the fluxes during the cooling tails of bursts , this potential systematic error would have to be added to the error budget of the measurements . however , for many of the sources we used for neutron - star radius measurements in papers i and ii , the burst flux measurements were carried over many epochs spanning the lifetime of rxte . in this case , as we will show below , this potential calibration uncertainty should not be treated as a separate source of systematic uncertainty as it is already inherently part of the measured scatter of observed fluxes . to demonstrate this point , we compute the mean and the variance of the ( burst ) flux measured by an instrument ( e.g. , rxte / pca ) . let the source flux for each burst be @xmath37 , with a mean of @xmath38 and an intrinsic dispersion @xmath39 . let the instrument have an inferred overall efficiency factor @xmath40 ( e.g. , as reported by a calibration team ) . if this efficiency factor has an unknown systematic bias , which may also be variable in time , then the observed countrate flux will be @xmath41 where we have denoted the mean bias in the calibration over the time span of the observations by @xmath42 and the potential time - variable part by @xmath43 . by definition , @xmath44 . when an observation is carried out and standard calibration procedures are applied , the observed source flux will be inferred to be equal to @xmath45 which will be related to the true source flux by @xmath46 clearly , the observations reproduce the true source flux if there is no bias in the calibration . the mean inferred flux of the source is equal to @xmath47 where we have used the fact that @xmath48 because the time evolution of the detector response and the variability of the source are uncorrelated . the variance in the inferred burst flux is @xmath49 and the mean of the square of the inferred fluxes is given by @xmath50 ^ 2 \right\rangle \nonumber \\ & = & \left(1+\frac{b}{r}\right)^2 \langle s_{i,\rm true}^2\rangle + \left\langle \left(\frac{v_i}{r}\right)^2 s_{i,\rm true}^2 \right\rangle + 2\left(1+\frac{b}{r}\right ) \left\langle \frac{v_i}{r } s_{i,\rm true}^2 \right\rangle.\end{aligned}\ ] ] since @xmath51 , we find that @xmath52 + \frac{1}{r^2}\left\langle v_i^2s_{i,\rm true}^2 \right\rangle \;.\end{aligned}\ ] ] the term in the square bracket is equal to the true variance of the source flux @xmath53 = \sigma_{\rm true}^2 $ ] . the average in the second term is @xmath54 , where we have used the definition of the variance of the time - variable bias @xmath55 as a result , @xmath56 or in terms of fractional errors , @xmath57\;.\end{aligned}\ ] ] keeping only terms up to second order in the dispersions , the final result is @xmath58 this expression shows that the measured fractional variance in the burst fluxes is equal to the quadrature sum of the intrinsic fractional variance and the fractional variance of the calibration bias . in other words , the observed variance in the burst fluxes is an upper limit to the intrinsic scatter . if the statistical properties of the calibration uncertainties could be measured independently , this contribution could be subtracted from the dispersion of measured fluxes . in this paper , we quantified the systematic uncertainties that arise from the flux calibration of the x - ray instruments used in neutron star mass and radius measurements . we used simultaneous observations of the burster gs 1826@xmath0238 with rxte / pca and chandra hetg / acis - s as well as with rxte / pca and xmm-_newton_/epic - pn . we performed a bayesian statistical analysis to measure the level of offset in the flux calibration between these instruments . in the case of the comparison between chandra and rxte , we found no evidence for a significant offset between the measured fluxes and concluded that the most likely value of the flux ratio is @xmath59 . as we describe in the appendix , the effect of pileup is unlikely to change the value of this ratio in a statistically significant way . in contrast , we found a 14.0@xmath10.3% difference between the fluxes inferred using xmm-_newton_/epic - pn and rxte / pca . this discrepancy is persistent in all five analyzed bursts and is independent of the source spectrum or countrate . the flux discrepancy between xmm-_newton_/epic - pn and other detectors has been noted in earlier studies of different sources . nevalainen et al . ( 2010 ) reported a difference between flux measurements with the epic - pn , chandra acis ( zeroth - order data ) , and epic mos detectors at a level of 5@xmath010% . according to their results , acis flux measurements are significantly higher than epic - pn , by 11.0@xmath10.5% . similarly , the difference in the flux measurements between the epic - pn and mos detectors has been noticed earlier ( see e.g. , mateos et al . 2009 and xmm - newton calibration documents xmm - newton - soc - cal - tn-0018 and xmm - newton - soc - cal - tn-0052 ) ; mos yields fluxes that are approximately 7% higher than epic - pn ( see also read et al . 2014 ) . most recently , using 64 clusters of galaxies , schellenberger et al . ( 2014 ) showed that the effective area cross - calibration uncertainty between acis and epic - pn is energy dependent and in the 2@xmath07 kev range stabilizes at a 15% level a similar result is also presented by guainazzi et al . ( 2015 ) . given that the calibration errors of the detectors may be variable in time , it is possible that the excellent agreement between rxte and chandra is a coincidence . we showed , however , that even in this case , the spread of the measured burst fluxes reported in papers i and ii already accounts for this potential systematic uncertainty . we , therefore , conclude that , while an overall bias may still exist in the measurements , potential time - dependent flux calibration uncertainties do not increase the error budget of the radius measurements we reported in these earlier papers . one final consideration is whether the difference in the measured flux between xmm-_newton _ epic - pn and the other two instruments also affects the neutron star radii determined through the observations of quiescent low - mass x - ray binaries . the flux levels in question are very different in the observations of these dim sources : they are approximately six orders of magnitude below the peak burst fluxes and the crab nebula , and about three orders of magnitude below other typical calibration sources ( e.g. , clusters of galaxies ) . given the signal - to - noise ratio of the spectra that is achieved with even the longest observations of quiescent low - mass x - ray binaries , it is not possible to discern any measured flux differences between instruments , even if this difference persisted at the 14% level . this supports the lack of evidence for flux differences presented in the earlier studies ( catuneanu et al . 2013 ; heinke et al . 2014 ; guillot et al . 2013 ) , which compared neutron star radii derived from the quiescent sources using xmm-_newton _ and chandra data . to account for pileup in the chandra hetg / acis - s spectra , we used a correction that was developed independently by one of the co - authors , herman marshall , for mildly piled - up calibration data . the approach relies on the measurement of counts per pixel around the ir - m and si - k edges , at which the spectral response changes rapidly . denoting the observed counts per detector pixel per frame in wavelength bin @xmath60 as @xmath61 , we define the degradation of the effective area as the value of the parameter @xmath63 and the uncertainty associated with its determination was found using an observation of mk 421 ( observation i d 4148 ) performed on 26@xmath027 october 2002 with the letgs / acis - s for over 91 ks . the data were downloaded from _ tgcat _ ( huenemoerder et al . 2011 ) . the countrate spectra were rebinned by a factor of two so that each spectral bin corresponds closely to one acis pixel . the source was extremely bright so that there were of order 5000 counts per spectral bin in both the + 1 and -1 orders near 2 kev . the count spectrum was divided by the arf obtained from tgcat to generate spectra in photons/@xmath64/s / for both the @xmath65 and @xmath66 orders . when the two orders are compared ( see the top panel of figure [ final_overplot ] ) without being corrected for pileup , there are clear edges at ir - m ( 6.0 ) and si - k ( 6.74 ) in the effective - area - corrected spectra . these features arise from differential pileup , where @xmath61 may increase significantly over a few spectral resolution elements at an edge in the effective area of the instrument , causing increased pileup and a larger fractional loss of events with a corresponding decrease in apparent photon flux . we devised a model - independent method for estimating the parameter @xmath63 in equation ( [ eqa1 ] ) , which depends on the very smooth and slowly varying photon spectrum of mk 421 . using the ratio of fluxes in the 5.5 - 5.9 range to those in the 6.1 - 6.5 range ( straddling the ir - m edge ) , we compute the value of @xmath63 that matches the ratios based on the average @xmath40 values in the two narrow bands . we find @xmath67 for @xmath66 and @xmath68 for @xmath65 and , therefore , adopt @xmath69 and a range of acceptable values that are @xmath70 away from this value . figure [ final_overplot ] shows the sensitivity of the spectral inversion to @xmath63 ; for @xmath71 , the spectra still show edges , while , for @xmath72 , the data are over - corrected to form an excess over a smooth spectral model . ( a global spectral fit to a double power law indicates that the photon spectrum should drop monotonically by only 1% per through this wavelength range , so the assumption that the fluxes are equal should be good to 1% . ) using equation ( [ eqa1 ] ) , which is verified to @xmath73 cts / frame / pixel , a @xmath74% uncertainty in @xmath63 results in at most a @xmath75% uncertainty in @xmath76 , which is smaller than the expected calibration uncertainty in @xmath77 . this method required the hrma effective area correction implemented in 2005 , which corrected the ir - m edge caused by a thin contaminant layer on the hrma . note that , the correction given in equation ( [ eqa1 ] ) is valid for low to modest pileup , i.e. , @xmath78 cts / frame / pixel and the data used to compute the coefficient @xmath63 ranged up to about 0.10 cts / s / pixel . this yields a maximum correction of about 50% . in our comparison between rxte and chandra , we took into account the pileup losses in the predicted countrates rather than correcting the observed rates for pileup . if we denote by @xmath79 the number of photons per detector pixel per frame that should have been recorded in wavelength bin @xmath60 in the absence of pileup , we can then invert the above expression to obtain houck , j. c. , & denicola , l. a. 2000 , in astronomical society of the pacific conference series , vol . 216 , astronomical data analysis software and systems ix , ed . n. manset , c. veillet , & d. crabtree , 591+
in the long theoretical preparation for the lhc , a wide spectrum of options for the new physics at the tev scale has been considered . a major role has been played by supersymmetric models . several schemes have been investigated in which supersymmetry is broken in a hidden sector with no renormalizable interactions with the observable sector and is communicated to the latter by a variety of mechanisms @xcite . below the scale @xmath4 at which supersymmetry breaking is communicated , sfermion masses are typically described by the model - independent effective lagrangian operator @xmath5 where @xmath6 is a standard model ( sm ) singlet hidden chiral superfield whose @xmath7-term vev breaks supersymmetry , @xmath8 , and @xmath9 is a generic light , observable chiral superfield , for example an mssm one . such an effective , supersymmetric description holds if @xmath10 . different models are characterized by different origins for the above operator . surprisingly enough , a simple and attractive possibility has been neglected in the almost three decades of phenomenological studies of supersymmetry : the possibility that the operator in eq . ( [ eq : sfermioneffective ] ) arises from the renormalizable , tree level exchange of heavy vector superfields , as in fig . [ fig : diagram ] . this is the communication mechanism that we call tree level gauge mediation ( tgm ) @xcite . besides its simplicity ( compare for example with the cumbersome set of two loop diagrams generating the operator in eq . ( [ eq : sfermioneffective ] ) in ordinary , loop gauge mediation ) , this framework is also motivated by the necessary presence of superheavy vector fields in grand unified theories ( guts ) . on the verge of the lhc era , we believe it is worth filling this lacuna and spell out the consequences of tgm , also in the light of its peculiar predictions . one may wonder how such a simple possibility could have been missed . the reason might be that well known arguments seem to prevent it . the main obstacle is represented by the supertrace formula @xcite and its consequences . in the context of the tree level , renormalizable spontaneously broken supersymmetric theory underlying fig . [ fig : diagram ] , we must have @xmath11 for the supertrace of the squared masses of the fields in the model . eq . ( [ eq : str ] ) holds separately for each set of conserved quantum numbers @xcite . if the action of the gauge generators on the full set of chiral superfields , @xmath12 , is traceless , as in the case we are going to consider , the supertrace vanishes . this represents a potential phenomenological problem . still , tgm leads to a viable spectrum , as we will see . while the explicit construction in the next sections is all we need to get to our results , we find useful , in this introduction , to review what the potential problem is and illustrate how tree level gauge mediation solves it . we can see the potential problem at two different levels . first , eq . ( [ eq : str ] ) holds in particular when applied to all fields with the quantum numbers of the sm fermions . let us consider then the case of the mssm . in this case the fields with the quantum numbers of the sm fermions are the sm fermions themselves , @xmath13 , and their supersymmetric partners , the sfermions @xmath14 . from @xmath15 we then conclude that the sum of the squared masses of fermions and sfermions should coincide . this is in clear contradiction with the experimental bounds on the sfermion masses , giving @xmath16 . this is however far from being the end of the story , as any realistic complete theory of supersymmetry breaking is likely to involve additional fields on top of the mssm ones . this is the case of our tgm framework , where the positive contribution to the supertrace from the mssm fermions and sfermions is compensated by an opposite contribution from extra fields with quantum numbers within the ones of the sm fermions , @xmath17 , so that @xmath18 , in agreement with the supertrace formula . the extra chiral superfields will get heavy supersymmetric mass terms . their negative contribution to the supertrace is due to the fact that their scalar components get negative @xmath19 soft masses , which however represent only negligible corrections to their much larger , positive supersymmetric mass term . as we will see , this can be obtained without ad hoc model building efforts . the supertrace formula has stronger implications than the ones outlined above , which should also be addressed . let us consider the fields with the @xmath20 quantum numbers of the @xmath21 quarks only . the latter set will contain at least the three down - type sm quarks and their scalar partners and possibly the extra fields we need to compensate the supertrace formula . let us project the supertrace formula in the flavor space of down - type fields along the direction corresponding to the lightest @xmath21-quark mass eigenstate , the down quark . note that when restricted to a given set of quantum numbers , the trace on the right - hand side of eq . ( [ eq : str ] ) can be non - vanishing . assume now that the only u(1 ) factor in the gauge group is the sm hypercharge , @xmath22 . we then obtain @xcite @xmath23 where @xmath24 is the lightest @xmath21-sfermion mass eigenstate , @xmath25 is the down quark mass , @xmath26 , @xmath27 is the hypercharge gauge coupling and @xmath28 the hypercharge @xmath29-term . ( [ eq : constraint ] ) represents a serious phenomenological problem , even in the presence of the extra fields invoked above . if @xmath30 , in fact , eq . ( [ eq : constraint ] ) would force a down sfermion mass to be smaller than about @xmath31 , in contrast with the lowest experimental limits of a few hundreds gev . if @xmath32 , the constraint would be even stronger . if @xmath33 , the constraint would be loosened , but one could repeat the argument for the up quarks and squarks . for which the @xmath28 contribution to the relation analogous to eq . ( [ eq : constraint ] ) would have opposite sign , leading to an even stronger bound for the lightest up squark . in order to bypass this problem , an extra u(1 ) factor , giving the same sign on both down and up fields , is needed . such an extra u(1 ) factor is present `` by definition '' in the tgm scheme . it is the u(1 ) factor associated to the heavy vector exchange in fig . [ fig : diagram ] ( as @xmath6 is a sm singlet , the heavy vector must also be a sm singlet ) . we therefore have all the ingredients needed to overcome the potential problem set by the supertrace formula . as we will see , those ingredients naturally combine in phenomenologically viable schemes . another potential problem is represented by the fact that gaugino masses arise at the loop level and are therefore potentially suppressed with respect to the sfermion masses by a large loop factor , thus pushing the sfermions out of the reach of the lhc and introducing a significant fine - tuning in the determination of the higgs mass . we will list in section [ sec : gauchi ] a number of gaugino mass enhancement factors that can compensate fully or partially that loop factor . a minimal model of tree - level gauge mediation has been presented in @xcite , solving the supersymmetric flavor problem and predicting the ratio of different sfermion masses to be different from msugra and other schemes . in this paper we would like to take a broader point of view and study the general implementation of tgm . this will allow to establish the general properties of tgm , to set up the guidelines for model building and to identify what are the hypotheses under which the peculiar predictions on soft masses of the minimal model hold . moreover , we would like to present a few new approaches to the @xmath3-problem , both in the context of well known ( giudice - masiero , nmssm ) and new solutions . in particular , in section [ sec : tree ] we will discuss what are the conditions under which heavy vector superfields can act as tree - level messengers of supersymmetry breaking and obtain a general expression for the tree level contribution to the supersymmetry breaking lagrangian , in particular to the sfermion soft masses . in section [ sec : oneloop ] , we will consider the one - loop contributions to soft masses , concentrating mostly on gaugino masses and the enhancement factors compensating their loop suppression . in section [ sec : guidelines ] we consider the possibility to obtain a phenomenologically viable model from the general formalism previously introduced . we will see that clear model building guidelines emerge , leading to peculiar predictions for the pattern of mssm sfermion masses and we will identify the assumptions underlying such predictions . in section [ sec : mu ] , we will discuss a few new approaches to the @xmath3-problem , before summarizing in section [ sec : conclusions ] . the paper also contains two appendixes . in appendix a , we outline the procedure to integrate out vector superfields and address a few minor issues , such as the generalization to the non - abelian case and the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives . in appendix b , we provide an example of a superpotential achieving supersymmetry breaking , so(10 ) breaking to the sm , ensuring that only the mssm fields survive at lower energy ( in particular providing doublet - triplet splitting ) and solving the @xmath3-problem . such a superpotential is not aimed at being simple or realistic , but it represents a useful existence proof . in this section we discuss the conditions under which heavy vector superfields can act as tree - level messengers of supersymmetry breaking in the context of a generic , renormalizable , @xmath34 globally supersymmetric gauge theory in four dimensions . then we recover the general expression for the tree level contribution to the sfermion soft masses . we discuss their origin both in the context of the full , renormalizable theory , and in an effective theory approach . we start from a lagrangian described by a canonical khler @xmath35 and gauge kinetic function and by a generic superpotential @xmath36 function of the chiral superfields @xmath37 , with no fayet - iliopoulos term . we follow the conventions in @xcite . we will denote by @xmath38 , @xmath39 , @xmath40 the scalar , spinor , and auxiliary component of @xmath41 and by @xmath42 , @xmath43 , @xmath44 the vector , spinor , and auxiliary component of @xmath45 . the gauge group @xmath46 ( assumed for simplicity to be simple with a single gauge coupling @xmath47 ) is broken by the scalar component vev @xmath48 to the subgroup @xmath49 at a scale @xmath50 , at which the theory is approximately supersymmetric . in the phenomenological applications we have in mind , @xmath49 contains the sm gauge group @xmath51 , @xmath46 is a grand - unified group ( for example so(10 ) or @xmath52 ) , and the breaking scale is of the order of the gut scale . correspondingly , the vector superfields split into light and heavy ones , associated to the orthonormalized generators @xmath53 and @xmath54 respectively : @xmath55 , @xmath56 , @xmath57 . the heavy vector superfields acquire a squared mass matrix given by @xmath58 we choose the basis of heavy generators @xmath59 in such a way that the above mass matrix is diagonal , @xmath60 the heavy vector superfields become massive by eating up a corresponding number of goldstone chiral superfields . it is then convenient to split the chiral superfields as follows @xmath61 where @xmath62 , @xmath63 are the goldstone superfields associated to the generators @xmath59 and @xmath64 , @xmath65 is an orthonormal basis in the space of the `` physical '' chiral fields @xmath66 , @xmath67 . in the supersymmetric limit , @xmath68 is orthogonal to @xmath69 and @xmath69 does not mix with the physical superfields . the physical components of the massive vector superfield @xmath45 are @xmath42 , @xmath43 , @xmath70 , @xmath71 , all with mass @xmath72 . the imaginary part of @xmath73 , the goldstone boson , becomes as usual the longitudinal component of the massive gauge boson @xmath42 and the spinors @xmath70 and @xmath43 pair up in a dirac mass term . this spectrum can be split by supersymmetry breaking corrections , as we will see in section [ sec : gauginovector ] . as for the physical chiral superfields @xmath74 , their supersymmetric mass matrix is given by @xmath75 again , we choose the basis @xmath76 in such a way that the above mass matrix is diagonal and positive , @xmath77 the scalar and fermion components of @xmath66 can be split by supersymmetry breaking corrections , which can also induce a mixing with the scalar and fermion components of the heavy vector superfields . supersymmetry is supposed to be broken at a much lower scale than @xmath78 , where some of the fields @xmath66 get an @xmath7-term , @xmath79 , @xmath80 . as a consequence , @xmath68 satisfies with good approximation the @xmath7-term and @xmath29-term conditions at the scale @xmath78 , @xmath81 and @xmath82 ( see eq . ( [ eq : dh ] ) ) for each @xmath83 . the @xmath7-terms induce a non vanishing vev for the @xmath29-terms @xmath84 of the heavy vector superfields . the stationary condition for the scalar potential @xmath85 , @xmath86 , together with the gauge invariance of the superpotential give @xmath87 with the light @xmath29-terms still vanishing . clearly , only generators @xmath59 that are singlets under the unbroken group @xmath49 can contribute to such @xmath29-term vevs . note also the condition @xmath88 which implies that the goldstone superfields @xmath69 do not get @xmath7-term vevs ( the @xmath29-term condition implies that in the supersymmetric limit they do not get scalar vev either ) . the latter relation also follows from the gauge invariance of the superpotential . in turn , the @xmath29-terms above give rise to tree level soft masses for the scalar components @xmath89 of the chiral superfields @xmath74 @xmath90 provided that both @xmath91 and the scalars @xmath92 are charged under the ( broken ) gauge interaction associated to @xmath59 and provided that @xmath59 is a singlet under @xmath49 . the complete list of tree level soft terms obtained from the gauge dynamics can be more conveniently recovered in the effective theory below @xmath78 . before discussing it , let us observe that this theory must necessarily satisfy the supertrace formula @xmath93 . in the case of the soft terms in eq . ( [ eq : mmd ] ) , this simply follows here from @xmath94 . in particular , the tracelessness condition implies that positive soft masses are accompanied by negative ones in eq . ( [ eq : dhsfermion ] ) . this is a potential phenomenological problem , which has long been considered as an obstacle to models in which supersymmetry breaking terms are generated , as here , at the tree , renormalizable level . however , it has recently been shown @xcite that such a potential problem can be easily solved by adding a large positive supersymmetric mass term to the chiral superfields whose tachyonic nature would be problematic . as mentioned , the generation of the sfermion masses can be conveniently seen in the effective theory below @xmath78 , where the heavy vector and the goldstone chiral superfields have been integrated out . in this theory , the chiral degrees of freedom are the @xmath66 . the gauge group is @xmath49 and it is unbroken ( we neglect electroweak symmetry breaking ) . as a consequence , there is no @xmath29-term contribution to supersymmetry breaking . the scalar masses arise in this context from @xmath7-terms vevs through an effective khler operator , as we will see in a moment . the vector superfields can be integrated out by solving the equations of motion @xmath95 @xcite . in appendix [ sec : eom ] we illustrate the details of such a procedure in a general case , we explicitly write the resulting effective theory at the leading order , and we make a few general remarks on the approximations involved in using @xmath95 and on the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives @xcite . for the present purposes , we are only interested in the terms in the effective lagrangian relevant to ( sizable ) soft supersymmetry breaking . those are the ones following from the effective tree level contribution to the khler potential in eq . ( [ eq : keff2]a ) : @xmath96 where we remind that @xmath66 has no vev in its scalar component . the operator in eq . ( [ eq : dk ] ) can be seen to arise from the diagram on the left - hand side in fig . [ fig : diagram ] . ( 400,130)(-80,-20 ) ( -60,50)(-10,50)46 ( -90,75)(-60,50 ) ( -90,25)(-60,50 ) ( 20,75)(-10,50 ) ( 20,25)(-10,50 ) ( -95,80 ) [ ] @xmath66 ( 27,80 ) [ ] @xmath97 ( -30,62 ) [ ] @xmath85 ( -98 , 28 ) [ ] @xmath98 ( 29 , 28 ) [ ] @xmath66 ( 53,50 ) [ ] @xmath99 ( 120,50)(180,50)46 ( 90,75)(120,50 ) ( 90,25)(120,50 ) ( 210,75)(180,50 ) ( 210,25)(180,50 ) ( 86,80 ) [ ] @xmath6 ( 83 , 28 ) [ ] @xmath100 ( 150,62 ) [ ] @xmath85 ( 218,80 ) [ ] @xmath101 ( 220 , 28 ) [ ] @xmath102 ( 250 , 50 ) [ ] @xmath103 ( 300,20)(300,80)46 ( 330,105)(300,80 ) ( 270,105)(300,80 ) ( 330,-5)(300,20 ) ( 270,-5)(300,20 ) ( 266,108 ) [ ] @xmath6 ( 264 , -3 ) [ ] @xmath100 ( 313,50 ) [ ] @xmath85 ( 339,108 ) [ ] @xmath104 ( 340 , -3 ) [ ] @xmath9 as mentioned , the only possible source of supersymmetry breaking in the effective theory are the @xmath7-term vevs of the chiral superfields @xmath66 . we remind that such @xmath7-term vevs must belong to non - trivial representations of the full group @xmath46 , in order to play a role in tgm . the only terms in the lagrangian containing such @xmath7-term vevs , at the tree level and up to second order in @xmath91 , @xmath105 , and @xmath106 , arise from the superpotential and from the operator in eq . ( [ eq : dk ] ) : @xmath107 where @xmath108 is the superpotential in the effective theory , @xmath109 let us consider the different terms in eq . ( [ eq : treesoft ] ) in turn . the first term in the second line reproduces the contribution to the soft scalar masses in eq . ( [ eq : mmd ] ) . the second term gives rise to an additional contribution , only relevant to superfields that are gauge partners of the goldstino superfield ( and have the same quantum numbers under @xmath49 as some of the generators of @xmath46)-term contribution to the scalar potential using eq . ( [ eq : mixing ] ) below . ] . all in all , we have @xmath110 .\ ] ] note that the soft terms do not actually depend on the gauge coupling or on the normalization of the generators @xmath111 , as @xmath112 is also proportional to @xmath113 . the second term in the first line of eq . ( [ eq : treesoft ] ) is a gauge - generated yukawa interaction with coupling @xmath114 , usually absent in models of supersymmetry breaking . from a phenomenological point of view , such tiny yukawa couplings might play a role in neutrino physics , where they could represent naturally small dirac neutrino yukawa couplings @xcite . finally , the first term in eq . ( [ eq : treesoft ] ) , has to do with the existence of a hidden sector in the effective theory . in the phenomenological applications we have in mind , the light spectrum will contain the mssm chiral superfields , as part of a light , `` observable '' sector . the latter will be charged under the residual gauge group @xmath115 . on the other hand , the supersymmetry breaking superfields do not feel the residual gauge interactions . in the effective theory , therefore , the supersymmetry breaking sector is hidden from the observable sector from the point of view of gauge interactions . in order for the supersymmetry breaking sector to be hidden also from the point of view of superpotential interactions , it is sufficient to make sure that the first term in eq . ( [ eq : treesoft ] ) does not induce a direct coupling between the two sectors . to be more precise , we can write the chiral superfields of the effective theory , @xmath66 , as @xmath116 the superfield @xmath6 is the only one getting an @xmath7-term vev , @xmath117 . its fermion component is the goldstino and therefore @xmath6 is a massless eigenstate of the mass matrix @xmath118 in eq . ( [ eq : mc0 ] ) . the remaining mass eigenstates are divided in two groups , the heavy ones , @xmath119 with masses @xmath120 , and the light , or observable , ones @xmath121 , with masses @xmath122 . in order to hide supersymmetry breaking from the observable sector also from the point of view of superpotential interactions , we require that @xmath123 ( at least for the renormalizable part of the superpotential ) . we can then see supersymmetry breaking as arising in a hidden sector and then communicated from the to the observable sector by the diagrams on the right - hand side of fig . [ fig : diagram ] . this can perhaps be considered as the simplest way to communicate supersymmetry breaking : through the tree level renormalizable exchange of a heavy gauge messenger . since heavy gauge messengers at a scale not far from the planck scale are automatically provided by grand - unified theories , this possibility is not only simple but also well motivated . the reason why it has not been pursued in the past is an apparent obstacle arising from the supertrace theorem that , as mentioned , can be easily evaded by providing heavy , supersymmetric masses to some of the superfields . such mass terms can naturally arise in the context of grand - unified theories , as we will see . we end this section with some comments on integrating out heavy chiral superfields and on the corresponding possible tree level contributions to @xmath124-terms and soft scalar masses . the heavy vector superfields may not be the only fields living at the scale @xmath78 , as chiral superfields could have mass terms of similar size or get it after gauge symmetry breaking . such chiral fields should also be integrated out in order to get the effective theory below the scale @xmath78 . in general , we want to integrate out all the heavy chiral superfields @xmath125 . since their masses @xmath126 are assumed to be much larger than the supersymmetry breaking scale , it will still be possible to write the effective theory in a manifestly supersymmetric way . in order to integrate them out , let us write the superpotential as @xmath127 where @xmath128 is at least trilinear in its argument . the equations of motion @xmath129 give @xmath130 the effective superpotential for the light fields @xmath6 and @xmath9 is therefore @xmath131 a contribution to the effective khler is also induced @xmath132 the effective contributions to the superpotential and to the khler in eqs . ( [ eq : weff ] ) and ( [ eq : keffchiral ] ) may give rise to `` chiral - mediated '' tree - level @xmath124-terms and ( negative ) additional contributions to soft scalar masses respectively . the latter should be subleading with respect to the ( positive ) vector mediated contributions in eq . ( [ eq : mm ] ) , at least in the case of the mssm sfermions . such tree level contributions could only arise in the presence of trilinear superpotential couplings in the form @xmath133 . in the following we will consider the case in which such a coupling is absent , @xmath134 so that the chiral - mediated tree tree level contributions also vanish . this is often the case , as illustrated by the model in @xcite . in this section we consider the one loop contributions to soft masses , focusing mostly on gaugino masses and the enhancement factors compensating their loop suppression . gaugino masses do not arise at the tree level . they are however generated at the one - loop level , as in standard , `` loop '' gauge mediation models . the suppression of gaugino masses by a loop factor with respect to scalar masses represents a potential phenomenological problem . given the present experimental limits on gaugino masses , a loop factor enhancement would make the sfermions heavier than @xmath135 , beyond the reach of the lhc and heavy enough to introduce a serious fine - tuning problem , thus approaching the split supersymmetry regime @xcite . however , it turns out that the loop hierarchy between gaugino and scalar soft masses is typically reduced or eliminated , as we will see in this section . we calculate gaugino masses in the full theory above @xmath78 . there are two types of one loop diagrams contributing to gaugino masses , depending on whether the degrees of freedom running in the loop are components of the heavy vector superfields ( including the goldstone superfields ) , as in fig . [ fig : mg]a , or physical chiral superfields , as in fig . [ fig : mg]b . correspondingly , we will distinguish a `` vector '' and a `` chiral '' contribution to the light gaugino masses , @xmath136 the latter may easily dominate on the former , as we will see . the source of supersymmetry breaking entering the diagrams of fig . [ fig : mg]a and 2b is a tree level splitting among the components of the heavy vector and chiral superfields respectively . we now examine the two contributions in eq . ( [ eq : gauginos ] ) in turn and write the known results @xcite in a form general enough to be suitable for the following discussion of their quantitative importance compared to the tree level scalar soft terms . ( 200,100)(50,5 ) ( 150,50)(40,180,360)38 ( 150,50)(40,0,45 ) ( 150,50)(40,90,45 ) ( 150,50)(40,180,135 ) ( 150,50)(40,90,135 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath138 ( 152,25)[t]@xmath139 ( 146,94)(154,86 ) ( 154,94)(146,86 ) ( 110,65)[r]@xmath140 ( 192,65)[l]@xmath140 ( 130,90)[r]@xmath141 ( 174,90)[l]@xmath141 ( 152,98)[b]@xmath142 ( 151,-10)[b]@xmath143 ( 200,100)(50,5 ) ( 150,50)(40,180,360)3 ( 150,50)(40,90,0 ) ( 150,50)(40,90,180 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath137 ( 150,25)[t]@xmath144 ( 146,94)(154,86 ) ( 154,94)(146,86 ) ( 115,75)[r]@xmath141 ( 190,75)[l]@xmath141 ( 152,98)[b]@xmath142 ( 151,-10)[b]@xmath143 + ( 300,100)(0,-10 ) ( 150,50)(40,180,360)3 ( 150,50)(40,90,0 ) ( 150,50)(40,90,180 ) ( 70,50)(110,50 ) ( 230,50)(190,50 ) ( 68,50)[r]@xmath137 ( 232,50)[l]@xmath137 ( 146,14)(154,6 ) ( 154,14)(146,6 ) ( 115,75)[r]@xmath141 ( 190,75)[l]@xmath141 ( 113,25)[r]@xmath145 ( 188,25)[l]@xmath145 ( 151,28)[t]@xmath146 ( 151,-10)[b]@xmath147 in the supersymmetric limit , the fields @xmath42 , @xmath43 , @xmath70 , @xmath71 form a massive vector multiplet with mass @xmath72 . once supersymmetry is broken , this spectrum is split by corrections to the fermion and scalar masses , which may also mix them with the components of the physical chiral superfields . here , we are interested to the supersymmetry breaking fermion mass term in the form @xmath148 , which is the source of the vector contribution to gaugino masses through the diagrams in fig . [ fig : mg]a . the mass term @xmath149 vanishes in the supersymmetric limit because of the gauge invariance of @xmath150 . the situation is different in the presence of supersymmetry breaking , when the gauge invariance of @xmath150 gives @xmath151 note also the more general expression for the mixed supersymmetry breaking terms @xmath152 before showing the expression for the gaugino masses induced by @xmath153 , let us remind that the heavy vector representation is in general reducible , under the unbroken gauge group @xmath49 , to a set of irreducible components , each with a single value of the mass . let us call @xmath154 the value of the mass in the representation @xmath155 and denote @xmath156 if @xmath59 , @xmath54 belong to the representation @xmath155 . in the limit @xmath157 , the supersymmetry breaking source @xmath153 can be treated as a perturbation in the one loop computation of gaugino masses . at the leading order in @xmath153 , the diagram in fig . [ fig : mg]a generates a contribution to light gaugino masses given by @xmath158 where @xmath159 is the dynkin index of the representation @xmath160 of the generator @xmath111 . the above contribution to gaugino masses arises at the scale @xmath78 where the heavy vectors live . let us now discuss the relevance of the above contribution to gaugino masses . first , let us note that in order for @xmath161 to be non - vanishing we need the following two conditions to be verified at the same time @xmath162 as it can be seen from eqs . ( [ eq : dmvector ] ) and ( [ eq : fgauge ] ) . in particular , we need at least one irreducible ( under the full group @xmath46 ) chiral superfield multiplet to get vev in both its scalar and @xmath7 components . at the same time , we need @xmath163 in order for the tree level contribution to scalar masses to be generated . the conditions in eqs . ( [ eq : con1],[eq : con2 ] ) may force the vector contribution to gaugino masses to vanish . on top of that , @xmath161 is always suppressed by a loop factor @xmath164 compared to the typical scalar mass in eq . ( [ eq : mm ] ) . if @xmath161 was the only contribution to gaugino masses , this would lead to an hierarchy between gaugino and scalar soft masses . moreover , the present experimental lower limits on gaugino masses , @xmath165 , would force the sfermion masses to be heavier than @xmath135 . however , as we will see in a moment , the chiral contribution to gaugino masses can be significantly larger than the vector contribution , thus reducing or even eliminating the loop suppression with respect to soft scalar masses . in this case , the vector contribution to gaugino masses typically ends up to be subdominant . the chiral contribution to gaugino masses arises from the one loop diagram in fig . [ fig : mg]b , as in ordinary loop gauge mediation . the scalar and fermion components of the chiral superfields entering the loop are split by a supersymmetry breaking scalar mass term @xmath166 . as a consequence of eqs . ( [ eq : hide ] ) and ( [ eq : pure ] ) , the supersymmetry breaking couples directly only to the heavy chiral fields and @xmath146 can be treated as a perturbation in the calculation of gaugino masses . the mass term @xmath146 is then given by @xmath167 which adds to the supersymmetric scalar mass term @xmath168 ( in the notation of eq . ( [ eq : mc0diag ] ) ) . the physical chiral superfield representation under the unbroken gauge group @xmath49 is in general reducible to a set of irreducible components , each with its own mass @xmath169 . let us denote @xmath170 at the leading order in @xmath146 , the diagram in fig . [ fig : mg]b generates a contribution to light gaugino masses given by @xmath171 each of the contributions in the sum in the rhs of eq . ( [ eq : gauginochiral ] ) arises at at the scale @xmath169 at which the corresponding chiral superfield lives . if the heavy chiral superfields split in two conjugated representations @xmath172 with a mass term in the form @xmath173 , eq . ( [ eq : gauginochiral ] ) still holds with @xmath174 and a factor 2 multiplying the rhs . let us now discuss the size of the typical chiral contribution to gaugino masses @xmath175 and compare it with the typical size of the tree level scalar soft masses @xmath176 in eq . ( [ eq : mm ] ) . let us consider for simplicity the case in which the scalar masses are due to the exchange of a single heavy vector and the irreducible ( under @xmath49 ) components of the physical chiral superfields have definite charges @xmath177 under the corresponding generators . as for the dynamics giving rise to gaugino masses , let us assume that there are no bare mass terms in the superpotential , i.e. @xmath178 . then both @xmath179 and @xmath180 arise from the same trilinear term in @xmath36 . under the above assumptions , we have @xmath181 where @xmath9 is the charge of the scalar acquiring the mass @xmath182 . while the loop factor @xmath164 suppresses @xmath175 compared to @xmath182 by a @xmath183 factor , the expressions in eqs . ( [ eq : comparison ] ) may give rise to several enhancements of @xmath184 reducing or even eliminating the loop hierarchy : * in the context of grand unified theories the heavy vectors contributing to the soft scalar masses are either one ( as in the case of the minimal possibility so(10 ) , see section [ sec : guidelines ] ) or a few , unless the unified group is very large . on the other hand , the gaugino masses may always get contribution from several chiral messengers . * sfermion and gaugino masses depend on different group factors . sfermions can get a mild suppression if @xmath185 . this is indeed what turns out to happen in simple models , as we will see in section [ sec : guidelines ] . * the heavy vector masses whose exchange generates @xmath182 collect all the vevs breaking the corresponding charge @xmath9 . the scalar mass @xmath182 is therefore suppressed by all such vevs . on the other hand , gaugino masses are only suppressed by the vevs related to supersymmetry breaking by superpotential interactions @xmath186 . unless some of then have @xmath187 , the vevs suppressing gaugino masses will be a subset of the vevs suppressing scalar masses , thus leading to an enhancement of gaugino masses . in the presence of an hierarchy between the vevs related to supersymmetry breaking and some of the other , @xmath9-breaking vevs , this enhancement can be quite large . * different couplings @xmath186 can appear in the numerator and denominator of the expression @xmath188 . this is likely to be the case as a consequence of the relation @xmath189 , which can be satisfied without the need of cancellations only in the case in which the fields charged under @xmath9 do not have vevs in both the @xmath7 and scalar components . if the couplings appearing in the numerator and the denominator are hierarchical , gaugino masses can be sizably enhanced ( or , in this case , further suppressed ) . the study of simple models shows that indeed the enhancement factors above can naturally arise ( see section [ sec:1610 ] below and @xcite ) . in particular , the first two factors reduce the hierarchy @xmath184 by factors 3 and @xmath190 respectively . the third factor gives at least a factor @xmath191 enhancement . the @xmath183 hierarchy arising from the loop factor is thus reduced by a factor 10 , which is enough to bring the sfermions within the reach of the lhc . such a milder hierarchy may even be necessary in the light of the bounds on the higgs mass , which require the stops not to be too light . still , the residual @xmath192 hierarchy can then be easily fully eliminated by the remaining two factors in the above list . besides gaugino masses , which can be seen to arise from one loop corrections to the gauge kinetic function , a number of soft terms can be generated or get a contribution from the one - loop corrections to the khler . the latter can be computed by using the general results in @xcite , which give @xmath193 - 2 \operatorname{tr}\left [ m^2_v \left ( \log \frac{m^2_v}{\lambda^2 } - 1 \right ) \right ] \right ] , \ ] ] where @xmath194 are functions of the chiral superfields , @xmath195 is our canonical khler @xmath196 and the indexes run on the heavy vector and chiral superfields . as in the case of gaugino masses , the soft terms might get a contribution from both heavy vector and chiral superfields running in the loop . as the contribution to one loop soft terms are highly model dependent , we just remind and collect their general expression in terms of @xmath197 . let us expand @xmath197 in terms of powers of @xmath9 and @xmath6 around @xmath68 . the relevant terms are @xmath198 where @xmath199 , @xmath200 , are hermitian , @xmath201 , @xmath202 symmetric and all are dimensionful . we have omitted @xmath203 terms , which are well - known to destabilize the hierarchy @xcite . their absence can be ensured for example by requiring that there are no light chiral fields with the same quantum numbers as @xmath6 . the first term @xmath199 gives rise to the following `` @xmath124-terms '' @xmath204 ( and to a two loop contribution to scalar soft masses ) , where @xmath205 is the scalar component of @xmath9 . the second term @xmath201 generates a contribution to the `` @xmath3-term '' in the superpotential @xmath206 and the fourth term @xmath202 a contribution to the `` @xmath207-term '' @xmath208 a more comprehensive discussion of the @xmath3-term and the @xmath3 problem can be found in section [ sec : mu ] . finally , @xmath200 gives 1-loop contributions to soft scalar masses @xmath209 that add to the tree level contributions in eq . ( [ eq : mm ] ) . additional one - loop contributions to soft scalar masses can come from an induced 1-loop fayet - iliopoulos term @xcite associated for example to the heavy @xmath49-singlet generators , in particular to those involved in the mediation of supersymmetry breaking at the tree level . such terms vanish if the heavy chiral mass matrix and the matrix of their couplings to the spurion @xmath6 are diagonal in the same basis ( in which case the condition in eq . ( [ eq : pure ] ) is also automatically satisfied ) or if the latter matrix of couplings is hermitian in one basis in which the mass matrix is diagonal @xcite . this completes the list of the soft terms arising at one loop . two loop corrections to the scalar soft masses can also arise , of course , as in standard loop gauge mediation , and be sizable in the presence of an enhancement of one - loop gaugino masses @xcite . we now consider the possibility to obtain a phenomenologically viable model from the general formalism discussed so far . we will see that clear model building guidelines emerge from this analysis , leading , in economical schemes , to peculiar predictions for the pattern of mssm sfermion masses . in particular , we will identify the assumptions underlying such predictions . in a phenomenologically viable model , the unbroken gauge group @xmath49 should contain the sm group , @xmath210 , and the light superfield content should contain the mssm spectrum , @xmath211 , in standard notations , where @xmath212 is the family index . we assume that the full gauge group @xmath46 is a simple , grand - unified group , motivated by the well known successful predictions of the sm fermion gauge quantum numbers , of the strong coupling in the mssm , and of the unification scale in the phenomenologically allowed region . the candidates for the unified group @xmath46 in a four - dimensional theory are su(@xmath213 ) , @xmath214 , so(4@xmath215 + 2 ) , @xmath216 , and the exceptional group @xmath52 @xcite . in the following we will focus on the smallest ( or unique ) representatives of each class , su(5 ) , so(10 ) , and @xmath52 . we want the mssm sfermions to get a positive , @xmath217 mass through tree level gauge mediation . the general form of such mass terms is given in eq . ( [ eq : mm ] ) . the latter contains two contributions , corresponding to the two diagrams on the right - hande side in fig . [ fig : diagram ] . in order for the second contribution to play a role for sfermion masses , the corresponding chiral superfields should live in the same unified multiplet as the supersymmetry breaking source @xmath6 . this will not be the case in the models we consider ( as a consequence , for example , of a matter parity telling the supersymmetry breaking multiplet from the matter ones ) . on the other hand , the second contribution might contribute to the higgs masses , if some of the gauge generators have the same quantum numbers ( which is not the case in so(10 ) , the unified group we will consider in greater detail ) . the mssm sfermions then get their tree level soft masses from the first term in eq . ( [ eq : mm ] ) only . in order for @xmath218 to be non - vanishing , the heavy generator @xmath59 must be a sm singlet , since @xmath91 is . we therefore need a group @xmath46 with rank 5 at least . this means that su(5 ) can not give rise to tree level gauge mediation , while so(10 ) and @xmath52 are in principle suitable . let us first consider the `` minimal '' option , so(10 ) , which has also the well known virtue to be able to accommodate a whole mssm family in a single irreducible spinorial representation . we will make a few considerations on the @xmath52 option at the end of this section . in so(10 ) there is exactly one ( up to a sign ) ortho - normalized heavy sm - singlet generator , @xmath219 , where @xmath220 is the su(5 ) invariant so(10 ) generator . the quantum numbers of the so(10 ) representations with dimension @xmath221 under @xmath222 are given in table [ tab : x ] . @xmath223 the values of the @xmath222 quantum numbers are crucial because the soft terms turn out to be proportional to those charges . from eq . ( [ eq : mm ] ) we obtain in fact @xmath224 where @xmath225 is the @xmath222-charge of the sfermion @xmath14 and @xmath78 is the mass if the vector superfield associated to the generator @xmath222 ( note that the gauge coupling and the normalization of the generator @xmath226 cancel in eq . ( [ eq : mmf ] ) ) . in order to predict the pattern of the tree level sfermion masses , we then just need to specify the embedding of the three mssm families into so(10 ) , which we will do through their su(5 ) embedding into three light @xmath227 , @xmath212 . we use two constraints to determine the embedding of the @xmath228 into so(10 ) representations . the first one is related to quite a nice feature eq . ( [ eq : mmf ] ) : the soft terms turn out to be family - universal , thus neatly solving the supersymmetric flavor problem . provided , of course , that the three families of each of the mssm matter multiplets are embedded in the same type of so(10 ) representation , which we will assume in order to ensure that family - universality indeed holds . on top of that , we want the mssm sfermion soft masses in eq . ( [ eq : mmf ] ) to be positive in order to avoid spontaneous symmetry breaking of color , electric charge , or lepton number at the scale @xmath182 . clearly , the standard embedding of a whole family into a 16 of so(10 ) would not work , as it would lead to negative masses for the sfermions in either the @xmath229 or the 10 of su(5 ) . this is in turn related to the tracelessness of the so(10 ) generators , and in particular of @xmath222 . as a consequence , whatever is the so(10 ) representation in which we choose to embed a given mssm matter multiplet with positive soft mass , that representation will necessarily contain extra fields with negative soft masses . this apparent obstacle can be easily overcome by splitting the so(10 ) representation containing the mssm multiplet through so(10 ) breaking , in such a way that the extra fields with negative soft masses acquire a large supersymmetric mass term . the negative soft mass will then represent a negligible supersymmetry breaking correction to that large ( positive ) mass . it turns out that such a splitting is actually expected to arise , as will see , a fact that reinforces the logical consistency of tree level gauge mediation . we are now ready to discuss the embeddings of the three @xmath230 and @xmath231 of su(5 ) containing the light mssm families in so(10 ) . as @xmath232 is positive , the possible choices depend on the sign of @xmath233 . we limit ourselves to the so(10 ) representations with @xmath221 , as in table [ tab : x ] . there are then only two possibilities : * @xmath234 . in this case we need to embed the @xmath230 s and @xmath231 s into so(10 ) representations containing @xmath235 and @xmath236 of su(5 ) with positive charges under @xmath222 . from table [ tab : x ] we see that the only possibility is to use three @xmath237 and three @xmath238 , @xmath212 , where we have explicitly indicated the su(5 ) decomposition , and to embed the @xmath231 s into the @xmath239 s , @xmath240 , and the @xmath230 s into the @xmath241 s , @xmath242 . the spare components @xmath243 , @xmath244 get negative soft masses and need to acquire a large supersymmetric mass term . * @xmath245 . in this case we need the @xmath230 s and @xmath231 s to have negative charges under @xmath222 . the only possibility is then to use three @xmath239 s as before and three @xmath246 , @xmath212 , with @xmath247 and @xmath248 . the spare components @xmath249 , @xmath250 , get negative or vanishing soft masses and need to acquire a large supersymmetric mass term . in both cases the chiral content of the theory is still given by three 16 of so(10 ) . we have implicitly neglected the possibility of mixed embeddings in which , for example , the @xmath251 s of su(5 ) are a superposition of the @xmath251 s in the @xmath241 s and @xmath239 s of so(10 ) . while this possibility is in principle not excluded , it would in general introduce a dependence of the sfermion soft masses on mixing parameters that are in general flavor violating , thus possibly spoiling the flavor universality result . the two possibilities above give rise to two definite predictions for the patter of sfermion soft masses at the scale @xmath78 : @xmath252 m^2_{\overline{5 } } = 2 m^2_{10 } \quad\text{if } \quad f_0^\dagger x f^{\phantom{\dagger}}_0>0 \\ m^2_{\overline{5 } } = \frac{3}{4 } m^2_{10 } \quad\text{if } \quad f_0^\dagger x f^{\phantom{\dagger}}_0<0 . \notag\end{gathered}\ ] ] to summarize , the latter predictions are based on the following hypotheses : `` minimal '' unified gauge group so(10 ) , embedding of the mssm families in the so(10 ) representations with dimension @xmath221 not containing the goldstino , and absence of mixed embeddings to automatically preserve flavor - universality . the predictions on the ratios @xmath253 in eq . ( [ eq : prediction ] ) are peculiar enough to make a possible experimental test at the lhc a strong hint for tree level gauge mediation . as for the source of supersymmetry breaking , @xmath117 , we need @xmath6 to have a non - vanishing charge under @xmath222 . if we limit ourselves again to representations with @xmath221 , the only possibility is that @xmath6 has a component in the `` right - handed neutrino '' direction of a 16 or a @xmath254 . with the sign conventions we adopted , a component in a 16 gives a positive contribution to @xmath233 , while a component in a @xmath254 gives a negative contribution . we now want to show that the two embeddings of the light mssm families described above can be obtained in a natural way . we have to show that it is possible to split the so(10 ) representations in which the mssm fields are embedded in such a way that the extra fields ( with negative soft masses ) get a heavy supersymmetric mass term from so(10 ) breaking . it will turn out that the so(10 ) breaking vevs of a @xmath255 , essential to break so(10 ) to the sm ( unless representations with @xmath256 are used to reduce the rank ) just provide the needed splitting . the fact that such vevs make heavy precisely the components of the so(10 ) representations that get a negative soft supersymmetry breaking mass reinforces the logical consistency of this framework . in the following , we first discuss the @xmath257 embedding in a general , top - bottom perspective , obtaining a generalization of the model in @xcite , and discuss the conditions for a pure ( non mixed ) embedding . we then discuss the possibility of a @xmath258 