Subset (2)

section · 216k rows

Split (3)

train · 203k rows

article stringlengths 0 745k	abstract stringlengths 0 92.7k
since the discovery of accelerated expansion of universe , there have been numerous efforts on theoretical and observational grounds to explain cause of the cosmic acceleration . the most popular theoretical model , which explains the cosmic acceleration and fits well to the currently available observational data , is the standard @xmath0cdm ( cosmological constant @xmath0 + cold dark matter ) model @xcite . in this model , @xmath0 is a candidate of dark energy ( de ) which is believed to be responsible for the current accelerated expansion of the universe . this model , however , suffers from some theoretical problems @xcite , which motivated the researchers to propose new models of de . one such problem is the coincidence problem which refers to the fact that there is no explanation for the same order of the energy densities of dark matter ( dm ) and vacuum energy today . to alleviate the coincidence problem , the interaction between dm and de components of the universe has been proposed / studied in the literature ( see @xcite for recent reviews ) . the interaction in the dark sector is quite appealing since dm and de are the dominant components in the overall energy budget of the universe . moreover , it has recently been shown that the current observational data can favor the late - time interaction between dm and de @xcite . in this paper , we shall study a cosmological scenario where dm and de are not conserved separately rather allowed to interact via a free coupling parameter . + the physics of neutrinos has important implications on the formation of the large - scale structure , big bang nucleosynthesis , cosmic microwave background , and in other cosmological information sources ( see @xcite for review ) . the sum of the active neutrino masses ( @xmath1 ) as well as the effective number of relativistic degrees of freedom ( @xmath2 ) , are still the unknown quantities , and these have received the attention of researchers in different cosmological contexts lately . within the framework of @xmath0cdm model , planck collaboration @xcite has measured @xmath3 ev and @xmath4 at 2@xmath5 confidence level . an extension of @xmath0cdm model has provided @xmath6 ev @xcite . constraints on these parameters have also been performed in @xmath7 gravity @xcite , holographic de @xcite , parametric dynamical de models @xcite , scalar field de @xcite , and in other pioneer works @xcite . estimates for mass of neutrinos have been carried out in other contexts as well such as the solar and atmospheric experiments @xcite , and from the reactor neutrino oscillation anomalies @xcite . + + given the motivation and the recent indication in favor of an interaction in dark sector , probing the interacting scenario in the dark sector with massive neutrinos would be useful to explore new perspectives in this direction . therefore , the aim of this paper is to investigate the constraints on @xmath1 in a cosmological scenario , where dm and de are allowed to interact via a free coupling parameter , by using the latest observational data . in the next section , we briefly review the coupled dm and de scenario in the presence of the massive neutrinos . in section iii , we present the data sets used and the statistical results . we summarize findings of the study in the final section . we consider a spatially - flat friedmann - robertson - walker ( frw ) universe , and assume that dm and de are not conserved separately . the energy conservation law for the interacting dm - de components reads as @xmath8 leading to @xmath9 the friedmann equation can be written as @xmath10 where @xmath11 is the hubble parameter with an over dot denoting derivative with respect to the cosmic time . further @xmath12 , @xmath13 , @xmath14 , @xmath15 , and @xmath16 stand for the energy densities of photons , neutrinos , baryons , cold dm and de , respectively . + let us consider that the dm particles can undergo dilution throughout the cosmic history with a small deviation from the standard evolution characterized by a constant @xmath17 as follows @xmath18 and consider that the de is described by an equation of state parameter @xmath19 , being @xmath20 featuring a quintessence field , and @xmath21 a phantom field . from the eq . ( [ dotrho ] ) and ( [ rho0a3 ] ) we find @xmath22.\end{aligned}\ ] ] the parameter @xmath17 is the coupling constant that characterizes the interaction between dm and de , and @xmath23 implies the absence of interaction while @xmath24 corresponds to the energy transfer from dm to de . hence , the parameter @xmath17 quantifies the deviation from non - interacting case , without the need of considering a specific interaction term between dm and de . the above scenario was originally introduced in @xcite . later , it was generalized and analyzed with observational data in @xcite . + we consider the scalar linear perturbations in the present work as described in @xcite ( and references cited therein ) . here let us briefly describe the cosmological evolution of the massive neutrinos . assuming that the neutrinos decoupling in the early universe occurs instantly , the zeroth - order distribution function for massive neutrinos is given by the fermi - dirac distribution , @xmath25^{-1 } , \label{nu_distro}\end{aligned}\ ] ] where @xmath26 , @xmath27 , are the temperature and scale factor at the neutrino decoupling epoch and , @xmath28 is the co - moving momentum . the factor @xmath29 and @xmath30 are the neutrino mass and number of spin degrees of freedom while @xmath31 and @xmath32 are the planck and the boltzmann constants , respectively . + the unperturbed energy density is given by @xmath33 where @xmath34 , and @xmath35 is given by eq . ( [ nu_distro ] ) . the distribution comprising of a zeroth - order distribution plus a perturbation can be described by @xmath36 . thus , the perturbed energy density in fourier space reads as @xcite @xmath37 where @xmath38 is the perturbation in @xmath39 . the full boltzmann equation that gives the evolution of @xmath38 , is described in @xcite . + in the next section , we discuss the observational constraints on the free parameters of the model under consideration in the presence of massive neutrinos . on the basis of neutrino masses , we consider the following two models . + * model i. a model with three active neutrinos subject to the condition that @xmath40 @xcite . it is the case , normally considered in the literature to investigate the effects of @xmath1 on a particular cosmological scenario . the base parameters set for the model i is @xmath41 model ii. a model with three massive neutrinos ( @xmath42 , with @xmath43 1,2,3 ) in addition to a non - standard effective number of relativistic degrees of freedom @xmath44 ( interpreted as a dark radiation ) that can affect the cmb in several forms @xcite . the base parameters set for the model ii is @xmath45 one may notice that the above two models carry free parameters in addition to the six parameters @xmath46 , @xmath47 , @xmath48 , @xmath49 , @xmath50 and @xmath51 of the @xmath0cdm cosmology . therefore , these models are simple extensions of the standard @xmath0cdm model . in what follows , first we briefly describe the observational data sets used to constrain the parameters of the models under consideration . + planck cmb * : we use the full planck 2015 data @xcite comprised of temperature ( tt ) , polarization ( ee ) and the cross correlation of temperature and polarization ( te ) together with the cmb lensing power spectrum . + * jla * : we use the latest joint light curves " ( jla ) sample @xcite , comprised of 740 type ia supernovae in the redshift range @xmath52 $ ] . + * bao * : we use the baryon acoustic oscillation ( bao ) measurements from the six degree field galaxy survey ( 6df ) @xcite , the main galaxy sample of data release 7 of sloan digital sky survey ( sdss - mgs ) @xcite , the lowz and cmass galaxy samples of the baryon oscillation spectroscopic survey ( boss - lowz and boss - cmass , respectively ) @xcite , and the distribution of the lymanforest in boss ( boss - ly ) @xcite . + * cc+@xmath53 * : we use the cosmic chronometers ( cc ) data set comprising of 30 measurements spanning the redshift range @xmath54 , recently compiled in @xcite . furthermore , we also include the recently measured new local value of hubble constant given by @xmath55 km s@xmath56 mpc@xmath56 as reported in @xcite . + we modified the publicly available class @xcite and monte python @xcite codes for the models under consideration , and constrained the free parameters of the models by utilizing the combined data set : planck cmb + jla + bao + cc + @xmath53 . we used metropolis hastings algorithm with uniform priors on the model parameters to obtain correlated markov chain monte carlo ( mcmc ) samples from class / monte python code , and analyzed these samples by using the getdist python package @xcite . + l l l l parameter & 68% cl & 95% cl & best - fit value + & @xmath57 & @xmath58&@xmath59 + @xmath60 & @xmath61 & @xmath62&@xmath63 + @xmath64 & @xmath65 & @xmath66&@xmath67 + @xmath68 & @xmath69 & @xmath70&@xmath71 + @xmath72 & @xmath73 & @xmath74&@xmath75 + @xmath76 & @xmath77 & @xmath78&@xmath79 + @xmath80 & @xmath81 & @xmath82&@xmath83 + @xmath84 & @xmath85 & @xmath86&@xmath87 + @xmath88 & @xmath89 & @xmath90&@xmath91 + & @xmath92 & @xmath93&@xmath94 + @xmath95 & @xmath96 & @xmath97&@xmath98 + @xmath99 & @xmath100 & @xmath101&@xmath102 + [ tab1 ] _ one - dimensional marginalized distribution , and two - dimensional contours with 68% cl and 95% cl for some selected parameters of the model i. _ , width=325 ] l l l l parameter & 68% cl & 95% cl & best - fit value + & @xmath103 & @xmath104&@xmath105 + @xmath60 & @xmath106 & @xmath107&@xmath108 + @xmath64 & @xmath109 & @xmath110&@xmath111 + @xmath68 & @xmath112 & @xmath113&@xmath114 + @xmath72 & @xmath115 & @xmath116&@xmath117 + @xmath76 & @xmath118 & @xmath119&@xmath120 + @xmath121 & @xmath122 & @xmath123&@xmath124 + @xmath125 & @xmath126 & @xmath127&@xmath128 + @xmath129 & @xmath130 & @xmath131&@xmath132 + @xmath133 & @xmath134 & @xmath135&@xmath83 + @xmath84 & @xmath136 & @xmath137&@xmath138 + @xmath88 & @xmath139 & @xmath140&@xmath141 + & @xmath142 & @xmath143&@xmath144 + @xmath95 & @xmath145 & @xmath146&@xmath147 + @xmath99 & @xmath148 & @xmath149&@xmath102 + [ tab2 ] first , we discuss the results for the model i. table [ tab1 ] displays the main results of the statistical analysis , and figure [ contour1 ] shows the contour plots for some selected parameters of the model i. we note that @xmath150 ev at 2@xmath5 confidence level in the presence of the interaction in the dark sector characterized by a coupling constant @xmath151 . this shows that interaction in the dark sector is mildly favored where the de is characterized by a quintessence field at more than 1@xmath5 confidence level . recently similar results for @xmath17 have been reported in @xcite , but with de favored by a phantom field . here it deserves to mention that the presence of massive neutrinos was not considered in @xcite , and the full planck data set as well . + for the model ii , the table [ tab2 ] summarizes the results obtained from the joint data analysis using planck cmb + jla + bao + cc + @xmath53 data . in figure [ contour2 ] , we present the contour plots for some selected parameters of the model ii . the likelihoods for the masses @xmath152 of the three active neutrinos are considered individually as shown in table [ tab2 ] and figure [ contour2 ] . it is customary in the literature to consider the constraints on the sum of the masses of the three active neutrinos , but here in model ii , we have constrained the three active neutrino masses individually in order to achieve a more robust and accurate analysis . we notice that @xmath153 ev and @xmath154 both at 2@xmath5 level . it favors that the big bang nucleosynthesis limits the extra radiation to @xmath155 at 95@xmath156 confidence level @xcite . thus , our constraints on @xmath157 in the context of an interacting dark sector indicate that the possibility of the presence of an extra radiation ( dark radiation ) is negligible . further , we see that the coupling constant @xmath151 , and the de is characterized by quintessence field ( @xmath20 ) at 2@xmath5 confidence level . + we notice that the mean values of the hubble parameter today @xmath53 are given approximately by 70.0 km s@xmath56 mpc@xmath56 and 71.0 km s@xmath56 mpc@xmath56 for the model i and model ii , respectively . in figure [ h0 ] , we display a schematic representation of 68% cl of @xmath53 for the model i ( blue color ) , ref . @xcite ( red color ) and model ii ( green color ) . it is interesting to observe that the model ii shows a significant consistency at 1@xmath5 level with the local measurement @xmath55 km s@xmath56 mpc@xmath56 of the hubble constant with a 2.4 @xmath156 determination @xcite . it means that the model ii can provide a possible alternative to reconcile the tension between the local and global determinations of the hubble constant ( see @xcite for some recent proposals to reconcile the tension ) . however , further detailing investigation should be developed to improve the constrains on @xmath53 in order to alleviate the tension on @xmath53 considering the coupled de model . _ a schematic representation of 68% cl of @xmath53 for the model i ( blue color ) , ref . @xcite ( red color ) and model ii ( green color ) , where the rectangles length - wise ( along horizontal axis of the frame ) display a graphical view of the 1@xmath5 cl of @xmath53 in each case . _ , width=321 ] in this work , we have investigated a cosmological scenario where dm and de are allowed to have mutual interaction via the free coupling parameter @xmath17 while assuming constant equation of state parameter @xmath158 for the de . we have constrained this interaction in the dark sector in the presence of massive neutrinos using the latest observational data from planck cmb , jla , bao , cc , and @xmath53 by considering two models . in model i , mass of one neutrino is allowed to be a free parameter while in model ii , masses of three active neutrinos are set free with dark radiation for a robust analysis with the latest observational data . the results of the statistical analysis are displayed in table [ tab1 ] and table [ tab2 ] for the model i and model ii , respectively . the main findings of the study are summarized in the following . * in both the models , we observe that the mean value of the coupling parameter @xmath17 is approximately 0 . this shows that interaction in the dark sector in mildly favored by the present observational data within the framework of the models under consideration . * the equation of state parameter @xmath158 of de is observed to have values greater than @xmath159 at around 2@xmath5 confidence level in both the models . it shows that the de from quintessence field is favored by the observational data in the models under consideration . * in model i , we obtain @xmath150 ev while in model ii we have constrained masses of the three neutrinos individually giving rise to @xmath153 ev and @xmath154 both at 2@xmath5 level . we notice that the possibility of the existence of dark radiation is negligible . * we find that the value of @xmath53 obtained in model ii is significantly consistent at 1@xmath5 level with the recent local measurement of @xmath53 reported in ref . @xcite ( see figure [ h0 ] ) . therefore , the model ii can provide a possible alternative to reconcile the tension between the local and global determinations of the hubble constant . a mutual interaction between dm and de has been investigated in different contexts , and it has recently been shown to have a strong potential to be an extension of the standard @xmath0cdm cosmology @xcite . obviously , the other models of coupled de are also worth exploring with massive neutrinos . s.k . gratefully acknowledges the support from serb - dst project no . emr/2016/000258 and dst fist project no . sr / fst / msi-090/2013(c )	we consider the possibility of an interaction in the dark sector in the presence of massive neutrinos , and investigate the observational constraints using the most recent cmb anisotropy data in combination with type ia supernovae , baryon acoustic oscillations , and hubble parameter measurements . we find that a coupled quintessence with massive neutrinos is mildly favored by the present observational data . the model with massive neutrinos is found to be a promising one to alleviate the current tension between local and global determinations of hubble constant .
there is considerable interest in the development of techniques for three - dimensional optical imaging of biological systems systems . in this context , of particular importance is the imaging of mesoscopic systems , where the photon transport mean free path is of the same order as the system size @xcite . for such systems none of the available optical ballistic imaging modalities @xcite-@xcite or the diffuse optical tomography @xcite can be used . on the mesoscopic length scale , applications to biological systems include imagining of engineered tissues , semitransparent organisms , or superficial tissues . this article is the second in a series devoted to the problem of optical imaging in the mesoscopic scattering regime . in ref . @xcite , we have proposed a novel imaging technique that uses angularly - selective measurements of scattered light intensity to reconstruct the attenuation coefficient of an inhomogeneous medium , assuming that the single - light scattering is dominant . the advantages of this single - scattering optical tomography ( ssot ) technique include the linearity , well - posedness , and two - dimensional character of the associated inverse problem , as well as the possibility to perform image reconstruction based on single - projection measurements . here we generalize the ssot technique to simultaneously reconstruct the internal scattering and absorption properties of the medium . we begin by presenting a brief review of the ssot formalism introduced in ref . we assume that the light transport in an inhomogeneous medium is described by the time - independent radiative transport equation ( rte ) for the specific intensity @xmath0 of light at the position @xmath1 and flowing in the direction @xmath2 , @xmath3 i({\bf r } , \hat{\bf s } ) = \mu_s({\bf r } ) \int a(\hat{\bf s},\hat{\bf s}^{\prime } ) i({\bf r } , \hat{\bf s}^{\prime})d^2\hat{s}^{\prime } \ , \ \ { \bf r}\in v \ .\ ] ] here @xmath4 and @xmath5 are the absorption and scattering coefficients , and @xmath6 is the scattering kernels normalized such that @xmath7 for all @xmath2 . the rte ( [ rte ] ) is equivalent to the integral equation @xmath8 where @xmath9 is the ballistic component of the specific intensity , and the ballistic green s function @xmath10 is expressed as @xmath11 with @xmath12 \ ] ] the angularly - averaged ballistic green s function . here @xmath13 is the extinction ( attenuation ) coefficient . the light transport in a mesoscopic system is described by the first - order scattering approximation to the rte . this corresponds to the assumption that light propagating in the inhomogeneous medium is just single scattered , and consists in replacing @xmath14 by @xmath9 in the right - hand side of eq . ( [ rte_int ] ) . consider that the medium is illuminated by a light beam of intensity @xmath15 entering the slab at the point @xmath16 and in the direction @xmath17 , and that an angularly - selective detector registers the ray exiting the slab through the opposite surface at the point @xmath18 and in the direction @xmath19 ( as shown in fig . [ fig : sketch2 ] ) . the intensity measured in a such experiment is denoted by @xmath20 . within the single - scattering approximation , a relationship between the scattering and absorption coefficients of the medium and the measured light intensity is derived in the form @xmath21 d\ell -\ln\left [ \frac{\mu_s({\bf r}_{21})}{\bar \mu_s } \right ] = \phi({\bf r}_2,\hat{\bf s}_2 ; { \bf r}_1,\hat{\bf s}_1 ) \ .\ ] ] here the integral @xmath22 of the attenuation function is evaluated along the _ broken ray _ ( br ) ( shown in fig . [ fig : sketch2 ] ) , corresponding to single - scattered photons and uniquely defined by the source and detector positions and orientations , @xmath23 is the linear coordinate on this ray , @xmath24 is the ray turning point , and @xmath25 is the average ( background ) value of the scattering coefficient . the data function @xmath26 is defined as @xmath27 \ , \ ] ] where @xmath28 , the angles @xmath29 and @xmath30 are defined by @xmath31 , @xmath32 is the polar angle of @xmath33 , and the scattering kernel @xmath34 is assumed position - independent and known . ( [ ssot_eq ] ) is applied to optical imaging in the following manner . the selection of incidence and detection points and incidence and detection directions defines a slice in which image reconstruction is performed . in fig . [ fig : sketch2 ] , this slice coincides with the @xmath35-plane of the laboratory frame . assuming that the @xmath36-coordinate is fixed , the absorption and scattering coefficients can be regarded as two - dimensional functions of variables @xmath37 . on the other hand , the data are four - dimensional , depending , in general , on two spatial and two angular variables , corresponding to the source and detection y - coordinate , and source and detection direction , respectively . this enables the simultaneous reconstruction of scattering and absorption . by utilizing multiple incident beams and detecting light exiting the medium at different points , and by varying the incident and exit angles , it is possible to collect enough data to reconstruct the absorption and scattering coefficients in a given slice . three - dimensional reconstruction is then performed slice - by - slice . in ssot , simultaneous reconstruction of scattering and absorption can be in fact realized without scanning all parameter space . it is enough to keep the incidence direction ( defined by the incidence angle @xmath38 ) fixed , to scan the incidence point @xmath39 , and for each such source realization to scan the detection point @xmath40 , for each detection position considering just one detection direction , corresponding to the detection angle @xmath41 , if @xmath42 , or to the angle @xmath43 , if @xmath44 , where @xmath45 is fixed and @xmath46 is the slab thickness . the incidence and detection angles are the angles between the @xmath47-axis of the laboratory frame and the unit vectors @xmath48 and @xmath19 , respectively . this principle is schematically illustrated in fig . [ fig : sketch1 ] for a rectangular sample illuminated by a normally incident beam . in the presence of scattering , both ballistic and scattered rays are present . to avoid the detection of the ballistic component of the transmitted light , the angularly - selective source and detectors are not aligned with each other . for a light beam entering the sample at position @xmath49 and at the incident angle @xmath38 , consider pairs of detections defined by the detection position and detection angle @xmath50 and @xmath41 , and @xmath51 and @xmath43 , respectively , where @xmath52 , @xmath53 , and @xmath54 . this two measurements corresponds to rays single - scattered at the same position @xmath55 within the sample , where @xmath56 . such rays are shown by the same color in fig . [ fig : sketch1 ] . writing eq . ( [ ssot_eq ] ) for the two situations and then subtracting the corresponding equations , one obtains the following equation : @xmath57 d\ell -\int_{{\rm br}(y_{2}^ { ' } , -\beta_2 ; y_1 , \beta_1)}\mu_t[y(\ell),z(\ell ] d\ell = \phi(y_2 , \beta_2 ; y_1 , \beta_1)-\phi(y_{2}^ { ' } , -\beta_2 ; y_1 , \beta_1).\ ] ] here the integrals of the total attenuation coefficient are along the two single - scattered rays described above , and @xmath58 and @xmath59 are the data functions corresponding to the two detections . ( [ sim_ssot_eq1 ] ) can be solved with respect to @xmath60 . then eq . ( [ ssot_eq ] ) written for one of the detections can be solved for @xmath61 $ ] . from these two solutions , @xmath62 and @xmath63 can be determined . in what follows , we illustrate simultaneous reconstruction of scattering and absorption in ssot based on the algebraic inversion of eq . ( [ ssot_eq ] ) . we note that a more sophisticated image reconstruction algorithm based on a inverse formula is also possible and will be presented elsewhere . the forward data is obtained by solving eq . ( [ rte_int ] ) numerically , along the lines presented in detail in ref . @xcite , generalized here for the case of a homogeneously scattering medium . for isotropic scattering ( @xmath64 ) , ( [ rte_int ] ) takes the form @xmath65 here @xmath66 is the density of electromagnetic energy satisfying the following integral equation @xmath67 where @xmath68 is the `` ballistic density '' . the scattered component of the intensity , @xmath69 , is computed by solving first eq . ( [ u_u_b ] ) and then substituting the numerical solution @xmath70 into ( [ i_u ] ) . we emphasize that this numerical approach is non - perturbative and includes all scattering orders , similarly to the experimental situation when all scattered light is detected . equation ( [ u_u_b ] ) is discretized on a rectangular grid and solved by methods of linear algebra . the energy density @xmath70 and the scattering and absorption coefficients are assumed constant within each cubic cell . the corresponding values @xmath71 , where @xmath72 is the center of the @xmath73-th cubic cell , obey the algebraic system of equations @xmath74 here @xmath75 is the disctretization step , @xmath76 , and @xmath77 is the radius of a sphere of equivalent volume to that of a cell , introduced to compute the diagonal matrix elements of the system ( [ u_u_b_disc ] ) @xcite . the system of equations ( [ u_u_b_disc ] ) is solved by direct matrix inversion , and then the specific intensity is calculated according with the discretized version of ( [ i_u ] ) , @xmath78 where the summation is performed only over such cells that are intersected by the ray exiting from the detection point @xmath18 in the direction @xmath19 . the data function is calculated in terms of the average of the specific intensity over the cell , @xmath79 . to model noise in the measured data , @xmath80 was scaled and rounded off so that it was represented by 16-bit unsigned integers , similar to the measurement by digital ccd cameras . then a statistically - independent positively - defined random variable was added to each measurement @xmath81 . the random variables were evenly distributed in the interval @xmath82 $ ] , where @xmath73 is the noise level and @xmath83 is the average measured intensity ( a 16-bit integer ) . the date function is calculated using the discretized version of eq . ( [ data_def ] ) , @xmath84.\ ] ] image reconstruction for the attenuation coefficient is obtained using eqs . ( [ ssot_eq ] ) and ( [ sim_ssot_eq1 ] ) , which are discretized on the same grid as the one used for obtaining the forward solution , except that in this case planar slices with fixed @xmath36-coordinates are used . the discrete version of ( [ sim_ssot_eq1 ] ) is @xmath85 where the index @xmath86 corresponds to a given realization of the source and detection pair , and @xmath87 , with the matrix element @xmath88 given by the length of the intersection of the detected ray @xmath89 with the @xmath73-th cubic cell . @xmath90 , with @xmath91 the data function corresponding to the ray @xmath89 . ( [ ssot_2d_d ] ) is solved for @xmath92 by regularized svd pseudoinverse @xcite , namely @xmath93 where @xmath94 here @xmath95 is the step function , @xmath96 is a small regularization parameter , and @xmath97 and @xmath98 are the singular functions and singular values , respectively , of the matrix @xmath99 , obtained by solving the symmetric eigenproblem @xmath100 . further , the scattering coefficient is determined from eq . ( [ ssot_eq ] ) , discretized as @xmath101 = \phi^{(1)}_\nu(y_2,\beta_2 ; y_1,\beta_1 ) . \,\ ] ] finally , the absorption coefficient is obtained as @xmath102 . we considered a rectangular isotropically scattering sample of dimensions @xmath103 , @xmath104 and @xmath105 . the background scattering coefficient is chosen such that the optical depth @xmath106 is @xmath107 . this corresponds to the mesoscopic scattering regime in which the image reconstruction method of ssot is applicable . the background absorption coefficient is set equal to @xmath108 . the target is a set of inclusions concentrated in the layers @xmath109 , @xmath110 and @xmath111 . image reconstruction is performed in slices @xmath112 separated by the distance @xmath113 . the reconstruction area inside each slice is @xmath114 , @xmath115 , with the field of view @xmath116 . for each slice , the sources are normally incident ( @xmath117 ) on the surface @xmath118 at the positions @xmath119 , @xmath120 , @xmath118 , with @xmath73 integers . the detectors are placed on the opposite side of the sample at positions @xmath119 , @xmath121 , @xmath122 and measure the specific intensity exiting the surface @xmath122 at the angle of @xmath123 , for @xmath124 , and @xmath125 , for @xmath126 , with respect to the @xmath47-axis . first , we considered the case of purely absorbing inhomogeneities , spatially modulated as shown in fig . [ mu_a_h_0.04 ] in the column marked `` model '' . the scattering coefficient is constant throughout the sample and equal to the background value . the absorption coefficient for the inhomogeneities in the slice @xmath109 is set to @xmath127 and @xmath128 , for the outer and inner square , respectively . in slices @xmath110 and @xmath111 there are more absorbing inhomogeneities , of absorbing coefficient @xmath129 , going from the outmost to the innermost square . thus , the contrast of @xmath130 ( the ratio of @xmath130 in the target to the background value ) varies from @xmath131 for the outmost squares in each slice to @xmath132 for the innermost square . the results of image reconstruction for the total attenuation coefficient @xmath130 and absorption coefficient @xmath133 for various noise levels @xmath73 are presented in figs . [ mu_t_h_0.04 ] and [ mu_a_h_0.04 ] . only the slices containing inhomogeneities are shown . the other slices present no features , which means that no cross - talk exists between various slices , as it was also demonstrated previously @xcite . it can be seen that the spatial resolution of images depends on the noise level and can be as good as one discretization step , @xmath75 . note that image reconstruction is in very good quantitative agreement with the model ( all panels in each figure are plotted using the same color scale ) and stable in the presence of noise . note also that the two - angle measurement scheme considered here enables better image reconstruction of the attenuation coefficient than the single - angle scheme considered previously @xcite . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for a homogeneously scattering sample with @xmath134 and @xmath135 , and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the absorbing inhomogeneities are placed . the contrast in @xmath130 varies from @xmath131 to @xmath132 . , height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for a homogeneously scattering sample with @xmath134 and @xmath135 , and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the absorbing inhomogeneities are placed . the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] consider now the case when scattering inhomogeneities are also present in the system and are spatially modulated as shown in fig . [ mu_s_low_0.04 ] in the column marked `` model '' , the absorption being modulated as described above . the scattering coefficients for the inhomogeneities in the slice @xmath109 is set to @xmath140 and @xmath141 , for the outer and inner square , respectively , and in this slice the the absorbing and scattering inhomogeneities overlap with each other . in the slice @xmath110 , there are more scattering inhomogeneities as compared to the slice @xmath109 , the scattering coefficient is @xmath142 , going from the outermost to the innermost inhomogeneity , and the absorbing and scattering inhomogeneities do not overlap . in the slice @xmath111 , the absorbing and scattering inhomogeneities overlap , the scattering coefficient is modulated the same as in slice @xmath110 except that its value for the innermost inhomogeneity is larger , @xmath143 . for this sample , the contrast of @xmath130 varied from @xmath131 for the utmost squares in slice @xmath144 to @xmath145 for the innermost square in slice @xmath146 . imagine reconstruction in this case is presented in figs . [ mu_t_low_0.04]-[mu_a_low_0.04 ] . very good image quality is obtained for both the total attenuation coefficient and scattering coefficient , image reconstruction for scattering being less influenced by the noise in the data function . on the other hand , image quality for absorption is notably lower . however , most of the relevant features are legible . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath148 to @xmath138 , the contrast in @xmath133 varies from @xmath138 to @xmath139 , and the contrast in @xmath130 varies from @xmath131 to @xmath145.,height=377 ] ( color online ) image reconstruction for the scattering coefficient @xmath147 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath148 to @xmath138 , and the contrast in @xmath133 varies from @xmath138 to @xmath139.,height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath148 to @xmath138 , and the contrast in @xmath133 varies from @xmath138 to @xmath139.,height=377 ] further , consider stronger scattering inhomogeneities and perform image reconstruction for a sample in which the scattering coefficient of the inhomogeneities is spatially modulated the same as in the previous case , but it is increased by a factor of @xmath149 , the absorbing inhomogeneities having the same characteristics . thus , in this case , the contrast in the scattering coefficient varies from @xmath138 for the outmost inhomogeneity in each slice to @xmath150 for the innermost inhomogeneity in slice @xmath146 . the contrast in @xmath130 varies from @xmath131 to @xmath151 . the results are presented in figs . [ mut_0.04]-[mua_0.04 ] . very good image reconstruction is obtained for both the total attenuation and scattering coefficients , but imagine reconstruction for absorption is very poor . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , the contrast in @xmath133 varies from @xmath138 to @xmath139 , and the contrast in @xmath130 varies from @xmath131 to @xmath151.,height=377 ] ( color online ) image reconstruction for the scattering coefficient @xmath147 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath135 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] the reconstructed image quality is determined by two factors , the amount of scattering in the system and the noise in the data . in particular , for stronger scattering , the single - scattering approximation we employ may be inaccurate , leading to poor image reconstruction . in order to separate the influence of these factors on the image quality , we perform image reconstruction based on a data function corresponding only to single - scattered light , obtained using the so - called inverse crime . this consists of generating data using the same model that the inverse solver is based on . specifically , instead of solving rte numerically and using the solution to calculate the data function according to the definition ( [ data_def ] ) , the data function is calculated from ( [ sim_ssot_eq1 ] ) , derived within the single - scattering approximation of rte , by replacing the extinction and scattering coefficients by those of the model . in this case , the influence of the amount of scattering in the sample on the image quality is eliminated , the only influence coming from the noise in the data . image reconstruction for the same sample that was analyzed in figs . [ mut_0.04]-[mua_0.04 ] is presented in figs . [ mut_0.04_ik]-[mua_0.04_ik ] . by comparing these two sets of results , one can see that the imagine quality for the attenuation and scattering coefficients is almost the same in both cases . therefore , it can be concluded that the single - scattering approximation works very well for the scattering strength considered and that , in this scattering regime , the most influence on the image quality comes from the noise in the data . moreover , figs . [ mut_0.04_ik]-[mua_0.04_ik ] show that the reconstructed coefficients experience various levels of influence from the noise in the data . while the reconstructed attenuation and scattering coefficients are very little influenced by the noise , the absorption coefficient experiences a much stronger influence , imagine quality being very poor even for a noise level of @xmath152 in the measured intensity . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for various noise levels @xmath73 , for a data function corresponding only to single - scattered light and calculated according with ( [ sim_ssot_eq1 ] ) ( inverse crime ) . all the sample parameters are as for fig . [ mut_0.04].,height=377 ] ( color online ) image reconstruction for the scattering coefficient @xmath147 for various noise levels @xmath73 , for a data function corresponding only to single - scattered light and calculated according with ( [ sim_ssot_eq1 ] ) ( inverse crime ) . all the sample parameters are as for fig . [ mus_0.04].,height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for various noise levels @xmath73 , for a data function corresponding only to single - scattered light and calculated according with ( [ sim_ssot_eq1 ] ) ( inverse crime ) . all the sample parameters are as for fig . [ mua_0.04].,height=377 ] to explain the various levels of influence of the noise in the data on the image quality , we perform a rough error propagation analysis . assume that the scattered intensity @xmath153 is determined with an error @xmath154 . from eq . ( [ mu_svd ] ) and the definition ( [ data_def ] ) of the data function , it follows that the noise in the data results in a noise @xmath155 in the total attenuation coefficient given by @xmath156 here , we have assumed that the relative error in determining the scattered intensity is the same for the symmetric measurements used to reconstruct @xmath130 , and the factor @xmath157 results from using the the difference in the data functions corresponding to these measurements . on the other hand , the noise @xmath158 in the scattering coefficient @xmath159 is estimated from eqs . ( [ ssot_2d_d ] ) and ( [ delta_mut ] ) to be @xmath160 here @xmath73 labels the cell where the detected rays corresponding to the source - detection realization @xmath161 are single scattered . finally , the absorption coefficient is determined as the difference between the total attenuation coefficient and the scattering coefficient with the error @xmath162 that verifies @xmath163^{1/2}>\frac{\mu_{sn}}{\mu_{an}}\frac{\delta i_s}{i_s}.\ ] ] this expresses the fact that for samples where the absorption is stronger than the scattering , very large noise to signal levels in the reconstructed absorption coefficient result , even for low noise levels in the data . this result completely explains the image reconstruction for the absorption coefficient presented above . the difference between figs . [ mu_a_low_0.04 ] and [ mua_0.04 ] is that the maximum value of the ratio @xmath164 is increased from @xmath165 ( for fig . [ mu_a_low_0.04 ] ) to @xmath166 ( for fig . [ mua_0.04 ] ) , leading to pronounced noise in the reconstructed image . also , although stronger scattering inhomogeneities are present in slice @xmath144 ( second row ) in fig . [ mua_0.04 ] , better image quality is obtained compared with the slice @xmath167 ( first row ) , since in this case the absorbing and scattering inhomogeneities do not overlap and @xmath164 is smaller . the physical interpretation of this result is that the ssot data function carries more signature of the scattering coefficient than of the absorption . in other words , in ssot , the scattering coefficient is privileged as compared with the absorption coefficient . this fact originates from the rte equation we employ , where the scattering coefficient has a stronger contribution than the absorption coefficient , and is also expressed by eqs . ( [ data_def ] ) and ( [ sim_ssot_eq1 ] ) , showing that the scattered intensity decays exponentially with the absorption coefficient , but has a stronger dependence on the scattering coefficient . in this case , besides the attenuation of the specific intensity as a result of absorption and scattering of photons from a given mode into other modes , there is also amplification of a given mode due to scattering of photons from other modes into that mode . we note that this dependence of the measured intensity of the scattering and absorption coefficients of the sample is different from the case of diffuse optical tomography . in the case of diffusive light propagation , the scattering of photons into and out various directional modes does not affect the light intensity , the diffusion equation has similar contributions from the scattering and absorption coefficients , and the light intensity emerging from the sample decays exponentially with both @xmath147 and @xmath133 . to verify the conclusion presented above , we perform image reconstruction for a stronger absorbing sample . figs . [ mut_0.04_ha]-[mua_0.04_ha ] present image reconstruction for a sample where the absorption and scattering are spatially modulated as for figs . [ mut_0.04]-[mua_0.04 ] , but the absorption coefficient is increased by a factor of @xmath168 , such that it becomes comparable to the scattering coefficient . indeed , the image reconstruction for the absorption coefficient presented in fig . [ mua_0.04_ha ] is markedly better than in fig . [ mua_0.04 ] and comparable to that for the scattering coefficient presented in fig . [ mus_0.04_ha ] . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , the contrast in @xmath133 varies from @xmath138 to @xmath139 , and the contrast in @xmath130 varies from @xmath138 to @xmath165.,height=377 ] ( color online ) image reconstruction for the scattering coefficient @xmath147 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath134 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] finally , we perform image reconstruction for a stronger scattering sample , characterized by an optical depth of the background of @xmath170 and an additional contrast in the scattering coefficient of up to @xmath150 . this is a borderline case when scattering is sufficiently strong so that the single - scattering approximation of ssot may be expected to be inaccurate . the results for imagine reconstruction obtained for the case when the scattering and absorption have comparable strengths are presented in figs . [ mut_0.08_ha]-[mua_0.08_ha ] . we obtain that even in this scattering regime the most relevant features in the reconstructed scattering and absorption coefficients remain legible . ( color online ) image reconstruction for the total attenuation coefficient @xmath130 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath171 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , the contrast in @xmath133 varies from @xmath138 to @xmath139 , and the contrast in @xmath130 varies from @xmath138 to @xmath165.,height=377 ] ( color online ) image reconstruction for the scattering coefficient @xmath147 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath171 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] ( color online ) image reconstruction for the absorption coefficient @xmath133 for an inhomogeneously scattering and absorbing sample and for various noise levels @xmath73 . the rows show the slices @xmath109 , @xmath136 and @xmath137 , where the inhomogeneities are placed . the background scattering and absorption coefficients are set such that @xmath171 and @xmath169 , the contrast in @xmath147 varies from @xmath138 to @xmath150 , and the contrast in @xmath133 varies from @xmath138 to @xmath139 . , height=377 ] we have demonstrated that the ssot technique enables simultaneous reconstruction of scattering and absorption properties of mesoscopic systems . in particular , we have shown that while accurate , qualitative imagine reconstruction of scattering is always possible , good image reconstruction for absorption can be realized under the condition that scattering and absorption have comparable strengths . these conclusions have been reached under the assumption that the light propagating in the mesoscopic systems is just single scattered , but without making any assumption of measuring just single - scattered light . we have argued that better image quality for scattering as compared to absorption is possible since the ssot data function carries a stronger signature of scattering than of absorption . simultaneous reconstruction of scattering and absorption of mesoscopic systems can be experimentally implemented by appropriately choosing the wavelength of the illuminating beam , such that the effects of absorption and scattering have comparable strengths @xcite . alternatively , the absorption characteristics of the sample can be recovered through fluorescent ssot , which will be the subject of a future study . this work was supported by the national science foundation under grant no .	we demonstrate that simultaneous reconstruction of scattering and absorption of a mesoscopic system using angularly - resolved measurements of scattered light intensity is possible . image reconstruction is realized based on the algebraic inversion of a generalized radon transform relating the scattering and absorption coefficients of the medium to the measured light intensity and derived using the single - scattering approximation to the radiative transport equation .
since the first investigations of electron - positron pair creation in strong electric fields , also known as the schwinger effect , there has been a lot of theoretical investigations of it . however it has not yet been possible to measure it directly due to the exponentially damped pair creation rate @xmath1 where @xmath2 is the field strength @xmath3 normalized by the critical electric field @xcite @xmath4 as laser powers in future may get closer to reaching this critical field strength , investigations of the effect are of interest . however there is the possibility that different strong field processes , such as qed cascades , will set in which might prevent reaching critical intensities @xcite . + as pointed out in @xcite it might be crucial for the detection of the schwinger effect to evaluate if one can distinguish a qed cascade triggered by an electron which was produced in the schwinger process from one which was triggered by vacuum impurities . to do so one first has to evaluate the properties of the pairs produced via the schwinger effect . the simplest field configurations taken into account for cascade calculations are uniformly rotating electrical fields . these are approximative models for the electric fields in the anti - nodes of circularly polarized standing waves . + the momentum spectrum of produced electron - positron pairs for rotating fields has been recently investigated numerically for pulsed fields with the help of the real - time dirac - heisenberg - wigner ( dhw ) formalism @xcite as well as analytically for constant rotating fields with help of the semiclassical wkb approximation for spinor qed @xcite . numerical methods usually need a lot of computation time . semiclassical methods can help to give a better understanding of special features of the momentum spectrum , e.g. interference effects . + for one - component fields depending solely on time there exist a lot of semiclassical investigations using e.g. the wkb - approximation @xcite or the world - line instanton method @xcite . in addition to that there are exact solutions for fields depending on lightcone variables @xcite . while one - component fields can be used to describe linearly polarized laser fields , one needs two - component fields to study pair creation of circularly polarized fields . there exist some semiclassical studies of more involved fields such as electric fields with two or three components depending on space @xcite or fields depending on space and on time @xcite . the exponential factor of rotating electric fields was studied in @xcite with the imaginary time method and in @xcite with the world - line instanton method . + in @xcite the wkb method was used to compute the momentum spectra for more than one - component fields in scalar qed . the assumption was made that for one pair of turning points the results of spinor and scalar qed are the same at the leading non perturbative order . this was based on the fact that it was shown to be true for one - component fields in @xcite . + in the present work we show that for two - component fields there is a difference in the pair creation rate of spinor and scalar qed even if there is only one pair of turning points . to do so we start from the dirac equation and find a semiclassical approximation which is a generalization of the wkb - approximation for the one - component case . we find that in the two - component case there is a factor depending on the spin of the produced pairs compared to the scalar case . this factor gives rise to a different rate for particles in different spin states . + looking at examples of rotating electric fields we find that particles of one spin direction dominate the produced pairs depending on the sense of rotation of the field for a certain range of pulse lengths and frequencies . for the constant rotating field a generalization of the analytic results for the scalar case @xcite is found , so that the momentum spectrum can be calculated analytically . + this work is arranged as follows : in section [ sec : scmomspec ] we derive the semiclassical momentum spectrum for two - component electric fields depending on time . these results are used in section [ sec : rotating ] to study the momentum spectrum of the constant rotating field . the total pair creation rate of this field is investigated in section [ sec : totalparticleyield ] . section [ sec : conclusions ] contains our conclusions and remarks . in order to keep the main text and ideas more clear , we have relegated some of the technical calculations to the appendix . in appendix [ app : ansatz ] we motivate our ansatz for the coefficients used in section [ sec : scmomspec ] . the analytic calculations necessary for the computation of the momentum spectra of the constant rotating field can be found in appendix [ app : rotana ] . here we compute the semi classical momentum spectrum for spinor qed for fields with two time - dependent components . the corresponding scalar case has been treated in @xcite . we start from the dirac equation @xmath5\gamma^\mu - mc\right)\psi(\vec{x},t)=0.\end{aligned}\ ] ] we can go to the squared version by substituting @xmath6\psi(\vec{x},t),\end{aligned}\ ] ] for which we find @xmath7\psi(\vec{x},t)=0,\end{aligned}\ ] ] where @xmath8\end{aligned}\ ] ] and @xmath9 for solely time - dependent fields with the general potential @xmath10 the field strength tensor only has non vanishing components for @xmath11 or @xmath12 . + we now choose to work in the weyl basis , i.e. @xmath13 in this basis we find @xmath14 if we now use the ansatz @xmath15 and @xmath16 is the canonical momentum , we find the four coupled equations @xmath17\psi_1^\pm(t)\mp{{\text{i}}}\hbar\ , c\dot{p}_{x - y}(t)\,\psi_2^\pm(t)&=0 , \label{eq : coupleddirac1}\\ \left[\hbar^2\partial_t^2+{{\mathcal{e}}}(t)^2\pm{{\text{i}}}\hbar\ , c\dot{p}_z(t)\right]\psi_2^\pm(t)\mp{{\text{i}}}\hbar\ , c\dot{p}_{x+y}(t)\,\psi_1^\pm(t)&=0 , \label{eq : coupleddirac2}\end{aligned}\ ] ] where we defined @xmath18 and introduced the kinetical momentum @xmath19 because of the chiral properties of the weyl - basis we can interpret @xmath20 and @xmath21 as different spin states of the produced pairs . + for the two - component case @xmath22 we can now make the ansatz ( for a motivation see appendix [ app : ansatz ] ) @xmath23 where we defined the spin parameter @xmath24 and @xmath25 as well as the integrals @xmath26 using the ansatz ( [ eq : ansatz1])-([eq : ansatzdot2 ] ) in the dirac equation ( [ eq : coupleddirac1])-([eq : coupleddirac2 ] ) leads to coupled equations for the coefficients @xmath27 where @xmath28 as shown in appendix [ app : ansatz ] this solution reduces correctly to the known results of the one - component case for @xmath29 . + the number of produced electron - positron pairs as a function of the momentum @xmath16 can be found as the transmission probability @xmath30 if we now use appropriate boundary conditions @xcite @xmath31 we can by iteratively using eqs . ( [ eq : alpha ] ) and ( [ eq : beta ] ) find @xmath32 note that this was done by analogy with @xcite . these integrals are dominated by the regions around the turning points which are given by @xmath33 according to eq . ( [ eq : sce ] ) the @xmath34 are found in complex conjugated pairs . following from eq . ( [ eq : multiint ] ) by deforming the contour we extract the singularities for the turning points @xmath35 lying in the upper half plane of the imaginary plane . by analogy with @xcite we consider general singularities of order @xmath36 @xmath37 which leads to @xmath38 using the fact that @xmath39 one can now change variables to @xmath40 to find @xmath41 where @xcite @xmath42 this leads to @xmath43 for all the examples covered in this work we deal with simple turning points , i.e. @xmath44 . thus the momentum spectrum of the pair creation rate ( [ eq : trans ] ) in the semiclassical approximation takes the form @xmath45 observe that unlike in the case of one - component fields the number of pairs produced in both spin states is not the same . in fact looking at examples we will see that the factor @xmath46 usually elevates the pair creation rate for one spin and suppresses the other . + comparing to the one - component case of @xcite we find that the factor @xmath47 is replaced by the factor @xmath46 in the two - component case . accordingly we find from eq . ( [ eq : comptoonecomp ] ) @xmath48 it is also noteworthy that the case of scalar qed which was treated in @xcite for general time dependent fields is included in this result if one sets @xmath49 . [ sec : rotating ] in this section we compute the pair creation rate for the rotating electric field which is described by @xmath50 where @xmath51 defines the sense of rotation . it is possible to compute the integrals @xmath52 analytically in terms of elliptic integrals ( this was shown for the scalar case in @xcite ) . to do so we first look at the turning points . these are given by @xcite @xmath53 we find an infinite number of turning points . they all have the same distance to the real axis and thus the integral @xmath54 defined in eq . ( [ eq : k_s ] ) has the same imaginary part for each of this pairs of turning points and we thus define @xmath55 . \label{eq : kappa}\end{aligned}\ ] ] since the imaginary parts of the turning points are equidistant from each other , the real parts of @xmath56 are multiples of @xmath57-{{\text{re}}}[k_s(t_{k}^+ ) ] \label{eq : theta}\end{aligned}\ ] ] for different turning points ( after subtraction of a global phase ) . the exact form of the integrals for @xmath54 is derived in appendix [ app : rotana ] and is given in eqs . ( [ eq : krot])-([eq : k_xyrot ] ) as functions of the elliptic integrals eqs . ( [ eq : k])-([eq : pi ] ) . + however since there is an infinite number of turning points , which give the same contribution to the momentum spectrum , the sum in eq . ( [ eq : momentumspectrum ] ) diverges . this only happens because the fields rotates infinitely long which is unphysical . if we however consider a rectangular pulse , i.e. we turn the field on at time @xmath58 and off at @xmath59 , only turning points which satisfy @xmath60 contribute to the momentum spectrum . we will therefore study the rotating rectangular pulse given by ) of the multiple integral iteration ( [ eq : multiint ] ) is only correct for analytic potentials . for potentials with discontinuities in their derivatives the semiclassical limit is dominated by the first term @xcite . however since we are interested in the rotation and not in the effects off turning on and off the field we will treat the potential as if it was analytical . note however that switching on and off effects have been recently studied in the context of pair creation in an exponentially decreasing field @xcite . [ fn : nonanalytical ] ] @xmath61 where we defined the rectangular box function @xmath62 and set @xmath63 for convenience . + using the criterion ( [ eq : criterium ] ) and eq . ( [ eq : rotatingcompltp ] ) we find that for every full period of @xmath64 inside the pulse length @xmath65 , one turning point contributes fully to the momentum spectrum . if we now introduce the parameter @xmath66 , the momentum spectrum ( [ eq : momentumspectrum ] ) with @xmath67 turning points taken into account , i.e. for @xmath68 , takes the form @xmath69 where @xmath70 and @xmath71 are defined in eqs . ( [ eq : kappa ] ) and ( [ eq : theta ] ) respectively . + from eq . ( [ eq : msrectangularpulse ] ) we find that interference effects occur for @xmath72 . these effects can for example be seen in the momentum spectra in the spinor and scalar case which are plotted in fig . [ fig : ms_t = infty ] for two different ranges of parameters . + the parameters of on the left side of fig . [ fig : ms_t = infty ] are comparable to the ones chosen in @xcite . in addition to the interference effects reported in that work one can also see that the spinor particle rate is increased with respect to the scalar one for particles with one spin and decreased for the others . as explained in section [ sec : scmomspec ] this is due to the factor @xmath73 . looking at the solution of @xmath74 given in eq . ( [ eq : k_xyrot ] ) we find that there is a global factor of @xmath75 which controls the sense of the rotation of the field ( [ eq : erot ] ) . we thus find that the sense of the rotation determines which spin state the spectrum is dominated by . + on the right side of fig . [ fig : ms_t = infty ] a part of the spectrum is plotted for parameters which are closer to the ones of present - day laser systems , i.e. a pulse length @xmath76 and a laser frequency @xmath77 while the amplitude field strength is still @xmath78 . although we still find interference effects the pattern is to narrow to be resolved by a detector . we therefore look at the non - interference part of the spectrum @xmath79 which is plotted in fig . [ fig : t ] . one sees that the difference between the spin states lies only in the interference effects for this range of parameters . accordingly the non - interference part of the spectrum is the same for either spin state and the scalar case . we also find that the radial momentum @xmath80 is tightly confined by the laser frequency @xmath36 . ) for an amplitude field strength @xmath78 and transversal momentum @xmath81 in scalar qed ( dashed ) as well as in spinor qed ( solid ) . the two different spin states are plotted as a thin red line . the left side shows the spectrum for @xmath82 and a pulse length of @xmath83 , which are parameters comparable to the ones of @xcite . one sees that one spin state depending on the sense of rotation @xmath75 dominates the spectrum . on the right sight one can see a part of the spectrum for @xmath76 and @xmath84 ( which corresponds to a frequency of @xmath85 ) as a function of the displaced radial momentum @xmath86 . for these parameters the spectra of the two spin states are merely displaced with respect to the scalar one and each other . observe the scale of the plot which lets conclude that interference effects will be smeared out by a detector . a plot omitting these effects for a wider range of momenta can be found in fig . [ fig : t ] . , scaledwidth=100.0% ] ) for an amplitude field strength @xmath78 and transversal momentum @xmath81 in scalar qed ( dashed ) as well as in spinor qed ( solid ) . the two different spin states are plotted as a thin red line . the left side shows the spectrum for @xmath82 and a pulse length of @xmath83 , which are parameters comparable to the ones of @xcite . one sees that one spin state depending on the sense of rotation @xmath75 dominates the spectrum . on the right sight one can see a part of the spectrum for @xmath76 and @xmath84 ( which corresponds to a frequency of @xmath85 ) as a function of the displaced radial momentum @xmath86 . for these parameters the spectra of the two spin states are merely displaced with respect to the scalar one and each other . observe the scale of the plot which lets conclude that interference effects will be smeared out by a detector . a plot omitting these effects for a wider range of momenta can be found in fig . [ fig : t ] . , scaledwidth=100.0% ] ) for an amplitude field strength @xmath78 , transversal momentum @xmath81 , a pulse length of @xmath76 and for @xmath68 which corresponds to a frequency of @xmath87 . the left side shows the spectrum for @xmath88 in scalar qed ( dashed ) as well as in spinor qed ( solid ) . the two different spin states are plotted as a thin red line . as already seen in fig . [ fig : ms_t = infty ] for this range of parameters the only difference between the spin states and the scalar case are interference effects . on the right side the spinor spectrum is plotted for @xmath89 . one sees that the radial momentum @xmath80 of the produced pairs depends characteristically on the laser parameters . , scaledwidth=100.0% ] ) for an amplitude field strength @xmath78 , transversal momentum @xmath81 , a pulse length of @xmath76 and for @xmath68 which corresponds to a frequency of @xmath87 . the left side shows the spectrum for @xmath88 in scalar qed ( dashed ) as well as in spinor qed ( solid ) . the two different spin states are plotted as a thin red line . as already seen in fig . [ fig : ms_t = infty ] for this range of parameters the only difference between the spin states and the scalar case are interference effects . on the right side the spinor spectrum is plotted for @xmath89 . one sees that the radial momentum @xmath80 of the produced pairs depends characteristically on the laser parameters . , scaledwidth=100.0% ] the total particle production per volume can be calculated by integrating over the momentum spectrum @xmath90 this integration normally has to be performed numerically . + for the rotating rectangle pulse we study parameters for which @xmath68 . as discussed in section [ sec : rotating ] for this cases @xmath67 pairs of turning points contribute to the pair creation rate . in fig . [ fig : totalyield ] the total particle yield for spinor qed is plotted for @xmath91 . for @xmath92 we also show how the total yield is distributed with respect to spin . as already mentioned in [ sec : rotating ] one finds that the pair creation rate is dominated by pairs with one spin depending on the sense of rotation of the field @xmath75 for a wide range of pulse lengths @xmath65 . for @xmath93 the ratio of spin states goes to @xmath0 . this is due to the fact that for longer pulses the effect of the rotation is smaller . accordingly for @xmath93 the results are asymptotic to those of the non - rotating rectangle pulse with length @xmath65 which has a total pair creation rate of on page . ] @xmath94 where @xmath95 is the compton volume . since the non - rotating rectangle pulse is an one - component field , half of the particles have either spin . we find that for @xmath96 there are no interference effects in the total particle number . this is clear since only one turning point contributes . as in the momentum spectrum we see that there are interference effects in the total particle number for @xmath72 . + of the rotating rectangle pulse for @xmath78 and @xmath97 as a function of the pulse length @xmath65 . for @xmath92 we also plotted the pair creation in the respective spin states . one sees that the pair creation is dominated by one of them for a wide range of parameters since the line for one spin direction overlaps the sum of both up to @xmath98 . observe that there are interference effects for @xmath72 . ] we introduce a semiclassical treatment of the dirac - equation for two - component time - dependent electric fields . in the limit of one - component fields our result reduces to the well known wkb - approximation . we use this method to investigate the momentum spectrum and total pair creation of schwinger pair production semiclassically . + we find that in difference to the one - component case , the pair creation rate for spinor qed can be dominated by pairs of one spin direction . due to this fact the pair creation rate of spinor qed differs from the one of scalar qed ( treated in @xcite for two - component fields ) even for the case of one ( dominant ) pair of turning points . this is evident in rotating field pulses , where the spin state that most pairs are created in , depends on the sense of rotation of the field for a certain range of pulse lengths and frequencies . + for a constant rotating field the momentum spectrum can be computed analytically . however one has to use a rectangle pulse to get a finite pair creation rate . if one chooses the parameter @xmath68 the momentum spectrum turns out to be rotationally invariant in the @xmath99 plane as was also found for the sauter pulse for high enough @xmath100 @xcite . this can be explained by the fact that for long enough pulses the effects of the pulse shape get less important and the effects of rotation are more apparent . as for the sauter pulse we find interference effects in the momentum spectrum of the produced pairs . these effects become stronger for higher @xmath100 . in addition to this we also find interference effects in the total particle number . + as was discussed in @xcite , the characteristic momentum spectra for rotating pulses could be a decisive fingerprint for schwinger pair creation if the qed cascade seeded by a created pair can be distinguished from one seeded by vacuum impurities . we showed that additionally to the form of the momentum spectrum , pairs have a preferred spin direction in the parameter range chosen in @xcite . for pulse lengths and frequencies reached by laser systems at the moment , both the characteristic interference patterns and the spin dependence could not be resolved by a detector . however the radial momentum spectrum of created pairs is very narrow . depending on the parameters of future laser systems these results can also help with the differentiation discussed above . the authors thank antonino di piazza , christoph h. keitel , hagen kleinert and clment stahl for fruitful discussions . es is supported by the erasmus mundus joint doctorate program by grant number 2012 - 1710 from the eacea of the european commission . the ansatz ( [ eq : ansatz1])-([eq : ansatzdot2 ] ) can be derived from the usual treatment for the one - component case ( see e.g. @xcite ) . there one uses the wkb - solution and introduces the coefficients @xmath101 and @xmath102 for the positive and negative solution of the zeroth order in @xmath103 respectively . putting this in the dirac equation one finds coupled differential equations where the time derivative of @xmath101 depends only on @xmath102 and vice versa . + since the zeroth order wkb solution in the two - component case is equivalent to the one - component case , we can make the ansatz @xmath104 where we introduced the coefficients @xmath105 where @xmath106 and used @xmath107 defined in eq . ( [ eq : kint ] ) . we now want to choose these coefficients such that the differential equations for @xmath108 and @xmath109 take the same form as in the one - component case . to do so we use ( [ eq : firstansatz1 ] ) and ( [ eq : firstansatz2 ] ) in the coupled dirac equations ( [ eq : coupleddirac1 ] ) and ( [ eq : coupleddirac2 ] ) to find @xmath110 as discussed before we now want to choose @xmath105 such that the last term in ( [ eq : ansatzindirac1 ] ) and ( [ eq : ansatzindirac2 ] ) vanishes . this leads to @xmath111 additionally eqs . ( [ eq : ansatzindirac1 ] ) and ( [ eq : ansatzindirac2 ] ) should not contradict each other and thus @xmath112 by analogy we find @xmath113 and @xmath114 using @xmath115 in eq . ( [ eq : fb2ndeq ] ) , sorting by terms proportional to the derivative of the momenta and requiring @xmath116 and @xmath117 to be independent we find two equations for @xmath118 . setting these equal leads to @xmath119 a solution of this equation is given by @xmath120 using this and @xmath121 in eq . ( [ eq : fb2ndeq ] ) we find @xmath122 one can check that the solutions ( [ eq : fb1fb2 ] ) and ( [ eq : fa1fa2 ] ) also fulfill eq . ( [ eq : fa2ndeq ] ) . this means we are left with the three equations combining @xmath123 and @xmath124 as well as @xmath125 and @xmath126 respectively . + if we now use eq . ( [ eq : fa1fa2 ] ) in the first equation of ( [ eq : fa1steq ] ) we find @xmath127f_{\alpha1}^\pm(t).\end{aligned}\ ] ] we can integrate this equation to find @xmath128 where the integral @xmath129 is defined in eq . ( [ eq : k_xy ] ) . + using eq . ( [ eq : fa1fa2 ] ) again we find @xmath130 one can now check if the two solutions ( [ eq : fa1sol ] ) and ( [ eq : fa2sol ] ) fulfill the remaining equation namely the second equation of ( [ eq : fa1steq ] ) . this is the case and thus we can use them in our ansatz . by analogy we find @xmath131 we can now put these results into eq . ( [ eq : ansatzindirac1 ] ) to find @xmath132}{2 cp_\parallel(t){{\mathcal{e}}}(t)}\frac{c_\beta^\pm}{c_\alpha^\pm}{{\text{e}}}^ { { { \text{i}}}k_{{{\pm}}}(t)}\beta^\pm(t ) . \label{eq : alphac}\end{aligned}\ ] ] to get the value of @xmath133 we can compare to the one - component case of @xcite . to do so we use @xmath134 to find @xmath135 using this in eq . ( [ eq : alphac ] ) we find @xmath136 this result is equivalent to eq . ( 14 ) of @xcite if @xmath137 . using this result in eqs . ( [ eq : firstansatz1 ] ) and ( [ eq : firstansatz2 ] ) leads to the ansatz used in section [ sec : scmomspec ] , i.e. eqs . ( [ eq : ansatz1])-([eq : ansatzdot2 ] ) [ app : rotana ] in this appendix we compute the integrals @xmath138 and @xmath139 given by eqs . ( [ eq : kint ] ) and ( [ eq : k_xy ] ) for the constant rotating field ( [ eq : erot ] ) . note that the imaginary part of @xmath138 was already computed in @xcite . for this purpose we introduce the adiabatic parameter @xmath140 + for the calculation of @xmath138 we first change the variable of the integral to the dimensionless @xmath141 such that eq . ( [ eq : kint ] ) takes the form @xmath142 we now change the variable to @xmath143 where for the rotating electric field defined in eq . ( [ eq : erot ] ) the kinetical momenta are given by @xmath144 for this special field configuration we find ( for details see @xcite ) @xmath145 where we defined @xmath146 one finds that the sign in eq . ( [ eq : dxdt ] ) changes at @xmath147 and that the new variable takes the values @xmath148 and @xmath149 . thus performing the substitution ( [ eq : sub ] ) the integral ( [ eq : kint2 ] ) takes the form @xmath150 where @xmath151 is a global phase factor which stems from integrating from @xmath152 to the turning point @xmath153 . the first integral corresponds to the integral parallel to the imaginary @xmath154-axis and will give rise to the imaginary part of @xmath138 while the second one gives the real contribution . + it is possible to solve the integrals in eq . ( [ eq : kwithints ] ) leading to ( see @xcite eqs . 3.141.2 and 3.141.5 ) @xmath155+\phi , \label{eq : krot}\end{aligned}\ ] ] were we use the elliptic integrals ( see @xcite eqs . 8.111 - 112 ) @xmath156 we can compute the integral @xmath139 in an analogous way . from eq . ( [ eq : k_xy ] ) follows @xmath157 this can be brought into the form @xmath158 where @xmath159 is again a global phase . so that we find ( see @xcite eqs . 3.131.3 , 3.131.6 , 3.137.4 and 3.137.6 ) @xmath160+\phi_{xy}\\ & + 2k\frac{\,g}{\sqrt{1-y_+^2}}\left[\left(1-y_-y_+\right){{\text{\textbf{k}}}}\left(\sqrt{\frac{y_-^2-y_+^2}{1-y_+^2}}\right)+({y_-^2 - 1})\frac{y_+}{y_-}{{\boldsymbol{\pi}}}\left(\frac{1}{y_-^2}\frac{y_-^2-y_+^2}{1-y_+^2},\sqrt{\frac{y_-^2-y_+^2}{1-y_+^2}}\right)\right ] . \label{eq : k_xyrot } \end{split}\end{aligned}\ ] ]	we semiclassically investigate schwinger pair production for pulsed rotating electric fields depending on time . to do so we solve the dirac equation for two - component fields in a wkb - like approximation . the result shows that for two - component fields the spin distribution of produced pairs is generally not @xmath0 . as a result the pair creation rates of spinor and scalar quantum electro dynamics ( qed ) are different even for one pair of turning points . for rotating electric fields the pair creation rate is dominated by particles with a specific spin depending on the sense of rotation for a certain range of pulse lengths and frequencies . we present an analytical solution for the momentum spectrum of the constant rotating field . we find interference effects not only in the momentum spectrum but also in the total particle number of rotating electric fields .
in @xcite for every automorphism @xmath13 of a non - abelian free group @xmath10 of finite rank @xmath14 an _ index _ @xmath15 has been defined , which counts in a natural way attracting fixed points at the gromov boundary @xmath16 and the rank of the fixed subgroup @xmath17 of @xmath18 . if @xmath19 , then @xmath20 is simply the number of attractors of the homeomorphism @xmath21 induced by @xmath18 . as main result of @xcite the index inequality @xmath22 has been proved , which strengthens the celebrated scott conjecture , proved in @xcite , and also extends some well known consequences of nielsen - thurston theory for surface homeomorphisms to free group automorphisms , in particular after passing to the _ stable index _ @xmath23 ) of the associated outer automorphism @xmath24 , defined below in ( [ eq : index - inequality - star ] ) as sum of @xmath25 for suitable representatives @xmath26 of a positive power of @xmath24 . the main difference to surface homemorphisms , however , where the analog indices always sum up to give via gauss - bonnet the maximal possible value postulated in ( [ eq : index - inequality - star ] ) , is that @xmath15 and even @xmath27 may well be strictly smaller than @xmath28 . ever since it has been an open question which precise value of @xmath29 @xmath30 can be realized as stable index @xmath31 by any @xmath32 , in particular if one restricts to automorphisms @xmath24 of @xmath10 which are _ irreducible with irreducible powers ( iwip ) _ , also called _ fully irreducible _ ( see section [ sec : prelim - clp ] ) . for any given @xmath32 its representatives @xmath33 are partitioned into _ isogredience classes _ , where isogredient automorphisms are conjugate by inner automorphisms and hence have conjugate @xmath16-dynamics and thus equal indices . the _ stable index list _ of @xmath24 is defined as the longest sequence ( up to permutation ) of positive indices @xmath34 , given by pairwise non - isogredient representatives @xmath26 of some power @xmath35 , for any fixed @xmath36 . the inequalities @xmath37 have been shown in @xcite . handel and mosher ( * ? ? ? * question 6 in 1.5 ) have asked explicitly which such values are realized as stable index list by iwip automorphisms of @xmath10 . we denote such a ( potential ) index list by @xmath38 $ ] , where the @xmath39 are usually given in decreasing order . for the `` maximal '' case , i.e. @xmath40 , a complete answer to this question has been given by masur and smillie @xcite : for @xmath41 any list @xmath42 $ ] of positive @xmath43 , with @xmath44 ( other than the single exceptional case @xmath45 $ ] for @xmath46 , see section [ sec : proof - discussion ] below ) , can be realized as index list of an iwip automorphism which is _ geometric _ , i.e. @xmath24 is induced by a pseudo - anosov homeomorphism of a surface with one boundary component . on the other hand , if @xmath47 , then any iwip @xmath24 is known not to be geometric . the purpose of this paper is to show that the analogue of masur and smillie s result holds also in the non - maximal case : [ thm : main ] let @xmath41 , and let @xmath48 be any list of positive numbers from @xmath49 which satisfy : @xmath50 then there exists ( and we give an explicit construction ) an iwip automorphism @xmath32 which realizes the given list of values @xmath39 as stable index list . for @xmath51 the statement of the theorem had already been proved c. pfaff @xcite . other special cases were also known , for example the single element list @xmath52 for any @xmath41 ( see @xcite ) . a further discussion , including some experimental data obtained by the first author , is given in section [ sec : proof - discussion ] below ( compare also ( * ? ? ? * section vi ) ) . already in @xcite the relationship between the index of @xmath24 and the branching index of a forward limit @xmath53-tree tree @xmath54 of @xmath24 has been exploited ( compare also @xcite ) . if @xmath24 is iwip , then such @xmath54 in the thurston boundary @xmath55 of ( unprojectivized ) outer space @xmath56 is unique up to rescaling , and for non - geometric @xmath24 the isometric @xmath10-action on @xmath54 is free and has dense orbits . for a suitable exponent @xmath36 there is a natural 1 - 1 correspondence , between isogredience classes of representatives @xmath26 of @xmath35 with @xmath57 on one hand , and @xmath10-orbits of branch points @xmath58 of @xmath54 on the other , where @xmath59 is precisely equal to the number of directions at @xmath58 . an exposition of this relationship is given in 8 of @xcite . this correspondence can be carried one step further by using the fact that @xmath54 is obtained as projective limit ( in @xmath60 ) of simplicial metric trees @xmath61 with free isometric @xmath10-action , which occur naturally as universal cover of a train track representative @xmath62 of @xmath24 ( see [ sec : prelim - clp ] ) . such train track representatives carry an _ intrinsic gate structure _ which allows one to define a gate index at every vertex of @xmath63 and a gate index list by considering all periodic vertices of @xmath63 with 3 or more gates . there is a natural relationship between the gates of @xmath61 and the branching directions of @xmath54 , which in the absence of so called _ periodic inps _ ( see [ sec : prelim - clp ] below ) becomes a 1 - 1 correspondence . again , see 8 of @xcite for more details . the problem of realizing a given list @xmath64 $ ] as in theorem [ thm : main ] as stable index list of an iwip automorphism can hence be subdivided into the following subproblems : 1 . [ problem : graph ] construct a graph @xmath63 with vertices @xmath65 and define a gate structure @xmath66 on @xmath63 which realizes the given list of the values @xmath39 as gate indices at the vertices @xmath67 . [ problem : iwip - but - inp ] define a map @xmath68 which respects the gate structure @xmath66 and is `` iwip up to inps '' . [ problem : control - inp ] control the periodic inps of @xmath69 . subproblems ( [ problem : graph ] ) and ( [ problem : iwip - but - inp ] ) are solved below in sections [ sec : graph ] and [ sec : map - iwip - but - inp ] . subproblem ( [ problem : control - inp ] ) , which is the hardest and conceptually the most interesting , requires a new tool , called _ long turns _ , which has been provided and investigated by the first two authors in the `` companion paper '' @xcite . in section [ sec : legalizing - map ] we give a brief summary of this method and provide the concrete tools that allow us in section [ proofs ] to apply the results of @xcite in order to obtain a _ legalizing _ train track morphism @xmath70 . it is then shown how theorem 1.1 of @xcite ( quoted in section [ proofs ] in an appropriate version ) can be applied to solve the left - over subproblem ( [ problem : control - inp ] ) for the resulting train track map @xmath71 . _ acknowledgements : _ this paper was intended by the authors to be joint with catherine pfaff : a large part of it is rooted in our weekly discussions with catherine , during the months before she left marseille . we regret that she declined despite our insistence to be coauthor of the paper . we also would like to point the reader to the thesis work of sonya leibman @xcite , which came only very recently to our attention . some of her results seem to be very interesting to the context of the work presented here ; in particular , there is an overlap of the results of her section 5.2 ( lemma 5.4 ) and our subsection 7.1 below . we will use in this paper the same terminology as set up in sections 2 and 3 of @xcite : a _ graph _ @xmath63 is always connected , without vertices of valence @xmath72 or @xmath73 , and moreover it is finite , unless it is the universal covering of a finite connected graph . the _ edges _ @xmath74 of @xmath63 come in pairs @xmath75 which differ only in their orientation , and @xmath76 contains precisely one of the two elements in each pair . a _ gate structure _ @xmath66 on @xmath63 is a partition of the edges @xmath77 into equivalence classes @xmath78 ( called _ gates _ ) , where equivalent edges must have the same initial vertex @xmath79 . two edges @xmath80 with same initial vertex form a _ turn _ @xmath81 , which is called _ legal _ ( with respect to @xmath66 ) if @xmath82 and @xmath83 belong to distinct gates , and _ illegal _ if they belong to the same gate . the turn @xmath81 is called _ degenerate _ if @xmath84 . a path @xmath85 _ crosses over _ a _ gate turn _ @xmath86 if for some @xmath87 one has @xmath88 or @xmath89 . the path @xmath90 is _ legal _ if , for each @xmath87 , the edges @xmath91 and @xmath92 belong to different gates of @xmath66 ( i.e. @xmath90 crosses only over legal turns ) . the _ gate index _ @xmath93 at a vertex @xmath79 is given by @xmath94 , where @xmath95 denotes the number of gates at @xmath79 . a _ graph map _ @xmath96 maps vertices to vertices and edges to ( possibly unreduced ) edge paths . the map @xmath97 has _ no contracted edges _ if for any edge @xmath82 of @xmath63 the combinatorial length (= number of edges traversed ) of @xmath98 satisfies @xmath99 . in this case @xmath97 induces a well defined map @xmath100 on @xmath74 which maps the edge @xmath82 to the initial edge of the path @xmath98 . the _ transition matrix _ @xmath101 of @xmath97 is defined as non - negative matrix , where @xmath102 counts the number of times that @xmath98 crosses over @xmath83 or over @xmath103 . the equality @xmath104 is a direct consequence of the definition of the transition matrix . recall that a non - negative matrix @xmath105 is called _ primitive _ if some positive power @xmath106 is _ positive _ , i.e. all coefficients of @xmath106 are strictly positive . a graph map @xmath96 is a _ train track morphism _ , with respect to a given gate structure @xmath66 on @xmath63 , if it has no contracted edges , and if @xmath97 maps every legal path to a legal path . it is shown in @xcite that a train track morphism has the additional property that at every periodic vertex @xmath79 of @xmath63 any illegal turn is mapped to an illegal turn , or equivalently : @xmath97 induces at every periodic vertex @xmath79 of @xmath63 a bijective map from the gates at @xmath79 to the gates at @xmath107 . note that in this paper all train track morphisms that occur have only periodic vertices ; indeed , each vertex is a fixed point . for a graph @xmath63 without preassigned gate structure , a _ train track map _ @xmath108 in the classical sense as defined by bestvina and handel @xcite ( and hence in particular any _ train track representative _ of a given automorphism of @xmath10 ) is a graph map with no contracted edges with the property that for any @xmath109 and any edge @xmath82 , @xmath110 is a reduced path . as legal paths are reduced , any train track morphism @xmath111 is always a classical train track map . conversely , every classical train track map @xmath62 is a train track morphism with respect to the _ intrinsic _ gate structure @xmath112 on @xmath63 , defined by @xmath97 through declaring two edges @xmath113 with same initial vertex to lie in the same gate if and only if for some @xmath114 the edge paths @xmath110 and @xmath115 have non - trivial initial subpaths in common . notice however that , for a train track morphism @xmath97 with respect to some gate structure @xmath66 , the intrinsic gate structure @xmath116 may be strictly finer than the given gate structure @xmath66 . a reduced path @xmath117 in @xmath63 is a _ periodic inp _ for a train track morphism @xmath96 if @xmath90 and @xmath118 are legal and for some @xmath114 the path @xmath119 is homotopic relative endpoints to @xmath120 . the _ gate - whitehead - graph _ @xmath121 of a train track morphism @xmath96 at a vertex @xmath79 of @xmath63 has the gates @xmath78 of @xmath66 at @xmath79 as vertices and a ( non - oriented ) edge connecting @xmath78 to @xmath122 if and only for some edge @xmath82 of @xmath63 and some integer @xmath114 the edge path @xmath110 crosses over the gate turn @xmath123 . recall also that an automorphism @xmath32 is called _ iwip _ ( or _ fully irreducible _ ) if no positive power of @xmath24 fixes the conjugacy class of any proper free factor of @xmath10 . in this and the following sections , let @xmath124 and @xmath125 be given as in theorem [ thm : main ] . in this section we will build a graph @xmath63 with @xmath126 which has precisely @xmath5 vertices @xmath65 , and has at each vertex @xmath67 precisely @xmath127 gates : the gate structure @xmath66 on @xmath63 realizes the given list @xmath128 as gate index list . note that from the inequalities in theorem [ thm : main ] we obtain @xmath129 as initial assumption on the number of gates in @xmath63 . we divide the possible index lists in three different cases : 1 . the _ even case _ : @xmath130 is even ( that is to say @xmath131 is an integer @xmath132 ) . 2 . the _ odd case _ ( non - maximal ) : @xmath130 is odd and smaller than @xmath133 ( alternatively : @xmath134 ) . the _ maximal odd case _ : @xmath135 ( i.e. @xmath136 ) . we consider a circle which is subdivided at vertices labeled @xmath137 , to obtain oriented edges labeled @xmath138 such that @xmath139 starts at @xmath67 and ends at @xmath140 ( for @xmath141 understood modulo @xmath5 ) . note that if @xmath142 then @xmath143 is a loop edge at the sole vertex @xmath144 of @xmath63 . at each vertex @xmath67 we add @xmath145 germs of edges to this circle . in the odd and maximal odd cases we remove one of these germs at @xmath144 such that in any cases the number of germs is even . we group these germs arbitrarily into pairs to form @xmath146 oriented edges @xmath147 . here @xmath146 is the largest integer @xmath148 , with @xmath149 exactly if the index list is equal to @xmath150 $ ] . in the even and odd cases let @xmath151 , where we note that @xmath152 . we add @xmath153 oriented edges @xmath154 which are loops at the vertex @xmath144 . in the maximal odd case we set @xmath155 and add a single edge @xmath156 which is a loop at @xmath144 . finally , in the odd case we add an extra edge @xmath157 which is a loop at @xmath144 . the graphs @xmath63 defined above are connected , without vertices of valence @xmath72 or @xmath73 , and have fundamental group @xmath10 . the _ oriented edge set _ is given by @xmath158 in the even and maximal odd cases , and by @xmath159 in the odd case . in all cases we have @xmath160 and @xmath161 , with @xmath162 if and only if we are in the odd case with index list @xmath150 $ ] . we define the gate structure @xmath66 on @xmath63 in such a way that each gate consists of a single edge , except for the following gates , all based at @xmath144 : * in the even case : @xmath163 , @xmath164 ; * in the odd case : @xmath165 , @xmath164 and @xmath166 ; * in the maximal odd case : @xmath167 . notice that in the maximal odd case @xmath168 and @xmath169 belong to distinct gates . as a consequence , at every vertex @xmath67 there are precisely @xmath170 gates , so that we obtain : [ prop : graph - index - realization ] the gate structure @xmath66 on @xmath63 realizes the given list of values @xmath171 as gate indices at the vertices @xmath172 of @xmath63 . the following will be used crucially in the subsequent sections : [ lem : hamiltonian - circuit ] let @xmath63 be the graph equipped with the gate structure @xmath66 as defined above . 1 . [ item : ueuprime ] for each edge @xmath173 in @xmath76 there exists a legal loop @xmath174 in @xmath63 which starts in @xmath175 does not end in @xmath175 , does not pass through @xmath176 or @xmath177 and passes exactly once through @xmath82 ( we allow @xmath178 or @xmath179 to be trivial ) . [ item : v ] for each gate turn @xmath180 , except for gate turns involving @xmath181 in the maximal odd case , there exists a legal loop @xmath79 which starts in @xmath182 , does not end in @xmath182 , does not pass through @xmath156 or @xmath168 , and crosses over the gate turn @xmath183 . [ item : alphabeta ] in the even and odd cases , for any edge @xmath82 in @xmath175 there exist legal paths @xmath184 and @xmath185 that do not pass through any of the @xmath186 or through @xmath82 ( and neither through their inverses ) . furthermore , @xmath187 is a legal path which ends in @xmath188 , and @xmath185 is a legal loop based at @xmath144 that does not start in @xmath175 or @xmath189 , and does not end in @xmath175 . we allow @xmath184 to be trivial . [ item : alphabetaprime ] symmetrically , in the even and odd cases , for each edge @xmath190 in @xmath188 there exist legal paths @xmath191 and @xmath192 that do not pass through any of the @xmath186 or through @xmath82 ( nor through their inverses ) , such that @xmath193 is legal and starts in @xmath175 , while the legal loop @xmath192 is based at @xmath144 but does not start in @xmath189 and does not end in @xmath175 or @xmath188 . we allow @xmath191 to be trivial . the above statements ( [ item : ueuprime ] ) and ( [ item : v ] ) are easy to verify if one keeps in mind that at every vertex there are @xmath194 gates , and that every vertex @xmath67 can be reached from @xmath144 by any one of two disjoint paths on the circle @xmath195 , so that it is easy to avoid any given edge in @xmath63 . concerning statement ( [ item : alphabeta ] ) , if @xmath196 or @xmath197 , we let @xmath184 be trivial . otherwise we set @xmath198 . in the odd case we let @xmath199 . in the even case , there is at least one edge @xmath200 ( or @xmath201 ) exiting from @xmath144 . let @xmath202 be the endpoint of @xmath200 ( or of @xmath201 ) , and set @xmath203 if @xmath204 and @xmath205 otherwise . for statement ( [ item : alphabetaprime ] ) the paths @xmath191 and @xmath192 can be chosen symmetrically to @xmath184 and @xmath185 in the above case ( [ item : alphabeta ] ) . in this section we construct for the graph @xmath63 a train track morphism @xmath68 with respect to the gate structure @xmath66 specified in the last section . the morphism @xmath69 will be `` fully irreducible up to inps '' in that it has primitive transition matrix @xmath206 and connected gate - whitehead - graphs at every vertex ( compare ( * ? ? ? * propositions 4.1 and 4.2 ) ) . below we will consider graph maps @xmath207 with the following properties : @xmath208 [ lem : property - star ] let @xmath209 and @xmath210 be graph maps which satisfy the above properties @xmath208 . 1 . [ item : composition- * * ] then the composition @xmath211 satisfies @xmath208 as well . [ item : composition - iwg ] moreover , for any vertex @xmath67 of @xmath63 the gate - whitehead - graph @xmath212 contains both gate - whitehead - graphs @xmath213 and @xmath214 as subgraphs . statement ( [ item : composition- * * ] ) is a direct consequence of the definition of properties @xmath208 . statement ( [ item : composition - iwg ] ) has been proved under slightly more general hypotheses as proposition 3.10 in @xcite . the properties ( [ * : fix - vertex ] ) and ( [ * : fix - gate ] ) above imply that a map which satisfies @xmath208 acts as identity on the set of gate turns . as a consequence one derives easily that statement ( [ item : composition - iwg ] ) of lemma [ lem : property - star ] can actually improved to @xmath215 . we define below several graph maps on @xmath63 where we use the following : [ convention : no - untouched - edges ] in this and the following sections , in the definition of a graph map @xmath216 we always use the convention that any edge with no explicitly defined image is mapped identically to itself . for any edge @xmath217 let @xmath178 and @xmath179 be as in lemma [ lem : hamiltonian - circuit ] ( [ item : ueuprime ] ) . define @xmath218 by : @xmath219{rcl}a_1&\mapsto&ueu'a_1\\ e&\mapsto&eu'a_1ue \end{array}\ ] ] note that the @xmath220-image of @xmath82 passes through @xmath156 and that the @xmath220 image of @xmath156 passes through @xmath82 . for any gate turn @xmath221 of @xmath63 , except for gate turns involving @xmath181 in the maximal odd case , let @xmath79 be as in lemma [ lem : hamiltonian - circuit ] ( [ item : v ] ) and define @xmath222 by : @xmath223 let @xmath224 be the composition of all these maps @xmath220 ( with @xmath173 ) and @xmath225 , where we do not care about the order of the composition . define @xmath226 through @xmath227 . [ prop : h - positive ] the map @xmath68 is a train track morphism with respect to @xmath66 . furthermore @xmath69 fixes every vertex @xmath67 of @xmath63 , maps each gate of the gate structure @xmath66 to itself , and is a homotopy equivalence . in addition , the transition matrix @xmath206 is positive , and the gate - whitehead - graph @xmath228 of @xmath69 at any vertex @xmath79 of @xmath63 is connected . we first consider the maps @xmath220 with @xmath229 and @xmath225 as defined above . properties ( [ * : train - track])-([:e - image ] ) of @xmath208 above are easily verified directly . for property ( [ : homot - eq ] ) the reader can check directly that the map given by @xmath230{rcl } a_1&\mapsto&\bar u ' \bar e \ , \bar u a_1 ^ 2\\ e&\mapsto & \bar u \ , \bar a_1 u e \end{array}\ ] ] is a homotopy inverse of @xmath220 . the fact that it is not a train track map with respect to @xmath66 is irrelevant . for @xmath225 a homotopy inverse is given simply by @xmath231 . from lemma [ lem : property - star ] it follows now that the maps @xmath224 and @xmath69 have this property @xmath208 , which is the statement of the first paragraph to be shown . in order to show that @xmath206 is positive , we use the equality @xmath232 from section [ sec : prelim - clp ] and condition ( [ * : e - image ] ) of property @xmath208 to obtain that the image @xmath233 of any edge @xmath82 crosses over @xmath156 , and that the image @xmath234 of @xmath156 crosses over every edge @xmath235 . hence the image @xmath236 of any edge @xmath82 passes through all edges of @xmath63 . this proves that the transition matrix @xmath206 is positive . from lemma [ lem : property - star ] we know that the gate - whitehead - graph of @xmath69 contains that of @xmath225 , for each gate turn @xmath221 . it follows from the above definition of @xmath225 via @xmath237 and from the definition of @xmath79 in lemma [ lem : hamiltonian - circuit ] ( [ item : v ] ) that in the even and the odd cases the gate - whitehead - graph of @xmath69 at each vertex of @xmath63 is a complete graph and thus connected . in the maximal odd case there are no maps @xmath225 for the gate turns involving the gate @xmath181 . but in this case the gate turn @xmath238 is crossed over by @xmath239 . this is enough to get that the gate - whitehead - graph of @xmath69 at @xmath144 is connected . the goal of this ( and the following ) section is to construct a train track morphism @xmath240 with respect to @xmath66 which is a homotopy equivalence and is `` legalizing '' . this notion has been introduced in @xcite , and is now briefly summarized : a pair @xmath241 of non - trivial legal paths @xmath90 and @xmath118 in @xmath63 is called in @xcite a _ long turn _ if the _ branches _ @xmath90 and @xmath118 start at the same vertex but have distinct initial edges . the set of long turns in @xmath63 , with both branches of length equal to some integer @xmath242 , is denoted by @xmath243 . the long turn @xmath241 can be _ legal _ or _ illegal _ , according to whether its _ starting turn _ @xmath244 , formed by the initial edges of @xmath90 and @xmath118 respectively , is legal or illegal ( as defined for turns in the traditional sense , see [ sec : prelim - clp ] ) . if neither @xmath245 is a subpath of @xmath246 nor conversely , then the long turn is called _ @xmath247-long _ , and the long turn , obtained from @xmath245 and @xmath246 through erasing from both the maximal common initial subpath , is called the _ @xmath247-image _ of @xmath241 and denoted by @xmath248 . a train track morphism @xmath247 is called _ legalizing _ if for some sufficiently large constant @xmath242 every long turn @xmath249 is @xmath247-long , and if @xmath248 is legal ( or , equivalenty , if the starting turn of @xmath248 is legal ) . to avoid a misunderstanding , we d like to point out that any non - degenerate turn @xmath81 in the classical sense can be considered alternatively as long turn with branches of length 1 . in particular , if @xmath81 is @xmath247-long , then it has both , an image turn @xmath250 , as well as an image long turn @xmath251 , which furthermore has a starting turn @xmath252 that is again a turn in the classical sense . however , in general @xmath253 and @xmath252 will be quite different : for example , @xmath250 may well be degenerate , while @xmath252 is by definition always non - degenerate . to construct the desired train track morphism @xmath247 we define now train track morphisms @xmath254 . in each of the cases considered below the `` variable '' @xmath255 denotes a non - degenerate illegal turn in @xmath63 , interpreted here as long turn with branch length 1 . the reader can verify directly that all of the maps @xmath254 defined below satisfy the statements ( [ * : train - track ] ) , ( [ * : fix - vertex ] ) and ( [ * : fix - gate ] ) of property @xmath208 from section [ sec : map - iwip - but - inp ] . we use again convention [ convention : no - untouched - edges ] . we first deal with the : 1 . let @xmath256 be an illegal turn in @xmath63 with @xmath257 , and with @xmath258 or @xmath259 where @xmath260 . let @xmath184 and @xmath185 be as in lemma [ lem : hamiltonian - circuit ] . define : @xmath261{rcl}a_i&\mapsto & a_i e\alpha\\ e&\mapsto&a_i \beta a_i e \end{array}\ ] ] the illegal turn @xmath256 is @xmath254-long and mapped by @xmath262 to the long turn @xmath263 which is legal . symmetrically , let @xmath264 be an illegal turn in @xmath63 with @xmath265 , and with @xmath266 or @xmath259 where @xmath260 . let @xmath191 and @xmath192 be as in lemma [ lem : hamiltonian - circuit ] . define : @xmath261{rcl}a_i&\mapsto & \alpha ' ea_i\\ e&\mapsto&ea_i \beta ' a_i \end{array}\ ] ] the illegal turn @xmath264 is @xmath254-long and mapped by @xmath262 to the long turn @xmath267 which is legal . the we have one more illegal turn @xmath268 . define : @xmath269{rcl } a_1&\mapsto&a_1\bar d c_1\cdots c_\ell\\ d&\mapsto&da_1\bar d \end{array}\ ] ] the illegal turn @xmath268 is @xmath254-long and mapped by @xmath262 to the long turn @xmath270 which is legal . in the there is only one illegal turn @xmath271 . as the rank @xmath272 is greater or equal to @xmath273 , there is at least one edge @xmath274 which starts from some vertex @xmath275 and ends at some @xmath202 . we set @xmath276}=c_1\cdots c_\ell$ ] if @xmath277 and @xmath276}=c_1\cdots c_{k-1}$ ] if @xmath278 . we furthermore set @xmath279}=1 $ ] if @xmath204 , and @xmath279}=c_{k'}\cdots c_\ell$ ] if @xmath280 . we define : @xmath261{rcl } a_1&\mapsto&c_1\cdots c_\ell c_{[1,k]}b_1c_{[k',\ell]}a_1\\ c_1&\mapsto&c_1\cdots c_\ell c_{[1,k]}b_1c_{[k',\ell]}a_1c_1\\ b_1&\mapsto&b_1c_{[k',\ell]}a_1c_{[1,k]}b_1 \end{array}\ ] ] [ lem : easy - check ] in the maximal odd case , every long turn of length equal to @xmath281 with starting turn @xmath282 is @xmath254-long and mapped by @xmath262 to a legal long turn . let @xmath283 be the long turn under consideration . we first observe that , if @xmath284 , then @xmath285 has starting turn @xmath286 and hence is legal . on the other hand , if @xmath287 , then we note similarly that @xmath285 is legal unless @xmath288 . however , in this last case we compute @xmath289}a_1c_1 , c_k\cdots c_\ell c_{[1,k]}b_1c_{[k',\ell]}a_1c_1 ) \,\,\ , { \rm if } \,\,\ , \ell \neq 1 \ , , \ ] ] and @xmath289}a_1c_1 , c_{[1,k]}b_1c_{[k',\ell]}a_1c_1 ) \,\,\ , { \rm if } \,\,\ , \ell = 1,\ ] ] which are both legal long turns . we now verify : [ lem : homotopy - inverses ] each of the above defined maps @xmath254 is a homotopy equivalence . for each case of the map @xmath254 we list below a map @xmath290 ; the reader can verify directly that they are homotopy inverses of the maps @xmath254 . even and odd cases : \(1 ) @xmath291{rcl } a_i&\mapsto & e \alpha \bar a_i \bar \beta \\ e&\mapsto & \beta a_i \bar \alpha \ , \bar e a_i \bar \alpha \end{array}\ ] ] \(2 ) @xmath291{rcl } a_i&\mapsto & \bar \beta ' \bar a_i \alpha ' e \\ e&\mapsto & \bar \alpha ' a_i \bar e \ , \bar \alpha ' a_i \beta ' \end{array}\ ] ] \(3 ) @xmath291{rcl } a_1&\mapsto & a_1 \bar c_\ell \ldots \bar c_1 d c_1 \ldots c_\ell \bar a_1 \\ d&\mapsto & d c_1 \ldots c_\ell \bar a_1 \end{array}\ ] ] maximal odd case : @xmath291{rcl } a_1&\mapsto & \bar c_{[k ' , \ell ] } \bar b_1 \bar c_{[1 , k ] } a_1 \bar c_\ell \ldots \bar c_1 a_1 ^ 2 \bar c_\ell \ldots \bar c_1 a_1 ^ 2 \\ c_1&\mapsto & \bar a_1 c_1 \\ b_1&\mapsto & \bar c_{[1 , k ] } \bar a_1 c_1 \ldots c_\ell \bar a_1 c_{[1 , k ] } b_1 \end{array}\ ] ] we thus have proved : [ prop : local - legalizing ] let @xmath292 in the even and odd cases and @xmath293 in the maximal odd case , and let @xmath255 be any illegal turn of @xmath63 . for each long turn @xmath294 of @xmath63 , with branch length @xmath295 and with starting turn @xmath255 , the map @xmath254 is a train track morphism with the property that @xmath294 is @xmath254-long and mapped by @xmath262 to a legal long turn . furthermore , @xmath254 is a homotopy equivalence which fixes every vertex of @xmath63 and every gate of @xmath66 . proposition [ prop : local - legalizing ] is the main ingredient we need to obtain the desired legalizing map . this is done through an application of proposition 7.1 of @xcite which we quote now , for the convenience of the reader in a slightly weakened form and with terminology adapted to the present paper : [ prop : quote-1 ] let @xmath63 be a graph with a gate structure @xmath66 . assume that there exists a constant @xmath296 , and assume furthermore : 1 . [ item : hyp - legalizing ] for each illegal long turn @xmath255 with branch length @xmath295 there exists a train track morphism @xmath297 such that @xmath255 is @xmath254-long and mapped by @xmath262 to a legal long turn . [ item : hyp - positive ] there exists a train track morphism @xmath226 which satisfies @xmath298 for any edge @xmath82 of @xmath63 . [ item : hyp - hom - eq ] all the above maps @xmath254 and @xmath69 are homotopy equivalences . then there exists a legalizing train track morphism @xmath299 which is obtained as a composition of the @xmath254 and @xmath69 . before going back to the situation considered in the previous sections , we first quote the main result of @xcite , in a slightly strengthened version due to remark 6.6 of @xcite and adapted to the circumstances here : [ thm : quote-2 ] let @xmath63 be a graph with gate structure @xmath66 , let @xmath300 be a train track morphism with positive transition matrix @xmath301 and gate - whitehead - graph at every vertex that is connected . let @xmath240 be a legalizing train track morphism with respect to the gate structure @xmath66 which is a homotopy equivalence that fixes every vertex of @xmath63 and every gate of @xmath66 . then : 1 . [ item : thm - quote-2-iwip ] the map @xmath302 is a train track representative of an iwip automorphism @xmath32 . [ item : thm - quote-2-no - inp ] there is no periodic inp in @xmath63 for the train track map @xmath303 . in particular there are no non - trivial @xmath304-periodic conjugacy classes in @xmath10 . [ item : thm - quote-2-index ] the stable index list for @xmath24 is given by the gate index list for @xmath63 defined by @xmath66 at the @xmath97-periodic vertices of @xmath63 . we will now go back to the graph @xmath63 from the previous sections , i.e. with gate structure @xmath66 that realizes the given index list from theorem [ thm : main ] as gate indices . we will show below how to use the previously derived train track morphisms @xmath69 and @xmath254 on @xmath63 via the above quoted results from @xcite to finish the proof of theorem [ thm : main ] . we first observe : [ cor : g - legalizing ] let @xmath63 be the graph with gate structure @xmath66 defined in section [ sec : graph ] for the given list of gate indices . then there exists a legalizing train track morphism @xmath305 which is a homotopy equivalence and fixes each vertex and each gate of @xmath63 . we use proposition [ prop : local - legalizing ] to obtain the hypothesis ( [ item : hyp - legalizing ] ) of proposition [ prop : quote-1 ] , where we note that if @xmath254 legalizes a long turn @xmath306 with branch length @xmath307 , then it also legalizes any long turn @xmath255 with branch length @xmath308 which contains @xmath306 as `` subturn '' . we note that hypothesis ( [ item : hyp - positive ] ) of proposition [ prop : quote-1 ] is satisfied by the train track morphism @xmath226 from proposition [ prop : h - positive ] . hypothesis ( [ item : hyp - hom - eq ] ) is satisfied , as has been shown in propositions [ prop : h - positive ] and [ prop : local - legalizing ] . thus proposition [ prop : quote-1 ] assures us the existence of the legalizing map @xmath247 as product of @xmath69 and the @xmath254 . since by propositions [ prop : h - positive ] and [ prop : local - legalizing ] the latter are all homotopy equivalences that fix every vertex of @xmath63 and every gate of @xmath66 , this proves the claim . we note that the map @xmath68 from section [ sec : map - iwip - but - inp ] satisfies by proposition [ prop : h - positive ] all of the requirements of the map @xmath97 in theorem [ thm : quote-2 ] . by corollary [ cor : g - legalizing ] the same is true for the map @xmath247 obtained in corollary [ cor : g - legalizing ] . hence part ( [ item : thm - quote-2-iwip ] ) of theorem [ thm : quote-2 ] shows that @xmath309 represents an iwip automorphism @xmath24 of @xmath310 , and part ( [ item : thm - quote-2-index ] ) assures that the stable index list of @xmath24 is equal to the gate index list of @xmath63 with respect to @xmath66 , which is built in section [ sec : graph ] to realize the given list of values @xmath311 , see proposition [ prop : graph - index - realization ] . [ rem : intrinsic - disappeared ] there is a subtlety in the above proof which we would like to point out to the reader , concerning the topic `` given gate structure '' versus `` intrinsic gate structure '' ( as defined in section [ sec : prelim - clp ] ) . it shows up in relation to two aspects which are relevant in our context : 1 . the index of the automorphism represented by a train track morphisms @xmath96 with respect to a gate structure @xmath66 depends on the intrinsic gate structure @xmath116 , which may well be strictly finer than @xmath66 . 2 . a map @xmath240 , which is a train track morphism with respect to two distinct gate structures @xmath66 and @xmath312 , may well be legalizing with respect to @xmath66 but not with respect @xmath312 . ( in this case , however , @xmath313 must be strictly finer than @xmath312 ) . in the situation considered above , both potential problems are resolved as follows by the application of theorem [ thm : quote-2 ] : the train track morphism @xmath69 constructed in section [ sec : map - iwip - but - inp ] may indeed well have an intrinsic gate structure @xmath314 that is strictly finer than the previously defined gate structure @xmath66 . however , in lemma 5.9 of @xcite it has been shown that a train track morphisms @xmath247 with respect to a gate structure @xmath66 which is legalizing ( with respect to @xmath66 ) satisfies indeed @xmath315 . now , since the composition of any train track morphism with a legalizing train track morphism is again legalizing ( see ( * ? ? ? * proposition 5.8 ) ) , by the same argument as before we obtain automatically @xmath316 . we will now discuss some further aspects of the index of free group automorphisms : handel and mosher ( * ? ? ? * , question 5 ) ask what values for the _ index deficit _ @xmath317 for any ( iwip ) automorphism @xmath32 are possible , and whether for @xmath318 the maximal index deficit goes to @xmath319 . from our theorem [ thm : main ] it follows directly that for every @xmath320 every value @xmath321 in @xmath322 with @xmath323 is achieved as index deficit for some iwip @xmath24 . the maximal index deficit is hence equal to @xmath324 , which indeed tends to @xmath319 for @xmath325 . this result has independently been obtained also by sonya leibman @xcite . we now consider in some detail the results of masur and smillie @xcite , in particular their theorem 2 on p. 291 : the translation of the terminology used there for quadratic differentials and pseudo - anosov homeomorphisms to the usual terminology for free group automorphisms is not completely evident . we give here a bit of translation help : in the absence of punctures on the surface @xmath105 , a @xmath326-pronged singularity in @xcite translates into an isogredience class of automorphisms @xmath26 with @xmath327 . in this case we would have to translate the genus @xmath247 of @xmath105 , multiplied by 2 , into the rank @xmath272 of the free group , except that without punctures @xmath328 is not free , which explains the summand @xmath329 in the index equality in part ( a ) of theorem 2 of @xcite . now , the @xmath330 punctures which masur and smillie admit in their theorem 2 appear nowhere explicitly , but in fact they can be essentially anywhere : if a puncture lies outside of the singularities and outside the separatrices raying out from them , then it lies on some regular leaf of the stable foliation , and hence it becomes a `` @xmath73-prong singularity '' , thus contributing a value @xmath331 to the given list . the automorphisms @xmath26 in the corresponding isogredience class have two attracting fixed points on @xmath16 and @xmath332 , which adds up to @xmath333 if a puncture coincides with a singularity , say with @xmath326 prongs , then , by the analogous reasoning , we obtain @xmath334 . this explains also why a value @xmath335 , which they admit , does not lead to negative index of the corresponding isogredience class : any singularity with a single prong only must coincide with one of the punctures ! however , in the context of the paper here we have to add a further restriction : a pseudo - anosov surface homeomorphism induces an iwip automorphism if and only if the surface has only one puncture , so that we have in our context always the condition @xmath336 . we come now to the 4 exceptional cases listed in part ( c ) of their theorem 2 : the first case @xmath337 can not be realized by a pseudo - anosov map with non - orientable stable foliation , but according to their theorem 2 there must be a realization by a pseudo - anosov with orientable stable lamination . the last case , @xmath338 requires more than one puncture , so that it is ruled out by the previous paragraph . the third case , @xmath339 , adds up to @xmath340 , so that for @xmath341 one has @xmath342 : in this case all automorphisms are geometric , and hence there is no chance to realize the corresponding index list @xmath343 $ ] . however , in the paper here we always assume @xmath41 , so that this case does not occur . it remains the second exceptional case : @xmath344 . in this case we have @xmath345 . from @xmath336 we deduce the following two possibilities for the index list : @xmath346 $ ] or @xmath347 $ ] , according to which of the two singularities coincides with the puncture . the former index list is alternatively realized by the case @xmath348 from @xcite ( with the puncture at the singularity with only 1 prong ) , which satisfies all conditions of theorem 2 of @xcite . the other index list , @xmath347 $ ] , however , leads always back to their second exceptional case @xmath344 , and hence can not be realized by a geometric automorphism . it remains as last `` left - over mystery '' of the index realization problem for @xmath41 the question whether the index list @xmath349 $ ] can perhaps be realized by a parageometric automorphism of @xmath350 . our realization result naturally leads to the question of frequency of the different index lists . thanks to the program developped by the first author in python and sage we were able to do the following numerical experiments . fix a finite alphabet @xmath351 with @xmath272 letters and let @xmath352 be the free group on @xmath351 . with convention [ convention : no - untouched - edges ] , an elementary nielsen automorphism is given by @xmath353 with @xmath354 , @xmath355 and @xmath356 . recall that elementary nielsen automorphisms form a generating set of @xmath357 and @xmath358 . each line of the table corresponds to a sample size of @xmath359 computed random automorphisms . computations were made at the math department in marseille , without compiling the sage code nor looking for serious optimization . note that those automorphisms commonly involve words with several thousands of letters . note also that those frequencies are note completly significant . first , rivin @xcite ( see also @xcite ) proved that the frequency of iwips goes to @xmath360 when the number of elementary nielsen automorphisms in the product goes to infinity ( but @xmath361 or even @xmath362 are quite small compare to infinity ) . different tests with the above data may lead to slightly different results . however , what seems to be significant is that : 1 . automorphisms with small indices are much more frequent that automorphisms with high indices . 2 . automorphisms with index greater than half the theoretical maximal ( @xmath363 ) almost never occur . in particular the maximal index @xmath28 never occured in our tests out of thousands of tries . several index lists seem to share positive frequency . the smallest index list : @xmath150 $ ] , is not always the most frequent . institut de mathmatiques de marseille + universit daix - marseille + 39 , rue frdric joliot curie + 13453 marseille cedex 13 + france + mailto:[email protected] [ ] + mailto:[email protected] [ ]	for any surface @xmath0 of genus @xmath1 and ( essentially ) any collection of positive integers @xmath2 with @xmath3 masur and smillie @xcite have shown that there exists a pseudo - anosov homeomorphism @xmath4 with precisely @xmath5 singularities @xmath6 in its stable foliation @xmath7 , such that @xmath7 has precisely @xmath8 separatrices raying out from each @xmath9 . in this paper we prove the analogue of this result for automorphisms of a free group @xmath10 , where `` pseudo - anosov homeomorphism '' is replaced by `` fully irreducible automorphism '' and the gauss - bonnet equality @xmath11 is replaced by the index inequality @xmath12 from @xcite .
over the last few years new observations and missions , energetic gamma - ray experiment telescope ( egret ) @xcite , _ chandra _ @xcite , tev- astronomy @xcite , revolutionized the measurement of radiation from a variety of objects in the universe . in most cases the _ primary _ source of the radiation is believed to be accelerated charged particles , often of remarkably high energies , such as @xmath0 or even higher @xcite , usually referred to as the ultra high energy cosmic rays ( uhecr ) . the accelerated particles themselves ( we will also use the term cosmic rays , crs ) are in many cases generated by shock waves ( shocks ) . note , that the latter are the major events where the huge energy of stars , supernovae s ( sn ) or black holes is released in bulk gas motions and ultimately dissipated . the most successful particle acceleration mechanism is , perhaps the diffusive shock acceleration ( dsa ) @xcite , which is a variant of the original fermi idea ( 1949 ) , also known as the first order fermi acceleration process . according to this mechanism particles gain energy by bouncing off hydromagnetic disturbances frozen into the converging upstream and downstream flow regions near a shock . clearly , the understanding of this mechanism is critical for radiation models since the primary particle spectrum is one of their most important `` input '' characteristics . the success of this mechanism has been due to the following appealing features 1 . it reproduces the power - law energy spectra with an index remarkably similar to that inferred from the observations of the galactic cosmic rays ( crs ) @xcite . 2 . convincing , direct observational evidence of its operation at interplanetary shocks and the earth bow shock @xcite 3 . the absence of any obvious intrinsic limitation to the maximum particle energy achievable by this process often , however , this mechanism is applied in its simplified version , the so called test particle ( tp ) or linear approximation ( that neglects the backreaction of accelerated particles on the shock structure and produces a simple @xmath1 particle energy distribution ) . although more realistic , some nonlinear theories do exist , they suffer from parameterization of particle transport coefficients such as the pitch - angle scattering , spatial diffusivity and plasma heating . also , it should be emphasized that the statement ( 2 ) above is only a prima facie evidence for the responsibility of this mechanism for galactic crs , alluded to in ( 1 ) . despite similarity between physical parameters @xcite , the life - times and extensions of shocks in these two environments are vastly different ( up to a factor @xmath2 ) . this requires different approximations for these shocks . namely , the treatment of large astrophysical shocks such as supernova remnant ( snr ) shells , shocks in the lobes of radio galaxies or even larger shocks in clusters of galaxies _ must _ include the back reaction of accelerated particles on the shock structure , and thus be _ intrinsically nonlinear_. this , in turn , necessitates the selfconsistent treatment of the above mentioned anomalous transport phenomena . from the perspective of the widely accepted snr shock hypothesis of the origin of galactic crs , perhaps , the most critical impact of this treatment would be on the statement ( 3 ) above . it implies that the maximum energy is limited only by the time available for acceleration and by the size of accelerating object ( roughly not to be exceeded by the diffusive escape length for particles , which grows with energy ) . this is becoming a `` hot '' issue in view of the lack of evidence for tev - protons in snr shocks ( despite the signature of gev - protons , egret ) , which might indicate that the particle spectrum either cuts off somewhere between gev and tev energies ( in a currently uncovered by any instrument energy range ) or probably significantly steepens there . we must await observations from gamma - ray large area space telescope ( glast ) and the next generation of the imaging atmospheric cherenkov telescopes that are currently being built @xcite to see where and how the spectrum disappears or changes . to understand how this _ may _ happen , one needs to consider all the crucial requirements for this mechanism , which are * the ability of some fraction of thermal particles downstream to make the first step to acceleration , to return upstream ( injection problem ) * good confinement of accelerated particles near the shock front ( to continue energy gain ) * the ability of the shock to withstand the pressure of accelerated particles ( the shock smoothing , or shock robustness problem ) these requirements are also associated with the main difficulties of the theory and are , in fact , strongly related . we consider them briefly , in order . injection is the process of initial particle energization . within the widely accepted `` thermal leakage '' scenario of injection , the way protons enter the shock acceleration process is physically the same as when they are accelerated afterwards ( see , ref.@xcite ) . after thermalization downstream , certain fraction of particles will catch up with the shock . their further leakage upstream generates alfven waves via the cyclotron resonance @xmath3 where @xmath4 is the cosine of the proton pitch angle , and @xmath5 is the particle momentum . these waves do two things . first , they self - regulate the thermal particle leakage ( injection ) by trapping particles downstream when their leakage becomes too strong and therefore the wave amplitude too large as well @xcite . second , the large amplitude waves scatter already accelerating particles in pitch angle , thus ensuring their diffusive confinement near the shock . according to the above resonance condition , waves excited by protons ( with @xmath6 are too long to scatter electrons , so that they interact with them adiabatically ( @xmath7 ) . therefore , electrons need a separate injection scenario . one suggestion is that they might be injected via scattering on self - generated whistler waves @xcite . according to another mechanism suggested in ref.@xcite , electrons are extracted directly from the thermal pool near the shock front via their interaction with proton - generated lower - hybrid waves and electrostatic barrier formed there due to this interaction . it should be noted that when the both species become relativistic and , if the synchrotron cooling time is longer than the acceleration time , the electron spectrum should be identical to that of the protons . however , it is electrons for which we have convincing direct evidence of fermi acceleration in snr shells . there are well documented measurements of nonthermal electron emission in radio , @xmath8 rays and possibly also in @xmath9 rays , presented in refs . @xcite , @xcite and @xcite , respectively . although electrons play perhaps dynamically unimportant role in the shock structure , they should trace the dynamically important proton spectra . as in the injection phase , further particle confinement near the shock is supported by self - generated alfven wave turbulence , since the energetic particle distribution ahead of the shock is ion - cyclotron unstable ( see , ref.@xcite ) . due to rapid pitch - angle scattering on these waves , particles cross the shock repeatedly , thus gaining energy . clearly , the scattering frequency ( turbulence level ) sets up the acceleration rate . note that the existing background magneto - hydrodynamic ( mhd ) turbulence in the interstellar medium ( ism ) would support only very slow acceleration . a long - standing problem is that the wave generation process is likely to be so robust that wave amplitude may far exceed the level admissible by the quasi - linear theory . it is usually expected that the turbulence saturates at a level @xmath10 . thus , particle scattering must occur via strongly nonlinear wave - wave and wave - particle interactions so that the conventional quasi - linear description should be replaced by a non - perturbative approach . mhd simulations significantly advanced recently ( see , eg @xcite and references therein ) but they would not suffice alone to self - consistently describe the wave generation by turbulently confined particles . the main difficulty is , as it will be seen in the sequel , the enormous extension of particle and wave spectra . it seems desirable to combine simulations in restricted but critical parts of the phase space ( as the short - wave , low - energy part that controlls the injection and the long - wave , high energy part where particle losses occur ) with an analytical approach in extended but tractable parts of the phase space where particle and wave spectra and even the flow profile exhibit relatively simple , scale invariant behaviour @xcite . it should be evident that a purely numerical approach would encounter serious difficulties if applied to the shock acceleration of uhecrs . indeed , even if we had numerical solution of the problem of cr acceleration in snr shocks , up to energies @xmath11 ev , we would need to extend its dynamical range by a factor of @xmath12 . since the integration domain in configuartion space is typically proportional to the maximum energy , this would mean @xmath13 times larger phase space . it seems natural to use this large size of the phase space instead of letting computer to process decade after decade of it with basically the same physics . the analytic approach that we introduce in sec.[kin - th ] below utilizes the corresponding small parameters and therefore appears particularly suitable to the problem of acceleration of uhecrs . due to the secular growth of the pressure of accelerated particles , the linear solution becomes invalid for shocks of sufficient life time , scale @xmath14 and magnetic field @xmath15 ( the most critical parameter is @xmath16 since the maximum particle energy scales as @xmath17 , provided that enough time is available for acceleration , as discussed , in ref.@xcite ) . the main nonlinear effect is due to the back reaction of energetic particles on the flow which reduces the velocity jump at the flow discontinuity ( subshock ) through the deceleration and heating of the plasma in front of it ( in the so called cr precursor ) , fig.[fig : setup ] . the total shock compression , however , increases due to the decrease in the adiabatic index @xmath18 caused by the presence of relativistic particles and , even more importantly , by their escape from the system ( akin to radiative shocks ) . at the same time the maintenance of a finite subshock within the global shock structure ( subshock itself , plus cr precursor ) is critical to the injection process and thus for acceleration in general . a mathematical limitation of the tp approximation can be obtained in terms of particle energy and injection rate @xmath19 , namely , @xmath20 @xcite , which states that the pressure of accelerated particles must remain smaller than the shock ram pressure . it is also clear from this condition that the above three issues are strongly coupled . injection ( @xmath19 ) and particle confinement ( which sets @xmath21 ) determine the shock structure and are , in turn , regulated by it . the acceleration time scale @xmath22 ( or particle diffusivity @xmath23 ) and the precursor turbulent heating rate are also implicitly involved in this `` feedback loop '' . there are indeed too many shock variables to obtain reliable prediction by simply scanning the parameter space . therefore , they need to be calculated self - consistently before the particle spectra can be calculated and compared with the observations . a typical nonlinearly accelerating shock is illustrated in fig.[fig : setup ] . the most `` visible '' nonlinear effects are : ( 1 ) the deceleration of the flow upstream ( see the flow profile on the top of fig . [ fig : setup ] ) by the pressure of accelerated particles , ( 2 ) subshock reduction , as a consequence ( its strength @xmath24 may become small compared to the total compression @xmath25 ( 3 ) bending of the b - field due to the compression of its tangential component because of the frozen in condition , @xmath26 , while the normal component is conserved due to @xmath27 . initially , the back reaction of accelerated particles ( crs ) on the shock structure was studied within the two - fluid model ( tfm ) . this model treats crs as a second fluid carrying the momentum and energy across the shock , but not the mass . complete solutions given in refs.@xcite indeed revealed a very strong back reaction of the crs onto the bulk plasma flow , leading to a bifurcation of the simple linear solution into a strongly nonlinear ( efficient ) solution with the acceleration efficiency approaching ( in strong shocks ) 100% due to the nonlinearly enhanced shock compression . this gives rise to a formally diverging ( in reality strongly dependent on the maximum energy ) cr pressure which makes the particle losses at the highest energies as important dynamically as injection , and strongly related to it through their feedback on the subshock strength . moreover , these losses are controlled by the alfven turbulence which , in turn , depends on the cr distribution in the shock precursor and thus on the losses themselves . unfortunately , the tfm , being a hydrodynamic theory , can not be closed in such a way that these essentially kinetic effects are properly represented . the earlier kinetic theories also demonstrated that _ upon accumulating enough cr energy , strong shocks develop into the nonlinear regime _ the minimal kinetic theory that captures bifurcations of the dsa can be formulated as follows . we describe the distribution of high energy particles ( crs ) by the so called diffusion - convection equation ( see , refs.@xcite ) . the gaseous discontinuity ( the subshock ) is assumed to be located at @xmath28 and , for convenience , we flip the @xmath29- coordinate in fig.[fig : setup ] , so that the upstream side is @xmath30 half - space . thus , the flow velocity in the shock frame can be represented as @xmath31 where the ( positive ) flow speed @xmath32 jumps from @xmath33 downstream to @xmath34 across the subshock and then gradually grows upstream up to @xmath35 . in a steady state the equation can be written as @xmath36 where @xmath37 is the isotropic ( in the local fluid frame ) part of the particle distribution . this is assumed to vanish far upstream ( @xmath38 ) , while the only bounded solution downstream is obviously @xmath39 . the most plausible assumption about the cosmic ray diffusivity @xmath40 is that of the bohm type , @xmath41 ( the particle momentum @xmath5 is normalized to @xmath42 ) . in other words @xmath23 scales as the gyroradius , @xmath43 . the reference diffusivity @xmath44 depends on the @xmath45 level of the alfvenic turbulence that scatters the particles in pitch angle . the minimum value for @xmath44 would be @xmath46 if @xmath47 . note that the replacement of this plain parameterization by a selfconsistent solution for the spectrum of turbulence driven by an inhomogeneous distribution of accelerated particles is a challenge to plasma physics . the existing quasi - linear approaches clearly fail in the efficient acceleration regime due to unacceptable wave amplitudes . the determination of @xmath32 in eq.([c : d ] ) requires three further equations . the first one is the conservation of the momentum flux in the smooth part of the shock transition ( @xmath30 , in the cr - precursor ) @xmath48 where @xmath49 is the pressure of the high energy particles @xmath50 it is assumed here that there are no particles with momenta @xmath51 ( they leave the shock vicinity because there are no mhd waves with sufficiently long wave length @xmath52 , since the cyclotron resonance requires @xmath53 ) . the momentum region @xmath54 can not be described by equation ( [ c : d ] ) and the behavior of @xmath55 at @xmath56 is described by the injection parameters @xmath57 and @xmath58 @xcite . the plasma density @xmath59 can be eliminated from equation ( [ mom : c ] ) by using the continuity equation @xmath60 . finally , the subshock strength @xmath61 can be expressed through the mach number @xmath62 at @xmath63 @xmath64 where the precursor compression @xmath65 and @xmath66 is the adiabatic index of the plasma . the system of equations ( [ c : d],[mom : c],[c : r ] ) describes in a self - consistent manner the particle spectrum and the flow structure , although under parameterization of such critical quantities as @xmath67 and @xmath68 . our poor knowledge of the maximum momentum @xmath68 is related to the prescribed form of particle diffusivity @xmath69 . it is useful to reduce this system to one integral equation @xcite . a key dependent variable is an integral transform of the flow profile @xmath32 with a kernel suggested by an asymptotic solution of the system ( [ c : d])-([mom : c ] ) which has the form @xmath70\ ] ] where @xmath71 is the flow potential and the spectral index downstream @xmath72 . the integral transform is as follows @xmath73 du(\psi ) \ ] ] and it is related to @xmath74 through the following formula @xmath75 the physical meaning of the function @xmath76 is very simple . it reflects the degree of shock modification . namely , a function @xmath77 is an effective flow velocity upstream as seen by a particle with momentum @xmath5 that diffusively escapes ahead of the shock to a point @xmath29 where the flow speed is @xmath78 . once @xmath76 is found , both the flow profile and the particle distribution can be determined by inverting transform ( [ u ] ) and integrating equation ( [ q(p ) ] ) . now , using the linearity of equation ( [ mom : c ] ) ( @xmath79 ) , we derive the integral equation for @xmath80 by applying the transformation ( [ u ] ) to the @xmath81 derivative of equation ( [ mom : c ] ) @xcite . the result reads @xmath82 ^{-1}\nonumber \\ & \times & \frac{u(t_{0})}{u(t')}\exp \left [ -\frac{3}{rr_{\rm{s}}}\int _ { t_{0}}^{t'}\frac{dt''}{u(t'')}\right ] \label{int : eq}\end{aligned}\ ] ] where @xmath83 , @xmath84 . here the injection parameter @xmath85 is related to @xmath86 by means of the following equation @xmath87 \right\ } ^{-1}\label{nu}\end{aligned}\ ] ] the equations ( [ c : r],[int : eq],[nu ] ) form a closed system that can be easily solved , numerically . asymptotic analytic solutions are also available @xcite . the physical quantities involved are : the far upstream mach number @xmath62 ( which is given , external parameter ) ; internal parameters : @xmath88 ( the mach number at the subshock ) , the injection rate @xmath19 , the particle maximum momentum @xmath89 , and particle diffusivity @xmath90 . these internal parameters must be determined self - consistently , but currently are parameterized or calculated using some simplifying assumption . for example , assuming that there is no turbulent heating in the shock precursor ( which is doubtful in such a turbulent environment ) , the parameter @xmath91 [ see eq.([c : r ] ) ] . the presence of bifurcation in this acceleration process is best seen in variables @xmath92 . the quantity @xmath93 is a measure of shock modification produced by crs , in fact @xmath94 [ eq.([mom : c ] ) ] and may be regarded as an order parameter . the injection rate @xmath67 characterizes the cr density at the shock front and can be tentatively treated as a control parameter . it is convenient to plot the function @xmath95 instead of @xmath96 [ using equations ( [ nu ] ) and ( [ c : r ] ) ] , fig.[fig : nuofr ] . in fact , the injection rate @xmath19 at the subshock should be calculated given @xmath97 with the self - consistent determination of the flow compression @xmath86 on the basis of the @xmath96 dependence obtained . however , in view of a very strong , even nonunique dependence @xmath96 , this solution can be physically meaningful only in regimes far from criticality , when @xmath98 ( test particle regime ) or @xmath99 ( efficient acceleration ) . there are , however , self - regulating processes that should drive the system _ towards critical region _ ( where @xmath96 dependence is very sharp ) @xcite . first , if @xmath19 is subcritical it will inevitably become supercritical when @xmath68 grows in course of acceleration . once it happened , however , the strong subshock reduction ( equation [ [ c : r ] ] ) will reduce @xmath19 as well and drive the system back to the critical regime , fig.[fig : bif ] . the maximum momentum @xmath68 is subject to self - regulation as well . indeed , when @xmath99 , the generation and propagation of alfven waves is characterized by strong inclination of the characteristics of wave transport equation towards larger wavenumbers @xmath100 on the @xmath101 plane due to wave compression . thus , considering particles with @xmath102 inside the precursor , one sees that they are in resonance with waves that must have been excited by particles with @xmath103 further upstream but , there are no particles with @xmath103 . therefore , the required waves can be excited only locally by the same particles with @xmath102 which substantially diminishes the amplitude of waves that are in resonance with particles from the interval @xmath104 . ( the left inequality arises from the resonance condition @xmath105 and the frequency conservation along the characteristics @xmath106 ) . this will worsen the confinement of these particles to the shock front . the quantitative study of this process is another challenge to the theory of plasma turbulence . what can be inferred from fig.[fig : nuofr ] now , is that the decrease of @xmath68 straightens out and rise the curve @xmath95 , so that it returns to the monotonic behaviour . however once the actual injection becomes subcritical ( and thus @xmath107 ) then @xmath68 will grow again restoring the two extrema on the curve @xmath95 . also the turbulent precursor heating straightens up the bifurcation diagram and returns it to the critical state . we illustrate this in fig.[fig : heat ] for different heating efficiencies @xmath108 , introduced phenomenologically through the decrease in the flow mach number across the cr precursor @xmath109 here the first term is due to the familiar adiabatic compression while the second describes the turbulent heating , assumed to be proportional to the cr pressure contrast @xmath110 ( source of the free energy for turbulence ) as well as to the precursor length @xmath111 . the plots in figs . [ fig : nuofr],[fig : heat ] demonstrate that parameterization approaches in which the internal parameters ( like @xmath112 and @xmath113 in many simulations ) are treated as given may be useful only when ( and if ) these parameters are well outside the critical region which is marked by very sharp or even nonunique parameter dependence . for example , as seen from fig.[fig : heat ] , if @xmath112 ( which is essentially unknown ) lies within the interval @xmath114 , then there are no means to calculate the flow structure ( @xmath86 ) , and thus the particle spectrum with a reasonable accuracy . to treat the injection rate in the critical region as a `` control '' parameter is equally useless , figs . [ fig : nuofr],[fig : heat ] . it appears to be more productive to _ assume _ a _ marginal state _ in which the maximum and the minimum of @xmath95 merge , at least in a sense of an averaged ( in time and space ) process . this means that @xmath115 at some @xmath116 . these two equations and the dependence @xmath95 itself , not only determine @xmath117 and @xmath118 but also provide an additional relation that involve other parameters of the problem which clearly enter the function @xmath119 . these are the mach number @xmath62 , the heating rate @xmath112 and the maximum momentum @xmath120 . for example , given @xmath62 and @xmath112 we can easily calculate @xmath121 @xcite . these results show , that at least in shocks with high mach numbers and no significant turbulent heating , the nonlinear effects may significantly limit the maximum energy achievable by this acceleration mechanism . before any relevance of this result to , the lack of observational evidence of tev protons in snr can be seriously debated , a number of difficult issues must be addressed which we discuss briefly in the next section . instead of relying on parameterization , and guessing about the possible values of relevant parameters it was suggested that , due to the self - regulation , the system should evolve precisely into the critical point within given parameter space @xcite . a comprehensive evaluation of this suggestion and the study of self - regulation mechanisms poses a serious challenge to plasma physics . note , however , that the critical self - organization approach was proven very useful in describing transport phenomena in laboratory plasmas , when the turbulence is generated by transport driving gradients , the pressure or density gradients , @xcite . the most important current issue is the construction of adequate links between the internal parameters ( and so the alfvenic turbulence ) , and their dependence upon the external parameters . this is an interdisciplinary problem in high energy astrophysics and nonlinear plasma physics . it combines hydrodynamics , particle kinetics with a strong emphasis on the theory of dynamical chaos and ergodicity ( injection ) , alfvenic turbulence , collisionless nonlinear plasma phenomena and , in particular , the theory of collisionless shocks . currently not all the aspects of this problem are well understood . we believe that we understand how the acceleration operates in the test particle ( linear ) regime , and have an analytic description of nonlinearly modified shocks including particle spectra , although under a prescribed bohm - type particle diffusion . injection can be calculated currently only for a relatively strong subshock , while we need it for varying shock parameters of dynamically evolving cr shocks . turbulence dynamics , associated plasma heating and the turbulent transport of crs are understood to even lesser extent . a systematic determination of parameters involved should allow the self - consistent calculation of particle spectra . specifically , this requires detailed understanding of several physical phenomena in a shock environment , which include : 1 . generation of waves by energetic protons , wave transport and spectral evolution 2 . turbulent transport of protons , in space and momentum , injection 3 . electron injection triggered by proton generated turbulence 4 . heating by alfvenic and ( magneto- ) acoustic turbulence , which controls the global shock structure the resolution of these issues will lead to the further progress in our understanding of particle acceleration and associated emission . some of the presented results have been obtained in collaboration with l. drury and h. vlk . we also acknowledge helpful discussions with f. aharonian and t. jones . r. c. hartman , d. l. bertsch , s. d. bloom , 79 ( 1999 ) john p. hughes , cara e. rakowski , anne decourchelle , electron heating and cosmic rays at a supernova shock from chandra x - ray observations of e0102.2 - 7219 , submitted to preprint : astro - ph/0007032 f. a. aharonian , 225 ( 1999 ) r. d. blandford physica scripta _ _ _ _ * * t85 , * * 191 ( 2000 ) l. oc . drury , rep . . phys . * 46 , * 973 ( 1983 ) r. d. blandford and d. eichler , , 1 ( 1987 ) l. a. fisk , * 76 , * 1662 , 1971 m. a. lee , * 87 , * 5063 ( 1982 ) c. f. kennel * 89 , * 5436 ( 1984 ) k. b. quest , , 9649 ( 1988 ) m. a. malkov , * 58 , * 4911 ( 1998 ) a. levinson , , 73 ( 1992 ) a.a . galeev , m.a . malkov , h.j . vlk , j. plasma phys . * 54(1 ) , 59 ( 1995 ) d. a. green 209 ( 1991 ) k. koyama , r. petre , e. v. gotthelf nature 378 * , 255 ( 1995 ) t. tanimori , l25 ( 1998 ) w.c . mller and d. biskamp * 84 * , 475 ( 2000 ) m.a . malkov , l.oc . drury , nonlinear theory of diffusive acceleration of particles by shock waves _ to appear in rep . on progr . in phys . malkov 638 ( 1997 ) l. oc . drury and h. j. voelk , 344 ( 1981 ) w.i . axford , e. leer and j.f . mckenzie , 317 ( 1982 ) m.a . malkov , p.h . diamond , h.j . vlk , * * 533 ( * * 2000 ) p.h . diamond , y.m . liang , b.a . carreras , p.w . terry , , 2565 - 2568 ( 1994 )	one of the main features of astrophysical shocks is their ability to accelerate particles to extremely high energies . the leading acceleration mechanism , the diffusive shock acceleration is reviewed . it is demonstrated that its efficiency critically depends on the injection of thermal plasma into acceleration which takes place at the subshock of the collisionless shock structure that , in turn , can be significantly smoothed by energetic particles . furthermore , their inhomogeneous distribution provides free energy for mhd turbulence regulating the subshock strength and injection rate . moreover , the mhd turbulence confines particles to the shock front controlling their maximum energy and bootstrapping acceleration . therefore , the study of the mhd turbulence in a compressive plasma flow near a shock is a key to understanding of the entire process . the calculation of the injection rate became part of the collisionless shock theory . it is argued that the further progress in diffusive shock acceleration theory is impossible without a significant advance in these two areas of plasma physics . # 1_proc . # 1th int . cosmic ray conf . _
soil hydraulic properties , represented by the soil water retention @xmath0 and hydraulic conductivity @xmath1 functions , dictate water flow in the vadose zone , as well as partitioning between infiltration and runoff . their evaluation has important implications for modeling available water resources and for flood forecasting . it is also crucial in evaluating the dynamics of chemical pollutants in soil and in assessing the potential of groundwater pollution . + soil hydraulic functions can be described by several mathematical expression @xcite , among them the van genuchten function @xcite . the determination of the parameters defining the van genuchten soil water retention function @xcite is usually done using laboratory experiments , such as the water hanging column @xcite . the hydraulic conductivity function can be estimated either in the laboratory , or in situ using infiltration tests . among the large number of existing infiltration tests @xcite , the single @xcite or double ring infiltrometers @xcite provide the field saturated hydraulic conductivity by applying a positive water pressure on the soil surface , while the disk infiltrometer @xcite allows to reconstruct the hydraulic conductivity curve , by applying different water pressures smaller than or equal to zero . for infiltration tests , the volume of infiltrated water versus time is fitted to infer the soil hydraulic conductivity at or close to saturation . these tests are time - consuming and difficult to apply to landscape - scale forecasting of infiltration . furthermore , their analysis involve various simplifying assumptions , partly due to the ignorance of the shape of the infiltration bulb . this lack of knowledge on the form of the infiltration bulb has to be filled to get accurate informations on the soil water retention @xmath0 function and consequently on hydraulic conductivity @xmath1 function . this can be done by water content sensing . vereecken @xcite and evett and parkin @xcite give a state of the art on the different techniques available for soil moisture measurements . among the large panel presented , geophysical methods take an important part , mainly because they are contact free and/or easy to use . the most commonly used hydro - geophysical methods are electrical resistivity measurements @xcite and electromagnetic methods @xcite . this paper focuses on the use of ground - penetrating radar ( gpr ) as a tool for monitoring water infiltration in soil . for few decades gpr has been known as an accurate method to highlight water variation in soils @xcite . different techniques are available in the literature for monitoring water content in soils using gpr . tomography imaging between boreholes during an infiltration has been done by binley @xcite and kowalsky @xcite among others . many advances were done during the last years on off - ground gpr using full waveform inversion , for instance to invert soil hydraulic properties ( lambot @xcite and jadoon @xcite ) . grote @xcite and lunt @xcite used two - way travel time variations from a reflector at a known depth to monitor water content variation with time . finally , multi - offset gpr survey techniques , i.e. cmp or warr , were carried out during infiltration processes in the works of greaves @xcite or mangel @xcite . the work presented here is based on mono - offset monitoring of infiltration with on - ground surface gpr as related by haarder @xcite , moysey @xcite , lai @xcite , dagenbach @xcite and saintenoy @xcite . haarder @xcite used a constant offset on - ground gpr coupled with dye tracing to exhibit preferential flows . they found that a gpr was able to map deep infiltration comparing to dye tracer , but they did not manage to resolve the infiltration patterns ( by - pass flow , fingering ... ) . moysey @xcite studied the infiltration inside a sand box from the surface with on - ground gpr . he used the reflection from the wetting front as well as from the ground wave and the bottom of the box , to monitor the water content . he also modelled his experiment and estimated the van genuchten parameters using semblance analysis . as lger @xcite , he found that the most poorly constrained parameter was @xmath2 . lai @xcite used a joint time frequency analysis coupled with grayscale imaging to measure infiltration and drainage in controlled conditions in laboratory . they were able to follow the peak frequency of the gpr wavelet associated with the wetting front using time frequency analysis and then determined the rate of water infiltration in unsaturated zone . saintenoy @xcite monitored the wetting bulb during an infiltration from a porchet infiltrometer . they were able to identify the dimension of the bulb with time and good agreement was found with modelling . on the continuity of those studies , we present a method for monitoring the wetting front during infiltration using on - ground gpr with fixed offset inside a ring infiltrometer . the objectives of this paper were i ) to check if the proposed method is accurate enough to monitor wetting front during infiltration with different boundary conditions , ii ) to invert saturated hydraulic conductivity using the model of mualem - van genuchten @xcite , and iii ) to analyze the uncertainties using a simplified mc uncertainty analysis . the method has been tested on synthetic examples and on two field data sets . in this study we consider one - dimensionnal vertical water flow in a soil , described by the one - dimensional richard s equation @xcite . its expression in term of water content is @xmath3~ , \label{rich}\end{aligned}\ ] ] where @xmath4 is the hydraulic conductivity as a function of water content , and @xmath5 is water diffusivity ( childs and georges - collis @xcite ) , expressed in terms of water content as @xmath6 . several mathematical functions exist to model the hydraulic properties of porous media @xcite . we chose the van genuchten model @xcite with the relation of mualem @xcite , giving the following expression for the water retention curve : @xmath7 where @xmath8 is the saturated water content , @xmath9 , the residual water content , and @xmath10 and @xmath2 , two fitting parameters which are respectively linked to the matric potential and the slope of the water retention curve at the inflexion point . the hydraulic conductivity function is described by @xmath11^{\frac{n}{n-1}}\right]^2~,\end{aligned}\ ] ] with @xmath12 the saturated hydraulic conductivity , @xmath13 the effective saturation and @xmath14 a factor that accounts for pore tortuosity . the @xmath14 parameter has an influence on the shape of the hydraulic conductivity function . however in this study we concentrated on the inversion of only one parameter , the saturated hydraulic conductivity . we fixed @xmath14 equal to 0.5 as reported in @xcite . several empirical and conceptual relationships exist to convert soil dielectric permittivity to volumetric water content . using the fact that the experiments presented here have been made in a quarry of fontainebleau sand , considered as pure silica , we used the crim relation @xcite , which relates the relative dielectric permittivity of bulk media , @xmath15 , to the volumetric summation of each components of it . thus for a tri - phasic medium comprising water , air and silicium , we obtain @xmath16 where @xmath17 are respectively the relative dielectric permittivity of water and silica , @xmath18 the porosity and @xmath19 the volumetric water content . surface gpr consists in a transmitting antenna , being a dipole , positioned on the surface , that emits short pulses of spherical electromagnetic ( em ) wave in response to an excitation current source , and a receiving antenna , also located at the surface , which converts the incoming em fields to an electrical signal source to be treated . following the works of annan @xcite , the velocity of electromagnetic waves is @xmath20 where @xmath21 is the loss factor as a function of the dielectric permittivity , frequency and electrical conductivity , @xmath22 is the real part of the relative dielectric permittivity , @xmath23 the relative magnetic permeability and @xmath24 is the velocity of em waves in air equal to @xmath25 . considering the case of non magnetic soil with low conductivity , in the range of 10 mhz to 1 ghz , the real part dominates the imaginary part of the dielectric permittivity and neglecting debye @xcite effect , equation ( [ v ] ) reduces to : @xmath26 we used this equation to compute the travelling time of an em wave through a layer of soil of known thickness with a given dielectric permittivity . numerous techniques are available for simulating gpr data , e.g. ray - based methods ( e.g. cai and mcmechan @xcite or sethian and popovici @xcite ) , time - domain finite - difference full - waveform methods ( e.g. kunz and luebbers @xcite or kowalsky @xcite ) , or finite differences time domain ( fdtd ) ( e.g. irving and knight @xcite ) . we used the gprmax 2d codes of giannopoulos @xcite , which uses fdtd modelling to solve the maxwell equations in 2 dimensions . we studied infiltration of a 5-cm thick water layer inside of a single ring infiltrometer in a sandy soil . the scheme of the apparatus is presented in figure [ schema ] . the single ring infiltrometer was a 1-mm thick aluminum cylinder with a 60-cm diameter , approximately 20-cm high , buried in the soil to a depth of 10 cm . gpr antennae ( namely the transmitter t and the receiver r ) were set up at a variable distance from the edge of the cylinder , noted @xmath27 in figure [ schema ] . in all our field experiments , we used a mala ramac system with antennae centered on 1600 mhz , shielded at the top . the inner part of the cylinder was covered with a plastic waterproof sheet . this allowed us to fill the cylinder with water and create an initial 5-cm thick water layer , while preventing infiltration into the sand before starting data acquisition . the beginning of the acquisition was launched by pulling away the plastic sheet to trigger water infiltration . the gpr system was set to acquire a trace every 10 s. with this apparatus , we performed two types of infiltration : i ) a falling head infiltration consisting of pulling away the plastic sheet and leaving water to infiltrate into the sand freely with no additional refill , and ii ) a constant head infiltration , when water was continuously added to the ring to maintain a 5-cm thick water layer during the infiltration experiment . in the following examples , we will show that gpr data acquired every 10 s during the infiltration experiment can be used to estimate the saturated soil hydraulic conductivity , @xmath12 . in all gpr data presented below , we subtracted the average trace and applied an automatic gain control ( agc ) to the data in order to make them clearer . the van genuchten parameters , @xmath28 of the sand have been determined in laboratory by several classical hanging water column experiments . we assumed arbitrarily a @xmath29 uncertainty for all the measured parameters . the sand was considered homogeneous . its initial water content , @xmath30 , and porosity , @xmath18 , of the soil were determined using gravimetric measurements on field samples . infiltration experiments were simulated by solving richards equation ( eq . ( [ rich ] ) ) using hydrus-1d . the soil profile was 50 cm deep , assumed to be homogeneous , and divided into 1001 layers . we used either an atmospheric boundary condition ( bc ) with no rain and no evaporation at the soil surface , for the falling head infiltration , or a constant pressure head of 5 cm to the top node , for the constant head infiltration , and for both case free drainage bc at the bottom . to simulate the 5-cm layer of water , the initial condition was set to a 5 cm pressure head in the top node . we simulated the first 10 minutes of the experiment with a time step of 10 s , i.e. , with 60 water content snapshots . using the crim relation ( eq . [ crim ] ) , each water content snapshot was converted to permittivity profiles ( made of 1001 points ) , considering a three - phase media : sand ( considered as pure silica ) , water , and air . each one of these permittivity profiles were the input for the gprmax2d program @xcite . gprmax2d gave simulated gpr monitoring of the infiltration process . we then picked the maximum amplitude of the signal to get the two way travel ( twt ) time of the wetting front reflection . our inversion algorithm was based on the comparison between the arrival times of the wetting front reflection observed in the radargrams acquired during the water infiltration experiment and the arrival times of these reflections computed from the theoretical water content profiles modeled by hydrus-1d . if a suitable relationship between water content and dielectric permittivity is known , water content profiles , obtained by the resolution of the richards @xcite equation ( done by hydrus1d in our case ) , can be transformed to a 2d series of reflection coefficients : @xmath31 where @xmath32 and @xmath33 are the relative dielectric permittivity at the infiltration time @xmath34 for two successive model cells centered at depth @xmath35 and @xmath36.the effective depth where the reflection coefficient is calculated is @xmath37 . knowing the dielectric permittivity of each layer of the profile , the electromagnetic wave velocity ( eq . [ huism ] ) and travel time can be computed . the travel time is used to interpolate reflection coefficients to a constant sampling interval . we used this depth to time conversion to compute a ricker signal in this time interval . the center frequency of the ricker was set to 1000 mhz , central frequency of the gpr signal recorded on the field . we derived it twice with respect to time to simulate the transformation made by the emitter and the receiver in real antennae . we then performed the convolution between this pseudo - gpr signal and the reflectivity to obtain @xmath38 where @xmath39 is the output signal , @xmath40 is the reflectivity and @xmath41 is the input source of the antenna . some remarks have to be made about the comparison between 1d - temporal convolution and real electromagnetic signal . first of all , our inversion algorithm is based on the assumption that soil can be represented as a stack of homogeneous layers . the assumption of horizontal interfaces forces the reflection coefficient ( equation ( [ refl ] ) ) to be expressed as a normal incidence case . secondly , we considered that the 2-d plane waves computed by fdtd algorithm ( modelling ) and 3-d plane waves ( experiments ) could be treated as a 1-d temporal convolution . third we neglect relaxation effects occurring when propagating an electromagnetic wave in water saturated sand . we used the twt time obtained from the radargram ( modelled or experimental ) as data to be fitted to derive the saturated hydraulic conductivity , assuming the other 4 van genuchten parameters and initial water content were known . using hydrus-1d , we generated 60 water content snapshots using the saturated hydraulic conductivity in the range from @xmath42 to @xmath43 , with a step of @xmath44 . for each value of @xmath12 , we calculated the twt time using our convolution algorithm and we computed the root mean square error ( rmse ) between these times and the data as an objective function , to be computed as function of saturated hydraulic conductivity . the @xmath12 value which corresponds to the minimum of the objective function was used as inverted value . the set of hydrodynamical parameters used for this numerical example is presented in table [ param ] . the permittivity profiles , resulting from water content conversions from hydrus-1d to permittivity and which were used as input of gprmax2d program @xcite are presented in figure [ algo1]-a . the simulated gpr monitoring of the infiltration process is shown in figure [ algo1]-b . the horizontal axis is the number of traces simulated with gprmax2d , two traces being separated by 10 seconds , as permittivity profiles are . the vertical axis is the twt time of the em wave coming back to the receiver . on the profile presented in figure [ algo1]-b , we denote one particular reflection , labeled a. its arrival time is increasing as the wetting front moves deeper . this reflection is interpreted as coming from the wetting front . the reflections labeled @xmath45 and @xmath46 are primary and secondary multiples of reflection @xmath47 . the reflection labeled @xmath48 is the wave traveling in air directly between the two antennae . after the 40@xmath49 trace , the 5-cm layer of water has been infiltrated , and drainage is starting . as a consequence , the permittivity of the upper part of the medium decreases and the velocity increases ( eq . [ huism ] ) . the twt time of reflection @xmath47 increases more slowly , creating a change of slope in the reflection time curve ( fig . [ algo1]-b ) . in figure [ algo1]-c , we display two curves : the twt time of the maximum peak of reflection @xmath47 ( obtained from figure [ algo1]-b ) and the twt time calculated by the convolution algorithm . the result of the convolution algorithm is in good agreement with the gprmax2d modelling . we used the twt time obtained from the radargram of figure [ algo1]-b as data to be fitted to derive the saturated hydraulic conductivity , assuming the other 4 van genuchten parameters and initial water content were known ( see table [ param ] ) . the rmse was minimized for @xmath50 , which has to be compared with the value set as input , i.e. , @xmath51 . this result confirms the ability of our algorithm to invert saturated hydraulic conductivity . the experiment took place in a quarry of fontainebleau sand in cernay - la - ville ( yvelines , france ) . the middle of the antennae was positioned 11 cm away from the cylinder wall ( @xmath52 in fig . [ schema ] ) . the 5-cm water layer was fully infiltrated after about 10 minutes , although in certain areas of the soil surface this time has been slightly shorter . the sand parameters measured by the hanging water column are given in table [ param ] and initial volumetric water content is @xmath53 . the recorded gpr data are shown in fig . [ r1 ] . in this profile , we denote three particular reflections . the one interpreted as coming from the infiltration front , labeled a , is visible during the first 30 minutes of the acquisition , with an arrival time varying from 2 ns down to 9 ns . the other reflections come from the cylinder and are interpreted in @xcite . we determined the arrival time of the a reflection peak and inverted the saturated hydraulic conductivity using the same algorithm as for the synthetic case . we obtained the minimum of the objective function for @xmath51 . in parallel , we also carried out disk infiltrometer experiments , using the multi - potential method @xcite . we obtained a value of the saturated hydraulic conductivity of @xmath54 . we attempted to evaluate the uncertainty in the saturated hydraulic conductivity retrieved from gpr data fitting by using a modified monte carlo method . we qualified this method as `` modified monte carlo '' in the sense that it is nor the tarantola method @xcite and neither the adaptive method proposed by the guide to the expression of uncertainty in measurement @xcite published by the joint committee for guides in metrology ( jcgm ) . we consider five major uncertainty sources , four from the van genuchten parameters , @xmath10 , @xmath2 , @xmath9 , @xmath8 and one from the initial water content @xmath30 . we do assume that all uncertainties can by described by gaussian distribution probability function centered on the value found by several water hanging column experiments with a standard deviation of @xmath29 of this value . with this definition we obtained the following set of a priori density function for experimental case : @xmath55 , @xmath56 , @xmath57 , @xmath58 , and @xmath59 , where the @xmath60 stands for the gaussian / normal probability density function and the @xmath61 and @xmath62 represent the mean and standard deviation . we generate multiple sets of parameters by sampling each gaussian distribution , @xmath63 , where the subscript `` @xmath64 '' is the iteration number . for each set the value of @xmath12 minimising the objective function was computed by our inversion procedure presented above . we generated enough sets of parameters such as the histogram of @xmath12 values look like a gaussian function with a stabilized standard deviation . we used this standard deviation as uncertainty on @xmath12 . we did not consider the uncertainties on radargram picking , because we evaluated it has a very weak influence comparing to the other uncertainties considered . using our analysis , we found in the case of falling head infiltration that @xmath12 was equal to @xmath65 . this narrow range of possible values is in agreement with disk infiltrometer value , and clearly shows the accuracy of our method . in this second case , a water layer of 5 cm above the ground was kept constant during the entire experiment . similarly as above , using the same van genuchten parameters as in the first synthetic example ( table [ param ] ) , we modeled infiltration of water inside a ring infiltrometer by applying a constant pressure head of 5 cm to the top node during 10 minutes . the permittivity profiles are presented in fig . [ algo2]-a , with each curve plotted every 10 s as in the previous case . [ algo2]-b shows the radargram simulated with gprmax2d . as can be seen , the reflection labeled @xmath47 describing the position of the infiltration front , is returning at increasing times , because infiltration is being constantly fed by the constant ponding depth , contrary to the previous falling head case . in fig . [ algo2]-c , we computed the twt time of the wetting front using the convolution algorithm and picking the @xmath47 reflection from the radargram in fig . [ algo2]-b . we inverted for the saturated hydraulic conductivity by minimizing the differences between the arrival times of the wetting front reflection obtained by the convolution algorithm and the arrival times picked from the simulated radargram in fig . [ algo2]-b . the objective function was minimized for @xmath66 , to be compared with the value used for simulating the data : @xmath51 . the experiment took place in the same quarry of fontainebleau sand as the previous experiment . the middle of the antennae was positioned in the middle of the ring ( x = 30 cm in fig . [ schema ] ) . the gpr data are shown in fig . [ r2 ] and were recorded during 80 minutes ( only a part of the radargram is presented ) . we used the van genuchten parameters determined in the laboratory using the hanging column experiments ( table [ param ] ) and we measured on sand core samples an initial volumetric water content of @xmath67 . in the profile presented in fig . [ r2 ] , the arrival time of reflection @xmath47 ranges from 0 at the beginning of the experiment to about 6 ns after 10 min . we picked the arrival time of the @xmath47 reflection peak and computed the objective function using the same procedure as described before . we obtained the minimum of the objective function for @xmath68 . again , this value has to be compared with the one obtained by the disk infiltrometer experiment , @xmath54 . using the same procedure as presented in the earlier field example , we found a range of possible values for the saturated hydraulic conductivity , @xmath69 . despite the fact that we are not in the same range as the disk infiltrometer method the discrepancy is very small and allows us to conclude on the good accuracy of our method . the results presented above indicate clearly that a commercial surface gpr can be used as a tool for monitoring the wetting front . although the use of surface - based gpr data to estimate the parameters of unsaturated flow models is not new @xcite , our method gives accurate values of the saturated hydraulic conductivity with uncertainties comparable or smaller than those obtained with disk infiltrometer measurements . a distinct advantage of our approach is the simplicity of the algorithm and its rapidity to converge , which is very encouraging for more complicated models ( stack of non - homogeneous layers ) . the discrepancy between saturated hydraulic conductivity determined by disk infiltrometry and that obtained with our gpr algorithm comes from different phenomena . first of all , the van genuchten parameters determined from the water hanging column experiment are obtained with saturation coming from the bottom of the soil samples , whereas in our case , the infiltration is a ponded one , thus coming from the top . despite the fact that we upgraded the single ring infiltrometer by the use of gpr to monitor the wetting front , we still suffer from the problem of entrapped - air , which causes reduction of saturated water content and hydraulic conductivity . this issue can not be fixed with ponded infiltration . disk infiltrometer measurement monitoring may cause less problems , working with negative matric potentials @xcite . during our modeling , we considered our soil as an homogeneous and isotropic one . real soils exhibit heterogeneities , triggering preferential flows . even in the case of our quarry of fontainebleau sand , differences in packing and compaction could lead to creation of preferential flow paths . one of the way to solve this issue could be to use a dual porosity model @xcite and a monte carlo procedure to generate a high number of soil models with different parameters , as we did with the single porosity model in hydrus-1d , and performed statistical analysis on the saturated hydraulic conductivity obtained . an other source of error , already discussed above , comes from the assumption that a 3d infiltration monitored by 3d electromagnetic waves can be treated as a 1-d temporal convolution . this limitation will be studied in future works , using hydrus 2d/3d to simulate 2d axisymmetrical infiltration and 2d infiltration . the results represent a promising step toward application of multi - parameters inversions . a first study in that direction was presented in lger @xcite . this research investigated the use of on - ground surface gpr to monitor the wetting front during infiltration inside a ring infiltrometer . we showed by modeling and experiments that a standard gpr device was able to monitor the displacement of the water front in the soil . we tested in synthetic cases the ability of our algorithm to invert the saturated hydraulic conductivity , knowing the other van genuchten parameters and the initial water content . two infiltration experiments were performed , falling head infiltration and constant head infiltration , in a quarry of fontainebleau sand . the retrieved saturated hydraulic conductivity was comparable to that obtained with disk infiltrometer experiments . uncertainty analysis accounting for all the van genuchten parameters , was performed using a modified monte carlo method , and proved the robustness of our algorithm . although results retrieved with gpr were in agreement with disk infiltrometry tests , we stress that further research is needed to improve our algorithm so as to determine the whole set of soil hydrodynamic parameters . angulo - jaramillo , r. , j .- vandervaere , s. roulier , j .- l . thony , j .- gaudet , and m. vauclin ( 2000 ) , field measurement of soil surface hydraulic properties by disc and ring infiltrometers : a review and recent developments , _ soil and tillage research _ , _ 55_(1 ) , 129 . binley , a. , p. winship , r. middleton , m. pokar , and j. west ( 2001 ) , high - resolution characterization of vadoze zone dynamics using cross - borehole radar , _ water ressources research _ , _ 37_(11 ) , 26392652 . bipm , iec , ifcc , ilac , iso , iupac , iupap , and oiml . ( 2011 ) , _ evaluation of measurement data \| supplement 2 to the - guide to the expression of uncertainty in measurement " - models with any number of output quantities . _ , joint committee for guides in metrology , jcgm:102 . dagenbach , a. , j. buchner , p. klenk , and k. roth ( 2013 ) , identifying a parametrisation of the soil water retention curve from on - ground gpr measurements . , _ hydrological earth system science _ , _ 17_(1 ) , 611618 . gerke , h. , and m. t. van genuchten ( 1993 ) , a dual porosity model for simulating the preferential movement of water and solutes in structured porous media , _ water ressources research _ , _ 29_(2 ) , 305319 . jadoon , k. , l. weihermller , b. scharnagl , m. kowalsky , m. bechtold , s. hubbard , h. vereecken , and s. lambot ( 2012 ) , estimation of soil hydraulic parameters in the field by integrated hydrogeophysical inversion of time - lapse ground - penetrating radar , _ vadose zone journal _ , _ 11_(4 ) . kowalsky , m. , p. dietrich , g. teutsch , and y. rubin ( 2001 ) , forward modeling of ground - penetrating radar using digitized outcrop images and multiple scenarios of water saturation , _ water ressources research _ , _ 37_(6 ) , 16151626 . kowalsky , m. , s. finsterle , j. peterson , s. hubbard , y. rubin , e. majer , a. ward , and g. gee ( 2005 ) , estimation of field - scale soil hydraulic and dielectric parameters through joint inversion of gpr and hydrological data , _ water ressources research _ , _ 41_(11 ) , w11,425.1w11,425.19 . lai , w. , s. kou , and c. poon ( 2012 ) , unsaturated zone characterization in soil through transient wetting and drying using gpr joint time - frequency analysis and grayscale images , _ journal of hydrology _ , _ 452 - 453 _ , 113 . lambot , s. , e. slob , m. vanclooster , and h. vereecken ( 2006 ) , closed loop gpr data inversion for soil hydraulic and electric property determination . , _ geophysical research letters _ , _ 33_(21 ) , l21,405.1l21,405.5 . lambot , s. , e. slob , j. rhebergen , o. lopera , k. jadoon , and h. vereecken ( 2009 ) , remote estimation of the hydraulic properties of a sand using full - waveform integrated hydrogeophysical inversion of time - lapse , off ground gpr data , _ vadose zone journal _ , _ 8_(3 ) , 743754 . lger , e. , and a. saintenoy ( 2012 ) , surface ground - penetrating radar monitoring of water infiltration inside a ring infiltrometerenetrating radar monitoring of water infiltration inside a ring infiltrometer , in _ 14th international conference on ground penetrating radar _ , shanghai . lger , e. , a. saintenoy , and y. coquet ( 2013 ) , estimating saturated hydraulic conductivity from surface ground - penetrating radar monitoring of infiltration , in _ 4th international conference hydrus software applications to subsurface flow and contaminant transport problems _ , pragues . mangel , a. , s. moysey , j. ryan , and j. tarbutton ( 2012 ) , multi - offset ground - penetrating radar imaging of a lab - scale infiltration test , _ hydrological earth system science discussions _ , _ 16 _ , 40094022 . roth , k. , r. schulin , h. fluhler , and w. attinger ( 1990 ) , calibration of time domain reflectometry for water content measurement using a composite dielectric approach , _ water ressources research _ , _ 26_(10 ) , 22672273 . zhou , q. , j. shimada , and a. sato ( 2001 ) , three - dimensionnal spatial and temporal monitoring of soil water content using electrical resistivity tomography , _ water ressources research _ , _ 37_(2 ) , 273285 . * 8\|@l@\| + & @xmath30 & @xmath9 & @xmath8 & @xmath10 & @xmath2 & @xmath12 & retrieved @xmath12 + & & & & ( @xmath70 ) & & ( cm / min ) & ( cm / min ) + numerical & 0.17 & 0.07 & 0.43 & 0.019 & 8.67 & 0.120 & 0.121 + field & 0.09 @xmath710.01 & 0.062 @xmath710.003 & 0.39 @xmath710.01 & 0.023 @xmath710.001 & 6.7 @xmath710.3 & 0.108 @xmath710.01@xmath72 & 0.120 @xmath710.013 + + & @xmath30 & @xmath9 & @xmath8 & @xmath10 & @xmath2 & @xmath12 & retrieved @xmath12 + & & & & ( @xmath70 ) & & ( cm / min ) & ( cm / min ) + numerical & 0.17 & 0.07 & 0.43 & 0.019 & 8.67 & 0.120 & 0.119 + field & 0.07 @xmath710.02 & 0.062 @xmath710.003 & 0.39 @xmath710.01 & 0.023 @xmath710.001 & 6.7 @xmath710.3 & 0.108 @xmath710.01@xmath72 & 0.089 @xmath710.009 +	in this study we show how to use gpr data acquired along the infiltration of water inside a single ring infiltrometer to inverse the saturated hydraulic conductivity . we used hydrus-1d to simulate the water infiltration . we generated water content profiles at each time step of infiltration , based on a particular value of the saturated hydraulic conductivity , knowing the other van genuchten parameters . water content profiles were converted to dielectric permittivity profiles using the complex refractive index method relation . we then used the gprmax suite of programs to generate radargrams and to follow the wetting front using arrival time of electromagnetic waves recorded by a ground - penetrating radar ( gpr ) . theoretically , the 1d time convolution between reflectivity and gpr signal at any infiltration time step is related to the peak of the reflected amplitude recorded in the corresponding trace in the radargram . we used this relationship to invert the saturated hydraulic conductivity for constant and falling head infiltrations . we present our method on synthetic examples and on two experiments carried out on sand soil . we further discuss on the uncertainties on the retrieved saturated hydraulic conductivity computed by our algorithm from the van genuchten parameters .