embedding . let us consider the embedding associated to the case @xmath259 . we assume the existence of a matter parity symmetry that tells matter superfields from higgs superfields . let 16 , @xmath254 be the so(10 ) multiplets breaking so(10 ) to su(5 ) ( we can always choose the basis in the space of the 16 ( @xmath254 ) representations in which a single 16 ( @xmath254 ) gets a vev in its scalar component ) . the most general renormalizable superpotential involving 16 , @xmath254 , @xmath239 , @xmath241 , @xmath212 , and invariant under a matter parity under which the so(10 ) higgs fields 16 , @xmath254 are even and the the matter fields are odd is @xmath260 where @xmath261 takes care of providing a vev to the 16 , @xmath254 in the sm - singlet direction and does not depend on the matter fields ( but can involve additional even fields , where @xmath222 is an so(10 ) singlet . ] ) . the term @xmath262 is just what needed to split the su(5 ) components of the @xmath237 and of the @xmath238 and make heavy the unwanted components @xmath243 and @xmath263 . once 16 acquires a vev @xmath85 in its singlet neutrino component , in fact , a mass term is generated for those components , @xmath264 the singlet neutrinos @xmath265 remain light at the renormalizable level but can get a mass at the non - renormalizable level through the operator @xmath266 . it is remarkable that the components acquiring a large mass are precisely those that get a negative soft mass term . on the other hand , this is only true in the limit in which the @xmath267 mass term in eq . ( [ eq : w1 ] ) can be neglected . in the presence of a non negligible @xmath267 , in fact , the full mass term would be @xmath268 which would give rise to a mixed embedding of the light @xmath230 s in the @xmath239 s and @xmath241 s . in order to abide to our assumptions , which exclude the possibility of mixed embeddings , such a @xmath267 term should be absent . this can be easily forced by means of an appropriate symmetry . let us however relax for a moment that assumption in order to quantify the deviation from universality associated to a small , but non - negligible @xmath267 . the mssm sfermions in the @xmath235 of su(5 ) receive in this case two contributions to their soft mass , a positive one associated to the components in the @xmath241 s , proportional to @xmath269 , and a negative one associated to the components in the @xmath239 s , proportional to @xmath270 . the soft mass matrix for the light sfermions in the @xmath235 of su(5 ) can be easily calculated in the limit in which the @xmath267 mass term can be treated as a perturbation . in this limit , the light mssm fields in the @xmath235 of su(5 ) are in fact @xmath271 and their soft scalar mass matrix at the scale @xmath78 is @xmath272 where @xmath176 is defined below . the mixed embedding induced by the mass term @xmath267 leads to flavor - violating soft - terms . setting @xmath273 allows to preserve the flavor blindness of the soft terms and to satisfy the fcnc constraints without the need of assumptions on the structure of the flavor matrices @xmath274 and @xmath267 . we therefore assume that @xmath267 is vanishing or negligible . we then have @xmath275 , @xmath276 , with the extra components @xmath243 and @xmath244 obtaining a large supersymmetric mass term @xmath277 , as desired . the soft masses for the light sfermions are [ eq : prediction1 ] @xmath278 as anticipated in eq . ( [ eq : prediction ] ) . the reason for the factor @xmath279 will become clear in a moment . we now need to identify the embedding of the mssm higgs superfields and obtain the mssm superpotential for them , in particular the mssm yukawa interactions . it is useful to discuss the yukawa interaction in su(5 ) language . the up quark yukawa interactions arise from the su(5 ) operator @xmath280 where @xmath281 contains the mssm up higgs . as @xmath276 , the operator in eq . ( [ eq : l1 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath281 has a component into a @xmath282 of so(10 ) , @xmath283 , with @xmath284 where @xmath285 measures the size of the @xmath281 component from 10 representations of so(10 ) ( a basis in the space of the 10 representations can always be chosen such that @xmath281 is contained in a single one , the @xmath282 ) . the operator in eq . ( [ eq : l1 ] ) will then originate as @xmath286 the down quark and charged lepton yukawa interactions arise at the renormalizable level from the su(5 ) operator @xmath287 where @xmath288 contains the mssm down higgs . as @xmath276 and @xmath275 , the operator in eq . ( [ eq : l2 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath288 has a component into a @xmath289 of so(10 ) , @xmath290 , with @xmath291 where @xmath292 measures the size of the @xmath288 component from 16 representations of so(10 ) . the operator in eq . ( [ eq : l2 ] ) will then originate as @xmath293 it is tempting ( and economical ) to identify the @xmath289 with 16 , the field whose vev breaks so(10 ) to su(5 ) , in which case @xmath294 and the mass of the heavy extra components @xmath243 and @xmath244 in eq . ( [ eq : unwanted1 ] ) turns out to be proportional to the corresponding light fermion masses ( up to non - renormalizable corrections needed to fix the light fermion mass ratios ) . having introduced the mssm higgs fields , let us now discuss their soft mass terms . to summarize the previous discussion , with our @xmath221 representation content , the up ( down ) higgs superfield @xmath295 ( @xmath296 ) can be embedded in either 10 s or 16 s ( @xmath254 s ) of so(10 ) , in both cases through the embedding into a @xmath281 ( @xmath288 ) of su(5 ) . we have denoted by @xmath285 ( @xmath297 ) the overall size of the @xmath295 ( @xmath296 ) component in the 10 s . the overall size of the component in the 16 s ( @xmath254 s ) is then measured by @xmath298 ( @xmath292 ) . correspondingly , the higgs soft masses get two contributions from the first term in eq . ( [ eq : mmf ] ) proportional to two different @xmath222 charges : [ eq : mmh ] @xmath299 let us now consider gaugino masses . a general discussion of all possible contributions to gaugino masses in the embedding we are considering and in the presence of an arbitrary number of so(10 ) representation with @xmath221 would be too involved . we then consider a few examples meant to generalize the case considered in @xcite and to illustrate the general properties discussed in section [ sec : oneloop ] . let us begin by illustrating in more detail the structure of supersymmetry breaking . with the representation content of table [ tab : x ] , supersymmetry breaking can be associated to the @xmath7-term vevs of superfields in 16 , @xmath254 , 45 , 54 representations ( the ones containing sm singlets ) . however , only the 16 , @xmath254 , whose singlets have non - vanishing @xmath222-charges , can contribute to tree level soft masses . let us call @xmath300 , @xmath301 the matter parity even superfields in the 16 and @xmath254 representations of so(10 ) . in a generic basis , we can parametrize the vevs of their singlet components as @xmath302 the @xmath29-term condition for the @xmath222 generator requires @xmath303 while gauge invariance gives @xmath304 sfermion masses are proportional to @xmath305 ( due to the factor 5 in the definition of @xmath176 ) , where @xmath306 by definition in the case we are considering . note that @xmath176 is suppressed by _ all _ vevs contributing to @xmath222 breaking . let us now comment on the vector contribution to gaugino masses . let us assume to begin with that the @xmath254 s do not break supersymmetry . without loss of generality we can then assume that supersymmetry breaking is only associated to @xmath307 . the gauge invariance condition then gives @xmath308 , i.e. a vev for both the @xmath7-term and scalar components is not allowed . since the @xmath7-term and scalar components belong to different irreducible representations , no vector contribution to gaugino masses is generated by the 16 s . a vector contribution can still be generated by the @xmath7-term vev of a 45 , for example , for which the gauge invariance condition does now prevent a vev in both the scalar and @xmath7-term component . or , it can be generated by the @xmath7-terms of the 16 s if some of the @xmath254 also breaks supersymmetry and cancels the contribution of the 16 to eq . ( [ eq : gaugeinv ] ) . let us next consider the chiral contribution to gaugino masses . the massive components @xmath243 and @xmath263 of the matter superfields will act as chiral messengers if they are coupled to supersymmetry breaking . let us then consider as before the case in which the @xmath254 s do not break supersymmetry , supersymmetry breaking is provided by the @xmath7-term vev @xmath7 of the singlet component of the @xmath309 and is felt by the chiral messengers through the @xmath310 interaction . let @xmath311 be the field whose vev gives mass to the @xmath243 , @xmath263 through the @xmath262 interaction , as in eq . ( [ eq : w1 ] ) . and let us assume that additional @xmath300 s and @xmath301 s get vevs in their scalar components . the chiral messengers @xmath243 , @xmath263 have therefore a supersymmetric mass @xmath312 and their scalar components get a supersymmetry breaking term mass term @xmath313 . the induced one loop chiral contribution to gaugino masses is then @xmath314 the tree level soft mass of the stop ( belonging to the 10 of su(5 ) ) is @xmath315 we can then compare stop and gaugino masses ( before radiative corrections ) . their ratio is particularly interesting , as the gaugino mass @xmath175 is at present bounded to be heavier than about @xmath316 , while @xmath317 enters the radiative corrections to the higgs mass . therefore , the ratio @xmath318 should not be too large in order not to increase the fine - tuning and not to push the stops and the other sfermions out of the lhc reach . from the previous equations we find @xmath319 eq . ( [ eq : ratio ] ) illustrates all the enhancement factors discussed in section [ sec : oneloop ] that can compensate the loop suppression of gaugino masses . the factor 3 corresponds to the number of chiral messenger families ( @xmath320 for @xmath321 ) contributing to gaugino masses , to be compared to the single vector messenger generating sfermion masses at the tree level . the factor @xmath190 comes from the ratio of charges @xmath322 suppressing the stop mass in eq . ( [ eq : comparison ] ) . the factor @xmath323 is the ratio of the vev suppressing gaugino masses ( the one related to supersymmetry breaking though superpotential interactions , @xmath324 ) , and the combination of vevs suppressing sfermion masses ( all of them ) . note that in the presence of hierarchies of vevs , the factor @xmath325 can be large . finally @xmath326 represents a combination of couplings that can further enhance ( or suppress , in this case ) gaugino masses . all in all , we see that the loop factor separating @xmath317 and @xmath175 is partially compensated by a combination of numerical factors : @xmath327 ( leading to @xmath328 for @xmath329 ) becomes at least @xmath330 ( leading to @xmath331 for @xmath329 ) . a largish value of the factors @xmath325 or @xmath326 can then further reduce the hierarchy and even make @xmath332 , if needed . let us now consider the second type of embedding identified above , corresponding to @xmath333 . the most general renormalizable superpotential involving 16 , @xmath254 and @xmath239 , @xmath334 , @xmath212 and invariant under matter parity is @xmath335 the term @xmath336 is just what needed to split the su(5 ) components of the @xmath237 and of the @xmath246 and make heavy the unwanted components @xmath249 and @xmath337 . once 16 acquires a vev @xmath85 , in fact , a mass term is generated for those components , @xmath338 it is remarkable that also in this case the components acquiring a large mass are precisely those that get a negative soft mass term . on the other hand , this is only true in the limit in which the @xmath267 mass term in eq . ( [ eq : w2 ] ) can be neglected . in order to abide to our pure embedding assumption , we will neglect such a term . let us note , however , that such a term should arise at some level in order to make the @xmath339 s components heavy . note that the @xmath340 s do not affect gauge coupling unification at one loop and can therefore be considerably lighter than the gut scale , consistently with the required smallness of @xmath267 . the soft masses for the light sfermions are now @xmath341 unfortunately , the embedding we are discussing can not be implemented with renormalizable interactions and @xmath342 representations only . the problem is obtaining the yukawa interactions . let us consider the up quark yukawas , arising as we saw from the su(5 ) operator in eq . ( [ eq : l1 ] ) . given its size , we expect at least the top yukawa coupling to arise at the renormalizable level . as in the present case @xmath343 , the operator in eq . ( [ eq : l1 ] ) can arise at the renormalizable level from a so(10 ) invariant operator only if @xmath281 has a component in a so(10 ) representation coupling to @xmath344 . and the lowest dimensional possibility containing the 5 of su(5 ) is the 210 . for this reason , we do not pursue this possibility further here , although models with large representations are not a priori excluded . we close this section with a few considerations about the possibility to identify the unified group with @xmath52 . such a possibility looks particularly appealing in the light of what above . we have seen in fact that the most straightforward possibility to realize tree level gauge mediation in so(10 ) requires the matter superfield content to include three @xmath345 , @xmath212 . this is precisely what @xmath52 predicts . the fundamental of @xmath52 , in fact , a representation of dimension 27 , decomposes as @xmath346 the matter content needed by the @xmath345 embedding can therefore be provided in the context of @xmath52 by three matter @xmath347 , @xmath212 , and the @xmath289 and @xmath282 needed to accommodate the higgs fields can also be provided by a single higgs @xmath348 . all yukawas can then in principle follow from the single @xmath52 interaction @xmath349 we postpone the analysis of this promising possibility to further study . in this section , we discuss a few approaches to the @xmath3-problem in the context of tree level gauge mediation . let us remind what the @xmath3-problem is . any supersymmetric extension of the sm must contain two higgs doublet chiral superfields @xmath350 , @xmath351 , with hypercharges @xmath352 , within the light spectrum @xmath9 . moreover , the lagrangian must contain a mass term for their higgsino ( fermion ) components , @xmath353 . it should also contain a corresponding term for the scalar components , @xmath354 , where @xmath207 is a dimension two parameter . the higgsino mass @xmath3 is constrained to be in the window @xmath355 by the present bounds on chargino masses and by naturalness considerations . this coincides with the window for the supersymmetry breaking scale in the observable sector , @xmath356 . it is then tempting try to establish a connection between these two a priori independent scales , in such a way that @xmath357 when @xmath358 , thus making the coincidence of the two scales not accidental . this is the @xmath3-problem . in the absence of such a connection , there would be no reason why @xmath3 should not be of the order of a much larger , supersymmetry conserving scale such as the gut or the planck scale . or , if a symmetry or some other independent principle suppressed @xmath3 , there would be no reason why @xmath3 should not be much smaller . fermion mass terms such as @xmath353 belong to the list of possible soft supersymmetry breaking mass terms @xcite . the reason why they are usually omitted from the mssm effective soft supersymmetry breaking lagrangian is that they can be always reabsorbed in the superpotential ( through appropriate additions to the scalar soft lagrangian ) . moreover , most models of supersymmetry breaking , including the ones we are considering , do not generate such supersymmetry breaking fermion mass terms . we can then assume that the higgsino mass term arises from a corresponding term in the superpotential . the problem is then to relate the coefficient of that ( supersymmetric ) superpotential term , @xmath359 , to the supersymmetry breaking scale in the observable sector , which in our case is given by @xmath360 . we discuss in the following three possible connections . one is peculiar of tree level gauge mediation , the other two have been considered in other contexts , but have specific implementations in tree level gauge mediation . we classify them according to the dimension @xmath29 of the so(10 ) operator from which the @xmath3 term arises . note that we are not addressing the origin of the smallness of @xmath182 and @xmath3 compared to the plank scale , just their connection . the three options we consider are : @xmath361 : : : @xmath3 comes from the operator @xmath362 . it is the supersymmetry breaking scale to be derived from @xmath3 , and not viceversa : @xmath363 , where @xmath364 , and @xmath365 . @xmath366 : : : @xmath3 comes from the operator @xmath367 . the light sm singlet @xmath368 gets a vev from a potential whose only scale is @xmath182 , so that @xmath369 . @xmath370 : : : @xmath3 comes from the operator @xmath371 , so that @xmath372 . let us discuss each of those possibilities in turn . such a possibility was anticipated in @xcite , where however no concrete implementation was given . let us consider the @xmath257 embedding . as discussed in section [ sec:1610 ] , @xmath350 is a superposition of the `` up higgs - type '' components in the @xmath254 s and @xmath236 s ( with @xmath374 ) in the model . analogously , @xmath351 will be a superposition of the `` down higgs - type '' components in the ( @xmath375 ) @xmath376 s and @xmath236 s . the only possible @xmath373 origin of the @xmath3-term in the context of the full so(10 ) theory are then @xmath19 mass terms for the above @xmath254 s , @xmath377 s , and @xmath236 s . as said , we do not address the origin of such a small parameter in the superpotential , as we do not address here the smallness of the supersymmetry breaking scale . the latter can for example be explained by a dynamical mechanism . we want however to relate such mass parameters , in particular the coefficient of a @xmath378 mass term , to the supersymmetry breaking scale . this is actually pretty easy , as the tree level gauge mediation embedding we are considering provides all the necessary ingredients and the result arises from their simple combination . we have seen in fact that the model needs a @xmath377 , @xmath254 pair to get a vev in the sm singlet direction of the scalar component , in order to break so(10 ) to the sm . moreover , we have seen that an independent @xmath309 , @xmath379 pair is required to break supersymmetry trough the @xmath7-term vev of the sm singlet component in the @xmath309 . the simplest way to achieve such a pattern is through a superpotential like @xmath380 where @xmath222 , @xmath6 are so(10 ) singlets and @xmath381 . this is a generalization of an example in @xcite . finally , we have just reminded that the light higgses may have a component in @xmath377 , @xmath309 , @xmath254 , @xmath379 . let @xmath382 be the coefficient of the @xmath296 component in the @xmath309 and @xmath383 the coefficient of the @xmath295 component in the @xmath254 . then a @xmath3 parameter is generated in the form @xmath384 from the @xmath385 term in eq . ( [ eq : wvev ] ) . the parameter @xmath386 is therefore required to be in the window @xmath387 . in the limit @xmath388 , supersymmetry is unbroken and @xmath377 , @xmath254 acquire a vev that can be rotated in the sm singlet component @xmath389 . a non - vanishing @xmath3 , on the other hand , triggers supersymmetry breaking and induces an @xmath7-term vev for the singlet component of the @xmath309 , @xmath390 , with @xmath391 . we therefore have @xmath392 providing the desired connection between @xmath3 and the supersymmetry breaking scale . tree level gauge mediation plays a crucial role not only in providing the ingredients ( and no need to stir ) but also because it is the very so(10 ) structure providing the heavy vector messengers to relate in a single irreducible representation ( the 16 ) supersymmetry breaking ( the @xmath7-term vev of its sm singlet component ) and the down higgs entering the @xmath3-term ( the lepton doublet - type component of the 16 ) . in the appendix [ sec : w ] we provide an existence proof of a ( perturbative ) superpotential that i ) implements the mechanism above , thus breaking supersymmetry and so(10 ) to su(5 ) , ii ) further breaks su(5 ) to the sm , iii ) makes all the fields that are not part of the mssm spectrum heavy , in particular achieves doublet - triplet splitting . this is an implementation of the nmssm solution of the @xmath3-problem ( see e.g. @xcite and references therein ) . as we will see , the implementation of such a solution in the context of tree level gauge mediation avoids some of the problems met in ordinary gauge mediation . in order to implement the nmssm solution of the @xmath3-problem , an explicit term @xmath359 should be forbidden , for example by a symmetry ; the light fields @xmath9 should include a sm singlet @xmath368 , coupling to the higgses through the superpotential interaction @xmath393 ; and @xmath368 should develop a non - zero vev . a @xmath3 parameter will then be generated , @xmath394 . in the absence of terms linear or quadratic in @xmath395 in the superpotential , the scale of a vev for @xmath368 can only be provided by the supersymmetry breaking terms in the soft lagrangian , @xmath396 , in which case @xmath397 , as desired . in order to generate a non - zero vev for @xmath368 , one would like to have a negative soft mass for @xmath368 at the weak scale , along with a stabilization mechanism for large values of the fields . in ordinary gauge mediation this is not easy to achieve . while the stabilization can be simply provided by a @xmath398 term in @xmath150 , as in the nmssm ( or by a quartic term in @xmath399 extensions of the mssm @xcite ) , the soft mass term of @xmath368 vanishes at the messenger scale because @xmath368 is typically a complete gauge singlet . a non - vanishing negative mass term is generated by the rge running but it is typically too small . another problem is that the higgs spectrum can turn out to be non - viable @xcite . a sizable soft mass can still be generated by coupling @xmath368 to additional heavy fields . such possibilities can be implemented in our setup by promoting @xmath368 to an so(10 ) singlet and coupling it to the higgses through a @xmath400 or a @xmath401 coupling to the so(10 ) representations containing ( a component of ) the higgs fields . tree level gauge mediation offers a different avenue . a sizable , negative soft mass term for @xmath368 can in fact be generated by embedding @xmath368 in a @xmath254 of so(10 ) ( this is the only choice within the fields in table [ tab : x ] ) . on the other hand , the stabilization of the potential for @xmath368 is not straightforward . a sizable @xmath398 term is not expected to arise , as it should involve a so(10 ) operator with three @xmath254 . however , the @xmath398 term can be replaced by a term involving a second light singlet @xmath213 , @xmath402 the latter can come from a @xmath403 coupling , if @xmath213 is in the 126 singlet , or from a @xmath404 coupling , where @xmath213 is the @xmath405 singlet and @xmath406 gets a vev . the scalar potential for @xmath407 can be written as @xmath408 where @xmath409 is the mssm scalar potential with @xmath410 , @xmath411 , and @xmath412 or @xmath176 depending on whether @xmath213 comes from a 126 or a 16 . we have neglected the @xmath124-terms , which play a role in explicitly breaking @xmath413-symmetries that could lead to massless states . the potential above has a minimum with a sizable @xmath414 , and a @xmath3 parameter whose size is controlled by @xmath326 . finally , let us consider the possibility to generate the @xmath3 parameter through a @xmath370 correction to the khler in the form @xmath415 , as in the giudice masiero mechanism @xcite . the @xmath7-term vev @xmath416 of @xmath6 would give in this case @xmath417 . we show first that the operator above can not arise at the tree level from integrating out heavy vector or chiral superfields . the corrections to the khler obtained by integrating out heavy vector superfields are given in eqs . ( [ eq : keff2 ] ) . all terms are at least of second order in @xmath418 and no trilinear term is present . moreover , no sizable trilinear term can be obtained through the vev of @xmath66 , as by definition the scalar components of @xmath66 do not get a vev ( and an @xmath7-term vev would give an additional @xmath419 suppression ) . a similar conclusion can be obtained for the corrections one obtains by integrating out chiral superfields @xmath119 with mass @xmath420 . we have seen in section [ sec : tree ] that the equations of motion allow to express @xmath119 in terms of the light fields as in eq . ( [ eq : phih ] ) . since @xmath128 contains terms at least trilinear in the fields , the expression for @xmath119 is at least quadratic in the light fields . when plugging eq . ( [ eq : phih ] ) in the canonical khler for @xmath119 one gets again terms that contain at least four light fields , with none of them getting a vev in the scalar component . therefore , no operator @xmath421 can be generated at the tree level by integrating out heavy fields . let us now consider the possibility that the @xmath370 operator above is obtained at the one loop level . this possibility raises two issues . first , @xmath3 would be suppressed compared to , say , the stop mass @xmath317 by a loop factor @xmath422 . as for the case of gaugino masses vs sfermion masses , such a large hierarchy would lead to sfermions beyond the reach of the lhc and a significant fine - tuning . however , as we will see , this problem can be overcome in the same way as for the gaugino masses . we will see in fact in an explicit model that @xmath3 and @xmath423 get a similar enhancement factor . the second issue is the well known @xmath3-@xmath207 problem . @xmath207 is a dimension two parameter generated , as @xmath3 , at the one loop level . therefore , we expect an order of magnitude separation between @xmath424 and @xmath3 : @xmath425 . this is however tolerable in a scheme in which @xmath426 , with @xmath427 and @xmath428 . the explicit model will show that the above pattern can be achieved in the large @xmath429 regime . in turn , the large @xmath429 regime raises a new issue . the minimization of the mssm potential shows in fact that large @xmath429 corresponds to small @xmath430 , while in the situation we want to reproduce , @xmath431 , we expect @xmath432 . in order to make @xmath429 large we therefore need to cancel the contribution to @xmath207 we get at one loop with an additional contribution , at least in the specific example we consider . such a cancellation may not be required in different implementations of the one - loop @xmath370 origin of the @xmath3 parameter . that is why we believe it is worth illustrating the example below despite the cancellation that needs to be invoked . let us consider as before a model involving the following @xmath374 fields : @xmath377 , @xmath254 , @xmath309 , @xmath379 , @xmath236 , with @xmath389 , @xmath433 , @xmath434 . let us denote the coefficients of the @xmath295 and @xmath296 components in the above so(10 ) representations as follows : @xmath435 , @xmath436 , @xmath437 , @xmath438 , @xmath439 , @xmath440 , where @xmath441 , @xmath442 , @xmath443 , @xmath444 , etc . the notation is in agreement with the definition of @xmath445 , @xmath446 in section [ sec:1610 ] . the @xmath3 and @xmath207 parameters , as the gaugino masses , get a vector and a chiral one - loop contribution , see eqs . ( [ eq : dkloop],[eq : dkloop2],[eq : mu],[eq : bmu ] ) . the vector contribution turns out to be [ eq : muvec ] @xmath447 as in the case of gaugino masses , the vector contribution to @xmath3 is suppressed with respect to the sfermion masses by a full loop factor . we therefore need a larger chiral contribution in order to reduce the hierarchy between @xmath3 and @xmath317 . let us then consider the one - loop chiral contribution associated to the superpotential @xmath448 that is easily found to be vanishing because of a pq symmetry of the superpotential . such a pq symmetry can however be broken by adding a term @xmath449 to the above superpotential , coming for example from the non - renormalizable so(10 ) operator @xmath450 after @xmath254 gets its vev ( note that @xmath451 would give @xmath452 ) . en passant , the singlet mass term in eq . ( [ eq : wpq2 ] ) is nothing but the right - handed neutrino majorana mass term entering the see - saw formula for light neutrino masses . note however that no light neutrino mass is generated here , as the light lepton doublets do not have yukawa interactions with the `` right - handed neutrinos '' , @xmath265 . once the pq symmetry is broken by the mass term in eq . ( [ eq : wpq2 ] ) , the @xmath3 and @xmath207 parameters get a chiral one - loop contribution given by [ eq : muchi ] @xmath453 where @xmath454 , @xmath455 are the top and bottom yukawa couplings respectively and the functions @xmath13 , @xmath47 are given by @xmath456 we have assumed the yukawa couplings @xmath274 , @xmath457 to be hierarchical in the basis in which the down yukawa matrix is diagonal . we can see from eq . ( [ eq : muchi ] ) that the one loop chiral contribution to @xmath3 is comparable to the corresponding contribution to @xmath423 if i ) @xmath458 , which corresponds to the large @xmath429 regime ( remember that the bottom mass is given by @xmath459 , where @xmath460 ) ; ii ) @xmath461 , @xmath462 ; iii ) @xmath463 ; iv ) @xmath464 . if the above conditions are satisfied , @xmath465 and both parameters can easily be enhanced , as explained in section [ sec : gauchi ] , for example because @xmath466 . the only non - trivial condition is the large @xmath429 one . remember in fact that @xmath429 is determined by @xmath207 through the minimization of the mssm potential , which gives @xmath467 therefore large @xmath429 , i.e. small @xmath468 , requires a small @xmath207 . this is in contrast with the situation we want to reproduce , @xmath431 . the rge evolution of @xmath207 from the scale at which it is generated ( @xmath469 ) down to the electroweak scale can reduce the value of @xmath207 but not enough to make it as small as we need . a significant rge contribution would in fact require @xmath470 , in contrast with the @xmath471 we are trying to reproduce . we are then forced to invoke a cancellation between the one - loop contribution to @xmath207 in eq . ( [ eq : muchi]b ) and an additional contribution . for example , a tree level contribution to @xmath207 can be obtained as in appendix b or in @xcite . in this paper we have considered what may be regarded as one of the simplest ways to communicate supersymmetry breaking from a hidden to the observable sector , through the tree level , renormalizable exchange of superheavy gauge ( gut ) messengers , and we have studied the general properties of such a tree level gauge mediation ( tgm ) scheme . we have first of all obtained the general structure of the tree - level soft terms arising from a supersymmetry breaking source that is part of a non - trivial gauge ( gut ) multiplet . this is most conveniently done in the effective theory in which the heavy vector superfields associated to the broken generators are integrated out at the tree level ( en passant , in appendix a , we summarized the procedure to integrate out vector superfields and addressed a few minor issues , such as the generalization to the non - abelian case and the role of gauge invariance in a consistent supersymmetric generalization of the expansion in the number of derivatives ) . the scalar soft terms then obtain the two contributions in eq . ( [ eq : mm ] ) , corresponding to the two diagrams on the right - hand side of fig . [ fig : diagram ] . only the first contribution is relevant for scalars that are not in the same gauge multiplet as the scalar partner of the goldstino ( or have not the same sm quantum numbers as some of the gut generators ) . because of the tracelessness condition , such a contribution gives both positive and negative soft masses . this potential phenomenological problem , which has long been considered as an obstacle to tree level supersymmetry breaking , is automatically solved in the models we consider because the fields getting a @xmath19 negative soft mass also get an @xmath472 positive , supersymmetric mass . gaugino masses do not arise at the tree level , but can be generated at the one - loop level , as in ordinary gauge mediation . they receive two contributions , from loops involving heavy vector or chiral superfields . the loop factor suppression of gaugino compared to sfermion masses must be at least partially compensated if the sfermions are to be within the lhc reach and the split - supersymmetry regime is to be avoided . we calculated in full generality the vector and chiral contributions to gaugino masses corresponding to the diagrams in fig . [ fig : mg ] . we have seen that the vector contribution is always suppressed by a full loop factor and is typically subdominant ( often vanishing ) . on the other hand , the chiral contribution is typically larger . we listed four potential enhancement factors that can ( do ) compensate , at least partially , the loop suppression : a larger number of ( chiral ) messengers contributing to gaugino masses than ( vector ) messengers contributing to sfermion masses ; group theoretical factors that in practice turn out to enhance gaugino masses ; the fact that sfermion masses are suppressed by all the vevs with non - vanishing gauge coupling to the vector messengers , while gaugino masses are suppressed only by the vevs that are related to supersymmetry breaking through superpotential interactions ; ratios of yukawa couplings appearing in the expression for the gaugino masses . in minimal models the first two factors partially compensate the @xmath422 loop factor , reducing it to the level of a tolerable ( and possibly necessary ) one order of magnitude hierarchy between gauginos and sfermions . the last two factors are more model - dependent but can give rise to larger enhancements . the general analysis of the tgm scheme allowed us to define the guidelines to obtain phenomenologically viable models from the general formalism and to identify the assumptions underlying the peculiar predictions one obtains . clear model building guidelines emerge , identifying so(10 ) and @xmath52 as the `` minimal '' grand - unified groups , while su(5 ) is found not to have the necessary structure ( rank @xmath473 ) to realize the tgm scheme . the so(10 ) possibility turns out to be quite appealing . it turns out in fact that the so(10 ) breaking vevs of a @xmath255 , important to break so(10 ) to the sm , typically make heavy precisely the components of the so(10 ) representations that need to be made heavy because of their negative soft supersymmetry breaking masses . this reinforces the logical consistency of the tgm framework . in so(10 ) , the tree level sfermion soft masses turn out to be proportional to their charges under the su(5)-invariant so(10 ) generator @xmath222 . we find two possible embeddings of the mssm superfields into so(10 ) representations , depending on whether @xmath474 is positive or negative . in the first case , @xmath475 , the three mssm families are embedded in three @xmath239 and three @xmath241 , @xmath212 . the quark doublets , the up quark singlets , and the lepton singlets , unified in 10 s of su(5 ) , are embedded in the @xmath239 s , while the lepton doublet and down quark singlets , unified in @xmath235 s of su(5 ) , are embedded in the @xmath241 s . they all get positive soft masses . the spare components in the @xmath239 s and @xmath241 s get superheavy , positive , supersymmetric mass terms ( and tev scale negative soft masses ) . in the second case , @xmath476 , the three mssm families are embedded in three @xmath239 and three @xmath334 , @xmath212 . the mssm fields in 10 s of su(5 ) are embedded in the @xmath334 s , while the ones unified in @xmath235 s of su(5 ) , are embedded in the @xmath239 s . as before , they all get positive soft masses . in both cases the chiral content of the theory is still given by three 16 of so(10 ) . an important property of the tgm soft terms is that they turn out to be family universal , thus solving the supersymmetric flavor problem . this property only depends on the hypothesis that the three mssm families are embedded in the same so(10 ) representations . mixed embeddings , in which the mssm fields are superpositions of fields in inequivalent so(10 ) representations , are also possible , but can spoil the flavor universality property . each of the two possible flavor - universal embeddings leads to specific and peculiar predictions for the soft masses at the gut scale : @xmath1 in the @xmath234 case and @xmath2 in the @xmath245 case , where @xmath477 and @xmath478 are common and family - independent soft masses for the fields in the @xmath235 and @xmath236 of su(5 ) respectively . the latter predictions are only based on i ) the use of the `` minimal '' unified gauge group so(10 ) , ii ) the embedding of the mssm families in the so(10 ) representations with dimension @xmath221 not containing the goldstino , and iii ) the absence of mixed embeddings to automatically preserve flavor - universality . the predictions on the ratios @xmath253 in eq . ( [ eq : prediction ] ) are determined by group theory factors and are peculiar enough to make a possible experimental test at the lhc a strong hint for tree level gauge mediation . the embedding into three @xmath257 s has the advantage that the large top yukawa coupling can be accounted for by a renormalizable superpotential interaction involving only low - dimensional ( @xmath479 ) representations for the chiral superfields . in the @xmath258 case , a @xmath480 representation of so(10 ) must be used to reproduce the top yukawa coupling at the renormalizable level . the @xmath52 option is also quite appealing , as the matter superfield content of the @xmath257 embedding is precisely the one obtained from three fundamentals @xmath347 of @xmath52 . the latter decompose in fact as @xmath481 under so(10 ) . we have postponed the investigation of this promising possibility to further study . finally , we have illustrated three possible approaches to the @xmath3-problem in tgm , which we classify according to the dimension @xmath29 of the so(10 ) operator from which the @xmath3-term arises . the @xmath373 option provides a new approach to the @xmath3-problem , peculiar of tgm . the idea is that the supersymmetry breaking scale turns out to coincide with the @xmath3 scale because supersymmetry is triggered by the same @xmath373 so(10 ) operator from which the @xmath3-term arises . we have provided an explicit realization of such a possibility in appendix b. the superpotential shown there also achieves supersymmetry breaking , so(10 ) breaking to the sm , and ensures that only the mssm fields survive below the breaking scale ( in particular it provides doublet - triplet splitting ) . while it is not meant to be simple or realistic , that superpotential represents a useful existence proof . the @xmath366 option is nothing but the nmssm solution of the @xmath3 problem , in which the @xmath3-term is obtained from the vev of a sm singlet superfield stabilized at the supersymmetry breaking scale . we pointed out that the above singlet can easily get a sizable , predictable , negative soft mass term in tgm . this makes giving a vev to the singlet easier than in ordinary gauge mediation ( where its soft mass usually vanishes before rge running ) , provided that the singlet potential can be made stable . the @xmath370 option is nothing but the giudice - masiero mechanism realized at the loop level , as in gauge mediation . the consequent loop hierarchy between the @xmath3-term and the sfermion masses can be reduced exactly as for the gaugino masses . we provided an explicit example , which however needs an extra contribution to the @xmath207 parameter in order to give rise to the necessary large @xmath429 . we thank andrea brignole and claudio scrucca for useful discussions and acknowledge partial support from the rtn european program `` universenet '' ( mrtn - ct-2006 - 035863 ) . we thank the galileo galilei institute for theoretical physics , where part of this work was done , for the hospitality . in this appendix , after a few general comments , we write the effective theory one obtains by integrating heavy vector superfields at the tree level and in unitary gauge in a generic , non - abelian , @xmath34 globally supersymmetric theory with renormalizable khler @xmath195 and gauge - kinetic function ( the superpotential @xmath150 is allowed to be non - renormalizable ) . the general prescription has been studied in @xcite . in particular , it has been shown in @xcite that the usual expansion in the number of derivatives @xmath482 can be made consistent with supersymmetry by generalizing @xmath482 to the parameter @xmath483 where @xmath484 is the number of fermion bilinears and @xmath485 the number of auxiliary fields from chiral superfields . with such a definition , a chiral superfield @xmath486 has @xmath487 and @xmath488 integrations and supercovariant derivatives have @xmath489 . such an expansion makes sense when supersymmetry breaking takes place at a scale much smaller than the heavy superfield mass @xmath4 and in particular when the @xmath7-terms and fermion bilinears from heavy superfields being integrated out are much smaller than @xmath4 . in the presence of vector superfields one should further assume that the @xmath29-terms and gaugino bilinears are small and should generalize eq . ( [ eq : n ] ) to account for the number @xmath490 of gauginos and the number @xmath491 of vector auxiliary fields . we claim that the correct generalization is @xmath492 according to which a vector superfield @xmath85 has @xmath487 . note that the double weight of @xmath29-terms compared to @xmath7-terms is consistent with eq . ( [ eq : dh ] ) . with such a definition , the initial lagrangian has @xmath493 , except for the gauge kinetic term , which has @xmath494 . chiral and vector superfields can then be integrated out at the tree level by using the supersymmetric equations of motion @xmath495 up to terms with @xmath496 when integrating out chiral superfields and @xmath497 when integrating vector superfields ( with the missing terms originating from the gauge - kinetic term having @xmath498 ) . from a physical point of view , we are interested not only in the expansion in @xmath215 but also , and especially , in the expansion in the power @xmath386 of @xmath499 . it is therefore important then to remark that using eqs . ( [ eq : susyeom ] ) amounts to neglecting terms with @xmath500 when integrating chiral superfields and @xmath501 when integrating out vector superfields . we are now ready to present out results on the effective theory obtained integrating out the heavy vector superfields in a generic supersymmetric gauge theory as above . we are interested in operators with dimension up to 6 ( @xmath502 ) in the effective theory . we can then use the equation @xmath503 . neglecting higher orders in @xmath386 , the latter equation can be rewritten as @xmath504 where @xmath66 is defined in eq . ( [ eq : goldstones ] ) , @xmath505 , the indices run over the broken generators , and @xmath506 is a function of the light vector superfields : @xmath507 ( m^2_{v2})_{ab } = \frac{g^4}{3 } \phi_0^ * t_a^h v^l v^l t^h_b \phi_0 + ( a \leftrightarrow b ) . \end{gathered}\ ] ] in order to solve eq . ( [ eq : vh ] ) for @xmath508 , we need to invert the field - dependent matrix @xmath506 . in the wess - zumino gauge for the light vector superfields , we get @xmath509 the effective contribution to the khler potential is @xmath510 where [ eq : keff2 ] @xmath511 \delta k^1_\text{eff } & = - 2 g^3 ( m_{v0}^2)^{-1}_{ab } ( \phi'^\dagger t^h_a \phi ' ) ( \phi'^\dagger \{v^l , t^h_b\ } \phi ' ) { \stepcounter{mysubequation}\tag{\theequation\protect{{\alph{mysubequation}}}}}\\ \delta k^2_\text{eff } & = - \frac{4}{3 } g^4 ( m_{v0}^2)^{-1}_{ab } ( \phi'^\dagger t^h_a \phi ' ) \phi'^\dagger ( t^h_b v^l v^l + v^l t^h_b v^l + v^l v^l t^h_b ) \phi ' { \stepcounter{mysubequation}\tag{\theequation\protect{{\alph{mysubequation}}}}}\\ & - g^4 ( m_{v0}^2)^{-1}_{ab } \left ( ( \phi'^\dagger \ { t^h_a , v^l \ } \phi ' ) ( \phi'^\dagger \ { t^h_b , v^l \ } \phi ' ) + \frac{1}{3 } ( \phi'^\dagger [ t^h_a , v^l ] \phi ' ) ( \phi'^\dagger [ t^h_b , v^l ] \phi ' ) \right ) . \notag\end{aligned}\ ] ] in recovering eq . ( [ eq : keff2]c ) we have used the identity @xmath512 where @xmath513 are the structure constants of the gauge group , the latin indices refer to broken generators and the greek one refers to an unbroken one . we are interested to soft supersymmetry breaking terms arising from eqs . ( [ eq : keff2 ] ) when some of the auxiliary fields get a vev . the relevant terms should contain up to two @xmath7-terms and one @xmath29-term ( eq . ( [ eq : dh ] ) ) . the only relevant terms are therefore those in ( [ eq : keff2]a ) . in this appendix we provide an example of a superpotential achieving supersymmetry breaking , so(10 ) breaking to the sm , ensuring that below the scale of this breaking only the mssm fields survive ( in particular providing doublet - triplet splitting ) and solving of the @xmath3-problem . we do not aim at being simple or realistic , we just aim at providing an existence proof . we include in this example representations with dimension @xmath514 . it would be interesting to obtain a dynamical supersymmetry breaking , in particular the @xmath7-term vev of a 16 of so(10 ) . so(10 ) will broken to the sm at a scale @xmath515 . below this scale only the mssm fields survive , in particular the higgs triplets are made heavy via a generalization of the dimopoulos - wilczek mechanism @xcite . the @xmath3-term is present in the theory in the form of a @xmath373 operator present at the gut scale and triggers supersymmetry breaking . @xmath516 is generated at the tree - level and turns out to be of the same order as the sfermion masses . the superpotential we use is @xmath517 where @xmath518 here we denote the fields according to their so(10 ) representation , except the so(10 ) singlet fields @xmath519 . the mass parameter @xmath386 is of the order of the tev scale ( we do not discuss the origin of such a small parameter here ) , while all other mass parameters are near the gut scale @xmath520 let us discuss the role of the different contributions to the superpotential and anticipate the vacuum structure and the spectrum . @xmath521 is responsible for supersymmetry breaking and the breaking of so(10 ) to su(5 ) : as we are going to show below , this part of the superpotential generates @xmath472 vevs for the scalar components of @xmath377 and @xmath522 along the su(5 ) singlet direction @xmath523 and @xmath524 and a supersymmetry breaking vev for the @xmath7-term component of @xmath525 along the su(5 ) singlet direction @xmath526 . it also provides small supersymmetry breaking vevs for the @xmath7-term component of @xmath222 @xmath527 and for the @xmath29-term of the vector superfield corresponding to the u(1)@xmath528 generator of so(10 ) @xmath529 this @xmath29-term vev will generate sfermion masses along the lines of section [ sec:1610 ] . this superpotential appears in eq . ( [ eq : wvev ] ) and is a generalization of an example in @xcite . @xmath530 contains the mssm yukawa couplings and provides supersymmetry breaking masses for heavy chiral superfields that will generate gaugino masses at 1-loop as in ordinary gauge mediation . the mssm matter is embedded in both the @xmath239 and the @xmath241 , as explained in section [ sec:1610 ] . the mssm higgs fields are linear combinations of different fields and have components in different representations , @xmath531 therefore the first term in @xmath530 contains the up - type yukawas , while the second and third terms provide down - type and charged lepton yukawas . the second term also gives a large mass to the additional fields @xmath532 and @xmath533 the latter are also the only fields that couple to the @xmath7-term vev in the @xmath525 and act as one - loop messengers of supersymmetry breaking . while this gives a subleading contribution to sfermion masses , it is the only source of gaugino masses in this model . + the role of @xmath534 is the breaking of su(5 ) to the standard model gauge group . it provides a large vev for the @xmath535 along the @xmath536 direction @xmath537 as needed for the dimopoulos - wilzcek mechanism . also @xmath538 and the sm singlet in the @xmath539 take large vevs . @xmath128 merely gives large masses to components in the @xmath525 and @xmath540 . note that since the @xmath541 does not contain su(5 ) singlets , the neutrino component in the @xmath525 stays massless as it should , being the dominant component of the goldstino superfield . @xmath542 takes care of the higgs sector : it keeps the mssm higgs doublets light and gives a large mass to the corresponding triplets . its last term provides the @xmath207 term because @xmath6 gets a small supersymmetry breaking vev and both @xmath543 and @xmath544 have components in the @xmath236 . the @xmath3 term is contained in @xmath521 , because @xmath544 has a component also in the @xmath525 . we are interested in a vacuum that does not break the sm gauge group . thus only that part of the superpotential which involve su(5 ) singlets is relevant for the determination of the ground state . we denote the singlets in @xmath545 by @xmath546 ( which is different than the notation used in the main text ) and the singlets in the @xmath547 by @xmath548 , where @xmath549 are the properly normalized fields corresponding to the @xmath536 and @xmath550 generators in so(10 ) . the relevant part of the superpotential is @xmath551 & + { \overline{s}^{\prime \prime}}\left ( -\frac{{\lambda_{3 } \,}}{2 } t + \frac{{\lambda_{3 } \,}}{2 } \sqrt{\frac{3}{2 } } b + { \lambda_{4 } \ , } u\right ) s + { \overline{s}}\left ( -\frac{{\lambda_{5 } \,}}{2 } t + \frac{{\lambda_{5 } \,}}{2 } \sqrt{\frac{3}{2 } } b + { \lambda_{6 } \ , } u \right ) { s^{\prime \prime}}\\[-1 mm ] & + { m_{45}}(b^2 + t^2 ) + { m_{54}}v^2 + { \lambda_{7 } \ , } v \left ( \frac{1}{2 } \sqrt{\frac{3}{5 } } t^2 - \frac{1}{\sqrt{15 } } b^2 \right ) . \notag\end{aligned}\ ] ] the @xmath7-term and @xmath552-term equations show that susy is broken ( @xmath553 ) and that all vevs are determined except @xmath554 , for which there exist three solutions , all yielding @xmath555 . this tree - level degeneracy is lifted by one - loop corrections which select the solution with @xmath556 . one can check that the vevs are given by @xmath557 s = m - \frac{m^2}{4 m } \left(\frac{1}{\lambda_1 ^ 2 } - \frac{1}{50 g^2}\right ) \qquad { \overline{s}}= m - \frac{m^2}{4 m } \left(\frac{1}{\lambda_1 ^ 2 } + \frac{1}{50 g^2}\right ) \notag \\ u = - \frac{3 \sqrt{5 } } { 2 } \frac{{\lambda_{5 } \,}}{{\lambda_{6 } \ , } { \lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \qquad { u^\prime}= - \frac{3 \sqrt{5 } } { 2 } \frac{{\lambda_{3 } \,}}{{\lambda_{4 } \ , } { \lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \\[1 mm ] v = \frac { \sqrt{15 } { m_{45}}}{{\lambda_{7 } \ , } } \qquad b = \frac{\sqrt{30}}{{\lambda_{7 } \ , } } \sqrt{{m_{45}}{m_{54 } } } \notag \\ f_{{s^\prime } } = - m m \qquad f_z = \frac{m^2}{2 { \lambda_{1 } \ , } } \qquad d_x = - \frac{m^2}{10 g}. \notag \end{gathered}\ ] ] in order to identify the light ( with respect to @xmath558 ) we can set @xmath559 and consider the supersymmetric limit . most fields are at the gut scale , with the light ones being the mssm ones , the goldstino superfield @xmath560 , and the right - handed neutrinos in the @xmath239 , which can easily be made heavy through a non - renormalizable superpotential operator @xmath561 . the mssm matter fields are embedded in the @xmath249 and in the @xmath562 , as desired . the higgs doublets are embedded into the @xmath563 and @xmath564 according to @xmath565 with normalization factors @xmath566 and @xmath567 , where @xmath568 , @xmath569 denote the sm component with the quantum numbers of @xmath296 , @xmath296 in the so(10 ) representation @xmath570 . after switching on @xmath386 the soft supersymmetry breaking terms and @xmath3-term are generated . the @xmath3-term is already present in the high energy lagrangian and is of order @xmath386 , the vev of @xmath552 generates sfermion and higgs masses of order @xmath571 and the vev of @xmath572 gives rise to a @xmath207 term of order @xmath571 . the heavy fields @xmath244 and @xmath243 act as messengers of susy breaking to the gauginos who get masses of order @xmath573 . the goldstino will be mainly the fermion in @xmath560 but gets also small contributions from the gaugino corresponding to the @xmath574 generator and the fermion in @xmath6 . the corresponding scalar will get a mass of order @xmath571 . h. p. nilles , phys . b * 115 * ( 1982 ) 193 ; a.h . chamseddine , r. arnowitt and p. nath , phys . * 49 * , 970 ( 1982 ) ; r. barbieri , s. ferrara and c. a. savoy , phys . b * 119 * , 343 ( 1982 ) ; l.e . ibez , phys . b * 118 * , 73 ( 1982 ) ; l.j . hall , j.d . lykken and s. weinberg , phys . d * 27 * , 2359 ( 1983 ) ; n. ohta , prog . phys . * 70 * , 542 ( 1983 ) . m. dine , w. fischler and m. srednicki , nucl . b * 189 * , 575 ( 1981 ) ; m. dine and w. fischler , phys . b * 110 * , 227 ( 1982 ) ; m. dine and a. e. nelson , phys . d * 48 * , 1277 ( 1993 ) [ arxiv : hep - ph/9303230 ] ; m. dine , a. e. nelson and y. shirman , phys . rev . d * 51 * , 1362 ( 1995 ) [ arxiv : hep - ph/9408384 ] . l. randall and r. sundrum , nucl . b * 557 * , 79 ( 1999 ) [ arxiv : hep - th/9810155 ] ; g. f. giudice , m. a. luty , h. murayama and r. rattazzi , jhep * 9812 * , 027 ( 1998 ) [ arxiv : hep - ph/9810442 ] . d. e. kaplan , g. d. kribs and m. schmaltz , phys . d * 62 * , 035010 ( 2000 ) [ arxiv : hep - ph/9911293 ] ; z. chacko , m. a. luty , a. e. nelson and e. ponton , jhep * 0001 * , 003 ( 2000 ) [ arxiv : hep - ph/9911323 ] . m. nardecchia , a. romanino and r. ziegler , jhep * 0911 * ( 2009 ) 112 [ arxiv:0909.3058 [ hep - ph ] ] . s. ferrara , l. girardello and f. palumbo , phys . d * 20 * ( 1979 ) 403 . s. dimopoulos and h. georgi , nucl . b * 193 * ( 1981 ) 150 . s. weinberg , _ the quantum theory of fields . 3 : supersymmetry _ , cambridge university press . j. wess and j. bagger , _ supersymmetry and supergravity _ , princeton university press . l. brizi , m. gomez - reino and c. a. scrucca , nucl . b * 820 * ( 2009 ) 193 [ arxiv:0904.0370 [ hep - th ] ] . i. affleck , m. dine and n. seiberg , nucl . b * 256 * ( 1985 ) 557 ; a. pomarol and s. dimopoulos , nucl . b * 453 * ( 1995 ) 83 [ arxiv : hep - ph/9505302 ] ; r. rattazzi , phys . b * 375 * , 181 ( 1996 ) [ arxiv : hep - ph/9507315 ] ; n. arkani - hamed , m. dine and s. p. martin , phys . b * 431 * , 329 ( 1998 ) [ arxiv : hep - ph/9803432 ] . k. a. intriligator and n. seiberg , nucl . * 45bc * ( 1996 ) 1 [ arxiv : hep - th/9509066 ] ; k. a. intriligator and n. seiberg , class . quant . grav . * 24 * ( 2007 ) s741 [ arxiv : hep - ph/0702069 ] . n. arkani - hamed , l. j. hall , h. murayama , d. tucker - smith and n. weiner , phys . d * 64 * ( 2001 ) 115011 [ arxiv : hep - ph/0006312 ] ; d. a. demir , l. l. everett and p. langacker , phys . lett . * 100 * , 091804 ( 2008 ) [ arxiv:0712.1341 [ hep - ph ] ] . n. arkani - hamed and s. dimopoulos , jhep * 0506 * ( 2005 ) 073 [ arxiv : hep - th/0405159 ] ; g. f. giudice and a. romanino , nucl . b * 699 * ( 2004 ) 65 [ erratum - ibid . b * 706 * ( 2005 ) 65 ] [ arxiv : hep - ph/0406088 ] . j. hisano , h. murayama and t. goto , phys . d * 49 * , 1446 ( 1994 ) . i. l. buchbinder , s. kuzenko and z. yarevskaya , nucl . b * 411 * ( 1994 ) 665 ; b. de wit , m. t. grisaru and m. rocek , phys . b * 374 * ( 1996 ) 297 [ arxiv : hep - th/9601115 ] ; a. pickering and p. c. west , phys . b * 383 * ( 1996 ) 54 [ arxiv : hep - th/9604147 ] ; m. t. grisaru , m. rocek and r. von unge , phys . b * 383 * ( 1996 ) 415 [ arxiv : hep - th/9605149 ] ; a. brignole , nucl . b * 579 * ( 2000 ) 101 [ arxiv : hep - th/0001121 ] . j. polchinski and l. susskind , phys . d * 26 * ( 1982 ) 3661 . s. dimopoulos and g. f. giudice , phys . b * 393 * ( 1997 ) 72 [ arxiv : hep - ph/9609344 ] . g. r. dvali , g. f. giudice and a. pomarol , nucl . b * 478 * ( 1996 ) 31 [ arxiv : hep - ph/9603238 ] . r. slansky , phys . rept . * 79 * ( 1981 ) 1 . m. frigerio , p. hosteins , s. lavignac and a. romanino , nucl . b * 806 * ( 2009 ) 84 [ arxiv:0804.0801 [ hep - ph ] ] . l. calibbi , m. frigerio , s. lavignac and a. romanino , arxiv:0910.0377 [ hep - ph ] . l. girardello and m. t. grisaru , nucl . b * 194 * ( 1982 ) 65 . n. arkani - hamed , s. dimopoulos , g. f. giudice and a. romanino , nucl . b * 709 * , 3 ( 2005 ) [ arxiv : hep - ph/0409232 ] . u. ellwanger , c. hugonie and a. m. teixeira , arxiv:0910.1785 [ hep - ph ] . p. langacker , n. polonsky and j. wang , phys . d * 60 * ( 1999 ) 115005 [ arxiv : hep - ph/9905252 ] . a. de gouvea , a. friedland and h. murayama , phys . d * 57 * ( 1998 ) 5676 [ arxiv : hep - ph/9711264 ] . g. f. giudice and a. masiero , phys . b * 206 * ( 1988 ) 480 . s. dimopoulos and f. wilczek , nsf - itp-82 - 07 ( unpublished ) ; m. srednicki , nucl . b * 202 * ( 1982 ) 327 .
maxwel introduced , in 1871 , a notorious being , known as maxwell s demon nowadays , to discuss the `` limitations of the second law of thermodynamics''@xcite . such a demon distinguishes the velocities of the gas particles and controls a tiny door on a partition of the gas container to create a temperature difference , which breaks the clausius statement of the second law of thermodynamics ( slot ) . to reveal the essence of the maxwell s demon , leo szilard proposed a single - particle heat engine @xcite , named as szilard heat engine ( she ) . the demon in she distinguishes the positions of the particles after the partition has been inserted . with the help of the demon , she can absorb heat from a single heat resource and convert it into work without apperant evoking other changes in the cycle . szilard pointed out that the slot was no longer violated if one considered the entropy increase during the measurement . brillouin generalized the szilard s argument and identified the thermodynamic entropy with the informational entropy firstly @xcite . however , the measurement could be carried out without any change in entropy @xcite . acutally , it was realized that a logically irreversible process must be `` accompanied by dissipative effects '' @xcite in the physical realization of the information processing , which is known as landauer s erasure principle . bennett used this point of view in the study of the maxwell s demon paradox and pointed out that the erasure of the demon s memory instead of the measurement was logically irreversible and thus must be accompanied by dissipative effects @xcite . with these observations , the conventional cycle presented by szilard is indeed not a thermodynamic cycle because the demon s memory has not been erased to complete the cycle . the slot will be saved if one considers the erasure process to finish the cycle of the demon . as people believe the essence of information should be discussed in the framework of quantum mechanics , various quantum versions of she have been proposed with different views about quantum measurements . one proposal is semi - classical @xcite . the working substance in this proposal is quantum mechanic while the demon is considered as a classical controller whose role is to extract information through measurement and control the system . the paradox of maxwell s demon was solved by arguing landauer s erasure principle . however , the final solution should include the md in the cycle and treat also the md in a quantum fasion@xcite . for she , it is also proved the existence of md will not violate the slot in ref . @xcite , where md is modeled as a two - level system . the focus of the study in both classical and quantum mechanical frameworks is the erasure process , which is crucial to solving the maxwell s demon paradox . in an odinary way , the demon and the working substance are in contact with the same heat bath . after erasing the demon by applying some work , one will find that she can not extract work in a cycle at all and the slot is not violated . however , a more general erasure should be done with a lower - temperature heat bath which is also called a heat sink . in this suituation , we turn the she into a thermal dynamic cycle , where the non - violation of slot can be proved by illustrating non - exceeding of carnot s efficiency . it is realized that the effective temperature of the md s initial state actually charasticrizes the error in the control of the heat engine@xcite . our previous work in ref . @xcite emphasized the functions of the demon with errors in the study of the single - particle she . but the erasure schemes in refs . @xcite are irreversible . therefore , the efficiencies of the heat engines in these papers can not reach the carnot cycle s efficiency . one purpose of the present paper is to establish an optimal scheme of the thermodynamic cycle with a reversible erasure process , assisted by demon . it is shown that the partition - removing process is not always reversible , which leads to the lower efficiency . we aso find the existence of the optimal expansion position to improve the efficiency of the single particle she to the carnot cycle s efficiency . the other purpose of this paper is to reveal the role of the quantum statistical properties of the working substance . we generalize our previous works about demon - assisted quantum heat engine by using a multi - particle working substance , which is the ideal bose or fermi gases , and find that the ratio of the work extracted to the working temperature has some discontinuous behavior , and that discontinuous behavior is closely related to the degenerate - ground - state phenomenon . the paper is organized as follows : in sec . [ sec2 ] we describe the model of quantum multi - particle she and the working scheme briefly . in sec . [ sec3 ] , we study in details the five steps of the working scheme : insertion , measurement , controlled expansion , removing and erasure separately and calculate the work applied and heat transferred in each step . in sec . [ sec4 ] , the efficiency of the engine is evaluated . it is found that our heat engine s efficiency can not exceed the carnot cycle s efficiency . for single - particle she , we optimize the scheme to make the efficiency of the engine reach the one of carnot s . in sec . [ sec5 ] , we discuss the behavior of the engine in low - temperature regime and show the ratio of the work extrated to the working temperature is closely related to the degenerate point ( whose definition can be seen in this section ) and the particles statistical properties . conclusions and remarks are given in sec . [ sec6 ] . ( color online ) modeling the chamber of gas by a one - dimensional infinite square well . ( a ) the chamber is modeled as a one - dimensional infinite square well with width @xmath0 where the particles are confined . the inner lines in the well represent the single - particle energy levels of the well . ( b ) after the insertion , the chamber is split into two chambers , which are modeled as two 1-d infinite square wells with widths @xmath1 and @xmath2 respectively . the inner lines represent the single - particle energy levels . , width=302 ] ( color online ) the states of the chamber and the demon ( a ) before and ( b ) after the expansion process . ( a ) after the measurement , the demon has revealed the number of the particles in the right compartment @xmath3 and recorded this information in its memory . if the demon is error - free , then its inner state ( or memory ) will be @xmath4 . however , when the demon contains some errors , there would be some mistakes in the demon s memory . for example , the sub - figure(2 ) shows a situation when @xmath5 and the demon is in state @xmath6 . thus , there are total @xmath7 situations after the measurement for a system with @xmath8 particles . here , the total @xmath9 situations are listed for @xmath10 . ( d ) after the measurement , the partition acts as a piston under the control of the demon . if the demon is in the state @xmath4 , the piston will move to a new position @xmath11 . such a movement is called controlled expansion . here the results of the expansion for all @xmath9 situations are listed for @xmath10.,width=302 ] the working substance is modeled as a collection of @xmath8 particles confined in a chamber , which is described as a one - dimensional infinitely deep square well with width @xmath0 illustrated in fig . [ chamber](a ) . generally , the particles can satisfy any distribution , bose distribution or fermi distribution . we calculate the canonical partition function @xmath12 as follows , where the subscript @xmath13 denotes the number of the particles in the well and @xmath1 represents the width of the well . its explicit expression is given in the appendix [ appendix1 ] . in the following calculations , we do not specify the concrete partition functions in most cases . the results obtained are valid for both bose and fermi distributions except for the case we claim specifically . in the model , maxwell s demon is described as a quantum system with @xmath14 energy levels to record the complete information about how many particles in each compartment after the insertion process . we may use a demon with number of energy levels less than @xmath14 , but it can only record partial information of the working substance , which decreases the work extract . we consider in the present paper only the demon with @xmath14 levels , with @xmath4 be its @xmath3-th eigen - state and @xmath15 be the energy of state @xmath4 . the energy of @xmath16 is set to be the zero point of energy , i.e. , @xmath17 . for a demon without errors , its initial state is a pure state , namely , @xmath18 . in this paper , we study the demon with errors generally , whose initial state can be written as @xmath19 when @xmath20 vanish for all @xmath21 , this demon returns to the one without errors , where@xmath16 is the standard state . for a demon with errors , a larger population @xmath22 in the state @xmath16 means a more efficient demon @xcite . the thermodynamic cycle consists the five steps . at the beginning , the working substance is in equilibrium with the heat resource at a higher temperature @xmath23 , and the demon is initially in the state @xmath24 ( see eq . ( [ ini_demon ] ) ) . in the first step , a partition is inserted at the position @xmath1 , which splits the initial well into two with widths being @xmath1 and @xmath2 respectively ( see fig . [ chamber](b ) ) . after the insertion , there are @xmath14 possible situations and one does not know which situation happens . suppose @xmath3 to be the number of the particles in the right compartment . then the @xmath14 situations correspond to @xmath25 . in the second step , the demon measures the working substance and records this information in its memory . for the error - free demon , it would be in the state @xmath4 for the @xmath3-th situation after the measurement . however , the demon with errors may record wrong information . for example , the working substance is in 0-th situation while the demon is in the state @xmath6 as illustrated in fig . [ expansion](a ) . in the third step , the partition acts as a piston and expands under the control of the demon . specifically , if the demon is in the state @xmath4 , it will move the piston to the position @xmath11 ( see fig . [ expansion](b ) ) . actually , one can optimize the expansion position @xmath11 to get the best efficiency . in the fourth step , the partition is removed and the working substance returns to its initial state . after these four steps , the working substance absorbs heat from the heat resource and converts it into work . it seems to violate the slot . however , the demon has not returned to its initial state to complete the cycle . thus , an additional step , step five , is needed to erase the memory of the demon . the effect of this step saves the slot as pointed out by landauer s erasure principle . in this section , we analyze in detail the thermodynamic cycle of the quantum multi - particle she . we calculate the heat absorbed and work applied in each step . in the first step , the partition is inserted isothermally , which means the working substance is in contact with a heat resource and the process is quasi - static . before the insertion , the state of working substance @xmath26 is @xmath27 where @xmath28 is the inverse temperature and the boltzmann constant is set to be unit , i.e. , @xmath29 , and @xmath30 represents the hamiltonian of @xmath8 particles confined in the infinite square well with width @xmath0 . @xmath31 is the corresponding partition function and its explicit expression is presented in appendix [ appendix1 ] . after the insertion , the chamber is split into two parts with widths @xmath1 and @xmath2 respectively and the substance is in an equilibrium state as @xmath32 here , @xmath33 describes an equilibrium state of the left chamber with similar meaning as those in @xmath34 . @xmath35 is the probability to find @xmath3 particles in the right chamber with insertion position being @xmath1 . as the process is isothermal , the work applied in this step is the difference of the free energies after and before the insertion , i.e. , @xmath36 and the heat absorbed is @xmath37 the free energy and the entropy are given in terms of the partition functions as @xmath38 and @xmath39 respectively . as shown in ref . @xcite , in quantum mechanics framework , the insertion work @xmath40 is no longer zero . moreover , it was proved in ref . @xcite that @xmath41 . in the second step , the total system is isolated from the heat bath . the demon finds out the number of the particles in the right compartment . this measurement process aims at establishing the correlation between the working substance and the demon . for the demon with no error , the state of the demon after the measurement is @xmath4 when there are @xmath3 particles in the right compartment . for the demon with errors , its initial state is a mixed state . thus , we should appoint the final states of the demon for all situations . let the final state be @xmath42 when the demon is initially in the state @xmath43 and the number of the particles in the right compartment is @xmath3 . mathematically , @xmath44 is a map from the set @xmath45 to itself . if the demon is error - free , the final state will be @xmath4 when there are @xmath3 particles in the right compartment , which leads the first constraint of @xmath46 , i.e. , @xmath47 . with this notation , the operator representing this measurement is @xmath48,\ ] ] where @xmath49 represents the @xmath11-th eigen - state of the working substance when there are @xmath3 particles in the right compartment . for a physical operation , @xmath50 should be unitary ( so that it is a quantum nondemolition pre - measurement ) , which leads to another constraint of @xmath51 . that is , for a fixed @xmath3 , when @xmath52 runs over from @xmath53 to @xmath8 , @xmath51 should also run over from @xmath53 to @xmath8 . thus , @xmath54 is a permutation of @xmath45 which satisfies @xmath47 . when @xmath55 , there is only one function of such kind and the corresponding operator @xmath50 represents a controlled - not operation . for @xmath56 , such function exists . one realization is @xmath57 . after the measurement , the state of the total system becomes @xmath58 , namely @xmath59 where @xmath60 is the inverse of the map @xmath46 . it follows from eq . ( [ state_aft_mea ] ) that there is a possibility that the demon is in the state @xmath4 while the number of particles in the right compartment is not @xmath3 . that is exactly what we mean ` error ' within the demon . due to the isolation from the heat bath , there is no heat transferred , @xmath61 , and the work applied is just the energy change during the measurement : @xmath62\delta_{j},\ ] ] which compensates for the energy difference between the initial state and the final state of the demon during the measurement . therefore , if all the energy levels of the demon are degenerate , one will not apply any work to perform the measurement . thus the work applied may not always be non - zero @xcite . in this step , the partition in the chamber acts as a piston and will be moved to the position according to the memory of the demon . specifically , if the state of the demon is @xmath4 , the finial position will be @xmath11 , which is called expansion position . this expansion process is slow enough and the working substance is in contact with the heat resource during the total expansion process . thus the state of the total system after the controlled expansion is @xmath63 however , the ` impenetrability ' of the piston does not make this process an isothermal process since the transition between two states with different numbers of the particles in the right compartment is forbidden . thus we should deal with this expansion process for each situation separately . for the situation when the demon is in state @xmath43 and there are @xmath3 particles in the right compartment , the work applied @xmath64 is @xmath65 and the heat absorbed @xmath66 is @xmath67 the probability of this situation is @xmath68 . thus , the total work applied and the total heat absorbed are : @xmath69 respectively . one can prove that even if the controlled expansion is not isothermal , the heat absorbed also satisfies the relationship @xmath70 , where @xmath71 is the entropy change of the total system . since the states with different numbers of the particles in the right compartment are orthogonal to each other , the entropies of the total system before and after the controlled expansion are @xmath72 respectively , where @xmath73 . thus , we have the following relationship @xmath74 ( color online ) the removing process . ( a ) before the removing , the state of the working substance may not be an equilibrium state because of the ` impenetrability ' of the partition ( see the sub - section [ expansion ] ) . ( b ) when the partition is lifted a little , it is no longer impenetrable and the particles can fly between the two sides of the chamber freely to reach the equilibrium state . ( c ) the partition keeps being lifted and finally be removed isothermally . , width=302 ] after the controlled expansion , the partition will be removed slowly while the working substance is in contact with the heat resource . because the state in eq . ( [ state_aft_exp ] ) before removing is not a thermal equilibrium state ( it is because the controlled expansion is not an isothermal process , see the last subsection ) , one should carefully deal with this step . once lifted a little , the partition is no longer ` impenetrable ' and the particles can fly between the two sides of the chamber freely to reach an equilibrium state . thus the removing process consists of two sub - steps . after the first sub - step finished , the partition has been lifted to create a small slit with width @xmath75 through which the particles can fly from one side to the other . whatever small the width @xmath75 of the slit is , the working substance can reach its equilibrium state . thus , we can let the width of the slit tend to zero , i.e. , @xmath76 , which makes this sub - step a thermalization process of the working substance while the partition remains still . thereafter the state of the system after the first sub - step becomes @xmath77 where @xmath78 . during this process , there is no work applied ( since @xmath76 ) and the heat absorbed is exactly the difference between the inner energies of the total system before and after the first sub - step , @xmath79\left[u(\rho_{n - i}^{l}(l_{j}))+u(\rho_{i}^{r}(l - l_{j}))\right].\ ] ] after the thermalization , the working substance is in a thermal equilibrium state , which makes the rest sub - step , the second sub - step , an isothermal process . after the removing , the working substance returns to its initial state and the correlation between the working substance and the demon no longer exists . during the second sub - step , the work applied and the heat absorbed are @xmath80 , \\ q_{\mathrm{rev } } & = & t_{1}\left[s(\rho_{s0})-\sum_{j=0}^{n}p_{j}^{(1)}s(\rho_{s , ins}(l_{j}))\right],\end{aligned}\ ] ] respectively , and the state of the total system after the removing process is @xmath81 one can see that the state of the demon does not return to its initial state , which does not make the above four steps a thermodynamic cycle . thus , an addition step , erasure , is needed . ( color online ) erasing the memory of a three - energy - level demon reversibly . suppose that the populations of the three levels after the removing process the partition are @xmath82 , @xmath83 and @xmath84 as in ( a ) and the initial state s populations of the three levels are @xmath85 , @xmath86 and @xmath87 as in ( d ) . in step 1 ( a - b ) , the demon is isolated from the heat bath and the energy level spacings are adjusted from @xmath15 to @xmath88 for @xmath89 . during this process , the populations of the levels remain constant . the new level spacings are chosen so that the demon has an effective temperature @xmath90 after the adjusting , which implies @xmath91 and @xmath92 . in ( b ) , the minus sign in front of @xmath93 is due to the negative value of @xmath93 . in step 2 ( b - c ) , the demon is in contact with a real heat bath at temperature @xmath94 and the energies of the levels are adjusted . in this step , the populations of the levels change while its energy spacings change . the goal of this step is to let the populations of the levels return to be @xmath22 , @xmath95 and @xmath96 . thus the new level spacings are @xmath97 and @xmath98 . in step 3 ( c - d ) , the demon is isolated from the heat bath again and its level spacings are adjusted to return to its initial values . the isolation guarantees that the populations of the levels are not changed . thus the state of the demon has returned to its initial state . , width=302 ] in eq . ( [ state_aft_rev ] ) , the state of the demon after the removing process is @xmath99 which is different from the initial state of the demon . as part of the engine , the demon should be re - initialized to complete the cycle . because the entropy of @xmath100 is different from the entropy of @xmath24 , such a re - initialization process , or erasure process , can not be realized by a unitary evolution and must be a logical irreversibility process accompanied by dissipative effects @xcite . a natural way to complete the re - initialization process is to make use of a thermalization process . it can be shown that the heat bath used in the thermalization process is a heat sink instead of a heat resource . because the entropy of @xmath100 is larger than the entropy of @xmath24 , the heat must be transferred from the demon to the heat bath . it seems that after a series of complex tasks , the demon needs to be cooled down . according to thermodynamics , the best re - initialization process is a reversible one which does not cost additional dissipative work . before going into the general discussion , we first consider a three - level demon as an example to find out how to re - initialize or erase the demon reversibly . the three - level demon considered is shown in fig . [ fig_era ] , where the three states are @xmath16 , @xmath6 and @xmath101 respectively . the energy of the state @xmath16 is set to be zero and the energy of the state @xmath4 is @xmath15 for @xmath89 . for a demon in its initial state , the populations of the three levels are @xmath22 , @xmath95 and @xmath96 respectively ( see fig . [ fig_era](d ) ) . to be more concrete , we assume that @xmath85 while @xmath86 and @xmath87 . suppose that the populations become @xmath82 , @xmath83 and @xmath84 respectively after the removing ( see fig . [ fig_era](a ) ) . in order to re - initialize it by using a reversible scheme by means of a heat bath with @xmath90 , we adjust the demon s energy level spacings adiabatically ( see fig . [ fig_era ] ( a)-(b ) ) . after being adjusted , the level spacings are changed from @xmath15 ( @xmath89 ) to @xmath88 ( see fig . [ fig_era](b ) ) . during this process , the population of each state remains unchanged . thus , the new level spacings satisfy the following relationship to ensure that the effective temperature of the demon is @xmath94 , i.e. , @xmath102 for @xmath89 , which implies @xmath91 and @xmath92 . next , the demon is kept in contact with the heat bath at temperature @xmath94 without any undesired irreversible heat transfer , and we adjust the energies of its levels isothermally ( see fig . [ fig_era ] ( b)-(c ) ) . in this process , the population of each level changes as the level spacings change and the final populations are adjusted to be @xmath22 , @xmath95 and @xmath96 respectively , which are the initial populations of the demon . therefore the new level spacings @xmath103 ( @xmath89 ) satisfy the relationship @xmath104 which implies @xmath97 and @xmath98 ( see fig . [ fig_era](c ) ) . after that , the population of each state has returned to its initial value while the level spacings do not . thus we need another adiabatical adjusting process as in the first step to re - initialize the level spacings ( see fig . [ fig_era ] ( c)-(d ) ) . these three sub - steps complete the re - initialization process and are all reversible , which makes the total re - initialization process reversible . the above example inspires us to construct the reversible re - initialization scheme in a more general case . in the following discussion , a detailed scheme of the re - initialization process is presented . in the first sub - step , we adjust the energy level of the state @xmath43 for @xmath105 adiabatically while the energy of @xmath16 is kept fixed . the corresponding new state is denoted as @xmath106 . during this sub - step , the demon is isolated from the heat sink . the target is to let the demon s effective temperature be @xmath94 , which is the temperature of the heat sink . thus , the new level spacing @xmath107 satisfies @xmath108 and the demon after the first sub - step is in the state @xmath109 it should be noticed that eq . ( [ new_energy ] ) always has a solution except for some extreme conditions , for example , @xmath110 or @xmath111 . if these extreme conditions happen , we will use some tricks to deal with it . in the case of @xmath111 , we will introduce a small positive quantity @xmath112 and let @xmath113 be @xmath114 . next we will use @xmath113 instead of @xmath94 to calculate the parameters of the scheme such as @xmath88 and @xmath103 . such a scheme is not reversible . however , if @xmath112 tends to zero , the scheme will tend to behave as a reversible scheme . in the following , we assume that eq . ( [ new_energy ] ) does have solution . in this sub - step , there is no heat transfer and the work applied is @xmath115 in the second sub - step , demon is in contact with the heat sink and the @xmath52-th level s energy is adjusted from @xmath107 to @xmath116 isothermally , where @xmath116 satisfies @xmath117 and @xmath118 is the inverse temperature of the heat sink . thus , the state of the demon after this sub - step is @xmath119 where @xmath120 represents the state corresponding to @xmath106 after the adjusting . the work applied and the heat transferred during this process are @xmath121,\\ q_{\mathrm{ini}}=t_{2}\left[h\left(\{p_{i}\}\right)-h\left(\left\ { \sum_{i=0}^{n}p_{i}(l)p_{f_{i}^{-1}(j)}\right\ } \right)\right],\end{aligned}\ ] ] respectively . here @xmath122 represents the shannon entropy of the probability distribution @xmath123 . in the last sub - step , the demon is isolated from the heat sink again and the energies of all of its levels are adjusted to its initial values . thus the final state of the demon is @xmath124 , which is indeed its initial state . during this sub - step , no heat transfers and the work applied is @xmath125 these three sub - steps together make up the re - initialization scheme . such a scheme involves adjusting the energies of the demon s levels , which is not easily realized . in a concrete experiment , if the quantum system that functions as the maxwell s demon has unchangeable energy levels , we will only use an irreversible scheme as in refs . as we know in thermodynamics , such an irreversible re - initialization scheme reduces the efficiency of the heat engine because of the dissipative effects . thus , when it comes to the maxwell s demon paradox , one needs a reversible re - initialization scheme without dissipative effects to get a good physical picture . combining the results obtained in the above section , one finally gets the total heat transferred and the work extracted in the thermodynamic cycle : \1 . the heat absorbed from the heat resource is @xmath126.\nonumber \end{aligned}\ ] ] \2 . the heat released to the heat sink is @xmath127.\nonumber \end{aligned}\ ] ] \3 . the work extracted is @xmath128 which can be checked to satisfy the first law of thermodynamics , i.e. , @xmath129 . firstly , it is emphasized that the result is general without referring to the statistical properties of the working substance , since @xmath130 has contained all the information of the distribution . thus eq . ( [ q1 ] ) , eq . ( [ q2 ] ) and eq . ( [ w ] ) are correct both for the bose system and the fermi system . secondly , one can optimize the expansion position @xmath131 to maximize the efficiency of she . consider a two - particle she as an example , in which the demon is of three - level . the function @xmath51 we choose is @xmath132 . thus the heat absorbed from the heat resource is @xmath133 thereafter , the optimized expansion position @xmath11 should satisfy @xmath134 working out the above equation , one will get the optimized expansion positions @xmath135 , @xmath136 and @xmath137 . thirdly , an upper limit of the heat absorbed from the heat resource @xmath138 in eq . ( [ q1 ] ) is estimated . use the method presented in the appendix [ appendix2 ] , one gets @xmath139\nonumber \\ & = & t_{1}\left[h\left(\{\sum_{i=0}^{n}p_{i}(l)p_{f_{i}^{-1}(j)}\}\right)-h\left(\{p_{i}\}\right)\right].\end{aligned}\ ] ] therefore the efficiency of the she has the upper limit : @xmath140 thus , it is clear that the efficiency of the she can not exceed the carnot cycle s efficiency when the erasure process or re - initialization process is considered , which saves the slot . it can be seen more clearly when @xmath55 , which represents a single - particle she . for a single - particle she , there is only one realization of @xmath51 , and the corresponding measurement process is indeed a controlled - not operation as in ref . @xcite , i.e. , @xmath141 for @xmath142 while @xmath143 and @xmath144 . the formulas eq . ( [ q1 ] ) and eq . ( [ q2 ] ) are reduced to @xmath145,\eqno(\ref{q1}')\ ] ] and @xmath146\eqno(\ref{q2}')\ ] ] respectively . it is easy to maximize @xmath138 to find that @xmath135 and @xmath136 satisfy @xmath147 and the maximum of @xmath138 is @xmath148.\label{q1max}\ ] ] thus , it is found that @xmath149 is always positive whatever the insertion position is . combining eq . ( [ q1max ] ) and eq . ( [ q2 ] ) , one gets the best efficiency as @xmath150 then it is concluded that for a single - particle she , one can always choose a good expansion position to let the she reach the carnot cycle s efficiency even if the demon contains some errors . recall that in the step 4 of the thermodynamic cycle , the removing process , one needs to thermalization the working substance firstly . however , for a single - particle she , it is found that the expansion position can always be chosen that the thermalization process is not needed . thus , a single - particle she can always be a reversible heat engine with two heat baths , and its efficiency reaches the carnot cycle s efficiency . on the other hand , for a multi - particle she , the thermalization process is always needed in the removing process , which makes the heat engine irreversible . thus , a multi - particle she s efficiency can not always reach carnot cycle s efficiency . ( color online ) work extracted @xmath151 as the function of insertion position @xmath152 and the working temperature @xmath23 . ( 1 ) bose case with @xmath153 . ( 2 ) fermi case with @xmath153.,scaledwidth=100.0% ] ( color online ) work extracted @xmath151 as the function of insertion position @xmath152 and the working temperature @xmath23 . ( 1 ) bose case with @xmath153 . ( 2 ) fermi case with @xmath153.,scaledwidth=100.0% ] + \(1 ) bose case \(2 ) fermi case + at the end of this section , a concrete example is presented to show the difference between the bose system and the fermi system . in this example , a four - particle system is considered as the working substance . the demon we chosen is a five - level quantum system with some errors . its initial state is @xmath154 the temperature of the heat sink is fixed that @xmath90 . denote the higher temperature and the insertion position as @xmath23 and @xmath1 respectively . then the work extracted by the heat engine during a thermodynamic cycle can be numerically calculated for both the bose case and the fermi case . the results are shown in fig . [ bvf ] . as one can see , the patterns in fig . [ bvf ] for different distributions are quite different . in fact , in the next section , it is found that the difference between the bose case and the fermi case is most significant in low - temperature regime . in this section , the behavior of a multi - particle she when the working temperature @xmath23 is very low is studied . in sec . [ sec4 ] , a necessary condition for the heat engine to extract positive work is found that @xmath155 . thus , if @xmath23 considered is very small , the temperature of the heat sink will be @xmath111 . the demon in this case is free of errors because it is natural to get a pure initial state if the demon is re - initialized by a zero - temperature heat sink . the motivation to concern the demon without error is to see the difference between the bose system and fermi system more clearly without the effect of the demon s errors . as mentioned in the step 5 of sec . [ sec3 ] , it is less likely for one to re - initialize such a demon by using a reversible scheme ; however , one can use an asymptotic scheme to re - initialize the demon . ( color online ) asymptotic erasure scheme for @xmath111 . , width=302 ] now , we describe the asymptotic scheme in details firstly . similar to the re - initialization scheme in sec . [ sec3 ] , the asymptotic scheme also contains three sub - steps . the three sub - steps are illustrated in fig . [ fig_era1 ] . in the first sub - step , the energies of the levels @xmath4 for @xmath156 are adjusted from @xmath15 to a very small value @xmath157 adiabatically . during this sub - step , there is no heat transferred . in the second sub - step , the demon is in contact with the heat sink at temperature @xmath111 until it is in equilibrium with the heat sink . during this sub - step , there is no work applied and the finial state of the demon is a pure state for its temperature is zero when @xmath158 . in the third sub - step , the energies of the demon s levels are re - adjusted adiabatically to its initial value . thus , these three sub - steps complete the re - initialization process of the demon . if we use this scheme , @xmath138 in eq . ( [ q1 ] ) will not change and @xmath159 in eq . ( [ q2 ] ) will be revised to be the heat transferred in the second sub - step . due to the irreversibility of the second sub - step , this scheme is not reversible . nevertheless , if one lets @xmath160 tend to @xmath53 , @xmath159 will tend to @xmath53 , which makes the scheme reversible . thus , eq . ( [ q1 ] ) and eq . ( [ q2 ] ) will still hold if one uses this asymptotic scheme . in the following , this asymptotic scheme is used in which the relationships , @xmath161 and @xmath162 , hold . since the initial state of the demon is a pure state , which corresponds to the conditions @xmath111 and @xmath163 , the work extracted can be given by the relationship @xmath162 and ( [ q1 ] ) as @xmath164\nonumber \\ & = & t_{1}\sum_{i=0}^{n}p_{i}(l)\ln\left(\frac{p_{i}(l_{i})}{p_{i}(l)}\right).\label{w'}\end{aligned}\ ] ] this is just the same as that in ref . @xcite and it can be calculated by a different way as in ref . then we will use eq . ( [ w ] ) to study the bose system and the fermi system separately . first of all , we study the behavior of @xmath165 when @xmath166 . according to eq . ( [ w ] ) , one gets @xmath167 here , the expansion position @xmath11 have been optimized as in sec . thus @xmath168 is always no less than @xmath130 . for a deterministic distribution , @xmath169 , @xmath165 in eq . ( [ a3_w / t ] ) becomes to be @xmath170 . due to the optimization of @xmath131 , we have @xmath171 , or , @xmath172 . thus , it is concluded that @xmath165 is always @xmath53 when @xmath173 is a deterministic distribution . in other words , the necessary condition for a non - zero @xmath165 is that the informational entropy @xmath174 of the distribution @xmath173 is non - zero . using the joint entropy theorem @xcite , one gets @xmath175 where @xmath176 is the entropy of the working substance after the insertion and @xmath177 represents the working substance s entropy in the condition that the number of the particles in the right compartment is @xmath3 . due to the non - negative property of entropy , one finds @xmath178 . therefore if the entropy of the working substance is zero , one will find that @xmath179 and @xmath180 . for a system with its ground state non - degenerate , its entropy always tends to zero when temperature tends to zero , which is referred to as the third law of thermodynamics . thus , it is found that , if the insertion position @xmath1 makes the ground state of the working substance non - degenerate , one will always have @xmath181 however , if the insertion position @xmath1 makes the ground state of the working substance degenerate , there will be a chance that @xmath165 tends to non - zero . such a position @xmath1 is called a degenerate point . thus , the relationship between the low - temperature behavior of @xmath165 and the degenerate point has been established . the similar result is found in ref . @xcite . then , we look for the degenerate points for both the ideal bose system and the ideal fermi system . for the ideal bose system , all particles condense in the ground state of the system when temperature is zero . thus the only chance for the ground state to be degenerate is that the energy of the left compartment s single - particle ground state is the same as that of the right compartment , which happens when the widths of the two compartments are equal to each other . thereafter it is found that there is only one degenerate point for the ideal bose system , i.e. , @xmath182 . ( color online ) degenerate ground states for a fermi gas . when the fermi surface contains more than one single - particle states , there is a chance that the ground state of the fermi gas is degenerate . for example , the states in ( 1 ) and ( 2 ) are two ground states of the system . , width=302 ] for a fermi system , the particles satisfy the pauli s exclusion principle . when temperature is zero , all the energy levels under the fermi surface are occupied . thus if the @xmath3-th single - particle energy level of the left compartment and the @xmath52-th single - particle energy level of the left compartment are both on the fermi surface , the ground state of the total gas may be degenerate , see fig . [ fermi_degen ] . the index @xmath3 and @xmath52 should satisfy the relationship that @xmath183 equals to the number of the particles @xmath8 . thus there are @xmath8 situations corresponding to @xmath184 . thereafter it is found that there are @xmath8 degenerate points for an ideal fermi system with @xmath8 particles . as the two compartments are modeled as two one - dimensional infinite square wells with widths @xmath1 and @xmath2 respectively , one can workout the spectrum of them as in eq . ( [ spectrum ] ) . thus the conditions of the degenerate points can be written as @xmath185 therefore the positions of the fermi system s degenerate points are @xmath186 so far we have revealed the qualitative difference of the work extracted by the engine between the bose type and the fermi type working substance at low temperature . that difference in the quantum statistical behavior between the bose and the fermi working substance will disappear as the temperature @xmath23 increase . the numerical simulation in fig . [ fig2 ] quantitively illustrates the statistical behavior at finite temperature . when @xmath23 tends to zero , the numerical calculations confirm the above results . similar result is also found in ref . in summary , the multi - particle she has been studied in a complete quantum framework , where the re - initialization process of the demon is discussed in detail . the errors are introduced in the demon so that one can use a finite - temperature heat sink to cool down the demon while the traditional demon must be cooled down by a zero - temperature heat sink . for a single - particle she , there is a scheme that the whole cycle of the she can be revisible . we illustrated that the efficiency of such a scheme is indeed the efficiency of carnot cycle s and other schemes have lower efficiency . and it is proved that she can not exceed the limit of the slot . on the other hand , the properties of the she with bose or fermi particles are studied . in the high - temperature regime , these two distributions make no difference . both of them reduce to the result of maxwell distribution . in contrast , the bose system and the fermi system differ from each other in the low - temperature regime . the origin of the difference is the difference of the degenerate points of these two distributions . the she whose insertion position is a degenerate point extracts more work in a cycle . if the temperature is absolute zero , the ratio of the work extracted to the working temperature will show some uncontinuous properties . this work was supported by nsfc through grants 10974209 and 10935010 and by the national 973 program ( grant no . 2006cb921205 ) . in this appendix , the explicit expressions of the multi - particle partition functions for the ideal bose gas and the ideal fermi gas are presented . let @xmath187 be the canonical partition function for @xmath13 particles , where @xmath188 represents the inverse temperature of the system and @xmath1 represents some parameters of the system , e.g. , in our paper @xmath1 means the width of the infinite square well . to present the explicit expression of @xmath187 , the explicit expression of the single - particle partition function @xmath189 is given firstly . then the relationship between the multi - particle partition function and the single - particle partition function is shown . the single - particle partition function does not depend on which distribution the particle satisfies . once one knows the single - particle spectrum of the system , one can write down its canonical partition function . for example , the single - particle spectrum in a one - dimensional infinite square well with width @xmath1 is @xmath190 where @xmath191 is the mass of the particle and @xmath3 is the index of the energy levels . then the single - particle partition function of this system is @xmath192 when it comes to multi - particle partition function , we first take the two - particle system as an example . in this condition , one gets @xmath193,\ ] ] where the indices @xmath188 and @xmath1 have been omitted for simplicity . here , the summation , @xmath194 , can be viewed as a single - particle partition function with spectrum @xmath195 . thus , it is found that @xmath196 can be decomposed into some single - particle partition functions . in general , all the partition functions of the ideal gas , @xmath197 , can be decomposed into some single - particle partition functions by this ways . the next task is to workout the coefficients in the decomposition . in the following discussion , the system considered has discrete spectrum and the index of the energy levels is denoted by @xmath198 . let @xmath199 be the number of the particles in the @xmath3-th level . to be convenient , we denote @xmath200 as @xmath201 for @xmath202 and the multi - particle partition function becomes to be @xmath203 here , the summation condition is that @xmath204 and @xmath205 for bose case or @xmath206 for fermi case . it is not difficult to show that @xmath207 in fact , if @xmath208 , one will find that @xmath209 and @xmath210 are nothing but the grand canonical partition functions for ideal bose case and ideal fermi case respectively . in mathematics , @xmath209 and @xmath210 are the generating functions for @xmath211 and @xmath212 respectively and they have the following simple relationship @xmath213 next , the single - particle partition function with spectrum @xmath214 , @xmath215 , is denoted as @xmath216 for @xmath217 . its generating function is @xmath218 thus , it is found that @xmath219 i.e. , @xmath220 by indefinite integration , one finds that @xmath221 where @xmath222 is the constant of integration and it is not difficult to find that @xmath223 . so , there is the following important relationship @xmath224 compare the coefficient of @xmath225 in both the two sides of eq.([h(t ) ] ) and one can find that @xmath226 where @xmath227 and the summation condition is @xmath228 which corresponds a summation for all the young diagrams with @xmath13 boxes . the @xmath229 is defined as @xmath230 and the @xmath231 is the product of some single - particle partition functions as @xmath232 thus we have decomposed the multi - particle partition function into some single - particle partition functions for bose case . for fermi case , using the relationship in eq.([h_e ] ) , one finds that @xmath233 also , compare the coefficient of @xmath225 in both the two sides of eq.([e(t ) ] ) and it is found that @xmath234 where the summation condition is the same as in the bose case . in this appendix , the upper limit of @xmath138 in eq . ( [ q1 ] ) is estimated . to this end , a two - particle she is used as an example firstly , in which the demon considered is of three - level . the function @xmath51 is chosen as @xmath132 . thus the heat absorbed from the heat resource is the one shown in eq . ( [ example_eff ] ) . to estimate the upper limit of @xmath138 , eq . ( [ example_eff ] ) should be re - written as where @xmath236 satisfy the constraints @xmath237 for all @xmath3 and @xmath52 . thus the original optimization problem becomes to be an optimization problem with constraints . this optimization problem with constraints has a maximum solution @xmath149 since @xmath138 in eq . ( [ b1 ] ) has an upper bound @xmath238 and the domain of @xmath239 is closed . substitute these constraints by the weaker constraints @xmath240 for all @xmath52 , then we get another optimization problem with constraints . after solving this problem we will get another maximum @xmath241 . because the constraints in the latter problem are weaker than the constraints in the former one , we have @xmath242 . thus @xmath241 is a upper limit of @xmath138 . the calculation of @xmath241 can be performed in terms of lagrange multipliers . namely , one only needs to solve an optimization problem without constraint in which the target function is @xmath243 where @xmath244 , @xmath245 and @xmath246 are three lagrange multipliers . solving this constraint - free optimization problem , one finds that the maximum solution @xmath241 is reached when @xmath236 satisfies the following relationships @xmath247 for a general situation , this trick can also be used to find an upper limit of the heat absorbed from the heat resource @xmath138 . to this end , one only needs to maximize @xmath138 under the constraints @xmath248 for all @xmath52 . thus the upper limit of @xmath138 is given as
in this paper , we propose a linear threshold agent - based model ( abm ) for the spread of political revolutions on social networks in dictatorial regimes . in our abm , nodes of the network can be in two states : active , i.e. , participating in the revolution , or inactive . transitions from the inactive to the active state are governed by a growth process that uses a traditional linear threshold mechanism : an individual @xmath1 may change from the inactive to the active state if the fraction of neighbours in the active state exceeds the linear threshold @xmath2 . linear threshold models have been studied before in the context of influence maximization @xcite , enforcement of unpopular norms @xcite , and general `` complex contagion '' processes @xcite . the development and analysis of linear threshold processes has been highly influenced by more established epidemiology @xcite and rumour spreading @xcite models , with one key difference : whereas a contact with a single `` infected '' individual is enough to spread a contagion in most epidemiology or rumour spreading applications , i.e. in `` simple contagion '' processes , multiple contacts with infected individuals are generally necessary to spread a contagion in a linear threshold process . indeed , epidemiological and rumour spreading models can be considered to be a special case of linear threshold models where the threshold parameter is chosen sufficiently small such that one infected neighbour enables propagation . most of the linear threshold models that have been studied in the context of social spreading or contagion processes @xcite consider evolution in discrete time . in contrast , our abm is chosen to evolve in continuous time since that facilitates the comparison of the abm with simplified population - level models , which is a major goal of this paper . the main justification for our linear threshold modelling approach in the context of political revolutions is that the decision to join a political revolution can be assumed to be a collective action problem @xcite : if individuals act unilaterally they are subject to retaliation by the regime , whereas if they act collectively then the regime loses the ability to punish due to a lack of resources . the linear threshold modelling approach is consistent with the collective action principle , since an individual transitions from an inactive to an active state only after the number of neighbours in the active state has been observed to reach the critical fraction @xmath2 , and the individual deems the revolution of sufficient size to consider participation . in order to incorporate communication network structure explicitly , we represent interactions between individuals @xmath1 by edges @xmath3 in an undirected graph @xmath4 . this contrasts with previous attempts to model political revolutions using linear threshold models that assume homogeneous mixing @xcite , i.e. that assume that @xmath4 is the complete graph . the goals and contributions of this paper are as follows . we start by presenting the linear threshold abm for the spread of political revolutions on social networks , and then derive a hierarchy of simplified ordinary differential - equation ( ode ) models of varying degree of sophistication that characterize the solutions of the linear threshold abm , see table [ tab : hierarchy ] . although the linear threshold abm allows us to explicitly model the effects of the communication network on the dynamics of the political revolution it models , the inherent complexity of modelling the state of all individuals simultaneously makes this approach difficult to analyze and expensive to simulate . for these reasons there is significant interest in simplified aggregate or population - level dynamical models , which can provide more cost - effective dynamical simulations , and can give insight in the qualitative dynamics of the abm and its parameter regimes , through dynamical analysis of the simplified model . it is a significant challenge to incorporate actual network structure in the ode models , and this is often important since details of the network structure may determine the qualitative behaviour of abm solutions as the abm parameters are changed . in this paper , we present two new effective ways to incorporate network structure into the one - compartmental ode that approximates the dynamical evolution of the expected fraction of the population that participates in the revolution in the linear threshold abm model . these approaches make use of the degree distribution of the graph or samplings of the network , and we call the resulting population - level ode models the _ binomial visibility function _ ( bvf ) and the _ empirical visibility function _ ( evf ) models . the evf model is significantly less expensive than the bvf , but we demonstrate its equivalence with the bvf in the limit of large network and sample size . extensive numerical tests show that these ode models perform remarkably well to predict qualitative behaviour of the abm ; in fact , they often perform as well as or better than a previously specified higher - order model that has many compartments and is much more expensive computationally @xcite , and that we refer to as the _ degree approximation _ ( da ) model . .summary of hierarchy of models from highest ( network - level ) to lowest ( population - level ) complexity . [ cols="<,<,<,<,^",options="header " , ] [ tab : sum_stats ] it is interesting to consider how the physical contact and facebook networks differ on some common network measures , see table [ tab : sum_stats ] . comparing the physical contact and facebook social networks along these admittedly limited measures nevertheless highlights substantial structural differences between these networks . the physical contact network appears more homogeneous than the facebook social network , in the sense that the facebook social network can be visually grouped into distinct communities , whereas the physical contact network can not ( fig . [ fig : vis ] ) . the network measures presented in figs . [ fig : deg_distn]-[fig : clust_coeff ] and table [ tab : sum_stats ] are generally supportive of this observation . for example , the cumulative degree distributions displayed in fig . [ fig : deg_distn ] show that the physical contact network has an approximately normal distribution with a relatively small standard deviation , and hence , a thin tail . in contrast , the facebook social network has a much broader distribution ( possibly scale - free or exponential , although this can not be determined conclusively with such a small network size ) with a relatively large standard deviation , and hence , a fat tail . similarly , it is known that online social networks often have larger clustering coefficients than offline social networks . in section [ sec : interp ] we use the facebook and physical contact networks as representatives of online and offline social networks , respectively , to investigate differences in propagation properties that may arise between online and offline social networks within the linear threshold propagation model . consider a population of individuals who are represented by nodes @xmath5 , and whose interactions are represented by edges @xmath6 , of the graph @xmath4 with degree distribution @xmath7 . an individual @xmath8 at time @xmath9 can be in one of two states , i.e. @xmath10 ( _ inactive _ in the revolution ) or @xmath11 ( _ active _ in the revolution ) . we assume that the network @xmath12 is static so that the dynamics of the abm can be fully specified by providing rules for the transition of individuals from an inactive state to an active state and vice versa , together with parameter values and an initial condition . for the moment , we set aside the issue of the choice of parameters and initial conditions and restrict the remainder of this section to the specification and justification of the transition rules . to specify the growth process of the abm , we use the standard linear threshold model that has been used before in the context of social spreading or contagion processes such as opinion formation , technology adoption , marketing , rioting , and political movements @xcite . as mentioned in section [ sec : intro ] , the choice of individuals to join a revolution is a collective action problem @xcite : individuals are averse to unilateral action against the regime for fear of severe retaliation , but are willing to take action collectively in the belief that the regime will lack sufficient resources to punish the entire collective . thus , an individual @xmath8 will decide to join the revolution if he or she believes that it has grown sufficiently large to reduce the risk of retaliation from the regime to an acceptable level . it is reasonable , therefore , to assume that individuals will decide to join the revolution only after a large enough fraction of their neighbours in the social network have done so . this behaviour is captured by the following transition rule : if individual @xmath8 has @xmath15 neighbours @xmath16 , if @xmath10 , and if @xmath17 i.e. , if @xmath1 is inactive at time @xmath9 and has at least a fraction @xmath18 of its neighbours that are active , then node @xmath1 transitions from state 0 to state 1 at time @xmath19 , where @xmath20 is the first arrival time of a poisson process with rate @xmath21 . we say that nodes that satisfy equation can `` see '' the revolution , i.e. the revolution is visible to them . alternatively , we say that these nodes are `` considering joining the revolution '' . parameter @xmath21 determines the timescale of the growth process . while most of the linear threshold models that have been studied in the context of social spreading or contagion processes @xcite consider evolution in discrete time by choosing @xmath22 constant , we have specified @xmath22 so that equation evolves in continuous time . specifying @xmath22 in this way is often considered in biology and epidemiology @xcite , since it ( a ) facilitates the comparison of the abm with population - level models ( as the svf model described in section [ sec : consist_abm_svf ] ) , and ( b ) eliminates the problem of choosing a suitable time - step for iterating the discrete - time process . furthermore , we note that by choosing @xmath22 to be the first arrival time of a poisson process , we are assuming that the decision making process is a markov , or memoryless , process . specifically , conditioned on the state of the system at time @xmath9 , we assume that the likelihood of an individual joining the revolution in the time interval @xmath23 $ ] is independent of the state of the system at any time @xmath24 . in words , the future is independent of the past , given the present . we assume that once an individual @xmath8 has become active the regime will attempt to arrest or disperse him or her , thus returning @xmath1 to an inactive state . as in the previous section , we assume that this is a memoryless process . we further assume that the regime can only arrest or disperse protesters so long as the total fraction of active protesters remains less than the regime s finite police capacity @xmath25 . the transition rule can then be characterized by the following : if @xmath26 and @xmath27 i.e. , if @xmath1 is active at time @xmath9 and the fraction of active individuals is less than the regime s police capacity @xmath28 , then the node @xmath1 transitions from state 1 to state 0 at time @xmath29 , where @xmath30 is the first arrival time of a poisson process with rate @xmath31 , which determines the timescale of the decay process . this fully specifies the evolution of the abm , which we simulate using gillespie s algorithm , see appendix [ sec : gillespie ] . 0.48 , @xmath28 , @xmath21 , and @xmath31.,title="fig : " ] 0.48 , @xmath28 , @xmath21 , and @xmath31.,title="fig : " ] before deriving a population - level ode for the linear threshold abm and establishing the consistency of the svf model from @xcite with the abm in the following section , it is useful to first review some basic properties of the svf model . for example , we will show later that the parameter regimes of the svf are predictive of the parameter regime transitions in the abm . the svf model is given by the equation @xmath32 with parameters @xmath33 and @xmath34 . the growth and decay of the fraction of protesters @xmath35 are modelled by the _ growth _ and _ decay _ functions @xmath36\rightarrow\mathbb{r}^+$ ] , respectively . given their roles in determining the rate of growth and decay , the parameters @xmath21 and @xmath31 are called the _ protesters enthusiasm _ and _ police efficiency _ parameters in @xcite . the growth and decay in the number of protesters are modulated by the _ visibility _ and _ policing _ terms that are taken to be step functions @xmath37 and @xmath38 , where the indicator function is defined as @xmath39 these functions are parametrized by _ visibility _ and _ police capacity _ parameters @xmath40 and @xmath28 , respectively . in words , the fraction of protesters can grow only when the protest movement is sufficiently large to be `` visible '' to the general population , i.e. when protests are large enough that their existence can not be masked by the regime through censorship . conversely , the fraction of protesters can decay only so long as the fraction of protesters is less than the regime s finite capacity to disperse or arrest protesters . this is consistent with the idea that individuals who consider joining the revolution are faced with a collective action problem @xcite : if they choose to act unilaterally then they face severe retaliation from the regime , so they wait to act until the revolution becomes visible and if they act collectively then the regime loses the ability to punish due to a lack of resources . depending on the values of the parameters @xmath40 , @xmath28 , @xmath21 , and @xmath31 , the dynamics of the svf fall into one of four main regimes , see fig . [ fig : svf_summary ] . the equilibrium of _ total state control _ @xmath41 and the equilibrium of the _ realized revolution _ @xmath42 are present in all parameter regimes and are always locally asymptotically stable . different parameter regimes are distinguished by the behaviour of equation when @xmath43 lies between @xmath44 and @xmath28 . region ii ( @xmath45 ) is characterized by an interval of stable equilibria @xmath46 bounded by two unstable equilibria @xmath47 . since parameters in region ii are characterized by a relatively weak regime ( low @xmath28 ) and relatively small visibility ( low @xmath40 ) , region ii is interpreted in @xcite as the _ failed state _ parameter region . in contrast , regions iii0 , iiie , and iii1 ( @xmath48 ) are characterized by a relatively strong regime ( large @xmath28 ) and relatively large visibility ( large @xmath40 ) and differ in the value of the quantity @xmath49 , i.e. in the ratio of protester s enthusiam to the sum of protester s enthusiasm and police efficiency . in region iii0 , i.e. when @xmath50 , the interval @xmath51 lies in the basin of attraction of the equilibrium of total state control @xmath41 . analogously , in region iii1 , i.e. when @xmath52 , the interval @xmath51 lies in the basin of attraction of the equilibrium of the realized revolution @xmath42 . finally , in region iiie , i.e. when @xmath53 , the interval @xmath54 becomes the basin of attraction of a new equilibrium @xmath55 , which is called the _ equilibrium of civil unrest _ in @xcite . because of the contribution of the interval @xmath51 to the stability of the equilibrium of total state control , of the realized revolution , or of civil unrest , regions iii0 , iii1 , and iiie are called the _ stable police state _ , _ unstable police state _ , and _ meta - stable police state _ , respectively . for a full description , interpretation , and analysis of the svf model we refer the reader to @xcite . to show consistency of the abm proposed in section [ sec : spec_abm ] with the svf model of @xcite we begin by defining @xmath56 to be the fraction of nodes ( in the abm model ) that are expected to be in state 1 , i.e. the expected fraction of active nodes at time @xmath9 conditioned on information at time @xmath57 : @xmath58.\ ] ] the fraction of nodes that are expected to be in state 0 ( inactive ) at time @xmath9 , conditioned on information at time @xmath57 , is then @xmath59 . we now write the change in @xmath60 from time @xmath57 to time @xmath9 as @xmath61 where @xmath62 , and the expected growth and decay of the fraction of active nodes are modelled by the non - negative growth and decay functions @xmath63 , and @xmath64 , respectively . in order to obtain the one - compartment model we will need to approximate the quantities @xmath65 and @xmath66 in terms of @xmath67 , @xmath68 , and @xmath28 . for notational convenience we begin by considering the case where @xmath69 . the pool of individuals that can go from active to inactive at time @xmath57 is @xmath67 . active nodes become inactive at the first arrival time of a poisson process with rate @xmath31 , provided that the fraction of active nodes does not exceed the regime s police capacity , i.e. @xmath70 . thus , @xmath71 { \mbox { } } p(r_a;\beta ) = c_2 { \mbox { } } r_a(t_0 ) { \mbox { } } p(r_a(t_0 ) ; \beta ) { \mbox { } } \delta t+ o(\delta t),\ ] ] where we take @xmath38 , as in equation . next , let @xmath72,\ ] ] where @xmath73 denotes the neighbours of individual @xmath8 . in words , we let @xmath74 be the fraction of the total population at time @xmath57 that can see the revolution . we call @xmath74 the visibility function of the linear threshold abm . in the chosen notation for @xmath75 it is emphasized that @xmath75 depends on @xmath67 , i.e. on the fraction of nodes that are active at time @xmath57 . at time @xmath76 , the pool of individuals that can go from inactive to active , i.e. that are considering joining the revolution , is therefore approximately @xmath77 . since inactive nodes that can see the revolution become active at the first arrival time of a poisson process with rate @xmath21 , we have @xmath78 combining equations - and dividing by @xmath79 gives @xmath80 taking the limit as @xmath81 yields @xmath82 this establishes a general population - level ode model that approximates the linear threshold abm , with visibility function @xmath83 . to close the model , the visibility function has to be specified , guided by equation ( [ eq : nu ] ) . observe that if we substitute @xmath83 by the step function @xmath84 , then this exactly recovers the svf model of @xcite , see equation . so , provided that the step function @xmath85 is a suitable choice for @xmath86 , which we argued in the appendix of @xcite based on considering a simplified averaged threshold process that ignores correlations , this confirms the consistency of the abm and the svf . the crux of the argument used in @xcite , which we discuss further in sec . [ sec : bvf ] , is that , for real networks @xmath86 can be expected to be a steep sigmoidal function of @xmath43 , which can be approximated by a step function . in this paper , we test these assumptions explicitly on the empirical physical contact and facebook social networks discussed above using abm simulations , and we derive new improved ode models that form a more accurate representation of @xmath86 than a step function , taking into account the actual network structure of the empirical networks . the step function assumption was chosen in @xcite because it is useful when little information is known about the underlying network structure . for example , without having to specify an underlying communication network , the svf model provides a potential mechanism by which networks with increased visibility @xmath40 are more susceptible to political revolution . the following section explores the relationship between the abm and the svf model numerically for empirical networks by using a fitting procedure to attempt to estimate the network - dependent @xmath87 for which @xmath85 best approximates @xmath86 . given this estimate for @xmath88 , we compare abm simulations to solutions of equation in order to verify the consistency of the abm and the svf numerically and to investigate the abm behaviour in the parameter regimes of the svf model as described in fig . [ fig : svf_summary ] . section [ sec : avf ] then expands on this by proposing and analyzing a more sophisticated approximation for the visibility function @xmath86 . we conclude this section by noting that the above derivation can easily be modified for the case where the @xmath18 are not uniform . in this case the growth and decay functions @xmath89 and @xmath90 , respectively , can be computed by averaging over the distribution of the @xmath18 s . ( grey ) and solution to equation ( black ) . ( right ) @xmath91 error between abm realizations and the solution to equation versus @xmath40 . the optimal @xmath92 is found by fitting equation ( the svf model ) to @xmath93 abm simulations with parameters @xmath94 , @xmath95 , @xmath96 , @xmath97 , and @xmath98 ( region iii0 , @xmath99 ) . , title="fig : " ] ( grey ) and solution to equation ( black ) . ( right ) @xmath91 error between abm realizations and the solution to equation versus @xmath40 . the optimal @xmath92 is found by fitting equation ( the svf model ) to @xmath93 abm simulations with parameters @xmath94 , @xmath95 , @xmath96 , @xmath97 , and @xmath98 ( region iii0 , @xmath99 ) . , title="fig : " ] + ( grey ) and solution to equation ( black ) . ( right ) @xmath91 error between abm realizations and the solution to equation versus @xmath40 . the optimal @xmath92 is found by fitting equation ( the svf model ) to @xmath93 abm simulations with parameters @xmath94 , @xmath95 , @xmath96 , @xmath97 , and @xmath98 ( region iii0 , @xmath99 ) . , title="fig : " ] ( grey ) and solution to equation ( black ) . ( right ) @xmath91 error between abm realizations and the solution to equation versus @xmath40 . the optimal @xmath92 is found by fitting equation ( the svf model ) to @xmath93 abm simulations with parameters @xmath94 , @xmath95 , @xmath96 , @xmath97 , and @xmath98 ( region iii0 , @xmath99 ) . , title="fig : " ] as a function of @xmath2 by fitting equation ( the svf ) to @xmath93 abm simulations with parameters @xmath95 , @xmath96 , @xmath97 , and @xmath98 ( region iii0 , @xmath100 ) . ( blue circles ) facebook social network . ( red diamonds ) physical contact network . line of best fit for networks given by solid ( @xmath101 , @xmath102 ) and dashed ( @xmath103 , @xmath104 ) lines , respectively . ] , @xmath105 ) . ( right ) parameter region iii1 ( @xmath106 , @xmath107 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and bvf model ( dashed black ) with parameters @xmath108 , @xmath95 , @xmath96 , and @xmath109.,title="fig : " ] , @xmath105 ) . ( right ) parameter region iii1 ( @xmath106 , @xmath107 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and bvf model ( dashed black ) with parameters @xmath108 , @xmath95 , @xmath96 , and @xmath109.,title="fig : " ] + , @xmath105 ) . ( right ) parameter region iii1 ( @xmath106 , @xmath107 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and bvf model ( dashed black ) with parameters @xmath108 , @xmath95 , @xmath96 , and @xmath109.,title="fig : " ] , @xmath105 ) . ( right ) parameter region iii1 ( @xmath106 , @xmath107 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and bvf model ( dashed black ) with parameters @xmath108 , @xmath95 , @xmath96 , and @xmath109.,title="fig : " ] in this section we simulate the abm on the physical contact and facebook networks for a fixed set of model parameters and find the optimal @xmath40 for equation that minimizes the difference between the abm realizations and the output of equation . the closeness of the output of equation and the abm realizations for broad ranges of parameter values demonstrates that the svf model does indeed approximate the abm , at least for non - pathological parameter choices . we begin by fixing the initial condition @xmath110 is implemented by choosing @xmath111 nodes uniformly at random to be active . ] and the parameters @xmath112 , @xmath28 , @xmath21 , and @xmath31 . to compare realizations of the abm with the solution to equation we use gillespie s algorithm ( summarized in appendix [ sec : gillespie ] ) to simulate @xmath93 realizations of the abm , recording @xmath60 at each @xmath113 , and then find the @xmath40 that minimizes the difference ( in @xmath114 norm ) between the abm realizations and the solution to equation ( see appendix [ sec : solnsvf ] ) . the results of this procedure for abm simulations performed on the physical contact and facebook social networks with parameters @xmath98 , @xmath94 , @xmath95 , @xmath96 , and @xmath97 are given in fig . [ fig : fit_alpha_ode ] . for these parameters we find the optimal @xmath40 is given by @xmath115 and @xmath116 for the facebook and physical contact networks , respectively . equivalently , the fitted value for the visibility threshold @xmath44 is given by @xmath117 and @xmath118 , respectively . holding the initial condition @xmath98 and parameters @xmath95 , @xmath96 and @xmath97 constant , we are able to repeat the fitting procedure to find fitted values of @xmath40 for @xmath119 . these results are displayed in fig . [ fig : fit_multi_theta_alpha ] and illustrate that for a large range of individual thresholds @xmath2 the facebook social network has a slightly lower fitted visibility threshold @xmath120 than does the physical contact network . observe that , for the parameters employed in figs . [ fig : fit_alpha_ode]-[fig : fit_multi_theta_alpha ] ( @xmath94 , @xmath95 , @xmath96 , and @xmath97 ) , the svf model lies in the stable police state dynamical regime ( region iii0 , @xmath121 ) . we note that this is critical to the fitting procedure , since the parameter @xmath40 appears in the solution to equation only when the parameters @xmath40 , @xmath28 , @xmath21 , and @xmath31 lie in region iii0 , see appendix [ sec : solnsvf ] . we then also attempt to approximate the abm by the svf model in regions iiie and iii1 by using the fitted @xmath92 displayed in fig . [ fig : fit_multi_theta_alpha ] . figures [ fig : fit_alpha_iiie1]-[fig : fit_fail_regions_alpha_iiie1 ] show some representative results . as illustrated in fig . [ fig : fit_alpha_iiie1 ] , for many combinations of parameters the svf model is able to approximate the behaviour of the abm in parameter region iiie ( left panels of fig . [ fig : fit_alpha_iiie1 ] ) and parameter region iii1 ( right panels of fig . [ fig : fit_alpha_iiie1 ] ) . however , we observe that under certain circumstances stochastic effects and details of the network structure become important and the svf ode model fails to approximate , even in a qualitative sense , the behaviour of the abm . specifically , the svf model fails to capture the behaviour of the abm when the initial condition is small ( @xmath122 ) , when the initial condition is near the boundary between the basins of attraction of @xmath41 and @xmath42 , and when parameters @xmath21 , @xmath31 , @xmath40 , and @xmath28 are close to the boundary between parameter regions iii0 , iiie , or iii1 in fig . [ fig : svf_regimes ] , see figs . [ fig : fit_fail_r0_alpha_iiie1 ] , [ fig : fit_fail_r0_regions_alpha_iiie1 ] , and [ fig : fit_fail_regions_alpha_iiie1 ] , respectively . these are well - known limitations of population - level ode models that attempt to approximate complex dynamics @xcite . note that these plots also provide comparisons with the binomial visibility function ( bvf ) model , which we introduce in section [ sec : avf ] . in the previous section we observed that setting @xmath123 in equation exactly recovers the svf model . therefore , a natural strategy for improving on the ability of the svf model to approximate the aggregate behaviour of the abm is to specify an alternative visibility function that better captures the structure of the underlying network . section [ sec : avf ] proposes two such visibility functions and compares their ability to approximate the abm to the ability of the svf model to approximate the abm . ) , or ( right ) the initial condition is near the boundary between the basins of attraction of @xmath41 and @xmath42 ( @xmath124 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( black dashed ) with parameters @xmath108 , @xmath95 and @xmath125 ( region iii1 , @xmath126 ) . in the right panels , the bvf model performs much better than the svf model.,title="fig : " ] ) , or ( right ) the initial condition is near the boundary between the basins of attraction of @xmath41 and @xmath42 ( @xmath124 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( black dashed ) with parameters @xmath108 , @xmath95 and @xmath125 ( region iii1 , @xmath126 ) . in the right panels , the bvf model performs much better than the svf model.,title="fig : " ] + ) , or ( right ) the initial condition is near the boundary between the basins of attraction of @xmath41 and @xmath42 ( @xmath124 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( black dashed ) with parameters @xmath108 , @xmath95 and @xmath125 ( region iii1 , @xmath126 ) . in the right panels , the bvf model performs much better than the svf model.,title="fig : " ] ) , or ( right ) the initial condition is near the boundary between the basins of attraction of @xmath41 and @xmath42 ( @xmath124 ) . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( black dashed ) with parameters @xmath108 , @xmath95 and @xmath125 ( region iii1 , @xmath126 ) . in the right panels , the bvf model performs much better than the svf model.,title="fig : " ] and @xmath42 , i.e. when @xmath127 . ( left ) facebook social network @xmath128 ( @xmath129 ) and ( right ) physical contact network @xmath130 ( @xmath131 ) : equation has solution in region iiie , while abm has solution in region iii0 . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( dashed black ) with parameters @xmath132 , @xmath95 , @xmath96 , and @xmath133 , and initial condition @xmath109 ( region iiie , @xmath134 ) . the bvf model performs much better than the svf model . ] ( @xmath135 ) with parameters in region iii1 ( @xmath136 , @xmath137 ) : equation has solution in region iii1 , while abm has solution in region iiie . ( right ) physical contact network @xmath130 ( @xmath117 ) with parameters in region iiie ( @xmath138 , @xmath139 ) : the svf has solution in region iiie , while abm has solution in region iii1 . time traces of @xmath93 abm simulations ( grey ) , solution to the svf model ( solid black ) , and solution to the bvf model ( dashed black ) with parameters @xmath94 , @xmath95 , and @xmath96 and with intial condition @xmath109 . the bvf model performs much better than the svf model . ] due to the difficulty of analyzing and computational cost of simulating the abm , as well as the limitations of the svf in approximating the abm , we propose two alternative choices for visibility functions ( different from the step - visibility function @xmath140 ) that can lend insight into how the abm process behaves on different networks . specifically , sections [ sec : bvf]-[sec : evf ] introduce the binomial and empirical visibility functions , respectively ( bvf and evf ) . these two visibility functions can be seen to be equivalent in the limit of large network and sample size , as shown in appendix [ sec : comp_bvf_evf ] . section [ sec : comp_abm_bvf_evf ] demonstrates numerically that the use of the bvf / evf in equation represents an improvement over the use of the step visibility function @xmath140 ( svf ) when approximating the abm for real networks . consider the following derivation . suppose that @xmath8 is a node in the network @xmath4 with degree @xmath15 , and that the fraction of nodes active in the revolution is @xmath43 . also , assume that the states of @xmath1 s neighbours are active with probability @xmath43 independently so that the probability of @xmath1 having @xmath141 neighbours active in the revolution is @xmath142 the probability that the fraction of @xmath1 s neighbours exceeds the linear threhsold @xmath2 is , therefore , @xmath143 where @xmath144 is the cumulative distribution function for the binomial distribution with @xmath145 trials and probability of success @xmath146 evaluated at @xmath147 . this quantity can be used to approximate the function @xmath148 ( the expected fraction of the population that can see the revolution ) . in @xcite we considered @xmath15 in equation to be the average degree over all nodes and argued that equation is a sigmoidal function that can be approximated by step - visibility function @xmath140 for some appropriate visibility parameter @xmath40 ( see appendix [ sec : justsvf ] for a brief summary of this argument ) . one possible improvement on adopting @xmath149 in equation ( resulting in the svf ) is to actually adopt @xmath150 instead , where @xmath15 is the average degree of the network . however , we can improve on this even further in terms of using more relevant information about the real graph by choosing @xmath86 to be equal to @xmath151 we call the function @xmath152 in equation the _ binomial visibility function _ ( bvf ) , see fig . [ fig : bvf ] . note the steep sigmoidal shape of @xmath153 for the two empirical networks of fig . [ fig : bvf ] , further justifying the choice of a step visibility function as an appropriate approximation in @xcite . we note that computing the binomial visibility function on some grid of @xmath154 discrete @xmath43-values @xmath155 requires calculating a double sum for every @xmath156 . in practice , this can be very expensive , especially for degree distributions @xmath7 with fat tails . for this reason , the following section introduces a complementary but less expensive approximation to the visibility function that is equivalent to the binomial visibility function of equation in the limit of large network and sample size . 0.48 ( curves from left to right ) . the binomial and empirical visibility functions are computed on the grid @xmath157,title="fig : " ] 0.48 ( curves from left to right ) . the binomial and empirical visibility functions are computed on the grid @xmath157,title="fig : " ] we now propose to reconstruct the binomial visibility function @xmath153 empirically by sampling the underlying network directly . specifically , for fixed @xmath2 we approximate @xmath158 by a discrete function @xmath159 where @xmath160 . since @xmath161 and @xmath162 , we set @xmath163 and @xmath164 . we calculate @xmath165 for @xmath166 as follows . 1 . seed the network with @xmath167 active nodes in expectation . ( generate @xmath168 random numbers drawn from the uniform distribution on @xmath169 $ ] : @xmath170 for @xmath171 . if @xmath172 then node @xmath173 is active , otherwise node @xmath173 is inactive . ) 2 . determine the fraction of nodes that can see the revolution . 3 . repeat 1 - 2 a total of @xmath174 times and set @xmath165 to be the average over the realizations . ( we choose @xmath93 in our numerical experiments . ) we denote the linear interpolation between the @xmath165 by the function @xmath175 @xmath176 \mbox { for some } j=1,\ldots,\mu\end{array}\right.\ ] ] which we call the _ empirical visibility function _ , see fig . [ fig : bvf ] . we now make several brief remarks about the procedure for calculating the empirical visibility function , outlined above . firstly , we remark that this procedure is easily extended to the case where @xmath18 are specified for each individual node . however , to keep our analysis as simple as possible we continue to restrict ourselves to the case where @xmath112 . in appendix [ sec : comp_bvf_evf ] we show that the empirical and binomial visibility functions are equivalent in the limit of large network and sample size . figure [ fig : bvf ] shows that they can also be expected to give similar results for finite network and sample size . finally , we remark that compared to the binomial visibility function , the procedure for calculating the empirical visibility function is in practice significantly less costly to implement , especially for networks with fat - tailed degree distributions . in this section we compare the ability of the bvf model to approximate the abm with that of the svf model . we note that since the bvf and evf are equivalent in the limit of large network and sample size and are very close in practice ( see fig . [ fig : bvf ] ) , the observations we make about the behaviour of the bvf model also apply to the behaviour of the evf model . we further note that the behaviour of the bvf / evf model has parameters regimes that induce behaviour analogous to those observed in the svf model . since the bvf / evf have sigmoidal shape , this follows directly from the analysis presented in appendix b of @xcite for visibility and policing functions of general sigmoidal shape ( the results from this analysis are summarized in appendix [ sec : ext ] ) . we observe four types of outcomes when comparing the ability of the bvf model to approximate the abm to the ability of the svf model to approximate the abm : 1 . the bvf model may succeed better in approximating the dynamics of the abm than the svf model , even if the improvement is only qualitative in nature , see figs . [ fig : fit_fail_r0_alpha_iiie1]a and b ( right panel ) , [ fig : fit_fail_r0_regions_alpha_iiie1 ] , and [ fig : fit_fail_regions_alpha_iiie1 ] ( left panel ) , 2 . the bvf model may complement the dynamics predicted by the svf model , see fig . [ fig : complement ] , 3 . the bvf model may produce qualitatively similar predictions to the svf model , see figs . [ fig : fit_alpha_iiie1 ] , [ fig : fit_fail_r0_alpha_iiie1]b ( left panel ) and [ fig : fit_fail_regions_alpha_iiie1 ] ( right panel ) , or 4 . both the bvf and the svf models fail to approximate the dynamics of the abm either qualitatively or quantitatively , see figs . [ fig : fit_fail_r0_alpha_iiie1]b ( left panel ) and [ fig : fail ] . for each of these outcomes , the approximation of the abm by the bvf model is no worse than the approximation by the svf model , and , in many cases the approximation of the abm by the bvf model is much better than the approximation by the svf model . therefore , we conclude that the bvf model , and hence the evf model , represents an improvement over the svf model with respect to their ability to approximate the behaviour of the abm . this is no surprise , since the bvf / evf model takes important information of the real network into account . 0.48 abm simulations ( grey ) , solution to the svf model ( black ) , and solution to the bvf model ( dashed black ) , with parameters @xmath94 , @xmath95 , and @xmath96 and with intial condition @xmath177 . , title="fig : " ] 0.48 abm simulations ( grey ) , solution to the svf model ( black ) , and solution to the bvf model ( dashed black ) , with parameters @xmath94 , @xmath95 , and @xmath96 and with intial condition @xmath177 . , title="fig : " ] in the previous section we introduced an extension to the svf model . in particular , we showed that the bvf / evf model is better able to approximate the abm than the svf model . however , both the bvf / evf and svf models are one - dimensional ordinary differential equation . this section now compares the performance of the bvf / evf models as approximations to the abm to the performance of a higher order ( i.e. , higher dimensional ) compartmental model which we modify from @xcite . we call this model the _ degree approximation _ ( da ) because it compartmentalizes individuals based on their state as well as their degree . we demonstrate numerically that the bvf model , and hence the evf model , is no worse at approximating the aggregate behaviour of the abm than the da . this highlights the usefulness of the bvf / evf model , especially since the da is in practice much more computationally expensive to solve and much harder to analyze than the bvf / evf model . before we detail the approach which we modify from @xcite , it is useful to introduce some notation . we define @xmath178 to be the set of nodes with degree @xmath15 and @xmath179 to be the set of nodes that have at least one neighbour of degree @xmath15 , @xmath180 respectively , where we denote the degree distribution of @xmath179 by @xmath181 , i.e. @xmath182 as mentioned above , the degree approximation aggregates individuals according to their state and degree . thus , conditioning on the state of the system at time @xmath57 , we define @xmath183\ ] ] to be the fraction of nodes with degree @xmath15 that are expected to be in state 1 at time @xmath9 . since nodes and edges are neither created nor destroyed , the fraction of nodes with degree @xmath15 that are expected to be in state 0 at time @xmath9 is given by @xmath184 . analogously , we define @xmath185 to be the fraction of nodes expected to be in state 1 at time @xmath9 . applying the notation introduced above with the rules characterizing the dynamics of the process we find an equation analogous to equation @xmath186 where the expected growth and decay of the fraction of active nodes is modelled by the nonnegative growth and decay functions @xmath187 and @xmath188 , respectively . in order to close the degree approximation we need to approximate the quantities @xmath189 and @xmath190 in terms of @xmath191 , @xmath192 , and @xmath28 . for notational convenience we first suppress the time argument , and consider the case where @xmath112 . in order to approximate @xmath189 or @xmath190 in terms of @xmath193 , @xmath2 , and @xmath28 , we are now required to make an assumption about how the states of nodes sharing a common neighbour are correlated . specifically , as for the bvf and evf models , for any fixed @xmath194 we assume that the states of any two neighbours of @xmath1 are independent . this assumption implies that the probability of @xmath1 having exactly @xmath141 active neighbours is @xmath195 , where @xmath196 is the fraction of nodes in @xmath197 with state 1 . so , the probability of @xmath1 having at least @xmath198 active neighbours is @xmath199 it follows that @xmath200 as in equation , we choose @xmath201 combining equations - and dividing by @xmath79 gives @xmath202 now , taking the limit as @xmath81 yields @xmath203 abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] + abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] abm simulations ( grey ) , solution to bvf model ( dashed black ) and da ( solid black).,title="fig : " ] in section [ sec : comp_abm_bvf_evf ] we showed that the approximation of the abm by the bvf / evf model is no worse , and usually much better , than the approximation of the abm by the svf model . in this section we show that the approximation of the abm by the bvf / evf model is no worse than the approximation of the abm by the da given in equation . specifically , we observe that the bvf and da model produce similar qualitative predictions for most choices of parameters @xmath2 , @xmath28 , @xmath21 , and @xmath31 , see for example fig . [ fig : evf_da ] . also , although this is an atypical result , we present one set of parameters where the bvf model outperforms the da on the facebook social network @xmath128 , and where the bvf model complements the da on the physical contact network @xmath130 , see fig . [ fig : evf_da_2 ] . we observe that , for the specific parameters and networks used in the simulations of figs . [ fig : evf_da]-[fig : evf_da_2 ] , the da does not significantly outperform the bvf model . it was confirmed in extensive additional simulations that this occurs generically for large parts of the parameter space of the models . this is worth emphasizing once more , since the bvf model is much easier to analyze and much less costly to solve than the da . this supports our proposition that the bvf ( or evf ) model is a powerful tool for approximating and analyzing the abm and a general technique for formulating tractable models for spreading processes on social networks that take real network characteristics into account . 0.48 abm simulations ( grey ) , solution to the bvf model ( dashed black ) and the da ( solid black).,title="fig : " ] 0.48 abm simulations ( grey ) , solution to the bvf model ( dashed black ) and the da ( solid black).,title="fig : " ] in epidemiological models the _ basic reproduction number _ @xmath0 is defined to be the number of secondary infections caused by a single infected individual introduced into a population that is entirely susceptible @xcite . note that in this context exposure to one infected individual is sufficient for a new infection . if @xmath204 then an infected individual is , on average , unable to replace himself and the outbreak terminates on its own in short order . on the other hand , if @xmath205 then an infected individual is , on average , more than able to replace himself and the outbreak may spread into a full - blown epidemic we define the basic reproduction number for our linear threshold model of political revolution analogously : the basic reproduction number is the average number of individuals that become active in the revolution due directly to the introduction of a single active individual into a population that is otherwise completely inactive . in this section , we show that the basic reproduction number in our linear threshold abm is related to the initial slope of the bvf / evf by the equation @xmath206 and discuss efficient ways to compute @xmath0 . we begin by deriving an expression for @xmath0 . suppose that we activate individual @xmath8 in the population , and suppose that individual @xmath1 has degree @xmath15 . further , suppose that this individual has @xmath207 neighbours who can see the revolution ( i.e. , the linear threshold criterion is satisfied ) once individual @xmath1 is activated . let @xmath208 be a neighbour of @xmath1 s that can see the revolution , and let @xmath209 and @xmath210 be the first arrival time of poisson processes with rates @xmath21 and @xmath31 , respectively . the probability that individual @xmath211 becomes active before individual @xmath1 becomes inactive is @xmath212 c_2\exp(-c_2 t ) dt\\ & = \frac{c_1}{c_1+c_2 } = c^*.\end{aligned}\ ] ] since we only consider activations that result directly from @xmath1 s activation ( and not activations caused by a combination of the activation of @xmath1 and the subsequent activation of @xmath1 s neighbours ) , for our purposes the behaviour of each of @xmath1 s neighbours is independent . thus , the expected number of @xmath1 s neighbours which become active is @xmath213 . recall that the social network @xmath214 has @xmath168 nodes , @xmath215 edges , degree distribution @xmath7 and secondary degree distribution @xmath181 . for the single activated individual @xmath1 with degree @xmath15 the expected number of neighbours who can see the revolution is @xmath216 = k \sum_j \rho_{k , j}{\mathbb{i}_{\ { j\theta \leq 1\}}}.\ ] ] averaging over all possible degrees @xmath15 now yields the basic reproduction number @xmath217 = \frac{c_1}{c_1+c_2 } \sum_{k , j } k \rho_k \rho_{k , j}{\mathbb{i}_{\ { j\theta \leq 1\ } } } = c^*\sum_{k=1}^\infty \sum_{j=1}^{\lfloor \theta^{-1 } \rfloor } k \rho_k \rho_{k , j}.\ ] ] it is now possible to prove the relationship given in equation . writing @xmath218 it suffices to show that @xmath219 since @xmath220 \nonumber \\ & \implies v_b'(0;\theta,\rho_k ) = \sum_{k=1}^\infty k \rho_k { \mbox { } } { \mathbb{i}_{\{\theta k \leq 1\ } } } = \sum_{k=1}^{\lfloor \theta^{-1}\rfloor } k \rho_k . \label{eq : vbprime}\end{aligned}\ ] ] we note that together with equation , equation gives an exact and cheap way to compute @xmath0 . from the undirected social network @xmath4 form the directed social network @xmath221 , where @xmath222 in this case , the number of edges emanating from all nodes of degree @xmath15 is @xmath223 . so , the number of edges emanating from nodes of degree @xmath15 and incident on nodes of degree @xmath141 is @xmath224 . thus , the number of edges incident on nodes of degree @xmath141 is @xmath225 equivalently , the number of edges incident on nodes of degree @xmath141 is @xmath226 putting these two expressions together , we find @xmath227 which completes the proof . intuitively , this makes sense since the number of nodes that can see the revolution when one node is activated can also be approximated by @xmath228 in this section we investigate experimentally the differences in spreading behaviour that occur under the linear threshold model for the spread of political revolutions when applied to some empirical online and offline social networks , searching for some initial quantitative evidence that political revolutions may be facilitated by the network structure of online social networks of social media . it is often assumed that the connectivity of modern online social networks was an important factor in the spread of political revolutions in the past decade , e.g. , in the arab spring revolutions of 2011 @xcite , while traditional offline social networks ( using in - person physical contact , or phone or mail interaction for safe communication ) featured a different connectivity structure that was often severely restricted by censorship of the regime . unfortunately , representative samples of the offline , traditional communication networks that were in existence prior to the adoption of new media technologies are unavailable at sufficient scale for countries affected by the arab spring , or indeed for any country . as opposed to the online networks of social media which are , by their nature , digitally stored and available , the offline social networks of pre - internet societies have not been recorded at scale , simply because it was impractical in terms of cost and effort . this is a serious roadblock when investigating the effects of the structure of new media networks on the dynamics of political revolution , and if one wants to compare with propagation on pre - internet social networks , it is necessary to identify proxy networks that are likely to be reasonable approximations to pre - internet social networks , in terms of network structure . in this section , we use simulations of our abm model on two empirical networks as a starting point to investigate differences in propagation properties that may arise between online and offline social networks within the linear threshold propagation model . we choose the small physical contact network @xmath130 between individuals from @xcite as a representative for offline social networks , and we choose the facebook network @xmath128 as a representative for online social networks . 0.48 of @xmath93 realizations of abm simulations on facebook and physical contact networks with @xmath229 $ ] , @xmath230 , @xmath95 , @xmath96 , and @xmath231 ( region iii1 , @xmath232 ) . realizations are allowed to run until time @xmath233 . basic reproduction numbers @xmath0 for facebook social and physical contact networks are @xmath234 and @xmath235 , respectively.,title="fig : " ] 0.48 of @xmath93 realizations of abm simulations on facebook and physical contact networks with @xmath229 $ ] , @xmath230 , @xmath95 , @xmath96 , and @xmath231 ( region iii1 , @xmath232 ) . realizations are allowed to run until time @xmath233 . basic reproduction numbers @xmath0 for facebook social and physical contact networks are @xmath234 and @xmath235 , respectively.,title="fig : " ] we now briefly examine the differences in how the political revolution spreads on the facebook social and physical contact networks via direct simulation of the abm . in particular , choosing parameters @xmath230 , @xmath95 , @xmath96 , and @xmath231 in the unstable police state region ( region iii1 , @xmath236 ) , we simulate @xmath93 realizations of the abm for each initial condition @xmath237 from time @xmath238 until time @xmath239 . the average final size ( at @xmath239 ) for each initial condition @xmath240 is recorded and displayed in fig . [ fig : r0_rfinal ] , which demonstrates that for these parameters the political revolution described by our abm propagates to a much greater extent on the facebook social network , i.e. the online proxy network , than on the physical contact network , i.e. the offline proxy network . as an important illustration of the usefulness of the basic reproduction number @xmath0 we defined in sec . [ sec : r0 ] , we find that this is consistent with a difference in @xmath0 for the two networks : the basic reproduction number is @xmath241 ( above the value of 1 which is expected to be required for propagation ) for the facebook network , and @xmath242 for the physical contact network . these results indicate that the offline social network is less conducive to spreading the revolution in the abm than the online social network : it has a smaller basic reproduction number , and in simulations a larger initial population of revolutionaries is required to spread the revolution . this provides some initial quantitative evidence that the spread of revolutions under a linear threshold process may occur more easily on modern online social networks than on traditional offline networks . while this is an interesting first observation , the next section discusses limitations and further investigations that are required to address this intriguing but complex question more comprehensively . in this paper we developed a linear threshold agent based model ( abm ) to model the spread of a political revolution in a dictatorial regime . we showed that this model is consistent with previous simple step visibility function ( svf ) ode model developed in @xcite . using the relationship between these two models as a template we developed a hierarchy of models of varying complexity that approximate the behaviour of the abm , see table [ tab : hierarchy ] . of all the models we have identified , we find that the bvf and evf models ( models of moderate complexity ) offer the optimal combination of low computational complexity ( cost ) , ease of analysis , and ability to approximate the behaviour of the abm . specifically , we find that for most parameters and initial conditions the bvf / evf model is better able to approximate average abm behaviour than the svf . also , for most networks the bvf / evf model is much less costly to solve and much easier to analyze than the degree approximation from @xcite . importantly , the analysis of the simple ode models in terms of stability of solutions for various parameter regimes directly gives insight in the qualitative dynamics of the linear threshold abm for the spread of political revolutions . we extended the concept of the basic reproduction number @xmath0 from epidemiology @xcite to the linear threshold abm , we showed how it is related to the slope of the empirical or binomial visibility functions at @xmath41 , and we provided efficient ways to compute or estimate it . analogously to epidemiological models , when @xmath205 we expect the political revolution to spread , and when @xmath243 we expect it to die out . thus , computing this quantity for a network can give an indication of how the abm will behave on that network without the need to perform simulations , as we have demonstrated for empirical networks . the facebook and physical contact networks we consider as case studies , i.e. the online and offline proxy networks , provide initial support to the hypothesis that the adoption of online social media may facilitate the spread of political revolutions by effectively changing the connectivity structure of the population in a way that makes linear threshold spreading more effective . specifically , we find that for certain parameters the online proxy network is more susceptible to the linear threshold spreading process than the offline proxy network . moreover , we find that the different behaviour of these two networks is consistent with the basic reproduction number calculated for these two networks . in addition to studying the abm and its approximations on more and larger online and offline networks , much work remains to be done in the actual modelling of the political revolution process . for example , throughout this manuscript we have assumed that the linear threshold of individuals is constant for the population , that the underlying communication networks are static , that the graph is undirected , and that the nodes that are initially activated are chosen uniformly at random . each of these assumptions represents a major simplification of reality that needs to be addressed to study further aspects of the spread of political revolutions on online social networks . for example , consider that in our model individuals estimate the current participation in the revolution by sampling their neighbours . since individuals with greater sample size , i.e. larger degree , can form more accurate estimates of the current participation in the revolution , they should be willing to join the revolution at a lower linear threshold @xmath2 than individuals who are more uncertain in their estimate of revolution size . we would therefore expect the linear threshold to vary from individual to individual as a function of their network degree . or , consider that one of the principal strengths of new media is the ability of individuals to search for both content and like - minded individuals . thus , we would expect that the underlying communication network should be changing on the same time scale as the revolution . finally , one might consider that the nodes that are predisposed to be active at the initial stages of a revolution , e.g. nodes that represent activists , may also have larger degree ( if , for example , they are charismatic and allowed to accrue followers ) or smaller degree ( if , for example , they are the specific target of regime censorship ) . there is significant value in the kind of parsimonious abm model on static networks that we have considered in this paper , because much can be learned from this type of model and it is easier to analyze and interpret than more complex models . nevertheless , extensions along the lines sketched above are important avenues for further study . the simulations in this paper were performed in matlab . all matlab code and data files will be made publicly available online upon publication of this paper . 10 j.c . lang and h. de sterck . . , 69:1221 , 2014 . d. kempe , j. kleinberg , and e. tardos . maximizing the spread of influence through a social network . in _ proceedings of the ninth acm sigkdd international conference on knowledge discovery and data mining _ , pages 137147 , 2003 . d. centola , r. willer , and m. macy . . , 110(4):10091040 , 2005 . d. centola , v.m . eguiluz , and m.w . cascade dynamics of complex propagation . , 374:449456 , 2007 . d. centola and m.w . 113(3):702734 , 2007 . d.j . watts . a simple model of global cascades on random networks . , 99(9):57665771 , 2002 . ames , d.b . george , c.p . hampson , a.r . kanarek , c.d . mcbee , d.r . lockwood , j.d . achter , and c.t . , 278(1724):35443550 , 2011 . s. bansal , b.t . grenfell , and l.a . meyers . . , 4(16):879891 , 2007 . the mathematics of infectious diseases . , 42(4):599653 , 2000 . p. lind , l. da silva , j. andrade , and h. herrmann . spreading gossip in social networks . , 76(3):036117 , 2007 . l. zhao , q. wang , j. cheng , y. chen , j. wang , and w. huang . . , 390(13):26192625 , 2011 . m. granovetter . . , 83(6):14201443 , 1978 . t. kuran . . , 44(1):748 , 1991 . m. nekovee , y. moreno , g. bianconi , and m. marsili . . , 374(1):457470 , january 2007 . nahed eltantawy and julie b wiest . the arab spring | social media in the egyptian revolution : reconsidering resource mobilization theory . , 5:18 , 2011 . habibul haque khondker . role of the new media in the arab spring . , 8(5):675679 , 2011 . zeynep tufekci and christopher wilson . social media and the decision to participate in political protest : observations from tahrir square . , 62(2):363379 , 2012 . s. khamis and k. vaughn . . , 14 , 2011 . julian mcauley and jure leskovec . . in _ advances in neural information processing systems 25 _ , pages 548556 , 2012 . marcel salath , maria kazandjieva , jung woo lee , philip levis , marcus w feldman , and james h jones . . , 107(51):2202022025 , 2010 . attila grsoy and murat atun . neighbourhood preserving load balancing : a self - organizing approach . in _ euro - par 2000 parallel processing _ , pages 234241 . springer , 2000 . kiss , d.m . green , and r.r . the effect of contact heterogeneity and multiple routes of transmission on final epidemic size . , 203(1):124136 , 2006 . r. pastor - satorras and a. vespignani . epidemic dynamics in finite size scale - free networks . , 65(3):035108 , 2002 . j. lindquist , j. ma , p. driessche , and f.h . effective degree network disease models . , 62(2):143164 , 2010 . in this appendix , we briefly discuss the network extraction protocol for the physical contact network @xmath130 from @xcite . the network presented in @xcite was constructed by distributing wireless sensors to students , teachers , and staff at a u.s . high school during a one day period from approximately 08:00 to 16:30 . when two wireless sensors are in proximity of one another , i.e. when they are less than approximately 3 m apart , they register an interaction with a temporal resolution of 20s . data are given for each seperate interaction in csv format with three columns : i d of first wireless sensor , i d of second wireless sensor , duration of interaction ( measured in 20s increments ) . a weighted undirected network @xmath244 is formed by connecting each pair of individuals with an edge whose weight is given by the total amount of time they spent in proximity to one another . in order to admit a comparison with the unweighted undirected facebook subnetwork , we de - weight network @xmath244 by discarding all edges whose weight is less than the minimum duration @xmath245 , and by weighting all remaining edges equally . we denote the largest connected component of the resulting network by @xmath246 , and for convenience we refer to this network as the _ physical contact network_. the choice of the minimum duration @xmath245 is important since it determines , for example , the average degree of the physical contact network @xmath130 . we choose @xmath245 such that the average degree of @xmath130 is close to the average degree of @xmath128 , for the following reason . in the simulations carried out in this paper to compare propagation under the linear threshold abm on @xmath130 and @xmath128 , we seed the networks with a fixed percentage of active nodes and use the same threshold @xmath2 on both networks . we want to calibrate the average degree of @xmath130 to the average degree of @xmath128 , such that , on average , nodes in @xmath130 have the same chance as nodes in @xmath128 to satisfy the linear threshold criterion and see the revolution . after this calibration , differences in propagation between @xmath130 and @xmath128 ( for the same @xmath2 ) are only due to the differences in network structure that go beyond the average degree . this motivates us to choose @xmath247 sensor measurements , equivalent to 11 minutes and 20 seconds , in order to match the average degrees of @xmath130 and the facebook subnetwork as closely as possible . as argued in appendix a of @xcite , the expression for the visibility function given in equation , with @xmath15 taken to be the average degree of the network , can be expected to have a steep sigmoidal shape , which can be approximated by a step function . indeed , fig . [ fig : binom ] shows that @xmath248 has a clear sigmoidal shape , with a steep transition from 0 to 1 . in @xcite it was also argued that increased average degree @xmath15 can be expected to co - occur with decreased threshold @xmath2 , since the more neighbors one has the more certain one can be about the true extent of the revolution because large samples are more reliable . under this assumption , @xmath248 can be considered a single - parameter sigmoidal function , which can be approximated by the single - parameter step visibility function ( svf ) @xmath140 , with parameter @xmath40 indicating where the step transition occurs . on ( top left ) @xmath2 , ( top right ) @xmath15 , and ( bottom ) negatively correlated @xmath2 and @xmath15.,title="fig:",width=240 ] on ( top left ) @xmath2 , ( top right ) @xmath15 , and ( bottom ) negatively correlated @xmath2 and @xmath15.,title="fig:",width=240 ] + on ( top left ) @xmath2 , ( top right ) @xmath15 , and ( bottom ) negatively correlated @xmath2 and @xmath15.,title="fig:",width=240 ] we conclude this section with a few remarks regarding the assumption made in sec . [ sec : bvf ] and in @xcite that the states of an individual s neighbors are uncorrelated . while this assumption is applied in many models of spreading processes , for example rumour spreading processes @xcite and epidemiological processes @xcite , this assumption was not verified in @xcite . in this paper we show that this assumption results in a valid approximation by comparing simulation results of the svf model , and also the bvf / evf models ( which were derived in sec . [ sec : avf ] under the same assumption ) , with abm simulations for empirical networks ( which do not make this assumption ) , finding consistent results for large parts of parameter space . in this section , we summarize the parameter regime and solution stability analysis that was developed in appendix b of @xcite for the one - dimensional ode model @xmath249 with generic sigmoidal visibility function @xmath250 as in fig . [ fig : sigmoid ] . since the newly derived bvf / evf in this paper have sigmoidal shape , the stability analysis of @xcite applies to the bvf / evf models . with a fast transition from 0 to 1 that follows a sigmoidal profile . ] as is shown in @xcite , the dynamics of model with generic sigmoidal visibility and policing functions closely follows the corresponding dynamics of model in region ii ( an open interval of equlibria @xmath251 ) , regions iii0 and iii1 ( one equilibrium @xmath252 ) , and region iiie ( three equilibria @xmath252 ) , thus establishing the equivalence in terms of dynamic behaviour of models ( with step functions ) and ( with sigmoidal functions , including the bvf / evf models ) . we summarize the phase portraits of the different regions for model in fig . [ fig : summaryext ] , which are analogous to the phase portraits of the svf model shown in fig . [ fig : svf_stability ] . see appendix b of @xcite for a detailed derivation and further explanation . for the bvf / evf models , the values of equilibrium quantities like @xmath253 and @xmath254 depend on @xmath2 , @xmath28 and the network structure . -axis ( @xmath255 $ ] ) for model . closed ( open ) circles represent locally asymptotically stable ( unstable ) equilibria . left ( right ) arrows indicate regions where @xmath256 ( @xmath257 ) . the phase portraits are similar to the case of the svf model in fig . [ fig : svf_stability ] . ] it is also important to point out how the svf stability region plot of fig . [ fig : svf_regimes ] generalizes to the bvf / evf models . for a given empirical network , the bvf / evf model does not feature a range of visibility parameters @xmath40 like the svf model , but each choice of @xmath2 results in a different bvf / evf function . for a fixed @xmath2 , the different phase portrait behaviours of fig . [ fig : summaryext ] occur depending on the values of @xmath28 , @xmath21 and @xmath31 . as a result , for a given empirical network , the parameter space for the bfv / evf models as a function of @xmath2 , @xmath28 , @xmath21 and @xmath31 can be divided in regions as for the svf model in fig . [ fig : svf_regimes ] , where @xmath2 replaces @xmath40 as the parameter on the horizontal axis . the precise shape of the bvf / evf parameter regions would depend on the actual empirical network considered . for example , the boundary between the ii and iii regions is in general not a straight line , but needs to be determined numerically . following @xcite , we numerically simulate the abm by implementing gillespie s algorithm . in this appendix we give a brief overview of this algorithm . we first introduce some notation . as above , let @xmath60 denote the fraction of individuals in the population that are expected to be active in the revolution at time @xmath9 . for each node @xmath258 we let @xmath259 if @xmath1 can see the revolution at time @xmath9 , i.e. if @xmath1 satisfies equation , and let @xmath260 otherwise . furthermore , we let @xmath261 and @xmath262 denote the first arrival times of independent poisson processes with rates @xmath21 and @xmath31 , respectively . gillespie s algorithm is based on the fact that the sum of two independent poisson variables is also a poisson variable with rate equal to the sum of the rates of the original processes . it follows that @xmath263\ ] ] is a poisson process with rate @xmath264.\ ] ] the first arrival time of this process , therefore , is an exponential random variable @xmath265 with rate @xmath266 . at time @xmath267 the state of exactly one of the nodes will change . moreover , since @xmath261 and @xmath262 are independent , the probability that the state of node @xmath1 will change is @xmath268 the gillespie algorithm then proceeds iteratively in three steps . 1 . find the time @xmath265 of the next event by drawing @xmath265 from an exponential distribution with rate @xmath266 . 2 . determine which node changes state by drawing one node from @xmath269 , where node @xmath8 is drawn with probability @xmath270 . update @xmath271 and re - calculate @xmath272 , @xmath35 , @xmath273 . since the complete analytic solution to the svf model was not provided in @xcite but is useful in the current paper , we provide it here . the solution to equation is as follows . in region i ( @xmath274 ) and region ii ( @xmath275 ) @xmath276\\ 1 - ( 1-r_0)e^{-c_1(t - t_0 ) } & \mbox{if } r_0 > 1-\alpha \end{array } \right . .\ ] ] in regions iii0 ( @xmath277 ) , iiie ( @xmath53 ) , and iii1 ( @xmath278 ) @xmath279 \mathbb{i}_{\{t < t_{\alpha}\ } } \\ \hspace{5 mm } + ( 1- \alpha ) e^{-c_2(t - t_\alpha ) } \mathbb{i}_{\{t\geq t_{\alpha}\ } } & \mbox{if } r_0 \in [ \beta , 1-\alpha ] \mbox { and } c^ * \leq 1-\alpha < \beta\\ \\ c^*+ ( r_0-c^*)e^{-(c_1+c_2)(t - t_0 ) } & \mbox{if } r_0 \in [ \beta , 1-\alpha ] \mbox { and } 1-\alpha < c^ * < \beta\\\\ \left [ c^*+ ( r_0-c^*)e^{-(c_1+c_2)(t - t_0 ) } \right ] \mathbb{i}_{\{t < t_{\beta}\ } } \\ \hspace{5 mm } + \left [ 1 - ( 1- \beta ) e^{-c_1(t - t_\beta ) } \right ] \mathbb{i}_{\{t\geq t_{\beta}\ } } & \mbox{if } r_0 \in [ \beta , 1-\alpha ] \mbox { and } 1-\alpha < \beta \leq c^*\\\\ 1 - ( 1-r_0)e^{-c_1(t - t_0 ) } & \mbox{if } r_0 > \beta \end{array } \right . , \ ] ] where @xmath280 we briefly argue that the empirical and binomial visibility functions are equivalent in the limit of large network and sample size , i.e. as @xmath281 , see also fig . [ fig : bvf ] . let @xmath8 be a node in the network @xmath4 with degree @xmath15 . now , fix @xmath2 and suppose that we are calculating @xmath282 via the algorithm presented in sec . [ sec : evf ] . the @xmath283^th^ iteration of step 1 of this algorithm selects @xmath284 nodes uniformly at random to be active and results in @xmath285 nodes that can see the revolution . note that by the law of large numbers @xmath286 = \mathbb{e}\left[\frac{\mathcal{v}_1}{n}\right ] \mbox { as $ rep\rightarrow\infty$,}\end{aligned}\ ] ] where @xmath287 denotes convergence in probability . since the @xmath284 activated nodes are chosen uniformly at random , the statuses of the neighbours of @xmath1 are independent . therefore , the probability that @xmath1 can see the revolution is @xmath288 it follows that the expected fraction of nodes that have degree @xmath15 and can see the revolution is @xmath289,\ ] ] and hence , the expected fraction of nodes that can see the revolution is @xmath290 = \mathbb{e } \left[\sum_k\rho_k \mbox{bincdf}\left(k-{\lceil \theta k \rceil};k , 1-\frac{\mathcal{n}_l}{n}\right)\right].\ ] ] we conclude the proof by observing that by the continuous mapping theorem @xmath291 and @xmath292 implies @xmath290 \stackrel{p}{\rightarrow } \sum_k\rho_k \mbox{bincdf}\left(k-{\lceil \theta k \rceil};k , 1-r_j\right ) = v_b(r;\theta,\rho_k)\mbox { as } n\rightarrow\infty,\ ] ] and hence , @xmath293
knotted proteins provide an ideal ground for examining how amino acid interactions ( which are local ) can favor their folding into a native state of non - trivial topology ( which is a global property ) . some of the mechanisms that can aid knot formation are investigated here by comparing coarse - grained folding simulations of two enzymes that are structurally similar , and yet have natively knotted and unknotted states , respectively . in folding simulations that exclusively promote the formation of native contacts , neither protein forms knots . strikingly , when sequence - dependent non - native interactions between amino acids are introduced , one observes knotting events but only for the natively - knotted protein . the results support the importance of non - native interactions in favoring or disfavoring knotting events in the early stages of folding . a much debated problem in the thermodynamics and kinetics of protein folding @xcite is the process of knot formation in proteins @xcite . the topological entanglement found in naturally - occurring proteins presents several differences from that of compact or collapsed flexible polymers @xcite . firstly , the overall percentage of knotted native states in the protein data bank ( pdb ) is lower than for globular flexible chains with the same degree of polymerization @xcite . in addition , the free energy landscape of proteins , unlike that of homopolymers or random heteropolymers , is sufficiently smooth to ensure that the same knot in the same protein location is observed in folded structures @xcite . these distinctive features are arguably the consequence of concomitant effects , including the propensity of polypeptides to form secondary structure elements , which enhance their local order compared to a typical collapsed polymer chain @xcite , as well as functionally - oriented evolutionary mechanisms @xcite . to gain insight into such mechanisms as well as to highlight general physico - chemical mechanisms favoring knot formation , an increasing number of experimental and numerical studies of knotted proteins have been carried out @xcite . notably , the latest experimental results have clarified that native knots can form spontaneously and efficiently from an unknotted initial state . in particular , yeates and coworkers @xcite successfully designed a protein which can refold into the natively - knotted structure , albeit much more slowly than an unknotted counterpart . finally , the recent study of mallam and jackson showed that newly translated , knot free , yibk molecules fold spontaneously to the native trefoil - knotted state , and that chaperones can significantly speed up the folding process @xcite . numerical simulations have been a valuable complement to experiments , particularly regarding the characterization of the pathways leading to the self - tying of knotted proteins . in particular , folding simulations based on simplified protein representations and/or force fields , have indicated two main mechanisms leading to knot formation : the threading of one of the termini through a loop @xcite , or by slipping a pseudoknot through a loop @xcite . the two mechanisms are not mutually exclusive , as reported by sulkowska _ @xcite for protein mj0366 from _ methanocaldococcus jannaschii_. importantly , in coarse - grained ( c@xmath0-trace ) simulations where folding was promoted by exclusively favoring native contacts it was seen that the yield of folding trajectories was low and knots would form at late folding stages . specifically , for the deeply - knotted protein yibk , only 1 - 2% of the trajectories reached the native state and the native knot was formed when about @xmath1 of the native contacts had already been established . arguably , the coarse - grained nature of the models contributes to the observed low folding yield . however , even when employing atomistic representations , purely native - centric models tend to favor a rather late onset of knotting @xcite . for example , a relatively late stage of knot formation @xmath2 was also estimated from the analysis of the free - energy landscape of protein mj0366 , despite being half as long as yibk and with a shallower knot @xcite . these findings are aptly complemented by the early observation of wallin _ et al . _ @xcite that the low yield of purely native - centric models can be dramatically enhanced by promoting the attraction of specific protein regions that are not in contact in the native state . in particular , it was seen that the formation of the native knot in yibk was particularly facilitated by introducing an _ ad hoc _ non - native attraction between a loop and the c - terminus that , by threading the former , produced a sizeable population of knots even at early folding stages , @xmath3% @xcite . these results , together with recent experimental ones @xcite , pose the question of whether natively - knotted proteins can form non - trivial entanglements during early folding stages and whether non - native interactions can provide a general mechanisms for such self - tying events . clarifying such aspects is important to advance the understanding of some of the general mechanisms aiding knot formation . it must however , be borne in mind that there could co - exist independent pathways where knots are established at different stages of the folding process and that active mechanisms , such as interactions with chaperones , could be involved _ in vivo _ @xcite . in this study we address these questions by simulating the early folding process of two transcarbamylase proteins that are structurally very similar and yet their native states are differently knotted @xcite . specifically , one of them is a trefoil - knotted n - acetylornithine carbamoyltransferase ( aotcase ) , while the other is an unknotted ornithine carbamoyltransferase ( otcase ) , see fig . [ figure_1 ] . their evolutionary relatedness has posed the interesting question of understanding the source of their different native topology , with particular regards to the role of specific loop regions whose `` virtual '' excision or addition alters the native topology @xcite . here , for both the otcase and aotcase , more than a hundred folding simulations are carried out using a coarse - grained model and two different energy functions . we first evolve a fully extended configuration and steer it towards the native state by exclusively promoting native contact interactions . in a second set of simulations we enrich the energy function by adding quasi - chemical and electrostatic non - native interactions . in both cases , the stochastic evolution is followed up to the formation of about 35% of the native contacts and the knotted topology of the partially - folded structures is monitored . we find that knot occurrence is negligible for both the aotcase and otcase when the energy function favoring only native contacts is used . a dramatic difference in knotting propensity is instead seen when the quasi - chemical non - native interactions are added . in this case , the level of self - entanglement remains negligible for the otcase but is greatly enhanced for the aotcase , which is natively knotted . notably , all proper and improper knots formed in these early folding stages have the correct ( native ) trefoil topology and chirality . the analysis of the repeated knotting / unknotting events observed in the simulated trajectories indicates that knotting usually results from the threading of the c - terminal through loops present in the loose protein globule . the threading events are favored by the effective ( non - native ) attraction of the c - terminus to other protein regions . additional folding simulations carried out for `` in silico mutants '' of the aotcases , provide further support for the effect of the c - terminus chemical composition on the knotting propensities . the present investigation of knot formation in proteins that are only partially folded is focused on two transcarbamylases : aotcase , pdb entry 2g68 , and otcase , pdb entry 1pvv . the monomeric units of these two proteins have nearly the same length ( 332 and 313 amino acids , respectively ) and are evolutionarily related . this is established from their significant sequence homology ( same cath code , 3.40.50.1370 @xcite and sequence identity equal to @xmath440% ) . despite their statistically - significant structural alignability ( mistral p - value @xmath5 @xcite ) they have a different knotted topology @xcite , see fig . [ figure_1 ] . specifically , the otcase is unknotted , while the aotcase contains a right - handed trefoil knot . the knotted region , established using the algorithm of ref . @xcite , corresponds to the k172-g255 segment and is at a distance of about 80 amino acids from the c terminus , which is the nearest one in sequence . because the biological units of the two proteins have a different oligomeric state it is assumed that , consistently with other multimeric knotted proteins , the folding of the monomeric units precedes their assembly . we shall therefore limit considerations to the monomeric units . for the purpose of our study , both proteins are described with a simplified structural model , where each amino acid is represented by one interaction center , coinciding with the c@xmath0 atom . the folding dynamics is simulated using the stochastic monte carlo ( mc ) approach introduced and validated in ref . the mc evolution entails a brief relaxation from the initial fully - extended state using both local and non - local moves . after this stage , indicated with a shaded region in fig.[figure_2 ] and related ones , the mc dynamics proceeds exclusively through local moves . the moves amplitude is sufficiently small that no chain crossings occurs , as verified _ a posteriori . the stochastic scheme is analogous to the kink - jump dynamics @xcite which , by virtue of the local character of the moves , is known to provide a physically - viable description of biopolymers kinetics in thermal equilibrium @xcite . _ the folding simulations are carried out using two alternative energy functions . the first one promotes exclusively the formation of native contacts between pairs of amino acids . following refs . @xcite , the strength of the attractive interaction between any given native amino acid pair is derived from the strength of their hydrogen bonding in the native state and is expressed in thermal units , @xmath6 , at the nominal monte carlo temperature of @xmath7k . the second energy function includes electrostatic and non - native pairwise interactions in addition to the native - centric potential . the relative strength of the non - native interactions is set according to the quasi - chemical potentials of miyazawa and jernigan ( mj ) @xcite that reflect the statistical propensity of amino acid pairs to be in contact in proteins native states . following , again , ref.s @xcite , the average strength of the added non - native interactions is set to one tenth of the native one . no change to the mc temperature was made using this second potential because previous studies have shown that the addition of non - native interactions modifies the effective temperature of the system by less than @xmath8 @xcite . consistently with this fact we have verified that the native states of both aotcase and otcase remain stable when evolved with the mc scheme using the second type of potential . in fact , the asymptotic value for average fraction of native contacts was @xmath9% . the fact that this value is smaller than the asymptotic one , @xmath10% , of the first type of potential indicates that the strength non - native interactions while weak enough to maintain stable the native state , can nevertheless compete with the native attractive interactions . the addition of the quasi - chemical interactions are therefore expected to be capable of accounting for sequence - specific non - native contact propensities in the partly - folded state . the sequence - dependent character of the non - native interaction differentiates the present approach from the early one of wallin _ et al . _ @xcite where the knot - promoting effect of non - native interactions was probed by systematically introducing attractive interactions between various pairs of segments of yibk . for each of the two proteins and for each of the two energy functions , we generated @xmath4150 mc trajectories each consisting of @xmath11 mc moves per amino acid ( corresponding to 3000 units of mc time reported in fig . [ figure_2 ] and fig . [ figure_3 ] ) , for a total of 75,000 cpu hours . the acceptance rate of the local mc moves for both proteins was nearly constant throughout the simulation and approximately equal to 50% . the entanglement of the structures sampled during the mc evolution was established using a combination of two unrelated knot detection schemes , which are described in the materials and methods . this choice was made to maximize the robustness of the criterion used to establish the knotted state of an open chain - which is mathematically properly defined only after the chain termini are joined by a segment or arc thus giving a circular chain ( chain closure operation ) . as described in the materials and methods section we used many alternative closures for each chain and adopted a majority rule to single out chains that have non trivial self - entanglements . these chains can correspond to proper , fully developed knots ( fully accommodated in the knotted structures ) as well as improper ones ( partly accommodated in the closing arcs ) . in order to illustrate how the folding process differs for the two potential energy functions , we show in fig . [ figure_2 ] the evolution of the gyration radius ( upper panel ) and the fraction of native contacts ( lower panel ) of the knotted aotcase , obtained by employing these two potentials . it is seen that the effect of the non - native and electrostatic interactions results in the higher compactness of the protein globule , since the gyration radius in this case lies systematically below the one calculated with the purely native - centric model . at any given stage of the mc evolution , the fraction of established native contacts is systematically lower for non - native interactions thus indicating that the latter introduce frustration that slows down the folding dynamics . analogous results hold for the unknotted otcase , see top panel of fig . [ figure_3 ] . the data in fig . [ figure_3 ] represent how the average radius of gyration and fraction of formed native contacts evolve in the course of the folding simulations of the two transcarbamylases . it is seen that , over the duration of the simulation , the average fraction of formed native contacts grows to @xmath12% on average ( the maximum value in all trajectories was @xmath13% . overall , the probability of formation of native contacts decreases with the sequence separation of the amino acid pair , so that native @xmath14-helical contacts are substantially more probable than inter - strand ones @xcite . from the steady increase of the fraction of native contacts , @xmath15 , it is extrapolated that folding completion would occur on time - scales at least ten times larger than considered here . from the same figure it emerges that the radius of gyration gradually decreases to about 25 @xmath16 which is 30% larger than the native one , consistently with the less compact character of the partially folded states . to characterize the overall propensity of the two proteins to form knots while still largely unfolded , we extracted configurations at regular intervals of the simulations and analyzed their topological state . the knotting propensity of aotcase , which is natively knotted , is illustrated in the upper panel of fig . [ figure_4 ] which portrays the fraction of configurations that are properly or improperly knotted during a given time - window of the mc dynamics . it is seen that when the purely native - centric potential is used , the knotting propensity is always negligible . however , when non - native interactions are added , the number of knotted configurations increases to a definite fraction of the total : at the end of the simulations , when only 35% of the native contacts are formed , the fraction of conformations that are properly or improperly knotted is about 1% . notice that this value is comparable with the yield of purely native - centric coarse - grained models where pathways with a late - stage formation of knots are typically observed @xcite . within the limitations of coarse - grained approaches , the results indicate that sequence - dependent non - native interactions can produce a detectable fraction of configurations with the correct native entanglement already at early folding stages . the analysis was repeated for the natively - unknotted otcase and the results are shown in the lower panel of fig . [ figure_4 ] . the contrast with the case of the atocase is striking . in fact , the knotting probability is negligible not only for the purely - native case , but remains so even when non - native interactions are introduced . the results indicate that the sequence - dependent non - native interactions promoted by the quasi - chemical potential increase dramatically the incidence of knots in the partially folded state of the aotcase compared to the otcase . equivalently , the model calculations indicate that the two related natively - knotted and unknotted proteins have different knotting propensities already in the partially folded state , and that this difference can be ascribed to sequence - specific non - native interactions . to further elucidate the mechanisms responsible for these differences we monitored several parameters in the course of the simulations of the two transcarbamylases . more specifically , we searched for systematic differences in the interactions that the termini of these proteins establish with other parts of the peptide chain and for preferential locations of fully - developed knots . in this respect , it is important to point out that , in the course of the simulations for the aotcase with non - native interactions , the process of knot formation is not irreversible . as illustrated in fig . [ figure_5 ] , trefoil knots can be formed and untied during each simulation . in particular , fully - developed knots can persists for up to one tenth of the duration of our simulations . notably , only knots with the correct ( i.e. native ) chirality are observed . by analysing the configurations preceding and following the knotting / unknotting events it is found that the change in topological state is typically caused by the threading of the helical c - terminus through loops formed by various protein regions . among the independent knotting events we identified six for which the stochastic closure returned a non - trivial topology in at least 80% of the cases or the portion accommodating the knot had a depth of at least 20 amino acids . the analysis of the loop threading events leading to such persistent fully - developed knots showed that they involved the c terminus interacting with amino acids 55 - 65 , 90 - 110 , or 125 - 155 . only in one case it was observed the threading of the n - terminal through a loop involving segments 252 - 277 . representative configurations are shown in the left panel of fig . [ figure_6 ] where it is clear that the helical c - terminus of the aotcase typically points towards the rest of the protein , and the stiffness of the helix facilitates the threading of various loop regions . at variance with the above situation , for both potential energy functions the otcase c - terminus is typically exposed to the solvent and pointing away from the rest of the protein , as in the example shown in the right panel of fig . [ figure_6 ] . the terminus also shows a lower propensity to form @xmath14-helices . an analogous situation is found for the aotcase in the presence of only native interactions . the effect is quantitatively illustrated in fig . [ figure_7 ] , which portrays the average strength of the non - native interaction potential energy between the amino acids in the c - terminal @xmath14-helix of the aotcase or the otcase and all the other residues in the chain . it is seen that the average non - native attraction is consistently stronger in the case of the knotted protein by about 15% , which is a sizable amount considering that the average is taken over all sampled structures , irrespective of their compactness and knotted topology . the same figure also illustrates that , if the amino acids in the c - terminal @xmath14-helix of protein aotcase are replaced by neutral hydrophilic residues ( glu , gln , asn ) , then the average non - native interaction energy becomes slightly repulsive , explaining the tendency of the @xmath14-helix to point outwards . this provides an additional argument in support of the picture where the hydrophobicity character of the residues plays a role in the folding of knotted proteins @xcite . finally , in fig . [ figure_8 ] we plot the average electrostatic potential energy between the @xmath14-helix and the rest of the chain and show that , overall , the electrostatic interaction is about two orders of magnitude smaller than that of the quasi - chemical potential . to further clarify the role of the c - terminal @xmath14-helix in the knotting of the partially - folded states of the aotcase , we have carried out another set of simulations on a mutant protein . the last 25 residues of the otcase ( which are involved in the c - terminal @xmath14-helix ) were substituted with the same number of residues that form the c - terminal @xmath14-helix in the knotted aotcase . apart from such a substitution , all other attributes , namely the length of the chain and the native contact map , were identical to the unknotted protein . the outcome of the simulation is summarized in fig . [ figure_9 ] , from which we can conclude that the mutant protein forms a substantial fraction of knotted configurations in the presence of non - native interactions in its early stage of the folding . a simplified coarse - grained protein model was used for a comparative study of early folding stages of two evolutionarily related transcarbamylase proteins : an aotcase ( pdb i d 2g68 ) and an otcase ( pdb i d 1pvv ) . the two proteins are well - alignable in sequence and structure and yet possess differently knotted native states : aotcase is trefoil - knotted while otcase is unknotted . the role of sequence - dependent non - native interactions in promoting the correct native topology in the early folding stages was investigated by using two different energy functions in the simulations : one with a purely native - centric potential energy ( favoring only native contact interaction ) and one with the additional contribution of quasi - chemical and electrostatic non - native interactions . we found that in the absence of quasi - chemical interactions , neither protein shows an appreciable propensity to self - entangle in the early folding stages . however , once non - native interactions are introduced , the natively - knotted aotcase does show a strongly enhanced propensity to form proper and improper knots with the correct native topology and chirality . by contrast , the knot enhancement effect is completely absent in the natively - unknotted otcase . inspection of the ensemble of folding trajectories of the two proteins suggest that knotting in the partly unfolded aotcase mostly results from the approach of the hydrophobic c - terminal @xmath14-helix to other regions of the protein which are eventually threaded through . because of the non - compact character of the partly - unfolded structures , the @xmath14-helix in the c - terminal can also retract from the threaded regions , so that various events of formation / disruption of proper and improper knots can be observed in a given trajectory . the influence of the c - terminal region on the different knotting propensity of the two carbamylases is further supported by the folding simulations for an `` in silico mutant '' of the otcase obtained by replacing the c - terminal sequence with the one of the knotted aotcase . in fact , the folding simulations based on the quasi - chemical non - native interactions yielded a portion of knotted structures similar to the natively - knotted aotcase , from which the c terminus was taken . we recall that , the seminal study of wallin _ et al . _ @xcite , which was based on a coarse - grained model of yibk with _ ad hoc _ non - native interactions , had suggested the relevance of non - native interactions for steering the early formation of the native knotted topology ( by the threading of a loop by the c - terminal ) . our results , which are consistent with these conclusions , provide additional elements to the picture by taking advantage of a coarse - grained framework where the folding of two carbamylases with similar structures , and yet different knotted topology are compared on equal footing . in particular our results indicate that non - native interaction propensities that are encoded in the primary sequence , can favor or disfavor the formation of knots already at early folding stages . the necessarily limited scope of the coarse - grained approach used here does not clarify whether additional pathways leading to a late - stage formation of knots can be present in transcarbamylases . it would be most interesting to address this standing issue in future studies within the present , or alternative comparative schemes . we have adopted the coarse - grained model developed in ref.s @xcite , in which the effective degrees of freedom are the amino acid residues , represented by a spherical bead located at the position of the corresponding c@xmath17 atom . the potential energy of the model consists of bonded and non - bonded terms . the bonded part of the of the potential consists of stretching potentials of the pseudo - bond and pseudo - angle , as well as potential for pseudo - torsions : @xmath18 @xmath19 the bond - stretching potential has the form @xmath20 where @xmath21 is the distance between the residues @xmath22 and @xmath23 , while @xmath24 is the equilibrium length of the c@xmath17-c@xmath17 pseudo - bond . @xmath19 the double - well pseudo - angle potential is given by @xmath25 , \label{equation_angle}\ ] ] where @xmath26 is the pseudo - angle formed by the residues @xmath22 , @xmath23 and @xmath27 , while @xmath28 and @xmath29 are the equilibrium values of the helical and the extended pseudo - angles , respectively . @xmath19 the torsion - angle potential for pseudo - torsions is of the form @xmath30 v_\text{n } , \label{equation_torsion}\ ] ] with @xmath31 being dihedral angle between the planes ( @xmath22,@xmath23,@xmath27 ) and ( @xmath23,@xmath27,@xmath32 ) . the constants @xmath33 and @xmath34 are depending only on the type of the middle two residues @xmath23 and @xmath27 and are adopted from karanicolas and brooks @xcite . all the values of the spring and energy constants appearing in the expressions of the bonded energy terms in eqs . ( [ equation_bond ] ) , ( [ equation_angle ] ) and ( [ equation_torsion ] ) can be found in ref . @xcite . the non - bonded part of the potential consists of three terms , comprising native , non - native and electrostatic interactions , respectively : @xmath35 @xmath19 @xmath36 denotes the g@xmath37-type potential developed by karanicolas and brooks @xcite . within such an approach , the native contact map is defined on the basis of the network of the hydrogen bonds in the native state , as well as on the degree of the proximity of the backbone atoms side - chains . namely , two residues are defined to be in the native contact if the hydrogen bond between them is stronger than -0.5 kcal / mol , or if any of their non - hydrogen side - chain atoms are within the distance of 4.5 @xmath16 , in the native state . we do not consider contacts between residues with a distance in sequence smaller than 3 amino - acids . these interactions are described by the following functional form : @xmath38,\ ] ] where @xmath39 is the native - state separation of residues @xmath22 and @xmath23 . the strength of the interaction @xmath40 is chosen to be that of their hydrogen bond , for the residues that are hydrogen - bonded in the native state , while for the side - chain interacting residues it is a value proportional to the corresponding mj contact potential @xcite and it was suitably renormalized in order to match the hydrogen bond native contact energy scale . the residues in @xmath41-sheets or hairpins are often in contact via multiple hydrogen bonds , so in order to stabilize these structures within the g@xmath37-model , the additional network of weaker hydrogen bonds ( having the strength @xmath42 ) involving the neighboring pairs @xmath43 , @xmath44 , @xmath45 and @xmath46 was introduced around each residue pair @xmath47 that was found to interact via either two hydrogen bonds or a hydrogen bond and a side - chain contact . @xmath19 depending on the amino acid type of the residues @xmath22 and @xmath23 , the non - native interactions in the model developed by kim and hummer in ref . @xcite can be both attractive and repulsive . repulsive interactions are applied between amino acid pairs that interact less favorably with each other than with the solvent and _ vice versa_. for a pair of residues that experiences an effective attractive interaction with a strength @xmath48 , the non - native interaction potential is given by @xmath49\,,\ ] ] with @xmath50 , where @xmath51 and @xmath52 are van der waals radii of the residues @xmath22 and @xmath23 . for pairs of residues that effectively repel each other , so that @xmath53 , the non - native potential energy function takes the following form : @xmath54 + 2 \varepsilon_{ij } , & r_{ij } < \sqrt[6]2\,\sigma_{ij}\\ -4 \varepsilon_{ij } \bigg [ \big(\frac{\sigma_{ij}}{r_{ij}}\big)^{12}- \big(\frac{\sigma_{ij}}{r_{ij}}\big)^{6 } \bigg ] , & r_{ij } \geq \sqrt[6]2\,\sigma_{ij } \end{array } \label{veff } \right .