core collapse supernovae produce huge bursts of mev neutrinos which carry most of the gravitational energy ( @xmath2 ) of a massive star in a timescale of a few seconds . the released energy ( @xmath3 ergs ) is expected to be roughly equipartioned among the different flavors of neutrinos emerging promptly from the dense core , a few hours before the optical signal from the stellar envelope . the neutrino burst s time , flavor and energy structure carry information about the explosion mechanism @xcite and they can also be used to study vacuum and matter neutrino oscillations @xcite , the neutrino mass hierarchy @xcite or even exotic physics phenomena related to large extra dimensions and couplings to axions @xcite . since the sn1987a explosion when 24 neutrinos were detected by the kamiokande - ii , imb and baksan , many other experiments were designed @xcite to observe and study sn neutrinos . these detectors can also provide an early alarm for a sn explosion offering the chance to the astronomical community to make unprecedented observations of the early turn - on of the supernova light curve . for this purpose a supernova early warning system ( snews @xcite ) already exists providing also a trigger to detector facilities which are not able to trigger by themselves on a supernova signal . at the time of this writing , supernova neutrinos are the only neutrinos detected beyond our solar system . however , a lot of effort has been invested in observing high energy neutrinos from galactic and extragalactic sources . very large volume neutrino telescopes are designed @xcite and are already operating @xcite . such telescopes are consisting of thousands photomultiplier tubes ( pmt ) and instrumented volumes of the order of km@xmath4 . neutrino interaction products , mainly muons and electrons , are detected through the cherenkov light produced in water or in ice , providing a neutrino detection energy - threshold of about 100 gev . however , it has been shown @xcite that bursts of low energy ( mev ) sn neutrinos are detectable in neutrino telescopes in ice , by observing a global rise of the pmts counting rates which exceeds significantly the background rate . since undersea neutrino detectors are suffering from background noise , due to the @xmath1 natural radioactivity of the sea - water , it was generally believed that only very dense detector configurations could observe sn signatures @xcite . in this paper we demonstrate that deep undersea neutrino telescopes equipped with optical sensors that comprise many small pmts , are capable of detecting burst of sn neutrinos without imposing any restrictions on the detector architecture . in this study , we assume an underwater neutrino telescope following the mediterranean km3net architecture @xcite . the km3net detector will consist of several hundreds of vertical structures ( detection units - dus ) , which carry photo - sensors and devices for calibration and environmental measurements , arranged vertically on storeys . each storey will support two photo - sensors . the photo - sensor unit is a digital optical module ( dom ) consisting of a 17 inch diameter pressure resistant glass sphere , housing 31 3-inch small photomultiplier ( small - pmt ) tubes , their high - voltage bases and their interfaces to the data acquisition system with nanosecond timing precision @xcite . the segmentation of the photocathode area in such a multi - pmt dom aids in distinguishing single - photon from multi - photon hits , providing thus a better background rejection and trigger efficiency . the front - end electronics will utilize time over threshold ( tot ) measurements that carry amplitude and timing information , which allow for reconstruction of the original signal . the readout system will support the broadcasting of all the digitized data to the shore . in this work , we studied the sn discovery potential of such a @xmath5-telescope configuration , deployed in a mediterranean site of 3500 m depth , consisting of 154 dus , with a mean horizontal separation of 180 m. each du was arranged in 20 storeys ( 6 m long bars ) , with a vertical distance between them of 40 m. each storey carries two photo - sensors , one at each end of the bar . assuming that the expected number of background events in a time interval @xmath6 is @xmath7 and that @xmath8 events have been observed in this time period , the condition that there is less than @xmath9 probability the observed events to be produced solely by the fluctuating background is : @xmath10 in case that @xmath8 is large , the above formula can be replaced by its gaussian approximation , @xmath11 , where @xmath12 is the excess of events with respect to the expected background . obviously , the aim of this study is to define an analysis strategy in order to maximize the following quantity : @xmath13 in the case that the excess @xmath14 is due to neutrinos from a sn explosion , it will depend on the proximity of the sn to earth , i.e. it will be inverse proportional to the squared distance ( @xmath15 ) . consequently , the @xmath16 confidence level ( cl ) requirement in ( [ eq_m ] ) determines the maximum distance of a typical sn in order to be detected , or equivalently , given the distance of the sn , the statistical significance of the observation can be evaluated . the sn explosion model used in this work , is the so called `` garching model '' @xcite which assumes an initial star with mass of 8 - 10 solar masses , which represent 30% of all the studied sne . among the interactions of these sn neutrinos in water , inside the instrumented volume , we consider only the inverse beta decay ( ibd ) of the electron antineutrino ( @xmath0 ) which is the most dominant process . we used the ibd total cross section @xmath17 as well as the differential cross sections @xmath18 ( where @xmath19 is the angular difference between the positron direction and the neutrino direction ) from @xcite , which are very accurate for the energy range of sn neutrinos . the time - integrated electron antineutrino differential flux ( integrated for an observation time interval , @xmath6 , which starts with the bounce ) is expressed as : @xmath20 where @xmath21 and @xmath22 are the luminosity and the energy of the antineutrino respectively , whilst @xmath23 is the distance of the sn from the earth . @xmath24 is the energy spectrum of the electron antineutrino as parametrized in @xcite , where @xmath25 is the mean energy of the neutrino and @xmath26 a parameter representing the amount of spectral pinching . the time evolution of @xmath21 , @xmath25 and @xmath26 are given by the garching model . the integration time @xmath6 is chosen as in @xcite in order to maximize the discovery potential for sn detection ( i.e. eq . ( [ eq_dp ] ) ) and it is found to be @xmath27 , using the time structure of the neutrino burst predicted by the garching model . the positron emitted in sn neutrino induced ibd process is sufficiently energetic ( mean energy 17.19 mev ) to produce cherenkov radiation in the sea water , which is detectable by the photo - sensors of a neutrino telescope . the response of the detector to the ibd positrons and background processes is simulated by utilizing the hours @xcite package ( hellenic open university reconstruction and simulation package ) , which describes in detail all the relevant physical processes ( e.g. secondary interactions , propagation of cherenkov photons ) , the response of the pmts and the digitization electronics . using the simulated events we evaluated the effective volume , @xmath28 , for ibd interactions and the interaction density , @xmath29 , which expresses the number of ibd neutrino interactions per unit volume , occurring inside an observation time interval @xmath6 , caused by the explosion of a sn at distance @xmath23 . the effective volume is estimated as the water volume , around a dom , in which we have simulated ibd positron production , weighted by the fraction of the interactions observed by the dom , under certain observation criteria . the interaction density is found by estimating the integral , @xmath30 by monte carlo integration , where @xmath31 is the free proton density in sea water at the detector deployment depth and the cross sections are taken as described in section 4 . as an example , for a sn explosion at a distance of @xmath32 and for @xmath27 the interaction density at a depth of 3500 m is estimated to be @xmath33 . finally the expected number of observed ibd interactions , from a sn explosion at a distance @xmath23 is evaluated as , @xmath34 , as a function of the pmt multiplicity for three type of events : @xmath1 decays , atmospheric muons and sn neutrinos . the sn explosion is assumed to occur at a distance of 10 kpc and the @xmath5-telescope is assumed to be deployed at a depth of 3500 m containing 6160 doms.,width=377 ] as already mentioned , the main background source to sn neutrinos is the natural radioactivity of @xmath1 in the sea water . the @xmath1 decay rate is about @xmath35 , five orders of magnitude greater than the expected ibd rate from a sn at 10 kpc . however , the released @xmath1 decay energy is relatively small with respect to the energy of the sn neutrino induced ibd positron . the effective volume corresponding to the @xmath1 decays , is thus much smaller than the effective volume for sn neutrino ibd interactions . furthermore the products of @xmath1 decays have less probability to activate synchronously several small pmts of the same dom . we define thus , as the main observation criterion , the pmt multiplicity , which is the number of pmts in the same dom that are simultaneously active within a coincidence time window of 100 ns . fig . [ fig_events ] presents the total number of active doms due to @xmath1 decays , collected during an observation time interval of @xmath27 , as a function of the pmt multiplicity , in comparison with the corresponding number due to sn neutrino induced ibds , assuming that the sn explosion occurred at a distance of 10 kpc . although the background due to @xmath1 decays is significantly reduced at high pmt multiplicities , it is expected that atmospheric muons will still contribute to the background up to very high pmt multiplicities , due to the causal connection between hits produced by muons . in order to quantify this muonic background we have used hours with generation input from the mupage package @xcite with an energy threshold as low as 50 gev . the contribution of this type of background is also shown as a function of the pmt multiplicity in fig . [ fig_events ] . ( a ) , @xmath36 ( b ) and @xmath37 ( c ) . ( d ) : the probability that a dom activated by an atmospheric muon , with pmt multiplicities greater that 5 , has simultaneously ( within a time window of 100 ns ) @xmath38 active neighboring doms , with multiplicity greater than 2.,width=377 ] as shown in fig . [ fig_events ] , a high pmt multiplicity rejects the @xmath1 background . a further suppression of this background can be achieved when we apply selection criteria , based on the amplitude of the small - pmt signals . [ fig_cuts ] ( a , b and c ) shows the probability of a dom , activated by a @xmath1 decay or sn neutrino induced ibd , to exhibit a certain pmt multiplicity , for three different small - pmt pulse - amplitude thresholds , @xmath39 . as an example , in fig . [ fig_cuts](a ) a small pmt is considered active when its pulse amplitude exceeds an amplitude - threshold corresponding to the average pulse amplitude produced by one photoelectron ( pe ) , @xmath40 . the probability of the @xmath1 background to produce high pmt multiplicity events decreases faster than the signal , as the threshold requirement is getting higher . however we found that this requirement has not any significant effect on the background produced by atmospheric muons . on the other hand , it is very improbable that the @xmath1 background and the sn neutrino induced ibds will activate small pmts in different doms ( especially when high pmt multiplicity is required ) . on the contrary , muon tracks produce very often time - correlated hits in different doms , with high pmt multiplicity . this is demonstrated in fig . [ fig_cuts](d ) , where it is shown the probability that a dom activated by an atmospheric muon , with pmt multiplicity greater that 5 , has simultaneously ( within a time window of 100 ns ) @xmath38 active neighbor doms , with pmt multiplicity greater than 2 . rejecting thus all the active doms with pmt multiplicities greater than 5 , when neighboring with at least one , simultaneously - active dom , with pmt multiplicity at least 3 , the atmospheric muon background is reduced by 94% whilst only 10% of the signal is lost . decays and atmospheric muon events . the error bars correspond to the rms of the tot distribution.,width=377 ] we used fully simulated signal ( @xmath41 sn neutrino induced ibds ) and background ( @xmath42 @xmath1 and @xmath43 atmospheric muon ) events in order to estimate ( using eq . ( [ obs ] ) ) the number of signal events from a sn explosion at 10 kpc , as well as the expected number of @xmath1 and atmospheric muon background events , for various sets of event selection criteria . in the simulation we have assumed that the response function of the pmts produced by a single pe is described by the function @xmath44 that the charge resolution is 35% , the time jitter is 1 ns and the rise time of the pulse is 2.5 ns . we have also assumed that the digitization electronics provide the arrival time and the tot of the pulse at the 30% of the mean pulse amplitude corresponding to one pe . in fig . [ fig_pulse ] it is shown the mean value of the pmt - waveform time over threshold duration as a function of the corresponding number of photoelectrons , for three type of events ( ibd interactions , @xmath1 decays and atmospheric muon events ) . as an example , the condition @xmath45 selects pulses corresponding to more than 1 pe . the set of the optimal selection criteria , which are maximizing the sensitivity of the detector as expressed in eq . ( [ eq_dp ] ) are tabulated in table [ table ] . with these event selection criteria we expect 118.7 signal events from a sn explosion at 10 kpc on top of 0.8 @xmath1 and 15.4 atmospheric muon background events . for this expected background , an observation with 5@xmath46 significance requires , according to eq . ( [ eq_m ] ) , at least 39 total observed events ( 22.8 signal events ) . consequently , the maximum distance of a sn explosion that could be observed by this detector is estimated , by scaling proportionally to the squared distance , to be ( @xmath47 ) 22.8 kpc . .selection criteria that maximize the sensitivity of the detector . [ cols="^,^",options="header " , ] [ table ] the km3net project is supported by the eu in fp6 under contract 140 no . 011937 and in fp7 under grant no . 212525 . 00 georg g. raffelt , arxiv : astro - ph/0701677 a. dighe , nucl . phys . proc . suppl . 143 , 449 ( 2005 ) a. dighe and a. y. smirnov , phys . d 62,033007 ( 2000 ) s. hannestad and g. raffelt , phys . 87 , 051301 ( 2001 ) k. scholberg arxiv : astro - ph/0701081 y. giomataris and j.d . vergados arxiv : hep - ex/0503029v2 p. antonioli et al , new j. phys . 6 , 114 ( 2004 ) the km3net collaboration , http://www.km3net.org the icecube collaboration , http://icecube / wisc.edu f. halzen et al , phys . d volume 49 , number 4 ( 1994 ) f. halzen et al , phys . d volume 53 , number 12 ( 1996 ) j. ahrens et al , astroparticle physics 16 ( 2002 ) 345 - 359 l. kopke arxiv:1106.6225v l. demirors , m. ribordy and m. salathe arxiv:1106.1937v2 c. pryor et al , the astrophysical journal , 329:335 - 338 http://www.km3net.org/ppm/km3net_fl_rep_ppm-description.pdf , 2010 l. hudepohl et al , phys . rev . lett . 104,251101 ( 2010 ) a. strumia , f. vissani physics letters b 564 ( 2003 ) 42 - 54 m.t . keil , et al , the astrophysical journal , 590:971 - 991 , 2003 a. tsirigotis , et al . instr and meth . a 626 , ( 2011 ) , s185 g.carminati et al . , commun . * 179 * ( 2008 ) 915	we study the potential of a very large volume underwater mediterranean neutrino telescope to observe neutrinos from supernova ( sn ) explosions within our galaxy . the intense neutrino burst emitted in a sn explosion results in a large number of mev neutrinos inside the instrumented volume of the neutrino telescope that can be detected ( mainly ) via the reaction @xmath0 . in this study we simulated the response of the underwater neutrino telescope to the electron antineutrino flux predicted by the garching model for sn explosions . we assumed that the neutrino telescope comprises 6160 direction sensitive optical modules , each containing 31 small photomultiplier tubes . multiple coincidences between the photomultiplier tubes of the same optical module are utilized to suppress the noise produced by @xmath1 radioactive decays and to establish a statistical significant signature of the sn explosion . supernova , neutrino telescope 95.55.vj
the standard description of electron - phonon interaction ( epi ) in metals is based on the so - called born - oppenheimer @xcite ( or adiabatic ) theorem . its main statement says that electrons are not sensitive to the motion of ions and are influenced only by their static electric field . one can to say that fast electrons , due to electroneutrality , follow slow ions . narrow bands and a strong electron - phonon interaction make their velocities comparable in order of magnitude . an appropriate mathematical description is the standard feynman - dyson perturbation theory , where phonons are considered to be uneffected by the epi . however , there are some processes which in higher orders violate this approximation . mathematically this can be taken into account by virtue of the so - called vertex corrections . these effects were considered by migdal @xcite who showed that their contributions are small , of the order of a small parameter @xmath0 , where @xmath1 is a characteric energy , defined by the bandwidth @xmath2 or the fermi energy @xmath3 . it was noted by migdal himself @xcite that at @xmath4 the vertex corrections are not small ( so they are important for optical conductivity and raman scattering ) . this case was considered in detail by engelsberg and schrieffer @xcite . migdal s theorem is also violated for a one - dimensional fermi surface @xcite . traditionally nonadiabatic corrections were considered to be small because the migdal parameter @xmath5 was small @xcite . but recently some materials ( for example , fullerenes and high-@xmath6 superconductors ) where @xmath7 were discovered . so migdal s theorem about the smallness of the nonadiabatic corrections @xcite can be violated . this gives grounds for not only discussing these corrections , but even examining an antiadiabatic limit @xmath8 @xcite . but even with small @xmath5 there are contradictory conclusions about the importance of such corrections in the normal and superconducting state ( the authors of refs.@xcite consider them to be negligible , while the authors of @xcite think the opposite is true ) . the most intriguing question is whether the violation of adiabaticity leads to an increase of the interelectron interaction ( and , as a result , to an increase of @xmath6 ) . there are contradictory results about the effect of the vertex corrections on @xmath6 . for example , takada @xcite reported some increase of @xmath6 for small parameters @xmath9 . even more significant increase was claimed in @xcite . at the same time , in refs . @xcite@xcite a decrease of @xmath6 was obtained due to the vertex corrections . one can distinguish between the two sorts of the nonadiabatic corrections to the electron self - energy : 1 ) the vertex ones , resulted from the higher order corrections of perturbation theory , as vertex function @xmath10 even at an infinite electron ( or hole ) band @xmath11 @xcite , and 2 ) the corrections due to the finite band width , @xmath12 @xcite . nonadiabatic corrections in the superconducting state enter the expression for @xmath6 in two ways : indirectly through renormalization of @xmath13 and directly in the equation for the order parameter @xmath14 . at the same time , analytical calculation of anomalous ( crossing ) diagrams is rather difficult . therefore before proceeding to calculations in the superconducting state we want to clarify the role of such corrections in the normal state . the main goal of this paper is to compare different approaches to calculation of nonadiabatic corrections to the electron quasiparticle self - energy in the normal state ( or more precisely to the mass renormalization factor @xmath13 ) . there are two methods to take nonadiabatic corrections into account . one of them , which can be called migdal s one , is based on the solution to the bethe - salpeter equation for the vertex function @xcite in the ladder approximation . in the lowest approximation , the first correction to the unity vertex is determined by the diagram in fig.1 . then with the obtained vertex function the self - energy @xmath15 is calculated @xcite . this traditional method does not allow to take into account higher - order diagrams analytically because of complexity of the integration over momenta . in 1989 the authors of the ref.@xcite considered a non - ladder approximation , using the ward identity . later y. takada used the same idea in his method @xcite , which he called the gauge - invariant self - consistent ( gisc ) method . in this method , based on the ward identity @xcite @xmath16 the vector term is neglected and the scalar vertex function is chosen to be a functional of the self - energy , which is supposed to be independent from momentum . recently there appeared another method @xcite , aiming at improving the gisc approach as it takes into account the momentum dependence of the vertex function . so , when comparing these methods with the migdal s one , we also discuss the validity of these assumptions . in fact , for the self - energy we calculate all the diagrams up to , and including the second order . the parameter @xmath17 is taken to be small . for simplicity , we consider the model of the fermi liquid with the usual electron - phonon hamiltonian @xcite and einstein phonon @xcite spectrum with the so - called eliashberg spectral function @xmath18 , where @xmath19 is the phonon frequency independent from momentum , and @xmath20 is the electron - phonon interaction constant . it is assumed that the density of states is constant @xmath21 for @xmath12 where @xmath22 is the bandwidth . the chemical potential is zero , which corresponds to a half - filled band . it is assumed that @xmath23 . we are interested in the quasiparticle self - energy @xmath15 on the imaginary axis which is determined by the diagram in fig.1 . the phonon green s function is @xmath24 , where @xmath25 , and the electron green s function is @xmath26 , where @xmath27 , @xmath28 is the bare electron spectrum . the vertex function itself ( two - point one ) is only of academic interest . we are , however , interested in the quasiparticle self - energy @xmath15 which affects many physical observables . assuming that @xmath29 weakly depends on @xmath30 , we take @xmath31 , and then @xmath32 the methods differ in the choice of the vertex function @xmath33 , and this will be discussed later . one should note that @xmath34 enters the integral for @xmath15 , and all @xmath35 s contribute to it . for simplicity we will obtain results for @xmath36 and the variables will be changed as follows : @xmath37 , @xmath38 , remaining on the imaginary axis . in the method which can be called migdal s , the first correction to the unity vertex function @xmath39 is considered . this is shown in fig1 . it has been estimated in many papers @xcite , but now it is becoming especially significant @xcitebecause the cases were found where as the parameter @xmath40 may be not small . the diagram in fig.1 corresponds to the expression @xmath41 assuming that @xmath42 and expanding @xmath43 we get for @xmath36 @xmath44 where @xmath45 , @xmath46 , @xmath47 , @xmath48 . this is a nonanalytical function at @xmath49 , @xmath50 , i.e. the result depends on the order of taking the limits . let us define the dynamical and static vertex functions @xmath51 from ( [ 1315 ] ) one can get @xmath52,\ ] ] @xmath53 it is evident that the nonanalyticity mentioned above gives @xmath54 at @xmath55 @xmath56 which satisfies the ward identity ( [ eq1.5 ] ) . at @xmath57 one has @xmath58 which complies with refs . @xcite . to calculate the self - energy we expand the last term of ( [ 1315 ] ) in @xmath59 , leaving the first two terms as they are . then @xmath60}. \label{eq4}\end{aligned}\ ] ] in fig . 2a we show @xmath61 ( numerical ( [ 1315 ] ) and approximate analytical ( [ eq4 ] ) results ) for the case @xmath62 as well as the result of the numerical integration of eq.(4 ) . one can see that there is a good agreement between these plots , so the expansion in small @xmath59 is well justified . the self - energy ( [ eq1 ] ) consists then of three terms : @xmath63 the first order term is @xcite @xmath64 the vertex correction @xmath65 is obtained when one substitutes @xmath66 and @xmath67 ( [ eq4 ] ) into ( [ eq1 ] ) @xmath68 @xmath69 @xmath45 , @xmath48 , @xmath70 . @xmath71 @xmath72 where @xmath73\}.\end{aligned}\ ] ] for @xmath74 , @xmath75 it gives @xmath76 . \label{591}\end{aligned}\ ] ] the plot of @xmath77 is shown in fig.2b . the numerical calculations ( the exact integration with eq.([1315 ] ) ) give similar results for the small @xmath78 s . the existing discrepancy for larger frequencies is due to our approximation for @xmath33 . another term is the so - called rainbow diagram . as the inner part of the rainbow diagram corresponds to @xmath79 , the second order rainbow diagram equals @xmath80 ^ 2 \sigma^{(1)}(i\omega_n - i\omega_{\nu } ) , \label{eqd05}\end{aligned}\ ] ] which results in @xmath81 @xmath82 at @xmath74 , @xmath75 , it equals @xmath83 for @xmath74 , @xmath75 , @xmath84 one can get @xmath85 one can see that the last expression has three terms with different denominators . when , for example , @xmath86 , the main contribution comes from the term @xmath87 ( even for an infinite bandwidth @xmath2 ) . in fig.2b there are shown for comparison analytical and numerical plots for the self energy correction @xmath88 . it is seen that the approximation used to calculate the correction is well justified . if one assumes @xmath89 , @xmath90 one can see that these corrections lower the renormalization function . had this tendency persist in all higher orders , this would have resulted in an instability ( see below ) . following the paper@xcite , the authors of ref.@xcite neglected the momentum dependence of the vertex function , considering @xmath91 , and interpolated it between zero - frequency and infinite - frequency limits : @xmath92 where @xmath93 , and @xmath94 this results in an effective interaction @xmath95 combined with the contribution from the rainbow diagram , which is the same as in the migdal method ( [ sigma_r ] ) , this gives @xmath96 the non - ladder approximation which was also discussed in @xcite was developed in detail by takada , and this is considered below . there has been a series of papers , e.g. ref . @xcite , which considered the effect of the vertex corrections on @xmath6 not only for an isotropic epi , but also for spin ( antiferromagnetic ) fluctuations with a pronounced scattering at the wave vector @xmath97 . in the paper @xcite the authors used the following equations for the self - energy @xmath98 $ ] : @xmath99[i\omega _ { n^{\prime \prime } } ^{2}-(\epsilon _ { { \bf k% } ^{\prime \prime } } /z_{n^{\prime \prime } } ) ^{2}][i\omega _ { n^{\prime \prime \prime } } ^{2}-(\epsilon _ { { \bf k}^{\prime \prime \prime } } /z_{n^{\prime \prime \prime } } ) ^{2 } ] } , \label{416}\end{aligned}\ ] ] where @xmath100 , and @xmath101 is a geometrical factor being a complicated function of momenta which is considered in detail in ref.@xcite . for an isotropic epi and a three - dimensional spherical fermi - surface it is approximated by its value on the fermi surface @xmath102 . it gives @xmath103 @xmath104 , @xmath105 , @xmath106 , and correction to the vertex function equals @xmath107 one can find that @xmath108 and so does not depend from @xmath109 . however , for @xmath110 which give the main contribution to the self - energy it presents the correct order of the correction to the vertex function ( compare with eq.([eq4 ] ) ) and the self - energy . from eq . ( 2 ) one has for @xmath74 , @xmath111 @xmath112 this is similar to eq . ( [ 592 ] ) , but has a different numerical coefficient , due to neglecting the @xmath35-dependence in the vertex function ( [ 418 ] ) . in the paper @xcite the authors started with the same equation as in migdal method , but made a series of approximate assumptions ( for example , expanded in small @xmath109 s to take integrals ) and gave the following estimate for the vertex function : @xmath113\arctan \frac{q}{m } \nonumber \\ & & -[q - m\arctan \frac{q}{m}]\frac{(\frac{1}{m_{0}}+1)[(\frac{1}{m_{0}}% + 1)^{2}+2m^{2}]}{[(\frac{1}{m_{0}}+1)^{2}+m^{2}]^{2}}].\end{aligned}\ ] ] here they set one of the external electronic frequencies to zero . in this case this result is similar to that in the migdal s approximation ( [ eq4 ] ) . however , to calculate @xmath15 one needs dependences on both external frequencies . there are some indications ( see , e.g. @xcite ) that for a small hole doping strong coulomb correlations renormalize the epi , giving rise to the strong forward ( small-@xmath109 ) scattering peak , while the backward scattering is strongly suppressed . with this idea in mind , in ref . @xcite the electron - phonon coupling constant was assumed to have a cut - off in a momentum space @xmath114 . an approximation for @xmath115 was used @xcite and then @xmath116 was set to @xmath117 giving the following result @xcite for @xmath13 : @xmath118 for @xmath119 it becomes @xmath120 which gives the correct order of the correction to the self - energy , but underestimates it ( compare with eq.([592 ] ) ) . probably this is due to the expansion in small @xmath109 s . according to takada s gauge - invariant self - consistent ( gisc ) method , the vertex function can be chosen as a functional of the self - energy @xcite @xmath121 $ ] . this choice is based on the ward identity @xcite ( [ eq1.5 ] ) valid for all @xmath122 s and @xmath123 s . in ref . @xcite takada proposes neglecting the vector term in the ward identity and choosing @xcite @xmath124=\gamma _ { t}(i\omega _ { n},i\omega _ { n}-i\omega _ { \nu } ) = 1+(-\frac{\sigma _ { t}(i\omega _ { n})}{2i\omega _ { n}}-\frac{\sigma _ { t}(i\omega _ { n}-i\omega _ { \nu } ) } { 2(i\omega _ { n}-i\omega _ { \nu } ) } ) , \ ] ] which corresponds to the estimates of refs . @xcite in this limit . if this method worked it would allow one to calculate also superconducting diagrams without complicated integrations over momenta . the set of equations ( 1,2 ) is suggested to be solved by iterations . at the first step @xmath125 , @xmath126 give @xmath127 from eq.(2 ) . at the second step @xmath128 $ ] ; and the correction @xmath129 for @xmath111 gives @xcite @xmath130 for small frequencies it results in @xmath131 which has the incorrect sign and order of magnitude of the vertex correction ( compare with ( [ 592 ] ) ) . during this calculation , the correction to the vertex @xmath129 is of the order of @xmath20 for any @xmath109 , which contradicts the migdal s theorem @xcite and can be valid only for @xmath132 ( see e.g. ref . so in takada s method the correction @xmath133 does not have the small parameter @xmath134 . this can be explained by the fact that the region @xmath55 where @xmath135 gives in fact a small contribution to @xmath15 . if , on the other hand , @xmath57 , the vertex @xmath33 is of the order of @xmath136 , which gives @xmath137 ( see ( [ 591 ] ) ) . to explain this result , one can find the vector vertex function @xmath138 in the lowest approximation and show that it really can not be neglected . the diagram in corresponds to @xmath139 if @xmath140 is substituted by @xmath141 and @xmath142 - by @xmath143 . using electron - hole symmetry , in our model @xmath144 @xmath145 neglecting the @xmath59-dependence in the last term of ( [ 1323 ] ) one can get @xmath146 we can find regions where scalar or vector terms in the ward identity dominate . if one uses at @xmath75 the expression ( [ eq4 ] ) for @xmath147 , one can find that the terms @xmath148 and @xmath149 from ( [ eq1.5 ] ) are of the same order if @xmath150 . if , however , @xmath151 , the vector term dominates and can not be neglected . taking into account a criticism of the gisc method @xcite , f. cosenza , l. de cesare , and m. fusco girard @xcite tried to improve the gisc method and did not neglect the vector term in the ward identity . they employed the simplest choice of the solution to the ward identity @xcite @xmath152}{% ( i\omega _ { n^{\prime } } -i\omega _ { n})^{2}-\|\alpha ( \vec{k},\vec{k^{\prime } } % ) \|^{2}\|\vec{k}-\vec{k^{\prime } } \|^{2 } } , \label{cdg}\ ] ] where @xmath153 . with @xmath154 taken from eq.([s1 ] ) one gets for @xmath155 @xmath156 which is different from the migdal s estimation @xcite @xmath157 ( see ( [ eq4 ] ) ) . substituting approximation ( [ cdg ] ) into the equation for the self - energy ( [ eq1 ] ) one can get @xmath158 which is different from ( [ 591 ] ) and gives @xmath159 this can be explained by the fact that the ward identity is a one equation for two functions ( the scalar and vector vertex functions ) and thus allows for multiple solutions @xcite . so there is no particular preference in using ( [ cdg ] ) instead of any other solution . [ seresultspage ] so in the paper we considered consistently the contribution of nonadiabatic effects to the electron self - energy in the normal state at @xmath36 in second orders of the epi constant @xmath20 . several methods of taking into account nonadiabatic corrections were compared which can be summarized in the table ( @xmath160 ) . the results following from the approximations of v. n. kostur and b. mitrovi @xcite , e. cappelluti , c. grimaldi , l. pietronero , and s. strssler @xcite give correct estimations of the order of magnitude and sign of the nonadiabatic corrections to the self - energy . the analytical results were compared with numerical calculations . it is shown that the gisc method @xcite and its generalization @xcite give for the self - energy overestimated and underestimated results respectively . this is connected with the fact that the leading contribution to @xmath15 comes from the region @xmath161 where @xmath162 . this means that in the ward identity one may not neglect the vector term as proposed by takada . in the framework of the standard ( migdal s ) approach , in the lowest order , there is a calculable parameter @xmath163 ( the ratio of the frequency of an incoming phonon to its momentum ) for which the contributions of the vector and scalar terms become of the same order . they are equal when @xmath164 . so for the most metals , where the main contribution to the self energy comes from the region @xmath165 , the gisc method may not be applied . one should also be careful when choosing a solution of the ward identity , so as not to underestimate the scalar term , like in ref . it seems that gisc method can work in systems with long range interaction , for example , in doped semiconductors . the difference in the renormalization factors @xmath13 in the table are due with the different approximations used in the calculations of the momentum and frequency dependence of vertex function . we can illustrate this in fig . 3 where we represent the results for @xmath166 in the takada s ( @xmath109-independent ) approximation@xcite , the migdal approach for different @xmath109 ( eqs . ( 4,27 ) of the present paper ) , and the @xmath109-independent kostur , mitrovi vertex function . we see that if the former overestimates the vertex corrections , the latter underestimates them . for a stable system the condition @xmath167 for @xmath168 must be satisfied @xcite ( it is equivalent to @xmath169 ) . however , as one can see from the table , the nonadiabatic corrections reduce the renormalization function @xmath13 , favoring the tendency towards instability . the critical @xmath170 at which it takes place ( see eqs . ( 10 - 16 ) ) is given by @xmath171 for @xmath172 , where @xmath173 and @xmath174 were defined above . the dependence of @xmath20 on the parameter @xmath134 is shown in fig.4 . thus , for a stable system there are constraints on @xmath175 for a fixed migdal parameter@xcite . this can explain , for example , the existence of pbbi with @xmath176 . at the same time , in some papers , e.g. @xcite , it is stated that @xmath23 independently from the migdal parameter . as one sees from , it may be the case only for @xmath177 . this possible violation of the analytical properties of the one - particle green s function could result in corresponding nonanaliticity of the two - particle green s function , and thus in a charge response function . this can lead to a charge instability . it follows from our analysis that different approximations to the vertex function give in the normal state quantitatively ( and sometimes even qualitatively ! ) different estimates for the self - energy . this makes the conclusions of these theories doubtful , as the same approaches were used there to calculate @xmath6 in the superconducting state . this means that only direct migdal - type calculations can give an answer to the question : do nonadiabatic effects enhance critical temperature of the superconducting transition ? we thank t. dahm , o. gunnarsson , m.l . kuli , v.v . losyakov , i.i . mazin , e.g. maksimov , and n. schopohl for helpful discussions . y. takada , j. phys . solids * 54 * , 1779 ( 1993 ) , techn . rep . of issp , 2732 * , september 1993 , y. takada , j. superconductivity * 8 * , 628 ( 1995 ) ; y. takada , t. higuchi , phys . rev . * b * * 52 * , 12720 ( 1995 ) fig . 2 . a ) plot of @xmath180 ( eq.(4 ) , dashed line ) and @xmath181 ( eq.(7 ) , solid line ) for @xmath182 , @xmath183 . b ) corresponding frequency dependence of @xmath77 ( dashed line ) and @xmath184 ( solid line ) . @xmath185 in the takada s approximation@xcite ( long dashed line ) ; the migdal approach for @xmath186 ( solid lines , from top to bottom ) ; and the kostur , mitrovi @xcite vertex function ( short dashed line ) .	we investigate effects of a nonadiabatic electron - phonon(boson ) interaction on the quasiparticle self - energy in the lowest order in the coupling constant . existing approaches either overestimate , or underestimate these effects because of different approximations for momentum and frequency dependences of the vertex corrections . the connection between the nonadiabaticity and a possible instability of the interacting fermi system is discussed as well .
the casimir effect is a consequence of the field - theoretical description of the quantum vacuum and results directly from the quantization of the electromagnetic field . traditionally derived for two infinite uncharged metallic plates placed only a few nanometers apart @xcite , this quantum - mechanical effect has also been investigated in the context of atoms interacting with surfaces @xcite . casimir - polder ( cp ) forces have been a subject of research of its own , both in view of nano - technological applications @xcite and motivated by the possibility of probing fundamental forces at the submicron scale @xcite . indeed , the effects of a dispersive potential due to a macroscopic surface on an atom , both at zero and finite temperature , are already quite well established @xcite . moreover , experiments with bose - einstein condensates ( bec ) near a surface have attracted a special attention since the early days of microtraps @xcite : if , in one hand , the understanding of the influence of vacuum forces is crucial to the operations in atomic microtraps , on the other , the measurements based on quantum optics of ultracold gases are very accurate , making them a very desirable candidate to probe vacuum forces @xcite . in fact , the influence of the solid - state substrate on the atomic dynamics of a cold gases has been the target of several experiments . as it has been shown , an atomic cloud near a rough surface ( typical distances of 1 @xmath0 m ) may undergo a matter - wave anderson localization in a random potential @xcite . also , by rotating a corrugated plate separated of a few microns from a bec , the nucleation of quantized vortices is theoretically predicted @xcite . more recently , bender et al . have used bec to harvest information about the shape of the potential landscape of a solid grating , in an excellent agreement with the theoretical predictions @xcite . an important and recent activity with ultracold atoms includes mechanical coupling via vacuum forces . experiments have put in evidence the resonant coupling of mechanical bec modes to a micromechanical oscillator @xcite , as well as the backaction of the atomic motion onto the membrane @xcite . there are also theoretical propositions of how creating a force in a graphene sheet due to highly excited ( rydberg ) atomic states @xcite . actually , the interest in the family of fullerenes has growing since the recent propositions of sympathetic cooling of carbon nanotubes , via cp interaction , by a laser cooled atomic gas @xcite . however , to the best of our knowledge , a completed treatment of the coupling of the flexural modes of a graphene membrane and a bec via vacuum forces has not yet been discussed . in this paper , we will provide a microscopic description of the problem . from a monolayer graphene sheet . the a ripple at the position @xmath1 provoques a small deformation in the casimir - polder potential at a position @xmath2 in the bec and vice - versa . ] benefiting from its remarkable mechanical and transport properties @xcite , monolayer graphene is a natural candidate to perform atom - surface interfaces . due to thermal fluctuations , the membrane may undergo mechanical out - of - plane vibrations ( flexural phonons ) , which can be well described within the kirchoff - plate theory of elasticity @xcite . a recent experimental study makes use of a cavity optomechanical protocol to cool down the zero - point flexural mode of a suspended graphene sheet @xcite . by previously cooling the graphene sheet ( with the help of in a dilution refrigerator , for example ) , the flexural modes can be quantized . thus , in this paper , we investigate the dynamics of a two - dimensional bec interacting with a monolayer graphene sheet via the casimir - polder potential ( see fig . ( [ fig1 ] ) for a schematic illustration ) . we proceed to a full quantization of the system in order to harvest the coherent coupling between the two phonon modes , leading to a new type of polariton excitation : a purely acoustic phonon - polariton . we show that for sufficiently large separation distances @xmath3 , heating of the condensate via vacuum fluctuations is negligible , showing that the polaritons may exist in the strong - coupling regime . in sec . ii , we present the governing equations in terms of the mean - field equations . the details of the cp potential for a @xmath4rb condensate are provided in sec . iii . in sec . iv , the dispersion relation of the phonon - polariton modes and the bec heating rate are derived . finally , in sec . v some concluding remarks are stated . in a general way , the potential between a single neutral atom and a surface is given by @xmath5vacuum where @xmath3 is the separation distance between the atom and the surface and @xmath6 is the strength of the interaction , which depends on the both the atomic polarizability and on the electromagnetic properties of the surface . let @xmath7 be the condensate order parameter , normalized to the number of atoms such as @xmath8 , where @xmath9 is the plane coordinate . the dynamics of the two - dimensional bec can then be described in terms of the gross - pitaevskii equation @xmath10 where @xmath11 , with @xmath12 , is the effective 2d coupling constant , @xmath13 is the atomic scattering length and @xmath14 is the transverse harmonic length . the atoms in the bec feel a vacuum - field potential of the form @xmath15^{\nu/2 } } , \label{potential1}\ ] ] where @xmath16 is a small deformation on the graphene surface ( ripple ) located at the position @xmath2 and @xmath17 is the quantization surface . on the other hand , the mechanical properties of the graphene sheet can be easily derived from the kirchoff - love plate theory @xcite . the lagrangian density of the sheet can be written as @xmath18 , with @xmath19 standing for the graphene mass density , and the potential energy per unit surface can be expressed as @xmath20 where @xmath21 is the bending stiffness , @xmath22 ( @xmath23 ) is the tension along the @xmath24 ( @xmath25 ) direction to the clamping with the substrate and @xmath26 is the local curvature . the term @xmath27 is the restoring force in the harmonic approximation , which acts as a charge on the surface of the sheet through the cp potential , @xmath28 from the euler - lagrange equations , we can obtain the kirchoff - love equation for the flexural modes @xmath29 where we assume , for simplicity , the self - suspended case @xmath30 . eqs . ( [ gp1 ] ) and ( [ plate1 ] ) form a self - consistent set of equations for the condensate field @xmath31 and for the deformation field @xmath16 , being the governing equations to be quantized in sec . rb atom near the surface of a graphene sheet . for short distances , the potential scales as @xmath32 ( oblique line ) . the dependence on the distance as @xmath33 ( horizontal line ) is obtained for separations larger than @xmath34 m . ] in the absence of coupling ( or , equivalently , in the limit of very large distances @xmath35 ) , eqs . ( [ gp1 ] ) and ( [ plate1 ] ) can be easily solved by taking the bogoliubov prescription @xmath36 $ ] and the expansion @xmath37 , which respectively yield to two stable modes , @xmath38 here , @xmath39 and @xmath40 are the condensate sound speed and healing length , respectively , @xmath41 is the specific stiffness and @xmath42 . in agreement with an experimentally feasible situation , we choose a @xmath4rb condensate with an areal density of @xmath43 @xmath44 , trapped along the @xmath45-direction by a harmonic trap frequency @xmath46 hz , @xmath47 nm , which yields @xmath48 @xmath0 m and @xmath49 mm / s . on the other hand , for the graphene sheet we have @xmath50 mg / m@xmath51 and @xmath52 ev @xcite , leading to @xmath53m@xmath51/s . this simply means that the bec phonons and the ripples in the graphene sheet propagate at very different frequencies . in fact , for wavevectors of the order @xmath54m@xmath55 ( typical of the bec phonon - like excitations ) , @xmath56 . such a picture is very similar to what happens in semiconductor microcavities , where strong coupling - allowing for the formation of exciton - polaritons - occurs between two modes of very different energy scales : photons and excitons . because of their very small effective mass , photons exhibit a parabolic dispersion relation , while the excitonic dispersion is almost flat for the relevant wavevectors @xcite . the coupling is then provided via the dipolar interaction between the photons and the excitons . in the next section , we show that the same thing happens here , where the bogoliubov excitations in the condensate , possessing a very flat dispersion , are the analogue of the semiconductor excitons , while the mechanical vibrations of the graphene sheet work as massive photons . in what follows , we quantize eqs . ( [ gp1 ] ) and ( [ plate1 ] ) to show that strong coupling between the condensate phonons ( bogolons ) and the flexural modes in the graphene ( flexural phonons ) is possible , leading to the formation of a sort of phonon - polaritons . the theoretical approaches to determine the casimir - polder energy shift are usually based on second - order perturbation theory @xcite . here , we will discuss how to evaluate the casimir - polder potential between a single graphene sheet and a @xmath4rb atom within the formalism of macroscopic qed @xcite . to do so , we shall neglecting the possible effects that may arise from the finite size of graphene and assume it to be infinitely extended . for planar structures , the casimir - polder potential of an atom in a ground state @xmath57 at a distance @xmath3 away from the macroscopic body can be written as @xcite @xmath58 , \end{array } \label{eq : ucp_1}\ ] ] where integration is done along the imaginary frequency axis @xmath59 and @xmath60 the wave vector in the plane of the interfaces . we have defined @xmath61 , which is the @xmath45 component of the wave number in a medium with permittivity @xmath62 ( the index 0 refers to the medium in which the atom is placed ) . here , @xmath63 is the atomic polarizability defined for an isotropic atom as @xmath64 the latter is valid in the zero temperature limit and for atoms in the ground state , which will be the case of the present manuscript . all the relevant electromagnetic features of the graphene sheet are cast in the reflection coefficients @xmath65 and @xmath66 in eq . [ eq : ucp_1 ] . in ref . @xcite , the reflection coefficients have been calculate assuming that the dynamics of the quasiparticles in graphene can be described within the @xmath67dimensional dirac model . imposing appropriated boundary conditions to the em field , it is possible to explicitly determine the reflection coefficients . taking the contribution of the electrons near the dirac cone , one obtains , for a self - suspend graphene sheet in vacuum , @xmath68 where we have defined @xmath69 and @xmath70 , with @xmath71 being the fermi velocity and @xmath72 the speed of light ; @xmath73 is the fine structure constant . more elaborated models could be performed , however , for the present conditions , eq . ( [ eq : reflcoefg ] ) provides a very good approximation @xcite . there are two regimes according to the atom - surface distances : the near - field , nonretarded limit , and the far - field ( retarded ) limit . in the nonretarded limit , @xmath74 , and for the usual fresnel reflection coefficients , successive approximations yield @xmath75 , and we obtain a fitting to eq . ( [ eq : ucp_1 ] ) with @xmath76 hz@xmath0m@xmath77 . the cp potential for a ground - state atom is attractive for a non - magnetic medium @xcite . we notice that the sign of the cp potential of a ground - state atom with a conducting plate can be easily understood with an image - dipole model . however , if a resonant coupling between the atomic dipole and a surface excitation occurs , it is possible to obtain a repulsive force @xcite . the retarded limit corresponds to the situation where the atom - surface distance @xmath3 is large when compared to the effective transition wavelength . in this situation , one finds the approximation @xmath78 also to be valid , with @xmath79 hz@xmath0m@xmath80 . the dependence of the cp potential on the separation distance @xmath3 as resulting from the numerical integration of eq . ( [ eq : reflcoefg ] ) is depicted in fig . ( [ fig2 ] ) . in the following , we will operate in the distance range @xmath81 m , deep in the limit @xmath82 . such a choice is a consequence of three major constraints : i ) transverse trap size , as the two - dimensional bec approximation is only valid if the transverse size of the bec is much smaller than the distance @xmath3 , ii ) the de - confinement effect associated to the cp potential , and iii ) the heating of the condensate for small distances . in the following , we assume condition i ) to be satisfied ( which is true for @xmath83 hz ) and discuss the limitations imposed by ii ) and iii ) in sec . v. in order to proceed to a microscopic description of the coupling between the bogolons and the flexural phonons , we quantize the theory . the total hamiltonian of the system can be defined as @xmath84 , where @xmath85\hat \psi ( \mathbf{x } ) , \label{hb}\ ] ] is the condensate hamiltonian , which can be solved in the fourier basis with @xmath86 and @xmath87 . here , @xmath88 represent the bosonic annihilation operator satisfying the canonical commutation relation @xmath89=\delta_{\mathbf{k}\mathbf{k}'}$ ] . using the bogoliubov approximation @xmath90 , with @xmath91 standing for the @xmath92 fluctuations , the condensate hamiltonian can be simply given by ( expressing the summation in terms of the absolute value @xmath93 only ) @xmath94 } . \end{array } \label{hb2}\ ] ] here , @xmath95 , @xmath96 is the chemical potential and @xmath97 is the condensate zero - point energy . similarly , the hamiltonian for the flexural modes easily follows from the canonical quantization of eq . ( [ plate1 ] ) @xmath98.\ ] ] expressing the phonon operator in the form @xmath99 with two polarizations @xmath100 and satisfying the normalisation condition @xmath101 , with @xmath102 being the membrane mass , such that the phonon operators obey the bosonic commutation relation @xmath103=\delta_{\mathbf{k}\mathbf{k}'}\delta_{\sigma , \sigma'}$ ] , the graphene hamiltonian reads @xmath104 the quantization of the flexural modes is relevant for temperature scales given of the cryogenically cooled atom chip , on which the graphene sheet is supposed to be suspended . typical experimental conditions involving a dilution refrigerator allow to cool down a carbon nanotube down to 4 k @xcite . additionally , according to the recent experiments of ref . @xcite , optomechanical cooling can be used to bring the graphene down to 50 mk , which corresponds to roughly 40 flexural phonon quanta . the interaction hamiltonian is given by a convolution @xmath105 assuming small deformations , we can taylor expand the potential to first order in @xmath106 , @xmath107 , which yields @xmath108 with @xmath109 being the fourier transform of the potential and @xmath110 the density operator . the first term can be separated into the condensate and the fluctuation contributions , leading to a change in free - particle energy @xmath111 the second term of eq . ( [ separation ] ) is of the order of the condensate depletion @xmath112 , and therefore we should neglect it . by making the substitutions @xmath113 and @xmath114 in eq . ( [ hb2 ] ) , and proceeding to a bogoliubov - valatin transformation to the excitation operators in the usual way , @xmath115 , we obtain the modified condensate hamiltonian as @xmath116 , where @xmath117 is the renormalized bogolubov spectrum and the transformation coefficients read @xmath118 the second term of eq . ( [ hint ] ) can be evaluated by noticing that in the bogoliubov approximation @xmath119 , which then yields @xmath120 where @xmath121 is the @xmath122dependent coupling @xmath123 the latest two terms in eq . ( [ hint1 ] ) do not conserve the total number of excitations , and therefore will be neglected in the present discussion . this approximation remains valid here as long as the interaction is small compared to the single - particle energy , @xmath124 , which we will confirm a posteriori . finally , the total hamiltonian of the system can be given as @xmath125 where @xmath126 is the interaction - induced energy shift . the later hamiltonian can be diagonalized using the so - called hopfield transformations @xmath127 where @xmath128 and @xmath129 and @xmath130 respectively denote the destruction operators for the lower ( l ) and upper ( u ) polaritons . in order to be bosonic , the operator in eq . ( [ pol ] ) must satisfy the normalization condition @xmath131 , which allow us to write the hamiltonian in the decoupled form @xmath132 where the eigenenergies are simply given by @xmath133,\ ] ] with @xmath134 . the two - mode polariton dispersion is depicted in fig . ( [ fig_dispersion ] ) . the graphene ( condensate ) fraction of the lp ( up ) mode almost 1 near the bottom of the dispersion @xmath135 ; on the contrary , for @xmath136 the lp ( up ) mode is essentially bogoliubov- ( graphene- ) like . moreover , the branches repeal each other at the wave vector @xmath137 . the effective rabi frequency @xmath138 measures the strength of the coupling and is equal to @xmath139 hz for @xmath140 m , being much smaller that the free - energy scale of the system . rb atoms and a graphene sheet interacting via the casimir - polder potential @xmath141 . panels a ) : upper ( red solid line ) and lower ( blue solid line ) polariton . panel b ) : amplification of the rectangle depicted in panel a ) . the bare dispersions for the flexural modes ( dot - dashed red line ) and for the condensate ( dot - dashed blue line ) are also shown . both panels are obtained for @xmath140 m . ] in order to demonstrate the conditions for which strong coupling effectively occurs , we must quantify the heating induced in the condensate . as illustrated in fig . ( [ fig_coupling ] a ) ) , the casimir - polder interaction modifies the trapping potential . if @xmath3 is small enough such that the first excited state of the trap lies above the cut - off energy @xmath142 , where @xmath143 and @xmath144 is the solution of the equation @xmath145 , the heating may cause the particles to scape the trap @xcite . in order to avoid such a situation , we should require the restriction @xmath146 , where the critical distance is @xmath147 . for @xmath148 hz , we obtain @xmath149 m . in this situation , the heating rate rules out trap losses and lis related to the creation of in - plane excitations only ( condensate depletion ) , which we calculate with the help of fermi s golden rule @xmath150 where the delta function accounts for the resonant terms only . here , the initial state @xmath151 describes the condensate in the ground state and the graphene phonon with energy @xmath152 and polarization @xmath153 ; the final state @xmath154 contains an extra excited state above with flat dispersion @xmath155 . by proceeding that way , we capture only the inelastic processes , which after some simple algebra yields the following heating rate @xmath156 where @xmath157 is the bec static structure factor . the strong coupling regime is achieved for @xmath158 , which corresponds to separation distances of the order of 1.5 @xmath0 m , deep in the regime where the cp potential considered here is valid ( see fig . ( [ fig_coupling]b ) ) . the weak coupling situation is also possible for shorter distances , where the heating of the bec becomes quite appreciable . due to fast variation of the cp potential , the transition between the strong and weak coupling regimes is quite sensitive . this requires a condensate to be confined at the sub - micro size corresponding to large values of @xmath159 , therefore safely lying in the regime where trap deconfining becomes less critical . , the condensate particles can escape as a consequence of the mechanical heating . panel b ) effective rabi frequency ( black solid line ) and heating rate ( red solid line ) . the shadowed region represents the strong - coupling regime . the dashed vertical line depicts the critical distance @xmath160 ( see text ) . ] we have demonstrated that the casimir - polder due to the interaction of a bose - einstein condensate and a graphene sheet can be perturbed by the mechanical out - of - plane vibrations and by the bogoliubov excitations . at cryogenic temperatures , the kirchoff - love flexural modes can be quantized , therefore coupling to the bogoliubov excitations of the condensate . as a result , a phonon - polariton is formed for sufficiently large separation distances , for which the effective rabi frequency ( the coupling strength ) dominates over the heating rate . our results may motivate a scheme for the sympathetic cooling of monolayer graphene via vacuum - fluctuation forces . in a feasible experimental situation , such a cooling scheme may be implemented in combination with an optomecanical cooling protocol , in order to pre - cool the membrane down to a temperature of few tens of mk @xcite . we may think that the sympathetic cooling via the casimir - polder interaction may then be used to further cool the zero - point mode , since the condensate temperature is of a few tens of nk . in fact , the major interest in cooling the mechanical modes of graphene lies in the perspective of controlling its transport properties , as the electronic response at low temperatures is intimately related by the scattering with the flexural phonons @xcite .	we consider the mechanical coupling between a two - dimensional bose - einstein condensate with a graphene sheet via the vacuum fluctuations of the electromagnetic field which are at the origin of the so - called casimir - polder potential . by deriving a self - consistent set of equations governing the dynamics of the condensate and the flexural ( out - of - plane ) modes of the graphene , we can show the formation of a new type of purely acoustic quasi - particle excitation , a phonon - polariton resulting from the coherent superposition of quanta of flexural and bogoliubov modes .
nanomaterials are tiny platforms on which a beautiful interplay between structure and property can be appreciated . for the last decade , advanced nanofabrication techniques have put out various nanostructured materials with novel geometry @xcite , many of which exhibit unprecedented properties not seen in macroscopic structures . among such nanostructures , carbon nanotubes have drawn great deal of attention . the salient feature of carbon nanotubes is their mechanical robustness and resilience @xcite : due to the extremely large stiffness , for instance , their thermal conductivity becomes higher than even that of diamond @xcite . besides , their flexibility in bending @xcite , twisting @xcite , radial compression @xcite , and the associated variations in the physical properties hold promise for developing nanoelectromechanical devices @xcite . towards successful implementations of such ideas , computational studies have been playing a vital role in complementing experimental observations , often difficult and incomplete for nanomaterials @xcite . in view of structure - property relations , radial deformation is expected to elicit untouched behavior of multiwall carbon nanotubes ( mwnts ) ; see fig . [ fig_01 ] for a microscopic image of a typically synthesized mwnt @xcite . the radial deformation disturbs both the equal spacings between the concentric walls in mwnts and the in - plane hexagonal lattice within each wall . in particular , spatial modulation in the wall - wall separation may enhance ( or hinder ) the electron charge transfer between neighboring walls if they locally get close to ( or pull away from ) each other @xcite ; this implies significant changes in various observables of mwnts under external pressure . despite the broad interest , theoretical efforts on the mechanics of radially pressurized mwnts remain limited @xcite . this scarcity is mainly because atomistic - based simulations of mwnts having many walls are computationally very expensive . in the present work , we employ a powerful alternative , the continuum elastic thin - shell theory approach , to analyze the stable cross - sectional shapes of mwnts under hydrostatic pressure . such approaches are often plagued with inaccurate or inconsistent elastic moduli . here , we carefully select the intra - wall and wall - wall elastic constants for an accurate prediction of the critical pressure @xmath0 above which the original circular cross - section transforms into radially corrugated ones . in the corrugation modes , each wall exhibits a wavy structure in the circumferential direction along the tube axis . such the pressure - induced corrugation is attributed to the mismatch in the mechanical stability between the flexible outmost walls with large tube diameters and rigid innermost walls with small diameters . although atomistic simulations may provide precise estimations of physical quantities in general , they often demand huge computational resources in systems of interest . on this background , the thin shell theory based analysis for the carbon nanotube mechanics has long been developed @xcite . along the continuum thin - shell method , a mwnt is mapped onto a set of @xmath1 continuum elastic hollow tubes of radii @xmath2 ( @xmath3 ) . a point on the circle corresponding to the cross - section of the @xmath4th tube is described by @xmath5 in terms of the polar coordinates ; @xmath6 is the circumferential angle around the tube axis . under pressure @xmath7 , the point moves to @xmath8 \cos\theta - v_i ( \theta , p ) \sin \theta , \;\right . \\ & & \left . \ ; \left[r_i+u_i(\theta , p)\right ] \sin\theta + v_i ( \theta , p ) \cos \theta \ ; \right ) . \label{eq_02}\end{aligned}\ ] ] , and the innermost tube diameter , @xmath9 , are estimated as 12 and 3.0 nm , respectively . after ref . @xcite.,width=162 ] if the deformation amplitudes , @xmath10 and @xmath11 , are sufficiently small , the mechanical energy of the @xmath4th tube per unit axial length is given by @xmath12 here , @xmath13 and @xmath14 are respectively the in - plane and bending strains of the @xmath4th wall , both depending on @xmath10 , @xmath11 and their derivatives with respect to @xmath6 @xcite . the constant @xmath15 denotes the in - plane stiffness , @xmath16 the flexural rigidity , and @xmath17 the poisson ratio of each wall . for quantitative discussions , the values of @xmath15 and @xmath16 must be carefully determined . in the conventional thin - shell theory for macroscopic objects , @xmath15 and @xmath16 are related to the young s modulus @xmath18 of the wall and its thickness @xmath19 as @xmath20 , @xmath21 $ ] . however , for carbon nanotubes , the wall is made out of a monoatomic graphitic layer and consequently the notion of a wall thickness becomes elusive . hence , the macroscopic relations for @xmath15 and @xmath16 noted above fail since there is no unique way of defining the thickness of the graphene wall @xcite . thus the values of @xmath15 and @xmath16 should be evaluated ab - initio from direct measurements or computations of carbon sheets , without reference to the macroscopic relations . in actual calculations , we substitute @xmath22 nn / nm , @xmath23 nn@xmath24 nm , and @xmath25 along with the prior work @xcite based on the density functional theory . it should be noted that the values of @xmath15 and @xmath16 are essentially tube - diameter dependent . neverthelss , such the dependences become negligible when the tube diameter exceeds 1 nm , above which the elastic constants of carbon nanotubes converge to those of a planar graphene sheet @xcite . on this background , we will take into account only the nanotubes whose diameters are larger than 1 nm , which allows to fix the values of @xmath15 and @xmath16 as noted above . the stable cross - sections of mwnts under @xmath7 minimize the mechanical energy @xmath26 of the whole system that is described by @xcite @xmath27 where @xmath28 and @xmath29 with @xmath30 . the term @xmath31 accounts for the van der waals ( vdw ) interaction energy of adjacent pairs of walls , and @xmath32 is the negative of the work done by @xmath7 during cross - sectional deformation . note that @xmath26 is a functional of @xmath10 and @xmath11 ; therefore , the variational method allows to obtain the optimal solutions of @xmath10 and @xmath11 that minimize @xmath26 under a given @xmath7 . for @xmath33 , the outside walls have large diameters , and consequently are very flexible and susceptible to mechanical instabilities under radial pressure . the contrast in the radial rigidity between the outermost walls and the innermost ones triggers a non - trivial cross - sectional deformation observed in radially pressurized mwnts . in fact , the linearized buckling analysis leads to the conclusion that immediately above a critical pressure @xmath0 , the circular cross section of mwnts becomes radially deformed as described by @xcite @xmath34 the solution ( [ eq_05 ] ) represents a wavy structure of a mwnt s cross - section , called the radial corrugation with a mode index @xmath35 . the integer @xmath35 indicates the wave number of the corrugated walls , being uniquely determined by the one - to - one relation between @xmath35 and @xmath0 @xcite . we will find below that @xmath35 depends systematically on @xmath1 and the innermost tube diameter @xmath36 as presented in fig . [ fig_03 ] . it is known that in the axial buckling of mwnts , the buckling load is sensitive to the vdw interaction between adjacent walls @xcite . likewise , @xmath0 is thought to depend on the strength of vdw interactions ; this is the reason why we address the rigorous formulation of the interaction . using both a _ discrete - pairwise _ summation and a _ continuum _ approach , we derive the coefficients @xmath37 in eq . ( [ eq_03 ] ) through a first order taylor approximation of the vdw pressure @xcite associated with the vdw potential @xmath38 $ ] . here , @xmath39 is the distance between a pair of carbon atoms , @xmath40 is the equilibrium distance between two interacting atoms , and @xmath41 is the well depth of the potential @xcite . the resulting equilibrium spacing between neighboring walls , used here to define the geometry of the mwncts , is @xmath42 . the derivative @xmath43 is the force between the two atoms . there exist several continuum models for the vdw interactions @xcite . in ref . @xcite , expressions for the pressure and the vdw interaction coefficients were obtained by integrating the continuum vdw force and its derivative on curved wall surface , while disregarding the vectorial nature of the force . the significance of the vectorial nature of the force was addressed in ref . @xcite , where analytical expressions for the vdw pressure were obtained by considering only the component of the vdw force normal to the wall . it was emphasized in ref . @xcite that for a two - walled carbon nanotube , the pressure exerted on inner wall is different from the pressure on the outer wall . the pressures on the inner and outer walls for a concentric two - walled tube with radii @xmath44 and @xmath45 are given below ( with positive signs for compression ) : @xmath46 where @xmath47 with @xmath48 nm@xmath49 being the area density of carbon atoms . we used the notations : @xmath50 where @xmath51 , @xmath52 , @xmath53 , @xmath54 and @xmath55 . ( bottom).,width=219 ] ( bottom).,width=215 ] in the following , we obtain analytical expressions for @xmath56 linearizing the formula ( [ eq_formula ] ) for the pressure . the formula takes into account correctly the normal - to - wall component of vdw forces , and avoids common assumptions like @xmath57 @xcite and @xmath58 @xcite ; therefore , the resulting @xmath56 will be also free from unnecessary assumptions . in the present work , infinitesimal deformation is considered . hence , the linearzed pressure is needed for vdw energy calculation . note that the vdw energy depends quadratically on the change in spacing between two adjacent walls . consider two consecutive walls with radii @xmath59 and @xmath60 , where the subscripts @xmath4 and @xmath61 correspond to @xmath62 and @xmath63 respectively . the vdw energy stored due to symmetric perturbation along the positive direction of pressure is given by @xmath64 where @xmath65 is the mean radius and @xmath66 is the vdw pressures on the @xmath4th wall . the corresponding linearized pressure is given by @xmath67 . ( [ eq_stored ] ) , @xmath68 describes the length of the infinitesimal element on which the pressure is acting . using the linearized pressure and comparing with eq . ( [ eq_03 ] ) , we get the expressions for vdw coefficients as @xmath69 where @xmath70 see appendix for the derivatives of @xmath71 . note that the @xmath56 is symmetric . -dependence of critical pressure curve @xmath0 . immediately above @xmath0 , the original circular cross section of mwnts gets radially corrugated.,width=170 ] these analytical expressions for the stiffness coefficients associated with the vdw interactions obtained with the continuum approximation are compared with the coefficients obtained using a discrete pairwise calculation . in the later approach , the pressure coefficients are computed by summing the component normal to the wall of the linearized vdw forces acting on an atom @xmath72 , _ i.e. _ the normal component of @xmath73 , where the linearized force @xmath74 is evaluated at the actual distance between the pair of interacting atoms @xmath72 and @xmath75 , @xmath76 . in ref . @xcite , the linearization is done at the equilibrium distance of the potential . noting that most of the interacting atom pairs are not at the equilibrium distance , it is apparent that the linearization should be done at the actual distance between each pair of interacting atoms . the discrete pressure can be obtained by dividing the force exerted on one atom by the area per atom . the vdw force on one atom is obtained by considering the interactions between this atom with all the atoms on a neighboring wall . the effect of the relative lattice placement of interacting walls is averaged out by integrating the interactions over a symmetric triangle of the representative hexagonal cell of graphene . the normalized pressure derivatives @xmath77 and normalized vdw interaction coefficients @xmath78 are plotted against normalized radius @xmath79 in fig . [ fig_cij ] . a good match is observed between the _ discrete - pairwise _ and the _ continuum _ approach . the interaction coefficient for two horizontally placed graphene walls , @xmath80 , is obtained from the second derivative of the analytic continuum interaction potential @xcite . for larger radii , @xmath56 approaches to @xmath80 , corroborating our results for nanotubes . figure [ fig_03 ] plots @xmath0 as a function of @xmath1 for various values of @xmath9 . an initial increase in @xmath0 at small @xmath1 ( except for @xmath81 nm ) is attributed to the enhancement of radial stiffness of the entire mwnt by encapsulation . this stiffening effect disappears with further increase in @xmath1 , resulting in decay or convergence of @xmath82 . it is noteworthy that mwnts practically synthesized often show @xmath9 larger than those presented in fig . [ fig_03 ] . in fact , the mwnt depicted in the image of fig . [ fig_01 ] gives @xmath83 nm , for which @xmath82 lies at several hundreds of mpa as estimated from fig . [ fig_03 ] . such degree of pressure applied to mwnts is easily accessible in high - pressure experiments , supporting the feasibility of our theoretical predictions . the @xmath9-dependence of @xmath0 for various fixed @xmath1 is provided in fig . [ fig_04 ] in a normalized manner ; when @xmath84 , we obtain @xmath85 in agreement with previous studies on single - walled nanotubes @xcite . it also deserves comment that a radial pressure large enough to cause corrugation can be achieved by electron - beam irradiation ; the self - healing nature of eroded carbon walls gives rise to a spontaneous contraction that exerts a high pressure on the inner walls @xcite -dependence of normalized critical pressure curve @xmath0 . inset : cubic behavior of the @xmath0 curve when @xmath84 . the dashed line showing the power - law @xmath85 is a visual reference.,width=188 ] figure [ fig_05 ] shows the stepwise increases in the corrugation mode index @xmath35 , defined by eq . ( [ eq_05 ] ) . for all @xmath9 , the deformation mode observed just above @xmath0 abruptly increases from @xmath86 to @xmath87 at a certain value of @xmath1 , followed by the successive emergence of higher corrugation modes with larger @xmath35 . these successive transitions in @xmath35 at @xmath33 originate from the mismatch in the radial stiffness of the innermost and outermost walls . a large discrepancy in the radial stiffness of the inner and outer walls results in a maldistribution of the deformation amplitudes of concentric walls interacting via vdw forces , which consequently produces an abrupt change in the observed deformation mode at a certain value of @xmath1 . a possible physical consequence of radial corrugation is a pressure - driven change in quantum transport of carriers in radially corrugated nanotubes . it is known that mobile carriers whose motion is confined to a thin curved layer behave differently from those on a conventional flat plane because of the curvature - induced electromagnetic field @xcite . another interesting issue is the effect of atomic lattice registry on non - linear deformation ( _ i.e. _ large deformation amplitude ) regimes @xcite . in the latter case , the degree of commensurance in atomic structures between neighboring carbon layers plays a prominent role in determining the optimal morphology of highly corrugated mwnts , in which the interwall spacings partially vanish . crumpling or twisting caused by radial pressure is expected , and in fact were observed in preliminary calculations @xcite . the thin - shell theory has been employed to derive the critical hydrostatic pressure @xmath0 above which the cross section of pressurized mwnts transforms into radially corrugated patterns . the value of @xmath0 lies on the order of several hundreds of mpa for practically synthesized mwnts with @xmath33 and @xmath88 nm , indicating the experimental feasibility of the prediction . the enegetically favored corrugation pattern strongly depends on @xmath1 and @xmath9 , where a larger @xmath1 and smaller @xmath9 give a larger corrugation mode index @xmath35 . our results provide a clue in developing mwnt - based devices operating under high pressure . we acknowledge m. rahimi , e. osawa , y. suda , k. yakubo and t. mikami for helpful discussions . hs is thankful for the financial supports provided by kakenhi and hokkaido gas co. , ltd . sg acknowledges the support of the spanish ministry of science and innovation through the juan de la cierva program . ma acknowledges the support of the european research council ( fp7/2007 - 2013)/erc grant agreement nr 240487 and the prize `` icrea academia '' funded by the generalitat de catalunya . numerical simulations were partly carried out using the facilities of the supercomputer center , university of tokyo . it follows from eq . ( [ eq_fpm ] ) that @xmath89 all @xmath90 functions obey the recurrence relation @xmath91 from complete elliptic integrals of the first and second kind , @xmath92 and @xmath93 , respectively ; @xmath94 and @xmath95 in the current notation . hence , the required derivatives of @xmath96 in eq . ( [ eq_appdx ] ) are given below : @xmath97 other derivatives are given by @xmath98 and @xmath99 . 99 s. tanda , t. tsuneta , y. okajima , k. inagaki , k. yamaya and n. hatakenaka , nature * 417 * ( 2002 ) 397 . m. ozawa , h. goto , m. kusunoki and e. osawa , j. phys . chem . b * 106 * ( 2002 ) 7135 . j. onoe , t. nakayama , m. aono and t. hara , appl . * 82 * ( 2003 ) 595 . y. tatewaki , t. hatanaka , r. tsunashima , t. nakamura , m. kimura and h. shirai , chem . asian j. * 4 * ( 2009 ) 1474 . v. saranathan , c. o. osuji , s. g. j. mochrie , h. noh , s. narayanan , a. sandy , e. r. dufresne and r. o. prum , pnas * 107 * ( 2010 ) 11676 . h. shima and m. sato , _ elastic and plastic deformation of carbon nanotubes _ , ( pan stanford pub . , 2011 ) _ in press_. p. kim , l. shi , a. majumdar and p. l. mceuen , phys . rev . lett . * 87 * ( 2001 ) 215502 . a. sears and r. c. batra , phys . b * 69 * ( 2004 ) 235406 . i. arias and m. arroyo , phys . rev * 100 * ( 2008 ) 085503 ; m. arroyo and i. arias , j. mech . solids * 56 * ( 2008 ) 1224 . x. huang , j. zou and s. zhang , appl . phys . lett . * 93 * ( 2008 ) 031915 ; j. zou , x. huang , m. arroyo and s. zhang , j. appl . phys . * 105 * ( 2009 ) 033516 . d. b. zhang , r. d. james and t. dumitric , phys . b * 80 * ( 2009 ) 115418 . f. khademolhosseini , r. k. n. d. rajapakse and a. nojeh , comp . sci * 48 * ( 2010 ) 736 . b. i. yakobson , c. j. brabec , and j. bernholc , phys . * 76 * ( 1996 ) 2511 . i. palaci , s. fedrigo , h. brune , c. klinke , m. chen , and e. riedo , phys . * 94 * ( 2005 ) 175502 . c. gmez - navarro , j. j. senz , and j. gmez - herrero , phys . * 96 * ( 2006 ) 076803 . m. barboza , a. p. gomes , b. s. archanjo , p. t. araujo , a. jorio , a. s. ferlauto , m. s. c. mazzoni , h. chacham and b. r. a. neves , phys . rev * 100 * ( 2008 ) 256804 . e. m. diniz , r. w. nunes , h. chacham and m. s. c. mazzoni , phys . rev . b * 81 * ( 2010 ) 153413 . l. x. dong , a. subramanian and b. j. nelson , nano today * 2 * ( 2009 ) 12 . s. roy and z. gao , nano today * 4 * ( 2009 ) 318 . h. rafii - tabar , _ computational physics of carbon nanotubes _ , ( cambridge univ . pr . , 2007 ) a. okita , y. suda , a. oda , j. nakamura , a. ozeki , k. bhattacharyya , h. sugawara , and y. sakai , carbon * 45 * ( 2007 ) 1518 . v. zlyomi , j. koltai , . rusznyk , j. krti , . gali , f. simon , h. kuzmany , . szabados and p. r. surjn , phys . b * 77 * ( 2008 ) 245403 . m. ahlskog , o. herranen , a. johansson , j. leppniemi and d. mtsuko , phys . b * 79 * ( 2009 ) 155408 . z. xu , l. wang , and q. zheng , small * 4 * ( 2008 ) 733 . s. h. yang , l. l. feng and f. feng , j. phys . d : appl . * 42 * ( 2009 ) 055414 . x. huang , w. liang and s. zhang , nanoscale res . * 6 * ( 2011 ) 53 . c. q. ru , phys . b * 62 * ( 2000 ) 16962 . k. n. kudin , g. e. scuseria , and b. i. yakobson , phys . b * 64 * ( 2001 ) 235406 . a. pantano , d. m. parks , and m. c. boyce , j. mech . . solid . * 52 * ( 2004 ) 789 . y. huang , j. wu , and k. c. hwang , phys . b * 74 * ( 2006 ) 245413 . k. chandraseker and s. mukherjee , comp . * 40 * ( 2007 ) 147 . j. l. sanders jr . , _ quart . _ * 21 * ( 1963 ) 21 . s. s. gupta , f. g. bosco and r. c. batra , comp . * 47 * ( 2010 ) 1049 , and references therein . h. shima and m. sato , nanotechnology * 19 * ( 2008 ) 495705 ; phys . a * 206 * ( 2009 ) 2228 . x. q. he , s. kitipornchai and k. m. liew , j. mech . solids * 53 * ( 2005 ) 303 . w. b. lu , b. liu , j. wu , j. xiao , k. c. hwang , s. y. fu , and y. huang , appl . phys . lett . * 94 * ( 2009 ) 101917 . l. a. girifalco , m. hodak , and r. s. lee , phys . b * 62 * ( 2000 ) 13104 . ru , j. mech . solids , * 49 * ( 2001 ) 1265 . d. y. sun , d. j. shu , m. ji , f. liu , m. wang , and x. g. gong , phys . rev . b * 70 * ( 2004 ) 165417 . f. banhart and p. m. ajayan , nature * 382 * ( 1996 ) 433 . l. sun , f. banhart , a. v. krasheninnikov , j. a. rodrguez - manzo , m. terrons , and p. m. ajayan , science * 312 * ( 2006 ) 1199 . h. shima , m. sato , k. iiboshi , s. ghosh and m. arroyo , phys . rev . b * 82 * ( 2010 ) 085401 . h. shima , h. yoshioka and j. onoe , phys . b * 79 * ( 2009 ) 201401(r ) ; physica e * 42 * ( 2010 ) 1151 ; h. shima , s. ono and h. yoshioka , eur . j. b * 78 * ( 2010 ) 481 . s. ono and h. shima , phys . b * 79 * ( 2009 ) 235407 ; physica e * 42 * ( 2010 ) 1224 ; j. phys . ( 2011 ) _ in press_. h. taira and h. shima , j. phys . : condens . mat . * 22 * ( 2010 ) 075301 ; _ ibid . _ , * 22 * ( 2010 ) 245302 ; j. phys . a : math . theor . * 43 * ( 2010 ) 354013 . m. f. yu , m. j. dyer , j. chen , d. qian , w. k. liu and r. s. ruoff , phys . b * 64 * ( 2001 ) 241403 . b. liu , m. f. yu and y. g. huang , phys . b * 70 * ( 2004 ) 161402 . n. marom , j. bernstein , j. garel , a. tkatchenko , e. joselevich , l. kronik and o. hod , phys . * 105 * ( 2010 ) 046801 . m. rahimi , m. arroyo , h. shima and m. sato , _	the elastic radial deformation of multiwall carbon nanotubes ( mwnts ) under hydrostatic pressure is investigated within the continuum elastic approximation . the thin - shell theory , with accurate elastic constants and interwall couplings , allows us to estimate the critical pressure above which the original circular cross - section transforms into radially corrugated ones . the emphasis is placed on the rigorous formulation of the van der waals interaction between adjacent walls , which we analyze using two different approaches . possible consequences of the radial corrugation in the physical properties of pressurized mwnts are also discussed . carbon nanotube , high pressure , elastic deformation , radial buckling , van der waals interaction
for the analysis of the experimental data the simplest two - parameter fermi ( 2pf ) distribution is often used : @xmath1^{-1},\ ] ] where @xmath2 is the half - density radius and @xmath3 is the diffuseness parameter of the distribution . here @xmath4 is for neutron and @xmath5 for proton distributions . there are two opposite pictures in description of a two - component finite fermi system with 2pf . the first one is the so - called `` neutron skin - type '' distribution having the neutron half - density radius larger than the proton half - density radius , @xmath6 , and equal diffuseness parameters @xmath7 . the second one is the `` neutron halo - type '' distribution having @xmath8 and @xmath9 . a mixture of the neutron `` skin - type '' and `` halo - type '' distributions having @xmath6 and @xmath9 is also possible . it is found that the experimental data can be reproduced by a variety of these 2pf distributions with different values of @xmath10 and @xmath11 @xcite . in that sense there is an ambiguity in theoretical description of the experimental data . in the present paper we study this ambiguity within the extended thomas - fermi approximation ( etfa ) based on the direct variational method . the nucleon densities @xmath12 are generated by the profile functions which are obtained from the requirement that the energy of the nucleus should be stationary with respect to variations of these profiles . in general , the total energy of a nucleus includes the kinetic and potential energies and is given by @xcite @xmath13 , \label{etot}\ ] ] where the potential energy density , @xmath14 , includes the nn - interaction , @xmath15 , the spin - orbit part of the nn - interaction , @xmath16 , and the coulomb energy density , @xmath17 : @xmath18 in the framework of etfa , the kinetic energy density is given by the sum of the neutron and proton contributions @xcite @xmath19 where @xmath20 . \label{eqkin}\ ] ] here @xmath21 is the bare nucleon mass . the semiclassical consideration gives the value of the parameter @xmath22 in eq . ( [ eqkin ] ) @xcite . in the asymptotic limit @xmath23 , the semiclassical particle density @xmath24 with @xmath22 falls off to zero significantly faster than the one from the quantum - mechanical calculation , where one has @xmath25 . we will use both values of the parameter @xmath26 to study the neutron halo and skin appearances and their superposition in nuclei . for the effective nucleon - nucleon interaction we will use the skyrme force @xcite @xmath27 \nonumber \\ & + & \frac{t_3}{12}\rho^\nu \left[\left(1+\frac{x_3}{2}\right ) \rho^2 -\left(x_3+\frac{1}{2}\right)\left(\rho_n^2+\rho_p^2\right)\right ] \nonumber \\ & + & \frac{1}{4}\left[t_1\left(1+\frac{x_1}{2}\right)+t_2\left(1+\frac{x_2}{2}\right)\right]\epsilon_{kin}\rho \nonumber \\ & + & \frac{1}{4}\left[t_2\left(x_2+\frac{1}{2}\right)-t_1\left(x_1+\frac{1}{2}\right)\right ] \left(\epsilon_{kin , n } \rho_n + \epsilon_{kin , p } \rho_p\right ) \nonumber \\ & + & \frac{1}{16}\left[3t_1\left(1+\frac{x_1}{2}\right)-t_2\left(1+\frac{x_2}{2}\right)\right]\left(\mathbf{\nabla}\rho\right)^2 \nonumber \\ & -&\frac{1}{16}\left[3 t_1\left(x_1+\frac{1}{2}\right)+t_2\left(x_2+\frac{1}{2}\right)\right ] \left[\left(\mathbf{\nabla}\rho_n\right)^2+\left(\mathbf{\nabla}\rho_p\right)^2\right],\end{aligned}\ ] ] where @xmath28 is the total density of nucleons . the spin - orbit part of the skyrme force is written as @xmath29.\ ] ] here @xmath30 , @xmath31 , @xmath32 , @xmath33 , @xmath34 , @xmath35 , @xmath36 , @xmath37 , @xmath38 , @xmath39 are the parameters of the skyrme interaction and the effective mass of a nucleon is @xmath40.\ ] ] the coulomb energy density is taken in the well - known form @xcite @xmath41 the unknown values of @xmath24 can be evaluated from the condition of equilibrium . the corresponding condition implies that the total energy @xmath42 reaches a minimum value for a given number of neutrons , @xmath43 , and protons , @xmath44 , @xmath45 following the direct variational method @xcite the trial function for @xmath46 is taken as a power of the fermi function @xmath47^{-\xi } , \label{rhoq}\ ] ] where @xmath48 , @xmath49 , @xmath50 and @xmath51 are the variational parameters . the nuclear radius for nucleons is defined as @xmath52 the difference between the neutron and proton root mean square radii gives the neutron - skin thickness @xmath53 from the definition of trial functions it is seen that the value of @xmath0 can be described by the different radii @xmath2 ( skin effect ) and the different diffusenesses @xmath3 ( halo effect ) of neutron and proton distributions , see also @xcite . within the leptodermous approximation @xmath54 one can separate the above contributions and the value of @xmath0 is written as @xmath55 where @xmath56 and @xmath57 are caused by the skin and halo effects , respectively . the corresponding values are given by ( see appendix a ) @xmath58 \left ( \frac{a}{r}\right)^2 \right . \nonumber \\ & & - \left . \frac{1}{3}\left(75\kappa_0 ^ 3(\xi)-204\kappa_0(\xi)\kappa_1(\xi)+81\kappa_2(\xi)\right)\left(\frac{a}{r}\right)^3 \right\ } , \label{drhor}\end{aligned}\ ] ] and @xmath59 \frac{a}{r } \right . \nonumber \\ & & + \left . \frac{1}{2}\left(75\kappa_0 ^ 3(\xi)-204\kappa_0(\xi)\kappa_1(\xi)+81\kappa_2(\xi)\right ) \left(\frac{a}{r}\right)^2 \right\ } , \label{drhoa}\end{aligned}\ ] ] where we have introduced the notations : @xmath60 , @xmath61 , @xmath62 , and @xmath63 , and @xmath64 are the generalized fermi integrals ( see appendix a ) . from the skin effect , @xmath65 ( solid lines ) and the halo effect , @xmath66 ( dashed lines ) versus the parameter @xmath26 . the calculations have been performed with skm parametrization for nuclei @xmath67sn and @xmath68pb . ] in fig . [ fig1 ] the partial contributions to the neutron - skin thickness from the skin effect , @xmath65 , and the halo effect , @xmath66 , are shown . one can see from fig . [ fig1 ] that the halo partial contribution increases and the skin partial contribution decreases with the increase of parameter @xmath26 . close to the semiclassical approach region , @xmath69 , the partial contributions from the halo effect are negative and therefore the partial contributions from the skin effect are greater than the unity . the reason for that is the high value of the parameter @xmath51 for the asymmetric fermi function , ( see below fig . [ fig2 ] ) , so the densities fall off too quickly in the outer surface . versus @xmath26 . the calculations have been performed for nuclei @xmath67sn and @xmath68pb using skm parametrization . ] this means that the halo component is negative and reduces the difference between the neutron and proton root mean square radii . in the region @xmath69 one has more preferably proton halo effect as reported in @xcite . for the values of @xmath70 , where the tails of the densities are close to that obtained in the quantum mechanical calculations , the halo and skin components of @xmath0 are found to be approximately equal . from the skin effect , @xmath65 ( solid lines ) and the halo effect , @xmath66 ( dashed lines ) versus the parameter @xmath26 . the calculations have been performed for @xmath67sn with the following skyrme force parametrizations : 1 - siii , 2 - skm , 3 - sly230b . ] the sensitivity of the partial contributions @xmath65 and @xmath66 on the skyrme force parametrization is shown in fig . [ fig3 ] . it is seen from fig . [ fig3 ] that the calculated values change of about 20% for different parametrizations . between the rms radii of the neutron and proton distributions as a function of @xmath71 . the solid and dashed lines are calculations with @xmath25 and @xmath72 for skm parametrization of the skyrme force . the experimental data are taken from ref . ] the dependence of @xmath0 on the asymmetry parameter @xmath73 together with the experimental data @xcite are shown in fig . [ fig4 ] . as it is seen from the figure the calculation with the `` quantum mechanical '' value @xmath25 ( solid curve ) , in general , agrees better with the experimental data than for @xmath74 ( dashed curve ) . the obtained dependencies are approximately linear for the considered values of isotopic asymmetry parameter @xmath73 . in fig . [ fig5 ] the analogous calculations for tin isotopes together with the experimental data are shown as a function of mass number . similarly to the previous cases , the calculation with @xmath25 describes the experimental data better than the one with @xmath74 . between the rms radii of the neutron and proton distributions as a function of @xmath75 for sn isotopes . the solid and dashed lines are calculations with @xmath25 and @xmath72 for skm force . the experimental data are taken from ref . ] the neutron - to - proton density ratio in the peripheral region with respect to the bulk value is described by the peripheral halo factor @xcite . following @xcite we will also use the theoretical estimation of the halo factor as @xmath76 in fig . [ fig6 ] we show the calculations of the theoretical halo factor for two values of the parameter @xmath26 together with the experimental data @xcite . as in the previous figures the solid lines correspond to the calculation with @xmath25 and dashed lines correspond to @xmath74 . sn and @xmath68pb as a function of the distance from the center of the nucleus calculated using the eq . ( [ fhalo ] ) . the values of the halo factor deduced from experiment are marked by crosses @xcite.,title="fig:"]sn and @xmath68pb as a function of the distance from the center of the nucleus calculated using the eq . ( [ fhalo ] ) . the values of the halo factor deduced from experiment are marked by crosses @xcite.,title="fig : " ] the left panel is the case for tin isotope @xmath77sn , and the right panel is for @xmath68pb . it is seen from fig . [ fig6 ] that the calculation with @xmath25 describes the experimental data better than the one with @xmath74 . we have considered the neutron skin and halo effects for medium and heavy nuclei within the extended thomas - fermi approximation using effective skyrme - like forces and the direct variational method . the nucleon densities @xmath78 and @xmath79 are generated by the profile functions which are obtained by the requirement that the energy of the nucleus should be stationary with respect to variations of these profiles . using the leptodermous properties of the profile nucleon densities @xmath80 and @xmath79 , we have obtained the analytical expression for the isovector shift of the rms radii @xmath0 as a superposition of the two terms . the first one , @xmath81 , describes the neutron `` skin - type '' distribution and the second one , @xmath82 , describes the neutron `` halo - type '' distribution . numerical calculations show that the partial contributions to the neutron - skin thickness from the skin effect , @xmath65 , and the halo effect , @xmath66 , depend on the parameter @xmath26 . with the inscrease of @xmath26 from @xmath72 to @xmath25 the partial contributions to the neutron - skin thickness from the skin effect decreases from 100% to about of 50% and the halo component increses from zero to about of 50% . the calculated isovector shift of the rms radii @xmath0 and halo factor @xmath83 give satisfactory description for the corresponding experimental data at @xmath25 . in this case we have superposition of the neutron skin and halo effects with approximately equal contributions . in this appendix , we will consider the radius for nucleons as @xmath84 where @xmath85 whithin the leptodermous approximation @xmath86 one has @xmath87 where @xmath64 are the generalized fermi integrals derived in ref . @xcite @xmath88}. \label{ki}\ ] ] inserting eqs . ( [ i2 ] ) and ( [ i4 ] ) into the eq . ( [ arms ] ) one gets @xmath89 from eq . ( [ a6 ] ) the isovector shift of the rms radii @xmath0 reads @xmath90 one can rewrite last terms of eq . ( [ a7 ] ) using the relations @xmath91 @xmath92 finally , we have @xmath93	the neutron skin and halo distributions in medium and heavy nuclei are calculated within the extended thomas - fermi approximation . calculations are carried out for the effective skyrme - like forces using the direct variational method . the analytical expression for the isovector shift of the rms radii @xmath0 as a sum of skin- and halo - like terms is obtained . the contribution of halo and skin terms to @xmath0 are found to be approximately equal .
a region of 100 arcmin@xmath3 around the hubble deep field north ( hdfn 10-arcmin field ) was observed by the uk radio interferometer merlin and the usa vla @xcite . 92 sources brighter than 40 @xmath1jy were found in the combined images , which reach a noise level of @xmath4jy using beam sizes of @xmath5 . these are the only observations apart from the hst images which provide morphological information . 8.4-ghz data are also available for most of the field @xcite allowing the radio spectral indices to be calculated . we employed data access , crossmatching and manipulation tools now available via the astrogrid workbench and the eurovo . these include the vizier and aladin services , topcat and a cut - out extractor for radio images which uses the parseltongue scripting interface developed by radionet . we compare the radio data with the _ chandra _ cdf(n ) source list @xcite and other goods @xcite hdfn observations to investigate the properties of active galaxies at @xmath6 : 1 . can radio emission originate from a process different from the x - ray source in the same galaxy ? if so , how much radiation has a common origin and how much is separate ? 2 . are high - redshift starbursts just analogues of local ulirgs or are they a distinct phenomenon only seen in the early universe ? the presence of rest - frame fir emission @xcite , the hst images @xcite and other non - radio data were used for supporting information but not as the primary classification criteria . we were thus able to classify the specific origins of the radio emission in each galaxy , which may be different from the sources of optical and other radiation from the same object . compact bright peaks with a flat radio spectrum are probably agn , especially if radio lobes extend beyond the optical limits . extended emission with a steep non - thermal spectrum is likely to be of starburst origin and in some cases optical knots of starformation can be seen . the latter are often associated with _ iso _ , _ spitzer _ or scuba sources @xcite and with interacting or distorted galaxies in the _ hst _ images . . shows two typical sources . 3 out of every 4 @xmath1jy sources are starbursts , including those with x - ray counterparts . three sources have both starburst and agn radio characteristics . 55 of the radio sources are among the 100 x - ray sources detected by _ chandra _ in the same field @xcite . a _ chandra _ source with luminosity @xmath7 w is probably an agn . if it has a hard photon index ( @xmath8 ) , this indicates the presence of an obscured ( type-2 ) agn @xcite . 64 type-2 agn were identified in the 10-arcmin field @xcite . over half of the x - ray sources with merlin+vla counterparts have @xmath9 w , 18 of which are classed as type-2 agn . these are shown by a blue a in fig . 2 . the radio emission is starburst dominated in at least 11 of the radio counterparts . the redshift distribution of radio+x - ray source classifications is shown in fig . 2 . at @xmath10 , under 1/3 x - ray sources in the hdfn have radio counterparts . this rises to over 1/2 at higher redshift and in most of these the high x - ray luminosity indicates the presence of an agn . the @xmath11 relationship for starbursts at @xmath12 @xcite breaks down at higher @xmath0 where we see a dramatic increase in the scatter in fig . this is the case even if only the soft - band or the de - absorbed hard - band x - ray luminosity is considered . half of the 18 x - ray selected type-2 agn with radio counterparts are at @xmath13 , all but one of which are radio starbursts . two also have radio agn cores . a large proportion of these objects are scuba sources , whose x - ray properties are discussed in detail by @xcite . all 4 of the radio agn identified with x - ray type-2 agn , with no starburst signature , are at lower redshifts . the mean angular size of the radio sources in the hdfn is @xmath14 , corresponding to a typical extent of 810 kpc for starbursts and the inferred starformation rates are 10002000 m@xmath15 yr@xmath16 for the higher - redshift sources . this is an order of magnitude more exended and more vigourous than the behaviour of any objects at @xmath2 . at @xmath13 , there is a strong association between the presence , but not the power , of faint radio and x - ray emission . the extended radio emission is starburst - dominated in 3/4 of the objects @xcite whilst 18 of the x - ray sources are hard enough and bright enough to be agn2 @xcite . 11 of these x - ray agn2 have radio starburst characteristics . their radio emission mechanisms must be dominated by star - formation whilst the agn provides most of the x - ray luminosity . the high-@xmath0 population of @xmath1jy radio sources contains a significant fraction of starbursts an order of magnitude more active and more extended than local ulirgs , and they simultaneously host highly x - ray - luminous obscured agn . , d. m. , et al . , 2003 , aj , 126 , 539 , d. m. , bauer , f. e. , chapman , s. c. , a smail , i. , blain , a. w. , brandt , w. n. , ivison , r. j. 2005 , apj , 632 , 736 , 2002 , aj , 124 , 2351 borys , c. , scott , d. , chapman , s. , halpern , m. , nandra , k. , pope , a. 2004 , mnras , 355 , 485 , m. a. , 2002 , a&a , 384 , l19 , d. , et al . , cesarsky , c. j. , chanial , p. , aussel , h. , franceschini , a. , fadda , d. , chary , r. r. , 2002 , a&a , 384 , 848 giavalisco , m. et al . , 2004 , apj , 600 , l93 , t. w. b. , et al . , 2005 , mnras , 358 , 1159 , p. , allen , m. g. , rosati , p. , walton , n. a. , 2004 , a&a , 424 , 545 richards , e. a. , kellermann , k. i. , fomalont , e. b. , windhorst , r. a. , partridge , r. b. , 1998 , aj , 116 , 1039 serjeant , s. et al . , 2003 , mnras , 344 , 887	we use virtual observatory methods to investigate the association between radio and x - ray emission at high redshifts . fifty - five of the 92 hdf(n ) sources resolved by combining merlin+vla data were detected by _ chandra _ , of which 18 are hard enough and bright enough to be obscured agn . the high-@xmath0 population of @xmath1jy radio sources is dominated by starbursts an order of magnitude more active and more extended than any found at @xmath2 and at least a quarter of these simultaneously host highly x - ray - luminous obscured agn .
let @xmath1 be a surface of genus @xmath2 . we can express @xmath1 as the boundary @xmath3 of a 3-dimensional handlebody @xmath4 by choosing @xmath0 disjoint , embedded circles , @xmath5 , linearly independent in homology : each of these is filled so as to bound a disc in @xmath4 . conversely , the handlebody @xmath4 determines the equivalence class of the @xmath0-tuple of attaching circles @xmath5 under the equivalence relation generated by isotopies , permutations and _ handleslides _ ( see , for instance , @xcite ) . handleslides are not possible when @xmath6 ( the attaching circle for a genus 1 handlebody is unique up to isotopy ) , so from now on we shall assume that @xmath7 . a handleside is a replacement @xmath8 where @xmath9 is disjoint from @xmath10 and isotopic to another circle @xmath11 , disjoint from @xmath12 , such that @xmath13 bounds an embedded pair of pants in @xmath1 ; see figure 1 . this condition implies that the circles @xmath14 are again linearly independent in homology , and hence determine a handlebody @xmath15 . there is a diffeomorphism @xmath16 acting as the identity on @xmath17 , since , once the @xmath18 curves @xmath19 have been collapsed , @xmath9 becomes isotopic to @xmath20 . @xmath21 at 220 325 @xmath21 at 220 115 @xmath11 at 165 245 @xmath9 at 165 32 @xmath20 at 330 245 @xmath20 at 323 50 @xmath22 at 220 60 @xmath23 at 350 20 , @xmath20 and @xmath21 bound an embedded pair of pants . in the lower diagram , @xmath9 intersects @xmath20 in two points ; this pattern of intersection is the one we shall use in the proof of our theorem.,title="fig:",scaledwidth=70.0% ] fix , once and for all , a complex structure @xmath24 on @xmath1 . the @xmath0-fold symmetric product @xmath25 is then a complex manifold and hence also a differentiable manifold . in working with @xmath25 one should be aware that its differentiable structure depends on @xmath24 but its diffeomorphism type does not . the @xmath0-tuple of disjoint attaching circles @xmath5 determines an embedded torus in @xmath25 , @xmath26 its _ heegaard torus_. here @xmath27 is the quotient map ; it restricts to an embedding on @xmath28 since the circles @xmath29 are disjoint . likewise , the @xmath0-tuple @xmath30 determines a torus @xmath31 these tori are totally real , that is , their tangent spaces contain no complex lines ; indeed , if we choose an area form @xmath32 on @xmath1 then the complement of the diagonal in @xmath25 has an induced symplectic form @xmath33 , and this form makes the two tori lagrangian . note , however , that the push - forward of a 2-form by a finite holomorphic map is not globally a smooth 2-form but rather a current , with singularities along the branch locus in this case , the diagonal . we now recall that for @xmath34 one has @xmath35 by an isomorphism equivariant under the actions of the mapping class group on the two sides . write @xmath36 for the class corresponding to @xmath37 , or for its image in @xmath38 ( this leaves a sign ambiguity which can be fixed by giving an alternative description : @xmath36 is poincar dual to the divisor @xmath39 , embedded by @xmath40 for some fixed @xmath41 ) . when @xmath42 , write @xmath43 for ( the image in real cohomology of ) the generator of the cyclic group of classes in the summand @xmath44 invariant under the mapping class group . this time the sign ambiguity can be fixed by saying that @xmath43 is the pullback by the abel jacobi map of an ample class ( the theta - divisor ) on the jacobian . suitable references include @xcite . our first , preliminary result is as follows . [ forms ] suppose @xmath45 is a @xmath0-tuple of disjoint , linearly independent attaching circles , and @xmath46 another @xmath0-tuple differing from the first by a handleslide . for real numbers @xmath47 with sufficiently small absolute value , and arbitrary positive area forms @xmath48 on @xmath1 , there exist symplectic forms @xmath49 on @xmath25 , taming the complex structure @xmath50 , such that 1 . @xmath49 agrees with the product form @xmath33 in neighbourhoods of the heegaard tori @xmath51 and @xmath52 ; and 2 . @xmath49 represents the class @xmath53 . one can in fact take @xmath49 to be khler . a short proof can be given using smoothing theory for currents , as in the author s earlier unpublished note @xcite which has now been incorporated as the last section ( [ smoothing ] ) of this paper . an immediate consequence of this smoothing theory ( corollary [ smoothing cor ] ) is that there are forms satisfying the proposition in the class @xmath36 . from this it is easy to deduce the result : one can perturb such a form by adding a small multiple of a closed @xmath54-form supported near the diagonal @xmath55 and poincar dual to @xmath55 . since @xmath56 = \eta - \theta$ ] , can be described algebro - geometrically as @xmath57 . hence @xmath56= c_1(t \operatorname{sym}^n(\sigma))=(n+1-g)\eta - \theta$ ] @xcite . ] these perturbed forms do the job . this paper actually contains a second proof of the proposition , independent from that given in section [ smoothing ] . the second proof is , however , valid only when @xmath58 , and the forms it produces are not necessarily khler . it can be found in section [ isotopies ] , subsumed in the proof of our main theorem , which we now proceed to state . for a form @xmath49 as in proposition [ forms ] , the tori @xmath51 and @xmath52 are lagrangian . one may ask whether they are isotopic through lagrangians , or indeed hamiltonian - isotopic . to make these into well - posed questions , we suppose that @xmath9 and @xmath20 intersect transversely in precisely two points , as in figure 1 . we also fix @xmath48 . the answers to the questions are then independent of the particular form @xmath49 and depend only on the parameter @xmath47 . indeed , it follows from moser s lemma and the convexity of the set of symplectic forms taming @xmath50 that , when such forms are cohomologous , they are related by symplectomorphisms supported away from @xmath59 . we answer these questions when @xmath58 . the answers depend on the areas @xmath60 and @xmath61 of the half - disc @xmath23 enclosed by arcs in @xmath9 and @xmath20 and the the annular region @xmath22 enclosed by @xmath21 and the remaining portions of @xmath9 and @xmath20 ( see figure 1 ) . [ main ] suppose @xmath45 is a @xmath0-tuple of disjoint , linearly independent attaching circles , and @xmath46 another @xmath0-tuple differing from the first by a handleslide . assume that @xmath62 is a transverse intersection consisting of precisely two points . fix a positive area form @xmath48 on @xmath1 , and let @xmath49 be a khler form satisfying the conclusions of the last proposition . suppose that @xmath47 is strictly positive . then 1 . @xmath51 is lagrangian - isotopic to @xmath52 . @xmath51 is hamiltonian - isotopic to @xmath52 if and only if @xmath63 3 . a lagrangian isotopy ( or indeed a hamiltonian one , when the area constraint is satisfied ) may be constructed as the product of the constant isotopy on the common factor @xmath64 of the two tori and an isotopy in @xmath65 , where @xmath66 is an ( arbitrary ) embedded pair of pants containing @xmath67 in its interior . the proof of lagrangian isotopy is constructive . the idea is to exploit the fact that @xmath9 and @xmath20 become isotopic in @xmath68 , the surface obtained by surgering out @xmath21 . the genus 2 case contains the heart of the matter . when @xmath1 has genus 2 , the surgery is mimicked symplectically by associating with @xmath21 a hypersurface @xmath69 which is the total space of an @xmath70-bundle @xmath71 . the pullback by @xmath72 of an area form on @xmath68 agrees with the symplectic form on @xmath73 restricted to @xmath74 . the preimage under @xmath72 of a lagrangian isotopy between the images of @xmath9 and @xmath20 in @xmath68 is then a lagrangian isotopy in @xmath73 . one needs to be able to exert enough control over @xmath74 to arrange that the isotopy begins at @xmath75 and ends at @xmath76 . note that the forms used in the proof are _ not _ the perturbed , smoothed currents mentioned above , but rather , symplectic structures arising from presentations of symmetric products as symplectic sums . however , as already noted , the existence of lagrangian or hamiltonian isotopies is a function of @xmath48 and @xmath47 and not of the particular form @xmath49 . it is an open question whether one can find a form making @xmath51 and @xmath52 into hamiltonian - isotopic lagrangians when @xmath77 and @xmath78 . it is conceivable that we have here an instance of _ symplectic fragility _ , the phenomenon whereby non - isotopic lagrangians become isotopic as soon as a perturbation parameter is switched on @xcite , though this seems rather unlikely since we shall find that the floer - theoretic properties of @xmath79 do not change as @xmath47 decreases to @xmath80 . in the gauge - theoretic interpretation of symmetric products as moduli spaces of vortices , the natural symplectic form represents ( up to scale ) a class @xmath53 where @xmath47 decreases to @xmath80 as the ` stability parameter ' tends to infinity @xcite . we shall use theorem [ main ] to re - prove the handleslide - invariance of ozsvth szab s heegaard floer homology ( * ? ? ? * section 9 ) ( the argument still draws on finiteness lemmas from @xcite ) . the precise statement uses notions from @xcite which we do not recapitulate here . [ heegcor ] let @xmath81 be a pointed heegaard diagram , weakly admissible for a @xmath82-structure @xmath83 , and suppose that @xmath84 differs from @xmath85 by a handleslide . suppose that the circle that is being replaced intersects its replacement non - trivially in the pattern of theorem [ main ] . assume that @xmath86 lies outside the handlesliding region . then there is a canonical isomorphism @xmath87 between the heegaard floer homology groups for the @xmath82-structure @xmath83 , and similarly for @xmath88 , @xmath89 and @xmath90 . referring to figure 1 , the ` handlesliding region ' means the union of the region bounded by @xmath21 and arcs in @xmath9 and @xmath20 and that bounded by arcs in @xmath9 and @xmath20 . _ plan of the paper . _ in section [ review ] we review the relationship between degenerations of symplectic manifolds and the symplectic sum operation ; this is then applied to symmetric products ( section [ symm ] ) , and employed to construct the lagrangian and hamiltonian isotopies of the theorem ( section [ isotopies ] ) . section [ non - iso ] proves the hamiltonian non - isotopy clause , and section [ heegaard ] covers the application to heegaard floer homology . the symplectic forms used up to that point tame the natural complex structure on @xmath25 but are not actually khler . it has already been mentioned that in section [ smoothing ] we give a different construction , by smoothing currents , which does produce khler forms and has some independent interest . the question of whether handleslides lead to hamiltonian - isotopic heegaard tori was posed to me by peter ozsvth . thanks to him , and to ron fintushel , robert lipshitz and ivan smith for productive conversations . i am grateful to selman akbulut and the other organisers of the 14th gkova geometry topology conference for a mathematically stimulating week in an idyllic setting . in this section we review the notion of symplectic sum and its relation to degenerations of symplectic manifolds @xcite . so as to simplify the definitions ( slightly ) we restrict the discussion to complex manifolds . if @xmath91 is a complex @xmath92-manifold , a _ nodal degeneration _ of @xmath91 is a pair @xmath93 where @xmath94 is a complex @xmath95 manifold and @xmath96 a holomorphic map which is a topological fibre bundle over @xmath97 , with fibre @xmath98 ( we write @xmath99 for @xmath100 ) . thus the critical locus @xmath101 is contained in the zero - fibre @xmath102 . we suppose that this locus is smooth and @xmath103-dimensional , and that the ( complex ) hessian on @xmath104 is everywhere non - degenerate . as a complex space , @xmath102 has a normal crossing along @xmath105 : locally near a point of @xmath105 , @xmath102 is equivalent to a neighbourhood of the origin in @xmath106 . let @xmath107 be the normalisation of @xmath102 . the normalisation is an intrinsic construction but , put simply , it is the complex manifold obtained by replacing each patch @xmath108 by @xmath109 . the preimage @xmath110 is the disjoint union of two divisors @xmath111 and @xmath112 which are identified with one another via @xmath92 . their normal bundles are dually paired with one another by means of the hessian @xmath113 on @xmath104 . choose closed tubular neighbourhoods @xmath114 and @xmath115 in @xmath116 for @xmath111 and @xmath112 respectively , and put @xmath117 , so that @xmath118 is an @xmath70-bundle over @xmath119 . there is an orientation - reversing diffeomorphism @xmath120 , which arises because because we can think of @xmath121 as the equator in the @xmath122-bundle @xmath123 , and @xmath124 as the equator in @xmath125 ; but using the hessian we have @xmath126 this discussion makes sense symplectically . if we are given a symplectic form @xmath127 on @xmath94 , taming the complex structure , the manifold @xmath116 inherits a symplectic form @xmath128 for which @xmath129 . we are thus in a position to form gompf s _ symplectic sum _ @xmath130 as in @xcite , the symplectic manifold obtained from @xmath131 by gluing @xmath121 to @xmath124 via @xmath132 . [ sum ] the fibre @xmath133 of a nodal degeneration @xmath93 , when equipped with the restriction of a khler form @xmath127 on @xmath94 , is symplectomorphic to the symplectic sum @xmath134 . we refer to @xcite for a complete treatment ; here we aim mainly to draw attention to an important aspect of the geometry of degenerations , their _ vanishing cycles_. the vanishing cycle @xmath135 is the set of points @xmath136 such that symplectic parallel transport @xmath137 over the ray @xmath138\subset \c$ ] lands in the critical set @xmath105 in the limit @xmath139 . the limiting parallel transport defines an @xmath70-bundle @xmath140 , and the crucial point for us will be that @xmath141 correspondingly , in the symplectic sum @xmath142 , let @xmath143 denote the common image of @xmath121 and @xmath124 . since it arises as the boundary of a tubular neighbourhood of @xmath105 , it comes equipped with an @xmath70-bundle @xmath144 . notice that @xmath145 is naturally identified symplectically with @xmath146 . on the other hand , symplectic parallel transport into @xmath102 defines a symplectomorphism @xmath147 . combining these observations we find a symplectomorphism @xmath148 . but both @xmath74 and @xmath143 are coisotropic submanifolds in symplectic manifolds ( in fact , they are @xmath70-bundles over a common symplectic manifold ) . there is a diffeomorphism @xmath149 covering the identity map on @xmath105 , and by the coisotropic neighbourhood theorem , this extends to a symplectomorphism between neighbourhoods of @xmath74 and @xmath143 . it remains to see that these can be made compatible with the identification @xmath148 . this requires closer inspection of the local model for the symplectic sum , and we shall not give the details here ( see @xcite ; also compare ( * ? ? ? * section 2 ) ) . let @xmath150 be a simple closed curve , and let @xmath151 be the surface obtained by excising a tubular neighbourhood of @xmath152 and gluing in a pair of discs . let @xmath41 and @xmath153 be points in the respective interiors of those discs . fix complex structures on @xmath1 and @xmath151 , and an integer @xmath154 . define two maps @xmath155 by @xmath156 , @xmath157 . denote by @xmath158 the complex blow - up of @xmath159 along the locus @xmath160 . let @xmath161 be the natural lifts of @xmath162 and @xmath163 . the images @xmath164 and @xmath165 are disjoint . let @xmath166 . there is a natural isomorphism @xmath167 ( see ( * ? ? ? * section 3 ) ) . thus @xmath168 identifies the divisors @xmath169 and @xmath170 , and under this identification , their normal bundles are dual . we may therefore form the symplectic sum @xmath171 ( as a smooth manifold ; we will say more about symplectic forms presently ) . there is a diffeomorphism @xmath172 canonical up to isotopy . the diffeomorphism - type of a symmetric product of a riemann surface does not depend on its complex structure . moreover , because the space of complex structures is simply connected , there are canonical diffeomorphisms between them ( up to isotopy ) . thus we can choose complex structures on @xmath1 and @xmath151 as we wish . to prove the proposition , we invoke proposition [ sum ] . what we are asserting is that there is a nodal degeneration of @xmath173 to a complex space whose normalisation is @xmath158 , in which @xmath169 and @xmath170 are the two preimages of the normal crossing divisor , and that the pairing of normal bundles is induced by the hessian along the critical locus of the degeneration . let @xmath174 be a nodal degeneration of @xmath1 , i.e. a holomorphic lefschetz fibration such that @xmath175 , with a single critical point lying over @xmath80 . we arrange that the vanishing cycle , taken along the vanishing path @xmath176 $ ] , is ( the isotopy class of ) @xmath152 . one can then form an associated family , @xmath177 , the relative hilbert scheme of @xmath92 points , as in donaldson smith @xcite . their crucial observation ( later re - proved by ran @xcite ) is that this space is globally smooth ; indeed , it is a nodal degeneration of its smooth fibre . there is a natural morphism @xmath178 to the relative symmetric product , whose restriction to @xmath97 is biholomorphic onto the restriction of @xmath179 to @xmath97 ; hence @xmath180 is a nodal degeneration of @xmath173 . the structure of the relative hilbert scheme was described in elementary terms in @xcite ; in particular , it was explained that the normalisation of the zero - fibre @xmath181 is precisely @xmath182 where @xmath183 is the normalisation of the zero - fibre of @xmath174 . the two distinguished divisors inside it are the divisors @xmath169 and @xmath170 described in the discussion above ( with @xmath41 and @xmath153 the points of @xmath184 lying over the node of @xmath102 ) , so @xmath181 is obtained by identifying @xmath169 with @xmath170 via @xmath168 . thus our result follows from the general theory of section [ review ] . a case to keep in mind is the symmetric square , @xmath73 . in this case , @xmath185 is the blow - up of @xmath186 at the point @xmath187 . the two divisors @xmath169 and @xmath170 are disjoint copies of @xmath151 . if @xmath152 is a separating curve , so that @xmath188 is a disjoint union @xmath189 , the assertion is that @xmath73 is the fibre sum @xmath190 here @xmath191 is the blow - up of @xmath192 at @xmath193 ; the proper transforms of the factors in the product give embeddings of @xmath194 and @xmath195 into the blow - up , each of self - intersection @xmath196 . on the other hand , @xmath197 contains a copy of @xmath194 of self - intersection @xmath198 , embedded by the map @xmath199 . likewise , @xmath200 contains a copy of @xmath195 of self - intersection @xmath198 , embedded by the map @xmath201 . the fibre sums @xmath202 and @xmath203 are taken along these two pairs of surfaces . ( note : the manifold appearing here is written as a sum of three different manifolds along two pairs of surfaces , whereas the proposition expresses it as a self - fibre sum of a disconnected manifold along a disconnected surface . ) this remark is due to r. fintushel . when @xmath195 has genus 1 , the description can be simplified because the second fibre sum has no topological effect . thus , if we write @xmath204 ( connected sum ) we have @xmath205 indeed , we know that @xmath206 . the summand @xmath207 is the total space of a non - trivial @xmath208-bundle over @xmath209 ( the bundle projection is the abel jacobi map ) . to perform the second fibre sum , we remove the tubular neighbourhood of a square @xmath196 torus in @xmath210 ( the proper transform of @xmath211 in the second summand ) . we glue in the complement of a tubular neighbourhood of a square @xmath198 section @xmath212 of @xmath213 . this complement is a tubular neighbourhood of another section @xmath214 , which must have square @xmath196 ( since the fact that @xmath215 \cdot [ s_0 ] = 0 $ ] forces @xmath215=[s_0]-[\mathrm{fibre}]$ ] ) . thus we glue into @xmath210 the same piece that we removed , without changing the gluing . we now describe the how the symplectic sum description affects the relevant cohomology classes , adding subscripts to the notation to track which surface we are considering . [ coh ] assume @xmath151 is connected . let @xmath216 be the pullback of @xmath217 from @xmath159 to its blow - up , minus @xmath47 times the class poincar dual to the exceptional divisor . under the diffeomorphism @xmath218 of the previous proposition , @xmath219 pulls back to @xmath220 . closely related results were obtained in @xcite , notably proposition 3.14 . the relevance of @xmath219 is that it will be the class of our symplectic form . see @xcite for accounts of blowing up symplectic or khler manifolds . we shall give a direct proof in a ` prototypical ' case , then explain how to obtain the general result from this . consider @xmath207 , thought of as @xmath221 ( symplectic sum along two square - zero @xmath222-spheres ) . one has @xmath223 , the generators being the classes @xmath224 of the exceptional divisor and @xmath225 of the proper transform of a ( generic ) line in @xmath226 . the fibre sum description corresponds to a nodal degeneration @xmath227 ; the homology cycles in @xmath228 project to cycles in the central fibre of @xmath229 , and ( by thinking of the symplectic sum construction ) one can see explicitly that these projected cycles are homologous in @xmath229 to cycles in @xmath207 . indeed , @xmath224 intersects each of the two square - zero @xmath230s in a point , and so deleting a neighbourhood of @xmath231 has the effect of removing two discs from @xmath224 ; in the symplectic sum , the two boundary circles of this 2-holed sphere are identified to form a torus @xmath232 in @xmath233 , of square @xmath196 , homologous in @xmath229 to the projection of @xmath224 . starting from @xmath225 one similarly gets a square @xmath198 torus in @xmath207 . note that both @xmath232 and @xmath234 can be thought of as sections of the abel jacobi fibration @xmath213 , @xmath235\mapsto x+y$ ] ( @xmath236 means the sum under the group law of an elliptic curve ) . one now sees that @xmath232 is dual to @xmath237 , @xmath234 to @xmath238 . thus in this case , the symplectic class @xmath239 does indeed correspond to @xmath240 ( but here @xmath241 ) . to obtain the general formula , it suffices to do so when @xmath242 . indeed , the construction of the degeneration of @xmath1 , and hence of its relative hilbert scheme , is natural under diffeomorphisms of @xmath1 which act trivially near @xmath152 . thus @xmath243 has to be invariant under the corresponding subgroup of the mapping class group . it must also restrict to the @xmath242 version of @xmath244 when restricted to @xmath245 ( where @xmath86 is a basepoint ) . bearing in mind that @xmath246 independent of @xmath247 , and that @xmath43 and @xmath36 restrict to @xmath245 in the obvious way , this shows that it is enough to prove the formula when @xmath242 . observe that we can embed a one - holed torus @xmath248 into @xmath1 as a neighbourhood of @xmath152 . the restriction of @xmath243 to @xmath249 is then independent of the topology of @xmath1 . from this observation and the case worked out above one obtains the @xmath242 result . when @xmath152 disconnects @xmath1 , the lemma is still true but rather trivial . in fact , in this case the weight of the blow - up makes no difference to the induced class @xmath219 on @xmath173 . these assertions are simple algebraic consequences of the naturality of the construction of @xmath185 from @xmath1 under diffeomorphisms of @xmath1 supported away from @xmath152 . we prove statements ( 1 ) and ( 3 ) of theorem [ main ] . the argument also contains one of our two proofs of proposition [ forms ] , though this one requires @xmath58 and does not produce khler forms . _ step 1a : reduction to genus 2 . _ if @xmath1 has genus @xmath250 , we can find a simple closed curve @xmath251 separating @xmath1 into a genus @xmath222 part containing @xmath9 , @xmath20 and @xmath21 , and a genus @xmath252 part containing @xmath253 . thus @xmath254 , where @xmath68 has genus 2 and contains the handleslide . fix closed discs @xmath255 and @xmath256 so that @xmath257 , and points @xmath258 , @xmath259 . the region of @xmath25 of interest to us is the open set @xmath260 . we claim that there are khler forms on @xmath25 which restrict to this region as the sum of forms pulled back from the two factors . once this claim is established , it will suffice to prove the theorem for a genus 2 surface . however , the reduction will not be perfect : the genus 2 surface has a puncture ( far from the handleslide region ) , and we have to use the symplectic forms produced in the course of proving the claim . for any @xmath261 , there are symplectic ( or indeed khler ) forms @xmath262 on @xmath263 , representing @xmath244 , such that @xmath264 . moreover , @xmath262 restricts to each connected component @xmath265 as the sum of two forms pulled back from the factors . note that @xmath53 is a khler class ( indeed , @xmath36 is ample and @xmath43 is the pullback by the abel jacobi map of an ample class on the jacobian ) . take a khler form @xmath266 on @xmath267 representing @xmath244 , and let @xmath268 be its restriction to @xmath269 , embedded by @xmath270 . similarly , take a khler form @xmath271 on @xmath272 representing @xmath244 , and let @xmath273 be its restriction to @xmath274 , embedded by @xmath275 . define @xmath262 , on the connected component @xmath276 , by @xmath277 . given such a form @xmath262 , pull it back to @xmath278 ( the blow - up of @xmath279 along @xmath280 ) and add a closed form supported near the exceptional divisor so as to obtain a symplectic form @xmath281 . the weight of the blow - up is unimportant , because of the remark following the proof of lemma [ coh ] . we may assume that @xmath282 , so that @xmath281 may be used in forming the symplectic sum @xmath283 . now , there is a diffeomorphism @xmath284 ( canonical up to isotopy ) , and thus @xmath285 is a symplectic structure on @xmath173 . in setting up @xmath286 we need to fix tubular neighbourhoods of @xmath287 of @xmath169 and @xmath288 of @xmath170 . it is clear that we can choose @xmath287 and @xmath288 to be disjoint from the open subset @xmath289 of @xmath290 , and hence we can ensure that @xmath286 restricts to this subset as the natural inclusion into @xmath25 . with this duly arranged , the symplectic form @xmath128 fulfils our claim . _ step 1b : completing the reduction to genus 2 . _ in this step we complete the reduction to genus @xmath222 by showing that we can choose _ any _ @xmath24-positive symplectic form representing the class @xmath244 on @xmath291 . as things stand , our form is constrained to be the restriction to @xmath292 of a @xmath24-positive symplectic form @xmath293 on @xmath267 . by an inductive argument , it will suffice to show that we can choose the khler form freely ( within our preferred cohomology class ) on the divisor @xmath294 . thus we have a @xmath24-positive symplectic form on a manifold @xmath295 and a codimension-2 @xmath24-holomorphic submanifold @xmath296 . the boundary of a tubular neighbourhood is a contact type hypersurface ( convex , as seen from the inside ) . but , by an application of gray s stability theorem left to the reader one can always isotope the symplectic form so that its restriction to @xmath297 is a prescribed @xmath24-positive symplectic form . this does the trick . _ step 2 : the ( punctured ) genus 2 case . _ now assume @xmath68 is closed and has genus 2 , and has a basepoint @xmath41 far from any of the @xmath29 . we have to prove the theorem for @xmath298 . write @xmath299 for the torus obtained by surgering out @xmath21 . the ` scar ' left by the surgery is a pair of discs in @xmath299 , containing points @xmath300 and @xmath225 in their interiors . in @xmath68 , the surgery is done in a region disjoint from the curves @xmath9 and @xmath20 , which therefore have images @xmath301 and @xmath302 in @xmath299 . at this point we should pin down the choice of @xmath24-positive symplectic form @xmath303 on @xmath291 . we choose it to be a symplectic structure arising as the pullback via a diffeomorphism @xmath304 of a khler form on @xmath305 . by lemma [ coh ] , we can assume that it represents the cohomology class @xmath244 . more specifically , we take a khler form on @xmath306 representing @xmath244 , and form its ordinary blow up in such a way that the exceptional divisor has weight @xmath47 . we have to make sure that the induced area forms on @xmath307 and @xmath308 agree with one another . for this , notice that , by the symplectic neighbourhood theorem , @xmath307 ( say ) has a neighbourhood of form @xmath309 , with a product symplectic form @xmath310 . think of the projection @xmath311 as a symplectic fibre bundle . by thurston s patching argument ( * chapter 6 ) , we can replace the original symplectic form by a new closed 2-form @xmath312 , symplectic on the fibres of @xmath41 , agreeing with the @xmath313 over the annulus @xmath314 , such that @xmath315 is a freely chosen positive area form @xmath316 . for suitable 2-forms @xmath317 on @xmath318 , supported in @xmath319 , the form @xmath320 , will then be symplectic . thus , replacing @xmath313 by @xmath320 near @xmath307 , we can adjust arrange that the forms on @xmath307 and @xmath308 agree . we also need to be more precise about the symplectic sum and the diffeomorphism @xmath218 . to do this , we need to construct suitable tubular neighbourhoods of @xmath307 and @xmath308 . given a compact subset @xmath321 , there is a natural framing of the restricted normal bundle @xmath322 ( more precisely , an identification with the trivial bundle with fibre @xmath323 ) . thus we can construct a symplectically trivial tubular neighbourhood of @xmath307 over the subset @xmath324 . similarly for @xmath308 . the symplectic sum description provides a hypersurface @xmath325the vanishing cycle of the degeneration whose isotropic foliation is by circles , and whose space of isotropic leaves is identified with @xmath299 . is naturally identified with the singular locus in the central fibre of the degeneration . the quotient map to the leaf - space corresponds to the limiting parallel transport map @xmath72 from proposition [ sum ] . ] thus there is an @xmath70-bundle @xmath326 ( topologically trivial , in fact ) such that @xmath327 for an area form @xmath328 on @xmath299 . we want to make sure that @xmath329 , and moreover , that @xmath330 when @xmath331 , @xmath332 . for this we have to set up the diffeomorphism @xmath218 correctly . we have already set up tubular neighbourhoods of @xmath333 and @xmath334 ; points @xmath335 in the boundary of the tubular neighbourhood ( with @xmath336 ) should map under @xmath218 to @xmath337 . this is a matter of smooth rather than symplectic topology ; it can be arranged by the method of ( * ? ? ? * lemma 3.16 ) , for instance . note furthermore that , because of the way we have set up the tubular neighbourhoods of @xmath307 and @xmath308 , we end up with a symplectic form @xmath303 on @xmath73 which is product - like in @xmath338 . our symplectic forms on @xmath25 are not necessarily khler ; the symplectic sum operation does not in general preserve the khler category . however , it is easy to see that they can be taken to tame the natural complex structure @xmath339 . _ step 3 : lagrangian isotopy . _ take an isotopy @xmath340}$ ] of circles in @xmath299 , from @xmath301 to @xmath341 . it is , of course , a lagrangian isotopy with respect to @xmath328 . the preimages @xmath342 are tori in @xmath291 . they are lagrangian , because if @xmath343 and @xmath344 are tangent to @xmath342 at a point @xmath345 then @xmath346 . moreover , we set things up in step 2 so that @xmath347 for @xmath348 . thus we have a lagrangian isotopy from @xmath349 to @xmath76 . notice that this isotopy stays inside a compact subset of @xmath350 , so it is again compatible with step 1 . this completes the proof of clauses ( 1 ) and ( 3 ) of theorem [ main ] . _ step 4 : hamiltonian isotopy . _ we begin the proof of statement ( 2 ) of theorem [ main ] . the difference between lagrangian and hamiltonian isotopy is measured by the _ flux _ @xcite . given a lagrangian isotopy @xmath351 , its ( infinitesimal ) flux at time @xmath352 is a class @xmath353 which can be obtained by extending the vector field generating the isotopy to a globally - defined symplectic vector field , dualising this to get a closed 1-form , and taking its cohomology class restricted to @xmath354 . for the isotopy in step 3 , one checks that the flux @xmath355 is equal to @xmath356 , where @xmath357 is the flux of the isotopy @xmath358 in @xmath299 with respect to @xmath328 . thus our lagrangian isotopy in @xmath291 is hamiltonian if and only if @xmath358 is a hamiltonian isotopy in @xmath299 . now , the complement @xmath359 has three components , of which two are homeomorphic to the open disc . the only obstruction to making @xmath358 hamiltonian is that the two disc - components have equal @xmath328-area . we have now proved the existence of hamiltonian isotopies under the condition that two discs have the same area with respect to the area form @xmath328 on the surgered surface . however , the stated criterion concerned areas with respect to the area form @xmath48 on @xmath1 , and we need to relate these conditions . in stating the theorem , we noted that the existence of a hamiltonian isotopy depends only on @xmath47 and @xmath48 ( and , _ a priori _ , on @xmath24 ) . the problem has an extra symmetry , however : we can apply self - diffeomorphisms to @xmath1 , pulling back @xmath24 , provided that they act trivially on @xmath360 . the only invariants of area forms @xmath48 under such diffeomorphisms are the areas of @xmath22 , @xmath23 and @xmath361 . given area - forms @xmath362 and @xmath363 , we can certainly find a self - diffeomorphism @xmath364 , trivial along @xmath365 , such that @xmath366 in a neighbourhood of @xmath251 . now replace @xmath362 by @xmath367 and apply moser s deformation argument to the path @xmath368 . we thus see that the existence of hamiltonian isotopies depends on @xmath48 only through the areas of @xmath22 and @xmath23 ( the area of their complement is clearly irrelevant ) . our construction can be interpreted , then , as saying that there is a function @xmath369 such that a hamiltonian isotopy exists if @xmath370 . in section [ non - iso ] we shall show that a hamiltonian isotopy can _ only _ exist if @xmath371 , which must therefore coincide with @xmath372 . in this section we complete the proof of theorem [ main ] by showing that the tori @xmath51 and @xmath52 can only be hamiltonian - isotopic when @xmath60 is related in a precise way to @xmath61 . we do this by showing that if the area constraint fails there is a discrepancy between the ranks of the lagrangian floer homology groups @xmath373 and @xmath374 . these would be isomorphic if a hamiltonian isotopy existed . we work over the universal novikov ring @xmath375 of the field @xmath376 , i.e. , the ring of formal ` series ' @xmath377 where @xmath378 is a function such that @xmath379 $ ] is finite for all @xmath380 . this coefficient ring is used to record the areas of holomorphic discs . the well - definedness and invariance of floer homology for monotone lagrangians of minimal maslov index 2 is not automatic , but in the case of heegaard tori it follows from a cancellation theorem from @xcite ( theorem 3.15 ) alongside the transversality argument outlined in @xcite . the self - floer homology of @xmath51 is as large as it conceivably could be : @xmath381 for any form @xmath49 as in proposition [ forms ] . the relevant moduli spaces are computed by ozsvth szab in ( * ? ? ? * lemma 9.1 ) . the discs contributing to the floer - theoretic differential come in pairs @xmath382 , with @xmath383 and @xmath384 elements of the same moduli space . these come from discs in @xmath1 itself , of equal @xmath48-areas and disjoint from the diagonal @xmath385 . thus they have equal @xmath386-area when @xmath386 is a form as in proposition [ forms ] representing @xmath36 . when @xmath387 the lemma therefore follows from ozsvth szab s calculation . now perturb @xmath386 to @xmath49 with @xmath388=\eta+\lambda\theta$ ] by adding a small multiple of a closed @xmath54-form supported near @xmath55 and representing the dual of @xmath389 $ ] . this perturbation does not change the areas , and so @xmath383 still cancels @xmath384 and the argument goes through . the floer homology of @xmath79 is usually smaller : either @xmath390 , or @xmath391 again , the proof rests on a calculation from @xcite ( lemma 9.4 ) . we set things up as in ( * ? ? ? * figure 9 ) , intersecting @xmath75 with @xmath392 , where @xmath393 is a small , transverse , hamiltonian perturbation of @xmath21 such that @xmath394 . let @xmath395 and @xmath396 . floer s complex is then a tensor product @xmath397 , where @xmath398 ( the ` uninteresting ' part ) corresponds to intersections among the curves @xmath399 and their @xmath48-hamiltonian translates , while @xmath400 has four generators : @xmath401 , @xmath402 , @xmath403 and @xmath404 . we set up the labelling so that the only potentially non - zero matrix entries in the differential @xmath405 on @xmath400 are @xmath406 and @xmath407 . now , writing elements of @xmath375 as sums @xmath408 with @xmath409 , ozsvth szab s calculation shows that @xmath410 where @xmath411 and @xmath412 are two particular discs and the integrals are with respect to @xmath49 . the first disc , @xmath411 , is the product of the disc @xmath23 shown in figure 1 and a constant disc at @xmath343 . it does not intersect the diagonal @xmath55 , and so ( using perturbed khler forms @xmath49 as in the proof of the last lemma ) its area is @xmath413 , regardless of @xmath47 . the second disc , @xmath412 , arises from a branched double covering of a disc by the annulus @xmath22 . by riemann hurwitz , its intersection number with @xmath55 is 2 . hence , using the formula @xmath389 = \eta - \theta$ ] ( which was mentioned following proposition [ forms ] ) , we have @xmath414 the homology of @xmath400 is zero unless the areas of @xmath411 and @xmath412 are equal , that is , unless @xmath415 . but @xmath416 has rank @xmath417 ( as in the proof of the previous lemma ) , so the result follows . the ` only if ' clause in statement ( 2 ) of theorem [ main ] follows immediately . the ` if ' clause also follows , by the argument explained in the last paragraph of section [ isotopies ] . we now turn to the proof of corollary [ heegcor ] , freely drawing on the language of heegaard floer homology developed by ozsvth szab in @xcite . before we begin the proof , we make some remarks on the relation of heegaard floer theory to lagrangian floer homology . it is a feature of lagrangian floer homology that the moduli spaces defining the differential depend on the lagrangians and the almost complex structure but do not directly involve a symplectic form . the primary role of symplectic forms is to fix the area of the pseudo - holomorphic ` whitney discs ' in a given homotopy class . having an _ a priori _ bound on area is essential for compactness , and for this reason it is important to work with lagrangian ( not merely totally real ) submanifolds . lacking a convenient symplectic form , ozsvth and szab bound the areas of whitney discs by treating @xmath25 as a ` symplectic orbifold'the quotient of @xmath418 , with its product symplectic form , by the action of the symmetric group and estimating areas ` upstairs ' . the symplectic class plays a second role in floer theory : the areas of the holomorphic discs are typically recorded in a novikov ring of coefficients , just as in section [ non - iso ] . this is useful because there may be infinitely many homotopy classes of whitney discs , but there are only finitely many index 1 whitney discs with a given area . ozsvth szab are able to do without novikov rings by showing that , for a fixed @xmath82-structure @xmath83 , and under the assumption of ` weak admissibility ' of the pointed heegaard diagram @xmath419 with respect to @xmath83 , there are only finitely many homotopy classes of whitney discs , between intersection points @xmath420 with @xmath421 , with a given maslov index , fixed intersection number with the divisor @xmath422 , and additionally satisfying a positivity constraint automatically satisfied by pseudo - holomorphic discs ( * ? ? ? * lemma 4.14 ) . now suppose @xmath49 is as in proposition [ forms ] . lagrangian floer homology @xmath423 , in any of its possible algebraic variants , can be computed via any almost complex structure taming @xmath49 that satisfies a regularity condition . ozsvth and szab show that to attain regularity it is sufficient to consider ( paths in ) a particular class of almost complex structures , those that are small ` nearly symmetric ' perturbations of the standard integrable structure ( * ? ? ? * theorem 3.4 ) ( these still tame @xmath49 ) . the periods are controlled by the maslov index and intersection numbers with @xmath424 , and the groups are set up so as to keep track of these intersection numbers . heegaard tori are monotone lagrangians of minimal maslov index 2 . the troublesome feature of lagrangian floer theory in such a case is that bubbling - off of discs spoils compactness of the moduli spaces involved in the definition of the groups ( more precisely , the matrix coefficients @xmath425 where @xmath136 is a generator for the floer complex and @xmath405 the usual floer - theoretic ` differential ' ) , cf . ozsvth and szab show that , though maslov index 2 discs can bubble off in the relevant 1-dimensional moduli space , they always do so in cancelling pairs ( * ? ? ? * theorem 3.15 ) , which , as explained in @xcite , implies that @xmath426 . we shall set up a ` continuation isomorphism ' @xmath427 and a similar one for the other variants of heegaard floer homology . by hypothesis @xmath428 is transverse to @xmath429 , and perturbing the @xmath84 curves slightly we may assume that @xmath428 is transverse to @xmath429 . choose any small @xmath58 and an area form @xmath48 satisfying the area constraint of theorem [ main ] ( 2 ) . we then consider a symplectic form @xmath49 as in proposition [ forms ] and use theorem [ main ] to obtain a hamiltonian isotopy @xmath430 } $ ] with @xmath431 and @xmath432 . by a general principle in lagrangian floer homology , the hamiltonian isotopy @xmath433}$ ] induces a continuation isomorphism @xmath434 ( floer homology with novikov - ring coefficients ) . because the minimal maslov index is 2 , the continuation - map moduli spaces can not bubble here ( and the same goes for those used in proving the composition law for continuation maps ) . however , it is not immediately apparent that the continuation map makes sense for @xmath435 ( say ) , because the @xmath49-area of whitney discs might not be controlled by maslov index and intersection number with @xmath424 . we deal with this issue now . the continuation map can be understood as follows . we have a hamiltonian isotopy @xmath436}$ ] , and we may extend this to a family @xmath437 where @xmath438 for @xmath439 and @xmath440 for @xmath441 ( this family will not be smooth , so strictly we should reparametrise @xmath176 $ ] before we extend so as to obtain a smooth family ) . we consider the trivial @xmath25-bundle @xmath442 \to \r\times [ 0,1]$ ] , and make it into a hamiltonian fibration by giving endowing it with a closed 2-form @xmath443 . here @xmath225 is the @xmath444-coordinate , @xmath352 the @xmath176$]-coordinate ; @xmath445 is a family of functions , non - zero only for @xmath446 $ ] , generating @xmath447 ; and @xmath448 is a cut - off function , equal to @xmath449 near @xmath450 and to @xmath80 near @xmath451 . there are lagrangian boundary conditions @xmath452 and @xmath453 over the boundary of the strip . the continuation map is defined by counting index-0 pseudo - holomorphic sections of this fibration subject to the lagrangian boundary conditions ( for an almost complex structure making the projection holomorphic , and translation - invariant for @xmath454 ) . see @xcite for details of the fibre - bundle approach to floer theory . we need to estimate the energies of sections @xmath343 subject to the lagrangian boundary conditions and asymptotic to intersection points @xmath455 as @xmath456 and @xmath457 as @xmath458 . these energies are cohomological in nature ; they are invariant under compactly supported homotopies of @xmath343 , for instance . because of this , we can estimate the energies by considering a degeneration of the domain in which a disc is pinched off , as in figure 2 . that is , we consider a 1-parameter family of surfaces @xmath459 , with @xmath460 $ ] , pulling back our hamiltonian fibration and lagrangian boundary conditions by a family of diffeomorphisms @xmath461 . we arrange that the surfaces @xmath462 limit to the nodal surface @xmath463 shown in figure 2 as @xmath464 . conformally , we can view @xmath463 as the nodal union of a triangle @xmath299 and a disc @xmath23 . we can arrange that there is a limiting hamiltonian fibration over @xmath463 ( still smoothly trivialised ) , and that over the three edges of the triangle , we have constant lagrangian boundary conditions @xmath465 , @xmath466 and @xmath467 , as shown . over the boundary of the disc we have the lagrangian boundary condition @xmath468 , where @xmath225 now parametrises the boundary anticlockwise . consider triangles @xmath469 , subject to the boundary conditions given by @xmath465 , @xmath470 and @xmath467 , of maslov index @xmath80 , and with fixed intersection number with @xmath424 . assume that @xmath299 is asymptotic to intersection points @xmath471 , as above , whose associated @xmath82-structure is @xmath83 . according to ( * ? ? ? * section 9.2 ) , the number of finite - energy holomorphic triangles @xmath299 satisfying these conditions is finite . on the other hand , homotopy classes of sections over @xmath23 , subject to the boundary conditions @xmath468 and mapping the node to a fixed @xmath472 , correspond to @xmath473 ( by extending them to sections of a trivial fibration over a larger disc , with constant lagrangian boundary condition @xmath465 ) , and so again the energy is controlled by the intersection number with @xmath424 . each homotopy class of sections over @xmath463 can be smoothed out to a homotopy class of sections of @xmath462 for @xmath352 finite , and it is easy to see that every homotopy class of sections over @xmath474 arises this way . the smoothing leaves the energy unchanged . hence , going back to the continuation map picture ( over @xmath474 ) we conclude that the energy of index-0 pseudo - holomorphic strips with fixed asymptotics and fixed intersection number with @xmath424 is bounded . thus the continuation map is well - defined . more accurately , the continuation map is defined initially on @xmath475 , but it respects the subcomplexes @xmath476 and so induces maps on @xmath476 and @xmath477 . for @xmath478 one defines the continuation map using strips whose intersection number with @xmath424 is zero . by using the same degeneration to obtain the necessary energy bounds , one sees that these continuation maps are chain maps , and that the continuation map associated with the reverse isotopy @xmath479 is an inverse up to chain - homotopy . hence all the continuation maps are quasi - isomorphisms . note finally that , because theorem [ main ] produces hamiltonian isotopies which are canonical ( up to homotopy with fixed endpoints ) , the continuation isomorphisms on heegaard floer homology are also canonical . @xmath467 at 280 605 @xmath465 at 200 515 @xmath466 at 370 515 @xmath468 at 215 450 @xmath299 at 280 560 @xmath23 at 280 460 the continuation isomorphisms on heegaard floer homology which we have associated with a handleslide are the same as the isomorphisms constructed by ozsvth szab using holomorphic triangles . to see this one applies a gluing theorem for holomorphic sections to the degeneration of figure 2 . this final section reproduces the note @xcite , which has not previously been published . we consider branched coverings @xmath480 of complex manifolds that is , holomorphic maps which are proper , surjective , and finite . the branch locus @xmath481 of such a map is @xmath482 a @xmath483 khler form @xmath128 on @xmath484 can be pushed forward in the sense of currents , that is , of 2-forms with coefficients of class @xmath485to a closed current @xmath486 on @xmath487 which is smooth on @xmath488 . the following result is essentially due to varouchas @xcite . [ var ] let @xmath489 be a branched covering of complex manifolds , and @xmath128 a khler form on @xmath484 . let @xmath490 be a neighbourhood of the branch locus in @xmath487 . then there exists a khler form @xmath491 on @xmath487 , representing the class @xmath492=\pi_[\omega]\in h^2(x';\r)$ ] , such that @xmath493 . [ smoothing cor ] let @xmath1 be a riemann surface with positive area form @xmath48 . let @xmath494 be the projection map . suppose that @xmath495 is an open subset containing the diagonal @xmath55 . then there exists a khler form @xmath128 on @xmath173 , representing the class @xmath496 $ ] , such that outside @xmath497 , @xmath128 is the smooth push - forward @xmath498 of the product form . let @xmath484 be a complex manifold . a khler cocycle * on @xmath484 is a collection @xmath499 , where @xmath500 is an open cover of @xmath484 , and @xmath501 is a a function , such that for all @xmath502 , one ascribes to the cocycle a property ( continuity , smoothness , etc . ) possessed by all the @xmath503 . khler cocycles are , by definition , upper semicontinuous . condition ( 1 ) means that the @xmath222-current @xmath507 is strictly positive on @xmath504 ; ( 2 ) means that these currents agree on overlaps , and are therefore restrictions of a @xmath222-current @xmath128 on @xmath484 ( closed and strictly positive ) . if the cocycle is @xmath483 then @xmath128 will be a khler form . let @xmath508 be open subsets of @xmath509 with @xmath510 let @xmath511 be continuous , strictly plurisubharmonic , and smooth on @xmath127 . then there exists a function @xmath512 , again continuous and strictly plurisubharmonic , equal to @xmath218 on @xmath513 and smooth on @xmath514 . [ global smoothing ] let @xmath499 be a continuous khler cocycle on the complex manifold @xmath484 . suppose that @xmath517 , with @xmath518 and @xmath519 open , and that the functions @xmath520 are smooth . then there exists a continuous function @xmath521 supported in @xmath522 and a locally finite refinement @xmath523 so that the family @xmath524 is a smooth khler cocycle . refine the cover @xmath525 to a countable , locally finite cover @xmath526 with the property that @xmath527 for definiteness suppose both @xmath528 and @xmath529 are infinite ; say @xmath530 , so that the labels @xmath531 are pairs @xmath532 with @xmath533 for @xmath528 , or @xmath534 for @xmath529 . find open subsets @xmath535 such that @xmath536 still covers @xmath484 . set @xmath537 the sets @xmath538 exhaust @xmath539 . let @xmath540 be the khler cocycle induced from @xmath499 by the refinement . we prove the claim by induction on @xmath92 . apply the ` lemme principal ' to @xmath549 and to the function @xmath550 , obtaining a new function @xmath551 ; let @xmath552 , extended by zero to all of @xmath484 , and use ( 2 ) to define the new cocycle . we have to verify ( 1 ) , i.e. to prove smoothness of @xmath544 at each @xmath553 . if @xmath554 then @xmath555 , but @xmath556 was already smooth . if @xmath557 then , near @xmath136 , @xmath558 , which is the sum of a pluriharmonic function and a smooth plurisubharmonic one . but a pluriharmonic function is automatically smooth . by a similar argument , @xmath544 is strictly plurisubharmonic . now define a function @xmath559 by @xmath560 ( the sum is locally finite ) . then @xmath561 defines a khler cocycle . it is smooth , since on one hand , on @xmath562 , the original cocycle was smooth and has not been modified , while on the other hand , @xmath563 so smoothness on @xmath564 is guaranteed by ( 1 ) . hence @xmath565 has the required properties . each fibre @xmath566 , being finite , has a neighbourhood which is a disjoint union of open balls . hence , using the @xmath567-lemma , one can find a smooth khler cocycle @xmath568 on @xmath484 such that each @xmath504 contains a fibre of @xmath569 , with @xmath570 . one can then find a locally finite cover @xmath571 of @xmath487 such that @xmath572 . a general property of branched covers is that the push - forward @xmath573 of a continuous function @xmath574 is again continuous ( it is given by @xmath575 , where the points @xmath136 are taken with multiplicities ) . the family @xmath576 on @xmath487 is thus a continuous khler cocycle : plurisubharmonicity is clear away from @xmath577 , since @xmath578 , hence everywhere by density ; similarly for pluriharmonicity on overlaps . as for strictness , let @xmath579 be the standard khler form in a complex chart centred at a point @xmath580 . then , for a test @xmath581-form @xmath328 supported near @xmath136 , one has @xmath582 , and using this one verifies strict positivity of @xmath583 . now let @xmath584 be a smaller open neighbourhood of @xmath577 . apply the ` global smoothing ' lemma ( [ global smoothing ] ) to @xmath576 on @xmath487 , taking @xmath585 and @xmath586 . the output is a function @xmath587 , as well as a refinement of @xmath588 , such that @xmath589 is a well - defined @xmath222-form with the right properties . for any smooth , closed test form @xmath590 one has @xmath591 , hence @xmath592 represents the zero cohomology class .	a @xmath0-tuple of disjoint , linearly independent circles in a riemann surface @xmath1 of genus @xmath0 determines a ` heegaard torus ' in its @xmath0-fold symmetric product . changing the circles by a handleslide produces a new torus . it is proved that , for symplectic forms with certain properties , these two tori are hamiltonian - isotopic lagrangian submanifolds . this provides an alternative route to the handleslide - invariance of ozsvth szab s heegaard floer homology .