\ ] ] the effective lennard - jones interaction strength @xmath40 between residues @xmath22 and @xmath23 in the coarse - grained model of kim and hummer @xcite is defined as @xmath55 the coefficients @xmath56 are negative and coincide with the entries of the mj matrix @xcite , while @xmath57 is an offset parameter . hence , eq . ( [ veff ] ) defines a statistical knowledge - based potential which measures the preference of residue - residue interactions relative to residue - solvent interactions . the parameter @xmath58 scales the strength of the lennard - jones interaction compared to the physical electrostatic interactions . the two free parameters @xmath58 and @xmath57 are fitted in order to correctly reproduce the binding affinity of the broad set of experimentally well - characterized protein complexes @xcite . as an illustration , the effective interaction strength between the neutral hydrophilic residues glu and gln is @xmath59 kcal / mol , so that effective interaction is repulsive , while for the hydrophobic residues ile and leu is @xmath60 kcal / mol , so that the effective interaction is attractive . @xmath19 @xmath61 is the long - range electrostatic interaction between residues @xmath22 and @xmath23 and it is modeled by debye - h@xmath62ckel - type of the potential @xmath63 } { r_{ij}}\,,\ ] ] where @xmath64 and @xmath65 are the electrostatic charges of residues @xmath22 and @xmath23 , @xmath66 is the debye screening length , @xmath67 is dielectric vacuum constant and @xmath68 is the relative dielectric constant of water in near - ambient conditions . the non - native interactions in the force field of our coarse - grained model introduce frustration and make the potential energy surface quite rugged . as a consequence , even within such a simple model , molecular dynamics ( md ) simulations of the folding for chains of several hundreds amino acids are very expensive . within the simulation time intervals which were accessible to our computer resources , such md trajectories did not allow to monitor significant changes in the chain conformations . to cope with this problem , we have simulated mc dynamics , using an algorithm which combines different types of moves , namely : @xmath19 local crankshaft moves @xcite , that consist of the rotation of a randomly selected single bead around the axis defined by its nearest neighbors . the angle of the rotation was randomly selected in the interval @xmath69 , + @xmath19 local end - point moves , in which the last 10 residues on both terminals are rotated rigidly with respect to the rest of the chain by up to @xmath70 around a random axis passing through the most interior bead of the end segment . @xmath19 local cartesian moves , that involve the displacement of the coordinates of a single randomly selected bead in the chain , within a sphere of radius @xmath71 , + @xmath19 global pivot moves @xcite , where one amino acid is picked at random and the chain portion involving all amino acids with smaller ( or alternatively larger ) sequence index are rotated by up to @xmath70 around a random axis passing through the picked amino acid . the moves were accepted or rejected according to the standard metropolis criterion . mc algorithms based on local crankshaft and end moves are commonly employed in the polymer physics @xcite to study _ dynamic properties _ , since it is was shown that they can mimic the intrinsic dynamics of a polymer in solution @xcite at a much lower computational cost of md simulations @xcite . however , we emphasize that since our computational scheme is based on the mc evolution , we can not establish a quantitative mapping between the simulation time and physical time . we have generated 150 independent mc trajectories for each system , starting from the same stretched coil configuration . in the early stage of the folding , these types of moves were attempted with the constant probabilities of 0.8 , 0.1 and 0.1 , respectively . such an algorithm generates a rapid collapse of the chain from a fully stretched configuration to one in which the gyration radius is reduced from about 70 @xmath16 to about 30 @xmath16 in about @xmath72 mc steps per particle , that corresponds to 100 units of mc time ( see fig . [ figure_3 ] ) . during such collapse , no knot was observed . at the end of this stage , the acceptance rate of the global pivot moves typically dropped below 1% . after this happened , we switched off the pivot moves , except from those involving few residues near the the two terminal of the chain . from this point on , the conformational changes of the chain are driven by the local crank - shaft and cartesian moves with an overall acceptance of about 50% . when computing the fraction native contacts @xmath15 along the calculated mc trajectories , we adopt a criterion according to which two residues with index difference @xmath73 are said to be in contact if the distance of their @xmath74 is less than @xmath75 . the conformations visited during the monte carlo dynamics were topologically classified by computing the alexander determinants after suitable closure into a ring @xcite . for robustness , two alternative closure schemes are used : the minimally - interfering closure @xcite and a modified version of the stochastic one @xcite . the minimally - interfering closure is used first because it is very computationally effective , in that it entails a single , optimally chosen closure . in case of positive knot detection we further validate the non - trivial entanglement by performing 100 closures where each terminus is prolonged far out of the protein along a stochastically chosen direction , and the end of the prolonged segments are closed by an arc ( that does not intersect the protein ) . the stochastic exit directions are picked uniformly among those that are not back - turning . specifically , they must form an angle of more than 90@xmath76 with the oriented segment going from each terminus to the @xmath74 at a sequence distance of 10 . if the majority of the stochastic closures return non - trivial alexander determinants , than the conformation is non - trivially entangled . such conformations can correspond to both proper , fully developed knots , and improper ones . the two can be distinguished using knot localization criteria @xcite : proper knots are entirely accommodated within the original protein chain , while improper ones span the exit segments . the knot type was determined using the scheme of ref.@xcite , which is based on the knotfind algorithm . we thank p. carloni , a.v . finkelstein and l. tubiana for valuable discussions and suggestions . pf is also a member of the interdisciplinary laboratory for computational science ( lisc ) , a joint venture between university of trento and fondazione bruno kessler . t is supported by the _ provincia autonoma di trento _ , through the aurorascience project . simulations were performed on the aurora supercomputer at the lisc . 99 finkelstein av , ptitsyn ob ( 2002 ) , protein physics : a course of lectures , academic press . creighton te ( 1992 ) , proteins : structures and molecular properties , w. h. freeman . mansfield ml ( 1994 ) are there knots in proteins ? nature struct biol 1 : 213 - 214 . mansfield ml ( 1997 ) fit to be tied . nature struct biol 4 : 166 - 167 . taylor wr ( 2000 ) a deeply knotted protein structure and how it might fold . nature 406 : 916 - 919 . nureki o , shirouzu m , hashimoto k , ishitani r , terada t _ ( 2002 ) an enzyme with a deep trefoil knot for the active - site architecture . acta crystallog d 58 : 1129 - 1137 . michel g , sauve v , larocque r , li y , matte a , cygler m ( 2002 ) the structure of the rlmb 23s rrna methyltransferase reveals a new methyltransferase fold with a unique knot . structure 10 : 1303 - 1315 . taylor wr , lin k ( 2003 ) a tangled problem . nature 421 : 25 . lua rc , grosberg ay ( 2006 ) statistics of knots , geometry of conformations , and evolution of proteins . plos comput biol 2 : doi : 10.1371/journal.pcbi.0020045 . delbruck m ( 1962 ) knotting problems in biology . mathematical problems in the biological sciences , edited by r e bellman , proceedings of symposia in applied mathematics , american mathematical society , providence , rhode island 14 : 55 - 63 . frish hl , wasserman e ( 1961 ) chemical topology . journal of american physical society 83 : 3789 - 3795 . vologodskii av , lukashin av , frank - kamenetskii md , anshelevich vv ( 1974 ) the knot problem in statistical mechanics of polymer chains . sov phys - jetp 39 : 1059 - 1063 . sumners dw , whittington sg ( 1988 ) knots in self - avoiding walks . journal of physics a - mathematical and general 21 : 1689 - 1694 . janse van rensburg ej , sumners dw , wasserman e , whittington sg ( 1992 ) entanglement complexity of self - avoiding walks . journal of physics a - mathematical and general 25 : 6557 - 6566 . micheletti c , marenduzzo d , orlandini e , sumners dw ( 2006 ) knotting of random ring polymers in confined spaces . j chem phys 124 : 064903 . michels jpg , wiegel fw ( 1986 ) on the topology of a polymer ring . proc r soc lond a403 : 269 - 284 . micheletti c , marenduzzo d , orlandini e ( 2011 ) polymers with spatial or topological constraints : theoretical and computational results . phys rep 504 : 1 . marenduzzo d , orlandini e , stasiak a , sumners dw , tubiana l and micheletti c ( 2009 ) dna dna interactions in bacteriophage capsids are responsible for the observed dna knotting proc natl acad sci usa 106 : 22269 - 22274 . taylor wr ( 2007 ) protein knots and fold complexity : some new twists . comput biol chem 31 : 151 - 162 . mallam al , jackson se ( 2012 ) knot formation in newly translated proteins is spontaneous and accelerated by chaperonins . nat chem biol 8 : 147 - 53 . king np , jacobitz aw , sawaya mr , goldschmidt l , yeates to ( 2008 ) structure and folding of a designed knotted protein . proc natl acad sci u s a 107 : 20732 - 7 . yeates to , norcross ts , king np ( 2007 ) knotted and topologically complex proteins as models for studying folding and stability . current opinion in chemical biology 11 : 595 - 603 . mallam al ( 2009 ) how does a knotted protein fold ? febs j 276 : 365 - 375 . virnau p , mallam a , jackson s ( 2011 ) structures and folding pathways of topologically knotted proteins . j phys : condens matter 23 : 33101 - 33118 . noel jk , sulkowska ji , onuchic j ( 2010 ) slipknotting upon native - like loop formation in a trefoil knot protein . proc natl acad sci usa 107 : 15403 - 15408 . virnau p , mirny la , kardar m ( 2006 ) intricate knots in proteins : function and evolution . plos comput biol 2 : doi : 10.1371/journal.pcbi.0020122 . potestio r , micheletti c , orland h ( 2010 ) knotted vs. unknotted proteins : evidence of knot - promoting loops . plos comput biol 6 : doi : 10.1371/journal.pcbi.1000864 . shakhnovich e ( 2011 ) protein folding : to knot or not to knot ? nat mater 10 : 84 - 6 . mallam al , jackson se ( 2005 ) folding studies on a knotted protein . j mol biol 346 : 1409 - 1421 . mallam al , jackson se ( 2006 ) probing nature s knots : the folding pathway of a knotted homodimeric protein . j mol biol 359 : 1420 - 1436 . mallam al , jackson se ( 2006 ) a comparison of the folding of two knotted proteins : ybea and yibk . j mol biol 366 : 650 - 665 . mallam al , jackson se ( 2008 ) knotted fusion proteins reveal unexpected possibilities in protein folding . molecular cell 30 : 642 - 648 . mallam al , morris er , jackson se ( 2009 ) exploring knotting mechanisms in protein folding . proc natl acad sci usa 105 : 18740 - 18745 . mallam al , rogers jm , jackson se ( 2010 ) experimental detection of knotted conformations in denaturated proteins . proc natl acad sci usa 107 : 8189 - 8194 . wallin s , zeldovich kb , shakhnovich ei ( 2007 ) the folding mechanics of a knotted protein . j mol biol 368 : 884 - 893 . sulkowska ji , sulkowski p , onuchic j ( 2008 ) dodging the crisis of folding proteins with knots . proc natl acad sci usa 106 : 3119 - 3124 . faisca pfn , travasso rdm , charters t , nunes a , cieplak m ( 2010 ) the folding of knotted proteins : insights from lattice simulations . phys biol 7 : 16009 - 16020 . orengo ca , michie ad , jones s , jones dt , swindells mb , et al . ( 1997 ) cath - a hierarchical classification of protein domain structures . structure 5 : 1093 - 1109 . micheletti c , orland h ( 2009 ) mistral : a tool for energy - based multiple structural alignment of proteins . bioinformatics 25 : 2663 - 2669 . kolesov g , virnau p , kardar m , mirny la ( 2007 ) protein knot server : detection of knots in their structures . nucleic acids res 35 : w425-w428 ( doi : 10.1093/nar / gkm312 ) . quake sr ( 1995 ) fast monte carlo algorithms for knotted polymers . phys rev e 52 : 1176 - 1180 . shuler ke , orwoll ra , stockmayer wh . stochastic models for chain dynamics , in advances in chemical physics : stochastic processes in chemical physics , volume 15 ( ed k. e. shuler ) , john wiley & sons , inc . , hoboken , nj , usa . de pablo jj , escobedo fa ( 2007 ) monte carlo methods for polymeric systems , in advances in chemical physics : monte carlo methods in chemical physics , volume 105 ( eds i. prigogine and s. a. rice ) , john wiley & sons , inc . , hoboken , nj , usa . best rb , hummer g ( 2010 ) coordinate - dependent diffusion in protein folding . proc natl acad sci usa 107:1088 - 1093 . kim yc , hummer g ( 2008 ) coarse - grained models for simulations of multi - protein complexes : application to ubiquitin binding . j mol biol 375 : 1416 - 1433 . miyazawa s , jernigan rl ( 1996 ) residue - residue potentials with a favorable contact pair term and an unfavorable high packing density term for simulation and threading . j mol biol 256 : 623 - 644 . a beccara s , krbi t , covino r , and faccioli p ( 2012 ) dominant folding pathways of a ww domain . proc natl acad sci usa 109 : 2330 - 2335 . ivankov dn , garbuzynskiy so , alm e , plaxco kw , baker d , finkelstein av ( 2003 ) contact order revisited : influence of protein size on the folding rate . protein sci 12 : 2057 - 2062 . karanicolas j , brooks cl ( 2002 ) the origins of the asymmetry in the folding transition states of protein l and protein g. protein sci 11 : 2351 - 2361 . kremer k , binder k ( 1988 ) monte carlo simulation of lattice models for macromolecules . comput phys rep 7 : 259 - 310 . madras n , sokal ae ( 1988 ) the pivot algorithm : a highly efficient monte carlo method for the self - avoiding walk . j stat phys 50 : 109 - 186 . jorgensen wl , tirado - rives j ( 1996 ) monte carlo vs molecular dynamics for conformational sampling . j phys chem 100 : 14508 - 14513 . tubiana l , orlandini e , micheletti c ( 2011 ) profiling the arc entanglement of compact ring polymers : a comparison of different arc - closure schemes with applications to knot localization . prog theor phys 191 : 192 - 204 . k. millett , a. dobay and a. stasiak ( 2005 ) linear random knots and their scaling behaviour . macromolecules 38 : 601 - 606 . tubiana l , orlandini e , micheletti c ( 2011 ) multiscale entanglement in ring polymers under spherical confinement . phys rev lett 107 : 188302 . traces of aotcase ; the n and c termini are colored in red and blue , respectively . bottom panel : monte carlo evolution of the fraction of native contacts of aotcase . the overall fraction is shown in red , while the the fraction of formed native contacts involved in @xmath14-helices and @xmath41-sheets are shown in green and blue , respectively.,width=453 ] -helix and the rest of the chain , in the knotted aotcase and unknotted otcase . * the upper curve shows the same quantity for a mutant of the natively knotted protein in which all residues in the c - terminal @xmath14-helix are replaced by hydrophilic residues.,width=453 ]
cognitive radio ( cr ) is a promising approach to efficiently utilize the scarce rf spectrum resource . in this paradigm , knowledge about spectrum occupancy in time , frequency , and space that is both accurate and timely is crucial in allowing cr networks to opportunistically use the spectrum , and avoid interference to a primary user ( pu ) . in particular , information about pu location will enable several key capabilities in cr networks including improved spatio - temporal sensing , intelligent location - aware routing , as well as aiding spectrum policy enforcement . the localization problem in cr networks is in general different from localization in other applications such as wireless sensor networks ( wsn ) and global positioning system ( gps ) , in which the target to be localized cooperates with the localization devices . in contrast , a pu does not communicate directly with the crs during the localization process . this scenario is referred to as non - cooperative localization and does not allow the use of conventional techniques such as time - of - arrival ( toa ) and time - delay - of - arrival ( tdoa ) ranging . several techniques have been proposed in the literature for estimating the position of active pus in a cr network . these schemes include both range - based @xcite and range - free @xcite algorithms . range - free positioning schemes , such as centroid localization @xcite , have attracted a lot of interest because of their simplicity and robustness to changes in wireless propagation properties such as path loss . this characteristic makes them suitable candidates for systems requiring coarse - grained , but reliable and cost - effective techniques . on the other hand , range - based techniques such as multi - lateration @xcite , offer better estimates , but require accurate information about the target s distance that can only be achieved either through cooperation with the target or a very precise knowledge of the path loss model . in this paper , we focus on analyzing the performance of a particular , low complexity , coarse - grained localization algorithm , referred to as weighted centroid localization ( wcl ) @xcite . in this technique , pu location is approximated as the weighted average of all secondary user positions within its transmission range . strictly speaking , wcl is not an entirely range - free technique because it requires additional information aside from simple connectivity , namely received signal strength ( rss ) measurements . wcl and wcl - like methods are suitable for pu localization because of their low computational complexity . they also do not require cooperation from the pu and only use the readily available rss information . in addition , wcl always converges to a solution , as opposed to some range - based techniques , and do not require estimation of the path loss exponent beforehand . recent papers on wcl have proposed improvements on the basic algorithm to enhance its performance for specific scenarios @xcite . all previous wcl studies reported in the literature have been analyzed either via simulation or field experiments @xcite . to the best of our knowledge , there has been no analysis of the probability distribution of the localization error for even the simplest version of wcl in the presence of shadowing in both wsn and cr literature . we present the first theoretical analysis of the error distribution of wcl in this paper . the presented analytical framework models varying levels of shadowing , including both independent and correlated shadowing environments , and also takes inaccuracy of sensor node positioning into consideration . using this analysis in conjunction with numerical simulations we observe and quantify the robustness of wcl to variable channel conditions and node placements . we also investigate variable node participation as a simple modification that can improve wcl performance . in the latter part of this paper we propose and investigate a distributed cluster - based implementation of wcl in the context of cr networks . previous work primarily assumes a centralized fusion node aggregating data from other nodes . in the context of wsn where wcl was first utilized @xcite , the primary motivation was node self - localization where a node applies wcl using measurements from beacon nodes in close proximity to it . this approach naturally dictates a centralized system where the node to be localized acts as the fusion center and only requires minimal interaction with the beacons . in the case of pu localization in cr networks , however , where the primary motivation is estimating the position of a non - cooperative node , there no longer exists any particular node that has all the requisite information to perform wcl . in this case a fusion node is arbitrarily chosen . this approach does not scale very well in terms of communication overhead and transmit power , which limits its application to low node densities or small network size . there is therefore a need to create distributed implementations of wcl for this technique to be practically applied in cr localization . a standard approach for distributing control in sensor networks is clustering . several algorithms exist in the literature for creating either disjoint @xcite or overlapping @xcite clusters . in @xcite , a distributed technique based on the concept of mean - shift @xcite . however , the technique requires repeatedly forming a virtual cluster until it converges to the target location . this introduces additional communication overhead and also requires several iterations to converge . we present an implementation which saves total transmit power needed for localization process communication , and investigate its accuracy as well as its effect on per node computational complexity and total transmit power consumption . the rest of the paper is organized as follows . the system model is introduced in section [ sec : systemmodel ] . the theoretical analysis of centralized wcl accuracy is presented in section [ sec : analysis ] , while a practical and distributed implementation of wcl is proposed and analyzed in section [ sec : distributed ] . numerical results for various scenarios used to evaluate performance of wcl algorithms are presented in section [ sec : evaluation ] . finally , the paper is concluded in section [ sec : conclusion ] . this section presents the general framework for our analysis . we first review the assumptions about the system setting and propagation model , and then present the basic wcl algorithm . we adopt the pu - centric model @xcite in which the pu transmitter is assumed to be at the center area of a circle . the rationale behind this choice is that in a practical scenario , sensor nodes that obtain enough received power ( i.e. above sensitivity threshold ) approximately form a circle around the primary transmitter . thus , sensors within this circle are used to jointly estimate the position of the pu . an implicit assumption in this model is that the pu is surrounded by enough nodes so that edge effects can be disregarded . assume @xmath0 sensor nodes in a circle of fixed radius @xmath1 . the number of sensors within the circle is jointly affected by several factors , such as transmit power of the pu ( which effectively determines @xmath1 ) , node density , node selection methodology , etc . @xmath0 is a design parameter for wcl deployment , the choice of which will have significant influence on the localization accuracy , as will be shown in our analysis . the pu located at the center has a coordinate defined as @xmath2^t=[0 , 0]^t$ ] . for the @xmath3th sensor node , we define its 2-dimensional location as @xmath4^t$ ] where @xmath5 . sensors are placed independently in both dimensions . we assume sensors obtain information about their own locations , which is modeled as imperfect . the measured location of the @xmath3th sensor is distributed as @xmath6 , where @xmath7 defines standard deviation of the error in sensor node self - localization . we further assume the self - localization errors are independent among sensors and are independent from received powers . we adopt the following channel model for received signal strength used in wcl . the received power of the @xmath3th node from the primary transmitter , @xmath8 , is given by @xmath9 in which the first two terms characterize path loss and the last term describes shadowing effect . the path loss is characterized by three constants : reference power @xmath10 , reference distance @xmath11 and path loss exponent @xmath12 . denote collection of shadowing variables as @xmath13 $ ] , then its distribution is characterized by @xmath14 . we consider the following shadowing cases _ = _ s^2 * i_n , & i.i.d case , + \{_}_ij = _ i-_j /x_c , & correlated case , [ shadowing ] * where @xmath15 is the correlation distance @xcite within which the shadowing effects of nodes are correlated , and @xmath16 is the identity matrix with dimension @xmath0 . the theoretical analysis presented in section [ sec : analysis ] is based on the centralized architecture , which will also give us an upper bound on the performance . in each localization period , all sensors involved in the wcl will send their rss measurements to the fusion center , where an estimate will be made about the coordinate of the primary transmitter . the considered algorithm is a modification of relative - span wcl proposed in @xcite . in this scheme , the weights are guaranteed to be non - negative in db scale . the 2-dimensional estimated location of pu is formed as @xmath17 the weight factor of node @xmath3 is defined as @xmath18 , @xmath19 where @xmath20 is a constant calculated from average received power of the node on the border with some margin , to guarantee the probability that receive power of all nodes in the area below @xmath20 is sufficiently small , say 1% . @xmath21 is the maximum received power among nodes , and @xmath22 is the span of received power . the localization error is then given by @xmath23^t$ ] , where @xmath24 and @xmath25 are the 1-dimensional errors along the x and y - axis respectively . finally , the performance of the wcl scheme is evaluated by distance error , which is given by @xmath26 this section presents the analysis of wcl error using probability theory , ultimately leading to analytical expressions for the error distribution , first of wcl applied to 1-dimensional coordinates , and finally the 2-dimensional wcl error given by ( [ error - dist ] ) . first we derive the 1-dimensional localization error for the i.i.d . shadowing case . we define a constant @xmath27 for each node @xmath3 as @xmath28 one can verify that @xmath29 . the location estimation in one dimension , say , x - axis , is naturally given by @xmath30 based on our i.i.d assumption among all @xmath8 s , one can verify that @xmath31 for the statistical property of @xmath32 , since @xmath33 s are also i.i.d and are independent of @xmath8 s , the mean of @xmath32 is given by @xmath34 . the variance of the @xmath3th term of @xmath32 is given by rcl [ variance a ] [ ( p_i - p _ ) x_i ] & = & [ ( p_i - p_)^2 ] [ x_i^2 ] - [ p_i - p_]^2 [ x_i]^2 + & = & _ l^2 _ i^2 + _ s^2 _ i^2 + _ l^2 _ s^2 . thus the variance of @xmath32 is derived as @xmath35 . from central limit theorem , we approximate @xmath32 as a single gaussian variable , i.e. @xmath36 , which is verified by numerical simulations that are not presented here due to space limit . since @xmath32 and @xmath37 are correlated due to summation over the same set of variables , i.e. @xmath8 s , @xmath24 is a ratio of two correlated gaussian variables . in order to obtain statistical properties of @xmath24 , we first calculate correlation coefficient @xmath38 of @xmath32 and @xmath37 , which is given by @xmath39/(\sigma_a \sigma_b ) = \operatorname{\mathbb{e}}\left [ \sum_{i=1}^{n } \sum_{j=1}^{n } x_i s_i s_j \right]/(\sigma_a \sigma_b ) . \label{correlation - coeff1}\ ] ] note that = _ i _ s^2 , & @xmath40 , + 0 , & @xmath41 , [ correlation ] thus the correlation coefficient is obtained as @xmath42 where @xmath43^t$ ] , and @xmath44 is the @xmath45th norm of the vector . from the previous analysis we know that @xmath24 is the ratio of two dependent gaussian random variables with non - zero means for which the closed - form pdf is available in ( * ? ? ? * eqn 7.14 ) . based on the pdf of @xmath24 , we calculate its mean @xmath46 and variance @xmath47 by numerical integration . alternatively , we also calculate @xmath46 and @xmath47 through approximation results of @xcite , which assume @xmath24 follows normal distribution . the approximation results are given by rcl m__p & & ( m_a / m_b ) + _ b^2 m_a / m_b^3 - _ ab _ a _ b /m_b^2 + _ _ p^2 & & _ b^2 m_a^2/m_b^4 + _ a^2/m_b^2 - 2 _ ab _ a _ b m_a /m_b^3 . [ approx ] the gaussian approximation ( [ approx ] ) is verified via numerical simulations . specifically , the mean and variance calculated from integration of the closed - form pdf are compared with the approximated mean and variance , for different node numbers . our results show that for @xmath48 , the approximation error for the means is essentially zero and less than 3% for the variances . up to this stage , we are able to calculate the mean and variance of the estimation error in the x - axis as @xmath49 , @xmath50 . the same analysis applies for calculating the mean and variance of the estimation error along the y - axis , namely , @xmath51 and @xmath52 . note that the estimation errors in x - axis and y - axis are correlated , therefore the statistical properties of the 2-dimensional error , @xmath53 , can not be obtained directly . in this case , since @xmath8 s in ( [ est x ] ) are correlated , thus the distribution of @xmath24 needs to be evaluated again . since @xmath37 in ( [ est x ] ) is the linear combination of correlated gaussian variables , it is still gaussian distributed @xcite , i.e. @xmath54 . thus , @xmath55 , and @xmath56 is given by @xmath57 , where @xmath58 is the @xmath59th element of the covariance matrix of @xmath60^{t}$ ] , defined as @xmath61 = \operatorname{\mathbb{e}}[s_i s_j ] = \sigma_s^2 \lambda_{ij } , \label{covariance_b}\ ] ] where @xmath62 . for the distribution of @xmath32 , it follows directly that @xmath63 . its variance is given by @xmath64 , where @xmath65 is the @xmath59th element of the covariance matrix of @xmath66^{t}$ ] , which can be represented by @xmath67[(p_j - p_{min})x_j- x_j \mu_j]\ } \ieeenonumber \\ = \operatorname{\mathbb{e}}[x_i s_i x_j s_j ] , \label{covariance_a}\ ] ] where @xmath68 $ ] is given by = _ s^2 ( _ i^2 + _ l^2 ) , & @xmath40 , + _ i _ j _ s^2 _ ij , & @xmath41 . [ correlation ] based on central limit theorem , we approximate @xmath32 as a gaussian random variable , i.e. @xmath69 . the approximation is validated via simulations , and it applies for @xmath70 . the correlation coefficient @xmath38 is given by @xmath71/(\sigma_a \sigma_b ) = ( \sigma_s^2 \sum_{i=1}^{n } \sum_{j=1}^{n } \overline{x}_i \lambda_{ij } ) / ( \sigma_a \sigma_b ) . \label{correlation coeff}\ ] ] substituting the expressions for @xmath72 and @xmath73 into ( [ correlation coeff ] ) , after some simple manipulation , @xmath74 where @xmath75 and @xmath76 is a @xmath77 vector of all @xmath78 s . since the location estimate @xmath24 is formed by @xmath79 , we use the closed - form pdf given by ( * ? ? ? * eqn 7.14 ) to calculate @xmath46 and @xmath47 by numerical integration . gaussian approximation can also be used similarly to i.i.d shadowing case in sec . [ subsubsec : iidshadowing ] . in our previous analysis , we obtained the statistical properties of 1-dimensional location estimation error for both i.i.d . and correlated shadowing case . from gaussian approximation of 1-dimensional localization error ( [ approx ] ) , we know that @xmath80 , where @xmath81^t$ ] , @xmath82,\ ] ] and @xmath83 is the correlation coefficient between 1-dimensional errors , the calculation of which is presented in the appendix . the calculation of localization error @xmath53 involves computing 2-norm of @xmath84 . since the two variables in @xmath84 are dependent gaussian r.v.s with different variances , we perform the following transformations . first we form two independent gaussian variables from @xmath84 , by the standard technique of decorrelation @xcite , @xmath85^t$ ] , where @xmath86 is the orthogonal eigenvalue decomposition matrix of @xmath87 , which satisfies @xmath88 , where @xmath89 is a diagonal matrix . therefore , one can verify that ( * ? ? ? * eqn 2.1 - 159 ) @xmath90 . then by the orthogonal - invariant property of 2-norm @xcite , we know @xmath91 . in order to obtain the pdf of @xmath92 , let us first consider pdf of @xmath93 . @xmath94 and @xmath95 are two chi - square variables with one degree of freedom . then the pdf of @xmath96 is the convolution of these two chi - square pdfs , denoted as @xmath97 . the pdf of the localization error @xmath98 is given by @xmath99 . as a result , the closed - form pdf of @xmath53 , for given sensor node positions , is given by rcl p_e_l(x ) & = & ( - ) + & & _ i=0^ _ l=0^ , x 0 [ pdf_2d ] where @xmath100 , @xmath101 and @xmath102 , @xmath103 are means and variances of @xmath104 and @xmath105 , respectively , and @xmath106 and @xmath107 are gamma function and @xmath108th - order modified bessel function of the first kind , respectively . the mean and variance of @xmath53 , denoted as @xmath109 and @xmath110 , are calculated using their definitions , by numerical integration over the above pdf . note that the above calculations are based on a particular fixed node placement . if the nodes are randomly placed within the area according to some distribution , say uniform distribution , integrating @xmath111 times over all node coordinates gives the theoretical average performance . based on the theoretical analysis , wcl requires large number of nodes in order to achieve a high level of accuracy . this causes large communication overhead in terms of number of messages and transmit power consumption when implemented in a centralized manner . we now present a practical distributed implementation of wcl to address this problem in the context of cr networks . clustering is commonly used in implementation of distributed algorithms . the challenge in designing these algorithms is to reduce the amount of information exchanged among clusters . the proposed algorithm accomplishes this by splitting the task into two phases . first , the cluster with the highest average rss is selected . this cluster then acts as the coordinator for the second phase which performs the actual wcl algorithm . nodes , @xmath112 , in the cr network are assumed to be organized into clusters , @xmath113 , using the clustering algorithm presented in @xcite . this particular algorithm is based on geographic boundaries and generates clusters which have a predefined shape ( hexagonal in this case ) . the regularity in the shape of the cluster is beneficial for the first step of the distributed algorithm since wcl works best for a radially symmetric distribution of nodes . however , the proposed algorithm would work for any arbitrary clustering algorithm as long as nodes are clustered based on proximity . the pu is assumed to be randomly placed in the entire area and can thus be in any of these clusters . each cluster selects one node , @xmath114 , to be the cluster head . the cluster head maintains a table of all adjacent clusters , @xmath115 , and the location of its member nodes , @xmath116 . once a pu is detected by a node , using an existing pu detection method such as energy detection @xcite , it transmits its rss measurement to the corresponding cluster head which then participates in the distributed wcl algorithm outlined in the next section . using a predetermined fusion rule , such as a majority vote , the cluster head makes a decision on the presence of the pu . all clusters that decide that the pu is present , participate in the distributed wcl algorithm detailed in the next subsection . these clusters form a set defined as @xmath117 . if in addition , some prior knowledge of the pu transmit power is available , a situation typical of a cr network coexisting with a licensed incumbent network , then a predetermined threshold can be used to further reduce the number of clusters in @xmath117 . the threshold can be chosen based on the average node spacing to get a certain desired average number of nodes . we investigate the effect of this number on wcl accuracy in sec . [ subsec : variableparticipation ] . the first phase , which selects the head cluster , is shown in algorithm [ alg : selection ] . the objective is to select the cluster with the highest average rss , @xmath118 . this is done in a distributed manner by comparing the average rss between neighboring clusters . however , directly implementing this procedure would require a large amount of inter - cluster communication . to solve this issue , a cluster only communicates its average rss to the neighboring cluster , @xmath119 , in the direction of the pu relative to the cluster s centroid . although techniques for estimating the angle of arrival of a transmitter could be beneficial , these methods require cluster heads to have multiple antennas which increases complexity of sensors . in order to get an approximation of the pu direction , we propose a metric , @xmath120 , shown in line 5 of algorithm [ alg : selection ] . we get @xmath120 by subtracting the wcl result of all nodes within the same cluster from the geometric centroid of those nodes . this vector approximates the rss gradient within the cluster . it corresponds to an approximate direction of the pu since on average the rss in ( [ received power ] ) decreases as distance to pu increases . therefore , the negative of the gradient is directed towards the point with maximum rss . once the cluster with the highest average rss is selected , algorithm [ alg : mwcl ] is performed . in this step , the final wcl calculation is done using nodes within a given radius , @xmath121 , of the selected cluster s strongest node ( sn ) , @xmath122 . @xmath121 is adaptively chosen to be the minimum of the distance to the closest map edge , @xmath123 , and the cluster radius @xmath124 . this reduces the border effect , which is the tendency of the location estimate to be biased towards the center when the pu is close to the map border , by ensuring a radially symmetric node distribution even when the pu is close to the network edge . in order to improve the localization accuracy , additional nodes from adjacent clusters are also included in the wcl calculation . using the location of @xmath122 and @xmath121 , these adjacent clusters can reduce the amount of information exchanged by only transmitting the location and rss measurements of nodes that fall within this circle . in the case of large clusters , each cluster could also choose to calculate a weighted centroid using its own nodes and transmit the result of this calculation only . since wcl is essentially an averaging algorithm , this technique allows even algorithm [ alg : mwcl ] to be distributed . @xmath125 @xmath126 @xmath127 \mathbf{l}(n_j ) } { \sum_{n_j\in c_i}(\operatorname{p}(n_j ) - \min(\operatorname{p}(n_j))}$ ] @xmath128 @xmath129 select @xmath130 proceed to algorithm [ alg : mwcl ] @xmath131 @xmath132 @xmath133 @xmath134 \mathbf{l}(n_j ) } { \sum_{n_j \in c^*_{wcl}}\left[\operatorname{p}(n_j ) - \min(\operatorname{p}(n_j)\right]}$ ] * function * @xmath135 * return * @xmath136 , @xmath137 * end function * the power efficiency and computational complexity of this distributed approach could further be improved by taking into account a practical mobility model of the pu . there has been a lot of work such as @xcite , which discussed how to incorporate past observations of the target position and a mobility model to improve localization . however , applying these methods introduces complexity to wcl and defeats the purpose of such a low complexity scheme . intuitively , assuming the pu continuously transmits during successive wcl estimations , a pu would most likely be in the same cluster as it was in the previous wcl calculation , or it could move to an adjacent cluster . by considering this fact , algorithm [ alg : selection ] can further limit the clusters involved in choosing the head cluster , which is the most power consuming phase , by only taking the previous cluster head and its adjacent clusters as @xmath117 . we analyze the total transmit power and computational complexity of cwcl and the proposed dwcl algorithm . in order to do so , we first need to investigate the transmit power on a single link of length @xmath138 . the minimum transmit power is obtained from ( [ received power ] ) as @xmath139 , where @xmath140 is the minimum received power to correctly receive the message . for the computation complexity estimates , we assume the number of operations ( ops ) for addition , substraction and comparison is 1 ; for multiplication and division is 10 . in cwcl , the total number of messages is the sum of reporting messages from nodes to the fusion center , thus , @xmath0 . for the transmit power consumption , we assume the fusion center is roughly at the center of a circle area of radius @xmath1 , and nodes are placed according to a uniform distribution . in this case , one can verify that , @xmath141 , the distance from the @xmath3th node to the fusion center satisfies @xmath142 , where @xmath143 is the uniform distribution from @xmath32 to @xmath37 . thus the average total power consumption is expressed as @xmath144 = \operatorname{\mathbb{e } } [ p_{r,\min } \sum_{i=1}^{n } ( d_i^{\gamma}10^{-s_i/10 } ) ] \approx n p_{r,\min } \operatorname{\mathbb{e}}[d_i^{\gamma } ] \operatorname{\mathbb{e}}[10^{-s_i/10}],\ ] ] where we use statistical average to replace sum of realizations . note that @xmath145 = \operatorname{\mathbb{e}}[{(d_i^2)}^{\gamma/2}]$ ] , which is the @xmath146th moment of an uniform variable . in order to calculate @xmath147 $ ] , we need the distribution of @xmath148 , which is a function of central gaussian variable @xmath149 . thus we can derive the desired pdf as @xcite @xmath150 finally , the average total power consumption is given by @xmath151 = n p_{r,\min } \operatorname{\mathbb{e}}[d_i^{\gamma } ] \int_0^{\infty } x p(x ) d x.\ ] ] from the definition of wcl estimate , number of ops of cwcl is @xmath152 . assume that : @xmath153 is the average number of nodes in one cluster ; @xmath154 is the average number of clusters ; @xmath155 is the number of neighboring clusters for each cluster ; @xmath156 is the average percentage of clusters who decide they are the head cluster in step 7 of algorithm 1 ( clusters that have to perform step 8) . the total number of messages for algorithm 1 is @xmath157 . to calculate transmit power , in step 1 ) we use the same procedure as cwcl ( with a smaller radius ) ; step 2 ) and 3 ) are communication among cluster heads , we use the average distance between centers of two clusters as the estimate of the average distance among cluster heads . in algorithm 2 , the total number of messages is @xmath158 and the calculation of power consumption for this process is the same as step 2 ) of algorithm 1 . in terms of computational complexity , the number of ops for algorithm 1 is @xmath159 while the number of ops for algorithm 2 is @xmath160 . in this section , we evaluate the performance of cwcl algorithm and its proposed distributed implementation . for cwcl algorithm , using the theoretical framework presented in section [ sec : analysis ] , we consider impact of sensor node positioning error , different levels of shadowing @xmath161 and different correlation distances @xmath162 . the effect of node placement is investigated under random grid and uniform distribution . prior work on wcl uses mean localization error as the performance metric of interest . in @xcite , a slightly modified performance metric where the mean error is normalized by the transmission range @xmath1 of pu is adopted . simulations in these prior work have shown that the accuracy of wcl improves as the density of nodes is increased . this result is intuitive since the node spacing decreases as the density of nodes is increased . in this work , we introduce a novel performance metric where the mean localization error @xmath163 is normalized with respect to the average node spacing @xmath164 . the motivation for choosing this metric is to eliminate the effect of geometry scaling which intuitively improves accuracy . in order to isolate the effect of shadowing on wcl performance we consider a fixed grid placement of nodes with perfect positioning ( i.e. @xmath165 ) and fixed pu position . the normalized mean localization error for this scenario is shown in fig . [ fig : fixgridsquare_shadow_mean ] . for given shadowing variance , as the number of nodes increases , there is a slight improvement in the accuracy , which is a consequence of averaged shadowing over many nodes . note that this gain saturates at approximately 200 nodes or more . we also observe that wcl is quite robust to shadowing , since for an additional @xmath166 of shadowing from @xmath167 to @xmath168 , normalized error increases only 5% . therefore wcl is well suited for pu localization in severely shadowed environments . the results for mean localization error under correlated shadowing are shown in fig . [ fig : fixgridsquare_corr_mean ] . assuming fixed value of @xmath169 , and different levels of @xmath162 normalized to @xmath164 are presented . in highly correlated shadowing such as @xmath170 , increasing the number of nodes always worsens the error . for the high node density case , e.g. 400 nodes , the average error is as high as 47% of @xmath164 . we conclude that when all nodes are included in the wcl calculation , correlation has a degrading effect to accuracy . further investigation shows that the correlation distance resulting in the worst performance is proportional to the number of nodes involved . the impact of positioning accuracy of sensing nodes is presented in fig . [ fig : imperfect location knowledge ] with a fixed grid placement of nodes in an i.i.d . shadowing environment with @xmath171 . analysis shows that errors in sensor node positioning decrease the wcl accuracy but the impact is marginal . for example , @xmath7 of @xmath172 m increases the error for about @xmath78% . ) as a function of variance in the participating nodes positioning error , @xmath7 . for this scenario , nodes are placed on a grid with equal spacing ( fixed grid placement ) in an area with @xmath173 m . pu is at the center of the area . ] the previous subsections considered fixed node placement and pu location . realistically , node placement can be random and the pu also appears at random locations relative to the sensor nodes . next we consider two cases of random node placement : random grid and uniformly distributed placement . in the random grid scenario , the sensor nodes are still on a grid , but the pu can appear uniformly within the center box . a grid distribution of nodes can be practically implemented if nodes are explicitly deployed for the purpose of acting as a dedicated localization sensor network . evenly placed beacons have been shown to provide better accuracy for wcl in wsns @xcite . results for random grid placement in fig . [ fig : randomgridanduniformrandomplacement ] show approximately 10% performance loss compared to fixed pu location scenarios . the uniformly distributed placement has been used in @xcite to more accurately model scenarios where the sensor node positions are not predefined , such as in mobile ad - hoc networks . the results for uniformly distributed placement are also shown in fig . [ fig : randomgridanduniformrandomplacement ] . it is observed that randomness in the node placement can increase the error as much as three times compared to the fixed node placement . ) with varying correlation distance normalized to the node spacing . results of two placements , random grid ( rg ) and uniform random ( ur ) , are shown in an area with @xmath173 m . for both placements , @xmath174 . ] the results given so far assumed an ideal circular coverage area for the pu . works such as @xcite indicated that in practical scenarios , the circular model can not be guarranteed due to several non - idealities such as fading , shadowing , and interference in the system . intuitively , the degree - of - irregularity ( _ doi _ ) parameter determines how closely the coverage area can be approximated by a circle . _ = 0 means the transmission range is perfectly circular . the model used in our simulation is the _ model briefly described in @xcite where the transmission range is a correlated gaussian r.v . centered at @xmath175 with variance @xmath176 . in fig . [ fig : varydoi ] the effect of _ doi _ is investigated in the presence of uncorrelated medium level shadowing ( @xmath177 ) . the result shows that _ doi _ between 0% and 40% increases the error by 4% . _ doi _ is not a significant impairment for wcl because nodes closest to the pu , where the effect of irregularity is negligible , are weighted more than nodes closer to the edge of the coverage area . however , a slight increase in the normalized mean error can be observed as more nodes are used at high doi ( 30% or 40% ) . the denser the node distribution , the more nodes are included in wcl , but the more nodes there are , the more deviation from a perfect circular coverage coverage area , leading to lower accuracy . with varying degree - of - irregularity ( doi ) . for this scenario , nodes are uniformly and independently distributed ( uniform random placement ) in an area with average @xmath175 . ] from fig . [ fig : randomgridanduniformrandomplacement ] we observe that in correlated shadowing increasing the number of nodes also increases the error . one solution to this problem is to reduce the number of participating nodes used in wcl for this scenario . this could be achieved by ignoring nodes farther away from the pu to reduce the number of nodes resulting in less degradation . in fig . [ fig : varyingparticipation ] we investigate varying levels of node participation for a fixed average node spacing , @xmath164 . in this case the total number of nodes is 100 , and the density remains the same while the number of involving nodes is varied . to select which nodes are included in the wcl calculation , nodes are sorted according to their rss values . the nodes with high received power are included in wcl . as a result , increasing the participation for the uncorrelated case always improves the accuracy . thus , 100% participation is optimal . however , even with a little correlation , this number is greatly reduced and about 10% of the total number of nodes is optimal . another selection scheme was previously proposed in @xcite where only nodes whose rss is within 15% of the stronges node s rss are used . the localization accuracy of wcl is also compared to other techniques suitable for cr scenarios . this includes lateration , centroid and strongest node ( sn ) . lateration is a range - based technique that uses least - squares estimation using distance estimates to the target . the centroid algorithm , on the other hand , is a range - free technique which calculates the average of all node positions in range of the pu . finally , the sn technique simply selects the node with the highest rss and uses its location as an estimate of the pu position . results in fig . [ fig : comaprison with other techiques ] show that wcl performs better under higher @xmath161 compared to lateration . although this seems counter - intuitive at first , the result can be explained by taking note that range - based techniques require an accurate estimate of the distance to the target to provide accurate results . in the case of non - interactive localization , where the only available information is the rss from the pu , shadowing induces a very large error in the range estimates . under low shadowing ( @xmath178 db ) , the lateration method is more accurate than wcl . at 0 db shadowing ( not shown in the figure ) , the range - based technique will achieve perfect localization . this result supports the claim that wcl is a viable method in the context of non - interactive localization because of its robustness . the centroid algorithm performs slightly worse than wcl because it does not utilize rss information . however , the performance difference becomes smaller at higher level of shadowing . . for this scenario , nodes are uniformly and independently distributed ( uniform random placement ) in an area with @xmath175 ( uncorrelated shadowing ) . note that the centroid technique is unaffected by @xmath179 . ] [ fig : dwcl compare ] presents simulation results that compare normalized mean error of cwcl , sn and dwcl under different shadowing environments . sn is included in the comparison because it forms the basis of the head cluster selection for dwcl so it acts as an upperbound to the error . we observe that the proposed dwcl performance is between cwcl and sn algorithms . it is also interesting to note that the accuracy of dwcl degrades relative to cwcl and approaches the performance of sn when shadowing is high . the reason behind this is that for cluster head selection , dwcl is dependent on the sn algorithm . the accuracy of the sn degrades at large shadowing shown in fig . [ fig : comaprison with other techiques ] . [ fig : fixgrid_nodepower_theo ] compares the transmit power consumption for cwcl and dwcl with different number of clusters . in the simulation we set @xmath180 which is the mean received power at the border of the area of @xmath175 , assuming @xmath181 dbm , @xmath182 m and @xmath183 in ( [ received power ] ) . we observe that dwcl with 16 clusters consumes about @xmath184dbm less power for each node compared with cwcl . the analysis also shows that transmit power of each node is reduced as the number of node increases , since the average communication distance is getting smaller in dwcl . [ fig : fixgrid_cpu_theo ] compares computational complexity estimated in terms of number of ops . it shows that distributed algorithms require about three times more operations than cwcl , however total transmit power is reduced . in this paper we present the first analytical framework for the performance of wcl for localization of pu transmitters in cr networks . we derived the closed - form expression for the error distribution of wcl parameterized by node density , node placement , shadowing variance , correlation distance and errors in sensor nodes positions . using the theoretical results in conjunction with numerical simulations , the robustness of wcl against various physical conditions is investigated and quantified . based on the above analysis , we provide guidelines in terms of number of nodes and node placement required to achieve certain localization accuracy using wcl and propose variable node participation for improved wcl performance . to facilitate the deployment of wcl in an energy efficient context , a distributed cluster - based implementation of wcl is also proposed , which does not assume a fusion center that collects all rss information from nodes . the distributed method achieves comparable accuracy with its centralized counterpart , and greatly reduces total power consumption . for notational convenience we define @xmath186 and form the vector @xmath187^{t}$ ] , so that @xmath188 with @xmath189 . note that this definition includes both i.i.d and correlated shadowing case . the first term in ( [ dimensional correlation ] ) then becomes @xmath190 = \operatorname{\mathbb{e}}\left[\frac{\mathbf{x}^{t } \mathbf{q } \mathbf{q}^{t } \mathbf{y}}{\mathbf{1}^{t } \mathbf{q } \mathbf{q}^{t } \mathbf{1}}\right].\ ] ] define three new variables @xmath191 , @xmath192 and @xmath193 , and their vector form @xmath194^{t}$ ] . now the first term in ( [ dimensional correlation ] ) becomes @xmath195 $ ] . as shown in sec . ii , @xmath196 , and @xmath197 . @xmath198 can be approximated as a single gaussian variable as shown in sec . [ subsubsec : iidshadowing ] since it is the same as @xmath32 in ( [ est x ] ) . we denote its distribution as @xmath199 , similarly @xmath200 . the distribution of @xmath201 is given as @xmath202 , where @xmath203 and @xmath204 . as a result @xmath205 , where @xmath206^t$ ] is the mean vector and @xmath207 is the covariance matrix . to derive @xmath207 we find @xmath208 as @xmath209 - m_1 m_2 = \operatorname{\mathbb{e}}[(\sum_{i=1}^{n } x_i q_i)(\sum_{j=1}^{n } y_j q_j ) ] - m_1 m_2 = \overline{\textbf{x}}^{t } \mathbf{\omega } \overline{\textbf{y } } - m_1 m_2\ ] ] where in the last the equation we used the fact @xmath210 = \overline{x}_i \overline{y}_j \mathbf{\omega}_{ij}$ ] . similarly we calculate all entries of @xmath207 . in order to simplify the calculation of ( [ reform ] ) , we perform the following transformation to @xmath211 , @xmath212 , thus @xmath213 . clearly @xmath214 , where @xmath215 is the @xmath3th column of @xmath216 . then we have @xmath217 = \operatorname{\mathbb{e}}[\frac{(m_1 + \boldsymbol{\omega}_{1}^{t } \widetilde{\textbf{t}})(m_2 + \boldsymbol{\omega}_{2}^{t } \widetilde{\textbf{t}})}{(m_3 + \boldsymbol{\omega}_{3}^{t } \widetilde{\textbf{t}})^2}].\ ] ] since the expectation in ( [ results ] ) involves multiplication and division of three correlated random variables , the calculation can not be done directly . in order to reduce the number of variables involved , first we form @xmath218 $ ] , then perform qr factorization to reduce dimensionality , obtaining @xmath219 , in which @xmath86 is orthogonal and @xmath220 is upper triangular @xmath221.\ ] ] note that since @xmath225 , denominator of ( [ results ] ) can further be expressed as @xmath226 \widetilde{\textbf{t}}\right)^{2 } \doteq \left(m_3 + r_{11 } \widetilde{t}_{1}\right)^{2 } , \label{den2_new}\ ] ] where we use the fact that @xmath227 is the first column of @xmath220 ; and @xmath228^{t}$ ] . similarly , numerator of ( [ results ] ) can also be further expressed as therefore , the expectation in ( [ reform ] ) becomes @xmath229 = \frac { r_{12 } r_{13 } } { r_{11}^2 } \operatorname{\mathbb{e}}\left[\frac{\widetilde{t}_{1}^2 + \left(\frac{m_1}{r_{12 } } + \frac{m_2}{r_{13}}\right ) \widetilde{t}_{1 } + \left(\frac{m_1 m_2 + r_{22 } r_{23}}{r_{12 } r_{13}}\right)}{\left(\widetilde{t}_{1 } + \frac{m_3}{r_{11}}\right)^{2}}\right ] . \label{exp1_new}\ ] ] after factorization ( [ exp1_new ] ) becomes @xmath230 = \frac { r_{12 } r_{13 } } { r_{11}^2 } \!\!\left(\!\!1 + \frac{\left(\frac{m_1}{r_{12 } } + \frac{m_2}{r_{13}}\right)-2 \frac{m_3}{r_{11}}}{\frac{m_3}{r_{11 } } } + \frac{\left(\frac{m_1 m_2 + r_{22 } r_{23}}{r_{12 } r_{13}}\right)+\left({\frac{m_3}{r_{11}}}\right)^2-\left(\frac{m_1}{r_{12 } } + \frac{m_2}{r_{13}}\right ) \frac{m_3}{r_{11}}}{1+\left({\frac{m_3}{r_{11}}}\right)^2 } \right).\ ] ] substituting ( [ corr - coeff_new ] ) into ( [ dimensional correlation ] ) gives the result . j. wang , p. urriza , y. han , and d. abri , `` performance analysis of weighted centroid algorithm for primary user localization in cognitive radio networks , '' in _ proc . asilomar conference on signals , systems , and computers _ , pacific grove , ca , usa , nov . 710 , 2010 . h. chen , q. shi , r. tan , h. poor , and k. sezaki , `` mobile element assisted cooperative localization for wireless sensor networks with obstacles , '' _ ieee trans . wireless commun . _ , vol . 9 , no . 3 , pp . 956 963 , mar . 2010 s. liu , y. chen , w. trappe , and l. j. greenstein , `` non - interactive localization of cognitive radios based on dynamic signal strength mapping , '' in _ proc . conf . on wireless on - demand network systems and services _ , snowbird , ut , usa , feb . 24 , 2009 . r. behnke , j. salzmann , r. grossmann , d. lieckfeldt , d. timmermann , and k. thurow , `` strategies to overcome border area effects of coarse grained localization , '' in _ proc . 6th workshop on positioning , navigation and communication _ , hannover , germany , eu , mar . 19 , 2009 . q. zhou , x. li , and y. xu , `` mean shift based collaborative localization with dynamically clustering for wireless sensor networks , '' in _ proc . conf . on communications and mobile computing _ , vol . 2 , yunnan , china , jan . 68 , 2009 . m. k. simon , _ probability distributions involving gaussian random variables : a handbook for engineers , scientists and mathematicians_. 1em plus 0.5em minus 0.4emnew york , ny , usa : springer - verlag , 2006 . r. chandra , j. padhye , a. wolman , and b. zill , `` a location - based management system for enterprise wireless lans , '' in _ proc . symposium on networked systems design and implementation _ , cambridge , ma , usa , apr . 1113 , 2007 .
within the last fifteen years or so , the dynamical mean field theory approach ( dmft)@xcite and its cluster generalizations@xcite have become some of the most powerful techniques to study strongly correlated electrons . in these approaches , a single - site hybridized to a bath or a cluster hybridized to a bath must be solved . the bath of non - interacting electrons is determined self - consistently . at the heart of the dmft approach then , one finds so - called impurity solvers . there are now very powerful impurity solvers , for example continuous time quantum monte carlo ( ctqmc ) methods@xcite . these methods are exact within statistical errors and , for the one - band hubbard model , certain versions@xcite at the single site level do not suffer from sign problem . yet , these approaches require sizeable computational resources and , in addition , real frequency information must be obtained through analytical continuation of data with statistical uncertainties , an ill - posed problem@xcite . it is thus still of great interest to work with approximate solvers that are reliable and do not suffer from statistical uncertainties . this facilitates the calculation of real - frequency quantities with pad approximants@xcite or directly in real - frequency and also enables one to quickly explore phase diagrams and pinpoint interesting regions of parameter space where state of the art solvers would be useful . among possible approximate solvers , one finds exact diagonalization , slave bosons , non - crossing approximation ( nca ) , numerical renormalization group ( nrg ) and others@xcite . they all have advantages and disadvantages . for example , exact diagonalization can consider only a limited number of bath sites , nca is limited to high temperatures and nrg to low energies . here we consider iterated perturbation theory ( ipt ) @xcite , an interpolation approach that generalizes the original@xcite ipt applicable only at half filling . this method has been , and still is , wildly used@xcite . the interpolation is constructed so that the self - energy recovers both the exact result in the atomic limit and the high - frequency limit of the hubbard model . there is one parameter however that can not be determined from these constraints . there have been several proposals to fix this parameter . at @xmath1 one can impose that luttinger s theorem be satisfied ( ipt-@xmath2 ) as was done in [ ] but when this condition is applied at finite temperature , the results are not satisfactory@xcite . another very popular approach for non - zero temperature fixes the occupation @xmath3 of the non - interacting part of the anderson impurity problem used in the perturbative calculation to be equal to the lattice occupation @xmath4 @xcite ( ipt-@xmath3 ) . this condition is arbitrary since there is no general principle relating these two numbers , but it turns out to be quite satisfactory in the case of correlated metal i.e. @xmath5@xcite . @xmath6 is the coupling for which the metal to insulator mott transition occurs at half - filling . despite this success , it is known@xcite that when @xmath7 is larger than the critical value for the mott transition at @xmath8 ( @xmath9 ) , then ipt breaks down for @xmath10 at low @xmath11 when the condition @xmath4 is applied . this happens even if , in principle , ipt is constructed to respect the atomic limit @xmath12 . it has been proposed@xcite that preserving the third moment of the spectral weight improves the results . here we show that ipt-@xmath3 is unsatisfactory for @xmath13 close to half - filling for both @xmath10 and @xmath14 . we propose a way to circumvent this problem by using the fact that when @xmath7 is large enough , the double occupancy becomes almost temperature independent in the paramagnetic state with a value that is , to a high degree of accuracy , a simple function of the density . this provides us with a condition different from @xmath4 that allows one to close the ipt equations even for large coupling . this approach , ipt-@xmath0 , is applicable at all temperatures contrary to the approach that enforces luttinger s theorem . it can in principle be improved further by enforcing the third - moment sum rule@xcite . in sec . [ dmft ] we summarize the dmft approach , the solvers that we use and the manner in which fermi liquid parameters are extracted . [ breakdown ] demonstrates the failure of ipt at large coupling . in this section and throughout the text , numerical examples are obtained with the 3-dimensional fcc lattice . amongst lattice presenting electronic frustration , the fcc lattice is important because of its prevalence in nature . our main contribution appears in sec . [ sec : double ] where we show that double occupancy can be accurately determined from simple arguments at very strong coupling and then used to fix the remaining parameter in ipt . we call this approach ipt-@xmath0 . fermi liquid parameters , density of states , chemical potential , energy , and specific heat on the fcc lattice are calculated with both ipt-@xmath0 and ctqmc as benchmark examples . resistivity and optical conductivity obtained with ipt-@xmath0 are physically reasonable . appendix a contains details on the three dimensional adaptive integrator we developed for both ipt and ctqmc calculation . appendix b contains details of the implementation of ipt-@xmath0 . appendix c explains how to calculate the different non - interacting functions and appendix d gives details on the calculation of the optical conductivity . we study the one - band hubbard model , @xmath15where @xmath16 is the hopping matrix between sites @xmath17 and @xmath18 , @xmath7 the on - site coulomb repulsion , @xmath19 the creation ( annihilation ) operator for an electron of spin @xmath20 on site @xmath17 and @xmath21 is the number operator . the dynamical mean - field theory ( dmft ) provides a solution of the hubbard model that describes the mott transition in three dimensions and has predictive power for real materials@xcite . drawing from ideas on the solution of the hubbard model in infinite dimension@xcite , the self - energy in this approach depends only on frequency . one first solves the problem of a single site with the hubbard @xmath7 , hybridized with an infinite bath of non - interacting electrons , the so - called anderson model . one extracts the frequency - dependent self - energy of the anderson model , which is then taken as the self - energy in the lattice green s function . the bath is determined self - consistently by requiring that projection of the lattice green s function on a single site is identical to the single - site green s function of the anderson model . the anderson impurity problem can be solved numerically with a very high precision . dmft has been justified with a variety of approaches@xcite including a variational one@xcite . the single - site dmft is exact in infinite dimension@xcite . benchmarks against the bethe ansatz solution in one - dimension shows that dmft can be an accurate solution of the hubbard model also in lower dimensions @xcite . mathematically , the partition function for the anderson impurity problem is given by the imaginary - time grassmann path integral @xmath22\,\mathrm{e}^{-s_{0}-\int_{0}^{\beta } d\tau \int_{0}^{\beta } d\tau ^{\prime \dag } ( \tau ) \delta ( \tau , \tau ^{\prime } ) \psi ( \tau ^{\prime } ) } , \label{eq : z}\]]with @xmath23 @xmath24 the bath hybridization and @xmath25 the action of the impurity , which consists of a single site with repulsion @xmath7 . the self - consistency condition in matsubara frequency reads @xmath26 ^{-1}. \end{split } \label{eq : scc}\]]with @xmath27 the self - energy . on the fcc lattice the single - particle dispersion is given , with lattice spacing @xmath28 , by @xmath29 $ ] . in single - site dmft , the self - energy is local ( @xmath30 ) which enables one to transform the integrals over the brillouin zone entering the self - consistency relation into integrals over the non - interacting density of states @xmath31 . however , in the case of a 3d fcc lattice , there is no analytic form for @xmath31 and its accurate numerical calculation using a monte - carlo binning procedure is cumbersome and only produces a fixed finite numbers of points . on the other hand , if we used a lorentzian as an approximation for the delta function , the band - edges and the van - hove singularities would suffer from accuracy problems that could be transferred to the dmft calculation . we thus performed the calculation with the full @xmath32 space integration . we devised an adaptive 3d fifth order gaussian quadrature for a cube . this integration method is explained in appendix [ appen_integrate ] . for comparisons in calculation of transport properties , we have nevertheless calculated @xmath31 using monte carlo integration as explained in appendix [ mc_function ] . the resulting non - interacting density of state for the fcc lattice with nearest - neighbor hopping only is shown in fig . [ fig : dos_u0 ] . we have used two `` impurity solvers '' for the auxiliary anderson model . they are described in the following subsections . the first method is the numerically exact continuous time quantum monte carlo method ( ctqmc ) @xcite , a finite temperature approach that relies on the monte carlo summation of all diagrams obtained from the expansion of the partition function in powers of the hybridization @xmath33 . this method does not have a sign problem , and does not have errors associated with time discretization or bath parametrization . it is therefore exact within statistical errors but computationally expensive . we refer to the literature for an explanation of the approach@xcite . we describe the second approach , iterated perturbation theory ( ipt ) , in more details since it is the focus of this paper . ipt is an approximation method that relies on an interpolation from @xmath34 order perturbation theory for the anderson impurity problem @xcite . the interpolation preserves the correct high - frequency limit for the self - energy and is exact in both the non - interacting and the atomic limits . we only consider paramagnetic solutions . the self - energy in this approach is parametrized by @xmath35 where @xmath36 with @xmath37 and @xmath33 the hybridization function . the constants @xmath38 and @xmath39 @xmath40 where @xmath41 and @xmath42 , are chosen such that one recovers the exact solution in the atomic limit as well as the exact result for arbitrary @xmath7 in the high - frequency limit . the green s function used to obtain the density @xmath43 is @xmath44 in eq . , @xmath45 is the chemical potential of the lattice that is determined by fixing the value of @xmath43 while @xmath46 is the chemical potential determined by the fictitious density @xmath3 . we need an additional equation to fix @xmath46 . this problem has been studied carefully in refs . setting @xmath47 is not a good option . indeed , as mentioned in the introduction , fixing luttinger s volume works only at very low temperature@xcite . a widely used approach @xcite consists in fixing @xmath4 . we call this approach ipt-@xmath3 . one can also modify the formula for the interpolated self - energy by requiring that the third moment , appearing in the high - frequency expansion of the green s function , be satisfied exactly . in this case , the deficiencies of ipt-@xmath3 at strong coupling are not as severe . we will see below that requiring that double - occupancy takes its exact value is an easier solution that does not modify the simplicity of the original scheme and gives accurate results . ipt can be implemented efficiently , as described in appendix . [ appen_ipt ] , so that the solution can be obtained in a very short time . in this section , we first define the physical parameters that will be used to demonstrate the breakdown of ipt-@xmath3 . then we take advantage of the existence of the exact ctqmc impurity solver to characterize the solution of the dmft equation . the last subsection demonstrates that for @xmath7 much larger than the bandwidth , ipt-@xmath3 fails to reproduce even qualitatively the exact solution . at low @xmath11 and finite doping , it is known that dmft predicts a fermi liquid regime no matter how strong the interaction@xcite . in other words , a quasiparticle peak always appears at @xmath48 at low @xmath11 except at half - filling when @xmath49 . we characterize the fermi liquid with three parameters namely the effective chemical potential , @xmath50 the quasiparticle weight , @xmath51 and the scattering rate @xmath52 , where real and imaginary parts are defined by @xmath53 . in the dmft treatment of the hubbard model , luttinger s theorem is satisfied at @xmath1 when @xmath54 takes the value of the non - interacting chemical potential that gives the same density . all of the above parameters are calculated with the self - energy on the real frequency axis and thus , in principle , one needs to perform an analytical continuation from the data in matsubara frequencies and then extrapolate to zero temperature . in practice , we calculate the values of the self - energy for a few very low temperatures and use them to extrapolate to zero frequency and zero temperature . for the retarded @xmath55 and @xmath52 we take @xmath56 at the smallest positive @xmath57 for three low temperatures and extrapolate to @xmath1 using the fact that @xmath58 . for @xmath59 , the spectral definitions of the self - energies @xmath60 allow one to prove @xmath61}{\omega_n}}\big|_{\omega_n\rightarrow 0 } = \int{\frac{d\omega'}{\pi}}{\frac{\sigma(\omega')}{\omega'^2 } } = { \frac{{\partial}\sigma'(\omega)}{{\partial}\omega}}\big|_{\omega = 0}\ ] ] which , for a linear dependence of @xmath62 $ ] on @xmath63 , also follows from the cauchy - riemann relation for holomorphic functions of a complex variable . this last equation with three low temperatures allows us to calculate @xmath64/\omega_{n=0}$ ] , extrapolate to @xmath1 and obtain @xmath59 . consider the one - band hubbard model on the fcc lattice , where the single particle dispersion is given by @xmath29 $ ] . we present the dmft results obtained with the ctqmc impurity solver for @xmath65 and @xmath66 , below and above @xmath6 for the mott transition at half - filling . the bandwidth is @xmath67 for the 3d fcc lattice . fig.[fig : fermi_ctqmc ] displays the fermi liquid parameters . in fig.[fig : fermi_ctqmc]-(a ) the red solid line shows the non - interacting chemical potential as a function of density . the effective chemical potential @xmath68 is shown with blue circles for @xmath65 and black points for @xmath66 . the dashed lines indicates the position of the band edges for the 3d fcc lattice . as expected , except at half - filling for @xmath66 , luttinger s theorem is satisfied . in fig.[fig : fermi_ctqmc]-(b ) , the extrapolated scattering rate @xmath69 is negligibly small , except for @xmath66 at @xmath8 . for @xmath65 , ( not shown ) it is of the order @xmath70 and has essentially no density dependance . the value of @xmath59 , shown in in fig.[fig : fermi_ctqmc]-(c ) , behaves as expected : for @xmath9 , @xmath59 vanishes when the occupation approaches half - filling while it is close to the non - interacting value @xmath71 when the lattice is almost empty or full . we can also see that even for coupling as large as @xmath66 , the absence of particle - hole symmetry in the dispersion relation still leads to a value of @xmath59 that is not symmetric with respect to half - filling . clearly , electronic frustration plays an important role in the doped mott insulator . the ipt equations eqs . - do not determine the value of @xmath3 . as mentioned previously , for @xmath1 the requirement that luttinger s theorem be satisfied ( ipt-@xmath2 ) provides an additional independent equation , except at half - filling for @xmath9 . however , luttinger s theorem is in general not satisfied at finite temperature and the method becomes inaccurate . the condition @xmath4 has thus been proposed@xcite ( ipt-@xmath3 ) . it gives satisfactory results for correlated metals ( not for the insulator at half - filling ) . the results for the low - temperature extrapolations of @xmath72 and @xmath59 for @xmath65 and @xmath66 are shown in fig . [ fig : fermi_ipt ] . below the mott transition , @xmath65 , the @xmath4 results ( brown ( @xmath73 ) ) are shown . one can detect only a very small difference with the solid red line . luttinger s theorem is thus essentially satisfied . on the other hand , ipt-@xmath3 for @xmath66 ( kaki ( @xmath74 ) ) gives non - physical results not only@xcite for @xmath10 but , quite generally , close to half - filling . not only is luttinger s theorem strongly violated , but for a large range of densities , @xmath10 , @xmath54 is outside the band . many properties of the fermi liquid are proportional to functions of the non - interacting system evaluated at @xmath54 . but these functions are zero outside the band and so if @xmath54 is outside the band we obtain zero . for example , this would predict an insulator away from half - filling . the situation is not better for @xmath59 , especially around half - filling where it vanishes for a finite range of densities when @xmath10 . this demonstrates that at low @xmath11 , when @xmath7 is large , ipt-@xmath3 in its simplest form can not be applied . we must thus search for a new condition to explore this region of parameter space imposing exact results such as sum - rules , whenever possible , is desirable for any physical theory . whereas the condition @xmath4 is not required by any fundamental principle , the self - energy must always obey @xmath75 where @xmath76 is double occupancy . enforcing this consistency condition between single - particle properties , such as @xmath77 and @xmath78 , and a two - particle property , @xmath0 , has been successful in other approaches , such as the two - particle - self - consistent theory.@xcite . in the regime of interest here , strong coupling , @xmath0 can be accurately estimated and is only very weakly dependent on temperature , as discussed in the following subsection . there , we assess the accuracy of the approach . for double occupancy @xmath79 plotted as a function of temperature . we define @xmath80 for fillings @xmath14 and @xmath81 for @xmath10 . the various densities are represented by different symbols : @xmath82 ( black ( @xmath83 ) ) , @xmath84 ( blue ( @xmath85 ) ) , @xmath86 ( red ( @xmath87 ) ) , @xmath88 ( green ( @xmath89 ) ) , @xmath90 ( yellow ( + ) ) , @xmath91 ( cyan ( @xmath92 ) ) , @xmath93 ( magenta ( @xmath94 ) ) , @xmath95 ( brown ( @xmath96 ) ) and @xmath97 ( kaki ( @xmath98 ) ) . the largest deviations from the naive value , occurring close to @xmath8 , are less than @xmath99 in absolute value . ] as a function of density obtained from ctqmc for @xmath66 for three temperatures : @xmath100 ( black ( @xmath87 ) ) , @xmath101 ( blue ( @xmath102 ) ) and @xmath103 ( red ( @xmath104 ) ) . on this scale , the naive value of @xmath0 is very accurate . the inset is a zoom for densities @xmath105 . ] for very large @xmath7 , it is easy to guess that @xmath0 should depend only very weakly on temperature . in addition , the value of @xmath0 can be estimated quite accurately . indeed if @xmath14 there are unoccupied sites in the lattice and since @xmath7 is large , @xmath0 should naively be zero . for @xmath10 , @xmath0 is necessarily non - zero . if we start from the half - filled , mott insulator , with one electron per site and add electrons , they must go to a site which is already occupied . thus @xmath0 should simply equal the excess number of electrons measured from half - filling , @xmath106 . these estimates are called @xmath107 . in reality for @xmath14 there are corrections of order @xmath108 to double occupancy , giving rise to exchange interaction , and @xmath0 is slightly larger than zero . for analogous reasons , @xmath0 for @xmath10 will always be a bit larger than @xmath109 i.e. @xmath110 . this is obvious for models with particle - hole symmetry , but it will be true as well , even in the absence of particle - hole symmetry . we can verify our estimates with the ctqmc results for @xmath66 . within ctqmc , @xmath0 is calculated directly on the impurity by the monte - carlo sampling . fig.[fig : d_occ ] displays @xmath79 as a function of temperature for different densities . we see that @xmath79 is small and that the @xmath11 dependence is on the third significant digits . so , for all practical purposes , we can assume @xmath0 to be independent of @xmath11 although it differs from @xmath107 . the values of @xmath0 obtained are very close to the naive expectation but always slightly larger . in fig.[fig : d_occ_n ] we show the double occupancy @xmath0 as a function of the density for three different temperature from @xmath111 to @xmath112 . this figure confirms again that even if we have some dependence on @xmath11 , it is quite small and the result is a fairly simple function of the density . very similar results have been obtained in ref . [ ] in the case of a 3d simple cubic lattice . we thus have a rather simple constraint that we can take into account in ipt to fix all parameters . we call this approach ipt-@xmath0 . in the next subsection , we will assess the accuracy of this approach and verify how the results are modified when the exact value of @xmath0 is used instead of the naive one . we first set @xmath0 to its naive values , i.e zero for @xmath113 and @xmath109 for @xmath10 . the results for @xmath54 and @xmath59 are shown in fig . [ fig : fermi_ipt ] as cyan stars . clearly , the value obtained for @xmath54 is in much better agreement with luttinger s theorem than in the case ipt - n@xmath114 , although it is still incorrect for densities near half - filling . the biggest improvement is that we avoid values of @xmath54 outside the band . furthermore , in the case of @xmath59 , ipt-@xmath0 is close to the ctmqc results while for ipt - n@xmath114 it is quite far@xcite from the correct result , leading in particular to an un - physical insulator over a finite range of densities for @xmath10 . as we now show , one can improve the results further by using an accurate value of @xmath0 . that value can be obtained from a number of methods , in particular ctqmc . it can be computed quite accurately and does not require a large number of matsubara frequencies . even if ctqmc is available , it may be desirable to use ipt-@xmath0 because the calculation can either be done directly in real frequency or analytically continued from matsubara frequencies using simple methods such as pad approximants@xcite , whereas with ctqmc , maximum entropy@xcite is necessary . in addition , since @xmath0 has negligible temperature dependence in strong coupling , only one value of @xmath0 may be sufficient with ipt-@xmath0 to compute other quantities for a wide range of temperatures . since @xmath0 is not completely @xmath11 independent , we use an average called @xmath115 calculated between @xmath116 and @xmath103 for the purpose of comparison with the naive approach . it is calculated from the arithmetic mean of the numerical values of @xmath117 . note that the values of @xmath117 for each @xmath43 are taken as the arithmetic mean of the last four dmft iterations . the results are shown in fig . [ fig : fermi_ipt ] as magenta lozenges . we see that the results for @xmath54 are in excellent agreement with luttinger s theorem except very close to half - filling where it deviates , but not too much . surprisingly , @xmath59 is not as accurate as that obtained from the naive estimate of @xmath0 . but if we look at the results for densities between 0.8 and 1.2 , the difference between @xmath118 and @xmath119 is small and constant and @xmath118 correctly extrapolates to zero at half - filling . a small difference in @xmath0 can have a quantitative impact on @xmath59 , without affecting qualitative trends . for example for @xmath120 , the @xmath0 given by ctqmc is @xmath121 instead of the naive @xmath122 , whereas for @xmath123 , ctqmc gives @xmath124 instead of @xmath125 . we could imagine that because these quantities are at low @xmath11 , it would be better to take a @xmath0 that is in the low temperature range . if we do this , we obtain a @xmath115 slightly larger . we then find that @xmath54 is not really affected while @xmath59 is a little bit worse than that obtained from the average @xmath0 over the larger @xmath11 range . some tuning of @xmath0 would allow us to get a best possible set of @xmath54 and @xmath59 , but that is clearly not the purpose of the exercise . finally , for @xmath52 fig . [ fig : fermi_ipt]-(b ) shows that ipt-@xmath0 is qualitatively correct while being always smaller than ctqmc . it is instructive to look at the density of states obtained from ipt with fixed @xmath0 and pad analytical continuation@xcite . we show @xmath126 and @xmath100 as typical values in fig.[fig : dos ] . we compare to the ctqmc values obtained from maximum entropy analytical continuation of @xmath127 . that green s function is an average over several converged dmft iterations . the maximum entropy implementation that was used here is somewhat crude and thus we must not really focus on the details . the ctqmc results have different errors at different scale , i.e very precise at low @xmath63 , fluctuating at intermediate @xmath63 while at large @xmath63 the results are analytical . hence , we choose the weight of the entropy term based on heuristic considerations , depending on the real - frequency range we are interested in . as was noted previously@xcite , in ipt there are states in the mott gap at finite frequency , but their weight is small compared to the states everywhere else , namely near zero frequency and in the lower and upper hubbard band . overall , ipt with fixed @xmath0 compares well with ctqmc , but , at low temperature , what really matters is the region near @xmath48 . we thus zoom on this region in fig . [ fig : dos]-(b ) . there , we see that in the vicinity of @xmath48 , when @xmath128 , we are quite close to the ctqmc values . when we use the naive @xmath0 , the quasi - particle peak is shifted a little bit to the right and so is this why the low @xmath11 results are different even if the shape and values of the peaks are similar . we also compare the results for an integrated quantity , @xmath129 , that is obtained in general by solving @xmath130 with @xmath131 the fermi function and @xmath132 the density of states . in our case , this is a byproduct of the dmft calculation . no analytical continuation is involved . in fig . [ fig : mu_t ] we show the results for three densities . in fig . [ fig : mu_t](a ) and ( b ) we note that the numerical values obtained with the different methods differ by at most about 10% . the best results are for ipt-@xmath115 since the curves are qualitatively very similar to the ctqmc ones with a derivative quantitatively quite close for all @xmath11 . the absolute difference between ipt-@xmath115 and ctqmc is almost doping and temperature independent . the derivative with respect to temperature for both ipt-@xmath3 and ipt-@xmath107 is not as good . at high enough temperature , all methods give similar results . [ fig : mu_t](c ) shows the result closer to half - filling , comparing ctqmc with ipt-@xmath115 , the best ipt method . as already discussed , ipt-@xmath3 gives un - physical results in the vicinity of half - filling . in this section we compare internal energy and specific heat in ipt-@xmath0 with ctqmc . within ctqmc , energy can be calculated quite accurately with a reasonable number of matsubara frequencies . indeed , it was shown by haule@xcite that the kinetic energy @xmath133 is proportional to the average perturbation order @xmath134 for a given set of parameters . as already discussed , the double occupancy is calculated directly by ctqmc and thus the total energy is given by @xmath135 in general , for a many - body system , the energy is given by the thermal average , in the grand - canonical ensemble , of the hamiltonian . for the hubbard model we may write @xmath136g_k(i\omega_n ) . \end{split}\ ] ] for ipt-@xmath0 , we use directly the imaginary time expression . since we generally compute green s functions only for positive imaginary time , we use the equivalent expression @xmath137 where we used that for the fcc lattice @xmath138 . once the energy is calculated , the specific heat @xmath139 is given by @xmath140 . to obtain values for all temperatures , we have considered the fixed @xmath0 as the one given by the average over @xmath141 of the ctqmc results . this is what we called ipt-@xmath115 previously . we use ipt-@xmath0 for brevity . we present the comparison in fig.[fig : energy ] for different densities . note the change in color code ( see the legend ) . in fig.[fig : energy]-(a ) we display results for densities @xmath113 while in fig.[fig : energy]-(b ) , the density is above half - filling @xmath142 . the energy scale for the latter case is shifted by a filling dependent quantity @xmath143 so that the various fillings can be displayed on the same scale . in fig.[fig : energy]-(c ) and ( d ) we zoom on @xmath144 and @xmath145 respectively without energy shift to emphasize the differences between ipt-@xmath0 and ctqmc . comparing those differences in fig.[fig : energy]-(c ) with those in fig.[fig : energy]-(d ) , we see that the further we are from half - filling , the better the agreement . in fig.[fig : cv ] we plot the specific heat as predicted by ipt-@xmath0 . once again we consider separately @xmath113 in fig.[fig : cv]-(a ) and @xmath142 in fig.[fig : cv]-(b ) . the closer we are to half - filling , the lower the temperature at which the peak that signals the appearance of the fermi liquid regime appears . despite the strong coupling , the particle - hole asymmetry is noticeable . in fig.[fig : cv]-(c ) we can compare the ipt-@xmath0 and the ctqmc results for three densities @xmath146 , @xmath147 and @xmath148 . the ctqmc results for the energy are not completely smooth especially at low @xmath11 because of statistical errors in the data . thus , if we were to calculate the specific heat @xmath139 for ctqmc , we would , for many of the values of @xmath43 , have to be very careful before calculating the derivative . here , we just performed a simple derivative on the raw ctqmc data to verify the trend . from the results for the energy , fig.[fig : energy]-(a ) we already knew that even if the change of curvature in the coherent - incoherent region exists also for ctqmc , it is much smoother than the one obtained from ipt . this is apparent indeed in @xmath139 . nevertheless , the transition temperature for the coherent - incoherent crossover given by the position of the maximum is quite similar for both methods even if , for ctqmc , its exact value is hard to really pinpoint due to the statistical errors in the raw data . we end this subsection by noting that the inadequacy of ipt-@xmath3 also shows when the energy is calculated . one finds an increase of the energy with decreasing @xmath11 at low @xmath11 , which is of course non physical since this corresponds to a negative specific heat . , as calculated in appendix [ mc_function ] . ] as calculated by ipt-@xmath0 for different values of density : @xmath82 ( black ( @xmath83 ) ) , @xmath84 ( blue ( @xmath85 ) ) , @xmath86 ( red ( @xmath87 ) ) , @xmath88 ( green ( @xmath89 ) ) , @xmath91 ( cyan ( @xmath92 ) ) , @xmath93 ( magenta ( @xmath94 ) ) , @xmath95 ( brown ( @xmath96 ) ) and @xmath97 ( kaki ( @xmath98 ) ) . the resistivity are largest close to half - filling where they exhibit low coherence temperatures . ] analytical continuation of response functions obtained with ctqmc is in general very difficult . with ipt , one can calculate response functions by first analytically continuing the self - energy with pad ? approximants and using the real - frequency expressions in terms of spectral weight . in this section , we obtain results for the electrical resistivity @xmath149 within ipt-@xmath0 to verify whether they are physically sensible at strong coupling . with ipt-@xmath3 they are not : one obtains an insulator at finite filling on the electron - doped side . since vertex corrections vanish in single - site dmft , the conductivity can be obtained from @xmath150 where @xmath151 and @xmath152 . we have restored the lattice spacing @xmath153 to exhibit the units . because the self - energy is local , we can also replace the triple sum over @xmath32 by a one dimensional integral over an energy variable . @xmath154 where the so called transport function that includes the effect of the lattice is @xmath155 . for a simple cubic lattice in any dimension , the calculation of this function can be brought in the form of a one - dimensional integral@xcite , but on the fcc lattice this is not possible . if one wishes to use the expression eq . instead of keeping the full three dimensional integrals in eq . , @xmath156 must be calculated numerically . we explain in appendix [ mc_function ] an efficient way to do this . the result for the fcc lattice is shown in fig . [ fig : x_func ] . we have tested both ways of obtaining @xmath157 , i.e. eq . and eq . and they give essentially the same answer . they can not give exactly the same number because @xmath158 is calculated for a fixed number of points and we must thus interpolate between these points when performing the integral over @xmath159 . this adds another source of numerical error not present in eq . . but , contrary to the non - interacting density of states ( fig . [ fig : dos_u0 ] ) , @xmath158 is a smooth function ( fig . [ fig : x_func ] ) and thus the process of interpolation will only induce a negligible error . when the integrals are performed in the order shown in eq . , i.e. integrate over @xmath160 first , the resulting integrand for the @xmath159 integration is smooth and @xmath158 need not be obtained with extreme accuracy . for these reasons , for the conductivity we preferred to use eq . . using eq . increases dramatically the calculation time here contrary to @xmath127 or @xmath161 discussed in the next section . at low temperature and finite doping , the conductivity is proportional to @xmath162 . thus , if @xmath54 is such that its value is outside the non - interacting band @xmath163 , @xmath164 and thus @xmath165 . in an exact implementation of single - site dmft , a null conductivity can only happen at half - filling for @xmath49 . at any finite doping , there is a quasi - particle peak and luttinger s theorem is respected ( fig.[fig : fermi_ctqmc]-(a ) ) . as ipt-@xmath3 fails with respect to luttinger s theorem ( fig.[fig : fermi_ipt]-(a ) ) in that case @xmath149 start diverging at low @xmath11 for the densities with @xmath54 outside the band while it should exhibit the @xmath166 behavior of a fermi liquid . we show in fig.[fig : resis ] the result for ipt-@xmath0 . we see that it has the correct @xmath166 behavior at low temperature . for clarity , we omitted values of densities between @xmath167 and @xmath168 . nothing very different happens there . we still have the low temperature @xmath166 behavior and , as is already obvious from the figure , the absolute values of @xmath149 obtained are larger and larger when we approach half - filling , the mott insulating state . note that close to half - filling , even though there is a tendency for the resistivity to saturate at very high temperature @xcite , this occurs at values of the resistivity much larger than the mott - ioffe - regel limit @xmath169 . this is characteristic of incoherent transport in strongly correlated systems . and @xmath167 , as calculated from ipt-@xmath0 for three different temperatures using two analytical continuation approaches for each temperature . the maximum entropy results are represented with solid lines and the pad analytical continuations with dashed lines . the broadest zero - frequency peak ( red ( @xmath170 ) ) is for the largest temperature , @xmath171 , and the narrowest one ( black ( @xmath102 ) ) for the lowest temperature , @xmath100 . the intermediate case , @xmath172 is in blue ( @xmath74 ) . the features in the optical conductivity can be identified with transitions between the fermi level and peaks in the single - particle density of states . ] we analyzed the performance of our new solver ipt-@xmath0 for fermi surface properties at @xmath173 ( @xmath59,@xmath174,@xmath175 ) , for integrated quantities like chemical potential @xmath45 , energy , specific heat , resistivity and frequency - dependent functions such as the density of states . to finish , we look at the optical conductivity . appendix . [ appen_opt_conduc ] explains how it is calculated using the susceptibility @xmath161 in bosonic matsubara frequency and analytical continuation . as a first check , which does not depend on analytical continuation , we verify the @xmath176-sum rule @xmath177 ( see for example ref . [ ] ) . in the case of nearest - neighbor hopping , this quantity can be related to the kinetic energy . for a fcc lattice one finds @xmath178 where @xmath179 is the average kinetic energy . for ctqmc the average kinetic energy is obtained from @xmath180 where , as already mentioned , @xmath134 is the average perturbation order obtained directly from the monte - carlo simulation . to calculate @xmath181 one needs the dressed green s function , or equivalently the self - energy ( @xmath182 ) . a very large number of numerical operations is necessary , as explained in appendix . [ appen_opt_conduc ] , but analytical continuation is unnecessary . for all the tests we did , the ratio @xmath183 agreed with @xmath184 up to the third digit . for example , for @xmath126 , @xmath185 and @xmath66 the ratio is 0.6668 , while for @xmath186 it is 0.6664 . analytical continuation of ctqmc is problematic , especially since we have a very wide frequency range given the large value of @xmath7 . hence we display only results obtained with analytical continuation of ipt-@xmath0 and check for consistency with what is expected from the density of states . results for density @xmath167 , interaction @xmath187 and three temperatures is shown in fig . [ fig : opt_conduc ] . solid and dashed lines correspond respectively to maximum entropy @xcite and pad analytical continuation . there is some quantitative disagreement but the qualitative information is the same . at low temperature ( @xmath185 black lines ) there is a clear peak around @xmath188 which corresponds to transitions between the lower hubbard band and the quasi - particle peak appearing in the density of state for a similar density in fig.[fig : dos ] . at larger temperature ( blue lines @xmath189 ) , the decrease of the peak near @xmath188 in @xmath190 corresponds to the disappearance of the quasi - particle peak and loss of coherence . that loss of coherence for @xmath186 is signaled by the maximum in @xmath139 observed in fig . [ fig : cv ] . at an even larger temperature , ( red lines @xmath191 ) the @xmath188 peak has disappeared since we are now in the incoherent regime . the peak for transition to the upper hubbard band around @xmath192 is always visible . in addition to being numerically inexpensive , ipt provides a method where analytically continued results can be reliably obtained directly in real frequencies or from pad approximants@xcite instead of maximum entropy methods required when quantum monte carlo is used as an impurity solver . however , for large interaction strengths in doped mott insulators , the popular condition for ipt where one imposes @xmath4 fails at low temperature for a broader regime than previously expected@xcite . as a solution , we propose that one should instead enforce the exact relation eq . between double occupancy and single particle quantities . further improvements are expected if one also enforces the third moment of the spectral weight@xcite . our new method , ipt-@xmath0 , can be used for any coupling if @xmath0 is known . double occupancy @xmath0 can be obtained quite accurately by a number of methods and is negligibly dependent on temperature for large coupling . for example , in the strong coupling regime one can use exact diagonalization of small clusters or slave bosons@xcite while , at weak coupling , methods such as two - particle - self - consistent theory ( tpsc)@xcite give good results . also , in both regimes , quantum monte - carlo methods can be used . in the very large @xmath7 limit , the naive estimate @xmath122 for @xmath14 and @xmath193 for @xmath10 leads to qualitatively correct results , except very close to half - filling . our approach has been benchmarked on the fcc lattice for a number of observables . the large - particle hole asymmetry of that lattice survives in observable quantities even for interaction strength equal to twice the bandwidth . the authors thank dominic bergeron for the maximum entropy codes that we used for analytical continuation of the ctqmc data and for discussions . this work was partially supported by nserc ( l .- f.a . and a .- m.s.t . ) , the tier i canada research chair program ( a .- m . s. t. ) , and universit de sherbrooke . a .- m.s.t is grateful to the harvard physics department for support and p. smon for hospitality during the writing of this work . partial support was also provided by the mit - harvard center for ultracold atoms . simulations were performed using a code based on the alps library @xcite on computers provided by cfi , mels , calcul qubec and compute canada . portions of the hybridization expansion impurity solver developed by p. smon were inspired by the code gracefully provided by e. gull and p. werner . as explained in the main text , for accuracy in the dmft iteration we need to perform integrals over a three dimensional brillouin zone . in this appendix we use symmetry and ideas from gaussian quadrature , adaptive methods , and statistics , to devise an accurate and fast integrator . we first obtain a quadrature of order five and then explain how we can make it adaptive . we need a normalized triple integral over a cube of length @xmath194 centered at a point @xmath195 $ ] . if we put the origin at this point , the integral takes the form @xmath196 we considered a normalized integral because the integrals we need to solve are over k - space and thus need normalization . we first show that fourteen appropriately chosen points and only two weights can give us an approximation of order five . usually when one develops a gaussian quadrature , only the order is specified and the points and weights are obtained . in 3d , this may be very cumbersome so we start with points symmetrically placed and we will show that they give a good approximation . take points on each axis and on the diagonals of the cube , @xmath197 , @xmath198 , @xmath199 and @xmath200 as illustrated in fig . [ fig : cube ] . by symmetry , there are only two weights @xmath201 and @xmath202 . thus the integral is approximated by @xmath203\\ & + w_2f(\pm b , \pm b , \pm b ) . \end{split}\ ] ] to determine the numbers @xmath153 , @xmath204 , @xmath201 and @xmath202 we require that every polynomial of order five or less should be integrated exactly by this scheme . in 3d , this corresponds to many different polynomials but we have only four unknowns and apparently too many equations . this is where symmetry comes into play . first , since we integrate over a cube from @xmath205 to @xmath206 , all odd polynomials integrate to zero . also , for example , a polynomial of the form @xmath207 is equivalent to @xmath208 , or @xmath209 is equivalent to @xmath210 and @xmath211 and so on for every type of polynomials . thus we only have to consider four different polynomials i.e. @xmath212 , @xmath209 , @xmath213 and @xmath207 . + + taking @xmath214 , the integral gives one and thus we obtain the first equation @xmath215 taking @xmath216 , the integral gives @xmath217 and we obtain @xmath218 similarly , with @xmath219 , we have @xmath220 and we find @xmath221 finally , taking @xmath222 , @xmath223 and we obtain @xmath224 solving these equations , we obtain @xmath225 to make the method adaptive , we take our cube and split it in eight . if we take one of these cubes , and put the origin in its center we now have the integral over a cube from @xmath226 to @xmath227 centered at @xmath228 . we can thus use eq . but with @xmath229 and the points @xmath230 , @xmath231 , @xmath232 and @xmath233 . we can do this for each of the eight cubes obtaining the new approximation for the integral @xmath234 . the process can be repeated . each of the eight cubes can be subdivided again with integrals from @xmath235 to @xmath236 . when one subdivision has converged , this part is stopped . the calculation has converged when all subdivisions have converged . the convergence criterion requires a detailed discussion . assume that we aim at a relative error @xmath237 . for an adaptive method , we need absolute error . indeed , with a simple 1d adaptive integration method where one subdivides the interval in half , if one wishes an absolute error @xmath238 one usually imposes absolute error @xmath239 on each of the two sub - intervals . to determine the absolute error in our case , we first estimate the value of the integral by performing three subdivisions , i.e. using @xmath240 cubes . let us call the resulting integral @xmath241 . then we take for the absolute error needed for the adaptive integration method @xmath242 . if we imagine launching the integrator from scratch , one would ask for @xmath243 accuracy in each of the @xmath244 sub - cubes when we do a division . this often leads to a final answer that is more accurate than desired . for heavy numerical calculations it is desirable to optimize the choice of the error in each subinterval to minimize the computation time while maintaining the final desired accuracy . in our case , we claim that it suffices to require the absolute error within each subinterval to be @xmath245 instead of @xmath246 , as we might have naively expected . indeed , if we consider each value on the sub - cube as a random variable @xmath247 and want an error @xmath238 on the original cube i.e. on the sum @xmath248 , an error @xmath245 for each sub - cube suffices is we assume that the errors on the @xmath247 s are independent and uniformly distributed ( iud ) . indeed , in that case @xmath249 and thus @xmath250 . this hypothesis of an iud is of course not rigorous , but we have extensively tested this choice for the error with many different integrands with known integrals . by taking advantage of the fact that statistical hypothesis on the errors become reasonable since the integral is high dimensional , our approach is faster . we also checked that our approach is more precise and faster than using three adaptive 1d integrators . additional speedup can be obtained by taking into account the symmetry of the integrand . , see text . ] in this section we detail how we implemented ipt-@xmath0 . ipt as an approximative solver is fast , but it needs to also be implemented in the fastest possible way . in ipt , we are solving a system of two nonlinear equations with two unknowns : @xmath46 and @xmath45 . the equations are @xmath251 and @xmath252 where @xmath43 and @xmath0 are fixed numbers for a particular set of parameters . the self - energy must be calculated using eqs . , . we show how to do this efficiently . we first start with guesses for the hybridization function @xmath253 , and the two chemical potentials @xmath46 and @xmath45 . then we calculate the impurity model loop as shown in fig . [ fig : loop_ipt ] . once the loop has been converged , we use the self - energy to calculate the lattice green s function @xmath254 . with it , the new hybridization function can be calculated @xmath255 and finally @xmath33 , @xmath45 and @xmath46 are fed back to the impurity model loop . this is repeated until global convergence is reached . it must be specified here that @xmath33 is a function of both @xmath45 and @xmath46 but in the present algorithm , once @xmath33 is fed to the impurity loop , it is considered to be independent while @xmath45 and @xmath46 are iterated until we obtain the correct @xmath43 and @xmath0 . however , once the system is close to convergence , the difference between @xmath33 in the inner loop and the correct @xmath256 becomes really small and once convergence is reached , it is indeed the same function . this is also the approach that was adopted originally in [ ] . this approach is much faster than fixing @xmath45 and @xmath46 , converging the entire dmft calculation , calculating the new @xmath43 and @xmath0 , updating @xmath45 and @xmath46 and converging again the dmft calculation until we obtain the correct @xmath43 and @xmath0 . we have tested and used the two methods and they give the same results . the fast method can , for some particular parameter set , become unstable and thus , in these cases , the long , more rigorous method , may be used . to perform this calculation , we see from fig . [ fig : loop_ipt ] that we must calculate fourier transforms from matsubara frequencies to imaginary time and back . this must be calculated numerically and we now explain how to do it efficiently . the first necessary step is to calculate @xmath257 . we can show that the function @xmath253 for a one band model with a dispersion relation @xmath258 behaves asymptotically like @xmath259 , similar to what was previously obtained@xcite . for the 3d fcc lattice with nearest - neighbor hopping @xmath260 . we can use this to define @xmath261 that gives the asymptotic high - frequency behavior of @xmath262 . we need it because the fourier transform necessary to get @xmath257 must be approximated by a finite sum and thus the function must be convergent at least as fast as @xmath263 , while @xmath264 . we thus consider the function @xmath265 instead and add the missing terms analytically . + + @xmath266 in the first sum , imaginary time has been discretized in @xmath267 bins so that @xmath268 . there is then a maximum and minimum matsubara frequency . in the second equality of eq . , fft stands for fast fourier transform . the last term , @xmath269 can be calculated analytically using complex analysis . we find , for @xmath270 , @xmath271 where @xmath272 is the fermi function and @xmath273 . hence , in terms of fft s , we obtain @xmath274 we must remember that fft does not give the value at @xmath275 . we will come back to that point later . we also need @xmath276 . this can be done using the antiperiodic property or by using a procedure similar to @xmath257 . in that case , we obtain ( again for @xmath270 ) @xmath277 where @xmath278 with these two green s function we can calculate the second order contribution to the aim self - energy that appears in ipt eq . . in this equation , we need to perform an integral . to obtain an asymptotic behavior in matsubara frequencies that decays instead of being periodic , we can not do a direct integration@xcite . the trick here is to perform a cubic spline interpolation of @xmath279 and then fourier transform that spline interpolation . to obtain the spline , one needs two conditions to solve the system of equations . in our case , it suffices to find the derivatives at @xmath280 and @xmath281 . we will show later that making the fourier transform of the spline is really accurate and introduces a minimum of numerical errors . up to now , we have only considered @xmath282 and thus what we need are the derivative at @xmath283 and at @xmath284 . if , for the moment , we consider a paramagnetic system we can write @xmath285 . the derivative is thus @xmath286 we calculate the derivative of the green s functions from their definition @xmath287 by defining @xmath288 and @xmath289 , we obtain @xmath290}{\frac{1}{\beta}}\text{fft}(f_1(i\omega_n))\\ & + { \frac{z_1 ^ 2}{z_1-z_2}}f(-z_1)\text{e}^{-z_1\tau } + { \frac{z_2 ^ 2}{z_2-z_1}}f(-z_2)\text{e}^{-z_2\tau } \\ { \frac{dg_0(-\tau)}{d\tau } } & = \text{e}^{-i\pi j[1/n - 1]}{\frac{1}{\beta}}\text{fft}(f_2(i\omega_n))\\ & + { \frac{z_1 ^ 2}{z_1-z_2}}f(z_1)\text{e}^{z_1\tau } + { \frac{z_2 ^ 2}{z_2-z_1}}f(z_2)\text{e}^{z_2\tau}. \end{split}\ ] ] with these two equations , we can get the derivatives at @xmath280 . to obtain the green s function and its derivative at @xmath281 , we use the spectral representation of the green s function to show that @xmath291 , @xmath292 and @xmath293 . we now have everything we need to calculate the derivative of eq . for @xmath280 and @xmath281 . knowing @xmath294 and @xmath295 we can calculate the coefficients of the spline . we need the matsubara - fourier transform of functions represented by that spline on the right - hand side of @xmath296 let us call @xmath297 the piecewise cubic spline for @xmath298 . the method is presented in details in appendix e of [ ] . integrating by parts , the result for fermionic frequencies is @xmath299 where ifft is the inverse fast fourier transform . this result is what we needed since it has the correct high - frequency behavior where in principle the first three terms are exact while the last one is obtained from a numerical inverse fourier transform . since the latter is the coefficient of @xmath300 , errors do not adversely affect the high - frequency behavior . even though for the dmft iterations we found that the way to obtain accurate results was to use the adaptive method described in appendix [ appen_integrate ] , we show here how to calculate @xmath301 , the non - interacting density of states and @xmath158 . the latter quantity appears in calculations of the conductivity and transport properties in general and it is in this context that we used the results presented here . both quantities have the general form @xmath302 . since our band structure is not simple , we can not perform the integral analytically . the question is thus , how do we treat the delta function in a numerical calculation ? + + a simple approach would be to replace the delta function by a lorentzian and perform the integral using an adaptive scheme . but , this approximation for the delta function gives tails at the edges of the band . a monte carlo scheme is preferable not only because the sharpness of the delta function is maintained without tails , but also because one does not need to do a triple integral for each value of @xmath159 : the complete function of @xmath159 is be obtained at once . + + we first choose for how many energy @xmath159 points we want to know the function . this number defines a number of bins , shown as dashed lines in fig . [ fig : mc_scheme ] . one then generates a random point @xmath303 , calculates @xmath258 and locates the bin where this number belongs . for example , in fig . [ fig : mc_scheme ] , the random @xmath258 belongs to the bin @xmath43 . we then add to this bin @xmath304 . this is equivalent to approximating the delta function by a rectangle of finite width . we continue this process @xmath305 times and divide , at the end , the numbers in the bins by @xmath305 . the function is thus given by the numbers in the bins , each bin corresponding to a particular energy @xmath159 . accuracy and smoothness can be improved by increasing the number of random points . for the optical conductivity , we need the current - current correlation function in bosonic matsubara frequencies . in dmft , vertex corrections vanish@xcite , hence we have @xmath306 where @xmath307 are bosonic matsubara frequencies while @xmath63 are fermionic . to compute this , we again use the convolution theorem and fft , as described for ipt in appendix [ appen_ipt ] . the above equation can be written as @xmath308 the only difference with eq . is that here we have bosonic frequencies . using again the cubic spline trick we obtain an expression similar to eq . . for @xmath309 @xmath310 while for @xmath311 @xmath312 . \end{split}\ ] ] with @xmath313 , @xmath314 , @xmath315 and @xmath316 the coefficients of the cubic spline . we also need to go from matsubara frequencies to imaginary time for @xmath317 and for @xmath318 in eq . . we proceed to obtain the analog of eq . . @xmath319 has a similar structure since the asymptotic behavior of the self - energy is @xmath320 . this time the poles are at @xmath321 where @xmath322 . + + once again , the values at @xmath323 are found using the spectral representation of @xmath324 . the expressions are @xmath325 @xmath326 @xmath327 @xmath328 we can thus calculate @xmath318 appearing in eq . and its first derivative at @xmath280 and @xmath281 to find the cubic spline interpolation . the sum over wave vectors @xmath32 requires some comments . like for the dmft calculation , we use the adaptive scheme in appendix [ appen_integrate ] . however , if we look at @xmath318 in eq . , in principle for each @xmath329 we need to perform the integral over @xmath32 independently , but the whole point of fft is to obtain all @xmath329 points at the same time . our solution is to launch the integrator for all @xmath329 at the same time and keep in memory the estimate of the integral for each @xmath329 while we refine the estimate . once the integrator converges for @xmath330 , we conclude that this @xmath32 grid is the one for all @xmath329 and stop the calculation . since we keep all values in memory , we have the function for all @xmath329 calculated for the grid in @xmath32 space appropriate for @xmath330 . we verified that by converging the calculation for other values of @xmath329 we obtain the same answer . we have also tried the other way around , where we interchange the integral over @xmath32 and the fourier transform in eq . . in this case , we converge the zero frequency and , once again , the results are essentially the same . once we have calculated @xmath331 , we need to obtain the real frequency representation since the optical conductivity is given by @xmath332 . in the case of ctqmc , we use the maxent analytical continuation scheme developed in bergeron _ [ ] for the conductivity . for ipt results we use both pad ? and maxent to find @xmath333 .