in this article we study of the statistical behavior of two non - equilibrium systems . the first one is a quasi one - dimensional gas introduced in ref . @xcite . there , the authors studied the biased diffusion of two species in a fully periodic @xmath0 rectangular lattice half filled with two equal number of two types of particles ( labeled by their charge @xmath1 or @xmath2 ) . an infinite external field drives the two species in opposite directions along the @xmath3 axis ( long axis ) . the only interaction between particles is an excluded volume constraint , i.e. , each lattice site can be occupied at most by only one particle . the second system is a one - dimensional spin system introduced in ref . @xcite , where the authors consider a chain of @xmath4 ising spins with nearest neighbor ferromagnetic interaction @xmath5 . the chain is subject to spin - exchange dynamics with a driving force @xmath6 that favors motion of up spins to the right over motion to the left . both systems evolve with formation of domains which grow in time . for intermediate times where the size of the domains is much smaller than the total size @xmath4 of the system , the domain size distribution exhibit a dynamic scaling . in this work we will be interested in the statistical properties of these domains in the scaling regime . for each system there are two kind of domains . for the gas system , there are domains filled with particles and empty domains . for the spin system there are domains of up spins and domains of down spins . we are interested in the statistical properties of these domains regardless of their type ( filled or empty , up or down ) . let us define @xmath7 as the probability density function that the distance between the external borders of @xmath8 consecutive domains is @xmath9 at time @xmath10 . let @xmath11 be the average of @xmath9 and the relative spacing between borders is @xmath12 . the scaling probability density function is defined as @xmath13 in particular @xmath14 is the scaling nearest neighbor edges distribution , i.e. the domain size distribution . in the scaling regime , the scaling probability density functions @xmath15 do not depend on the time @xmath10 . numerical simulation of both systems performed in @xcite show that their scaling domain size distribution function is well fitted by @xmath16 this distribution is known in the random matrices theory as the wigner surmise and it describes the spacing distribution between eigenvalues in the gaussian and circular orthogonal ensembles ( goe and coe respectively ) @xcite . this fact suggest a possible connection between the statistical behavior of the borders of domains in non - equilibrium systems and the eigenvalues in random matrices in a similar way as it occurs with the eigenvalues of the gaussian orthogonal / unitary ensemble and the vicious random walk @xcite . the main objective of this paper is to explore this possible connection between non - equilibrium domain systems and random matrices , in particular the circular orthogonal ensemble . in sections [ sec : gas ] and [ sec : spin ] , we summarize the main results found in ref . @xcite on the quasi one - dimensional gas , and in ref . @xcite on the spin system . in section [ sec : rand - mat ] , we recall some facts about the random matrix theory , relevant for our purposes . in section [ sec : res ] , we compare the statistical behavior of the domains for each system with the statistical behavior of the eigenvalues of random matrices from the circular orthogonal ensemble . also , we will consider a simple model of independent intervals , which turns out to better describe the statistics of the domains in both the gas and the spin systems . this system was described briefly in the introduction , more details are found in ref . the system evolves in time according to the following dynamical rules : 1 . @xmath4 particles are randomly inserted in a @xmath0 rectangular lattice , @xmath17 particles @xmath18 and @xmath17 particles @xmath19 , the remaining sites are empty . periodic boundary conditions are imposed in both directions of the lattice . let the @xmath3 axis be the long axis of length @xmath4 . two neighbor sites are chosen at random . the contents of the sites are exchanged with probability @xmath20 if the neighbors sites are particle - hole , but if they are particle - particle the content are exchanged with probability @xmath21 . the exchanges which result in @xmath22 particles moving in the positive / negative @xmath3 direction are forbidden due to the action of the external field . a time unit correspond to @xmath23 attempts of exchange . the system was studied by computer simulation in refs . we also performed several simulations of this system with a parameter @xmath24 . the following qualitative results are found from the simulations : * for low @xmath21 values the system remains homogeneous , i.e. , the system evolves without domain formation . * for high @xmath21 values the system evolves as follows . for early times small domains form everywhere due to mutual obstruction of the opposite species . after a long time , the system settles into a non - equilibrium steady state ( ness ) in which only one macroscopic domain survives , so that , its length is about @xmath17 and it contains almost no holes . in the low density region the traveler particles leak out of the domain at one end and later rejoin it at the other end . the charge distribution of this macroscopic domain is not trivial . * for intermediate times , when the average size of the clusters is much smaller than the size @xmath4 of the system , the system shows dynamical scaling . following ref . @xcite , we can map the quasi one - dimensional lattice into a one - dimensional lattice . this approximation described in ref . @xcite is a coarse grained description . for any configuration on the @xmath0 lattice we construct an effective one - dimensional one , with occupation numbers zero or one on a @xmath4 sites line , as follows . at each site @xmath25 , we assign @xmath26 if there are @xmath27 or less particles in the @xmath28 sites around it , including the @xmath25-th column of the original lattice . we assign 1 otherwise . in the coarse grained configuration a domain is a simple consecutive sequence of ones and its size is just the length of this string . under this coarse grained description , the authors of ref . @xcite considered the `` residence distribution '' of the clusters @xmath29 , defined as @xmath30 in terms of the domain size distribution @xmath31 . using the residence distribution , they found @xcite that the average length of domains @xmath32 grows in time with an exponent of at least @xmath33 . the numerical results from the simulations of ref . @xcite , show that the scaling residence distribution @xmath34 is well fitted by @xmath35 remembering that @xmath36 is , by definition , proportional to @xmath37 , this implies that the scaling domain size distribution @xmath38 is properly fitted by the wigner surmise for the orthogonal ensembles of random matrices , eq . ( [ dgoe ] ) . we performed our own simulations to confirm this , the results will be shown in section [ sec : res ] . in ref . @xcite , the authors proposed an approximate model to describe this system , in the coarse grained description , using the coalescing random walk ( crw ) where the particles execute independent random walks , suffering a fusion reaction @xmath39 when two particles meet . this approximation is useful because the full hierarchy of correlations for this diffusion - limited reaction has been solved , see ref . this approximation is done under the following assumptions . * the numbers of domains is equal to the number of walkers in the crw . * the walkers describe random walks over a one dimensional ring with @xmath40 sites , in that way , @xmath40 its equal to the sum of the length of the domains . * the distances between walkers are equal to the length of the domains . * when two walkers meet they coalesce , representing the disappearance of a domain . * the traveling particles of the original model are neglected in the crw description . under this approximation the spacing distribution is the same found as in the coarse grained description , but the average length of domains grow in time with an erroneous exponent of @xmath41 . in order to solve this problem , in ref . @xcite , the authors introduced the model of the interacting random walkers ( irw ) , where neighboring walkers experience stronger attraction with decreasing separation . the probability of a walker moving to the right is @xmath42 where @xmath43 is the distance to its right nearest neighbor and @xmath44 is the probability of staying . in the same way , the probability of moving to the left is @xmath45 where @xmath46 is the distance to its left nearest neighbor and @xmath47 . the factor @xmath48 is chosen in ref . @xcite in order to give the best fit to the simulation data and the dependence with @xmath49 is justified in ref . @xcite . in this model , @xmath50 grows with an exponent of @xmath33 , see ref . the spacing distribution is still given by ( [ dgoe ] ) . we performed a few simulations of this interacting random walk changing the constant 0.06 in eqs . ( [ eq : qr ] ) and ( [ eq : ql ] ) . this has the effect of changing the growth exponent of the domains , however the scaling domain size distribution @xmath38 remains unchanged , it is still given by the wigner surmise ( [ dgoe ] ) . in the introduction we briefly presented this system , for more details see ref . @xcite . the lattice has a length @xmath40 with @xmath51 spins up ( `` @xmath1 '' ) and @xmath52 spins down ( `` @xmath2 '' ) with @xmath53 . periodic boundary conditions are imposed . the spin - flip events are : 1 . @xmath54 @xmath55 @xmath56 @xmath57 . 2 . @xmath58 @xmath55 @xmath59 @xmath60 . @xmath61 @xmath55 @xmath62 @xmath63 . 4 . @xmath64 @xmath55 @xmath65 @xmath63 . where the transition probability rate for a process from left to right is proportional to @xmath66 . the constant @xmath5 is the nearest neighbor coupling between spins , @xmath6 is the energy associated to an external field which drives the up ( `` @xmath1 '' ) spins to the right and the down ( `` @xmath2 '' ) spins to the left , and @xmath67 the thermal energy ( temperature times boltzmann constant ) . as explained in @xcite , the microscopic dynamics of the lattice of spins may be mapped onto one for an array domain dynamics , which provides a good approximation in the regime @xmath68 . in this approximation , domains of up spins move spontaneously to the right , and domains of down spins move to the left . the rates for such processes are independent of the domain size . the algorithm used for the numerical simulation is the following : 1 . set up a random array of alternating down and up spins , with @xmath69 spins up and @xmath70 spins down . 2 . choose a domain at random . 3 . if the domain is down , move it to the left ( i.e. , reduce the size of its left neighbor by one and increase its right neighbor by one ) , otherwise move it to the right . if one of the neighbor domains is zero size , then the domain is removed . update the clock by 1/number of domains . repeat steps 26 . this simplified system allows much better statistics than it would be possible with the true microscopic system @xcite , and its simulation is more easy to implement and it reproduces the same asymptotic behavior . from the simulation , the following qualitative results are observed : * for early times , little domains form everywhere . * at intermediate times , some domains disappear while other domains grow . in this time regime , the system shows dynamical scaling . * for later times , only two macroscopic domains remain , which move in opposite directions . the system falls in a non - equilibrium steady state . in ref . @xcite , the authors derive an analytical solution in the case @xmath71 , nevertheless we can derive another exact solution for the dynamical domain model with @xmath72 . let @xmath73 be the frequency of finding a domain of spin up with length @xmath74 and another domain of spin down with length @xmath75 at time @xmath10 . the master equation for @xmath73 is @xmath76 let @xmath77 be the distance between nearest neighbor sites of the lattice . in the continuous limit , @xmath78 , the above probability is a function of @xmath79 , @xmath80 . it satisfies the master equation @xmath81 if at @xmath82 all domains have length @xmath83 , the above equation is subject to the initial condition @xmath84 and the boundary conditions @xmath85 and @xmath86 . the appropriate scaling solution for this differential equation is @xmath87 as a consequence , the length of domains of spin up and of domains of spin down have the same probability distribution function found in the quasi one - dimensional gas . the above equation can be written as the product of two distributions with the form of the wigner surmise ( [ dgoe ] ) . the fact that the domain size distribution @xmath38 for a domain of up or down spins is well described by the wigner surmise ( [ dgoe ] ) was also confirmed in ref . @xcite from the numerical simulation results . historically , the initial motivation to introduce random matrices was to study the statistics of the energy levels of some quantum systems , in nuclear physics initially , then for quantum systems when their classical counterpart is chaotic , for a review see ref . @xcite . some local properties , such as the spacing between consecutive high energy levels are well described by the random matrix theory . for quantum chaotic systems with time - reversal invariance and integral total angular momentum , the high energy levels , for a given value of the spin and the parity , have spacing distributions and correlations that are well described by the gaussian orthogonal ensemble ( goe ) of random matrices , when the energy levels are rescaled such that the average interspacing is unity . for large matrices , the ( properly rescaled ) statistics of the eigenvalues of the goe in the bulk is the same as the one for the circular orthogonal ensemble ( coe ) . in this work we preferred to work with the circular orthogonal ensemble rather than the gaussian orthogonal ensemble since the former one is periodic and that way we avoid boundary condition problems ( in the goe the statistics of the eigenvalues near the edge are different than in the bulk ) . the coe is an statistical ensemble of @xmath88 unitary symmetric matrices @xmath9 invariant under the transformations @xmath89 where @xmath90 is any @xmath88 unitary matrix . the eigenvalues of @xmath9 are of the form @xmath91 , with @xmath92 . the probability density function for the eigenvalues is @xmath93 with @xmath94 . the partition function is @xmath95 @xcite . it should be noted that the eigenvalue probability density function of the coe is equal to the boltzmann factor of an equilibrium system of particles moving on a circle and interacting through a log - potential , the two - dimensional coulomb potential : @xmath96 with @xmath97 up to an additive constant , the log - potential is scale invariant , thus in these log - gases the density plays only the trivial role of fixing the length scale of the problem . in the other systems we study here , it is important to rescale the lengths in order to have the interparticle average spacing equal to one . for the non - equilibrium systems , this has been possible due to the fact that they exhibit a scaling regime . the coe is a homogeneous system , the density of eigenvalues is constant on the circle , equal to @xmath98 . it is convenient to rescale the lengths such that the average spacing between eigenvalues is unity . let @xmath99 be the rescaled eigenvalues ( note a small abuse of language here , since the eigenvalues are really @xmath100 not @xmath101 ) . analytic expressions for the eigenvalue spacing distributions are known , however they are complicated , see ref . @xcite . it turns out that a very accurate approximate expression for the nearest neighbor spacing distribution @xmath38 is given by the wigner surmise ( [ dgoe ] ) . in the limit of large matrices , @xmath102 , the correlation function between two ( rescaled ) eigenvalues @xmath3 and @xmath103 is given by @xmath104 \frac{ds}{dr}(r)+\left[s(r)\right]^2\ ] ] with @xmath105 and @xmath106 . we performed simulations for the gas , the spin , the crw , the irw and the circular orthogonal ensemble ( coe ) of random matrices with the following parameters . for the gas , the spin and the crw systems , we used a lattice with 1000 sites , and for the irw the lattice had 500 sites . the random matrix simulations were performed with 200@xmath107200 matrices . for the non - equilibrium systems , the simulations were carried out long enough to reach the scaling regime . the initial density of particles in the crw and the irw was @xmath108 and @xmath109 respectively . in the scaling regime , we build histograms for the domain sizes ( i.e. the nearest neighbor spacings for the crw , irw and coe ) . the data to build the histograms was taken at three different times for each system in order to verify the existence of a proper scaling regime . these times were @xmath110 , @xmath111 and @xmath112 for the gas system ; @xmath113 , @xmath114 and @xmath115 for the spin system ( the time unit of each system was defined in the previous sections ) ; @xmath116 , @xmath117 and @xmath118 for the crw , and @xmath117 , @xmath119 and @xmath118 for the irw . for the irw and the crw systems , the monte carlo time unit corresponds to a trial to move all the particles of the system . to have appropriate statistics , we performed 20000 simulations ( realizations ) of the gas , the spin and the coe systems , and 50000 realizations of the crw and the irw systems . figure [ g0 ] shows the scaling domain size distribution @xmath38 for the gas and the spin systems , obtained from the simulations . it is compared with the scaling nearest neighbor distribution of random walkers in the irw and the crw , and the scaled nearest neighbor distribution of eigenvalues of the circular orthogonal ensemble , which we also obtained from simulation of these systems . we also present the wigner surmise , eq . ( [ dgoe ] ) , which is know to numerically reproduce accurately the nearest neighbor distribution of eigenvalues of the circular orthogonal ensemble ( it is exact for 2@xmath1072 matrices ) . as we can see in figure [ g0 ] , the wigner surmise reproduces correctly the nearest neighbor distribution @xmath120 for the gas , the spin , the irw , the crw and the coe systems . this suggest a possible connection between the random matrices and the out of equilibrium systems . motivated by this possible link between the non - equilibrium statistical systems and the random matrix theory ( i.e. the equilibrium log - gas system ) , we explore the possibility if not only the nearest neighbor distribution is the same for all these systems , but also the other spacing distributions @xmath121 for @xmath122 and the pair correlation function @xmath123 . the approach here is to consider the edges of the domains in the gas and spin system as fundamental entities , and to study their spacing distributions and correlation function , in the scaling regime . with this study we wish to know if the similitude between these systems goes beyond the nearest neighbor distribution , or if it is simple a coincidence that the nearest neighbor distribution is the same for all systems . this also can give an indication about how much information on each system is contained in the nearest neighbor distribution . from our simulations , we computed the other spacing distributions @xmath15 and the correlation function @xmath123 of all these systems in order to compare them . the results are presented in the following subsections . [ ! h ] [ ! h ] and @xmath124 sites respectively . ] in ref . @xcite , the domain size distribution of the quasi one - dimensional gas was described by the approximate models of random walkers , the irw and crw . we now compare the other spacing distribution functions and the correlation function . the results are shown in figure [ g1 ] . the crw and irw reproduce correctly the nearest neighbor distribution @xmath120 . but , both the irw and the crw fail describing the other spacing distribution functions @xmath125 with @xmath122 , and , as consequence , they also fail describing the two - point correlation function @xmath126 , since the correlation function can be expressed as @xmath127 we notice in particular the fact that the correlation function of the gas system shows an oscillation around @xmath128 ( and probably more for higher values of @xmath129 ) . on the contrary , the crw and irw models show no oscillation in the correlation function @xmath123 . for these systems @xmath123 seems to be a monotonic function . the spacing distributions @xmath15 for @xmath130 of the irw and crw are somehow similar , but show higher maxima values than those of the gas system . furthermore , the maxima of the irw and the crw are located at higher values of @xmath131 than the corresponding ones of the gas system . we conclude that the crw and irw systems are not statistically equivalent to the quasi one - dimensional gas , although these models were proposed in ref . @xcite to describe ( approximately ) the quasi one - dimensional gas . now , in figure [ g3 ] , we compare the spacing distributions @xmath15 and the correlation function @xmath123 of both the gas and spin systems with the ones for the random matrix coe . the correlation functions @xmath123 of the gas and the spin system seem to be very similar , almost identical , with only small differences of a few percents in relative difference . they both exhibit the oscillation near @xmath128 mentioned above . on the contrary , the correlation function for the coe does not show any oscillation , and it is very different from the correlation function of the gas and spin system . the spacing distributions @xmath15 are somehow similar for the gas and the spin system , with some small differences between them . but , in any case , they differ much from the ones of the coe . the distributions @xmath15 , for @xmath130 , are more `` disperse '' in the gas and spin systems than the circular orthogonal ensemble , i.e. they have a width at half height larger than the ones for the coe . also , the maxima for each distribution of the gas and spin system are located at smaller values of @xmath131 than for the coe . the maxima values are also smaller for the gas and spin system than for the coe . it is possible to conclude that the gas and spin systems have a similar statistical behavior but this behavior is different from the coe . they only coincide in the nearest neighbor distribution . [ ! h ] ) and coe for a lattice with @xmath124 sites . in the coe we took @xmath132 matrices of @xmath133 . ] in the previous subsection , we have shown that the crw and the irw systems do not describe properly the statistics of the spacing distributions of the gas system . but do they have a similar behavior as the one of the coe ? in figure [ g4 ] , we compare the correlation functions and spacing distribution functions of both the crw and irw systems with coe . the correlation function @xmath123 of the crw and irw differs from the one of the coe . although the three correlation functions are monotonous ( they have no oscillations ) , around @xmath134 , the correlation function of the coe is smaller than the one of the irw and the crw . the first two distribution functions @xmath38 and @xmath135 of the three systems , irw , crw and coe , are very similar . however from @xmath136 , the coe spacing distributions @xmath15 start to differ from the corresponding ones of the crw and the irw . the spacing distributions of the crw and the irw seem more `` disperse '' ( as defined above ) than the ones of the coe . the maxima values for the spacing distributions in the coe are larger than for the irw and crw . also , these maxima are located at smaller values of @xmath131 for the irw and crw than for the coe . we conclude that the crw and the irw systems are not statistically equivalent to the coe . on the other hand the crw and the irw show very similar spacing distribution functions and correlation functions . thus , another important conclusion is that the interaction proposed in ref . @xcite for the irw do not change considerably the statistical behavior of the domains of the system from the crw in the scaling regime . the interaction between neighbors , eqs . ( [ eq : qr ] ) and ( [ eq : ql ] ) , change the growth exponent of the domains . but , once rescaled , the statistics of the walker spacing distributions and correlations of the crw and irw are very similar . [ ! h ] sites and in coe we take @xmath132 matrices of @xmath133 . ] the results presented above show that the statistics of the edges of the domains of the out of equilibrium gas and spin systems , properly rescaled in the scaling regime , are not described by the coe , thus are not described by an equilibrium statistical system of particles with a log - potential interaction . furthermore , for the gas systems , the approximate models of crw and irw do not reproduce correctly higher order spacing distributions and the correlation function . in this section , we propose a model that describe better the statistics of the domains of the out of equilibrium systems in the scaling regime when properly rescaled . we have found numerical evidence that suggest that the statistical behavior the domains of both the gas and spin systems can be fairly well described by the independent interval approximation ( iia ) , where the correlation between domains is neglected . with this simplification , it is possible calculate the distributions @xmath121 , for any @xmath130 , see refs . @xcite . for example , for @xmath137 , the probability to have two domain edges separated by a distance @xmath131 , knowing that there is one edge in between at a distance @xmath129 from the first edge is @xmath138 , if the domains are considered independent . therefore , @xmath139 it is useful to introduce here the laplace transform of the distribution functions @xmath140 taking the laplace transform of equation ( [ eq : p1-iia ] ) , we have @xmath141 if the wigner surmise ( [ dgoe ] ) is assumed for @xmath38 , its laplace transform is @xmath142 where @xmath143 is the complementary gaussian error function . using the inverse laplace transform of expression ( [ eq : p1-lap ] ) or from a direct calculation of eq . ( [ eq : p1-iia ] ) , we find @xmath144 with @xmath145 . more generally , under this approximation , the laplace transform of @xmath15 is simply @xmath146 since the pair correlation function can be obtained from eq . ( [ pkycdepares1 ] ) , then @xmath123 is given by the sum over convolutions of @xmath147 , see ref . @xcite . in terms of the laplace transform @xmath148 of the correlation function , we have @xmath149 with @xmath150 given by eq . ( [ eq : laplace - wigner ] ) . we have @xmath151 the inverse laplace transform of the above expression can be computed numerically to obtain @xmath123 . in this independent interval approximation , the joint probability density function to find borders in positions @xmath152 in a line of length @xmath4 is given by @xmath153 in compact form @xmath154 where we have considered periodic boundary conditions and we defined @xmath155 . the partition function @xmath156 is the normalization constant @xmath157 comparing ( [ pjoin ] ) with a boltzmann factor with an inverse temperature @xmath94 , we see that , under the independent interval approximation , the statistics of the domain edges is equivalent to a system with @xmath40 particles which interact according to the potential @xmath158\,.\ ] ] thus , it is equivalent to a statistical equilibrium system of particles on a circle interacting through a nearest neighbor pair potential . the partition function and correlations can be computed by means of the laplace transform as explained in general for this kind of systems in ref . we already computed the correlation function @xmath123 in eqs . ( [ pkycdepares2 ] ) and ( [ eq : g - r - iia ] ) . the partition function can be obtained as follows . doing a change of variable @xmath159 in eq . ( [ eq : part - iia ] ) we have @xmath160 this is an @xmath40-fold convolution product of @xmath161 . thus the laplace transform of the partition function is @xmath162 we recall that the circular orthogonal ensemble is equivalent to a system of particles on a circle interacting through a log - potential , see eq . ( [ eq : log - pot ] ) . from the equations ( [ viia ] ) and ( [ eq : log - pot ] ) , it is clear the difference between the independent domain approximation and the circular orthogonal ensemble , in the first system there are only nearest neighbor interactions , while in the second one every single particle interact with all the other particles , the pair potential is not restricted to nearest neighbors . additionally , the functional form of both potentials is different . in figure [ g2 ] , we compare the gas and spin system correlation functions and spacing distributions with the theoretical predictions from the independent interval approximation . we notice that this approximation reproduce more closely the spacing distributions and the correlation function . the iia correlation function has an oscillatory behavior near at @xmath128 as it occurs in the gas and spin system . although the predictions are not identical , the iia gives a good approximation for the correlation function and spacing distributions with an error of a few percents ( 5% ) deviation from the numerical results for the gas and the spin system . the fact that the independent interval approximation reproduce much better the correlation function and spacing distributions of the gas and the spin system than the other approaches considered ( crw , irw and coe ) , suggest that in the gas and spin systems the domains are not strongly correlated . [ ! h ] sites . ] we studied two out of equilibrium systems , the gas and spin system , which present a formation of growing domains . we studied the statistical properties of the edges of the domains in the scaling regime and compared them with several other models . since in the scaling regime the statistical properties of the edges of the domains are time independent , when properly rescaled , we tried to find if there exists an equilibrium statistical system which reproduce these statistics . a first candidate was the circular orthogonal ensemble of random matrices ( coe ) , since the scaling size distribution of the domains satisfies the wigner surmise . studying the other spacing distributions @xmath15 and the correlation function , we found that only the nearest neighbor distribution ( @xmath163 ) coincides for the gas system , the spin system , the coe and the models of coalescing and interacting random walkers ( crw and irw ) which were introduced in ref . @xcite to explain the properties of the gas system . the distribution functions @xmath15 for @xmath130 and the correlation function are different for each of these systems . the nearest neighbor distribution for the coalescing random walk is obtained by solving the diffusion equation with an absorbing condition in @xmath164 @xcite , and the scaling solution turns out to be given by eq . ( [ dgoe ] ) . as a side note , it is interesting to notice that , up to a normalization , this is the wave function of the first excited state of the harmonic oscillator , an analogy that has been used in the context of persistence , see ref . our work shows that it is merely a coincidence that this nearest neighbor distribution turns out to be the wigner distribution from the random matrices theory . we found a model which provides a better approximation for the gas and the spin system domain statistics . this model is the independent interval approximation , even though in this approximation the correlation between domains is not taken into account as it occurs in interacting and coalescing random walk . the independent interval approximation is statistically equivalent to an equilibrium system of particles with only nearest neighbor interactions . the fact that the wigner surmise , eq . ( [ dgoe ] ) , reproduces correctly the nearest neighbor spacing distribution of all the different systems we considered here , but that the other spacing distributions differ from system to system , seems to show that the nearest neighbor distribution has some kind of universality while the other distributions do not . however , it can also be interpreted as an indication that the nearest neighbor distribution does not contain enough information about the statistics of the system : several different systems could share the same nearest neighbor distribution , while the finer details and differences between them are contained in the correlation function and the other spacing distributions . this is the most important result of this paper , because in many cases complex systems are mapped onto more simple systems with the only criteria of the nearest neighbor distribution similitude . the authors thank p. j. forrester and f. van wijland for valuable comments and observations . this work was partially supported by an ecos nord / colciencias action of french and colombian cooperation . 99 j. mettetal , b. schmittmann and r. zia , coarsening dynamics of a quasi one - dimensional driven lattice gas , europhysics lett . * 58 * , 653659 ( 2002 ) . s. j. cornell and a. j. bray , domain growth in a one - dimensional driven diffusive system , phys . rev . e * 54 * , 11531160 ( 1996 ) . v. spirin , p. l. krapivsky and s. redner , coarsening in a driven ising chain with conserved dynamics , phys . e * 60 * , 26702676 ( 1999 ) . m. l. mehta , random matrices , @xmath165 , academic press ( 1991 ) . m. e. fisher , walks , walls , wetting and melting , j. stat . phys . * 34 * , 667 ( 1984 ) . p. j. forrester , exact solution of the lock step model of vicious walkers , j. phys . a : math . * 23 * , 12591273 ( 1990 ) . j. forrester , exact results for vicious walker models of domain walls , j. phys . a : math . gen . * 24 * , 203218 ( 1991 ) . i. georgiev , b. schmittmann and r. zia , cluster growth and dynamic scaling in a two - lane driven diffusive system , j. phys . a : math . gen . * 39 * , 34953509 ( 2006 ) . d. ben - avraham , m. a. burschka , and c. r. doering , statics and dynamics of a diffusion - limited reaction : anomalous kinetics , nonequilibrium self - ordering , and a dynamic transition , j. stat . phys . * 60 * , 695728 ( 1990 ) . b. derrida , c. godrche , and i. yekutieli , scale - invariant regimes in one - dimensional models of growing and coalescing droplets , phys . a * 44 * , 62416251 ( 1991 ) . d. ben - avraham and e. brunet , on the relation between one - species diffusion - limited coalescence and annihilation in one dimension , j. phys . a. * 38 * , 32473252 ( 2005 ) . b. derrida and r. zeitak , distribution of domains sizes in the zero temperature glauber dynamics of the one - dimensional potts model , phys . e. * 54 * , 25132525 ( 1996 ) . a. alemany and d. ben - avraham , inter - particle distribution functions for one - species diffusion - limited annihilation , @xmath166 , phys . a. * 206 * , 1825 ( 1995 ) . b. derrida , v. hakim and r. zeitak , persistent spins in the linear diffusion approximation of phase ordering and zeros of stationary gaussian processes , phys . * 77 * , 28712874 ( 1996 ) . s. n. majumdar , c. sire , a. j. bray and s. j. cornell , nontrivial exponent for simple diffusion , phys . rev . lett . * 77 * , 28672870 ( 1996 ) . p. l. krapivsky and e. ben - naim , domain statistics in coarsening systems , phys . e * 56 * , 37883798 ( 1997 ) . e. ben - naim and p. l. krapivsky , domain number distribution in the nonequilibrium ising model , j. stat . phys . * 93 * , 583601 ( 1998 ) . z. w. salsburg , r. w. zwanzig and j. g. kirkwood , molecular distribution functions in a one - dimensional fluid , j. chem . phys . * 21 * , 10981107 ( 1953 ) . s. n. majumdar and c. sire , survival probability of a gaussian non - markovian process : application to the @xmath167 dynamics of the ising model , phys . 77 * , 14201423 ( 1996 ) . k. oerding , s. j. cornell , and a. bray , non - markovian persistence and nonequilibrium critical dynamics , phys . e * 56 * , r25r28 ( 1997 ) .	we study the statistical behavior of two out of equilibrium systems . the first one is a quasi one - dimensional gas with two species of particles under the action of an external field which drives each species in opposite directions . the second one is a one - dimensional spin system with nearest neighbor interactions also under the influence of an external driving force . both systems show a dynamical scaling with domain formation . the statistical behavior of these domains is compared with models based on the coalescing random walk and the interacting random walk . we find that the scaling domain size distribution of the gas and the spin systems is well fitted by the wigner surmise , which lead us to explore a possible connection between these systems and the circular orthogonal ensemble of random matrices . however , the study of the correlation function of the domain edges , show that the statistical behavior of the domains in both gas and spin systems , is not completely well described by circular orthogonal ensemble , nor it is by other models proposed such as the coalescing random walk and the interacting random walk . nevertheless , we find that a simple model of independent intervals describe more closely the statistical behavior of the domains formed in these systems . * keywords : * systems out of equilibrium , random matrices . * pacs : * 05.40.-a , 05.40.fb , 05.50.+q , 45.70.vn .
survival analysis is a commonly - used method for the analysis of failure time such as biological death , mechanical failure , or credit default . within this context , death or failure is also referred to as an event " . survival analysis tries to model time - to - event data , which is usually subject to censoring due to the termination of study . the main goal is to study the dependence of the survival time @xmath0 on covariate variables @xmath1 , where @xmath2 denotes the dimensionality of the covariate space . one common way of achieving this goal is hazard regression , which studies how the conditional hazard function of @xmath0 depends on the covariate @xmath3 , which is defined as @xmath4 according to the definition , the conditional hazard function is nothing but the instantaneous rate of failure at time @xmath5 given a particular value @xmath6 for the covariate @xmath7 . the proportional hazards model is very popular , partially due to its simplicity and its convenience in dealing with censoring . the model assumes that @xmath8 in which @xmath9 is the baseline hazard function and @xmath10 is the covariate effect . note that this model is not uniquely determined in that @xmath11 and @xmath12 give the same model for any @xmath13 . thus one identifiability condition needs to be specified . when the identifiability condition @xmath14 is enforced , the function @xmath9 , the conditional hazard function of @xmath0 given @xmath15 , is called the baseline hazard function . by taking the reparametrization @xmath16 , @xcite introduced the proportional hazards model @xmath17 see @xcite and references therein for more detailed literature on cox s proportional hazards model . here the baseline hazard function @xmath9 is typically completely unspecified and needs to be estimated nonparametrically . a linear model assumption , @xmath18 , may be made , as is done in this paper . here @xmath19 is the regression parameter vector . while conducting survival analysis , we not only need to estimate @xmath20 but also have to estimate the baseline hazard function @xmath9 nonparametrically . interested readers may consult @xcite for more details . recent technological advances have made it possible to collect a huge amount of covariate information such as microarray , proteomic and snp data via bioimaging technology while observing survival information on patients in clinical studies . however it is quite likely that not all available covariates are associated with the clinical outcome such as the survival time . in fact , typically a small fraction of covaraites are associated with the clinical time . this is the notion of sparsity and consequently calls for the identification of important risk factors and at the same time quantifying their risk contributions when we analyze time - to - event data with many predictors . mathematically , it means that we need to identify which @xmath21s are nonzero and also estimate these nonzero @xmath21s . most classical model selection techniques have been extended from linear regression to survival analysis . they include the best - subset selection , stepwise selection , bootstrap procedures @xcite , bayesian variable selection @xcite . please see references therein . similarly , other more modern penalization approaches have been extended as well . @xcite applied the lasso penalty to survival analysis . @xcite considered survival analysis with the scad and other folded concave penalties . @xcite proposed the adaptive lasso penalty while studying time - to - event data . among many other considerations is @xcite . available theory and empirical results show that these penalization approaches work well with a moderate number of covariates . recently we have seen a surge of interest in variable selection with an ultra - high dimensionality . by ultra - high , @xcite meant that the dimensionality grows exponentially in the sample size , i.e. , @xmath22 for some @xmath23 . for ultra - high dimensional linear regression , @xcite proposed sure independence screening ( sis ) based on marginal correlation ranking . asymptotic theory is proved to show that , with high probability , sis keeps all important predictor variables with vanishing false selection rate . an important extension , iterative sis ( isis ) , was also proposed to handle difficult cases such as when some important predictors are marginally uncorrelated with the response . in order to deal with more complex real data , @xcite extended sis and isis to more general loss based models such as generalized linear models , robust regression , and classification and improved some important steps of the original isis . in particular , they proposed the concept of conditional marginal regression and a new variant of the method based on splitting samples . a non - asymptotic theoretical result shows that the splitting based new variant can reduce false discovery rate . although the extension of @xcite covers a wide range of statistical models , it has not been explored whether the iterative sure independence screening method can be extended to hazard regression with censoring event time . in this work , we will focus on cox s proportional hazards model and extend sis and isis accordingly . other extensions of sis include @xcite and @xcite to generalized linear models and nonparametric additive models , in which new insights are provided via elegant mathematical results and carefully designed simulation studies . the rest of the article is organized as follows . section [ sec : cox ] details the cox s proportional hazards model . an overview of variable selection via penalized approach is given in section [ sec : variable - selection - cox ] for cox s proportional hazards model . we extend the sis and isis procedures to cox s model in section [ sec : sis - cox ] . simulation results in section [ sec : simulation ] and real data analysis in section [ sec : realdata ] demonstrate the effectiveness of the proposed sis and isis methods . let @xmath0 , @xmath24 , and @xmath7 denote the survival time , the censoring time , and their associated covariates , respectively . correspondingly , denote by @xmath25 the observed time and @xmath26 the censoring indicator . for simplicity we assume that @xmath0 and @xmath24 are conditionally independent given @xmath7 and that the censoring mechanism is noninformative . our observed data set @xmath27 is an independently and identically distributed random sample from a certain population @xmath28 . define @xmath29 and @xmath30 to be the censored and uncensored index sets , respectively . then the complete likelihood of the observed data set is given by @xmath31 where @xmath32 , @xmath33 , and @xmath34 are the conditional density function , the conditional survival function , and the conditional hazard function of @xmath0 given @xmath3 , respectively . let @xmath35 be the ordered distinct observed failure times . let @xmath36 index its associate covariates @xmath37 and @xmath38 be the risk set right before the time @xmath5 : @xmath39 . consider the proportional hazards model , @xmath40 where @xmath9 is the baseline hazard function . in this model , both @xmath9 and @xmath20 are unknown and have to be estimated . under model , the likelihood becomes @xmath41 where @xmath42 is the corresponding cumulative baseline hazard function . following breslow s idea , consider the least informative " nonparametric modeling for @xmath43 , in which @xmath44 has a possible jump @xmath45 at the observed failure time @xmath46 , namely , @xmath47 . then @xmath48 consequently the log - likelihood becomes @xmath49 maximizer @xmath45 is given by @xmath50 putting this maximizer back to the log - likelihood , we get @xmath51,\nonumber\ ] ] which is equivalent to @xmath52 by using the censoring indicator @xmath53 . this is the partial likelihood due to @xcite . maximizing @xmath54 in ( [ simlik ] ) with respect to @xmath20 , we can get an estimate @xmath55 of the regression parameter . once @xmath55 is available , we may plug it into ( [ hj ] ) to get @xmath56 . these newly obtained @xmath57s can be plugged into ( [ leastinfo ] ) to obtain our nonparametric estimate of the baseline cumulative hazard function . in the estimation scheme presented in the previous section , none of the estimated regression coefficients is exactly zero , leaving all covariates in the final model . consequently it is incapable of selecting important variables and handling the case with @xmath58 . to achieve variable selection , classical techniques such as the best - subset selection , stepwise selection , and bootstrap procedures have been extended accordingly to handle cox s proportional hazards model . in this section , we will focus on some more advanced techniques for variable selection via penalization . variable selection via penalization has received lots of attention recently . basically it uses some variable selection - capable penalty function to regularize the objective function while performing optimization . many variable selection - capable penalty functions have been proposed . a well known example is the @xmath59 penalty @xmath60 , which is also known as the lasso penalty @xcite . among many others are the scad penalty @xcite , elastic - net penalty @xcite , adaptive @xmath59 @xcite , and minimax concave penalty @xcite . denote a general penalty function by @xmath61 , where @xmath62 is a regularization parameter . from derivations in the last section , penalized likelihood is equivalent to penalized partial likelihood : while maximizing @xmath54 in ( [ simlik ] ) , one may regularize it using @xmath63 . equivalently we solve @xmath64 by including a negative sign in front of @xmath54 . in the literature , @xcite , @xcite , and @xcite considered the @xmath59 , scad , and adaptive @xmath59 penalties while studying time - to - event data , respectively , among many others . in this paper , we will use the scad penalty for our extensions of sis and isis whenever necessary . the scad function is a quadratic spline and symmetric around the origin . it can be defined in terms of its first order derivative @xmath65 for some @xmath66 and @xmath67 . here @xmath68 is a parameter and @xcite recommend to use @xmath69 based on a bayesian argument . the scad penalty is plotted in figure [ scadplot ] for @xmath69 and @xmath70 . the scad penalty is non - convex , leading to non - convex optimization . for the non - convex scad penalized optimization , @xcite proposed the local quadratic approximation ; @xcite proposed the local linear approximation ; @xcite presented the difference convex algorithm . in this work , whenever necessary we use the local linear approximation algorithm to solve the scad penalized optimization . the penalty based variable selection techniques work great with a moderate number of covariates . however their usefulness is limited while dealing with an ultra - high dimensionality as shown in @xcite . in the linear regression case , @xcite proposed to rank covariates according to the absolute value of their marginal correlation with the response variable and select the top ranked covariates . they provided theoretical result to guarantee that this simple correlation ranking retains all important covariates with high probability . thus they named their method sure independence screening ( sis ) . in order to handle difficult problems such as the one with some important covariates being marginally uncorrelated with the response , they proposed iterative sis ( isis ) . isis begins with sis , then it regresses the response on covariates selected by sis and uses the regression residual as a working " response to recruit more covariates with sis . this process can be repeated until some convergence criterion has been met . empirical improvement over sis has been observed for isis . in order to increase the power of the sure independence screening technique , @xcite has extended sis and isis to more general models such as generalized linear models , robust regression , and classification and made several important improvements . we now extend the key idea of sis and isis to handle cox s proportional hazards model . let @xmath71 be the index set of the true underlying sparse model , namely , @xmath72 , where @xmath73s are the true regression coefficients in the cox s proportional hazards model ( [ coxmodelwu ] ) . first let us review the definition of sure screening property . we say a model selection procedure satisfies sure screening property if the selected model @xmath74 with model size @xmath75 includes the true model @xmath71 with probability tending to one . for each covariate @xmath76 ( @xmath77 ) , define its marginal utility as the maximum of the partial likelihood of the single covariate : @xmath78 here @xmath79 is the @xmath80-th element of @xmath81 , namely @xmath82 . intuitively speaking , the larger the marginal utility is , the more information the corresponding covariate contains the information about the survival outcome . once we have obtained all marginal utilities @xmath83 for @xmath84 , we rank all covariates according to their corresponding marginal utilities from the largest to the smallest and select the @xmath85 top ranked covariates . denote by @xmath86 the index set of these @xmath85 covariates that have been selected . the index set @xmath86 is expected to cover the true index set @xmath71 with a high probability , especially when we use a relative large @xmath85 . this is formally shown by @xcite for the linear model with gaussian noise and gaussian covariates and significantly expanded by @xcite to generalized linear models with non - gaussian covariates . the parameter @xmath85 is usually chosen large enough to ensure the sure screening property . however the estimated index set @xmath86 may also include a lot of unimportant covariates . to improve performance , the penalization based variable selection approach can be applied to the selected subset of the variables @xmath87 to further delete unimportant variables . mathematically , we then solve the following penalized partial likelihood problem : @xmath88 where @xmath89 denotes a sub - vector of @xmath81 with indices in @xmath86 and similarly for @xmath90 . it will lead to sparse regression parameter estimate @xmath91 . denote the index set of nonzero components of @xmath92 by @xmath74 , which will serve as our final estimate of @xmath71 . @xcite pointed out that sis can fail badly for some challenging scenarios such as the case that there exist jointly related but marginally unrelated covariates or jointly uncorrelated covariates having higher marginal correlation with the response than some important predictors . to deal with such difficult scenarios , iterative sis ( isis ) has been proposed . comparing to sis which is based on marginal information only , isis tries to make more use of joint covariates information . the iterative sis begins with using sis to select an index set @xmath93 , upon which a penalization based variable selection step is applied to get regression parameter estimate @xmath94 . a refined estimate of the true index set is obtained and denoted by @xmath95 , the index set corresponding to nonzero elements of @xmath94 . as in @xcite , we next define the conditional utility of each covariate @xmath80 that is not in @xmath95 as follows : @xmath96 this conditional utility measures the additional contribution of the @xmath80th covariate given that all covariates with indices in @xmath95 have been included in the model . once the conditional utilities have been defined for each covariate that is not in @xmath95 , we rank them from the largest to the smallest and select these covariates with top rankings . denote the index set of these selected covariates by @xmath97 . with @xmath97 having been identified , we minimize @xmath98 with respect to @xmath99 to get sparse estimate @xmath100 . denote the index set corresponding to nonzero components of @xmath100 to be @xmath101 , which is our updated estimate of the true index set @xmath71 . note that this step can delete some variables @xmath102 that were previously selected . this idea was proposed in @xcite and is an improvement of the idea in @xcite . the above iteration can be repeated until some convergence criterion is reached . we adopt the criterion of either having identified @xmath85 covariates or @xmath103 for some @xmath104 . @xcite noted that the idea of sample spliting can also be used to reduce the false selection rate . without loss of generality , we assume that the sample size @xmath105 is even . we randomly split the sample into two halves . then apply sis or isis separately to the data in each partition to obtain two estimates @xmath106 and @xmath107 of the true index set @xmath108 . both these two estimates could have high fsrs because they are based on a simple and crude screening method . yet each of them should include all important covariates with high probabilities . namely , important covariates should appear in both sets with probability tending to one asymptotically . define a new estimate by intersection @xmath109 . the new estimate @xmath110 should include all important covaraites with high probability as well due to properties of each individual estimate . however by construction , the number of unimportant covariates in the new estimate @xmath110 is much smaller . the reason is that , in order for an unimportant covariate to appear in @xmath110 , it has to be included in both @xmath106 and @xmath107 randomly . for the new variant method based on random splitting , @xcite obtained some non - asymptotic probability bound for the event that @xmath111 unimportant covaraites are included in the intersection @xmath110 for any natural number @xmath111 under some exchangeability condition on all unimportant covariates . the probability bound is decreasing in the dimensionality , showing a blessing of dimensionality " . please consult @xcite for more details . we want to remark that their theoretical bound is applicable to our setting as well while studying time - to - event data because theoretical bound is based on splitting the sample into two halves and only requires the independence between these two halves . while defining new variants , we may use the same @xmath85 as used in the original sis and isis . however it will lead to a very aggressive screening . we call the corresponding variant the first variant of ( i)sis . alternatively , in each step we may choose larger @xmath106 and @xmath107 to ensure that their intersection @xmath112 has @xmath85 covariates , which is called the second variant . the second variant ensures that there are at least @xmath85 covariates included before applying penalization in each step and is thus much less aggressive . numerical examples will be used to explore their performance and prefer to the first variant . in this section , we conduct simulation studies to show the power of the ( i)sis and its variants by comparing them with lasso @xcite in the cox s proportional hazards model . here the regularization parameter for lasso is tuned via five fold cross validation . most of the settings are adapted from @xcite and @xcite . four different configurations are considered with @xmath113 and @xmath114 . and two of them are revisited with a different pair of sample size @xmath115 and dimensionality @xmath116 . covariates in different settings are generated as follows . * 1 : @xmath117 are independent and identically distributed @xmath118 random variables . * 2 : @xmath117 are multivariate gaussian , marginally @xmath118 , and with serial correlation @xmath119 if @xmath120 . here we take @xmath121 . * 3 : @xmath117 are multivariate gaussian , marginally @xmath118 , and with correlation structure @xmath122 for all @xmath123 and @xmath124 if @xmath125 and @xmath104 are distinct elements of @xmath126 . * 4 : @xmath117 are multivariate gaussian , marginally @xmath118 , and with correlation structure @xmath127 for all @xmath128 , @xmath122 for all @xmath129 , and @xmath130 if @xmath125 and @xmath104 are distinct elements of @xmath131 . * 5 : same as case 2 except @xmath115 and @xmath116 . * 6 : same as case 4 except @xmath115 and @xmath116 . here , case 1 with independent predictors is the most straightforward for variable selection . in cases 2 - 6 , however , we have serial correlation such that @xmath132 does not decay as @xmath133 increases . we will see later that for cases 3 , 4 and 6 , the true coefficients are specially chosen such that the response is marginally independent but jointly dependent of @xmath134 . we therefore expect variable selection in these situations to be much more challenging , especially for the non - iterated versions of sis . notice that in the asymptotic theory of sis in @xcite , this type of dependence is ruled out by their condition ( 4 ) . in our implementation , we choose @xmath135 for both the vanilla version of sis ( van - sis ) and the second variant ( var2-sis ) . for the first variant ( var1-sis ) , however , we use @xmath136 ( note that since the selected variables for the first variant are in the intersection of two sets of size @xmath85 , we typically end up with far fewer than @xmath85 variables selected by this method ) . for any type of sis or isis , we apply scad with these selected predictors to get a final estimate of the regression coefficients at the end of the screening step . whenever necessary , the bic is used to select the best tuning parameter in the regularization framework . in all setting , the censoring time is generated from exponential distribution with mean @xmath137 . this corresponds to choosing the baseline hazard function @xmath138 for @xmath139 . the true regression coefficients and censoring rate in each of the six cases are as follows : * 1 : @xmath140 , and @xmath141 for @xmath142 . the corresponding censoring rate is 33% . * 2 : the coefficients are the same as case 1 . the corresponding censoring rate is 27% . * 3 : @xmath143 , and @xmath141 for @xmath144 . the corresponding censoring rate is 30% . * 4 : @xmath145 and @xmath141 for @xmath146 . the corresponding censoring rate is 31% . * 5 : @xmath147 , and @xmath141 for @xmath142 . the corresponding censoring rate is 23% . * 6 : the coefficients are the same as case 4 . the corresponding censoring rate is 36% . in cases 1 , 2 and 5 the coefficients were chosen randomly , and were generated as @xmath148 with @xmath149 and @xmath150 with probability 0.5 and -1 with probability 0.5 , independent of @xmath151 . for cases 3 , 4 , and 6 , the choices ensure that even though @xmath152 , we have that @xmath134 and @xmath153 are marginally independent . the fact that @xmath134 is marginally independent of the response is designed to make it difficult for the common independent learning to select this variable . in cases 4 and 6 , we add another important variable @xmath154 with a small coefficient to make it even more difficult . we report our simulation results based on 100 monte carlo repetitions for each setting in tables 1 - 7 . to present our simulation results , we use several different performance measures . in the rows labeled @xmath155 and @xmath156 , we report the median @xmath59 and squared @xmath157 estimation errors @xmath158 and @xmath159 , respectively , where the median is over the 100 repetitions . in the row with label @xmath160 , we report the proportion of the 100 repetitions that the ( i)sis procedure under consideration includes all of the important variables in the model , while the row with label @xmath161 reports the corresponding proportion of times that the final variables selected , after further application of the scad penalty , include all of the important ones . we also report the median model size ( mms ) of the final model among 100 repetitions in the row labeled mms . c\|c\|c\|c\|c\|c\|c\|c + & & & & & & & + @xmath155 & 0.79 & 0.57 & 0.73 & 0.61 & 0.76 & 0.62 & 4.23 + @xmath156 & 0.13 & 0.09 & 0.15 & 0.1 & 0.15 & 0.1 & 0.98 + @xmath160 & 1 & 1 & 0.99 & 1 & 0.99 & 1 & + @xmath161 & 1 & 1 & 0.99 & 1 & 0.99 & 1 & 1 + mms & 7 & 6 & 6 & 6 & 6 & 6 & 68.5 + + & & & & & & & + @xmath155 & 2.2 & 0.64 & 4.22 & 0.8 & 3.95 & 0.78 & 4.38 + @xmath156 & 1.74 & 0.11 & 4.71 & 0.29 & 4.07 & 0.28 & 0.98 + @xmath160 & 0.71 & 1 & 0.42 & 0.99 & 0.46 & 0.99 & + @xmath161 & 0.71 & 1 & 0.42 & 0.99 & 0.46 & 0.99 & 1 + mms & 7 & 6 & 6 & 6 & 7 & 6 & 57 + [ case1 ] we report results of cases 1 and 2 in table 1 . recall that the covariates in case 1 are all independent . in this case , the van - sis performs reasonably well . yet , it does not perform well for the dependent case , case 2 . note the only difference between case 1 and case 2 is the covariance structure of the covariates . for both cases , vanilla - isis and its second variant perform very well . it is worth noticing that the isis improves significantly over sis , when covariates are dependent , in terms of both the probability of including all the true variables and in reducing the estimation error . this comparison indicates that the isis performs much better when there is serious correlation among covariates . while implementing the lasso penalized cox s proportional hazards model , we adapted the fortran source code in the r package glmpath . " recall that the objective function in the lasso penalized cox s proportional hazards model is convex and nonlinear . what the fortran code does is to call a minos subroutine to solve the corresponding nonlinear convex optimization problem . here minos is an optimization software developed by systems optimization laboratory at stanford university . this nonlinear convex optimization problem is much more complicated than a general quadratic programming problem . thus generally it takes much longer time to solve , especially so when the dimensionality is high as confirmed by table 3 . however the algorithm we used does converge as the objective function is strictly convex . table 1 shows that lasso has the sure screening property as the isis , however , the median model size is ten times as large as that of isis . as a consequence , it also has larger estimation errors in terms of @xmath155 and @xmath156 . the fact that the median absolute deviation error is much larger than the median square error indicates that the lasso selects many small nonzero coefficients for those unimportant variables . this is also verified by the fact that lasso has a very large median model size . the explanation is the bias issue noted by @xcite . in order for lasso to have a small bias for nonzero coefficients , a smaller @xmath162 should be chosen . yet , a small @xmath162 recruits many small coefficients for unimportant variables . for case 2 , the lasso has a similar performance as in case 1 in that it includes all the important variables but has a much larger model size . c\|c\|c\|c\|c\|c\|c\|c + & & & & & & & + @xmath155 & 20.1 & 1.03 & 20.01 & 0.99 & 20.09 & 1.08 & 20.53 + @xmath156 & 94.72 & 0.49 & 100.42 & 0.47 & 94.77 & 0.55 & 76.31 + @xmath160 & 0 & 1 & 0 & 1 & 0 & 1 & + @xmath161 & 0 & 1 & 0 & 1 & 0 & 1 & 0.06 + mms & 13 & 4 & 8 & 4 & 13 & 4 & 118.5 + + & & & & & & & + @xmath155 & 20.87 & 1.15 & 20.95 & 1.4 & 20.96 & 1.41 & 21.04 + @xmath156 & 96.46 & 0.51 & 102.14 & 1.77 & 97.15 & 1.78 & 77.03 + @xmath160 & 0 & 1 & 0 & 0.99 & 0 & 0.99 & + @xmath161 & 0 & 1 & 0 & 0.99 & 0 & 0.99 & 0.02 + mms & 13 & 5 & 9 & 5 & 13 & 5 & 118 + [ case3 ] results of cases 3 and 4 are reported in table 2 . note that , in both cases , the design ensures that @xmath134 is marginally independent of but jointly dependent on @xmath153 . this special design disables the sis to include @xmath134 in the corresponding identified model as confirmed by our numerical results . however , by using isis , we are able to select @xmath134 for each repetition . surprisingly , lasso rarely includes @xmath134 even if it is not a marginal screening based method . case 5 is even more challenging . in addition to the same challenge as case 4 , the coefficient @xmath163 is 3 times smaller than the first four variables . through the correlation with the first 4 variables , unimportant variables @xmath164 have a larger marginal utility with @xmath153 than @xmath154 . nevertheless , the isis works very well and demonstrates once more that it uses adequately the joint covariate information . .the average running time ( in seconds ) comparison for van - isis and lasso . [ cols="^,^,^,^,^",options="header " , ] now we try to provide some understanding to the significance of these selected genes in predicting the survival information in comparison to other genes that are not selected . we first fitted the cox s proportional hazard model with all these eight genes . estimated coefficients are given in table [ tablenb ] , estimated baseline survival function is plotted in figure [ fignb ] , and the corresponding log-(partial)likelihood ( [ simlik ] ) is -129.3517 . the log - likelihood corresponding to the null model without any predictor is -215.4561 . a @xmath165 test shows the obvious significance of the model with the eight selected genes . table [ tablenb ] shows that there are two estimated coefficients that are statistically insignificant at @xmath166 .. next for each one of these eight genes , we remove it , fit cox s proportional hazard model with the other seven genes , and get the corresponding log - likelihood . the eight log - likelihoods are -137.5785 , -135.1846 , -129.4621 , -142.4066 , -156.4644 , -158.3799 , -141.0432 , and -129.8390 . their average is -141.2948 , a decrease of log - likelihood by 11.9431 , which is very significant with reduction one gene ( the reduction of the degree of freedom by 1 ) . in comparison to the model with the eight selected genes , @xmath165 tests shows significance for all selected genes except a_23_p31816 and hs150167.1 . this matches the p - values reported in table [ tablenb ] . finally we randomly select 2 genes out of the genes that are not selected , fit the cox s proportional hazard model with the above eight genes plus these two randomly selected genes , and record the corresponding log - likelihood . we repeat this process 20 times . we find that the average of these 20 new log - likelihoods is -128.3933 , an increase of the log - likelihood merely by 0.9584 with two extra variables included . comparing to the model with the eight selected genes , @xmath165 test shows no significance for the model corresponding to any of the 20 repetitions . the above experiments show that the selected 8 genes are very important . deleting one reduces a lot of log - likelihood , while adding two random genes do not increase very much the log - likelihood . we have developed a variable selection technique for the survival analysis with the dimensionality that can be much larger than sample size . the focus is on the iterative sure independence screening , which iteratively applies a large - scale screening that filters unimportant variables by using the conditional marginal utility , and a moderate - scale selection by using penalized partial likelihood method , which selects further the unfiltered variables . the methodological power of the vanilla isis has been demonstrated via carefully designed simulation studies . it has sure independence screening with very small false selection . comparing with the version of lasso we used , it is much more computationally efficient and far more specific in selection important variables . as a result , it has much smaller absolute deviation error and mean square error . oberthuer , a. , berthold , f. , warnat , p. , hero , b. , kahlert , y. , spitz , r. , ernestus , k. , knig , r. , haas , s. , eils , r. , schwab , m. , brors , b. , westermann , f. and fischer , m. ( 2006 ) . customized oligonucleotide microarray gene expressionbased classification of neuroblastoma patients outperforms current clinical risk stratification . _ journal of clinical oncology _ , * 24 * 50705078 .	variable selection in high dimensional space has challenged many contemporary statistical problems from many frontiers of scientific disciplines . recent technological advances have made it possible to collect a huge amount of covariate information such as microarray , proteomic and snp data via bioimaging technology while observing survival information on patients in clinical studies . thus , the same challenge applies in survival analysis in order to understand the association between genomics information and clinical information about the survival time . in this work , we extend the sure screening procedure @xcite to cox s proportional hazards model with an iterative version available . numerical simulation studies have shown encouraging performance of the proposed method in comparison with other techniques such as lasso . this demonstrates the utility and versatility of the iterative sure independence screening scheme . ,
metal - rich stars ( [ fe / h]@xmath1 + 0.25 dex ) found in the solar neighborhood , including giant planet hosts , are objects that have migrated from the inner disk ( i.e inside the solar galactocentric radius ) by the effect of radial mixing ( sellwood & binney 2002 ; haywood 2008b ) . this is an intriguing fact , but since it has been proposed that the prevalence of jovian planets on metal - rich stars could be due to the enhanced probability of forming planetesimals in an environment enriched in metals , why should one bother about the galactic origin of the host stars ? the first reason is that the correlation between stellar metallicities and the presence of giant planets is made up of stars of different galactic origins , which could bear some importance on the formation of giant planets themselves . the second is that the study of the effect of radial mixing permits new insights on two particularities that otherwise do not fit well into the scenario of metal - enhanced planet formation , namely the normal metallicity distribution of giants stars with planets ( pasquini et al . ( 2007 ) , takeda et al . ( 2008 ) ) , and the fact that , at intermediate metallicities , giant planets seems to favor thick disk stars rather than thin disk objects ( haywood , 2008a ) . both arguments are the subject of this letter , which is laid out as follows : in @xmath22 we briefly discuss the evidences of radial mixing and show how metallicity depends on the galactic orbital parameters of planet host stars . in 3 , we show that ( 1 ) the difference in the metallicity distribution of planet host giants and dwarfs and ( 2 ) the difference in the number of planet hosts between the thin and thick disks can be both explained as a consequence of radial mixing . in 4 , we discuss the implication of these results , and demonstrate that if the percentage of giant planets among stars is dependent upon the galactocentric distance , a correlation between stellar metallicity and the presence of planet is a natural outcome of radial migration . finally , we discuss what galactic property , dependent on galactocentric distance , could explain these new results . the suggestion that stars migrate in the galactic disk has been around since at least the seventies . a few dynamical processes have been claimed responsible for this phenomenon : random scattering ( wielen 1977 ) , churning by spirals ( sellwood & binney 2002 ) or perturbations by an orbiting satellite ( quillen et al . 2009 ) , but specific observational evidence for any such processes remain elusive . the only direct evidence so far that such processes may be active is the differentiation encrypted in the orbital parameters of solar neighborhood stars considered as a function of metallicity . an example is given in fig . 4 of haywood ( 2008b ) , which shows the clear difference in the distribution of apo and pericenters for metal - poor and metal - rich stars . other indirect evidences exist . one is the increasing metallicity dispersion with age ( haywood 2006 , 2008b ) , which testifies that the older the star , the greater the distance over which it may have migrated , and therefore come from a region with significantly different mean stellar abundance . the second is that the terminal metallicity reached by the local chemical evolution is about 0.2 dex , not 0.4 or 0.6 dex . this is evidenced by the fact that there are no young stars above [ fe / h]@xmath30.2 dex in the solar neighborhood . the solar radius has simply not reached this state of chemical evolution , and super metal - rich objects must have formed elsewhere , the inner disk being the most probable site . the galactic aspects of the bias towards metal - rich stars among planet hosts have been seldom investigated . in particular , the evidence that metal - rich stars of the solar vicinity must have come from the inner disk has been barely discussed . the exceptions are ecuvillon et al . ( 2007 ) and haywood ( 2008a ) . several studies investigated whether planet host stars have properties that could differ from those of common field stars , but it has been so far rather unfruitful ( see udry & santos 2007 ) . robinson et al . ( 2006 ) found host stars to be overabundant in si and ni , but this has not been confirmed so far ( see gonzalez & laws 2007 ) . the only clear evidence of a difference has been reported in haywood ( 2008a ) , where it was shown that at intermediate metallicities ( [ fe / h]@xmath4 - 0.3 ) , most stars known to harbor giant planets belong to the thick disk rather than to the thin disk ( in the ratio 10/2 ) . we come back to this point in the next section . in any case , there is no reason to suspect that the metal - rich host stars are not being affected by radial mixing and that their origin would be different from other stars of the same metallicities . is there direct evidence of radial mixing among planet host stars ? 1a shows the pericenter - apocenter distribution for stars in the gcs ( geneva - copenhagen survey ) catalogue ( nordstrm et al . 2004 ) as a density - surface plot . the analysis of metallicity distribution of the stars in this plot shows that metal - poor thin disk populates preferentially the upper right part of the diagram ( r@xmath58 kpc ) , or outer orbits , while the metal - rich stars occupy mainly inner orbits ( see haywood 2008b , fig . this is interpreted as an effect of radial mixing . the metal - poor stars show a specific kinematic signature , having a component in the direction of rotation significantly larger than the lsr ( haywood 2008b ) . a similar signature have been obtained by schoenrich & binney ( 2009 ) in modeling the effect of churning , which basically allows stars to swap between circular orbits of different angular momentum . their fig . 3 shows that within a few gyr , the mean orbital radius of a star can change by several kpc , both inward or outward . so we do expect that stars coming from inside the solar circle ( respectively outside ) to populate inner ( outer ) orbits . more details on the observational signatures and other consequences are given in haywood ( 2008b ) . sellwood & binney ( 2002 ) describe the effect of churning while schoenrich & binney ( 2009 ) discuss the chemical evolution aspects . red symbols on fig.1a represent known host - planet stars for which orbital parameters from the gcs catalogue are available . the asymmetry in the distribution of orbital parameters is clearly apparent in the stars with detected planets , with the overwhelming number of objects ( 79% ) having r@xmath6 8 kpc . the histograms on the right illustrate the metallicity distribution for the two groups of stars . it clearly shows that stars with planets are subject to a differentiation , the group with r@xmath78 kpc having a metallicity distribution similar to dwarfs hosting planets in general , while the group at r@xmath88 kpc has a metallicity centered on [ fe / h]=0 . it shows that planet hosts follow the general behavior of the metal rich population and that the specific high metallicities can reasonably be attributed to a common origin in the inner disk . after the discovery that the presence of jovian planets is more frequent around metal - rich stars , several studies have explored how the formation of giant planets could be favored in a circumstellar disk enhanced in metals ( ida & lin 2004 , mordasini et al . the results of the previous section now leads to the following preliminary question : how do we know that the higher percentage of giant planets detected on metal - rich stars is due to their metallicity and not to some other factor also linked with their origin in the inner disk ? the question is relevant , because any measurable property of inner disk stars other than metallicity would be correlated with the presence of planet . the obvious a priori response is that metallicity is a measurable parameter , and intrinsic to the star . but there could be others however , which , although not measurable on the stars , could be no less important , such as , for example , the surface density of molecular hydrogen in the inner galactic disk regions . we now show that there are two cases where the planet - metallicity correlation is not verified , for which radial mixing provides a simple explanation , suggesting that metallicity may not be the relevant parameter . we will show in the next section that a bona - fide planet - metallicity correlation can be obtained in the context of radial mixing with no metallicity dependence of any kind . the first case where the planet - metallicity correlation breaks down is the metallicity distribution of giant host stars . while the planet - host dwarf metallicity distribution is known to be skewed towards metal - rich objects , giant hosts are known to have a metallicity which is more like the field stars distribution , see in particular takeda et al . ( 2008 ) and pasquini et al . ( 2007 ) . figure [ fig : amr ] shows the age - metallicity relation for field dwarfs and giants as derived by takeda ( 2007 ) , and takeda et al . ( 2008 ) . the progressive enlargement of the metallicity with age is well in accord with the effect of radial mixing , as noted in haywood ( 2006 , 2008b ) . the metallicity dispersion is smaller for giants : this is expected given their age distribution , fig . [ fig : amr ] showing that most giants have ages smaller than 1 gyr . since radial mixing is a secular process , its effect increases with time : the contamination by stars from the inner and outer disk is proportional to age . because the sample of giants contains mostly young stars , it is little polluted by old , metal - rich , wanderers of the inner disk . the squares on fig . [ fig : amr ] show the planet host giant stars from takeda et al . ( 2008 ) , completed with the few other massive ( [email protected] m@xmath9 ) objects available from the exoplanet database . the figure illustrates that these objects , when older than about 2 gyr , are mostly metal - rich ( [ fe / h]@xmath10.0 dex ) , while the 14 stars younger than 2 gyr have a mean metallicity of -0.04 dex . in the sample of takeda et al . , 7 host giants ( out of 10 ) are younger than 1 gyr , and all are younger than 3 gyr . the explanation for the difference between the dwarf and giant distributions comes out as a natural galactic effect : the giant sample contains a limited bias towards metal - rich objects because it is much younger than the dwarf sample , and then much less contaminated by radial mixing . pasquini et al . ( 2007 ) have suggested that the mass of the convective envelope could play a role . if the excess of metals is due to pollution at the surface of stars , it could be diluted when the dwarf becomes a giant . we propose instead that the excess of metals is intrinsic to the star , and that the age is the determining factor , producing a selective effect on the origin of the stars . at intermediate metallicities ( -0.7@xmath4[fe / h]@xmath4 - 0.3 dex ) , stellar populations in the solar vicinity can be divided into two groups : the thin and the thick disks , which differentiate both by their @xmath10-elements content and their asymmetric drift . at these metallicities , the thin disk is solar in @xmath10-elements , but rotates faster than the lsr , while the thick disk is enriched in @xmath10-elements ( [ @xmath10/fe]@xmath10.1 dex ) but lags the lsr . while the local metal - rich stars may be attributed to migration from the inner disk , the metal - poor end can be attributed to stars that came from the outer disk ( see haywood 2008b ) . it has been shown in haywood ( 2008a ) that in this metallicity interval , giant planets are found preferentially on thick disk stars . this is illustrated in fig . [ fig : fehvrot ] , where 10 stars with giant planets are compatible with being either thick disk or transition objects between the thin and the thick disks . only 1 dwarf , hd 171028 , and 1 giant , hd 170693 , are compatible with being a member of the metal - poor thin disk with an origin in the outer disk . as commented in haywood ( 2008a ) , this is significant , because the number of thin disk objects at these metallicities is expected to be higher or equal to the number of thick disk stars . on fig . [ fig : fehvrot]a , 6 objects having [ fe / h]@xmath4 - 0.2 dex are thin disk objects ( smaller symbols below the line on plot ( a ) ) . the rotation lag and @xmath10-element content of these stars ( plot b ) support the view that they are bona - fide solar radius objects , with no specific indication that they would come from the outer disk . the search of new giant planet hosts in this metallicity range with no bias in favor or against either the thin disk and the thick disk is highly desirable to confirm this trend , but we think the difference between the two groups is significant . finally , it should be noted that the galactocentric radii of origin of thick disk stars ( those with [ fe / h]@xmath4 - 0.2 dex and [ @xmath10/fe]@xmath10.15 dex on fig . [ fig : fehvrot ] ) is not clear . according to schoenrich & binney ( 2009 ) , we should expect the most metal - rich ( at [ fe / h]@xmath1 - 0.4,-0.5 dex ) and @xmath10-elements enhanced thick disk objects to come from the inner disk . this could be the case in particular for hip 3497 , hip 26381 , hip 58952 , hip 62534 , which all have metallicities above [ fe / h]=-0.5 , and relatively high level of @xmath10 abundance . this is an interesting possibility since in this eventuality even thick disk objects could originate from the inner disk . if metallicity was the determining factor for the presence of giant planet , we should not expect a difference between the number of planet host stars of the thin and thick disks . since the metal - poor thin disks objects are expected to come from the outer disk , it is again suggested that the distance to the galactic center plays a role . we are now facing the following picture : stars that come from the inner disk are noticeably rich in giant planets , while stars that come from the outer disk seems to be less favored in this respect . this new information changes considerably how we envisage the correlation between metallicity and the presence of giant planets . for the surprising point here is not the fact that most host - planet stars are metal - rich , since they come from a region where most stars are metal - rich , but the very fact that most would come from the inner disk . we are led to conclude that the distance to the galactic center must somehow play a role in setting the percentage of giant planets , with two new questions : ( 1 ) what of the correlation between stellar metallicity and planet , and ( 2 ) what is the parameter linked to the galactocentric distance which could influence the percentage of giant planets ? \(1 ) in this new scheme , the well - admitted correlation between metals and the presence of giant planets could be the mere reflection of the galactic origin of the stars , but does not necessarily implies an effect of the metallicity on the formation of giant planets . to illustrate this prediction , we make the following simple estimate . we adopt a local ( e.g for stars born at the solar galactocentric radius ) gaussian metallicity distribution centered on [ fe / h]=-0.1 dex , with intrinsic dispersion 0.1 dex . we assume that about 4% ( grenon 1989 ) of the stars at the solar radius come from the inner disk , sampling a metallicity distribution centered on [ fe / h]=+0.35 , with dispersion + 0.2 dex ( a higher dispersion takes into account , in a simplified way , the fact that stars come from different inner radii , and therefore from regions where the mean metallicity is not strictly 0.35 dex ) . given a metallicity gradient of about 0.07 to 0.1 dex.kpc@xmath11 , which is about what is measured ( maciel & costa 2008 ) a mean metallicity [ fe / h]=+0.35 dex can be expected towards the galactic center at about 3 to 5 kpc from the sun . we assume that the percentage of host - planet stars in the inner disk is 25% ( as measured on the most metal - rich objects of the solar neighborhood ) and _ independent _ of metallicity . we also assume a 4% of metal - poor stars centered on [ fe / h]=-0.4 dex , with a dispersion in metallicity of + 0.2 dex , with no giant planets . finally , an error on measured metallicities is simulated with a random gaussian with 0.15 dex dispersion . the metallicity distribution generated with these parameters is given in fig.[fig : fehdist ] ( a ) . on plot ( b ) , we show the proportion of stars with giant planets obtained with our assumptions . the thick line is the fit made by udry & santos ( 2007 ) on the observational distributions ( 3.01@xmath1210@xmath13}$ ] ) . as can be seen , a good correlation between the presence of a giant planet and the metallicity of the stars is obtained , providing an honest fit to the observed rate . \(2 ) some factor linked to galactocentric distance , but not metallicity , must play a role in setting the percentage of giant planet . a candidate could be the density of dust in the inner disk , because dust is thought to favor the formation of planetesimals . however , there is , as yet , no evidence for a difference between the distributions of dust and metals in our galaxy , so that we do not expect dust to lead to different patterns than metallicity . a better candidate is molecular hydrogen . it is foremost a fundamental ingredient for the formation of giant planets , being the principal constituent of stellar disks and jupiters . its main structure in the galaxy , the molecular ring , is thought to contain 70% of h@xmath14 gas inside the solar circle ( clemens et al . 1988 , jackson et al . 2006 ) , thereby providing a huge reservoir for star ( h@xmath14 is known to be directly linked to star formation ( kennicutt 2008 ) ) and planet formation . the most interesting aspect however , is the fact the molecular ring reaches a maximum density at 3 - 5 kpc from the sun , corresponding to the distance where stars with metallicity in the range ( + 0.3,+0.5 ) dex are expected to be formed preferentially . interestingly , the mean surface or volume density of h@xmath14 at a galactocentric distance of 4 to 5 kpc is 2 to 5 times its local value ( nakanishi & sofue 2006 ) , in proportion with the rate of giant planets on metal - rich ( 25% ( roughly @xmath15 10% ) ) and solar metallicity stars ( 4% ) . a final indication may comes to support our views . stars hosting only neptunian and/or super - earth should be less prone to having an origin in the inner disk , if they can form in environment less dense in h@xmath14 . which means that we should not expect a predominance of metal - rich stars among neptunian / super - earth host stars . among the 12 objects on which neptunian or super - earth planets have been discovered , 7 with no jovian planets have metallicities -0.28 , -0.33 , -0.31 , -0.31 , -0.05 , -0.1 , and -0.15 ( gj 674 , gl 581 , hd 4308 , hd 40307 , hd 69830 , hd 285968 , hd 7924 ) according to the exoplanet database . the 5 stars also harboring jovian planets ( hd 75732 , gl 876 , hd 47186 , hd 160691 , hd 181433 ) have metallicities + 0.29 , -0.12 , + 0.23 , + 0.28 and + 0.33 dex , amply confirming the possibility that the first group of stars could be genuine solar radius objects , and the second wanderers from inside the galaxy . bensby t. , feltzing s. , lundstrm i. , ilyin i. , 2005 , a&a , 433 , 185 clemens d. p. , sanders d. b. , scoville n. z. , 1988 , apj , 327 , 139 ecuvillon a. , israelian g. , pont f. , santos n. c. , mayor m. , 2007 , a&a , 461 , 171 gilli g. , israelian g. , ecuvillon a. , santos n. c. , mayor m. , 2006 , a&a , 449 , 723 gonzalez g. , laws c. , 2007 , mnras , 378 , 1141 grenon m. , 1989 , ap&ss , 156 , 29 haywood m. , 2006 , mnras , 371 , 1760 haywood m. , 2008a , a&a , 482 , 673 haywood m. , 2008b , mnras , 388 , 1175 ida s. , lin d. n. c. , 2004 , apj , 616 , 567 jackson j. m. , et al . , 2006 , apjs , 163 , 145 kennicutt r. c. , jr . , 2008 , aspc , 390 , 149 maciel w. j. , costa r. d. d. , 2008 , arxiv , arxiv:0806.3443 mordasini c. , alibert y. , benz w. , naef d. , 2008 , aspc , 398 , 235 nakanishi h. , sofue y. , 2006 , pasj , 58 , 847 nordstrm b. , et al . , 2004 , a&a , 418 , 989 pasquini l. , dllinger m. p. , weiss a. , girardi l. , chavero c. , hatzes a. p. , da silva l. , setiawan j. , 2007 , a&a , 473 , 979 quillen a. c. , minchev i. , bland - hawthorn j. , haywood m. , 2009 , arxiv , arxiv:0903.1851 reddy b. e. , tomkin j. , lambert d. l. , allende prieto c. , 2003 , mnras , 340 , 304 reddy b. e. , lambert d. l. , allende prieto c. , 2006 , mnras , 367 , 1329 robinson s. e. , laughlin g. , bodenheimer p. , fischer d. , 2006 , apj , 643 , 484 schoenrich r. , binney j. , 2008 , arxiv , arxiv:0809.3006 sellwood j. a. , binney j. j. , 2002 , mnras , 336 , 785 takeda y. , 2007 , pasj , 59 , 335 takeda y. , sato b. , murata d. , 2008 , pasj , 60 , 781 udry s. , santos n. c. , 2007 , ara&a , 45 , 397 wielen r. , 1977 , a&a , 60 , 263	the correlation between stellar metallicity and the presence of giant planets is well established . it has been tentatively explained by the possible increase of planet formation probability in stellar disks with enhanced amount of metals . however , there are two caveats to this explanation . first , giant stars with planets do not show a metallicity distribution skewed towards metal - rich objects , as found for dwarfs . second , the correlation with metallicity is not valid at intermediate metallicities , for which it can be shown that giant planets are preferentially found orbiting thick disk stars . none of these two peculiarities is explained by the proposed scenarios of giant planet formation . we contend that they are galactic in nature , and probably not linked to the formation process of giant planets . it is suggested that the same dynamical effect , namely the migration of stars in the galactic disk , is at the origin of both features , with the important consequence that most metal - rich stars hosting giant planets originate from the inner disk , a property that has been largely neglected until now . we illustrate that a planet - metallicity correlation similar to the observed one is easily obtained if stars from the inner disk have a higher percentage of giant planets than stars born at the solar radius , with no specific dependence on metallicity . we propose that the density of @xmath0 in the inner galactic disk ( the molecular ring ) could play a role in setting the high percentage of giant planets that originate from this region .
in recent years , large area avalanche photodiodes ( laapd ) have been successfully applied as photodetectors for a variety of scintillators @xcite , including liquid xenon ( lxe ) @xcite . lxe scintillation light is in the vuv region , centered at 178 nm @xcite , which makes it difficult to detect efficiently . within the context of the r&d for the xenon dark matter project @xcite , we are testing different photodetectors directly coupled to lxe , including uv sensitive photomultipliers ( hamamatsu r9288 and r8520 ) , microchannel plate pmts ( burle 85006 ) , silicon photomultiplier ( sipm ) @xcite and laapds from advanced photonix , inc.(api ) . in this paper we report our first results on the detection of lxe scintillation with a 16 mm diameter laapd , mounted on a custom designed ceramic support @xcite . the interest in laapds for lxe is in their high quantum efficiency ( qe ) at 178 nm , as originally reported by @xcite . their negligible radioactive contamination is also attractive for low background experiments based on lxe , such as direct dark matter searches ( xenon @xcite and zeplin @xcite ) and neutrinoless double beta decay searches ( exo @xcite ) . on the other hand , for a practical application of laapds , especially for lxe based dark matter searches , several issues remain to be addressed . the primary concern is the capability of a light readout based on laapds to achieve the low energy threshold of a few tens of kev required for high sensitivity . other issues include long term stability of operation , impact on lxe purity , as well as readout complexity and cost . the experiments presented in this paper aimed at confirming the high qe for lxe scintillation reported in the literature , and at verifying the compatibility of laapds immersed in the liquid with the requirement to detect tiny ionization signals , at the single electron level . the 16 mm diameter , windowless laapd tested in these experiments was mounted inside a gridded ionization chamber , filled with high purity lxe . a photograph of the assembled electrodes and laapd is shown in fig . [ apd_s ] . the cathode is a 6 cm diameter stainless steel plate with either a gamma ray source ( @xmath1bi ) or an alpha source ( @xmath2am ) deposited on its center . the shielding grid and the anode grid are made with stretched wires on stainless steel frames , with a thickness of about 0.8 mm . the wire spacing is 2 mm , and the wire diameter is 60 @xmath3 m . the separation between cathode and grid is 1 cm , which defines the maximum drift gap for ionization electrons . the grid to anode separation is 3 mm . the electrodes spacing was maintained with rings made of ptfe , for its high reflectivity in the uv region @xcite . the laapd , originally glued by api on a thin ceramic substrate , was mounted on a ptfe disk , facing the anode grid at a distance of 6 mm above . we note that the laapd had been exposed to air for several months prior to its use in lxe . hermetic feedthroughs are used for cathode and grid hv lines and for anode signal readout via a charge sensitive amplifier . additional feedthroughs are used to bias the laapd and to connect its output to a separate charge sensitive amplifier . a schematic drawing of the detector system and electronics readout is shown in fig . [ daq ] . the ionization electrons , which are created from gamma rays or alpha particles in the drift region , are drifted upward by the applied electric field , and are collected by a charge sensitive pre - amplifier ( clearpulse model 580 @xcite ) connected to the anode . the scintillation light hits the laapd and produces photoelectrons , which are amplified by the avalanche process . the avalanche electrons are collected by an amptek 250 pre - amplifier @xcite . the charge and light signals , either from clearpulse 580 or amptek a250 , are fed into a low - noise shaping amplifier ( ortec 450 @xcite ) . the amplified signals are further fed into a pc - based multi - channel analyser ( mca ) for spectroscopy analysis . known test pulses are used to calibrate the daq system , for both light and charge signals . the capacitances in the pre - amplifiers were calibrated with a silicon detector . we used an open bath cooling apparatus with a liquid nitrogen and alcohol mixture to condense the xenon gas . the vessel enclosing the assembled detector was filled with high purity lxe , covering completely the laapd . as shown in fig . [ daq ] , we used a high temperature getter to purify xe gas continuously via a diaphragm pump . the system , developed for xenon prototypes , is described in @xcite . the gain of the laapd was measured in a different set - up configuration , which optimizes the light collection efficiency by placing the laapd very close ( about 3.5 mm ) to the source plate , at -95@xmath4 ( fig . [ gain ] ) . the direct scintillation from a 5.5 mev alpha source ( @xmath2am ) was measured as a function of applied voltage on the laapd up to about 1500 v. unitary gain was determined from the average amplitude in the 300 to 500 v range @xcite . the typical reduction in applied voltage , for a constant gain , when reducing temperature @xcite was observed , corresponding to an average voltage variation of about 2.4 v/@xmath4 at a gain of 100 . the scintillation light spectrum of the @xmath1bi radiation in lxe was measured , as shown in fig . [ bi - spe ] . due to the small liquid xenon sensitive volume , the energy from most of the 1064 kev gamma rays of @xmath1bi is not fully deposited . the second peak on the spectrum is mostly contributed by the 976 kev internal conversion electrons . as the gamma rays interact at any point in the liquid xenon active volume , light collection in the laapd varies accordingly to the interaction position . using ptfe reflectors , the variation in light collection can be reduced to less than 1% , without compromising the energy resolution . the energy resolution for the 976 kev electrons of @xmath1bi is 7.5% ( @xmath0 ) , which is comparable to the energy resolution obtained earlier by using a pmt in the same chamber with similar geometry @xcite . the spectrum was accumulated at zero electric field to maximize light output from lxe . to better evaluate the laapd performance in liquid xenon , larger amounts of scintillation light are necessary . the 5.5 mev alpha particles from @xmath2am provide typically one order of magnitude larger scintillation pulse and their interaction is very localized in liquid xenon , contributing to a clear scintillation light spectrum . to get the best possible energy resolution , the gain of laapd and shaping time on the amplifier were optimized and finally a gain of 57 was selected for this measurement as shown in fig . [ alpha - spe ] . a very good energy resolution of 2.6% ( @xmath0 ) ( with ptfe walls ) and 3.0% ( @xmath0 ) ( without ptfe walls ) was obtained . to compare the performance of laapd with other photon detection devices , the alpha spectrum was also measured using a 2-inch diameter hamamatsu r9288 pmt with less than 20% qe at 178 nm wavelength . a value of 2.5% energy resolution was obtained . the energy resolution as a function of laapd gain can be written as follows , @xmath5 the first term in this equation is contributed from electronic noise . @xmath6 is the noise equivalent charge , which can be measured from the distribution of a known test pulse . @xmath7 is the number of primary electron - hole pair produced from the scintillation light in the photodiode surface . @xmath8 is the laapd gain . the second term is from the fluctuations of the gain and is inherent to the electron avalanche process of the laapd , where @xmath9 is the excess noise factor . the experimental value for @xmath9 is approximately written as @xmath10 @xcite for @xmath11 . the third term @xmath12 is contributed from the scintillation process in the liquid xenon , including statistical fluctuations of scintillation photon production , and photoelectrons created in the laapd . using the gain measurement with the laapd at 3.5 mm from the source plate , we fit the results with the above equation with @xmath6 and @xmath12 as free parameters ( fig . [ res_vs_g ] ) , while fixing @xmath9 from the experimental value @xcite . the noise equivalent charge @xmath6 from the fit agrees well with the measured value with a calibrated test pulse . from the fitted values , we infer that the statistical fluctuations contribute 1.8% to the energy resolution , with a contribution of about 1.6% ( @xmath13 , @xmath14 ) from fluctuations in the number of photoelectrons . the quantum efficiency ( @xmath15 ) of a laapd can be calculated by the following equation , @xmath16 here @xmath7 is the number of photoelectrons from the laapd , @xmath17 is the number of photons reaching the laapd surface , @xmath8 is the laapd gain , @xmath18 is the number of electron charges detected by the pre - amplifier and @xmath19 is the total number of scintillation photons produced by an event . @xmath19 is approximately equal to @xmath20 , where @xmath21 is the energy of the event , and @xmath22 is the average energy required to produce a scintillation photon in liquid xenon . the @xmath22 values are 21.6 ev @xcite and 19.6 ev @xcite for gamma and alpha events respectively . @xmath23 is the light collection efficiency , which is defined as the percentage of the total lxe light yield reaching the laapd surface . we have estimated the light collection efficiency by using a light tracing simulation program in geant4 @xcite with the assumptions listed in table 1 . the estimated light collection efficiency from the simulation is 7.0 @xmath24 0.7% for the structure with 1 cm thick ptfe wall between the cathode and grid . the error indicates the different reflectivity values used in the simulations . in the case of no ptfe walls , the light collection efficiency was calculated by simply using the solid angle for alpha particles and ignoring the reflectivity of stainless steel . as result , 3.3% light collection efficiency was obtained . considering a 20% stainless steel reflectivity , 4.1% light collection efficiency was obtained . in order to estimate the number of photons incident on the laapd surface , an average value for light collection efficiency was used , as presented in table 2 . from the above considerations , we estimated the qe of the laapd for different measurements , which are shown in table 2 . the main uncertainty is from the estimation of the light collection efficiency . the different values obtained from electron and alpha events may be due to the uncertainty in @xmath22 and the laapd gain values . in conclusion , we use the average of the qe values from the electron and alpha measurements , which is @xmath25 . the liquid xenon scintillation light yield depends on the strength of the applied electric field @xcite . the dependence of the scintillation yield of liquid xenon for alpha particles has been measured by the columbia group several years ago , using an external pmt coupled to a lxe volume via a caf@xmath26 window @xcite . in the current setup , we were able to measure this field dependence with a laapd . for comparison , we also measured this field dependence with a hamamatsu r9288 pmt immersed in liquid xenon , in the same chamber . [ field_dep ] shows the combined results . the laapd gain varies considerably with temperature , and our simple alcohol - ln@xmath26 cooling bath does not keep the liquid temperature stable enough to avoid temperature dependent gain variations . the gain of the pmt is not much affected by such small temperature fluctuations . the data from the laapd and the pmt are in good agreement , but the result with the laapd has more fluctuations due to its acute temperature dependence , which is obvious from the curve . one challenge involving photon detection devices immersed in liquid xenon is their compatibility with the high purity required for electron drift , if a combined charge and light readout is implemented . our experience with lxe detectors shows that the light yield of lxe is not very sensitive to the purity level , unlike the charge yield . many efforts have been made in the past to detect both ionization and scintillation , using pmts , in liquid xenon @xcite . currently we have developed a xenon recirculation and purification system @xcite , which continuously removes impurities from the liquid xenon during experiment . depending on the size of the detector , we can achieve a sufficient purity level for drifting ionization electrons within two to tens of hours . during the first experiment with the laapd immersed in the lxe , we used this recirculation system to purify the xenon , continuously . we measured the 976 kev peak position of the @xmath1bi to monitor the charge collection . within a few hours of recirculation and purification , we achieved a high charge collection of 75% . the ionization spectrum of @xmath1bi at 1 kv / cm drift field shown in fig . [ bi_qspe ] is comparable to that measured in a liquid xenon ionization chamber with an external pmt @xcite . from the second experiment , we observed the same level of charge collection even without using the recirculation system , which indicates that the laapd is clean and does not bring any impurities into the liquid xenon . in this paper , we have demonstrated the operation of a laapd in liquid xenon to detect scintillation light from gamma rays and alpha particles . we have achieved the best energy resolution , for alpha particles and fast electrons , from liquid xenon scintillation light detected by a laapd . the inferred value of the qe is 45% , which is lower than that previously reported in @xcite . a recent article also reports a higher qe value from a measurement in gas xenon @xcite . we have repeated the measurements with the laapd sample and confirm that the results are reproducible . a possible explanation for the lower qe is the intrinsic quality of the laapd . this explanation appears consistent with the subsequent qe measurement of the same laapd at room temperature , using an indepedent setup . the qe was inferred from the ratio of the current output from the laapd and a calibrated pin diode , by irradiating uv light from a xenon lamp to the devices @xcite . the measured value ( @xmath27 ) is very close to what we measured in liquid xenon . we believe that the qe of this device should not change significantly with temperature . we intend to test a new laapd of the same type as used in @xcite to investigate if the qe we measured was specific to the sensor used . the gain of the laapd is much lower than that of pmts . using an external amplification system limits the energy detection threshold , which is very crucial for a sensitive dark matter detector . the scintillation light from a wimp recoil of a few tens of kev energy may produce not enough light to be detected by the laapd . on the other hand , properties of laapds , such as their compact size , very high qe , and compatibility with low radioactive background and high lxe purity requirements , make them attractive for applications of lxe detectors in particle physics , medical imaging and astrophysics . this work was supported by a grant from the national science foundation to the columbia astrophysics laboratory ( grant no . phy-02 - 01740 ) . one of the authors ( p. majewski ) acknowledges the support by the north atlantic treaty organization under a grant awarded in 2003 . lopes , et al . , ieee trans . vol.48 , no . 3 , ( 2001 ) 312 . m. kapusta , et al . , ieee trans . 4 , ( 2002 ) 1693 - 1398 . m. moszynski , et al . instr . and meth . a 497 ( 2003 ) 226 . solovov , et al . instr . and meth . a 488 ( 2002 ) 572 . j. jortner , et al . , j. chem . phys . 42 ( 1965 ) 4250 . e. aprile , et al . , astro - ph/0207670 . e. aprile , et al . , physics/0501002 . a. bolozdynya , private communication . n. smith , proceedings of the 4th international workshop on the identification of dark matter , york , 2002 , eds . n. spooner and v. kudryavtsev . exo experiment , see online http://www - project.slac.stanford.edu / exo/. m. yamashita , et al . instr . meth . a 535 , 692 - 698 ( 2004 ) . http://www.ortec - online.com/. e. aprile , et al . , astro - ph/0407575 . fernandes , et al . , ieee trans . nucl . 4 , ( 2004 ) 1575 - 1580 . solovov , et al . , ieee trans . 4 , ( 2000 ) 1307 - 1310 . m. yamashita , ph.d . thesis , `` dark matter search experiment with double phase xe detector '' , pp.42 , waseda university , japan , july 2003 . fernandes , et al . instr . meth . a 531 ( 2004 ) 566 - 568 . t. doke , et al . 41 ( 2002 ) 1538 - 1545 . t. doke , et al . instr . and meth . a 420 , 62 - 80 ( 1999 ) . s. agostinelli et al . , nucl . instr . meth . a , 506 ( 2003 ) , 250 - 303 . a. baldini , et al . , physics/0407033 . seidel , et al . instr . and meth . a 489 ( 2002 ) 189 - 194 . e. aprile , et al.,ieee trans . 2 , ( 1990 ) 553 - 558 . e. aprile , et al.,ieee trans . 50 , no . 5 , ( 2003 ) 1303 - 1308 . e. aprile , et al.,ieee trans . 51 , no . 5 , ( 2004 ) 1986 - 1990 . e. aprile , et al . a 302 ( 1991 ) 177 - 185 . r. chandrasekharan , et al . , physics/0412155 . p. shagin , university of minnesota , private communication . t. takahashi , et al . rev . a , vol.12 , pp.1771 , 1975 . bi at zero electric field . laapd gain is about 146 and the temperature is -95@xmath4 . the peak of the 976 kev electron line is fitted at about channel 330 , giving a resolution of 7.5% ( @xmath0 ) . the 570 kev gamma line can also be seen clearly at about channel 190.,scaledwidth=90.0% ] am scintillation light energy spectrum from 5.5 mev alpha particles detected by the laapd in lxe . the lxe temperature is -95@xmath4 and laapd gain is 57 . the alpha peak is fitted with a gaussian function . the energy resolution is 2.6% for the case with the ptfe wall ( left ) and 3.0% for the case without the ptfe wall ( right ) . the peaks are normalized to the energy of the alpha particles.,scaledwidth=90.0% ] , and p1 is @xmath12 in eq.1 . the noise contributions from electronic noise ( dashed line , first term in eq.1 ) and excess noise factor term ( thick line without data points , second term in eq.1 ) , from the fitting parameters , as functions of laapd gain are also plotted . , scaledwidth=90.0% ] am 5.5 mev alpha particles . the measurements were carried out with either a pmt or a laapd , immersed in liquid xenon at around -95@xmath4 . the error bars on the laapd data are from the laapd gain variation due to temperature fluctuations.,scaledwidth=90.0% ] bi radiation in liquid xenon at 1 kv / cm drift field with an laapd immersed in the liquid . the peaks are from 570 kev gamma rays ( channel 244 ) and 976 kev internal conversion electrons ( channel 486 ) . the sharp peak on the right of the charge spectrum is from a known test pulse . based on the peak values , we estimated the charge collection at about 75% , by assuming the average energy needed to produce an electron ion pair in liquid xenon is 15.6 ev @xcite . the energy resolution for the 570 kev gamma rays and 976 kev electron peaks are 9.5% and 6.5% separately.,scaledwidth=90.0% ]	scintillation light produced in liquid xenon ( lxe ) by alpha particles , electrons and gamma - rays was detected with a large area avalanche photodiode ( laapd ) immersed in the liquid . the alpha scintillation yield was measured as a function of applied electric field . we estimate the quantum efficiency of the laapd to be 45% . the best energy resolution from the light measurement at zero electric field is 7.5%(@xmath0 ) for 976 kev internal conversion electrons from @xmath1bi and 2.6%(@xmath0 ) for 5.5 mev alpha particles from @xmath2am . the detector used for these measurements was also operated as a gridded ionization chamber to measure the charge yield . we confirm that using a laapd in lxe does not introduce impurities which inhibit the drifting of free electrons . , , , , , , laapd , liquid xenon , dark matter
observational and theoretical efforts to understand the emission from high energy x - ray sources ( e.g. agn and black hole x - ray binaries ) have left no doubt about the existence of astrophysical plasmas with electron temperatures of order @xmath1 . as soon as the first hot accretion flow model was constructed ( shapiro , lightman & eardley 1976 , hereafter sle ) it was realized that the production and annihilation of @xmath0 pairs may play an important role in such solutions . since then , enormous progress has been made in our understanding of the nature of @xmath0 equilibria , first in the context of static , isothermal hot plasma clouds ( e.g. lightman 1982 ; svensson 1982 , 1984 ; sikora & zbyszewska 1986 ; kusunose 1987 ; bjrnsson & svensson 1991a ) , and later for more realistic accretion disk models ( e.g. kusunose & takahara 1988 , 1989 , 1990 ; tritz & tsuruta 1989 ; white & lightman 1989 , 1990 ; bjrnsson & svensson 1992 ; bjrnsson et al . 1996 ; kusunose & mineshige 1996 ) . until now , most treatments of pair equilibria ( with the exception of the papers by zdziarski [ 1986 ] , kusunose & takahara [ 1989 ] , and kusunose & zdziarski [ 1994 ] ) have considered non - magnetic plasmas , in which the only photon production mechanisms are bremsstrahlung , double compton scattering and pair annihilation ( several authors include also an external soft photon source , e.g. zdziarski et al . [ 1994 ] ) . however , there exists now a consensus that magnetic fields are universal in accretion flows and play an important role in their energy balance . in the presence of reasonably strong magnetic fields , synchrotron radiation dominates over any other photon production mechanism at mildly relativistic temperatures ( @xmath2 ) , and the presence of synchrotron photons can change significantly the results obtained for non - magnetized plasmas . an external source of photons , as in the case of a hot corona above a thin disk ( e.g. haardt & maraschi 1991 , hereafter hm ) , will also modify the standard @xmath0 pair equilibria . recently , models for agn and galactic x - ray binaries based on pair - dominated coronae have been extensively discussed in the literature ( e.g. liang 1991 ; zdziarski et al . 1994 ; skibo et al . 1995 ; stern et al . 1995a , 1995b ; poutanen , krolik & ryde 1997 ) . however , to our knowledge , most authors have concentrated on numerical computations of spectra , and have not given enough attention to more basic questions such as the existence and properties of pair equilibria for a range of relevant parameters and assumptions . kusunose & mineshige ( 1991 ) investigated a two - temperature disk corona , but they explored a relatively narrow parameter range . it is also not clear how their results are related to one - temperature pair - dominated coronae or a popular scenario ( suggested originally by hm ) in which part of the disk viscous energy is dissipated in the corona . moreover , since in their model the corona is heated through its own differential rotation , the dynamics of the coronal gas must be treated self - consistently , taking into account the radial energy advection term , especially for the high accretion rates considered in the paper . in this paper we systematically investigate the existence and properties of pair - dominated equilibrium solutions for three different scenarios : one - temperature static thin disk coronae , heated by the viscous dissipation of the gravitational energy stored in the disk , two - temperature coronae , and hot accretion flows , similar to those introduced by sle and ichimaru ( 1977 , see also narayan & yi 1994 and abramowicz et al . 1995 ) . the latter problem was already discussed by bjrnsson et al . ( 1996 ) and kusunose & mineshige ( 1996 ) . our treatment is different only in that we include synchrotron cooling in our calculation , which was left out by the previous authors . the detailed discussion of our initial assumptions and basic equations used for each of the three scenarios are contained in [ basiceq ] . we find that pair - dominated solutions are easy to obtain only in the case of one - temperature coronae , while both two - temperature coronae and hot accretion flows always have negligible pair densities ( see [ res ] ) . these results are discussed in [ disc ] . the main conclusions are summarized in [ sum ] . following bj@xmath3rnsson & svensson ( 1991 , 1992 ) we divide the problem of finding equilibrium accretion flow solutions in the presence of @xmath0 pairs into two main parts , one dealing with radiative transfer and pair balance in a static plasma cloud , and the other describing the details of the dynamics and energetics in an accretion flow . bj@xmath3rnsson & svensson ( 1991 ) have shown that the physical properties of a non - magnetized plasma cloud are completely determined by specifying only two parameters . these are the proton optical depth , @xmath4 , and the dimensionless luminosity , or compactness parameter , @xmath5 , where @xmath6 is the size of the cloud , @xmath7 is the proton ( i.e. ionized electron ) number density , and @xmath8 is the radiative emissivity per unit volume . in this ( non - magnetic ) case , the mapping between the two parts of the calculation is particularly simple . the accretion flow is treated as a sequence of plasma clouds , whose size is equal to the scale height of the flow . at each point in the flow one needs only to specify @xmath9 and @xmath10 in terms of the usual accretion parameters : the mass of the accreting object @xmath11 , the radius @xmath12 , where @xmath13 , and the accretion rate @xmath14 , normalized to the eddington value @xmath15 . in this paper , we attempt to treat a more general problem in which the accretion flow contains strong magnetic fields , which lead to non - negligible cooling via synchrotron radiation . we also allow for external heating and cooling sources which are relevant in a corona heated through dissipation of the thin disk gravitational energy and cooled via comptonization of thin disk radiation . when these effects are properly taken into account , in addition to the parameters @xmath10 and @xmath9 , one needs to specify also the particle density @xmath7 or , equivalently , the cloud size @xmath6 , to completely determine the properties of a plasma cloud . with this modification , the mapping between the cloud and flow properties is no longer as straightforward as in bj@xmath3rnsson & svensson ( 1991 , 1992 ) , and depends on the details of the initial assumptions . nevertheless , the mapping method is still a convenient technique for solving the problem of electron - positron pairs in accretion flows . pair equilibrium requires that @xmath16 where @xmath17 is the positron number density , and the rate of pair annihilation on the left hand side is balanced by the sum of pair production rates via @xmath18@xmath18 , @xmath18@xmath19 , @xmath20@xmath20 , @xmath20@xmath18 , and @xmath20@xmath19 collisions , respectively , on the right hand side . we neglect pair escape from the cloud . the analytical expressions for the pair annihilation and electron - electron pair creation rates in a thermal plasma are given by svensson ( 1982 ) and white & lightman ( 1989 ) . according to svensson ( 1982 ) , the @xmath18@xmath19 rate can be ignored , since it is generally negligible compared with the @xmath18@xmath18 rate . the last three terms in equation ( [ pairb ] ) depend on the photon spectral density and , therefore , involve the details of the radiative transfer which we discuss below . in a magnetized plasma , thermal synchrotron and bremsstrahlung are the most important photon production mechanisms ( in this paper we ignore double compton scattering ) . in addition , there might be an external source of soft photons , e.g. blackbody emission from a thin disk irradiating a hot corona . the expressions for thermal bremsstrahlung emissivity are taken from svensson ( 1982 ) . the self - absorbed synchrotron emission is modeled using the formalism of mahadevan et al . ( 1995 ) and narayan & yi ( 1995 ) . the magnetic field strength , @xmath21 , is calculated under the assumption that the gas pressure is in equipartition with the magnetic field pressure in the cloud . for a tangled isotropic magnetic field , this implies that [ bfield ] = n_p m_e c^2 ( _ p + ( 1 + 2 z ) _ e ) , where @xmath22 and @xmath23 are dimensionless proton and electron temperatures ( both in electron mass energy units ) , and @xmath24 is the pair fraction . the second term on the right side of equation ( [ bfield ] ) takes into account the pressure from pairs , as well as ionized electrons . in general , most of the primary photons are too soft to produce pairs . however , when inverse compton scattering is important , many of the soft photons are boosted up to the wien peak . in that case collisions between photons in the wien peak as well as between wien and bremsstrahlung photons can dominate pair - production . we follow the approach of svensson ( 1984 , hereafter s84 ) and consider the radiation field as a sum of a flat bremsstrahlung continuum and a wien distribution . in this special case , the photon - particle and photon - photon pair production rate equations can be integrated analytically , and the resulting expressions depend only on the number density of the wien photons , @xmath25 . in order to compute @xmath25 , we solve the photon balance equation [ photonb ] f_b n_^b + f_s n_^s + f_d n_^d = n_/t_esc , where @xmath26 , @xmath27 , and @xmath28 are the rates of bremsstrahlung , synchrotron , and external soft photon production respectively , and @xmath29 , @xmath30 , and @xmath31 represent the fractions of emitted photons that are scattered into the wien peak before escaping from the cloud . the details of how we compute these quantities are discussed in appendix [ apa ] . finally , we approximate the photon escape time as @xmath32 ( s84 ) , where @xmath33 is the total scattering optical depth of the cloud , and the factor @xmath34 incorporates the klein - nishina effect , averaged over the wien photon distribution . for fixed @xmath35 , @xmath10 and @xmath7 , equations ( [ pairb ] ) , ( [ bfield ] ) , and ( [ photonb ] ) can be solved to obtain the equilibrium pair fraction @xmath24 and the cloud compactness , @xmath9 , which in general includes contributions from comptonized bremsstrahlung and synchrotron emission , as well as comptonization of the external soft radiation ( see appendix [ apa ] ) . as was pointed out by s84 , in non - magnetized plasmas the results are practically independent of @xmath7 . the only dependence comes from the determination of the bremsstrahlung self - absorption energy , @xmath36 , defined so that at energies below @xmath36 , the local bremsstrahlung spectrum is a blackbody . the value of @xmath36 is computed using the expression derived by s84 . since the bremsstrahlung energy spectrum is nearly flat , the total emission , as well as the pair production rate , are dominated by the high - energy end of the spectrum , and have at best a logarithmic dependence on @xmath36 , and therefore , @xmath7 . in the presence of magnetic fields , however , the value of the proton number density becomes more important , because it determines the strength of the magnetic field ( see equation ( [ bfield ] ) ) as well as the value of the synchrotron self - absorption energy , @xmath37 , computed as described in esin et al . ( 1996 ) . as opposed to the case of bremsstrahlung radiation , synchrotron emission is generally dominated by the self - absorbed part , so that the dependence on @xmath37 is no longer negligible . to compare our calculations with previous work , in fig . [ zfig ] we plot the equilibrium pair fraction ( panel a ) and corresponding compactness ( panel b ) as functions of the gas temperature @xmath38 for a single - temperature plasma cloud of a fixed size . there is no magnetic field , so that bremsstrahlung is the only source of soft photons . different curves correspond to different values of @xmath10 . the results are very similar to those of svensson ( 1982 , 1984 ) . for each value of @xmath10 , there are two equilibrium solutions for temperatures below some critical value , @xmath39 ; one branch of the solutions ( indicated by the solid lines ) is pair - dominated , the other ( short - dashed lines ) is practically pair - free . above @xmath40 , the pair production rate is greater than the pair annihilation for any value of @xmath41 and no equilibrium solution is possible . the critical gas temperature is itself a function of the proton optical depth , increasing with decreasing @xmath10 , until it reaches a maximum at @xmath42 . the pair - dominated solution branch exists only for @xmath43 , where @xmath44 is the critical temperature at which the pair fraction reaches its thermal equilibrium value , and the compactness reaches the blackbody limit ( the dashed line in fig . [ lfig](b ) ) . at low @xmath10 , the equilibrium values of @xmath41 on the pair - free solution branch converge to a single curve ( fig . [ zfig](a ) ) , since in this regime , pair production is dominated by particle - particle collisions which have the same scaling with density as the pair annihilation rate ( they both vary as @xmath45 ) . the compactness on the low-@xmath41 branch is dominated by bremsstrahlung and as a result , varies as @xmath45 ( fig . [ lfig](b ) ) . the high-@xmath41 branch , on the other hand , is dominated by photon - photon processes ; as a result the pair density is proportional to @xmath46 , while the compactness is independent of the proton optical depth , converging to the wien equilibrium values for @xmath47 . the only significant difference between our results and those of s84 is that we overestimate the pair fraction in the temperature range @xmath48 , where double compton scattering , which we neglect here , becomes an important source of photons . [ zsfig ] shows how the pair equilibria in a plasma cloud change in the presence of magnetic fields . the magnetic field strength depends on both the electron and proton temperatures ( see equation ( [ bfield ] ) , and for simplicity the results shown in fig . [ zsfig ] were calculated under the assumption that @xmath49 . a comparison of figs . [ zfig ] and [ zsfig ] shows that synchrotron cooling significantly changes the shapes of both @xmath41 and @xmath9 pair equilibria curves . on the low-@xmath41 branch , the cloud compactness increases strongly for @xmath50 as compared to the pure bremsstrahlung case . synchrotron cooling causes the critical temperature to decrease considerably in denser clouds , with @xmath51 . by contrast , on the high-@xmath41 branch the compactness is practically unchanged , since it follows the wien equilibrium curve . but because of the extra cooling per lepton at temperatures where synchrotron emission is important ( @xmath52 ) , the equilibrium pair fraction in a magnetic plasma is up to @xmath53 times smaller than in non - magnetic plasmas . this figure also illustrates the point made previously , that the properties of pair equilibria in magnetized plasmas are _ not _ independent of the cloud size . for a fixed proton optical depth , a larger cloud has smaller @xmath7 . therefore , the value of the equipartition magnetic field is smaller as well , reducing the importance of synchrotron cooling . a comparison between the short - dashed curve computed for a cloud of size @xmath54 and a corresponding curve with the same value of @xmath10 but @xmath55 shows this effect . it is clear that the dependence on @xmath6 is not negligible , though it is not very strong . the low-@xmath41 compactness branch is affected the most , and an increase in @xmath6 by seven orders of magnitude , causes the value of @xmath9 to decrease by at most a factor of @xmath56 . to relate the pair equilibrium solutions calculated for static plasma clouds to a realistic accretion scenario , we must express @xmath7 , @xmath10 , and @xmath9 in terms of the accretion parameters @xmath57 and @xmath58 , and impose an appropriate energy balance condition . below we describe these mapping relations for two different accretion flow geometries : a corona above a thin disk ( [ onetcor ] , [ twotcor ] ) and a hot , optically thin accretion flow ( [ adaf ] ) . first we consider the simplest scenario , namely a static , hot , one - temperature corona above a cool , geometrically thin , optically thick disk . the study of such coronae was pioneered by hm . the corona is heated through viscous dissipation of the gravitational energy stored in the thin disk ( perhaps via magnetic reconnection , e.g. field & rogers 1993 ) , and in turn heats the disk by irradiation . using the formalism of hm , it is easy to show ( see appendix [ apb ] ) that the total emission per unit area from the thin disk , @xmath59 , is equal to @xmath60 here @xmath61 is the total viscous dissipation per unit area of the thin disk ( e.g. shakura & sunyaev 1973 ; frank , king , & raine 1992 ) with mass accretion rate @xmath62 , @xmath63 is the fraction of viscous energy dissipated directly in the corona , @xmath64 is the total scattering optical depth in the corona , @xmath65 is the fraction of radiation emitted or scattered in the corona that is incident on the thin disk , and @xmath66 is the fraction of incident radiation that is reflected by the disk . in general , @xmath65 and @xmath66 depend on the temperature and optical depth of the corona , the ionization state of the thin disk , and the spectrum of irradiating flux . however , since their exact values are not important in light of other approximation made in this work , we adopt the values @xmath67 and @xmath68 suggested by hm . this prescription constrains the disk emission to be always in the range @xmath69 , regardless of the exact values of @xmath63 and @xmath64 . ( note that @xmath59 is the disk emission as seen by the corona , not by a remote observer . ) we assume that the thin disk radiates as a black body with a color temperature @xmath70 to estimate the characteristic size of the corona , we assume that it is in hydrostatic equilibrium , so that @xmath71 where @xmath6 is the scale height of the corona , @xmath72 is the mass density , and @xmath73 includes contributions from the magnetic and gas pressures . it is interesting to note that though in pair - free plasma the contribution of electrons to the mass density in the corona is entirely negligible , if the pair fraction is large enough , namely @xmath74 , pairs must be properly taken into account . this allows us to solve for the scale height : @xmath75},\ ] ] where @xmath76 is the ratio of the magnetic field pressure to the total pressure in the corona . in plasmas with no magnetic fields @xmath77 . when magnetic fields are taken into account we always assume equipartition between the gas and magnetic field pressure , corresponding to @xmath78 . knowing @xmath6 , allows us to write the energy balance equation for the corona , by equating compactness @xmath9 and viscous dissipation ( converted into a dimensionless compactness - like quantity ) : @xmath79 to compute the compactness @xmath9 of the corona , we include contributions from comptonized bremsstrahlung and synchrotron radiation , as well as inverse compton scattering of the thin disk photons ( see [ ppbal ] and appendix [ apa ] for details ) . once the parameters describing the accretion disk ( @xmath80 , @xmath58 , @xmath81 ) and the corona ( @xmath63 , @xmath10 ) are fixed , we can solve equations ( [ pairb ] ) through ( [ otcorep ] ) to obtain the equilibrium pair fraction and temperature of the corona . we begin by solving pair and photon balance equations for a range of gas temperatures @xmath82 . as in the case of a static plasma cloud described in [ ppbal ] above , we obtain two equilibrium curves for @xmath83 and @xmath84 , slightly modified due to the dependence of the scale height @xmath6 on @xmath85 and @xmath41 , and the increase in cooling through comptonization of the thin disk photons . then we calculate the dimensionless viscous dissipation rate ( the right hand side of equation ( [ otcorep ] ) ) as a function of temperature , again obtaining two solutions for each value of @xmath85 . finally , we solve for @xmath85 such that equation ( [ otcorep ] ) is satisfied , i.e. the cooling of the gas in the corona is balanced by viscous dissipation . the results are described in [ rescor1 ] . depending on the microphysical details of how viscosity dissipates gravitational energy , it is feasible that only the protons in the corona are heated directly ( e.g. sle , rees et al . since protons generally can not radiate , viscous energy must be transfered to the electrons via coulomb collisions . the coulomb energy transfer is proportional to the difference between the electron and proton temperatures , so for the gas to be in thermal equilibrium we must have protons hotter than electrons , i.e. @xmath86 . such two - temperature plasmas have been often invoked in the literature in the context of hot accretion flows ( sle ; rees et al . 1982 ; narayan & yi 1995 ; nakamura , kusunose , matsumoto & kato 1997 ; di matteo , blackman & fabian 1997 , to give just a few examples ) and it is important to investigate how the results for single - temperature coronae are modified when proton - electron energy transfer is explicitly taken into account . the only difference between one- and two - temperature coronae is that the latter must satisfy two energy balance equations instead of one . energy balance for the protons requires that viscous heating is equal to the coulomb energy transfer : @xmath87 for the electrons to be in thermal equilibrium , we must have @xmath88 the quantity @xmath89 is the rate of coulomb energy transfer per unit volume . in our calculations we use the expression for @xmath89 derived by stepney & guilbert ( 1983 ) @xmath90,\end{aligned}\ ] ] where @xmath91 is a modified bessel function of order @xmath92 , @xmath93 , and the coulomb logarithm is taken to be @xmath94 . to obtain equilibrium solutions for a two - temperature corona , we begin by solving equations ( [ pairb])([h ] ) , ( [ ttcorp ] ) , and ( [ qie ] ) for a range of @xmath35 , to obtain @xmath95 , @xmath96 and @xmath97 . imposing electron energy balance then allows us to determine @xmath35 , such that equation ( [ ttcore ] ) is satisfied . this procedure is very similar to that described in [ onetcor ] , except that we have an extra energy balance equation , which allows us to solve for @xmath35 and @xmath98 separately . in addition to static coronae above a thin disk , we also investigate pair production in pure optically thin , hot accretion flows . such flows were first proposed by sle and ichimaru ( 1977 ) and later extensively studied by narayan & yi ( 1994 , 1995 ) , abramowicz et al . ( 1995 ) , chen ( 1995 ) , chen et al . ( 1995 ) and others . recently , kusunose & mineshige ( 1996 ) and bjrnsson et al . ( 1996 ) addressed the problem of pairs in the context of these flows . in this paper we extend their calculations to include the effects of strong thermal synchrotron cooling . as in two - temperature coronae , the protons in hot accretion flows are heated by viscous dissipation of gravitational energy , which is parameterized in the usual way through the viscosity parameter @xmath99 . however , we allow for the possibility that only a fraction @xmath100 of the dissipated energy is transfered to the electrons via coulomb collisions . the rest of the energy is stored in the gas as entropy and advected inward with the accretion flow . when the accreting object is a black hole , the stored energy is then lost inside the horizon . with these assumptions , the energy balance equation for the protons takes the form @xmath101 where @xmath102 is the viscous dissipation rate per unit area of the flow , while the energy equation for the electrons is the same as in [ twotcor ] above : @xmath103 to compute the rate of viscous dissipation , as well as all other relevant quantities in the accreting gas ( e.g. @xmath7 , @xmath6 , equipartition magnetic field strength @xmath21 , and total pressure @xmath73 ) , we use the self - similar solutions derived by narayan & yi ( 1994 , 1995 ) . these solutions allow us to express various gas properties as functions of the usual input parameters , @xmath80 , @xmath58 , @xmath57 , and @xmath99 as well as the new advection parameter @xmath104 . as opposed to the two - zone models discussed above , in hot accretion flows the mass accretion rate @xmath57 , which determines the rate of viscous dissipation , and the proton optical depth @xmath10 , which specifies the cooling rate of the gas , are not independent parameters . consequently , to obtain equilibrium accretion flow solutions it is sufficient to specify @xmath80 , @xmath57 , @xmath58 and @xmath99 . we solve for equilibrium solutions in the following manner . for a fixed @xmath104 , the scale height of the flow and the characteristic sound speed at a given radius @xmath58 are completely determined by the self - similar solutions . this allows us to solve for the electron temperature @xmath35 for which the pair balance and electron energy balance equations , ( [ pairb])([photonb ] ) and ( [ adafe ] ) , are satisfied . we then solve for the value of @xmath104 at which the protons are in thermal equilibrium , determined by equation ( [ adafp ] ) . the aim of this work is to explore whether equilibrium pair - dominated accretion flow solutions exist for different flow geometries and initial assumptions . we present here the results of our calculations for the three models described in [ enbal ] . [ corfig ] illustrates how equilibrium solutions for a one - temperature corona above a thin disk are determined . on it , we plot the compactness of the corona in pair equilibrium , @xmath105 , and the corresponding dimensionless viscous dissipation rate as functions of the gas temperature @xmath85 . as in the case of a static plasma cloud , both cooling and heating rates have two solutions , which we refer to as the high-@xmath41 and low-@xmath41 solutions respectively , for each value of @xmath85 . to emphasize the distinction between the high-@xmath41 and low-@xmath41 solutions , the former are plotted in fig . [ corfig ] as solid and dashed curves , while the latter are shown by dotted and dot - dashed curves . energy balance requires that cooling must be balanced by viscous heating ; therefore , the point where the two curves intersect , marked by a circle in the figure , is an equilibrium solution for the corona . because the plot shows two different solution branches simultaneously , it is important to remember that valid solutions correspond only to the intersections of solid lines with dashed lines or dotted lines with dot - dashed lines . the other intersections do not give a solution , since the crossing lines correspond to different values of @xmath41 . on panel ( a ) we show our calculations for proton optical depth @xmath106 , while panel ( b ) corresponds to @xmath107 . the values of other relevant parameters are given in the figure caption . note that the relative slopes of the cooling and heating curves are such that for any choice of @xmath57 and @xmath10 there is only one intersection point . we do not find multiple solutions for any choice of parameters . we are interested in pair - dominated solutions , i.e. we wish to find regions of the parameter space in which the high-@xmath41 branches of cooling and viscous heating have an intersection point . one major trend is clear already from comparing the two panels in fig . [ corfig ] ; we see that for a fixed mass accretion rate in a thin disk , a corona with a relatively high proton optical depth is pair - free ( fig . [ corfig](a ) ) , while a lower value of @xmath10 yields a pair - dominated solution ( fig . [ corfig](b ) ) . this trend is easy to understand . the rate of viscous heating is determined by the mass accretion rate in the disk and is relatively independent of @xmath10 in the corona ( if we ignore the slight dependence through the vertical scale height ) , while the cooling rate on the low-@xmath41 branch , dominated by comptonization of the thin disk radiation , decreases linearly with @xmath10 . thus , for fixed @xmath81 , @xmath58 , and @xmath63 , we can find a critical proton optical depth , @xmath108 , below which the corona is pair - dominated , i.e has @xmath109 . these maximum values of @xmath10 for pair - dominated coronae are plotted in fig . [ taufig](a ) as a function of @xmath81 . the solid line shows our results for a plasma with equipartition magnetic fields . for comparison , we plot @xmath108 for non - magnetized coronae as a dashed line . the two lines are practically identical at high @xmath57 , where energy balance in the corona is dominated by comptonization of disk photons , but begin to diverge at lower @xmath57 where internal cooling in the corona becomes more important . the other three panels in fig . [ taufig ] illustrate the dependence of @xmath108 on different input parameters . [ taufig](b ) shows how @xmath108 varies as a function of @xmath63 , the fraction of the gravitational energy of the gas in the thin disk that is dissipated in the corona . lower @xmath63 means that for the same value of @xmath81 the corona is heated less . to compensate for this , we are driven to lower values of @xmath108 to ensure that equation ( [ otcorep ] ) is satisfied for @xmath110 . increasing @xmath58 , the radial distance from the accretor , has a similar effect on @xmath108 at larger radii the viscous dissipation rate decreases , which leads to @xmath108 being smaller for a given @xmath81 ( see fig . [ taufig](c ) ) . finally , fig . [ taufig](d ) shows that the results are very similar for low - mass and high - mass accreting black holes . the slight difference between the two curves in due to the fact that the mass of the central object determines the physical scale ( @xmath6 ) of the accretion flow ( see discussion in section [ ppbal ] and fig . [ zsfig](b ) ) . in studying pair - dominated corona solutions , two quantities of special interest are the equilibrium temperature and the total optical depth to electron scattering , @xmath33 , since they determine to a large degree the shape of the emitted spectrum . [ hfig ] shows the dependence of these parameters on @xmath10 for different values of the thin disk mass accretion rate . on every curve , the solution with @xmath110 ( which corresponds to @xmath111 ) is marked with a solid dot . to the right of the dot , pairs are not important . here the scattering optical depth of the corona is simply equal to @xmath10 , and the gas temperature increases with decreasing optical depth , to keep up with the cooling . in this regime , our @xmath112 curve is qualitatively similar to the results of hm , except that for each value of @xmath81 , there is a maximum allowed @xmath10 , for which a corona can still be in thermal equilibrium . above this value , the bremsstrahlung and synchrotron emissivity of the gas in the corona is larger than viscous heating , and thermal equilibrium is not possible . for @xmath113 , on the other hand , the radiative transfer in the corona is dominated by pairs . in this regime , @xmath64 initially _ increases _ with decreasing @xmath10 , due to copious pair production . correspondingly , @xmath85 decreases , since pairs more than make up for dropping numbers of primary ionized electrons . in the limit @xmath114 , the gas in the corona reaches wien equilibrium , where @xmath115 , and as a result , both @xmath64 and @xmath85 converge to constant values , determined solely by @xmath81 . a single temperature corona is gravitationally bound when the gas temperature is below the virial value , which we obtain by equating the thermal energy and gravitational energy per proton in the gas : @xmath116.\ ] ] when @xmath117 the gas in the corona simply escapes from the system on the dynamical time scale . thus , the solutions calculated here are consistent only for @xmath118 . this requirement is equivalent to restricting the scale height of the corona ( see equation ( [ h ] ) ) to @xmath119 . to check whether this condition is satisfied , we have plotted in fig . [ hfig](c ) @xmath120 curves corresponding to equilibrium corona solutions . the thin solid line is drawn for @xmath121 , the stability limit for a corona with equipartition magnetic fields ( @xmath122 ) . only those parts of the curves that lie below this line correspond to gravitationally bound solutions . we have seen earlier that the solutions with @xmath123 exist only to the left of the dots . in other words , pair - dominated and stable coronae exist when @xmath124 and @xmath113 . these two conditions are mutually exclusive for @xmath125 . the dot - dashed curves in fig . [ hfig ] show the properties of accretion disk coronae around a supermassive black hole , computed for @xmath126 . as we have seen already , the results are not very sensitive to the black hole mass and there are no qualitative differences between @xmath127 and @xmath128 . the gas in a two - temperature corona must satisfy both electron and proton energy balance equations , which require that in thermal equilibrium , compactness , viscous heating and coulomb energy transfer rate are all equal to one another . first we concentrate on the balance between the two latter quantities , which did not play any role in single - temperature coronae . the viscous dissipation rate is determined by the mass accretion rate in the thin disk and @xmath63 , and is independent of the amount of matter in the corona . the rate of proton cooling via coulomb collisions , however , varies as a product of the proton and electron@xmath129positron densities in the corona , i.e. it is proportional to @xmath130 . on the low-@xmath41 branch , pair density is negligible , so the coulomb energy transfer rate scales simply as @xmath45 . on the high-@xmath41 branch , we have @xmath115 , so that this quantity is linear in @xmath10 . in either case , the protons can be in thermal equilibrium only for relatively high values of @xmath10 . on the other hand , in [ rescor1 ] we showed that electron energy conservation in pair - dominated corona solutions requires relatively low values of @xmath10 . it would appear that the two conditions are mutually exclusive . of course in this simplistic argument we have neglected the dependence of both proton cooling and heating rates on particle temperatures and pair density . to get the exact answer we have solved for two - temperature corona solutions numerically , following the procedure described in [ twotcor ] , exploring the entire @xmath131 parameter space . the results are shown on fig . [ twotfig ] for two values of @xmath63 . equilibrium solutions are allowed only in the shaded region above the solid ( @xmath132 ) or dashed ( @xmath133 ) line . a comparison between figs . [ twotfig ] and [ taufig](b ) clearly shows that these regions lie far above the @xmath108 line , and therefore , must have very few pairs . and indeed , we find that in every case , when an equilibrium corona solution is found , it corresponds to @xmath134 . in other words , we have been unable to find equilibrium pair - dominated two - temperature corona solutions for any choice of parameters . our conclusions disagree with those of kusunose & mineshige ( 1991 ) , the only authors , to our knowledge , who investigated @xmath0 pair equilibrium states in a two - temperature corona . this discrepancy stems most likely from different assumptions made in the two studies about the energy source in the corona . in our model , the coronal gas is heated by dissipating the gravitational energy of material in the disk , while kusunose & mineshige considered the situation where viscous heating is due to differential rotation of the corona itself . in a sense , their model corresponds to an accretion flow with external soft photon irradiation , rather than a canonical thin disk corona . in fact , the pair equilibrium solutions found by kusunose & mineshige ( 1991 ) are very similar to those discovered previously by kusunose & takahara ( 1988 , 1990 ) and white & lightman ( 1989 ) in their study of an sle accretion disk . figs . [ adaf1fig ] and [ adaflfig ] demonstrate how we solve for equilibrium solutions in two - temperature hot accretion flows . at a fixed @xmath104 , @xmath57 , and @xmath58 , the scale height of the accreting gas is fully determined , and therefore , we can compute @xmath41 , @xmath9 , and @xmath135 ( the dimensionless coulomb energy transfer rate ) as functions of the electron temperature . compactness and coulomb energy transfer are plotted in fig . [ adaf1fig ] as solid@xmath129dotted and dashed@xmath129dot - dashed lines respectively . as usual , we have two branches for each value of @xmath35 , one with negligible pair density ( dotted and dot - dashed lines ) , and the other dominated by pairs ( solid and dashed lines ) . the transition between the two branches occurs at @xmath136 on panel ( a ) and at @xmath137 on panel ( b ) . the heating and cooling of the electrons ( and pairs ) in the flow are balanced only at the intersection points marked in the figure . the two panels show that we can find both high-@xmath41 and low-@xmath41 solutions to equation ( [ adafe ] ) , depending on the values of @xmath104 and @xmath57 . it remains now to ensure that the proton energy balance equation ( [ adafp ] ) is satisfied . since the viscous dissipation rate has the strongest dependence on the advection parameter @xmath104 , in fig . [ adaflfig ] we plot it ( solid line ) as a function of @xmath104 together with the equilibrium proton cooling rate ( dashed lines ) , obtained by considering electron energy balance as discussed above . the results displayed on panels ( a ) , ( b ) , and ( c ) differ only by their values of @xmath57 , listed in the figure caption . it is clear from fig . [ adaflfig ] that there are at most two values of @xmath104 for which the gas is in thermal equilibrium . the two equilibrium points correspond to the well known solution branches for optically thin accretion flows : the cooling - dominated branch ( @xmath138 ) discovered by sle , and the advection - dominated branch ( @xmath139 ) proposed by ichimaru ( 1977 ) and rediscovered by narayan & yi ( 1994 , 1995 ) and abramowicz et al . [ adaflfig](c ) shows that at high accretion rates , both solutions disappear . this feature of optically thin accretion flow solutions was pointed out previously by many authors ( e.g. narayan & yi 1995 ; chen et al . 1995 ) . the question we ask here is whether the properties of the optically thin accretion flow solutions are modified because of the presence of electron - positron pairs . in fig . [ adaflfig ] , the cooling rate corresponding to the low-@xmath41 compactness branch ( see fig . [ adaf1fig ] ) is indicated by long - dashed lines ; short - dashed lines show high-@xmath41 compactness curves . the actual values of the pair fraction @xmath41 corresponding to the compactness curves are plotted in fig . [ adafzfig ] . as before , solid dots mark thermal equilibrium solutions . note that in every case these equilibria have @xmath140 , implying that pairs do not play an important role in these accretion flows . to compare these results to previous work on pairs in optically thin accretion flows , it is instructive to plot the thermal equilibrium solutions for optically thin accretion flows at a fixed radius on a @xmath141 plane . the resulting curves computed for three different values of @xmath99 are shown in fig . [ sigmafig ] ( thick lines ) . for comparison , thin lines show equilibrium solutions computed assuming @xmath142 . consistent with the findings of kusunose & mineshige ( 1996 ) and bjrnsson et al . ( 1996 ) , the effects of pairs become more pronounced for larger values of @xmath99 ; and are practically negligible for @xmath143 . this is mainly because pair production is more efficient in flows with higher density , and therefore , mass accretion rate , and the critical @xmath57 above which thermal equilibrium solutions do not exist , scales roughly as @xmath144 ( narayan & yi 1995 ; abramowicz et al . however , even for @xmath145 , pairs contribute no more than @xmath146 of the total optical depth ( see fig . [ sigmafig ] ) . in fact , we find that the presence of magnetic fields in the flow reduces the importance of pairs even further , as compared with pure bremsstrahlung solutions , since , for a given @xmath99 , magnetized flows are restricted to lower values of @xmath57 due to extra cooling via synchrotron radiation . though the results displayed in fig . [ sigmafig ] were computed for @xmath127 we find practically identical results for accretion flows around supermassive black holes . from the analysis presented in this paper we conclude that pair - dominated equilibrium solutions exist only in single - temperature disk plus corona models . both two - temperature coronae and single - phase hot accretion flow models have pair fractions @xmath24 significantly below unity . what is the special feature of single - temperature coronae that allows them to be pair - dominated ? the answer seems to be that it has fewer constraints then the other two models . the gas in the corona is heated via viscous dissipation of energy in an underlying thin disk . the rate of such heating is completely determined by the properties of the thin disk ( namely , mass accretion rate @xmath81 and radius @xmath58 ) and is virtually independent of the amount or temperature of matter in the corona , which specify the cooling rate . thus , for a given value of @xmath81 , the energy balance in the corona allows us to solve for either @xmath10 or @xmath85 , but not both . one of these parameters has to be specified as an initial condition . this extra degree of freedom allows us to adjust the mass accretion rate in the disk and the optical depth of the corona to obtain thermal equilibrium solutions with @xmath123 . we find that such solutions are restricted to @xmath113 , where the critical value of the proton optical depth decreases with decreasing @xmath81 ( see fig . [ taufig ] ) . in a two - temperature corona , heating and cooling rates are still relatively independent of each other ; however , proton energy balance imposes an important additional constraint on the model . the proton cooling rate via coulomb scattering between protons and electrons ( as well as pairs ) is sensitive to the number density of particles in the corona , while proton heating through viscous dissipation is independent of @xmath10 . consequently , protons are in thermal balance only for relatively high values of @xmath10 . but at these values of @xmath10 , electron energy balance requires @xmath147 ( fig . [ twotfig ] ) . in the single - phase hot accretion flow model , both viscous heating and radiative cooling rates are determined by the mass accretion rate in the flow . in this model , @xmath57 specifies both the optical depth and temperature of the accreting gas ; one can no longer adjust these variables independently in order to obtain pair - dominated solutions . we find that for a realistic range of values for the viscosity parameter @xmath99 , the pair density in hot accretion flows is negligible , both for the cooling - dominated ( sle ) and the advection - dominated ( ichimaru 1977 , narayan & yi 1994 ) solution branches . in the two - temperature accretion flow solution we have considered , the coulomb energy transfer bottleneck also plays a role in limiting the effects of pairs . even in one - temperature coronae , we find that pair - dominated solutions exist only over a relatively narrow range of the proton optical depth , for each value of @xmath81 ( see fig . [ hfig](c ) ) . it is bounded from above by @xmath108 , imposed by the thermal equilibrium balance equation for the gas , and from below by the requirement that the hot gas in the corona should be gravitationally bound . the latter condition can be somewhat relaxed , if we assume that the corona is held in place not by gravity , but rather by magnetic field loops anchored in the thin disk . in fact , the most popular recent scenario for the dissipation of the thin disk gravitational energy in the corona is based on reconnection of the magnetic field loops generated in the disk and transported by buoyancy into the corona ( galeev , rosner & vaiana 1979 ; field & rogers 1993 ) . in this picture , if the magnetic energy density dominates the energy density of the gas , magnetic fields may prevent the escape of particles from the corona even if their temperature is above virial . on the other hand , if the cooling of the gas in the corona is dominated by inverse compton scattering of the thin disk photons , electrons and positrons can undergo a considerable radiative acceleration along the accretion disk poles ( see for example liang & li [ 1995 ] , li & liang [ 1996 ] ) , which may allow the escape of pairs with temperatures below virial . when we consider a one - temperature accretion disk corona with @xmath148 , an important question to ask is what keeps electrons and protons at the same temperature . if viscous dissipation preferentially heats the protons , as is often assumed in the literature ( e.g. sle ; rees et al . 1982 ; narayan & yi 1995 ) , the electrons will be considerably cooler than the heavier particles . our results for two - temperature coronae clearly show that coulomb energy transfer is too inefficient to equilibrate @xmath35 and @xmath98 . one way to solve this problem is to introduce some other more efficient energy transfer mechanism between the protons and electrons in the corona ( e.g. begelman & chiueh 1988 ) . on the other hand , recent work on dissipation mechanisms for mhd waves by quataert ( 1998 ) and gruzinov ( 1998 ) indicate that in magnetically dominated plasma it is electrons , rather than the protons that are preferentially heated . in this case , the protons might even be cooler than the electrons ( if the accretion timescale is shorter than the coulomb energy transfer timescale ) . however , since it is the latter that contribute to cooling and pair - production , @xmath98 is practically irrelevant , so that the corona can be treated as a single - temperature gas cloud . finally , we would like to emphasize that the results for thin disk coronae presented in this paper apply only to steady - state , homogeneous corona scenarios such as discussed , for example , by haardt & maraschi ( 1991 , 1993 ) , as opposed to time - dependent flare models in which a corona contains many localized active regions ( e.g. haardt , maraschi & ghisellini 1994 ; stern et al . 1995b , nayakshin & melia 1997a , b ; di matteo , celotti & fabian 1997 , 1998 ) . in the latter case , many of the constraints imposed to find the equilibrium solutions need not be satisfied . in particular , the protons may transfer only part of the viscously dissipated energy to the electrons and pairs . in that case , the protons will experience a net heating , until they reach virial temperature and escape from the system , perhaps in the form of an outflow or a jet . after that more material will be evaporated into the corona from the disk , and the cycle will be repeated . similarly , the pairs themselves may be periodically ejected from the corona driven either by radiative acceleration ( liang & li 1995 ) or simply through heating above their virial temperature , in the absence of energy balance . such transient or time - dependent situations are beyond the scope of the paper . we find that the properties of pair equilibria in static plasma clouds , taking into account synchrotron emission and comptonization of external soft photons differ only in quantitative details from the results of s84 , who restricted his study to comptonized bremsstrahlung and double compton emission . for every value of the proton optical depth , @xmath10 , there exists a critical electron temperature @xmath149 below which there are two pair equilibrium solution branches , one of which is pair - dominated , and the other practically pair - free . on the pair - free branch , particle - particle pair production processes dominate and the pair fraction @xmath150 is independent of @xmath10 . by contrast , on the pair - dominated branch the electron ( and positron ) scattering optical depth is constant , so that @xmath115 . the extra sources of photons considered here ( synchrotron and external photons ) do not change the general character of the solutions , but they do decrease the value of @xmath151 so that pair equilibria are restricted to lower temperatures than in s84 . we studied in this paper three popular accretion scenarios involving hot plasmas to assess the importance of pairs . our results for the three models are summarized below : * in the case of a single - temperature corona above a thin disk , the cooling of the gas must be balanced by viscous heating . the latter quantity is determined by the mass accretion rate , @xmath62 , in the thin disk and the fraction @xmath63 of the viscous energy which is dissipated in the corona . the compactness on the low-@xmath41 branch is due primarily to inverse compton scattering of synchrotron and thin disk photons , so that @xmath152 . as a result , for every value of the mass accretion rate in the thin disk , we find that there is a critical @xmath108 in the corona below which the equilibrium solution is pair - dominated ( see fig . [ taufig ] ) . for larger values of @xmath10 , pairs in the corona are unimportant . thus , while pure pair coronae are allowed , such coronae have very little mass and are dominated by energy deposition from the disk . * a two - temperature corona above a thin disk must satisfy two energy conservation equations , one for electrons and one for protons . that is , the coulomb energy transfer rate must balance both the rate of viscous energy input into the protons and the the cooling rate of the electrons . since the coulomb energy transfer is a two - body process , it is proportional to @xmath45 on the low-@xmath41 pair equilibrium branch and goes as @xmath153 on the high-@xmath41 branch . we find that the extra constraint on the energy balance ( compared to a single - temperature corona ) restricts solutions to relatively high values of the proton optical depth . only at such @xmath10 is coulomb coupling efficient enough . however , the critical @xmath10 is generally above the maximum @xmath108 up to which pair - dominated coronae are possible ( see figs . [ taufig ] and [ twotfig ] ) . thus , we conclude that , if viscous dissipation heats primarily the protons in the corona , the equilibrium solutions are generally pair - free . * finally , we have considered pair equilibria in hot accretion flows described by the self - similar solution of narayan & yi ( 1995 ) . we find that the energy transfer bottleneck due to coulomb coupling conspires with the advective cooling in these flows to limit the effects of @xmath154 pairs . we agree with the results of earlier investigators ( bjrnsson et al . 1996 and kusunose & mineshige 1996 ) who concluded that for reasonable values of the viscosity parameter ( @xmath155 ) there are no pair - dominated solutions on either the advection - dominated or cooling - dominated ( i.e. sle ) branches . previous work dealt only with comptonized bremsstrahlung cooling . we have included comptonized synchrotron here , but recover the same results as before with only minor quantitative differences ( fig . [ sigmafig ] ) . i would like to thank ramesh narayan for numerous useful discussions and suggestions . this work was supported in part by national science foundation graduate research fellowship , and by nasa through axaf fellowship grant # pf8 - 10002 awarded by the axaf science center , which is operated by the smithsonian asprophysical observatory for nasa under contract nas8 - 39073 . we treat inverse compton scattering of photons by thermal electrons using the approximate analytical formulae derived by s84 and esin et al . ( 1996 ) . when the plasma cloud is optically thin , @xmath156 , the probability @xmath157 that a photon will suffer exactly @xmath158 scatterings before escaping can be approximated by a poisson formula : @xmath159 where @xmath160 is the mean number of scatterings ( rybicki & lightman 1979 ) . if the initial photon energy , @xmath161 is small compared with the thermal energy of the electrons in the cloud , on average , each scattering will increase the photon energy by a factor @xmath162 ( rybicki & lightman 1979 ) . this process saturates when the photon energy becomes comparable to the average energy of the wien distribution , @xmath163 . thus , the number of scatterings corresponding to saturation is @xmath164 . then the fraction of photons with initial energy @xmath165 that are scattered into the wien peak before escaping is simply @xmath166 where @xmath167 , is the incomplete gamma function ( see esin et al . [ 1996 ] for a detailed derivation of this formula ) . the energy gain through comptonization is characterized by the average energy enhancement factor for each photon , @xmath168 , defined as the ratio of the photon final and initial energies : @xmath169 + \frac{3 \theta_e}{x } p ( j_m+1,s).\ ] ] for an optically thick cloud , @xmath170 , we use a different expression for probability @xmath157 , derived by sunyaev & titarchuk ( 1980 , see also s84 ) , @xmath171 since in this regime each photon will undergo many scatterings before escaping , we can treat @xmath158 as a continuous variable ( the expression for @xmath157 is normalized accordingly ) . then the fraction of photons scattered into the wien peak can be written as an integral : @xmath172}.\ ] ] note that this expression differs slightly from that derived by s84 ( equation c3 ) , since we assume that the average energy of wien photons is @xmath173 , rather than @xmath35 . the average energy enhancement factor for photons is then @xmath174 + \frac{3 \theta_e}{x } f ( x).\ ] ] in our calculations we simply use equations ( [ f ] ) and ( [ xi ] ) when @xmath156 and equations ( [ fp ] ) and ( [ xip ] ) when @xmath175 , with a suitable interpolation across @xmath176 . bremsstrahlung photon emissivity is given by a normalized expression ( s84 ) @xmath177 where @xmath36 is the bremsstrahlung self - absorption energy , computed by comparing the bremsstrahlung and compton scattering absorption coefficients ( for details see appendix d of s84 ) . then the emission weighted fraction @xmath29 is simply @xmath178 while the total energy amplification factor for bremsstrahlung emission is @xmath179 in the optically thick limit , equations ( [ fb ] ) and ( [ xib ] ) can be evaluated analytically ( s84 ) : @xmath180,\ ] ] and @xmath181 where @xmath182 is the emission averaged compton @xmath183-parameter . in the optically thin regime , we integrate equations ( [ fb ] ) and ( [ xib ] ) numerically . the rate of wien photon production due to compton upscattering of bremsstrahlung photons is simply @xmath184 . we use the expression for @xmath185 , volume emissivity of bremsstrahlung photons via @xmath186 and @xmath187 interactions , derived by s84 ( equations a1-a3 and a22a ) . total comptonized bremsstrahlung cooling rate per unit volume is @xmath188 , where @xmath189 is the thermal bremsstrahlung cooling rate per unit volume ( svensson 1982 ) . thermal synchrotron emission is generally strongly self - absorbed , so that most of the emission comes out near the self - absorption frequency @xmath37 , calculated as described in esin et al . when calculating the effect of comptonization , this means we can treat synchrotron emission as a monochromatic source of photons , characterized by a photon production rate @xmath190 . here @xmath191 is the synchrotron cooling rate per unit volume , for which we employ the expressions derived by mahadevan et al . ( 1995 ) and esin et al . . then the rate of wien photon production due to upscattering of synchrotron photons is @xmath192 . similarly , total comptonized synchrotron cooling rate is just @xmath193 . since thin disk emission has a thermal spectrum , we can treat it in exactly the same way as the synchrotron emission as a monochromatic source of photons with energy @xmath194 . the characteristic temperature @xmath195 is given by equation ( [ td ] ) . then the rate of production of the thin disk photons per unit volume of the corona is @xmath196 , where cooling rate per unit area of the disk , @xmath59 , is given by equation ( [ disken ] ) . consequently , we can write the rate of upscattering of disk photons into the wien peak as @xmath197 . the cooling rate of gas in the corona through inverse compton scattering of thin disk photons is then @xmath198 . following hm we approximate disk and corona as two uniform adjacent slabs , characterized by different temperatures and densities . the rate of gravitational energy release per unit area is @xmath102 , of which a fraction @xmath63 is dissipated directly in the corona , while the remaining @xmath199 is dissipated within the thin disk . the cold optically thick disk radiates as a blackbody , with a total flux @xmath59 . on the other hand , the optically thin corona cools by emitting comptonized synchrotron and bremsstrahlung radiation , as well as through inverse compton scattering of the thin disk photons . total cooling rate per unit ares of the corona can be written as @xmath200 , where the first term stands for the internal coronal emission and @xmath168 is amplification factor due to comptonization of the thin disk emission . the energy balance for the corona requires that @xmath201 we assume further that a fraction @xmath65 of all photons emitted of scattered in the corona is directed downwards towards the thin disk , while the remaining radiation escapes towards the observer . then the total flux incident onto the thin disk can be written as @xmath202,\ ] ] where @xmath64 is the optical depth of the corona to electron scattering , and @xmath203 is the fraction of the thin disk photons scattered at least once before escaping . this incident radiation is partly absorbed by the thin disk , heating it further , and partly reflected . we write the absorbed fraction as @xmath204 . note that equation ( [ qinc ] ) above differs from the corresponding expression in hm in two ways . firstly , we include the internal emission from the corona . secondly , we take into account the fact that the corona not only amplifies thin disk emission through compton scattering , it also acts as a reflector for the thin disk photons . in the limit when comptonization is not important , i.e. @xmath205 , the presence of the corona still raises the equilibrium temperature of the disk through the greenhouse effect . with this , we can write down the energy balance equation for the disk : @xmath206 = q_d.\ ] ] combining equations ( [ corbal ] ) and ( [ diskbal ] ) , we obtain the expression for total disk emission , which is independent of the details of inverse compton scattering in the corona : @xmath207	we investigate equilibrium accretion flows dominated by @xmath0 pairs . we consider one- and two - temperature accretion disk coronae above a thin disk , as well as hot optically thin two - temperature accretion flows without an underlying thin disk ; we model the latter in the framework of advection - dominated accretion flows ( adafs ) . in all three cases we include equipartition magnetic fields . we confirm the previous result that the equilibrium density of pairs in two - temperature adafs is negligible ; and show that the inclusion of magnetic fields and the corresponding synchrotron cooling reduces the pair density even further . similarly , we find that pairs are unimportant in two - temperature coronae . even when the corona has significantly enhanced heating by direct transfer of viscous dissipation in the thin disk to the corona , the inefficient coulomb coupling between protons and electrons acts as a bottleneck and prevents the high compactness required for pair - dominated solutions . only in the case of a one - temperature corona model do we find pair - dominated thermal equilibria . these pair - dominated solutions occur over a limited range of optical depth and temperature . = 11 versim#1#2 = 12
the realization of spinor bose - einstein condensation in optical traps@xcite has stimulated enormous interest in magnetic properties of quantum bose gases@xcite . in optical traps , the hyperfine degree of freedom of confined atoms , such as @xmath0 , is released and therefore the atom can exhibit magnetism . more intriguingly , an exchange - like spin - spin interaction can be present between atoms . in the @xmath1 @xmath0 atoms , the interaction is ferromagnetic@xcite , so the @xmath0 gas appears to be a prototype of itinerant - boson ferromagnet@xcite . ferromagnetism is one of the central research themes in condensed matter physics @xcite . two types of ferromagnetism have already been intensively studied : local - moment ferromagnetism and itinerant - electron ferromagnetism . although particles in these two systems obey different statistics , they both share some common features . for example , both ferromagnets have a curie point , above which the susceptibility conforms to curie - weiss law . nonetheless , from the theoretical point of view , the origin of curie - weiss law is quite different for these two systems . in insulators it is due to local thermal spin fluctuations and can be easily explained in the mean - field approximation . on the other hand , in itinerant - electron ferromagnets the curie - weiss law may be caused by the mode - mode coupling between spin fluctuations and the theoretical treatment is much more complicated @xcite . an appropriate theory is the self - consistent renormalization ( scr ) theory @xcite which goes beyond the hartree - fock approximation and the random - phase approximation . the scr theory succeeds in explaining various magnetic properties of itinerant - electron ferromagnets and is also extended to treat the specific heat @xcite . the @xmath0 gas provides opportunity to study the third type of ferromagnetism . ho @xcite , ohmi and machida @xcite have studied its ground state properties and the spin - wave spectrum . the long wavelength spectrum is linear in @xmath2 , the wave vector , as in the two former cases . in our previous papers , we have investigated the finite - temperature properties , especially the curie point @xcite . we suggest that the phase diagram in itinerant bosons should be more complicated than the other two ferromagnets , because the bose system has an intrinsic phase transition , other than the ferromagnetic transition . an interesting conclusion we arrived is that its curie point , @xmath3 , is never below the bose - einstein condensation temperature , @xmath4 , regardless of the magnitude of the ferromagnetic coupling @xcite . kis - szabo _ et al _ got the same point later @xcite . however , thermodynamics of the itinerant - boson ferromagnet has not yet been investigated systematically so far . the purpose of this paper is to calculate the thermodynamic quantities of ferromagnetic bosons . as in the fermion case , the specific heat and magnetic susceptibility are of the most interest . in section 2 , we introduce the mean - field approximation to deal with ferromagnetic interaction , taking the spin-1 bose gas as an example . in section 3 , phase transitions are discussed by calculating the free energy and specific heat . in section 4 , the susceptibility above the curie point is calculated . a summary is given in the last section . the spin-1 bose gas with ferromagnetic couplings is described by the following hamiltonian , @xmath5 where @xmath6 is the quantum field operator for annihilating an atom in spin state @xmath7 at site @xmath8 . for a spin-1 gas , @xmath9 . the parameter @xmath10 denotes the external magnetic field . the last term represents the ferromagnetic exchange between two different bosons meeting at site @xmath8 and @xmath11 is the exchange constant . @xmath12 are the spin operators , which can be expressed via the @xmath13 pauli matrices , for example , @xmath14 within the mean - field approximation , we treat the spin - dependent interactions as a molecular field except of a particle with itself , @xmath15 where @xmath16 is the ferromagnetic order parameter . then the effective hamiltonian for the grand canonical ensemble reads , @xmath17 \hat n_{{\bf k}\sigma } + \frac{1}{2}\overline{m}^2i_s n , \ ] ] where @xmath18 is the kinetic energy for free particles , @xmath19 is called the molecular field , similar to the stoner theory for fermion gases@xcite ; @xmath20 is the chemical potential ; @xmath21 is the total particle number . the grand thermodynamic potential can be worked out in a standard way , @xmath22\nonumber\\ & = & -\frac{(k_bt)^{\frac{5}{2}}vm^{\frac{3}{2}}}{(2\pi\hbar^2)^{\frac{3}{2}}}\sum_\sigma f_{\frac{5}{2}}\left ( \frac{\mu+\sigma h}{k_bt } \right ) + \frac{1}{2}\overline{m}^2i_s n,\end{aligned}\ ] ] where @xmath23 , @xmath24 is the mass of particle , and @xmath25 is the polylogarithm function defined by @xmath26 where @xmath27 . the mean - field self - consistent equations are derived from the grand thermodynamic potential , [ e - sf ] @xmath28+n_0~;\end{aligned}\ ] ] where @xmath29 is the density of particles , @xmath30 is the density of condensed one and @xmath31 is the magnetization . @xmath30 is zero unless the temperature is below the bec point @xmath32 . in our previous investigations , we showed that the system exhibits two phase transitions , the bose - einstein condensation ( bec ) and the ferromagnetic transition @xcite . the condensation temperature @xmath4 and the curie temperature @xmath3 are calculated by solving the self - consistent equations . we find that @xmath3 is never below @xmath4 for all systems with a finite ferromagnetic exchange ( @xmath33 ) . however , one can get another solution to the eqs . ( [ e - sf ] ) , with @xmath34 at all temperatures . it means the system does not undergo a ferromagnetic transition at all , but remains in paramagnetic ( pm ) state at low temperatures . actually , whether there exists a curie point in the ferromagnetic bose gases is still a controversial question . some researchers suppose that the bose gas can not be magnetized spontaneously at low temperatures even if the ferromagnetic exchange is present @xcite . . the two curves cross at the temperature @xmath35 , which is just the fm transition point.,scaledwidth=35.0% ] in order to single out the physically correct solution , one has to compare the free energy of the the ferromagnetic ( fm ) state and the pm state . the relation between the free energy and the grand thermodynamic potential has the form : @xmath36 for computational convenience , the temperature @xmath37 and exchange interaction @xmath38 are re - scaled , as did in ref . [ 8 ] , by the following formula : @xmath39}^{-\frac{2}{3}}{t}/{t_0}$ ] and @xmath40}^{-\frac{2}{3}}i_s / ( k_b t_0 ) $ ] , where @xmath41 is the condensation temperature of ideal spin-1 bose gas . hereinafter , all the numerical results are obtained by setting @xmath42 . figure 1 shows the free energy of unit volume for the gas with @xmath43 . it shows clearly that the free energy of fm state is lower than that of the pm state at the low temperature region , which demonstrates that the fm state should be more stable than pm state . therefore , the low temperature state has a spontaneous magnetization . in experiments , the total spin of the ferromagnetic spinor condensate is observed to be conserved , which is called the _ spin conservation rule _ in some literatures @xcite . however , the spin conservation rule holds only globally , not locally . in the theoretical treatment of ref . [ 20 ] , the spin conservation rule is imposed by introducing a lagrangian multiplier . it is overconstrained in some sense , so that the spontaneous magnetization can not be established . recent experiments and theories indicate some domain structures should be formed and each domain is magnetized @xcite , where the conservation law for the total spin can be restored naturally . the fm transition is induced by the fm coupling and the transition temperature is about @xmath44 for the bose gas with @xmath43 . when the temperature goes down further , the bec then occurs , which is the intrinsic phase transition of bose gases . to demonstrate different features of the two transitions , we now calculate the specific heat of unit volume , @xmath45 where @xmath46 is the internal energy @xmath47 as shown in fig . 2 , for the system with @xmath43 , the specific heat exhibits a jump discontinuity at @xmath44 , from the pm state to the fm state . this is a characteristic feature of the landau - type of second - order phase transition . and similar behaviors have been observed in the specific heat of ferromagnetic insulators or itinerant - fermion ferromagnets @xcite . the bec occurs at @xmath48 , where the specific exhibits a bend . but specific heat is continuous at the bec point , similar to that of a free bose gas . the results indicate that the critical behaviors are different at the two transition points on the mean - field level . and @xmath49 . the dotted vertical lines serve to guide the eye to see the transition points.,scaledwidth=35.0% ] for a ferromagnet , the susceptibility above the curie point is of special interest . as already studied , the susceptibility is well described by curie - weiss law both in the insulating ferromagnet and the itinerant - electron ferromagnet . in this section we calculate the susceptibility for the itinerant - boson ferromagnet . the susceptibility can be derived from eqs . ( [ e - sf ] ) . differentiating both sides of the two equations and removing the term @xmath50 , the following equation are deduced , @xmath51 where @xmath52\nonumber \\ & & -\left(\frac{k_btm}{2\pi\hbar^2}\right)^{\frac{3}{2 } } \frac{{\left [ { f_{\frac{1}{2 } } \left ( { \frac{{\mu + h}}{k_bt } } \right ) - f_{\frac{1}{2 } } \left ( { \frac{{\mu - h}}{k_bt } } \right ) } \right]^2 } } { { \sum\limits_\sigma { f_{\frac{1}{2 } } \left ( { \frac{{\mu + h\sigma } } { k_bt } } \right ) } } } ~.\end{aligned}\ ] ] above the curie point , the magnetization @xmath53 ( then @xmath54 ) diminishes correspondingly when the external field @xmath10 tends to zero . so the second term in the above equation is omitted and then @xmath25 has a simple form : @xmath55 thus the zero - field susceptibility of unit volume is given by @xmath56 the susceptibility @xmath57 is a function of the coupling @xmath38 and temperature @xmath37 . figure 3 shows @xmath58 and @xmath57 versus @xmath59 at different given temperatures . as shown in the inset of fig . 3 , the susceptibility becomes larger as the coupling @xmath59 increasing . it is physically reasonable since the the system with larger @xmath59 can be magnetized more easily . at a given temperature , @xmath57 diverges as @xmath59 approaches a critical value . it is worth noting that the inverse of the susceptibility is in a good linear relationship with the coupling . and @xmath60.,scaledwidth=35.0% ] the susceptibility versus temperature is shown in fig . 4 . one can immediately find that the susceptibility meet quite well with the curie - weiss law in a very large temperature region . seeing that the curie - weiss law is very difficult to be derived for the itinerant-_fermion _ ferromagnet , it is really surprising that we get it for the itinerant-_boson _ ferromagnet just based on the mean - field approximation . and @xmath61 . , scaledwidth=35.0% ] in order to discuss the curie - weiss law in a more explicit way , we proceed to carry out a semi - analytical calculation to deduce the linear dependence of @xmath58 on the temperature . the first step is to analyze the temperature dependence of @xmath25 . it is quite complicated , because the chemical potential @xmath20 is an implicit function of the temperature . we consider a limit case that the parameter @xmath38 is quite small , when @xmath62 is close to @xmath32 . so @xmath20 is close to zero in the vicinity of @xmath62 . according to the asymptotic behavior of the polylogarithm function : @xmath63 and @xmath64 as @xmath65 , we get the following equations from eqs . ( 7a ) and ( 13 ) respectively , @xmath66.\ ] ] and @xmath67 substitute eqs . ( [ e15 ] ) and ( [ e16 ] ) into eq . ( [ e14 ] ) , we get @xmath68 in the vicinity of @xmath3 which is only slightly larger than @xmath69 , eq . ( [ e17 ] ) could be further simplified to @xmath70~.\end{aligned}\ ] ] thus the effective fm transition temperature is defined as @xmath71 so far the curie - weiss law is derived . we note that the derivation is only valid in small @xmath38 cases . in the high temperature limit , one can also easily prove that @xmath72 is linearly dependant on @xmath37 . in this case , @xmath73 has a quite large value , so that @xmath74 according to eq . ( 6 ) . combining eqs . ( 7a ) , ( 13 ) and ( 14 ) , it yields @xmath75 we estimate this equation holds in the range of @xmath76 . in summary , we calculate thermodynamic quantities of the spinor bose gas with ferromagnetic interactions . such kind of investigations has already been performed intensively for the ferromagnetic fermions , while few as yet for bosons . based on a mean - field approximation , we show that the system undergos a ferromagnetic phase transition first , then the bose - einstein condensation with the temperature decreasing . the specific heat shows a jump discontinuity at the curie point and a bend at the bose - einstein condensation temperature , indicating that critical behaviors are different near the two transition . the more surprising result is that the mean - field theory yield the magnetic susceptibility which satisfies perfectly the curie - weiss law over a wide range of temperature . this work is supported by the national natural science foundation of china ( grant no . 10504002 ) , the fok yin - tong education foundation , china ( grant no . 101008 ) , and the ministry of education of china ( grant no . ncet-05 - 0098 ) . < num > d.m . stamper - kurn _ et al_. , phys . * 80 * , 2027 ( 1998 ) ; j. stenger _ et al_. , nature ( london ) * 396 * , 345 ( 1998 ) . barrett , j.a . sauer , and m.s . chapman , phys . lett . * 87 * , 010404 ( 2001 ) . klausen , j.l . bohn , and c.h . greene , phys . a * 64 * , 053602 ( 2001 ) . t.l . ho , phys . rev . lett . * 81 * , 742 ( 1998 ) ; t. ohmi and k. machida , j. phys . soc . jpn . * 67 * , 1822 ( 1998 ) . t. isoshima , k. machida , and t. ohmi , phys . a * 60 * , 4857 ( 1999 ) . m. ueda , phys . a * 63 * , 013601 ( 2000 ) . q. gu and r.a . klemm , phys . a * 68 * , 031604 ( 2003 ) ; q. gu , k. bongs , and k. sengstock , phys . a * 70 * , 063609 ( 2004 ) . k. kis - szabo , p. szepfalusy , and g. szirmai , phys . a * 72 * , 023617 ( 2005 ) ; g. szirmai , k. kis - szabo , and p. szepfalusy , eur . j. d * 36 * , 281 ( 2005 ) . lett . * 95 * , 180403 ( 2005 ) . j. mur - petit _ et al . _ , a * 73 * , 013629 ( 2006 ) . q. gu and h. qiu , phys . lett * 98 * , 200401 ( 2007 ) . h. schmaljohann _ et al . _ , lett . * 92 * , 040402 ( 2004 ) ; j. kronjger _ et al . _ , a * 72 * , 063619 ( 2005 ) . et al . _ , lett . * 92 * , 140403 ( 2004 ) ; m .- s . _ , nature physics * 1 * , 111 ( 2005 ) . sadler_et al . _ , nature * 443 * , 312(2006 ) . p. mohn , _ magnetism in the solid state : an introduction _ ( springer - verlag , berlin , 2003 ) . t. moriya , _ spin fluctuations in itinerant electron magnetism _ ( springer - verlag , berlin , 1985 ) . murata and s. doniach , phys . * 29 * , 285 ( 1972 ) ; t. moriya and a. kawabata , j. phys . soc . jpn . * 34 * , 639 ( 1973 ) . y. takahashi , j. phys . : condens . matter * 11 * , 6439 ( 1999 ) ; y. takahashi and h. nakano , j. phys . : condens . matter * 16 * , 4505 ( 2004 ) . t. isoshima , t. ohmi , and k. machida , j. phys . . jpn . * 69 * , 3864 ( 2000 ) ; w. zhang , s. yi , and l. you , phys . rev . a * 70 * , 043611 ( 2004 ) .	thermodynamics of a spin-1 bose gas with ferromagnetic interactions are investigated via the mean - field theory . it is apparently shown in the specific heat curve that the system undergoes two phase transitions , the ferromagnetic transition and the bose - einstein condensation , with the curie point above the condensation temperature . above the curie point , the susceptibility fits the curie - weiss law perfectly . at a fixed temperature , the reciprocal susceptibility is also in a good linear relationship with the ferromagnetic interaction .