consider a finite - parameter analytic family of analytic plane vector fields @xmath4 such that @xmath0 has a limit periodic set @xmath5 . the cyclicity of @xmath5 is , roughly speaking , the maximal number of limit cycles of @xmath1 which tend to @xmath5 as @xmath6 . the roussarie s finite cyclicity conjecture claims that _ every limit periodic set occurring in an analytic finite - parameter family of plane analytic vector fields , has a finite cyclicity _ if true , the conjecture would imply the finitness of the maximal number @xmath7 of the limit cycles , which a plane polynomial vector field of degree @xmath8 can have . therefore it plays a fundamental role in all questions related to the second part of the 16th hilbert problem and its ramifications . recall that a polycycle of the vector field @xmath0 is a topological polygon composed of separatrices and singular points . a @xmath9-saddle cycle of @xmath0 ( or a hyperbolic @xmath9-graphic ) denoted @xmath10 , is a polycycle composed of @xmath9 distinct saddle - type singular points @xmath11 , @xmath12 and separatrices ( heteroclinic orbits ) connecting @xmath13 to @xmath14 as on fig . [ dessin9 ] . @xmath9-saddle cycles , period orbits and weak foci or centers are the simplest limit periodic sets . the finite cyclicity of period orbits and weak foci is well known and follows from the gabrielov theorem @xcite . the finite cyclicity of one - saddle loops is due to roussarie @xcite . _ the purpose of the present paper is to prove the finite cyclicity of a two - saddle cycle , under finite - parameter analytic deformation _ , see theorem [ mainth ] . several special cases of this result were earlier proved , under different genericity assumptionss either on @xmath0 or on the family @xmath1 , by cherkas , mourtada , el morsalani , dumortier , roussarie , rousseau , jebrane , oadek , li , caubergh , luca and others @xcite , see also @xcite for survey of the results and references up to 1996 . the finite cyclicity of a k - saddle cycle ( any k ) , under finite - parameter analytic deformation was recently announced by mourtada @xcite . for _ generic families _ of vector fields the analitycity can be relaxed . as it is shown by ilyashenko and yakovenko @xcite , kaloshin @xcite _ any nontrivial elementary polycycle occurring in a generic @xmath9 - parameter family of @xmath15 vector fields has finite cyclicity_. in contrast to the aforementioned papers we shall not use the asymptotic expansions of the corresponding dulac maps . instead of this , we evaluate the number of the limit cycles near @xmath16 in a complex domain , by making use of a suitable version of the argument principle . this approach was initiated by the author in @xcite , where we studied cyclicity of hamiltonian two - loops . as it is well known , the limit cycles of planar systems close to hamiltonian are closely related to the zeros of associated abelian integrals depending on a parameter ( the so called weakened 16th hilbert problem ( * ? ? ? * arnold , p.313 ) ) . zeros of complete elliptic integrals were successfully studied by topological arguments in a complex domain ( the argument principle ) after the pioneering work by g. s. petrov @xcite , see also oadek @xcite for a description of the method . it has been used in a more general context in several papers , e.g. @xcite , and in @xcite the idea has been used to replace abelian integrals by the true poincar return map . in the present paper we shall find a relation between the fixed points of the poincar first return map and the fixed points of holonomies of the separatrices of the saddle points , which correspond to complex limit cycles . to count such fixed points is a question on the zeros of families of _ analytic functions _ which is easily solved . the main technical tool is lemma [ mainlemma ] in which we show that the connected components of the zero locus of the imaginary part of a dulac map are smooth semianalytic curves . this allows to estimate the variation of the argument of the displacement map along the border of an appropriate complex domain , and finally to apply the argument principle in order to evaluate its zeros in the domain . note that previously the relation between the monodromy and the dulac map was used by roussarie to compute the bautin ideal associated to the dulac map @xcite . this combined with @xcite also implies the finite cyclicity of one - saddle cycles . the paper is organized as follows . in section [ dulacsection ] we provide the necessary technical background , and prove the main technical lemma [ mainlemma ] . in section [ onesaddle ] , we give a new self - contained proof of roussarie s theorem about the finite cyclicity of one - saddle cycles . the origin of our method is then explained in section [ petrovtrick ] where we give a brief account of a local version of the so called `` petrov trick '' . the same method is then easily adapted in section [ mainsection ] to show that the cyclicity of @xmath16 is finite . consider an analytic family of plane real analytic foliations @xmath17 , @xmath18 , having a non - degenerate isolated saddle point . an appropriate translation analytically depending on @xmath3 will place the saddle point at the origin . the foliation @xmath17 has two analytic separatrices , transversally intersecting at the saddle point , and depending analytically on @xmath3 @xcite . therefore a further real bi - analitic change of the variables @xmath19,analytically depending on @xmath3 , will identify them to the axes @xmath20 and @xmath21 as on fig.[fig1a ] , so @xmath22 where the dots replace higher order terms in @xmath19 with coefficients depending on @xmath3 . the number @xmath23 is the hyperbolic ratio of the saddle point . from now on we shall suppose that the foliation ( [ foliation ] ) is analytic and depends analytically in @xmath3 in a neighborhood of the origin in @xmath24 . for @xmath25 sufficiently small , let @xmath26 , @xmath27 be open complex discs centered at @xmath28 and @xmath29 , parameterized by @xmath30 and @xmath31 respectively . the ( real ) dulac map is the germ of analytic map at @xmath32 @xmath33 defined as follows : if @xmath34 then @xmath35 is the intersection with @xmath36 of the orbit @xmath37 of ( [ foliation ] ) , passing through @xmath30 , see [ fig1a ] ( ii ) . this geometric definition of @xmath38 allows to control to a certain extent its analytic continuation in a complex domain . the dulac map allows an analytic continuation on some open subset of the universal covering space @xmath39 of @xmath40 , depending on @xmath3 . let us parameterize @xmath39 by polar coordinates @xmath41 , @xmath42 . the following result is well known ( e.g. ( * ? ? ? * appendix a ) ) [ continuation ] there exists @xmath43 and a continuous function @xmath44 such that the dulac map allows an analytic continuation in the domain @xmath45 the geometric content of theorem [ continuation ] is as follows . let @xmath46 be a continuous family of paths contained in the leaves of @xmath17 , and connecting @xmath47 to @xmath48 . for @xmath49 we suppose that @xmath50 is the real orbit of @xmath17 contained in the first quadrant @xmath51 , and connecting @xmath52 to @xmath48 , see fig.[fig1a ] ( ii ) . the above theorem claims that this family of orbits allows an extension to a continuous family of paths @xmath46 , contained in the leaves of @xmath17 , and connecting @xmath53 to @xmath54 . the family is defined for all @xmath55 which belong to the domain ( [ domain ] ) . each path starts at @xmath52 and terminates at a unique point on @xmath56 , denoted @xmath57 . although the paths @xmath46 are not unique , their relative homotopy classes are uniquely defined . to the axes @xmath58 parameterized by @xmath31 and @xmath30 , we associate holonomy maps @xmath59 defined by two closed paths contained in the axes @xmath60 and @xmath61 and based at @xmath28 , @xmath29 respectively . we shall make the convention , that each closed path makes one turn around the origin of the axe in which it is contained , in a positive direction ( recall that the axes are parameterized by @xmath31 and @xmath30 respectively ) . it is easily seen that in the case of a linear foliation of the form @xmath62 we have @xmath63 in the general case of a nonlinear foliation of the form ( [ foliation ] ) the dulac map @xmath38 is only asymptotic to @xmath64 , while the holonomy maps are analytic in @xmath65 and @xmath66 the dulac map @xmath38 is a transcendental multi - valued map . for @xmath67 let @xmath68 be the result of the analytic continuation of @xmath38 along an arc of radius @xmath30 and angle @xmath69 . similarly , for @xmath70 let @xmath71 be the result of the analytic continuation of @xmath38 along an arc of radius @xmath31 and angle @xmath69 . [ monodromy ] for every sufficiently small @xmath67 , @xmath70 , @xmath72 holds @xmath73 * proof . * consider , instead of @xmath38 the underlying path @xmath74 . the loop @xmath75 has the same origin as @xmath76 so they can be composed and the resulting loop @xmath77 starts at @xmath78 and terminates at @xmath79 . in the special linear case ( [ linear ] ) with @xmath80 the foliation is a fibration , the paths @xmath81 represent relative cycles in the fibers of @xmath82 , and the path @xmath77 is closed and represents a vanishing cycle . the claim of lemma [ monodromy ] is then the classical picard - lefschetz formula . in the general case the result follows `` by deformation '' . indeed , in the linear case ( [ linear ] ) with @xmath80 the family of closed paths @xmath83 is defined for all sufficiently small @xmath31 , and @xmath84 is a closed path which makes one turn around the origin in the axe @xmath21 in a positive direction . note that the paths @xmath85 are bounded away from the origin in @xmath86 . it follows that @xmath87 defines the holonomy @xmath88 of the separatrix @xmath21 , and this property holds true also for every sufficiently small deformation of ( [ linear ] ) . the homothety @xmath89 transforms ( [ foliation ] ) to a small deformation of ( [ linear ] ) which completes the proof of the first identity ( but see also @xcite ) . the second identity in lemma [ monodromy ] is proved in a similar way . @xmath90 consider the universal covering @xmath91 and the zero locus @xmath92 of the imaginary part of the dulac map @xmath38 corresponding to the domain ( [ domain ] ) @xmath93 in the case of a linear foliation ( [ linear ] ) the zero locus is therefore a union of half - lines : @xmath94 to describe @xmath95 in the case of a general foliation of the form ( [ foliation ] ) , with hyperbolic ratio @xmath96 , consider the germs of real analytic sets at the origin in @xmath97 @xmath98 where @xmath99 is the holonomy map associated to the separatrix @xmath20 . [ mainlemma ] the zero locus @xmath100 of the imaginary part of the dulac map in the domain ( [ domain ] ) is a union of connected components @xmath101 , indexed by @xmath102 . * each set @xmath103 , ( [ clk ] ) , is a germ of a real analytic curve of @xmath104 , which is smooth at the origin and tangent to the line @xmath105 there . * each connected component @xmath101 is projected on the plane @xmath106 under the map @xmath107 ( [ universal ] ) to the connected component of @xmath108 tangent to the half - line ( [ line ] ) , @xmath109 , at the origin . for a general bi - holomorphic map @xmath99 , vanishing at the origin , the set @xmath110 coincides with the origin itself . the above lemma shows , however , that for the monodromy map @xmath99 of a saddle point of a real - analytic plane vector field , the set @xmath103 , ( [ clk ] ) , is a germ of a real analytic curve of @xmath104 , which is smooth at the origin . the position of the connected components of @xmath108 tangent to the half - lines ( [ line ] ) , @xmath109 ( [ line ] ) , @xmath109 is shown on fig . [ dessin1 ] . the above lemma is the main technical result of the present paper . the analyticity of the zero locus @xmath95 is responsible for the algebraic - like behavior of the dulac map . + * proof of lemma [ mainlemma ] * let @xmath111 and suppose that for some @xmath112 , @xmath113 . as the dulac map is real along @xmath114 , then @xmath115 is complex conjugate to @xmath116 and hence @xmath117 if the point @xmath118 is seen as the inverse image of @xmath119 with respect to the dulac map @xmath120 , then the point @xmath121 is the result of the analytic continuation of the map @xmath120 along a suitable closed path of @xmath48 , starting and terminating at @xmath122 . if we put @xmath123 then @xmath124 are two values of the multivalued map @xmath125 and hence , by lemma [ monodromy ] , they differ by a power of the monodromy @xmath99 @xmath126 or equivalently @xmath127 clearly every such relation will correspond to a connected component @xmath101 of @xmath95 . as @xmath101 is an analytic set of real dimension one , then @xmath103 is an analytic set of dimension one too . it can be defined therefore by each of the following equivalent relations @xmath128= { \operatorname{re } } ( \bar{z } ) \}\ ] ] or @xmath129= { \operatorname{im}}(\bar{z } ) \}.\ ] ] as @xmath130 = -1 $ ] , then the linear part of the complex - analytic function @xmath131 can not be identically zero , and therefore @xmath132 is a real analytic curve , smooth at the origin . it follows from ( [ nonlinear ] ) that the projection of @xmath101 under @xmath107 on the plane @xmath106 is tangent to the half - line ( [ line ] ) , @xmath109 at the origin . @xmath90 let @xmath133 be a relatively compact domain , with piece - wise smooth boundary , and @xmath134 an analytic function which allows a continuation to the closure @xmath135 . denote by @xmath136 the number of the zeros of @xmath137 in @xmath138 , counted with multiplicity . if we assume that @xmath137 does not vanish along the border @xmath139 , then the increment of the argument @xmath140 of @xmath137 along @xmath141 oriented counter - clockwise is well defined . @xmath140 equals the winding number of the curve @xmath142 about the origin and the classical argument principle states that @xmath143 more generally , if @xmath137 has zeros on @xmath139 , isolated or not , the variation of the argument @xmath140 might be not well defined . we say that @xmath144 is a regular zero of @xmath137 if @xmath145 , and @xmath137 allows an analytic continuation in a neighborhood of @xmath52 in @xmath146 . if we assume that @xmath137 has only regular zeros in @xmath135 , then @xmath140 is well defined as a sum of the increments of the argument of @xmath147 between consecutive zeros of @xmath137 . indeed , the increments are finite , because the border @xmath148 is piece - wise smooth . the argument principle can be reformulated as follows [ argument ] let @xmath149 be a relatively compact domain with piece - wise smooth boundary . if @xmath150 is a continuous function , analytic in @xmath151 , and having only regular zeros in @xmath135 , then @xmath152 * proof . * there always exists a polynomial @xmath153 , such that @xmath154 has no zeros in @xmath135 , so we need to verify ( [ argument2 ] ) for polynomials only . the set @xmath151 is open , connected and oriented , with piece - wise smooth boundary , which therefore has no self - intersections and has an induced orientation . the inequality @xmath155 allows to `` remove '' the zeros along @xmath139 and hence formula ( [ argument1 ] ) implies ( [ argument2 ] ) . @xmath90 in the present paper the first inequality in ( [ argument2 ] ) will be used to bound the number of the zeros @xmath156 . for this we shall need estimates on the variation of the argument @xmath157 along any compact segment @xmath158 of a curve . more precisely , let @xmath159 be a compact segment of a smooth real analytic curve . let @xmath160 be an open set containing @xmath158 and @xmath161 , @xmath162 , be a germ of a family of complex - analytic functions in @xmath163 at @xmath164 . for every fixed @xmath3 such that the function @xmath165 is not identically zero , the variation of its argument @xmath166 is well defined . [ variation ] let @xmath158 be a compact segment of a real analytic curve and let @xmath167 be a family of functions analytic in a neighborhood of @xmath158 , and depending analytically in @xmath3 . there exists @xmath43 , such that @xmath168 the above result follows from the following theorem due to gabrielov @xcite let @xmath169 be real analytic varieties and consider the canonical projection @xmath170 . for every relatively compact semianalytic set @xmath171 , the number of the connected components of the pre - images @xmath172 is bounded from above uniformly over @xmath173 . * proof of theorem [ variation ] . * the number of the isolated zeros of @xmath165 along @xmath158 counted with multiplicity is uniformly bounded in @xmath3 at @xmath164 ( franoise - yomdin theorem @xcite ) . on an interval between two zeros of @xmath174 the variation of the argument divided by @xmath175 is bounded by the number of the zeros of the imaginary part of @xmath165 divided by two , plus the sum of the multiplicities of the zeros of @xmath165 at the end of the interval . the imaginary part of @xmath165 is a real analytic function in @xmath176 and the gabrielov theorem implies that the number of the connected components of @xmath177 is uniformly bounded in @xmath3 at @xmath164.@xmath90 let @xmath1 , @xmath2 be a germ of an analytic family of analytic plane vector fields , such that @xmath0 has a one - saddle cycle ( homoclinic saddle loop ) @xmath178 . the first - return map associated to @xmath178 is a composition of a dulac map @xmath179 and a transport map @xmath180 , see fig . [ dessin2 ] . we assume that the dulac map is in a normal form as in section [ scontinuation ] . the limit cycles of @xmath1 near @xmath178 correspond to the zeros of the displacement map @xmath181 near @xmath182 . an appropriate choice of the local coordinates on the croos - sections @xmath56 and @xmath48 brings the transport map to the form @xmath183 . alternatively , we could choose simply @xmath184 ( without supposing that the dulac map is in the normal form of section [ scontinuation ] ) . we shall bound the number of the zeros of @xmath165 in the domain @xmath185 delimited by the circle @xmath186 , and the connected components @xmath187 and @xmath188 of the zero locus of the imaginary part of the dulac map , as it is shown on fig . [ dessin4 ] . we shall suppose that @xmath189 is so small , that @xmath174 is analytic in @xmath190 for all @xmath191 , such that @xmath192 ( theorem [ continuation ] ) and it is analytic even on the closure of @xmath190 except of course at @xmath182 , where @xmath174 is only continuous . indeed , @xmath193 while @xmath180 is holomorphic at @xmath182 , so @xmath194 where @xmath195 is analytic and @xmath196 . if the family of functions @xmath165 is sufficiently general , then @xmath197 , and in the case when @xmath198 we may replace @xmath165 by the new family @xmath199 , @xmath200 , for which the limit at @xmath182 is the parameter @xmath201 . after this preparation , we may prove the finite cyclicity of the homoclinic loop @xmath178 . for this we apply proposition [ argument ] ( the argument principle ) to the family of functions @xmath165 in the domain @xmath190 . in the course of the computation , it will be supposed that @xmath189 is sufficiently small , @xmath202 is sufficiently small with respect to @xmath203 , and @xmath3 is such that @xmath204 . we may encode this choice of the parameters by the `` physical '' notation @xmath205 the hyperbolic ratio of the saddle point will be not bigger than one only in a suitable semi - analytic set in the parameter space , and will be bigger than one in another ( complementary ) semi - analytic set . after eventual exchanging of @xmath56 and @xmath48 , it will be also supposed that the hyperbolic ratio of the saddle point is not bigger than one for all parameter values . along the circle @xmath206 with angle close or strictly less than @xmath175 the variation of the argument of @xmath165 is uniformly bounded in @xmath3 ( theorem [ variation ] ) . along the curve @xmath207 the imaginary part of @xmath165 equals the imaginary part of the transport map @xmath208 . therefore the zeros of @xmath209 along @xmath207 are exactly the intersection points of @xmath207 and the segment @xmath210 . according to lemma [ mainlemma ] we have @xmath211 as @xmath212 is an analytic family of analytic functions , then by gabrielov s theorem , the number of such fixed points is uniformly bounded in @xmath3 on @xmath213 $ ] . to conclude , we have only to check that the family @xmath167 has regular zeros along the border of the domain @xmath190 . this is indeed the case , when @xmath214 , as @xmath215 . we conclude that the number of isolated zeros of the family of functions @xmath216 in the domain @xmath190 is uniformly bounded by some integer , say @xmath217 . finally , note that the condition @xmath218 can be removed . indeed , if for some @xmath219 , @xmath220 , @xmath221 , the function @xmath222 has at least @xmath223 zeros in @xmath190 , then it has at least @xmath223 zeros in @xmath190 in a sufficiently small neighborhood of @xmath219 , in contradiction with the preceding estimate . to resume , we proved the following classical result + * theorem ( roussarie@xcite ) * _ every homoclinic saddle loop ( a one - saddle cycle ) occurring in an analytic finite - parameter family of plane analytic vector fields , may generate no more than a finite number of limit cycles within the family . _ let us note that our method , exactly as the roussarie s theorem allows to compute more precisely the cyclicity of @xmath178 . we shall not enter into details here . just to illustrate this , note that if the hyperbolic ratio @xmath224 is strictly bigger than one , then the overall increase of the argument of the displacement map along the border of @xmath190 is strictly less than @xmath175 ( this computation is omitted ) and the cyclicity of @xmath178 is zero . the content of this section is not necessary for the proof of our main result theorem [ mainth ] , but it aims to shed some light on the origin of the method , used to bound the limit cycles near the saddle loop in the preceding section . with the same notations as in section [ onesaddle ] , consider the analytic family of analytic vector fields @xmath225 defining a holomorphic foliation @xmath17 of the form @xmath226 where @xmath227 is a function and @xmath228 is an analytic family of differential one - forms , both analytic in a neighborhood of the saddle loop @xmath178 . for definiteness , we put the saddle point at the origin in @xmath104 , so @xmath229 . we shall further suppose that the saddle loop @xmath178 is contained in the level set @xmath230 , and the interior of @xmath178 is filled up by a continuous family of periodic orbits @xmath231 , parameterized by @xmath232 , where @xmath233 is the restriction of @xmath227 on the cross - section @xmath56 . the displacement map is approximated by the usual poincar - pontryagin formula as follows @xmath234 where @xmath235 tends to zero as @xmath3 tends to zero , uniformly in @xmath236 in every compact interval in which the displacement map is defined . the zeros of @xmath174 correspond to limit cycles and , at least far from @xmath237 , they are approximated there by the zeros of the complete abelian integral @xmath238 we make the assumption ( actually justified by the roussarie s theorem @xcite ) , that this is so also in a neighborhood of @xmath239 ( corresponding to limit cycles close to the saddle loop @xmath178 ) . thus , it makes a sense to prove the finiteness of the maximal number of the zeros of the abelian integral @xmath240 , which tend to @xmath241 as @xmath242 tends to the origin in the parameter space . this follows of course from a well known general result of varchenko and khovansky . we shall use , however , a different idea due to g.s . petrov @xcite , who showed that the analogous global problem for complete elliptic integrals of second kind is of algebraic nature . this observation has been used in several papers by petrov to evaluate the precise number of zeros of complete elliptic integrals , and hence of limit cycles of perturbed hamiltonian vector fields , see for instance oadek @xcite . we are ready to describe the local version of the petrov method . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ consider the sector @xmath243 for a fixed sufficiently small @xmath189 and all sufficiently small @xmath244 the abelian integral @xmath245 allows an analytic continuation in @xmath246 . to bound the number of its zeros on @xmath246 ( and hence on @xmath247 ) we apply the argument principle to the domain @xmath246 . along the circle @xmath248 the increase of the argument of @xmath249 is bounded uniformly in @xmath3 ( due to gabrielov s theorem ) . along the segment @xmath213 $ ] the abelian integral allows two analytic continuations @xmath250 . as @xmath251 is real - analytic on @xmath247 then @xmath252 and by the picard - lefschetz formula @xmath253 where @xmath254 is a continuous family of cycles , vanishing at the origin as @xmath255 tends to zero . the imaginary part of @xmath240 on @xmath210 is therefore an analytic function , and by gabrielov s theorem again , its zeros are uniformly bounded in @xmath3 on the closed interval @xmath213 $ ] . this implies that the increase of the argument of @xmath240 on @xmath210 is also uniformly bounded in @xmath3 which combined to the argument principle shows the finiteness of the maximal number of zeros _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the proof of the finite cyclicity of the one - saddle loop from the preceding section , may be seen as a generalization of the petrov method . indeed , the picard - lefschetz formula corresponds to the claim of lemma 1 , and by lemma 2 the zeros the analytic abelian integral ( [ impart ] ) correspond to the fixed points ( complex limit cycles ) of the holonomy map @xmath99 of the separatrix . as it is well known , the holonomy map of a separatrix is analytic , which implies the finite cyclicity of the saddle loop @xmath178 . the main result of the paper is the following [ mainth ] every heteroclinic saddle loop ( a two - saddle cycle ) occurring in an analytic finite - parameter family of plane analytic vector fields , may generate no more than a finite number of limit cycles within the family . using the notations used in the preceding sections , suppose that the vector field @xmath0 has a two - saddle loop @xmath16 . consider the dulac maps @xmath256 associated to the corresponding foliation , as on fig.[fig4 ] . each map @xmath257 is a composition of a `` local '' dulac map ( as in section [ dulacsection ] ) and two real - analytic transport maps . from this it follows that lemma [ mainlemma ] applies to @xmath257 , @xmath258 , too . from now on we choose a real - analytic local variable @xmath52 on the cross - section @xmath56 thus identifying @xmath56 to an open disc centered at @xmath259 . we shall also suppose that @xmath260 . the functions @xmath261 , @xmath258 are multivalued on the cross - section @xmath56 and have critical points at @xmath262 , @xmath263 , respectively . the functions @xmath264 are real - analytic . the limit cycles of @xmath1 near @xmath16 correspond to the zeros of the displacement map @xmath265 near @xmath182 . let @xmath266 , @xmath258 be the hyperbolic ratios of the saddles . we shall suppose , upon exchanging eventually the roles of @xmath56 and @xmath48 , that @xmath267 . denote the zero loci of the imaginary parts of the dulac maps @xmath268 , @xmath269 by @xmath270 and @xmath271 respectively . we shall bound the number of the zeros of @xmath165 in the complex domain @xmath272 of the universal covering of @xmath273 defined as follows ( without loss of generality we assume that @xmath274 ) . * if @xmath275 , the domain @xmath272 is bounded by the circle @xmath276 and by @xmath277 as it is shown on fig.[dessin5 ] . * if @xmath278 then necessarily @xmath279 . the domain @xmath272 is bounded by the circle @xmath246 , by the interval @xmath280 $ ] , and by @xmath281 , as it is shown on fig.[dessin6 ] . in the course of the proof the parameters @xmath203 and @xmath3 will be chosen as in the one - saddle case : the constant @xmath203 will be sufficiently small , @xmath43 will be sufficiently small with respect to @xmath203 , and @xmath282 will be such that @xmath283 , see ( [ constants ] ) . like in section [ onesaddle ] we shall suppose , without loss of generality , that the analytic functions @xmath284 where @xmath285 are not identically zero . this will guarantee that for generic values of @xmath3 the displacement map will have only regular zeros in the closure of @xmath272 , so the argument principle ( proposition [ argument ] ) can be applied . * proof of theorem [ mainth ] . * it follows from the definition of the domain @xmath286 that the displacement map @xmath161 is analytic there . to count the zeros ( corresponding to real and complex limit cycles ) of the displacement map in @xmath287 we apply proposition 1 ( the argument principle ) to the family of functions @xmath165 . to evaluate the variation of the argument of the displacement map along the border of @xmath272 we repeat the arguments of section [ onesaddle ] . consider first the case @xmath275 , fig.[dessin5 ] . the connected component of the zero locus of the imaginary part of @xmath288 which is tangent to the line @xmath289 through @xmath290 intersects the circle @xmath246 transversally , and along this circle the variation of the argument of @xmath165 is uniformly bounded in @xmath3 ( theorem [ variation ] ) . the imaginary part of @xmath161 restricted to @xmath270 equals the imaginary part of @xmath291 and hence @xmath292 vanishes along @xmath293 exactly at the intersection points @xmath294 according to lemma [ mainlemma ] these intersection points are the solutions of the equation @xmath295 where @xmath296 are the holonomies of the separatrices intersecting @xmath56 and related to the saddle points @xmath297 and @xmath290 . by gabrielov s theorem , the number of such fixed points is uniformly bounded in the disc @xmath298 . consider now the second case @xmath278 , @xmath279 , see fig.[dessin6 ] . along this circle @xmath246 the variation of the argument of @xmath165 is uniformly bounded in @xmath3 ( theorem [ variation ] ) . along the interval @xmath280 $ ] the imaginary part of @xmath299 vanishes identically , and the imaginary part of @xmath161 restricted to this interval equals the imaginary part of @xmath291 . therefore the zeros of @xmath209 along @xmath280 $ ] are exactly the intersection points of @xmath300 and @xmath280 $ ] . by lemma [ mainlemma ] , and like in ( [ 1loop ] ) , these intersection points are the solution of the equation @xmath301 where @xmath302 is the holonomy of the separatrix intersecting @xmath56 and related to the saddle points @xmath290 . by gabrielov s theorem , the number of such fixed points is uniformly bounded . finally , the zeros of @xmath209 along @xmath303 and @xmath304 are evaluated as in the case @xmath275 . this completes the proof of theorem [ mainth].@xmath90 the identity ( [ 8loop ] ) which determines complex limit cycles `` responsible '' for the cyclicity of the double loop @xmath16 is the main new ingredient in the proof with respect to the one - saddle case . indeed , solutions of ( [ 8loop ] ) are fixed points of the holonomy @xmath305 which , for @xmath164 , is generated by a closed loop @xmath306 contained in the complexified separatrix of @xmath16 intersecting the cross - section @xmath56 . the topological type of this separatrix near @xmath16 is a disc with two punctures , corresponding to the two saddle points @xmath307 and @xmath308 . clearly @xmath306 makes one turn around each of them , but depending on the orientation we have two possibilities shown on fig . [ dessin8 ] ( i ) and ( ii ) . a simple computation on a model example shows that the loop @xmath306 associated to the holonomy @xmath305 is the figure eight - loop on fig . [ dessin8 ] ( i ) . the reader will recognize in the loop @xmath306 a key ingredient in the proof of the local boundedness of the number of zeros of pseudo - abelian integrals in @xcite . although the result of theorem [ mainth ] is existential , the proof we use leads to effective upper bounds on the number of the bifurcating limit cycles . this possibility is explored in @xcite , where we show that the cyclicity of a hamiltonian two - loop is bounded by the number of the zeros of _ a pair _ of associated abelian integrals , a phenomenon which also explains the appearance of alien limit cycles in @xcite .
dense self - avoiding polymers in two dimensions ( 2d ) have been considered theoretically @xcite , by means of computer simulation @xcite and more recently even in real experiments @xcite . it is now generally accepted @xcite ( with the notable exception of refs . @xcite and @xcite ) that these chains adopt compact and segregated conformations at high densities , i.e. as first suggested by de gennes @xcite the typical chain size @xmath19 scales as @xmath20 with @xmath21 being the chain length , @xmath22 the monomer number density and @xmath0 the spatial dimension . we assume here that monomer overlap and chain intersections are strictly forbidden @xcite . compactness and segregation are expected to apply not only on the scale of the total chain of @xmath21 monomers but also to subchains comprising @xmath13 monomers @xcite , at least as long as @xmath13 is not too small . the typical size @xmath23 of subchains should thus scale as @xmath24 with a flory exponent @xmath25 set by the spatial dimension . interestingly , the direct visualization of chain conformations is possible for dna molecules @xcite , nanorope polymer chains @xcite or brushlike polymers @xcite adsorbed on strongly attractive surfaces or confined in thin films by means of fluorescence microscopy @xcite or atomic force microscopy @xcite . the experimental verification of the various conformational properties discussed in this paper , such the one described by eq . ( [ eq_compact ] ) , is thus conceivable for these systems @xcite . compactness does obviously _ not _ imply gaussian chain statistics @xcite ( as incorrectly stated , e.g. , in @xcite ) nor does segregation of chains and subchains impose disk - like shapes minimizing the average perimeter length @xmath26 of chains and subchains @xcite . the contour boundaries are in fact found to be highly irregular as is revealed by the snapshots presented in fig . [ fig_snapshot ] and fig . [ fig_snapblob ] . elaborating a short communication made recently @xcite we present here theoretical arguments and molecular dynamics simulations demonstrating that the contours are in fact _ fractal _ , scaling as @xmath27 with @xmath28 being the fractal line dimension @xcite . our work is based on the pioneering work by duplantier who predicted a contact exponent @xmath29 @xcite and a more recent paper by semenov and one of us ( a.j . ) in contrast to many other possibilities to characterize numerically the compact chain conformations the perimeter length is of interest since it can be related to the intrachain structure factor @xmath16 with @xmath30 being the wave vector . it is thus accessible experimentally , at least in principle , by means of small - angle neutron scattering experiments to all polymers which can be appropriately labeled @xcite . specifically , it will be demonstrated that due to the generalized porod scattering of _ compact _ objects @xcite the structure factor of dense 2d polymers should scale in the intermediate wave vector regime as @xmath31 for sufficiently long chains and _ not _ as @xmath32 as numerous authors have assumed @xcite . the present paper is organized as follows . in sec . [ sec_algo ] we recall the computational model used for this study . our mainly numerical results are presented in sec . [ sec_results ] . we confirm first the compactness of the chain conformations by considering the typical size of chains and subchains ( sec . [ sub_rs ] ) . corrections to this leading power - law behavior due to chain - end effects caused by the confinement of the chains will be analyzed in sec . [ sub_rscorrect ] . that @xmath33 does _ not _ imply gaussian chain statistics will be emphasized by the scaling analysis of various intrachain properties such as the histograms of inner chain distances @xmath34 in sec . [ sub_theta ] , the bond - bond correlation functions @xmath35 and @xmath11 in sec . [ sub_bond ] or the single chain structure factor @xmath16 in sec . [ sub_fq ] . two ( related ) scaling arguments will be given in sec . [ sub_ls ] and at the end of sec . [ sub_fq ] , demonstrating the scaling of the perimeter length @xmath26 , eq . ( [ eq_keyls ] ) . the analytic calculation of the structure factor @xmath16 for 2d melts is relegated to the appendix . a discussion of possible consequences of the observed static properties for the dynamics of dense 2d solutions and melts concludes the paper in sec . [ sec_concl ] . our numerical results are obtained by standard molecular dynamics simulations of monodisperse linear chains at high densities . the coarse - grained polymer model hamiltonian is essentially identical to the standard kremer - grest ( kg ) bead - spring model @xcite which has been used in numerous simulation studies of diverse problems in polymer physics @xcite . the non - bonded excluded volume interactions are represented by a purely repulsive lennard - jones potential @xcite @xmath36 + \epsilon \mbox { for } r/\sigma \le 2^{1/6 } \label{eq_algolj}\ ] ] and @xmath37 elsewhere . the lennard - jones potential does _ not _ act between adjacent monomers of a chain which are topologically connected by a simple harmonic spring potential @xmath38 with a spring constant @xmath39 and a bond reference length @xmath40 . both constants have been calibrated to the finite extendible nonlinear elastic " ( fene ) springs of the original kg model . lennard - jones units are used throughout this paper ( @xmath41 ) . the classical equations of motion of the multichain system are solved via the velocity - verlet algorithm at constant temperature using a langevin thermostat with friction constant @xmath42 @xcite . we focus in this presentation on melts of density @xmath43 at temperature @xmath44 . due to the excluded volume potential monomer overlap is strongly reduced as may be seen from the pair correlation function @xmath45 shown in fig . [ fig_gr ] . the bonding potential , eq . ( [ eq_algospring ] ) , prevents the long range correlations which would otherwise occur at such a high density for a 2d system of monodisperse lennard - jones beads @xcite . as shown in the inset of fig . [ fig_gr ] we have in fact a dense _ liquid _ and the oscillations of @xmath46 decay rapidly with an exponential cut - off . the parameters of the model hamiltonian and the chosen density and temperature makes chain intersections impossible , as can be seen from the snapshot of chain 1 " presented in fig . [ fig_snapshot ] . we simulate thus self - avoiding walks " in the sense of the first model class discussed in ref . @xcite . monodisperse systems with chain lengths @xmath21 ranging between @xmath47 up to @xmath1 have been sampled using periodic square boxes containing either @xmath48 or @xmath49 monomers . the larger box of linear length @xmath50 corresponds to @xmath51 chains of length @xmath1 . some conformational properties discussed below are summarized in the table . except the systems with @xmath1 , all chains have diffused over at least 10 times their radius of gyration @xmath8 providing thus sufficiently good statistics . note that our largest chain is about an order of magnitude larger than the largest chains used in previous computational studies of dense 2d melts : @xmath52 by baumgrtner in 1982 @xcite , @xmath53 by carmesin and kremer in their seminal work in 1990 @xcite , @xmath53 by nelson _ et al . _ in 1997 @xcite , @xmath54 by polanowski and pakula in 2002 @xcite , @xmath55 by balabaev _ et al . _ in 2002 @xcite , @xmath56 by yethiraj in 2003 @xcite and @xmath56 by cavallo _ et al . _ in 2005 @xcite . the presented data was obtained on ibm power 6 with the lammps version 21may2008 @xcite . it is part of a broader study where we have systematically varied density , system size and friction coefficient to confirm the robustness of theory and simulation with respect to these parameters . since the chain length is computationally the limiting factor fixing the number of blobs " @xcite at a given density we present data at the largest density , i.e. the largest number of blobs , where we have been able to equilibrate chains of @xmath1 figure [ fig_rs ] confirms that 2d chains are indeed compact as stated in eq . ( [ eq_compactn ] ) and eq . ( [ eq_compact ] ) and as shown in fig . [ fig_snapshot ] for chains and in fig . [ fig_snapblob ] for subchains of arc - length @xmath13 . the typical size @xmath23 of subchains is characterized by either the root - mean - square end - to - end distance @xmath57 or the radius of gyration @xmath58 @xcite . as indicated in the sketch we consider a subchain between two monomers @xmath59 and @xmath60 and average over all pairs @xmath61 possible in a chain of length @xmath21 following @xcite . averaging only over subchains at the curvilinear chain center ( @xmath62 ) slightly reduces chain end effects , however the difference is negligible for the larger chains , @xmath63 , we focus on . the limit @xmath64 corresponds obviously to the standard end - to - end vector @xmath7 and gyration radius @xmath8 of the total chain . open symbols refer to subchains of length @xmath65 with @xmath66 ( squares ) and @xmath1 ( spheres ) , stars to total chain properties ( @xmath64 ) . in agreement with various numerical @xcite and experimental studies @xcite the presented data confirms that the chains are compact , i.e. @xmath3 ( thin lines ) , and this on all length scales with @xmath67 . the segregation of the chains may also be shown by computing the average number @xmath68 of chains in contact with a reference chain , i.e. having at least one monomer closer than a distance @xmath69 to a monomer of the reference chain ( sec . [ sub_ls ] ) . depending weakly on @xmath70 , we find @xmath71 , as one may expect for 2d colloids and in agreement with fig . [ fig_snapshot ] . at variance to any open non - segregated polymer - like structure @xmath68 does thus _ not _ increase with chain length @xmath21 . an alternative way to confirm this statement is to count the centers of mass around the reference chain s center of mass by integrating the center - of - mass pair - correlation function @xmath72 , a structureless function without oscillations ( not shown ) . one verifies that 6 chains are found in a shell of about @xmath73 and this irrespective of @xmath21 . being characterized by the same flory exponent @xmath74 as their three dimensional ( 3d ) counterparts does by no means imply that 2d melts are gaussian @xcite . this can be directly seen , e.g. , from the different probability distributions of the intrachain vectors @xmath75 presented in fig . [ fig_theta ] . to simplify the plot we have focused on the two longest chains @xmath66 and @xmath1 we have simulated . as illustrated in panel ( a ) , * @xmath76 characterizes the distribution of the total chain end - to - end vector ( @xmath77 , @xmath78 ) , * @xmath79 the distance between a chain end and a monomer in the middle of the chain ( @xmath77 , @xmath80 ) , * @xmath81 the distribution of an inner segment vector between the monomers @xmath82 and @xmath83 , * while the segmental size distribution " @xmath84 @xcite _ averages _ over all pairs of monomers @xmath85 for @xmath65 . the second index @xmath13 indicated in @xmath86 characterizes the length of the subchain between the two monomers @xmath59 and @xmath87 . as shown in panel ( b ) , all data for different @xmath21 and @xmath13 collapse on three distinct master curves if the axes are made dimensionless using the second moment @xmath88 of the respective distribution . the only relevant length scale is thus the typical size of the subchain itself . the distributions are all non - monotonous and are thus _ qualitatively _ different from the gaussian ( thin line ) expected for uncorrelated ideal chains . confirming duplantier s predictions @xcite we find @xmath89 with @xmath90 being the scaling variable and the contact exponents @xmath4 , @xmath5 and @xmath6 ( dashed lines ) describing the small-@xmath91 limit where the universal functions @xmath92 become constant . especially the largest of these exponents , @xmath93 , is clearly visible . the contact probability for two monomers of a chain in a 2d melt is thus strongly suppressed compared to ideal chain statistics ( @xmath94 ) . the rescaled distributions show exponential cut - offs for large distances . the redner - des cloizeaux formula @xcite is a useful interpolating formula which supposes that @xmath95 the constants @xmath96 and @xmath97 with @xmath98 being the gamma function @xcite are imposed by the normalization and the second moment of the distributions @xcite . this formula is by no means rigorous but yields reasonable parameter free fits as it is shown by the solid line for @xmath99 . obviously , @xmath100 for very large subchains @xmath101 ( not shown ) . as can be seen , the rescaled distributions @xmath84 and @xmath102 become identical if @xmath103 . it is for this reason that the exponent @xmath93 is central for asymptotically long chains as will become obvious below in sec . [ sub_ls ] and sec . [ sub_fq ] . the two exponents @xmath104 and @xmath105 are only relevant if the measured property specifically highlightes chain end effects as in the example given in the next subsection . the log - log representation chosen in fig . [ fig_rs ] masks deliberately small corrections to the leading power law due to chain end effects which exist in 2d as they do in 3d melts @xcite . these are revealed in fig . [ fig_rsb ] presenting @xmath106 and @xmath107 _ vs. _ @xmath13 and @xmath108 and @xmath109 _ vs. _ @xmath21 using log - linear coordinates and the same symbols as in fig . [ fig_rs ] . the reduced radius of gyration ( bottom data ) becomes in fact rapidly constant and chain length independent . as emphasized by the bottom horizontal line we find @xmath110 with @xmath111 for @xmath112 . interestingly , we observe _ non - monotonous _ behavior for @xmath106 with a decay for @xmath113 . due to this decay @xmath108 ( stars ) is systematically _ below _ the corresponding internal chain distance @xmath106 . if fitted from the chain end - to - end distances one obtains an effective segment size @xcite @xmath114 as shown by the dashed line . ( the index @xmath115 indicates that we refer to the total chain . ) this value corresponds to the ratio @xmath9 given in the table . it confirms similar observations made in previous simulations using much shorter chains @xcite . if on the other hand the effective segment size @xmath116 is obtained from the internal distances this yields @xmath117 as indicated by the top solid line . this value is consistent with a ratio @xmath118 as one expects in any dimension due to @xcite @xmath119 where we have _ assumed _ @xmath120 for _ all _ @xmath121 up to @xmath122 . since this assumption breaks down for @xmath123 ( as seen in fig . [ fig_rsb ] ) eq . ( [ eq_rend2rgyr ] ) is not in conflict with eq . ( [ eq_bestar ] ) . note that the integral over @xmath121 in eq . ( [ eq_rend2rgyr ] ) is dominated by subchains with @xmath124 and that thus large subchains of order @xmath125 are less relevant for the gyration radius of the total chain @xmath126 . hence , the non - monotonous behavior observed for @xmath57 should be barely detectible for the radius of gyration in agreement with eq . ( [ eq_bg ] ) . that a naive fit of @xmath116 from the total chain end - to - end distance @xmath7 leads to a systematic _ underestimation _ of the effective segment size of asymptotically long chains is a well - known fact for 3d melts @xcite . however , both estimations of @xmath116 merge for 3d melts if sufficiently long chains are computed ( as may be seen from eq . ( 16 ) and fig . 4 of ref . apparently , this is not the case in 2d since if @xmath57 or @xmath127 are plotted as a function of @xmath128 a nice scaling collapse of the data is obtained for large @xmath91 and for @xmath129 , i.e. @xmath130 ( since a very similar scaling plot is presented in the inset of fig . [ fig_p1 ] this figure is not given . ) hence , chain end effects may _ not _ scale away with @xmath131 as they do in 3d . a simple qualitative explanation for eq . ( [ eq_rsnscaling ] ) is in fact readily given by considering an ideal chain of bond length @xmath116 squeezed into a more or less spherical container of size @xmath132 . for @xmath133 it is unlikely that the subchain interacts with the container walls and @xmath134 . with @xmath123 the chain will feel increasingly the confinement reflecting it back from the walls into the center of the container reducing thus the effective segment length @xmath135 associated with the chain end - to - end distance . since the scaling function @xmath136 must be universal , it should be possible to express the ratio @xmath137 and thus the ratio @xmath138 in terms of the dimension @xmath139 and universal compact exponents @xmath104 , @xmath105 and @xmath93 . the two statistical segment sizes @xmath116 and @xmath135 should thus be related . at present we are still lacking a solid theoretical proof for the latter conjection . let @xmath140 denote the normalized tangent vector connecting the monomers @xmath141 and @xmath142 of a chain . the bond - bond correlation function @xmath143 has been shown to be of particular interest for characterizing the deviations from gaussianity in 3d polymer melts @xcite . ( as above we average over all pairs of monomers @xmath61 with @xmath60 . ) the reason for this is that @xcite @xmath144 and that thus small deviations from the asymptotic exponent @xmath145 are emphasized @xcite . @xmath35 is presented in fig . [ fig_p1 ] for different chain lengths . apparently , the bond pairs are aligned only for small arc - lengths with @xmath146 . for larger @xmath13 the bonds are _ anti - correlated _ with two characteristic peaks . the first anti - correlation peak visible in panel ( a ) is due to the local backfolding of the chain contour which can be directly seen from chain 1 drawn in fig . [ fig_snapshot ] . note that the chain length dependence disappears for @xmath147 . the second anti - correlation peak is shown in panel ( b ) where @xmath148 is plotted as function of @xmath128 . using eq . ( [ eq_p1rs ] ) this corresponds exactly to the scaling expected from eq . ( [ eq_rsnscaling ] ) with an associated universal function scaling as @xmath149 . in agreement with the qualitative explanation mentioned at the end of sec . [ sub_rs ] , the peak at @xmath150 can be attributed to the confinement of the chain which causes long segments to be reflected back , i.e. @xmath35 must be anti - correlated for large @xmath91 . ( data for @xmath1 is not presented here due to its insufficient statistics . ) we stress finally that altogether this is a rather small effect and essentially @xmath151 for @xmath152 and @xmath153 as already obvious from panel ( a ) . hence , the first legendre polynomial confirms that to leading order @xmath154 for sufficiently large chains and segments . conceptually more important for the present study is the fact that the second legendre polynomial @xmath155 given in the main panel of fig . [ fig_p2 ] reveals a clear power law behavior over two orders of magnitude in @xmath13 ( dashed line ) . this power law is due to _ ( a ) _ the return probability @xmath156 after @xmath13 steps and _ ( b ) _ the nematic alignment " of two near - by bonds . the alignment of bonds is investigated in the inset of fig . [ fig_p2 ] where the second legendre polynomial is plotted as a function of the distance @xmath157 between the mid - points of both bonds . averages are taken over all intrachain bond pairs with @xmath158 $ ] using a bin of width @xmath159 . since @xmath160 becomes rapidly chain length independent we only indicate data for @xmath1 . the vertical axis is rescaled with the phase volume @xmath161 . as can be seen , the orientational correlations oscillate with @xmath157 and this with a rapidly decaying amplitude . these oscillations are related to the oscillations of the pair correlation function @xmath45 shown in fig . [ fig_gr ] and reflect the local packing and wrapping of chains composed of discrete spherical beads . due to both the oscillations and the decay only bond pairs at @xmath162 matter if we compute @xmath11 , i.e. if we sum over all distances @xmath157 at a fixed curvilinear distance @xmath13 . following eq . ( [ eq_theta ] ) one thus expects @xmath163 for @xmath103 . the agreement of the data with this power law is excellent and provides , hence , an independent confirmation of the contact exponent @xmath6 . as obvious from fig . [ fig_snapshot ] and fig . [ fig_snapblob ] the conformations of chains and subchains are neither perfectly spherical nor extremely elongated . having discussed above the chain size we address now the chain shape as characterized by the average _ aspherity _ of the gyration tensor . the gyration tensor @xmath164 of a subchain between the monomers @xmath59 and @xmath165 is given by @xmath166 with @xmath167 being the @xmath168-component of the subchains s center of mass . we remind that the radius of gyration @xmath169 discussed in sec . [ sub_rs ] is given by the trace @xmath170 averaged over all subchains and chains with eigenvalues @xmath171 and @xmath172 obtained from @xmath173 the ratio of the mean eigenvalues @xmath174 for @xmath64 is given in the table . decreasing slightly with @xmath21 this aspect ratio " approaches @xmath175 for our longest chains which corresponds to a reduced principal eigenvalue @xmath176 . it should be noted that gaussian chains and dilute good solvent chains in 2d are characterized by an aspect ratio @xmath177 and @xmath178 , respectively @xcite . our chains are thus clearly less elongated . the asphericity of the inertia tensor of 2d objects may be further characterized by computing the moments @xcite @xmath179 which are plotted in fig . [ fig_asph ] for subchains ( @xmath65 ) and total chains ( @xmath64 ) using the same symbols as in previous plots . @xmath180 describes the mean ellipticity and @xmath181 the normalized variance of @xmath171 and @xmath172 @xcite . obviously , @xmath182 for rods and @xmath183 for spheres . note that taking the first and the second moments of the ellipticity @xmath184 of _ each _ subchain yields qualitatively similar results ( not shown ) . as one expects , both moments do not depend on whether a chain or a subchain is considered . in agreement with yethiraj @xcite they decrease weakly with @xmath21 and @xmath13 . unfortunately , it is difficult to determine precisely the plateau values one expects for asymptotically large chains and subchains , and the horizontal lines with @xmath185 are merely guides to the eye . note that yethiraj @xcite indicates @xmath186 for @xmath187 . considering that the latter value has been obtained at a slightly smaller monomer volume fraction both @xmath181-values are compatible . we remind that in two dimensions @xmath188 and @xmath189 for gaussian chains @xcite and @xmath190 and @xmath191 for dilute good solvent chains according to refs . @xcite . these values are definitely _ larger _ than our respective estimates , eq . ( [ eq_deltavalues ] ) , and segregated chains in 2d melts are thus clearly more axisymmetric . the above analysis has been motivated by recent experimental work on the conformational properties of _ dilute _ dna molecules investigated using fluorescence microscopy @xcite . a similar characterization of the chain shapes at higher semidilute densities appears therefore feasible , at least in principle . as shown in fig . [ fig_snapshot ] , 2d chains adopt irregular shapes with perimeters _ not _ appearing to be smooth , i.e. characterized by a line dimension @xmath192 , but clearly fractal ( @xmath193 ) @xcite . in this subsection we analyze quantitatively this visual impression confirming the announced key result eq . ( [ eq_keyls ] ) for the average perimeter length @xmath26 of chains ( @xmath64 ) and subchains ( @xmath65 ) . we define a perimeter monomer as having at least one monomer not belonging to the same chain or subchain closer than a reference distance @xmath194 essentially set by the monomer density ( see below ) . specifically , we have used @xmath195 in figs . [ fig_snapshot ] , [ fig_snapblob ] and [ fig_ls ] and for the data listed in the table . the number of such perimeter monomers is called @xmath196 , its mean number @xmath197 with @xmath198 being the probability that a monomer of a subchain is on its perimeter . ( note that for a continuous chain model a slightly different box counting " method must be used to obtain a finite perimeter length @xcite . ) the main panel of fig . [ fig_ls ] presents @xmath26 using the same symbols as in fig . [ fig_rs ] . all data collapses on the same master curve , confirming nicely the announced exponent @xmath199 ( dashed line ) and thus a fractal line dimension @xmath200 . this result holds provided that chain and subchain lengths are not too small ( @xmath201 ) . having just confirmed eq . ( [ eq_keyls ] ) numerically we have still to give a theoretical argument to show where it stems from . using the return probability @xmath202 measured in fig . [ fig_p2 ] a simple scaling argument can be given following semenov and johner @xcite . the key point is that a monomer in a long subchain can not `` distinguish '' if the contact is realized through the backfolding of its own subchain or by another subchain of length @xmath13 . since the probability of such a contact , @xmath203 , must be proportional to @xmath156 times the number @xmath13 of monomers in the second subchain we have @xmath204 which is identical to eq . ( [ eq_keyls ] ) . an alternative , but related derivation will be given in section [ sub_fq ] . the fluctuations of the perimeter length are characterized in the inset of fig . [ fig_ls ] . we present here the histograms @xmath205 of the number of perimeter monomers for different subchains of lengths @xmath13 for chains of length @xmath66 . assuming that the first moment @xmath26 of the histogram sets the only scale all histograms are successfully brought to a scaling collapse . please note that the fractality of the perimeter does not imply that the histograms have to be broad . in fact , they decay rather rapidly as indicated by the two phenomenological fits and all moments of the distributions exist . the configurations corresponding to small perimeters are rather strongly suppressed ( solid line ) . at variance to this , the decay for @xmath206 is found to be successfully described by a gaussian ( dash - dotted line ) . the perimeter fluctuations of different contour sections of these configurations are apparently only weakly coupled , if at all . the influence of the distance @xmath70 used to define a perimeter monomer is investigated in fig . [ fig_dll ] where we present the relative error " @xmath207 of the distributions @xmath205 . ( the relative error for @xmath64 is listed in the table . ) for clarity , only data for one chain length @xmath66 is given for several @xmath70 as indicated in the figure . obviously , if @xmath70 is too large ( @xmath208 ) all subchain monomers are considered to be perimeter monomers , @xmath209 , and the perimeter length can not fluctuate ( @xmath210 ) . with increasing @xmath13 the fluctuations increase first ( @xmath211 ) and level then off in the limit of large @xmath13 , i.e. @xmath212 in agreement with the scaling found in the inset of fig . [ fig_ls ] . if , on the other hand , @xmath70 is too small ( @xmath213 ) not all monomers clearly on the contour are detected , as can be checked by looking at snapshots similar to fig . [ fig_snapshot ] . in this case the fluctuations first decay until the subchain length is sufficiently large that the number of detected perimeter monomers becomes proportional to the true number . hence , the relative errors of too small and too large @xmath70 essentially merge for large @xmath13 or become parallel . the specific value of @xmath70 is thus inessential from the scaling point of view . however , computationally it is important to choose a parameter @xmath70 allowing to probe the asymptotic scaling behavior for as broad an @xmath13-range as possible . it is for this technical reason that @xmath195 has been chosen above . a method allowing to verify the fractal dimension of the chain contour not requiring such an artificial parameter is presented in fig . [ fig_grinter ] . we show here the radial pair correlation function @xmath214 between monomers on _ different _ chains as a function of the distance @xmath157 between the monomers . the bold line indicates the pair correlation function @xmath45 between _ all _ monomers already presented in fig . [ fig_gr ] . the same normalization is used for @xmath214 as for @xmath45 , i.e. @xmath215 for large distances where both monomers must necessarily stem from different chains . obviously , @xmath216 for all distances @xmath157 . for small distances @xmath162 the pair correlation function @xmath214 measures the probability that two monomers from different chains are in contact . we remind that for _ open _ chains , e.g. self - avoiding chains in 3d melts , @xmath214 becomes rapidly chain length independent . our chains are _ compact _ , however , and only a fraction of the chain monomers is close to its contour . hence , @xmath214 must decrease with @xmath21 for @xmath162 . this is clearly confirmed by the data presented in the main panel . from eq . ( [ eq : pint ] ) we know already the probability for two monomers from different chains to be close to each other , i.e. for both monomers to be close to the chain contour . one expects thus to find @xmath217 for small distances @xmath157 of the order of a few monomer diameters . this scaling is perfectly demonstrated by the data collapse presented for all chain lengths in the inset of fig . [ fig_grinter ] . neither the intrachain size distributions @xmath86 , nor the bond - bond correlation functions @xmath35 and @xmath11 or the contour length @xmath26 are readily accessible experimentally , at least not for classical small - monomer polymers which can not be visualized directly by means of fluorescence microscopy or atomic force microscopy . it is thus important to demonstrate that @xmath93 is measurable in principle from an analysis of the intrachain structure factor @xmath218\rangle$ ] as announced in eq . ( [ eq_keyfq ] ) in the introduction and as shown now in fig . [ fig_fq ] and fig . [ fig_fqkratky ] . figure [ fig_fq ] presents the unscaled structure factors @xmath16 as a function of the wave vector @xmath18 for a broad range of chain lengths @xmath21 as indicated . as one expects , @xmath16 becomes constant for very small wave vectors ( @xmath219 ) , decreases in an intermediate wave vector regime ( @xmath220 ) and shows finally the non - universal monomer structure for large @xmath18 comparable to the inverse monomeric size ( bragg peak " ) . the first striking result of this plot is that @xmath16 does _ not _ become chain length independent in the intermediate wave vector regime as it does for ( uncollapsed ) polymer chains in 3d . the second observation to be made is that with increasing chain length the decay becomes _ stronger _ than the power - law exponent @xmath221 indicated by the thin line corresponding to gaussian chain statistics . since for an _ open _ polymer - like aggregate or cluster of inverse fractal dimension @xmath74 without any sharp surface the structure factor must indeed scale as @xcite @xmath222 several authors @xcite have argued that an exponent @xmath221 should be observed for 2d polymer melts . however , eq . ( [ eq_fqopen ] ) does not hold for _ compact _ structures where strong composition fluctuations ( of the labeled monomers of the reference chain with respect to unlabeled monomers ) at a thus well - defined surface or perimeter must dominate the structure factor leading to a generalized porod scattering " @xcite . since the exponent @xmath33 for 2d melts does not refer to their gaussian open chain statistics but rather to their compactness , eq . ( [ eq_fqopen ] ) is thus inappropriate . quite generally , the scattering intensity @xmath223 of compact objects is known to be proportional to their surface " @xmath224 which implies @xcite @xmath225 where we have used that @xmath226 and @xmath227 . for a 2d object with a smooth perimeter , i.e. for @xmath192 , this corresponds to the well - known porod scattering @xmath228 . as indicated by the dash - dotted line in fig . [ fig_fq ] , this yields a too strong decay not compatible either with our data . obviously , this is to be expected since we already know from sec . [ sub_ls ] that the perimeter is fractal ( @xmath229 ) and the power - law slope @xmath230 must be a lower bound to our data . if we assume , on the contrary , in eq . ( [ eq_fqsurface ] ) a fractal line dimension @xmath231 , as demonstrated analytically in eq . ( [ eq : pint ] ) , this yields directly the key result eq . ( [ eq_keyfq ] ) anticipated in the introduction . using @xmath232 we thus have @xmath233 as indicated by the dashed line . this power law gives a reasonable fit for the largest chains we have computed . the representation of the structure factor used in fig . [ fig_fq ] is not the best one to check the asymptotic power - law exponents and does not allow to verify the @xmath21-scaling implied by eq . ( [ eq_fqsurface ] ) . we have thus replotted our data in fig . [ fig_fqkratky ] using a kratky representation with vertical axis @xmath234 and a reduced wave vector @xmath235 . using the measured radius of gyration @xmath8 given in the table this obviously allows to collapse all data in the guinier regime for @xmath236 where @xcite @xmath237 the observed data collapse is , however , much broader in @xmath238 and the more the larger the chain length . the deviations observed for large @xmath238 are due to ( chain length independent ) physics on scales corresponding to the monomer size ( bragg peak " ) already seen in fig . [ fig_fq ] . the debye formula for gaussian chains @xcite is given by the thin line which becomes constant in this representation for @xmath239 in agreement with eq . ( [ eq_fqopen ] ) . at variance to this , a striking decay of @xmath240 is observed over a decade in @xmath238 confirming observations by yethiraj @xcite and cavallo _ et al . _ @xcite using much shorter chains . due to the scaling of @xmath240 as a function of @xmath238 this decay implies the chain length dependence seen for the unscaled structure factor in fig . [ fig_fq ] . with increasing chain length our data approaches systematically the power law @xmath241 given by eq . ( [ eq_keyfq ] ) and indicated by the dashed line . even longer chains obviously are warranted to unambiguously show the predicted asymptotic exponent @xmath242 in a computer experiment . in the preceding two paragraphs we have used the fractal line dimension @xmath231 derived _ via _ eq . ( [ eq : pint ] ) together with the generalized porod scattering scaling eq . ( [ eq_fqsurface ] ) to demonstrate the key result eq . ( [ eq_keyfq ] ) . interestingly , the structure factor can be computed directly _ without _ the scaling argument eq . ( [ eq : pint ] ) using that @xmath243 for @xmath133 as discussed at the end of sec . [ sub_theta ] . for asymptotically long chains the structure factor thus can be well approximated as @xmath244 using the fourier transform @xmath245 of @xmath246 . within the redner - des cloizeaux approximation , eq . ( [ eq_redner ] ) , this yields an analytic formula , eq . ( [ ap_fq1 ] ) , given in the appendix . readily computed numerically , this theoretical prediction is represented by the solid line in fig . [ fig_fqkratky ] . since eq . ( [ eq_redner ] ) and eq . ( [ eq_gs2fq ] ) are both approximations this result is not strictly rigorous . however , by construction our formula must yield the guinier regime , eq . ( [ eq_guinier ] ) , for small @xmath247 and since for large @xmath238 only the @xmath93-exponent matters for large @xmath21 it is only around the hump @xmath248 where deviations could be relevant . fig . [ fig_fqkratky ] shows that in practice our approximation agrees well for all @xmath238 as long as the wave vector @xmath18 does not probe local physics ( bragg regime ) . as shown in the appendix , the redner - des cloizeaux approximation eq . ( [ ap_fq1 ] ) reduces to a power law for wave vectors @xmath239 corresponding to the power - law regime of eq . ( [ eq_theta ] ) , @xmath249 in agreement with the key eq . ( [ eq_keyfq ] ) given in the introduction . equation ( [ eq_fqpower ] ) is represented by the dashed lines in fig . ( [ fig_fq ] ) and fig . ( [ fig_fqkratky ] ) . comparing eq . ( [ eq_fqsurface ] ) with eq . ( [ eq_fqpower ] ) demonstrates that 2d melts are characterized by a fractal line dimension @xmath250 hence , using a slightly more physical route as the scaling argument given in sec . [ sub_ls ] we have confirmed the fractal line dimension of the chain perimeter @xmath224 . by labeling only the monomers of sub - chains ( which corresponds to a scattering amplitude @xmath251 ) the above argument is readily generalized to the perimeter length @xmath26 of arbitrary segments of length @xmath65 . using scaling arguments and molecular dynamics simulation of a well - known model hamiltonian we investigated various static properties of linear polymer melts in two dimensions . we have shown that the chains adopt compact conformations ( @xmath252 ) . due to the segregation of the chains the ratio of end - to - end distance @xmath7 and gyration radius @xmath8 becomes @xmath9 ( fig . [ fig_rsb ] ) and the chains are more spherical than gaussian phantom chains ( fig . [ fig_asph ] ) . more importantly , it is shown that the irregular chain contours can be characterized by a fractal line dimension @xmath253 ( figs . [ fig_ls ] and [ fig_grinter ] ) . this key result has been demonstrated analytically using two different scaling arguments given in sec . [ sub_ls ] and sec . [ sub_fq ] , both based on the numerically tested power - law scaling of the intrachain size distribution @xmath254 for small distances @xmath255 with @xmath6 ( fig . [ fig_theta ] ) . compactness and perimeter