due to its unique electronic properties , the study about the graphene system has recently drawn great attentions.@xcite . the charge carriers in single - layer graphene behave like `` relativistic '' chiral massless particles and possess a gapless linear energy spectrum . effectively , in low energy regime , graphene can be viewed as a two - dimensional ( 2d ) dirac fermion system . for such dirac - like quasiparticles , the transport properties are dramatically different from the conventional semiconductors . there is hope to utilize these peculiar transport characteristics to develop novel relativistic electronic devices@xcite . recently , another 2d dirac fermion system has been confirmed experimentally.@xcite it is called topological metal ( tm ) , _ i.e. _ , the surface states of a strong topological insulator ( sti)@xcite . the strong topological insulator is a new state of matter protected by the time - reversal invariance . though still with a bulk gap , it is distinguished from the normal insulator by the topological @xmath0 invariance . this distinction results in the existence of the gapless electronic states on the sample surfaces . the surface state is the tm mentioned above , which can be effectively viewed as a 2d chiral dirac fermion system . due to the time - reversal symmetry and the topology of the bulk gap , the topological surface states are robust against disorder . because of the similar energy dispersions , the transport property of the tm is very similar to that of a graphene . however , there are important differences between the two systems . tm is a surface state of semiconductor alloy , _ e. g. _ , @xmath1 or @xmath2 , thus more stable than a fragile thin graphene sheet . furthermore , the spin of the dirac fermion of a tm is the real spin , while for graphene it is the pseudo spin . hence , the magnetic or spin control of the electronic transport on a tm can be much more effective than that on a graphene . is @xmath3 ( @xmath4 ) . @xmath5 denote the interface between region @xmath6 and @xmath7 . , width=302 ] in this work , we study the transport properties with ferromagnetic barriers on top of a topological surface . as shown in fig.1 , for a ferromagnetic barrier , a ferromagnetic insulator is put on the top of the topological insulator to induce an exchange field via the magnetic proximity effect . meanwhile , a metal gate on the top of the insulator is used to control the electrostatic potential . we find that by adjusting the magnetization orientation in the ferromagnetic insulator and the gate voltage , the ferromagnetic barrier can be tuned into various transmission regions . the behaviors of the transmission probability , as a function of the incident angle of the electron , in different regions are rather different . an interesting issue is about the klein tunneling . it is well known that , for dirac fermion , merely with electrostatic potential one can not block the electron transport , and the normal incidence is at least partially transmitted . but for the ferromagnetic barrier , because the induced exchange field , not only could the angle of klein tunneling be controlled , but also the klein tunneling could even be blocked . we carefully analyze the transmission probabilities in various regions , for both single and double barriers . based on these analyses , we propose that electron beam collimation , wavevector filtering , magnetic and electric switch , wavevector - based in - plane spin valve could be realized in such structures . these transport phenomena are quite different from the case with ferromagnetic barrier on graphene@xcite . this paper is organized as follows . in sec . ii , we introduce the model and formalisms . in sec . iii , we discuss the analytical and numerical results of the transport properties of single and double ferromagnetic barriers . and we then show how to construct novel electronic devices based on the phenomena obtained in our theory . finally , a brief summary is given in sec . the dirac fermion of the topological surface can be described as : @xmath8 where @xmath9 is the effective exchange field induced via magnetic proximity effect . in this work , we only consider the in - plane exchange field . the main consequence of the z component of the exchange field is to open an energy gap and its effects have been investigated in a recent theoretical study@xcite . if the fermi level of the whole system is inside the gap , the transport of the barrier is completely blocked . no tunneling is permitted . otherwise , the effect is just a change of the position of the fermi surface . but in our ferromagnetic barrier model , adjusting the gate voltage will also change the fermi surface . so for simplicity and without loss of generality , we merely consider an in - plane exchange field . the sketch of the barrier structure is shown in fig.1 . take the single barrier structure for example , the whole system consists of three different regions and the interfaces between these regions are labeled by @xmath5 . with the above hamiltonian , the eigenfunctions for region @xmath10 are @xmath11 @xmath12 where @xmath13 , @xmath14 , and @xmath15 . @xmath16 is the potential of this region , @xmath17 is the energy of the incident electron . due to the translation invariant along the y - axis , the momentum @xmath18 is conserved , i.e. @xmath19 . in order to solve the transport problem , we use the standard transfer matrix method@xcite . by matching the wave functions at the interface @xmath20 , we get the transfer matrix @xmath21 the transfer matrix at the interface @xmath22 can be represented as @xmath23 where @xmath24 is the same as @xmath25 except the region index . finally , we will get the total transfer matrix for the single barrier @xmath26 with the transfer matrix , the reflection and translation coefficients can be got through following equation @xmath27 @xmath28 is just the transmission probability we want . double barrier structure could be treated in similar way . the only difference is that we have four interfaces here . @xmath29 with the transmission probability , in the linear transport regime and for low temperature , we can get the conductance density ( the conductance per unit length along the interface ) of the barrier structure@xcite @xmath30 here , @xmath31 , @xmath32 is the fermi energy of the system , @xmath33 is the incident angle of the electron and @xmath34 is the transmission probability through the barrier . for a single barrier , an analytical expression of the transmission probability could be given . @xmath35 where @xmath36 . formally , it is the same as the one for the graphene potential barrier except a shift of the fermi surface due to the in - plane exchange field . . @xmath17 is the energy of incident electron . @xmath37 is the electrostatic potential of region 1(2 ) . @xmath38 is the in - plane rotational angle of the exchange field . , width=302 ] as mentioned above , for a ferromagnetic barrier , changing the gate voltage will alter the fermi surface , the in - plane exchange field will also cause a shift . the key issue about the transport is that , during the tunneling , the energy and @xmath18 of the incident electron are conserved . as shown in fig . 2 , in the incident region ( region 1 in fig.1 ) , the range of @xmath18 values @xmath39 is plotted . but not all the incident electrons are allowed to transmit , since the fermi surface has been changed in the ferromagnetic barrier ( region 2 in fig . the @xmath18 range of the ferromagnetic barrier is @xmath40 . therefore , the permitted incident angle of the electron is determined by the overlap of the @xmath18 ranges of adjacent regions . it means that , through adjusting the gate voltage and the direction of the magnetization of the ferromagnetic barrier , we can tune the acceptable range of @xmath18 , and rather different transmissions will appear accordingly . is the incident angle of the electron . @xmath38 is the in - plane rotation angle of the magnetization of the barrier . here , @xmath41 , @xmath42 , @xmath43 , @xmath44 , @xmath45 , @xmath46 , @xmath47 . , title="fig:",width=340 ] + and other parameters are the same as fig . 3 . , title="fig:",width=340 ] + [ cols="^ " , ] , @xmath48 , @xmath49 and @xmath50 . , width=302 ] for the double barrier structure , we also could understand the transport properties through the conservation of the @xmath18 and the energy . as mentioned above , depending on the voltage and magnetization , single ferromagnetic barrier could be tuned into three different transmission statuses : total transmission , partial transmission , and the blockade situations . the transport behavior of the double barrier system is more complex than that of the single one , since that the two barriers can be tuned into different transport situations . here , we investigate two interesting examples . one is the case that two ferromagnetic barriers are both in the partial transmission status . in this case , both barriers have their own transmission sector of the incident angle of the electron . only when the two transmission sectors overlap , the double barrier structure becomes conducting . hence , as shown in fig . 7 , the direction of magnetization of the second barrier is fixed @xmath51 . only if the angle of the first barrier @xmath52 is tuned in parallel , the transmission of the double barrier structure is nonzero . the corresponding conductance is also given in fig . 7 . essentially , in this situation the double barrier system becomes a special wavevector - dependent in - plane spin valve system , for which the parallel configuration is the only conducting state . the blockade of the transmission is mainly because the mismatch of the wavevectors , while for normal spin valve the primary reason is the different spin dependent potentials . therefore , for a normal spin valve , the conductance will vary continuously with the angle between the magnetizations . but , as shown in fig . 7 , for the in - plane spin valve on tm the change of the conductance is rather sharp and the the conductance of the non - parallel configuration is exactly zero . these special transport properties of the double barrier structure on tm could also be tested in transport experiments as a distinction between the tm and normal semiconductors . for a single barrier , it is hard to directly observe the partial transmission region through the conductance . this is because that the change of position of the transmission sector will not induce an obvious change in the conductance . but in the double barrier system this can be done . as a comparison ( see in fig . 8) , we also calculate the case that both barriers are in the total transmission status . in this situation , the angle dependent transmission probability is much more complicated . in conclusion , we have investigate the transport properties of the ferromagnetic barriers on the surface of a topological insulator . the single and double barrier structures have been carefully considered . we show that the gate voltage and the magnetization could be used to manipulate the fermi surface of the barrier . due to the conservation of energy and @xmath18 of an incident electron during the transport process , adjusting the fermi surface will tune the single ferromagnetic barrier into three different transmission statuses : total transmission , partial transmission , and the transmission blockade . in the total transmission case , all the incident angles are allowed . the angle of the klein tunneling and position of the resonance vary with the direction of the magnetization , and meanwhile new resonance peaks appear . in the partial transmission situation , only the incident angles in a transmission sector are allowed . the position and width of the transmission sector can be adjusted through the gate voltage and the magnetization . therefore , the electron collimation and wavevector filter could be realized based on this phenomenon . with proper gate voltage and magnetization , one can even separate the fermi surfaces , reaching the blockade status . in this case , no transport can occur , even the klein tunneling . we propose that both magnetic and electric switchs , based on the single ferromagnetic barrier on topological surface , could be realized in this situation . we also study the transport of the double barrier structure . in this system , we can construct a special wavevector - dependent spin valve when both barriers are in the partial transmission status . the spin valve is conducting only if the magnetization of the two barriers are parallel . we emphasis that these unique transport properties of tm could be used to distinguish tm from the normal semiconductors and the graphene systems . hence , the observations of these transport properties provide confirmation of the existence of tm . a. h. castro neto , f. guinea , n. m. r. peres . k. s. novoselov , and a. k geim , rev . 81 , 109 ( 2009 ) c. w. beenakker , rev . mod . phys . 80 , 1337 ( 2009 ) m. i. katsnelson , mater . today 10 , 20 ( 2007 ) f. rana , ieee transactions on nanotechnology , 7 , 91 ( 2008 ) ; j. r. williams , l. dicarlo , c. m. marcus , science , 317 , 638 ( 2007 ) ; m. c lemme , t. j. echtermeyer , m. baus , h. kurz , ieee electron device letters , 28 , 4 , 284 ( 2007 ) ; d. hsieh et al . , nature 452 , 970 ( 2008 ) d. hsieh , et al . , arxiv:0904.1260 y. xia et al . , nat . phys . 5 , 398 ( 2009 ) y. l. chen et al . science 325 , 5937 ( 2009 ) x. -l . qi , t. hughes , and s .- c . zhang , phys . b 78 , 195424 ( 2008 ) j. e. moore and l. balents , phys . b 75 , 121306 ( 2007 ) l. fu , c. l. kane , and e. j. mele , phys . rev . 98 , 106803 ( 2007 ) l. fu and c. l. kane , phys . b 76 , 045302 ( 2007 ) j. c. y. teo , l. fu , and c. l. kane phys . rev . b 78 , 045426 ( 2008 ) h. zhang et al phys . 5 , 438 ( 2009 )	we study the transmission through single and double ferromagnetic barriers on the surface of a topological insulator . by adjusting the gate voltage and magnetization oreintation , the ferromagnetic barrier can be tuned into various transmission regions , where the wavevector - dependent tunnelings are quite different . we find that the klein tunneling can be manipulated or even be turned off . these special properties offer the possibility to control electron beams on the `` topological metal '' . various novel devices , such as electronic collimation , wavevector filter , magnetic and electric switchs , and wavevector - based spin valve , may be constructed based on our observed phenomena .
the simplest investigations into a many - body system like a bose - einstein condensate comprise studies of thermal equilibrium properties , and the physics of small fluctuations around thermal equilibrium . for one - dimensional systems , very similar behavior is found using either photons in optical fibres or ultra - cold atoms in waveguides . although techniques are not yet as experimentally advanced in the latter case , preliminary theory and some experimental measurements have already taken place . the atomic systems have the advantage that relatively long interaction times , large interaction strengths and low losses are possible , thus potentially allowing stringent tests of underlying quantum correlations . in this paper , we extend previous studies of correlations to include the experimentally realistic case of atoms in a waveguide with a harmonic longitudinal confining potential . the treatment is at finite temperature , and makes use of exact results for the uniform gas , together with a local density approximation . for strong radial confinement , these types of system are examples of one - dimensional quantum ( @xmath0 ) gases @xcite . they have the important property that in many cases their energy eigenstates are exactly solvable @xcite , resulting in a greatly increased fundamental understanding of the relevant quantum field theory . for this reason , the study of 1d systems plays an important role in the physics of quantum many - body systems . it is possible to make first - principle predictions without introducing added approximations like perturbation theory . this permits direct experimental tests of the underlying many - body quantum physics , as has been demonstrated in photonics with squeezed solitons in optical fibres @xcite . for ultra - cold atomic systems with repulsive interactions , the most interesting and exciting feature is the predicted transition of an interacting gas of bosons to a ` fermionized ' tonks - girardeau @xcite regime at large coupling strengths and low densities . we have recently made use of the known exact solutions to the uniform one - dimensional ( 1d ) interacting bose gas problem to calculate the exact local second - order correlation function at all densities and interaction strengths @xcite . this is the most direct indication of ` fermionic ' behaviour , since this correlation function is strongly reduced at low density and strong coupling similar to the case of fermions , where it vanishes exactly , due to the pauli exclusion principle . the first experimental evidence of reduced or ` antibunched ' correlations in a @xmath0 bose gas has recently been demonstrated in ref . however , current experiments typically take place in traps , with a longitudinal trapping potential . provided the trap potential varies slowly , this environment is sufficiently close to a uniform one so that the exact solutions can still be used locally , in an approximation called the local density approximation . in this paper , we make use of the local density approximation ( lda ) to calculate the density profile and finite temperature local pair correlation function of a @xmath0 bose gas trapped in a harmonic potential . the results are valid for sufficiently low longitudinal trap frequencies , and make use of the exact solutions to the plane - wave lieb - liniger model @xcite at finite temperature , together with the hellmann - feynman theorem . we mostly focus on regimes with quantum degeneracy . this requires temperatures @xmath1 , where @xmath2 is the temperature of quantum degeneracy of the trapped sample as a whole , @xmath3 is the total number of particles , and @xmath4 is the axial trap frequency . our main results show how ` fermionization ' can be readily detected through a simple measurement of the pair correlation averaged over the trap . this is very close to the correlation at the trap center as predicted @xcite using the lieb - liniger uniform model , and in principle can be measured via photoassociation of trapped atoms , or other related two - body inelastic processes whose rates are governed by the local pair correlations @xcite . in addition , an indirect measure of the pair correlations can be obtained @xcite via the measurement of three - body recombination rates as recently demonstrated experimentally @xcite . one dimensional quantum field theories have the important and useful property that they are often exactly solvable . this is not generally the case for higher - dimensional quantum field theories . thus , the study of these 1d models can lead to an insight into the nature of quantum field theory for interacting particles , that is not possible from the usual perturbative approaches . in this section , we review the physics of these exact solutions for interacting bosons in the uniform 1d case , and introduce the theoretical framework for treating a non - uniform gas within the local density approximation . the study of exact solutions for the one - dimensional bose gas started with girardeau s seminal work @xcite on hard - core , or impenetrable bosons . in this model , there is a remarkable and exact correspondence between the measurable correlation functions of free fermions , and those of strongly interacting bosons . in the @xmath0 bose gas model with a delta - function interaction , solved by lieb and liniger @xcite , the particles can pass through each other , so they are no longer impenetrable . this provides a realistic description of a wave - guide with transverse dimensions larger than the ` core ' of a particle in the wave - guide . under these circumstances , there may only be a single relevant transverse mode , yet particles are able to exchange their positions as they propagate past each other . thus , we start by reviewing the theory of a gas of @xmath3 bosons interacting via a delta - function potential in one dimension . the @xmath0 bose gas has a short - range repulsive interaction between particles which is characterized by just one coupling constant . in second quantization , the hamiltonian is @xmath5 where @xmath6 is the bosonic field operator , @xmath7 is the atom mass , @xmath8 is the coupling constant , and @xmath9 is the trapping potential which we assume is harmonic with @xmath10 , while @xmath4 is the trap oscillation frequency in the axial direction . to treat the uniform gas we set @xmath11 . for bose gases in highly elongated cylindrical traps ( @xmath12 , where @xmath13 is the frequency of the transverse harmonic potential ) such that the sample can be described by the above @xmath0 model , the coupling constant @xmath14 is expressed through the @xmath15 scattering length @xmath16 @xcite . for a positive scattering length @xmath16 which is much smaller than the amplitude of transverse ( @xmath17-direction ) zero point oscillations , or the transverse harmonic oscillator length , @xmath18 one has @xmath19 the @xmath0 regime is reached if @xmath20 is much smaller than the thermal de broglie wavelength @xmath21 and a characteristic length scale @xmath22 @xcite responsible for short - range correlations . on the same grounds as at @xmath23 @xcite , one finds that for fulfilling this requirement it is sufficient to satisfy the inequalities @xmath24 where @xmath25 is the @xmath0 ( linear ) density in the center of the trap , @xmath26 . we now give a brief overview of the uniform ( @xmath11 ) bose gas problem describing a gas of @xmath3 bosons interacting via a pair - wise repulsive delta - function potential in a @xmath0 box of length @xmath27 with periodic boundary condition @xcite . in the thermodynamic limit ( @xmath3 , @xmath28 , while the @xmath0 linear density @xmath29 is kept constant ) , the solution to the energy eigenstates is found @xcite using the bethe ansatz @xcite . in this solution , all relative wave - functions are assumed to have a plane - wave form except for finite changes in gradient at each collision where two particle coordinates are equal : @xmath30 \notag \\ & \times\hat{\psi}^{\dagger}(z_{1})\hat{\psi}^{\dagger}(z_{2}) ... \hat{\psi } ^{\dagger}(z_{n})\left\| 0\right\rangle\end{aligned}\ ] ] here , we have used units in which @xmath31 , while introducing the lieb - liniger notation @xcite of @xmath32 . also , @xmath33 is the sign function , and @xmath34 is the `` quasi - momentum '' . the quasi - momenta are determined from the delta - function slope - change requirements , that is , @xmath35 is determined by the boundary conditions at @xmath36 . in the limit of a large sample , and defining @xmath37 $ ] , one can approximate @xmath38 by a continuous function @xmath39 which is the density of quasi - momenta . the distribution of quasi - momenta is then obtained as the solution to the following integral equation : @xmath40 here , the kernel function @xmath41 is given by @xmath42 and @xmath43 is the maximum quasi - momentum which determines the particle number density @xmath29 via @xmath44 the corresponding ground - state energy is given by @xmath45 and is often written as @xmath46 , being an implicit function of @xmath47 , via a dimensionless function @xmath48 of the parameter @xmath49 . restoring the physical units , this gives an energy per particle : @xmath50 where @xmath51 the dimensionless parameter @xmath52 which characterizes the strength of interactions is in fact the only parameter needed to describe the uniform @xmath53 bose gas at zero temperature . the limit of @xmath54 corresponds to the weakly interacting gross - pitaevskii ( gp ) regime , where the mean - field bogoliubov theory works well . the opposite limit of @xmath55 corresponds to the strongly interacting or tonks - girardeau ( tg ) regime , and as @xmath56 one regains girardeau s results for impenetrable bosons . the solution to the ground state energy @xmath57 can be used , together with the hellmann - feynman theorem @xcite , for calculating an important observable the normalized local pair correlation @xmath58 the pair correlation is found by taking the derivative of the ground state energy with respect to the coupling constant @xmath14 , owing to the fact that @xmath59 so that @xmath60 the pair correlation @xmath61 for the zero temperature uniform @xmath0 bose gas has been calculated using the lieb - liniger exact solution @xcite for @xmath48 . the results are given in ref . @xcite . here , we will extend these results ( see section iii ) to the case of a trapped ( non - uniform ) bose gas using the local density approximation , and to finite temperatures as well . the excited states of the uniform 1d bose gas can be calculated in a similar way , with each excited state corresponding to the removal of a quasi - momentum with @xmath62 called a hole and the creation of a quasi - momentum with @xmath63 . in 1969 , yang and c.p . yang @xcite worked out the finite - temperature density matrix solution for the lieb - liniger model , by constructing the free energy and taking into account the entropy of all the different excited states . this was used in a subsequent work @xcite to calculate numerically the pressure of the gas as a function of temperature . at thermal equilibrium , we now assume that the density of quasi - momenta @xmath39 has no upper cut - off , and that it consists of two types of terms occupied quasi - momenta [ with density @xmath64 and unoccupied or ` hole ' quasi - momenta [ with density @xmath65 , i.e. @xmath66 . the overall integral equation now has the form ( using units in which @xmath31 and @xmath32 ) @xmath67 \notag \\ & = 1+\int_{-\infty}^{\infty}k(k - p)f_{p}(p)dp.\end{aligned}\ ] ] hence , the particle density @xmath47 is obtained from the occupied or particle quasi - momenta @xmath68 while the total energy is now @xmath69 however , there is also an entropy involved , since there are many wavefunctions that are nearly the same , within a given range of values of @xmath70 and @xmath71 . in fact , the number of choices compatible with a given @xmath72 value is @xmath73 ! } { \left [ f_{p}(k)ldk\right ] ! \left [ f_{h}(k)ldk \right ] ! } .\ ] ] thus , the entropy is @xmath74 dk.\ ] ] minimizing the total free energy @xmath75 gives the thermal equilibrium distribution of holes and particles , where we choose temperature to be in energy units , so that @xmath76 . the minimization at a fixed average particle number requires the use of a lagrange multiplier @xmath77 , and gives the result that the distribution @xmath78 satisfies the integral equation @xmath79 = 1+\int_{-\infty}^{\infty } k(k - p)f_{p}(p)dp , \label{yy1}\ ] ] where the excitation spectrum @xmath80 is calculated from a second integral equation @xmath81 here , @xmath77 can be shown to coincide with the chemical potential of the system , while the entropy and the free energy per particle are found from : @xmath82 dk \notag \\ & + \frac{1}{nt}\int_{-\infty}^{\infty}f_{p}(k)\varepsilon(k)dk , \label{entropy}\end{aligned}\ ] ] @xmath83 in addition , using the thermodynamic identity @xmath84 , one can arrive at the following simple result for the pressure of the gas @xmath85 to calculate the pair correlation @xmath61 for a finite temperature gas one can again use the hellmann - feynman theorem @xcite . here , we consider the canonical partition function @xmath86 , where the trace is over the states of the system with a fixed particle number @xmath3 , at temperature @xmath87 . taking the derivative of @xmath88 with respect to the coupling constant @xmath14 we obtain @xmath89 = \frac{l}{2}\left\langle \psi^{\dagger}(z)\psi^{\dagger}(z)\psi(z)\psi(z)\right\rangle .\ ] ] introducing the free energy per particle @xmath90 and restoring the physical units , this gives @xcite : @xmath91 here , @xmath92 is a dimensionless temperature parameter , with @xmath93 being the temperature of quantum degeneracy for a uniform gas . hence , we have @xmath94 the pair of the dimensionless parameters @xmath52 and @xmath95 completely characterize the properties of a finite temperature uniform gas . alternatively , the local pair correlation @xmath61 can be calculated within the grand canonical formalism . here , we consider the grand canonical partition function @xmath96 $ ] , where @xmath97 is the grand canonical thermodynamic potential and @xmath98 is the pressure . the trace is over the states of the system , at a fixed chemical potential @xmath77 and temperature @xmath87 . taking the derivative of @xmath99 with respect to the coupling @xmath14 we obtain : @xmath100\right\ } \notag \\ & = \frac{l}{2}\left\langle \psi^{\dagger}(z)\psi^{\dagger}(z)\psi ( z)\psi(z)\right\rangle \,.\end{aligned}\ ] ] thus , the normalized pair correlation @xmath61 can be calculated using @xmath101 this requires the use of eq . ( [ p ] ) for the pressure , which in turn is found after solving the yang - yang integral equations ( [ yy1 ] ) and ( [ yy2 ] ) . the local pair correlation for a finite temperature uniform gas has been first calculated in ref . @xcite using the exact solutions to the yang - yang integral equations ( [ yy1 ] ) and ( [ yy2 ] ) , together with eqs . ( [ fovern ] ) and ( [ g2-canonical ] ) . in sections iv - vii , we will extend these results to the case of a trapped gas using the local density approximation . in a quasi - uniform approximation , we suppose that the system can be divided into small regions of size @xmath102 which is larger than a characteristic short - range correlation length @xmath22 . in each of this regions we assume that the inhomogeneity of the gas is negligible so that it can be treated as a uniform gas . in this case , the trapping potential @xmath9 is replaced by a step - like function @xmath103 which is constant within each region from @xmath104 to @xmath105 and undergoes step - like changes at the boundaries of the adjacent regions . here , the size @xmath102 takes the role of the length @xmath27 from the yang - yang solution that applies to each region . we now consider an ansatz in which the overall density matrix has the structure of an outer - product of canonical solutions , with @xmath106 being the average number of particles in the @xmath107-th region : @xmath108 next we look for an approximate solution in which the effective hamiltonian is assumed to introduce no coupling between the regions . to obtain this we must now minimize the total free energy given by @xmath109 this requires us to include a constraint on the total particle number : @xmath110 hence , it is appropriate to use a lagrangian formulation with @xmath111 we note here that the lagrangian @xmath112 is now simply a sum over independent regions , with each term corresponding to that for a single uniform bose gas . as we are only constraining the total particle number , not the number in each region , the chemical potential is the same for each term . since there is no explicit coupling between the regions , the lagrangian is minimized when we satisfy the yang - yang equations in each separate region , but with the same ( global ) chemical potential @xmath113 at all locations . in more detail , we have shown that for a large system , where the density profile varies in a smooth way , the system behaves locally as a piece of a uniform gas . this can be described locally as a uniform gas with chemical potential equal to the local effective chemical potential @xmath114 where @xmath113 is the global equilibrium chemical potential . for the lda to be valid , the short - range correlation length @xmath115 should be much smaller than the characteristic inhomogeneity length @xmath116 . these length scales depend on the displacement from the trap center @xmath117 , and the lda validity criterion reads : @xmath118 the short - range correlation length @xmath115 is defined locally via the density distribution @xmath119 . at low temperatures , the correlation length @xmath120 can in general ( irrespective of the interaction strength ) be expressed via the local chemical potential @xmath121 : @xmath122 in the weakly interacting gross - pitaevskii ( gp ) regime the relation between the chemical potential @xmath121 and the density @xmath119 is @xmath123 , and we obtain that the correlation length coincides with the healing length @xmath124 . in the strongly interacting tonks - girardeau ( tg ) regime one has @xmath125 , so that @xmath126 , neglecting the numerical factor of order @xmath127 . at high temperatures , @xmath115 is of the order of the thermal de broglie wavelength @xmath128 . the condition ( [ lda - criterion ] ) is sufficient for using the lda for calculating the density profiles and local correlation functions . the reason is that these correlations , in particular the two - particle correlation @xmath129 , are determined by the contribution of excitations which have energies of the order of the chemical potential and wavelength of the order of @xmath22 . however , this is not the case for all correlation functions . for example , calculation of the finite - temperature single - particle correlation function would require a strong lda condition in which the sample size was much larger than the phase correlation distance @xcite . in this sense , one may call eq . ( [ lda - criterion ] ) the ` weak ' lda criterion . however , within the lda , no correlation function can be calculated reliably over distance scales that are comparable to the sample size . thus , the ` weak ' criterion of validity of the lda requires that variations of the density occur on a length scale that is much larger than @xmath115 , in which case the gas is treated locally as a piece of a uniform gas . from the definition of @xmath130 , one can easily see that the lda is easier to satisfy in the center of the trap where the density profile is almost flat than near the tails of the distribution where the density drops rapidly . however , for measurements that average over an entire trap , it is the central region that plays the most important role . here , we analyze the implications of the lda criterion ( [ lda - criterion ] ) for a zero temperature gas . at @xmath23 , a uniform @xmath0 bose gas can be characterized by a single dimensionless interaction parameter @xmath52 , eq . ( [ gamma ] ) . depending on its value , one has two well - known and physically distinct regimes of quantum degeneracy . for @xmath54 , i.e. at weak couplings or high densities , the gas is in a coherent or gross - pitaevskii ( gp ) regime . in this regime , long - range order is destroyed by long - wavelength phase fluctuations @xcite and the equilibrium state is a quasi - condensate characterized by suppressed density fluctuations . for strong couplings or low densities , @xmath131 , the gas reaches the strongly interacting or tonks - girardeau ( tg ) regime and undergoes `` fermionization '' @xcite . the term `` fermionization '' is used here in the sense that the wave function strongly decreases as particles approach each other . for a trapped ( non - uniform ) gas one can introduce a local interaction parameter @xmath132 which changes with the density distribution @xmath119 and can be used for characterizing the local properties of the gas . from the definition of @xmath133 it is clear that as one moves from the center of the trap towards the tails of the density distribution where @xmath134 , the gas either enters the tg regime where @xmath135 , or else the lda itself breaks down . moreover , in the @xmath23 case the lieb - liniger solution within the lda gives a density profile that vanishes [ @xmath136 beyond a certain distance @xmath137 from the origin @xcite . this distance is called the thomas - fermi radius and it is determined from the condition @xmath138 which gives : @xmath139 since @xmath119 vanishes exactly at @xmath140 @xcite , it is clear that the lda criterion ( [ lda - criterion ] ) can only be satisfied up to a certain maximum distance from the trap center @xmath141 , displaced from @xmath137 by @xmath142 ( @xmath143 ) . we would like therefore to determine the displacement @xmath142 such that the lda is valid for @xmath144 and breaks down beyond @xmath141 . as we are interested in calculating the density profiles and the local two - particle correlation function , we will only focus on the ` weak ' lda criterion , eq . ( [ lda - criterion ] ) . first , we rewrite the inhomogeneity length scale from eq . ( [ lda - criterion ] ) in the following equivalent form @xmath145 using the explicit expression @xmath146 and taking its derivative , we obtain at @xmath141 ( @xmath143 ) : @xmath147 combining eqs . ( [ l - c - mu ] ) and ( [ l - inh - approx ] ) , one can rewrite the lda criterion @xmath148 in the following equivalent form ( again neglecting numerical factors of order one ) : @xmath149 next , we note that in the limiting gp and tg regimes the derivative @xmath150 is equal , respectively , to one and two , so in general @xmath151 . therefore , the exact numerical value of this quantity can be replaced by unity in all regimes , and the lda criterion becomes @xmath152 finally , expanding @xmath121 near the edge of the cloud where @xmath153 ( @xmath143 ) , we obtain that @xmath154 , so that the criterion of applicability of the lda is reduced to a simple requirement : @xmath155 for @xmath156 , the local value of @xmath133 at @xmath157 is @xmath158 we see that if @xmath159 then @xmath160 at @xmath157 , implying that the gas stays in the gp regime at all locations @xmath117 until the lda breaks down . if , on the other hand , @xmath161 then @xmath162 at @xmath157 so that the gas first approaches the tg regime and then the lda breaks down . for @xmath163 the gas is in the tg regime at all locations @xmath117 , until the lda breaks down . this is because @xmath133 is always larger than @xmath164 , and hence @xmath135 . in the limiting cases of @xmath156 and @xmath163 , the lda criterion ( [ lda - criterion - final ] ) can be conveniently rewritten in terms of @xmath165 and the total number of particles @xmath3 . in doing so we use the fact that the chemical potential is given by @xmath166 for @xmath167 , and by @xmath168 for @xmath163 . in addition , we use the known relationship between the peak density @xmath169 and @xmath3 in each case [ see eqs . ( [ tf - gp - density ] ) and ( [ tf - tg - density ] ) below ] . as a result , we obtain that the lda criterion ( [ lda - criterion - final ] ) can be rewritten as follows , in the gp and tg regimes , respectively : @xmath170 , \label{lda - gp } \\ \frac{\delta z}{r } & \gg\left ( \frac{2m\omega_{z}}{\pi^{2}\hbar^{2 } n^{2}(0 ) } \right ) ^{2/3}=\frac{1}{n^{2/3}},\;\;\;\;\;[\gamma(0)\gg1 ] . \label{lda - tg}\end{aligned}\ ] ] as we see , for any small but finite @xmath165 in the gp regime , the right hand side of eq . ( [ lda - gp ] ) can be made small by increasing the total number of particles @xmath3 . for a fixed coupling @xmath14 and a constant density @xmath171 [ such that @xmath165 stays constant ] , the increase of the particle number @xmath3 has to be accompanied by a reduction of the trap frequency @xmath172 . thus , the ratio @xmath173 can also be made small , so that that the lda criterion in the gp regime [ @xmath156 ] is satisfied for almost the entire sample , up to the location @xmath175 very close to the edge of the cloud . similar considerations apply to the tg regime [ @xmath176 , where the requirement on the ( large ) total number of particles @xmath3 is less stringent than in the gp regime . here we discuss the local pair correlation @xmath177 in a zero temperature trapped gas within the lda . the calculations are done using the solution to the lieb - liniger equation ( [ llequation ] ) and the hellmann - feynman theorem @xcite , eq . ( [ g2derivative ] ) . here , @xmath61 is now replaced by @xmath178 and @xmath179 is to be understood as the local value of @xmath133 . thus , to calculate @xmath178 as a function of the distance from the trap centre , one can use the uniform results in which the interaction parameter @xmath180 is found from the density profile @xmath119 , for different values of @xmath165 . the implementation of the lda , using the local effective chemical potential @xmath181 , eq . ( [ mu - z ] ) , is carried out by means of first calculating the chemical potential @xmath77 as a function of @xmath47 , and then inverting this dependence for obtaining @xmath119 . this gives @xmath119 [ and hence @xmath133 ] as a function of @xmath121 , for a given value of the interaction parameter @xmath164 at the trap center . depending on the value of the coordinate - dependent interaction parameter @xmath183 , eq . ( [ gamma - z ] ) , we have the following limiting behavior of the pair correlation function . in the gross - pitaevskii ( gp ) limit of a weakly interacting gas , the pair correlation in the uniform case is @xmath184 , @xmath185 @xcite . for a trapped gas , replacing @xmath186 by @xmath187 gives : @xmath188 where we have used the relationship between @xmath189 and @xmath119 and the fact that in the gp regime the density profile is given by the familiar thomas - fermi parabola : @xmath190 and @xmath191 for @xmath140 . here , the peak density @xmath169 and the radius @xmath192 are given by : @xmath193 in the tonks - girardeau ( tg ) limit of strong interactions , the uniform gas pair correlation is @xmath194 , @xmath195 @xcite . in the trapped gas case , replacing @xmath186 by @xmath189 gives @xmath196 where we again used the relationship between @xmath189 and @xmath119 and the fact that the density profile in the tg regime is given by the square root of parabola : @xmath197 and @xmath191 for @xmath140 . here , the peak density @xmath169 and the radius @xmath198 are @xmath199 of a zero temperature @xmath200 bose gas in a harmonic trap as a function of the dimensionless coordinate @xmath201 , for different values of the interaction parameter @xmath202 . the solid lines are the results of the exact numerical solution of the lieb - liniger equations within the lda . the dashed and the dotted lines are the analytic results given by the thomas - fermi parabola in the gp regime and the square root of parabola in the tg regime , respectively.,width=226 ] in fig . [ density - zerot ] we present the density profiles @xmath119 as a function of the dimensionless coordinate @xmath201 , for different values of @xmath164 . the full lines represent the results of the numerical calculation within the lda , which reproduce the results of ref . @xcite , while the dashed and the dotted lines represent , respectively , the above analytic results in the gp and tg regimes . bose gas at zero temperature , @xmath178 , as a function of the displacement from the trap center @xmath201 , for different values of @xmath203 . the full lines are the results of numerical calculation , while the dashed lines for @xmath204 and @xmath205 are the respective analytic results of eqs . ( [ g2gp ] ) and ( [ g2tg ] ) shown for comparison . , title="fig:",width=245 ] bose gas at zero temperature , @xmath178 , as a function of the displacement from the trap center @xmath201 , for different values of @xmath203 . the full lines are the results of numerical calculation , while the dashed lines for @xmath204 and @xmath205 are the respective analytic results of eqs . ( [ g2gp ] ) and ( [ g2tg ] ) shown for comparison . , title="fig:",width=247 ] figure [ g2-zerot ] shows the local pair correlation @xmath178 as a function of @xmath201 , for different values of @xmath165 . as we see , for the case of weak interactions @xmath156 , the pair correlation is close to unity in the central bulk part of the distribution . this is an expected result for the coherent or gp regime . as one approaches the tails of the distribution , where the density is reduced and @xmath133 becomes larger than one , the gas locally approaches the tg regime of ` fermionization ' . here , the pair correlation is suppressed below the coherent level , @xmath206 . for the cases where @xmath207 , including the tg regime of @xmath208 , the pair correlation is suppressed across the entire sample . in the limit @xmath209 , the gas acquires pronounced `` fermionic '' properties so that the wave function strongly decreases as particles approach each other , thus resulting in @xmath210 . an obvious choice of dimensionless interaction and temperature parameters for describing a trapped @xmath0 bose gas within the lda consists in using the local value of the interaction parameter @xmath165 and the reduced temperature @xmath211 in the trap center . these are the same parameters that are used in the uniform gas treatment @xcite , eqs . ( [ gamma ] ) and ( [ tau1 ] ) , except that now we define them via the local density @xmath119 . thus , in general , we define the local interaction parameter @xmath189 and the local reduced temperature @xmath212 , according to : @xmath213 where @xmath214 is the local temperature of quantum degeneracy that corresponds ( locally ) to the conditions where the mean inter - particle separation becomes of the order of the thermal de broglie wavelength . the values of these parameters at the trap center , @xmath165 and @xmath211 , completely characterize all relevant properties of the gas within the lda , including the associated density profiles @xmath119 , the resulting total number of particles @xmath3 , as well as the correlation functions and the thermodynamic properties . a completely equivalent pair of the interaction and temperature parameters , which is , however , more suitable for practical purposes is the local value of @xmath133 and a new temperature parameter @xmath215 defined via : @xmath216 according to this definition , the temperature is measured in units of the characteristic energy @xmath217 . the advantage of using @xmath215 as the dimensionless temperature parameter is that it is independent of the density and gives a direct measure of the global temperature of the gas , which in equilibrium is the same for the entire sample . this allows us to easily explore the `` interaction temperature '' parameter space [ @xmath218 in a systematic way . for example , considering different values of @xmath165 while @xmath215 is kept constant would correspond to physical conditions under which the peak density of the gas @xmath169 is varied while the absolute temperature @xmath87 is kept unchanged . a novel experimental technique that implemented this approach for achieving a bose - einstein condensation in a @xmath15 gas has recently been demonstrated in ref . @xcite . other alternative choices are possible for characterizing the interactions and temperature of a trapped gas in dimensionless units . for example , to characterize the system at different temperatures @xmath87 while the total number of particles @xmath3 is kept constant , one can define an alternative pair of global parameters which are more suitable for this case ( see sec . vi ) . here , the global temperature parameter can be defined as @xmath219 , where @xmath220 is the global temperature of quantum degeneracy of a trapped gas ( in energy units , @xmath221 , where @xmath222 is the boltzmann constant ) . irrespective of the interaction strength , a harmonically trapped bose gas at @xmath223 obeys the classical boltzmann statistics , whereas for @xmath224 quantum statistical effects become important . this is clearly seen in the limit of a trapped ideal gas ( @xmath225 ) and in the opposite limit of a strongly interacting gas ( @xmath226 ) . in the latter case , the problem maps onto the trapped gas of non - interacting fermions @xcite . so , in both limits @xmath227 appears explicitly as the temperature of quantum degeneracy for the trapped sample as a whole . here we analyze the local density approximation for a finite temperature gas , and obtain simple criteria for its validity in the limiting cases of very high and very low temperatures , @xmath223 and @xmath228 . in the high temperature limit , @xmath223 [ in which case @xmath229 , the density distribution @xmath119 can be approximated by a gaussian profile in all regimes , as the interaction between the particles is negligible compared to their thermal kinetic energies . for @xmath3 particles at temperature @xmath87 in a harmonic trap of frequency @xmath4 , the density profile is determined by the thermal distribution for a classical ideal gas described by boltzmann statistics : @xmath230 where the radius @xmath231 characterizes the width of the gaussian and is given by @xmath232 in this high - temperature limit , the correlation length is given by the thermal de broglie wavelength @xmath128 , so that the lda criterion ( [ lda - criterion ] ) gives : @xmath233 since @xmath234 implies that @xmath235 , the above lda criterion can be satisfied for all locations @xmath117 from the trap center up to distances equal to several characteristic widths @xmath231 . for sufficiently large total number of particles @xmath3 , the ratio @xmath236 will be even larger so that the lda will be valid for even larger distance from the trap center . in the opposite limit of low temperatures @xmath228 [ in which case @xmath237 for @xmath156 , and @xmath238 for @xmath163 ] , the density profile @xmath119 can be approximated by two contributions . the first one is for the central bulk part which will be close to the @xmath23 density profile up to a certain distance @xmath141 ( @xmath143 ) from the trap center . here @xmath137 is the zero - temperature thomas - fermi radius , eq . ( [ r - tf - gen ] ) . the second contribution is for the tails of the distribution which can be approximated by a thermal gaussian . as before , we will focus on the ` weak ' lda condition ( [ lda - criterion ] ) involving the correlation length @xmath115 . for the central part of the sample , up to distances @xmath141 ( @xmath143 ) , we can use the lda criterion derived for the zero temperature gas , eq . ( [ lda - criterion - final ] ) . in the weakly and strongly interacting limits this can be rewritten as before in terms of the interaction parameter @xmath164 and the total number of particles @xmath3 [ see eqs . ( [ lda - gp ] ) and ( [ lda - tg ] ) ] . for the gaussian tails of the distribution , i.e. at distances @xmath240 , where the local correlation length is given by @xmath128 , we use the above high - temperature result , eq . ( [ lda - hight ] ) , and rewrite it in terms of the thomas - fermi radius @xmath137 and the global chemical potential @xmath113 . as a result , the lda criterion for the gaussian tails of a low - temperature gas can be written in the following form : @xmath241 in the gp regime [ @xmath242 , where @xmath243 , this gives : @xmath244 thus , in order that the lda works in the tails of the density distribution ( @xmath245 . this requirement can always be satisfied with a sufficiently large number of particles @xmath3 . for example , for @xmath246 [ which can be obtained , for example , with @xmath247 and @xmath248 , according to the relationship @xmath249 valid in this regime ] one would need to have @xmath250 in order to satisfy the lda criterion for distances @xmath251 . in the tg regime [ @xmath163 ] , where @xmath252 , eq . ( [ lda - tails ] ) again reduces to the condition given by eq . ( [ lda - tails - gp ] ) . thus , in the tg regime the validity of the lda in the tails of the distribution again requires that @xmath253 . however , now we have @xmath254 , for @xmath228 . to summarize , in the low temperature limit the lda criterion can be easily satisfied for the central bulk part of the density distribution and for the gaussian tails . this leaves the question of validity of the lda in the low - density region near @xmath255 , where the density may vary more rapidly . we note , however , that at small finite temperatures the variation of the density profile around @xmath255 is more smooth than in the @xmath23 limit , so that the lda criterion may still be satisfied in this region , in contrast to the @xmath256 case where the lda necessarily breaks down as one approaches the edge of the cloud at @xmath255 . more importantly , the lda becomes valid again for distances past the small critical region around @xmath255 , i.e. in the tails of the density distribution . this means that the results of calculation of the pair correlation function @xmath178 at small finite temperatures should be valid everywhere except in a small region around @xmath255 . at high temperatures the lda criterion becomes less restrictive , and can be satisfied for the entire sample . the local pair correlation @xmath178 , eq . ( [ g2z ] ) , as a function of the location @xmath117 from the trap center is calculated using eq . ( [ g2-grand ] ) in which @xmath77 is replaced by the local chemical potential @xmath257 and where @xmath47 is the local density @xmath119 . the calculation is based on iterative numerical solution of the yang - yang exact integral equations @xcite for the excitation spectrum and for the distribution function of `` quasi - momenta '' , ( [ yy1 ] ) and ( [ yy2 ] ) . for a given set of values of @xmath121 , @xmath87 and @xmath14 , this gives the resulting density profile @xmath119 , eq . ( [ nzt ] ) , and the pressure @xmath98 , eq . ( [ p ] ) . differentiating @xmath98 with respect to @xmath14 gives the local pair correlation @xmath178 . a convenient way to implement the numerical algorithm for solving the yang - yang equations is via a dimensionless coordinate @xmath258 where the length scale @xmath231 is the thermal width of the classical gaussian distribution @xmath259 , given by eq . ( [ r - t ] ) . using the dimensionless coordinate @xmath260 , the local chemical potential can be rewritten as @xmath261 after setting up a lattice of @xmath260 values , \{@xmath262 } , the solution to the yang - yang equations proceeds as in the case of a uniform gas , with the input parameters being an array of the values of the local chemical potential @xmath263 , the temperature @xmath87 , and the coupling parameter @xmath14 , as described above . the final numerical results are then presented in terms of the dimensionless parameters @xmath165 and the temperature parameter @xmath211 ( or @xmath215 ) , where we note that @xmath264 this makes the output results scalable with respect to the physical parameters , rather than dependent of their absolute values . the total number of particles in the system is calculated from the resulting density profile @xmath119 via @xmath265 using the dimensionless coordinate @xmath260 and eq . ( [ nzovern0 ] ) this can be rewritten as @xmath266 so that the dimensionless ratio @xmath267 is given by @xmath268 this gives a relationship between the global and local dimensionless parameters and allows us to present the final results in a scalable fashion , rather than in terms of the absolute values of @xmath3 , @xmath87 , @xmath4 , and @xmath269 . here , the desired values of @xmath267 can be achieved by varying the ratio @xmath270 of the input parameters @xmath113 and @xmath87 . in order to understand the results for the pair correlations @xmath178 of a trapped @xmath0 bose gas , we first recall the classification of the regimes of a _ uniform _ gas . in ref . @xcite , these were identified using the results for the local pair correlation @xmath271 in terms of the interaction parameter @xmath52 and the reduced temperature @xmath95 . here , we give a brief summary of these results , except that we rewrite them in terms of the parameters @xmath52 and @xmath272 , instead of @xmath52 and @xmath95 . this is completely equivalent to the original pair . the new parameters @xmath52 and @xmath215 are more suitable for exploring the properties of the _ trapped _ gases , as discussed in sec . iv a. bose gas in the ( @xmath273-@xmath215 ) plane . the labels tg , gp , dq and dc refer to the tonks - girardeau , gross - pitaevskii , decoherent quantum and decoherent classical regimes , respectively . although all transitions are continuous , for purposes of discussion we classify the distinct regimes as follows : + tg : @xmath274 , @xmath275 ; tg(high - t ) : @xmath276 ; + gpa : @xmath277 , @xmath278 ; gpb : @xmath279 ; + dq : @xmath280 ; dc : @xmath281 . , width=245 ] the diagram representing these different regimes for a uniform 1d bose gas in the parameter space ( @xmath52-@xmath215 ) is shown in fig . [ regionstgamma ] . the regimes are classified as follows : _ strong coupling regime _ : in the strong coupling tg regime of ` fermionization ' , where @xmath195 and the temperature @xmath282 ( @xmath283 or @xmath284 ) , we only have a small correction compared to the zero temperature result @xcite : @xmath285;\,\,\,t\ll \gamma ^{-2},\,\gamma \gg 1;\,\,\,\text{tg}. \label{tg - low - t}\ ] ] in the case of strongly interacting nondegenerate bosons , where @xmath286 and the temperature @xmath287 ( @xmath288 or @xmath289 ) , we have the regime of high - temperature ` fermionization ' . despite the temperature @xmath287 , the local pair correlation is strongly suppressed ( @xmath290 ) @xcite : @xmath291 _ gross - pitaevskii regime : _ in the gp regime , where @xmath292 , the chemical potential is @xmath293 and at temperatures @xmath294 ( @xmath295 or @xmath296 ) the finite temperature correction to the zero temperature result is again very small @xcite : @xmath297 for @xmath298 ( @xmath299 or @xmath300 ) , the finite temperature correction is the leading one . it is important to recognize that the upper bound for the gp regime extends only up to temperatures of the order of @xmath301 ( @xmath302 or @xmath303 ) , and not to @xmath304 . here , the temperature @xmath305 is responsible for the presence of the quantum degeneracy , while @xmath306 for the presence of phase coherence . thus , for @xmath307 , the finite - temperature gp regime lies within the temperature interval @xmath308 ( @xmath309 or @xmath310 ) , and the pair correlation here is given by @xcite : @xmath311 _ decoherent regime : _ @xmath312 . due to the existence of two characteristic temperatures in the @xmath0 uniform gas , @xmath313 and @xmath305 , at temperatures @xmath87 higher than @xmath314 one has two sub - regions . for temperatures in the interval @xmath315 ( @xmath316 or @xmath317 ) the gas is in the decoherent quantum ( dq ) regime , while for @xmath287 ( @xmath318 or @xmath319 ) the gas is in the decoherent classical ( dc ) regime . in both cases the local pair correlation is close to @xmath320 @xcite : @xmath321 @xmath322 the result in the dc regime remains valid for large @xmath186 , provided @xmath323 ( or @xmath324 ) @xcite , and we can combine the required conditions on temperature via @xmath325 . in a harmonically trapped finite - temperature @xmath0 bose gas we again have a strong coupling regime , weak coupling gp regime , and a decoherent regime . the results for the local pair correlation @xmath178 in the first two regimes are easily obtained from eqs . ( [ tg - low - t])-([gpb ] ) by replacing the interaction parameter @xmath186 by the local @xmath117-dependent value @xmath189 of the trapped sample . however , it is convenient to rewrite the results for the local correlation @xmath326 in the trap center in terms of @xmath327 and the temperature parameter @xmath328 , where @xmath329 is the global temperature of quantum degeneracy of the sample as a whole . _ strong coupling regime . _ in the strong coupling tg regime , where @xmath330 and @xmath228 , the density profile is given by the thomas - fermi result , eq . ( [ tf - parabola - tg ] ) , and this allows one to establish the relationship between the temperature parameters @xmath215 and @xmath331 , using @xmath332 valid in this regime . thus , @xmath333 and eq . ( [ tg - low - t ] ) transforms into : @xmath334 , \label{tg - low - t - global}\ ] ] where @xmath335 , and @xmath330 . for the regime of high - temperature fermionization at @xmath223 , the density profile is given by the thermal gaussian , eq . ( [ thermal - gaussian ] ) , so that @xmath336 and hence @xmath337 . therefore , eq . ( [ tg - high - t ] ) transforms into : @xmath338 where @xmath339 . _ gross - pitaevskii regime . _ in the gp regime ( @xmath340 ) , for temperatures @xmath228 , the density profile is given by the thomas - fermi inverted parabola , eq . ( [ tf - parabola ] ) . we then have @xmath249 or @xmath341^{-3/2}/3 $ ] , so that eq . ( [ gpa ] ) transforms into @xmath342 where @xmath343 ( @xmath344 ) and @xmath340 . similarly , eq . ( [ gpb ] ) transforms into @xmath345 where @xmath346 ( @xmath347 ) . it is important to emphasize that for small @xmath327 and making the temperature sufficiently low ( @xmath348 ) the gas is always in the gross - pitaevskii regime . _ decoherent regime . _ for @xmath223 , a harmonically trapped @xmath0 bose gas is in the decoherent classical regime . here , the density profile is given by the thermal gaussian , eq . ( [ thermal - gaussian ] ) , so that @xmath349 and hence @xmath350 . accordingly , eq . ( [ dc - eq ] ) transforms into : @xmath351 where @xmath352 , neglecting a numerical factor of the order of one . thus , the validity of this result requires @xmath353 for @xmath354 , and @xmath355 for @xmath356 . and the local pair correlation @xmath357 as a function of the distance from the trap center @xmath358 , for @xmath359 ( @xmath360 ) and @xmath361 . this corresponds to case 1b in the diagram of fig . [ linesofconstt ] below.,title="fig:",width=215 ] and the local pair correlation @xmath357 as a function of the distance from the trap center @xmath358 , for @xmath359 ( @xmath360 ) and @xmath361 . this corresponds to case 1b in the diagram of fig . [ linesofconstt ] below.,title="fig:",width=215 ] in the case of small @xmath327 and at temperatures @xmath362 , one has a crossover from the classical decoherent ( @xmath223 ) regime to the gp ( @xmath228 ) regime . the properties of the gas in this region can be treated as containing ( locally ) features of the decoherent quantum regime of the uniform gas . an example illustrating this behaviour is given in fig . [ dq ] where we plot the density profile @xmath119 and the local pair correlation @xmath178 as a function of @xmath358 where @xmath363 is the thomas - fermi radius in the gp regime . these are calculated numerically using the solution to the yang - yang integral equations , with a value of @xmath364 at the trap center . the temperature in this example is @xmath359 , which is intermediate between the decoherent classical and the gp regimes . locally , the tails of the density profile are in the decoherent classical regime . on the other hand , the central part has features of the decoherent quantum regime . the figure for @xmath178 shows that fluctuations well above the coherent level , with @xmath365 , can occur even at temperatures below the transition to a quantum gas . however , with further temperature reduction well below @xmath227 , the density profile shrinks and one always has a coherent gp regime . for smaller values of @xmath52 , the temperature at which the coherent regime emerges must in general be calculated numerically . as a function of the reduced temperature @xmath215 , for different values of the interaction parameter @xmath273 . the solid lines are the exact numerical results , while the dashed lines correspond to the approximate analytic result of eq . ( [ eq : g2edge ] ) . , width=226 ] in fig . [ g2edgepic ] , we illustrate different regimes by plotting the local pair correlation @xmath271 as a function of the temperature parameter @xmath366 , for different values of @xmath186 . for sufficiently large @xmath52 the pair correlation approaches a universal function of the parameter @xmath215 : @xmath367 this is because by increasing @xmath52 one can always reach locally the condition @xmath368 . then , for @xmath369 one has ( locally ) the regime of high - temperature fermionization and can use eq.([tg - high - t ] ) , whereas for @xmath370 the sample will be in the decoherent classical regime described by eq.([dc - eq ] ) . this has an interesting consequence at sufficiently low temperatures @xmath369 . for @xmath195 , which is always the case for far tails of the density distribution , the pair correlation remains suppressed below the coherent level ( @xmath371 ) rather than approaches the value of @xmath372 . this occurs despite the gas is locally not quantum degenerate at low density . one thus sees that fermionization in which the bose gas develops antibunching with @xmath373 is an explicitly low - temperature phenomenon , when the temperature is scaled relative to the interaction strength . however , for @xmath55 the suppression of pair correlations is not temperature - independent . instead , the numerical results for increasing @xmath179 converge to a single universal function of @xmath215 . in the case of a trapped gas , the same diagram of fig . [ regionstgamma ] also describes the spatial variation of the gas within the lda . the parameter @xmath52 now becomes position dependent , @xmath133 , due to the dependence on the density @xmath119 . in this diagram , any point on the ( @xmath374-@xmath215 ) plane can be thought of as representing the interaction parameter @xmath164 evaluated at the trap center , and the dimensionless temperature @xmath375 . this is sufficient to completely characterize the properties of the trapped gas . the subsequent local values of @xmath133 of such a gas as one moves from the trap center towards the tails of the density distribution can be represented by a horizontal line drawn in the direction of increasing @xmath133 at constant @xmath215 . , except with four horizontal lines at different temperatures @xmath215 . this explores different density profiles in the parameter space [ @xmath203-@xmath215 ] , where the points 1a-1c , 2a-2c , 3a-3c , and 4a-4c ( marked by circles ) are representative examples corresponding to different values of @xmath203 at the trap center , at different temperatures @xmath215 . , width=226 ] this is shown in fig . [ linesofconstt ] , where the four horizontal lines correspond to four different temperatures @xmath215 , while various points along each line represent different ` initial ' values of the interaction parameter @xmath164 . the interval in the left lower corner of the diagram shows the displacement ( notice the logarithmic scale ) for which the local value of @xmath183 is increased by a factor of @xmath377 . this corresponds to a @xmath377 fold decrease in the density @xmath119 . for any given distribution with the value of @xmath165 in the center , this interval helps to immediately determine what fraction of the density profile relative to the peak density @xmath169 is contained within a certain regime . in the very far tails ( @xmath378 ) of any density distribution , where @xmath119 vanishes and @xmath379 , we always enter either the dc or the high - temperature tg regime , depending on the temperature @xmath215 . in addition , by considering a sample at any fixed temperature @xmath215 , while the peak density is increased [ @xmath165 is decreased ] , one can always reach the situation where the bulk of the density distribution is in the gp regime where @xmath380 . physically , this can be achieved by adding more particles to the system while maintaining the same global temperature @xmath87 , under constant coupling @xmath14 . from fig . [ g2edgepic ] it is clear that the density required may be relatively high , with @xmath381 being necessary to have a limiting value of @xmath382 at @xmath383 , as an example . for @xmath384 , there is no coherent gp regime over the entire range of temperatures . to illustrate different examples , we now calculate the density profiles @xmath385 and the local pair correlations @xmath178 as a function of the distance from the trap center @xmath117 . the distance @xmath117 is conveniently plotted in units of the thomas - fermi radius in the gp regime , @xmath363 , given by by eq . ( [ tf - gp - radius ] ) . the relationship between @xmath358 and the dimensionless coordinate @xmath386 is : @xmath387 . for a gas with a given coupling constant @xmath14 , the temperature parameter @xmath215 gives the measure of the absolute temperature @xmath87 . in this sense , the examples with @xmath388 and @xmath389 in fig . [ linesofconstt ] represent high- and low - temperature limits , which we analyze separately . and the local pair correlation @xmath178 as a function of @xmath358 for a harmonically trapped @xmath0 bose at different temperatures @xmath215 : @xmath390 ( first column ) and @xmath391 ( second column ) . the values of the interaction parameter @xmath165 in the trap center for each of the curves are as follows : 1a @xmath392 , 1b @xmath393 , 1c @xmath394 , 2a @xmath395 , 2b @xmath396 , and 2c @xmath397 . the dashed lines represent the thomas - fermi inverted parabola , eq . ( [ tf - parabola ] ) . the density profiles corresponding to the lower density cases 1c , and 2c are well approximated by the thermal ( gaussian ) distribution for a classical ideal gas , eq . ( [ thermal - gaussian ] ) , and are omitted from the graphs for clarity . the respective pair correlations for these low density cases are almost constant along the entire sample and are given by the value of @xmath178 in the tails of the distribution @xmath378 . depending on the temperature @xmath215 , these values can be determined using the results of fig . [ g2edgepic ] at @xmath135 [ see the curve for @xmath398 . , height=283 ] in fig . [ nz - g2 - 12 ] we present examples of calculated density profiles @xmath399 and local pair correlations @xmath178 , for the high - temperature cases of @xmath390 and @xmath391 . the examples shown correspond to the points marked by circles 1a-1c and 2a-2c in the diagram of fig . [ linesofconstt ] . for each temperature @xmath215 , the sequence of points 1a , 1b and 1c ( and similarly for 2a , 2b and 2c ) corresponds to a decreasing peak density of the gas [ increasing values of @xmath165 ] , while the absolute temperature @xmath87 is kept constant . this can be achieved by decreasing the total number of particles @xmath3 in the sample , at constant @xmath87 . the examples 1c and 2c represent a low - density ( non - degenerate ) gas in the decoherent classical ( dc ) regime . the corresponding density profiles are well approximated by a thermal gaussian , eq . ( [ thermal - gaussian ] ) , and are omitted from the graphs for clarity . the respective second - order correlation functions @xmath178 , display large thermal ( gaussian ) density fluctuations with @xmath401 . moving along the horizontal lines in the direction of decreasing @xmath165 ( starting from the points 1c or 2c for each temperature @xmath215 ) corresponds to increasing peak densities of the gas . as a result one crosses the respective boundaries and enters different regimes of quantum degeneracy shown in fig . [ regionstgamma ] . here , the limiting regime as @xmath402 at constant @xmath215 is the gp regime where the density profiles are well approximated by the thomas - fermi parabola ( [ tf - parabola ] ) ( see the graphs corresponding to 1a and 2a ) , while the pair correlation in the bulk of the density profile is close to that of the coherent level @xmath403 . the intermediate values of @xmath165 are represented by the examples 1b and 2b which have density profiles that are intermediate between a gaussian and the inverted parabola . the respective pair correlations @xmath178 also take intermediate values @xmath404 . in the example 2b , however , the central part of the density profile is in the gp regime , so that the departures form the coherent level of fluctuations are only seen in the tails of the density profiles . , except for : @xmath405 ( first column ) and @xmath406 ( second column ) . the values of the interaction parameter @xmath165 in the trap center for each of the curves are as follows : 3a @xmath407 , 3b @xmath408 , 3c @xmath409 , 4a @xmath410 , 4b @xmath411 , and 4c @xmath412 . the dashed lines represent the thomas - fermi inverted parabola in the gp regime , eq . ( [ tf - parabola ] ) , while the dotted lines correspond to the thomas - fermi square root of parabola in the tg regime , eq . ( [ tf - parabola - tg ] ) . , height=283 ] next , we consider the low - temperature behaviour , in which evidence for the tonks - girardeau ` fermionization ' can occur . fig . [ nz - g2 - 34 ] represents examples of the density profiles and pair correlations for a gas with lower values of the temperature parameter @xmath215 : @xmath405 and @xmath413 . the examples shown correspond to the points 3a-3c and 4a-4c in the diagram of fig . [ linesofconstt ] . as wee see , for @xmath414 the gas is in the tonks - girardeau regime . comparing this with the earlier high - temperature examples , we see that for a given density with @xmath414 , achieving the tonks - girardeau regime requires lower temperatures , @xmath389 . here again , the low - density [ large @xmath165 ] examples of 3c and 4c have density profiles that are well approximated by the thermal gaussian , eq . ( [ thermal - gaussian ] ) , and are omitted from the graphs for clarity . however , the pair correlations do not display large thermal fluctuations , but rather are suppressed below the coherent level , @xmath415 . this reflects the fact that the gas is in the regime of high - temperature `` fermionization '' . the example 4c corresponds to the lowest temperature parameter @xmath215 , which at constant peak density [ or constant @xmath165 ] corresponds to the largest interaction strength @xmath14 , hence the smallest value of @xmath178 . the example 4b is deep enough in the tg regime , and we see that the density profile is close to the thomas - fermi square root of parabola , eq . ( [ tf - parabola - tg ] ) , while the pair correlation is well below the coherent level @xmath416 . the example 4a is for a smaller value of @xmath417 [ higher peak density ] , which is closer to the boundary with the gp regime . as a result , the shape of the density profile departs from the respective tg result and is intermediate between the tg and gp parabolas , while the pair correlation in the central part of the density distribution increases . finally , the examples 3b and 3a are for intermediate values of @xmath215 and @xmath165 which are not well described by analytical approaches . in all these examples the limiting behaviour of the pair correlation @xmath357 in the far tails of the density distribution is described by a universal function of @xmath215 , as discussed earlier ( see fig . [ g2edgepic ] ) . the overall conclusion that can be drawn from this analysis is that the local pair correlation @xmath178 can vary between a broad range of values between zero and two and has a rich built - in structure . it provides far more sensitive information about the regimes of trapped @xmath0 bose gases than the respective density profiles . here we investigate the properties of a trapped gas at different temperatures @xmath87 and constant total number of particles @xmath3 . since the overall picture in terms of the density profiles and the behaviour of the local pair correlation has already been understood in terms of the diagram of figs . [ regionstgamma ] and [ linesofconstt ] , it is now sufficient to only monitor the changes in the temperature parameter @xmath215 and the value of the interaction parameter @xmath165 under conditions when @xmath3 is kept constant , and then map these changes into the [ @xmath215-@xmath165 ] plane . thus , for a given system with the coupling @xmath14 , trap frequency @xmath4 , and the total number of particles @xmath3 , our task consists of calculating the density profiles @xmath119 at different temperatures @xmath87 , with the constraint that the total number of particles remains unchanged . once this is done , we identify the respective values of the dimensionless temperature parameter @xmath215 and the local value of @xmath165 and plot these on the [ @xmath215-@xmath165 ] plane of fig . [ regionstgamma ] . more specifically , instead of performing this analysis for absolute values of physical parameters , we first identify new dimensionless variables for the temperature and interaction strength that are more suitable under these conditions . the new parameters we introduce are the global interaction parameter @xmath418 and the global reduced temperature @xmath419 : @xmath420 @xmath421 where @xmath2 . bose gas as in fig . [ regionstgamma ] , except that the curved lines @xmath422 represent the locations of the interaction parameter @xmath203 and the reduced temperature @xmath215 , for four different ( fixed ) values of the global interaction parameter @xmath423 while the global temperature @xmath424 is changing . this represents four different samples with fixed total number of particles @xmath3 and varying absolute temperature @xmath87 . for each point on a given line , there exists an associated density profile with the peak density @xmath169 corresponding to the respective value of @xmath425 , and the local values @xmath119 corresponding to the values of @xmath426 in the horizontal direction to the right . the values of the global interaction parameter @xmath423 for each line are : ( 1 ) @xmath427 , ( 2 ) @xmath428 , ( 3 ) @xmath429 , and ( 4 ) @xmath430.,width=226 ] the definition of the global interaction parameter @xmath418 , eq . ( [ tildegamma - def ] ) , relies on the identity @xmath431 using the definitions of the local parameters @xmath211 and @xmath165 , we see that @xmath418 is the square root of the ratio of two energy scales , @xmath432 and @xmath433 , as in eq . ( [ tildegamma - def ] ) ( see also ref . @xcite ) . the definitions of the dimensionless temperature and interaction parameters @xmath419 and @xmath418 both include the total number of particles @xmath3 . this is more suitable for analyzing the properties of the gas under conditions of changing temperature at constant @xmath3 . in fig . [ linesofconstn ] we present the results of calculation of the density profiles for four different ( fixed ) values of the global interaction parameter @xmath418 while the temperature @xmath419 is changed within a broad range of values , typically between @xmath434 . the results are summarized by plotting the path of the resulting local values of the interaction parameter @xmath165 at the trap center and the reduced temperature @xmath215 , in the parameter space of fig . [ regionstgamma ] . as a function of the temperature @xmath435 at constant total number of particles this is the same data as in fig . [ linesofconstn ] except plotted in the @xmath436-@xmath424 plane , where @xmath437.,width=226 ] for quantitative purposes , we also present the same data in the @xmath331-@xmath438 plane , which is shown in fig . [ theta - gamma0 ] . while fig . [ linesofconstn ] identifies the local regimes of the gas with constant @xmath3 , fig . [ theta - gamma0 ] helps to understand the properties of the gas in terms of the variation in the global temperature parameter @xmath419 . note that when @xmath418 is kept constant , the variations in the temperature parameters @xmath419 and @xmath215 are essentially equivalent and scale as @xmath439 @xmath440 , according to eq . ( [ global - local - relationship ] ) . there are simple approximate relations between the global and local interaction parameters @xmath418 and @xmath165 at high and low temperatures . at high temperatures ( @xmath441 ) the relationship is given by : @xmath442 . \label{tildegamma - gamma0-hight}\ ] ] at low temperatures ( @xmath443 ) , and in the limiting gp [ @xmath156 ] and tg [ @xmath163 ] regimes , the relationship between @xmath418 and @xmath165 becomes independent of @xmath419 and is given , respectively , by : @xmath444 , \\ \tilde{\gamma } & \simeq\frac{1}{\pi}\gamma ( 0),\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;[\theta\ll1,\;\gamma ( 0)\gg1].\end{aligned}\ ] ] this implies that the gp and tg regimes can equivalently be defined via @xmath445 or @xmath165 . the gp regime corresponds to @xmath446 or @xmath156 , while the tg regime will correspond to @xmath447 or @xmath163 . from fig . [ theta - gamma0 ] we see that at high temperatures and constant @xmath445 , the local interaction parameter @xmath165 varies according to @xmath448 , i.e. linearly in the logarithmic scale of fig . [ theta - gamma0 ] and in agreement with eq . ( [ tildegamma - gamma0-hight ] ) . this means that the response of the peak density @xmath169 to temperature changes at constant @xmath3 follows the power law of @xmath449 , which is an expected result for the thermal distribution of a classical ideal gas , eq . ( [ thermal - gaussian ] ) . as the temperature is reduced , the response of @xmath169 to the temperature changes becomes modified , and the modification is quite different depending on the interaction strength @xmath418 . for weak interactions ( @xmath450 ) , as the temperature @xmath87 is reduced below @xmath451 ( @xmath452 ) , the peak density @xmath169 first increases more rapidly than in a thermal gas , and then the growth is saturated as the temperature is reduced further ( see curves ( 1 ) and ( 2 ) in fig . [ theta - gamma0 ] ) . at very low temperatures ( @xmath443 ) , the peak density @xmath171 approaches a constant value independent of temperature . this is a typical behaviour found in a weakly interacting gas that undergoes quasi - condensation and reaches the gp regime . for intermediate and strong interactions ( @xmath453 ) , on the other hand , the response of @xmath171 to the temperature reduction is different . instead of an initial speed up , the growth of the peak density @xmath169 directly goes to the regime of saturation , once the temperature is reduced below the global temperature of quantum degeneracy @xmath227 ( see curves ( 3 ) and ( 4 ) ) . at very low temperatures , @xmath169 again approaches a constant value independent of the temperature and the gas ends up in the tg regime . from the paths of the curves ( 3 ) and ( 4 ) in fig . [ linesofconstn ] , we see that achieving the tg regime from a high temperature classical gas by means of reducing the temperature @xmath87 at constant @xmath3 requires large values of @xmath445 in the first place . this can be achieved by having a relatively small total number of particles @xmath3 or a small trap frequency @xmath172 , according to eq . ( [ tildegamma - def ] ) . while the pair correlation @xmath178 provides detailed information about the local correlation properties of a trapped gas , its measurement as a function of @xmath117 may not be an easy task in practice . here , one usually probes the pair correlation @xmath454 within a finite volume , e.g. via the measurement of the rates of two - body inelastic processes within the entire sample . this means that one probes the integrated or averaged correlation properties of the gas , as has been demonstrated in a recent experiment of ref . . we are therefore motivated to study the average pair correlation defined via : @xmath455 here , an interesting question arises of whether this average correlation has a simple relationship with the local pair correlation at the trap center @xmath456 . the reason to expect this is the fact that @xmath178 under the integral in eq . ( [ g2_av_def ] ) is multiplied by @xmath457 which vanishes rapidly as one approaches the tails of the density profile . the function @xmath178 near the trap center , on the other hand , varies slowly and can be approximated by the value of @xmath456 . therefore , we can approximate @xmath178 under the integral by a constant @xmath326 , thus reducing eq . ( [ g2_av_def ] ) to @xmath458 ( solid lines ) and the normalized average pair correlation @xmath459 ( dashed lines ) as a function of the interaction parameter @xmath460 , for four different temperatures @xmath215 . for each temperature , the respective lines monitor the values of @xmath326 and @xmath461 as one moves along the lines of constant @xmath215 in fig . [ linesofconstt].,width=245 ] thus , the average pair correlation @xmath462 can be expressed via the local pair correlation @xmath326 using a simple relationship , eq . ( [ g2_av ] ) . note that this also requires an independent evaluation of the integral of the squared density , @xmath463 . introducing a normalized average pair correlation , @xmath459 , we obtain @xmath464 in fig . [ g2_0 ] , we plot the local pair correlation at the trap center @xmath456 and the normalized average pair correlation @xmath461 as a function of the interaction parameter @xmath327 , for four different temperatures @xmath215 . each line monitors the values of @xmath326 and @xmath459 as one moves along the lines of constant @xmath215 in fig . [ linesofconstt ] . here , the sequence of points along the horizontal axis refers to the value of @xmath327 of the associated density profile @xmath399 , for which we first calculate the pair correlation @xmath178 as a function of @xmath117 [ which includes the plotted values of @xmath326 ] and then evaluate the integral in eq . ( [ g2_av_def ] ) to obtain the average correlation @xmath462 , and hence @xmath459 . as we see , in the limit of small @xmath327 the pair correlation approaches the coherent level of fluctuations with @xmath466 , while at large @xmath327 it can take any value between zero and two , depending on the temperature @xmath215 . by comparing the full and dashed lines in fig . [ g2_0 ] , we see that the normalized average pair correlation @xmath459 can indeed be well approximated by the local pair correlation in the trap center @xmath326 . this is an important result and may have useful applications in practice . for example , it gives a direct justification of the analysis performed in ref . @xcite where the results of the measurements of three - body recombination rates in a bulk trapped sample have been compared with theoretical predictions @xcite for a uniform gas . here , we return to the analysis of section v , with the reference to fig . [ linesofconstt ] , and complete it by providing the results of calculation of the total number of particles @xmath3 . more specifically , we give the results for the dimensionless ratio @xmath467 as a function of the local interaction parameter @xmath165 taken along the horizontal lines of fig . [ linesofconstt ] , i.e. at four different ( fixed ) values of the temperature parameter @xmath215 . this is shown in fig . [ n - gamma0 ] . for a given trap frequency @xmath4 and coupling @xmath14 , each line corresponds to monitoring the variation in the total number of particles @xmath3 as a function of the peak density @xmath169 , at constant temperature @xmath87 . as a function of @xmath202 , for four different values of the temperature parameter @xmath215 . for a given coupling @xmath14 and a fixed @xmath87 , this monitors the variation of @xmath468 as one moves along the respective horizontal line in fig . [ linesofconstt ] . here , each point along the horizontal line is being referred to the value of @xmath203 of the associated density profile , for which we first calculate the density @xmath119 as a function of @xmath117 and then evaluate the resulting total number of particles @xmath469 to form the dimensionless ratio @xmath470.,width=226 ] fig . [ n - gamma0 ] can also be viewed as giving the variation of the interaction parameter at the trap center @xmath165 as a function of @xmath471 , which in turn corresponds to monitoring the change in the peak density of the gas @xmath169 as the total number of particles @xmath3 is varied at constant temperature @xmath87 . starting from the regime of low particle numbers ( @xmath472 or high temperatures @xmath473 ) , we see that the increase in @xmath3 results in a linear increase of the peak density , @xmath474 . this is an expected result for the thermal density distribution of a classical ideal gas , eq . ( [ thermal - gaussian ] ) , and corresponds to the linear dependence of @xmath475 on @xmath165 as seen in fig . [ n - gamma0 ] . as the number of particles is increased further and the ratio @xmath476 goes through the critical region @xmath467 @xmath477 ( corresponding to temperatures of the order of the global temperature of quantum degeneracy , @xmath478 ) the growth of the peak density @xmath169 speeds up , for the lines corresponding to @xmath390 and @xmath391 . this speed up is most prominent in the first case corresponding to very week interactions , and reflects the fact that the gas undergoes quasi - condensate formation where the added particles mostly go into the trap center rather than into the tails of the density distribution . with further increase of @xmath471 , the growth of the peak density @xmath169 is slower than in the classical gas case and the situation is now reversed : the added particles mostly go into the tails . here , the gas is deep in the gp regime and the density distribution is given by the inverted parabola , eq . ( [ tf - gp - density ] ) . for the cases of @xmath405 and @xmath406 , which at constant @xmath87 correspond to stronger inter - particle interactions , the change in the slop of the respective curves in fig . [ n - gamma0 ] around @xmath471 @xmath477 represents the fact that the growth of the peak density @xmath169 only slows down when the number of particles is increased past this critical region . here , the gas goes first through the tg regime where the density profile is given by the square root of inverted parabola , eq . ( [ tf - tg - density ] ) , and eventually it enters the gp regime ( see fig . [ linesofconstt ] ) with gradual transformation of the shape of the density profile to the gp parabola . apart from providing additional information about the properties of trapped gases , fig . [ n - gamma0 ] can also serve for quantitative analysis relevant to practice . as an example , we consider a gas of @xmath479rb atoms ( @xmath480 kg , @xmath481 nm ) with the aim of identifying a set of physical parameters that correspond to the conditions of the point 3b in fig . [ linesofconstt ] . here , @xmath408 and @xmath405 , which is an example of a moderately `` fermionized '' gas . the pair correlation @xmath178 is below the coherent level @xmath416 , yet the gas is not in the extreme low - temperature tg regime and therefore the density profile can not be approximated by the square root of parabola . in this sense , this example would be easier to realize in practice than the extreme tg regime . we first consider the trapping potential with @xmath482 hz and @xmath483 khz ( @xmath484 m and @xmath485 m ) . with the @xmath479rb scattering length of @xmath486 nm , giving @xmath487 j / m , and with the value of @xmath408 that we are aiming at , this set of parameters results in the peak @xmath0 density of @xmath488 m @xmath489 ( and hence @xmath490 m@xmath491 ) . at this stage , we can identify that the conditions of eq . ( [ 1d - regime ] ) for achieving the @xmath0 regime are satisfied . next , the aimed value of @xmath405 , together with the value of @xmath14 found above , gives the required absolute temperature @xmath492j ( in energy units , or @xmath493 nk ) . we next refer to the results of fig . [ n - gamma0 ] and read off the value of the dimensionless ratio @xmath494 ( or equivalently @xmath495 ) , which corresponds to @xmath408 on the respective @xmath405 line . finally , using the values of @xmath4 and @xmath87 , we find that the required total number of particles here is @xmath496 . we note that all of the above parameter values are close to the conditions realized in recent experiments @xcite . in summary , we have obtained predictions for the correlations and density profiles of a one - dimensional trapped bose gas at finite temperature . this allows previous results for the uniform 1d bose gas to be applied to the experimentally relevant case of a harmonic trap . the calculations use a local density approximation which is asymptotically correct in the limit of a large trap with a sufficiently slowly varying trap potential we find that , in this limit , there is a similar classification of different coherence regimes as in the uniform case . remarkably , the density variation across the trap does not cause a dramatic change in the average correlation function compared to the value at the trap center . this is because the correlations are found to be relatively uniform in the high density region near the trap center , which dominates any trap - averaging measurement . this is particularly useful for experiments that measure correlation functions through averaging a nonlinear interaction over the length of the trap , which is the simplest currently available procedure . k.k . and p.d . acknowledge the australian research council for the support of this work . d.g . and g.s . acknowledge support from the ministre de la recherche ( grant aci nanoscience 201 ) , from centre national de la recherche scientifique ( cnrs ) , and from the nederlandse stichting voor fundamenteel onderzoek der materie ( fom ) . the research was also supported in part by the national science foundation under grant no . phy99 - 07949 . laboratoire kastler brossel is a research unit ( umr 8552 ) of universit pierre et marie curie and ens , associated with cnrs . lptms is a research unit ( umr 8626 ) of cnrs and universit paris sud . 99 a. grlitz , j. m. vogels , a. e. leanhardt , c. raman , t. l. gustavson , j. r. abo - shaeer , a. p. chikkatur , s. gupta , s. inouye , t. rosenband , and w. ketterle , phys . lett . * 87 * , 130402 ( 2001 ) ; f. schreck , l. khaykovich , k. l. corwin , g. ferrari , t. bourdel , j. cubizolles , and c. salomon , phys . lett . * 87 * , 080403 ( 2001 ) ; m. greiner , i. bloch , o. mandel , t. w. hnsch , and t. esslinger , phys . lett . * 87 * , 160405 ( 2001 ) . b. paredes , a. widera , v. murg , o. mandel , s. flling , i. cirac , g. v. shlyapnikov , t. v. hnsch , and i. bloch , nature * 429 * , 277 ( 2004 ) ; see also : m. a. cazalilla , phys . a * 70 * , 0416604(r ) ( 2004 ) . e. h. lieb and w. liniger , phys . rev . * 130 * , 1605 ( 1963 ) ; e. h. lieb , _ ibid _ , * 130 * , 1616 ( 1963 ) . note that the value of @xmath61 at zero temperature is also implicit in the interaction energy calculated here . s. j. carter , p. d. drummond , m. d. reid , and r. m. shelby , phys . * 58 * , 1841 ( 1987 ) ; m. rosenbluh and r. m. shelby , phys . lett . * 66 * , 153 ( 1991 ) ; p. d. drummond , r. m. shelby , s. r. friberg , and y. yamamoto , nature * 365 * , 307 ( 1993 ) . an alternative exact numerical approach using stochastic gauge methods in imaginary time has been recently developed and used for calculating _ non - local _ pair correlations in : p.d . drummond , p. deuar , and k.v . kheruntsyan , phys . * 92 * , 40405 ( 2004 ) . well below the temperature of quantum degeneracy @xmath497 in the gp and tg regimes ( @xmath77 is the chemical potential ) , and otherwise @xmath22 is of the order of @xmath128 . in the gp regime @xmath22 coincides with the healing length , @xmath498 , and in the tg regime @xmath499 , where @xmath47 is the linear ( @xmath0 ) density . the global interaction parameter @xmath500 can alternatively be defined via @xmath501 , i.e. via the ratio of the thermal de broglie wavelength @xmath502 at the global temperature of quantum degeneracy , @xmath2 , to the @xmath0 scattering length @xmath503 , including a numerical factor @xmath504 for convenience . this would correspond exactly to the conventional definition of the dimensionless interaction parameter @xmath52 in the _ uniform _ @xmath0 bose gas problem , if one defines this via the thermal de broglie wavelength @xmath505 @xmath506 at the respective temperature of quantum degeneracy of the uniform gas , @xmath507 , and the @xmath0 scattering length @xmath508 : @xmath509 .	we calculate the density profiles and density correlation functions of the one - dimensional bose gas in a harmonic trap , using the exact finite - temperature solutions for the uniform case , and applying a local density approximation . the results are valid for a trapping potential which is slowly varying relative to a correlation length . they allow a direct experimental test of the transition from the weak coupling gross - pitaevskii regime to the strong coupling , ` fermionic ' tonks - girardeau regime . we also calculate the average two - particle correlation which characterizes the bulk properties of the sample , and find that it can be well approximated by the value of the local pair correlation in the trap center .
the filled skutterudite compound pros@xmath0sb@xmath1 displays superconductivity with @xmath6 and has an effective mass @xmath7 . @xcite this compound is the first example of a pr - based heavy fermion superconductor ( all other known heavy fermion superconductors are intermetallic compounds of the rare earth element ce or the actinide element u ) . inelastic neutron scattering experiments , along with an analysis of magnetic susceptibility @xmath8 and specific heat @xmath9 ) data @xcite for a cubic crystalline electric field ( cef ) , yield a pr@xmath10 energy scheme consisting of a nonmagnetic @xmath11 doublet ground state ( ) , a @xmath12 triplet first exited state ( ) , and higher energy @xmath13 triplet ( ) and @xmath14 singlet ( ) excited states . the heavy fermion properties of the pr - based compounds prinag@xmath15 and prfe@xmath0p@xmath1 have been attributed to the interaction of the charges of the conduction electrons with the electric quadrupole moments of the pr@xmath10 @xmath11 nonmagnetic doublet ground state in the cef . @xcite the evidence for a pr@xmath10 @xmath11 ground state in pros@xmath0sb@xmath1 indicates that the electric quadrupolar fluctuations may be responsible for the heavy fermion state in this compound and could also be involved in the superconductivity . for magnetic fields @xmath16 above @xmath17 t , sharp features in measurements of the normal state electrical resistivity @xcite @xmath18 , magnetization @xcite @xmath19 , specific heat @xcite @xmath20 and thermal expansion @xcite @xmath21 of pros@xmath0sb@xmath1 indicate that a phase transition is induced at high fields . the origin of the high - field ordered phase ( hfop ) is still under investigation but may be related to the crossing of the zeeman levels of the @xmath11 and @xmath12 cef states and a corresponding change of the ground state at high fields . @xcite in this report , we present further results of our investigation of the electrical resistivity measurements up to and magnetization measurements with magnetic field @xmath22 $ ] and [ 001 ] crystallographic directions up to . the pros@xmath0sb@xmath1 samples studied were single crystals grown in sb flux . @xcite x - ray diffraction measurements confirmed the cubic lafe@xmath0p@xmath1-type structure . the @xmath23 measurements were made with a linear research lr 700 4-wire ac bridge operating at 16 hz with constant currents of ( @xmath24 ) and ( @xmath24 ) , and in a transverse geometry in a @xmath25he-@xmath26he dilution refrigerator @xcite ( @xmath24 ) and a @xmath25he cryostat ( @xmath24 ) . the @xmath27 measurements were performed in a faraday magnetometer in a @xmath25he refrigerator with a gradient field of . the behavior of electrical resistivity @xmath28 below and is summarized in fig . [ fig : rhot&h ] . figs . [ fig : rhot&h](a)[ref . 6 ] and ( b ) show @xmath18 data in various constant fields and fig . [ fig : rhot&h](c ) shows isotherms of @xmath29 . in this temperature range , the phonon contribution to @xmath28 is negligible . @xcite the data in figs . [ fig : rhot&h](b ) and ( c ) were measured on a different pros@xmath0sb@xmath1 sample and in a different cryostat . the absolute value of the resistivity is sample - dependent in the present work , probably due to irregular sample shapes and the presence of microcracks in the samples . the @xmath30 data in figs . [ fig : rhot&h](b ) and ( c ) are normalized by comparing the @xmath18 data at 10 t for both samples , which differ by a factor of @xmath31 . the sharp drops in @xmath18 below are due to the superconducting transition . above @xmath32 , a kink starts to develop in the @xmath18 curves and becomes most pronounced between and , then gradually subsides as @xmath16 increases . this kink is apparently related to the occurrence of a high field ordered phase which will be discussed later ( fig . [ fig : phasediagram ] ) . the @xmath18 data between and almost overlap with each other ( fig . [ fig : rhot&h](a ) and ref . 6 ) . the @xmath29 data below 1 k reach a maximum at @xmath32 as shown in fig . [ fig : rhot&h](c ) , while the dome - shaped feature becomes more pronounced as @xmath33 decreases . two sharp kinks ( @xmath34 and @xmath35 ) become easily identified below and mark the boundary of the high field ordered phase . @xcite the magnetization @xmath19 measurements were performed with the applied magnetic field oriented along the [ 111 ] and [ 001 ] crystallographic directions of a single crystal of pros@xmath0sb@xmath1 for 0 t @xmath36 5.5 t. the inset of fig . [ fig : peh111 ] only shows the isothermal @xmath19 data for @xmath37 $ ] . a kink ( at @xmath34 ) appears in @xmath19 curves above 4.5 t and can not be resolved above 0.8 k for @xmath38 5.5 t. this feature is associated with the high field ordered phase . we found that the fields where the @xmath19 kinks occur are the same for @xmath22 $ ] and [ 001 ] ( i.e. , no anisotropy ) . in contrast , tenya et al . reported small but noticeable anisotropy in the location of the boundary of the high field ordered phase . @xcite a peak effect ( pe ) in the superconducting state that had an onset at @xmath32 1.25 t and disappeared at @xmath32 below @xmath39 was also observed ( fig . [ fig : peh111 ] ) . according to recent thermal conductivity measurements on a pros@xmath0sb@xmath1 single crystal in a magnetic field , @xcite there are two distinct superconducting phases in the @xmath40 plane with two - fold and four - fold symmetries . the phase boundary between the two - fold and four - fold symmetry superconducting phases is at for . the onset field of our pe occurs at a higher field and for the fields below the pe region no anomaly is observed in our @xmath19 curves . it is not clear whether the pe is related to the two - fold and four - fold symmetry superconducting phases @xcite or to pinning from crystal defects or impurities . recent transverse - field muon spin rotation measurements @xcite with @xmath41 indicated that pros@xmath0sb@xmath1 has a nearly isotropic superconducting energy gap in the superconducting state . evidence for a high field ordered phase ( hfop ) was derived from the kinks in magnetoresistivity @xmath23 data , @xcite kinks in magnetization @xmath27 data , @xcite peaks in specific heat @xmath42 data , @xcite and peaks in thermal expansion @xmath43 data . @xcite the @xmath28 vs @xmath16 isotherms reveal that @xmath28 is enhanced in the hfop , over a linear interpolation of @xmath28 from outside this region ( fig.[fig : rhot&h](c ) ) . from our new set of @xmath23 measurements ( fig.[fig : rhot&h ] ) , the upper boundary of the hfop can be determined . the @xmath40 phase diagram , depicting the superconducting and the hfop regions , is constructed in fig . [ fig : phasediagram ] . since an increase in the magnetic field would induce mixing of the ground state and the excited state , the hfop may be driven by the crossing of the upper level of the @xmath11 doublet and the lowest level of the @xmath12 triplet states at and another crossing between the lowest levels of the @xmath11 and @xmath12 states ( change of the ground state ) at . @xcite calculations of @xmath23 for the case in which a magnetic exchange interaction and an aspherical coulomb interaction @xcite between the zeeman split 4f and conduction electrons has been added to the cef hamiltonian with a @xmath11 ground state at @xmath44 t agree qualitatively with the experimental @xmath28 vs @xmath16 isotherms . @xcite the calculated @xmath28 vs @xmath16 isotherms are plotted in fig . [ fig : acsrhoh ] and compared with the experimental @xmath28 vs @xmath16 isotherms in fig . [ fig : rhot&h](c ) . in this calculation , we found that a @xmath14 ground state will not produce the same features we observe in @xmath45 measurements . this provides further support for a @xmath11 ground state in pros@xmath0sb@xmath1 . in fig . [ fig : phasediagram ] , the dashed line below and that connects the peaks in @xmath46 ( ref . 6 ) and intersects the hfop is a measure of the zeeman splitting between the pr@xmath10 ground state and the first excited state . the nearly temperature - independent boundary of the hfop at @xmath32 14.5 t resembles the antiferroquadrupolar ordered phase observed in prpb@xmath47 , @xcite which also has a @xmath11 ground state . it is plausible that the hfop in pros@xmath0sb@xmath1 has the same origin . however , the antiferroquadrupolar ordered phase observed in prpb@xmath47 has strong anisotropy at high fields . we did not observe such an anisotropy in the magnetization data of pros@xmath0sb@xmath1 along [ 001 ] and [ 111 ] below 5.5 t , although the anisotropy at these fields may not be large enough to resolve . neutron scattering experiments @xcite on our powdered samples of pros@xmath0sb@xmath1 did not detect any signs of an hfop . [ fig : hc1h001 ] shows the lower critical field @xmath5 vs @xmath33 determined for @xmath48 $ ] . @xmath49 is quite small , @xmath32 . the magnetic susceptibility @xmath50 in the meissner state was found to be ( or -1/[0.613(4@xmath51 ) ] ) . the demagnetization factor of the sample is estimated to be @xmath52 , @xcite resulting in an estimate for the superconducting volume fraction of . the ginzburg - landau parameter is estimated to be @xmath53 with the upper critical field @xmath54 and @xmath55 . this value is a factor of 10 larger than that estimated from the relation @xmath56 , where the penetration depth @xmath57 was taken from @xmath58sr measurements @xcite and the coherence length @xmath59 was estimated from the initial slope of @xmath60 . @xcite the discrepancy may arise from error in the estimates of @xmath5 due to the residual field in the superconducting magnet or to different qualities of samples used in the various measurements . in summary , we have reported electrical resistivity and magnetization measurements on pros@xmath0sb@xmath1 at high magnetic fields and determined the @xmath40 phase diagram below . aspherical coulomb and magnetic exchange scattering between 4f and conduction electrons can qualitatively describe the dome shape features observed in @xmath29 data and support the evidence that @xmath11 is the ground state at zero magnetic field . the high field ordered phase ( hfop ) boundary is determined from kinks in @xmath23 and @xmath27 data . the hfop is confined to a region on the @xmath40 plane between and , and below . measurements of @xmath27 for @xmath61 and @xmath62 [ 001 ] and [ 111 ] directions did not exhibit appreciable anisotropy . in analogy with the behavior of prpb@xmath47 , the hfop may be associated with antiferroquadrupolar order . the @xmath27 measurements revealed a small value for the lower critical field @xmath49 of @xmath32 oe , and a value for the upper critical field @xmath63 of @xmath32 , yielding a ginzburg - landau parameter @xmath64 31 . a peak effect was observed which had an onset at @xmath32 between and and disappeared at a field that tracked the @xmath39 curve which had an offset of @xmath32 . no evidence in the @xmath27 data was found for a cross - over between two superconducting phases with different order parameter symmetry as reported by izawa et al . this research was supported by the department of energy under grant no . defg-03 - 86er-45230 , the national science foundation under grant no . dmr-00 - 72125 , and the nedo international joint research program , and the work at the nhmfl pulsed field facility ( los alamos national laboratory ) was performed under the auspices of the nsf , the state of florida and the us department of energy . the pros@xmath0sb@xmath1 single crystal used in the magetization meausrements had an irregular bar shape with @xmath69 2.2 mm @xmath70 0.72 mm @xmath70 0.61 mm and @xmath71 . it was assumed ro be equivalent to an ellipsoid with @xmath72 .	the filled skutterudite compound pros@xmath0sb@xmath1 , the first example of a pr - based heavy fermion superconductor , displays superconductivity with @xmath2 k and has an effective mass @xmath3 , where @xmath4 is the free electron mass . for magnetic fields above 4.5 t , sharp features in the normal state electrical resistivity , magnetization , specific heat , and thermal expansion data suggest the occurrence of a phase transition at high fields . this high field ordered phase in the normal state may originate from a combination of crystalline electric field enhanced zeeman splitting and quadrupolar ordering . we present an investigation of the electrical resistivity and magnetization of pros@xmath0sb@xmath1 as a function of temperature between and and magnetic field up to . the data reveal a detailed phase boundary of the high field ordered phase as well as the lower critical field @xmath5 and the onset field of the peak effect in the superconducting state of pros@xmath0sb@xmath1 . pacs number : 74.70.tx , 65.40.-b , 71.27.+a , 75.30.mb keywords : pros@xmath0sb@xmath1 , magnetoresistance , magnetization , peak effect , @xmath5 , high field ordered phase
multifractals have been largely employed in the characterization of time series . this tool have been successfully applied in many different areas as economics @xcite , meteorology @xcite , geology @xcite , or biomedical @xcite . several algorithms have been used to find the multifractal spectrum of the time - series , for instance , wavelet analysis @xcite and dfa ( detrended fluctuation analysis ) @xcite . however , despite the large use of multifractals as a time - series analysis technique there is no simple geometrical examples of multifractal sets . some years ago it was introduced a multifractal partition of the unit square @xcite . this mathematical object was originally developed to model multifractal heterogeneity in oil reservoirs and to study percolation on complex lattices . afterwards , a generalization of this object to random partitions was performed to improve the model @xcite . in this work we explore the unidimensional version of this model . indeed , the multifractal partition of the square , cube or hypercube are , in essence , derived from a multifractal partition of a line segment . cantor sets , peano curves or sierpinski models are very useful subjects to grasp the fundamentals of fractals . on the other side , the study of multifractal sets lack simple geometrical examples . this paper intend to fulfill this vacancy in the literature . the work is organized as follows . in section @xmath8 we introduce the partition of a line segment that produces a multifractal set . we focus our attention on the case of a constant cutting ratio @xmath0 . the binomial distribution naturally arises in the construction of the partition algorithm . in section @xmath9 we analytically derive the multifractal spectrum @xmath5 of the partition and find its limit cases @xmath6 and @xmath7 . we use a method similar to the boxcounting algorithm . finally in section @xmath10 we give our final remarks . in this section we introduce a partition of a line segment . we expose the method of construction of the partition as a recursive algorithm . we call the sets at step @xmath2 as father sets respect to son - sets at step @xmath11 . each father set always give origin to two son sets . the sum of lengths of the son sets is equal to the length of the father set , that means , the dynamic rule of the algorithm is length preserving . at each step of the algorithm we use a ratio @xmath0 to perform the partition of the father sets . in the next paragraphs we will detail the initial steps of the algorithm , but the multifractal , that shall be shown in the next section , is properly defined in the limit of @xmath2 going to infinity . in figure [ fig1 ] we show a representation of the five initial steps in the construction of our partition . the step @xmath12 corresponds to the segment of length @xmath13 itself . at step @xmath14 the segment is partitioned into two pieces of length @xmath15 and @xmath16 . at step @xmath17 the two pieces are each one partitioned into two new sets giving origin to @xmath10 sets and so on . in table [ tab1 ] we show the iteration step @xmath2 , the number of sets for each step @xmath18 and the partition length ( as a fraction of @xmath13 ) inside each step . we notice that the length distribution is trivially done by a binomial rule , that means , the partition gives raise , at step @xmath2 , to a binomial length distribution : @xmath19 .the main quantities of the algorithm of the partition of a segment line according to a ratio @xmath0 . the iteration step @xmath2 , the total number of sets at each step @xmath20 , the full length partition as a fraction of @xmath13 . [ cols="^,^,^",options="header " , ] we remark some differences between the exposed process and the algorithm construction of cantor set and koch curve @xcite . initially the cited fractals do not conserve length along the dynamic formation , the measure divergence in these sets is originated in the algorithm process itself , the cantor set subtracts subsets and the koch adds subsets at each step . in addition , in these fractal models all sets , at the same step of their generating algorithm , have the same length . by that reason these models produce monofractal sets . we call @xmath1-sets the set of all elements , at the step @xmath2 , that has the same length @xmath21 . by construction there are @xmath4 elements in a @xmath1-set . in the next section , for the limit of @xmath22 we will determine the fractal dimension of each @xmath1-set and characterize the set of all @xmath1-set as a multifractal object . to find the length partition of a line segment in the case of variable @xmath0 we proceed as follows . consider that @xmath23 is present @xmath24 times at the partition and that at step @xmath2 we have @xmath25 . in this case the length distribution is done by : @xmath26 in figure [ fig1 ] we assume a fixed rule in the partition . each time a father set is cut into two new pieces the largest son set is always situated at the right position . we notice that this is an arbitrary choice that will not change the length distribution of the object . on the other side , the topology of the partition ( a subject that we do not consider in this paper ) will change since the neighborhood properties will change . we cite that the bidimensional version of the multifractal has a topological treatment @xcite that can not be easily extended to the unidimensional object . in fact , the topology of a line segment is sort of trivial since each open set in the segment has always two neighbours . in the cited work @xcite we report that the bidimensional multifractal version of the object we work in this paper has a power - law distribution of neighbours . in this section we estimate the fractal dimension for each @xmath1-set that results from the partition of a segment . we calculate the fractal dimension of each @xmath1-set , @xmath5 , with help of the definition : @xmath27 in this definition @xmath28 is the number of open balls , of size @xmath29 , necessary to cover a given @xmath1-set . in order to use ( [ dnn ] ) we generate a proper coverage of the segment @xmath13 . to create an adequate coverage of the partition we take at step one @xmath30 for @xmath31 integers . therefore @xmath0 is a rational number @xmath32 ( we take @xmath33 ) . in addition , at each step @xmath2 the size of the segment is @xmath34 and the size of the open balls used to cover the @xmath1-sets is : @xmath35 in this way , @xmath28 , the total length of each @xmath1-set is @xmath36 and as a result @xmath5 is done by : @xmath37 in figure [ fig2 ] we computationally test the finite size effect over the multifractal spectrum ( [ dk ] ) . we use as a case study @xmath38 and @xmath39 and as a consequence @xmath40 . in the simulations we take @xmath41 , @xmath42 and @xmath43 to follow the convergence of the spectrum . in this figure we plot @xmath44 in the horizontal axis to compare the results of several @xmath2 . we observe in the data that in the limit of large @xmath2 the spectrum seems to touch the line @xmath45 which suggests that the @xmath1-set of largest dimension is dense in the line . it is interesting to estimate @xmath5 in the limits of @xmath1 going to @xmath46 and @xmath2 . using equation ( [ dk ] ) we calculate @xmath5 for @xmath47 : @xmath48 summarizing , we have : @xmath49 in a similar way we estimate : @xmath50 we use the notation @xmath51 instead of @xmath52 because he multifractal is only defined in the limit of @xmath53 , and therefore @xmath52 is in fact an accumulation point . it is worth to note that in the limit of @xmath51 or @xmath54 the fractal dimension is not necessarily @xmath46 . one could think that @xmath55 for @xmath52 because the multifractal of the corresponding @xmath1-set for @xmath52 is composed by just one point and the dimension of a single point is zero . this argument is not valid because the multifractal is only properly defined in the limit of @xmath22 , where the variable @xmath1 is not discrete , but continuous . we show the spectrum of fractal dimension for several ratios @xmath0 in figure [ fig3 ] . we display in this plot the spectrum @xmath5 versus @xmath44 for some @xmath56 and @xmath57 as indicated in the figure . in this estimation we use @xmath58 . the relations ( [ eq5 ] ) ( [ eq6 ] ) are verified in the curves of the figure . we find the maximum of @xmath5 for any @xmath0 using the property of the maximum of the binomial distribution @xcite . in this way the maximum of @xmath5 is at @xmath59 or @xmath60 . to compare this result with the maximum of curves in figure [ fig3 ] we use the normalization @xmath44 and the maximum is at @xmath0 . we check in the figure , for instance , that the maxima for the cases @xmath38 and @xmath61 and @xmath62 are at positions @xmath63 and @xmath64 respectively . these values are indicated by successive arrows in the plot . the degenerate case @xmath65 corresponds to the standard partition of a segment which generates the unidimensional lattice . in this limit all elements of the partition have the same size . following this rule we cover the line segment with @xmath3 equal elements of length @xmath66 and the dimension of the object is not a fractal , indeed @xmath67 in this case there is just one @xmath1-set that trivially have the same dimension of the line segment itself . in this work we introduce an object that arises from an infinite partition of a line segment . we present the construction algorithm of the object as a recursive sequence . in the case of a constant partition ratio @xmath0 the length distribution of its elements , at step @xmath2 of the algorithm , follows a binomial distribution . at step @xmath2 , the partition is formed by @xmath3 sets grouped in @xmath68 @xmath1-sets , that means , sets whose elements share the same length @xmath69 . in the limit of @xmath53 each @xmath1-set is a monofractal set . the multifractal set is composed by the totality of infinite monofractal sets . in addition we find the fractal spectrum @xmath5 and estimate where is its maximum . finally we find the values of @xmath5 for the limits @xmath6 and @xmath7 . in this work we adapted to a line segment an algorithm of generation of multifractal objects that was initially defined on a square . the route from two dimensions to one dimension is a challenge . at one side we loose the attractive idea of modeling bidimensional geological formations . otherwise , once we go to an unidimensional version of the problem some points became more clear : the generalization of this object to any dimension , its formation algorithm and the spectrum of fractal dimensions . we think we have also attained our aims by showing to a general reader an illustrative ans simple geometrical multifractal set . however , some questions about this multifractal set remain open . is the largest set of the multifractal dense in the line ? the simulations indicate a positive answer , but we lack a rigorous proof of that . following equation ( [ multidist ] ) , is there a simple spectrum of dimensions for the case of non constant partition ? the answer seems positive if we could define appropriate @xmath1-sets . a more general question would be : if the set of ratios @xmath23 in equation ( [ multidist ] ) follows a normal distribution , what will be the resulting multifractal ? we conjecture that this new object would be usefull as a benchmark to compare with empirical multifractals .	we report an algorithm for the partition of a line segment according to a given ratio @xmath0 . at each step the length distribution among sets of the partition follows a binomial distribution . we call @xmath1-set to the set of elements with the same length at the step @xmath2 . the total number of elements is @xmath3 and the number of elements in a same @xmath1-set is @xmath4 . in the limit of an infinite partion this object become a multifractal where each @xmath1-set originate a fractal . we find the fractal spectrum @xmath5 and calculate where is its maximum . finally we find the values of @xmath5 for the limits @xmath6 and @xmath7 . keywords : multifractal , binomial distribution , partition of a segment , spectrum of fractal dimensions .
although far from complete , the overall picture describing the angular momentum evolution of solar type stars is well established . despite modest mass - loss rates , @xmath3 for the sun , magnetised stellar winds strongly brake the rotation of the star . the decrease of the angular velocity then reacts back on the magnetic field by reducing the efficiency of the dynamo process . this picture can not be applied as such to early - type stars . first of all , the history of stellar rotation at intermediate and high masses is generally still poorly known ; besides , the presence of magnetic fields in early - type stars has not been established , except for some categories of chemically peculiar stars , as well as for a couple of particular cases ( donati et al . 1997 ; henrichs et al . 2000 ) . on the other hand , we know that early type stars can experience strong angular momentum losses as stellar winds with very large mass loss rate have been observed ( @xmath4 ) . in this paper , we investigate how such angular momentum losses will affect the stellar rotation , assuming magnetic fields are not dynamically relevant . this question has not yet received much attention although it might have important consequences for stellar structure and evolution as well as for the understanding of early - type stars activity . in the context of stellar evolution , the effect of mass - loss on rotation has to be investigated since the rotation strongly influences the stellar structure . however , current models of stellar evolution with rotation do not take this effect into account ( talon et al . 1997 , denissenkov et al . 1999 ) . as explained below , we have been confronted to the same question while investigating the origin of the very strong activity of herbig ae / be stars , a class of pre - main - sequence stars with masses ranging from @xmath5 to @xmath6 . a significant fraction of these objects is known to possess extended chromospheres , winds , and to show high levels of spectral variability . in addition , the presence of magnetic fields , first suggested by the rotational modulation of certain spectral lines ( catala et al . 1986 ) , has been recently supported by a direct detection at the surface of the herbig ae star , hd 104237 ( donati et al . detailed estimates of the non - radiative heating in the outer atmosphere of herbig ae / be stars ( catala 1989 , bouret et al . 1998 ) compared to observational constraints on the available energy sources strongly suggest that the rotation of the star is the only energy source capable of powering such activity ( see discussion in f.lignires et al . this led vigneron et al . ( 1990 ) to propose a scenario whereby the braking torque exerted by the stellar wind forces turbulent motions in a differentially rotating layer below the stellar surface . then , by invoking an analogy with stellar convection zones , these turbulent motions could generate a magnetic field which would transfer and dissipate the turbulent kinetic energy into the outer layers of the star . unlike herbig stars , there is at present no observational evidence of non - radiative energy input in the atmosphere of ob stars . their radiatively - driven winds do not require any additional acceleration mechanism and an eventual non - radiative heating would be very difficult to detect because most lines are saturated . however , they show various forms of spectral variability . recent observations seem to indicate that these phenomena are not due to an intrinsic variability of the wind but are instead caused by co - rotating features on the stellar surface ( massa et al . 1995 ) . as proposed in the literature ( howarth et al . 1995 ; kaper et al . 1996 ) , these corotating features could be due to a magnetic structuration of the wind . since ob stars rotate fast and possess strong winds , the vigneron et al . scenario might also explain these phenomena . the recent direct detection of a magnetic field in an early be star , @xmath7 cep , adds some credit to this hypothesis ( henrichs et al . 2000 ) . in this paper , we shall investigate the starting assumption of the vigneron et al . scenario namely that the braking torque of an unmagnetised wind generates strong enough velocity shear to trigger an instability in the subphotospheric layers . this is a crucial step in the scenario since the onset of the instability allows to transfer kinetic energy from rotational motions to turbulent motions . the possible connection between mass - loss and instability had been already suggested by schatzman ( 1981 ) . however , the onset of the instability is not really considered in this study since it is assumed from the start that the wind driven angular momentum losses induce a turbulent flux of angular momentum . here , we propose a simple model of angular momentum transport by a purely radial mass flux in order to estimate the angular velocity gradients induced by mass loss . then , the stability of these gradients is studied according to existing stability criteria . note that , as we neglect latitudinal flows , our model is best regarded as an equatorial model . in the absence of magnetic fields , the braking effect of a stellar wind is simply due to the fact that matter going away from the rotation axis has to slow down to preserve its angular momentum . thus , for the star s surface to be significantly braked , fluid parcels coming from deep layers inside the star have to reach the surface . since radial velocities induced by mass - loss are very small deep inside the star , this is expected to take a relatively long time , not very different from the mass - loss time scale . by contrast , we shall see that the formation of unstable angular velocity gradients near the surface take place in a very short time , much smaller than the braking time scale . the paper is organised as follows : first , we estimate the time scale characterising the braking of the stellar surface and relate it to the mass loss time scale ( sect . , we show that radial outflows in stellar envelopes generate differential rotation and we estimate the time scale of this process ( sect . the stability of these angular velocity gradients is considered ( sect . 4 ) and the results are summarised and discussed ( sect . 5 ) . generally speaking , the braking of the star s surface depends on the mass - loss mechanism and on the efficiency of angular momentum transfer inside the star . in this section , we shall make assumptions regarding these processes in order to estimate the braking time scale . however , before we consider these particular assumptions , it is interesting to show that the simple fact that the star adjusts its hydrostatic structure to its decreasing mass already implies that the star slows down as it loses mass . indeed , according to models of pre - main - sequence evolution ( palla & stahler 1993 ) , a @xmath8 pre - main - sequence star losing mass at a rate of the order of @xmath9 will have lost about one percent of its mass when it arrives on the main sequence . because mass is concentrated in the core of stars , one percent of the total mass corresponds to a significant fraction of the envelope . for pre - main - sequence models of @xmath5 to @xmath10 , this means that all the matter initially located above the radius @xmath11 is expelled during the pre - main - sequence evolution . but , during this mass - loss process , the star continuously adjusts its structure to its decreasing mass and , from the point of view of stellar structure , the loss of 1 percent of mass has a negligible effect on the stellar radius . thus , as represented on fig.1 , the sphere containing 99 % of the initial mass must have expanded significantly during its pre - main - sequence evolution . its moment of inertia has increased and , due to angular momentum conservation , its mean angular velocity has been reduced . thus , we conclude that the hydrodynamic adjustment of stars ensures that mass loss is accompanied by a mean braking of the remaining matter . now , if one wants to estimate the actual braking of the stellar surface , assumptions have to be made on the mass - loss process and on the efficiency of angular momentum transfer inside the star . in order to obtain the order of magnitude of the braking time scale , we assume that mass - loss is isotropic and constant in time and we consider two extreme assumptions regarding the efficiency of angular momentum transfer . first , we assume that the transfer of angular momentum is only due to radial expansion . then , angular momentum conservation states that the angular velocity @xmath12 of a fluid parcel located at a radius @xmath13 will have decreased by @xmath14 when it reaches the stellar surface . this occurs when all the matter above @xmath13 has been expelled so that the decrease of the surface angular velocity can be related to the mass - loss . we used the density structure of a @xmath8 pre - main - sequence model ( palla & stahler 1993 ) for the following calculation . second , we assume that an unspecified transfer mechanism enforces solid body rotation throughout the star so that angular momentum losses are distributed over the whole star and the braking of the stellar surface is less effective . in this case , global angular momentum conservation reads @xmath15 where @xmath16 is the total angular momentum . according to stellar structure models of @xmath5 to @xmath10 pre - main - sequence stars , the radius of gyration , @xmath17 is close to @xmath18 so that the angular velocity decrease is approximatively given by @xmath19 to simply relate the braking to the mass - loss , we also assumed that the decrease of the moment of inertia is proportional to the mass decrease , or equivalently that mass - loss induces an uniform density decrease throughout the star . albeit rough , this assumption is not critical for the conclusion drawn in this section . starting with an uniformly rotating star , fig . 2 presents the braking of the stellar surface as a function of the percentage of mass lost assuming solid body rotation ( solid line ) and radial expansion ( dashed line ) . this shows that , in the case of solid body rotation , the braking time scale is about ten times smaller than the mass loss time scale , whereas , it is hundred times smaller in the case of radial expansion ( the braking time scale is defined as the time required to decrease the surface angular velocity by a factor @xmath20 ) . we therefore conclude that the braking time scale is significantly smaller than the mass - loss time scale although both time scales does not differ by many orders of magnitude . in the next section , we study the formation of angular velocity gradients in a stellar envelope assuming the transport of angular momentum is only due to a radial mass flux . we shall see that this process generates strong near surface gradients in a time scale smaller by many orders of magnitude than the mass - loss time scale . non - uniform radial expansion tends to generate differential rotation and this is particularly true across the stellar surface where the steep density increase has to be accompanied by a steep decrease of the radial velocity . to show this let us follow the expansion of two spherical layers located at different depths in the star s envelope , and rotating at the same rate . after a given time , the outer layer will have travelled a much larger distance than the inner one so that its angular velocity will have decreased more than that of the inner layer . a gradient of angular velocity has then appeared between both layers . to estimate the gradient generated by this non - uniform expansion , we write down the angular momentum conservation assuming the transport of angular momentum is only due to a radial flow . consequently , angular momentum transfers by viscous stresses , meridional circulation , gravity waves or turbulence are neglected all together . although latitudinal inhomogeneities in the mass - loss mechanism or eddington - sweet circulation are expected to induce latitudinal flows , we note that the present assumption would be still justified in the equatorial plane if the meridional circulation is symmetric with respect to the equatorial plane . in the context of our simplified model , the angular momentum balance reads @xmath21 where @xmath22 is the specific angular momentum and @xmath23 is the radial velocity . up to mass loss rates of the order of @xmath24 , radial velocities carrying out a time - independent and isotropic mass - loss are much smaller than the local sound speed . consequently the star is always very close to hydrostatic equilibrium and continuously adjusts its structure to its decreasing mass . this means in particular that the temporal variations of the density near the surface are small , because the effect of mass - loss is distributed over the whole star through hydrostatic equilibrium . accordingly , temporal variations of the radial velocities are small and we shall neglect them as their inclusion would not modify our conclusions . then , the radial outflows satisfies @xmath25 where @xmath26 is given by stellar structure models . with the above assumptions , the specific angular momentum evolves like a passive scalar advected in a one dimensional stationary flow @xmath27 . the mathematical problem can be readily solved and we will do so in the following for a radial flow corresponding to a @xmath8 star with a mass loss rate of @xmath9 . but first we derive useful properties by studying the evolution of the angular momentum gradient in the general case . except when @xmath22 or @xmath28 are uniform in space , advection always modifies the distribution of the conserved quantity . while decelerated flows tend to sharpen @xmath22-gradients , accelerated flows smooth them out . to illustrate the process of gradient smoothing in an accelerated flow , the angular momentum evolution of two neighbouring fluid elements @xmath29 and @xmath30 has been represented in fig.3 . following the motions , the angular momentum gradient between these two points decreases because their separation @xmath31 increases while the angular momentum difference @xmath32 is conserved . this simple sketch can also be used to quantify the gradient decrease . we first note that @xmath29 and @xmath30 travel the distance separating @xmath33 from @xmath34 in the same time . thus , the time interval required by a to go from @xmath35 to @xmath33 is the same as the one used by @xmath30 to go from @xmath34 to @xmath36 . this is expressed by , @xmath37 , and then , @xmath38 taking the limit of vanishing separation between @xmath29 and @xmath30 , we conclude that the product @xmath39 is conserved following the motions , a property which can be readily verified by calculating the lagrangian derivative of @xmath40 . this property implies that angular momentum gradients can be completely smoothed out if they are advected in a strongly enough accelerated flow . this is particularly relevant near stellar surfaces where steep density gradients induce steep radial velocity gradients . to specify this effect we express the conservation of @xmath41 for the radial outflow given by eq . ( [ eq : mass ] ) . then , the evolution of the logarithmic gradients of the angular velocity following the flow between radii @xmath35 and @xmath34 reads @xmath42 where the quantity @xmath43 measures departures from uniform specific angular momentum . in this expression , @xmath44 decreases almost like @xmath26 in the vicinity of the photosphere since the density decreases considerably over distances small compared to the stellar radius . then , the above expression shows that any departure from constant specific angular momentum will have decreased by a factor @xmath45 after crossing @xmath46 density scale heights . moreover , initial departures can not be too large otherwise the corresponding differential rotation would be subjected to powerful instabilities . then , for any realistic values of the initial angular velocity gradients , fluid elements reaching the surface will have an angular velocity gradient close to @xmath47 . we confirmed the validity of this simple picture by solving the advection problem for the density profile corresponding to the stellar structure model of a pre - main - sequence @xmath8 star ( palla & stahler 1993 ) . again , the mass - loss rate has been fixed to @xmath48 . starting from a solid body rotation , the evolution of the rotation profile in the vicinity of the stellar surface is presented in fig.4 . the different curves correspond to increasing advection times , @xmath49 , @xmath50 , @xmath51 , @xmath52 , @xmath53 and @xmath54 years , respectively . we observe that the evolution is first rapid and then much slower . this is because the evolution starts with the rapid expansion of the outer layers which tends to set - up a profile of uniform specific angular momentum @xmath55 . once this is done , the evolution takes place on much larger time scales as it involves the much slowly expanding inner layers . the gradual set - up of the @xmath56 profile is confirmed by the evolution of the angular velocity logarithmic gradients . as shown on fig.5 , specific angular momentum gradients are rapidly smoothed out in the outer layers while the inner ones become affected after a long time . we can therefore conclude that starting with any realistic angular velocity profile in a stellar envelope , radial expansion will generate angular velocity gradients very close to @xmath57 near the surface . whether such gradients are actually present in the subphotospheric layers of mass - losing stars depends on how the characteristic time for the formation of these gradients compares with the time scales of other angular momentum transport processes . the angular velocity gradients being produced by radial acceleration , the associated time scale is @xmath58 where @xmath59 is the density scale height and @xmath60 is the mass - loss time scale . fig.5 shows that no more than one month is necessary to form gradients of the order of @xmath61 for a mass loss rate equal to @xmath9 . this time scale is smaller by many orders of magnitude than the mass - loss time scale or the braking time scale . this is because the formation of gradients requires radial displacements corresponding to a density scale height whereas a significant braking requires radial displacements of the order of the stellar radius . in this section , we investigate whether the differential rotation induced by the radial mass - loss is sufficient to trigger a hydrodynamical instability . the uniform angular momentum profile being marginally stable with respect to rayleigh - taylor instability , we consider its stability with respect to shear instabilities . when , as it is the case here , the vorticity associated with the velocity profile does not possess extrema , the study of shear instabilities is much complicated by the fact that finite amplitude perturbations involving non - linear effects have to be taken into account . existing stability criteria for such velocity profiles rely on laboratory experiments which shows critical reynolds numbers above which destabilization occurs . although the critical reynolds number depends on the particular flow configuration , its value is generally of order of @xmath62 . j.p . zahn ( 1974 ) proposed the following instability criterion @xmath63 where @xmath64 is the critical reynolds number , @xmath65 is the prandtl number comparing the thermal diffusivity @xmath66 to the viscosity @xmath67 and @xmath68 is the brunt - visl frequency which measures the strength of stable stratification in radiative interiors . the combined effect of the stable stratification and the thermal diffusion has been derived on phenomenological grounds . however , the way the criterion depends on this effect has been recently supported in the context of the linear stability theory ( lignires et al . 1999 ) . applying the criterion to an uniform angular momentum profile , we find that instability occurs if the rotation period at the surface is smaller than @xmath69 where the actual temperature gradient has been taken equal to @xmath70 , a typical value for radiative envelopes according to cox ( 1968 ) , and the radiative viscosity has been assumed to dominate the molecular one which is true in the envelope of intermediate mass stars . note also that this expression holds for a mono - atomic completely ionised gas in a chemically homogeneous star . when radiation pressure is taken into account , the upper limit of the rotation period has to be multiplied by @xmath71 , where @xmath7 is the ratio between the gas pressure and the total pressure . this factor remains very close to one for pre - main - sequence models from @xmath5 to @xmath10 . the above instability condition being easily met by early type stars , we conclude that mass - loss tends to impose an unstable differential rotation below the surface of early - type stars . we have studied the effect of mass - loss on the rotation of stellar envelopes in the simplified context of an unmagnetised spherically symmetric outflow . in the same way as an inhomogeneous distribution of a passive scalar tends to be smoothed out in a expanding flows , the radial gradients of specific angular momentum are smoothed out by radial expansion in stellar envelopes . this process becomes more and more effective as one approaches the surface because expansion becomes stronger and stronger . as shown in figs 4 and 5 , a profile of uniform angular momentum , @xmath56 is rapidly set - up in the outermost layers of the star and then pervades towards the interior on much larger time scales . these time scales being proportional to the mass - loss time scale , they vary very much from stars to stars . for a typical herbig star ( @xmath9 ) , an angular velocity gradient close to @xmath72 appears in one month . by contrast , for a solar - type mass - loss rate , @xmath73 , it would be @xmath74 times longer to reach the same level of differential rotation . according to existing stability criteria , the uniform angular momentum profile is subjected to shear instabilities provided the rotation period is shorter than @xmath75 days . the present one - dimensional model is admittedly not realistic at least because latitudinal variations occur in rotating stars and give rise to a meridional circulation . nevertheless , as already mentioned , neglecting latitudinal flows may be justified in the equatorial plane . in the following we discuss to which extent the neglected processes can prevent the formation of turbulent differentially rotating layers below the surface of rapidly rotating mass - losing stars . we first consider the angular momentum transport and then the stability problem . latitudinal variations could be inherent to the mass - loss mechanism as proposed for radiatively driven wind emitted from fast rotating stars ( owocki & gayley 1997 ) . such variations are likely to generate a meridional circulation ( maeder 1999 ) but , at the latitudes where outflows occur , we still expect radial expansion to play a major role in shaping the angular velocity profile . the eddington - sweet circulation driven by departure from sphericity operates on time scale larger than the kelvin - helmholtz time scale of the star . this time scale is of the order of the order of @xmath76 years for a @xmath8 herbig stars . consequently the eddington - sweet circulation is unlikely to prevent the formation of the differentially rotating subphotospheric layer for strong mass - loss rate . turbulent motions like those generated in thermal convection zones might prevent the formation of the differentially rotating layer . in radiative envelopes however , it is not clear whether turbulent motions ( not generated by the mass - loss process ) would be vigorous enough . strong magnetic fields could in principle prevent the formation of these gradients . note that this objection is not relevant in the context of the vigneron et al . scenario where the shear layer is first produced by an unmagnetised wind . finally , we also neglected the effect of pre - main - sequence contraction . this is justified because the negative radial velocities associated with the contraction are much smaller than the positive radial velocities induced by mass - loss in the vicinity of the stellar surface . in what concerns the stability of the uniform angular momentum profile , one has to remind that the stability criterion results from an extrapolation of laboratory experiments results . despite recent numerical simulations which contradict the validity of this extrapolation for a differential rotation stable with respect to the rayleigh - taylor instability ( balbus et al . 1996 ) , analysis of the experimental results reveal that for large values of the reynolds number the properties of the turbulent flow do not depend on its stability or instability with respect to the rayleigh - taylor criterion ( richard & zahn 1999 ) . the solution of this debate would have to wait numerical simulations with higher resolution or specifically designed laboratory experiments . the fact that the shear layer is embedded in an expanding flow may also affect the stability because expansion is known to suppress the turbulence . for example , this phenomenon is observed in numerical simulations of turbulent convection near the solar surface ( stein & nordlund 1998 ) . however , the growth time scale of the instability being of the order of the rotation period ( @xmath77 day for herbig stars ) , we do not expect the expansion to be rapid enough to suppress shear turbulence . we conclude that the present model of angular momentum advection by a radial flow supports the assumption made by vigneron et al . ( 1990 ) and f. lignires et al . ( 1996 ) that unmagnetised winds tend to force an unstable differential rotation in the subphotospheric layers of herbig ae / be stars . note that this work is part of an ongoing effort to assess the viability of the vigneron et al . f. lignires et al ( 1996 ) and f.lignires et al . ( 1998 ) have already studied the step further , once turbulent motions are generated at the top of the radiative envelope , their work pointing towards the formation and inwards expansion of a differentially rotating turbulent layer . in addition , three - dimensional numerical simulations are being performed to investigate whether magnetic fields can be generated in such a layer . bouret j .- c . , catala c. , 1998 , a&a 340 , 163 balbus s.a . , hawley j.f . , stone j.m . , 1996 , apj 467 , 76 catala c. , felenbok p. , czarny j. , talavera a. , boesgaard a.m. , 1986 , apj 308 , 791 catala , c. , 1989 , in : delache , s. laloe , c. magnan , j. , et al . @xmath78 iap meeting , modeling the stellar environment : how and why ? paris , p. 207 cox j.p . , 1968 , in : gordon and breach science publishers ( eds . ) , principles of stellar structure , p. 587 denissenkov p.a . , ivanova n.s . , weiss a. , 1999 , a&a 341 , 181 donati j .- f . , semel m. , carter b.d . , rees d.e . , cameron a.c . , 1997 , mnras 291 , 658 henrichs h.f . , de jong j.a . , donati j.f . , catala c. , shorlin s.l.s . , wade g.a . , veen p.m. , nichols j.s . , verdugo e. , talavera a. , kaper l. , 2000 , a&a in press kaper l. , henrichs h.f . , nichols j.s . , snoek l.c . , volten h. , zwarthoed g.a.a . , 1996 , a&as 116 , 257 lignires f. , catala c. , mangeney a. , 1996 , a&a 314 , 465 lignires f. , califano f. , mangeney a. , 1998 , geophys . astrophys . fluid dyn . 88 , 81 lignires f. , 1999 , a&a 349 , 1027 maeder a. , 1999 , a&a 347 , 185 massa d. , fullerton a.w . , nichols j.s . , owocki s.p . , prinja r.k . , st - louis n. , willis a.j . , altner b. , bolton c.t . , cassinelli j.p . , cohen d. , cooper r.g . , feldmeier a. , gayley k.g . , harries t. , heap s.r . , henriksen r.n . , howarth i.d . , hubeny i. , kambe e. , kaper l. , koenigsberger g. , marchenko s. , mccandliss s.r . , moffat a.f.j . , nugis t. , puls j. , robert c. , schulte - ladbeck r.e . , smith l.j . , smith m.a . , waldron w.l . , white r. l. , 1995 , apjl 452 , 53 owocki s.p . , gayley k.g . , 1997 , in : nota a. , amers h. ( eds . ) luminous blue variables : massive stars in transition . p. 121 palla f. , stahler s.w . , 1993 , apj 418 , 414 richard d. , zahn j .- p . , 1999 , a&a 347 , 734 schatzman e. , 1981 , in : g.e.v.o.n . and sterken c. ( eds . ) proceedings of the workshop on pulsating b stars ( nice , 1 - 5 june 1981 ) , p. 347 stein r.f . , nordlund a. , 1998 , apj 499 , 914 talon s. , zahn j .- p . , maeder a. , meynet g. , 1997 , a&a 322 , 209 vigneron , c. , mangeney , a. , catala , c. , schatzman , e. , 1990 , sol . 128 , 287 zahn j .- p . , 1974 , in : ledoux et al . ( eds . ) , stellar instability and evolution , p. 185	to explain the origin of herbig ae / be stars activity , it has been recently proposed that strong mass - losses trigger rotational instabilities in the envelope of fast rotating stars . the kinetic energy transferred to turbulent motions would then be the energy source of the active phenomena observed in the outer atmosphere of herbig ae / be stars ( vigneron et al . 1990 ; lignires et al . 1996 ) . in this paper , we present a one - dimensional model of angular momentum transport which allows to estimate the degree of differential rotation induced by mass - loss . gradients of angular velocity are very close to @xmath0 ( @xmath1 being the surface rotation rate and @xmath2 the stellar radius ) . for strong mass - loss , this process occurs in a short time scale as compared to other processes of angular momentum transport . application of existing stability criteria indicates that rotational instabilities should develop for fast rotating star . thus , in fast rotating stars with strong winds , shear instabilities are expected to develop and to generate subphotospheric turbulent motions . albeit very simple , this model gives strong support to the assumption made by vigneron et al . 1990 and lignires et al . 1996 .
t coronae borealis ( t crb ) is one of the well observed recurrent novae and characterized by a secondary , fainter maximum occurring @xmath6 days after the primary peak . historically , t crb bursted twice , in 1866 and 1946 , with the light curves very similar each other ( e.g. , @xcite ) . large ellipsoidal variations in the optical light curves during quiescent phase suggest that an m3@xmath74 red - giant component fills its roche lobe ( e.g. , @xcite ; @xcite ; @xcite ) . there have been debates on the nature of the hot component of this binary system . webbink ( 1976 ) and webbink et al . ( 1987 ) proposed an outburst mechanism of t crb based on their main - sequence accretor model . this accretion event model was investigated further by 3d numerical simulations ( @xcite ; @xcite ) . however , selvelli , cassatella , & gilmozzi ( 1992 ) have been opposed to the main - sequence accretor model from their analysis of _ iue _ data in its quiescent state , which indicates the existence of a mass - accreting white dwarf ( wd ) . the detection of x - rays and the presence of flickering in the optical light curves are also naturally explained in terms of accretion onto a wd . their estimated mass accretion rate in quiescent state is very high ( @xmath8 yr@xmath9 ) and is exactly required by the thermonuclear runaway ( tnr ) theory to produce a tnr event every 80 yr on a massive ( @xmath10 ) wd . thus , the 1866 and 1946 outbursts can be interpreted in terms of a tnr event on a very massive wd . rapid decline rates of the light curves indicate a very massive wd close to the chandrasekhar limit , @xmath11 ( @xcite , 1999 ) . assuming solar composition of the wd envelope , kato calculated nova light curves for wd masses of 1.2 , 1.3 , 1.35 and 1.377 @xmath12 and found that the light curve of the @xmath13 model is in better agreement with the observational light curve of t crb than the other lower mass models . recently , other observational supports for a massive wd in t crb have been reported . belczyski and mikolajewska ( 1998 ) derived a permitted range of binary parameters , @xmath14 , from amplitude of the ellipsoidal variability and constraints from the orbital solution of m - giants . in shahbaz et al . ( 1997 ) , a massive wd of @xmath15 is suggested from the infrared light curve fitting . combining these two permitted ranges of the wd mass in t crb , we may conclude that a mass of the wd is between @xmath16 , which is very consistent with the light curve analysis @xmath17 by kato ( 1999 ) . the secondary maximum in outbursts is not generally observed in classical novae or in other recurrent novae . selvelli et al . ( 1992 ) suggested a possibility of irradiation by a stationary shell around the system , although the presence of the shell is just a speculation . in this letter , we propose another possibility of the second peak : irradiation by a tilting accretion disk around a massive wd . the main results of our analysis are : 1 ) the first peak is naturally reproduced by the fast developing photosphere of the wd envelope based on the tnr model incorporated with a very massive wd ( @xmath18 ) . 2 ) the second peak is not fully reproduced by an irradiated m - giant model as simply estimated by webbink et al . ( 1987 ) ; 3 ) the second peak can be well reproduced if we introduce an irradiated tilting accretion disk around the wd in addition to the irradiated m - giant companion . such tilting instabilities of an accretion disk have been suggested by pringle ( 1996 ) for central stars as luminous as the eddington limit or more ( radiation - induced instability ) . because the maximum luminosity of t crb outbursts exceeds the eddington limit ( e.g. , @xcite ) , the radiation - induced instability may work well in t crb outbursts . our model is graphically shown in figure [ vmaxfig ] . the visual light is contributed to by three components of the system : 1 ) the wd photosphere , 2 ) the m - giant photosphere , and 3 ) the accretion disk surface . in the tnr model , wd envelopes expand greatly as large as @xmath19 or more and then the photospheric radius gradually shrinks to the original size of the white dwarfs ( e.g. , @xmath20 for @xmath21 ) . the optical luminosity reaches its maximum at the maximum expansion of the photosphere and then gradually darkens to the level in quiescent phase . since the wd envelopes reach a steady - state in the decay phase of novae ( e.g. , @xcite ) , we are able to treat the development of the envelope by a unique sequence of steady - state solutions with different envelope masses ( @xmath22 ) as shown by kato & hachisu ( 1994 ) . we have calculated such sequences for wds with various masses of @xmath23 , 0.7 , 0.8 , 0.9 , 1.0 , 1.1 , 1.2 , 1.3 , 1.35 and @xmath13 and obtained the optical light curves for the decay phase of tnr events . here , we choose @xmath13 as a limiting mass just below the mass at the sn ia explosion in w7 ( @xmath24 , @xcite ) . we have used the updated opal opacity ( @xcite ) , which has a strong peak near @xmath25 about @xmath26% larger than that of the original opal opacity ( @xcite ) which was used in kato & hachisu ( 1994 ) . the numerical method and various assumptions are the same as those in kato & hachisu ( 1994 ) . it should be noted here that optically thick winds blow when the wd envelope expands and the photospheric temperature decreases below @xmath27 . each wind solution is a unique function of the envelope mass @xmath22 if the wd mass is given . the envelope mass is decreasing due to the wind mass loss at a rate of @xmath28 and hydrogen shell burning at a rate of @xmath29 , i.e. , @xmath30 where @xmath31 is the mass accretion rate to the wd . integrating equation ( [ dmdt_envelope_mass ] ) , we follow the development of the envelope mass @xmath22 and obtain physical quantities such as the photospheric temperature @xmath32 , photospheric radius @xmath33 , photospheric velocity @xmath34 , wind mass loss rate @xmath35 , and nuclear burning rate @xmath36 . when the envelope mass decreases to below the critical mass , the wind stops and after that the envelope mass is decreased only by nuclear burning . we have assumed a black - body photosphere of the white dwarf envelope . after the optical peak , the photosphere shrinks with the envelope mass being blown off in the wind , and the photospheric temperature increases with the visual light decrease because the main emitting region moves blueward . based on our wind solutions , we have obtained visual magnitude of the wd photosphere with a window function given by allen ( 1973 ) . the photospheric surface is divided into 16 pieces in the latitudinal angle ( @xmath37 ) and into 32 pieces in the longitudinal angle ( @xmath38 ) as shown in figure [ vmaxfig ] . then , the contribution of each piece is summed up by considering the inclination angle to the viewer . a linear limb - darkening law ( the coefficient is @xmath39 , see @xcite ; @xcite ) is incorporated into the calculation . to construct a light curve , we have also included the contribution of the companion star irradiated by the wd photosphere . the surface of the companion star is assumed to fill the inner critical roche lobe as shown in figure [ vmaxfig ] . numerically dividing the latitudinal angle into 32 pieces ( @xmath40 ) and the longitudinal angle into 64 pieces ( @xmath41 ) , we have also summed up the contribution of each area considering the inclination angle to the viewer by assuming the same limb - darkening law as the wd photosphere , but we neglect the gravity - darkening effect of the companion star because the hemisphere to the wd is heated up by the irradiation . here , we assume that 50% of the absorbed energy is reemitted from the hemisphere of the companion with a black - body spectrum at a local temperature . the original ( non - irradiated ) photospheric temperature of the companion star is assumed to be @xmath42 k ( @xcite ) . if the accretion disk around the wd blocks the light from the wd photosphere , it makes a shadow on the surface of the companion star . such an effect is also included in our calculation . the orbit of the companion star is assumed to be circular . the light curves are calculated for five cases of the companion mass , i.e. , @xmath43 , 0.7 , 0.8 , 0.9 , and @xmath44 , as derived by belczyski & mikolajewska ( 1998 ) . since we obtain similar light curves for all of these five masses , we show here only the results for @xmath45 . in this case , the separation is @xmath46 , the effective radius of the inner critical roche lobe for the wd component is @xmath47 , the effective radius of the m - giant companion star is @xmath48 . we have included the luminosity coming from the accretion disk irradiated by the wd photosphere when the accretion disk reappears a few to several days after the optical maximum . the surface of the accretion disk absorbs photons and reemits in the same way as the companion does . here , we assume that the accretion disk surface emits photons as a black - body at a local temperature of the heated surface . we assume further that 25% of the absorbed energy is emitted from the surface , while the other is carried into interior of the accretion disk and eventually brought into the wd . the temperature of the outer edge is assumed to be @xmath49 k , which is never irradiated by the wd photosphere . an axisymmetric accretion disk with a thickness given by @xmath50 is assumed , where @xmath51 is the height of the surface from the equatorial plane , @xmath52 the distance on the equatorial plane from the center of the wd , @xmath53 the outer edge of the accretion disk , and @xmath54 is a numerical factor showing the degree of thickness . here , we assume @xmath55 during the strong wind phase because the flaring - up edge of the accretion disk is blown in the wind , but it increases to @xmath56 after the wind stops ( @xcite ) . the surface of the accretion disk is divided into 16 pieces logarithmically in the radial direction and into 32 pieces evenly in the azimuthal angle as shown in figure [ vmaxfig ] . the outer edge of the accretion disk is also divided into 32 pieces by rectangles . the luminosity of the accretion disk depends strongly on both the thickness of @xmath54 and the radius of @xmath53 . the size of the accretion disk is also assumed to be given by @xmath57 where @xmath58 is the effective radius of the inner critical roche lobe given by eggleton s ( 1983 ) formula . the viscous heating is neglected because it is rather smaller than that of the irradiation effects . we assume that the disk is tilting due to pringle s ( 1996 ) mechanism . the tilting disk is introduced by inclining the above disk by degree of @xmath59 with a precessing angular velocity of @xmath60 the initial phase of precession is assumed to be @xmath61 at the epoch of spectroscopic conjunction with the m - giant in front , i.e. , jd 243 1931.05 @xmath62 227.67@xmath63 at @xmath64 ( @xcite ) . to fit the first peak of the outburst light curve , we have calculated four cases of @xmath65-magnitude light curves with four different wd masses , i.e. , @xmath66 , @xmath67 , @xmath68 , and @xmath69 and found that both the @xmath13 and @xmath70 light curves are in much better agreement with the observational one than the other less massive ones as already shown by kato ( 1995 , 1999 ) . therefore , we adopt the @xmath21 in this letter . the optically thick wind stops about 100 days after the luminosity peak , around 1960 ( jd 2430000 + ) as shown in figure [ vmag1350va1 ] . here , we assume that the mass accretion rate ( @xmath31 ) to the wd is increased to a few times @xmath71 yr@xmath9 , because the disk is heated up by the wd photosphere ( also once engulfed by the wd photosphere ) . the hydrogen shell - burning vanishes 120 days after the peak when @xmath72 yr@xmath9 , while it continues for rather long period when @xmath73 yr@xmath9 as shown in table [ tbl-1 ] . the upper panel in figure [ vmag1350va1 ] shows the theoretical @xmath65-magnitude light curve ( solid line ) of the wd photosphere with @xmath74 yr@xmath9 and the companion without irradiation together with the observational points ( open circles ; taken from @xcite , b , c , d ) . two arrows indicate epochs of the spectroscopic conjunction with the m - giant in front ( @xcite ) . the ellipsoidal light variation is clearly shown in the later phase . the second panel depicts the @xmath65-magnitude including the effects of the companion irradiated by the wd . as already suggested by webbink et al . ( 1987 ) , this can not well reproduce the second peak of the light curves . the third panel shows the theoretical light curve further including the tilting accretion disk irradiated by the wd photosphere . it fully reproduces the observational light curve if we choose the fitting parameters of @xmath75 , @xmath56 , @xmath76 , @xmath77 , and @xmath61 . the radiation induced instability of the accretion disk sets in if the condition @xmath78 is satisfied ( @xcite ) . selvelli et al . ( 1992 ) estimated the accretion rate of t crb as @xmath79 yr@xmath9 , which meets condition ( [ radiation_condition ] ) . therefore , we can expect the radiation induced instability in t crb system . the growth timescale of warping is estimated by pringle ( 1996 ) and livio & pringle ( 1996 ) as the same timescale of the precessing period , i.e. , @xmath80 where we assume that the @xmath81-parameter of the standard accretion disk is @xmath82 . thus , the growth timescale is short enough to excite warping of the accretion disk . it should be noted here that the heated accretion disk surface , just after the maximum expansion of the wd photosphere , could drive a disk wind because the disk has once been engulfed by the wd photosphere . such a wind could exert an even larger back pressure on the disk than radiation and thus lead to tilting ( e.g. , @xcite ; @xcite ) . its growth timescale is much shorter than the timescale given by equation ( [ precession_time ] ) . the large inclination angle of the accretion disk such as @xmath83 is required to reproduce the observational light curve mainly because we need a reflection area as large as the companion , which is viewed from ours as shown in figure [ vmaxfig ] . about 10% faster precessing angular velocity is also required from the phase relation between the rising shoulder of the second peak near 1970 ( jd 2430000 + ) , a small dip near 2100 jd , and then a small bump near 2130 jd as shown in figure [ vmag1350va1 ] . these dips are caused by a large shadow on the companion cast by the accretion disk . this precession velocity is consistent with the fact that the radiation - induced precession is prograde , while the tidally induced precession is retrograde .	a new interpretation for the second peak of t coronae borealis ( t crb ) outbursts is proposed based on a thermonuclear runaway ( tnr ) model . the system consists of a very massive white dwarf ( wd ) with a tilting accretion disk and a lobe - filling red - giant . the first peak of the visual light curve of t crb outbursts is well reproduced by the tnr model on a wd close to the chandrasekhar mass ( @xmath0 ) , while the second peak is reproduced by the combination of the irradiated m - giant and the irradiated tilting disk . the derived fitting parameters are the wd mass @xmath1 , the m - giant companion mass @xmath2 ( @xmath3 is acceptable ) , the inclination angle of the orbit @xmath4 , and the tilting angle of the disk @xmath5 . these parameters are consistent with the recently derived binary parameters of t crb .
the accurate description of the single - particle ( s.p . ) strength in atomic nuclei is , to a large extent , an open issue ( for a recent discussion see , e.g. , ref . @xcite ) . whereas in light nuclei either _ ab - initio _ or shell model calculations are feasible , in the case of medium - heavy nuclei we miss a fully microscopic theory that is able to account for the experimental findings . modern self - consistent models ( either based on the mean - field hamiltonians or on some implementation of density functional theory ) do not reproduce , as a rule , the level density around the fermi surface . the reader can see , as a recent example , the results shown in ref . @xcite . moreover , the fragmentation of the s.p . strength is by definition outside the framework of those models . in the past decades , much emphasis has been put on the impact on the s.p . properties provided by the coupling with various collective nuclear motions . the basic ideas leading to particle - vibration coupling ( pvc ) models in spherical nuclei , or particle - rotation coupling models in deformed systems , have been discussed in textbooks @xcite . these couplings provide dynamical content to the standard shell model , in keeping with the fact that the average potential becomes nonlocal in time or , in other words , frequency- or energy - dependent . we will call self - energy , in what follows , the dynamical part of the mean potential arising from vibrational coupling . this contribution will be added to the static hartree - fock ( hf ) potential . in this way , one may be able to describe the fragmentation and the related spectroscopic factors of the s.p . states , their density ( which is proportional to the effective mass @xmath3 near the fermi energy ) , the s.p . spreading widths , and the imaginary component of the optical potential . in , e.g. , the review article @xcite one can find a detailed discussion about the points mentioned in the previous paragraph , together with the relevant equations and the results of many calculations performed in the 80 s for the single - particle strength ( mainly in @xmath1pb ) . these calculations are mostly not self - consistent and it is hard to extract from them quantitative conclusions because of the various approximations involved . certainly , they all agree qualitatively in pointing out that pvc plays a decisive role to bring the density of levels near the fermi energy in better agreement with experiment or , in other words , the effective mass @xmath3 close to the empirical value @xmath4 . this enhancement of the effective mass around the fermi energy , as compared to the hf value , is only one example of a phenomenon that can be explained by assuming that single - particle and vibrational degrees of freedom are not independent . other examples , although not treated in the current work , are worth to be mentioned in this introduction . several works have identified the exchange of vibrational quanta ( phonons ) between particles as one important mechanism responsible for nuclear pairing @xcite . in the approach of @xcite and references therein , one aims at explaining the properties of superfluid nuclei by taking into account both the pairing induced by the phonon exchange and the self - energy mentioned above ( cf . also the discussion in refs . we also remark that important developments are under way aiming at implementing _ ab - initio _ calculation schemes for open shell nuclei , based on self - consistent green s functions @xcite or on the unitary correlation operator method @xcite ) . along the same line , more complicated processes can be explained by starting from elementary single - particle and vibrational degrees of freedom , and treating their coupling within the framework of an appropriate field theory : the spreading width of nuclear giant resonances , or the anharmonicity of two - phonon states ( to mention only a few examples ) . the development of such a general many - body perturbation theory scheme could not avoid , so far , to resort to various approximations . in particular most of the calculations have employed simple , phenomenological coupling hamiltonians . recently , in order to calculate the s.p . strength , microscopic pvc calculations have become available , either based on the nonrelativistic skyrme hamiltonian @xcite or on relativistic mean field ( rmf ) parameterizations @xcite . the results seem to be satisfactory , in a qualitative or semi - quantitative sense , as they point to an increase of the effective mass around the fermi energy . the results are clearly sensitive to the collectivity of the low - lying phonons produced by the self - consistent calculations . it is still unclear whether the results will eventually be improved by a re - fitting of the effective interactions or by the inclusion of higher - order processes . one of the limitations of all the pvc models that have been introduced so far , lies in the fact that they discretize the s.p . continuum ( clearly , this means that the description of the vibrations themselves relies on the same approximation ) . although in ref . @xcite a scheme to calculate the self - energy in coordinate - space representation had been proposed , there is at present no available result for the s.p . strength ( let alone more complex physical observables ) that avoids the continuum discretization . consequently , the goal of the present work is to introduce for the first time a consistent description of pvc with a proper treatment of the continuum . in order to achieve this , the coordinate space representation is used . the current work is based on previous experience on how to treat the continuum within the linear response theory or random phase approximation ( rpa ) ( see refs . @xcite ) . the outline of the paper is as follows . in sec . [ formalism ] , we describe our formalism starting from the general formulation and stressing the implementation of proper continuum treatment . the main goal of the section is to display the equations that we have implemented and solved , and discuss the s.p . level density . in sec . [ results ] , the results for our three nuclei of choice are presented and discussed ; whenever possibile , they are compared with experimental data . finally , we summarize the paper and draw our conclusions in sec . [ summary ] . some details of the calculations are shown in a few appendices . the particle - vibration coupling ( pvc ) hamiltonian @xcite in coordinate space can be written as @xmath5 the density variation operator @xmath6 ( where the brackets denote the ground - state expectation value ) in second quantized form is given by @xmath7,\end{aligned}\ ] ] where @xmath8 and @xmath9 are the creation and annihilation operators , respectively , of a phonon @xmath10 having multipolarity @xmath11 and @xmath12 is the corresponding transition density , whereas @xmath13 is the residual force . if the total hamiltonian is @xmath14 , where the term @xmath15 describes uncoupled s.p . states and vibrations , the many - body perturbation theory @xcite can be applied . in particular , we assume that @xmath16 includes the hf hamiltonian for the nucleons and the independent boson hamiltonian for the phonons ( based on their rpa energies ) . we treat the term @xmath17 as a perturbation using the interaction picture . we define green s functions in space - time representation and we apply standard tools like the wick s theorem to obtain the dyson equation in terms of the unperturbed hf green s function @xmath18 and the perturbed green s functions @xmath19 : @xmath20 the hf green s function satisfies @xmath21 , where @xmath22 is the s.p . hf hamiltonian . the self - energy function is defined by @xmath23 where @xmath24 is the rpa response function ( or phonon propagator ) in the space - time representation , and is defined by @xcite @xmath25 since this is exactly the phonon vacuum . we also note that the use of the wick s theorem in the derivation of eq . ( [ hfdyson1 ] ) implies the use of the _ causal _ representation of the green s functions @xmath19 and @xmath18 , as well as of the rpa response function @xmath26 . the connection between the causal representation with the retarded and advanced representations is outlined in the appendices , where the causal functions will be denoted by @xmath27 . this label will be omitted in the main text , where we shall only make use of the causal functions . ) , in the approximation in which @xmath18 replaces @xmath19 . ] the fourier transform of eq . ( [ hfdyson1 ] ) is given by @xmath28 while the fourier transform of the self - energy reads @xmath29 due to the convolution theorem . a self - consistent solution of the dyson equation involves the iteration of the two previous equations until convergence is reached . in practice , this is almost never done . in our work , since we explore for the first time the proper continuum coupling , we limit ourselves to the first iteration by replacing @xmath19 with @xmath18 in eq . ( [ selfen ] ) . we restrict our investigation to spherical systems in which static pairing correlations vanish . by taking profit of the spherical symmetry , one can use partial wave expansions and arrive at @xmath30 where @xmath31_{jm}$ ] . the latter equation holds evidently for @xmath18 as well . the dyson equation can then be written as @xmath32 similarly , the rpa response function and the residual force can be represented as @xmath33 and consequently the self - energy can be calculated by @xmath34 ) or eq . ( [ hfdyson2 ] ) ] . ] in our work , we start from the hf green s function and rpa response function ( together with the residual force @xmath13 ) and we obtain the self - energy from eq . ( [ selfen2 ] ) . then , we also solve numerically the dyson equation in the form ( [ hfdyson3 ] ) : for every energy of interest this equation can be cast in matrix form with respect to @xmath35 and @xmath36 and solved as @xmath37 in this way , the perturbed green s function @xmath19 contains the pvc perturbation up to infinite order , in keeping with the fact that it can be expressed by the feynman diagrams represented in fig . [ dysondiag ] . as already mentioned , our goal is an implementation of pvc that treats the continuum properly . in the case of atomic nuclei , in particular when local functionals like those based on the skyrme interaction are used , considerable efforts have been made in this direction as far as the hf - rpa formalism is concerned . indeed , the green s function rpa has been formulated with skyrme forces , with or without @xcite the continuum ; the first self - consistent continuum calculations have been presented in ref . @xcite . in this context , proper treatment of the continuum means that the schrdinger equation including the hf mean field can be solved at any positive energy with the correct boundary conditions and , based on this , an exact representation of the hf green s function @xmath18 can be obtained . this unperturbed hf green s function can be written as @xmath38 where @xmath39 and @xmath40 are , respectively , the regular and irregular solutions of the radial hf equation at energy @xmath41 , @xmath42 ( @xmath43 ) are the larger ( smaller ) between @xmath35 and @xmath36 , and @xmath44 is the wronskian given by @xmath45 @xmath46 is the ( radial - dependent ) hf effective mass which is defined as usual , in terms of the skyrme force parameters , as @xmath47 in order to take properly into account the continuum effects also for the rpa phonons that lie above the threshold , the rpa response function appearing in the self - energy function will be calculated self - consistently , using the same skyrme hamiltonian used to compute the mean field . the details of the continuum rpa calculation have been given in previous papers @xcite . we simply recall that two - body spin orbit and coulomb terms , as well as spin - dependent terms , are dropped in the residual interaction . it is also necessary to convert the continuum rpa response function into the causal function , because normally the linear response theory ( rpa ) is formulated in terms of the retarded functions . this point is further discussed in appendix b. formally , the equations that appear in subsec . [ iiia ] are defined in a model space which does not have an upper bound : in fact , the integrals over energy extend in principle from @xmath48 to @xmath49 , and single - nucleon as well as phonon energies have only a natural lower bound . however , the self - energy function does not converge if the upper limit on @xmath50 [ eqs . ( [ selfen ] ) and ( [ selfen2 ] ) ] is extended to infinity , in keeping with the well - known ultraviolet divergence associated with the zero - range character of skyrme forces . to avoid this , one must introduce a cutoff @xmath51 . in order to make sure that only states below that cutoff contribute to the integrals in eq . ( [ selfen ] ) and ( [ selfen2 ] ) , one can use the following procedure . by considering the expression ( [ selfen2 ] ) for the self - energy function , one notices that the integral receives contribution from the poles of the causal hf green s function and the causal rpa response function . the positions of these discrete and continuum ( i.e. , branch - cut ) poles in the complex energy plane are schematically shown in fig . [ contour ] . the blue dots and line represent the poles of the rpa response function , while the black crosses and line represent the poles of the hf green s function . in order to pick up correctly the contribution of the poles below the cutoff , we must replace the integral @xmath52 in eq . ( [ selfen2 ] ) by an integral @xmath53 over an appropriate contour path . we have adopted the rectangular integration path displayed in fig . [ contour ] , which is similar to that employed in ref . it extends between @xmath54 to @xmath51 on the real axis , and from 0 to -@xmath55 on the imaginary axis . it can also be shown that in this way one can reproduce the correct spectral representation of the self - energy function [ cf . ( [ selfenrt3 ] ) ] in the limit of a discrete system . for the integration on @xmath56 in eq . ( [ selfen ] ) and eq . ( [ selfen2 ] ) . the blue dots and lines represent the poles of the rpa response function . the black crosses and line represent the poles of the hf green s function . for the parameters @xmath57 , @xmath55 and @xmath51 see the text . ] the level density associated with the hf single - particle levels can be defined by using the hf green s function @xmath18 as @xmath58 where @xmath26 is the upper limit of the integration . for bound states at negative energies , one is guaranteed that the result is stable with respect to increasing @xmath26 . the sign @xmath59 guarantees that the level density is positive , _ i.e. _ , the sign + ( - ) refers to particle ( hole ) states . this is equivalent to the definition @xmath60 and is normalized to 1 for bound states . in the absence of any potential , @xmath61 reduces to the free particle level density @xmath62 , obtained by replacing @xmath63 with the green s function for the free particle which satisfies @xmath64 . @xmath65 can be calculated either numerically or analytically using the same definition of eq . ( [ cgf ] ) , but with the use of the wave function for the free particle . this free particle level density is @xmath66 for large value of @xmath67 . for @xmath68 , @xmath69 diverges . on the other hand , @xmath61 tends to @xmath69 for large values of @xmath67 . it is then useful to introduce a new level density @xmath70 by subtracting @xmath62 @xcite , that is , @xmath71 in this way , the dependence on @xmath26 is eliminated also for positive energies ( cf . [ lvdhf ] ) . for @xmath72 , there is no contribution associated with the free particles . as mentioned above , @xmath73 coincides with the usual definition of the level density for the single - particle levels @xcite @xmath74 in an analogous way , we define the perturbed ( hf+pvc ) level density by using the solution of the dyson equation @xmath75 as @xmath76 the peaks of the perturbed level density provide renormalized single - particle energies which include the effect of the particle - phonon coupling . in fact , if this coupling is small one can expect a simple shift of the hf peaks . otherwise , the s.p . strength can be quite fragmented : the associated widths reflect the basic decay mechanisms that are the nucleon decay ( providing the so - called escape width , or @xmath77 ) and the spreading into the complicated configurations made up with nucleons and vibrations ( providing the spreading width @xmath78 ) . we shall present results for three nuclei : @xmath0ca , @xmath1pb and @xmath2o . the effective skyrme interaction sly5 @xcite is used to calculate the hf mean field . the calculation of the rpa response is carried out exactly as described in ref . @xcite in the limit of no pairing , using all the terms of the sly5 interaction , except for the two - body spin - dependent terms , the spin - orbit terms and the coulomb term in the residual p - h force . in the calculations , the angular momentum cutoff for the unoccupied continuum states is set at @xmath79 for @xmath0ca , and @xmath80 for @xmath2o and @xmath1pb , respectively . the radial mesh size is @xmath81 fm . the values of the parameters used in the contour integrations ( see the discussion in [ sub_contour ] ) are @xmath82 0.2 mev [ cf . ( [ selfen2 ] ) ] and @xmath83 [ cf . ( [ hfdyson3 ] ) ] . there are a few important issues that we wish to stress : 1 . due specifically to the zero - range character of the skyrme interaction , the self - energy diverges logarithmically as a function of the maximum energy of the phonons , as it has mentioned above . the first steps towards a systematic renormalization procedure have only recently been started to be worked out @xcite . in this work , we shall take the usual view that the important couplings are those associated with the collective low - lying states and giant resonances . we shall then include phonons associated with the multipolarities @xmath84 , @xmath85 , @xmath86 and @xmath87 , and set an upper cutoff on the phonon energies given by @xmath88 mev because no strong peaks are present above this value in the calculated rpa strengths . the way in which this cutoff is implemented has been described in detail in [ sub_contour ] . 2 . in the present scheme , the price to be paid for the exact continuum treatment , is that one can not discriminate between the inclusion of collective and non - collective phonons . actually , the diagrams shown in figs . [ selfendiag ] and [ dysondiag ] contain terms that violate the pauli principle , and these terms are larger when the phonons are non - collective . in other words , one could expect that in an exact calculation the correction of the pauli principle violation cancels , to a large extent , the contributions from non - collective phonons . although this point has never been clarified in the available literature , to our knowledge , on a quantitative basis , in most of the cases the usual view has been to take into account only the coupling to collective states . in the work that we quoted already ( the most similar to the present one ) , namely in the recent calculation of particle - vibration coupling in @xmath0ca and @xmath1pb of ref . @xcite , only phonons exhausting at least 5% of the isoscalar or isovector non - energy - weighted sum rules have been taken into account . therefore , we can expect that the effect of particle - vibration coupling is larger when we calculate it with the present method , as compared with ref . we shall come back to this issue below . the momentum - dependent part of the particle - hole interaction had previously been neglected in the calculation of the particle - vibration coupling ; then , in ref . @xcite it was shown that its effect is important ( at least in the case of the sly5 interaction ) , and that it can be reasonably accounted for within the landau - migdal ( lm ) approximation , by choosing the fermi momentum @xmath89 as 1.33 @xmath90 ( that is , at the value associated with the nuclear matter saturation density ) . the lm approximation will be adopted in the following , with the same value of @xmath89 . the first essential steps of our work consist in the calculation of the hf spectrum and of the rpa strength functions . the results are illustrated respectively in table [ splvtbca40 ] ( hf single - particle spectrum ) and in fig . [ str_ca40 ] ( isoscalar and isovector rpa strength functions associated with the multipolarities @xmath91 and @xmath87 ) . in table [ lwst_ca40 ] we give the theoretical energies and transition strengths for the low - lying collective @xmath85 and @xmath87 states , comparing them with available data . .neutron single - particle energies in @xmath0ca . [ cols="^,^,^,^,^ " , ] in fig . [ lvdo24 ] , we display our results for the level density of @xmath2o . before entering into some detail , we discuss the main emerging features and compare with what is known experimentally . in our calculation , the 1d@xmath92 and 2s@xmath93 states have a marked quasiparticle character , namely they are associated with a single narrow peak . it makes sense , therefore , to compare the experimental value of the gap with the hf and hf - pvc results for the energy difference between the 1d@xmath92 and 2s@xmath93 states , that is , 4.0 and 4.6 mev respectively . the hf - pvc result is in good agreement with the experimental value of 4.86 mev . in heavy nuclei , as a rule , the pvc shrinks the single - particle gap and increases the effective mass ( cf . the previous subsection ) , but this is not the case in light nuclei due to the specific effect of having only low angular momentum occupied states ( as already noticed in ref . @xcite ) . in the present case , while the pvc pushes the 1d@xmath92 orbital closer to the fermi energy ( -2.5 mev compared to the hf value of -1.3 mev ) , the 2s@xmath93 hole state is pushed further from it ( -7.1 mev compared to the hf value of -5.3 mev ) . the peak energies of the other orbitals obtained by using hf - pvc ( hf ) read 2.4 mev ( 4.3 mev ) for 1f@xmath94 , and -8.3 mev ( -7.5 mev ) for 1d@xmath95 . the net effect of pvc is a shift down of the states . the absolute value of the energies is expected to depend on the choice of the effective force . skyrme forces , as other mean - field frameworks , tend to predict larger binding in light neutron - rich nuclei as compared with the experimental findings , as it is clearly testified by the fact that @xmath96o turns out to be bound in many of these models . in the present case , the 1d@xmath92 state is bound while it should be a resonant state . we can nonetheless look at relative energy differences . the known states in @xmath97o taken from ref . @xcite are , in addition to the 1/2@xmath98 ground - state , a 5/2@xmath98 state at 2.79 mev and a 3/2@xmath98 at 4.04 mev ( leaving aside the state at 5.34 whose character is not clear , being either 3/2@xmath99 or 7/2@xmath99 ) . these are states that can decay to the @xmath100o ground state . in our calculation we can identify states below the energy threshold for this kind of decay : in particular the first 5/2@xmath98 state lies at 1.2 mev in our calculation . we now discuss , the couplings that produce fragmentation of most of the single - particle strength distributions . at variance with the state 2s@xmath93 , the state 1s@xmath93 is strongly fragmented . this fragmentation is chiefly caused by the configurations 1d@xmath101 and @xmath102 . by inspecting the energy difference , we can assume that the ivgqr and ivgdr play the main role for the fragmentation . the state @xmath103 is fragmented due to the coupling with the configuration @xmath104 : we expect that the isgor and ivgor play the main role by considering the energy matching . @xmath105 can also contribute to the fragmentation , yet to a minor extent . the main configuration giving rise to the fragmentation of the state @xmath106 is the configuration @xmath107 . here , due to the energy difference between the hole states @xmath106 and @xmath108 , the dipole excitations around @xmath109 mev do play the main role . in the case of the fragmentation of the @xmath108 state , the configuration @xmath110 is the most important one , and the low - lying @xmath84 state at 3.4 mev is the most relevant . finally , the energy shift of the state @xmath111 is mostly caused by the coupling with the configuration @xmath112 . ccccccc + & & + & & & & + @xmath113 & exp . & theory & & @xmath113 & exp . & theory + s@xmath93 & & 0.81 & & d@xmath92 & & 0.83 + d@xmath95 & & 0.78 & & & & + last but not least , we come back to the point raised above , namely that the ( quasi - particle - like ) states around the fermi energy , 1d@xmath92 and 2s@xmath93 , are quite pure ( see also table [ table_spect_o24 ] ) . in our calculation , there are not optimal energy matching of those states with other configurations due to the scarcity of low - lying collective excitations and large gaps between single - particle state . more generally , in our calculations the coupling of neutron states is mainly with the proton component of the phonon states ( due to the dominance of the neutron - proton interaction ) . therefore , neutron states on neutron - rich nuclei are expected to be more pure because proton excitations are pushed at higher energy as the neutron excess increases . recently , other calculations of the spectroscopic factors in light nuclei within the coupled - cluster approach have become available in ref . @xcite ( see also the critical discussion in ref . the idea that single - particle strength is not systematically pure in atomic nuclei , and that coupling with other degrees of freedom is quite relevant , is an old idea in nuclear physics . phenomenological calculations based on particle - vibration coupling ( pvc ) for spherical nuclei have been performed for several deacades , and they have been quite instrumental to point out some of the limitations of the pure mean - field approach ( like , e.g. , the enhancement of the effective mass around the fermi energy ) . microscopic pvc calculations based on the consistent use of an effective hamiltonian , have become available only recently . none of the mentioned calculations , to our knowledge , takes proper care of the continuum . in our work , for the first time , we have implemented a scheme based on coordinate - space representation in which the single - particle states , the vibrations , and their coupling are calculated with proper inclusion of the continuum . this is of special interest if weakly - bound nuclei close to the drip lines are to be studied . however , also in well - bound nuclei the present approach present advantages in the sense that resonant states can be properly studied . transfer to the continuum has been the subject of several experimental studies . in stable nuclei we obtain results that are in overall agreement with previous studies . we can , at the same time , better describe the fragmentation of the single - particle strength . the shifts of the single - particle states around the fermi energy , with respect to the hf values , are relatively large ( in keeping also with the fact that we can not restrict our coupling to collective states only ) . we obtain an overall agreement with experiment in @xmath1pb , while in the case of @xmath0ca our results point to the need of re - fitting the skyrme interaction that has been devised to work at the mean - field level and not beyond it . we have also applied our model to a neutron - rich nucleus , namely @xmath2o . this is a double - magic nucleus , and there are few low - lying states . because of this , and also since the neutron states energy would be more affected by coupling with protons , the neutron single - particle strength around the fermi energy is quite pure ( i.e. , spectroscopic factors are rather close to one ) . while this is in agreement with some experimental findings , certainly more detailed spectroscopic studies are needed to extract a global trend and a firm understanding . we thank f. barranco for discussions that initiated the present work , and p.f . bortignon for useful suggestions . this work has been partly supported by the italian research project `` many - body theory of nuclear systems and implications on the physics of neutron stars '' ( prin 2008 ) " . as it is stated in the main text , it is necessary to use the _ causal _ green s function for the dyson equation because this equation is based on the use of the wick s theorem , and the wick s theorem applies only to time - ordered products . on the other hand , the continuum hf green s function is given in the form of a _ retarded _ function . so we need to compute the _ causal _ green s function starting from the _ retarded _ green s function in order to use the continuum hf green s function in the dyson equation . the causal hf green s function is defined by @xmath114 the retarded green s function is instead defined by @xmath115 from these two definitions , and using @xmath116 , we can find that @xmath117 the fourier transform of eq . ( [ randn ] ) is expressed as @xmath118 the continuum hf green s function @xmath63 given by eq . ( [ cgf ] ) is regular for the complex energy @xmath41 . so the retarded green s function with a smearing width @xmath57 is expressed as @xmath119 where @xmath67 is the real part of the complex energy . then the continuum causal green s function can be expressed by @xmath120 in general , the rpa theory can be formulated in two ways . one is based on the _ causal _ function , while another one on the _ retarded _ function . both formulations give the same results for the physical part of the spectrum , namely for positive excitation energy . however , a complete rpa basis must include negative energy states and the two aforementioned formulations are different for negative energies . in order to construct the self - energy function of eq . ( [ selfen ] ) , we need to consider the rpa response function not only in the positive energy domain but also in the negative energy domain due to the required energy integration . so we can not use the _ retarded _ rpa response function for the self - energy function . the rpa equation for the response function is given by @xmath121 in any of the representations , @xmath122 being the unperturbed response function , and @xmath26 the rpa response function . in order to obtain the _ causal _ rpa response function by solving this equation , the _ causal _ unperturbed response function should be used . normally the continuum ( q)rpa is formulated by using _ retarded _ functions within the linear response theory . the continuum hf(b ) green s function is used to build the ( retarded ) unperturbed response function @xcite . it is therefore necessary to know how to convert the retarded unperturbed response function to the causal function in the continuum rpa formalism . we show it in the present appendix . the _ causal _ and _ retarded _ response function in rpa are defined by @xmath123 @xmath124 is the hf ground state ) . from these definitions , one can find the relation between them as follows , @xmath125 the fourier transformation of the latter equation gives @xmath126 where @xmath127 can be expressed by means of the retarded hf green s function as @xmath128 in the continuum rpa formalism , the continuum hf green s function is used as @xmath129 in eq . ( [ cunpres ] ) . here we show the spectral representations of the hf green s function and the rpa response function ( both causal and retarded ) . the difference will appear in the sign of the imaginary part @xmath57 . actually this is very important to obtain the proper self - energy function by using the contour integration , because this sign differencee produce changes in the position of the poles of the green s function and of the response function on the complex energy plane ( this fact is connected with the fact that the wick s theorem can be applied only for the causal function , as already mentioned ) . @xmath130 @xmath131 the self - energy function in the space - time representation is defined by eq . ( [ selfenrt ] ) . if we insert the hf and the rpa results in this definition , then the self - energy function can be expressed as @xmath132 the fouriter transform of eq . ( [ selfenrt2 ] ) gives @xmath133 here we show the explicit expression of the residual interaction within the so - called landau - migdal approximation . this residual force is used in the self - energy function [ eq . ( [ selfen2 ] ) ] . @xmath134 } \\ & \displaystyle { + \frac{1}{4}\left(t_1(1-x_1)+3t_2(1+x_2)\right)k_f^2 } \\ & \\ ( q\neq q ' ) & \displaystyle { t_0(1+\frac{x_0}{2 } ) } \\ & \displaystyle { + \frac{t_3}{12}\rho^\gamma\big [ ( \gamma+2)(\gamma+1)(1+\frac{x_3}{2 } ) - ( x_3+\frac{1}{2 } ) \left ( 2\gamma + \gamma(\gamma-1)\sum_\alpha\left(\frac{\rho_\alpha}{\rho}\right)^2 \right ) \big ] } \\ & \displaystyle { + \frac{1}{2}\left(t_1(1+\frac{x_1}{2})+t_2(1+\frac{x_2}{2})\right)k_f^2 } \end{array } \right.\end{aligned}\ ] ] 00 p. f. bortignon , g. col , h. sagawa , j. phys . g 37 , 064013 ( 2010 ) . m. stoitsov , m. kortelainen , s. k. bogner , t. duguet , r. j. furnstahl , b. gebremariam , and n. schunck , phys . c 82 , 054307 ( 2010 ) . a. bohr , b. r. mottelson , _ nuclear structure . ( w.a . benjamin , new york , 1971 ) . c. mahaux , p. f. bortignon , r. a. broglia , c. h. dasso , phys . 120 , 1 ( 1985 ) . f. barranco , r. a. broglia , g. gori , e. vigezzi , p. f. bortignon , and j. terasaki , phys . 83 , 2147 ( 1999 ) . a. v. avdeenkov and s. p. kamerdjev , jetp lett . 69 , 669 ( 1999 ) . a. pastore , f. barranco , r. a. broglia , and e. vigezzi , phys . c78 , 024315 ( 2008 ) . a. idini , f. barranco , r.a . broglia and e. vigezzi , phys . c 85 , 014331 ( 2012 ) . k. hebeler , t. duguet , t. lesinski , and a. schwenk , phys . c 80 , 044321 ( 2019 ) . m. baldo , u. lombardo , s. s. pankratov , and e. e. saperstein , j. phys . g : nucl . part . phys . 37 , 064016 ( 2010 ) . v. som , t. duguet and c. barbieri , phys . rev c 84 , 024312 ( 2012 ) . h. hergert and r. roth , phys . c 80 , 024312 ( 2009 ) . g. col , h. sagawa , p. f. bortignon , phys . c82 , 064307 ( 2010 ) . e. litvinova , p. ring , phys . c73 , 044328 ( 2006 ) . e. v. litvinova and a. v. afanasjev , phys . c84 , 014305 ( 2011 ) . g. col , p. f. bortignon and r. a. broglia , nucl . a649 , 335c ( 1999 ) . m. matsuo , nucl . a696 , 371 ( 2001 ) . k. mizuyama , m. matsuo , and y. serizawa , phys . c79 , 024313 ( 2009 ) . g. f. bertsch and s. f. tsai , phys . 18 , 125 ( 1975 ) . s. shlomo and g. f. bertsch , nucl . phys . a 243 , 507 ( 1975 ) . k. f. liu and n. van giai , phys . b 65 , 23 ( 1976 ) . a. l. fetter , j. d. walecka , _ quantum theory of many - particle systems _ ( mcgraw - hill , new york , 1971 ) . r. d. mattuck , _ a guide to feynman diagrams in the many - body problem _ ( dover publications , new york , 1992 ) . h. sagawa , prog . 142 , 1 ( 2001 ) . s. shlomo , nucl . a539 , 17 ( 1992 ) . e. chabanat , p. bonche , p. haensel , j. meyer and r. schaeffer , nucl . a643 , 441 ( 1998 ) . k. moghrabi , m. grasso , g. col and n. van giai , phys . 105 , 262501 ( 2010 ) ; k. moghrabi , m. grasso , x. roca - maza , and g. col , phys . rev . c85 , 044323 ( 2012 ) . t. kibdi and r.h . spear , atomic data and nuclear data tables 80 , 35 ( 2002 ) . cameron and b. singh , nuclear data sheets 102 , 293 ( 2004 ) . b. singh and j.a . cameron , nuclear data sheets 107 , 225 ( 2006 ) . cameron and b. singh , nuclear data sheets 94 , 429 ( 2001 ) . r. satchler , _ direct nuclear reactions _ ( oxford university press , oxford , 1983 ) . eckle , h. lenske , g. eckle , g. graw , r. hertenberger , h. kader , h.j . maier , f. merz , h. nann , p. schiemenz and h.h . wolter , nucl . a506 , 159 ( 1990 ) . harakeh , a. van der woude , _ giant resonances . fundamental high - frequency modes of nuclear excitation _ ( oxford university press , oxford , 2001 ) . s. kamerdzhiev , j. speth and g. tertychny , phys . 393 , 1 ( 2004 ) . s. raman , c.w . nestor and p. tikkanen , atomic data and nuclear data tables 78 , 1 ( 2001 ) . f.g . kondev and s. lalkovski , nuclear data sheets 112 , 707 ( 2011 ) . martin , nuclear data sheets 63 , 723 ( 1991 ) . kramer , h.p . block and l. lapiks , nucl . a679 , 267 ( 2001 ) . w.h . dickhoff and d. van neck , _ many - body exposed ! _ ( world scientific , singapore , 2005 ) . r.j . furnstahl and h.w . hammer , phys . b531 , 203 ( 2002 ) . jennings ( 2011 ) , arxiv:1102.3721 . brown and w.a . richter , phys . c74 , 034315 ( 2004 ) y. utsuno , t. otsuka , t. misuzaki and m. honma , phys . c60 , 054315 ( 1999 ) t. otsuka _ 87 , 082502 ( 2001 ) . et al . _ , phys . 100 , 152502 ( 2008 ) . et al . _ , b672 , 17 ( 2009 ) . a. gade _ et al . c77 , 044306 ( 2008 ) . m. stanoiou _ et al . _ , rev , c69 , 034312 ( 2004 ) . k. mizuyama and g. col , phys . c85 , 024307 ( 2012 ) . a. schiller _ et al . _ , 99 , 112501 ( 2007 ) . jensen , g. hagen , m. hjorth - jensen , b.a . brown , and a. gade , phys . 107 , 032501 ( 2011 ) . t. duguet and g. hagen , phys . c85 , 034330 ( 2012 ) .	in this paper we present a new formalism to implement the nuclear particle - vibration coupling ( pvc ) model . the key issue is the proper treatment of the continuum , that is allowed by the coordinate space representation . our formalism , based on the use of zero - range interactions like the skyrme forces , is microscopic and fully self - consistent . we apply it to the case of neutron single - particle states in @xmath0ca , @xmath1pb and @xmath2o . the first two cases are meant to illustrate the comparison with the usual ( i.e. , discrete ) pvc model . however , we stress that the present approach allows to calculate properly the effect of pvc on resonant states . we compare our results with those from experiments in which the particle transfer in the continuum region has been attempted . the latter case , namely @xmath2o , is chosen as an example of a weakly - bound system . such a nucleus , being double - magic and not displaying collective low - lying vibrational excitations , is characterized by quite pure neutron single - particle states around the fermi surface .
the general interest in networks of man - made and natural systems has lead to a careful analysis of various network instances using empirical , simulational , and theoretical tools . the emergence of this field is sometimes referred to as the birth of network science @xcite . in this paper , we use complex network concepts to analyze the statistical properties of public transport networks ( ptn ) of large cities . these constitute an example of transportation networks @xcite and share general features of these systems : evolutionary dynamics , optimization , embedding in two dimensional ( 2d ) space . other examples of transportation networks are given by the airport @xcite , railway @xcite , or power grid networks @xcite . routes and @xmath0 stations , some of them are shown in this map ( color online ) . [ fig1],width=302 ] while the evolution of a ptn of a given city is closely related to the city growth itself and therefore is influenced by numerous factors of geographical , historical , and social origin , there is ample evidence that ptns of different cities share common statistical properties that arise due to their functional purposes @xcite . some of these properties have been analyzed in former studies , however , the objective of the present study is to present a comprehensive survey of characteristics of ptns and to provide a comparative analysis . based on this empirical survey we are in the position to propose a growth model that captures many of the main ( statistical ) features of ptns . a further distinct feature of our study is that the ptns we will consider are networks of _ all _ means of public transport of a city ( buses , trams , subway , etc . ) regardless of the specific means of transport . a number of studies have analyzed specific sub - networks of ptns @xcite . examples are the boston @xcite and vienna @xcite subway networks and the bus networks of several cities in china @xcite . however , each particular traffic system ( e.g. the network of buses or trams , or the subway network ) is not a closed system : it is a subgraph of a wider transportation system of a city , or as we call it here , of a ptn . therefore to understand and describe the properties of public transport in a city as a whole , one should analyze the complete network , without restriction to specific parts . indeed , for the case of boston it has been shown that changing from the subway system to the network subway + bus " the network properties change drastically @xcite . urban public transport networks of general type have so far been analyzed mainly in two previous studies @xcite . in the first one , ref . @xcite , the ptns of berlin , dsseldorf , and paris were examined , whereas the subject of ref . @xcite were public transport systems of 22 polish cities . @xcite concentrated on the scale - free properties . for the cities considered , the node degree distribution was shown to follow a power law . moreover power laws were found for a number of other specific features describing the traffic load on the ptn . however , the statistics in this study was too small for definite conclusions . in ref . @xcite it was found that the node degree distribution may follow a power law or be described by an exponential function , depending on the assumed network representation . besides , a number of other network characteristics ( clustering , betweenness , assortativity ) were extensively analyzed . in the present paper , we analyze ptns of a number of major cities of the world ( see table [ tab1 ] ) @xcite . our choice for this data base was motivated by the requirement to collect network samples from cities of different geographical , cultural , and economical background . our current analysis extends former studies @xcite by considering cities with larger public transport systems ( the typical number of stops in the systems considered in ref . @xcite was several hundreds ) as well as by systematically analyzing different representations . the idea of different network representations naturally arises in the network science @xcite . for the ptn the primary network topology is given by the set of routes each servicing an ordered series of given stations ( see fig . [ fig1 ] as an example ) . for the transportation networks studied so far mainly two different neighborhood relations were used . in the first one , two stations are defined as neighbors only if one station is the successor of the other in the series serviced by this route @xcite . in the second one , two stations are neighbors whenever they are serviced by a common route @xcite . we will exploit both representations in our study . moreover , we introduce further natural representations ( described in detail in section [ ii ] ) which make the description of the ptns of table [ tab1 ] comprehensive . in particular , this includes a bipartite graph representation of a transportation network that reflects its intrinsic features @xcite . 1.5 mm .cities analyzed in this study . @xmath1 : urban area ( km@xmath2 ) ; @xmath3 : population ( million inhabitants ) ; @xmath4 : number of ptn stations ; @xmath5 : number of ptn routes ; @xmath6 : mean route length . type of transport which is taken into account in the ptn database : bus , electric trolleybus , ferry , subway , tram , urban train . [ cols="<,>,>,>,>,>,<",options="header " , ] let us first investigate the global topology of this model as function of its parameters . we first fix both the number of routes @xmath5 and the number of stations @xmath6 per route as well as the size of the lattice @xmath7 . this leaves us with essentially two parameters @xmath8 and @xmath9 , eqs . ( [ 6.1 ] ) , ( [ 6.2 ] ) . dependencies on @xmath5 , @xmath6 , and @xmath7 will be studied below . for the real - world ptns as studied in the previous sections , almost all stations belong to a single component , gcc , with the possible exception of a very small number of routes . within the network however we often observe what above we called the harness effect of several routes proceeding in parallel for a sequence of stations . let us first investigate from a global point of view which parameters @xmath8 and @xmath9 reproduce realistic maps of ptns . in fig . [ fig17 ] we show simulated ptns on lattices @xmath10 for @xmath11 , @xmath12 and different values of the parameters @xmath8 and @xmath9 . each route is represented by a continuous line tracing the path along its sequence of stations . for representation purposes , parallel routes are shown slightly shifted . thus , the line thickness and intensity of colors indicate the density of the routes . parameter @xmath9 governs the evolution of each single subsequent route . if @xmath13 each subsequent route is restricted to follow the previous one . the change of simulated ptns with @xmath9 for fixed @xmath14 is shown in the first row of fig . [ fig17 ] . for small values of @xmath15 the ptns obtained result in almost all routes following the same path with only a few deviations . increasing @xmath9 from @xmath16 to @xmath17 the area covered by the routes increases while the majority of the routes are concentrated on a small number of paths . further increasing @xmath9 to @xmath18 and beyond we find a wider distributed coverage with the central part of the network remaining the most densely covered area . this is due to the non - equilibrium growth process described by eqs . ( [ 6.1 ] ) , ( [ 6.2 ] ) . the parameter @xmath8 quantifies the possibility to start a new route outside the existing network . for vanishing @xmath14 the resulting network always consists of a single connected component , while for finite values of @xmath8 a few or many disconnected components may occur . the results found for @xmath14 and varying @xmath9 parameters are completely independent of the lattice size @xmath7 provided @xmath7 is sufficiently large . when introducing a finite @xmath8 parameter , however , new routes may be started anywhere on the lattice which results in a strong lattice size dependency . to partly compensate for this , the impact of @xmath8 has been normalized by @xmath19 in ( [ 6.1 ] ) . the dependence of the simulated ptn maps on @xmath8 for fixed @xmath18 is shown in the second row of fig . [ fig17 ] . for @xmath20 one observes the formation of a single large cluster with only a few individual routes occurring outside this cluster . slightly increasing @xmath8 beyond @xmath21 one finds a sharp transition to a situation with several ( two or more ) clusters . for much larger values of @xmath8 the number of clusters further increases and the situation becomes more and more homogeneous : the routes tend to cover all available lattice space area . * a * * a * from the above qualitative investigation we conclude that realistic ptn maps are obtained for small or vanishing @xmath8 and @xmath22 . to quantitatively investigate the behavior of the simulated networks on their parameters including @xmath5 and @xmath6 let us now compare their statistical characteristics with those we have empirically obtained for the real - world networks . in table [ tab5 ] we have chosen to list the same characteristics of the simulated ptn as they are displayed for the real - world networks in table [ tab2 ] . to provide for additional checks of the correlations between simulated and real - world networks , we present the characteristics in all @xmath23 , @xmath24 , and @xmath25-spaces . let us note that our choice of the underlying grid to be a square lattice limits the number of nearest neighbors of a given station in @xmath23-space to @xmath26 . moreover , as far as no direct links between these neighbors occur , the clustering coefficient in @xmath23-space vanishes , @xmath27 . nonetheless , as we discuss below , both characteristics display nontrivial behavior similar to real - world networks when displayed in @xmath24 and @xmath25-spaces . for reasons explained above we choose a vanishing parameter @xmath14 and @xmath18 and for comparison @xmath28 . the data shown in the table was obtained for simulated ptns of different numbers of routes , @xmath29 and route lengths @xmath30 . in the range of parameters covered in the table we observe only weak changes of the various characteristics . natural trends are that with the increase of the number of routes @xmath5 the maximal and mean shortest path length increases in all spaces . most pronounced this is observed in @xmath23-space , while it is weakest in @xmath25-space . a similar increase is observed in @xmath23-space when increasing the number of stations @xmath6 per route . choosing the values of @xmath5 in the range @xmath31 and @xmath32 , @xmath33 the average and maximal values of the characteristics studied here are found within the ranges seen for real - world ptns , see table [ tab2 ] . more detailed information is contained in the distributions of these characteristics and their correlations . we restrict the further discussion to simulated ptns described by @xmath34 , @xmath35 , and @xmath14 , @xmath18 , which appear to reproduce many of the characteristics of real - world ptns . in figure [ fig18 ] we display the mean shortest path length distribution for these selected ptns in @xmath23 , @xmath24 , and @xmath25-spaces . in @xmath23-space we observe two groups of distributions which correspond to the two route lengths @xmath32 and @xmath33 . the most probable values for the path length @xmath36 being of the order of the corresponding @xmath6 . in @xmath24 and @xmath25-spaces the distributions are very similar with most probable path lengths @xmath37 , @xmath38 . in all cases the distributions are well fitted by the asymmetric unimodal distribution ( [ 4.3 ] ) and resemble those of the real - world networks shown in fig . varying @xmath39 does not significantly change this picture . * a * * a * let us now examine the node degree distributions of the simulated ptns selected above . as explained above , the @xmath23-space degrees are restricted by the geometry of the underlying square lattice . thus of the representations discussed here one may observe non - trivial distributions only in @xmath24 , @xmath25 , and @xmath40-spaces . [ fig19]a shows the cumulative node degree distribution in @xmath24-space in semi - logarithmic scale . recall that for the majority of real - world ptns studied in section [ iii ] as well as in other works @xcite , the @xmath24-space node degree distribution was found to decay exponentially . all distributions shown in fig . [ fig19]a display two regions each governed by an exponential decay with a separate scale . note that increasing both @xmath6 and @xmath5 leads to an increase of the ranges over which these regions extend . comparing these results with those of fig . [ fig4]b for real - world ptns we find that all ranges observed there are also reproduced here . within the parameter ranges chosen here the current model does not seem to attain a power law node degree distribution in @xmath24-space . comparing the@xmath25-space node degree distributions for real - world and simulated ptns ( figs . [ fig4]c and [ fig19]b , correspondingly ) one finds a definite tendency to an exponential behavior with two different scales in both cases . note however that for the simulated ptns the scales increase with the number of routes @xmath5 while they decrease with the number of stations per route @xmath6 . the simulated results discussed so far concerned data obtained for individual instances of modeled ptns . one of the reasons for this was to reduce the computational effort required for the calculation of path lengths , betweennesses , and related global characteristics . furthermore , in particular for the simulations involving high number of routes some self averaging may be expected to occur . the latter assumption was tested and verified by ( i ) simulating a reasonable set of ptns with the same choice of parameters and ( ii ) by performing large - scale simulations calculating local characteristics . a result of the latter procedure involving averages over up to @xmath41 instances of simulated networks is shown in fig . [ fig20]a. there we show the node degree distribution of the station nodes in @xmath40-space , i.e. the bipartite network of routes and stations with the inherent neighborhood relation ( see fig . [ fig3 ] ) . as can be seen in the double logarithmic plot shown in fig . [ fig20]a a power - law like behavior of this distribution that extends over a wider range is found for small values of the parameter @xmath9 . this corresponds to a situation where one finds many routes to proceed in parallel ( compare with the maps shown in fig . [ fig17 ] ) . for the more realistic choices of the @xmath9 parameter the overall behavior of this distribution is described by an exponential decay . the scale of this decay strongly depends on @xmath9 . [ fig20]b shows that similar distributions for the real cities have oscillating character , which is caused by the fact that non - cumulative distributions are plotted . similarly , individual distributions for simulated ptns are in general non - monotonous , however the large number average of the distribution appears to be monotonously decreasing . nevertheless , comparing plots in figs . [ fig20]a and [ fig20]b one sees that in general the model is capable to reproduce the global decay properties of the station node degree distributions in @xmath40-space . in fig . [ fig21 ] we show the betweenness - degree correlation for the simulated ptn with @xmath42 , @xmath43 , @xmath44 , @xmath14 , and @xmath18 . there , we present the mean betweenness centrality @xmath45 in @xmath25 , @xmath24 , and @xmath40-spaces . corresponding plots for a real world network are shown in fig . [ fig14 ] . plots displayed for the simulated networks in figs . [ fig21]a - [ fig21]c qualitatively reproduce the behavior of @xmath45 observed for the real world networks in @xmath25 , @xmath24 , and @xmath40-spaces . @xmath23-space behavior can not be reproduced due to the restrictions caused by the geometry of the underlying square lattice . in figs . [ fig22]a and [ fig22]b we plot the cumulative harness distributions @xmath46 for two simulated networks with @xmath47 , @xmath33 , @xmath14 and different values of parameter @xmath9 : @xmath17 ( fig . [ fig22]a ) and @xmath48 ( fig . [ fig22]b ) . similar plots for real world networks are given in figs . [ fig15 ] and [ fig16 ] . the plots of fig . [ fig22 ] nicely reproduce two regimes empirically observed for the real - world ptn . in the first , the harness distribution is governed by a power law decay ( [ 4.12a ] ) , fig . [ fig22]a , whereas in the other one there is a tendency to an exponential decay ( [ 4.13 ] ) , fig . [ fig22]b. a prominent feature demonstrated by fig . [ fig22 ] is that one can tune the decay regime by changing the parameter @xmath9 . for small values of @xmath9 the probability of a route to proceed in parallel with other routes is high c.f . ( [ 6.2 ] ) . therefore , the number of `` hubs '' in the @xmath46 distribution of lines of several routes that go in parallel is large for small @xmath9 . this is reflected by a power - law decay of the distribution . alternatively , an increase of @xmath9 leads to a decrease of such hubs as shown by the exponential decay of their distribution . for the simulated ptn with @xmath47 , @xmath33 . a * : @xmath14 , @xmath17 , fig . * b * : @xmath14 , @xmath48 . compare with plots in figs . [ fig15 ] , [ fig16 ] for the real - world networks.,title="fig:",width=170 ] for the simulated ptn with @xmath47 , @xmath33 . fig . * a * : @xmath14 , @xmath17 , fig . * b * : @xmath14 , @xmath48 . compare with plots in figs . [ fig15 ] , [ fig16 ] for the real - world networks.,title="fig:",width=170 ] * a * * b * summarizing the comparison of the statistical characteristics of real world networks with those of simulated ones one can definitely state that the model proposed above captures many essential features of real world ptns . this is especially evident if one includes into the the comparison different network representations ( different spaces ) as performed above . this paper was driven by two main objectives towards the analysis of urban public transport networks . first , we wanted to present a comprehensive survey of statistical properties of ptns based on the data for cities of so far unexplored network size ( see table [ tab1 ] ) . based on this survey , the second objective was to present a model that albeit being simple enough is capable to reproduce a majority of these properties . especially helpful in our analysis was the use of different network representations ( different spaces , introduced in section [ ii ] ) . whereas former ptn studies used some of these representations , here within a comprehensive approach we calculate ptn characteristics as they show up in @xmath23 , @xmath24 , @xmath25 , and @xmath40-spaces . it is the comparative analysis of empirical data in different spaces that enabled , in particular , an adequate ptn modeling presented in section [ vi ] . the networks under consideration appear to be strongly correlated small - world structures with high values of clustering coefficients ( especially in @xmath23 and less in @xmath25-spaces ) and comparatively low mean shortest path values , as listed in table [ tab2 ] . standard network characteristics listed there correspond to the features a passenger is interested in when using public traffic in a given city . to give several examples , any two stops in paris are on the average separated by @xmath49 stations ( with a maximal value of @xmath50 ) and to travel between them one on average should do @xmath51 changes . evidence of correlations present in ptns are the power - law node degree distributions observed for many networks in @xmath23 and for some in @xmath24-space ( see table [ tab3 ] ) . currently , we find no explanation why some of the networks of our survey are governed by power - law node degree distributions whereas others follow an exponential decay . in the analysis of urban street networks a classification has been found @xcite that allows to discriminate between properties of different classes of city organization . let us note however that as a rule the latter analysis is performed for restricted regions of street networks i.e. either the historical or the suburban part . in the case of a ptn , however , one usually deals with a structure that spreads over all the city , covering both the inner and outer regions . besides looking on traditional network characteristics ( as described in sections [ iii ] - [ v ] ) we addressed here a specific feature which is unique for ptns and networks with similar construction principles . namely , we analyzed statistical distributions of public transport routes that go in parallel for a sequence of stations . as we have shown such distributions ( we call them harness distributions ) are well defined for the networks under consideration and may be also be used for a quantitative description of similar networks embedded in 2d or 3d space as cables , pipes , neurons , or ( blood- ) vessels , etc . the common statistical features of the networks considered emerge due to their common functional purposes and construction principles also reflected in the underlying bipartite structure @xcite . it is this structure that explains parts of the correlations present in ptns @xcite . the network growth model we present in section [ vi ] captures this structure describing network evolution in terms of adding public transport routes , each of them being a complete graph in @xmath24-space . our choice to use a self avoiding walk ( saw ) as a route model in lattice simulations was motivated by geographical observations and other reasons , as argued in section [ vi ] . in support of the scaling argument given there , one may note that the fractal dimension of a saw on a lattice does not change if a weak uncorrelated disorder is present , i.e. when some lattice sites can not be visited @xcite . in turn , this tells that the model is robust with respect to weak disturbances of the underlying lattice structure . further analysis of simulated ptns performed in section [ vi ] established strong similarities in the statistical characteristics of simulated and real - world networks . obviously , the two objectives in the ptn study we have so far achieved in this paper - the empirical analysis and the modeling - naturally call for an analytic approach . in particular , such approach may be used in parallel with numerical simulations to derive statistical properties of the model proposed in section [ vi ] . this will be a task for forthcoming studies . another natural continuation of this work will be to analyze different possibly dynamic phenomena that may occur on and with ptns . a particular task will be to study robustness of ptns to targeted attacks and random failures @xcite . 50 r. albert and a .- barabsi , rev . * 74 * , 47 ( 2002 ) . s. n. dorogovtsev and j. f. f. mendes , adv . phys . * 51 * , 1079 ( 2002 ) . m. e. j. newman , siam review * 45 * , 167 ( 2003 ) . s. n. dorogovtsev and s. n. mendes , _ evolution of networks _ ( oxford university press , oxford , 2003 ) . holovatch , o. olemskoi , c. von ferber , t. holovatch , o. mryglod , i. olemskoi , and v. palchykov , j.phys . stud . * 10 * , 247 ( 2006 ) . l. a. n. amaral , a. scala , m. barthlmy , and h. e. stanley , proc . , * 97 * , 11149 ( 2000 ) . r. guimera and l. a. n. amaral , eur j. b * 38 * , 381 ( 2004 ) . r. guimera , s. mossa , a. turtschi , and l.a.n . amaral , proc . usa * 102 * , 7794 ( 2005 ) . a. barrat , m. barthlemy , r. pastor - satorras , and a. vespignani , proc . usa * 101 * , 3747 ( 2004 ) . chi , r. wang , h. su , x .- xu , j .- s . zhao , w. li , and x. cai , chin . . lett . * 20 * , 1393 ( 2003 ) . y. he , x. zhu , and d .- b * 18 * , 2595 ( 2004 ) . w. li and x. cai , phys . e * 69 * , 046106 ( 2004 ) . w. li , q. a. wang , l. nivanen , and a. le mhaut , physica a * 368 * , 262 ( 2006 ) . p. sen , s. dasgupta , a. chatterjee , p. a. sreeram , g. mukherjee , and s. s. manna , phys . e * 67 * , 036106 ( 2003 ) . p. crucitti , v. latora , and m. marchiori , physica a * 338 * , 92 ( 2004 ) . r. albert , i. albert , and g. l. nakarado , phys . e * 69 * , 025103 ( 2004 ) . m. marchiori and v. latora , physica a * 285 * , 539 ( 2000 ) . v. latora and m. marchiori , phys . 87 * , 198701 ( 2001 ) . v. latora and m. marchiori , physica a * 314 * , 109 ( 2002 ) . k. a. seaton and l. m. hackett , physica a * 339 * , 635 ( 2004 ) . c. von ferber , yu . holovatch , and v. palchykov , condens . matter phys . * 8 * , 225 ( 2005 ) , e - print cond - mat/0501296 . j. sienkiewicz and j. a. holyst , phys . e * 72 * , 046127 ( 2005 ) , e - print physics/0506074 ; j. sienkiewicz and j. a. holyst , acta phys . polonica b * 36 * , 1771 ( 2005 ) . p. angeloudis and d. fisk , physica a * 367 * , 553 ( 2006 ) . zhang , k. chen , y. he , t. zhou , b .- b . su , y. jin , h. chang , y .- p . zhou , l .- c . sun , b .- h . wang , and d .- he , physica a * 360 * , 599 ( 2006 ) . c. von ferber , t. holovatch , yu . holovatch , and v. palchykov , physica a * 380 * , 585 ( 2007 ) . h. chang , b .- b . su , y .- p . zhou , and d .- he , physica a * 383 * , 687 ( 2007 ) . x. xu , j. hu , f. liu , and l. liu , physica a * 374 * , 441 ( 2007 ) . c. von ferber , t. holovatch , yu . holovatch , and v. palchykov , arxiv:0709.3203 . in _ traffic and granular flow 07_. springer ( 2008 ) ( to appear ) . c. von ferber , t. holovatch , and yu . holovatch , arxiv:0709.3206 . in _ traffic and granular flow 07_. springer ( 2008 ) ( to appear ) for links see http://www.apta.com . some numbers for the real - world ptns slightly differ from those from our letter @xcite . the reason is an improvement of the database . h. a. simon , biometrica * 42 * , 425 ( 1955 ) . d. de s. price , j. amer . * 27 * , 292 ( 1976 ) . barabsi and r. albert , science * 286 * , 509 ( 1999 ) ; a .- barabsi , r. albert , and h. jeong , physica a * 272 * , 173 ( 1999 ) . r. ferrer i cancho and r. v. sol , e - print cond - mat/0111222 ; s. valverde , r. ferrer i cancho , and r. v. sol , europhys . lett . * 60 * , 512 ( 2002 ) ; r. ferrer i cancho and r. v. sol , in _ statistical mechanics of complex networks _ , edited by r. pastor - satorras , m. rubi , and a. diaz - guilera ( lecture notes in physics vol 625 , springer , berlin , 2003 ) , p. 114 . m. t. gastner and m. e. j. newman , eur . j. b * 49 * , 247 ( 2006 ) . n. mathias and v. gopal , phys . e * 63 * , 021117 ( 2001 ) . r. ferrer i cancho and r. v. sol , proc . , * 100 * , 788 ( 2003 ) ; r. ferrer i cancho , physica a , * 345 * , 275 ( 2005 ) . a. cardillo , s. scellato , v. latora , and s. porta , phys . e * 73 * , 066107 ( 2006 ) . p. erds and a. rnyi , publ . math . ( debrecen ) * 6 * , 290 ( 1959 ) ; publ . inst . hung . sci . * 5 * , 17 ( 1960 ) ; bull . * 38 * , 343 ( 1961 ) . b. bollobs , _ random graphs _ ( academic press , london , 1985 ) . j. a. holyst , j. sienkiewicz , a. fronczak , p. fronczak , and k. suchecki , phys . e * 72 * , 026108 ( 2005 ) . a. fronczak , p. fronczak , and j. a. holyst , phys . e * 68 * , 046126 ( 2003 ) . u. brandes , j. math . sociology * 25 * , 163 ( 2001 ) . g. sabidussi , psychometrika * 31 * , 581 ( 1966 ) . p. hage and f. harary , social networks * 17 * , 57 ( 1995 ) . a. shimbel , bull . biophys . * 15 * , 501 ( 1953 ) . l. c. freeman , sociometry * 40 * , 35 ( 1977 ) . goh , b. kahng , and d. kim , phys . * 87 * , 278701 ( 2001 ) . m. e. j. newman , phys . lett . * 89 * , 208701 ( 2002 ) . m. e. j. newman , phys . e * 67 * , 026126 ( 2003 ) . m. e. j. newman , s. h. strogatz , and d. j. watts , phys . e * 64 * , 026118 ( 2001 ) . z. liu , y .- c . lai , n. ye , and p. dasgupta , phys . a * 303 * , 337 ( 2002 ) . newman , phys . e * 64 * , 016131 ( 2001 ) . x. li and g. chen , physica a * 328 * , 274 ( 2003 ) . ramasco , s.n . dorogovtsev , and r. pastor - satorras , phys . e * 70 * , 036106 ( 2004 ) . guillaume and m. latapy , physica a * 371 * , 795 ( 2006 ) . b. nienhuis , phys . lett . * 49 * , 1062 ( 1982 ) . d. volchenkov and ph . blanchard , phys . e * 75 * , 026104 ( 2007 ) ; d. volchenkov , condens . matter phys . ( 2008 ) , to appear . harris , z. phys . b * 49 * , 347 ( 1983 ) ; y. kim , j. phys . c * 16 * , 1345 ( 1983 ) ; v. blavatska , c. von ferber , and yu . holovatch , phys . e , * 64 * , 041102 ( 2001 ) ; c. von ferber , v. blavatska , r. folk , and yu . holovatch , phys . e * 70 * , 035104(r ) ( 2004 ) .	we use complex network concepts to analyze statistical properties of urban public transport networks ( ptn ) . to this end , we present a comprehensive survey of the statistical properties of ptns based on the data of fourteen cities of so far unexplored network size . especially helpful in our analysis are different network representations . within a comprehensive approach we calculate ptn characteristics in all of these representations and perform a comparative analysis . the standard network characteristics obtained in this way often correspond to features that are of practical importance to a passenger using public traffic in a given city . specific features are addressed that are unique to ptns and networks with similar transport functions ( such as networks of neurons , cables , pipes , vessels embedded in 2d or 3d space ) . based on the empirical survey , we propose a model that albeit being simple enough is capable of reproducing many of the identified ptn properties . a central ingredient of this model is a growth dynamics in terms of routes represented by self - avoiding walks .
the fact that quantum physics contradicts local - realism is one of the most seminal discoveries . in his pioneering work @xcite , john s. bell showed that the statistical behavior of a bipartite qubit system is irreconcilable with any local - realistic theory when the system is in a singlet state . he revealed that for such theories correlations are bounded by constraints which , however , can be violated if the setup is handled quantum physically . since then much effort has been made to fully understand the origins and conditions that allow to demonstrate and experimentally test this contradiction known as quantum nonlocality . nowadays , a variety of so - called bell inequalities is known . recent bell inequalities also cover systems with more than two degrees of freedom @xcite and/or more than two parties @xcite . yet , the relation between entanglement and nonlocality is not completely clarified apart from the fact that bell inequality violations are a consequence of the presence of entanglement @xcite . the existence of entangled states that do not violate any bell inequality @xcite has raised several questions . first , how far can bell inequalities be improved and is it possible for a given system to systematically derive the most powerful inequality or a set of those ( see @xcite for recent approaches ) ? second , which entangled states allow a local realistic description even if locc ( local operators and classical communication ) and povms ( positive operator valued measurements ) are used and what are their characteristics ? besides that , a connection between the violation of a bell inequality and the security of quantum cryptography was shown , attesting to the importance of further investigations @xcite . a nontrivial connection between nonlocality and entanglement was found by investigating the collins gisin linden massar popescu bell ( cglmp ) inequality @xcite . in particular , it was shown that maximal violation is achieved with non - maximally entangled states @xcite . similar features were also found in @xcite for neutral k mesons , which are two level systems oscillating and decaying in time . for this system entangled in strangeness , though having only two degrees of freedom one finds no violation of the chsh bell inequality if the system is prepared in the spin singlet state copiously produced in accelerator experiments @xcite . a violation is only achieved if a non maximally entangled state is used . furthermore , a relation between bell inequality violation and the violation of the @xmath1 symmetry ( @xmath2 - charge conjugation , @xmath3 - parity ) in particle physics was established @xcite . ( it should be noted that this particular system in high energy physics is considerably different from systems not being affected by decay processes ) . another example demonstrating the fundamental difference between entanglement and nonlocality is given in @xcite . here , it is shown that the simulation of non - maximally entangled states via so - called non - local boxes ( or popescu - rohrlich boxes ) requires more resources than the simulation of a bell state . in the present work we aim to study nonlocality and entanglement in the context of state space geometry . that is , we consider these properties using descriptive real vector representations for density matrices . our main motivation is the fact that geometric considerations of the state space can provide deeper insights into quantum physics . they are also ideal to obtain an impression of the volumes related to certain properties or the strength of particular criteria . in case of a single qubit a geometric representation is given by the well - known three dimensional bloch - vector where the state space is called the bloch - sphere . regarding multipartite systems one usually tries to find representations where local properties and correlations can be separated in different quantities in order to avoid unwanted complexity caused by high - dimensionality . hence , when it comes to studying entanglement and nonlocality of a certain system it suffices to investigate representatives of equivalence classes of states which are equivalent with respect to local unitary transformations . in this way , for bipartite qubits it was found that correlations can be summarized in a three dimensional vector lying within a regular tetrahedron @xcite . for the investigation of bipartite qudit systems we utilize a proposed generalization of this tetrahedron a regular simplex spanned by mutually orthogonal maximally entangled states @xcite . our goal is to visualize and compare the geometries of the boundaries of entanglement and nonlocality of low - dimensional subsections of this set of states . in order to do so , we choose certain classes of states for which via optimal entanglement witnesses we are able to exactly determine the boundary between separable and entangled states . for the same classes we then reveal all states that violate the cglmp bell inequality . determining , for a given state , whether it violates or obeys a certain bell inequality is a high dimensional nonlinear optimization problem which is difficult to solve even for low dimensional quantum systems . in sec . [ nonlocalstates ] we show how this problem can be solved utilizing an advantageous parameterization of the unitary group @xmath0 and robust numerical methods . our main results concerning the geometry of nonlocality and entanglement are given in sec . [ vischsh ] for bipartite qubits and in sec . [ visqutrits ] for bipartite qutrits . in sec . [ measure ] we extend our geometric considerations by incorporating a comparison between bell inequality violation and an entanglement measure . our results show that both quantities are so intrinsically different that not even for mixtures of states sharing the same amount of entanglement the relation is monotone . we begin by reviewing the foundations of bell inequalities . any local - realistic description of a bipartite system with @xmath4 outcomes and @xmath5 measurement apparatuses on each side implies the existence of @xmath6 probabilities @xmath7 with normalization @xmath8 determining the statistics of the observables @xmath9 which can take on the values @xmath10 @xcite . as it is known , probability distributions of this form can not reproduce the statistical behavior of composite quantum systems @xmath11 when the system is in a certain state and the observables are chosen appropriately . in order to reveal this , one constructs quantities @xmath12 that consist of linear combinations of joint probabilities @xmath13 and shows that they are bounded , i.e. @xmath14 if ( [ probdist ] ) and ( [ norm ] ) are assumed . if this bound can be exceeded when the system is treated quantum physically then the established inequality ( [ bellineq ] ) is called a bell inequality . states that can not be described by a local - realistic theory are said to be nonlocal . one way of detecting nonlocality of a certain state @xmath15 is to show that it violates a bell inequality . the set of nonlocal states is complementary to the set of local states whose elements allow probability distributions of the form ( [ probdist ] ) with ( [ norm ] ) for any number of apparatuses @xmath5 . it should be noted that proving locality of a state is generally difficult since , in principle , it has to be shown that any bell inequality is satisfied or an explicit local realistic description has to be found . unfortunately , also the problem of determining whether a particular bell inequality is violated or obeyed for a given entangled state is nontrivial . in detail , to show that a certain state @xmath15 possesses nonlocal correlations one has to consider quantum mechanical joint probabilities @xmath16 where @xmath17 and @xmath18 are orthonormal eigenvectors of the corresponding observables of alice @xmath19 and bob @xmath20 . these probabilities are then inserted into the quantity @xmath12 corresponding to a particular bell inequality ( [ bellineq ] ) which can then be rewritten in the form @xmath21 wherein @xmath22 is called the bell operator @xcite . in order to examine if the inequality is preserved or violated for a given @xmath15 one has find observables that yield the maximum of @xmath12 , i.e. @xmath23 this is in general a high - dimensional nonlinear constrained optimization problem for which analytic solutions are only known for a few special cases @xcite . most often , the quantity @xmath12 has to be maximized numerically for each given @xmath15 . numerical tractability and reliability is closely related to the formulation and parameterization of the problem . our problem can be brought into a computationally beneficial form in the following way : the outcome of @xmath12 depends on the choice of the observables @xmath24 and @xmath25 , i.e. the orthonormal bases @xmath26 and @xmath27 . since all orthonormal bases are local - unitarily related , our problem is equivalent to determining a set of unitary transformations @xmath28 and @xmath29 that maximizes @xmath12 . it is unknown whether or not there exists a restrictive set of unitaries that in all cases contains the global maximum : unbiased multiport beam splitters @xcite containing only few parameters were shown to be too restrictive in general @xcite . consequently , we have to take into account all possible unitary transformations , i.e. the unitary group @xmath0 . regarding our problem we utilize the recently introduced composite parameterization of @xmath0 @xcite allowing an efficient variation . this parameterization is composed of elementary two - dimensional rotations and one - dimensional phase transformations . explicitly , using one - dimensional projectors @xmath30 and generalized anti - symmetric @xmath31-matrices @xmath32 constructed with orthonormal basis vectors any unitary operator can be written as ] @xmath33 \cdot \left [ \prod_{l=0}^{d-1 } \mbox{exp}(i p_l \lambda_{l , l})\right ] \ .\end{aligned}\ ] ] the contained @xmath34 real parameters @xmath35 lie within the ranges @xmath36 $ ] for @xmath37 and @xmath38 $ ] for @xmath39 . gathered in a @xmath40 matrix @xmath41 this parameters are to be interpreted as follows : a diagonal entry @xmath42 represents a global phase transformation for the vector @xmath43 , while an off - diagonal entry @xmath35 represents an operation in the two - dimensional subspace spanned by @xmath43 and @xmath44 : an upper right entry @xmath35 is related to a rotation and the corresponding lower left entry @xmath45 performs a relative phase shift . in our case we can exploit this structure to conveniently eliminate the @xmath4 physically irrelevant phases that are related to bases in @xmath46 . that is , starting from the basis that was used to define @xmath47 for the observables , one can remove the right part of the parameterization without discarding any solution , i.e. it suffices consider the transformations @xmath48 \ .\end{aligned}\ ] ] optimal values for the parameters can now be determined via search algorithms such as differential evolution @xcite , simulated annealing @xcite or the nelder - mead method @xcite . with respect to this , the composite parameterization has various advantages compared to several other parameterizations . for instance , varying over @xmath0 by using arbitrary hermitian operators @xmath49 and computing @xmath50 as done in @xcite is computationally expensive @xcite ; thus , inappropriate for high - dimensional problems . computing @xmath47 however , is simple since it is only a product of the matrices @xmath51 , which explicitly read @xmath52 also the feature that the product @xmath53 leaves the subspace spanned by the basis vectors @xmath54 invariant for a fixed @xmath55 ( see @xcite ) improves the speed of the computation for large @xmath4 . efficiently , we multiply matrices that are smaller than @xmath40 and in each step @xmath56 the dimension is increased by one . another reason to use this particular parameterization besides its computational benefits is the fact that the parameters can directly be related to experimental setups , that is the upper right entries of ( [ parametermatrix ] ) correspond to settings of beam splitters and the lower left entries to phase shifters . a relevant bell inequality for bipartite systems of dimension @xmath40 is the collins gisin linden massar popescu bell inequality @xcite . for @xmath57 observables on each side @xmath58 and @xmath59 it was shown that @xmath60\\ -\bigl[&p(a_1=b_1-k-1)+p(b_1=a_2-k)\\ + & p(a_2=b_2-k-1)+p(b_2=a_1-k-1)\bigr]\bigr\}\end{aligned}\ ] ] with @xmath61 is bounded by @xmath62 for local - realistic theories . this inequality can be seen as a generalization of the well known chsh bell inequality @xcite since for @xmath63 they are equivalent . the cglmp - bell inequality was proven to correspond to facets of local - realistic correlations implied by ( [ probdist ] ) and ( [ norm ] ) for @xmath57 observables and therefore belongs to the class of tight bell inequalities @xcite . some known results on the cglmp inequality can be used to test the power and reliability of our proposed algorithm . as shown in @xcite with a supposed set of optimal observables one can attain the values @xmath64 - 1}\left(1-\frac{2k}{d-1}\right)\left(\frac{1}{\sin^2\left(\frac{\pi}{d}(k+\frac{1}{4})\right)}-\frac{1}{\sin^2\left(\frac{-\pi}{d}(k+\frac{3}{4})\right)}\right ) \end{aligned}\ ] ] for the maximally entangled state @xmath65 . by maximizing @xmath66 over all observables , i.e. the @xmath67 involved parameters @xmath35 as described in the previous section using the nelder - mead method we have been able to confirm these values up to dimension @xmath68 with an accuracy of @xmath69 . thus , the maximum of @xmath66 increases with dimension @xmath4 and lies within @xmath70 for @xmath63 and @xmath71 which is reached for large @xmath4 . for all investigated dimensions the algorithm has required only a few runs to find the value given by ( [ maxvalue ] ) due to the robustness of the nelder - mead method . similar approaches utilizing other parameterization also confirm ( [ maxvalue ] ) , however , since they are computationally less tractable they do not reach @xmath68 but only @xmath72 as in @xcite or @xmath73 as in @xcite . note that there are also techniques for deriving upper bounds on @xmath66 @xcite using semi - definite programming @xcite which can be used for small @xmath4 to prove that the found values are indeed global maxima . by maximizing the largest eigenvalue of @xmath74 we could also confirm the results in @xcite , namely that for @xmath75 the maximal violation of the cglmp - inequality is attained with non maximally entangled states . now , we come to the main issue of this paper , namely comparing entanglement and quantum nonlocality using geometric representations of the state space . for a single qubit such a representation is given by the three - dimensional bloch vector ( for generalized bloch vectors see @xcite ) . in a similar way , any density matrix of a bipartite qubit system can be written as @xmath76 with two three - dimensional real vectors @xmath77 and @xmath78 related the local properties and a real @xmath79 matrix @xmath80 related to correlations . with regard to studying entanglement and nonlocality , we are not interested in the local properties of the system . hence , local unitary transformations @xmath81 can be used to diagonalize the matrix @xmath80 ( for more details on this and the following steps see @xcite ) . the diagonal entries can be regarded as components of a three - dimensional real vector @xmath82 . due to the non - negativity condition @xmath83 the components of this vector must obey the inequalities and @xmath84 . ] @xmath85 these constraints restrict @xmath82 to lie within a regular tetrahedron spanned by the projectors of the four well - known bell states @xmath86 and @xmath87 . this tetrahedron represents the core of the state space we are interested in and for it , we determine the regions that are entangled or nonlocal , respectively . for @xmath63 it was proven that the ppt ( positive under partial transposition @xcite ) criterion is necessary and sufficient , i.e. iff all eigenvalues of a partially transposed density matrix are non - negative then the state is separable ( for @xmath75 it is only a necessary criterion ) . by means of this criterion one finds that all states outside the octahedron spanned by the points @xmath88 and @xmath89 are entangled . for @xmath90 and @xmath84 this is necessary and sufficient for entanglement . now , we complete this geometric picture introduced in @xcite by adding the geometry of quantum nonlocality . states of the tetrahedron that violate the chsh bell inequality ( cglmp inequality for @xmath63 ) can be determined analytically . according to the criterion given in @xcite the maximal attainable value for @xmath91 is @xmath92 , where @xmath93 and @xmath94 are the two largest eigenvalues of the matrix @xmath95 having the components @xmath96 . consequently , since in our case @xmath80 is a diagonal matrix with the entries @xmath97 and @xmath98 the bell inequality is violated iff at least one of the three inequalities @xmath99 holds . this means that each of the inequalities defines a cylinder with radius @xmath100 and if a density matrix lies outside of one it is nonlocal . the union of the exterior regions of these three cylinders belongs to the set of nonlocal states . the tetrahedron including the boundaries of separability and chsh violation is visualized in fig . [ tetranl ] . it descriptively demonstrates that not all entangled states violate the chsh inequality . note that it is also known that all entangled bipartite qubit states can be distilled and therefore do contain so - called hidden nonlocality @xcite . ( 350 , 420)(0,0 ) ( 0,0 ) ( 0,20 ) ( 0,0 ) [ ] @xmath101 ( 320,170 ) ( 0,0 ) [ ] @xmath102 ( 350,400 ) ( 0,0 ) [ ] @xmath103 ( -10,410 ) ( 0,0 ) [ ] @xmath104 a generalization of the tetrahedron for bipartite qudit systems for arbitrary dimension @xmath4 was introduced in @xcite . analogously to the qubit case , a regular simplex can be constructed using mutually orthogonal generalized bell states . this so - called `` magic simplex '' is given by the set of states @xmath106 where @xmath107 are the projectors of @xmath34 bell - type states generated by applying the weyl operators @xmath108 with @xmath109 on the maximally entangled state @xmath110 i.e. @xmath111 the simplex is a convex set located in a @xmath112 dimensional hyperplane in a @xmath34 dimensional real vector space of hermitian operators spanned by the operators @xmath113 . unlike as for two qubits , where every state @xmath15 has a representative in the tetrahedron , for @xmath75 not every @xmath15 on @xmath114 can be related to an element of the simplex @xcite . however , the derived class of states is of special importance in quantum key distribution @xcite and entanglement distillation protocols @xcite . moreover , these states are frequently studied in the context of ( bound ) entanglement @xcite and entanglement measures @xcite due their interesting features . a detailed discussion on the properties of @xmath105 and on how it is embedded in the state space can be found in @xcite . for the subsequent investigations let us note that local ( anti-)unitary transformations @xmath115 mapping @xmath105 onto itself are related to symmetries of @xmath105 , since equivalence classes @xmath116=\ { \rho ' \in \mathcal{w } \| \rho'=u_a\otimes u_b\ ; \rho\ ; u_a^{\dagger } \otimes u_b^{\dagger}\}\end{aligned}\ ] ] share the same properties with respect to separability and ( non-)locality . the set of all symmetries can be represented with a discrete classical phase space , see fig . [ ps ] . ( 182.42 , 154.54)(0,0 ) ( 0,0 ) of the simplex @xmath105 . all possible complete lines through the point @xmath117 for @xmath118 are drawn , representing one class of states which have the same geometry concerning separability and ( non-)locality . lines can be completed by points with @xmath119 because of the periodicity of the weyl operators implying @xmath120 for all @xmath121.,title="fig : " ] ( 152.18,35.38 ) ( 0,0 ) [ ] l - position ( 28.42,115.49 ) as shown in @xcite , states with components @xmath122 and @xmath123 , respectively , belong to the same equivalence class iff there exists a phase space transformation of the form @xmath124 with @xmath125 or @xmath126 such that @xmath127 with @xmath128 . quantum nonlocality has so far not been investigated for the magic simplex @xmath105 . in the present paper we aim to determine the states of the magic simplex @xmath105 that violate the cglmp bell inequality by using the novel approach introduced in sec . [ nonlocalstates ] . in particular , we want to exactly specify the implied boundaries , i.e. states @xmath15 that obey @xmath129 for arbitrary families of states this is in general difficult since it has to be done iteratively in small regions until a required precision is reached . however , for the considered classes of states one can exploit that the trace of the cglmp bell operator vanishes , i.e. @xmath130 which is a consequence of the specific form of the cglmp bell inequality : the quantity @xmath66 ( [ cglmpineq ] ) is composed of the probabilities @xmath61 and when it is rewritten as @xmath131 every @xmath132 corresponds to a one - dimensional projector @xmath133 which has trace @xmath100 . since there are equally many projectors with positive and negative prefactors in every term of the sum in ( [ cglmpineq ] ) and due to the linearity of the trace it follows that @xmath130 . consequently , we can exploit that @xmath134 ) \nonumber \\ & = \max_{\{\mathcal{b}_{i_d}\}}tr(\mathcal{b}_{i_d}\nu\ ; \tau ) \nonumber \\ & = \nu \max_{\{\mathcal{b}_{i_d}\}}tr(\mathcal{b}_{i_d}\ ; \tau ) \label{addnoise } \ , \end{aligned}\ ] ] which implies that if the maximal value of @xmath135 is known for a given state @xmath136 then the state @xmath137 with added uncolored noise @xmath138 lies on the boundary for @xmath139 . in order to compare our new results on the geometry of quantum nonlocality with the geometry of entanglement we use several results and techniques of precedent publications on the magic simplex @xmath105 . for a detailed discussion of how to decide separability for the simplex states via the ppt criterion , matrix realignment , optimal entanglement witnesses and entanglement measures we refer the reader to @xcite . in the following we investigate and illustrate low - dimensional sections of the @xmath105-simplex for bipartite qutrits ( @xmath118 ) . as a first example consider the so - called isotropic states @xmath140 the phase space transformation rules ( [ phasetrans ] ) imply that all such one parameter states with arbitrary @xmath141 but same @xmath142 have the same properties in terms of separability / entanglement and ( non-)locality . any state of this form is separable for @xmath143 and entangled for @xmath144 ( see @xcite and references therein ) . using ( [ addnoise ] ) and the compliance of our numerical results with ( [ maxvalue ] ) we find that the cglmp bell inequality is violated for @xmath145 . next , consider the two - parameter families of states of the form @xmath146 with arbitrary @xmath147 . any such state is local - unitarily equivalent to the state @xmath148 . for this set of states the ppt boundary reads @xmath149 which was derived by setting the eigenvalues of the partially transposed matrix @xmath150 to zero . there are also bound entangled states in this set . they can be found through optimal entanglement witnesses yielding the boundaries @xcite @xmath151 the boundary of the cglmp bell violation was obtained by computing @xmath152 for @xmath153 equally spaced points on the boundaries of positivity ( @xmath154 , @xmath155 and @xmath156 ) . again , ( [ addnoise ] ) was exploited to determine the values of @xmath142 and @xmath157 corresponding to states on the boundary , i.e. @xmath158 . the result is graphically illustrated in fig . the illustration suggests that the boundary of cglmp bell inequality violation describes a circle for @xmath159 and a line if one of the parameters is negative , i.e. @xmath160 or @xmath161 . note that this abrupt change in the shape only appears for the boundary of cglmp bell inequality violation but not for the boundary of separability . . all states lie within the ( green ) triangle which corresponds to the border of positivity . the ( blue ) ellipse corresponds to the ppt border , i.e. all ppt states lie inside . as shown in @xcite if one parameter is negative , there is also a small ( grey filled ) region of bound entanglement . states @xmath15 in the ( red ) filled area violate the cglmp bell inequality @xmath162 . interestingly , the geometry given by the cglmp bell operator changes its shape ( from a circle to a line ) at the transition from positive to negative parameters . ] ( 0 , 0)(0,0 ) ( -10,50 ) ( 0,0 ) [ ] @xmath142 ( -280,340 ) ( 0,0 ) [ ] @xmath157 conjectures on the exact specifications of these boundaries are contained in the considerations of the three - parameter families of states @xmath163 as the special case @xmath164 . as discussed in @xcite , depending on whether the index pairs @xmath165 form a line or not ( in the sense of the discrete classical phase space , fig . 2 ) , this three - parameter family of states is either local - unitarily equivalent to the state @xmath166 or @xmath167 the class of line states are graphically illustrated in fig . [ linenl ] , and a particular slice of the off - line states is visualized in fig . [ p00p01p10 ] ( b ) . , eq . ( [ line ] ) . as for bipartite qubit states , fig . [ tetranl ] , all states have to be within a ( green ) tetrahedron ( positivity condition ) . the boundary of ppt states forms a ( blue ) cone and thus all states beyond this surface are entangled . the tip of the cone touches the surface of positivity at @xmath168 . states @xmath15 beyond the shaded ( light red / blue ) surface violate the cglmp bell inequality @xmath169 . again , the geometric form changes from a sphere to a plane when one or more parameters become negative . ] ( 0 , 0)(0,0 ) ( -8.8,3.3 ) ( 0,0 ) [ ] @xmath157 ( -5.8,8.3 ) ( 0,0 ) [ ] @xmath170 ( -0.2,3.0 ) ( 0,0 ) [ ] @xmath142 for the states @xmath171 the ppt boundary is @xmath172 the boundary of separability obtained by optimal entanglement witnesses revealing further ( bound ) entangled states is given by @xcite @xmath173 and the equations one gets via permutations with respect to @xmath174 . the boundary of cglmp bell violation is deduced via ( [ addnoise ] ) from the values @xmath152 of @xmath175 equally spaced states on the boundaries of positivity ( @xmath176 ( and all parameter permutations of this term ) and @xmath177 ) . the resulting points apparently describe a spherical surface which meets the boundaries of positivity and ppt at the point @xmath168 . this sphere appears to be intersected by planes in the vicinity of the isotropic states , fig . [ linenl ] . when taking into account symmetries and the fact that ( [ maxvalue ] ) yields the boundary parameter @xmath178 for the isotropic states one can suppose a sphere with radius @xmath179 and center at @xmath180 . these specifications coincide with the numerical data up to the order @xmath69 . the planes are given by the function @xmath181 and the two additional equations given by permutations with respect to @xmath182 and @xmath170 . in contrast to the specifications of the sphere , these functions are not conjectured but result from the measurement settings given in @xcite and local unitary transformations constituted by the weyl operators . to show this one can use the local - unitarily equivalent state @xmath183 and write @xmath184 . and @xmath185 differ from that of @xmath186 . different states , even though they are local - unitarily equivalent require different measurement settings . ] here , the measurement bases in ref . @xcite yield @xmath187 and @xmath188 . thus , by setting @xmath189 we obtain ( [ plane ] ) up to a parameter permutation , which in practice can be realized using a local - unitarily modified bell operator @xmath190 since @xmath191 . for the remaining family of off - line states , eq . ( [ offline ] ) , the geometric form of the boundary of cglmp bell violation has a more complex shape . an uncovered suggestion on the exact form has therefore not been made . [ p00p01p10 ] ( b ) gives an impression of the fact that the boundaries of ppt , as well as cglmp bell violation have a different shape for the different types of states . for instance , in contrast to the class of line states where there is only one state @xmath168 on the boundary of positivity @xmath192 that does not violate the cglmp bell inequality , for the second type @xmath193 there is a whole region of states for @xmath192 that obeys the cglmp bell inequality . moreover , in comparison the entire region of nonlocal states in fig . [ p00p01p10 ] ( b ) is smaller than in fig . [ p00p01p10 ] ( a ) . for entanglement , one finds the opposite , which is a intriguing result . it is rather counterintuitive because the region of separable states of the subset @xmath194 already is larger than the ppt region of the same subset of off - line states @xmath195 which in addition also contains bound entanglement for @xmath196 . ( a ) ( b ) ( 0 , 0)(0,0 ) ( -8.0,2.2 ) ( 0,0 ) [ ] @xmath142 ( -0.2,2.2 ) ( 0,0 ) [ ] @xmath142 ( -13.0,7.5 ) ( 0,0 ) [ ] @xmath157 ( -5.5,7.5 ) ( 0,0 ) [ ] @xmath157 in @xcite , an entanglement measure for multipartite qudit systems was introduced , as well as a method how optimal bounds can be derived for it . here we apply this measure to bipartite qutrits . for any bipartite qutrit state @xmath197 , entanglement can be quantified via @xmath198 where @xmath199 is any renyi entropy . in our case we use the linear entropy @xmath200 . in general the infimum , i.e. the optimal decomposition is not known . the derivation of lower bounds on the measure are based on the observation that the linear entropy can be rewritten as @xcite @xmath201 using the anti - symmetric @xmath31-matrices ( [ sigmamatrices ] ) . @xmath202 is called the @xmath55concurrence and is an entanglement measure for pure states in its own right . the m - concurrence is generalized to mixed states via the convex roof construction ( [ measuredef ] ) to @xmath203 this expression has the lower bound @xmath204 for which the contained individual infima are known ( see @xcite ) @xmath205 with @xmath206-\sum_ix^i_{k_a , l_a , k_b , l_b},0]\end{aligned}\ ] ] where @xmath207 are the square roots of the eigenvalues of @xmath208 the derived bound depends on in which basis the state @xmath15 is written and is thus not invariant under local unitaries @xcite . consequently , the optimal lower bound for a state @xmath15 is given by the maximum over all @xmath209 . this problem can be reduced to a @xmath210-parameter optimization problem analogously to sec . note that there are also algorithms to numerically derive tight upper bounds on the convex roof @xcite , thus we know for which classes of states the exact value of the measure is obtained . -concurrence @xmath212 with bell inequality violation given by @xmath213 for the line states @xmath214 . the black lines parallel to the boundaries of separability correspond to the amount of entanglement , while the red curves correspond to the degree of bell inequality violation . the fact that both quantities can not be monotonically related can easily be seen by walking along one of red lines on the right side , i.e. @xmath215 . for instance , starting from one of the points where a red curve intersects the upper green boundary of positivity , the amount of entanglement increases first until @xmath216 but when @xmath157 becomes negative it decreases again . ] ( 0 , 0)(0,0 ) ( -0.2,2.3 ) ( 0,0 ) [ ] @xmath142 ( -4.9,6.8 ) ( 0,0 ) [ ] @xmath157 now , consider the class of line states with two equally weighted bell states according to ( [ eqmixed ] ) . we evaluated the introduced lower and upper bounds on a grid of @xmath217 points with a step size of @xmath218 and found that they coincide ( up to numerical precision ) for this particular class of states . the numerical data can be represented with an accuracy of @xmath69 by the expressions @xmath219 for the @xmath55-concurrence @xmath220 . the contour plot fig . [ measurecplot ] shows that all states lying on lines parallel to the boundaries of separability contain the same amount of entanglement . in comparison to this , the strength of cglmp - bell inequality violation is not related to the distance to the separable states . [ measurecplot ] is an illustrative example that bell inequality violations and the entanglement measures are non - monotonically related . another example is illustrated in fig . [ picturemeasure ] where the amount of entanglement of states on the border of cglmp bell inequality violation ( @xmath221 ) and of positivity are plotted . both results demonstrate that there exists no monotone function relating the bell inequality violation to the amount of entanglement . for pure states this was already known , however , the surprising thing is that the same statement is true also for mixtures of states that all share the same amount of entanglement , i.e. @xmath113 . , eq . ( [ line ] ) , in dependence of @xmath157 ( compare with fig . [ p00p01p10 ] ( a ) ) . the solid thin ( black ) curve is the amount of entanglement for all states lying on the boundary of positivity . the other graphs show the amount of entanglement for the boundary of cglmp bell inequality violation . in detail , the dotted ( red ) graph illustrates the amount of entanglement for @xmath161 . the solid thick ( green ) graph illustrates the amount of entanglement in the range @xmath216 to the separable state @xmath222 along the curved cglmp boundary , while the dashed ( blue ) graph illustrates the region @xmath223 . ] we presented the first detailed state space analysis of a bell inequality in the context of geometry . the analysis necessitated to ascertain whether a bell inequality is violated or obeyed for a given quantum state @xmath15 . in order to solve this , we used a novel composite parameterization of the unitary group @xmath224(d ) and robust numerical methods ( sec . [ nonlocalstates ] ) . by means of this we achieved compliance with the orthonormality constraints and reduced the numerical search to @xmath225 real parameters for each involved observable . our tests for the cglmp inequality ( sec . [ sectioncglmp ] ) showed that our method works within good numerical precision and that the used paramterization is beneficial with respect to numerical tractability . in sec . [ vischsh ] we began our geometric state space considerations concerning entanglement and nonlocality . we derived a regular tetrahedron spanned by the four bell states for bipartite qubits . for this tetrahedron , the sets of entangled and chsh inequality violating states were determined with analytic methods . the graphical illustration of the tetrahedron ( fig . [ tetranl ] ) provides a descriptive example that not all entangled states violate the chsh inequality . in sec . [ wsimplex ] we considered the state space geometry for the more general case of bipartite qudits . we focused on a generalization of the tetrahedron the so - called `` magic '' simplex @xmath105 a class of states which is of special interest for quantum key distribution and entanglement distillation protocols . we gave a short review on the properties and the symmetries of @xmath105 which are related to a classical discrete phase space ( fig . [ ps ] ) . in sec . [ visqutrits ] we determined the nonlocal states of the simplex for the special case of bipartite qutrits ( @xmath118 ) using the proposed numerical method . here , we exploited a particular property of states containing uncolored noise ( [ addnoise ] ) to further reduce the numerical effort . the boundaries of cglmp bell inequality violation and separability were visualized for the class of the so - called `` line '' states and for the class of the so - called `` off - line '' states in fig . [ nl1 ] to fig . [ p00p01p10 ] . as could be seen in all cases , both boundaries are not simply related by a constant shift or scaling factor but rather are fundamentally different . in sec . [ measure ] we considered the relation between entanglement measures and bell inequality violations . we determined the exact value of the @xmath55-concurrence @xcite for a particular class of line states with two equally weighted parameters . comparing the strength of cglmp bell inequality violation with the amount of entanglement , we geometrically demonstrated that both quantities are non - monotonically related . in summary , we have seen that geometric considerations of the state space are an insightful way of studying the manifestations of entanglement . as we have shown , those approaches are suitable for revealing fundamental differences between entanglement and quantum nonlocality . + * acknowledgements : * we thank t. adaktylos , a.m. adaktylos , a. gabriel , f. hipp , g. krizek , h. schimpf , m. meyer and h. waldner for inspiring discussions and careful reading of the manuscript . c.s . and m.h . acknowledge financial support from the austrian fwf , project p21947n16 . b.c.h . acknowledges the fellowship moel 428 .	we compare entanglement with quantum nonlocality employing a geometric structure of the state space of bipartite qudits . central object is a regular simplex spanned by generalized bell states . the collins gisin linden massar popescu bell inequality is used to reveal states of this set that can not be described by local - realistic theories . optimal measurement settings necessary to ascertain nonlocality are determined by means of a recently proposed parameterization of the unitary group @xmath0 combined with robust numerical methods . the main results of this paper are descriptive geometric illustrations of the state space that emphasize the difference between entanglement and quantum nonlocality . namely , it is found that the shape of the boundaries of separability and bell inequality violation are essentially different . moreover , it is shown that also for mixtures of states sharing the same amount of entanglement , bell inequality violations and entanglement measures are non - monotonically related .