fractality repeat for subchains of arc - lengths @xmath13 down to a few monomers due to the self - similar structure of the chains ( figs . [ fig_snapblob],[fig_rs ] and [ fig_ls ] ) . measuring directly the return probability of the chain after @xmath13 steps , the second legendre polynomial @xmath11 of the bond vectors decays as @xmath12 ( fig . [ fig_p2 ] ) . interestingly , as implied by the generalized porod scattering of a compact object with fractal surface " , eq . ( [ eq_fqsurface ] ) , the predicted fractal line dimension should in principle be accessible experimentally from the power - law scaling , eq . ( [ eq_keyfq ] ) , of the intrachain structure factor @xmath16 in the intermediate wave vector regime . computationally very demanding systems with chain lengths up to @xmath1 have been required to test the proposed scaling of the structure factor ( fig . [ fig_fqkratky ] ) . we would like to stress that our results are not restricted to a particular melt density , but should also hold for all densities provided that the chains are sufficiently long to allow a renormalization of all length scales in terms of semidilute blobs @xcite . this is of some interest since chain conformations of semidilute 2d solutions of large - monomer polymers ( such as dna or brushlike polymers ) are experimentally better accessible than dense melts @xcite . obviously , these macromolecules are rather rigid and in view of the typical molar masses currently used , deviations are to be expected from the asymptotic chain length behavior we focused on . following previous computational work @xcite it should thus be rewarding to reinvestigate the scaling of flexible and _ semiflexible _ chains in 2d semidilute solutions to see how finite-@xmath21 effects may systematically be taken into account . interestingly , the fractality of the perimeter precludes a finite line tension and the shape fluctuations of the segments are not suppressed exponentially @xcite , but may occur by advancing and retracting `` lobes '' in an `` amoeba - like '' fashion . this opens the possibility for a relaxation mechanism , specific to 2d polymer melts , in which energy is dissipated by friction at the boundary between subchains . following a suggestion made recently @xcite , the longest relaxation time @xmath256 of a chain segment should , hence , scale as @xmath257 rather than with @xmath258 as predicted by the rouse model which is based on gaussian chain statistics @xcite . as gaussian chain statistics is inappropriate to describe conformational properties of 2d melts , there is no reason why a modeling approach based on this statistics may allow to describe , e.g. , the composition fluctuations at the chain contour . since the latter can in principle be probed experimentally using the dynamical intrachain structure factor @xmath259 @xcite this is an important issue we are currently investigating . we thank the ulp , the iuf , the deutsche forschungsgemeinschaft ( grant no . kr 2854/1 - 1 ) , and the esf - stipomat programme for financial support . a generous grant of computer time by genci - idris ( grant no . i2009091467 ) is also gratefully acknowledged . we are indebted to a.n . semenov ( strasbourg ) , s.p . obukhov ( gainesville ) , m. mller ( gttingen ) , m. aichele ( frankfurt ) and a. cavallo ( salerno ) for helpful discussions . the intramolecular structure factor @xmath16 may be rewritten generally as @xcite @xmath260 using the fourier transform @xmath261 of the two - point intramolecular correlation function @xmath84 averaging over all pairs of monomers @xmath85 discussed in sec . [ sub_theta ] . as we have seen in sec . [ sub_theta ] , @xmath84 is well approximated by the distribution @xmath246 for @xmath133 . for asymptotically long chains it is justified to neglect chain - end effects ( @xmath123 ) , i.e. physics described by the contact exponents @xmath104 and @xmath105 . assuming thus translational invariance along the chain contour the structure factor is given approximately by @xmath262 the factor @xmath263 counting the number of equivalent monomer pairs separated by an arc - length @xmath13 . using the redner - des cloizeaux approximation , eq . ( [ eq_redner ] ) , for @xmath264 we compute first the 2d fourier transform @xmath265 with @xmath266 being an integer bessel function @xcite and @xmath267 , @xmath6 , @xmath268 , @xmath269 as already defined in sec . [ sub_theta ] . as can be seen from eq . ( 11.4.28 ) of ref . @xcite , this integral is given by a standard confluent hypergeometric function , the kummer function @xmath270 , @xmath271 with @xmath272 . according to eq . ( 13.1.2 ) and eq . ( 13.1.5 ) of @xcite the kummer function can be expanded as @xmath273 using eq . ( [ ap_g2 m ] ) this yields , respectively , the small and the large wave vector asymptotic behavior of the fourier transform of @xmath246 @xmath274 note that eq . ( [ ap_g2smallz ] ) implies @xmath275 as one expects due to the normalization of @xmath246 . after integrating over @xmath13 following eq . ( [ ap_fqg2 ] ) and defining @xmath276 one obtains for the guinier regime of the structure factor @xmath277 i.e. according to eq . ( [ eq_guinier ] ) we have , as one expects , @xmath278 this is consistent with eq . ( [ eq_rend2rgyr ] ) and the ratio @xmath279 with @xmath116 determined from subchains with @xmath133 , eq . ( [ eq_bers ] ) . ( [ ap_rg ] ) is of course slightly at variance with the measured ratio @xmath280 due the end - effects not taken into account in eq . ( [ ap_fqg2 ] ) . the power law behavior of the structure factor for large wave vectors announced in eq . ( [ eq_keyfq ] ) is obtained by integrating eq . ( [ ap_g2largez ] ) with respect to @xmath13 . this gives @xmath281 obviously , it is also possible to directly integrate eq . ( [ ap_g2 m ] ) with respect to @xmath13 according to eq . ( [ ap_fqg2 ] ) . this yields the complete redner - des cloizeaux approximation of the structure factor @xmath282 which can be computed numerically . using again the expansions of the kummer function , eq . ( [ ap_msmallz ] ) and eq . ( [ ap_mlargez ] ) , one verifies readily that eq . ( [ ap_fq1 ] ) yields the asymptotics for small and large wave vectors already given above . it is convenient from the scaling point of view to replace the variable @xmath283 used above by the reduced wave vector @xmath235 substituting @xmath284 as suggested by eq . ( [ ap_rg ] ) . this gives the curve represented by the solid line in fig . [ fig_fqkratky ] . ( [ ap_flargeq ] ) reexpressed in these terms is given by eq . ( [ eq_fqpower ] ) in the main text . note that due to this substitution the guinier limit of the redner - des cloizeaux approximation of the structure factor is correct by construction . .various conformational properties defined in the main text for 2d polymers at monomer number density @xmath285 as a function of chain length @xmath21 . the aspherity of the chains is characterized by the aspect ratio @xmath286 and the moments @xmath287 and @xmath181 of the eigenvalues @xmath171 and @xmath172 of the inertia tensor . note that @xmath288 . the mean perimeter length @xmath224 and its relative fluctuation @xmath289 given in the two last columns have been obtained assuming a reference distance @xmath195 ( as indicated in fig . [ fig_gr ] ) within which at least one monomer from another chain is to be found to identify a monomer as a perimeter monomer . [ tab_n ] [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] and monomer number density @xmath112 . due to the excluded volume interactions the monomers do barely overlap and chain intersections are strictly forbidden . one chain in the middle is fully drawn while for the other chains only the perimeter monomers interacting with other chains are indicated . a perimeter monomer is defined here as having at least one monomer _ not _ belonging to the same chain within a distance @xmath290 . numbers refer to an arbitrary chain index used for computational purposes . the chains are compact , i.e. , they fill space densely , and interact typically with about @xmath291 other chains . however , compactness does apparently not imply a disk - like shape which would minimize the perimeter length @xmath224 . quantitative analysis reveals that the contours are fractal with a fractal line dimension @xmath292 . [ fig_snapshot ] , width=453 ] . the solid line indicates the perimeter of this chain with respect to monomers of other chains . we consider 8 consecutive subchains of length @xmath293 and compute their respective perimeter monomers being close to monomers from other chains or subchains . the subchains are compact and of irregular shape , just as the total chains ( @xmath64 ) presented in fig . [ fig_snapshot ] . [ fig_snapblob ] , width=453 ] between all monomers for chains of length @xmath1 at monomer number density @xmath112 . note that @xmath294 for large monomer distances @xmath157 . the correlation function oscillates strongly with a period given by the bead diameter @xmath295 . as shown in the inset , the amplitude of these oscillations decays exponentially as expected for a dense liquid without long - range positional order @xcite . the arrow indicates the position of the reference length @xmath195 used for identifying a monomer as being in contact with a monomer from another chain or subchain . [ fig_gr ] , width=453 ] ( top data ) and radius of gyration @xmath58 ( bottom data ) of subchains " or chain segments " @xcite containing @xmath296 monomers as indicated by the sketch on the left hand side . the data is averaged over all @xmath13-subchains possible in a chain of length @xmath21 . open symbols refer to subchains with @xmath65 for @xmath66 ( squares ) and @xmath1 ( spheres ) , stars to overall chain properties ( @xmath64 ) . the indicated power - law slopes confirm eq . ( [ eq_compact ] ) , i.e. the chains are compact on all scales . [ fig_rs ] , width=453 ] described in the main text and the contact exponents @xmath297 predicted by duplantier @xcite . * ( b ) * scaling plots are presented for @xmath298 , @xmath299 and @xmath102 for @xmath66 ( squares ) and @xmath1 ( spheres ) and for @xmath84 for @xmath300 with @xmath293 ( triangles ) and @xmath301 ( diamonds ) . all data for different @xmath21 and @xmath13 collapse on the respective universal master curves if @xmath302 is plotted _ vs. _ the reduced distance @xmath90 with @xmath88 being the second moment of the corresponding distribution . the thin line indicates the gaussian distribution @xmath303 expected for ideal chains in 2d . the power laws @xmath304 observed for @xmath305 ( dashed lines ) confirm duplantier s prediction . the solid line at the bottom shows the redner - des cloizeaux formula , eq . ( [ eq_redner ] ) , for @xmath306 . [ fig_theta ] , width=377 ] described in the main text and the contact exponents @xmath297 predicted by duplantier @xcite . * ( b ) * scaling plots are presented for @xmath298 , @xmath299 and @xmath102 for @xmath66 ( squares ) and @xmath1 ( spheres ) and for @xmath84 for @xmath300 with @xmath293 ( triangles ) and @xmath301 ( diamonds ) . all data for different @xmath21 and @xmath13 collapse on the respective universal master curves if @xmath302 is plotted _ vs. _ the reduced distance @xmath90 with @xmath88 being the second moment of the corresponding distribution . the thin line indicates the gaussian distribution @xmath303 expected for ideal chains in 2d . the power laws @xmath304 observed for @xmath305 ( dashed lines ) confirm duplantier s prediction . the solid line at the bottom shows the redner - des cloizeaux formula , eq . ( [ eq_redner ] ) , for @xmath306 . [ fig_theta ] , width=453 ] as characterized by the root - mean - square end - to - end distance @xmath57 ( top data ) and the radius of gyration @xmath58 ( bottom data ) using the same symbols as in fig . [ fig_rs ] . interestingly , @xmath106 decays for @xmath113 due to chain - end effects . the solid horizontal lines indicate the effective segment sizes @xmath307 and @xmath111 obtained for @xmath308 from @xmath106 and @xmath107 , respectively . the dashed line corresponds to an apparent effective segment size @xmath309 from the end - to - end distances @xmath7 of our longest chains . [ fig_rsb ] , width=453 ] _ vs. _ curvilinear distance @xmath310 . main panel : the first legendre polynomial shows an anti - correlation at @xmath311 due to the local wrapping of the chain . inset : a second anti - correlation peak is visible at @xmath150 which is caused by the reflection of the confined chain . due to ( [ eq_p1rs ] ) this peak is related to the decay of the reduced subchain size @xmath312 visible in fig . [ fig_rsb ] . [ fig_p1 ] , width=453 ] as a function of curvilinear distance @xmath310 * ( a ) * and as a function of intrachain distance @xmath157 * ( b)*. * ( a ) * @xmath11 decays over two orders in magnitude as a power law ( dashed line ) with an exponent @xmath313 in agreement with a return probability given by eq . ( [ eq_preturn ] ) . * ( b ) * @xmath314 oscillates strongly with an exponentially decreasing amplitude . being dominated by a nematic peak at the origin ( @xmath162 ) it acts as a @xmath315-function . [ fig_p2 ] , width=453 ] ( squares ) and @xmath1 ( spheres ) using log - linear coordinates . the aspherity is characterized by the moments @xmath287 and @xmath181 of the eigenvalues @xmath171 and @xmath172 of the inertia tensor discussed in the main text . both moments decay weakly with @xmath13 and @xmath21 . the horizontal lines are guides to the eye indicating possible plateau values for asymptotically long chains . [ fig_asph ] , width=453 ] ( main panel ) and perimeter length distribution @xmath205 ( inset ) for @xmath195 . in agreement with eq . ( [ eq_keyls ] ) the perimeter scales as @xmath316 ( bold line ) . stars refer to perimeters of chains ( @xmath64 ) , open symbols to subchains of lengths @xmath65 in chains of length @xmath66 ( squares ) and @xmath1 ( spheres ) . the histograms for different @xmath13 collapse if rescaled using @xmath26 as presented in the inset for different segment sizes and @xmath66 . the distributions are lopsided decaying much sharper for small - perimeter segments ( @xmath317 ) . the exponential cut - offs indicated for both limits are phenomenological fits . [ fig_ls ] , width=453 ] as a function of @xmath13 for chains of length @xmath66 and for different distances @xmath70 as indicated . the relative error increases for small @xmath13 and too large @xmath70-values . it decreases for @xmath213 since these values are too small for the given density and not all perimeter monomers are detected . the relative error becomes constant for large @xmath13 irrespective of the value of @xmath70 . due to chain end effects the relative error increases slightly for @xmath318 . [ fig_dll ] , width=453 ] between monomers on different chains using the same symbols as in fig . [ fig_p2 ] for the chain length @xmath21 . also indicated is the pair correlation function @xmath45 between all monomers for @xmath1 ( solid line ) . measuring the probability that a monomer is close to the chain perimeter @xmath214 decreases strongly with @xmath21 . this decay is perfectly described by @xmath319 as shown by the scaling collapse of the data presented in the inset . [ fig_grinter ] , width=453 ] as a function of wave vector @xmath18 for different chain lengths @xmath21 as indicated . in striking contrast to 3d melts @xmath16 becomes chain length independent only for wave vectors corresponding to the monomer scale ( bragg peak " ) . to characterize the decay in the intermediate wave vector regime our data is compared with three power - law exponents @xmath221 , @xmath320 and @xmath230 indicated by the thin , the bold dashed line and the dash - dotted line , respectively . the first exponent corresponds to eq . ( [ eq_fqopen ] ) , the second exponent to eq . ( [ eq_keyfq ] ) and the last exponent to eq . ( [ eq_fqsurface ] ) , i.e. to the porod scattering of a compact 2d object with smooth surface ( @xmath321 ) . [ fig_fq ] , width=453 ] tracing @xmath234 as a function of the reduced wave vector @xmath247 for different @xmath21 using the same symbols as in fig . [ fig_fq ] . the debye formula ( top thin line ) corresponds to a plateau for @xmath239 . at variance to this a strong non - monotonous behavior is revealed by our data which approaches with increasing @xmath21 a power law exponent @xmath322 ( dashed line ) corresponding to a compact object of fractal line dimension @xmath323 . the porod scattering for a compact 2d object with smooth perimeter is given by the dash - dotted line , the fourier transform of the redner - des cloizeaux approximation by the solid line . [ fig_fqkratky ] , width=453 ]
triggering the collapse of the presolar cloud with an interstellar shock wave propagating away from a site of stellar nucleosynthesis is a favored explanation for the widespread evidence of short - lived radioisotopes ( slri ) in chondritic refractory inclusions ( lee et al . 1976 ; cameron and truran 1977 ; macpherson , davis and zinner 1995 ) and , much more rarely , in chondrules ( russell et al . 1996 ) found in primitive meteorites . the goal of this paper is to continue the theoretical exploration of the trigggering and injection scenario for slris , in the context of shock waves striking a dense molecular cloud core that could have collapsed to form the solar system . the dozen or so confirmed ( @xmath4be , @xmath5ca , @xmath6al , @xmath7fe , @xmath8mn , @xmath9pd , @xmath10hf , @xmath11i , and @xmath12pu ) or suspected ( @xmath13tc , @xmath14cl , @xmath15pb , and @xmath16nb ) short - lived radioisotopes may require a fairly involved history for their complete explanation ( goswami and vanhala 2000 ; meyer and clayton 2000 ; mckeegan & davis 2003 ; wadhwa et al . a stellar nucleosynthetic source ( a supernova or an agb star ) has been the leading explanation for most of these nuclei ( cameron 1993 , 2001 ; wasserburg et al . 1994 , 1995 , 1998 ; trigo - rodrguez et al . 2009 ; huss et al . 2009 ) , though there are other possibilities as well , in particular local production ( i.e. , in the solar nebula ) of short - lived radioisotopes produced during spallation reactions involving energetic particles emanating from protosolar flares ( shu et al . such local irradiation models appear to have a problem with being able to match the observed abundance ratio of @xmath6al to @xmath5ca ( srinivasan et al . 1996 ; sahijpal et al . 1998 ; lee et al . the observed abundance of @xmath6al thus seems to require its production by stellar nucleosynthesis ( mckeegan et al . however , this problem can be avoided by assuming that the refractory inclusions are shielded by a less refractory mantle during the irradiation , with the mantle being lost later on during heating of the inclusions , thereby yielding the approximate abundance ratio of @xmath6al to @xmath5ca observed in certain meteorites ( gounelle et al . 2001 ) . on the other hand , producing @xmath6al by multiple episodes of local irradiation would negate its use as a precise chronometer for the early solar system ( bizzarro et al . 2004 ; halliday 2004 ; krot et al . 2005 ; thrane et al . 2006 ) , which seems to be ruled out by the agreement of @xmath6al ages with those derived from the pb - pb dating system ( connelly et al . 2008 ) . evidence has also appeared for the presence of the short - lived isotope @xmath4be in an allende inclusion ( mckeegan et al . 2000 ) . because @xmath4be is thought to be produced only by nuclear spallation reactions , its existence has been used to argue strongly in favour of local irradiation ( mckeegan et al . 2000 ; gounelle et al . sahijpal & gupta ( 2009 ) have calculated that even if all of the @xmath4be was produced by local irradiation , then the amount of @xmath6al also produced by local irradiation was about 10% of the total amount of @xmath6al , so that the bulk of the @xmath6al was probably synthesized by a massive star . however , if irradiation is responsible for the @xmath4be , it is unclear if the irradiation occurred in the solar nebula , or in an earlier phase of evolution . arnould et al . ( 2000 ) point out that spallation can occur in the winds ejected from h - depleted wolf - rayet ( wr ) stars . desch et al . ( 2004 ) showed that the @xmath4be might well have originated from @xmath4be galactic cosmic rays that were stopped in the presolar cloud . evidence has also been advanced for the presence of live @xmath17be in ca , al - rich , refractory inclusions ( cais ) , which are believed to represent the earliest solids formed in the solar nebula that have survived relatively unaltered ( chaussidon et al . because of the extremely short half - life of @xmath17be of 53 days , this evidence , if correct , would require formation of @xmath17be in the solar nebula . desch & ouellette ( 2006 ) have disputed the @xmath17be claim , which has not been confirmed by other groups to date . the short - lived isotope @xmath7fe ( goswami and vanhala 2000 ) can not be produced in the appropriate amounts by spallation , and requires a stellar nucleosynthetic source ( tachibana & huss 2003 ) , as does the bulk of the @xmath6al observed to be polluting the interstellar medium . the half - life of @xmath7fe is 2.6 million years ( rugel et al . 2009 ) , roughly four times that of @xmath6al , so in any case , the evidence for live short - lived isotopes in refractory inclusions seems to require that no more than about 1 million years elapsed between the nucleosynthesis of some of the short - lived isotopes in a star and the formation of refractory inclusions in the solar nebula . solar - type stars are believed to form from the collapse of dense molecular cloud cores , which are supported against collapse primarily by magnetic fields ( e.g. , mouschovias , tassis , & kunz 2006 ) , though turbulence also plays a role ( e.g. , kudoh & basu 2008 ) . collapse of a quiescent cloud core begins once the magnetic field support decreases sufficiently through the process of ambipolar diffusion . recently kunz & mouschovias ( 2009 ) have shown that ambipolar diffusion leading to collapse and fragmentation is able to reproduce the observed distribution of molecular cloud core masses , i.e. , the initial core mass function , suggesting the importance of magnetic fields for star formation in general . ambipolar diffusion is estimated to require of order 10 myr to lead to collapse ( mouschovias , tassis , & kunz 2006 ) , a period considerably longer than the half - life of @xmath6al of 0.73 myr . if the solar system s @xmath6al was produced in a massive star , it may have been injected promptly into the protosolar cloud , which must have then collapsed and formed cm - sized solids , all within @xmath18 myr . this constraint assumes that the same stellar outflow that carried the @xmath6al may have triggered the collapse of the protosolar cloud and injected other newly - synthesized elements , including other short - lived isotopes ( cameron & truran 1977 ; boss 1995 ) . abundant observational support exists for the triggering of star formation by expanding supernova shells in upper scorpius ( preibisch & zinnecker 1999 ; preibisch et al . 2002 ) and the cygnus loop ( patnaude et al . 2002 ) , by superbubbles in ob associations ( oey et al . 2005 ; lee & chen 2009 ) , by ionization fronts associated with hii regions ( leppanen , liljestrom , & diamond 1998 ; healey , hester , & claussen 2004 ; hester & desch 2005 ; snider et al . 2009 ) , by generic external shocks ( tachihara et al . 2002 ) , and by protostellar outflows ( barsony et al . 1998 ; sandell & knee 2001 ; yokogawa et al . 2003 ) . here we consider generic shocks , with a special emphasis on supernovae and agb stars as shock sources . boss ( 1995 ) showed that shock fronts from a nearby agb star or a relatively distant supernova could trigger the collapse of a 3d dense cloud core and inject shock front material into the collapsing cloud . foster & boss ( 1996 , 1997 ) studied this process in greater detail for axisymmetric , 2d clouds , and pointed out the crucial role of the assumed isothermal shock front for achieving both goals of triggered collapse and injection . a supernova shock passes through three phases : ejecta - dominated , sedov blast wave , and radiative ( chevalier 1974 ) . the latter phase occurs at distances of about 10 pc , after which the shock front sweeps up a cool shell of gas and dust as it propagates . several recent amr studies ( nakamura et al . 2006 ; melioli et al . 2006 ) have confirmed the results of boss ( 1995 ) and foster & boss ( 1996 , 1997 ) that shock - triggered star formation is likely to occur when the supernova shock front has evolved into a radiative shock , i.e. , the shock front is able to cool so rapidly by radiation that the thin shock front gas is essentially at the same temperature as the ambient gas , which is the same situation as in the isothermal shocks considered by boss ( 1995 ) and foster & boss ( 1996 , 1997 ) . rayleigh - taylor ( r - t ) fingers were identified as the physical mechanism for achieving injection of dust grains and gas into the collapsing presolar cloud ( foster & boss 1997 ; vanhala & boss 2000 , 2002 ) . because the r - t fingers strike the outermost layers of the presolar cloud , inducing collapse , the r - t fingers do not reach the central regions until shortly after the central protosun and the early solar nebula have formed , possibly explaining the lack of @xmath6al in certain ( fraction unknown nuclear = fun ) refractory inclusions ( sahijpal & goswami 1998 ) that may have formed before the r - t fingers arrived . boss ( 2007 ) modeled the r - t finger injection process in the context of this scenario simply by imagining spraying the @xmath6al onto the surface of an existing solar nebula . a related but alternative scenario involves having a nearby ( @xmath19 0.1 pc ) supernova inject @xmath6al directly into the solar nebula ( rather than into the presolar cloud ) , as studied by ouellette , desch & hester ( 2007 ) . they found that the gas from a shock front could not be injected efficiently into a protoplanetary disk because of the disk s much higher density compared to the presolar cloud . chevalier ( 2000 ) had found the same result and attributed it to the hot shock gas not having enough time to cool down . as a result , ouellette et al . ( 2007 ) suggested that the @xmath6al resided primarily in dust grains that shot through the stalled shock - front gas and thereby penetrated into the disk . dust grains smaller than 0.1 micron would be deflected , but micron - sized and larger dust grains would be injected with a high efficiency ( ouellette et al . 2009 ) . supernovae are known to produce large amounts of dust grains , but theoretical models suggest that the newly condensed grains are essentially all smaller than 0.1 micron , and are sputtered to even smaller sizes in the reverse shock front driven into the expanding supernova remnant ( snr ) by the interstellar medium that the snr encounters ( bianchi & schneider 2007 ) . the models are in accord with dust extinction estimates for an observed reddened qso . an insignificant fraction of the total dust grain mass is contained in grains larger than 0.1 micron , presenting a problem for scenarios that rely on large dust grains for injection . however , nittler ( 2007 ) has argued that a sub - class of presolar grains appears to have been formed in a single supernova , conceivably the same supernova that produced many slris . these presolar grains have sizes of 0.1 to 10 micron , large enough to be injected into the disk , or the presolar cloud core . what fraction of the mass of the initial dust grain population these relatively large grains represent is difficult to determine , given the processing associated with detecting presolar grains in meterorites and the typical limitation to the study of grains larger than about 0.1 micron ( e.g. , amari , lewis , & anders 1994 ) . recently it has been suggested that @xmath7fe was injected by a supernova directly into the solar nebula roughly 1 myr after solar system formation ( bizzarro et al . 2007 ) , as in the ouellette et al . ( 2007 ) models . in this scenario , the @xmath6al was derived from the wind from a wolf - rayet ( wr ) star , a massive star that would later become a supernova and inject the @xmath7fe . wr winds do indeed carry large amounts of @xmath6al ( arnould , goriely , & meynet 2006 ) , with winds that are comparable in speed ( @xmath19 1500 km / sec ) to supernova shocks ( marchenko et al . 2006 ) , meaning that such winds would need to be slowed down by sweeping up interstellar gas and dust to speeds less than @xmath20 km / sec if they are to trigger cloud collapse rather than simply shred clouds to pieces ( foster & boss 1996 , 1997 ) . however , other groups have not been able to replicate the ni isotope data that forms the basis for the @xmath7fe scenario ( dauphas et al . 2008 ; regelous et al . nevertheless , wr stars should be considered as a possible source of @xmath6al in addition to that obtained from a supernova ( e.g. , gaidos et al . 2009 ) . williams & gaidos ( 2007 ) estimated that the likelihood of a protoplanetary disk being struck by a supernova shock was less than 1% , but considering that we do not know if the solar system s inventory of short - lived isotopes is rare or not , such an argument can not be considered decisive . in fact , it has been argued recently that significant @xmath6al is necessary for the development of technological civilizations ( gilmour & middleton 2009 ) . however , gounelle & meibom ( 2007 ) and krot et al . ( 2008 ) argued that injection could not have occurred directly onto a relatively late - phase , low mass solar nebula , as @xmath6al from a massive star supernova would have been accompanied by sufficient oxygen to lead to an oxygen isotope distribution in the solar nebula that would be distinct from that inferred for the sun based on mass - independent fractionation of carbonaceous chondrites and from that recently measured by the genesis mission ( mckeegan et al . krot et al . ( 2008 ) therefore argued that injection must have instead occurred into the presolar cloud , so that the sun and the solar nebula shared a common reservoir of oxygen isotopes that could then undergo fractionation . ellinger , young , & desch ( 2009 ) , however , pointed out that supernova explosions are not spherically symmetric , and so some degree of anisotropy in the ejecta is to be expected , perhaps enough to permit @xmath6al injection into the presolar nebula without disturbing the oxygen isotope ratios . while isothermal shock fronts are capable of simultaneous triggering and injection ( boss 1995 ; foster & boss 1996 , 1997 ; vanhala & boss 2000 , 2002 ) , it has been less clear what happens when detailed heating and cooling processes in the shock front are considered . vanhala and cameron ( 1998 ) found that when they allowed nonisothermal shocks in their models , they could not find a combination of target cloud and shock wave parameters that permitted both triggered collapse and injection to occur : they could trigger cloud collapse , or they could inject particles , but not both in the same simulation . such an outcome would be fatal to the triggering and injection hypothesis if definitive . however , vanhala & cameron s ( 1998 ) models employed a smoothed - particle hydrodynamics ( sph ) code , which has since been shown to be poor at resolving dynamical instabilities such as the rayleigh - taylor or kelvin - helmholtz instabilities ( agertz et al . 2007 ) . furthermore , vanhala and cameron s ( 1998 ) thermodynamical routines led to post - shock thermal profiles that were quite different from those of kaufman & neufeld ( 1996 ) , who found that for a 40 km / sec shock , the post - shock gas cooled down from a maximum of over 3000 k to less than 100 k within a distance of 0.001 pc . with a 25 km / sec shock in the vanhala & cameron ( 1996 ) sph code , however , the post - shock temperature rose to 3000 k and showed no signs of decreasing over a distance on the order of 0.5 pc . kaufmann & neufeld ( 1996 ) studied c - type shock fronts , which result in the most successful models of the shock emission from the kleinmann - low nebula in the orion molecular cloud . kaufman & neufeld ( 1996 ) assumed preshock magnetic field strengths in their c - shock models that are consistent with magnetic field strengths measured by zeeman splitting in molecular clouds ( crutcher 1999 ) . such nondissociative , magnetohydrodynamic c - type shocks appear to be the correct analogue for the shock speeds ( @xmath19 5 to 40 km s@xmath21 ) that we expect will be necessary to simultaneously achieve triggered collapse and injection . the relatively low postshock temperatures in c - type shocks are crucial for this scenario . postshock cooling depends sensitively on the detailed microphysics , e.g. , on the emission from rotational states of h@xmath3o , h@xmath3 , co , and oh , and hence on quantities such as the ratio of ortho- to para - hydrogen molecules ( neufeld & kaufmann 1993 ; kaufmann & neufeld 1996 ) . atomic species are also important coolants , as are dust grains . more recently morris , desch , & ciesla ( 2009 ) have reexamined the question of cooling shocked gas by h@xmath3o line emission , finding that the cooling rates in the optically thin limit are at least as high as those calculated by neufeld & kaufmann ( 1993 ) . boss et al . ( 2008 ) found that simultaneous triggering and injection was possible for a 20 km / sec shock striking a 1 @xmath2 cloud , provided that neufeld & kaufmann s ( 1993 ) molecular cooling functions were employed . gounelle et al . ( 2009 ) interpreted this result as meaning that only a very narrow range of shock speeds was consistent with the triggering and injection hypothesis . in this paper we examine what happens for a wide range of shock speeds ( and different mass clouds ) compared to the single shock speed considered by boss et al . ( 2008 ) , in order to determine the robustness of the shock wave trigger hypothesis . achieving adequate spatial resolution of shock - compressed regions of presolar clouds hit with supernova shock fronts , while minimizing the overall computational burden , is an insurmountable problem for a fixed grid hydro code of the type previously used on these problems ( boss 1995 ; foster & boss 1996 , 1997 ; vanhala & boss 2000 , 2002 ) . clearly the demands of the shock - triggered collapse and injection problem requires adoption of the adaptive mesh refinement ( amr ; e.g. , truelove et al . 1997 ; poludnenko et al . 2002 ) technique , which was designed to handle just this type of situation . amr techniques automatically insert new grid points in regions of strong physical gradients , and remove them in regions without strong gradients , in order to maximize the spatial resolution in the crucial regions while minimizing the computational burden . flash employs a block - structured adaptive grid approach using the paramesh package . advection is handled by the piecewise parabolic method ( ppm ) , which features a riemann solver at cell boundaries that handles shock fronts exceptionally well . in flash , ppm is incorporated in a form that is second - order accurate in space and time . we have tested the flash2.5 code s ability to reproduce the results of several different test cases that are relevant to the problem of triggering cloud collapse , namely the sod shock tube problem and the collapse of a pressureless sphere . while the performance of flash on the pressureless sphere collapse is not as accurate as with codes designed to study collapse problems ( e.g. , boss & myhill 1992 ) , flash does a superb job of handling the sod shock tube problem on a cartesian grid , when the shock flows parallel to one axis or at a 45 degree angle . while the standard flash test cases were run on cartesian grids , we have also reproduced the correct results for the sod shock tube and for pressureless cloud collapse on the cylindrical coordinate ( @xmath22 ) grid that is used in the present calculations . in the absence of cooling or an isothermal constraint , flash produces an adiabatic evolution with an effective @xmath23 . the flash equation of state routines were taken to be those for a simple perfect gas with a mean molecular weight of @xmath24 . we have adapted these flash routines to simulate isothermal shock - cloud interactions , where the entire computational grid is forced to remain isothermal , typically at 10 k , in order to compare with the results of foster & boss ( 1996 ) . in these models , @xmath25 is set equal to 1.01 , as a value of @xmath26 is prohibited by the riemann solver . in addition to isothermal models , we also present models that employed the same compressional heating and radiative cooling that was studied by boss et al . ( 2008 ) . our model for radiative cooling is based on the results of neufeld & kaufman ( 1993 ) , who calculated the radiative cooling caused by rotational and vibrational transitions of optically thin , warm molecular gas composed of h@xmath3o , co , and h@xmath3 and found h@xmath3o to be the dominant cooling agent . neufeld & kaufman s ( 1993 ) figure 3 shows that over the range of temperatures from 100 k to 4000 k , the total cooling rate coefficient @xmath27 can be approximated as @xmath28 erg @xmath29 s@xmath21 . the cooling rate @xmath30 , where @xmath31 is the number density of hydrogen molecules and @xmath32 is the number density of the molecular species under consideration . assuming that @xmath33 ( neufeld & kaufman 1993 ) , we take @xmath34 , leading to a radiative cooling rate of @xmath35 erg @xmath36 s@xmath21 , where @xmath37 is the gas density in g @xmath36 . boss et al . ( 2008 ) found that @xmath38 could be increased or decreased by factors of two without having a major effect on the outcome of shock triggering and injection , so the precise value of @xmath38 does not appear to be critical to the results . kaufman & neufeld ( 1996 ) found that the peak temperatures in their mhd shocks were typically of order 1000 k for shock speeds in the range from 5 km / sec to 45 km / sec . as a result , the temperatures in the present models as well as those of boss et al . ( 2008 ) have been restricted to values between 10 k and 1000 k. kaufman & neufeld ( 1996 ) also found a typical shock thickness of order 0.001 pc = @xmath39 cm for a 40 km / sec shock propagating in a magnetized gas with a preshock density similar to that of dense cloud cores . a similar shock thickness occurs in the models of boss et al . ( 2008 ) , using the neufeld & kaufman ( 1993 ) radiative cooling rate @xmath38 , though with a 20 km / sec shock speed and an unmagnetized cloud : evidently the higher shock speed in kaufman & neufeld ( 1996 ) is roughly compensated for by the presence of magnetic fields , compared to the boss et al . ( 2008 ) results . as in boss et al . ( 2008 ) , in the present models we used the two dimensional , cylindrical coordinate ( @xmath40 ) version of flash2.5 , with axisymmetry about the rotational axis ( @xmath41 ) . multipole gravity was used , including legendre polynomials up to @xmath42 . the cylindrical grid is typically 0.197 pc long in @xmath43 and 0.063 pc wide in @xmath44 , though in the higher shock speed models , the grid was extended to be 0.320 pc long in order to better follow the interaction . we set the number of blocks in @xmath44 ( @xmath45 ) to be 5 in all cases , while the number of blocks in @xmath43 ( @xmath46 ) has been varied from 5 to 20 . most models have had 15 blocks in @xmath43 , leading to an approximately uniform grid spacing in @xmath44 and @xmath43 . with each block consisting of @xmath47 grid points , this is equivalent to an initial grid of @xmath48 for most models . the number of levels of refinement ( @xmath49 ) has been varied from five to six . with five levels of refinement employed , flash is able to follow small - scale structures with the effective resolution of a grid 16 times finer in scale , or effectively @xmath50 , somewhat better than the highest resolution of @xmath51 used by vanhala & boss ( 2000 ) . with six levels , the resolution is increased by another factor of two in each direction . our target dense cloud cores consist of bonner - ebert ( be ) spheres ( bonnor 1956 ) , which are the equilibrium structures for self - gravitating , isothermal spheres of gas . be spheres are excellent models for the structure of pre - collapse dense molecular cloud cores seen in star - forming regions ( e.g. , shirley et al . 2005 ) . for the models that compare results to the isothermal models of foster & boss ( 1996 , 1997 ) , the target dense cloud core is a be sphere with a mass of 1.1 @xmath2 , a radius of 0.058 pc , a temperature of @xmath52 k , and a maximum density of @xmath53 g @xmath36 , located at rest near the top of the cylindrical grid . the be sphere is embedded in an intercloud medium with a density of @xmath54 g @xmath36 and a temperature of 10 k. the shock wave begins at the top of the grid and propagates downward at a specified speed toward the be sphere . the shock wave has a thickness of 0.003 pc with a uniform density of @xmath55 g @xmath36 , a mass of 0.016 @xmath2 , and a temperature of 10 k. for the models where cooling is included , the shock wave begins with a temperature of 1000 k and is followed by a post - shock wind with a density of @xmath54 g @xmath36 and temperature of 1000 k , also moving downward at the same speed as the shock wave . the assumed shock structure is the same as that used in the standard case of foster & boss ( 1996 , 1997 ) and investigated by boss et al . the shock structure was chosen to resemble the expected conditions in a planetary nebula wind ( e.g. , plait & soker 1990 ; see discussion in foster & boss 1996 ) . however , it is also consistent with a supernova shock that has swept - up considerable material and slowed down as a result . chevalier ( 1974 ) considered a supernova shock propagating into a cold ( t = 10 k ) medium with a number density of 1 @xmath36 . by 0.25 myr , the shock has slowed to a top speed of @xmath19 60 km / sec and has travelled a distance of @xmath19 2.5 pc . the amount of swept - up mass contained in the shock front that is incident on the target clouds used in the present calculations is 0.015 @xmath2 , quite close to the value of 0.016 @xmath2 in the standard case . the shock wave material is represented by a color field , initially defined to be equal to 1 inside the shock wave and 0 elsewhere , which allows the shock wave material to be tracked in time ( foster & boss 1997 ) . the slri are assumed to be contained primarily in dust grains of sub - micron size ( e.g. , bianchi & schneider 2007 ) , small enough for the grains to remain coupled to the gas . grains larger than this size could shoot through the shock front as it strikes the target cloud and increase the injection efficiency ( e.g. , ouellette et al . 2007 , 2009 ) , so injection efficiencies derived solely from the color field approach should be considered as lower bounds on the true injection efficiencies . we present results for several related studies with the flash code , namely a set of comparisons with the results on the standard case of foster & boss ( 1996 , 1997 ) , the stability of be - like spheres with higher central densities ( and higher masses ) in the absence of shock waves , and , finally , the effects of varying the shock wave speed across a wide range of values for a be - like target sphere with a mass of @xmath56 . we first used flash2.5 to reproduce the standard case of triggered isothermal collapse of foster & boss ( 1996 ) , and verified that flash2.5 was able to produce simultaneous triggered collapse and injection of shock wave material in the case of a 20 km / sec shock . table 1 lists four models that duplicate the standard case , with varied spatial resolution in flash2.5 . table 1 also gives the results of the models : the maximum density obtained ( @xmath57 ) in g @xmath36 , the fraction of the incident color field that is injected ( @xmath58 ) , and the final time of the model ( @xmath59 ) in seconds . once maximum densities of @xmath60 g @xmath36 are reached , even with five or six levels of refinement , flash is unable to follow the collapse to even higher densities , either because it does not have sufficient spatial resolution on the scale of the density maximum , which typically occupies only a few grid cells , or else because the poisson solver is unable to properly represent the gravitational potential of what has become in essence a point mass . as a result , the final times listed in table 1 do not correspond to the time when the density maximum is reached , as flash simply continues to do its best to evolve the entire cloud - shock system past that instant of time . once maximum densities of @xmath60 g @xmath36 are reached , it is appropriate to terminate these calculations anyway , because by such densities the collapsing cores have become optically thick and can no longer cool by our assumed molecular cooling law , which assumes optically thin clouds . a full radiative transfer treatment ( e.g. , boss & myhill 1992 ) is required to treat optically thick regions , a capability that does not exist in flash2.5 . finally , as in boss et al . ( 2008 ) , @xmath58 is defined to be that fraction of the initial color field that is incident on the initial target cloud and that is injected into regions of the collapsing cloud core with density greater than @xmath61 g @xmath36 . that is to say , if the total amount of color that is incident on the target cloud is 1 in dimensionless units , @xmath62 means that 0.001 is the amount of color that was injected . it is clear from table 1 that in all four models , the be sphere was triggered into self - gravitational collapse , given that maximum densities of @xmath60 g @xmath36 or higher were achieved in each case . this maximum density is well over a factor of @xmath63 times higher than the initial maximum density in the target be sphere of @xmath53 g @xmath36 , indicating that dynamical collapse has been induced in these cloud cores . it is also clear that as the spatial resolution in flash is increased , the collapsing cloud is able to reach slightly higher densities , implying that the highest resolution calculations are approaching the continuum limit . note also that the injection efficiency obtained for model fbd , 0.002 , is only slightly lower than the efficiency of 0.003 obtained for the 20 km / sec shock model with heating and cooling presented by boss et al . ( 2008 ) , which had the same spatial resolution as model fbd , showing that when nonisothermal processes are considered , the injection process becomes only somewhat less efficient than when isothermality is assumed . the evolution of these four isothermal models is very similar to that of the models to be presented below with compressional heating and radiative cooling , and so separate figures for the isothermal models are not displayed . we have also used flash to verify the long - term stability of the target clouds in the absence of a triggering shock front , as clouds that collapse on their own on time scales similar to the shock wave passage can not be considered to have been triggered into collapse . shocked - triggered collapse for the standard case occurs within a time span of about @xmath64 yrs ( foster & boss 1996 , 1997 ; boss et al . when the target be sphere used in the models in table 1 is evolved isothermally in flash2.5 without being struck by a shock wave , it does not collapse , but instead oscillates around its initial equilibrium structure , over a time period of at least @xmath63 yrs , sufficiently long to validate the claim that triggering has occurred in the foster & boss ( 1996 ) comparison models . in addition , we have tested the stability of higher mass cloud cores , obtained simply by multipying the initial densities of the standard case be sphere by factors of 2 , 3 , or 4 . formally speaking , these clouds are not true be spheres in equilibrium , but any deleterious effects of this simple computational convenience can be discerned by seeing if the clouds do indeed collapse on their own without being subjected to shock triggering . in the case of the two clouds where the densities had been increased by factors of 3 or 4 , the clouds did indeed begin to undergo dynamic collapse to densities of @xmath60 g @xmath36 or higher within times of @xmath65 yrs , showing their gravitational instability and unsuitability for the shock - triggering models . however , the cloud with twice the be sphere density , i.e. , an initial central density of @xmath66 g @xmath36 and a mass of 2.2 @xmath2 , remained stable in flash for at least @xmath67 yrs , proving its suitability for the present studies . the next section then presents the results for when a 2.2 @xmath2 cloud core is struck by shocks with varying speeds . we now turn to the main focus of this paper , a consideration of what happens when shock waves with a wide variety of speeds are allowed to strike a dense cloud core , while including a fully nonisothermal treatment with compressional heating and radiative cooling . table 2 lists the shock speeds employed for these models of a 2.2 @xmath2 target cloud core , as well as the results of the calculations , as in table 1 . several variations on the spatial resolution were also calculated : models v4 - 4 and v5 - 4 had four levels of adaptive mesh refinement , rather than five , while model v20 - 6 had six . the higher speed models ( v75l , v80l , v90l , and v100l ) had longer grids in the @xmath43 ( vertical ) direction , in order to better follow the evolution of the cloud as it is swept downward by the shock front . figure 1 depicts the initial density distribution for all of the nonisothermal models with varied shock speeds , showing the target cloud and the incoming shock front , where the color field is taken to be initially uniform with a value of unity . figures 1 through 3 show the evolution of model v20 , which typifies the results for models resulting in simultaneous collapse and injection . model v20 is also identical to model c of boss et al . ( 2008 ) , except for having a cloud mass of 2.2 @xmath2 instead of 1 @xmath2 , and behaves in a very similar manner , implying that the process works equally well for target clouds in this mass range . figure 2 shows that the model v20 shock front is able to compress the top edge of the target cloud , while rayleigh - taylor ( r - t ) fingers and kelvin - helmholtz ( k - h ) vortices form around the shock - cloud interface . the r - t fingers drive into the target cloud , while the k - h vortices tend to ablate material off the cloud edge and force it to join the downstream flow . the contours for the color field in figure 2 show that most of the color field impinging on the target cloud is diverted from the cloud by these vortices and disappears downstream . figure 4 shows a close - up of the shock - cloud interface of model v20 at the same time as figure 2 , providing a better look at the structure of the r - t and k - h features , as well as of the color field . figure 3 shows that model v20 is able to form a dense clump along the symmetry axis less than 0.1 myr after the evolution began . figure 5 shows a close - up of the dense clump at the same time as figure 3 , this time with the temperature field contoured in black , instead of the color field . temperatures rise above 100 k only in regions immediately adjacent to the shock - cloud interface , i.e. , in the region with the largest density gradients and hence the strongest compressional heating . except for this highly shocked region , then , the bulk of the cloud is able to remain nearly isothermal at the assumed background temperature of 10 k as a result of the cooling by molecular species such as h@xmath3o , co , and h@xmath3 . the strength of this molecular cooling explains why the results of the nonisothermal model presented by boss et al . ( 2008 ) are so similar to those of the isothermal standard test case of foster & boss ( 1996 , 1997 ) . figure 6 shows an even closer - in view of the density maximum of model v20 after a time of 0.1 myr . the velocity vectors indicate that the region in the vicinity of the density maximum is trying to collapse onto the clump with speeds as high as several km / sec , which is highly supersonic , considering that the sound speed in 10 k gas is 0.2 km / sec . evidently , a protostar with a maximum density of @xmath68 g @xmath36 has formed and is growing by the accretion of gas from the target cloud s envelope . meanwhile , the protostar has been accelerated by the shock front to a speed of order 1 km / sec , and is moving downward as a result . figure 7 plots the color field over the same region as figure 6 , showing that significant shock front material has been injected into the collapsing protostar and the infalling cloud envelope . multiple waves of shock front material should be accreted by the protostar , considering the velocity field evident in figure 7 , including the regions with even higher color densities than that within the density maximum , which appear likely to collapse onto the protostar within a few thousand more years . figure 8 , 9 , and 10 show the evolution of model v4 , where dynamic collapse was triggered , but no significant injection occurred . comparing figure 8 for model v4 with figure 2 for model v20 , at comparable phases of shock - cloud interaction , it is apparent that the much slower speed shock front in model v4 is unable to compress the target cloud s edge to the extent achieved by the shock in model v20 . in fact , by a time of 0.105 myr ( figure 9 ) , the cloud in model v4 has been clearly triggered into collapse by the shock front , yet the color field lags behind and seems unlikely to achieve a significant injection efficiency . figure 10 depicts a close - up of the density maximum for model v4 at the same time as in figure 9 , showing that the cloud has been triggered into roughly spherically symmetric collapse , judging from the density distribution and the velocity vectors . at the same time , the color field within the region plotted in figure 10 is essentially zero , implying that if any shock front material is to be accreted by the growing protostar , it must occur at some later phase in its evolution . finally , figures 11 , 12 , and 13 show the evolution for model v80l , where injection occurred , but dynamical collapse did not ensue . in this case , the strong shock front tends to shred the target cloud into streamers ( figure 11 ) , an excellent situation for injecting shock - front material into the same region ( figure 12 ) , but not well - suited for inducing sustained dynamic collapse . model v80l achieved a maximum density of @xmath69 g @xmath36 by the time shown in figure 11 of 0.056 myr , and by 0.1 myr ( figure 13 ) , the maximum density has dropped to @xmath70 g @xmath36 . figure 12 also demonstrates that even with the more vigorous shock compressional heating in model v80l , the molecular cooling is able to limit the shock - heated regions to the close vicinity of the shock - cloud boundary . in summary , figure 14 shows how the critical outcomes of achieving injection and sustained collapse depend on the assumed shock speed , for all the models listed in table 2 . low speed shocks can induce collapse , but not injection , while high speed shocks result in significant injection , but not in collapse . shocks falling in the range of about 5 km / sec to 70 km / sec appear to be able to simultaneously induce collapse and achieve injection of significant amounts of shock wave material . this result is consistent with the prediction by foster & boss ( 1996 ) that shock speeds of @xmath71 km / sec or higher would shred the target clouds and prevent the formation of a collapsing protostar . table 2 also shows that these basic results are relatively insensitive to the amount of spatial resolution employed , specifically to the number of amr grid levels allowed . model v5 with the standard 5 levels of amr and model v5 - 4 with 4 levels resulted in quite similar outcomes , as was the case for model v20 with 5 levels and model v20 - 6 with 6 levels . boss et al . ( 2008 ) found that for a 1.0 @xmath2 target cloud and a 20 km / sec shock , the injection efficiency @xmath58 was 0.003 . for the comparable 2.2 @xmath2 target cloud in model v20 , @xmath58 was 0.001 , a factor of three times lower for the higher mass cloud . this difference may be attributed to the fact that while the shock front was identical in both models , the larger mass of the target cloud in model v20 made the task of the shock front more difficult : evidently somewhat lower mass clouds are easier to trigger into collapse and pollute with shock front material than somewhat higher mass clouds , at least to the phase studied by these models ( i.e. , maximum densities less than @xmath60 g @xmath36 ) . estimates of the injection efficiency @xmath58 have dropped steadily as the spatial resolution and physical modeling have improved . boss ( 1995 ) found that about half of the impinging shock material ( @xmath72 ) entered the collapsing cloud in his coarsely - gridded , 3d isothermal models . foster & boss ( 1997 ) found @xmath58 between 0.1 and 0.2 in their relatively coarsely - gridded , 2d isothermal models , while vanhala & boss ( 2000 , 2002 ) found @xmath73 in their increasingly higher spatial resolution , 2d isothermal models . broadly speaking , a typical value of @xmath74 characterizes all of the successful triggering and injection models listed in table 2 , a value considerably lower than those previously found . this difference may be attributed to a number of factors , principally the improved spatial resolution of the current models and the inclusion of nonisothermal heating effects , both of which appear to have the effect of reducing @xmath58 compared to lower resolution , isothermal calculations . the superior shock - handling ability of the ppm hydrodynamics method that flash is based upon undoubtedly also plays a role . finally , there is the question of how @xmath58 is defined , and when it is evaluated : in vanhala & boss ( 2002 ) , e.g. , @xmath58 was typically evaluated at earlier times than in the present models , and the region over which the color field was considered to have been injected was liberally interpreted to extend quite some distance from the collapsing protostar ( e.g. , figure 3 in vanhala & boss 2002 ) . when the region expected to be accreted by the protostar is defined in a similar manner to that used in the present models , an estimate of @xmath75 results from the vanhala & boss ( 2002 ) models , an estimate more in line with the current values . nevertheless , such a low injection efficiency may still be in accord with a supernova as the source of the shock wave . based on the estimates of cameron et al . ( 1995 ) , foster & boss ( 1997 ) noted that the @xmath6al - containing gas and dust in a supernova shock wave would have to be diluted by a factor of @xmath76 in order to explain the inferred initial abundance of @xmath6al in the solar nebula , i.e. , @xmath77 of supernova shock - wave material should be injected into a 1 @xmath2 presolar cloud . more recently , takigawa et al . ( 2008 ) used detailed models of a faint supernova with mixing and fallback to attempt to match the inferred initial abundances of the slris @xmath6al , @xmath5ca , @xmath8mn , and @xmath7fe in the solar nebula , based on nucleosynthetic yield calculations ( e.g. , rauscher et al . they found that a dilution factor of @xmath78 and a time interval of 1 myr between the supernova explosion and the formation of the first refractory solids in the solar nebula was able to do a good job of matching all four initial abundances . the dilution factor @xmath79 is the ratio of the amount of mass derived from the supernova that ends up in the solar nebula to the amount of mass in the solar nebula that did not derive from the supernova , i.e. , the mass derived from the target cloud in the present models . takigawa et al . ( 2008 ) found that the best estimates for @xmath79 depended on the assumed mass of the pre - supernova star , ranging from @xmath80 for a 25 @xmath2 star to @xmath81 for a 20 @xmath2 star ; stars with masses of 30 and 40 @xmath2 led to intermediate values of @xmath79 . gaidos et al . ( 2009 ) , on the other hand , suggest a value of @xmath79 of at least @xmath82 for a 25 @xmath2 progenitor star , significantly larger than the estimate by takigawa et al . ( 2008 ) . trigo - rodrguez et al . ( 2009 ) have shown that a 6.5 @xmath2 agb star could have produced the inferred initial abundances of the slris @xmath6al , @xmath5ca , @xmath7fe , and @xmath9pd in the solar nebula , with a similar dilution factor of @xmath83 . the planetary nebulae formed by agb stars typically have slow wind speeds of 10 km / sec , while the fast winds caused by agb flashes can overtake the slow winds and produce swept - up shells with speeds of 30 km / sec ( frank & mellema 1994 , their figures 1 and 2 ) . in fact , the swept - up shell in the frank & mellema ( 1994 ) model closely resembles the structure assumed for the shock front here and in the standard cases of foster & boss ( 1996 , 1997 ) : a shock front with a number density of @xmath76 @xmath36 and a thickness of @xmath84 cm , leading to a total shock front mass of 0.016 @xmath2 impacting the target cloud . based on the present models , then , agb - derived shocks moving at speeds of @xmath19 30 km / sec should be able to trigger collapse and injection in the same manner as a supernova shock with similar properties . in the present models , the mass of the shock wave that is incident on the target cloud is 0.016 @xmath2 , so values of @xmath74 imply that about @xmath85 of shock front material is injected into the collapsing protostar . if the final result is a well - mixed @xmath86 protostar and protoplanetary disk , then the dilution factor @xmath79 is @xmath87 , considerably lower than required by either takigawa et al . ( 2008 ) or gaidos et al . in fact , the mismatch is even worse than this , because in the case of a supernova shock front , the material ejected by the supernova must be diluted by the swept - up , intervening interstellar medium that is necessary to slow down the shock front to speeds capable of achieving triggering and injection . a supernova shock launched at 1000 km / sec must snowplow at least 15 times more mass in order to slow below 70 km / sec . hence the dilution factor for a supernova shock must be decreased by this same factor , to @xmath88 . clearly there is a need to learn if these crudely estimated dilution factors can be increased to values closer to those advocated by takigawa et al . ( 2008 ) or gaidos et al . not all of the 2.2 @xmath2 target cloud will be accreted by the growing protostar , so the injection efficiency will be larger by a proportionate amount . in addition , if the bulk of the shock front material infalls somewhat later than the earliest arrivals ( as suggested by the color waves in figure 7 ) , then the shock front material may end up preferentially in the protoplanetary disk , rather than in the star , thereby increasing proportionally the dilution factor in the disk . since the present models do not include rotation of the target cloud or shock front , there is no possibility for a rotationally supported disk to form , and so the present models can not fully answer the question of the dilution factor appropriate for the solar nebula , as opposed to the presolar cloud as a whole . calculations are currently underway on the dc101 cluster at dtm that include rotation for 2d target clouds , in order to address this key question . given the lower injection efficiency for a 2.2 @xmath2 cloud compared to a 1.0 @xmath2 cloud , higher cloud densities evidently result in lower injection efficiencies , as might be expected . target clouds with lower initial densities should then have higher injection efficiencies , and their larger radii ( for a given mass cloud ) will also go in the direction of increasing the total amount of injected shock wave material . furthermore , allowing injection of shock wave material from behind the leading edge of the shock front ( here considered to be only 0.003 pc thick ) will also increase the amount injected . increasing the assumed density and thickness of the shock front should also lead to higher injection efficiencies . future work will include the study of shock fronts with different densities and thicknesses compared to the standard case of foster & boss ( 1996 , 1997 ) and employed in the present models , as well as different density and radii target clouds , in order to better ascertain the suitability of a wider range of possible shock fronts and target clouds for triggering and injection . coupled with these concerns over the low injection efficiency and the dilution factor are the implications for the star - formation environment where simultaneous triggering and injection might have occurred . looney , tobin , & fields ( 2006 ) showed that for the target cloud considered here and an injection efficiency of @xmath89 , due to geometric dilution alone , a supernova would have to occur within about 0.06 pc to 1.2 pc from the target cloud core in order to inject the desired amount of slris . for the much lower values of @xmath58 found here , @xmath74 , these distance estimates decrease by factors of 10 , to 0.006 pc to 0.12 pc . such distances are appropriate for the proplyds in the orion nebula that are being photoevaporated by the orion trapezium s four o stars ( e.g. , williams , andrews , & wilner 2005 ) . however , the proplyds have already collapsed to form protostars , and the density of the surrounding hii region is too low to slow down a supernova shock wave by snowplowing to the required speeds . thus it remains to be seen if a combination of target cloud and shock front parameters can be found that will increase the injection efficiencies sufficiently to produce a scenario that is consistent with observations of regions of high mass star formation ( e.g. , hester & desch 2005 ) . when cooling by appropriate molecular species is included , shocks with speeds in the range from 5 km / sec to 70 km / sec are able to trigger the gravitational collapse of otherwise stable , dense cloud cores , as well as to inject shock wave material into the collapsing cloud cores . this injected material consists of shock wave gas as well as dust grains small enough to remain coupled to the gas , i.e. , sub - micron - sized grains , which are expected to characterize supernova shock waves ( bianchi & schneider 2007 ) and to carry the slri whose decay products have been found in refractory inclusions of chondritic meteorites . evidently a radiative - phase supernova shock wave ( chevalier 1974 ) is able to cool sufficiently rapidly to behave in much the same way as a shock wave that is assumed to remain isothermal with the target cloud ( e.g. , boss 1995 ) . given that wolf - rayet star winds and supernova shocks both are launched with shock speeds on the order of @xmath90 km / sec , these shock waves can only trigger collapse after they have travelled some distance ( typically about 10 pc ) and been slowed down to 5 km / sec to 70 km / sec by the snowplowing of intervening interstellar cloud gas and dust . the distance a fast shock must travel in order to slow down to speeds consistent with simultaneous triggering and injection is inversely proportional to the mean density of the intervening material ( assuming this material to be moving much less than the shock speed ) ; typical interstellar medium densities lead to distances of a few pc , depending on the desired shock speed . an agb star wind with a typical speed of 10 to 30 km / sec could also have triggered collapse and injection without the need for snowplowing . the low injection efficiencies of the present models , however , point to the need to consider shock fronts and target clouds with different parameters , in order to learn if the injection efficiencies can be increased to the levels thought to be necessary to explain the observed abundances of fossil slris in meteorites . we are currently running three - dimensional ( 3d ) models on the xenia cluster at dtm . the need for a 3d treatment of the shock triggering process is evident from previous 3d studies of the r - t instability ( stone & gardiner 2007 ) and shock fronts ( stone & norman 1992 ; whalen & norman 2008 ) . the r - t `` sheets '' that form in axisymmetric 2d models become true r - t fingers in 3d , allowing better penetration into the target dense cloud cores . we will present the results of these ongoing 3d models as well as of models with varied shock fronts and target clouds in future papers . we thank the referee , steven desch , for several constructive reports that have led to a number of significant improvements in this paper . the calculations were performed primarily on the dc101 cluster at dtm . this research was supported in part by nasa origins of solar systems grant nng05gi10 g and nasa planetary geology and geophysics grant nnx07ap46 g , and is contributed in part to nasa astrobiology institute grant ncc2 - 1056 . the software used in this work was in part developed by the doe - supported asc / alliances center for astrophysical thermonuclear flashes at the university of chicago . v1 & 1.0 & 5 & 15 & 5 & 5.e-12 & 0.0 & 3.8e12 + v2 & 2.0 & 5 & 15 & 5 & 1.e-12 & 0.0 & 8.5e12 + v2.5 & 2.5 & 5 & 15 & 5 & 2.e-12 & 0.0 & 4.8e12 + v4 - 4 & 4.0 & 5 & 15 & 4 & 1.e-12 & 0.0 & 1.2e12 + v4 & 4.0 & 5 & 15 & 5 & 5.e-12 & 0.0 & 3.6e12 + v5 - 4 & 5.0 & 5 & 15 & 4 & 2.e-12 & 2.e-4 & 7.5e12 + v5 & 5.0 & 5 & 15 & 5 & 1.e-12 & 3.e-4 & 8.5e12 + v7 & 7.0 & 5 & 15 & 5 & 2.e-12 & 6.e-4 & 8.5e12 + v9 & 9.0 & 5 & 15 & 5 & 1.e-12 & 2.e-4 & 1.2e12 + v10 & 10.0 & 5 & 15 & 5 & 5.e-12 & 2.e-3 & 6.5e12 + v20 & 20.0 & 5 & 15 & 5 & 1.e-12 & 1.e-3 & 5.3e12 + v20 - 6 & 20.0 & 5 & 15 & 6 & 1.e-12 & 4.e-4 & 5.2e12 + v30 & 30.0 & 5 & 15 & 5 & 4.e-12 & 3.e-3 & 4.0e12 + v40 & 40.0 & 5 & 15 & 5 & 2.e-12 & 1.e-3 & 1.2e13 + v50 & 50.0 & 5 & 15 & 5 & 1.e-12 & 4.e-4 & 4.7e12 + v60 & 60.0 & 5 & 15 & 5 & 1.e-12 & 4.e-4 & 6.5e12 + v70 & 70.0 & 5 & 15 & 5 & 1.e-12 & 3.e-4 & 6.5e12 + v75 & 75.0 & 5 & 20 & 5 & 3.e-13 & 3.e-4 & 8.5e12 + v75l & 75.0 & 5 & 20 & 5 & 1.e-15 & 6.e-4 & 4.5e12 + v80l & 80.0 & 5 & 20 & 5 & 1.e-15 & 5.e-4 & 3.7e12 + v90l & 90.0 & 5 & 20 & 5 & 1.e-15 & 4.e-4 & 3.4e12 + v100l & 100.0 & 5 & 20 & 5 & 1.e-15 & 6.e-4 & 3.0e12 +