since observational evidence of star - to - star abundance inhomogeneity among the light elements of stars in the galactic globular clusters ( gcs ) was discovered ( e.g. , cohen 1978 ; peterson 1980 ; norris et al . 1981 ; leep et al . 1986 ) , the origin of the inhomogeneity has been extensively discussed both theoretically and observationally ( e.g. , sneden et al . 1992 ; norris & da costa 1995 ; smith et al . 2005 ; see gratton et al . 2004 for a recent review ) . so far the following two scenarios ( or working hypotheses ) have been proposed for the origin of the abundance inhomogeneity : ( 1 ) the primordial hypothesis and ( 2 ) the mixing hypothesis ( e.g. , sweigart & mengel 1979 ; cottrell & da costa 1981 ; freeman & norris 1981 ; smith 1987 ; suntzeff 1993 ; kraft 1994 ; denissenkov & weiss 1996 ) . the first hypothesis is that the observed inhomogeneity is due to the second generation of stars that formed from gas ejected from the first generation of evolved stars ( e.g. , agb stars ) ( `` primordial pollution '' or `` self - pollution '' scenarios , e.g. , cottrell & da costa 1981 ) . internal processes of stars , such as dredge - up of the cn - processing of envelope material in inner hydrogen - burning regions , are key for understanding the observed chemical inhomogeneity of gcs in the second mixing hypothesis ( e.g. , `` deep mixing '' ; smith 1987 ; kraft 1994 ; thoul et al . 2002 ) . although deep mixing is highly unlikely to occur in less evolved stars on the main sequence and subgiant - branch , star - to - star abundance inhomogeneity was found in these stars of some galactic gcs ( e.g. , cannon et al . 1998 for 47 tuc ) . the self - pollution scenario is accordingly suggested to be more promising than the mixing ( evolutionary ) scenario ( da costa et al . 2004 ; gratton 2004 for a recent review ) . stellar ejecta from agb stars and massive stars have been considered to play a key role in early chemical evolution of gcs for the self - pollution scenario ( e.g. , fenner et al . 2004 ; charbonnel & prantzos 2006 ; prantzos & charbonnel 2006 ; karakas et al . 2006 ; bekki & chiba 2007 ) . previous studies suggested that gcs are unlikely to retain agb ejecta owing to ram pressure stripping by warm / cold gas of the galactic halo and disk ( e.g. , frank & gisler 1976 ; smith 1996 ; gnedin et al . 2002 ) . these imply that gcs are even more unlikely to retain agb ejecta in protogalactic environments , where denser halo / disk gas can strip the agb ejecta efficiently from gcs during hydrodynamical interaction between the gas and the agb ejecta . these previous studies thus appear to suggest that self - pollution is _ not _ likely within gcs evolving in isolation , though numerical attempts have not yet been made to investigate the details of self - pollution processes within gcs . furthermore , the self - pollution scenario is suggested to have the following two problems in explaining physical properties of gcs . the first is that the observed large number fraction of cn - strong populations ( @xmath7 for ngc6752 ) , which correspond to the second generation of stars formed from agb ejecta , can not be explained by the scenario without assuming very unusual and unrealistic imfs ( e.g. , smith & norris 1982 ; dantona and caloi 2004 ) : the total mass of agb ejecta from the first generation of stars in a gc is only @xmath8% of the total mass of the gc with a canonical imf ( bekki & norris 2006 ) . the second is that the observed o@xmath9na and mg@xmath9al anticorrelations and c - depleted stars in cn - strong populations can not be reproduced quantitatively by chemical evolution models based on the self - pollution scenario ( fenner et al . this problem , however , could be due to the adopted agb models , which have some uncertainties in predicting chemical yields ( e.g. , denissenkov & herwig 2003 ; karakas & lattanzio 2003 ; ventura & dantona 2005a , b ; campbell et al . 2006 ) . the most serious problem among these is the first one the observed large fraction of stars that ( in this scenario ) formed from agb ejecta . the very large mass fraction of `` polluted '' second generation of stars in the self - pollution scenario requires that agb ejecta used for the formation of this generation of stars originate from stellar systems that had total masses much larger than those of the present gcs and have already disappeared for some reason . there could be two possible scenarios that can satisfy the above requirement . in the first place , we propose : ( i ) the gcs were initially more massive than the present - day gcs by a factor of @xmath10 , ( ii ) agb ejecta of the first generation of stars were consumed by the formation of the second generation with very high star formation efficiency , and ( iii ) most ( @xmath11% ) of the first generation of stars had been _ preferentially _ lost by some physical mechanisms such as tidal stripping due to the galactic tidal field . this scenario can be viable if enough of the agb ejecta can be converted into the second generation in less than @xmath12 yr after the completion of the first generation type ii supernovae ( sneii ) otherwise , age differences between the first and the second generation would cause a large age spread inconsistent with the observed tightness of the color - magnitude diagrams of gcs . in the second scenario , proposed by bekki ( 2006 ) : ( i ) proto - globular clouds are located in the central regions of low - mass proto - galaxies embedded in dark matter subhalos , ( ii ) agb ejecta of the host galaxy s field stars , which formed earlier and surround the clouds , goes into star formation within the clouds on a time scale of @xmath13 yr ( i.e. , until the intracluster gas is expelled by sneii ) , and ( iii ) the low - mass galaxies were destroyed by the galactic tidal fields to become the galactic halo components whereas the gcs were tidally stripped to form the galactic halo gcs . in this scenario there can be very small age differences ( @xmath14yr ) between stars with different abundances in a single gc . this pollution process would be better called `` external pollution '' rather than `` self - pollution '' , because most of agb ejecta originates from _ outside _ the gcs . although bekki & norris ( 2006 ) have recently discussed the origin of the possible helium overabundance of stellar populations of @xmath6 cen in the context of the external pollution scenario , the abundance inhomogeneities observed in `` normal '' gcs have not been extensively studied using previous chemical evolution models based on the _ external pollution _ scenario . the purpose of this paper is thus to discuss the origin of abundance inhomogeneity observed in the lighter elements ( e.g. , c , n , o , na , and al ) of gcs by using chemical evolution models based on the external pollution scenario . in particular , we discuss ( 1 ) the [ c / fe]-[n / fe ] anticorrelation , ( 2 ) cn - bimodality , ( 3 ) the o@xmath9na and mg@xmath9al anticorrelations , ( 4 ) the helium abundance , and ( 5 ) observable predictions of the models . we adopt the agb models recently developed by campbell et al . ( 2006 ) in which the third dredge up in low - mass agb stars can , in effect , be switched off and on due to the use or non - use of the ledoux criterion for convective boundaries . we mainly show the results of the models _ without _ the third dredge - up , because the models can better explain the observed abundance inhomogeneity if they are included in the present chemical evolution models . we realize that these models are ad - hoc , but they enable us to investigate the proposed formation mechanism . we consider gc formation in the central regions of low - mass , proto - galaxies embedded in dark matter halos at very high @xmath0 . the gc host galaxies can provide a deep gravitational potential that allows agb ejecta to be retained and used for star formation in proto - gc clouds . in this scenario , both ( i ) agb ejecta from field stellar populations ( i.e. , the main components of the galaxies ) and ( ii ) proto - galactic interstellar gas ( which falls into the central regions ) can be mixed with each other and subsequently used for gc formation . gas ejected from sneii with velocities of @xmath15 km s@xmath16 is assumed to be expelled from dwarfs during galaxy formation and thus irresponsible for the chemical evolution of forming gcs . after gc formation , the host galaxies merge with the proto - galaxy and are completely destroyed by the strong galactic tidal field . the field stellar components of the galaxies are dispersed into the galactic halo region to become old , metal - poor halo stars whereas gcs , which are not destroyed by the galaxy owing to their initial compactness , become the galactic halo gcs . star formation and chemical evolution are assumed to proceed until type ii supernovae ( sne ii ) formed within proto - gcs prevent further star formation in this scenario . therefore , star formation is assumed to continue for less than @xmath13 yr in the models in which stars with different masses form in a coeval way . age differences of stars are typically less than 5 myr for most models in the present study . evolution of chemical abundances depends mainly on ( 1 ) how much agb ejecta is mixed into fresh ism and consequently converted into stars and ( 2 ) the time scale of gas infall into central regions of galaxies . advantages and disadvantages of the external pollution scenario are discussed later in 4.1 . the total gas mass ( @xmath17 ) of a forming gc changes through the gas ejection of agb stars , gas infall , and star formation : @xmath18 where @xmath19 , @xmath20 , and @xmath21 are the gas ejection rate of the field agb stars , gas infall rate , and star formation rate , respectively . in order to calculate @xmath19 at each time step for the field stars ( which have a total mass of @xmath22 ) , we assume an imf in number defined as @xmath23 , where @xmath24 is the initial mass of each individual star and the slope @xmath25 corresponds to the salpeter imf . the normalization factor @xmath26 is a function of @xmath22 , @xmath27 ( lower mass cut - off ) , and @xmath28 ( upper mass cut - off ) : @xmath29 where @xmath27 and @xmath28 are set to be @xmath30 and @xmath31 , respectively . we adopt @xmath25 for all models in the present study . @xmath19 between @xmath32 and @xmath33 is accordingly described as : @xmath34 where @xmath35 and @xmath36 are masses of stars that turn off the main sequence ( thus becomes a agb star ) at the time @xmath33 and @xmath32 , respectively ( thus @xmath37 ) . @xmath38 describes the total gas mass ejected from an agb star with initial mass @xmath24 and final mass ( @xmath39 ) . we derive an analytic form of @xmath38 ( @xmath40 ) from the observational data by weidemann ( 2000 ) by using the least - square fitting method , and find : @xmath41 in order to calculate the main - sequence turn - off mass ( @xmath42 and @xmath43 ) at each time step , we use the following formula ( renzini & buzzoni 1986 ; norman & scoville 1988 ) : @xmath44 where @xmath45 is in solar units and time @xmath46 in years . we assume that the time @xmath47 corresponds to the epoch ( @xmath48 ) when the most massive agb star ( with @xmath49 ) starts to eject gas in the present models . therefore @xmath50 in the above equation ( 5 ) . the gas infall rate @xmath20 is described as @xmath51 where @xmath52 is the normalization factor for the total mass of infalling gas , @xmath53 represents the epoch of the maximum infall rate , and @xmath2 describes the time - scale of gas infall . the adopted gaussian distribution means that the gas infall rate increases , whether slowly or rapidly , depending on the adopted @xmath2 , for @xmath54 and then decreases for @xmath55 . the star formation rate @xmath21 is described as @xmath56 where @xmath57 ( a normalization constant ) is determined for a given @xmath58 so that most of the gas can be consumed within an order of @xmath13 yr . although we investigate a range of @xmath58 , we show the results for the models with @xmath59 which can better explain observations . observations ( e.g. , tightness of color - magnitude diagrams of gcs ) strongly suggest that the timescale of gc formation should be quite short ( an order of @xmath12 yr ) . we also consider that star formation is truncated owing to feedback effects of type - ii supernovae ( sne ii ) formed during gc formation : the present models with this truncation can keep abundance homogeneity in heavy elements ( e.g. , fe ) and thus can be consistent with physical properties of normal gcs in the galaxy . therefore we assume that gas can be converted into stars in gcs only for @xmath60 , where @xmath61 should be similar to or shorter than @xmath62 yr corresponding to the typical lifetimes of stars that explode as sne ii . we present the results of the representative models with @xmath63 yr and @xmath64 in the present study . the gas mass ( @xmath65 ) is given in units of @xmath22 ( the total mass of field stars ) . in the adopted imf , @xmath66 is 0.01 ( @xmath67 ) at @xmath68 for all models . the time evolution of the abundance of the @xmath69-th element is described as follows : @xmath70 where @xmath71 and @xmath72 are the @xmath69-th abundance of agb ejecta and infalling gas , respectively . @xmath71 is time - dependent , because of the varying timescales of evolution for the different masses of agb stars ( see equation ( 5 ) ) . we assume that @xmath72 is constant for each @xmath69-th element and thus refer to it simply as @xmath73 in the present study . in order to calculate @xmath71 , we use the latest chemical yield tables produced by campbell et al . ( 2006 ) in which h , he , c , n , o , na , @xmath74 mg , @xmath75 mg , @xmath76 mg , al , are listed for agb stars with the masses ranging from @xmath77 to @xmath78 for a given [ fe / h ] ( @xmath79 ) . these models follow normal agb evolution with hot bottom burning , but we produce two sets of yields - one with the normal third dredge - up and one without ( see below for details ) . the chemical yield model without ( with ) the third dredge - up are referred to as ya1 ( ya2 ) in the present study . table 1 shows the mass fraction of the above eleven species in agb ejecta from campbell et al . we consider that @xmath73 should be similar to typical values of old halo stars in the galaxy ( e.g. , goswami & prantzos 2000 ; venn et al . 2004 ) and the adopted values of @xmath73 are summarized in table 2 for four different models . the details of the models without third dredge - up are given in campbell ( 2006 ) . throughout this study , we assume that [ fe / h ] is the same between infalling gas and agb ejecta . this assumption is reasonable , because we consider that gas from sneii can be expelled from dwarf efficiently during galaxy formation . we do not include the effects of massive rotating stars on the chemical evolution of proto - gc clouds in the present study , because we intend to understand more clearly the roles of agb stars in the chemical evolution : we need to derive separately the roles of agb stars and those of massive rotating stars in the chemical evolution . we mainly show the results of the models that use the agb yields with no third dredge - up , though we investigate the models with and without third dredge - up . it should be stressed here that the models with no third dredge - up are purely hypothetical ones that have not yet been proved . previous studies based on self - pollution models with third dredge - up in agb stars failed to explain many of the observations of abundance inhomogeneity in gcs ( e.g. , fenner et al . 2004 ) . if our models can successfully reproduce some observations that were not explained by previous ones , we can obtain some insights into the `` culprits '' in the self - pollution or external pollution scenarios . the abundance inhomogeneity of gcs in the present models with no third dredge - up originates from pure hydrogen burning , which is available in hbb agb stars . the present models with third dredge - up can not explain at all most of the observed correlations between different abundances ( e.g. , the [ c / fe]-[n / fe ] anticorrelation ) . for example , these models can not produce stars with [ c / fe ] @xmath80 and [ n / fe ] @xmath81 , because [ n / fe ] increases with the increase of [ c / fe ] during the chemical evolution owing to the c - rich ejecta of agb stars . furthermore , the observed cn - bimodality of stars can not be reproduced in these models , owing to the lack of c - depleted , n - rich stars in these models . these failures would suggest that _ agb stars are not the polluters that are responsible for the observed abundance inhomogeneity in gcs_. we however consider that it is still worthwhile to investigate abundance properties of gcs by using different agb models ( i.e. , those with no third dredge - up ) in the present study for the above - mentioned reason . it should also be stressed here that the present models with no third dredge - up can better explain observations , because we adopt the star formation histories of proto - gc clouds in the external pollution scenario : successful reproduction of some observations in the present study is due to the _ combination of the adopted models both with no third dredge - up and star formation histories of the external pollution scenario_. further , at present there are no yields available for the so - called `` super - agb stars '' , which are more massive agb stars that experience thermal pulses after core carbon burning . the few calculations in existence ( e.g. , ritossa et al 1996 ; gil - pons et al 2005 ; siess 2006 ) show that these stars experience very hot hbb and very little dredge - up . qualitatively this is what is required to solve the gc problem , so our models without dredge - up may simulate to some extent the yields from these stars , while we await full evolution and nucleosynthesis calculations for the appropriate masses and compositions . we close this section with some comments about the reliability of agb star yields . we believe that the balance of evidence indicates that agb stars with dredge - up are not the site of the abundance inhomogeneities seen in globular cluster stars . however , there are still so many uncertainties that it may still be possible for these stars to produce the abundance profiles required . for example , the depth of dredge - up is still very uncertain , the temperature profile in the hbb envelope depends on the convection theory ( e.g. , ventura & dantona 2005a , b , c ) as well as associated parameters , mass - loss and of course various reaction rates . although previous studies ( ventura et al . 2002 ; ventura & dantona 2005c ) provided useful tables for stellar yields of agb stars , the lack of tables for some abundances ( e.g. , the lack of a table for al abundance in ventura & dantona 2005c ) currently does not allow us to investigate the o - na and mg - al anticorrelations in a fully self - consistent manner . it is thus our future study to compare the present results with those based on chemical evolution models using tables from works other than campbell ( 2006 ) . more comments on agb models with and without third dredge - up are given in the appendix b. one of the most important parameters that govern the chemical evolution of gcs in the external pollution scenario is the ratio ( @xmath1 ) of the total mass of the infalling gas ( @xmath82 ) to that of agb ejecta ( @xmath66 ) . we investigate the models with @xmath1 ( @xmath83 ) ranging from 0.1 to 100 and thereby isolated the best values of @xmath1 that can explain observations . the time scale of gas infall ( @xmath2 ) is also an important parameter that can control final abundance distributions of gcs in the present study . we investigate models with @xmath2 ranging from @xmath84 yr to @xmath62 yr and show important results of the models . although these @xmath1 and @xmath2 are the most important parameters , the initial gas mass @xmath85 can also change abundance distributions in a less dramatic manner for a given @xmath1 and @xmath2 . we investigate the models with @xmath86 and show the results of some representative models . table 3 summarizes the parameter values used for the 15 models investigated in the present study . we first present the results of the fiducial model m1 , which shows interesting and important results , in 3.1 . we then describe the dependences of the results on the two key parameters ( @xmath1 and @xmath2 ) in 3.2 . by changing @xmath1 , @xmath2 , and @xmath85 , we can construct a model that can explain reasonably well the abundance distribution for a specific gc . for example , the model m11 shows a bimodal [ n / fe ] distribution with the normalized number fraction of n - rich and c - depleted stars being equal to that of the n - normal and c - normal ones and thus is consistent with the observed cn distribution in ngc 6752 ( e.g. , smith & norris 1993 ) . since we are focusing on the general behaviors resulting from parameter dependences of the models , we do not discuss extensively the abundance distributions of individual gcs . we however just briefly discuss the origin of he abundance distributions of different stellar populations in @xmath6 cen in the appendix a , because this issue is quite topical and is currently being investigated extensively by different theoretical models ( e.g. , bekki & norris 2006 ; maeder & meynet 2006 ) . in order to compare the present models with previous ones ( e.g. , fenner et al . 2004 , ventura & dantona 2005c ) , we compare the present results with the same observational ones ( yong et al . 2003 for ngc 6752 ) that have been already compared also with these previous studies in 3 . we also compare the present results with other more recent observations on abundances of stars ( in particular , less evolved stars ) in gcs ( e.g. ramirez & cohen 2002 ; briley et al . 2004a , b ; carretta et al . 2005 ; cohen et al . 2005 ) and assess the viability of the external pollution scenario later in 4.1 2 shows the time evolution of the fiducial model m1 in comparison with the model m2 on the [ c / fe]-[n / fe ] plane . there are differences in initial abundances of infalling gas between models m1 and m2 ( e.g. , [ c / fe ] and [ n / fe ] ) . since these abundances are estimated for new stars formed at each time step in a gc , each location of the line represents abundances of stars with given ages rather than the mean abundances of the gc . as the proto - gc cloud is polluted by ejecta from more massive field agb stars ( @xmath87 ) , [ n / fe ] and [ c / fe ] in the cloud rapidly increases and decreases , respectively . gas infall increases the total gas mass of the cloud ( and thus the density ) and consequently increases gradually star formation in the cloud . as a result of this , new stars that formed from the gas , which is a mix of infalling gas and agb ejecta , can have higher [ n / fe ] and lower [ c / fe ] . as a larger amount of agb ejecta is accumulated in the cloud at later times , new stars formed later can have higher [ n / fe ] and lower [ c / fe ] than those formed earlier . thus stars with higher [ n / fe ] can show lower [ c / fe ] in the fiducial model m1 and this anticorrelation between [ n / fe ] and [ c / fe ] can be seen in most models . although the fiducial model can qualitatively explain the observed anticorrelation in fig . 2 ( e.g. , norris et al . 1981 ; smith & norris 1993 ) , it does not pass the observed location of the n - rich population in ngc 6752 . this quantitative inconsistency between the model and the observation can be seen in the comparative model m2 with initially small [ n / fe ] in the infalling gas that can significantly dilute [ n / fe ] during star formation within gcs . in the present model , [ n / fe ] always becomes significantly high ( @xmath88 ) _ when [ c / fe ] becomes low irrespective of the infall rate of n - poor gas . _ since only two data points ( i.e. , two stars ) shown in fig . 2 are obviously not enough to make any robust conclusions on the quantitative consistency of the present model with observations , we discuss the consistency of the model by using more data sets later in 4.1 . as a result of gradual increase ( decrease ) of [ n / fe ] ( [ c / fe ] ) due to agb pollution , new stars formed at different times can have different [ n / fe ] and [ c / fe ] and thus show abundance inhomogeneity in [ n / fe ] and [ c / fe ] . 3 , showing the normalized number distributions of stars with different [ n / fe ] and [ c / fe ] in m1 , demonstrates that the [ n / fe ] distribution has a clearer bimodality than the [ c / fe ] distribution . most n - rich stars with [ n / fe ] @xmath89 show a lower [ c / fe ] ( @xmath90 ) , as expected from chemical pollution from massive agb stars . owing to the adopted density - dependent star formation rate , the gas infall causes two peaks in the star formation rate in m1 . new stars formed during the first peak of star formation can have lower [ n / fe ] whereas those formed during the second peak can have higher [ n / fe ] due to a higher degree of agb pollution . the reason for the apparent absence of the bimodality in [ c / fe ] is that [ c / fe ] can not dramatically change within a short time scale ( @xmath13 yr ) so that new stars can show a much narrower [ c / fe ] distribution ( i.e. , the bimodality can not be evident for the adopted binning ) . the final _ mean _ [ c / fe ] and [ n / fe ] at @xmath91 yr are -0.09 and 0.85 , respectively . as the gas mass of the proto - gc cloud increases owing to accumulation of agb material in m1 , the he abundance ( y ) of the gas also increases in a monotonic fashion : the final _ mean _ y at @xmath91 yr is 0.26 whereas the final y of stars _ formed at @xmath91 yr _ is 0.33 . as a result of this , new stars formed later will have higher y than those formed earlier . since agb ejecta increases [ n / fe ] rapidly , new stars that form later and thus have higher y can show higher [ n / fe ] . 4 clearly shows a positive yet non - linear correlation between y and [ n / fe ] in new stars of gcs for m1 . this correlation can be seen in almost all models in the present study and thus is simply a result of h - burning . owing to the time - evolution of y in the gas cloud , new stars formed at different times show different y , which results in a y inhomogeneity in the gc . 5 clearly shows that as the proto - gc cloud is polluted by agb ejecta to a larger extent , newly formed stars show a larger degree of y inhomogeneity . furthermore , the location of the peak in the y distribution is also shifted to the higher values as the agb pollution proceeds : the peak value changes from 0.23 at @xmath92 myr to 0.25 at @xmath93 myr . these results imply that the time scale of gc formation is an important factor which can determine y distributions of gcs . since abundances in the present model either increase or decrease in a monotonic fashion as agb pollution proceeds , strong correlations or anticorrelations can be expected between different abundances of stars in a gc . 6 shows that y can anticorrelate with [ c / fe ] , [ o / fe ] , and [ na / fe ] , though the dependences are not necessarily linear . these correlations can be expected for any chemical evolution models of gcs using stellar yields of agb stars with no 3rd dredge - up . 7 shows the time evolution of m1 in the [ mg / fe]-[al / fe ] plane and the observations of bright red giant stars in ngc 6752 ( yong et al . 2003 ) . as a natural result of ongoing chemical pollution of the proto - gc cloud by field agb stars , the [ mg / fe ] and [ al / fe ] ratios of new stars within the gc decrease and increase , respectively , as agb pollution proceeds . although the observed [ mg / fe]-[al / fe ] anticorrelation can be well reproduced qualitatively , the observed slope of the anticorrelation and the ranges of [ mg / fe ] and [ al / fe ] are not consistent with the predicted values : a shallower slope of the anticorrelation , a wider range of [ mg / fe ] , a smaller range of [ al / fe ] are clearly seen in the model . this inconsistency can be seen in the model m3 with initially higher [ al / fe ] . although all models in the present study show [ mg / fe]-[al / fe ] anticorrelations , none of the models shows slopes as steep as the observed one for ngc 6752 . 8 shows the time evolution of m1 in the [ o / fe]-[na / fe ] plane as well as the observational plots of bright red giant stars in ngc 6752 . the model predicts a positive correlation between [ o / fe]-[na / fe ] , which is obviously inconsistent with the observed [ o / fe]-[na / fe ] anticorrelation . furthermore , the predicted [ na / fe ] is too low to match the observed values , and this inconsistency can be seen in all models of the present study . this suggests that the derived inconsistency is due to the adopted agb yield models in which no third dredge - up is assumed . also the present study with [ o / fe]-[na / fe ] correlation is in contrast with fenner et al . ( 2004 ) in which third dredge - up is assumed and the derived [ o / fe]-[na / fe ] anticorrelation is not quantitatively consistent with the observed one by yong et al . the fact that both fenner et al . ( 2004 ) and the present study fail to explain the observed [ o / fe]-[na / fe ] anticorrelation quantitatively may well suggest that agb pollution scenarios have a serious problem in terms of explaining the observed abundance inhomogeneity in gcs . we will discuss this point later in 4.1 . 9 shows the dependences of [ c / fe ] and [ n / fe ] number distributions normalized to their maximum values ( denoted as @xmath94 ) on @xmath1 , i.e. , the ratio of @xmath82 ( the total mass of infalling gas ) to @xmath66 ( the total mass of agb ejecta ) and thus controls the degree of agb pollution . fig . 9 clearly shows that both the shapes and the peak values of the distributions depend on @xmath1 , which suggests that @xmath1 is an important parameter for [ c / fe ] and [ n / fe ] distributions . bimodal distributions can be seen in [ n / fe ] for models with @xmath95 and the fraction of n - rich stars with [ n / fe ] @xmath89 is higher in the models with smaller @xmath1 ( i.e. , larger mass fraction of agb ejecta ) . as @xmath1 becomes larger , both the difference between smaller and larger peaks in [ n / fe ] and @xmath94 in the smaller peak become larger . the derived diverse distributions of [ n / fe ] can be compared with the observed diversity in the cn distributions of gcs ( norris 1988 ) . the [ c / fe ] distributions , on the other hand , do not depend so strongly on @xmath1 and the four models show almost monomodal distributions . this reflects the fact that chemical pollution by agb stars does not change [ c / fe ] as dramatically as [ n / fe ] . the spread in [ c / fe ] is however dependent on @xmath1 in the sense that the spread is larger for models with smaller @xmath1 ( i.e. , a larger degree of agb pollution ) . the predicted bimodal [ n / fe ] distribution and monomodal [ c / fe ] one are discussed later in 4.1 by comparing these predictions with the latest observations for less - evolved stars ( e.g. ms stars ) . fig . 10 shows that y distributions normalized to the maximum values ( @xmath94 ) depend on @xmath1 such that both the peak values and the spreads in the y distributions are larger for the models with smaller @xmath1 ( a larger degree of agb pollution ) . these results accordingly suggest that gcs with larger degrees of y inhomogeneity will show higher y. the results shown in figs . 9 and 10 therefore suggest that gcs with higher fractions of n - rich stars with [ n / fe ] @xmath89 can show higher y and larger degrees of y inhomogeneity . since the fraction of n - rich stars in a gc can be observationally quantified by an `` @xmath96 '' parameter by counting the number fraction of cn - strong stars vs cn - weak stars ( norris 1988 ) , the above predicted correlation can be tested by investigating correlations between the @xmath96 parameters and observable properties dependent on y inhomogeneity , such as color spreads between ms stars in the color - magnitude diagrams of gcs . 11 shows that [ n / fe ] distributions depend strongly on @xmath2 , which means that [ n / fe ] distributions can be controlled by the time - scale of gas infall . as seen in the @xmath1 dependences of [ n / fe ] , the models with higher peak values of [ n / fe ] show larger degrees of [ n / fe ] spreads in the distributions . it is significant that the model with @xmath97 shows an almost monomodal [ n / fe ] distribution with an apparently small dispersion of [ n / fe ] ( i.e. no clear bimodality ) . [ c / fe ] distributions do not depend so strongly on @xmath2 as [ n / fe ] so that models show a relatively narrow range of [ c / fe ] . it can be concluded from figs . 9 and 11 that ( i ) [ c / fe ] distributions are highly likely to be monomodal irrespective of @xmath1 and @xmath2 and ( ii ) [ n / fe ] distributions are likely to be bimodal and it depends on @xmath1 and @xmath2 how clearly the distributions show the bimodality . fig . 12 shows that the models with smaller @xmath2 ( i.e. , more rapid gas infall ) show larger degrees of y inhomogeneity . however a positive correlation between the peak y and the degrees of y inhomogeneity seen in models with different @xmath1 ( shown in fig . 10 ) can not be seen in these models with different @xmath2 . given the fact that mean y should not be different between models with the same @xmath66 ( or @xmath1 ) yet different @xmath2 , @xmath96 parameters can be correlated _ only _ with the degrees of y inhomogeneity ( i.e. , not with y ) . the results shown in figs . 10 and 12 thus suggest that y distributions are diverse depending on the two key parameters @xmath1 and @xmath2 . observational implications of these results are discussed later in 4.4 . thus the present study shows that y , [ c / fe ] , and [ n / fe ] distributions in gcs are predicted to be diverse if gcs are formed from the mixture of infalling gas and agb ejecta of field agb stars in forming dwarfs at high @xmath0 . the results shown in figs . 10 - 12 can provide some physical basis not only for the observed diversity in cn distributions of gcs ( e.g. , norris 1988 ) but also for the possible y spreads in ngc 2808 ( e.g. , dantona et al . 2005 ) and @xmath6 cen ( e.g. , bedin et al . 2004 ) . it should be stressed here that the diverse distributions come from _ the combination of the external pollution scenario and the adopted agb yield _ : neither the external pollution scenario assuming the third dredge - up in agb stars nor classical self - pollution scenarios assuming no third dredge - up are able to predict diverse distributions similar to the observed ones . it is more instructive for the present study to compare our model predictions on abundance inhomogeneity within gcs with observations for gc ms stars , rather than with the giant branch populations . comparison with near - ms stars is to be preferred , because we do not have to consider abundance changes due to deep mixing in evolved stars . that said , there are relatively few studies that show observational data sets on abundances of ms populations in gcs ( due to the difficulty in observing these faint objects by the current generation of telescopes ) . we thus use available near - ms data sets in order to understand how well the `` pure '' agb pollution scenario can explain the observations for galactic globular clusters , and note that the data for evolved stars ( i.e. , rgb and agb stars ) are not ideal for testing our models . we here mainly focus on relations between [ c / fe ] and [ n / fe ] using recent observational data sets from briley et al . ( 2004a , b ) , carretta et al . ( 2005 ) , cohen et al . ( 2002 ) , cohen et al . ( 2005 ) , and smith et al . ( 2005 ) . 13 shows the comparison between the predicted evolution of model in the [ c / fe]-[n / fe ] plane and the observed [ c / fe ] and [ n / fe ] values for the red giants in m13 , m10 , and ngc 7006 ( smith et al . 2005 ) . it is clear from this figure that ( i ) the observed anticorrelation between [ c / fe ] and [ n / fe ] can be qualitatively reproduced , ( ii ) the adopted `` pure '' agb - pollution scenario , however , can not explain the observed smaller [ c / fe ] ( [ c / fe ] @xmath98 ) , and ( iii ) it can not explain stars with high [ n / fe ] ( @xmath99 ) and low [ c / fe ] ( @xmath98 ) . these results clearly suggest that a combination of deep mixing ( e.g. , see the new mechanism discussed by eggleton et al 2006 ) and primordial pollution from agb stars in proto - gc clouds need to be considered for reasonable explanations of the observations . 14 shows the comparison between the predicted evolution of the models on the [ c / fe]-[n / fe ] plane and the observed [ c / fe ] and [ n / fe ] for less evolved stars such as ms populations , subgiants , and dwarfs in ngc 6397 ( carretta et al . 2005 ) , ngc 6752 ( carretta et al . 2005 ) , 47 tuc ( briley et al . 2004b ; carretta et al . 2005 ) , m15 ( cohen et al . 2005 ) , m13 ( briley et al . 2004a ) , and m5 ( cohen et al . 13 and fig . 14 clearly indicate that the models are much more consistent with observations for less evolved stars ( fig . 14 ) than for evolved ones ( fig . 13 ) , though the models still have difficulty in explaining stars with low [ c / fe ] ( @xmath98 ) . furthermore , the observed [ c / fe]-[n / fe ] anticorrelation and the apparently non - linear dependence of [ n / fe ] on [ c / fe ] are both consistent with the theoretical predictions . these derived consistencies may well provide support for the adopted star formation histories of proto - gc clouds in the external pollution scenario . 15 shows the comparison between the predicted [ c / fe ] and [ n / fe ] distributions and the observed ones for subgiants and stars at the base of rgb in m15 and for ms stars in 47 tuc . it is clear from fig . 15 that ( i ) both models and observations show almost monomodal distributions in [ c / fe ] , ( ii ) the predicted peak ( in normalized number fraction ) in the [ c / fe ] distribution is significantly higher than those of the observations , ( iii ) the predicted bimodal distribution is more similar to the observed one in 47 tuc than to the one in m15 , and ( iv ) the locations of the two peaks in [ n / fe ] are however not exactly the same as the observed ones . thus the observed wider spread in [ c / fe ] can not be explained so well by the present models , though the monomodal [ c / fe ] and bimodal [ n / fe ] distributions are quite well reproduced by the models . the present models also predict the mean values of [ c / fe ] and [ n / fe ] in gcs and thus can be compared with the ones derived from high - resolution spectroscopic studies of _ integrated spectra _ of gcs in nearby galaxies other than the galaxy . ongoing and future observational studies based on large ground - based telescopes with high - dispersion spectrographs enable us to derive [ c / fe ] and [ n / fe ] of gcs in other galaxies by model fitting ( e.g. , bernstein 2006 ) . recent preliminary results reported by strader et al . ( 2006 ) for a gc in m31 ( 034 - 096 ) show that [ c / fe ] and [ n / fe ] are roughly 0 dex and 0.6 dex , respectively , with the observational errors of about 0.15 dex . this result is more consistent with the present models with smaller @xmath1 ( that shows higher [ c / fe ] and smaller [ n / fe ] ) . we suggest that a statistical study of m31 gcs in terms of [ c / fe ] and [ n / fe ] can provide new insight into the differences in chemical pollution of proto - gc clouds by agb stars between m31 gcs and the galactic ones . one of the serious problems in any self - pollution scenario is that the number fraction of n - rich , c - depleted stars in a gc for a canonical imf can not be large enough to explain the observed fraction ( e.g. , smith & norris 1982 ; dantona & caloi 2004 ; prantzsos & charbonnel 2006 ) . this problem can be solved if we assume peculiar top - heavy imfs ( dantona & caloi 2004 ) , for which gcs can either be easily destroyed by the galactic tidal field during their evolution ( e.g. , chernoff & weinberg 1990 ) or be disintegrated quickly due to loss of a large amount of mass through type - ii supernovae explosion ( bekki & norris 2006 ) . since both n - rich and n - normal components can be formed from proto - gc clouds chemically polluted by field agb stars ( that do not become parts of gcs ) in the external pollution scenario , there are no serious problems on number fraction of n - rich , c - depleted stars in the scenario with a canonical imf . recently pasquini et al . ( 2005 ) have suggested that the whole proto - gc cloud needs to be chemically enriched by a previous generation of stars based on the observational data of the li abundance in 9 turn - off stars in ngc 6752 : this is consistent with the external pollution scenario . gcs can be formed during or after the formation of the field stars in dwarfs at very high @xmath0 in the external scenario . a consequence of this is that the field stars in dwarfs on average should be older and more metal - poor than gcs when gcs are formed in this scenario . accordingly , if dwarfs with gcs can be destroyed by the tidal field of the proto - galaxy to form the galactic field halo stars and the halo gcs soon after gc formation , the field halo stars are highly likely to be more metal - poor than the halo gcs in this scenario . it is a well known observational fact that the metallicities of the most metal - poor halos stars are lower than those of the most metal - poor gcs in the galaxy ( e.g. , freeman 1993 ) . the earlier formation of field halo stars in comparison with gcs in the external pollution scenario can naturally explain this observation . since proto - gc clouds are chemically enriched by old field stars ( not by the very first stars like pop iii stars ) of low - mass dwarf galaxies in the present models , the clouds can have abundances similar to field stars ( which can finally become halo stars ) and be uniformly enriched by @xmath100 and @xmath101process syntheses . therefore , gcs formed from these clouds can have abundances similar to those of field stars except for light elements and accordingly explain recent observations that have reported little differences in [ fe / h]-dependences of [ sr / fe ] , [ y / fe ] , [ ba / fe ] , and [ eu / fe ] in main sequence turnoff stars and subgiants at the base of the red giant branch ( rgb ) in some gcs ( james et al . stars in the models with no third dredge - up keep initial abundances of @xmath101process elements so that the models can be consistent with the observed possible evidences for homogeneous abundances of s - process elements in james et al . the external pollution from agb stars _ with third dredge - up _ would be inconsistent with the above observations by james et al . ( 2004 ) : future observations on the degree of homogeneity in the abundances of @xmath101process elements will give strong constrains on any agb - pollution scenario . recent abundance studies of the galactic open cluster have not yet revealed possible abundance spreads between stars of these clusters ( e.g. , friel et al . 2003 ; de silva et al . the external pollution scenario suggests that these clusters are highly unlikely to show abundance inhomogeneities , because they are formed from the galactic gas clouds in which agb pollution can not proceed owing to very shallow gravitational potentials of the clouds . hill et al . ( 2000 ) reported that lmc clusters do not show al - o anticorrelations , which implies that agb pollution did not proceed in these systems . the external pollution scenario thus suggests that these lmc clusters , which shows disky kinematics ( e.g. , freeman et al . 1983 ) , were formed not within dwarfs that were the lmc s building blocks but within the proto - lmc disk via a dissipative process at the formation epoch of the disk . the observed differences in some elemental abundances between the lmc clusters and the galactic gcs ( johnson et al . 2006 ) may well be understood in terms of clusters originating within _ the forming disk of the lmc_. as pointed out by fenner et al . ( 2004 ) , chemical evolution models of gcs based on agb - pollution have some difficulties in explaining quantitatively o - na and mg - al anticorrelations and correlations between @xmath102mg@xmath103 mg ( and @xmath104mg@xmath103 mg ) and light elements ( e.g. , [ o / fe ] and [ na / fe ] ) . although some improvement over previous models can be seen in the present models in which agb stars experience hbb but no third dredge - up , the observed o - na anticorrelation can not be explained by them . this incapability comes from the adopted stellar yield models of agb stars rather than from improper models for early star formation histories within gcs . given the fact that there are some uncertainties in model calculations of yields in agb stars ( ventura & dantona 2005c ; campbell et al 2006 ) , the above apparent failure does not necessarily mean that there is a serious problem with the adopted agb - pollution scenario . indeed , we remind the reader that no consistent solution currently exists . all proposals rely on tweaking of some inputs , be it overshoot , reaction rates or something else , or indeed a combination . one solution for the above problem would be to adopt agb models with little third dredge - up ( i.e. , not the models with no dredge - up ) in order to avoid unreasonably small values of [ na / fe ] in the models . this contrived way of modeling , however , should not be regarded as a necessary solution for this problem . it may say something about the characteristics of low - metallicity agb stars or perhaps higher mass super - agb stars maybe they do not suffer as much third dredge - up as expected by our models . previous observational studies showed that there exists variations in the strength of cn bands between the galactic gcs ( e.g. , norris & smith 1981 ; suntzeff 1981 ) . norris ( 1988 ) investigated the value of the `` @xmath96 '' parameter , which is defined as the ratio of the number of cn - strong to cn - weak stars in a given system , for each of 12 gcs with @xmath105 } \le -1.2 $ ] and thereby found that the values of the @xmath96 parameter range from 0.22 ( i.e. , smaller fraction of cn - strong stars ) to 3.29 ( i.e. , larger fraction ) . norris et al . ( 1984 ) also suggested that the observed anticorrelation between the behavior of the cn bands and that of the ch features in gcs ( e.g. , norris & cottrell 1979 ; norris & freeman 1982 ; norris et al . 1984 ; bell et al . 1983 ) is difficult to explain in terms of the primordial pollution ( or self - pollution ) scenarios . we have demonstrated that ( 1 ) the distributions of n and c are diverse depending on the two key parameters @xmath1 and @xmath2 , ( 2 ) n distributions show clear bimodality in some models , and ( 3 ) c and n abundances anticorrelate with each other in the external pollution scenario . these results provide some physical basis for the origin of the observed diversity and for the observed c - n anticorrelation , _ if we understand what determines the values of the above two key parameters in proto - galaxies_. owing to the lack of extensive numerical simulations on the dynamical fate of agb ejecta in the central region of proto - galaxies , it is unclear what physical processes in proto - galaxies ( e.g. , the effectiveness of stellar feedback effects ) can control @xmath1 and @xmath2 . we plan to derive @xmath1 and @xmath2 and their dependences on physical properties of proto - galaxies ( e.g. , masses and sizes ) in our future high - resolution simulations and thereby discuss the origin of the observed diversity of cn distributions in gc in the context of galaxy formation . recent photometric observations of stars in @xmath6 cen have discovered a double main sequence ( dms ) in the color magnitude diagrams ( cmds ) of its stellar populations ( anderson 1997 ; bedin et al . one of the most promising interpretation is that stars on the bluer main sequence ( bms ) of the dms represents a very helium - rich ( @xmath106 ) population ( norris 2004 ; piotto et al . . furthermore , dantona et al . ( 2005 ) have reported helium abundance variation among main sequence stars of ngc 2808 . green ( 1980 ) first showed that there is a possible correlation between the helium abundance of gcs and the locations of gcs with respect to the galactic center : inner gcs are more likely to show higher helium abundance . recently yong et al . ( 2005 ) have reported a possible trend between [ fe / h ] and [ al / fe ] for giants stars in ngc 6752 and thus suggested that this trend can result from a possible correlation between the h - he ratio and [ al / fe ] , which is a natural result of chemical pollution by h - burning ejecta . although the total number of observations reporting the possible abundance spread in helium is currently very small , these observations contain valuable information not only on horizontal branch morphologies but also on helium pollution processes in proto - gcs ( e.g. , dantona et al . 2005 , 2006 ) . important predictions of the present external pollution scenario on helium abundance properties of gcs are ( 1 ) spread in helium abundance between stars , ( 2 ) larger helium abundances in n - rich , c - depleted stars , and ( 3 ) smaller helium abundance in gcs formed from gas less polluted by field agb stars . furthermore , we can provide some prediction on color distributions of stars for the main - sequence ( ms ) populations of gcs by using the model developed by dantona et al . differences in y between stars with normal y ( = 0.24 ) and those with higher / lower y can be converted into color difference ( @xmath107 ) by using the following formula ( dantona et al . 2005 ) : @xmath108 fig . 16 clearly demonstrates that the color distributions of stars in the fiducial model m1 shows significant dispersions in @xmath109 of the ms . this dispersion can be observed as spreads in the cm relations of ms populations of gcs and thus as evidence for gc formation from agb ejecta . several authors have pointed out that he abundance inhomogeneity of stars in gcs can be imprinted on photometric properties of stellar populations on the horizontal branch ( hb ) of gcs ( e.g. , norris 2004 ; dantona et al . 2006 ; lee et al . these authors suggested that a gc with a he - rich population can ( e.g. , ngc 2808 ) show a characteristic hb color and luminosity distribution , a bimodal hb distribution , and faint , very hot hb stars . since a gc with a larger fraction of n - rich stars can have a higher degree of he abundance inhomogeneity in the external pollution scenario , the hb properties of gcs can be correlated with the fraction of n - rich stars ( i.e. , @xmath96 parameter ) . if the origin of the second parameter problem is related to the difference in helium abundance between gcs ( see catlan 2005 for a recent review ) , the external pollution scenario suggests that the origin of the problem can be closely associated with differences in agb pollution processes ( controlled mainly by the two parameters , @xmath1 and @xmath2 ) between different proto - gc clouds in the central regions of forming dwarfs at high @xmath0 . norris ( 1987 ) investigated possible correlations between the @xmath96 parameter and other physical properties of gcs ( e.g. , metallicities and luminosities ) and found that there exists a correlation between ellipticities ( @xmath110 ) of gc shapes and the values of the @xmath96 parameters ( see smith ( 2002 ) for recent confirmation of this correlation ) . this result means that if the flattened shapes are due to rotational kinematics of gcs ( rather than anisotropy of velocity dispersion ) , gcs with a larger amount of angular momentum are more likely to show larger @xmath96 . norris ( 1987 ) suggested that _ if global rotation of a gc positively correlate with rotation of individual stars within the gc and if evolutionary mixing responsible for the cn abundance inhomogeneity is driven by stellar rotation _ , the observed correlation can be readily explained . norris ( 1987 ) however pointed out that this explanation has a difficulty in explaining the existence of cn - enhanced stars near the main - sequence turnoff of 47 tuc ( bell et al . 1983 ) unless the effects of rotation are efficient for main - sequence stars . furthermore it is unclear why stars within gcs with a larger amount of intrinsic angular momentum would have a larger amount of stellar rotation in this scenario by norris ( 1987 ) . in the present external pollution scenario , the values of the @xmath96 parameters depend strongly on the relative mass fraction of agb ejecta with respect to the total gas mass ( i.e. , agb ejecta + fresh infall gas ) in proto - gcs : the higher the fraction is , the larger the @xmath96 becomes . one of possible explanations for the above @xmath110-@xmath96 correlation in the context of the external pollution is therefore that whenever a larger amount of agb ejecta can be transferred into proto - gc regions ( i.e. , central regions of their host dwarfs ) , the gas is highly likely to have a larger amount of angular momentum , from which gcs with a larger amount of angular momentum and thus with larger @xmath110 can be formed . we suggest that the above situation of the larger amount of agb ejecta with higher ( specific ) angular momentum is likely if field stars in the central regions of dwarfs ( thus , agb polluters for proto - gcs ) have rotational kinematics : agb ejecta from such stars rotating dwarfs can also have orbital angular momentum ( with respect to the centers of dwarfs ) so that gcs formed from the ejecta in the centers can have rotational kinematics and thus large @xmath110 after conversion of orbital angular momentum of gas into intrinsic spin angular momentum of gcs during gc formation . it is thus our future study to investigate ( i ) whether orbital angular momentum of agb ejecta from field stars rotating dwarfs can be really converted into intrinsic spin angular momentum of gcs and ( ii ) whether gcs have a larger amount of intrinsic spin angular momentum thus larger @xmath110 , if they are formed from gas with a larger fraction of agb ejecta . high - resolution numerical simulations on stellar and gas dynamics in the central @xmath111pc of dwarfs at high @xmath0 would be essential for this investigation . we plan to perform such chemodynamical simulations with agb feedback effects ( e.g. , thermal heating and return of agb ejecta to ism ) to understand the origin of the observed @xmath110-@xmath96 correlation in the galactic gcs . nuclear dynamics in proto - galaxies would be dependent on the depth of their gravitational potential wells thus on their masses and sizes . our future simulations will therefore enable us to understand whether very flattened gcs ( e.g. , @xmath6 cen ) can be formed in less or more massive dwarfs at high @xmath0 . recently , li & burstein ( 2003 ) have found that the nh absorption line is far stronger in ( three ) metal - rich gcs of m31 than it is in the galactic gcs at a given value of ch or [ fe / h ] . beasley et al . ( 2004 ) also have revealed that the near - uv cyanogen features of m31 gcs are strongly enhanced with respect to the galactic gcs for @xmath112 } < -0.3 $ ] . these observations imply that [ n / fe ] can be higher in some m31 gcs than in the galactic ones for a same metallicity range . it has been also revealed that m31 gcs having very strong nh absorptions are more luminous with @xmath113 ( e.g. , burstein et al . this implies that the possibly higher [ n / fe ] is due to a larger number of bright gcs in m31 , which can have the higher values for some physical reasons . it remains , however , totally unclear why these bright m31 gcs have high [ n / fe ] . the present models have shown that mean values of [ n / fe ] of gcs , which should be compared with the [ n / fe ] estimated from the integrated spectra of m31 gcs , can be as high as 0.9 . this result , combined with recent theoretical results and observational ones on nucleus formation of galaxies ( e.g. , bekki 2006 ; bekki et al . 2006 ; cte et al . 2006 ) , can provide the following answer for the above question . based on analytical calculations , bekki ( 2006 ) pointed out that agb ejecta can be more effectively trapped in nuclear region of more massive dwarfs embedded in dark matter halos so that agb pollution can proceeds to a greater extent in these dwarfs . inward radial gas - transfer into nuclear regions , which is indispensable for the formation of compact stellar systems , is demonstrated to be more efficient in more massive dwarfs ( bekki et al . photometric studies on nuclear properties of dwarfs based on the acs virgo cluster survey have recently revealed that stellar nuclei are more massive in more massive dwarfs ( cte et al . therefore we can claim that if some bright gcs in m31 originate from nuclei of brighter nucleated dwarfs , the observed higher [ n / fe ] can be naturally explained . thus the origin of the observed possible differences in [ n / fe ] between m31 gcs and the galactic ones could reflect the fact that the proto - m31 contained a larger number of brighter dwarfs where agb pollution could proceed more effectively in their central regions to form gcs with higher [ n / fe ] . these more massive dwarfs were destroyed to form the m31 s stellar halo during the hierarchical formation of m31 through merging of these dwarfs . because of the mass - metallicity relation of field stellar populations of dwarfs ( e.g. durrell et al . 1996 ) , the developed stellar halo can show a higher metallicity and is consistent with observations ( e.g. , reitzel et al . if [ n / fe ] of gcs can be controlled by nuclear dynamics of their host dwarfs at very high @xmath0 , as suggested by the present study , more luminous galaxies would have both a larger number of gcs with high [ n / fe ] and more metal - rich ( thus redder ) stellar halos . we thus suggest that future high - resolution spectroscopic studies for gc abundances in galaxies beyond the local group , combined with wide - field imaging of stellar halos in these galaxies , can give some constraints on the formation histories of gcs within dwarfs at very high @xmath0 . we have investigated the abundance inhomogeneity among the light elements ( e.g. , c , n , o , na , and al ) in stars in the galactic gcs by adopting the external pollution scenario and using latest stellar yield models of metal - poor agb stars with and without third dredge - up . in the external pollution scenario , gcs within a forming low - mass dwarf embedded in a dark matter halo can be formed from the mixture of ( i ) gas ejected from the field agb stars formed earlier in the dwarf and ( ii ) the interstellar gas infalling to the central region of the dwarf . we extensively investigated our models with no third dredge - up over wide parameter ranges . we summarise our principal results of the models as follows . \(1 ) the ratio of the total mass of infalling gas to that of agb ejecta during gc formation within a forming dwarf galaxy ( described by the @xmath1 parameter ) and the time scale of gas infall ( @xmath2 ) are the most important parameters that can determine abundance properties of gcs . \(2 ) both [ n / fe ] and [ c / fe ] can be diverse among stars within a gc owing to chemical pollution from field agb stars . [ n / fe ] distributions in some gcs can clearly show bimodality where s [ c / fe ] can be monomodal in most models . [ n / fe ] distributions depend on @xmath1 , such that models with smaller @xmath1 ( i.e. , larger mass fraction of agb ejecta used for gc formation ) show the [ n / fe ] bimodality more clearly . \(3 ) n - rich , c - poor stars in gcs also have higher he abundances owing to pollution from massive agb stars with he - rich ejecta . the number fraction of he - rich stars ( y @xmath3 ) is higher for the models with smaller @xmath1 and shorter @xmath2 for @xmath4 and @xmath5 yr . he abundances of stars can correlate with [ n / fe ] and [ al / fe ] and anticorrelate with [ c / fe ] , [ o / fe ] , and [ na / fe ] within gcs ( and a simple consequence of h - burning ) . \(4 ) although our models can much better explain the observed c - n and mg - al anticorrelations than previous theoretical models , the observed o - na anticorrelation can not by simply explained by any models with different parameters in this scenario . the inability of the present models to match the observations results from the adopted agb yields and thus suggests that models using different agb yields need to be explored to assess the viability of agb pollution scenarios ( for more details , see appendix b ) . \(5 ) extensive comparison between the models and the observations demonstrates that the predicted [ c / fe]-[n / fe ] anticorrelation can be more quantitatively consistent with observations for less evolved stars ( ms populations , subgiants , and dwarfs ) than for evolved ones ( e.g. , red giants ) . the comparison also demonstrates that the observed [ c / fe ] and [ n / fe ] distributions for less evolved stars are at least qualitatively consistent with the model predictions . \(6 ) one important prediction of our models is that gcs with higher degrees of abundance inhomogeneity should show a larger color spread ( e.g. , @xmath107 ) in their ms populations as the result of bluer stars formed from he - rich gas . accordingly , it would be worthwhile for future observational studies to investigate correlations between the @xmath96 parameters ( i.e. , the number ratio of cn - strong to cn - weak stars ) and @xmath107 in gcs . \(7 ) the values of the two key parameters ( @xmath1 and @xmath2 ) are suggested to be controlled by stellar and gas dynamics in the central regions of host galaxies of gcs . central stellar and gas dynamics ( e.g. , gas fueling rates to the central 10 pc ) have been suggested to be controlled by global properties ( e.g. , total masses and densities ) of galaxies . accordingly , abundance inhomogeneities of gcs may have fossil information on ( i ) formation and evolution histories of central regions of defunct dwarfs and ( ii ) physical properties of the dwarfs . thus we have shown that the cn - bimodality and the number fraction of c - depleted , n - rich populations of a gc can give strong constants on the short - term ( an order of @xmath62 yr ) star formation history of the proto - gc cloud of the gc , if the cloud is chemically polluted by agb stars . we accordingly suggest that any theoretical models for the origin of gc abundances need to explain not only the o - na and mg - al anticorrelations but also the bimodal cn abundance distributions in gcs . we are grateful to the anonymous referee for valuable comments , which contribute to improve the present paper . k.b . , j.c.l . , and j.e.n . acknowledge the financial support of the australian research council throughout the course of this work . swc would like to thank the australian partnership for advanced computing ( apac ) for the computing time granted under project g61 , `` element production by intermediate mass red - giant stars '' . anderson , j. , 1997 , ph.d . thesis , univ . california , berkeley beasley , m. a. , brodie , j. p. , strader , j. , forbes , d. a. , proctor , r. n. , barmby , p. , huchra , j. p. , 2004 , aj , 128 , 1623 bedin , l. r. , piotto , g. , anderson , j. , cassisi , s. , king , i. r. , momany , y. , carraro , g. , 2004 , apj , 605 , l125 bekki , k. , 2006 , mnras , 367 , l24 bekki , k. , chiba , m. , 2004 , a&a , 417 , 437 bekki , k. , freeman , k. c. , 2003 , mnras , 346 , l11 bekki , k. , norris , j. e. , 2006 , apj , 637 , l109 bekki , k. , chiba , m. , 2007 , submitted to apjl bekki , k. , couch , w. j. , shioya , y. , 2006 , apj , 642 , l133 bell , r. a. , hesser , j. e. , cannon , r. d. , 1983 , apj , 269 , 580 bernstein , r. et al . , 2006 , in globular clusters guides to galaxies , eds . t. richtler & d. geisler , ( eso / springer ) , in preparation briley , m. m. , cohen , j. g. , stetson , p. b. , 2004a , aj , 127 , 1579 briley , m. m. , harbeck , d. , smith , g. h. , grebel , e. k. , 2004b , aj , 127 , 1588 burstein , d. et al . 2004 , apj , 614 , 158 campbell , s. w. 2006 , phd thesis ( monash university ) campbell , s. w. , fenner , y. , karakas , a. i. , lattanzio , j. c. , gibson , b. k. , 2005 , nupha , 758 , 272 campbell , s. w. et al . , 2006 , in preparation cannon , r. d. , croke , b. f. w. , bell , r. a. , hesser , j. e. , stathakis , r. a. , 1998 , mnras , 298 , 601 carretta , e. , gratton , r. g. , lucatello , s. , bragaglia , a. , bonifacio , p. , 2005 , a&a , 433 , 597 catelan , m. 2005 , preprint ( astro - ph/0507464 ) charbonnel , c. , prantzos , n. , 2006 , preprint ( astro - ph/0606220 ) chernoff , d. f. , weinberg , m. d. , 1990 , 351 , 121 cohen , j. g. 1978 , apj , 223 , 487 cohen , j. g. , briley , m. m. , stetson , p. b. , 2002 , aj , 123 , 2525 cohen , j. g. , briley , m. m. , stetson , p. b. , 2005 , aj , 130 , 1177 cte et al . , 2006 , preprint ( astro - ph/0603252 ) cottrell , p. l. ; da costa , g. s. , 1981 , apjl , 245 , 79 da costa , g. s. , cannon , r. , croke , b. , norris , j. e. , 2004 , mmsai , 75 , 370 dantona , f. , caloi , v. , 2004 , apj , 611 , 871 dantona , f. , bellazzini , m. , caloi , v. , pecci , f. f. , galleti , s. , rood , r. t. , 2005 , apj , 631 , 868 dantona , f. , ventura , p. , caloi , v. , 2006 , mmsai , 77 , 168 denissenkov , p. a. , weiss , a. , 1996 , a&a , 308 , 773 denissenkov , p. a. , weiss , a. , wagenhuber , j. , 1997 , a&a , 320 , 115 denissenkov , p. a. , herwig , f. , 2003 , apj , 590 , l99 de silva , g. m. , sneden , c. , paulson , d. b. , asplund , m. , bland - hawthorn , j. , bessell , m. s. , freeman , k. c. , 2006 , aj , 131 , 455 duerr , r. , imhoff , c. l. , lada , c. j. , 1982 , apj , 261 , 415 dupree , a. k. , sasselov , d. d. , lester , j. b. , 1992 , apjl , 387 , 85 durrell , p. r. , harris , w. e. , pritchet , c. j. , 2000 , aj , 121 , 2557 eggleton , p. e. , dearborn , d. s. p. , lattanzio , j.c . , 2006 , science , 314 , 1580 fenner , y. , campbell , s. w. , karakas , a. i. , lattanzio , j. c. , gibson , b. k. , 2004 , mnras , 353 , 789 frank , j. , gisler , g. , 1976 , mnras , 176 , 533 freeman , k. c. 1993 , in the globular clusters - galaxy connection , eds . g. h. , smith , and j. p. , brodie , ( san francisco : asp ) , asp conf . 48 , p608 freeman , k. c. , norris , j. e. , 1981 , ara&a , 19 , 319 freeman , k. c. , illingworth , g. , oemler , a. , jr . , 1983 , apj , 272 , 488 friel , e. d. , jacobson , h. r. , barrett , e. , fullton , l. , balachandran , s. c. , pilachowski , c. a. , 2003 , aj , 126 , 2372 gil - pons , p. , suda , t. , fujimoto , m. y. , garca - berro , e. , 2005 , a&a , 433 , 1037 gnedin , oleg y. , zhao , h. , pringle , j. e. , fall , s. m. , livio , m. , meylan , g. , 2002 , apjl , 568 , 23 goswami , a. , prantzos , n. , 2000 , a&a , 359 , 191 gratton , r. , sneden , c. , carretta , e. , 2004 , ara&a , 42 , 385 gratton , r. g. , 2004 , mmsai , 75 , 274 green , e. m. , 1980 , in star clusters ; proceedings of the symposium , ( british columbia , dordrecht , d. reidel publishing co. ) , p.441 hilker , m. , richtler , t. , 2000 , a&a , 362 , 895 hill , v. , francois , p. , spite , m. , primas , f. , spite , f. , 2000 , a&a , 364 , l19 ivans , i. i. , sneden , c. , kraft , r. p. , suntzeff , n. b. , smith , v. v. , langer , g. e. , fulbright , j. p. , 1999 , aj , 118 , 1273 james , g. , francois , p. , bonifacio , p. , carretta , e. , gratton , r. g. , spite , f. , 2004 , a&a , 427 , 825 johnson , j. a. , ivans , i. i. , stetson , p. b. , 2006 , apj , 640 , 801 karakas , a. i. , lattanzio , j. c. , 2003 , pasa , 20 , 393 karakas , a. , fenner , y. , sills , a. , campbell , s. w. , lattanzio , j. , 2006 , accepted in apj kraft , r. p. , 1994 , pasp , 106 , 553 lee , y. et al . 2005 , apj , 621 , l57 leep , e. m. , wallerstein , g. , oke , j. b. , 1986 , aj , 91 , 1117 li , y. , burstein , d. , 2003 , apj , 598 , l103 loup , c. , forveille , t. , omont , a. , paul , j. f. , 1993 , a&as , 99 , 291 maeder , a. , meynet , g. , 2006 , a&a , 448 , l37 norman , c. , scoville , n. , 1988 , apj , 332 , 124 norris , j. e. , 1987 , apj , 313 , l65 norris , j. e. , freeman , k. c. , 1982 , apj , 254 , 143 norris , j. & smith , g. h. 1981 , astrophysical parameters for globular clusters , i.a.u . colloquium no . philip & d.s . hayes ( schenectady : l. davis press ) , p.109 norris , j. e. , 1988 , in the harlow - shapley symposium on globular cluster systems in galaxies ( 126th iau symposium ) , ( dordrecht , kluwer academic publishers ) , p. 93 . norris , j. e. , cottrell , p. l. , freeman , k. c. , da costa , g. s. , 1981 , apj , 244 , 205 norris , j. e. , freeman , k. c. , da costa , g. s. , 1984 , apj , 277 , 615 norris , j. e. , da costa , g. s. , 1995 , apj , 441 , l81 norris , j. e. , 2004 , apj , 612 , 25 oliveira , j. m. , van loon , j. t. , stanimirovic , s. , zijlstra , a. a. , 2006 , in preprint ( astro - ph/0609236 ) pasquini , l. , bonifacio , p. , molaro , p. , francois , p. , spite , f. , gratton , r. g. , carretta , e. , wolff , b. , 2005 , a&a , 441 , 549 peterson , r. c. , 1980 , apj , 237 , l87 piotto , g. et al . 2005 , apj , 621 , 777 pilachowski , c. a. , 1988 , apj , 326 , l57 prantzos , n. , charbonnel , c. , 2006 , preprint ( astro - ph/0606112 ) prantzos , n. , doom , c. , de loore , c. , & arnould , m. , 1986 , apj , 304 , 695 ramirez , s. v. , cohen , j. g. , 2002 , aj , 123 , 3277 reitzel , d. b. , guhathakurta , p. , gould , a. , 1998 , aj , 116 , 707 renzini , a. buzzoni , a. , 1986 , in spectral evolution of galaxies , ( dordrecht , d. reidel publishing co. ) , p.195 ritossa , c. , garcia - berro , e. , iben , i. jr , 1996 , apj , 460 , 489 siess , l. , 2006 , a&a , 448 , 717 smith , g. h. , 1987 , pasp , 99 , 67 smith , g. h. , 1996 , pasp , 108 , 176 smith , g. h. , norris , j. e. , 1982 , apj , 254 , 594 smith , g. h. , norris , j. e. , 1993 , aj , 105 , 173 smith , g. h. , shetrone , m. d. , bell , r. a. , churchill , c. w. , briley , m. m. , 1996 , aj , 112 , 1511 smith , g. h. , briley , m. m. , harbeck , d. , 2005 , aj , 129 , 1589 sneden , c. , kraft , r. p. , prosser , c. f. , langer , g. e. , 1992 , aj , 104 , 2121 sollima , a. , borissova , j. , catelan , m. , smith , h. a. , minniti , d. , cacciari , c. , ferraro , f. r. , 2006 , apj , 640 , l43 strader , j. et al . , 2006 , in proceedings of globular clusters guides to galaxies , eds . t. richtler & d. geisler , ( eso / springer ) , in preparation suntzeff , n. , 1993 , in the globular clusters - galaxy connection , eds . g. h. , smith , and j. p. , brodie , ( san francisco : asp ) , asp conf . vol 48 , p167 sweigart , a. v. , mengel , j. g .. , 1979 , apj , 229 , 624 thoul , a. , jorissen , a. , goriely , s. , jehin , e. , magain , p. , noels , a. , parmentier , g. , 2002 , a&a , 383 , 491 tsuchiya , t. , dinescu , d. i. , korchagin , v. i. , 2003 , apjl , 589 , 29 van den hoek , l. b. , groenewegen , m. a. t. , 1997 , a&as , 123 , 305 venn , k. a. , irwin , m. , shetrone , m. d. , tout , c. a. , hill , v. , tolstoy , e. , 2004 , aj , 128 , 1177 ventura , p. , dantona , f. , mazzitelli , i. , 2002 , a&a , 393 , 215 ventura , p. , dantona , f. , mazzitelli , i. , 2004 , mmsai , 75 , 335 ventura , p. , & dantona , f. , 2005a , a&a , 431 , 279 ventura , p. , & dantona , f. , 2005b , apj , 635 , l149 ventura , p. , & dantona , f. , 2005c , a&a , 439 , 1075 ventura , p. ; dantona , f. , 2006 , a&a , 457 , 995 weidemann , v. , 2000 , a&a , 363 , 647 wilking , b. a. , lada , c. j. , 1983 , apj , 271 , 698 woosley , s. e. , & weaver , t. a. 1995 , apjs , 101 , 181 yong , d. , grundahl , f. , lambert , d. l. , nissen , p. e. , shetrone , m. d. , 2003 , a&a , 402 , 985 yong , d. , grundahl , f. , nissen , p. e. , jensen , h. r. , lambert , d. l. , 2005 , a&a , 438 , 875 bedin et al . ( 2004 ) suggested that stars on the bluer main sequence ( bms ) of the double main sequence ( dms ) can be a very helium - rich ( @xmath106 ) population . a number of recent investigations have suggested that a higher degree of he enhancement in the bms is a viable explanation for the origin of the dms ( e.g. , lee et al . 2005 ; norris 2004 ; piotto et al . 2005 ) and recent analytical works by bekki & norris ( 2006 ) have suggested that he enhancement due to self - pollution ( or more correctly , external pollution ) of agb stars can explain the origin of the bms . it however remains unclear whether such agb pollution scenarios can _ quantitatively _ explain the observed abundance properties ( e.g. , [ c / fe ] , [ n / fe ] , and y ) in a self - consistent manner . a1 shows y , [ c / fe ] , and [ n / fe ] distributions of the model m14 in which very small @xmath1 ( i.e. , a higher degree of agb pollution ) , smaller @xmath2 ( i.e. , more rapid gas infall ) , and smaller @xmath85 ( i.e. , more rapid chemical enrichment ) are adopted so that the model shows a large fraction of he - rich stars . owing to the modeled very larger degree of agb pollution , stars in this model shows a large fraction of he - rich ( @xmath114 ) and n - rich ( @xmath89 ) stars . piotto et al . ( 2005 ) have reported that ( i ) the he - rich bms shows a rather large nitrogen abundance ( [ n / m ] @xmath115 ) and ( ii ) it is not very rich in c ( [ c / m]=0 ) . the result in fig . a1 implies that these observations can be due to agb pollution , though the c abundance of the bms is not so consistent with the predicted one . an important result in fig . a1 is that some fraction ( @xmath116 ) of stars show relatively normal y ( @xmath117 ) . this suggest that if the bms is formed from gas polluted by agb stars , stars with relatively normal y are also formed , though the mass fraction of the he - normal population is small . this may well provide a clue to the origin of the observed normal y ( = 0.246 ) for the metal - intermediate rrl stars ( that would have been formed with the bms ) in @xmath6 cen ( sollima et al . 2006 ) . very luminous gcs such as @xmath6 cen in the galaxy and g1 in m31 have long been suggested to originate from nuclei of ancient nucleated systems , where continuous gas infall and chemical enrichment were highly likely to occur ( e.g. , hilker & richtler 2000 ; bekki & freeman 2003 ; bekki & chiba 2004 ) . as demonstrated above , the total mass of agb ejecta used for the he - rich population in @xmath6 cen needs to be @xmath118 6 times larger than that of infalling gas not enriched by agb stars . it remains unclear whether this can really happen in the nuclei of dwarf galaxies that are considered to be progenitors of very luminous gcs like @xmath6 cen and g1 . current standard models of intermediate - mass agb stars with third dredge - up ( 3dup ) do not reproduce the observed c - n , o - na or mg - al inverse - correlations observed in most gcs . we note however that the models of ventura & dantona ( 2005b ) do come close , due to their use and development of an alternative theory of convection , which alters the evolution of their models significantly ( see e.g. , ventura & dantona 2005a ) . in the standard models the combination of hot bottom burning ( hbb ) and 3dup during agb evolution does predict enhancements of nitrogen , sodium and aluminium , and a reduction in oxygen all of which are needed to explain the observations however they also predict an increase of carbon and magnesium , giving positive c - n and mg - al correlations which is in direct violation of the observations . in addition to this the models do not _ quantitatively _ account for the large degree of o depletion and tend to over - produce na , thus falling short of the mark with the o - na inverse - correlation as well ( e.g. , denissenkov et al . 1997 ; denissenkov & herwig 2003 ; ventura et al . furthermore the standard agb models produce ejecta with an overall increase in cno nuclei of @xmath118 1 to 3 dex ( mainly due to the dredge - up of carbon ) , whilst observations show that the sum of cno nuclei is virtually constant , varying by a factor of only two or so at most ( e.g. , pilachowski 1988 ; smith et al . 1996 ; ivans et al . these discrepancies are based on confronting single stellar models with observations . taking the evolution of an entire cluster of stars into account in particular convolving with an imf presents further problems as the stars with the prerequisite hot hydrogen burning are few in number when using a standard imf ( e.g. , fenner et al . 2004 ) , whilst these stars need to contribute an amount of polluting material that is a large fraction of the gc mass . despite all these discrepancies it should be stressed that the agb pollution scenario should not be dismissed outright just yet , as the models are well known to have many serious uncertainties ( e.g. , the treatment of convection , overshoot , mass loss and the uncertainties in reaction rates ) . for example the non - standard models of ventura & dantona ( 2005b ) , which use a different convection model , can reproduce most of the abundance anomalies , although some overshooting and the adoption of a high mass - loss rate is required . in addition to the uncertainties in the input physics of all stellar models , numerical details can also significantly affect the results . for example , time - stepping during diffusive mixing can cause changes in the stellar structure , hence altering the temperature at the bottom of the convective envelope , and thus the resulting nucleosynthesis and yields ( siess , private communication ) . attempts have been made to reconcile the ( standard ) models with the observations but have universally ended with a need to ` tweak ' the relative effects of hbb and 3dup to an unpalatable degree ( e.g. , denissenkov & herwig 2003 ) , also suggested by fenner et al . the models using a different convection formalism ( ventura & dantona 2005b ) show a better agreement with observations but , as mentioned earlier , also need some overshooting and the adoption of a high mass - loss rate . with the results of all these stellar modelling studies in mind we have chosen to investigate the limiting case by utilising yields from agb models that do not experience any 3dup , which we have calculated ourselves . although ` turning off ' 3dup in our models is ad - hoc , it was surprisingly easy to achieve . by adding an extra physical condition for the determination of convective boundaries ( the molecular weight gradient ) via implementing the ledoux criterion rather than the schwarzschild criterion we found that no dredge - up at all occurred in our models . we note that we have not extended the convective boundaries past the formal ones given by the ledoux criterion . we leave the detailed description of the models for a separate paper , but now discuss the nucleosynthetic results relevant for the current problem . decoupling 3dup from hbb exposes the dependence of the yields on the interplay between the two processes and provides a possible solution to the mismatch between theory and observation . at the very least , by separating the effects it is a useful investigative tool . a salient result from our new stellar models , evident in table 1 , is that oxygen is now heavily depleted in the more massive stars . also of importance is the fact that there is now an opposite trend for mg ( in the same stars ) it is depleted instead of produced . furthermore , the c is depleted whilst the n yield remains high . all of these features derive from the lack of 3dup , as o is now not periodically replenished , c can not increase and fresh fuel for the mgal cycle is not available . the lack of 3dup also implies that the sum of cno nuclei will remain constant , as fresh cno is not added to the envelope and the cno burning cycles ( occurring at the base of the convective envelope ) conserve the number of cno nuclei . all these features indicate that a better fit to the observations is expected ( imf withstanding ) . also , as stated above , these models may simulate the effects of a previously ignored population of stars , the super - agb stars . the main discrepancy we now have in terms of comparing at a single mass is with na . at the high masses ( ie . temperatures ) required to deplete o and mg via the mgal and on cycles , na is not produced enough , or rather it is destroyed too much by proton captures , resulting in low yields . moving to lower masses ( temperatures ) improves the na situation but prevents o and mg from being depleted . interestingly this is the identical problem that ventura & dantona ( 2005b ) have with their best - fit ( for the na - o inverse - correlation ) model . denissenkov & herwig ( 2003 ) also report this problem . we note that the models of ventura & dantona ( 2005b ) are quite independent as they use a different treatment for convection . in a recent and very timely study ventura & dantona ( 2006 ) have investigated this problem in detail , finding that their stellar model requires either the addition of some overshooting or a reduction of the @xmath119na@xmath120ne reaction rate in order to produce the required na yield whilst maintaining the o destruction . the ventura and dantona ( 2005b , 2006 ) models , like our more ad - hoc models , otherwise match the abundance anomalies fairly well , with the strong exception of the mg isotopes , which seem resistant to any approach tried thus far . note that we do not suggest that real agb models of low metallicity do not suffer from the 3dup . rather we have used this as an investigative tool to simulate reduced 3dup , or sagb stars with little or no 3dup , or some other source of ( almost ) pure h - burning . clearly a thorough understanding , and reliable quantitative prediction of 3dup is still required , for stars of all masses and metallicities .	we numerically investigate abundance properties of the galactic globular clusters ( gcs ) by adopting a new `` external pollution '' scenario . in this framework , gcs are assumed to originate in forming low - mass dwarfs embedded in dark matter subhalos at very high redshifts ( @xmath0 ) and thus be chemically influenced by field agb stars of the dwarfs during early gc formation processes . gcs within a dwarf galaxy therefore can be formed from the mixture of ( i ) gas ejected from the field agb stars formed earlier in the dwarf and ( ii ) the interstellar gas infalling to the central region of the dwarf . in this external pollution scenario , the ratio of the total mass of infalling gas to that of agb ejecta during gc formation in a dwarf ( @xmath1 ) and the time scale of gas infall ( @xmath2 ) are the most important key parameters that can determine abundance properties of gcs . we mainly investigate the abundance inhomogeneity among light elements ( e.g. , c , n , o , na , and al ) of stars in gcs by using the latest stellar yield models of metal - poor agb stars _ with and without third dredge - up . _ our principal results for the models with no third dredge - up , which are more consistent with observations , are as follows . \(1 ) both [ n / fe ] and [ c / fe ] can be diverse among stars within a gc owing to chemical pollution from field agb stars . [ n / fe ] distributions in some gcs can clearly show bimodality whereas [ c / fe ] is monomodal in most models . [ n / fe ] distributions depend on @xmath1 such that models with smaller @xmath1 ( i.e. , larger mass fraction of agb ejecta used for gc formation ) show the [ n / fe ] bimodality more clearly . \(2 ) n - rich , c - poor stars in gcs also have higher he abundances owing to pollution from massive agb stars with he - rich ejecta . the number fraction of he - rich stars ( y @xmath3 ) is higher for the models with smaller @xmath1 and shorter @xmath2 for @xmath4 and @xmath5 yr . he abundances of stars correlate with [ n / fe ] and [ al / fe ] and anticorrelate with [ c / fe ] , [ o / fe ] , and [ na / fe ] within gcs in our models . \(3 ) although our model can much better explain the observed c - n and mg - al anticorrelations than previous theoretical models , it is in strong disagreement with the observed o - na anticorrelation . \(4 ) this model naturally provides an explanation for the large fraction of cn - strong stars without recourse to an implausible imf . based on these results for the above external pollution scenario , we discuss the long - standing problem of the cn - bimodality prevalent in the galactic gcs , the possible helium abundance inhomogeneity in these systems , and their horizontal branch morphologies . [ firstpage ] globular clusters : general globular clusters : individual ( @xmath6 centauri) globular clusters : individual ( ngc 6752) galaxies : star clusters galaxies : evolution galaxies : stellar content [ cols="^,^,^,^,^,^,^,^,^,^,^,^ " , ] * notes : * cols . ( 23 ) yield types for infalling gas and agb ejecta , ( shown in tables 1 and 2 ) respectively . cols . ( 46 ) model parameters .
the dirac equation is a fundamental base of the relativistic field theory . however , it is an important model in the non - relativistic solid state theory as well . superconductors with @xmath0pairing @xcite , the cohen - blount two - band model of narrow - gap semiconductors @xcite , @xcite , electronic spectrum of the carbon tubes form an incomplete list of the non - relativistic applications of this equation . during the last two years extremely much attention was payed to the problem of the electronic spectrum of graphene ( see for the review @xcite ) . two - dimensional structure of it and a presence of the cone points in the electronic spectrum make actual a comprehensive study of the external fields effect on the spectrum and other characteristics of the electronic states described by the dirac equation in the 2 + 1 space - time . we consider in this work the bound states of the 2 + 1 dirac equation due to the short - range perturbation . particular attention to this case stems from the effectiveness of short - range scatterers in contrast to the long - range ones : an effect of the latter is suppressed by the klein paradox @xcite . our work takes into account the obvious fact that the kohn - luttinger matrix elements of the short - range perturbation calculated on the upper and lower band wave functions are not equal in a general case . this means that in the perturbed dirac equation not only the potential but the mass perturbation can be present . the dirac equation describing electronic states in graphene reads @xcite @xmath1 where @xmath2 is the limiting velocity of the band electrons , @xmath3 are the pauli matrices , @xmath4 is the electronic spectrum gap , @xmath5 is the two - component spinor . the spinor structure takes into account the two - band nature.@xmath6 and @xmath7 are the local perturbations of the mass ( gap ) and the chemical potential . a local mass perturbation can be induced by defects in the graphene film or in the substrate @xcite . we consider here the delta function model of the perturbation:@xmath8 where @xmath9 and @xmath10 are respectively the polar coordinate radius and the perturbation radius . such short - range perturbation ( and the equivalent form @xmath11 with @xmath12 ) was used in the ( 3 + 1)-dirac problem for narrow - gap and zero - gap semiconductors in @xcite . the two - dimensional dirac problem with the scalar short - range perturbation ( [ delta ] ) ( but without the mass perturbation ) was considered in @xcite . the obtained there characteristic equation for the discrete spectrum energy contains one mistake . we correct it here and take account of the mass perturbation @xmath13 let us present the two - component spinor in the form@xmath14 \\ g_{j}\left ( r\right ) \exp\left [ i\left ( j+1/2\right ) \phi\right ] \end{pmatrix } , \label{spinor}%\ ] ] where @xmath15 is the pseudospin quantum number ; @xmath16 . in the opposite to the relativistic theory , this quantum number has nothing to do with the real spin and indicates the degeneracy in the biconic dirac point . the upper @xmath17 and @xmath18 components of the spinor satisfy the equations @xmath19 @xmath20 these equations have a symmetry : @xmath21 let us introduce the function @xmath22 it satisfies the equation:@xmath23 + \delta(r - r_{0})=0 \label{phi}%\ ] ] integrating in the vicinity of @xmath24@xmath25 we obtain the matching condition@xmath26 where @xmath27 the upper and lower component matching condition resulting from ( [ match2 ] ) reads@xmath28{c}% f_{j}^{+}\\ g_{j}^{+}% \end{array } \right ) = \overset{\wedge}{a}\left ( \begin{array } [ c]{c}% f_{j}^{-}\\ g_{j}^{-}% \end{array } \right ) , \label{matrixrelation}%\ ] ] where the matrix @xmath29 @xmath28{cc}% \cos\sqrt{a^{2}-b^{2 } } , & -\sqrt{\frac{a - b}{a+b}}\sin\sqrt{a^{2}-b^{2}}\\ \sqrt{\frac{a - b}{a+b}}\sin\sqrt{a^{2}-b^{2 } } , & \cos\sqrt{a^{2}-b^{2}}% \end{array } \right ) \label{matrix}%\ ] ] is orthogonal for @xmath30 it transmutes into the matrix@xmath28{cc}% \cosh\sqrt{b^{2}-a^{2 } } , & -\sqrt{\frac{b - a}{b+a}}\sinh\sqrt{b^{2}-a^{2}}\\ \sqrt{\frac{b - a}{b+a}}\sinh\sqrt{b^{2}-a^{2 } } , & \cosh\sqrt{b^{2}-a^{2}}% \end{array } \right ) , \label{matrix2}%\ ] ] when @xmath31 the general solution can be found solving the second - order equation obtained by excluding one of the spinor components from the equation set ( [ componenteq1 ] ) , ( [ componeq2 ] ) in the domains @xmath32 and @xmath33@xmath34 f_{j}=0 . \label{secondorder}%\ ] ] this equation is related to the bessel one . its general solution reads@xmath35 where @xmath36 @xmath37 and @xmath38 are the modified bessel functions . the constant @xmath39 in the domain @xmath32 , while @xmath40 in the domain @xmath41 expressing the @xmath42-component using ( [ componeq2 ] ) , we can write@xmath43 @xmath44 applying the matching condition ( [ match2 ] ) to the expressions ( [ phi- ] ) , ( [ phi+ ] ) we obtain the characteristic equation for the bound state energy levels : @xmath45 = \ ] ] @xmath46 \label{character1}%\ ] ] where @xmath47 this equation turns to the characteristic equation obtained in @xcite , for @xmath48 apart from the mistakenly omitted terms in the right hand side of ( [ character1 ] ) . in the opposite case of @xmath49 we have @xmath45 = \]]@xmath50 \label{character2}%\ ] ] we write these equations in another form making the symmetry ( [ symm ] ) manifest : @xmath51 = \ ] ] @xmath52 , \label{character1a}%\ ] ] @xmath53 = \]]@xmath54 \label{character2a}%\ ] ] making use of the bessel functions limiting forms for small arguments @xcite @xmath55 we can obtain a simple relation describing the asymptotic behaviour of the energy level , where the perturbation power approaches zero:@xmath56 , \label{approach}%\ ] ] where @xmath57 ( in units with @xmath58 @xmath59 this result conforms the well known general property of the two - dimensional quantum systems : a threshold for creation of the bound state is absent ; the point @xmath60 is the essentially singular point of the function @xmath61 one can see that the function @xmath62 approaches the point @xmath63 at some large enough value of @xmath64 . @xmath65 making use of the bessel function asymptotic behaviour @xcite , @xmath66 and the equation [ character2 ] , we can see that the function @xmath67 approaches the point @xmath68 when @xmath69 is large enough and @xmath70 . in the fig . 1 the electron bound state energy is presented as a function of the potential amplitude for the angular momentum quantum number @xmath71 , @xmath72 and @xmath30 inspecting this plot one can see that our analytic solution ( [ approach ] ) perfectly approximates approaching of the bound state energy value the upper band bottom , when @xmath73 approaches zero . in the fig . 2 the bound state energy is presented as a function of the mass perturbation amplitude @xmath73 for @xmath74 @xmath75 @xmath76 in the fig . 3 the electron bound state energy is presented as a function of the potential amplitude for the angular momentum quantum number @xmath71 , @xmath75 and @xmath77 we see that the energy dependence on @xmath78 is non - monotonic function , but approaching the upper band bottom takes place similarly to the case of @xmath30 in conclusion , we considered the bound electron states for the two - dimensional dirac equation with the short - range perturbation . the short - range perturbation is approximated by the delta function @xmath79 with different amplitudes in the upper and lower bands . we found the characteristic equation for the discrete energy levels . energy levels behaviour in dependence on the perturbation amplitudes was investigated both analytically and numerically . the obtained results can be useful for understanding of the graphene electron properties .	bound electron states in impure graphene are considered . short - range perturbations for defect and impurities of the types `` local chemical potential '' and `` local gap '' are taken into account .
theory of quantum information is a generalization of the shannon theory of information that takes into account the physical nature of the information carrier . previously , information was assumed to be encoded in `` classical '' physical states of a system that are distinct and infinitely copyable . classical physics remains only an approximation to the underlying quantum nature of matter . to understand the true limits that the laws of physics will place on our ability to communicate and process information , the quantum behavior of the information carrier must be addressed . in the classical theory of transmission through noisy channels , the maximum asymptotic rate that information may be transmitted through a discrete memoryless channel ( dmc ) is given by the shannon capacity theorem @xcite . the capacity for a dmc is unchanged even with the inclusion of a number of additional resources , most notably , a noiseless feedback channel @xcite . this results in a robust measure for the capacity of any memoryless channel . for noisy quantum channels the situation is somewhat different , and there currently exists a plethora of different capacities for any type of memoryless quantum channel . by demonstrating relationships between the various capacities , and invariance under the addition of certain resources , the number of different capacities may be reduced to a smaller number . any quantum channel may be used to send quantum information , in the form of intact quantum states or entanglement , or used to send classical information encoded in quantum states . a noisy quantum channel may also be augmented by the use of auxiliary resources , such as one - way or two - way classical side - channels , or prior shared entanglement between sender and receiver . a channel augmented by a one - way forward classical channel has been shown to have the same quantum capacity as a channel without any classical communication @xcite . however , augmentation by classical feedback or a two - way classical channel has been demonstrated to increase the quantum capacity in certain circumstances @xcite . the quantum channel capacities that most resemble the shannon capacity in both form and behavior appear to be the quantum and classical entanglement assisted capacities of the channel @xcite . the formula describing either entanglement assisted capacity is a quantum analogue of the shannon formula up to a constant factor , and both of these capacities are unchanged by the addition of a noiseless quantum feedback channel @xcite . for quantum capacities the number of capacities has been reduced to four , @xmath0 the unassisted quantum capacity , @xmath1 the capacity with a classical feedback channel , @xmath2 the capacity with a two - way classical side - channel , and @xmath3 the entanglement assisted quantum capacity . the quantum capacities obey the following relationships @xmath4 and @xmath5 , with known channels for which @xmath6 and @xmath7 . whether or not there exists a channel for which @xmath8 remains an open question . when sending classical information through quantum channels additional resources such as a quantum or classical feedback channel , or prior shared entanglement , may be utilized . the use of shared entanglement gives rise to the entanglement assisted capacity @xmath9 , which is equivalent to the classical capacity with a quantum feedback channel . whether classical feedback can increase the classical capacity is the question that is addressed in this paper . it is obvious that the capacity for a channel with feedback is at least as great as the channel without feedback , @xmath10 and to prove equality it is only necessary to show the reverse inequality . in this paper the use of classical feedback across non - entangled input states is shown to provide no increase in the classical capacity of noisy quantum channels . additionally , if the quantum channel is an entanglement breaking channel then feedback can not increase the classical capacity , even for feedback across input states that may be entangled between channel uses . these results give a partial analogy to the use of feedback in the shannon theory , and the remaining open case of feedback across entangled input states is examined , with a conjecture that such feedback protocols will not increase the channel capacity . a proof that feedback can not increase the classical capacity for the remaining case , of feedback across entangled input states , will result in two well defined classical capacities for memoryless quantum channels , the capacity @xmath11 where there is no prior shared entanglement , and the entanglement assisted capacity @xmath12 , thus simplifying our current understanding of communication through quantum channels . given a quantum channel @xmath13 , the shannon mutual information between a message memory and the average state after transmission is bounded above by the von neumann mutual information between the states . a quantum memory for the message states may be constructed , without loss of generality , using known orthogonal pure states @xmath14 corresponding to each message @xmath15 . the memory state acts classically , in the sense that it may measured without error and can be copied arbitrarily many times by measurement and state preparation . the total state consisting of the message memory and the input states is then , @xmath16 with each state @xmath17 corresponding to the message @xmath15 , where the message @xmath15 is sent with _ apriori _ probability @xmath18 . following transmission of the quantum state through the channel the resultant combined memory and output state is given by , @xmath19 any local measurements on the message memory and output state must give a distribution of measurement outcomes such that the shannon mutual information is bounded by , @xmath20 @xcite , with the von neumann mutual information between @xmath21 and @xmath22 defined by @xmath23 . here , @xmath24 is the shannon mutual information between the measurement outcomes on the memory state and the measurement outcomes on the output state of the channel @xmath25 . expanding the von neumann mutual information for the alphabet of states @xmath26 , with average density matrix @xmath27 , obtains the bound @xcite , @xmath28 where the inequality follows from the concavity of the conditional entropy @xmath29 , with @xmath30 the conditional entropy of @xmath21 given @xmath22 @xcite . the last line of ( [ eqn : holevo_deriv ] ) follows from the additivity of the entropy for product states @xmath31 . the holevo schumacher westmoreland ( hsw ) theorem states that a rate equal to the maximum over all such alphabets is asymptotically attainable with vanishing probability of error @xcite . therefore the classical information capacity for a quantum channel using a product state alphabet across channel uses , is given by , @xmath32\ ] ] which can be generalized to give the classical capacity of a quantum channel , @xmath33 with @xmath34 $ ] the hsw capacity for an alphabet in the hilbert space of maximum dimension @xmath35 , that is , an alphabet that is a product state over uses of blocks of @xmath36 channels but may be entangled across different channel uses within the same block . the additivity of the hsw capacity is still an open question , although for certain classes of channels it is known to be additive @xcite , @xmath37 . to derive the upper bound on the capacity we use the technique of attaching a copy of the message , encoded in mutually orthogonal pure states , to the message states to be transmitted . in addition , for each use of the feedback channel we add a correlated set of quantum operations , where each operation on the first output is correlated to a trace preserving operation on the second state . the proof works by induction , showing that for any single step of a feedback protocol the maximum increase of the mutual information between the sender and receiver can not exceed the hsw bound . the maximum mutual information generated for a multi - step feedback protocol can not exceed the sum of the mutual information gained from each step of the protocol , and hence the maximum rate for feedback codes utilized across non - entangled input states or feedback codes for entanglement breaking channels can not exceed the classical capacity of the channel without feedback . in order to prove the result for the most general type of protocol for product state inputs , the channels may be assumed to be of the form @xmath38 , where for multiple use of a single channel @xmath39 we can assume @xmath40 and @xmath41 , with @xmath42 and @xmath36 arbitrary . the initial total state is of the form , @xmath43 with message @xmath15 being sent with _ probability @xmath18 . the state @xmath44 is then sent through the first channel to produce the new state , @xmath45 this is then followed by the _ feedback operation _ , which without loss of generality may be represented by classically correlated operations on the feedback state @xmath46 and a combined operation on the memory state and next state to be transmitted @xmath47 , where the copy of the message memory must remain invariant under the operation @xmath48 . the total state after the feedback operation and transmission of the second state through the channel is then , @xmath49 where , @xmath50 \nonumber \\ & = \sum_{jk } b^{j}_{q_1}\omega\rho^i_{q_1}b^{j\dag}_{q_1 } \otimes a^{(ij)k}_{q_2 } \rho^i_{q_2 } a^{(ij)k\dag}_{q_2}.\end{aligned}\ ] ] for each @xmath51 the state is an action utilizing only local operations and classical communication ( locc ) . any locc action on a separable state necessarily results in a separable state , hence each @xmath52 is separable , and the convex sum of these separable states @xmath53 is also separable . as each state following the feedback operation is separable , it may be written as a convex sum over product states , @xmath54 where @xmath55 and @xmath56 . the total state in ( [ eqn : final_state ] ) , following both the feedback operation and the second transmission , may then be rewritten in the form , @xmath57 the mutual information between the message memory and the combined output states may be rephrased in terms of the reduced mutual information and the conditional mutual information , @xmath58 with the conditional mutual information defined in terms of the conditional entropies by , @xmath59 the conditional information for quantum states differs from the classical counterpart , in that , the two terms on the right hand side of ( [ eqn : cond_mut_def ] ) can each be negative , but only for entangled states . the first term on the right hand side of ( [ eqn : mutual_exp ] ) may be explicitly written as , @xmath60 where @xmath61 is the hsw capacity of the first channel . the first inequality follows from the fact that any quantum operation @xmath62 acting on part of a bipartite state can not increase the von neumann mutual information @xcite , and the second inequality follows from the definition of @xmath63 . the first inequality in ( [ eqn : first_bound ] ) incorporates any information gained from the measurement outcome during the feedback protocol . this may be seen by attaching an initially pure ancilla state @xmath64 which after the new operation on the output state and ancilla @xmath65 , gives a classical copy of the measurement outcome in the ancilla state @xmath66 . the mutual information between the message memory and the output state combined with the ancilla following the measurement operation @xmath65 , must also necessarily be less than the initial mutual information between the message memory and the initial output state and ancilla state . because the ancilla is initially in a product state , the first inequality in ( [ eqn : first_bound ] ) then follows from the additivity of the entropies of product states @xmath67 . obtaining the required bound on the final term of ( [ eqn : mutual_exp ] ) is only slightly more difficult , and we begin by expanding the terms according to the basic definition in terms of the conditional von neumann entropies , such that , @xmath68 where the inequality follows from the fact that conditioning can not increase the entropy in the first term @xcite . as the conditional entropy is concave , the second term @xmath69 is convex , and from the decomposition of each of the states @xmath52 as a separable state , the bound @xmath70 is obtained . substituting ( [ eqn : cond_term2 ] ) into ( [ eqn : conditional_mutual ] ) then gives , @xmath71 with @xmath72 , and the decomposition , @xmath73 the second inequality in ( [ eqn : second_chi ] ) follows from the fact that the hsw capacity is the maximum over all such ensembles . hence , the total capacity of the feedback protocol across the channels is bounded above by the separate hsw capacities of the channels , @xmath74 the same result follows for an arbitrary feedback protocol , across any product states inputs , by induction . the upper bound on any feedback protocol in the previous section is easily extended to include the class of input states that are non - entangled , that is , all separable states . this is due to the fact that any separable input states will remain separable after any locc feedback operation , and can therefore be written in the form of ( [ eqn : total_separable ] ) . therefore feedback across non - entangled input states can not increase the maximum asymptotic rate at which classical information may be sent through a memoryless quantum channel . an entanglement breaking quantum channel is one which can not transmit entanglement . if part of any bipartite entangled state is transmitted through the channel , then the bipartite state following transmission is necessarily separable . explicitly , for any initial state @xmath75 , the output @xmath76 given by , @xmath77 is always separable . the classical capacity for entanglement breaking channels has previously been shown to be additive , and hence the classical information capacity for such channels is simply @xmath78 @xcite . the proof that feedback can not increase the classical capacity of an entanglement breaking channel is straightforward . after the first state @xmath44 is sent through an entanglement breaking channel , the total state is necessarily separable and may be written in the form , @xmath79 for which the feedback operation will result in a new separable state , @xmath80 following transmission of the second state , the mutual information between @xmath81 and @xmath82 is bound by , @xmath83 with @xmath84 . the total mutual information over the two channels is therefore bound by , @xmath85 and consequently feedback can not increase the classical information capacity of entanglement breaking channels . the derivation from ( [ eqn : first_use_state2 ] ) to ( [ eqn : feed_subadd2 ] ) is essentially a less detailed , but otherwise almost identical , version of ( [ eqn : first_use_state ] ) to ( [ eqn : feedback_bound ] ) , with the major difference being that in ( [ eqn : first_use_state ] ) the alphabet states are product states rather than separable states after the use of the channel . it may be noted , however , that for a feedback protocol to possibly exceed the bound for the non - feedback capacity the _ average total states _ , @xmath86 and @xmath87 , must remain entangled between the states that have been sent and the states held by the sender , at some step of the protocol . any channel for which all possible ensembles do not obey this property must therefore have @xmath88 , which is the defining characteristic of any entanglement breaking channel . the use of classical feedback for the transmission of classical information through a memoryless quantum channel has been shown to give no increase in the capacity of the channel when the feedback is used across non - entangled input states . additionally , it has been shown that feedback can not increase the classical capacity of entanglement breaking channels . the question of whether or not feedback can increase the capacity of memoryless quantum channels when used across entangled input states remains open .	in shannon information theory the capacity of a memoryless communication channel can not be increased by the use of feedback from receiver to sender . in this paper the use of classical feedback is shown to provide no increase in the unassisted classical capacity of a memoryless quantum channel when feedback is used across non - entangled input states , or when the channel is an entanglement breaking channel . this gives a generalization of the shannon theory for certain classes of feedback protocols when transmitting through noisy quantum communication channels . quantum information , channel capacity , quantum channels , feedback .
in the following , we give the series expansions in the different phases for the ground - state energy per plaquette @xmath75 and the quasiparticle gaps @xmath107 , for positive and negative signs of the dimensionless parameter @xmath108 respectively . for the sake of clarity , we give below the numerical values of the coefficients with 16 digits . the ground - state energy per plaquette @xmath75 near @xmath26 ( @xmath109 , @xmath110 ) has been obtained up to order @xmath111 using operator perturbation theory @xcite , whereas quasiparticle gaps @xmath112 were obtained up to order @xmath113 using perturbative continuous unitary transformations . @xcite @xmath114\\ \nonumber\\ \delta^+ / j_{\rm{p } } = & \ 1 -1.658359213500126 \,t-2.029179606750063 \,t^2 - 3.107113095525145 \,t^3 - 8.042597266963313 \,t^4\nonumber\\&\ -19.16249885423558 \,t^5 -58.31720409052607 \,t^6 - 164.4421257647495 \,t^7 - 528.5318111014412 \,t^8\nonumber\\&\ -1615.453328025113 \,t^9 , \displaybreak[0]\\ \nonumber\\ \delta^- / j_{\rm{p } } = & \ 1 + 0.8291796067500631 \,t+0.1145898033750315 \,t^2 + 0.5110332556124590 \,t^3 + 0.4044760408194119 \,t^4\nonumber\\&\ + 0.9554382719956335 \,t^5 + 1.784752477740017 \,t^6 + 4.523961920423115 \,t^7 + 11.17294663306187 \,t^8\nonumber\\&\ + 31.20020681009114 \,t^9.\end{aligned}\ ] ] the ground - state energy per plaquette @xmath75 near @xmath57 ( @xmath115 , @xmath116 ) has been obtained up to order @xmath117 using a partitioning technique provided by lwdin @xcite , whereas quasiparticle gaps @xmath112 were obtained up to order @xmath111 using operator perturbation theory @xcite on appropriate periodic clusters . @xmath118\\ \nonumber\\ \delta^{\pm } / j_{\rm{e}}=&\ 6 - 4.472135954999579\cdot 10^{-1 } \,t^{-1}-5.916925468334595\cdot 10^{-2 } \,t^{-2}-1.532368203449658\cdot 10^{-2 } \,t^{-3}\nonumber\\&\ -1.826128841179490\cdot 10^{-3 } \,t^{-4}-2.141218566177655\cdot 10^{-4 } \,t^{-5}-7.057273308076495\cdot 10^{-5 } \,t^{-6}\nonumber\\&\ -7.713120260517609\cdot 10^{-6 } \,t^{-7}-1.822619200222926\cdot 10^{-6 } \,t^{-8}-1.739394034185955\cdot 10^{-7 } \,t^{-9}\nonumber\\&\ -1.068512192171555\cdot 10^{-7 } \,t^{-10}-1.583732801424164\cdot 10^{-8 } \,t^{-11}.\end{aligned}\ ] ] the ground - state energy per plaquette @xmath75 near @xmath71 ( @xmath119 , @xmath116 ) has been obtained up to order @xmath113 using partitioning techniques @xcite , whereas quasiparticle gaps @xmath112 were obtained up to order @xmath120 using perturbative continuous unitary transformations . @xcite @xmath121\\ \nonumber \\ \!\!\delta^+ / \left(-j_{\rm{e}}\right ) = & \ 1 - 7.331262919989905\cdot 10^{-1}\,t^{-1}+2.627515502196252\cdot 10^{-1}\,t^{-2}-1.016556730679853\cdot 10^{-1}\,t^{-3}\nonumber\\&\ + 1.966652521926989\cdot 10^{-2}\,t^{-4}-7.070423645089451\cdot 10^{-2}\,t^{-5}+1.098516898950685\cdot 10^{-1}\,t^{-6 } , \displaybreak[0]\\ \nonumber \\ \!\!\delta^- / \left(-j_{\rm{e}}\right ) = & \ 1 + 2.668737080010095\cdot 10^{-1}\,t^{-1}-2.048165955279984\cdot 10^{-1}\,t^{-2}-7.571315893653220\cdot 10^{-2}\,t^{-3}\nonumber\\&\ + 3.964262957878962\cdot 10^{-2}\,t^{-4}+7.257956045336907\cdot 10^{-2}\,t^{-5}-6.691149658050659\cdot 10^{-2}\,t^{-6}. \ ] ]	we examine the zero - temperature phase diagram of the two - dimensional levin - wen string - net model with fibonacci anyons in the presence of competing interactions . combining high - order series expansions around three exactly solvable points and exact diagonalizations , we find that the non - abelian doubled fibonacci topological phase is separated from two nontopological phases by different second - order quantum critical points , the positions of which are computed accurately . these trivial phases are separated by a first - order transition occurring at a fourth exactly solvable point where the ground - state manifold is infinitely many degenerate . the evaluation of critical exponents suggests unusual universality classes . quantum phases of matter are often well described by local order parameters and landau - ginzburg symmetry - breaking theory is an efficient tool to analyze transitions between these phases . however , in the late 1980s , a new class of phases that can not be understood in terms of local symmetries has emerged in the context of high - temperature superconductivity @xcite . these phases , dubbed topological because of their sensitivity to the system topology , have stimulated many studies in different domains ( see ref . for a recent review ) . one of the most intriguing properties of topologically ordered phases is that they are robust against local ( not too strong ) perturbations @xcite . this stability makes them especially appealing for quantum computation @xcite as well as good candidates for quantum memories @xcite . several experiments have been proposed to realize the so - called topologically protected qubits @xcite . in this perspective , a theoretical characterization of the robustness of topological phases under strong perturbations as well as the nature of the phase transitions signaling their breakdown is undoubtedly an important issue . thanks to recently proposed exactly solvable lattice models realizing various topological phases of matter @xcite , this program has been undertaken for several models @xcite . the main purpose of this letter is to go one step beyond , by studying the phase diagram of a paradigmatic 2d non - abelian model . we consider the levin - wen model @xcite on the honeycomb lattice with fibonacci anyons ( the golden string - net model ) in the presence of the same perturbation as the one introduced in ref . . we determine the extension of the _ doubled _ fibonacci ( dfib ) topological phase and show that it is separated from two other nontopological phases via second - order transitions that are analyzed in detail . _ hilbert space _ following the levin - wen construction , we consider a honeycomb lattice with anyonic degrees of freedom living on its edges . in the fibonacci string - net model , these local ( microscopic ) degrees of freedom can be in two different states @xmath0 or @xmath1 . the hilbert space @xmath2 is restricted to states that satisfy the so - called branching rules stemming from the non - abelian fusion rules @xmath3 at each vertex of the honeycomb lattice , the fusion rules must not be violated ; i.e. , if one edge is in state @xmath1 , then at least one of the two other edges must be in the same state . for an arbitrary trivalent graph with @xmath4 vertices , the dimension of the hilbert space is then given by @xmath5 where @xmath6 is the golden ratio ( see , e.g. , ref . ) . _ model. _ we study the following hamiltonian @xcite , @xmath7 the first term is the string - net hamiltonian introduced by levin and wen @xcite . it involves the projector @xmath8 onto states with no flux @xmath9 through @xcite . the second term is diagonal in the basis introduced above since @xmath10 is the projector onto state @xmath0 on edge @xmath11 . to help the reader to grasp the physical content of this hamiltonian , let us mention what happens if one replaces eq . ( [ eq : fusion ] ) by the simpler abelian @xmath12 fusion rule @xmath13 . the model then becomes kitaev s toric code @xcite on the honeycomb lattice , restricted to the charge - free sector ( because of the branching rules ) , in the presence of a magnetic field @xmath14 in the @xmath15 direction . indeed , one can write the hamiltonian in terms of pauli matrices , with @xmath16 and @xmath17 . identifying plaquette fluxes with spin @xmath18 variables as done in refs . for the square lattice , the hamiltonian can further be mapped onto the transverse - field ising model on the triangular lattice ( of plaquettes ) , with coupling @xmath14 and transverse - field @xmath19 . limiting cases. _ for convenience , let us set @xmath20 and @xmath21 . to our knowledge , contrary to the ladder geometry studied in ref . , the hamiltonian @xmath22 is exactly solvable only at the four points for which @xmath23 is a multiple of @xmath24 . in the following , we discuss the low - energy spectrum of @xmath22 as well as the corresponding phases around these special points and give some arguments in favor of transitions between them . @xmath25 @xmath26 : for @xmath27 and @xmath28 , the model reduces to the golden string - net model . ground states @xmath29 are flux - free states satisfying @xmath30 for all @xmath31 and thus have an energy per plaquette ( @xmath32 being the number of plaquettes ) . their degeneracy depends on the system topology which is the most salient property of a topological phase . for the fibonacci string - net model on any trivalent graph , the ground state is unique on a sphere @xcite whereas it is fourfold degenerate on a torus . interestingly , one can also compute the degeneracy of the @xmath33 excited states ( with energy @xmath34 ) @xmath35 where we introduced the famous fibonacci sequence defined for any integer @xmath36 by @xmath37 , with and @xmath38 . the integers @xmath39 depend on the surface considered . for instance , one has on a sphere whereas @xmath40 on a torus . equation ( [ eq : deg_fibo ] ) shows that for the fibonacci theory , _ an odd number of excitations can exist on a compact surface _ contrary to the charge - free toric code discussed above where fluxes are always created and annihilated by pairs . note that the binomial coefficient simply arises from the different ways to choose @xmath41 plaquettes carrying the flux excitations among @xmath42 . products of _ two _ @xmath43 s stem from the fact that the `` emergent '' flux excitations are not the microscopic fibonacci anyons but are achiral combinations of _ two _ fibonacci anyons ( details will be given in ref . ) . the non - abelian topological phase in the vicinity of @xmath26 is described by a dfib theory @xcite . excitations have a trivial topological spin @xcite and can fuse to the vacuum ( also called trivial particle ) @xcite . as such , they can also be considered as bosons @xcite and hence condense . @xmath44 @xmath45 : for @xmath28 and @xmath46 , the low - energy spectrum is very different . indeed , in this case , the ground - state manifold is @xmath47-fold degenerate and spanned by all states @xmath48 satisfying @xmath49 for all @xmath31 . as discussed above , this degeneracy depends on the topology through its indices ( @xmath50 ) so that one might be tempted to consider the system as topologically ordered . however , the local operator @xmath51 couples the ground states and splits the degeneracy for any @xmath52 . as a consequence , the system can not be considered as topologically ordered @xcite . owing to this huge degeneracy , we have not been able to analyze the vicinity of this point . however , numerical results obtained by exact diagonalizations clearly show that @xmath53 the degeneracy is lifted as soon as the coupling @xmath54 and @xmath55 the ground state for @xmath56 is unique and adiabatically connected to the polarized ground states found at @xmath57 and @xmath58 , respectively , ( see discussion below ) . this result is in stark contrast with the scenario described in ref . on the ladder where a gapless phase is observed for @xmath59 $ ] . in addition , as can be seen in fig . [ fig : dgse ] ( central panel ) , we found a jump in @xmath60 at @xmath45 ( for all system sizes ) indicating that the two gapped phases ( @xmath61 and @xmath62 ) are separated by a first - order phase transition . @xmath63 @xmath57 : for @xmath64 , the hamiltonian @xmath22 is diagonal in the canonical basis of states satisfying the branching rules . for @xmath65 , the ground state is unique whatever the topology and corresponds to the fully polarized state where all edges are in the state @xmath0 ( with eigenenergy @xmath66 , where @xmath67 is the total number of edges ) . first excited states are obtained by flipping six links around one hexagon . they behave as trivial hard - core bosons that become dynamical when the coupling @xmath68 is switched on . thus , near @xmath57 , the system is gapped but not topologically ordered , making the occurrence of a phase transition in the interval @xmath69 $ ] compulsory . @xmath70 @xmath71 : for @xmath72 and @xmath64 , the hamiltonian is also diagonal and the unique ground state is the fully polarized state where all edges are in the states @xmath1 . note that such a state would be forbidden by the abelian @xmath12 fusion rules . first excited states are obtained from the ground state by flipping a single link . as previously , these localized excitations are trivial hard - core bosons that become mobile when @xmath73 so that one expects a phase transition in the interval @xmath74 $ ] . _ phase diagram. _ to determine the zero - temperature phase diagram , we combined two different approaches . first , we performed high - order series expansions in the thermodynamical limit by means of several methods @xcite , around the exactly solvable points @xmath26 , @xmath24 , and @xmath58 described above . this yields the ground - state energy per plaquette @xmath75 as well as the quasiparticle dispersion from which the low - energy gap @xmath76 can be extracted . lengthy expressions of these series expansions can be found in the supplemental material . this method allows one to accurately compute the critical couplings for which the gap vanishes . these points are associated to second - order transitions but might not be relevant if first - order transitions are present ( see refs . for details about this issue in a similar context ) . second , we perform exact diagonalizations ( ed ) for lattices with periodic boundary conditions . as can be seen in figs . [ fig : gse ] and [ fig : gap ] , series expansions and ed data are in very good agreement except in the vicinity of the transition points where finite - order and finite - size effects are important . combining these two methods ( ed and series expansions ) we found that _ the dfib topological phase near @xmath26 ranges from to @xmath77 . _ as we shall now argue , we associate these two points to second - order transitions . the first piece of evidence pleading in favor of such a scenario follows from ed and is the behavior of @xmath78 that clearly decreases with the system size near these points ( see left and right panels in fig . [ fig : dgse ] ) . in addition , the position of the low - energy gap minimum as well as the topological degeneracy splitting shown in fig . [ fig : gap ] lie in the same region as the position of the minimum of @xmath78 . let us also note that we did not find any relevant level crossing in the excitation spectrum that could lead to a first - order transition . the second argument comes from the high - order perturbation theory . as can be seen in the supplemental material , the series behave very differently for positive and negative @xmath14 . so , let us first discuss the most favorable case @xmath65 ( @xmath27 ) . a close inspection of the series expansion near @xmath26 and @xmath57 reveals three important features : ( i ) the sign of each coefficient is the same in all series ; ( ii ) series of the ground - state energy intersect in two order - dependent points the possible merging of these points , in the infinite - order limit , being a signal of a second - order transition since series then have to be tangential at the critical point ; ( iii ) series of the gap intersect in a unique ( still order - dependent ) point , see , e.g. , the left part of fig . [ fig : gap ] . we emphasize that the value of the gap at this crossing point decreases when the order increases and eventually vanishes in the infinite - order limit . in fig . [ fig : scaling ] , we plotted the position of these crossing points as a function of the ( inverse ) order as well as the position of the minimum of the low - energy gap and of @xmath78 as a function of @xmath79 computed from ed results . as can be seen , all data seem to converge to the same point @xmath80 $ ] in the infinite - order ( size ) limit , providing a smoking - gun evidence of a second - order . of the corresponding series expansion [ crossing points of @xmath75 ( dots ) , crossing points of the gap ( triangles ) ] or @xmath79 [ minimum of the gap ( squares ) , minimum of @xmath78 ( diamonds ) ] . lines are power - law fits obtained by choosing @xmath81 such that it maximizes the correlations for the crossing points of @xmath75 ( solid lines ) that are the most accurate results in this work . these crossing points have been obtained from order @xmath36 series around @xmath26 and order @xmath82 series around @xmath57 . for the crossing points of the gap , we used order @xmath36 series around both limits . ] unfortunately , the case @xmath72 ( @xmath27 ) is more involved for three reasons . first , series expansions of @xmath75 and @xmath76 in this region have alternate signs so that the previous criteria based on crossing points can not be used . second , contrary to the case @xmath65 , the momentum minimizing the dispersion of the low - energy quasiparticles is not at the @xmath83 point and only belongs to the reciprocal lattice of @xmath84 systems , @xmath85 . the only system with such characteristics considered in this study is the @xmath86 lattice so that one can not perform any reliable analysis from ed data near @xmath87 . third , because of the nature of the low - energy states , maximum orders reachable around @xmath71 are smaller than around @xmath57 ( see supplemental material ) . nevertheless , using the gap series expansion around @xmath26 , it is possible to perform ( dlogpad ) resummations that lead to a position of the critical point @xmath88 $ ] . note that the same methods for @xmath65 would lead to @xmath89 $ ] . this value differs from the one proposed above only by a few percent and it overestimates the extension of the topological dfib phase . consequently , as we have no alternative approach , we roughly estimate that @xmath87 lies in the range @xmath90 $ ] . _ critical exponents. _ the obvious question that arises next concerns universality classes associated to the transition points . in the absence of a local order parameter , the only meaningful critical exponents for topological phase transitions are those associated to the spectrum . let us remind that for a second - order transition , the gap vanishes , at large linear system size @xmath91 and at the critical point , as @xmath92 where @xmath93 is the dynamical exponent . in the thermodynamical limit , one further has @xmath94 and @xmath95 . as already explained , ed are only useful quantitatively around @xmath81 although restricted to systems of small sizes . using the gap data , we find @xmath96 . a finite - size analysis of @xmath75 yields a surprisingly good data collapse for the @xmath86 and @xmath97 systems , with @xmath98 , @xmath99 , @xmath100 and @xmath101 @xcite . the rather large value of @xmath102 might be responsible for the quality of the data collapse . we emphasize that these values are compatible with the previous estimate of @xmath81 as well as with the hyperscaling relation @xmath103 . the above exponents are furthermore in agreement with dlogpad resummations of the series expansion around @xmath26 which yield @xmath104 $ ] . we discard results from resummations around @xmath57 because these results are not as well converged , since for a given order , clusters needed to compute series around this point are twice as small as those needed around @xmath26 . note that as usual , extracting @xmath102 from series of @xmath75 does not give any conclusive result . concerning the critical behavior at @xmath87 , we only use dlogpad resummation around @xmath26 and we obtain a gap exponent . _ outlook. _ it is difficult to provide some error bars concerning these values . to estimate these errors , we performed similar series expansion analysis for the fibonacci ladder ( for which exponents are known exactly @xcite ) and for the 2d @xmath12 string - net model ( having either ising or @xmath105 transitions depending on the sign of @xmath14 ) . the results we obtained @xcite lead us to conclude that critical exponents are to be considered with a precision of about @xmath106 . as a conclusion , we found two different second - order transitions with universality classes that , to the best of our knowledge , are as yet unknown in the context of topological phase transitions . let us mention that a critical dfib wave function has been proposed @xcite but its relevance for the present problem requires further studies @xcite . to gain more understanding about these transitions , different approaches could be used , e.g. , variational methods or monte carlo simulations , although a naive implementation of the latter should suffer from the sign problem . it would also be worth studying similar models with a dfib phase @xcite as well as different topological phases . finally , another important issue concerns transitions between two distinct topological phases @xcite . given the ubiquity of fibonacci anyons in many different physics domain such as topological quantum computation , condensed matter , or atomic physics @xcite , we hope that the present work will stimulate such investigations . we thank k. coester , b. douot , m. kamfor , and zuber for fruitful discussions . k. p. schmidt acknowledges esf and eurohorcs for funding through his euryi . 42ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/physrevb.39.11413 [ * * , ( ) ] link:\doibase 10.1103/physrevb.40.7387 [ * * , ( ) ] link:\doibase 10.1142/s0217979290000139 [ * * , ( ) ] @noop link:\doibase 10.1016/s0003 - 4916(02)00018 - 0 [ * * , ( ) ] link:\doibase 10.1063/1.3490195 [ * * , ( ) ] @noop link:\doibase 10.1063/1.1499754 [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/75/7/072001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.016803 [ * * , ( ) ] link:\doibase 10.1103/physrevb.71.045110 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.070602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.030502 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/84/17004 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.033109 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.081104 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.107203 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.195104 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/14/2/025005 [ * * , ( ) ] link:\doibase 10.1038/nphys1396 [ * * , ( ) ] @noop link:\doibase 10.1088/1367 - 2630/13/4/045014 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.134439 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.045414 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/14/1/015004 [ * * , ( ) ] link:\doibase 10.1088/1751 - 8113/46/10/105002 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.075107 [ * * , ( ) ] @noop link:\doibase 10.1007/s00220 - 009 - 0757 - 9 [ * * , ( ) ] @noop link:\doibase 10.1016/j.aop.2010.06.003 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/13/6/065001 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.125434 [ * * , ( ) ] link:\doibase 10.1016/j.aop.2010.08.001 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.045316 [ * * , ( ) ] link:\doibase 10.1016/j.aop.2008.11.002 [ * * , ( ) ] link:\doibase 10.1063/1.1724312 [ * * , ( ) ] \doibase doi:10.1088/0022 - 3719/10/8/031 [ * * , ( ) ] link:\doibase 10.1007/s100510050026 [ * * , ( ) ] link:\doibase 10.1016/j.aop.2008.04.011 [ * * , ( ) ] @noop link:\doibase 10.1103/physreva.86.041601 [ * * , ( ) ]
* in loving memory of my late father anwar * * my mother adiba * * jacob avner and daniel * the purpose of this book is to publicize both the random dynamics program as well as a possible approach to it using the theory of multicoloured random graphs , by giving a flavour of the properties of these graphs together with a hint of the associated mathematics . the random dynamics project is an approach to physics that differs from the usual bottom - up approach of building on known theory by unifying different phenomena into one theory . it is in the spirit of smolin s clearly - argued @xcite need for radical rather than incremental approaches to furthering our understanding of how things work . its premise is that all symmetries , including the most fundamental ones such as lorentz and gauge invariance , can be derived without having to be assumed , even if the most fundamental laws at very high energies do not possess them . many symmetries such as charge conjugation , parity and strangeness are considered to arise from the _ gauge field theories _ of electromagnetic and strong interactions , whilst others arise from taking the limits in physical models such as for small mass - energies . the _ gauge group _ is a group of gauge transformations between the possible gauges . _ gauge symmetry _ itself is perhaps not a true symmetry but a redundancy in our descriptions . however as well as its use in build theories it has been a mechanism in explaining fundamental issues such as why the weak charge depends on handedness and why there is parity violation in the standard model ( sm ) ? it has been argued @xcite that gauge invariance protects quarks and leptons from gaining a fundamental mass that is large compared to the electroweak scale , and so their gauge charges must depend on handedness ; furthermore parity conservation in the electromagnetic and strong interactions is derived . the standard model comprises three leptons @xmath0 , three neutrinos @xmath1 , three up - type quarks @xmath2 , three down - type quarks @xmath3 . these undergo gauge field theory interactions mediated by the force carrier bosons , the photon , gluons and @xmath4 and @xmath5 bosons . particles that have mass gain it through a condensing higgs boson . approximate sm symmetries such as chiral symmetry and gell - mann and neeman s @xmath6 flavour symmetry @xcite can be derived in the limit as quark masses become ever smaller @xcite . the handedness of a particle gives the direction of its spin vector along the direction of motion , and the sm prediction that neutrinos always have left - handed spin means that its spin vector always points in the opposite direction to its momentum vector . if there were only left - handed neutrinos , as required by the sm , then they would be massless for according to special relativity a neutrino with mass can not travel at the speed of light . then a faster moving observer could therefore overtake the spinning massive neutrino and would see it moving in the opposite direction and thus it would thus appear right - handed . so the non - zero neutrino mass and mixing that has been observed is evidence for new physics beyond the sm . although the tri - bimaximal mixing pattern may now have been excluded , the discrete family symmetry approach is still favoured after experimental measurements , it could still be a possible good first approximation to the observed data . the alternating group of even permutations of 4 objects @xmath7 is the simplest symmetry group to leading order that leads to tri - bimaximal mixing . altarelli , feruglio and merlo s speculation on the origin of @xmath7 as a subgroup of the modular group @xmath8 is in line with our suggestion that discrete sugroups in particle physics may arise as homomorphic images of this group . king and luhn have written a recent review article on discrete family symmetry approaches to modeling neutrino mass and mixings @xcite . on his website , holger nilesen , the originator of the random dynamics philosophy states : `` in the random dynamics scheme the predicament of the poor physicist is taken very seriously , as it is postulated that we simply can not deduce what the physics is like at the highest energy levels . therefore , in order to avoid carrying along the usual tacit assumptions in fabricating a theory of the fundamental physics , we consider a most general , `` generic '' mathematical structure which we equip with some general properties and notions . the ambition is to create a ( as explicit as possible ) list of assumptions , in order to pin down what conceptual luggage we can not do without , if we have the ambition of formulating a theory for the fundamental physics , which at our low energy level appears as the physics we observe , as described by the standard model . '' ( http://www.nbi.dk/ kleppe / random/////qa / poor.html ) one motivation for the top - down random dynamics program are the laws of thermodynamics which were at first built upon certain principles but are now understood to be a reflection of the behaviour of large numbers of randomly - moving particles . symmetry has been identified as a key theme in understanding the principles underlying natural phenomena in theoretical physics . it represents the idea that observed physics is invariant under some symmetry transformation . the grand unification program in particular postulates that a larger gauge group beyond that relevant to the standard model ( sm ) corresponds to a particular force and that at lower energies this group breaks down to the sm group corresponding to the forces already discovered . random dynamics assumes that any extension of the sm can not be inferred from sm itself . the random dynamics project requires starting from general assumptions about an abstract relational mathematical construct and using very generic arguments because the idea is that the details of the very complicated laws are almost unimportant , only that we study them in some limits such as that of low energy . we take seriously their proposal to begin with a random mathematical structure and put forward evidence as to why the family of multicoloured random graphs provide a concrete possibility . a _ simple , undirected graph _ is an ordered pair @xmath9 consisting of a set @xmath10 of vertices , and a set @xmath11 of edges or lines which are @xmath12-element subsets of @xmath10 so that an edge is a relation on pairs of vertices . complete graph _ is one which has an edge on every pair of vertices , and its complement the _ null graph _ is one which has only non - edges on every pair of vertices . an _ induced subgraph _ is one obtained by deleting vertices and the incident edges . a _ spanning graph _ is a subgraph of a graph that is a spanning tree ( having no cycles ) containing all the vertices of the over - graph . a bijection @xmath13 between the vertex sets of two graphs @xmath14 and @xmath15 is an _ isomorphism _ if @xmath16 . if @xmath17 , then the isomorphism is an _ automorphism_. an _ automorphism _ of a graph is a permutation of the vertex set preserving adjacency . choose a _ natural number _ @xmath18 , @xmath19 , and a _ countable _ ( that is finite or countably infinite ) set @xmath10 of vertices and colour the edges of the complete graph on @xmath10 from @xmath20 colours @xmath21 , selecting colours independently for each edge at random . by a random colouring of a graph we simply mean an arbitrary colouring . denote @xmath20-coloured random graphs by @xmath22 , where @xmath18 , @xmath19 , and @xmath23 denotes the countably infinite vertex set . as we will only consider infinite graphs we can omit @xmath23 . they are defined by an infinite iteration of the following _ one - point extension property _ for @xmath20 edge - coloured graphs : @xmath24 given finite disjoint sets @xmath25 of vertices there is a vertex @xmath26 such that all edges from @xmath26 to @xmath27 have colour @xmath28 @xmath29 . for @xmath30 the colours are conventionally characterized as edges and non - edges . the random two - coloured graph @xmath31 has a countably infinite number of vertices . its most common variant is as a simple graph , thus having only two possible adjacencies with no loops on single vertices or multiple edges between pairs of vertices being permitted . it is a remarkable fact @xcite that a countable two - coloured random graph is isomorphic to @xmath31 with probability @xmath32 . we denote the countable random graph with three adjacency types , which we call the _ triality graph _ , by @xmath33 . we can get from @xmath34 to @xmath31 by going colourblind in pairs of colours consecutively until there are only two colours left . there are three equivalent ways of characterizing @xmath31 up to isomorphism as a countable graph , these being the so - called one - point extension property , the injection property or a combination of universality and homogeneity @xcite . a _ universal _ structure is one that contains all finite substructures , and a homogeneous structure is one for which any isomorphism between finite substructures extends to an automorphism of the entire structure . to say that a structure is homogeneous is equivalent to asserting that it has the maximum amount of symmetry ; or that many parts of the structure look alike ; or that the structure looks the same when viewed from many positions within the structure we have mentioned the uniqueness of the countable random graph ; it also has a huge amount of symmetry . most graphs , certainly finite ones , are _ asymmetric _ , that is they possess a trivial automorphism group consisting solely of the identity element . by contrast the automorphism group of @xmath31 and its multicoloured versions are gigantic , being uncountably infinite in size , thereby indicating a colossal degree of symmetry possessed by these graphs . to see how remarkable and unexpected this is , contrast the situation with the finite case . construct a finite random graph on @xmath35 vertices by ordering the pairs of vertices in a countable sequence , and tossing a fair coin a total of @xmath36 times where heads means join the two vertices by an edge , and tails means do nothing . every @xmath35-vertex graph ( up to isomorphism ) occurs with non - zero probability , and the probability of a particular graph arising is inversely proportional to its symmetry as measured by the order of its automorphism group . the asymmetric graphs are overwhelmingly more numerous than the symmetric ones . the probability measure associated with a countable sequence of coin tosses is discussed in category and measure appendix in @xcite . to get a sense for the meaning of homogeneity beyond its definition consider that the pentagon or 5-cycle graph is homogeneous , but that the hexagon or 6-cycle graph is not because a pair of points two steps apart and a pair three steps apart are isomorphic as induced subgraphs but are not equivalent under automorphisms of the whole graph . the existence and uniqueness of homogeneous universal structures follow immediately from frass s theorem @xcite on relational structures . the basic concept in frass s work is the _ age _ age@xmath37 of a relational structure @xmath38 , which is the class of all finite structures ( over the same logical language ) which are embeddable in @xmath38 as induced substructures ; see the appendix on frass s theory of relational structures in @xcite . as we stated , homogenous structures are those with the maximum amount of symmetry and one of the most useful methods for constructing objects with a large amount of symmetry is based on this theorem . in the two - colour case , thinking of a two - coloured complete graph as a simple graph , the appropriate structure is _ rado s graph _ , or the erds - rnyi _ random graph _ @xcite @xmath31 ; it was richard rado @xcite who gave the first construction of @xmath31 . relational structures with large automorphism groups have a high degree of symmetry ; this is also implied by an orbit space with a small number of orbits of the acting automorphism group . a countable homogeneous relational structure of a given type is determined up to isomorphism by the isomorphism classes of its finite substructures . an important outcome of frass s theorem is that determining the numbers of orbits on @xmath35-sets or @xmath35-tuples of _ oligomorphic _ permutation groups ( those having finitely many orbits in their induced action on @xmath35-tuples for all @xmath35 ) is equivalent to enumerating the unlabelled or labelled objects in certain classes of finite structures characterized largely by the _ amalgamation property _ , which is a way of building larger structures from smaller ones ; see the appendix on frass s theory of relational structures in @xcite . not all graph classes amalgamate ; for example , @xmath35-colourable graphs ( those having chromatic number at most @xmath35 ) do not have the amalgamation property . the word _ oligomorphic _ is intended to capture the notion of ` few shapes ' , where the group orbits contain a finite number of structures of any given finite size ( that is , few ) and each orbit contains an isomorphism class of structures ( that is , shapes ) . it is a fair assumption that the fundamental principles underlying physics are simple conceptions , whilst the phenomena themselves can be very complicated . in random dynamics it is postulated that this complexity is best described through randomness . the ambitious goal of random dynamics is to _ derive _ all the known physical laws as an almost unavoidable consequence of a random system which it is assumed would be a very general , random mathematical structure containing non - identical elements and set - theoretical notions such as subsets . it is proposed that the physics of our experience is derived from this basic mathematical structure , but as we do not have access to fundamental scales such as the plack scale the requisite mathematical object(s ) should have a generic structure , without a priori notions of distance , geometry , differentiability etc . the ` random ' aspect of the dynamics represents our lack of knowledge . the infinite random graphs are simply defined and their open definition leaves much room for enhancing the graphs with a myriad of properties such as different topologies , required to model a particular physical fact . the surprising uniqueness ( up to isomorphism ) of the countably infinite random graph for a given number of edge - colours is of course reminiscent of this philosophy ; it means that given _ any _ finite graph , as more and more vertices are attached the eventual infinite graph is almost always isomorphic to @xmath31 . if even space and time are to be derived then paths in our connected random graphs may approximate the discrete time - slicing of points in the path integral formulation of quantum mechanics ( qm ) and thereby give the principle of locality in qm . it may be that there are graph - theoretic criteria on the allowed paths in the path integral formulation or the decoherent histories interpretation of qm that narrow the possibilities for the sets of projective decompositions to which the histories correspond , and thereby help in the formulation and solution of problems ; see the section on quantum locality in @xcite , where criticisms by bassi , kent and weinberg are mentioned together with references . it would be advantageous if the graphs are supplemented with as little extra information as possible , and certainly properties such as topologies can be defined from the graph structure itself . some of the mathematics , such as probability measures on the graphs , is already worked - out in the literature ( see the appendix on category and measure in @xcite ) and some will can be used to describe the observed world . the requirement that every useful physical model arises from the purported random mathematical structure requires us to take the most general , random mathematical structure @xmath38 . here again a graph is a very general structure linking with an edge only those points which are related and model theory provides a very general formalism and body of results giving substructures and in turn through their properties , primitive laws @xcite . the various number systems can be arrived at from the mathematical structures . so for example natural numbers count the number of substructures and subtracting the number of certain types of substructures from the set of other types gives the integers . integers also count the number of vertices in @xmath31 , whilst being of uncountable order the real numbers enumerate the number of elements of the group @xmath39 @xcite . eventually a manifold structure will be required ( see below ) , for which real numbers are necessary . the uncountable order of the automorphism group @xmath40 of the multicoloured random graph on @xmath20 colours is one source . complex numbers can arise from a definition of an action principle in physics with a complex exponent . the geometric meaning of the imaginary `` i '' could simply be the introduction of rotations , as its basic action on the complex plane would attest . we emphasized the fundamental nature of the rotation group in @xcite . physics at energies less than 300 gev is called ` infrared ' and above 1 tev is called ` ultraviolet ' . in the infrared limit an asymmetric @xmath41 theory , under suitable conditions , becomes @xmath42-invariant . there are many topologies and distance functions that are derivable from the random graphs . distance within a relational structure @xmath38 may be a function of the degree of difficulty in going from one substructure to another . random graphs have a _ topology of pointwise convergence _ of permutation groups that they support . a _ permutation group _ acting on a set @xmath43 is a subgroup of the symmetric group @xmath44 . the action is _ transitive _ if the only fixed subsets of @xmath43 are the empty set @xmath45 and the entire set @xmath43 , and _ primitive _ if in addition the only fixed partitions of @xmath43 are @xmath46 and the partition into singletons . thus a group is _ primitive _ if there are only the two trivial congruences , which are the relation of equality and the ` universal ' relation with @xmath47 for all @xmath48 . a group is _ @xmath35-transitive _ if it is transitive on the set of ordered @xmath49-subsets of @xmath43 and it is _ @xmath35-homogeneous _ if it is transitive on the set of unordered @xmath35-subsets of @xmath43 . let @xmath50 be a permutation group acting on a countably infinite set @xmath43 . the natural topology on permutation groups is the _ topology of pointwise convergence _ in which for a sequence @xmath51 of permutations @xmath52 on @xmath43 , @xmath53 if and only if @xmath54 such that @xmath55 . ( we shall assume that the permutation groups have countable degree ; for arbitrary degree we would have to consider limits of nets rather than sequences . ) endowing a group @xmath50 with this topology turns it into a _ topological group _ , that is multiplication and inversion are continuous , so that if @xmath56 and @xmath57 then @xmath58 and @xmath59 . further endowing @xmath50 with a metric turns it into a complete metric space . this in turn can potentially give a manifold - like structure . it maybe that the graph - theoretic content of manifolds will eventually have to be considered . a lucid introduction to both the theory of graphs in surfaces and to homology on graphs is @xcite but also see the more recent @xcite . one of the few assumptions permitted in the nielsen program is the existence of _ exchange forces _ which permute subsystems of the abstract set of points thereby transposing particle states . that is , in a pair of particle states , the exchange force will exchange the two different particles occupying these states with each other . there are a large variety of possible types of permutations of random graph vertices that may be brought to bear here . we mentioned possible uses of _ switching automorphisms _ and _ almost automorphisms _ in modeling physics in @xcite . a permutation @xmath60 of @xmath43 is an _ almost automorphism _ of a multicoloured graph @xmath14 on a vertex set @xmath43 if the colour of @xmath61 is equal to the colour of @xmath62 for all but finitely many @xmath12-element subsets of @xmath43 . that is a permutation of the vertex set of @xmath63 is an almost automorphism if the set of edges whose colour it alters is finite . we discuss switchings in a later chapter . just as force is a gradient of energy with respect to position coordinates , so exchange forces can generically be understood as the gradient with respect to the difference in energy depending on whether or not the pauli exclusion principle applies . exchange forces are central to the emergence of bosons and fermions , and so in rendering particles that are all different at fundamental scale but are effectively identical at low energies . one possible symmetry is a _ permutation _ of two sets of particles which corresponds to a boson symmetry between the two sets . this could be a local automorphism of two subsets of graph vertices . perhaps the permutation of two sets of fermions could be effected by anti - automorphisms . an _ anti - automorphism _ of a graph @xmath14 is an isomorphism from @xmath14 to the complementary graph @xmath64 wherein edges and non - edges are interchanged . the idea is that the particles are so restricted that they behave to all intents and purposes as if they were identical in almost all respects , and in this way they become subject to permutation symmetry operations . nielsen argues in his blog post that the addition of an extra particle to an already huge set of particles could be an argument for the @xmath65 symmetry for rotation of the superposition of the set of particles and thereby the particle should effectively have an @xmath65 symmetry . the random graph has stability properties such as for each vertex @xmath66 , @xmath67 . the starting point of the random dynamics project should be the consideration of as large a class of conceivable ` theories ' or ` models ' as possible . a probability measure might then be chosen for the set of models . the hypothesis is that a model chosen at random from this set will almost always contain the empirically known laws of nature ; that is all the presently known laws of physics will follow from almost any model which is sufficiently complicated enough , provided we go to some generic low energy limit . froggatt and nielsen state @xcite : `` at first , it may seem rather hopeless to try to derive low energy physics , or even the quantum field theory of glass , out of a totally random mathematical model . however we believe some concepts are so general that we can hope to find them relevant in almost any sufficiently rich and complicated ` model ' . for example it should be possible to introduce a crude concept of distance , or some sort of topology , on almost any mathematical structure @xmath68 which contains a huge number of ` elements ' having relations between themselves : the longer the chains of intermediate elements needed to establish a relation between two ` elements ' , the further apart the two ` elements ' are defined to be . for such a mathematical structure , rich in the sense of having a huge number of elements and also being rather repetitive , it must be possible to define the concept of small modifications @xmath69 $ ] of the structure @xmath68 . '' multicoloured random graphs would then seem to fit this approach and to provide us with a mathematical formalism that produces meaningful theory . so the idea behind the random dynamics program is that from almost all choices of the a priori fundamental laws based on structural stability , the resulting physical laws are unavoidable @xcite , so that a small variation of the basic dynamics that is of the equations of motion , should not lead to an essential change . there is one important example that rhymes with this idea . poincar invariance is a given subgroup of the diffeomorphism reparameterisation group of general relativity , which under the transformation @xmath70 relates two coordinate systems by a lorentz transformation @xmath71 followed by a translation @xmath72 . as we mentioned in @xcite , one reason for the fundamental importance of the poincar group is that the poincar - invariant linear wave equations that are based on special relativistic quantum theory ( those of klein & gordon , weyl , dirac , maxwell , proca , rarita & schwinger , bargmann & wigner , pauli & fierz ) are projections of an irreducible sub - representation of the ( universal cover @xmath73 of the ) poincar group @xcite . perhaps then the poincar group may be considered as being prior to both matter and spacetime . furthermore , physical quantities such as energy , momentum , and angular momentum can be _ defined _ as the conserved quantities associated to spacetime automorphisms via noether s theorem . within the random dynamics program itself , froggatt and nielsen study a very general quantum field theory ( qft ) , which is not even assumed to be lorentz invariant , in the limit of free and lowest energy approximation @xcite . a relativistic theory with just three and in the boson case in the form of the maxwell equations . however this works for one particle species on its own and does not , immediately at least , lead to lorentz invariance if many particle species are involved . the three space dimensions are derived from the fact that whilst the basic model assumes an arbitrary number of dimensions and has momentum degrees of freedom in all these dimensions , the velocity components in all but three dimensions turn out to be zero . thus almost all equations of motion would lead to internal symmetries such as parity , isospin , etc . , relativistic invariance and other special principles . then if known physics emerges as a consequence of all but those belonging to a special null measure set then there would be no need to consider laws of nature at all or at least only as approximations in some limit , and that the parameters of the fundamental dynamics are general rather than of a special type . relativity and 3-dimensionality of space could be derived in the limit of low energy per particle . more formally begin with the largest conceivable class of models and choose a measure on this class . using a probability measure we could ask for the probability that a given class of models is lorentz invariant . it could also lead to symmetries in certain limits ( * ? ? ? * section 2 , chapter 6 ) . the hope is that neither the choice of measure nor the choice of parameters used to specify a measure on all the lattice models is significant @xcite . the search would be for generic properties , derived from the topology on the set of objects or theories being studied . such properties would almost certainly be true when a measure exists on the set . a subset of a topological space is _ dense _ if it meets every nonempty open set . a space admitting a countable dense set is called _ separable_. a _ polish space _ is a topological space that is separable and completely metrizable , meaning that its topology is induced by a complete metric . a topological group is _ polish _ if the underlying topology of the group is that of a polish space , with the open subgroups forming a base of open neighbourhoods of the identity @xmath32 . a property @xmath74 of elements of a topological space @xmath75 is said to be _ generic _ if the set @xmath76 is the intersection of countably many dense open subsets of @xmath75 . for example , being irrational is a generic property of the real numbers . the notion of genericity is standard and already well - studied when considering homogeneous structures such as random graphs . one way to construct the countable random graph that is equivalent to the usual method is as a countable _ generic structure _ where generic means construction by finite approximations . random graphs are defined in a space of graphs with a given property . the potential difficulties with this process are firstly that of defining a measure on this space and secondly that of determining whether or not a limit exists . even if a limiting structure exists it could be a surprising one . that a random structure is not always equivalent to a generic structure is exemplified by triangle - free graphs @xcite . alternatively a _ generic structure _ is one that is residual in some natural complete metric space in the sense of the _ baire category theorem _ ( see category and measure appendix in @xcite ) . this says that we can put a natural metric structure on the class of all objects with a given _ age _ or smaller so that the isomorphism class of such objects is residual , or in other words that almost all objects look like the one we are interested in . endowing a topological group @xmath50 with a metric turns it into a complete metric space , so we can use the baire category theoretic notion of _ meagre_. the topology is generated by basic open sets of the form @xmath77 = \{g \in g : g\ \text{agrees with}\ p\ \text{on its domain } \}$ ] where @xmath78 is a @xmath32@xmath32 map from a finite subset of @xmath43 to @xmath43 . the automorphism group of any relational structure is a closed subgroup of @xmath44 and so is automatically a complete metric space , and therefore we can talk about generic automorphisms . generic automorphism _ is one whose conjugacy class within the group is large in the sense of baire category . one way of capturing the notion of @xmath79 being a ` typical ' element is to require @xmath79 to lie in certain _ dense open sets_. an example of a dense open set for any @xmath80 is the set @xmath81 . a homogeneous structure will have a generic automorphism , where a _ generic element _ of the automorphism group is one that lies in a comeagre conjugacy class relative to the given topology . let @xmath50 be a permutation group acting on a countably infinite set @xmath43 . for more on the notion of _ generic automorphism _ see @xcite . in mathematics , a lattice is a vector space . more accurately , a lattice @xmath82 [ mathbbl ] in a real finite - dimensional vector space @xmath10 is a subgroup of @xmath10 satisfying one of the following equivalent conditions : \(i ) @xmath82 is discrete and @xmath83 is compact ; ( ii ) @xmath82 is discrete and generates the @xmath84-vector space @xmath10 ; \(iii ) @xmath10 has an @xmath84-basis @xmath85 which is a @xmath86-basis of @xmath82 ( that is @xmath87 . so a lattice in @xmath88 is the set of integral linear combinations of @xmath35 linearly independent vectors . the lattices generated by vectors of norm @xmath32 are @xmath89 with the standard bilinear form . in this brief book we can only give a flavour of the random graph mathematics that can be put to use in modelling random dynamics . one question we could seek to answer is : given a certain natural topology on the set of all lattice theories show that lorentz invariance is a generic property in the low energy limit . a lattice could reasonably be taken to be a low energy limit of a graph . amongst the derived symmetries to which it is hoped that the random dynamics approach can be applied ( * ? ? ? * chapter vii ) are : \(i ) macroscopic scaling symmetry present for a very large number of molecules but absent at the atomic scale ; \(ii ) five symmetries derived from the sm , ( a ) charge conjugation invariance , ( b ) parity invariance , ( c ) time reversal invariance , ( d ) flavour conservation ( including any baryon and lepton number conservation ) , and ( e ) chiral symmetry ( including gell - mann - neeman @xmath6 ) . there are exact symmetries for example @xmath90 invariance , baryon number conservation and conservation of lepton number for each generation . approximate symmetries such as @xmath91 , @xmath74 , and @xmath92 , flavour and chiral symmetries are derived . given that parity and strangeness conservation are derivable from the strong and electromagnetic interactions there is no reason why weak interactions should conserve them . poincar symmetry , including rotational and space - time translational symmetries are considered part of the diffeomorphism symmetry of general relativity . so a proper account of their origin must go beyond known theory to consideration of say a pregeometric theory , as must any account of gauge invariance . but it is not merely symmetries whose origin random dynamics aims to uncover . the idea goes further in suggesting that almost any theory proposed at random will in one or more limits , say at low energies , possess the regularity in question . almost any model at the fundamental level would then be a theory of everything rather than one specific one . the observed physical laws would seem to be stable under changes of the fundamental model . this random dynamics philosophy overlaps with that of wheeler s ` law without law ' @xcite . the following discussion @xcite is preliminary and abstract . beginning with a random mathematical structure @xmath38 space - time must be introduced and physical objects must be identified . physical states or ` wavefunctions ' are defined on a topological space @xmath93 constructed from the structure . a topology on the random structure would give the concept of distance . the vector space of small modifications of @xmath38 is identified with the hilbert space of a generalised qm , and the topological space on @xmath38 is the configuration space of functions for a quantum field - type theory , that is , each point in the topological space corresponds to a field configuration . the field values over space - time become the coordinates for @xmath93 which then essentially becomes a differentiable manifold . then assume that it is possible to find a tangent vector space to @xmath93 . introduce a topological structure on the basis vectors or ` directions ' in the tangent space , so as to identify coordinates in @xmath93 with points or small regions in a pregeometrical space . these directions could form another complicated mathematical structure and again as for @xmath93 , two ` directions ' are close together if a simple mathematical operation connects them . groups of ` directions ' in the tangent space can form neighbourhoods or small overlapping structures associated with small regions in a pregeometrical space - time . associating field values , that is points of @xmath93 , with small regions of space - time ensures a local field theory in the long wavelength limit . next identify a certain configuration as the vacuum state and a symmetry group of gauge transformations including ones that might form a diffeomorphism group permuting points of the pregeometric space which may themselves be identified with space - time points . orbits under diffeomorphism group action could then form space - time manifolds . in random dynamics the action is assumed to be that of a discretized lattice - type qft rather than the usual continuum theory . the founders chose the description _ glass _ by analogy with the `` frozen in '' structure of a spin glass . the resulting qft glass is modelled as a feynman path integral with constant ( or in the terminology _ quenched _ ) random parameters . then quantum fluctuations are thought to smooth out the qft glass into a continuum theory . in building a field theory glass model , a generalised quantum field @xmath94 is defined on a set of quenched random space - time points @xmath95 and takes values in quenched random manifolds @xmath96 . quenched parameters are chosen randomly but are then fixed in the path integral . the field theory glass action @xmath97 = \sum_r \mathcal{s}_r(\phi(i)),\quad i \in r,\ ] ] is a sum over contributions @xmath98 from many small overlapping regions @xmath99 in space - time . a gauge theory for which the gauge symmetry group varies randomly across the regions , called a _ gauge glass _ , is constructed on which continuum gauge field degrees of freedom @xmath100 can be imposed . a _ gauge glass _ model has randomly varying parameters on the `` lattice '' , though the same random value at any given plaquette is retained throughout ; the value is then called _ quenched_. we take up the issue of the gauge group having nontrivial outer automorphisms in chapter [ oaut ] . let @xmath101 be the number of substructures @xmath43 of the relational structure @xmath38 and @xmath102 be the number of substructures @xmath103 of @xmath43 . clearly , these are natural numbers including zero . a coordinate system can be formed in which we can associate one substructure with each coordinate . the whole of @xmath38 would then be describable by a vector in a vector space with a basis in corespondence with the substructures . a mathematical structure such as a graph is a _ cayley object _ , @xmath104 , for the group @xmath50 if its point set are the elements of @xmath50 and right multiplication by any element of @xmath50 is an automorphism of @xmath104 . a cayley graph for a group @xmath50 takes the form @xmath105 , where @xmath68 is an inverse - closed subset of @xmath106 with vertex set @xmath50 and edge set @xmath107 . as @xmath50 is countable , we can enumerate the inverse pairs of non - identity elements of @xmath50 as @xmath108 . baire category theory can sometimes be used for building a homogeneous cayley object for a group @xmath50 , by showing that almost all @xmath50-invariant objects of the required type , that is a residual set of them , are homogeneous . in order to make sense of the notion of a residual set of cayley graphs for @xmath50 , specify cayley graphs by paths in a ternary tree , whereby the three descendents of a node or vertex at level @xmath35 correspond to including or excluding the inverse pair @xmath109 in one of the colour classes of @xmath50 ; call these colour classes red @xmath110 , blue @xmath111 and green @xmath112 . @xmath113 & { \bullet } { \{g_{n+1 } , g_{n+1}^{-1}\ } } { \in \mathfrak{b } } \ar@{-}[d ] & { \bullet}{\{g_{n+1 } , g_{n+1}^{-1}\ } } { \in \mathfrak{g } } \ar@{-}[dl]\\ & & { \bullet } \ar@{-}[d]\\ & & { \bullet } } \ ] ] the set of three - coloured cayley graphs is identified with the set of paths in the ternary tree . in baire category theory if an object is specified by a countable sequence of choices , then the existence of one such object with a given property @xmath74 can be proved by showing that @xmath74 holds for ` almost all ' choices . let @xmath114 denote the set of paths of countable length starting at the root of a tree @xmath115 , whose nodes at height @xmath35 are labelled by structures on @xmath116 . we define the distance between distinct paths @xmath78 and @xmath117 to be @xmath118 , where @xmath35 is the height of the last node at which @xmath78 and @xmath117 agree , and @xmath119 is any strictly decreasing function tending to zero . the complete metric space to which the baire category theorem will be applied arises from paths in rooted trees of countable height . a cauchy sequence in this space is a sequence of paths agreeing on increasingly longer initial segments , and so has a unique limiting path . in this way a complete metric space can be built . we require an interpretation of openness and denseness to formulate residual sets in this space . open ball _ consists of all paths in the tree containing a given vertex . a set @xmath68 of paths is open ( or _ finitely determined _ ) if each path in @xmath68 has a vertex such that every path through this vertex is in @xmath68 . a set @xmath68 is _ dense _ ( or _ always reachable _ ) if it meets every open ball , i.e. if all vertices lie on some path in @xmath68 . the triality graph is the frass limit of the class of all appropriately defined finite @xmath120-coloured graphs , so with a countable vertex set the isomorphism class of @xmath33 is residual in the set of @xmath120-coloured graphs on vertex set @xmath121 . thus it makes sense to talk of a residual set of cayley graphs for @xmath50 . the metric space has an underlying topological space , where the topology lies on the collection of inverse closed subsets of @xmath50 . for a physical interpretation we may be able to identify the types of substructures with possible paths in the feynman path integral . that is to say @xmath122 , where @xmath123 is the substructure corresponding to _ another possibility is that the path from one vertex to another in the graph is what identifies the discrete parts of a path integral . the transition amplitude from an initial to final state in usual feynman path integral is the funtional integral @xmath124 where @xmath125 is the integral of the lagrangian is over the path being integrated in between times @xmath126 and @xmath127 . the fields remain real or complex as usual but the coupling constants and squared masses are assumed complex giving @xmath128 . adding an imaginary part to the lagrangian density allows the action to be complex . the imaginary part @xmath129 of the action is a weight function of the usual real - action feynman path integral and only when @xmath129 is very negative will the factor @xmath130 dominate the feynman path integral . an action with both real and imaginary parts leads to the second law of thermodynamics . the imaginary part @xmath131 of the action can be thought of as the time - integral over the imaginary part of the lagrangian for @xmath132 . one consequence of introducing an imaginary part to the action used in the path integral , of a similar form to the usual real part but with different parameters , is a model determining not only equations of motion but also the initial conditions by typically providing a large factor in the probability thereby fixing the path obeyed by the equations of motion and the initial conditions . the imaginary part @xmath131 of @xmath133 fixes a state in both the future and past , for example the particle s position when it went past the double slit screen . two approaches to extracting probabilities and expectation values are then outlined @xcite . firstly allowing a natural definition of an average of quantity @xmath134 as a function of the fields @xmath135 @xmath136 } \mathcal{o}(\phi ) \mathcal{d } \phi}{\int e^{i\mathcal{s}[\phi ] } \mathcal{d } \phi}\ ] ] may give a complex probability in a path integral . the second approach inserts a series of projections onto small regions for operators and take the modulus square of the integrals between initial and final values . under certain assumptions the two approaches should approximately agree with each other . it is hoped that the feynman path integral formulation could arise from random dynamics plus a few extra assumptions @xcite , and it seems that taking an imaginary action is what is required @xcite because this determines which of the solutions of the equations of motion is the right one and thus determines initial conditions . in @xcite bennett , kleppe and nielsen discuss the possibility of random complicated structures leading to manifolds , via concepts of connectivity and nearness or resemblance of one substructure with another . the numbers of substructures determining the coordinates of the mathematical structure are given by a matrix which shows how to get to certain substructures from others . in particular , with as many rows and columns as there are number of types of substructures , that is paths . matrices that grow to reflect subgraphs on the way to constructing the random graph have also been used , as we shall now see . the edge - set of a simple graph on a given countable vertex set can be thought of as a countable zero - one sequence , leading to the fact that the class of graphs forms a complete metric space , of which the subclass of graphs satisfying the two - colour injection property ( equivalent to the _ one - point extension property _ @xmath137 for @xmath12 edge - coloured graphs ) is a residual subset @xcite . the metric space equivalent of the random graph is uniquely defined by a one - point extension property as it is for random graphs . a metric space @xmath43 is called a _ ( generalized ) urysohn space _ @xcite if whenever @xmath138 is a finite metric subspace of @xmath43 and @xmath139 is an arbitrary one - point metric space extension of @xmath140 , the embedding @xmath141 extends to an isometric embedding @xmath142 . up to isometry there is only one universal complete separable urysohn metric space , denote it @xmath143 , which contains an isometric copy of every separable metric space . the way to show that @xmath143 is the ` random polish space ' is to choose @xmath35-tuples of random real numbers as distances from @xmath35-dimensional euclidean space @xmath144 , take a measure for each dimension , and construct a countable space one point at a time using a suitable measure on extensions via a ` random metric ' between the @xmath145 point and the first @xmath35 points , ensuring that the triangle inequality which determines a cone in @xmath144 is satisfied each time . finally take the completion of this . we can say more precisely what a random metric on @xmath121 looks like . let @xmath146 be the first two points of the space . choose @xmath147 , to lie in @xmath148 , where the superscript denotes dimension . next , choose @xmath149 so that @xmath150 and @xmath151 satisfying @xmath152 now choose @xmath153 so that @xmath154 , @xmath155 and @xmath156 satisfying several inequalities of the form @xmath157 each cone at each new dimension depends on the previous one . by choosing a wide range of reasonable measures on all cones , the completion of @xmath121 equipped with a random metric is almost always isometric to @xmath143 . a. vershik @xcite attaches admissible vectors to distance matrices in order to mimic the one - point extension property . the indivisibility property of @xmath143 whereby if @xmath158 is an open ball then @xmath159 is equivalent to removing a row and some columns in the distance matrices stabilizing @xmath143 . it is possible to construct @xmath31 from @xmath143 . any group acting on @xmath143 with a countable dense orbit preserves the structure of @xmath31 on the dense orbit . topology is in a sense a generalisation of geometry where sets of points and their relation to each other is the central idea rather than studying solid objects . there is a natural topology for the ordinary two - colour random graph which may well have a physical interpretation , and that is the _ neighbourhood topology _ @xcite . in ( * chapter 8) we studied a filter , to be defined shortly , and a topology defined naturally from @xmath31 , generated by the vertex neighbourhoods in @xmath31 , and whose automorphism groups contain the automorphism group of @xmath31 . for such neighbourhood filters , @xmath31 has a ` universal ' property : any countable graph whose neighbourhood filter is non - trivial contains @xmath31 as a spanning subgraph . from the injection property of @xmath31 it follows that any finite vertex set of @xmath31 has a common neighbour , and this property characterizes the class of countable graphs containing @xmath31 as a spanning subgraph . a _ filter _ on a set is a family @xmath160 of subsets of the vertex set @xmath161 of @xmath31 satisfying * @xmath162 implies @xmath163 ; * @xmath164 , @xmath165 implies @xmath166 ( upward closed ) . a filter @xmath167 is _ trivial _ if it consists of all subsets of @xmath161 ; it is _ principal _ ( one - generated ) if it consists of all sets containing a fixed subset @xmath140 of @xmath167 ; and it is an _ ultrafilter _ if , for any @xmath168 , just one of @xmath43 and @xmath169 belongs to @xmath167 . if @xmath167 is a filter , then @xmath170 is a topology . since @xmath161 is countable , no non - discrete metric on @xmath161 can be complete . if the topology is not discrete then the set of complements of finite unions of discrete sets is a filter . let @xmath14 denote a graph and @xmath171 be the set of vertices attached to vertex @xmath172 in the graph . suppose that @xmath14 has the property that each vertex has a non - neighbour . then the filter generated by the closed neighbourhoods @xmath173 is equal to @xmath174 . let @xmath31 denote the countable random graph . [ prop8.2 ] the following three conditions on a graph @xmath14 are equivalent : * @xmath175 is nontrivial ; * @xmath14 contains @xmath31 as a spanning subgraph ; * @xmath176 . this result @xcite shows that @xmath177 is the unique maximal neighbourhood filter . but this uniqueness is only up to isomorphism . so part ( c ) really means that @xmath175 is contained in a filter isomorphic to @xmath177 . grafting qm onto classical objects such as random graphs can either be done by attaching hilbert space(s ) to vertices or edges , or as an integrand in a feynman path integral . heuristicallly the random mathematical structure @xmath38 can be interpreted as an affine space ; its basis of vectors is itself an affine space which must be interpreted as the space of paths . each unit volume in the vector space has a weight - function - induced probability density for the infinitesimal region around a point to be realized as a true structure . the algebraic structures relevant to applications could be rings rather than vector spaces . rings are flexible enough to allow linear superpositions of elements , are more general than algebras , and intersect algebras in the class of linear spaces ; see appendix b of @xcite . two relational structures with the same age have the same graded algebra up to isomorphism , hence the name _ age algebra_. in fact the homogeneous components are indexed by an age of the appropriate size and so algebras built out of the age are equal and not just isomorphic . the converse is not true . in ( * chapter 5 ) we found an isomorphism between two algebras with very different structures in the finite algebra case , one graded and the other semisimple , but which are isomorphic in the infinite case . let @xmath38 be a relational structure on a set @xmath43 , finite or infinite . for any natural number @xmath178 , let @xmath179 denote the vector space of all complex - valued functions from @xmath35-element subsets of @xmath43 to @xmath180 which are constant on isomorphism classes of @xmath35-element substructures of @xmath38 , and construct an algebra by @xmath181 there are different ways to define multiplication on @xmath182 , each of which could turn it into a commutative and associative algebra . this was studied in @xcite . when the relational structure is a graph , there is a separate combinatorial problem of how many graphs can be fitted into a larger graph on a given vertex set , and one approach to its resolution is exemplified by the following . denoting the number of @xmath183-vertex graphs @xmath184 that can be embedded as induced subgraphs of a @xmath185-vertex graph ( @xmath186 ) by a double angled bracket , we have that @xmath187 @xmath188 \llangle \gamma_{j-1 } : \gamma_i \rrangle\ ] ] where the sum is over all @xmath189 vertex graphs . this gives a recursive combinatorial expression , with the possible use of the following known identities:- \(1 ) the recurrence relation for bell numbers : + @xmath190 , ( 2 ) @xmath191 , \(3 ) @xmath192;@xmath193;and + @xmath194 , where the unsigned lah number @xmath195 counts the number of ways a set of @xmath35 elements can be partitioned into @xmath49 nonempty linearly ordered subsets , and @xmath196 is the stirling number of the first kind . _ combinatorial enumeration _ let @xmath38 be a countable homogeneous structure with age @xmath182 and let @xmath197 . by homogeneity , two finite subsets of @xmath38 lie in the same @xmath50-orbit if and only if the induced substructures are isomorphic . so the sequence enumerating _ @xmath35-element members of @xmath182 ( that is , up to isomorphism ) is identical with the sequence enumerating the @xmath50-orbits on unordered @xmath35-element subsets of @xmath38 . similarly , the number of _ labelled _ @xmath35-element members of @xmath182 ( that is , members of @xmath182 on the set @xmath198 ) is equal to the number of @xmath50-orbits on ordered @xmath35-tuples of distinct elements of @xmath38 . let @xmath50 be an oligomorphic permutation group acting on each of the sets @xmath199 ( all @xmath35-tuples of elements of @xmath43 ) , @xmath200 ( all @xmath35-tuples of distinct elements of @xmath43 ) , and @xmath201 ( all @xmath35-element subsets of @xmath43 ) . if @xmath202 and @xmath203 denote these numbers of orbits then @xmath204 , since each orbit on @xmath35-sets corresponds to at least one and most @xmath205 orbits on @xmath35-tuples . a permutation group is _ @xmath35-transitive _ if @xmath206 , and _ @xmath35-homogeneous _ if @xmath207 . also , @xmath208 where @xmath209 is the stirling number of the second kind , the number of partitions of an @xmath35-set into @xmath49 parts . for an orbit @xmath210 on @xmath35-tuples determines , and is determined by , a partition of @xmath116 into @xmath49 parts ( where @xmath183 and @xmath185 lie in the same part if @xmath211 ) and an orbit on @xmath49-tuples of distinct elements . the _ exponential generating function _ is given by @xmath212 . the series @xmath213 for a direct product ( in the intransitive action ) or a wreath product ( in the imprimitive action ) can be calculated from those of the factors : @xmath214 @xmath215 if @xmath216 where @xmath23 denotes countable infinity and the sequence @xmath217 is realized by a group @xmath50 then @xmath218 is realized by @xmath219 . let @xmath50 act transitively on @xmath43 , and @xmath220 be the stabilizer of the point @xmath221 . it can be shown that @xmath222 or equivalently differentiating an exponential generating function corresponds to shifting the sequence terms one place to the left , @xmath223 so there is an equivalence between @xmath224-orbits on @xmath35-tuples and of @xmath50 orbits on @xmath225-tuples . if @xmath50 is intransitive , then the derivative of @xmath226 gives the sum of @xmath227 , over a set of representatives @xmath228 of the orbits of @xmath50 . for any oligomorphic group @xmath50 , _ generalized stirling numbers _ denoted by @xmath229(n , k)$ ] , can be defined that obey @xmath230(n , k ) = f_{n}(g { \mathop{\mathrm{wr}}}s),\ ] ] and have the composition property @xmath231(n , l ) s[h](l , k ) = s[g { \mathop{\mathrm{wr}}}h](n , k),\ ] ] where @xmath232 denotes wreath product . the last result can be expressed in terms of infinite lower triangular matrices of generalized stirling numbers . there is a linear analogue of the combinatorics of sets and functions that applies to vector spaces over finite fields and linear transformations in which the gaussian ( or @xmath233-binomial coefficient ) replaces the binomial coefficient . these coefficients enumerate the number of @xmath49-dimensional subspaces of @xmath35-dimensional @xmath234-vector space @xcite . there are general theorems on growth rates of counting sequences , many of them requiring a primitive permutation group @xmath197 , for example if @xmath235 . primitivity can be completely described if @xmath50 is oligomorphic . within the area of probabilistic combinatorics , there are various concepts with suggestive titles such as the _ concentration of measure _ in discrete random processes that may be brought to bear but these would have to be studied before an assessment can be made of their utility to random dynamics program . with one eye on the philosophy of the path integral , one of the questions in random graph theory is can it be shown that some random variables are concentrated around either their expected values or their expected trajectories as an underlying random process evolves ? the differential equation method of wormald @xcite and its extension due to bohman @xcite uses solutions of differential equations to approximate the dynamical evolution of the random process . once space and time are derived position , momentum and translational invariance follow as well as energy and momentum conservation . rotational invariance may come with angular momentum conservation . nielsen and kleppe returned to the theme of a derivation of space in @xcite . they postulate the existence of an abstract , general phase space or state space together with a random generic hamiltonian @xmath236 and then examine the statistically expected functional form of the `` random @xmath237 '' . within phase space one state is identified as the key vacuum state given by a wave function ; classically this is a point but quantum mechanically it is a volume @xmath238 ( uncertainty principle ) . because of quantum mechanics , space is identified with half of the @xmath239 dimensions of the phase space of a very extended wave packet which is a superposition of the different field degrees of freedom . a wave function can be approximated by excitable displacements of the transversal directions of the @xmath240-dimensional manifold . the primacy of 3 + 1 dimensions is hinted at by the observation that if non - lorentz invariant terms are added to the weyl equation , only in 3 + 1 dimensions does lorentz invariance eventually emerge . random dynamics takes nature to be nonlocal because it is argued that locality only makes sense in the presence of a spacetime or at least a manifold , whilst field theory assumes locality that is all interactions occur in one spacetime point . this nonlocality is intended to be inherent in nature , as opposed to being of a dynamical origin , for example the purported quantum non - separability which arises as nonlocal correlations . locality is then explained using diffeomorphism symmetry ( that is invariance under reparametrization mappings ) and a taylor expansion of the action . translational invariance also follows from diffeomorphism symmetry . recall the postulate of the random dynamics project that any sufficiently general initial assumptions or axioms or model at the fundamental scale will lead to the same known physical laws . nielsen and co - workers have devised discrete spacetime so - called _ gauge glass _ models , inspired by lattice gauge theory , to derive criteria for the breaking of a large over - group to the standard model group . the name _ gauge glass _ was chosen to be analogous to a spin glass whose `` frozen in '' structure is reminiscent of that of glass . an explanation as to why the sm group is group @xmath241 @xmath242 defined by the set of matrices and why groups with complex representations are likely to be more important down to low energies than those with only real self - conjugate representations can be found in the paper @xcite by bennett , brene and nielsen . the complex representations should have complex conjugation as an automorphism ; it is the only automorphism of the sm group . a reflexive , symmetric and transitive binary relation on a set is an equivalence relation , and one that is reflexive , antisymmetric and transitive is a partial order . a _ lattice _ in _ lattice gauge theory _ is a set of elements with a partial order , the elements being lattice _ sites _ connected by lattice _ links_. the sites can be interpreted as charges or as sinks or sources of magnetic fields . link variables defined on the edges of the lattice are the fundamental variables of the lattice theory . denoted @xmath243 & { \bullet}_y ) } \in g$ ] they are elements of the gauge group @xmath50 describing a symmetry of the lattice gauge theory . the continuum space version of the link variable is a parallel transport operator between the points @xmath228 and @xmath244 : @xmath245 where is the path ordering operator and @xmath246 is a curve between points @xmath228 and @xmath244 . the operator @xmath247 is the wilson loop . for a scalar field @xmath248 interacting with a gauge field @xmath249 the term @xmath250 is a gauge invariant observable . the link variable connects the point @xmath35 and the point @xmath251 , where the index @xmath252 indicates the direction of a link in the hypercubic lattice with parameter @xmath253 . the lattice version of the equation for the parallel transport operator is @xmath254 & { \bullet}_y ) } = e^{i \theta_{\mu } ( n ) } \equiv u_{\mu } ( n).\ ] ] considering the continuum limit @xmath255 of this operator we have that @xmath256 where @xmath257 and @xmath258 generates @xmath259 . for example , if @xmath260 then @xmath261 and for @xmath262 @xmath263 where @xmath264 are the gell - mann matrices . the links couple together to form lattice _ plaquettes _ , denoted @xmath265 , where the plaquette variables @xmath266 and plaquette action @xmath267 contributions are defined below ; see figure [ latgth ] . to model the gauge invariant action of the gauge theory assume that bianchi identities are satisfied ; see next chapter . these identities apply also to plaquettes in more than 2 dimensions , for example to cubes in a @xmath120-dimensional lattice . the lattice action @xmath75 is invariant under the gauge transformations @xmath254 & { \bullet}_y ) } \to \lambda(x ) u(\xymatrix{{\bullet}_x \ar@{-}[r ] & { \bullet}_y ) } \lambda^{-1}(y)\ ] ] on a lattice where @xmath268 . in 4-dimensional euclidean space the analogue of the partition function is a path integral @xmath269 & { \bullet } } ) e^{-s[u(\xymatrix{{\bullet}_x \ar@{-}[r ] & { \bullet}_y)}]}.\ ] ] the lattice field theory is built in such a way that in the continuum limit it gives a regularized smooth gauge theory of fields @xmath270 . the coupling constant strength determines the quantum field fluctuations which in turn determine the vacuum phases . short range correlations correspond to the confinement - like phase in which the bianchi identities can be assumed negligible implying the approximate independence of the fluctuations of each plaquette in a lattice cell . in the coulomb phase the electromagnetic potential and photons have infinite range and so the bianchi identities lead to a correlation of plaquette variable fluctuations over distances of many lattice constants and more . @xmath271 & & { } \ar@{-}[d ] \\ & { } \ar@{-}[r ] & { \bullet } \ar@{-}[dd]_{l_4 } \ar@{-}[rr]^{l_3 } & & { \bullet } & { } \ar@{-}[l ] \\ \\ & { } \ar@{-}[r ] & { \bullet } \ar@{-}[rr]_{l_1 } & & { \bullet } \ar@{-}[uu]_{l_2 } & { } \ar@{-}[l ] \\ & & { } \ar@{-}[u ] & & { } \ar@{-}[u ] } \ ] ] different combinations of gauge field phases correspond to different vacuua . the phase transations at the _ multiple point _ of a phase diagram for a lattice gauge theory are first order and the phases are distinguishable at the lattice scale . commutator relations in non - abelian groups constrain the gauge coupling normalization but they are absent in abelian groups . rescalings of the gauge potential are only possible in the abelian sectors so here the gauge couplings have no physical significance and so the @xmath272 has no intrinsic unit of charge . by analogy with lattice gauge theory , consider links between sites and associate group - valued link variables corresponding to a gauge group @xmath50 which gives symmetries of the theory . introduce dynamics through an action @xmath75 composed additively of contributions from all primitive polygons ( plaquettes in the case of a regular cubic lattice ) . to achieve gauge invariance on the lattice , the individual `` plaquette '' contributions @xmath273 must be invariant functions of the `` plaquette '' elements which themselves are products of the link elements circumscribing the `` plaquettes '' : @xmath274 @xmath275 the class of an element here refers to the conjugacy class set of all elements equivalent to the element under conjugation : @xmath276 an example of a class function is @xmath277 $ ] where @xmath278 denotes the trace of some matrix representation , and @xmath279 where @xmath280 is the gauge coupling constant . in the simplest case , @xmath281 , @xmath282 and the link variables @xmath283 & { \bullet } ) } \in \mathbb{c}$ ] are modulus one complex numbers . in the lattice model , the lorentz gauge condition is @xmath284 & { \bullet}_y ) } = 1\ ] ] where the product is over all links @xmath285 & { \bullet}_y}$ ] . also we can write @xmath243 & { \bullet}_y ) } = e^{i \theta_{\square}}$ ] . these plaquette link variables are not independent but satisfy a bianchi identity ( see later ) . so the simplest lattice @xmath286 action takes the form @xmath287 . for the compact lattice qed , @xmath288 where @xmath289 is the bare electric charge . wilson s original simplest action for the lattice @xmath290 gauge theories @xmath291 where the sums run over all plaquettes of a hypercubic lattice and @xmath266 belongs to the adjoint representation of @xmath290 generalized by bhanot and creutz @xcite @xcite to @xmath292\ ] ] + where @xmath293 and @xmath294 are respectively the lattice constants and traces in the fundamental and adjoint representations of @xmath290 . the phase diagrams for the generalized lattice @xmath295 and @xmath296 theories using monte carlo methods have a triple ( boundary ) point of three first - order phase transitions , where the `` coulomb - like '' , @xmath297 and @xmath298 confinement phases meet . three phase border lines emanate from the triple point which separate the corresponding phases . the @xmath298 phase transition is due to a condensation of monopoles . the discrete @xmath297 phase transition occurs when lattice plaquettes change from the identity to nearby elements in the group . lattice gauge theory assumes a regular lattice , a given gauge group and a given form of the plaquette action throughout the lattice . a gauge glass model has randomly varying parameters on the `` lattice '' , though the same random value at any given plaquette is retained throughout , that is it is `` quenched '' . randomness can be introduced in a variety of ways:- \(i ) the sites can be irregularly distributed ( hence the quotation marks around `` lattice '' and `` quenched '' ) . however this possibility is not considered and the gauge glass lattice is taken to be a regular hypercubic lattice in @xmath35-dimensional space ( -time ) . \(ii ) fix the gauge group throughout the lattice but randomly vary the functional form of each plaquette action , for example in @xmath299\]]take @xmath300 to have a random value for each plaquette and the trace to be in the quark - like representation @xmath233 . \(iii ) if the gauge group has isomorphic subgroups , which are present on all links that join at a site , these are not in the first instance distinguishable in a gauge glass model . so in going around a plaquette the subgroups can either remain separate or can be mixed together . \(iv ) associate a new group in a random quenched manner to each link ; again this is not considered in the gauge glass models . regardless of the randomness , the action is constructed to be invariant under the gauge group @xmath50 but the ground vacuum state , which is determined by the minimum energy value , need not be . let @xmath301 be the energetically preferred value of plaquette group element . only if @xmath301 commutes with every element of the group @xmath50 will the vacuum be invariant under global gauge transformations . so the best surviving elements will belong to the center of the group , that is the set of elements that commute with each group element . the @xmath302 groups have such central elements , but the @xmath303 groups have a discrete part of the center divided out . the center of all groups of the form @xmath304 @xmath305 , for example @xmath306 is @xmath307 . the central elements are required to be continuous meaning that there are elements in the center that are arbitrarily close to the unit element . also the center must be connected so that the preferred plaquette values do not lie in a different nonconnected part of the center . the group @xmath241 has a compact connected center . the purported randomly chosen gauge group near the planck energy breaks down in a series of steps until the group remaining at the gauge glass level has the required properties of its central elements and has no multiple occurrences of isomorphic factors . according to the random dynamics philosophy , in reaching the group @xmath241 neither the specific structure of the `` lattice '' nor the choice of action for any given plaquette nor the group chosen as long as it is sufficiently large , are crucial . as to the origin of the gauge glass , associated to each link of a random lattice group variables which approximately stabilize the local action contributions @xcite . these new variables may eventually lead to gauge fields with massless gauge quanta for those parts of the symmetry that are not broken . in other words the large distance behaviour of gauge theories is stable , within certain limits , to the addition of gauge non - invariant interactions at small distances . _ switchings and switching automorphisms of graphs _ having proved theorems @xcite @xcite that connect random graph vertices and edges in a @xmath32@xmath32 correspondence with lattice points and links respectively we require at least one physical interpretation of the graph vertices and edges . we are potentially provided with one by the random dynamics program where the regular hypercubic lattice in normal lattice gauge theory is replaced by an amorphous glass - like structure simply by letting the lattice sites take random positions separated by distances of order the planck length . ( it could be that a lattice is the local version of a global random graph structure , in parallel to the random dynamics philosophy that what we observe at our low energy level can be interpreted as essentially correction terms of a description of the physics taking place at a higher energy level . ) this interpretation needs to be expanded upon , but one possible step thereafter is to understand some of the groups that are naturally supported by random graphs and then to explore whether or not any physical meaning can be attributed to such groups . the purpose of this subsection is to look at one type of such groups . let @xmath308 be the set of simple complete graphs on a fixed set of @xmath35 vertices and @xmath20 edge - colours and @xmath309 be an arbitrary element of this set . let @xmath310 and @xmath311 be distinct colours , and @xmath103 a subset of the vertex set @xmath43 of the graphs in @xmath308 . the operation @xmath312 interchanges the colours @xmath310 and @xmath311 whenever they occur on an edge with just one vertex in @xmath103 , leaving all other colours unchanged . note that @xmath313 and @xmath314 the _ switching group _ @xmath315 is the group of permutations of @xmath308 generated by the _ switching operations _ , which are defined to be the switchings on all vertex subsets of @xmath43 . a _ switching class _ is an orbit of @xmath315 on @xmath308 . the structure of the group of all switchings on @xmath35-vertex subgraphs having @xmath20 colours is given by @xmath316 here @xmath317 is the smallest non - trivial group consisting of the identity element and transpositions , and the alternating group @xmath318 of even permutations is the index-2 normal ( that is , invariant ) subgroup of the symmetric group @xmath319 of all permutations . the semi - direct product action here partitions the vertex set into two parts and the @xmath317 groups transpose the colours on edges crossing the partition . this structure is retained in the limit of an infinite number of vertices , as @xmath320 , and it specializes to @xmath321 in the two - colour case . if @xmath79 is any element of @xmath44 , the symmetric group of all permutations on @xmath43 , then @xmath322 . so @xmath44 normalises @xmath315 . note that @xmath44 also acts on the set @xmath308 , and hence so does the semi - direct product of these groups . let @xmath14 be an element of @xmath308 , that is , an @xmath20-coloured graph on @xmath43 . the _ group of switching automorphisms _ @xmath323 is defined by @xmath324 that is , it consists of all permutations of @xmath43 whose effect on @xmath14 can be undone by a switching . another way to think about this group is as follows : if @xmath325 is the stabilizer of @xmath14 ( so @xmath326 , and @xmath79 undoes the effect on @xmath14 of @xmath327 ) in the semi - direct product @xmath328 , then @xmath323 is the image of @xmath325 under the canonical projection of the semi - direct product onto @xmath44 . note that multicoloured graphs in the same switching class have the same group of switching automorphisms . for suppose that @xmath329 , so that @xmath330 , and let @xmath331 be any switching operation . then @xmath332 ; since @xmath333 is again a switching operation , @xmath79 is a switching automorphism of @xmath334 . that switching is an equivalence relation is easily seen : ( a ) reflexivity - switch with respect to the whole of set @xmath43 and everything outside @xmath43 , that is @xmath45 . ( b ) symmetry - switch with respect to @xmath335 ; and then @xmath103 again to recover the original configuration . ( c ) transitivity - switch with respect to @xmath336 then with respect to @xmath337 is the same as switching with respect to @xmath338 , the symmetric difference of @xmath339 and @xmath340 . _ a first presentation of @xmath315_. in ( * ? ? * chapter 4 ) , we derive two presentations of the switching groups @xmath315 for graphs on @xmath35 vertices and @xmath20 colours . the first comes out as @xmath341 . the second presentation leads us to define and study a type of generalized coxeter group , with the following as basic motivation . a _ coxeter group _ has generators @xmath342 and relations @xmath343 and @xmath344 , where @xmath345 . the finite coxeter groups are the finite real reflection groups @xcite . the coxeter group @xmath346 may be defined on a graph @xmath14 and is generated by vertices @xmath347 @xmath348 , such that @xmath349 + for example , @xmath350 is a coxeter group represented by the following diagram : @xmath351 & { \circ } \ar@{-}[r ] & { \circ } \ar@{-}[r ] & { } { \ldots } { \ldots } \ar@{-}[r ] & { \circ } } \ ] ] which has @xmath352 nodes indexed by @xmath348 , and where @xmath353 . _ a second presentation of @xmath315_. a second type of @xmath315 presentation is given in terms of involutions about vertices and @xmath120-cycles about edges , and can be recovered by inspection of the form of the switching groups . for example @xmath354 consists of @xmath120 generators @xmath355 that are intended to represent three @xmath120-cycles and two involutions @xmath356 and @xmath357 whose actions are shown in the following picture @xmath358_{a_2=(j k i ) \rightarrow \ j } \ar@{-}[dr]^{i\ \leftarrow a_1=(i j k)}\\ { b_1=(jk ) } \ar@{->}[r ] & { 1 } \ar@{-}[rr]_{k\ \leftarrow a_3 = ( k i j ) } & & { 2 } & { b_2=(ik ) } \ar@{->}[l ] } \ ] ] the presentation is given by @xmath359 = 1,\ b_i^2 = 1,\ [ b_i , b_j ] = 1,\\ b_1^{-1 } a_1 b_1 = a_1,\ b_1^{-1 } a_2 b_1 = a_2^{-1},\ b_1^{-1 } a_3 b_1 = a_3^{-1},\ b_2^{-1 } a_1 b_2 = a_1^{-1},\ b_2^{-1 } a_2 b_2\\ = a_2,\ b_2^{-1 } a_3 b_2 = a_3^{-1 } \rangle$ ] . extrapolating from @xmath120 to a finite number @xmath35 of vertices , we can give generators for @xmath360 as @xmath361 , where the @xmath362 are transpositions of @xmath12 colours on edges @xmath363 for all @xmath364 and the @xmath365 are @xmath120-cycles on colours on edge @xmath366 . the relations in the presentation are : @xmath367,@xmath368=1 $ ] @xmath369,@xmath370,@xmath371=1 $ ] @xmath372 , @xmath373 certainly @xmath360 satisfies this presentation . but we must show that this gives a defining set of relations for @xmath360 . if @xmath50 is a group defined by this presentation then @xmath374 and we also have the surjection @xmath375 . therefore we have equality and a verification that this is a presentation for @xmath360 . we can extrapolate further from @xmath376 to any finite number @xmath20 of colours , and use the presentation @xmath377 . this gives the generators for @xmath315 as : @xmath362 where @xmath362 is a @xmath12-cycle @xmath378 on colours on all edges @xmath366 for all @xmath379 , and @xmath380 for @xmath381 which are @xmath120-cycles @xmath382 of colours on all edges @xmath366 . then @xmath383 . the relations are : @xmath384 = 1\ ( \text{if}\ \{i',j'\ } \neq \{i , j\}),\quad b_i^{2}=1,\ [ b_i , b_{i ' } ] = 1\ ( \text{if}\ i \ne i')$ ] , @xmath385 each edge @xmath366 represents one alternating group in the direct product expansion of @xmath315 . to prove that this presentation defines precisely @xmath315 , first recall the presentation @xmath386 for fixed @xmath363 and for all @xmath49 . making the substitutions @xmath387 we see that the @xmath380 generate @xmath318 . then proceeding as for @xmath360 proves the presentation . how might this impact applications ? the obvious question now is that although a large variety of groups can be supported by the graph edges and vertices , the generators @xmath365 and @xmath362 of the switching groups and the groups themselves arise naturally as _ reducts _ ( or closed over - group of the automorphism group ) and given the @xmath32@xmath32 correspondence between graph vertices and edges and lattice sites and links , do these generators or switching groups have a natural application in lattice gauge theory or gauge glass theory ? _ formal appearance of gauge symmetry by definition _ regardless of whether or not gauge symmetry is a true symmetry , it has been a central notion in the formulation of the field theories that have had great success in predicting and explaining a good deal of fundamental physics . so we give a summary of the purely formal derivation of gauge symmetry appears in @xcite . define a euclidean action for @xmath272 lattice electrodynamics @xmath388 on a regular hypercubic euclidean space - time lattice with lattice spacing @xmath253 . define on each link , denoted @xmath389 , the variables @xmath390 the flux variables @xmath266 , for the plaquette denoted @xmath265 , is given by the product of four link variables on neighbouring links . if @xmath391 then with @xmath392 where @xmath280 is the gauge coupling constant , the first term of equation ( 1 ) gives a pure @xmath286 gauge theory which is invariant under the lattice gauge transformation @xmath393 , through @xmath394 & { \bullet}_y ) \to \lambda(x ) } u(\xymatrix{{\bullet}_x \ar@{-}[r ] & { \bullet}_y ) \lambda^{-1}(y)}\ ] ] here @xmath395 and @xmath252 refers to the direction of the link @xmath396 & { \bullet}_y } \ ] ] connecting the sites with coordinates @xmath397 and @xmath398 . gauge invariance is formally introduced by hand ( subscript ` h ' ) by expressing the action in terms of field variables @xmath399 & { \bullet}_y ) \in \operatorname{u(}1 ) } , \quad h(\xymatrix { \bullet } ) \in u(1)\ ] ] defined respectively on the links and sites , and where @xmath394 & { \bullet}_y ) : = h^{-1}(x ) } u_h ( \xymatrix{{\bullet}_x \ar@{-}[r ] & { \bullet}_y ) h(y).}\ ] ] the link variables are functions of oriented variables : @xmath400 & { \bullet}_y ) = } u_h^{-1 } ( \xymatrix{{\bullet}_x \ar@{->}[r ] & { \bullet}_y ) = } u_h^ { * } ( \xymatrix{{\bullet}_x \ar@{->}[r ] & { \bullet}_y ) .}\ ] ] the new site variable @xmath401 could represent a higgs field . the original field @xmath402 & { \bullet}_y } ) $ ] and action @xmath403 & { \bullet}_y } ) ] $ ] which is a function only of this field are both invariant under the artificial gauge transformation @xmath404 : @xmath405 & { \bullet}_y ) \to \lambda^{-1}(x ) } u_h ( \xymatrix{{\bullet}_x \ar@{-}[r ] & { \bullet}_y ) \lambda(y),}\ h(x ) \to \lambda^{-1 } ( x ) h(x).\ ] ] using the invariance of @xmath406 of equation ( 3 ) under a @xmath272 gauge transformation , the action in equation ( 5 ) for @xmath272 lattice electrodynamics exhibits formal gauge invariance in terms of the new variables as : @xmath407 = \beta \sum_{\square } re[u_{h \square } ] + \alpha \sum_{- } re[h^{-1}(x ) u_h(\xymatrix { { \bullet}_x \ar@{-}[r ] & { \bullet}_y ) h(y)}].\ ] ] the _ mean field approximation ( mfa ) _ of lattice gauge theory , characterises the physical behaviour of the fields locally _ at the lattice scale _ ignoring correlations over distances much larger than the lattice spacing and long wavelength effects . the fields correspond to a phase transition between confined and coulomb - like phases . the formal gauge symmetry of equation ( 5 ) is sufficiently close to the physics to produce a coulomb phase with a massless photon . to demonstrate this in the mean field approximation , link and site variables are assumed to have expectation values which are determined by self - consistency conditions , that is one link or site variable is allowed to fluctuate quntum mechanically whilst all fields on neighbouring links and sites are replaced by their mean values . working one link or site at a time spontaneously breaks the local gauge symmetry without needing to work in a fixed gauge . translational and rotational invariance of the vacuum are used to make the gauge - higgs mean field ansatz : @xmath408 & { \bullet}_y ) \rangle = v_{u_h}\quad \text{and}\quad \langle h(x ) \rangle = v_h,}\ ] ] where @xmath409 and @xmath410 assume the same real value for each link and site respectively . from equation ( 7 ) , the real expectation value @xmath409 is independent of the link orientation , and because @xmath411 ( as for an antiunitary operator ) it follows that @xmath412 . to calculate an effective single - link action replace all variables except for @xmath413 & { \bullet}_y ) } $ ] by their mean value in the action of equation ( 9 ) . then look for the unique self - consistent solutions of @xmath409 and @xmath410 for every coupling constant pairing @xmath414 , this being the physical solution with the lowest free energy , that is the largest value of @xmath415 where @xmath5 is the partition function @xmath416.\ ] ] the 3 phases in the mean field approximation are:- \1 ) _ strong coupling or confined phase _ , including the region of small @xmath217 and @xmath417 values and corresponds to the trivial mean field solution with @xmath418 . \2 ) _ higgs phase _ , including the region of large @xmath217 and @xmath417 values and corresponding to a solution with @xmath419 . \3 ) _ coulomb phase _ , including the region of large @xmath417 values and bounded @xmath217 and corresponding to a solution with @xmath420 but @xmath421 . in the coulomb phase there are long - range correlations corresponding to a massless photon due to the formal gauge invariance , that is the gauge properties of the vacuum . from equation ( 8) the vacuum expectation values ( vev ) @xmath409 and @xmath410 under global gauge transformations with a constant gauge function @xmath217 behave as : @xmath422 : @xmath423 and @xmath424 + and under local gauge transformations with a linear gauge function @xmath425 : @xmath426 and @xmath427 , + where @xmath428 is the appropriate lattice link vector . it is then argued @xcite that a phase containing a massless photon must have that ( i ) the gauge symmetry for linear gauge functions be spontaneously broken , that is the vacuum is not invariant under @xmath429 , but that ( ii ) the global part of the gauge symmetry must not be spontaneously broken , that is the vacuum is invariant under a constant @xmath430 . because the coulomb phase satisfies both @xmath421 and @xmath420 , it must contain a massless gauge particle ( the photon ) . so the dynamics of the gauge non - invariant action of equation ( 4 ) for @xmath286 lattice electrodynamics can generate a massless gauge symmetric theory with no parameter fine - tuning . this generalizes to a non - abelian lattice gauge theory with a gauge - symmetry violating lattice action of equation ( 4 ) which is effected by taking the @xmath431 link variables to be non - abelian group valued . in the non - abelian theory there is no region in @xmath414 parameter space corresponding to a coulomb phase , but this appearance of gauge symmetry `` ex nihilo '' also follows for a non - abelian group @xcite . thus the field theory glass model can thus be used to demonstrate the very general nature of the inverse higgs mechanism and the spontaneous appearance of gauge symmetry . a thought experiment to produce a monte carlo simulation of such a model begins with a randomly selected set of sites @xmath95 in four - dimensional space - time . a manifold @xmath96 is constructed at each point @xmath183 on the random lattice and varies from site to site . the basic dynamical variable is a generalized quantum field @xmath432 . the parameters of an action @xmath75 are randomly chosen numbers . monte carlo methods are used to study long wavelength properties . the claim is that the physical degrees of freedom , with the most long range correlations , are similar to the gauge degrees of freedom in the above translational invariant lattice models . ( translational invariance is also key to the proof of the nambu - goldstone theorem ) . so the introduction of variables such as @xmath433 and @xmath236 should be possible , leading to a precise formal gauge invariance , which in turn is realised as massless gauge particles or as low energy confinement . the non - poincar invariant field theory glass action is a sum @xmath97 = \sum_r \mathcal{s}_r(\phi(i)),\quad i \in r\ ] ] over quenched random terms from small overlapping space - time regions @xmath99 of the order of the planck size . each small region of space - time has an approximate local gauge symmetry of the action , together with a gauge group . assume that within the totality of space - time there is an overlapping area with a large density of such regions . the approximate local symmetry groups @xmath434 for a site @xmath435 now converts into exact gauge symmetries through the introduction of ` by hand ' variables @xmath436 , where the higgs field is defined on @xmath435 . define @xmath437 to have the same structure as the fundamental field @xmath135 , by @xmath438 where @xmath439 is the mapping of @xmath135 by an element @xmath440 from the direct product of all the local symmetry groups @xmath434 . the field @xmath441 at site @xmath183 is transformed only under the @xmath434 for which the gauge area contains @xmath183 . the action @xmath442 $ ] expressed in terms of new variables , @xmath443 = \mathcal{s}[\phi = \phi_h^{h}]$ ] is automatically stable under the formal gauge transformation @xmath444 because @xmath135 is also invariant under this artificial map @xmath445 the field theory glass expressed in terms of @xmath436 manifests a formal gauge symmetry in which the symmetry group changes _ randomly _ from place to place . any given gauge group @xmath446 is an approximate gauge symmetry of @xmath442 $ ] . to discuss the situation where the associated higgs field @xmath236 in the new action @xmath443 $ ] has local fluctuations and no long - range correlations , and so a vanishing ` mean value ' in the vacuum , focus on gauge areas within which the symmetry breaking terms of @xmath442 $ ] are small corresponding to a small @xmath217 in the previous lattice model . in order to achieve vacuum invariance under global transformations assume that @xmath447 . to describe deviations from the globally gauge invariant vacuum state , modify the vacuum state by a continuum yang - mills gauge field @xmath100 for @xmath446 with generators @xmath448 . consider all gauge areas containing @xmath183 with small - enough @xmath446-symmetry violating terms in @xmath442 $ ] to ensure that the correlation function between the higgs field at two points , @xmath449 and @xmath450 decays exponentially with the separation . assume @xmath100 is constant over all the gauge areas containing @xmath183 and having coordinates @xmath451 . in order to modify @xmath437 at each site so as to establish a new configuration corresponding to @xmath100 , apply to @xmath441 one gauge transformation @xmath452\ ] ] per gauge area containing @xmath453 . the @xmath454 are coordinates of the gauge area centre @xmath435 . the gauge transformation @xmath455 has the same dependence on gauge area center @xmath454 for every field variable @xmath456 . the above construction of a continuum gauge field fails for a group with outer automorphisms but can be salvaged if the field theory glass representations are not symmetric under the automorphism . whilst the sm group @xmath241 is symmetric under complex conjugation ( see next chapter ) , the chiral quark and lepton representations are not . for each continuum field configuration @xmath100 assign a new configuration of the field theory glass . the natural correspondence between gauge transformations on the @xmath100 and the field theory glass variables @xmath456 means that the effective action for @xmath100 is invariant under @xmath446 gauge group transformations . thus if @xmath446 has no outer automorphisms a yang - mills field theory can be established for it , and if it does have outer automorphisms then arbitrary choices would need to be made all over space - time . in conclusion the formal gauge invariance arises from the definition of what froggatt and nielsen call the ` human variables ' @xmath436 . importantly for potential applications of our connections between random graphs and lattices , the lattice used in the above derivations do not have to be regular but can just as well be a discretized model , which in random dynamics can be taken to be fixed sites distributed randomly in four - dimensional space - time . equally this mechanism should work for very general field theories including a `` field theory glass '' where coupling constants , the type and number of degrees of freedom and other parameters are chosen in a _ random quenched way _ , that is , fixed and not varied in the quantum mechanical functional integral - by analogy with the atomic binding in a real glass . from a philosophical point of view , if everything were totally random , then it would be hard to discern much physics ; it is because there are some constants or patterns despite the randomness that we can pinpoint physical laws . some of these could be the quenched random terms that have been identified in the random dynamics program . _ the multiple point principle ( mpp ) _ we saw how the continuum gauge field @xmath457 can be implemented on the gauge glass by a suitable modification of the gauge glass degrees of freedom @xmath441 . if the gauge field @xmath457 for the gauge group @xmath446 can only be implemented very sparsely in space - time rather than throughout space - time , then the effective continuum gauge field lagrangian density acquires a term of the form @xmath458 with a very small coefficient @xmath459 because @xmath457 would represent relatively few gauge glass degrees of freedom . this implies a strong gauge coupling constant @xmath460 , so the gauge degrees of freedom are confined close to the planck scale and become irrelevant to the low energy levels accessible in the laboratory . for a gauge group @xmath446 to survive to low energies it must be a good approximate symmetry of the fundamental field theory glass action @xmath461 $ ] . one of the assumptions of the random dynamics project is that there is an _ anti - unified _ gauge group @xmath462 ( see below ) on which the plaquette action of the lattice gauge theory is defined as a direct product of @xmath463 standard model group factors of the form @xmath464 , that at the planck scale ( @xmath465 gev ) breaks down to the _ diagonal subgroup _ @xmath466 due to a so - called _ confusion mechanism _ @xcite @xcite arising from group outer automorphisms - which of the identical but different groups being acted on by the outer automorphism to choose . the diagonal subgroup consists of ordered sets of sm group elements @xmath467 for which all the components are equal to each other . in fact the speculation is that there is a successive breaking of groups with many direct products down to a collection of groups with particularly few automorphisms such as @xmath241 itself , with @xmath468 being the penultimate group in this series . given the diagonal subgroup , the naive lattice continuum formula on an assumed fundamental lattice with lattice constant @xmath253 is @xmath469 & { \bullet}_{y_{\mu } } ) } \simeq exp(i g a a^{a}_{\mu } \frac{1}{2 } \lambda^a),\quad y_{\mu } = x_{\mu } + a \delta_{\mu},\ ] ] implies @xmath470 for such a diagonal subgroup , the original fundamental lagrangian density @xmath471 becomes @xmath472 with @xmath473 if the couplings for the ` fundamental ' fields @xmath474 are all assumed at the fundamental scale to be equal to a certain critical coupling at that scale then the observed coupling will obey @xmath475 if many vacua are degenerate then the corresponding phases meet at a certain point in the phase diagram of the theory . the phase diagram of any gauge theory is represented by a space which has axes given by bare coupling constants , and perhaps also by bare masses . the criticality of the gauge coupling constants at the planck scale means that at this scale the couplings lie at the boundary between the coulomb ( photon ) and the strong coupling ( confining ) phases . a calculation of critical couplings at a phase boundary , applying to both phases , can be done using the mean field approximation ( mfa ) . the mpp mechanism within the random dynamics program postulates the existence of a point in the phase diagram for a yang - mills lattice gauge theory of the standard model with gauge group @xmath468 where a maximum number of vacuum phases ( all with the same energy density ) come together so that the vacuum is maximally degenerate . the principle of multiple point criticality then requires that the running gauge coupling constants correspond to the multiple point critical values in the phase diagram of a lattice theory and so the mpp explains the fine tuning of the free parameter values of the standard model . the multiple point is where all or at least a maximum number of the phases of the lattice gauge theory meet . ` phase ' here means different behaviour , analogous to the 3 phases of water ; in the relevant phase diagram water and vapour are indistinguishable at the critical point and ice , water and vapour exist simultaneoulsy at the triple point . the multiple point principle was born in the lattice investigations of gauge theories , and particularly monte carlo simulations of @xmath272 , @xmath295 and @xmath296 gauge theories on the lattice which alluded to the existence of the triple critical point . there is a phase for each normal ( invariant ) subgroup of @xmath468 . the system is very stable because there is an entire range of parameter values for say the mean energy and volume per molecule . the observed coupling values are the critical multiple point values that the couplings take when the vacuum is maximally degenerate . the multiple point values of the plaquette action parameters for the diagonal subgroup of @xmath468 are in the continuum limit equal to the experimental gauge coupling values that have been extrapolated to the planck scale , thereby justifying use of the gauge group @xmath468 . grand unified theory models based on simple gauge groups @xmath476 and @xmath477 place left - handed fermions and their antiparticles in the same representation and seem to lead to families with a pair of almost mass degenerate particles . but inter - generation masses are not approximately degenerate @xcite . random dynamics bypasses considerations of such gauge groups , and further assumes that the chiral horizontal symmetry group commutes with the sm gauge group . random dynamics , through the confusion mechanism breakdown , does seem to favour the sm gauge group @xmath241 , and furthermore relates the number 3 of generations to the phenomenological values of the gauge coupling constants for the 3 invariant subgroups of the sm group @xcite , but this result essentially requires there to be no new phenomena between accessible energies and that of the planck scale . one of the most important conclusions of das and laperashvili s review @xcite of the multiple point principle ( mpp ) is the validity of an approximate universality of the critical couplings . they showed that values of the phase transition couplings can be crudely approximated without using any specific lattice , so that the details of both the lattice and the regularization are unimportant . critical couplings depend only on groups with some regularization . this approximate universality is essential in comparisons of the lattice phase transition couplings with the experimentally observed couplings . within the grand unified theories program it is assumed that different irreducible representations ( ir ) of the @xmath241 will combine into a single ir of a larger group , however this approach does not seem to produce a satisfactory explanation of the three family generation structure within the sm . family ( or horizontal ) symmetries are to be differentiated from gut symmetries which unify different members within a family . it was argued in @xcite that the fermion mass hierarchy suggests extensions of the sm which have the same ir as arise in the sm and which embed @xmath241 as a diagonal subgroup . the fact that no pair of correponding particles in different generations is degenerate makes it less likely that horizontal flavour groups can account for the generation gaps : if such pairs were positioned into multiplets of groups such as @xmath478 in an @xmath479 model it would suggest that these particles are degenerate . also in order for an over - group @xmath50 of @xmath241 to account for the gaps , corresponding particles from different generations should belong to inequivalent irs of @xmath50 , and furthermore single irs of the @xmath241 should form irs of @xmath50 . a more interesting possibility is the diagonal embedding suggested above . whilst the sm does not contain any quantum number which distinguishes the lepton and quark generations it is suspected that such a number must exist , and the 3-coloured random graph provides one approach to finding it . having @xmath480 many @xmath481 factors allows for different gauge quantum numbers for each of the @xmath480 families @xcite , an idea that mirrors the possibility that we have suggested that these are colours of the triality graph @xcite . we return to this point in the next chapter . _ tight packing _ for all of the successes of the standard model it predicts neither the number of generations nor the pattern of generation replication . it also leaves unresolved the large number of free parameters . a random dynamics - inspired model relating the smallness of the gauge coupling constants to the number of generations predicts 3 generations @xcite . in @xcite it was suggested that a tight packing interpretation of the leptons of the standard model in some space could be part of the solution to the family structure within the sm . a. kleppe has used a tightest packed three - dimensional lattice which basically represents the abelian sector of the standard model group , i.e. @xmath482 in an @xmath463-dimensional space corresponding to three generations , within the anti - grand unification scheme . the tight packing arises because each lattice site has a maximal number of neighbouring sites . dense packing means that each lattice site has the maximum number of nearest neighbours at the same critical distance . each lattice direction where the nearest neighbour distance is smaller than that corresponding to the critical coupling value , has a confining subgroup . the critical coupling is usually a function of the remaining @xmath272 degrees of freedom . but the critical distance in one direction is approximately independent of the nearest neighbour distance in other directions , so approximate mpp criticality is achieved when the nearest neighbour distance is critical in all directions . this means a maximal number of @xmath32-dimensional subgroups , or having the tightest possible packing of lattice points . this is the origin of tight packing in kleppe s argument . the lattice packing density is the ratio of the volume of a sphere to the volume of the fundamental region . in lattice packing a regular sphere is a repetition of one lattice crystal but in sphere - packing the position of every sphere is a variable . the origin and nature of the tight packing differs in the random graph context and the lattice gauge theory context , but it is interesting that the same general idea has been called upon . [ oaut ] the groups @xmath272 , @xmath302 @xmath483 and the sm group @xmath241 have an outer automorphism that corresponds to charge conjugation symmetry . for @xmath302 groups ( @xmath484 ) it is complex conjugation of the matrix elements . in the sm this symmetry is broken because only left - handed fermions and right - handed anti - fermions couple to the @xmath485 intermediate vector boson . note that @xmath486 , odd-@xmath35 spin and symplectic groups have no outer automorphisms . chirality breaks the charge conjugation symmetry of the gauge fields and ensures that gauge groups such as @xmath272 and @xmath6 survive the breakdown from larger to smaller groups . the confusion breakdown through a charge conjugation automorphism is not possible because left- and right - handed fermions appear in different representations . however there may be non - representation theoretic reasons for a phenomenon . the order-2 outer automorphism of @xmath39 might be taken to correspond to charge conjugation @xcite . the random dynamics program goes further assuming that a gauge field can be permuted with an automorphic image of itself , and that this can happen locally and not globally . the diagonal subgroup of the direct product of isomorphic groups is the maximal invariant subgroup of the permutation automorphisms ; it is symmetric under gauge transformations generated by constant gauge functions that correspond to the global part of a local gauge transformation . the diagonal subgroup of @xmath462 is stabilized by the automorphic permutation of the @xmath480 factors in @xmath462 . but then the matter field content of each factor must have the same structure . this and the known fact that what is normally taken to be a fermion generation gives the simplest arrangement of particles that avoids gauge anomalies suggests that the @xmath480 factors of @xmath462 are just repetitions of the @xmath487 , one for each generation . the assumption is that the final breakdown of gauge symmetry by confusion is caused by the automorphism that permutes the @xmath480 isomorphic product factors @xmath241 in @xmath462 . random dynamics suggests the existence of an intermediate gauge group @xmath488 at the confusion scale ; with one direct product factor @xmath489 for each quark - lepton generation . with such a direct product unified gauge group , it is possible that the separate gauge quantum numbers of @xmath490 , @xmath491 and @xmath492 could play the role of horizontal quantum numbers and be responsible for the different mass scales of the three generations . but what is the origin of these quantum numbers ? even within random dynamics there is the problem of understanding the origin of the @xmath241 and of the number of quark and lepton generations . we outlined a different type of solution based on the 3-coloured random graph in @xcite . random dynamics relates the number @xmath463 of generations to the phenomenological values of the gauge coupling constants for the three invariant subgroups of @xmath241 , a result that seems to require there to be no new physics between presently accessible energies and the planck scale , which is surely unlikely . there is an additivity of the inverse squared couplings of the group factors in the direct product @xmath468 in going to the diagonal subgroup of the form @xmath493 that is at the multiple point each term in this series become equal to the same value . for the abelian @xmath272 the deviation of the coupling in going from the multiple point of @xmath468 to the diagonal subgroup is 6 instead of 3 . when considering the action corresponding to @xmath494 the abelian part is treated separately from the non - abelian part . for the non - abelian subgroups the action is additive @xmath495 meaning the confining phases correspond to factorizable normal ( invariant ) subgroups so the phase diagram for @xmath494 is determined from those of the individual factors . the @xmath120-dimensionality of the abelian group @xmath482 gives the underlying lattice and the number of generations . this lattice brings together a maximal number of phases and is found using the principle of multiple point criticality . the lattice sites can correspond to quantized quantities such as charges . to introduce dynamics into this picture absorb the @xmath272 coupling into the metric @xmath496 on the space of charges . taking @xmath272 to be @xmath497 , the metric is used in forming the inner product of two @xmath498-variables . the @xmath272 viewpoint is that the lattice sites @xmath499 are identified with the origin . but by including the coupling into the definition of distance , the distance between two charges will be a function of the values of the charges . for confinement along @xmath500 the critical distance is longer than the lattice link distance . for every pair @xmath501 of subgroups of @xmath241 , where @xmath502 is a normal subgroup of @xmath503 , there is a phase , and all the phases contribute to the action . for example , @xmath504 begets @xmath505 as a lattice action @xcite . in every small space - time region there is an ambiguity in choosing how to represent the continuum gauge field . the difficulty in implementing a continuum yang - mills field @xmath457 on a gauge glass with gauge group @xmath446 having outer automorphisms , is in deciding which degrees of freedom of the gauge glass to identify with which degrees of freedom of @xmath457 @xcite . by contrast the the ambiguity in identifying the continuum field due to inner automorphisms is just the usual gauge ambiguity in defining the gauge field . it adds a background field but does not give rise to any ambiguity in the physics . for an inner automorphism , @xmath506 the transformed field @xmath507 is related to @xmath457 by a gauge transformation , with a constant gauge function : @xmath508 where @xmath509 and @xmath510 are group generators . inconsistencies corresponding to the inner automorphism @xmath511 can be gauge transformed away but not those corresponding to outer automorphisms . because a continuum yang - mills theory for a gauge group with outer automorphisms is unlikely to be implemented on a gauge glass , random dynamics predicts that the continuum gauge group must have no outer automorphisms that can be extended to true discrete symmetries . _ generalised outer automorphisms _ of a non - semisimple group are isomorphisms between two of its factor groups obtained by dividing out low order normal subgroups . they act as ordinary outer automorphisms of the lie algebra and thus of the continuum gauge fields @xmath457 . according to the random dynamics program a generalised outer automorphism causes an inconsistency in the identification convention for the gauge glass degrees of freedom , because of an ambiguity involving the identification of a factor group of the gauge group @xmath446 of the continuum gauge field . such an identification means that the group elements in the coset of the factor group must correspond to the same continuum field @xmath457 . there should then be an approximate symmetry under permutations of the the group elements inside a coset . what froggatt and nielsen have termed the ` confusion mechanism ' is supposed to collapse the gauge group @xmath446 with its non - trivial outer automorphisms near the planck scale . using a generalized lattice gauge theory helps to identify collapse mechanisms that break @xmath446 somewhere between the planck scale and the low energy scales accessible to us today . the following modifications to the lattice gauge theory , listed in @xcite are chosen to simulate the action for a gauge glass but the feynman path integral is stabilised : \1 ) taking the lattice sites to assume random positions separated by distances of the order of the planck length gives an amorphous ` glass - like ' structure instead of a regular hypercubic lattice . 2 ) the gauge group can vary from site to site . \3 ) the plaquette variables @xmath266 are formed from link variables for the gauge group @xmath446 , denoted @xmath283 & { \bullet})}$ ] , as is the plaquette action @xmath267 . the plaquette variables and action contributions are invariant under the ` confused gauge transformation ' @xcite : @xmath512 & { \bullet}_y ) \to f_{x , y}(\lambda(x))^{-1 } } $ ] @xmath243 & { \bullet}_y ) f_{y , x}(\lambda(y ) ) } $ ] ( ) + rather than simply under the usual gauge transformation @xmath512 & { \bullet}_y ) \to \lambda(x)^{-1 } } $ ] @xmath243 & { \bullet}_y ) \lambda(y ) } $ ] + where the @xmath513 are quenched randomly chosen outer automorphisms of the gauge group @xmath446 . \4 ) assume that @xmath267 is a random function of @xmath266 . gauge invariance requires @xmath267 to take the same value for all conjugate group elements @xmath266 ( those related by an inner automorphism ) . so @xmath267 can be expressed as a sum of the characters for the various representations @xmath99 of @xmath446 , @xmath514 where the coefficients @xmath515 are quenched random variables obeying @xmath516 and @xmath517 denotes the complex conjugate representation to @xmath99 . the character for representation @xmath99 is @xmath518 where @xmath519 is the representation matrix for @xmath266 in representation @xmath99 . the @xmath515 coefficients are chosen from a gaussian distribution , independently for each plaquette and for each pair of conjugate representations . for @xmath520 to be convergent and reasonably smooth , choose the distribution to have zero mean and a width which decreases with the dimension of the representation . \5 ) the link variables are group - valued and the group structure varies randomly from link to link , as in fig . 7.3 of @xcite . the plaquette variables @xmath266 are then constructed from common factor groups of the groups on the surrounding links . ambiguities arising from outer automorphisms are inherent in identifying factor groups of different but isomorphic link groups leading us to consider the confused gauge transformations of point ( 3 ) . how does the amorphous nature ( that is randomness ) of these modifications of the usual lattice gauge theory break , at the planck scale , the gauge groups present _ a priori _ in this gauge glass ? when the vacuum state is non - invariant under the global gauge symmetry , spontaneous symmetry breakdown occurs , as happens for example when gauge bosons acquire a mass according to the englert - brout - higgs - guralnik - hagen - kibble mechanism . the vacuum state of the theory in the classical approximation can be studied by considering individually the energetically preferred value of each plaquette variable @xmath266 . the link and plaquette variables in the vacuum state must belong to the center of the gauge group @xmath446 otherwise the vacuum will not be invariant under global gauge transformations and gauge symmetry will collapse . the following are required to prevent such a collapse : \(a ) _ absence of outer automorphisms . _ for a gauge group @xmath446 with non - trivial outer automorphisms the gauge glass action is invariant under the above confused transformation @xmath521 . assuming that the vacuum values of the corresponding confused link variables @xmath512 & { \bullet}_y ) } $ ] lie in the center of the group , the vacuum values will only be invariant under global gauge transformations @xmath522 if they obey @xmath523 for all outer automorphisms @xmath513 randomly chosen throughout the lattice . so the vacuum state is only symmetric under the subgroup which is left invariant by all the outer automorphisms of @xmath446 , and so @xmath446 spontaneously breaks down to a subgroup with no or few outer automorphisms . inner automorphisms may be transformed into background gauge fields which may in turn cause the gauge particles to gain masses through the englert - brout - higgs - guralnik - hagen - kibble mechanism . the confusion mechanism boils down to inconsistencies in identification for the gauge glass degrees of freedom , and it is possible to circumvent the mechanism and allow a gauge group with a few outer automorphisms to survive , by including matter degrees of freedom . \(b ) _ a non - trivial center . _ for a vacuum state to be invariant under the global gauge transformations of a group @xmath446 with a trivial center , all the plaquette variables @xmath266 must equal the identity element in @xmath446 . the coefficients of the gauge plaquette action @xmath524 may have either sign , corresponding either to an energy maximum or minimum . so in about half the cases of the plaquette energy being minimized , the group element will not be central meaning that the vacuum is not invariant under the global @xmath446-action , which is thus broken . therefore in a gauge glass , groups with a trivial center are likely to collapse and groups with a non - trivial center survive . \(c ) _ a connected center . _ applying a similar argument to ( b ) to a set of neighbouring plaquettes instead of just one , suggests that gauge groups with a topologically disconnected center tend to collapse . the plaquettes forming the surface of a generalised 3-cube of the amorphous lattice obey the bianchi identity , @xmath525 which arises from the fact that each link is contained in two plaquettes of the cell and the plaquette variables @xmath266 are thus not independent . if the @xmath266 lie in the center of the group @xmath446 then the product of the plaquette variables forming the cell surface must equal the unit element . the identity ( [ bianchi ] ) is not automatically satisfied when each plaquette variable is chosen to minimise its energy . the energetically preferred individual plaquette variables @xmath526 must belong to the center to stop collapse of @xmath446 , but due to the plaquette action being random we expect the product @xmath527 to be a random central element . if @xmath528 then the true vacuum values @xmath529 of the plaquette variables can not equal @xmath526 . if the group center were not topologically connected then for some cells @xmath530 would belong to a different connected component to that of the unit element . it follows that some of the deviations @xmath531 would be non - central , so as the @xmath526 are assumed to be central , some of the @xmath529 are non - central and the gauge group @xmath446 collapses . to prevent such a gauge symmetry breaking , @xmath446 must have a connected center . ( d ) _ conjugacy class space with only central singularities . _ this is described in @xcite . these four conditions favour the standard model group @xmath241 as surviving collapse in the gauge glass and surviving as a low energy symmetry group . it is the one group whose lie algebra has a non - trivial connected center . it s one outer automorphism is complex conjugation . regarding points ( a ) and ( b ) above , note that if @xmath532 then @xmath533 is a @xmath12@xmath32 surjection . in fact one way of knowing that the cartan triality automorphisms ( discussed at length in @xcite and @xcite ) are outer is because they act nontrivially on the centre @xmath534 of @xmath535 . the consistent string theories have either of the two gauge groups @xmath536 or @xmath537 , both of which have trivial centers . group transformations act in pairs on the two ends of a string but elements of the center act as the identity . can the strings be considered as graph edges and if so would it then be meaningful to consider in a string context the physical meaning behind the action of groups supported by the random graphs ? the outer automorphic transposition of the two @xmath538s in @xmath537 may provide an alternative to the specific model applying the confusion mechanism proposed as a global modification of the heterotic string model , wherein the two heterotic string @xmath538s are permuted across a confusion hypersurface of codimension one representing a global topology change in space - time @xcite , as a result of which the gauge group typically reduces to a single @xmath538 . _ group representation theory _ the standard model is based on the 12-dimensional algebra of @xmath539 and describes the forces between the gauge particles through a field theory . the three generations of basic fermions are three instances of a specific collection of representations for this lie algebra and thereby of the universal covering group @xmath540 , from which it is obtained by factoring out a discrete normal subgroup which is determined from a certain charge quantization rule @xcite . the pattern of the fundamental fermion charges suggests the importance of a particular group having this algebra , namely @xmath241 . ever since the ground - breaking work of wigner group representation techniques have provided one of the most effective set of tools used in the classification of particle properties . we mention just one such recent example , out of a vast literature going back over three - quarters of a century . in looking to select a criterion as to why the sm gauge group turns out as it is , the one chosen in @xcite is the maximisation of a modified ratio of the orders of the quadratic casimir for the adjoint representation over the quadratic casimir of a representation chosen so as to make the quadratic casimir the smallest possible non - trivial faithful representation . the lie gauge group would then have the smallest possible faithful representation in this sense of smallest , of the possible groups that gives the chiral weyl fermions ( leptons and quarks ) and the higgs boson . in @xcite and in previous work with bennett , nielsen begins with a group that is the semi - direct product of simple lie groups by the additive group of real numbers ; the number of abelian dimensions gives the dimension of the lie algebra . then he factors out of the center a normal discrete subgroup to get a group @xmath50 . the best known quadratic casimir is the angular momentum operator @xmath541 for @xmath542 . the target quantity is @xmath543 where the @xmath183 product runs over the normal simple groups with dimensions @xmath544 , the @xmath185 product is over abelian factors , @xmath545 is the dimension of @xmath50 , @xmath546 and @xmath547 are the `` charges '' for the @xmath530 and adjoint representations . the word charges is in quotation marks because being abelian the @xmath548 subgroups have no adjoint representations as the lie algebra basis of a lie group is only transformed trivially , so the second set of products in the above equation must be differently defined . the ratio @xmath549 is related to the dynkin index @xcite of a representation of a compact simple lie algebra . the target quantity has a value of at least 1 , and is precisely 1 for both @xmath538 and @xmath537 gauge groups @xcite . of the groups tested , the target quantity attains its maximum for the sm group . the conclusion is that a principle of small representations should be enough to imply a significant part of the details of the sm . despite the success of representation theory thus far the question can still be raised as to whether or not post - sm physics may not need new ideas and new approaches . our work has concentrated on the automorphism group @xmath550 of the triality graph , which being uncountably infinite in order , is less amenable to representation - theoretic techniques . however another criterion which is closer in spirit to our focus on the automorphism group of relational structures @xcite than representation theory provides , and that distinguishes the group @xmath241 from all other connected compact non semisimple groups with lie algebra dimensionality up to 12 is that it has few _ generalised outer automorphisms _ , in fact the least of any of the 192 connected non - semisimple groups with algebras of dimensionality 12 or less . brene and nielsen proposed studying the orders @xmath551 of pairs of normal subgroups in the definition of generalized outer automorphisms in order to select the standard model group @xcite , considering factoring out only discrete normal subgroups of the center of a starting group . groups with simple algebras have finite discrete center and so a finite number of possible pairs , and those with continuous centers such as @xmath272 have an infinite number of level pairs . nonetheless some groups that do contain @xmath272s have few outer automorphisms because such groups are obtained from their covering group by factoring out a discrete normal subgroup with nontrivial projections in both the @xmath272 factor and other group factors , and so the original generalized @xmath272 symmetry is destroyed . _ groups with triality and triplication _ despite the successes of the sm there are still around 20 parameters characterizing the couplings and particle masses which it does not explain . the family replicated gauge group model ( frggm ) was proposed in @xcite and @xcite as an extension of the sm . froggat and nielsen have presented @xcite a 5-parameter fit to the orders of magnitude of the quark - lepton masses and mixing angles in the frggm where the gauge group is @xmath552 or otherwise written @xmath553 ^ 3 \times [ \operatorname{su(}2)_l]^3 \times [ \operatorname{u(}1)_y]^3 \times [ u(1)_{(b - l)}]^3.\ ] ] in this extended sm model the gauge group and a gauged @xmath554 ( baryon number minus lepton number ) is extended to one set of gauge fields per fermion family . the 6 abelian gauge charges for the 3 families are the 3 weak hypercharges and the 3 @xmath555 charges , and these @xmath272 quantum charges generate the fermion mass hierarchy . higgs fields break down this family replicated gauge group to the sm group . in @xcite a set of higgs fields was constructed and vevs were assigned to them , which fit all the quark - lepton masses and mixing angles , by taking the fundamental couplings to be of order unity and the fundamental masses to be of order the planck mass ; see also @xcite . the frggm approach is called _ anti - grand unified theory_. it assumes that space - time is discrete at very small distances and was developed as an alternative to susy guts . there is a _ mass protection mechanism _ , whereby fermions would have zero mass from gauge ( charge ) conservation , were it not for the higgs mechanism . this is thought to be due to the existence ( of what are assumed to be gauge ) quantum numbers lying outside the sm which are assigned different values on the left - handed and right - handed weyl components of the sm quarks and leptons . it is hoped that an assignment of these chiral quantum numbers to the quarks and leptons can generate sm yukawa coupling constants suppressed by the appropriate combinations of higgs vevs so as to be in agreement with experiment . the frggm gauge quantum numbers are candidates for these mass - protecting chiral quantum numbers . one of the key elements of the random dynamics program and the mpp is the speculation of a successive breaking of groups from a large over - group to for example @xmath468 or an extension thereof , and finally to the standard model group @xmath241 . this final step is reminiscent of the phenomenon of _ groups with triality _ , discussed in @xcite , which is a generalization of cartan @xmath556 outer automorphism triality of the group @xmath306 . and the three representations of its double cover spin group @xmath535 . we conjectured the possibility of a unified triality whereby the embedding of three two - colour random graphs @xmath31 in one triality graph @xmath33 , three copies of the spin group @xmath557 in one spin group @xmath535 , three @xmath558 lattices in a leech lattice @xmath559 , and finally lie algebras with triality , are all related phenomena . we refer to @xcite and @xcite for discussions of these topics . in @xcite we motivated an argument for how the 3-family horizontal structure might arise from the triality graph , with the colours @xmath560 identifying the generations _ generational quantum numbers_. a different model inspired by random dynamics was given by d. l. bennett , h. b. nielsen and i. picek in @xcite . random dynamics takes as its starting point the idea that the most advantageous assumption that we can make about the fundamental laws is that they are random . that a fundamental theory is so beyond our reach that to all intents and purposes we can take its laws to be random . furthermore that the known laws of relativity and quantum mechanics emerge from this fundamental theory . despite its successes the project has thus far this project has been a niche endeavour . by identifying a candidate as the sought after random structure we hope to increase interest in this line of enquiry . the countably infinite random graph @xmath31 is the unique ( up to isomorphism ) connected structure that encodes the connections of every possible configuration of sets of vertices . multicoloured versions of the graph are natural extensions . the graph @xmath31 can in fact be reconstructed from its automorphism group @xmath39 ( as can the multi - coloured versions @xcite ) so that in the right context the group can be thought of as a ` prior ' construct . symmetries of the physical laws that seem to govern the wokings of nature are not in general symmetries of the world because the latter are subject to _ initial conditions _ , which are not invariant under symmetries of nature s laws @xcite . it would appear that many of the well - known symmetries arise naturally out of the sm . the @xmath561 theorem can be derived for any quantum field theory from a few assumptions ( * ? ? ? * chapter 5 ) . poincar and gauge invariance ( and perhaps translation invariance ) must be added separately and so these are regarded as the fundamental symmetries @xcite . nonetheless there have been attempts to derive even these , classified according to three different methods ( * ? ? ? * chapter 6 ) , ( i ) formal dynamical derivations that reveal the symmetry which then takes on a physical interpretation in some vacuum phase due to quantum fluctuations , ( ii ) from the renormalisation group in the limit of larger distances or lower energies , and ( iii ) as a consequence of string theory properties . the automorphism group @xmath550 of the 3-coloured ( triality ) random graph is connected to the modular group @xmath8 through the construction of @xmath33 ( * ? ? ? * theorem 6.6 ) as a homogeneous cayley graph for an index @xmath120 subgroup of the modular group @xmath562 modular groups of fractional linear transformations acting on the upper half - plane in @xmath180 take the form @xmath563 by taking @xmath180 instead of @xmath86 as the normed division algebra for the modular group it is a short step by complexification to its lorentz over - group @xmath564 and we arrive at lorentz invariance which is thus , as required by the random dynamics program , derived from a random structure without having to be assumed a priori . an intermediate group in the complexification is the group of conformal automorphisms of the upper half - plane @xmath566 of which the modular group @xmath8 is a discrete subgroup . the upper half - plane can be thought of as a hyperbolic plane . in @xcite and @xcite feingold , kleinschmidt , nicolai and palmkvist study isomorphisms between hyperbolic weyl groups and modular groups over integer domains in normed division algebras : @xmath567 @xmath568 @xmath569 @xmath570 where for example , @xmath566 is the identity component of the lorentz group of 3-dimensional spacetime , acting on the hyperboloid @xmath571 this being one representation of the hyperbolic plane . the isomorphism between the lorentz group @xmath572 and the complexified rotation group @xmath573 is the basis of the formalism that lies behind various self - dual formulations of general relativity @xcite , such as that of plebaski @xcite which uses self - dual two - forms instead of the dynamics of a metric , and of which the hamiltonian formulation of general relativity due to ashtekar @xcite is the phase space version as proved by jacobson and smolin @xcite . a model has been constructed by froggatt , das , laperashvili , nielsen and tureanu @xcite which uses this formulation as well as the multiple point principle to unify gravity with some weak gauge fields . the _ integer octonions _ @xmath574 form a discrete subring of the octonions @xmath575 and is modulo a rescaling , just the @xmath538 root lattice which we denote @xmath558 , this being the smallest even unimodular lattice embedded within an 8-dimensional euclidean space . the discrete subgroup @xmath576 of @xmath577 acts on the 9-dimensional hyperbolic space @xmath578 , where @xmath579 is the identity component of the lorentz group of 10-dimensional minkowski spacetime . the above four identifications must account for the noncommutativity of both quaternions @xmath580 and octonions @xmath575 , and nonassociativity of @xmath575 . various ` integer ' subrings of the division algebras are described together with their relationships to hyperbolic coxeter groups @xcite . an example of nonassociativity in physics is the special relativistic velocity composition law that relates the velocities of moving objects in different reference frames . the velocities add as if they are hyperbolic tangent ( @xmath581 ) functions because the lorentz transformation can be thought of as the application of a hyperbolic rotation through a hyperbolic angle ( which is velocity divided by @xmath310 ) . this is used in the gyrovector space description of thomas precession , where in abraham ungar s theory gyrovector addition is based on a type of bol loop called a _ gyrogroup_. we see again that the rotation groups and modular groups are fundamental , just as we emphasized in @xcite . as indeed are lattices ; feingold et al studied a type of weyl group of hyperbolic kac - moody algebras which are intimately linked with the four normed division algebras @xmath582 , making crucial use of integral lattices as well as the integer subrings of the division algebras and associated discrete matrix groups . we constructed @xcite @xmath33 as a homogeneous cayley graph for the complex leech lattice as a special case of a construction of random cayley @xmath20-coloured graphs from lattices @xmath82 in @xmath583 , for @xmath584 , identifying lattice vectors with graph vertices and vector pairs with edges such that the graphs are invariant under the lattice automorphism group @xmath585 . this particular link may have an application to global ( that is , non - space - time dependent ) symmetry properties of gauge glass theoretic models . in @xcite , chkareuli , froggatt and nielsen argue that lorentz non - invariant effects caused by the spontaneous breakdown of lorentz symmetry ( sbls ) are physically unobservable , and that application of this principle to the most general relativistically invariant lagrangian , with arbitrary couplings for all the fields , gives rise to both abelian and non - abelian massless vector gauge fields . more specifically to the appearance of a continuous symmetry in terms of conserved noether currents and to the massless vector fields gauging this symmetry . in other words , gauge invariant abelian and non - abelian theories can be derived from the requirement of the physical non - observability of the sbls rather than by using the yang - mills gauge principle . matrices of coupling constants form lie algebra representations corresponding to this symmetry , and in the non - abelian case the coupling constants for the vector field self- interaction terms are the structure constants for this algebra . therefore the vector gauge fields become a source of the symmetries , rather than local symmetries being a source of the gauge fields as in the usual formulation . imposing the condition of unobservability of sbls restricts the values of the coupling constants and mass parameters in the lagrangian density . the issue of which mechanisms could induce the sbls is taken up in @xcite , where it is argued that the non - observability of the sbls caused by the vacuum expectation values ( vev ) of scalar gauge fields could provide the origin of all observed internal symmetries . the presence of scalar fields can possibly induce non - zero classical fields as the vevs of the original vector fields . the conclusion is that a gauge symmetry phase is created in order to avoid a breakdown of lorentz invariance . in an older reference , assuming only the classical electrodynamics of a charged particle interacting with an electromagnetic plane wave field , kupersztych derives @xcite a lorentz - transformation operator that describes the motion of the spin of a particle having a magnetic moment precisely that of a dirac spin particle , and is simultaneously a gauge transformation ( which thus leaves invariant a plane wave field ) . this connects gauge invariance , relativistic invariance and electron spin which has no classical analogue . however , the derivation assumes the lorentz force law which is obeyed by any charged particle , including the pion which has no spin . yang - mills theory is a gauge theory based on the @xmath302 group but can more generally be based on any compact , semisimple lie group . for example , electroweak and strong force theories are respectively based on @xmath6 and @xmath586 . a purely formal derivation of gauge symmetry appears in @xcite and as a consequence of lorentz invariance in the model of @xcite . if lorentz invariance can be derived or arrived at from a random discrete mathematical structure , whether through our purported route or any other , is there an effective discrete content to group - theoretic gauge invariance ? consider the well - known subgroup relationship between orthogonal and unitary groups @xmath587 , and in particular that @xmath588 . being compact the orthogonal group has discrete subgroups that are equivalent to finite subgroups , amongst which is the coxeter group @xmath589 of permutation matrices ( which are themselves orthogonal matrices ) . note also that to get from @xmath42 to its subgroup @xmath590 simply consider rotations that stabilize an axis ; this simple fact allows us to go down the orthogonal group series . further that @xmath591 with the projection map picking matrices with @xmath592 determinants . orthogonal matrices with determinant -1 can not include the identity , so form a coset rather than a subgroup . the projection map splits so @xmath42 is the semi - direct product @xmath593 . two references for the classical groups are @xcite and @xcite . the matrices representing @xmath303 have determinant @xmath594 describing _ proper rotations _ rather than rotations combined with reflections . but the determinant of a permutation matrix equals the signature of the corresponding permutation . so there are natural maps from @xmath595 , and @xmath596 . so there are maps @xmath597 , @xmath598 and @xmath599 . the following identifications are known : + @xmath600 , @xmath601 , @xmath602 , @xmath603 . is there a series of mappings with an enlightening physical interpretation , from @xmath8 to the standard model group @xmath241 ? there is an interest in these types of questions that make use of the important discrete ingredients in continuous groups . one such is the froggatt - nielsen mechanism for explaining the hierarchies of fermion masses and mixings in terms of a @xmath272 symmetry @xcite . the fermion masses and mixings are given by yukawa couplings of the form @xmath604 , for two interacting fermionic fields , a higgs boson and a yukawa matrix . for their proper description we could either fine - tune correlations amongst the entries of the yukawa matrix @xmath103 or we could assume the basic ( bare ) coupling constants to be random ( but at small distances are all of the same order of magnitude ) and that the entries of @xmath103 itself contain a hierarchical pattern . but what would be the origin of this randomness ? the mechanism proposed introduces so - called _ flavon _ @xmath272 charged fields @xmath135 , which are neutral under the sm gauge group but that like the higgs , develop a vacuum expectation value ( vev ) to generate mass , via the introduction of an expansion parameter @xmath605 where @xmath393 denotes a high energy mass scale . the three generations of left - handed and right - handed quark fields ( up - type quarks are not distinguished from down - type quarks ) carry different charges under @xmath606 such that the usual yukawa terms have positive integer charges . the hierarchy in the yukawa matrix entries arises from powers of the flavon field compensating for these non - zero charges of the above trilinear yukawa terms . imposition of this mechanism reduces the number of free parameters if the new @xmath272 family symmetry is spontaneously broken down to a residual discrete @xmath607 symmetry . the article @xcite further take up the topic of flavons . the uniqueness of the multicoloured random graphs on a given set of colours in the infinite - vertex limit starting from arbitrary initial configurations would seem to suggest that the physical interpretation of any result derived therefrom would be independent of initial conditions . one physical interpretation that we have in mind is this : that the growth of a graph from an arbitrary initial configuration to the infinite random graph is mirrored by taking the limit of a large number of particles . there are many instances of limits of large numbers in physics , such as the hydrodynamical scaling law for an incompressible fluid being derived when the number of fluid molecules becomes very large @xcite . the intimate relationship between the modular and 3-braid groups , for example that the 3-braid group @xmath608 is the universal central extension of the modular group @xcite , offers a direction for further study . a key test for any theory that extends the sm model will be in its ability to better explain the sm coupling constants and mass parameters . it is possible that quantum field theory will only be able to do this through the introduction of a new ingredient such as a supplementary symmetry . the hope has been that application of the mpp gives information about these sm parameters . it is also possible that such additional restricting principles are not a priori clearly or directly related to the dynamical entities and only deeper work would reveal the connections . in @xcite we have proposed candidates and have tried to motivate them . it could be argued that the restrictions should , for reasons of simplicity , apply to the zero - particle states , that is the vacua . given that random dynamics is based upon an entirely different philosophy to the the more mainstream approaches of unification , we end with a philosophical point regarding choice of the best theory from alternatives , though within philosophy itself there is no consensus on an accepted theory of this subject . science operates through observation , analysis , reason , deduction and sometimes prediction . it functions according to a precision , which is why mathematics is often required for its proper explanation and description . scientific theories are never complete ; in physics this has led to the rise of so - called `` effective theories '' . as wells has pointed out @xcite to all intents and purposes all theories are effective theories , choosing thagard s criteria @xcite for _ best explanation _ : observational and mathematical consistency with known empirical facts are sine qua non and take precedence over other factors . conjecture is probably the most effective source of theories and knowledge grows by rearrangement combination , alteration and subsuming of previous principles . criticism and empiricism allows choice of theories from alternatives . good explanations have robustness @xcite . to this end the entire random dynamics program may be one of the most economical possibilities , if it could be developed to anywhere near the ambitious intentions of its founders , beginning with specific random structures . we have made a starting suggestion as to what such structures might be . d. l. bennett , a. kleppe and h. b. nielsen , _ random dynamics in starting levels _ , in `` what comes beyond the standard model ? '' proceedings of the 10th bled workshop in physics held from july 17 - 27 , 2007 in bled . edited by n. bortnik , h. b. nielsen , c. d. froggatt and d. lukman , 31 ( 2007 ) . c. d. froggatt , c. r. das , l. v. laperashvili , h. b. nielsen and a. tureanu , _ gravi - weak unification and multiple point principle _ , international conference of ras , protvino , russian federation , november 5 - 8 , ( 2013 ) . c. d. froggatt and h. b. nielsen , _ why do we have parity violation ? _ , arxiv : hep - ph/9906466v1 . updated version of discussion published in `` what comes beyond the standard model ? '' proceedings of the bled workshop in physics , 29 june - 9 july ( 1998 ) . c. d. froggatt and h. b. nielsen , _ derivation of lorentz invariance and three space dimensions in generic field theory _ , in `` what comes beyond the standard model 2000 , 2001 , 2002 ? '' proceedings of the 5th bled workshop in physics . edited by n. bortnik , h. b. nielsen , c. d. froggatt and d. lukman , 1 - 19 ( 2002 ) . c. d. froggatt and h. b. nielsen , _ trying to understand the standard model parameters_. invited talk by h.b . nielsen at the xxxi itep winter school of physics , moscow , russia , 1826 feb . , 2003 ; published in surveys high energy phys . 18 , 55 ( 2003 ) ; arxiv : hep - ph/0308144 . a. kleppe , _ a tight packing problem _ , in `` what comes beyond the standard model 2000 , 2001 , 2002 ? '' proceedings of the 5th bled workshop in physics . edited by n. bortnik , h. b. nielsen , c. d. froggatt and d. lukman , 52 - 70 ( 2002 ) . h. b. nielsen , _ initial condition model from imaginary part of action and the information loss problem _ , in proceedings of workshop `` black holes in general relativity and string theory '' , 24 - 30 august , 2008 , in veli losjinj , croatia , arxiv:0911.3859 ( 2009 ) . h. b. nielsen , _ small representation principle _ , in `` what comes beyond the standard models ? '' proceedings to the 16th bled workshop in physics , 14 - 21 of july 2013 , vol . 2 , dmfa zaloznistvo , ljubljana . edited by n. bortnik , h. b. nielsen , c. d. froggatt and d. lukman , 143 - 170 ( 2013 ) . h. b. nielsen and a. kleppe , _ towards a derivation of space _ , in `` what comes beyond the standard model ? '' proceedings of the 16th bled workshop in physics . edited by n. bortnik , h. b. nielsen and d. lukman , 14 - 21 ( 2013 ) . h. b. nielsen and m. ninomiya , _ future dependent initial conditions from imaginary part in lagrangian _ , in `` what comes beyond the standard model 2006 ? '' proceedings of the 9th bled workshop in physics . edited by n. bortnik , h. b. nielsen , c. d. froggatt and d. lukman , 87 - 124 ( 2006 ) . e. p. wigner , _ the role and value of symmetry principles and einstein s contribution to their recognition _ , in _ symmetries in science _ , edited by b. gruber and r. s. millman , ( plenum press , 13 - 21 , 1979 ) .	multicoloured random graphs : the random dynamics program the random dynamics program of holger nielsen and co - workers is a proposal to explain the origin of all symmetries , including lorentz and gauge invariance without appeal to any fundamental invariance of the laws of nature , and to derive the known physical laws in such a way as to be almost unavoidable . the philosophy of random dynamics is that the most useful assumption that we can make about the fundamental laws is that they are random and then to see how known physics like mechanics and relativity follow from them . it is believed that almost any model of these regularities would appear in some limit for example as energies become small . almost all theories or models at the fundamental level could then explain known physics and specific models would arise from chaotic fundamental physics which has as few starting assumptions as possible . this brief book continues the previous two in the series , the first being a derivation of pure mathematical results and the second containing a sketch of possible physical applications of some of the results in the first volume . in this third volume we focus on the ideas behind the froggatt - nielsen project and suggest how using the formalism and properties of random graphs can be useful in developing the theory , and point towards directions in which it can be more fully extended in future work . in the previous volume we sketched our proposal for the origin of the standard model family structure . here we look in greater detail at some of the other proposals of the random dynamics program and outline possible mathematics and in particular properties of multicoloured random graphs as possible starting points in a research project that applies the mathematics to the physics by judicious interpretation of the various elements of the random models as physical concepts . we would hope that the ideas presented here could be implemented into a detailed formal theory , as part of an exploration of the graph - theoretic and related group - theoretic content of random dynamics . in this prelude to an investigation of the mathematics that may be applied to the random dynamics program we retain the nomenclature used by nielsen and co - workers .
digital media , especially internet and smartphones , provides a new lens to study human behaviors in both the physical space and the virtual world . researchers are getting more interested in studying the relationship between online and offline human behaviors @xcite . for example , ginsberg et al . tried to predict the influenza epidemics using search engine query data @xcite ; bond et al . studied the relationship between facebook use and offline political mobilization @xcite ; zhao et al . investigated the relationship of human movements in cyberspace and physical space using a mobile phone dataset @xcite . in particular , zhao et al . found a superlinear scaling relation between the mean frequency of visit @xmath5 and its fluctuation @xmath6 , and there was a strong correlation for the average frequency of @xmath5 in both spaces @xcite . however , the high - resolution spatial behaviors in these research remained unexplored . in this study , we use the fine - grained smartphone data of @xmath0 users in a major city of china to study the interrelationship between online human behaviors and offline mobility in the perspective of network structures , especially the small world property and the fractal character . in previous research , most complex networks are found to be small - world @xcite . however , a series of recent studies discover that many complex networks are featured by fractal patterns @xcite . small - world and fractality are two fundamental structural properties of complex networks . yet , suggested by @xcite , there is a seemingly contradictory relationship between the small - world property and the fractality property in complex networks . the small - world property of complex networks suggests that the average diameter of the network @xmath7 increases logarithmically with the total number of nodes @xmath8 , that is @xmath9 , where @xmath10 is a characteristic length . however , the self - similarity of fractality requires a power relation between @xmath8 and @xmath7 , that is @xmath11 . we construct the mobility network and the attention network to compare their network structures . in the mobility network , nodes are physical locations and edges represent the movements between locations @xcite , and in the attention network , nodes are websites and edges represent the switch of users between websites @xcite . we use the network renormalisation method in general and the box - covering method in specific to investigate the network structure . the box - covering method is proposed by song et al . it is originated from the box - counting method which is widely used in calculating the fractal dimension @xcite . considering a network embedded in euclidean space , we cover the network with boxes of size @xmath1 , and the number of boxes needed is denoted as @xmath12 . thus the fractal dimension or box dimension @xmath13 can be given from @xmath14 . our findings reveal that the attention network is self - similar , whereas the mobility network shows a small world property . the intriguing implication is that there are structural differences between the physical mobility and online surfing for smartphone users . that is to say , human behaviors in physical space and cyberspace represent two universal classes . surprisingly , we find that the degree correlation of the mobility network shifts from positive to negative with the increasing of the length of box @xmath1 , which is found for the first time to our best knowledge , and we define and further explain such phenomenon as `` spatial - constrained hub '' . with these properties found above , we apply the community detection technique on the mobility network . to be specific , using renormalisation techniques , we map the spatial space into different communities , and measure the association between spatial communities and the websites visited by the moving citizens by computing the tf - idf values . thus we can identify the websites frequently visited by the users of a given community , and conversely , we can tell the communities frequently visited by the users of a given website . using this method , we find three examples of location - based behaviors , including the shopping , dating , and taxi - calling . the investigations in the local structure of the mobility network also help us further understand the spatial - constrained hub . finally , we explain the observed empirical patterns with a revised geometric network model . using the box - counting method , song et al . confirmed the self - similarity property of complex networks @xcite . we apply the same renormalization technique in mobility networks and attention networks to investigate the fractal pattern of the network structure . figure [ renormalization ] presents the renormalization process for the two networks . the attention network is much more dense than the mobility network . there are 9899 nodes and 39,083 edges in the mobility network ( @xmath15 ) , and there are 16,476 nodes and 144,909 edges in the attention network ( @xmath16 ) . the diameter of the mobility network is 15 , and the diameter of the attention network is only 10 . thus , it is natural that we need more boxes of the same length @xmath1 to cover the mobility network than the attention network . actually , it takes 15 steps to tile the mobility network into a single node , and it needs 10 steps only to tile the attention network into a single node . we find two universal classes of behaviors ( figure [ transition]b ) . for the attention network , the relationship between the number of boxes @xmath2 and the length of box @xmath1 shows a power law function , @xmath4 . thus , the structure of the attention network is self - similar or fractal . whereas for the mobility network , the number of boxes @xmath2 and the length of box @xmath1 have an exponential relationship . thus the structure of the mobility network demonstrates a small - world property . the small - world network is featured for its short diameter and a large clustering coefficient , which is efficient for transportation @xcite . this finding implies that there are two different dynamical mechanisms governing the evolution of the two networks . suggested by song et al.@xcite , the fractal property of networks originates from disassortativity , i.e. , the hubs tend not to connect directly with each other , on the contrary , the hubs tend to connect directly with non - hub nodes . thus , the degree correlation is negative for fractal networks . to gain insights into the mechanism that leads to the two universal classes of behaviors , we further inspect the change of degree correlation against the length of box @xmath1 . the degree correlation of a network is quantified by calculating the correlation @xmath17 between the degree @xmath18 for nodes and the average degree of their neighbors @xmath19 @xcite . as it has been shown in figure [ transition]b , @xmath1 = 3 is a turning point at which assortativity transfers to disassortativity ( @xmath20 ) . the curve of @xmath17 for both mobility network and attention network became largely flat when @xmath1 is larger than 4 . to better understand the transition of degree correlation under renormalization , the relationship between @xmath21 and @xmath19 are visualized for different @xmath1 . figure [ transition]c - d show the detailed transition process for the mobility network and the attention network , respectively . there is a common pattern in the renormalization process for both the mobility network and the attention network . with the increasing of @xmath1 , the degree correlation @xmath17 and @xmath19 decreases gradually . for example , in the attention network , when @xmath22 , the @xmath19 for @xmath23 is larger than 1000 , while when @xmath24 , the @xmath19 for @xmath23 is only 10 . figure [ transition]d demonstrates that the overall fractal structure of the attention network holds for different resolutions of @xmath1 . on the contrary , the structure of the mobility network is more complex . there are two layers of structure in the mobility network . to tile the network with a small @xmath1 will combine the locally connected nodes into super - nodes , when @xmath1 is smaller than 3 , even though the nodes within the range of 2 are collapsed into supernodes , the overall positive degree correlation ( small - world property ) are well maintained . when @xmath1 is larger than or equal to 3 , some of these super - nodes further collapsed into much larger ones , and we call them `` spatial - constrained hub '' . when the spatial - constrained hub appears , the topological structure of the mobility network significantly changes from assortativity to disassortativity . according to previous research @xcite , there are smaller nodes standing between hubs . since users are simultaneously travel in the attention network and mobility network , it is natural to infer that our online surfing behaviors within the attention network are also location - based . we will first find the communities of the mobility network , and then explore the underlying connection between the spatial communities and online surfing behaviors , in order to find out what websites users are visiting when they are moving within a community of the mobility network . using the boxing - covering method , we can detect the communities of the mobility network . according to figure [ transition]b , when @xmath25 , the degree correlation curve becomes relatively stable ( around 0.7 ) . thus , we tile the mobility network into communities using the box of length 4 ( see figure [ community ] ) . for each community , we calculate its tf - idf value for all the websites . thus , we can identify the most relevant websites for a given community , and qualitatively check the validity of the results . the assumption underlying these patterns is that the mobility behaviors in the physical space and the attention switching behaviors in the virtual space are inter - connected . as it has been demonstrated in figure [ three_examples ] , we find three examples of typical location - based behaviors : a. the red nodes and edges profile a community featured by the online shopping behaviors . these users move around chaoyang road which connects wangfujing business street and the eastern 5th ring road . this area is around the largest business circles in beijing and attracts a lot of consumers . with the aid of mobile internet , users can surf online to collect online information before they purchase offline . b. the blue nodes and edges mark the community which is featured by the online dating behaviors . it well depicts the footprints of online daters in haidian district , which is the home of many well - known universities ( e.g. , peking university , tsinghua university , and renmin university of china ) . c. the orange nodes and edges capture the locations of the users who use mobile texi - calling apps , especially the didi dache app ( www.xiaojukeji.com/ ) which is the most popular taxi apps in china . our findings reveal that the heavy use of taxi - calling service is primarily along the highway to changping district in the north - west of beijing where is not well covered by the city subway . one important feature of the mobility network lies in the fact that both the node location and edge - generating mechanisms are spatial - constrained . this pattern has been well demonstrated in figure [ three_examples ] . we can qualitatively observe that the probability of tie - generating is constrained by the spatial distance : although long distance connections are possible , there tends to be more edges within a short physical distance , and thus we can find apparent local clustering patterns . this feature has been theorized as the spatial - constrained attachment ( sca)@xcite . based on the sca mechanism , zhang et . al proposed a growing geometric graph model to explain the origin of scaling behaviors in cities @xcite . in the original geometric network model , there is a geometric graph growing in d - dimensional euclidean space . initially , there is only one node as the seed of the growing graph located in the center of the space . at each time @xmath26 a new node @xmath27 is generated , and @xmath27 will be added to the graph if only it is within the radius @xmath28 of at least an existing node @xmath29 , or else @xmath27 will not be added to the graph . if @xmath27 are within the radius of more than one node @xmath30 , an edge will be established between @xmath27 and each node of @xmath30 . since the new node @xmath27 will be connected with all the nodes within a given radius , this original model is named as model all . based on model all , more practical extensions can be implemented . for example , to add a parameter to tune the local clustering density , or to add multiple seeds in the initial stage @xcite . we further extend the geometric network model by changing the edge - generating mechanism based on model all . in model all , a new node @xmath27 will be connected with all the nodes within the length of radius @xmath28 , this rule of edge - building could be not practical in real life . one competing rule comes from the preferential attachment of the ba model @xcite in which new added node links preferentially to hubs . in our revised geometric models , we choose to connect the new node @xmath27 with only one node within the radius @xmath28 . there are primarily two scenarios : i. model max , to link the new node @xmath27 with the node @xmath31 which has the largest degree within the radius @xmath28 ; ii . model min , to link the new node @xmath27 with the node @xmath32 which has the smallest degree within the radius @xmath28 . we compare the structural patterns of these three geometric network models discussed above . the panel a , b , and c of figure [ geometric_model ] show the overall structure of three geometric networks generated with model all , model max , and model min respectively . to look into the local structure can help us gain more insights . the panel d , e , and f of figure [ geometric_model ] demonstrate the local connectivity generated by model all , model max , and model min respectively . it is easy to find that : the geometric network generated by model all is very dense relative to model max and model min ( panel d ) ; there are significantly more hubs in the network generated with model max , and note that there are also obvious disassortativity in the local structure ( there are nodes standing between hubs , which is a feature of fractal networks , see panel e ) ; there are few hubs in the local structure of the network generated with model min , and the links are somewhat like the traces of a random walk ( panel f ) . we also renormalize the geometric networks generated with three models using the box - covering method . figure [ geometric_model]g confirms two universal classes of behaviors by inspecting the relationship between the size of box @xmath1 and the number of boxes @xmath2 . the networks generated with model all and model min are fractal , since the relationship between @xmath2 and @xmath1 can be fitted with a straight line . while the network generated with model min shows an apparent small - world pattern that we have also found in the mobility network . we further explore the degree correlations of the generated networks . as figure [ geometric_model]h has shown , there is a positive relationship between @xmath19 and @xmath18 for model all and model min , just as what we have discovered for the mobility network . while there exists a negative relationship between @xmath19 and @xmath18 for model max . in addition , the diameters of the networks generated with three models grows identically with the number of nodes ( figure [ geometric_model]i ) . the findings confirms that model max can replicate the findings we have found in the attention network , while the model min can help explain the results of the mobility network . to compare how human move in the virtual world and physical world has been a major interest of scientific research . our study provides one approach to fulfill this task by constructing and comparing the mobility network and the attention network . using networks as a mathematical structure of the complex human behaviors , we can implement the tools of network science to lend us a new lens and gain deeper insights . by renormalizing the two networks , we confirm that the attention network is featured by its fractal patterns , and the mobility network is characterized by its small - world property . the study of zhao et al . concluded that there is a strong correlation between between human movements in cyberspace and physical world in the perspective of flux - fluctuation law@xcite . however , our study showed that the the human movements in two spaces represent two universal classes of behaviors . analyzing the change of degree correlation with the size of box @xmath1 reveals that there is a surprising transition of degree correlation from positive to negative for the mobility network , while the degree correlation of the attention network stays negative with the increasing of @xmath1 . this finding underlines the structural difference between online and offline human behaviors . previous studies on fractal networks construct a good theoretical framework on explaining the scaling property of self - similarity . for example , gallos et al . proposed a scaling theory of degree correlation for scale - free networks@xcite . the degree correlation can be captured by the probability of @xmath33 that two nodes of degree @xmath34 and @xmath35 are connected together by a link . gallos et al . showed that @xmath33 can be expressed as @xmath36 . to better estimate @xmath37 , they proposed a quantity of @xmath38 which measures to which extent the nodes of degree @xmath18 tend to link with more connected nodes , @xmath39 . for the scale - free network , @xmath40 . by estimating the correlation exponent @xmath37 and the power - law exponent @xmath41 , they constructed a phase diagram which classifies scale - free networks into three areas : below the line of @xmath42 are the random networks ( area i ) ; when @xmath43 is the area ii where the network has a weaker hub - hub correlation than the random networks but is not fractal ; above the line of @xmath44 is the area iii where the networks are fractal . with the theory , we calculate the power exponents for the mobility network(@xmath45 ) and the attention network ( @xmath46 ) , and the slopes of @xmath47 for the mobility network(@xmath48 ) and the attention network ( @xmath49 ) . thus , the @xmath37 of the mobility network and the attention network is 1.76 and 2.80 . therefore , the mobility network is nonfractal ( although its hub - hub correlation is weaker than random networks ) , and the attention network is fractal . however , it is also necessary to note that since the attention network and the mobility network are not perfectly scale - free network networks , such analysis may be problematic . the transition of degree correlation uncovers the fact that the mobility network is spatial - constrained . the transition happens during the renormalization process within which the nodes covered in a box are tiled into one supernode . there are two layers of structures for the mobility network . in the initial stage of network renormalizaiton , the local clustering are not fully destroyed , thus we can still observe positive degree correlation for the mobility network ( when @xmath50 ) . when @xmath51 , the local structure of the mobility network significantly changes from assortativity to disassortativity . using the box - covering algorithm , the city can be classified into different communities . however , neither the original fractal model nor the classical small - world model could demonstrate the transition of degree correlation . we then inspect the geographical distribution of three location - based behaviors . the spatial - constrained feature of the edge again indicates that the mobility network is different from the attention network for its spatial - constrained mechanisms , and we need to adopt the geometric network model to explain our findings . based on the original geometric network model ( model all ) , we propose two revised models ( model max and model min ) by changing the edge - generating mechanisms . the revised geometric models replicate what we have discovered in the data : both two universal classes of behaviors and the transition of degree correlation . our findings about how human move in both cyberspace and physical space can help develop commercial applications to find the best spot to place online and offline geomarketing advertisement . the increasing power of search engines has made advertisement more and more precise . computational advertising systems like google adsense collect and filter the attention of the most relevant users and sale it to companies . for mobile internet use , the telecommunications operators have the most detailed information about the users geographical locations and the websites they are visiting . for a city , it can be separated into functional areas using the network renormalization method . relating the moving citizens behaviors in cyberspace with their mobility in the physical space can help develop more accurate recommender . in summary , this research confirms that the human behavior in the virtual world and that in the physical world are structurally different . the human moving behaviors in the physical space and the virtual world represent two universal classes of human behaviors . there are two layers of structure in the mobility network , and only when the local structure is tiled using the box - covering method we can get a fractal structure . connecting new nodes with non - hub nodes can help make up the two - layers structure of the mobility network . only when the locally layer of clustered nodes are titled into spatial - constrained hubs under renormalization , the negative degree correlation can be observed . the spatial - constrained attachment mechanism determines its unique network structure . using the revised geometric models , we are able to replicate what we have found , and the growing of attention network and mobility network is not fully preferential attachment or anti - preferential attachment , but also spatial - constrained . mobile phone provides a new lens to investigate the human behaviors . eagle et al . analyzed the relationship between mobile phone use and the social - economic development of british cities@xcite . song et al . used the mobile phone data to study the predicability of human mobility@xcite . in this study , we primarily work on a fine - grained anonymous smartphone dataset collected from a randomly sampled @xmath0 users in a major city of china . the dataset profiles what websites or mobile apps users subsequently use in the virtual world as well as the physical locations . the dataset is anonymized , and we do not have the personal information of these users . based on these information , we construct two networks to trace how users move from one website to another ( attention network ) and from one location to another ( mobility network ) . in the mobility network , the nodes are mobile base stations ( @xmath52 ) , and the edges ( @xmath53 ) are users travel from one base station to another . in the attention network , the nodes ( @xmath54 ) are websites , and edges ( @xmath55 ) represent the switch of users between websites . figure [ spatial_mobility]a shows the spatial distribution of base stations . the base stations geographically distributed according to the shape of the city , and there are obvious clustering patterns . figure [ spatial_mobility]b shows the mobility network with heatmaps . the trajectory of human mobility vividly shows the transportation backbone of the city . to further describe the data , we calculate the following six variables for each individual based on their daily behavioral records : ( 1 ) the total number of records ( websites / base stations ) ; ( 2 ) the number of unique base stations ; ( 3 ) the number of sequentially visited base stations ( any two successive base stations are different ) ; ( 4 ) the total distance of navigation ; ( 5 ) the number of unique websites ; ( 6 ) the number of sequentially visited websites . it turns out that the probability distributions of all these variables are featured by the long - tail . we fit their power law exponents using the powerlaw module @xcite , and the results are shown in the figure [ power_law ] . apparently , the human navigation in both the real world and the virtual space are very heterogeneous . it is important to check how online behaviors correlate with the offline behaviors . we find that several pairs of variables are positively correlated ( see figure [ correlation_navigation ] ) . in addition to that , we investigated the effect of several variables on the amount of information download within one day . the total distance of real world movement positively with the total number of websites visited within one day ( figure [ correlation_navigation]c ) , and the amount of information download(figure [ correlation_navigation]i ) . the amount of information download positively correlates with the unique base stations ( figure [ correlation_navigation]g ) , sequential base stations ( figure [ correlation_navigation]h ) . the findings above reveals that those who are active online are largely proportionally active offline . yet , it does not mean that online and offline human behaviors are similar . we implement the box - counting method to renormalize the mobility network and the attention network @xcite . the network is covered with boxes such that the nodes within a box are at a distance smaller than the length of box @xmath1 . with the increasing of @xmath1 , we need a smaller number of boxes @xmath2 to cover the whole network . after we tile the entire network , each box was replaced by a single supernode , and connections between the supernodes are merged @xcite . the renormalization procedure is repeated until there is only one node left in the network . we use the term frequency inverse document frequency ( i.e. , tf - idf ) measurement to find what the websites simultaneously visited by the moving citizens . tf - idf is a statistic widely adopted to measure the importance of a term to a document in a corpus . it is the product of two statistics , term frequency and inverse document frequency . @xmath56 measures the frequency of a term appearing in a given document . if a term appears many times in a document , its @xmath56 value of this document will be large . @xmath57 measures the concentration of terms across documents . if that term appears in many documents , its @xmath57 value would be small . combining the information supplied by both @xmath56 and @xmath57 , we can select the most important words for a given document . in our case , we can detect which websites are important for a group of users moving around a group of mobile base stations . the authors thank zhang chang for the assistance in providing data and inspiring discussion . we acknowledge financial support for this work from the national social science foundation , grant number 15cxw017 . l.w and c.w designed research ; c.w and l.w performed research ; l.w and c.w collected and analyzed data ; c.w and l.w wrote the manuscript ; all authors discussed the results and reviewed the manuscript . the authors declare no competing financial interests . robert m bond , christopher j fariss , jason j jones , adam di kramer , cameron marlow , jaime e settle , and james h fowler . a 61-million - person experiment in social influence and political mobilization . , 489(7415):295298 , 2012 .	with the widespread use of mobile computing devices in contemporary society , our trajectories in the physical space and virtual world are increasingly closely connected . using the anonymous smartphone data of @xmath0 users in 30 days , we construct the mobility network and the attention network to study the correlations between online and offline human behaviors . in the mobility network , nodes are physical locations and edges represent the movements between locations , and in the attention network , nodes are websites and edges represent the switch of users between websites . we apply the box - covering method to renormalize the networks . the investigated network properties include the size of box @xmath1 and the number of boxes @xmath2 . we find two universal classes of behaviors : the mobility network is featured by a small - world property , @xmath3 , whereas the attention network is characterized by a self - similar property , @xmath4 . in particular , with the increasing of the length of box @xmath1 , the degree correlation of the network changes from positive to negative which indicates that there are two layers of structure in the mobility network . we use the results of network renormalisation to detect the community and map the structure of the mobility network . further , we locate the most relevant websites visited in these communities , and identify three typical location - based behaviors , including the shopping , dating , and taxi - calling . finally , we offer a revised geometric network model to explain our findings in the perspective of spatial - constrained attachment .
space debris consists of a range of human - made objects in a variety of orbits around the earth , representing the remnants of payloads and payload delivery systems accumulated over a number of decades . with the increasing accumulation of debris in earth orbit , the chance of collisions between the debris increases ( causing an increase in the number of debris fragments ) . more importantly , the chance of collisions between debris and active satellites increases , posing a risk of damage to these expensive and strategically important assets . this risk motivates the need to obtain better information on the population characteristics of debris ( distribution of sizes and masses , distribution of orbits etc ) . research into space debris is considered a critical activity world - wide and is the subject of the inter - agency space debris coordination committee , previously described in detail in many volumes , including a united nations technical report on space debris @xcite and a report from the committee on the peaceful uses of outer space @xcite . the american astronomical society maintains a committee on light pollution , radio interference and space debris , since space debris poses a risk to important space - based astrophysical observatories . the risks posed by space debris , and the difficulties inherent in tracking observations and orbit predictions , were illustrated starkly on the 10@xmath3 of february 2009 , when the defunct russian kosmos-2251 satellite collided with the active iridium-33 satellite at a relative speed of over 11 km / s , destroying the iridium satellite . even with the global efforts to track space debris and predict their orbits , this collision between two satellites of 900 kg and 560 kg , respectively , and of @xmath110 m@xmath4 debris size , was unanticipated . further exploration of ssa capabilities is required to continue to minimise collision risks and provide better early warnings following collision breakups that multiply debris numbers and randomise orbits . methods for obtaining data on space debris include ground - based ( radar observations and optical observations ) and space - based measurements . previously , large ground - based radio telescopes ( whose primary operations are for radio astronomy ) have been used in a limited fashion to track space debris . for example , the 100 m effelsberg telescope in germany , outfitted with a seven - beam 1.4 ghz receiver , was used for space debris tracking using reflected radiation generated by a high power transmitter @xcite . recently , a new method of ground - based space debris observation has been trialed , using interferometric radio telescope arrays ( also primarily operated for radio astronomy ) , in particular using the allen telescope array @xcite . this technique utilises stray radio frequency emissions originating from the earth that reflect off space debris and are received and imaged at the interferometric array with high angular resolution . scenarios such as this , with a passive receiver and a non - cooperative transmitter , are described as `` passive radar '' , a sub - class of the bi - static radar technique ( transmitter and receiver at different locations ) . this paper explores the possibilities offered by this technique , using a new low frequency radio telescope that has been built in western australia , the murchison widefield array ( mwa ) . the mwa is fully described in @xcite . briefly , the mwa operates over a frequency range of 80 - 300 mhz with an instantaneous bandwidth of 30.72 mhz , has a very wide field of view ( @xmath12400 deg@xmath4 at the lower end of the band ) , and reasonable angular resolution ( @xmath16 arcmin at the lower end of the band ) . recent observations with the mwa by @xcite have shown that terrestrial fm transmissions ( between 87.5 and 108.0 mhz ) reflected by the moon produce a significant signal strength at the mwa . @xcite estimate that the equivalent isotropic power ( eip ) of the earth in the fm band is approximately 77 mw . the ensemble of space debris is illuminated by this aggregate fm signal and will reflect some portion of the signal back to earth , where the mwa can receive the reflected signals and form images of the space debris , tracking their positions on the sky as they traverse their orbits . it should be noted that this technique makes use of the global distribution of fm radio transmitters , but that the technique is likely to work most effectively with receiving telescopes that are well separated from the transmitters . if the receiving array is in close proximity to an fm transmitter , the emissions directly received from the transmitter will be many orders of magnitude stronger than the signals reflected from the space debris . therefore , the technique can likely only be contemplated using telescopes like the mwa , purposefully sited at locations such as the radio - quiet murchison radio - astronomy observatory ( mro ) , located in murchison shire , 700 km north of perth in western australia @xcite . the mwa is a science and engineering precursor for the much larger and more sensitive low frequency component of the square kilometre array ( ska ) @xcite , which will also be located at the mro and is currently under development by the international ska organisation . in this paper we estimate the conditions under which the mwa can detect space debris , in passive radar mode in the fm band . in section 2 we consider a simple calculation of the expected signal strength at the mwa for one idealised scenario . we also present comprehensive but idealised electromagnetic simulations that agree well with the simple calculation and show that space debris detection is feasible with the mwa . for normal mwa observation modes , the detection of space debris of order @xmath10.5 m radius and larger appears feasible . if efforts are made to modify standard mwa observation modes and data processing techniques , then substantially smaller debris could be detected , down to @xmath10.2 m in radius in low earth orbits . in section 3 we present observational tests that show the basic technique to be feasible , but also illustrate challenges that must be overcome for future observations . in section 4 we discuss the utility of the technique , complementarity with other techniques , and suggest future directions for this work . as an example that illustrates the feasibility of this technique , we consider the mwa as the receiving element in a passive radar system in the fm band and consider a single transmitter based in perth , western australia ( specifically the transmitter for call sign 6jjj : australian communications and media authority [ acma ] licence number 1198502 ) . the transmitter is located at ( lat , long)=(@xmath5 , @xmath6 ) and transmits a mixed polarisation signal , with an omnidirectional radiation pattern in the horizontal plane , at 99.3 mhz over the fm emission standard 50 khz bandwidth and with an effective radiated power ( erp ) of 100 kw . the mwa , acting as receiver , is located at ( lat , long)=(@xmath7 , @xmath8 ) , @xmath1670 km from the transmitter . we consider an idealised piece of space debris ( a perfectly conducting sphere of radius @xmath9 , in metres ) at a distance of @xmath10 km from the mwa site . we denote the distance between the transmitter and the space debris as @xmath11 km . for simplicity , we model the radar cross section ( rcs , denoted by @xmath12 ) of the debris as the ideal backscatter from a conducting sphere in one of two domains that are approximations of mie scattering @xcite : 1 ) when the wavelength @xmath13 , the rcs is described by rayleigh scattering and @xmath14 ; and 2 ) when @xmath15 , @xmath16 , corresponding to the geometrical scattering limit . the greatest interest is in case 1 , as the space debris size distribution is dominated by small objects ( although the example given in the introduction shows that even the largest items of space debris pose serious unknown risks ) . for case 1 we obtain equation 1 , below , following the well - known formula for bi - static radar @xcite , @xmath17 where s is the spectral flux density of the received signal at the mwa in astronomical units of jansky ( 1 jy = 10@xmath18 wm@xmath19hz@xmath20 ) , @xmath21 is the erp in kw , @xmath22 is the transmitter gain in the direction of the space debris , @xmath23 is the bandwidth in khz over which the signal is transmitted , @xmath11 , @xmath10 and @xmath9 are as described above , and @xmath24 is the center frequency of the transmitted signal in mhz . for @xmath21=100 kw , @xmath10=1000 km , @xmath251200 km , @xmath22=0.5 , @xmath23=50 khz , @xmath9=0.5 and @xmath24=99.3 mhz , the spectral flux density , @xmath26 , is approximately 4 jy . the assumption of @xmath22=0.5 is based on an idealised dipole antenna transmitter radiation pattern which is omni - directional in the horizontal plane . fm transmitter antenna geometries vary significantly , with local requirements on transmission coverage dictating the directionality and thus the antenna geometry . phased arrays can be used to compress the radiation pattern in elevation and different antenna types can produce gain in preferred azimuthal directions . therefore , in reality , significant departures from a simple dipole is normal . propagation losses in signal strength due to two passages through the atmosphere are assumed to be negligible . due to the approximations described above , this estimate should be considered as order - of - magnitude only . however , the calculation indicates that the detection of objects of order 1 m in size is feasible . in a 1 second integration with the mwa over a 50 khz bandwidth , an image can be produced with a pixel rms of approximately 1 jy @xcite , making the reflected signal from the space debris a several sigma detection in one second . according to equation 1 , the mwa will be sensitive to sub - metre scale space debris . however , for fm wavelengths , sub - metre scale debris are in the rayleigh scattering regime , meaning that the rcs drops sharply for smaller debris , dramatically reducing the reflected and received power . in order to more systematically explore the detectibility of debris of different sizes and relative distances between transmitter , debris , and receiver , we have performed a series of electromagnetic simulations . the simulations performed take the basic form described above , in which a single fm transmitter ( idealised dipole radiation pattern for the transmitter , omni - directional in the horizontal plane ) at a given location was considered ( separate simulations were performed for 11 different transmitters located in the south - west of western australia ) , along with the known location of the mwa as receiving station . the debris were modelled as being directly above the mwa . a range of different debris altitudes was modeled : 200 km ; 400 km ; 800 km ; and 1600 km ( as for the calculation above , propagation losses due to the atmosphere are considered negligible ) . the debris was assumed to be a perfectly conducting sphere , with a range of radii : 0.2 m ; 0.5 m ; 1 m ; and 10 m. the simulations proceeded using the _ xfdtd _ code from remcom inc . _ xfdtd _ is a general - purpose electromagnetics analysis code based on the finite - difference , time - domain technique and can model objects of arbitrary size , shape and composition . the @xmath27 ( transmitters ) @xmath28 @xmath29 ( radii ) @xmath28 @xmath29 ( altitudes ) @xmath30 runs of _ xfdtd _ were calculated on a standard desktop computer over the course of several hours . the simulations compute output that includes the rcs according to the full mie scattering solution and the spectral flux density at the location of the mwa , for each combination of debris size , altitude and transmitter . figure 1 shows a selection of results from the simulations , for the perth - based 6jjj transmitter and the ranges of debris size and altitudes listed above . also shown is a single point representing the result of the simple calculation from equation 1 . the simple calculation overestimates , by a factor of approximately two , relative to the _ xfdtd _ simulation , the flux density at the mwa . this level of agreement is reasonable , given the simplifying assumptions made for equation 1 . figure 1 shows that debris with @xmath31 0.5 m can plausibly be detected up to altitudes of approximately 800 km with a one second observation . the solid line in figure 1 denotes the one sigma sensitivity calculation made for the simple example , above , based on information from @xcite , showing that a three sigma detection can be made for debris of @xmath320.5 m at 800 km . of course , if longer observations can be utilised , the detection thresholds can be reduced . for example , a ten second observation would yield a three sigma detection of debris with @xmath32 0.5 m at 1000 km . the solution to the detection of smaller ( @xmath110 cm ) scale debris with low frequency passive radar in the past has been to use erps of gigawatts , as with the 217 mhz navspacecom `` space fence '' @xcite . an alternative approach is to build larger and more sensitive receiving antennas . the natural evolution of the mwa is the much larger and more sensitive low frequency ska , which will have a factor of @xmath11000 more receiving area and be far more sensitive than the mwa . the low frequency ska may have great utility for space debris tracking later this decade . we verified the technique outlined above with an observation using the mwa during its commissioning phase . this is a subarray of 32/128 tiles with a maximum east - west baseline of roughly 1 km and a maximum north - south baseline of roughly 2 km , giving a synthesized beam of roughly @xmath33 at a frequency of @xmath1100 mhz . we observed an overflight of the international space station ( iss ) on 2012 november 26 , as shown in figure [ fig : track ] . the center frequency of our 30.72mhz bandpass was 103.4mhz , and we correlated the data with a resolution of 1s in time and 10khz in frequency . we used two different array pointings , as plotted in the right panel of figure [ fig : track ] . the first was from 12:20:00ut to 12:24:56ut and was pointing at an azimuth of @xmath34 ( due south ) and an elevation of @xmath35 . the second was from 12:24:56ut to 12:29:52ut at an azimuth of @xmath36 ( due east ) and an elevation of @xmath35 . as discussed in @xcite and verified by @xcite , the mwa has an extremely broad primary beam ( roughly @xmath37 fwhm ) which was held fixed during each observation . we observed the overflight of the iss between 12:23:00ut and 12:26:00ut , when it was at a primary beam gain of @xmath38% , traversing about @xmath39 on the sky . during the pass , the iss ranged in distance between approximately 850 km ( at the low elevation limit of the mwa , approximately 30@xmath40 ) and approximately 400 km ( near zenith at the mwa ) ( figure 2 ) ; its angular speed across the sky is approximately 0.5@xmath40 per second , corresponding to approximately five synthesised beams with this antenna configuration . to image the data , we started with an observation pointed at the bright radio galaxy 3c 444 taken before the iss observations . this observation was used as a calibration observation , determining complex gains for each tile assuming a point - source model of 3c 444 , which was sufficient to model the source given that its extent ( @xmath41 ) is the same as the fwhm of the synthesised beam of the mwa subarray at this frequency . while we changed pointings between this observation and those of the iss , we have previously found that the instrumental phases are very stable across entire days , and that only the overall amplitudes need to be adjusted for individual observations . antenna - based gain solutions were obtained using the casa task _ bandpass _ on a 2-minute observation . solutions were obtained separately for each 10khz channel , integrating over the full two minutes . the flux density used for 3c 444 was 116.7jy at the central frequency of 103mhz , with a spectral index of @xmath42 . a s / n@xmath43 was required for a successful per - channel , per - baseline solution , and the overall computed gain solutions for each antenna were of s / n@xmath44 . calibration solutions were applied to the iss observations using the casa task _ applycal _ and the corrected data were written to uv - fits format for imaging in ` miriad ` @xcite . for the two iss observations , we flagged bad data in individual fine channels ( 10khz wide ) associated with the centers and edges of our 1.28mhz coarse channels . because of the extremely fast motion of the iss , we created images in 1s intervals ( the integration time used by the correlator ) using ` miriad ` . we started by imaging the whole primary beam ( @xmath45 ) , separating the data into @xmath46mhz bandpasses . the bottom two bandpasses covered the fm band , while the top bandpass was above the fm band . we used a @xmath47 cell size and cleaned emission from fornax a. note that ` miriad ` does not properly implement wide - field imaging , but our images used a slant - orthographic projection such that the synthesized beam is constant over the image @xcite . the results covering the two time intervals and two bandpasses are shown in figure [ fig : images ] . the iss is readily visible as a streak moving through our images , but is only visible in the images that cover the fm band as would be expected from reflected terrestrial fm emission ( as in @xcite ) . we predicted the position of the iss based on its two - line ephemeris ( tle ) using the ` pyephem ` package . the predicted position agrees with the observed position to within the observational uncertainties . to determine more quantitatively the nature of the emission seen in figure [ fig : images ] , we imaged each 1s interval and each 10khz fine channel separately ( as before , this was done in ` miriad ` ) . we did not deconvolve the images at all ; most of them had no signal or the emission from the iss was very weak . the emission from the iss is not a point source ( smeared due to its motion ) and adding deconvolution for a non - stationary object to our already computationally - demanding task was not feasible . in each of our @xmath48 dirty images , we located the position of the iss based on its ephemeris . we then measured its flux density by adding up the image data over a rectangular region of @xmath49 ( comparable to our instrumental resolution ) in width and with a length appropriate for the instantaneous speed of the iss ( up to @xmath50 in a 1s interval ) , as seen in figure [ fig : zoom ] . the resulting dynamic spectra showing flux density as functions of time and frequency are shown in figure [ fig : dynspec ] . these spectra have been corrected for the varying primary beam gain of the mwa over its field - of - view based on measurements of the sources pks 2356@xmath5161 ( first observation , assuming a flux density of @xmath52jy at 97.7mhz ) and pks j0237@xmath511932 ( second observation , assuming a flux density of 35jy at 97.7mhz ) ; we have ignored the frequency - dependence of the primary beam gain for the purposes of this calculation . note that the emission seen in figure [ fig : dynspec ] is manifestly _ not _ radio - frequency interference ( rfi ) that is observed as a common - mode signal by all of our tiles . it is only visible at a discrete point in our synthesized images corresponding to the position of the iss ( figure [ fig : zoom ] ) . we ran the same routines used to create the dynamic spectra in figure [ fig : dynspec ] but using a position offset from the iss position , and we see almost no emission , with what we do see consistent with sidelobes produced by bright emission in individual channels , due to the iss . the iss has an approximate maximum projected area of @xmath11400 m@xmath4 , is of mixed composition ( thus not well approximated as perfectly conducting ) , has a highly complex geometry , and was at an unknown orientation relative to the transmitter(s ) and receiver during the observations . the detailed transmitter antenna geometries and gain patterns are not known . thus , it is very difficult to accurately predict the flux density we would expect from either simple calculations or electromagnetic simulations . + we can , however , unambiguously identify the origin of some of the transmissions reflected off the iss . for example , we consider the dynamic spectrum from our observations in figure 5 at the time 12:24:20 ut . at this time there are clearly a number of strong signals detected , plus a large number of much weaker signals . the five strongest signals at this time , in terms of the integrated fluxes ( jy.mhz , measured by summing over individual transmitter bands in the dynamic spectra to a threshold defined by the point at which the derivative of the amplitude changed sign ) , are listed in table 1 , along with transmitters that can be identified as the origin of the fm broadcasts @xcite . each of the five identified transmitters are local to western australia and , as can be seen from figure 2 , are all relatively close to the iss at this time . each of the five transmitters have omni - directional antenna radiation patterns in the horizontal plane . if we take the reported erp for a typical station from table [ tab : transmit ] and use it with equation [ eqn : flux ] to estimate the received flux , we find values of roughly @xmath53 , assuming the full @xmath54 of reflecting area for the iss . given its varying composition and orientation , the measured fluxes in table [ tab : transmit ] are reasonably consistent with this estimate . .the five strongest detected signals at 12:24:20 ut and transmitters identified with those signals[tab : transmit ] [ cols="^,^,^,>,>,^,^,^ " , ] many , but not all , of the much weaker signals detected can also be identified . however , it is clear that some of these weaker signals are from powerful transmitters at much larger distances , making unique identification of the transmitter more difficult ( fm broadcasts are made on the same frequencies in different regions ) . furthermore , it is clear that transmitters local to western australia , and of comparable power to those listed in table 1 , have not been detected in our observations . this is likely to be the result of the complex transmitter / reflector / receiver geometries mentioned above and the rapid evolution of the geometry with time , illustrated by the rapid evolution of the received signals identified in table 1 . the five strongest signals at 12:24:20 ut are strongest over a 5 second period , have a highly modulated amplitude over a further 30 second period , but are generally very weak or not detected over the majority of our observation . thus , it is highly likely that other transmitters of comparable strength would be detected at other times , when favourable geometries prevail . at 12:24:20 ut , it appears that the geometries were favourable for the central agricultural , northam and geraldton transmitters , but not for transmitters of comparable power in perth , bunbury and southern agricultural . we note that , while the strongest signals detected originated from transmitters with omni - directional transmitting antennas ( see above ) , the transmittors of comparable power not detected ( perth , southern agricultural and bunbury ) also have omni - directional antennas . thus , the most likely factor driving detection or non - detection is the transmitter / reflector / receiver geometry , rather than transmitting antenna directionality . the modulation in power of the reflected radiation is likely at least partly due to `` glints '' from the large , complex structure , which creates a reflected radiation pattern whose finest angular angular scale is @xmath55 , where @xmath56 is the wavelength of the radiation and @xmath57 is the extent of the object . for @xmath58 m ( full extent of the iss ) and @xmath59 m , the glints are @xmath60 in angular extent . at an altitude of @xmath1500 km , the smallest glint footprint on the surface of the earth is @xmath61 km . with an iss orbital speed of 8 kms@xmath20 , the glint duration at the mwa is of order one second . this corresponds well to some of the modulation timescales seen in figure 5 , although modulation exists on longer timescales ( 10 seconds and longer ) . these longer timescales may be due to glints involving structures smaller than the full extent of the iss ( single or multiple solar panels , for example ) . a full analysis of the modulation structure of the signals is extremely complex and can not be performed at a sufficiently sophisticated level to be useful at present . in the future , detailed electromagnetic simulations of complex reflector geometries may be used to gain insight into the modulations , but such an analysis is beyond the scope of the current work . in long integrations , such as required for epoch of reionisation experiments @xcite , the effect of having 10 pieces of space debris in the mwa primary beam at any given time with flux densities @xmath621 jy is the equivalent of a @xmath631 mjy additional confusion noise foreground component in the fm band . this contribution will have no discernable impact on the mwa s science goals for observations in the fm band . at metre size scales , of order 1000 pieces of debris are currently known and tracked . on average , up to approximately 10 such pieces will be present within the mwa field - of - view at any given time , allowing continuous , simultaneous detection searches and tracking opportunities . these estimated detection rates naturally lead to two primary capabilities . the first is a blind survey for currently unknown space debris , provided by the very large instantaneous field - of - view of the mwa . the second capability is the high cadence detection and tracking of known space debris . the wide field - of - view of the mwa means that large numbers of debris can be simultaneously detected and tracked , for a substantial fraction of the time that they appear above the local horizon . imaging with the mwa can provide measurements of the right ascension and declination ( or azimuth and elevation ) of the space debris , as seen from the position of the mwa ( convertable to topocentric right ascension and declination ) . the mwa will also be able to measure the time derivatives of right ascension and declination , giving a four vector that defines the `` optical attributable '' . measurements of the four vector at two different epochs allow an estimate of the six parameters required to describe an orbit . @xcite describe methods for orbit reconstruction using the optical attributable , taking into account the correlation problem ( being able to determine that two measurements of the four vector at substantially different times can be attributed to the same piece of space debris ) . the calculations and practical demonstrations presented above show that the basic technique using the mwa is feasible and worthy of further consideration as a possible addition to australia s ssa capabilities , particularly considering that the mwa is the precursor to the much larger and more sensitive low frequency ska . however , the calculations and observational tests presented here also point to the challenges that will need to be met in order to develop this concept into an operational capability . overall , these challenges relate to the non - standard nature of the imaging problem for space debris detection , compared to the traditional approach for astronomy . in radio interferomeric imaging , a fundamental assumption is made that the structure of the object on the sky does not change over the course of the observation . this assumption is broken in spectacular fashion when imaging space debris , due to their rapid angular motion across the sky relative to background celestial sources and due to the variation in rcs caused by the rapidly changing transmitter / reflector / receiver geometry . another standard assumption used in astronomical interferometric imaging is that the objects being imaged lie in the far field of the array of receiving antennas ; the wavefronts arriving at the array can be closely approximated as planar . it transpires that for space debris , most objects lie in the transition zone between near field and far field at low radio frequencies , for an array the size of the mwa . signal to noise is degraded somewhat when the standard far - field assumption is adopted , as an additional smearing of the imaged objects results . it is worth noting that this will be a much larger effect for the ska , with a substantially larger array footprint on the ground . thus , accounting for this effect for the mwa will be an important step toward using the ska for ssa purposes . for example , for an object at 1000 km and the mwa spatial extent of 3 km , the deviation from the plane wave assumption translates into more than a radian of phase error across the array at a frequency of 100 mhz , which produces appreciable smearing of the reconstructed signal with traditional imaging . it may be possible to use the near field nature of the problem to undertake 3-dimensional imaging of the debris , using the wavefront curvature at the array to estimate the distance to the debris . further significant work is required to implement modified signal and image processing schemes for mwa data that take account of these time - dependent and geometrical effects and sharpen the detections under these conditions . for example , the flexible approach to producing mwa visibility data ( the data from which mwa images are produced ) allows modifications to the signal processing chain to incorporate positional tracking of an object in motion in real - time . additionally , it is possible to collect data from the mwa in an even more basic form ( voltages captured from each antenna ) and apply high performance computing in an offline mode to account for objects in motion . this would also enable measurement of the doppler shifts of individual signals ( expected to be @xmath64khz , which is smaller than our current 10khz resolution ) that could enable separation of different transmitters at the same frequency and unambiguous identification of transmitters ( a doppler shift pattern fixes a one - dimensional locus perpendicular to the path of the reflecting object , so combining two passes allows two - dimensional localizations ) . that in turn would help with constraining the basic properties of the reflecting objects . finally , modifications to image processing algorithms can be applied in post - processing to correct for near - field effects , essentially a limited approximation of the same algorithms that can be applied earlier in the signal chain . the results of this future work will be reported in subsequent papers . once these improvements are addressed , the remaining significant challenges revolve around searching for and detecting signals from known and unknown space debris . these challenges are highly aligned with similar technical requirements for astronomical applications , in searches for transient and variable objects of an astrophysical nature . a large amount of effort has already been expended in this area by the mwa project , as one of the four major science themes the project will address @xcite . to illustrate the challenges , our flux estimates require ( for example ) greater than 4@xmath12 significance , but that is for a single trial . this is correct when looking to recover debris with an approximately known ephemeris , but blind searches have large trial factors that can require significantly revised thresholds . given the full @xmath65 field - of - view of the mwa , we must search over @xmath66 individual positions in each image to identify unknown debris . combined with 3000 channels and 3600 time samples ( for an hour of observing ; note that we might observe with a shorter integration time in the future ) this results in @xmath67 total trials . so our 4@xmath12 threshold ( which implies 1 false signal in @xmath68 trials ) would generate @xmath69 false signals , and we would need something like an @xmath70 threshold to be assured of a true signal . given the steep dependence of flux on object size ( equation [ eqn : flux ] ) this means our size limit may need to be increased significantly . however , knowledge of individual bright transmitters ( from the type of analysis presented in this paper ) means we can reduce the number of trials in the frequency axis substantially , and seeking patterns among adjacent time samples that fit plausible ephemerides will also help mitigate this effect . if the challenges above can be addressed and a robust debris detection and tracking capability can be established with the mwa , it could become a very useful element of a suite of ssa facilities focussed on southern and eastern hemispheric coverage centered on australia . in november 2012 an announcement was made via an australia - united states ministerial consultation ( ausmin ) joint communiqu that a us air force c - band radar system for space debris tracking would be relocated to western australia . this facility will provide southern and eastern hemispheric coverage , allowing the tracking of space debris to a 10 cm size scale . this system can produce of order 200 object determinations per day ( multiple sets of range , range - rate , azimuth and elevation per object ) and will be located near exmouth in western australia , only approximately 700 km from the mwa site . additionally , the australian government has recently made investments into ssa capabilities , via the australian space research program ( asrp ) : rmit univerity s `` platform technologies for space , atmosphere and climate '' ; and a project through eos space systems pty ltd , `` automated laser tracking of space debris '' . the eos project will result in an upgrade to a laser tracking facility based at mt stromlo observatory , near canberra on australia s east coast , capable of tracking sub-10 cm debris at distances of 1000 km . in principle , the c - band radar system , a passive radar facility based on the mwa , and the laser tracking facility could provide a hierarchy of detection and tracking capabilities covering the southern and eastern hemispheres . the c - band radar system could potentially undertake rapid but low positional accuracy detection of debris , with a hand - off to the mwa to achieve better angular resolution and rapid initial orbit determination ( via long duration tracks ) . the mwa could then hand off to the laser tracking facility for the most accurate orbit determination . such a diversity of techniques and instrumentation could form a new and highly complementary set of ssa capabilities in australia . a formal error analysis , describing the benefits of combining the data from such a diverse set of tracking assets will be the subject of work to be presented in a future paper . such a hierarchy of hand - off between facilities has implications for scheduling the mwa , whose primary mission is radio astronomy . currently the mwa is funded for science operations at a 25% observational duty cycle ( approximately 2200 hours of observation per year ) . this duty cycle leaves approximately 6500 hours per year available to undertake ssa observations ( assuming that funds to operate the mwa for ssa can be secured ) . even taking a conservative approach where only night time hours are used for observation , up to approximately 2200 hours could be available for ssa . it would be trivial to schedule regular blocks of mwa time , per night or per week , to undertake ssa observations . given the rapid sky coverage of the mwa and the short observation times effective for ssa , the entire sky could be scanned over the course of approximately one hour each night , or using multiple shorter duration observations over the course of a night . since the mwa can be rapidly repointed on the sky , the possibility of interupting science observations for short timescale follow - up ssa observations , triggered from other facilities such as the c - band radar , could be considered . a review of australia s capabilities in ssa was presented by @xcite , pointing out that better connections between the ssa community and latent expertise and capabilities in the australian astronomical community could be leveraged for significant benefit . in particular the authors suggested involving new radio astronomy facilities in ssa activities , given the advances in western australia connected with the ska . the development of a passive radar facility in the fm band using the mwa would be a significant step in this direction . immediate future steps are to repeat the iss measurements with the full 128 tile mwa , as an improvement on the 32 tile array used for the observations reported here , as well as to target smaller known satellites as tests and to implement the improvements described above . we acknowledge the wajarri yamatji people as the traditional owners of the observatory site . support for the mwa comes from the u.s . national science foundation ( grants ast career-0847753 , ast-0457585 , ast-0908884 , ast-1008353 , and phy-0835713 ) , the australian research council ( lief grants le0775621 and le0882938 ) , the u.s . air force office of scientic research ( grant fa9550 - 0510247 ) , the centre for all - sky astrophysics ( an australian research council centre of excellence funded by grant ce110001020 ) , new zealand ministry of economic development ( grant med - e1799 ) , an ibm shared university research grant ( via vuw & curtin ) , the smithsonian astrophysical observatory , the mit school of science , the raman research institute , the australian national university , the victoria university of wellington , the australian federal government via the national collaborative research infrastructure strategy , education investment fund and the australia india strategic research fund and astronomy australia limited , under contract to curtin university , the ivec petabyte data store , the initiative in innovative computing and nvidia sponsored cuda center for excellence at harvard , and the international centre for radio astronomy research , a joint venture of curtin university and the university of western australia , funded by the western australian state government . the electromagnetic simulations were performed by gary bedrosian and randy ward of remcom inc . the radio and television broadcasting stations internet edition , commonwealth of australia , 2011 . issn 1449 - 5686 ( http://www.acma.gov.au ) bowman , j. et al . 2013 , pasa , 30 , 31 dempsey , p. et al . 2012 , committee on the peaceful uses of outer space scientific and technical subcommittee , forty - ninth session ( vienna , 6 - 17 february 2012 ) item 14 of the draft provisional agenda , long - term sustainability of outer space activities ( a / ac.105/c.1/2012/crp.16 ) dewdney , p.e . , hall , p.j . , schlizzi , r.t & lazio , t.j.l . 2009 , ieee , 97 , 8 , 1482 farnocchia , d. , tommei , g. , milani , a. & rossi , a. 2009 , arxiv:0911.0149 haslam , c. g. t , salter , c. j. , stoffel , h. & wilson , w. e. 1982 , a&as , 47 , 1 johnston , s. et al . 2008 , ex.a . , 22 , 151 mckinley , b. et al . 2013 , aj , 145 , 23 newsam , g. & gordon , n. 2011 , conference technical papers of the 2011 advanced maui optical and space surveillance technologies conference ( http://www.amostech.com/technicalpapers/2011.cfm ) murphy , t. et al . 2013 , pasa , 30 , 6 ord , s. m. et al . 2010 , pasp , 122 , 1353 petersen , j.k . 2012 , handbook of surveillance technologies ( 3rd edition ) , crc press ( florida , usa ) isbn:978 - 1 - 4398 - 7315 - 1 ruitz , leushacke & rosebrock 2005 , proceedings of the 4th european conference on space debris ( esa sp-587 ) . 18 - 20 april 2005 , esa / esoc , darmstadt , germany . editor : d. danesy . , p.89 sault r.j . , teuben p.j . , & wright m.c.h . 1995 . in astronomical data analysis software and systems iv , ed . r. shaw , h.e . payne , j.j.e . hayes , asp conference series , 77 , 433 stratton , j.a . 1941 , electromagnetic theory , new york : mcgraw - hill tingay , s.j . 2013 , pasa , 30 , e007 technical report on space debris , 1999 , united nations publication , sales no . e.99.i.17 , isbn 92 - 1 - 100813 - 1 welch , j. et al . 2009 , ieee , 97 , 1438 williams , c. l. et al . 2012 , apj , 755 , 47 wills , n.j . 2005 , bistatic radar , scitech publishing inc . ( north carolina , usa ) isbn:1 - 891121 - 45 - 6	the murchison widefield array ( mwa ) is a new low frequency interferomeric radio telescope , operating in the benign radio frequency environment of remote western australia . the mwa is the low frequency precursor to the square kilometre array ( ska ) and is the first of three ska precursors to be operational , supporting a varied science mission ranging from the attempted detection of the epoch of reionisation to the monitoring of solar flares and space weather . in this paper we explore the possibility that the mwa can be used for the purposes of space situational awareness ( ssa ) . in particular we propose that the mwa can be used as an element of a passive radar facility operating in the frequency range 87.5 - 108 mhz ( the commercial fm broadcast band ) . in this scenario the mwa can be considered the receiving element in a bi - static radar configuration , with fm broadcast stations serving as non - cooperative transmitters . the fm broadcasts propagate into space , are reflected off debris in earth orbit , and are received at the mwa . the imaging capabilities of the mwa can be used to simultaneously detect multiple pieces of space debris , image their positions on the sky as a function of time , and provide tracking data that can be used to determine orbital parameters . such a capability would be a valuable addition to australian and global ssa assets , in terms of southern and eastern hemispheric coverage . we provide a feasibility assessment of this proposal , based on simple calculations and electromagnetic simulations that shows the detection of sub - metre size debris should be possible ( debris radius of @xmath0 m to @xmath11000 km altitude ) . we also present a proof - of - concept set of observations that demonstrate the feasibility of the proposal , based on the detection and tracking of the international space station via reflected fm broadcast signals originating in south - west western australia . these observations broadly validate our calculations and simulations . we discuss some significant challenges that need to be addressed in order to turn the feasible concept into a robust operational capability for ssa . the aggregate received power due to reflections off space debris in the fm band is equivalent to a @xmath2 mjy increase in the background confusion noise for the long integrations needed for epoch of reionisation experiments , which is insignificant .
quantum entanglement @xcite , the most characteristic feature of quantum mechanics , is one of the central concepts in quantum information theory and is the feature that distinguishes it most significantly from the classical theory . entanglement is now viewed as a physical resource , which provides a means to perform quantum computation and quantum communication @xcite . it should be emphasized that property of entanglement can be considered regardless of the nature of qubits@xcite . one of the most intensively investigated systems is clusters of coupling nuclear spins @xcite which received considerable attention as a platform for the practical implementation of a quantum computer ( qc ) by using nuclear magnetic resonance ( nmr ) technique @xcite . the strength of the coupling ( such as dipole - dipole , scalar or exchange interactions ) between different spins and interaction of the spins with environment determine the decoherence time . short decoherence times limit possible calculation times of the qc @xcite . later it has been proposed to used quadrupole nuclei , thus eliminating the requirement of interaction between spins and decreasing interaction of qubits with environment @xcite it was shown that a spin @xmath1 system is equivalent to a system of two magnetization vectors @xcite and can be represented using the pauli spin matrices 2x2 @xcite . it means that a single spin 3/2 is isomorphic to a system consists of two dipolar coupling spins 1/2 , which can be considered as qubits . this was experimentally confirmed by using nqr technique with quadrupole splitting @xcite . the feasibility of quantum computing based on a pure ( without external magnetic fields ) nqr technique was theoretically investigated in detail in @xcite . using the resonance excitation technique and the level - crossing method , it was proposed three quantum logic gates : _ _ a controlled @xmath2 , @xmath3 and @xmath4 . thus the method to synthesize qubits from a set of the spin states of a single particle with spin higher 1/2 has been developed . it is logical to raise the question of entanglement of these qubits . our present purpose is to investigate entanglement between the quantum states of quadrupole nuclei . we consider a nucleus with spin @xmath5 being in an internal electric field gradient ( efg ) and an external magnetic field when the quadrupole interaction energy is of the order of the magnitude or even greater than the zeeman one . results of computer simulations of entanglement dynamics are presented for real spin systems with a non - equadistant energy spectrum . let us consider a quadrupole nucleus with spin @xmath6 ( @xmath7 ) in the thermodynamic equilibrium with the density matrix @xmath8 here @xmath9 is the boltzmann constant , @xmath10 is the lattice temperature , @xmath11 is the partition function . in the general case , the hamiltonian @xmath12 can consist of the zeeman ( @xmath13 ) and the quadrupole ( @xmath14 ) parts,@xmath15 the zeeman interaction between the applied magnetic field , @xmath16 and nuclear spin is the external factor for a crystal and the direction of this field is chosen as the @xmath17-axis of the laboratory frame , @xmath18 , where @xmath19 is the strength of the external magnetic field . the part of the hamiltonian describing this interaction is @xmath20 where @xmath21 is the gyromagnetic ratio of the nucleus with spin _ _ @xmath6 , @xmath22 is the projection of the individual spin angular momentum operators @xmath23 on the @xmath17- axis . quadrupole coupling exists between a non - spherical nuclear charge distribution and an electric field gradient ( efg ) generated by the charge distribution in their surroundings . it is possible to reduce the efg symmetric tensor to a diagonal form by finding the principal axes frame ( paf ) with the @xmath24- and @xmath25-axises directed along the maximum and minimum of efg , respectively , @xmath26 , where @xmath27 and @xmath28 is the potential of the electric field . in the laboratory frame the quadrupolar hamiltonian can be presented in the following form @xmath29 u^{+}\left ( \theta\ , \varphi\right ) , \tag{4}%\ ] ] where @xmath30 @xmath31 is the quadrupole coupling constant of efg , @xmath32 are the raising and lowering operators of the spin , and @xmath33and @xmath34 refer to the polar and azimuthal angles determining the orientation of the laboratory frame @xmath17-axis _ _ in the paf coordinate system . the asymmetry parameter _ _ @xmath35 is defined as @xmath36 which may vary between @xmath37 and @xmath38 . using the hamiltonian ( 2 ) the density matrix ( 1 ) can be represented as a function of parameters @xmath39 and @xmath40:@xmath41 u^{+}\left ( \theta\ , \varphi\right ) \right\ } . \tag{7}%\ ] ] the density matrix ( 7 ) can be employed to obtain information on the dependence of entanglement on the magnetic field , quadrupole coupling constant , orientation of the crystal principal axis in the laboratory frame , and temperature . _ _ below we consider entanglement in a system of spins 3/2 . _ _ a suitable system for studying by nqr technique is a high temperature superconductor @xmath42 containing the @xmath43 and @xmath44 nuclei with spin @xmath5 possessing quadrupole moments @xmath45 @xmath46 and @xmath47 @xmath46 , respectively @xcite . there are two different locations of copper ions in this structure : the first are the copper ion sites at the center of an oxygen rhombus - like plane while the second one is five - coordinated by an apically elongated rhombic pyramid . the four - coordinated copper ion site , efg is highly asymmetric @xmath48 while the five - coordinated copper ion site , efg is nearly axially symmetric @xmath49 . the quadrupole coupling constant @xmath50 of @xmath43 in the four - coordinated copper ion site is @xmath51 mhz@xmath52 and in the five - coordinated copper ion site is @xmath53 mhz @xcite . an important measure of entanglement is the concurrence @xcite . the concurrence @xmath54 is usually used @xcite . for the maximally entangled states , the concurrence is @xmath55 , while for the separable states @xmath56 . the concurrence of a quantum system with the density matrix presented in the hilbert space as a matrix @xmath57 is expressed by the formula @xcite:@xmath58 where @xmath59 and @xmath60 @xmath61 are the square roots of the eigenvalues of the product @xmath62 where @xmath63 is the reduced density matrix , @xmath64 is the partial trace , and @xmath65 where @xmath66 is the complex conjugation of the reduced density matrix @xmath67 and @xmath68 an another important measure of entanglement is the entanglement entropy . in order to implement this measure of entanglement we first map the hilbert space of a spin 3/2 which is four dimensional onto the hilbert space of two spins 1/2 , @xmath69 and @xmath70 @xcite . the entropy of entanglement is defined as @xcite:@xmath71 here @xmath72 is the partial trace of @xmath73 over subsystem @xmath74 , and @xmath75 has a similar meaning . this entropy is related to the concurrence @xmath54 by the following equation @xcite @xmath76 where @xmath77 . the numerical simulations of entanglement of the spin states are performed for special case @xmath78 using the software based on the _ mathematica _ package . in the equilibrium the states of the system determined by eq . ( 7 ) are separable without applying external magnetic filed @xmath79 at any temperature . at large temperature @xmath80 and low magnetic field strength @xmath81 the concurrence is zero . entanglement appears in the course of temperature decrease and increasing the magnetic field . the concurrence and the entanglement of formation dependences on the magnetic field , quadrupole coupling constant , and temperature is qualitatively independent of the angles . as example , we present below the concurrence as a function of the parameters @xmath82 and @xmath83 for the concurrence maximum for a spin in the five - coordinated copper ion site ( @xmath84 at @xmath85and @xmath86 ( fig . the concurrence and the entanglement of formation increase with the magnetic field strength and inverse temperature and reache their maximum value . then the concurrence and the entanglement of formation decrease with increasing the magnetic field strength ( fig . _ _ another dependence of the concurrence and the entanglement of formation on temperature is observed ( fig . 3 ) . at a high temperature concurrence and the entanglement of formation are zero . with a decrease of temperature below a critical value the concurrence and the entanglement of formation monotonically increase till a limiting value . the critical temperature and limiting value are determined by a ratio of the zeeman and quadrupole coupling energies , @xmath87 the obtained results show that the entangled states can be generated between the states of a single nuclear spin @xmath5 . from a point of of view of quantum information processing the considered system is isomorphic to a system consisting of two dipolar coupling spins @xmath88 . the same quantum logical gates can be realized using the both systems . therefore the obtained entanglement can be considered as entanglement between qubits formed by states of a single particle . on the other hand , a single spin 3/2 isomorphic to a system consists of two dipolar coupling spins 1/2 and entanglement between the states of a spin 3/2 can be considered as entanglement between two effective spins 1/2 . the behavior of entanglement of a spin 3/2 is very similar to that for the system consisting of two dipolar coupling spins @xmath88 @xcite . it was obtained that in zero magnetic field the states of the both spin systems , two spins @xmath88 and spin @xmath5 , are in separable states . these systems become entangled with increasing the magnetic field . then , with a further increase of the magnetic field the spin states of the both systems tend to a separable one . it has recently been shown that , in a system of nuclear spins @xmath89 = 1/2 , which is described by the idealized xy model and dipolar coupling spin system under the thermodynamic equilibrium conditions , entanglement appears at very low temperatures @xmath90 @xmath91k @xcite . in a non - equilibrium state of the spin system , realized by pulse radiofrequency irradiations , estimation of the critical temperature at which entanglement appears in a system of spins @xmath88 gives @xmath92 k @xcite . the calculation for @xmath43 in the five - coordinated copper ion site of @xmath93 at @xmath94 , @xmath95 and @xmath96 @xmath53 mhz , gives that the concurrence appears at @xmath97 ( fig . 4).@xmath98this @xmath83 value corresponds to temperature @xmath99 mk . this estimated value of critical temperature is by three orders greater than the critical temperature estimated for the two dipolar coupling spins under the thermodynamic equilibrium @xcite . in conclusion , performing investigation has shown that entanglement can be achieved by applying a magnetic field to a single spin 3/2 at low temperature . concurrence and the entanglement of formation depend on the orientation between the external magnetic field and paf axes . at @xmath100 and @xmath101 the states are separable at any conditions . at @xmath95 the concurrence and the entanglement of formation reache their maximum value _ _ at @xmath102 and @xmath86 and@xmath103 . this effect can open a way to manipulate with the spin states by a rotation of the magnetic field or a sample . 2 ( color online ) concurrence ( a ) and the entanglement of formation ( b ) vs. magnetic field at @xmath10 = const for various quadrupole interaction constants : black solid line @xmath108 ; red dashed line @xmath109 ; green dotted line @xmath110 ; blue dash - doted line @xmath111 . fig . 3 concurrence ( a ) and the entanglement of formation ( b ) as a function of temperature at @xmath112 ( black solid line ) , @xmath113 ( red dashed line ) , and @xmath114 ( blue dotted line ) at @xmath95 , @xmath102 , @xmath86 temperature is given in units of @xmath115 .	entangled quantum states are an important element of quantum information techniques . we determine the requirements for states of quadrupolar nuclei with spins @xmath01/2 to be entangled . it was shown that entanglement is achieved at low temperature by applying a magnetic field to a quadrupolar nuclei possess quadrupole moments , which interacts with the electricfield gradient produced by the charge distribution in their surroundings .
why there is so little anti - matter in the universe , and why the matter coalesced in the way it did are two of the major problems facing cosmology . predictably , both have attracted a great deal of attention spawning a panoply of explanations and theories . some of these theories involve objects known as topological defects@xcite , regions of trapped primordial vacuum , an example of which is the cosmic string . a string approach is an appealing one since it can be used to address both questions : the wake left by strings moving through the universe can produce fluctuations which may lead to the accretion of matter into large scale structures@xcite@xcite , whilst their interaction with particles and the decay of string loops can provide mechanisms leading to baryon number violation@xcite and the observed matter bias@xcite . hence , it is important to understand how these strings form , both to predict how many we can expect to have been created and how likely the above processes are . the formation of topological defects is thought to proceed via the kibble mechanism@xcite . modern particle physics and the hot big bang model suggest that as the universe cooled it underwent several phase transitions in which the symmetry of the vacuum was broken into a successively smaller , and smaller group . during such a transition , it is possible for fields to acquire non - zero vacuum expectation values . how they do this depends on the order of the transition . if we consider a transition where a u(1 ) symmetry is broken , then following the transition all points in space will have a physically identical , non - zero vacuum expectation value , the only variation being in the difference in phase between any two points . by causality , we expect the phases to be uncorrelated on distances greater than the horizon length , and so there is a finite probability that the phase along a closed path through a number of horizon volumes will wind through some multiple of @xmath0 - an indication that the loop contains a string . in practice two points do not need to be separated by a horizon length for their phases to be uncorrelated . this should also be true if they are a thermal correlation length ( defined later ) apart ; usually a considerably smaller distance than the horizon size . this is the kibble mechanism , and it relies upon the so - called geodesic rule , which is that in passing between two domains of different phase , the phase will follow the shortest route . in the global case this has been verified both numerically@xcite and experimentally@xcite@xcite , though , for a local symmetry it has been argued@xcite that the presence of gauge fields may influence the path the phase takes , and may actually prevent it from following a ` geodesic ' . more recently , however , work has been done suggesting that , despite this , the geodesic rule holds in the local case for a first order phase transition@xcite . as the temperature falls below the critical temperature , for a while it is still possible for thermal fluctuations to restore the broken symmetry , and hence erase any topologically interesting configuration present at the time . the point at which it is thought that this ceases to be possible is referred to as the ginzburg temperature , @xmath1 , and is found by equating the free energy with the thermal energy ( for such a restoring fluctuation will have a high probability while the former is considerably less than the latter ) . brandenberger and davis@xcite demonstrated that given certain constraints on the parameters , the ratio of fluctuations in the scalar field to the background is less than one beneath a temperature just below the ginzburg temperature , regardless of whether gauge fields are , or are not , present . this means that topologically non - trivial configurations arising from thermal fluctuations will become stable to such fluctuations just under the ginzburg temperature . this adds weight to the arguments in favour of the kibble mechanism . however , since brandenberger and davis considered a linearized model , only valid for the short time immediately following a spinodal decomposition , before non - linear effects start to dominate , their analysis only holds for early times . to study the evolution of the fields at later times it is necessary to include the non - linearities . a flexible way to do this is to study the langevin equation associated with the classical field equations@xcite . this is the purpose of this paper . by studying the langevin equation for the system , we derive an equation for the probability distribution of the fields , @xmath2 which we use to analyse the evolution of the expectation values of the classical fields coupled to a thermal bath . this enables us to study not only the stability of configurations to fluctuations at and below the ginzburg temperature , but also the long time evolution of the fields . the flexibility of this method is demonstrated by the ease with which it is modified to include the expansion of the universe . our method seems to be the only one that allows the study of the effect that thermal fluctuations have on the development of the field , and the stability of defects formed , _ throughout _ the phase transition . other methods either concentrate on the start of the phase transition , or near its completion . although our ultimate aim is to study the case of an expanding universe with broken local symmetry , it is beneficial , for several reasons , to start with the more straightforward case of a global , non - expanding model . firstly , since the kibble mechanism has already been verified for this case , we know that any topologically non - trivial configurations present should be stable to fluctuations below @xmath1 , and hence we have a benchmark to check our results against . although this provides a useful test of our method , it is by no means a proof of its validity . secondly , since the amount of algebra involved is very dependent on the number of fields present , the global , non - expanding case provides the simplest , and hence clearest , demonstration of the method used throughout . consider the @xmath3 toy model @xmath4 where @xmath5 is a complex scalar field , @xmath6 is the u(1 ) gauge connection ( taken to be zero for now ) and @xmath7 . we adopt an effective potential of the form @xmath8 where @xmath9 and the temperature dependence reflects the fluctuations on a scale smaller than some correlation length , defined later . for sufficiently high temperatures this is symmetric about a global minimum at zero . however , as the temperature passes through some critical temperature , @xmath10 , the system undergoes a second order phase transition , breaking the u(1 ) symmetry , with new minima appearing at @xmath11 . any two points in the new vacuum will now have non - zero vacuum expectation values of equal moduli but random phase . setting @xmath12 , our equations of motion for @xmath13 and @xmath14 are @xmath15 we assume that @xmath14 is time independent and varies spatially over some length - scale @xmath16 . setting @xmath17 we obtain @xmath18 where @xmath19 and @xmath20 is explained below . note that we have replaced @xmath21 with @xmath22 c.f the mean square approximation to make the resulting equations more accessible . for the purpose of this analysis we consider an initial configuration varying spatially on a scale of the correlation length . since we do not want to become embroiled in a discussion of effects due to fluctuations on scales shorter than the thermal correlation length , @xmath23 $ ] , we consider a coarse - grained field where we have integrated out all modes associated with such . this leads to the effective potential mentioned earlier . hence , if we were to perform a fourier decomposition , then it would be of the form @xmath24 for some @xmath25 . we also assume that the mode corresponding to @xmath26 dominates , ( which we later show to be self - consistent ) and investigate a configuration with a length scale @xmath27 . by ( [ octopus ] ) we see that the earlier assumption that @xmath28 requires @xmath29 . to incorporate thermal fluctuations into our model , we modify ( [ octopus ] ) such that the equations describing the evolution of @xmath30 and @xmath13 over some small time interval @xmath31 are @xmath32 where @xmath33 and @xmath34 are the thermal fluctuations in @xmath13 and @xmath30 respectively . defining @xmath35 , @xmath36 , and similarly @xmath37 and @xmath30 , we can write @xmath38 where @xmath39 and @xmath40 are the probability measures for @xmath33 and @xmath34 respectively . since @xmath33 and @xmath34 are random fluctuations , we may assume that @xmath41=@xmath42 . expanding the integrands as taylor series , we find , after considerable algebra , that @xmath43 from which , assuming that @xmath33 and @xmath34 are independent of @xmath13 and @xmath30 , we obtain the following differential equation for p : @xmath44 one interpretation of this equation is as follows . if we move the first two terms on the right hand side over to the left , then we have a full derivative of @xmath45 . liouville s theorem states that for a closed system this derivative should be zero . however , our system is coupled to a thermal bath and so there is a flow of probability between the two , as demonstrated by the two non - zero noise terms , due to the bath , on the right hand side . this equation is clearly a rather forbidding equation to solve analytically . however , we can use it to derive equations governing the quadratic moments @xmath46 we choose to investigate these moments because their equations form a closed system with the mean field approximation we have taken . as a modification we set @xmath47 since this normalises @xmath48 , @xmath49 and @xmath50 , and gives the equations for the moments in a form where the relative sizes of terms are much more apparent ; @xmath51 where @xmath52 denotes @xmath53 , @xmath54 are the fluctuations and @xmath55 it should be noted that these equations are even nastier than they look at first glance , since @xmath56 contains a term proportional to @xmath57 . however , by assuming that the temperature varies at a much slower rate than the fields ( something which we will see is self - consistent later on ) it is possible to make some progress analytically . for the time being , we consider the case when fluctuations are absent , and treat @xmath56 as constant over a small time period . after a bit of substitution , we integrate the equations to obtain @xmath58 where @xmath59 , @xmath60 and @xmath61 , @xmath62 and @xmath63 are the initial values of @xmath48 , @xmath49 and @xmath50 . let the three roots of the polynomial on the right hand side be @xmath64 . hence @xmath65 taking initial conditions such that , at @xmath1 , @xmath66 $ ] , which is the minimum of the effective potential at that time , and @xmath67 , we find @xmath68 , implying that one of the roots is zero . furthermore , @xmath69 and @xmath56 are found to be negative , so the two non - zero roots must be positive . since @xmath70 is seen to be positive between @xmath71 and @xmath72 we expect @xmath48 to oscillate between these two . we can also make an estimate of the time period of these oscillations , @xmath73 , since @xmath74 where @xmath75 is the first complete elliptic integral and @xmath76}$ ] . these results closely agree with the corresponding numerical calculations . however , they are only valid for very small fluctuations , and so we turn to a numerical approach for a more detailed analysis . in order to study the effect of fluctuations on the evolution of the fields , it is necessary to make some estimate of the fluctuation terms . to do this we imagine @xmath5 coupled to some other field , @xmath77 , in thermal equilibrium , via an extra term in the lagrangian of the form @xmath78 by comparing the resulting equations of motion with those already obtained we find that @xmath79 since @xmath77 is in thermal equilibrium , we also have that @xmath80 , with corresponding number density @xmath81 where @xmath82 is the number of internal degrees of freedom ( 107 for a grand unified theory ) . taking @xmath83 to be a typical interaction time , such that @xmath84 , where @xmath85 is the interaction cross - section , we find the following approximation for the fluctuation terms @xmath86 as expected , the size of the fluctuations decreases with temperature . before we can carry out a calculation , we need to address the problem of choice of parameter values . since the model we are considering is a global one , we see from the definition of the potential that we must have @xmath87 . we now demonstrate why we are justified in assuming that the mode corresponding to @xmath20 dominates . figure 1 shows the evolution of the quadratic moment @xmath48 , ( corresponding to @xmath88 ) from the ginzburg time onwards , where we have taken @xmath89 , fluctuation coupling , @xmath90 and a range of different wavelengths . the most obvious feature is that the mode varying with wavenumber @xmath20 dominates those with longer wavelengths , consistent with our earlier assumption . however , one may be slightly alarmed at the fact that at least two of the curves look like they have no intention of converging to one ( corresponding to @xmath91 ) as one might expect . the reasons for this are twofold , and both somewhat of our own creation . the first is that in assuming that @xmath13 and @xmath14 vary spatially with fixed wavenumber , we alter the value of @xmath13 for which @xmath92 since we have essentially added two terms onto the derivative of the potential . this has the effect of raising the equilibrium value of @xmath13 . the second is that we have to make an arbitrary choice of @xmath14 ( taken throughout as @xmath93 ) , which effectively scales @xmath94 , and so has a similar effect to the first . conversely it could also be used to tune the expected equilibrium value to one by choosing a sufficiently small value of @xmath14 . this may raise questions over the validity of this method for studying the evolution of fields . however , the evolution of the fields is not qualitatively changed by taking different values of @xmath20 , @xmath94 or @xmath14 and so we argue that , as an approximation , our approach is still of interest . the potential we are using is , unfortunately , only a one loop approximation , and hence is not valid above the ginzburg temperature where higher loops dominate . our simulation therefore must run from the ginzburg time onwards , and so we can only investigate the stability of string configurations to thermal fluctuations and not their formation . we also only consider the case of a gut phase transition , since we expect one at a lower temperature , such as the electro - weak phase transition for instance , to be qualitatively the same , but slower due to the larger value of @xmath95 . we take the coupling @xmath96 to be between 0.01 and 1 , and @xmath97 . figure 2 shows the effect of the two couplings , @xmath98 and @xmath96 . the former controls the size of fluctuations , and it is seen , in 2a and 2b , that the larger @xmath98 , the quicker the rise of the lower bound , so fluctuations are , bizarrely , actually seen to help stabilise configurations , on average , by damping oscillations in the field . they also cause the upper bound to rise at a faster rate though this is not as pronounced , nor as important to the stability of domain structures . fig.2c reveals that decreasing @xmath96 decreases the frequency of oscillations , and also the asymptotic value for @xmath48 . the latter is because @xmath99 and we have already noted that the value of @xmath20 effects the limiting value . finally , fig.2d demonstrates that the effect of fluctuations decreases dramatically with @xmath96 . we see then that , since all curves move away from zero , any topologically non - trivial configuration is stable from the ginzburg temperature onwards , though the fields may take a long time to reach their equilibrium values . we also note that , in all cases considered , the oscillations occur on a much smaller timescale than the evolution towards the equilibrium value ; consistent with our earlier assumption . that the configurations formed in the above transition are stable against thermal fluctuations is nothing new . the kibble mechanism for the global case is already well accepted , since one can argue in favour of the geodesic rule just by demanding that the path followed minimizes the energy density . however , the presence of gauge fields may undermine this , since their presence in the gradient energy , @xmath100 , may make it equally favourable , energetically , to follow a longer path . luckily , the method used for studying the global case works equally well in the local one , the only drawback being a significant increase in the amount of algebra that has to be done . we start by writing the equations of motion in the form @xmath101 where @xmath102 ( note that this is gauge invariant ) . setting @xmath103 , @xmath17 and @xmath104 for convenience , we find @xmath105 where @xmath106 and @xmath34 , @xmath33 , @xmath107 , @xmath108 are the thermal fluctuations . proceeding in exactly the same manner as before , we arrive , after some very unpleasant algebra , at the equation for @xmath109 , @xmath110 it is important to note that the sum over indices in the last two terms is over all four indices , not the usual two . once more we see the violation of liouville s theorem via the coupling to the heat bath . from this it is straightforward to obtain the equations governing the quadratic moments . following the global method and defining @xmath111 , @xmath48 , @xmath49 , @xmath50 as before , plus @xmath112 we arrive at @xmath113 where @xmath114 and @xmath115 , @xmath116 , @xmath117 and @xmath118 are the thermal fluctuation terms . note that the terms involving @xmath119 have integrated to zero . clearly we are not going to get too far with an analytic approach this time , so we restrict ourselves to a numerical analysis . as before , our first preparation is to calculate the fluctuation terms . since the coupling of @xmath5 to @xmath77 is independent of gauge fields , the two new fluctuation terms , @xmath117 and @xmath118 must be zero . hence , the only non - zero fluctuation term is @xmath116 , which is unchanged . we consider once more an initial domain structure of length scale @xmath120 , and take @xmath121 , since we expect the gauge coupling to dominate . figure 3 shows the effect of varying @xmath96 and @xmath98 , the results being very similar to those observed in the global symmetry case . figures 3a and 3b once again demonstrate the effect of increasing the size of the fluctuations ; an increase in the rate of growth of the lower bound and a damping of oscillations , whilst fig.3c reveals how decreasing the size of the self - coupling decreases the frequency of oscillations , the effect of fluctuations decreasing in a similar manner ( fig.3d ) . the most important feature however is that , as in the global case , any non - trivial domain structure present at @xmath122 is seen to be stable against fluctuations at greater times . this reinforces the work by brandenberger and davis@xcite , and , similarly , the arguments in favour of the kibble mechanism . until now we have ignored the expansion of the universe , which we would expect to damp the amplitude of any oscillations present . to make the analysis more realistic it is necessary to include this expansion . taking first the case with a global symmetry , such a modification is straightforward , and leads to an extra term in the equation of motion for @xmath13 proportional to the hubble parameter @xmath123 , @xmath124 as expected , a damping term . the only effect this has on the equations for the quadratic moments @xmath48 , @xmath49 and @xmath50 , is to add @xmath125 to the right hand side of equation ( [ neq2 ] ) , and @xmath126 to that of ( [ neq3 ] ) , though @xmath56 is now written as @xmath127\\ \nonumber\end{aligned}\ ] ] where @xmath128 is the expansion parameter , and @xmath129 its value at @xmath1 is an unphysical quantity , we can without loss of generality take @xmath130 . ] . figure 4 shows the results for a small selection of values to illustrate the effects of varying the different parameters . once more we consider an initial domain structure of length scale @xmath120 . all four diagrams are seen to display the rapid damping due to the expansion of the universe . 4a and 4b demonstrate the effect of fluctuations . in 4b , where the fluctuations are suppressed , the lower bound on @xmath48 rises much more slowly than in the unsuppressed case , fig.4a , which actually overshoots its asymptotic value of one before reconverging . hence , it is seen that fluctuations actually make it less likely that a configuration will be erased , agreeing with our non - expanding simulations . the upper bound varies very little between the two . 4c shows the effects of reducing the self - coupling ; a longer period of oscillation , a less dramatic initial growth and a much gentler approach toward its asymptotic value . 4d demonstrates how for small values of @xmath96 the fluctuations have very little effect on the evolution of the fields . in summary , topologically non - trivial configurations are stable to thermal fluctuations . we also note that due to the scale factor now present in the equations of motion , the effect of @xmath26 and @xmath94 is rapidly damped out so that in all expanding cases considered , @xmath48 converges on one , corresponding to @xmath131 tending to @xmath132 , the long - time minimum of the effective potential . now including gauge fields once more , the equation of motion for @xmath6 is , predictably , modified in a very similar way to that for @xmath13 when we include expansion , the new version being @xmath133 in addition , the equations for the quadratic moments @xmath134 , @xmath135 and @xmath136 acquire near identical terms to those already acquired by those for @xmath48 , @xmath49 and @xmath50 , the only difference being an extra factor of two in the former case . once more , we illustrate four different values of the parameters , for a configuration varying on a correlation length scale . as in the non - expanding case , we take @xmath121 throughout . figures 5a and 5b demonstrate the effect of fluctuations . in fig.5a , where the fluctuation coupling is comparable to the self - coupling , the lower bound to fluctuations is seen to rise much quicker ( as was the case for a global symmetry ) than that in fig.5b where the fluctuations are suppressed . so much so in fact that it overshoots its expected limit , though long time studies show that it gradually bends back and converges to one . in fig.5c we see the effect of decreasing the self - coupling ; a less dramatic growth of the lower bound , whilst fig.5d reveals , once more , that the effect of fluctuations decreases dramatically with @xmath96 . much the same as in the previous three cases . comparing figures 4 and 5 , what we notice is that the presence of gauge fields heavily damps the initial growth leading to a lower upper bound and consequently smaller oscillations . by studying the langevin equations for the classical fields we have verified that for a u(1 ) model with broken global symmetry ( whether expanding or not ) , string configurations formed during a second order phase transition , are stable to thermal fluctuations below the ginzburg temperature . we have also shown this to be true in the case of a system with a local symmetry , reinforcing earlier work@xcite on the subject , and lending further support to the kibble mechanism for the formation of topological defects . the same method has also been used to study how the fields evolve at late times , with the scalar field gradually tending to its equilibrium value ; a process accelerated by the damping produced by an expanding universe . indeed , our method tracks the evolution of the field _ throughout _ the phase transition . other methods are only able to consider early or late times . in addition , we have seen that thermal fluctuations actually accelerate the early evolution of the field , and damp the amplitude of oscillations in the field as it tends to its asymptotic value , making it even less likely that a fluctuation will destroy a configuration . this work is still an approximation however , since we have had to assume @xmath28 , leading to an arbitrariness in the asymptotic value of the field in the non - expanding models . however , this should not affect the stability of configurations , and in the expanding case this problem is smoothed out anyway as the model scales with time . another approximation we have made is in neglecting the dissipation term necessary when a source of fluctuations is present@xcite@xcite . since any dissipation term would have a damping effect , it would only increase the stability of a non - trivial domain structure , and so including it , a further avenue of research , should only strengthen our results . this and the quantum field theoretical treatment are in progress@xcite finally , the flexibility of the langevin equation approach@xcite may make this method suitable for a number of other applications , such as the study of the evolution of seed magnetic fields following the breaking of a non - abelian symmetry@xcite , and the stability of defects to fluctuations in condensed matter systems such as @xmath137he@xcite . work on these subjects is in progress@xcite . 99 t.w.b.kibble , _ _ @xmath138 , + t.w.b.kibble , _ physics reports _ @xmath139 . r.brandenberger , _ phys.scripta _ @xmath140 , + n.turok , _ phys.scripta _ @xmath141 . a.vilenkin and e.p.s.shellard , _ ` cosmic strings and other topological defects ' , cup , cambridge _ , @xmath142 . r.h.brandenberger , a .- c.davis and a.m.matheson , _ nuc.phys _ @xmath143 , + w.b.perkins et al , _ nuc.phys . _ @xmath144 , + m.alford , j.march-russell and f.wilczek , _ nuc.phys . _ @xmath145 . r.h.brandenberger , a .- c.davis and m.hindmarsh , _ phys.lett . _ @xmath146 , + a .- c.davis and m.a.earnshaw , _ nuc.phys . _ @xmath147 . j.ye and r.brandenberger , _ _ @xmath148 . i.chuang , r.durrer , n.turok and b.yurke , _ science _ , @xmath149 , + b.yurke , a.pargellis , i.chuang and n.turok , _ physica _ @xmath150 , + m.bowick , l.chandlar , e.schiff and a.srivastava , _ science _ p.hendry et al . , _ j.low temp.phys._ @xmath152 . m.hindmarsh , a .- c.,davis and r.brandenberger , _ phys.rev . _ @xmath153 . s.rudaz and a.srivastava , _ mod.phys.lett . _ @xmath148 . r.brandenberger and a .- c.davis , _ phys.lett . _ @xmath154 . r.brandenberger , h.feldman and j.macgibbon , _ phys.rev . _ @xmath155 . m.gleiser and r.o.ramos , _ phys.rev _ d.boyanovsky , h.j.de vega , r.holman , d .- s.lee and a.singh , _ pitt_@xmath157 , _ lpthe_@xmath158 , _ cmu - hep_@xmath159 , dor - er/40682 - 77 . c.davis , a.p.martin and r.j.rivers , _ in preparation_. t.vachaspati , _ phys.lett . _ , * b*@xmath160 . w.h.zurek , _ proceedings of nato asi ` formation and interactions of topological defects ' _ ed . a .- c.davis and r.h.brandenberger , plenum press , to appear . c.davis and a.p.martin , _ in preparation_.	we analyse the evolution of scalar and gauge fields during a second order phase transition using a langevin equation approach . we show that topological defects formed during the phase transition are stable to thermal fluctuations . our method allows the field evolution to be followed throughout the phase transition , for both expanding and non - expanding universes . the results verify the kibble mechanism for defect formation during phase transitions . = 0.3 in = 0.2 in = -0.2 in = -0.2 in = 9.2 in = 6.55 in = -0.5 in pubnumber # 1pubnumberdamtp 94 - 87 makepubto @to maketitle makepub 2em title 1.5em 0.5em [ cols="^ " , ] 1em date 1.5em
one question which has recently received a great deal of attention in the literature is whether or not the mass of an agn s black hole is strongly related to it s radio luminosity . this question is of importance , because if it is established that radio - loud and radio - quiet quasars have different black - hole mass distributions , it may help explain why quasars of comparable optical luminosities can differ in their radio luminosity by many orders of magnitude . on the contrary , if radio - loud and radio - quiet quasars are found to have essentially identical black - hole mass distributions , then the search for the origin of radio loudness must move to some other physical parameter such as black - hole spin . using the spectral data of boroson & green ( 1992 ) , laor ( 2000 ) investigated the relation between black - hole mass and radio luminosity in the palomar - green quasar sample using the virial black - hole mass estimator . the results from this analysis pointed to an apparent bi - modality in black - hole mass , with virtually all of the radio - loud quasars containing black holes with masses @xmath4m@xmath1 , whereas the majority of quasars with black hole masses @xmath5m@xmath1 were radio quiet . a similar result was arrived at by mclure & dunlop ( 2002 ) using a sample of radio - loud and radio - quiet quasars matched in terms of both redshift and optical luminosity . however , the substantial over - lap between the black - hole mass distributions of the two quasar samples indicated in addition that black - hole mass could not be the sole parameter controlling radio power . in contrast to the studies outlined above , ho ( 2002 ) suggested that there was no clear relationship between radio power and black - hole mass , leading the author to conclude that radio - loud agn could be powered by black holes with a large range of masses ( @xmath6 ) . following their study of the black - hole masses and host - galaxy properties of low redshift radio - loud and radio - quiet quasars , dunlop et al . ( 2003 ) proposed an alternative view of the @xmath7 plane . they argue that the location of both active and non - active galaxies on the @xmath7 plane appears to be consistent with the existence of an upper and lower envelope , both of the approximate form @xmath8 , but separated by some 5 orders of magnitude in radio power . in this scheme the upper and lower envelopes delineate the maximum and minimum radio luminosity capable of being produced by a black hole of a given mass . however , the recent study by oshlack , webster & whiting ( 2002 ; hereafter oww02 ) of the black - hole masses of a sample of flat - spectrum radio - loud quasars from the parkes half - jansky flat spectrum sample of drinkwater et al . ( 1997 ) , casts doubt on the existence of any upper threshold in the @xmath7 plane . oww02 found that their flat - spectrum quasars harbour black - hole masses in the range @xmath9 , and therefore lie well above the upper @xmath8 boundary proposed by dunlop et al . the conclusion reached by oww02 following this result was that previous studies have actively selected against including powerful radio sources with relatively low black - hole masses , due to their concentration on luminous , optically selected radio - loud quasars . here we use the oww02 sample to re - examine the position of these flat - spectrum radio - loud objects on the @xmath7 plane when both doppler boosting effects and the likely geometry of the broad - line region are taken into account . the virial black - hole mass estimate uses the width of the broad hydrogen balmer emission lines to estimate the broad line region ( blr ) velocity dispersion . the black hole mass may then be calculated under the assumption that the velocity of the blr clouds is keplerian : @xmath10 where @xmath11 is the velocity dispersion of the blr clouds , usually estimated from the full - width half maximum ( fwhm ) of the h@xmath12 line , and @xmath13 is the radius of the broad - line region . the measurement of the radius of the blr is ideally achieved by reverberation mapping , in which the continuum and line variations of a number of sources are monitored over a number of years ( e.g. wandel , peterson & malkan 1999 ; kaspi et al . 2000 ) . unfortunately , reverberation mapping of quasars is extremely time consuming and it remains unrealistic that the black - hole masses of a large sample of quasars can be measured in this way . however , the radius of the blr is found to be correlated with the monochromatic agn continuum luminosity at 5100 , @xmath14 ( e.g. kaspi et al . 2000 ) . therefore , this correlation can be exploited to produce a virial black - hole mass estimate from a single spectrum covering the h@xmath12 emission line . here we adopt the calibration of the correlation between @xmath13 and @xmath14 from mclure & jarvis ( 2002 ; see also these proceedings ) , i.e. @xmath15^{(0.61 \pm 0.10)},\ ] ] which , when combined with blr velocity estimate from the @xmath16 fwhm , leads to a black - hole mass estimate given by : @xmath17 flat - spectrum radio samples unavoidably contain a mix of radio source populations including starbursts , giga - hertz peaked spectrum sources and compact steep spectrum sources . however one population is thought to dominate , the doppler boosted sources . these are preferentially selected in high - frequency samples because the superposition of many synchrotron self - absorbed spectra along our line - of - sight results in a flat - spectrum at high - radio frequencies . as the radio emission is propagating along our line - of - sight in these objects , the relativistic velocities associated with powerful radio sources means that face - on radio sources may undergo relativistic beaming which we see as a boost in the flux . from a statistical study of low - frequency and high - frequency selected radio sources , jackson & wall ( 1999 ) have shown that high - frequency selected flat - spectrum sources have an opening angle within @xmath18 of our line - of - sight . we note however that this value is essentially a mean value and that both smaller or larger opening angles are undoubtedly consistent with a doppler boosting paradigm . the opening angles may also depend on the intrinsic radio power of the source ( e.g. jackson & wall 1999 ) , thus the level of doppler boosting in a sample of flat - spectrum sources may have a wide distribution . however , keeping these caveats in mind , we can estimate the amount of doppler boosting the average flat - spectrum sources will exhibit , compared with the doppler boosting of the average quasar , if the maximum opening angle is known for each population . following the method of jarvis & rawlings ( 2000 ) , we take the maximum opening angle for which we observe a radio source as a quasar to be @xmath19 , as derived from the quasar fraction in low - frequency selected samples ( willott et al . 2000 ) . consequently , averaging over solid angle , the mean opening angle of steep spectrum radio - loud quasars is @xmath20 . where @xmath23 is the lorentz factor , @xmath24 and @xmath25 is the angle between the radio jet and the line of sight . therefore , adopting the conservative approach of substituting @xmath26 for the flat - spectrum sources ( many of the sources will have @xmath27 ) and @xmath28 for the steep spectrum quasar population , we find that the doppler boosting factor is @xmath29 . hence , the intrinsic radio luminosity of the flat - spectrum population is of the order @xmath30-times fainter than the intrinsic radio luminosity of the average steep - spectrum radio - loud quasar . in fig . [ fig1 ] we plot radio luminosity versus black - hole mass with both the original radio luminosities , without any boosting correction , and the same objects with the radio luminosity decreased by a factor of 100 . it is clear from fig . [ fig1 ] that following the correction for doppler boosting the vast majority of the flat - spectrum sources now lie within the @xmath31 envelope of the quasar population suggested by dunlop et al . ( 2003 ) . this evidence is in itself enough to account for the major discrepancy between the results of oww02 and previous work . however , in this section have have only applied an average doppler boosting correction factor to the flat - spectrum sample as a whole . obviously this average correction factor will constitute an overestimate , or underestimate , depending on the orientation of each individual object . in the next section we proceed to consider the likely effect upon the estimated black - hole masses of the flat - spectrum quasars due to their inclination close to the line of sight . an indication of when a quasar is ` misaligned ' may also come from the fwhm of the balmer broad lines . the naive assumption is that narrow ( @xmath32 km s@xmath33 ) broad lines imply black holes of lower mass . however , there is a wealth of evidence in the literature which supports the view that the blr has a disk - like geometry , at least for radio - loud sources ( e.g. brotherton 1996 ) . to account for a disk - like geometry we use a low - frequency radio selected quasar survey to predict what the mean fwhm of the broad balmer lines should be , given no spectral - index selection criteria . we use the quasars from the molonglo quasar sample ( mqs ; kapahi et al . 1998 ) for which line - width measurements are available in the literature ( baker et al . 1999 ) . the mean fwhm of the h@xmath12 line in the mqs is @xmath34 km s@xmath33 . in contrast , the mean fwhm of the balmer lines in the oww02 flat - spectrum sample is @xmath35 km s@xmath33 . we therefore choose to adopt a correction factor of two for the flat - spectrum fwhms to compensate for orientation effects . given that @xmath36 , this increases the black - hole mass estimates for the flat - spectrum sample by a factor of four . the predicted position of the flat - spectrum quasars on the @xmath7 plane after application of the inclination correction is shown in fig . [ fig1 ] , from which it can be seen that the flat - spectrum quasars are now even more consistent with the upper and lower radio power envelopes suggested by dunlop et al . ( 2003 ) . we have re - analysed the data of oshlack et al . ( 2002 ) on a sample of flat - spectrum radio - loud quasars . contrary to their conclusions we find that , by correcting for the effects of inclination upon both the radio luminosity and estimated black - hole mass , the black holes harboured by intrinsically powerful flat - spectrum quasars are of comparable mass to those found in other quasars of similar _ intrinsic _ radio luminosity , i.e. @xmath37 m@xmath38 . we also find that although many of the flat - spectrum quasars occupy the region of intrinsic radio luminosity comparable to the frii radio sources found in low - frequency selected radio surveys , some of the sources may occupy the lower - luminosity regime of radio - intermediate and radio - quiet quasars . therefore , we conclude that by consideration of source inclination and intrinsic radio power , flat - spectrum quasars may well be consistent with the @xmath39 relation found in previous studies . baker , j. c. , hunstead , r. w. , kapahi , v. k. , & subrahmanya , c. r. 1999 , apjs , 122 , 29 boroson , t. a. , & green , r. f. 1992 , apjs , 80 , 109 drinkwater , m. j. , webster , r. l. , francis , p. j. , condon , j. j. , ellison , s. l. , jauncey , d. l. , lovell , j. , peterson , b. a. , & savage , a. 1997 , mnras , 284 , 85 dunlop , j. s. , mclure , r. j. , kukula , m. j. , baum , s. a. , odea , c. p. , & hughes , d. h. 2003 , mnras in press ( astro - ph/0108397 ) franceschini , a. , vercellone , s. , & fabian , a. c. 1998 , mnras , 297 , 817 ho , l. c. 2002 , apj , 564 , 120 jackson , c. a. , & wall , j. v. 1999 , mnras , 304 , 160 jarvis , m. j. , & rawlings , s. 2000 , mnras , 319 , 121 jarvis , m. j. , & mclure , r. j. 2002 , mnras , 336 , l38 kapahi , v. k. , athreya , r. m. , subrahmanya , c. r. , baker , j. c. , hunstead , r. w. , mccarthy , p. j. , & van breugel , w. 1998 , apjs , 118 , 327 kaspi , s. , smith , p. s. , netzer , h. , maoz , d. , jannuzi , b. t. , & giveon , u. 2000 , apj , 533 , 677 lacy , m. , laurent - muehleisen , s. a. , ridgway , s. e. , becker , r. h. , & white , r. l. 2001 , apj , 551 , 17 laor , a. 2000 , apj , 543 , l111 mclure , r. j. , & dunlop , j. s. 2002 , mnras , 331 , 795 mclure , r. j. , & jarvis , m. j. 2002 , mnras , 337 , 109	in this proceedings we summarize our recent letter ( jarvis & mclure 2002 ) where we suggested that by correcting for the inevitable effects of inclination , the black - hole masses of flat - spectrum quasars ( fsq ) with intrinsically powerful radio jets are confined , virtually exclusively , to @xmath0 m@xmath1 . after considering realistic doppler boosting factors , many of the fsq would be more accurately classified as radio - intermediate or radio - quiet quasars . this range in radio luminosity suggests that the fsq are fully consistent with an upper boundary on radio power of the form @xmath2 . [ 1996/06/01 ] a&a
99 cullen s and perelstein m 1999 _ phys . rev . lett . _ * 83 * 268 abbott b _ et al . _ [ d0 collaboration ] 2001 _ phys . lett . _ * 86 * 1156 hoyle c d _ _ 2001 _ phys . lett . _ * 86 * 1418 anchordoqui l a , feng j l , goldberg h and shapere a d 2002 _ phys _ d * 65 * 124027 abazov v m _ [ d0 collaboration ] 2003 _ phys . _ * 90 * 251802 anchordoqui l a , feng j l , goldberg h and shapere a d 2003 _ phys . _ d * 68 * 104025 hannestad s and raffelt g g 2003 _ phys . _ d * 67 * , 125008 ( 2003 ) hannestad s and raffelt g g 2004 _ phys . _ d * 69 * , 029901 abazov v m _ [ d0 collaboration ] 2005 _ phys . _ * 95 * 161602	motivated by recent interest in tev - scale gravity and especially by the possibility of fast baryon decay mediated by virtual black holes , we study another dangerous aspect of spacetime foam interactions : lepton flavor violation . we correlate existing limits on gravity - induced decoherence in the neutrino sector with a lower bound on the scale of quantum gravity , and find that if spacetime foam interactions do not allow an @xmath0-matrix description the uv cutoff is well beyond the electroweak scale . this suggests that string theory provides the appropriate framework for description of quantum gravity at the tev - scale . what is the meaning of quantum gravity ? it means that spacetime itself is subject to quantum laws , necessitating inherent fluctuations in the fabric ( metric and topology ) of space and time . these microscopic boiling bubbles force on spacetime a foam - like structure @xcite . a heuristic example pictures spacetime to be filled with tiny virtual black holes that pop in and out of existence on a timescale allowed by heisenberg s uncertainty principle @xcite . these black holes conserve energy , angular momentum , and electric and color charge ( unbroken gauged quantum numbers ) , but they _ are believed _ not to conserve global quantum numbers . if this is the case , the transition between initial and final density matrices associated with black hole formation and evaporation is not factorizable into products of @xmath0-matrix elements and their hermitian conjugates . the evolution of such a quantum system is characterized by a superscattering operator @xmath1 that maps initial pure states to final mixed states , @xmath2 with @xmath3 @xcite . in other words , there _ may be _ a loss of quantum information across the black hole event horizons , providing an environment that can induce decoherence of apparently isolated matter systems . in recent years much attention has been devoted to tev - scale gravity models @xcite , as they provide an economic explanation of the hierarchy between the planck and electroweak mass scales . in the canonical example , the standard model ( sm ) fields are confined to a four dimensional world ( corresponding to our apparent universe ) , while gravity spills into large spatial compact dimensions without conflicting with experimental bounds @xcite . of particular interest is the question whether fast baryon decay can proceed via virtual black hole states in the spacetime foam @xcite . the process is envisioned as the simultaneous absorption of two quarks into the black hole , followed without memory of the initial state by the thermal emission of an antiquark and a lepton , entailing a change in the global baryon and lepton quantum numbers @xmath4 . the probability that two quarks in a proton of size @xmath5 pass within a fundamental planck length , within the heisenberg lifetime uncertainty of the black hole is @xmath6 where @xmath7 is the gravitational uv cutoff . thus , the present limit on the proton lifetime , @xmath8 yr @xcite , implies @xcite @xmath9 interactions through the higher dimensional qqql operator can be prevented if one separates the quark and lepton fields far enough in an extra dimension so that their wave function overlap is exponentially suppressed @xcite . within this picture one may imagine that quark and leptons are fields localized on different branes at the boundaries of a thick wall , but of course gauge and higgs fields are free to propagate inside the wall so that quarks and leptons interact with one another through sm interactions . this division will certainly inhibit fast baryon decay . however , violation of global quantum numbers for processes with spacetime foam black holes on the lepton brane should proceed without wave function suppression . if there is loss of quantum information across event horizons , an interesting quantitative measurement of interaction with the spacetime foam emerges in its effect on neutrino oscillations : interaction with the virtual black holes introduces a decohering process which can be strongly constrained by current observations @xcite . in this work we correlate the existing limits on gravity - induced decoherence with a lower bound on the scale of quantum gravity . we then examine how this bound can shed light on the nature of tev - scale gravity . the sm is based on the gauge group @xmath10 with three fermion generations . a single generation consists of five different representations of the gauge group : @xmath11 @xmath12 @xmath13 @xmath14 @xmath15 where the numbers in parenthesis represent the corresponding charges under @xmath16 the model contains a single higgs boson doublet , @xmath17 whose vacuum expectation value breaks the gauge symmetry @xmath18 into @xmath19 an important feature of the sm , which is relevant to spacetime foam interactions discussed here , is the fact that the sm also comprises an accidental global symmetry , @xmath20 where @xmath21 is the baryon number symmetry , and @xmath22 are three lepton flavor symmetries , with total lepton number given by @xmath23 . it is an accidental symmetry because we do not impose it . it is a consequence of the gauge symmetries and the low energy particle content . it is possible ( but not necessary ) , however , that effective interaction operators induced by the high energy content of the underlying theory may violate sectors of the global symmetry . this violation is already present in the absence of gravity , as has been evidenced by the discovery of neutrino flavor oscillations . if there are additional contributions of this kind from the gravity sector , then black hole - mediated interactions could violate @xmath24 in a distinct manner , as mentioned above and reviewed in what follows . measurements of flavor transformations in a neutrino beam can provide a clean and sensitive probe of interactions with the spacetime foam . without interference from the gravitational sector , oscillations in the neutrino sector provide a pure quantum phenomenon in which the density matrix has the properties of a projection operator , tr @xmath25 tr @xmath26 because black holes do conserve energy , angular momentum ( helicity ) , color and electric charge , any neutrino interacting with the virtual black holes needs to re - emerge as a neutrino . as an example , if spacetime foam black holes do not conserved @xmath27 neutrino flavor is randomized by interactions with the virtual black holes . the result of many interactions is to equally populate all three possible flavors . the ensuing discussion will be framed in the context of @xmath28 oscillations , and we will comment on other channels after presenting our results . consider two neutrino states @xmath29 and @xmath30 with masses @xmath31 and @xmath32 , and two flavor states @xmath33 and @xmath34 , where @xmath35 is the neutrino mixing angle , @xmath36 , @xmath37 , and @xmath38 denotes the transpose . in the absence of spacetime foam interactions the time evolution equation for the density matrix , @xmath39\ , \label{liouville}\ ] ] is governed ( in the mass basis ) by the hamiltonian @xmath40 , where @xmath41 , @xmath42 , and @xmath43 is the neutrino energy ( in natural units ) . the solution @xmath44 of eq . ( [ liouville ] ) , with initial conditions @xmath45 , gives the @xmath46 survival probability after propagation of a distance @xmath47 , @xmath48 = 1- s^2_{2\theta}\,\,(1-\cos kx)/2\ . \label{prob}\ ] ] here , @xmath49 is the @xmath46 state projector . transitions from pure to mixed states become possible by addition of a dissipative term @xmath50 $ ] into eq . ( [ liouville ] ) @xcite @xmath39 - { \cal d}[\rho]\ . \label{liouvilled}\ ] ] energy conservation , complete positivity ( assuring the absence of unphysical effects , such as negative probabilities , when dealing with correlated systems ) , and monotonic increase in the von - neumann entropy eventually lead to a modification of eq . ( [ prob ] ) @xcite , @xmath51 here , @xmath52 has dimension of energy , and its inverse defines the typical ( coherence ) length after which the system gets mixed . thus , for @xmath53 one expects significant deviations from eq . ( [ prob ] ) . we are interested in dissipative scenarios where decoherence effects vanish in the weak gravitational limit , @xmath54 and thus we take @xmath55 where @xmath56 is a dimensionless parameter which by naturalness is expected to be @xmath57 and @xmath58 atmospheric neutrinos are generated in the decay of pions and kaons resulting from cosmic ray collisions . most of them are produced in a spherical surface at about 10 - 20 km above ground level and they proceed towards the earth . production of muon neutrinos is dominated by the process @xmath59 ( and its charged conjugate ) . at energies @xmath60 gev the muons decay before reaching the surface of the earth , @xmath61 this decay chain then leads to a flavor ratio @xmath62 the zenith angle distribution observed by super - kamiokande shows a clear deficit of upward - going muon neutrinos , which is well explained by two flavor @xmath28 oscillations , with @xmath63 ev@xmath64 and @xmath65 @xcite . a best fit to data collected by the super - kamiokande atmospheric neutrino experiment @xcite , allowing for both oscillation and decoherence yields , for @xmath66 @xmath67 at the 90% cl @xcite . for larger @xmath68 eq.([tkappa ] ) generalizes to @xcite @xmath69 indeed , this represents a conservative bound . this can be seen from eq . ( [ osde ] ) : the analysis of data places bounds on @xmath70 and the neutrino energies are well above 1 gev . from eq . ( [ tkappan ] ) it is straightforward to see that , for @xmath71 _ the lower limit on the uv cutoff is well beyond the electroweak scale . _ we now comment on other possible oscillation channels . the ccfr detector at fermilab is sensitive to @xmath72 @xcite and @xmath73 @xcite flavor transitions . neutrino energies range from 30 to 600 gev with a mean of 140 gev , and their flight lengths vary from 0.9 to 1.4 km . a best fit to the data allowing for both oscillation and decoherence yields @xmath74 at the 99% cl @xcite . moreover , similar constraints on gravity - induced decoherence arise in the quark sector . the persistence of coherence in @xmath75 oscillations and neutron interferometry leads to @xcite @xmath76 the typical energy in these systems is @xmath77 in the near future , the possible observation of galactic anti - neutrino beams @xcite at the icecube facility may provide a major improvement in sensitivity to spacetime foam interactions @xcite . in summary , using an existing analysis of atmospheric neutrino data we have shown that if tev - scale gravity is realized in nature , interactions with virtual black holes are non - dissipative and there is therefore no loss of information . non - dissipative interactions are expected when gravity is embedded in string theory , so that an @xmath0-matrix description is possible . the existence of an @xmath0-matrix makes it no longer automatic that , _ e.g. , _ the @xmath78-violating @xmath79 and @xmath78-conserving @xmath80 outgoing channels have the same probability , as they would in thermal evaporation . thus , the problem of avoiding rapid baryon decay or large majorana neutrino masses is shifted to the examination of symmetries @xcite in the underlying string theory which would suppress the appropriate non - renormalizable operators at low energies . i would like to thank luis epele , jonathan feng , haim goldberg , concha gonzalez garcia , francis halzen , dan hooper , subir sarkar , al shapere , and tom weiler for valuable discussions .
dna methylation , cpg island , shore , and shelf , random forest classifier , dnase i hypersensitive sites , transcription factor binding sites , ewas epigenetics is the study of changes in gene expression or complex phenotype that are not associated with changes in dna sequence and but inherited through cell division . epigenetic markers often change within an individual over time and are cell type specific @xcite . epigenetics has been shown to play a critical role in cell differentiation , development , and tumorigenesis @xcite . dna methylation is probably the best studied epigenetic modification of dna , but our understanding of dna methylation is still in its infancy . in vertebrates , dna methylation occurs by adding a methyl group to the fifth carbon of the cytosine residue , mainly in the context of neighboring cytosine and guanine nucleotides in the genome ( 5-cg-3 dinucleotides or _ cpg sites _ ) mediated by dna methyl - transferases ( dnmts ) @xcite . dna methylation has been shown to play an important functional role in the cell , including involvement in dna replication and gene transcription , with substantial downstream association with development , aging , and cancer @xcite . cpg sites are underrepresented in the human genome relative to their expected frequency as a result of being a _ mutation hotspot _ , where the deamination of methylated cytosines often changes cpg sites into tpg sites @xcite . although cpg sites are mainly methylated across the mammalian genome @xcite , there are distinct , mostly unmethylated cg - rich regions termed cpg islands ( cgis ) that have a g+c content greater than @xmath2 @xcite . cgis account for @xmath3 of the genome and are often located in promoters and exonic regions in mammalian genomes @xcite . methylation patterns in cgis that are in promoter regions , where most previous studies have focused attention , have recently been shown to differ from methylation patterns elsewhere , indicating a specific biological role for these promoter cgis @xcite . cgis have been shown to co - localize with dna regulatory elements such as transcription factor binding sites ( tfbss ) @xcite and dna binding insulator proteins , such as ctcf , which protect downstream dna from upstream methylation activities @xcite . across the genome , dna methylation levels have been shown to be associated with gene regions , active chromatin marks @xcite , cis - acting dna regulatory elements , and proximal sequence elements @xcite , giving hints about the processes that regulate methylation and how methylation may in turn impact cellular phenotypes . the non - uniform distribution of cpg sites across the human genome and the important role of methylation in cellular processes imply that characterizing genome - wide dna methylation patterns is necessary to better understand the regulatory mechanisms of this epigenetic phenomenon @xcite . recent advances in methylation - specific microarray and sequencing technologies have enabled the assay of dna methylation patterns genome - wide and at single base - pair resolution @xcite . the current gold standard to quantify single site dna methylation levels across an individual s genome is whole genome bisulfite sequencing ( wgbs ) , which quantifies dna methylation levels at @xmath4 million ( out of @xmath5 million total ) cpg sites in the human genome @xcite . however , wgbs is prohibitively expensive for most current studies , is subject to conversion bias , and is difficult to perform in particular genomic regions @xcite . other sequencing methods include methylated dna immunoprecipitation ( medip ) sequencing , which is experimentally difficult and expensive , and reduced representation bisulfite sequencing ( rrbs ) , which assays cpg sites in small regions of the genome @xcite . as an alternative , methylation microarrays , and the illumina humanmethylation 450k beadchip in particular , measure bisulphite treated dna methylation levels at @xmath6 preselected cpg sites genome - wide @xcite ; however , these arrays assay less than @xmath7 of cpg sites , and this percentage is biased to gene regions and cgis . quantitative methods are needed to predict methylation status at unassayed sites and genomic regions . in this study , we examined measurements of methylation levels in @xmath8 individuals using the illumina 450k beadchip @xcite . within these methylation profiles , we examined the patterns and correlation structure of the cpg sites , with attention to characterizing methylation patterns in cgi regions . using features that included neighboring cpg site methylation status , genomic location , local genomic features , and co - localized regulatory elements , we developed a random forest classifier to predict single cpg site methylation levels . using this model , we were able to identify dna regulatory elements that may interact with dna methylation at specific cpg sites , providing hypotheses for experimental studies on mechanisms by which methylation is regulated or leads to biological changes or disease phenotypes . methylation status is a delicate epigenomic feature to characterize , and even more challenging to predict , because assayed dna methylation marks are ( a ) an average across the sampled cells , ( b ) cell type specific , ( c ) environmentally unstable , and ( d ) not well - correlated within a genomic locus @xcite . it is not clear whether either predicted or measured methylation status at specific cpg sites generalize well across platforms , cell types , individuals , or genomic regions @xcite . a number of methods to predict methylation status have been developed ( table s1 ) . most of these methods assume that methylation status is encoded as a binary variable , e.g. , a cpg site is either methylated or unmethylated in an individual @xcite . these methods have often limited predictions to specific regions of the genome , such as cgis @xcite . more broadly , all of these methods make predictions of average methylation status for windows of the genome instead of individual cpg sites . our model is an attempt to develop a general method for predicting methylation levels at individual cpg sites by removing the simplifying assumptions of current methods without sacrificing prediction accuracy . the relative success of these methods depends heavily on the prediction objectives . all of the studies that achieved prediction accuracy @xmath9 @xcite predicted average methylation status within cgis or dna fragments within cgis . most of the cpg sites in cgis are unmethylated across the genome @xcite for example , @xmath10 of cpg sites in cgis in cells from human brain were found to be methylated using a wgbs approach @xcite so it is not surprising that classifiers limited to these regions perform well . studies not limiting prediction to cgis uniformly achieved lower accuracies , ranging from @xmath11 to @xmath12 . many of these methods predicted binary methylation status ; one study used categorical methylation status @xcite . another study predicted average methylation levels as a continuous variable @xcite , and the predictions achieved a maximum correlation coefficient of @xmath13 and a root mean square error ( rmse ) of @xmath14 ; however , the study was limited to @xmath15 dna fragments instead of a genome - wide analysis . across these methods , features that are used for dna methylation prediction include : dna composition ( proximal dna sequence patterns ) , predicted dna structure ( e.g. , co - localized introns ) , repeat elements , tfbss , evolutionary conservation ( e.g. , _ phastcons _ @xcite ) , number of snps , gc content , alu elements , histone modification marks , and functional annotations of nearby genes . several studies only using dna composition features achieved prediction accuracies ranging from @xmath11 to @xmath16 @xcite . et al . _ used @xmath17 features including dna composition , dna structure , repeat elements , tfbss , evolutionary conservation , and number of snps ; zheng _ et al . _ included @xmath18 features including dna composition , dna structure , tfbss , histone modification marks , and functional annotations of nearby genes . two studies did not use any dna composition features @xcite . the relative contribution of each feature to prediction quality is not quantified well across these studies because of the different methods and prediction objectives . the majority of these methods are based on support vector machine ( svm ) classifiers @xcite . general non - additive interactions between features are not encoded when using linear kernels , as most of these svm - based classifiers used . if a more sophisticated kernel is used , such as a radial basis function kernel ( rbf ) , within the svm - based approach , the contribution of each feature to the prediction quality is not readily available . three studies included alternative classification frameworks : one found that a decision tree classifier achieved better performance than an svm - based classifier @xcite . another study found that a naive bayes classifier achieved the best prediction performance @xcite . a third study used a word composition - based encoding method @xcite . in our study , we choose to use a random forest classifier because it encodes non - additive interactions between features and because the relative contribution of each feature to prediction quality is quantifiable . our method for predicting dna methylation levels at cpg sites genome - wide differs from these current state - of - the - art classifiers in that it : ( a ) uses a genome - wide approach , ( b ) makes predictions at single cpg site resolution , ( c ) is based on a random forest classifier , ( d ) predicts methylation levels instead of methylation status , ( e ) incorporates a diverse set of predictive features , including regulatory marks from the encode project , and ( f ) allows the quantification of the contribution of each feature to prediction . we find that these differences substantially improve the performance of the classifier and also provide testable biological insights into how methylation regulates , or is regulated by , specific genomic and epigenomic processes . dna methylation profiles were measured in whole blood samples from @xmath8 unrelated individuals by illumina humanmethylation 450k beadchips at single cpg site resolution for @xmath19 cpg sites @xcite . single cpg site methylation levels are quantified by @xmath20 , the ratio of the methylated probe intensity and the sum of the methylated and unmethylated probe intensities , which ranges from 0 ( unmethylated ) to 1 ( methylated ) . within the single - site @xmath20 values across individuals , we controlled for probe chip position , sample age , and sample sex . after these data were preprocessed ( see materials and methods ) , @xmath21 cpg sites remained across the @xmath22 autosomal chromosomes . first we examined the distribution of dna methylation levels , @xmath20 , at cpg sites on autosomal chromosomes across all @xmath8 individuals ( figure 1a ) . the majority of cpg sites were either hypermethylated or hypomethylated , with @xmath23 of sites with @xmath24 and @xmath25 of sites with @xmath26 . using a cutoff of @xmath27 , across the methylation profiles and individuals , @xmath28 of these cpg sites have a _ methylated status _ ( @xmath29 ) . across the individuals , we observed distinct patterns of dna methylation levels in different genomic regions ( figure 1b ) . using cgis labeled in the ucsc genome browser @xcite , we defined _ cgi shores _ as regions @xmath30 kb away from cgis in both directions and _ cgi shelves _ as regions @xmath31 kb away from cgis in both directions @xcite . we found that cpg sites in cgis were hypomethylated ( @xmath32 of sites with @xmath26 ) and sites in non - cgis were hypermethylated ( @xmath33 of sites with @xmath24 ) , while cpg sites in cgi shore regions had variable methylation levels following a u - shape distribution ( @xmath34 of sites with @xmath24 and @xmath35 of sites with @xmath26 ) , and cpg sites in cgi shelf regions were hypermethylated ( @xmath36 of sites with @xmath24 ) . these distinct patterns reflect highly context - specific dna methylation levels genome - wide . individuals at cpg sites assayed on the 450k array . panel a : distribution of dna methylation values across all cpg sites . panel b : distribution of dna methylation values for cpg sites within cgis , cgi shores , cgi shelves , and non - cgi regions.,width=384 ] dna methylation levels at nearby cpg sites have previously been found to be correlated ( indicating possible co - methylation ) , particularly when cpg sites are within @xmath37 kb from each other @xcite ; these patterns are in contrast with the correlation among nearby genotypes due to linkage disequilibrium ( ld ) that often extends to large genomic regions from a few kilobases to @xmath38 mb @xcite . we quantified the correlation of methylation levels ( @xmath20 ) between neighboring pairs of cpg sites using the absolute value pearson s correlation across individuals . we found that correlation of methylation levels between neighboring ( or adjacent on the array ) cpg sites decreased rapidly to approximately @xmath39 within @xmath15 bp , in contrast to sharp decays noted within @xmath40 kb in previous studies with sparser cpg coverage ( figure 2a ) @xcite . we found the rate of decay in correlation to be highly dependent on genomic context ; for example , for neighboring cpg sites in the same cgi shore and shelf region , correlation decreases continuously until it is well below what is expected ( figure 2a ) . because of the over - representation of cpg sites near cgis on the array , an increase in correlation can be observed as neighboring sites extend past the cgi shelf regions , where there is lower correlation with cgi methylation levels than we observe in the background . bp , which is the length of the cgi shore & shelf regions . the solid horizontal line represents the background ( absolute value correlation or mse ) levels averaged from @xmath41 pairs of cpg sites from arbitrary chromosomes . panel a : the absolute value of the correlation between neighboring sites across all individuals ( y - axis ) . the lines represent cubic smoothing splines fitted to the correlation data . panel b : the mse was calculated for cpg sites ( y - axis ) for each pair of cpg sites within the genomic distance window.,width=384 ] to make this decay more precise , we contrasted the observed decay to the level of _ background correlation _ ( @xmath42 ) , which is the average absolute value correlation between the methylation levels of pairs of randomly selected cpg sites across chromosomes ( figure 2a , figure 3 ) . we found substantial differences in correlation between neighboring cpg sites versus arbitrary pairs of cpg sites at identical distances , presumably because of the dense cpg tiling on the 450k array within cgi regions . interestingly , the slope of the correlation decay plateaus after the cpg sites are approximately @xmath43 bp apart ( both for neighbors and for arbitrary pairs of a specified distance ) , but the distribution of correlation between pairs of cpg sites is not substantially different from the distribution of background correlation even within @xmath44 kb ( figure s1a ) . while this certainly suggests that there are may be patterns of methylation that extend to large genomic regions , the pattern of extreme decay within approximately @xmath43 bp across the genome indicates that , in general , methylation may be biologically manipulated within very small genomic regions . thus , neighboring cpg sites may only be useful for prediction when the sites are sampled at sufficiently high densities across the genome . kb ( panel a ) , @xmath45 mb ( panel c ) and @xmath46 mb ( panel e ) ; the right column plots show the mse patterns of cpg sites in relation to their genomic distances with distance range @xmath8 kb ( panel b ) , @xmath45 mb ( panel d ) and @xmath46 mb ( panel f ) . the solid horizontal lines represent the background correlation or mse level calculated from @xmath47 pairs of cpg sites from a different chromosome.,width=384 ] we repeated these experiments using mean squared error ( mse ) between cpg site levels to quantify patterns of decay of methylation within each individual , instead of across individuals as is measured with the correlation analyses ( see materials and methods ; figure 2b , figure 3 ) . in general , the mse trends echo the local patterns seen in the correlation analysis and also appear to be region specific . in cgi regions , the mse of neighboring sites was low and increased slowly with genomic distance . in contrast , mse in cgi shore and shelf regions increased rapidly to an mse higher than background mse ( @xmath48 ) , indicating that the edges of a single shore and shelf region are less predictive of each other than any two cpg sites at random . the individual - specific mse between neighboring sites ( figure s1b ) shows a much higher deviation from background distribution of mse at @xmath44 kb relative to correlation , indicating that the biological manipulation of methylation in larger genomic regions may be individual specific , such as being driven by genetic variants or environmental effects . as we observed that methylation patterns at neighboring cpg sites depended heavily on genomic content , we further investigated methylation patterns within cgis , cgi shores , and cgi shelves . methylation levels at cgis and cgi shelves were fairly constant genome - wide and across individuals cgis are hypomethylated and cgi shelves are hypermethylated but cgi shores exhibit a reproducible but drastic pattern of change ( figure 4a ) . cpg sites in cgi shores have a monotone increasing pattern of methylation status from cgis towards cgi shelves , and this pattern is symmetric in the cgi shores upstream and downstream of cgis . if we examine the mse between pairs of cpg sites methylation status in these regions , we find that mse within the cgi and within the cgi shelves is low , consistent with the variance we observed within dna methylation profiles in these regions ( figure 4b ) . additionally , we find that the mse between the cpg sites in the shelves appears to increase as the sites are further away from the cgi on the shelf , suggesting a circular dependency in methylation status between the ends of the shelf sequences . it is interesting that the cpg sites in the shore regions appear substantially more predictive of cpg sites in the shelf regions than those in the cgi regions , although this may indicate a less precise delineation of the shore and shelf regions relative to the cgi and cgi shore delineation . equal sized windows . panel a : the points represents the mean @xmath20 in cgis , cgi shores or cgi shelves across all sites in all individuals with a window size of @xmath8 bp . panel b : methylation levels of each cpg site in each cgi , cgi shore , and cgi shelf were compared with all the other sites in the same cgi region . x - axis and y - axis represent the genomic position of each cgi with a scale of 1:100 , i.e. one unit in matrix represents @xmath8 bp distance . the mse of each unit cell was calculated for all pairwise cpg sites with one site located in the relative scaled position on x - axis and the other one on y - axis , and then averaged over 100 individuals.,width=576 ] to quantify the amount of variation in dna methylation explained by genomic context , we considered the correlation between genomic context and principle components ( pcs ) of methylation levels across all @xmath8 individuals ( figure 5 ) . we found that many of the features derived from the cpg site s genomic context appear to be correlated with the first principal component ( pc1 ) . methylation statuses of upstream and downstream neighboring cpg sites and a co - localized dnase i hypersensitive ( dhs ) site are the most highly correlated features , both with pearson s correlation around @xmath49 ( figures 5 ) . ten genomic features all have correlation @xmath50 with pc1 , including co - localized active tfbss elf1 ( ets - related transcription factor 1 ) , maz ( myc - associated zinc finger protein ) , mxi1 ( max - interacting protein 1 ) and runx3 ( runt - related transcription factor 3 ) , suggesting that they may be useful in predicting dna methylation status ( figure s2 ) . that said , the features themselves are well correlated ; for example , active tfbs are highly enriched within dhs sites ( correlation @xmath51 ) @xcite . cpg sites . the correlation with each of the features was calculated between the first ten methylation pcs ( y - axis ) and all features ( x - axis).,width=672 ] these observations about patterns of dna methylation suggest that correlation in dna methylation is local and dependent on genomic context . thus , prediction of dna methylation status based only on methylation levels at neighboring cpg sites may not perform well , especially in sparsely assayed regions of the genome . using prediction features including neighboring cpg site methylation levels and a number of features characterizing genomic context , we built a classifier to predict dna methylation status , where status indicates no methylation ( 0 ) or complete methylation ( 1 ) at a cpg site . there were @xmath52 sites with neighboring cpg sites at arbitrary distances used in these analysis per individual . the @xmath53 features that we used for dna methylation status prediction fall into four different classes ( table s4 ) : * _ neighbors _ : one upstream and one downstream neighboring ( cpg sites assayed on the array and adjacent in the genome ) cpg sites genomic distances , binary methylation status , and @xmath20 values , * _ genomic position _ : binary values indicating co - localization with dna sequence annotations , including promoters , gene body , intergenic region , cgis , cgi shores and shelves , and snps ( in site probes ) ; * _ dna sequence properties _ : continuous values representing the local recombination rate from hapmap @xcite , gc content from encode @xcite , integrated haplotype scores ( ihs ) @xcite , and genomic evolutionary rate profiling ( gerp ) calls @xcite . * _ cres _ : binary values indicating co - localization with cis - regulatory elements ( cres ) , including dhs sites ( assayed in the gm12878 cell line , the closest match to whole blood ) and @xmath54 specific tfbss ( assayed in the gm12878 cell line ) @xcite . we used a random forest ( rf ) , which is an ensemble classifier that builds a collection of decision trees and combines the predictions across all of the trees to create a single prediction . the output from the random forest is the proportion of trees in the fitted forest that classify the test sample as a 1 . we threshold this output , @xmath55 $ ] , to @xmath56 using a cutoff of @xmath27 to find the predicted methylation status . we quantified generalization error for each feature set using a modified version of repeated random subsampling , where the training and test sets are within the same individual ( see materials and methods ) . we used prediction accuracy , specificity , sensitivity , and area under the receiver operating characteristic ( roc ) curve ( auc ) to evaluate our predictions ( see materials and methods ) . using all @xmath53 features and not restricting neighboring site distance , we achieved an accuracy of @xmath57 and an auc of @xmath58 . we considered the role of each subset of features ( table 1 ) . for example , if we only include _ genomic position _ features , the classifier had an accuracy of @xmath59 and auc of @xmath60 . when we included all classes of features except for _ neighbors _ , the classifier achieved an accuracy of @xmath61 and an auc of @xmath62 , supporting substantial improvement in prediction from considering the methylation status of neighboring cpg sites ( t - test @xmath63 ) . however , we also found that the additional features improve prediction substantially over just using the _ neighbors _ features , which has an accuracy of @xmath64 and an auc of @xmath65 ( t - test @xmath66 ) . because the correlation between neighboring cpg sites methylation levels decays rapidly with distance , we considered restricting the window size for neighboring cpg sites to @xmath45 kb , @xmath67 kb , and arbitrary distances on the same chromosome : predictions at cpg sites where one or both of the neighboring cpg sites were greater than this distance were excluded . there were @xmath68 sites ( @xmath69 ) with both upstream and downstream neighboring cpg sites within @xmath67 kb and @xmath70 sites ( @xmath2 ) with both neighboring sites within @xmath45 kb . we found that when we restricted prediction to @xmath67 kb and @xmath45 kb windows , the classifier using only the _ neighbors _ features had accuracies of @xmath71 and @xmath72 , and aucs of @xmath58 and @xmath73 , respectively ( table 1 ) . this result is consistent with the high correlation of neighboring cpg sites within @xmath45 kb . when we used all of the prediction features and restricted neighboring cpg sites to @xmath45 kb , the classifier achieved @xmath74 accuracy and an auc of @xmath75 . interestingly , while prediction accuracy , auc , and specificity all improved when restricting neighboring sites to shorter genomic windows , sensitivity , or @xmath76 , was increasingly worse , indicating that the statistical gain in restricting site distance is in eliminating type i errors ( false positives ) rather than in predicting a higher proportion of methylated sites correctly ( true positives ; table 1 ) . to determine how predictive methylation profiles were across individuals , we quantified the generalization error of our classifier genome - wide across individuals . in particular , we trained our classifier on 10,000 sites from one individual , and predicted cpg sites from the other @xmath77 individuals for all feature sets . the classifier performance was highly consistent across genomic regions and individuals regardless of the subset of cpg sites in the training set ( figure 6a and figure s3 ) . we found that the auc increased when restricting the distance of neighboring cpg sites , repeating earlier trends ( t - test between arbitrary distances and @xmath45 kb : @xmath78 ) . the color represents the density of the distribution at each point . the line is fitted by linear regression . panel d : 2d histogram of predicted methylation levels versus experimental methylation @xmath20 values with restricted to 1 kb neighboring cpg sites . the color represents the density of the distribution at each point.,width=576 ] to test the sensitivity of our classifier to the number of cpg sites in the training set , we investigated the prediction performance for different training set sizes . we found that training sets with greater than @xmath79 cpg sites had fairly similar performance ( figure s4 ) . throughout these experiments , we used a training set size of @xmath41 , in order to strike a balance between large numbers of training samples and computational tractability . we compared the prediction performance of our random forest classifier with an svm classifier with a radial basis function ( rbf ) kernel . we validated by using repeated random resampling , with the same training and test sets for both classifiers . svm classifiers had an accuracy of @xmath80 including all features and neighbors at an arbitrary distance ; the accuracy was @xmath81 when neighboring sites were restricted to @xmath67 kb window and @xmath82 when restricting to @xmath45 kb window ( table s5 ) . we found that the random forest classifier had better prediction accuracy than the svm classifier without restricting neighboring cpg site distance ( @xmath83 ; table s5 ) . when we restricted neighboring cpg sites to within @xmath45 kb , random forest and svm classifiers had similar prediction performance . cpg methylation levels @xmath20 in a dna sample represent the average methylation status across the cells in that sample and will vary continuously between 0 and 1 ( figure s5 ) . since the illumina 450k array enables precise methylation levels at cpg site resolution in each sample , we used our classifier to predict methylation levels at single cpg site resolution . we compared the prediction probability ( @xmath84 $ ] ) from our random forest classifier ( without thresholding ) with methylation levels ( @xmath85 $ ] ) from the array , and validated this approach using repeated random subsampling to quantify generalization accuracy ( see materials and methods ) . using all @xmath53 features used in methylation status prediction , but modifying the _ neighbors _ neighboring cpg site methylation status values to be continuous methylation levels @xmath20 , we trained our random forest classifier and evaluated the correlation coefficient ( @xmath86 ) and root - mean squared error ( rmse ) between experimental and predicted methylation values ( table 2 ; figure 6c and 6d ) . we found that the experimental and predicted methylation values had an @xmath87 , which increased to @xmath62 and @xmath65 when restricting neighboring cpg sites to within @xmath67 kb and @xmath45 kb , respectively . the rmse between experimental and predicted methylation values for the unrestricted experiment was @xmath88 , decreasing to @xmath89 and @xmath90 for the @xmath67 kb and @xmath45 kb restrictions . the correlation coefficient and the rmse indicate good recapitulation of experimental values using predicted methylation values . .performance of methylation level predictions using the random forest classifier . the rows represent distance restrictions on the pairs of neighboring sites ( _ arbitrary _ for none ) . r represents the correlation and rmse represents the root mean squared error of the predicted and actual methylation levels in the cpg sites . [ cols="<,<,<",options="header " , ] studies of dna methylation have focused on methylation within cgis at promoter regions , restricting predictions to cgi regions @xcite ; we and others have shown dna methylation has different patterns in different genomic regions @xcite . here we investigated regional dna methylation prediction for our genome - wide cpg site methods , restricted to cpgs within gene coding regions and cgis ( table s6 ) . for this experiment , prediction was performed on cpg sites with neighboring sites within @xmath45 kb distance because of the limited size of cgis . we found , within cgi regions , predictions of methylation status using our method had an accuracy of @xmath91 and an auc of @xmath92 . methylation level prediction within cgis achieved a correlation coefficient of @xmath65 and a rmse of @xmath93 . as with the related work on prediction within cgi regions , we believe the improvement in accuracy is due to the limited variability in methylation patterns in these regions ; indeed , @xmath94 of cpg sites in cgi regions have a @xmath95 ( table s6 ) . conversely , prediction of cpg methylation status within cgi shores had an accuracy of @xmath96 . this lower accuracy is consistent with observations of robust and drastic change in methylation status across these regions @xcite . prediction performance within various gene regions was fairly consistent , with @xmath97 accuracy for predictions of cpg sites in promoter regions , @xmath98 accuracy for predictions within gene body regions ( exons and introns ) , and @xmath99 accuracy within intergenic regions ( figure 6b ) . this pattern may in part reflect the biased density of cpg sites on the illumina 450k array . we evaluated the contributions of each feature to the overall prediction accuracy , as quantified by the gini index . the _ gini index _ measures the decrease in _ node impurity _ , or the relative entropy of the observed positive and negative examples before and after splitting the training samples on a single feature , of a given feature over all trees in the trained random forest . we computed the gini index for each of the @xmath53 features from the fitted rf for predicting methylation status . we found that upstream and downstream neighboring cpg site methylation status are the most important features ( table s2 , figure 7a ) . when we restrict the neighboring cpg sites to be within @xmath67 kb or @xmath45 kb , the gini score of the neighboring site status features increased in relation to the other features , echoing our observation that the non-_neighbor _ feature sets are less useful when a cpg site s neighbors are nearby , and thus more informative . in contrast , the gini score of the genomic distance to the neighboring cpg site feature decreased when the neighboring site distance was restricted , suggesting that neighboring genomic distance an important feature to consider when some neighbors may be more distant and correspondingly less predictive . we found that dhs sites are strongly predictive of an unmethylated cpg site ; the dhs site feature has the third most significant gini index across these experiments . this observation is consistent with a previous study showing that cpg sites in dhs sites tend to be unmethylated @xcite . cgi status is also an important feature , which is unsurprising given that most cpg sites in cgis are unmethylated . gc content , which also ranked highly based on gini index , may have a substantial contribution to prediction as a proxy for other important features , such as cgi status and cpg density . kb and @xmath45 kb . panel a : gini scores for the top 10 features . upmethy : upstream cpg site methylation status ; downmethy : downstream cpg site methylation status ; dhs sites : dnase i hypersensitive sites ; updist : genomic distance with upstream cpg site ; downdist : genomic distance with downstream cpg site ; cgi : cgi status ; gc_cont : gc content , maz , elf1 , mxi1 : tfbss of maz , elf1 and mxi1 . panel b : gini scores for the top 10 tfbs.,width=384 ] several tfs were among the most highly ranked features across experiments ( figure 7b ) , some of which are known to be associated with dna methylation , including elf1 , runx3 , maz , mxi1 , and max . indeed , the ets - related transcription factor ( elf1 ) has been shown to be overrepresented in methylated regions , associating dna methylation with hemapotopoiesis in hematopoietic stem cells @xcite . runx3 ( runt - related transcription factor 3 ) , a strong tumor suppressor associated with diverse tumor types , has been suggested to be associated with cancer development through regulating global dna methylation levels @xcite . runx3 expression is associated with aberrant dna methylation in adenocarcinoma cells @xcite , primary bladder tumor cells @xcite , and breast cancer cells @xcite . for another tumor suppressor transcription factor00 , mxi1 ( max - interacting protein 1 ) , expression levels ( specifically , lack of expression ) have been reported to be associated with promoter methylation levels and neuroblastic tumorigenesis @xcite . it has been suggested that suppression of maz ( myc - associated zinc finger protein ) may be associated with dna methyltransferase i , the key factor for _ de novo _ dna methylation @xcite . mxi1 and max ( myc - associated factor x ) both interact with c - myc ( myelocytomatosis oncogene ) , a well characterized oncogene , which has been shown to be methylation sensitive , meaning that the tf motifs contain cpg sites and , thus , tf binding is sensitive to methylation status at those sites @xcite . this suggests a potential regulatory relationship between max , mxi1 , and dna methylation that may extend to downstream cancer tumor development . we found that the feature rankings based on gini scores differed only slightly across experiments ( figure 7a and 7b ) , indicating the robustness of the gini index to quantify contribution to prediction accuracy . we found that the correlation between a binary feature and pc1 is proportional to the gini index of that feature ( figure 5 and table s2 ) . the variation in the gini index rankings for tfbss varied more than we expected based on the other features ( figure s6 ) . tfbss that co - occur with cpg sites more often tend to be more important for prediction , according to the gini index score . we found that the gini index of a binary feature has a log linear relationship with the number of co - occurrences of that binary feature with cpg sites in the data set : the more often a cpg site in the training set co - occurred with a tfbs , the higher the gini index rank of that cpg site . there were several outliers to this trend , including upstream and downstream cpg site methylation status , status of promoter , gene body , and non cgi regions , and three tfbss ( pol3 ( rna polymerase iii ) , rxr-@xmath100 ( retinoid x receptor alpha ) and c - fos ( a proto - oncogene ) ) . these features were less important than we would predict using the fitted linear regression model of log gini importance . this trend limits the strong conclusions that associate specific tfbss with dna methylation biochemically from a high gini index rank for that tfbs ; it may be that there are general relationships between tfbs and cpg sites that we are learning , but a relatively high tfbs frequency in these data will artificially inflate the rank of that tfbs in comparison to the others ( figure s6 ) . most cpg sites within tfbss have low average methylation levels ( table s3 ) . several tfbss have disproportionately high average methylation values : for example znf274 ( zinc - finger protein 274 ) and jund ( jun d proto - oncogene ) ; however , both of these outliers also have a low co - occurrence frequency with cpg sites in these data , suggesting that this finding may be an artifact . we characterized genome - wide and region - specific patterns of dna methylation . these region - specific patterns raise additional questions , including how these observations may resolve or at least suggest causal relationships between methylation and other genomic and epigenomic processes . with single nucleotide polymorphism ( snp ) associations with complex traits , it is likely that the genotype drives associated processes rather than the other way around ; the causal relationship is established by inductive logic , since it is biologically difficult to perform site - specific mutation . the dynamic nature of cpg site methylation means that no such causal relationship can be established inductively ; however , experiments can be designed to establish the impact of changing the methylation status of a cpg site @xcite . conditional analyses , such as those developed for dna , may prove to be illuminating for epigenomics @xcite , but the current data are still murky . for example , does a tfbs containing a cpg site prevent methylation when a transcription factor is actively bound , or does a methylated cpg site in a tfbs prevent a tf from binding to that site ? we built a random forest predictor of dna methylation levels at cpg site resolution . in our comparison between random forest classifiers and svm classifiers , we found that improvements of the random forest classifier include i ) better prediction for more sparsely sampled genomic regions , and ii ) biological interpretability that comes from the ability to readily extract information about the importance of each feature in prediction . the accuracy results for predictions based on this model are promising , and suggest the possibility of imputing cpg site methylation levels genome - wide in the future . for example , if we assay a set of individuals in an ewas study on the illumina 450k array , we may be able to impute the missing genome - wide cpg sites from wgbs assays . however , we are still far from the prediction accuracies currently expected for snp imputation for downstream use in gwas studies . our cross - sample analysis illustrates that including methylation profiles from other individuals as references , as is done for dna imputation @xcite , may improve accuracies substantially . however , because of biological , batch , and environmental effects on dna methylation , it is possible that precise imputation will require a much larger reference panel relative to dna imputation . as in gwas studies , all of these imputation methods will fail to predict rare or unexpected variants @xcite , which may hold a substantial proportion of association signal for both gwas and ewas @xcite . this work raises the additional question , then , of how best to sample cpg sites across the genome given the methylation patterns and the possibility of imputation ; for example , it may be sufficient to assay a single cpg site within a cgi and impute the others , given the high correlation between methylation values in cpg sites within the same cgi . we identified genomic and epigenomic features that were most predictive of methylation status for co - located cpg sites . the biological functions of cgi shore and shelf regions , and in particular the impact of methylation in these regions , are mostly unknown ; however , it has been shown there is substantial dna methylation variation in cgi shore regions relative to other regions in the genome , and these alterations may contribute to cancer development and tissue differentiation @xcite . we hope to better characterize the role of cgi shore and shelf regions with respect to enrichment of particular regulatory elements in the future to understand the cellular role of these regions and the specific , curious pattern of methylation found within them . one particularly important driver of methylation that we do not study carefully here is methylation quantitative trait loci ( meqtls ) , or genetic drivers of methylation @xcite . there is substantial work on the enrichment of meqtls within snps and genetic loci that appear to regulate gene transcription levels ( eqtls ) , dhs site status ( dsqtls ) , and others @xcite . the characterizations described here lead us to consider identifying qtls associated with deviations from cre - specific methylation patterns instead of single cpg sites , as has been done with methylation in cgi shore regions and associations with cancer @xcite . we investigated genome - wide methylation in 100 individuals profiled using the illumina 450k array . we identified patterns of correlation in dna methylation at cpg sites specific to cpg islands , cgi shores , and non - cgis , quantifying the variability within cgi shore regions and a pattern of correlation across the shelf regions by which correlation increases with distance . we built a random forest classifier to predict methylation as a binary status and as a continuous level at single cpg site accuracy , using as features neighboring cpg site information , genomic position , dna sequence properties , and cis - regulatory element co - location information . our method outperformed state - of - the art methylation classifiers , including our own version of an svm - based classifier . our approach quantifies features that are most predictive of cpg status : we found that neighboring cpg site methylation levels , location in a cpg island , and co - localized dhs sites and specific transcription factor binding sites were most predictive of dna methylation levels . we identify several tfbss , including elf1 , maz , mxi1 , and runx3 , that are highly predictive of methylation levels in whole blood . these predictive features may play a mechanistic role in methylation , either in regulating the methylation of cpg sites or as a downstream partner in modifying the cellular phenotype . illumina humanmethylation450k array data were obtained for 100 unrelated individuals from the twinsuk cohort @xcite all participants in the study provided written informed consent in accordance with local ethics research committees . the 100 individuals were adult unselected volunteers and included 97 female and 3 male individuals ( age range 2778 ) . whole blood was collected and dna was extracted using standard protocols . illumina humanmethylation450k array ( illumina 450k ) measured dna methylation values for more than @xmath101 cpg sites per individual at single - nucleotide resolution . the genomic coverage includes @xmath102 of reference sequence genes , with an average of @xmath103 cpg sites per gene region distributed across the promoter , 5utr , first exon , gene body , and 3utr , and @xmath104 of cpg islands @xcite . methylation values for each cpg site are quantified by the term @xmath20 , which is the fraction of methylated bead signal over the sum of methylated and unmethylated bead signal : @xmath105 where _ methy _ represents the signal intensity of the methylated probe and _ unmethy _ represents the signal intensity of the unmethylated probe . the quantity @xmath20 ranges from @xmath106 ( unmethylated ) to @xmath45 ( fully methylated ) . data quality control was implemented using r ( http://www.r-project.org/ ) ( version 2.15.3 ) . we removed @xmath107 cpg sites for whom the probes mapped to multiple places in the human genome reference sequence . cpg sites with missing values or detection p - values @xmath108 were excluded . methylation data from the x and y chromosomes were excluded , leaving @xmath21 cpg sites from @xmath8 individuals in downstream analyses . the data were controlled for array number , sample position on the array , age , and sex by taking the residual from a fitted linear regression model . the sum of residuals and intercepts of each site was scaled to @xmath109 $ ] by truncating all sites with values larger than @xmath45 to @xmath45 and all sites with values smaller than @xmath106 to @xmath106 . data quality was assessed to identify sample outliers and batch effects using principal component analysis ( pca ) @xcite , no obvious outliers were identified . the statistical analyses were implemented using r and bioconductor ( http://www.bioconductor.org/ ) ( version 2.15.3 ) . methylation correlations between cpg sites were assessed by the absolute value of pearson s correlation coefficient and mean square error ( mse ) : @xmath110 where @xmath111 and @xmath112 represent the methylation values of the two cpg sites being compared , @xmath113 represents the total number of cpg sites being compared . for _ neighboring _ cpg sites , pairs of cpg sites assayed on the array that were adjacent in the genome were sampled ; the genomic distance between the pairs of cpg sites were within the range @xmath114 bp to @xmath115 bp , where @xmath116 . the correlation and mse of a @xmath44 bp window was not computed , as there were too few cpg sites . the non - adjacent pair correlation or mse values are the average absolute value correlation or mse of @xmath117 pairs of cpg sites that were not immediate neighbors with their genomic distances in the same range as for the adjacent cpg sites . we performed pca on methylation values of cpg sites by computing the eigenvalues of the covariance matrix of a subsample of cpg sites using the r function svd . among the @xmath52 cpg sites that have complete feature information , @xmath118 sites ( every tenth cpg site ) were sampled along the genome across all autosomal chromosomes . pearson s correlation was calculated between each feature and first ten pcs . pca analysis was performed by plotting the principal component biplot ( scatterplot of first two pcs ) . the prediction performances were assessed by receiver operating characteristic ( roc ) curves and residual sum of squares ( rss ) . we used the randomforest package in r for the implementation of the random forest classifier @xcite ( version 4.6 - 7 ) . most of the parameters were kept as default , but ntree was set to @xmath119 to balance efficiency and accuracy in our high - dimensional data . we found the parameter settings for the random forest classifier ( including the number of trees ) to be robust to different settings , so we did not estimate parameters in our classifier . the gini index , which calculates the total decrease of node impurity ( e.g. , the relative entropy of the class proportions before and after the split ) of a feature over all trees , was used to quantify the importance of each feature : @xmath120 where @xmath121 represents the class and @xmath122 is the proportion of sites belonging to class @xmath121 in node @xmath123 . we compared the performance of the random forest with the support vector machine ( svm ) an alternative classifier frequently used in related work @xcite . we built an svm classifier with a radial basis function ( rbf ) nonlinear kernel , which generally yields more accurate results than the linear kernel that was used in most of the related classifiers @xcite . we used the svm implemented in the e1071 package in r @xcite . the parameters of the svm were optimized by 10-fold cross - validation using grid search . the penalty constant @xmath124 ranged from @xmath125 and the parameter @xmath126 in the kernel function ranged from @xmath127 . the parameter combination that had the best performance was used in comparisons with our random forest classifier ; specifically , we set @xmath128 and @xmath129 a comprehensive list of @xmath77 features were used in prediction ( table s4 ) . the _ neighbors _ features were obtained from data from the methylation 450k array ; the _ position _ features , including gene coding region category , location in cgis , and snps , were obtained from the methylation 450k array annotation file ; dna recombination rate data was downloaded from hapmap ( phaseii_b37 , update date jan192011 ) @xcite ; gc content data were downloaded from the raw data used to encode the gc5base track on hg19 ( update date apr242009 ) from the ucsc genome browser ( http://hgdownload.cse.ucsc.edu/goldenpath/hg19/gc5base/ ) @xcite , integrated haplotype scores ( ihs scores ) were downloaded from the hgdp selection browser ihs data of smoothedamericas ( update date feb122009 ) ( http://hgdp.uchicago.edu/data/ihs/ ) @xcite , and gerp constraint scores were downloaded from sidowlab gerp++ tracks on hg19 ( http://mendel.stanford.edu/sidowlab/downloads/gerp/ ) @xcite ; _ cres _ features : dnase i hypersensitive sites data were obtained from the dnase - seq data for the gm12878 cell line produced by crawford lab at duke university ( ucsc accession : wgencodeeh000534 , submitted date mar202009 ) and @xmath54 specific transcription factor binding sites chip - seq data were from the narrow peak files from gm12878 cell line that were available before june 2012 from the encode website @xcite . neighboring cpg site methylation status was encoded as methylated " when the site has a @xmath29 and unmethylated " when @xmath95 . for continuous features , the feature value is the value of that feature at the genomic location of the cpg site , and for binary features , the feature status indicates whether the cpg site is within that genomic feature or not . dhs sites were encoded as binary variables indicating a cpg site within a dhs site ; tfbss were included as binary variables indicating the presence of a co - localized chip - seq peak ; ihs scores , gerp constraint scores and recombination rates were measured in terms of genomic regions ; for gc content , we computed the proportion of g and c within a sequence window of @xmath43 bp , as this feature was shown to be an important predictor in previous study @xcite . among all 99 features , 97 of them ( excluding upstream and downstream neighboring cpg sites @xmath20 values ) were used for methylation status predictions , and all excluding upstream and downstream neighboring cpg sites methylation status were used for methylation level predictions . when limiting prediction to specific regions , e.g. , cgi , we excluded those features indicating co - location with that region type . our methylation predictions were at single cpg site resolution . for regional specific methylation prediction , we grouped the cpg sites into either promoter , gene body , intergenic region classes or cgi , cgi shore & shelf , non - cgi classes according to methylation 450k array annotation file , which was downloaded from the ucsc genome browser@xcite . the classifier performance was assessed by a version of repeated random subsampling ( rrs ) validation . within a single individual , ten times we pulled @xmath41 random cpg sites from across the genome into the training set , and we tested on all other held - out sites . the prediction performance for a single classifier was calculated by averaging the prediction performance statistics across each of the @xmath46 trained classifiers . we checked the performance with smaller training set of sizes @xmath8 , @xmath119 , @xmath130 , @xmath117 and @xmath41 sites in the same evaluation setup . in cross - sample analyses , we set the size of the training set to @xmath41 randomly chosen cpg sites to balance computational performance and accuracy . we then evaluated the consistency of methylation pattern in different individuals by training the classifier using @xmath41 randomly chosen cpg sites in one individual , and then using the trained classifier to predict all of the cpg sites in the remaining @xmath77 individuals . we quantified the accuracy of the results using the specificity ( sp ) , sensitivity ( se ) , accuracy ( acc ) , and the matthew s correlation coefficient ( mcc ) . these values were calculated as follows : @xmath131 @xmath132 @xmath133 @xmath134 where tp , tn , fp , fn represent the number of true positives , true negatives , false positives , and false negatives respectively for a particular threshold . receiver operating characteristic ( roc ) curves were plotted and we chose a cutoff of @xmath27 . the area under the roc curve ( auc ) was calculated ; the auc reflects the overall prediction performance considering both type i ( fps ) and type ii errors ( fns ) @xcite . we used the rocr package in r. to estimate continuous methylation levels ( @xmath20 ) , we used the classifier output of prediction probability directly as an estimate of a specific @xmath135 $ ] . prediction accuracy in this regression setting was evaluated using the correlation coefficient and root mean squared error ( rmse ) . @xmath136 @xmath137 where @xmath138 are the experimental and predicted values , respectively , @xmath115 , @xmath139 are the means of the experimental and predicted methylation levels , respectively , @xmath140 , @xmath141 are the empirical standard deviations of the experimental and predicted values , respectively . conceived the experiments : jtb , bee . designed the experiments : wz , bee . performed the experiments : wz . analyzed the data : wz , bee . wrote the paper : wz , jtb , bee . responsible for quality control and pre - processing : wz , jtb , bee . contributed valuable data : tds , pd , jtb . the authors acknowledge the twinsuk consortium for providing all methylation data on 100 individuals , dr . casey d brown for providing critical insights , and dr . susan murphy for helpful discussions . bee was funded through nih nigri r00 hg006265 .	recent assays for individual - specific genome - wide dna methylation profiles have enabled epigenome - wide association studies to identify specific cpg sites associated with a phenotype . computational prediction of cpg site - specific methylation levels is important , but current approaches tackle average methylation within a genomic locus and are often limited to specific genomic regions . we characterize genome - wide dna methylation patterns , and show that correlation among cpg sites decays rapidly , making predictions solely based on neighboring sites challenging . we built a random forest classifier to predict cpg site methylation levels using as features neighboring cpg site methylation levels and genomic distance , and co - localization with coding regions , cgis , and regulatory elements from the encode project , among others . our approach achieves @xmath0 prediction accuracy of genome - wide methylation levels at single cpg site precision . the accuracy increases to @xmath1 when restricted to cpg sites within cgis . our classifier outperforms state - of - the - art methylation classifiers and identifies features that contribute to prediction accuracy : neighboring cpg site methylation status , cpg island status , co - localized dnase i hypersensitive sites , and specific transcription factor binding sites were found to be most predictive of methylation levels . our observations of dna methylation patterns led us to develop a classifier to predict site - specific methylation levels that achieves the best dna methylation predictive accuracy to date . furthermore , our method identified genomic features that interact with dna methylation , elucidating mechanisms involved in dna methylation modification and regulation , and linking different epigenetic processes . [ 1995/12/01 ]
the study of continuous phase transitions has been one of the most fertile branches of theoretical physics in the last decades . each phase can usually be characterized by an order parameter . often , the choice of an order parameter is obvious . however , in some cases finding an appropriate order parameter is complicated . in particular , many magnetic systems experience the luttinger liquid ( ll ) phase for certain values of some non - thermal control parameter . luttinger liquid is the paradigm for the description of interacting one - dimensional ( 1d ) quantum systems@xcite . the correlation functions decay as power laws and the ground state has a quasi - long range order . in spite of all manifestations of this phase , the magnetic ordering in the ll phase is still unclear , i.e , all suggested local magnetic order parameters turn out to be zero , having no effect in determining the structure of the ll phase . in this work , we will concentrate on studying the probably nonlocal magnetic ordering of the ll phase . one dimensional quantum spin systems such as antiferromagnetic spin-1/2 chains@xcite , spin ladder systems@xcite , or bond alternating spin-1/2 af - f chains are good candidates for studying the ll phase . when these systems are placed in an external magnetic field , they can be mapped onto a 1d system of interacting spinless fermions@xcite . the filling of a fermionic band can thus be continuously tuned , making these systems suitable for probing the ll physics@xcite . in the case of isotropic bond alternating af - f spin-1/2 chains the whole band can be covered . in the absence of an external magnetic field ( @xmath0 ) , the model is mapped onto a nonlinear sigma model with a @xmath1 topological angle@xcite . this model is always gapful and can be regarded as a haldane gapped spin-1 chain@xcite . the gap decreases by increasing the magnetic field and goes to zero at the lower critical field @xmath2@xcite . at @xmath3 , the system enters the gapless phase and the fermionic band starts to be filled . this process continues until the upper critical field @xmath4 is reached . increasing the field beyond @xmath4 reopens the gap and the band is then completely filled . recently , it was demonstrated that @xmath5 is a unique system for controlling and probing the physics of ll @xcite . this sample is a spin ladder system with @xmath6 t and @xmath7 t@xcite . bond alternating af - f spin-1/2 chains that allow experimental access to the whole fermionic band are not known . however the @xmath8 system has been considered to be a suitable realization of bond alternating af - f spin chain . linked - cluster calculations and bulk measurements show that @xmath9 is also a realization of the spin-1/2 alternating af - f chain with nearly the same strength of antiferromagnetic and ferromagnetic couplings@xcite . other experimental samples of the af - f alternating spin-1/2 chain compounds have also been reported in refs . although it is not known whether these systems can experience the ll phase , theoretically , an isotropic bond alternating af - f spin chain , in its ground state , can enter the ll phase with quasi - long range order . in this paper , we look for an order parameter to describe the ll phase of the bond alternating af - f spin-1/2 chain . the hamiltonian of this model with anisotropic ferromagnetic coupling is given by @xmath10 where @xmath11 are spin-1/2 operators on the @xmath12-th site . @xmath13 and @xmath14 denote the ferromagnetic and antiferromagnetic couplings , respectively . @xmath15 is the anisotropy parameter and @xmath16 is a uniform magnetic field . the ground - state properties@xcite and low - lying excitations@xcite of this model have been thoroughly investigated by numerical tools and variational schemes . in particular , the string order parameter which was originally defined for the spin-1 heisenberg chains@xcite has been generalized to this system . the ground state has long - range string order , which is characteristic of the haldane phase . hida has shown that the haldane phase of the af - f alternating chain is stable against any strength of randomness@xcite . the ground state phase diagram of the af - f alternating chain in a longitudinal ( @xmath17 ) magnetic field is studied using numerical diagonalization and finite - size scaling based on conformal field theory@xcite . it is shown that the magnetic state is gapless and described by the ll phase . the model represented by the hamiltonian in eq . [ h ] which includes a transverse magnetic field has been studied recently using the numerical lanczos method@xcite . the main attention in this study was focused on the investigation of field - induced effects in the ground state phase diagram , present when the antiferromagnetic coupling @xmath14 dominates ( @xmath18 ) . the system has two critical fields and the energy gap in the intermediate region depends on the anisotropic parameter @xmath15 . for @xmath19 , the intermediate state is gapful and the ground state of the model has stripe - antiferromagnetic order@xcite . in this paper we consider again an anisotropic af - f chain in a transverse magnetic field . using the numerical exact diagonalization method and the analytical spinless fermion approach , we investigate the zero temperature and thermodynamic behavior of the model . we introduce a new mean field order parameter which can distinguish the ll phase from the other gapped phases ( fig . [ fig2 ] ) . in the specific heat curves versus temperature , a double peak structure appears in the intermediate region of the magnetic fields @xmath20 . the outline of the paper is as follows . in section ii we discuss the zero - temperature ground state phase diagram of the model . in section iii we present the results of the spinless fermion approach and the numerical full diagonalization results on the low - temperature behavior of the model . finally we conclude and summarize our results in section iv . in this section we briefly discuss the model ( 1 ) in the limiting case of the strong af coupling @xmath21 . in this limit the model can be mapped onto an effective spin - chain hamiltonian@xcite . at @xmath21 , the system behaves as a nearly independent block of pairs@xcite . indeed an individual block may be in a singlet or a triplet state with the corresponding energies given by @xmath22 for @xmath23 , one component of the triplet becomes closer to the singlet ground state such that for a strong enough magnetic field we have a situation when the singlet and @xmath24 component of the triplet create a new effective spin @xmath25 system . on the new singlet - triplet subspace , the original hamiltonian becomes the hamiltonian of a fully anisotropic xyz spin-1/2 chain in an effective magnetic field@xcite @xmath26 \nonumber \\ & + & h^{eff } \sum_{j=1}^{n/2 } \tau^{z}_{j}\ , , \label{effectivehamiltonian2}\end{aligned}\ ] ] where @xmath27 . at @xmath28 , the effective problem reduces to the theory of the @xmath29 chain with a fixed antiferromagnetic anisotropy of @xmath30 in a magnetic field@xcite . the gapped haldane phase at @xmath31 for the af - f alternating chain corresponds to the negatively saturated magnetization phase for the effective spin chain , whereas the massless ll phase of the af - f alternating chain corresponds to the finite magnetization phase of the effective spin-1/2 chain . the critical field @xmath32 where the af - f alternating chain is totally magnetized , corresponds to the fully magnetized phase of the effective spin chain . away from the isotropic point @xmath33 the effective hamiltonian ( [ effectivehamiltonian2 ] ) describes the fully anisotropic ferromagnetic xyz chain in a magnetic field that is directed perpendicular to the easy axis . in this case , it is found@xcite that a gapped stripe - antiferromagnetic phase exists for the intermediate values of the transverse magnetic field @xmath20 . to recognize the different phases induced by the transverse magnetic field in the ground - state phase diagram , we implemented the modified lanczos algorithm of finite - size chains @xmath34@xcite with @xmath35 and different values of the anisotropy parameter @xmath15 . the energies of a few lowest eigenstates were obtained for the chains with periodic boundary conditions . in fig . [ fig1 ] we have plotted the results of these calculations for different values of the anisotropy parameter @xmath36 , and chain length @xmath37 . as it is clearly seen from this figure , in the case of zero magnetic field the spectrum of the model is gapped . for @xmath38 the gap decreases linearly with @xmath16 and vanishes at the critical field @xmath39 . in the isotropic case @xmath40 , the spectrum remains gapless for @xmath41 and becomes once again gapped for @xmath42 . but , in the anisotropic case @xmath43 , the excitation spectrum is gapfull except at two critical field values @xmath44 and @xmath45 . in the intermediate region @xmath46 the spin gap which appears at @xmath47 , first increases vs external field and after passing a maximum decreases to vanish at @xmath48 . in conclusion , at @xmath49 , two quantum phase transitions in the ground - state phase diagram of the model have been identified with increasing transverse magnetic field @xcite . the first transition corresponds to the transition from the gapped haldane phase to the gapless ll phase ( or gapped stripe - antiferromagnetic for the anisotropic case ) . the other one is the transition from the gapless ll phase ( or gapped stripe - antiferromagnetic ) to the fully polarized phase . , for different anisotropy parameters @xmath50 and chain size @xmath37.,width=340 ] the behavior of the gap and the quantum phase space of the model lead us to investigate the thermodynamic properties of the model . in this respect , we implement the jordan - wigner transformation to fermionize the model . because of two types of coupling constant , we introduce two kinds of spinless fermion through the following jordan - wigner transformations : @xmath51 using the above transformations , the thermodynamic behavior of the isotropic ( @xmath52 ) heisenberg af - f spin-1/2 chains has been studied in the absence of a magnetic field@xcite . by the above transformations , the af - f spin chain is mapped onto a 1d system of interacting spinless fermions ; @xmath53 treating the hamiltonian @xmath54 in the mean field approximation , the interacting fermionic system reduces to a 1d system of new non - interacting dynamical quasi - particles . many mean field order parameters ( auxiliary fields ) might be considered . many of them are irrelevant , _ i.e. _ , they do not have a stable mean field solution , while some are relevant . in our system we have introduced the magnetization , ferromagnetic and antiferromagnetic dimers as mean field order parameters ; @xmath55 utilizing the above order parameters , the mean field hamiltonian is given by ; @xmath56 where , @xmath57 in the above , @xmath58 represents thermal averaging over the hartree - fock eigen - states . @xmath59 and @xmath60 are related to the magnetization , and @xmath61 and @xmath62 are ferromagnetic and antiferromagnetic exchange order parameters , respectively . it is clear that @xmath63 in the hilbert space is the af(f)-dimer order parameter , _ i.e. _ , @xmath64 using the following unitary transformations @xmath65 the diagonalized hamiltonian is given by @xmath66 equation([param ] ) clearly shows that the effect of the anisotropy appears in the dispersion relations . the dispersion relations of low - lying excitation read as , @xmath67^{\frac{1}{2}}+h,\end{aligned}\ ] ] where @xmath68 and @xmath69 . using the above order parameters , the thermodynamic functions such as the internal energy and the specific heat are expressed as follows : @xmath70 where , @xmath71^{-1}$ ] is the fermion distribution function ( choose @xmath72 ) . in order to obtain the thermodynamic behavior of the system , we need to know the whole temperature behavior of the order parameters for different values of @xmath16 . the order parameters satisfy a set of self consistent equations and it should be solved the equations ( [ or - pa ] ) , self - consistently . in fig.[fig2]-(a ) , we have plotted the ferromagnetic dimer order parameter of both anisotropic @xmath73 and isotropic ( @xmath52 ) af - f heisenberg spin-1/2 chains versus @xmath16 . the curves have been plotted close to the zero temperature . as we mentioned , at @xmath49 for different values of @xmath16 , our 1d system experiences three phases . in the haldane phase ( @xmath74 ) , the quantum fluctuations are strong enough to suppress the ferromagnetic order of the system and the af(f)-dimer parameters @xmath75 should be close to the classical value @xmath76@xcite . in this region the energy spectrum is gapped and the gap is decreased by increasing @xmath16 . versus @xmath16 , @xmath77 and @xmath78 , b ) exact diagonalization lanczos results of the real part of the @xmath61 for a chain size @xmath37.,width=302 ] versus @xmath16 , @xmath77 and @xmath78 , b ) exact diagonalization lanczos results of the real part of the @xmath61 for a chain size @xmath37.,width=302 ] the first quantum phase transition occurs at @xmath3 . at the same time the fermionic band starts to be filled where the excited state sticks to the ground state and the gap of the system is closed . for the first region there is no specific difference between the two anisotropic and isotropic cases . _ i.e. _ , anisotropy shows a marked behavior for the case of @xmath20 . for the case of @xmath52 and @xmath0 , the system is isotropic and has su(2 ) rotational symmetry . in the intermediate region this system is gapless and correlations decay in a power law@xcite . in this region , the luttinger liquid phase , the magnetic field suppresses quantum fluctuations and induces quasi long - range order in the 1d system . although the power law behavior of the correlations and other properties of the ll phase are manifest , the lack of an order parameter to show the behavior of the system in this phase is still felt . as it is obviously seen from fig.[fig2 ] , the f - dimer order parameter has a considerable value in this region and is sensitive to @xmath16 . the value of this order parameter falls sharply at the second critical field where the fermionic band is completely filled and the gap of the system is reopened . we have also plotted in fig.[fig2]-(b ) the f - dimer order parameter of the isotropic af - f heisenberg spin-1/2 chain by using the exact diagonalization lanczos results . the numerical lanczos method is implemented as a reference approach to see the accuracy of our spinless fermion method . as it is clearly observed from fig.[fig2]-(b ) the results of lanczos method , confirms that the defined mean field order parameter has considerable value for the intermediate values of field . we have also plotted in fig.[fig2 ] the f - dimer order parameter of anisotropic f - af spin-1/2 chain versus transverse magnetic field . regarding the anisotropy in ferromagnetic coupling , the anisotropic system is mapped to a 1d xyz model with u(1 ) symmetry . in the intermediate region ( @xmath20 ) the system has the magnetic long range order and spins are aligned stripe - antiferromagnetically . this phase is gapfull and the f - dimer order has no considerable value in the this phase . moreover , there is a big discrepancy between the values of the f - dimers of the isotropic and anisotropic cases , _ i.e. _ in the ll gapless phase the behavior of the f - dimer order parameter ( @xmath79 ) is different from those of the gapped phases . therefore , we can conclude that the above f - dimer order parameter can distinguish between the gapless ll phase and the other gapped phases . this means that experimenters might look for such behavior in @xmath79 at finite temperatures . _ i.e. _ , experimental results on this quantity may be used to distinguish between the isotropic chains and anisotropic ones ( gapped stripe - antiferromagnetic and gapless ll phases ) . in this section we will study the thermal behavior of the mean field order parameters of the heisenberg spin-1/2 af - f chains in both anisotropic and isotropic cases . the order parameters also have different behavior for different values of @xmath16 . for @xmath74 , in the haldane phase , an increase in temperature increases the thermal fluctuations and haldane ordering is suppressed . the solution of the eqs . [ or - pa ] shows that the af - dimer order parameter ( @xmath80 ) decreases with temperature . however , @xmath81 couplings supply an interaction between the unit cells and try to establish f - dimer order ( @xmath82 ) even at moderate temperatures ( scaled with @xmath81 ) , _ i.e. _ ( @xmath82 ) increases with temperature up to @xmath83 ( see fig . [ fig3 ] ) . above @xmath83 , fluctuations are large enough to decrease @xmath79 . the locations of the maximum value of ( @xmath79 ) with respect to @xmath16 are shown in fig.[fig3 ] . as it is obviously seen from this plot , there is a peak at @xmath16 very close to the first critical field . the location of this peak is a good candidate to find the first critical point . by increasing the magnetic field , the value of @xmath83 ( location of the maximum value of @xmath79 ) decreases . it is found that @xmath83 goes to a minimum ( fig.[fig4 ] ) when @xmath16 increases up to @xmath3 . increasing the magnetic field further and for @xmath84 , @xmath83 always maintains a minimum value and the value of @xmath79 decreases monotonically with increasing @xmath85 . increasing the magnetic field even further , for @xmath86 , @xmath83 also increases . versus @xmath16 , for @xmath77 and @xmath78 .,width=302 ] versus @xmath16 , with @xmath77 and @xmath78 .,width=302 ] the sf calculations also show that the behavior of @xmath83 is similar to the energy gap ( see fig.[fig1 ] ) for different values of the magnetic field @xmath16 with @xmath77 and @xmath78 . ( a ) @xmath52 ( b ) @xmath43.,width=302 ] for different values of the magnetic field @xmath16 with @xmath77 and @xmath78 . ( a ) @xmath52 ( b ) @xmath43.,width=302 ] for different values of the magnetic field @xmath16 with @xmath77 and @xmath78 . ( a ) @xmath52 ( b ) @xmath43.,width=302 ] for different values of the magnetic field @xmath16 with @xmath77 and @xmath78 . ( a ) @xmath52 ( b ) @xmath43.,width=302 ] for both values of @xmath15 and @xmath87 using the spinless fermion ( sf ) and full diagonalization ( fd ) approaches.,width=302 ] the discovery of gapless or gapped excitations have led to the investigation of the thermodynamic properties of the model . one of the most important thermodynamic functions is the specific heat . in this respect , we study the temperature - dependence of the specific heat of the model in different quantum regimes . we have calculated the specific heat of both isotropic and anisotropic cases , using two methods , the analytical sf and the numerical full diagonalization methods . in fig . [ fig5]a(b ) we have plotted the results of the sf method in the form of the specific heat @xmath88 versus @xmath85 with @xmath52 and @xmath43 for different values of @xmath16 . as it is clearly seen from this figure , in the haldane and paramagnetic regions ( @xmath74 and @xmath86 , respectively ) , there is only one peak on the specific heat curve . however , in the intermediate region ( @xmath46 ) a double peak is observed . at @xmath0 there is a schottky like peak in the specific heat which is expected from the haldane phase . by increasing @xmath16 , the peak becomes wider and goes to lower temperatures . increasing @xmath16 further , causes a shoulder to appear on the right hand side of the curve . for @xmath89 , increasing @xmath16 , an additional peak appears which is the signal for the paramagnetic phase . because of the two f and af interactions ( @xmath13 and @xmath14 ) , there are two kinds of quasi - particles in the system . these quasi - particles have two different dynamics . these dynamics bring about two energy scales in the system . the energy scales affect the behavior of the response functions . it is obviously seen from the plots of the specific heat @xmath88 that the energy scales create a double peak in the specific heat at two different temperatures . as a general statement , two energy scales can produce a double peak structure in the specific heat . we have also plotted in figs . [ fig6 ] a(b ) the specific heat of the model by using the full diagonalization method . we have computed all the eigenvalues of the energies for different values of the transverse magnetic field @xmath16 and anisotropy parameter @xmath15 . therefore , using these eigen - energies , we have computed the specific heat as a function of the temperature @xmath85 . the specific heat curves have been plotted for @xmath77 and @xmath90 and the anisotropy parameter @xmath91 , @xmath92 . as it is clearly seen from this figures , in the haldane and paramagnetic regions ( @xmath74 and @xmath86 , respectively ) , there is only one peak in the specific heat curve . to see the qualitative agreement between spinless fermion results and full diagonalization ones , we have plotted the specific heat versus @xmath85 for an intermediate value of @xmath16 ( for example @xmath87 ) . as it is seen from fig.[fig7 ] the mean field results are in agreement with the others , qualitatively . it is interesting that the numerical results confirm the existence of the double peak in the intermediate region ( @xmath46 ) . to summarize , we have studied the zero and finite - temperature behavior of the anisotropic alternating af - f heisenberg spin-1/2 chains in a transverse magnetic field . the numerical exact diagonalization method and analytical spinless fermion approach are applied to analyze the model . the first notable point is introducing a new mean field order parameter which can distinguish between a gapless ll phase and the gapped phases . this order parameter in the spin language of hilbert space is the f - dimer order parameter . in the isotropic case , the f - dimer order parameter has a considerable value in the ll region . we have shown that there is a big discrepancy between the values of the f - dimers of the isotropic and anisotropic cases , _ i.e. _ , in the ll gapless phase the behavior of the f - dimer order parameter ( @xmath79 ) is different from that of the gapped phases . therefore , we have concluded that the f - dimer order parameter can distinguish the gapless ll phase from the other gapped phases . the second notable point is found from the specific heat . we have obtained a double peak structure in the specific heat curves vs temperature . there are some questions behind the defined order parameter which are still controversial and we can investigate them in future . the most remarkable questions referred to the topological order and spontaneous symmetry breaking . _ i_.e . in the ll phase , where the real part of the f - dimer is not zero , how can describe topological order and what kind of spontaneous symmetry breaking is occurred . it is our pleasure to thank g. i. japaridze for his valuable comments and fruitful discussions . j. a. also would like to thank a. langari and f. shahbazi for their useful suggestions and comments . we are also grateful to b. farnudi for reading carefully the manuscript and appreciate his useful comments . j. a. was supported by the grant of shahrood university of technology . t. giamarchi , _ quantum physics in one dimension _ , ( oxford univ.press , oxford , 2004 ) . m. takahashi , _ thermodynamics of one - dimensional solvable models _ , ( cambridge university press : cambridge , 1999 ) . b. lake , d. a. tennant , c. d. frost and s. e. nagler , nat . mat . * 4 * , 329 ( 2005 ) . m. b. stone , w. tian , m. d. lumsden , g. e. granroth , d. mandrus , j .- h . chung , n. harrison , and s. e. nagler : phys . lett * 99 * , 087204 ( 2007 ) . m. hagiwara , y. narumi , k. kindo , t. c. kobayashi , h. yamakage , k. amaya , and g. schumauch , j. phys . soc . jpn . * 66 * , 1792 ( 1997 ) .	we have studied the zero and low - temperature behavior of anisotropic alternating antiferromagnetic - ferromagnetic heisenberg spin-1/2 chains in a transverse magnetic field . using the analytical spinless fermion approach , the thermodynamic behavior of the model has been studied . we have introduced a new order parameter to distinguish the gapless luttinger liquid phase from the other gapped phases . the exact diagonalization lanczos results are also used to compare with the spinless fermion ones . we have found a double peak structure in the specific heat curves for the region between the two quantum critical points . using the numerical full diagonalization results , the existence of the double peak structure is confirmed .
direct numerical simulations ( dns ) provide the ability to simulate turbulent flows with the least amount of physical modeling . this is accompanied by a strong computational demand due to the scale disparity which restricts dns often to simple geometries and moderate reynolds numbers . large - eddy simulation ( les ) offer a way to reduce this computational demand by modeling the small non - resolved length scales ( see , e.g. , @xcite ) . + on the other side of the spectrum are approaches based on the reynolds averaged navier - stokes ( rans ) equations . such rans based approaches are applicable for many engineering applications ( i.e. , complex geometries and high reynolds numbers ) , however , involves significant modeling assumptions that fail to predict complex flow phenomena with high accuracy , e.g. , pressure driven separation and the corresponding flow topology . hybrid rans / les methods offer an attractive alternative to rans or les stand - alone methods where through a suitable combination of rans and les only les is performed in the regions where it is needed and rans in the remainder . for an overview concerning hybrid les / rans approaches the reader is referred to @xcite . + in computational aeroacoustics ( caa ) hybrid approaches are common in simulating acoustic propagation , where perturbation equations ( such as the linearized euler equations ) are solved on top of a background flow . viscous effects or nonlinearity are often neglected as they play a minor role for sound propagation over small distances and when considering small acoustic perturbations . + in the current work , an extended version of piano ( perturbation investigation of aerodynamic noise ) is used . piano is a block - structured , high - order , low - dispersive and dissipative caa code @xcite . piano solves perturbation equations , e.g. , the linearized euler equations ( lee ) , acoustic perturbation equations ( ape ) @xcite , or non - linear euler equations in perturbation form ( penne ) @xcite over a time - averaged background flow . the non - linear euler equations have been augmented with fluctuating viscosity terms and therefore represent the fully non - linear navier - stokes equations formulated in perturbation form over a given base flow ( see , e.g. , @xcite ) . in the majority of cases , the base flow constitutes a steady background flow obtained with a rans simulation . + in the above described approach , piano is applicable as a zonal - hybrid tool for scale - resolving simulation of the flow . in contrast to the so - called embedded hybrid approach , the current method is referred to as `` overset '' to emphasize that a perturbation simulation is performed on top of a background flow . in fig . [ fig : overset ] the difference between the embedded ( left ) and overset approach ( right ) is illustrated . as opposed to embedded les , there exists an area where the rans and les overlap ( red color in fig . [ fig : overset ] ) with an overset les ( oles ) grid . as such , it constitutes a hybrid technique similar to the one proposed by terracol @xcite , however , formulated in primitive variables . an additional benefit of the here proposed method is that the computed nonlinear perturbations provide information about the correctness of the underlying base flow ( i.e. , a non - zero steady part of the perturbations should be interpreted as a correction of the underlying background flow ) . possible applications are slat noise ( i.e. , wing with extended slat ) or trailing - edge noise . + the overset approach is further illustrated with the aid of a cylinder in uniform flow at low reynolds number in fig . [ fig : overset2 ] ( see , e.g. , @xcite ) . an application to a decaying taylor - green vortex flow is presented in ref . the cylinder is here indicated by the black circle . the steady background flow is depicted in the lower plane of fig . [ fig : overset2 ] and represents the mean pressure field around the cylinder . the fluctuating pressure obtained with the overset technique is presented in the top layer , showing strong hydrodynamic pressure fluctuations in the cylinder wake and weaker ( as compared to the hydrodynamic ) acoustic pressure fluctuations whose propagation direction is mainly normal to the uniform flow . this cylinder in uniform flow illustrates a dnc application of piano ( i.e. , simultaneously resolving the turbulent scales ( dns ) including the radiated acoustic perturbations , see @xcite ) . as a comparison , piano in its traditional caa - modus would use a steady background flow on top of which the acoustic propagation is performed , e.g. , with the lee . + the present paper reports on the application of a novel perturbation approach ( denoted as overset approach ) to a generic fully developed turbulent channel flow . the derivation of the non - linear perturbation equations ( nlpe ) with fluctuating viscous terms is presented as well as information related to the numerical method . the ability to simulate turbulent flow with the overset approach is illustrated with a turbulent channel flow . results show an excellent agreement with reference dns data of moser et al . the ability of the overset method to successfully correct for an imperfect background flow is furthermore shown . + this paper is organized as follows . in sect . 2 the derivation of the governing equations ( the extended viscous nlpe ) is presented in the second section , followed by details concerning the numerical method in section 3 . hereafter , in sect . 4 the computational setup and results of the fully developed turbulent channel flow are presented . finally , in sect . 5 the main conclusions are summarized . in the first subsection , the viscous non - linear disturbance equations are presented which are applicable in the so - called overset dns approach . hereafter , the extension is made to overset les ( oles ) by presenting the subgrid - scale modeling is the second subsection . the navier - stokes equations in primitive perturbation formulation form constitutes the basis of the here used governing equations . the navier - stokes equations for a compressible flow read @xmath0= 0 \label{eq:01c}&\end{aligned}\ ] ] with density @xmath1 , velocity vector @xmath2 , and pressure @xmath3 . the heat flux vector and viscous stress tensor are denoted by @xmath4 and @xmath5 , respectively . furthermore , @xmath6 represents the specific heat ratio . einstein summation convention is implied for repeated indices . the use of primitive variables ( i.e. , density @xmath1 , velocity @xmath2 , and pressure @xmath3 ) has the advantage that it delivers directly the desired variable of interest in aeroacoustics : the fluctuating pressure ( see , e.g. , @xcite ) . we are primarily interested in applying the overset les approach to subsonic problems , for which a formulation in primitive perturbation variables is suitable . energy conservation of the mean - flow is guaranteed by a strongly conservative rans simulation ( see , e.g. , @xcite ) . preliminary test with a energy conserving skew - symmetric formulation of the nlpe do not indicate any improvement so far . however , to capture also supersonic simulation problems with possible shocks @xcite , also conservative perturbation formulations ( nlde ) will be studied with piano in future work . non - linear disturbance equations ( nlde ) starting from the navier - stokes equations in conservative form have been proposed by morris et al . @xcite and liu & long @xcite , or nlde based les by terracol @xcite . + the perturbation form of the above navier - stokes equations are obtained by substitution of a decomposition of the variables into a background part and a fluctuating part , i.e. , @xmath7 the base flow variables are indicated by the superscript `` 0 '' and perturbations by a prime . note that in eqs . ( [ eq:02 ] ) , @xmath8 and @xmath9 denote viscous background flow and fluctuating viscous contributions , respectively , they do not represent turbulent reynolds stresses . upon substitution of the above decomposition into the compressible navier - stokes equations [ eqs . ( [ eq:01a]-[eq:01c ] ) ] , one obtains after some rearrangement @xmath10= \vartheta^0 , \label{eq:03c}\end{aligned}\ ] ] where use has been made of the definition for the stress tensor @xmath11 ( with @xmath12 ) . in these equations , variables without superscript denote total variables and produce nonlinear terms when combined with primed variables . terms containing only background flow contributions have been collected in the right - hand side residual terms and are given by @xmath13 \label{eq:04a } & \\ & f^0_i = -\left[\frac{\partial v_i^0}{\partial t}+ v^0_j \frac{\partial v^0_i}{\partial x_j } + \frac{1}{\rho^0}\frac{\partial p^0}{\partial x_i } - \frac{1}{\rho^0}\frac{\partial \tau^0_{ij}}{\partial x_j}\right ] \label{eq:04b}\\ & \vartheta^0 = -\left[\frac{\partial p^0 } { \partial t } + v_i^0\frac{\partial p^0}{\partial x_i } + \gamma p^0\frac{\partial v^0_i}{\partial x_i } -(\gamma-1)\big ( \tau^0_{ij}\frac{\partial v^0_i } { \partial x_j}-\frac{\partial q^0_i } { \partial x_i}\big)\right].\label{eq:04c}\end{aligned}\ ] ] this set of equations [ eqs . ( [ eq:03a]-[eq:04c ] ) ] , the viscous non - linear perturbation equations ( nlpe ) , are applicable to simulate laminar and turbulent sound generation in the manner of a direct numerical simulation ( dns ) . the base flow part of the viscous stress tensor and heat flux vector are denoted by @xmath14 and @xmath15 , respectively . therefore the right - hand side residuals represent residual turbulent viscous stress and heat flux whose exact definition is related to the base flow definition . for a base flow obtained from a steady compressible rans simulation , where the variables indicated by `` 0 '' are favre averaged , one can write @xmath16.\label{eq:05c}\end{aligned}\ ] ] finally , the residual right - hand side terms become for a steady compressible rans base flow @xmath17 . \label{eq:06c}\end{aligned}\ ] ] the residual right - hand source terms , represent the residual turbulent viscous stresses and heat fluxes , whose exact definition depends on the definition of the base flow . + for the case that the base flow is obtained from a steady rans simulation , the storage requirement for background flow components can be reduced in the following way . with time derivatives equal zero for a steady rans , the momentum eq . ( [ eq:05b ] ) can be used to rewrite the divergence of the background stress tensor @xmath18 , i.e. @xmath19 the number of additional background flow variables that need to be permanently stored reduces from 13 ( 6 for the mean - flow stress tensor , three for the divergence of the mean - flow stress tensor , and 4 for the residual right - hand side terms ) to 10 ( 6 for the mean - flow stress tensor , three for the divergence of the mean - flow stress tensor , and 1 for @xmath20 ) . the extended viscous nlpe ( with memory - reduced momentum equations ) finally becomes @xmath21= \vartheta^0.\label{eq:08c}\end{aligned}\ ] ] where @xmath22 the above presented extended viscous nlde are preferred over the original viscous nlde as they require less memory storage ( see @xcite ) . equations ( [ eq:08a]-[eq:09 ] ) reduce to the penne equations as proposed by long @xcite when the viscous terms are neglected . + the computed perturbations provide information concerning the correctness of this underlying background flow , i.e. , a non - zero steady part of the perturbation provides a correction for the rans background flow . note that for aeroacoustic applications , the correct temporal evolution of the pressure fluctuations is of paramount importance and the ( residual ) mean pressure should be subtracted as this masks the acoustic content @xmath23 where @xmath24 is the computed perturbation pressure from piano . the in the previous subsection described navier - stokes equations can be made usable as overset les ( oles ) by extension with suitable subgrid - scale ( sgs ) models for the stress and heat flux contributions [ i.e. , @xmath25 and @xmath26 , respectively ] . note that then the equation variables implicitly represent favre - average variables ( i.e. , density - weighted averaging ) . for the flow variables this yields @xmath27 where spatial filtering is denoted by the overbar and favre averaging by a tilde . the subgrid - scale terms are defined in terms of favre - averaged flow variables , for the sgs stress term this yields @xmath28 where the exact definition depends upon the used sgs model . two sgs models will be described , explicit spatial filtering and the classical smagorinsky ( see , e.g. , @xcite ) . furthermore , an sgs - model based on the so - called flow simulation methodology ( fsm , see @xcite ) is discussed . spatial filtering plays a important role in caa codes . normally caa codes have low dissipative finite difference schemes , and besides this ( artificial ) numerical dissipation there is often no physical dissipation in the governing equations when using , e.g. , the lee . filtering is utilized to remove grid to grid oscillations , i.e. , spurious high - wavenumber oscillations . furthermore , it can also act as a sgs model by removing the high - frequency small scales through explicit spatial filtering of flow variables . such an approach is motivated by considering the smallest scales mainly responsible for dissipative action which can be modeled by removing the high frequency part of the solution . implicit les approaches based on explicit filtering has been employed by visbal et al . @xcite , rizzetta et al . @xcite , and marsden et al . @xcite . an implicit les approach based on spatial filtering ( as practiced e.g. , @xcite ) , is easily implemented by utilizing the finite - difference filters that are available in the piano code . the deviatoric part of the stress tensor is modeled with the eddy viscosity hypothesis , i.e. , @xmath29 where @xmath30 denotes the resolved - velocity strain - rate tensor and @xmath31 the eddy viscosity . the latter is computed with the classical smagorinsky model which reads @xmath32 here , @xmath33 represents the smagorinsky constant and @xmath34 the filter width which is related to the grid size , here taken to be @xmath35 . combination of the last two equations reveals that the modeled stress tensor is proportional with the resolved strain - rate tensor squared . for the perturbation approach , an additional term appears on the right hand side of the momentum equation @xmath36 with @xmath37 @xmath38 where use has been made from the relations @xmath39 and @xmath40 , i.e. , a similar decomposition as eqs . ( [ eq:02 ] ) . some rearrangement yields @xmath41 where variables without a superscript denote total variables . the first term on the right - hand side is a background flow contribution to the sgs and therefore added to eq . ( [ eq:09 ] ) , whereas the remaining terms represent linear and non - linear sgs contributions and are shifted to the left - hand side of momentum equation eq . ( [ eq:08b ] ) . + the subgrid - scale heat flux term @xmath42 in the energy equation is modeled in a similar way as is done above , by a proportionality coefficient @xmath43 which is related to the eddy viscosity by @xmath44 with @xmath45 the subgrid - scale prandtl number ( taken equal to 0.6 @xcite ) . the turbulent eddy viscosity is obtained as described above . the subgrid - scale heat flux term @xmath26 is modeled as @xmath46 and added to the right hand side source term @xmath20 [ eq.([eq:06c ] ) ] . a further sgs model is presented that is based on the flow simulation methodology ( fsm , originally developed by speciale @xcite ) , i.e. @xmath47 with @xmath34 proportional to grid size and @xmath48 represents the kolmogorov length scale . the ratio of these two variables is used to estimate the `` distance to dns '' @xcite with @xmath49 the contribution function that is used to damp the contributions from the rans model . for a course grid ( @xmath50 ) the rans model is fully used and the contribution function equals one , for a dns - like grid ( i.e. , @xmath51 ) the contribution function goes to zero . + the fsm adaptation reads @xmath52 with @xmath53 . the rans turbulent viscosity @xmath54 is estimated with the variables turbulent kinetic energy @xmath55 and dissipation @xmath56 . the modeled viscosity @xmath57 then becomes @xmath58 therefore , the fluctuating part of the modeled viscosity equals zero , i.e. , @xmath59 . the turbulent viscous stress becomes @xmath60 so that for the fluctuating stress tensor it follows @xmath61 as opposed to the smagorinsky sgs - model , the fsm based sgs - model is linear proportional to the resolved strain - rate tensor . the in section 2 presented governing equations are solved for the results in the remainder of this paper . the caa - code piano @xcite is a structured code , based on curvilinear , multi - block grids . in addition to the in section 2 presented viscous nlpe equations , it also supports the computation of sound propagation by the linearized euler equations ( lee ) , acoustic perturbation equations ( ape ) or non - linear euler equations in primitive disturbance form ( penne ) as governing equations . as is fairly standard in aeroacoustics , optimized finite - difference schemes are used for the numerical discretization . spatial gradients are approximated with the dispersion relation preserving ( drp ) scheme , as proposed by tam and webb @xcite . explicit time advancement is achieved by the 4th - order low - dispersion runge - kutta ( lddrk ) algorithm , as proposed by hu et al . artificial selective damping as well as different filters provides the possibility to remove contaminated high - frequency wave components . + new terms in the equations , as compared to piano in its traditional caa modus , are the viscous and heat flux terms . these additional terms involve second derivative operators which could be tackled by the successive application of the first derivative operator . the computation of viscous terms mainly involves computation of divergence of vectors and are calculated as such . the routine based on divergence calculations also serves as a building block for the anticipated future implementation of a full conservative formulation of the governing equations . the interested reader is referred to moghadam @xcite for a detailed description of the developed divergence computation . the fully developed turbulent channel flow at @xmath62 serves as a test case for the overset les method , where @xmath63 is the friction based reynolds number based on the mean wall - friction velocity @xmath64 . a comparison will be made mainly with corresponding dns reference data from moser et al . @xcite . the simulations have been carried out in a computational box of @xmath65 equal to @xmath66 . here , the @xmath67- and @xmath68-axes are in the streamwise and spanwise direction , respectively . the @xmath69-axis is the wall - normal direction , perpendicular to the plane spanned by the @xmath70-axes . the number of grid points @xmath71 equals @xmath72 . periodic boundary condition were enforced on the streamwise and spanwise directions . the channel walls were taken to be no - slip , isothermal and normal pressure gradient boundary conditions . the pressure gradient in the streamwise direction enters the simulation through the background flow ( cf . the residual right - hand side term for the momentum equations ) . the computational mesh is equally spaced in the @xmath67- and @xmath68-direction ( @xmath73 ) . a stretching has been applied to the wall - normal direction so as to have the minimum spacing at the wall of @xmath74 and a maximum at the channel centerline of @xmath75 . the simulations were performed on the north - german supercomputing alliance ( hlrn ) cluster and were parallelized on 24 processor for the used computational domain . the timestep was set equal to @xmath76 where typically about 600.000 timesteps were computed . + the simulations have been initiated with the aforementioned rans background flow , obtained with the dlr tau - code ( see , e.g. , @xcite ) . it constitutes a 3d rans simulation with the reynolds stress model jhh - v2 ( see , e.g. , @xcite ) . with the frpm code ( see , e.g. , @xcite ) realistic turbulent fluctuations ( that correspond to the turbulent kinetic energy distribution provided by the background flow ) were generate and added to the rans background flow to accelerate the development of turbulence . typically a total of 45 convective time units were simulated with the overset approach , and after an initial transient , the last 25 convective time units were sampled for statistics . + first , the utilised steady background flow is presented obtained with the dlr tau - code . -plane slice of steady rans background flow showing contour and vectors of the streamwise velocity component . ( right ) non - dimensional mean velocity profile @xmath77 of the steady rans background flow together with dns reference data ( moser et al . @xcite ) . ] -plane slice of steady rans background flow showing contour and vectors of the streamwise velocity component . ( right ) non - dimensional mean velocity profile @xmath77 of the steady rans background flow together with dns reference data ( moser et al . @xcite ) . ] figure [ fig : figrans ] ( left ) shows contours and vectors of the streamwise velocity component , where the typically full velocity profile of a turbulent channel flow can be appreciated . in the right plot of fig . [ fig : figrans ] , the non - dimensional mean velocity @xmath77 profile is depicted for the rans background flow together with the one for the dns reference data . clearly , the the logarithmic region of the non - dimensional mean velocity profile @xmath77 is substantially over - predicted . + in fig . [ fig : figrans2 ] the normal reynolds stress components given by the reynolds stress model jhh - v2 ( see , e.g. , @xcite ) are depicted as function of wall units . ( @xmath78 ) , @xmath79 ( @xmath80 ) , and @xmath81 ( @xmath82 ) , respectively.,scaledwidth=60.0% ] comparison of the reference dns data ( moser et al . @xcite ) with the steady tau rans reveals a qualitative agreement with significant quantitative differences . in section 2 , it was claimed that the overset approach is able to correct for such deficits in the background flow [ see figs . [ fig : figrans ] ( right ) and [ fig : figrans2 ] ] , provided sufficient resolution of the overset approach . this would show up as a non - zero mean part of the calculated fluctuations ( of the overset les ) . + in fig . [ fig : vortexshed ] a snapshot of vortical structures ( visualized with the @xmath83-criterion ) is depicted for the fully developed turbulent channel flow . -criterion for the fully developed turbulent channel flow at @xmath62.,scaledwidth=65.0% ] near the walls , small vortical structure are seen with increasing vortex size towards the centerline of the channel . + the universal velocity profile @xmath77 of the fully developed channel flow obtained with the overset les method is displayed in fig . [ fig : figles ] ( left ) . it compares the piano simulation with a reference dns of moser et al . @xcite , illustrating that the logarithmic region is correctly captured . in the right plot of fig . [ fig : figles ] the normal reynolds stresses are depicted for this turbulent channel flow . comparison between the piano and dns reference data reveals an excellent quantitative agreement . peak value and position of , e.g. , the streamwise reynolds stress component are captured very well . . ( left ) non - diemnsional mean velocity profile @xmath77 and ( right ) non - dimensional reynolds stress components for from top to bottom @xmath84 ( @xmath78 ) , @xmath79 ( @xmath80 ) , and @xmath81 ( @xmath82 ) . dns reference data from moser et al . ( left ) non - diemnsional mean velocity profile @xmath77 and ( right ) non - dimensional reynolds stress components for from top to bottom @xmath84 ( @xmath78 ) , @xmath79 ( @xmath80 ) , and @xmath81 ( @xmath82 ) . dns reference data from moser et al . ] comparison of the rans computed mean velocity and reynolds stresses [ fig . [ fig : figrans ] ( right ) and fig . [ fig : figrans2 ] , respectively ] and the piano results for the mean velocity and reynolds stresses ( see fig . [ fig : figles ] ) reveals that piano indeed is able to correct an imperfect background flow ( provided sufficient resolution ) . hybrid rans / les methods offer an attractive alternative to stand - alone rans or les methods where les is performed in the regions where it is needed and rans in the remainder . a reduction in computational demand results as compared to stand - alone les , facilitating the computation of more complex geometries and higher reynolds numbers . in computational aeroacoustics such hybrid approaches are common , where the acoustic propagation is computed with a perturbation approach ( such as linearized euler equations ) on top of a background flow . + in this contribution , we have presented an application of a novel perturbation approach ( denoted as an overset approach ) to a generic turbulent channel flow . the derivation of the non - linear perturbation equations extended with fluctuating viscous terms was presented as well as subgrid - scale modeling aspects which results in the overset les ( oles ) method . the application of the overset method was illustrated with a generic fully developed turbulent channel flow , the results of which show excellent agreement with reference direct numerical simulation validation data from moser et al . furthermore , the ability of the overset method to successfully correct for an imperfect background flow was shown . this work was funded by the deutsche forschungsgemeinschaft dfg ( german research funding organisation ) in the framework of the collaborative research centre crc880 ( sfb880 ) . computational resources were provided by the north - german supercomputing alliance hlrn , which is gratefully acknowledged . 00 p. sagaut , `` large eddy simulation for incompressible flows : an introduction '' , 3rd edition , springer , 2006 j. frhlich , d. von terzi , hybrid les / rans methods for the simulation of turbulent flows , prog . , vol . 44 , 2008 . delfs , m. bauer , m. , r. ewert , h.a . grogger , m. lummer , t.g.w . lauke , numerical simulations of aerodynamic noise with dlrs aeroacoustic code piano , vol . handbook version 5.2 , 2008 . r. ewert , w. schrder , `` acoustic perturbation equations based on flow decomposition via source filtering , '' j. comp . , vol . 188 , 2003 . long , `` a nonconservative nonlinear flowfield splitting method for 3-d unsteady fluid dynamics , '' aiaa 2000 - 1998 , 2000 . r. ewert , m. bauer , m. lummer , a review of state - of - the - art aeroacoustic prediction approaches . in : r. dnos , e. lecomte , e. kors , c. schram . aircraft noise , monograph vki lecture series 2012 - 02 , aircraft noise , von krmn institute for fluid dynamics , brussels , 2012 . ewert , r. , dierke , j. , neifeld , a. , moghadam , s. m. a. , `` linear and non - linear perturbation equations with relaxation source terms for forced eddy simulation of aeroacoustics sound generation , '' 20th aiaa / ceas aeroacoustics conference , altanta , usa , 2014 . moghadam , implementation of viscous terms into the computational aeroacoustics code piano , msc .- thesis tu braunschweig , 2012 . m. terracol , a zonal rans / les approach for noise sources prediction , flow turbul . 77 , 2006 . akkermans , r. ewert , m.s.a . moghadam , j. dierke , n. buchmann , `` overset dns with application to sound source prediction , '' in progress in hybrid rans - les modelling paper contributions to the 5th symposium on hybrid rans - les methods , 19 - 21 march 2014 , college station , texas , usa . notes on numerical fluid mechanics and multidisciplinary design , vol . 130 , 2015 . akkermans , n. buchmann , r. ewert , j. dierke , `` overset dns with application to application to homogeneous decaying turbulence , '' proceedings of stab symposium 2014 , in notes on numerical fluid mechanics and multidisciplinary design , 2014 ( accepted with minor modifications ) . moser , j. kim , n.n . mansour , `` direct numerical simulation of turbulent channel flow up to @xmath63=590 , '' phys . fluids , vol . 11 , 1999 . c. bailly , c. bogey , o. marsden , progress in direct noise , int . j. aeroacoust . , vol . 9 , 2010 . j. sesterhenn , `` a characteristic - type formulation of the equations for high order upwind schemes , '' comput . fluids , vol . 30 , 2001 . d. schwamborn , t. gerhold , r. kessler , dlr - tau code an overview , 1st onera / dlr aerospace symposium , 1999 . anderson , computational fluid dynamics : basics with applications , mcgraw - hill , 1995 . p. morris , l. long , a. banglore , q. wang , `` a parallel three - dimensional computational aeroacoustics method using nonlinear disturbance equations , '' j. comput . , vol . 133 , 1997 . j. liu , l.n . long , `` direct aeroacoustic and aerodynamic simulation of multi - hole engine liners using the non - linear disturbance equations , '' aiaa 98 - 2320 , 1998 . visbal , d.p . rizzetta,``large - eddy simulation on curvilinear grids using compact differencing and filtering schemes , '' j. fluid . 124 , 2002 . d. rizzetta , m. r. visbal , g.a . blaisdell `` a time - implicit high - order compact differencing and filtering scheme for large - eddy simulation , '' int . 42 , 2003 . o. marsden , c. bogey , c. bailly , direct noise computation of the turbulent flow around a zero - incidence airfoil , aiaa j. , vol . 46 , 2008 . e. garnier , n. adams , p. sagaut , `` large eddy simulation for compressible flows '' , springer , 2009 . speziale , `` turbulence modeling for time - dependent rans and vles : a review , '' aiaa j. , vol . 36 , 1998 c.k . tam , webb , j.c . , dispersion - relation - preserving finite difference schemes for computational aeroacoustics , j. comp . 107 , 1993 . hussaini , j. manthey , low - dissipation and low - dispersion runge - kutta schemes for computational acoustics , j. comput . 124 , 1996 . cecora , r. radespiel , b. eisfeld , a. probst , differential reynolds - stress modeling for aeronautics , " aiaa j. , vol . 53 , 2015 . r. ewert , `` broadband slat noise prediction based on caa and stochastic sound sources from a fast random particle - mesh ( rpm ) method , '' comput . fluids , vol . 37 , 2008 .	in this contribution , we present an application of a novel perturbation approach ( denoted as an overset approach ) to a generic turbulent channel flow . the derivation of the governing equations ( the non - linear perturbation equations extended with fluctuating viscous terms ) is presented as well as subgrid - scale modeling aspects which results in the overset les ( oles ) method . the application of the overset method is illustrated with a generic fully developed turbulent channel flow , the results of which show excellent agreement with reference direct numerical simulation data . the ability of the overset method to correct for an imperfect background flow is furthermore shown .
the presence of aberrant conformations of the amyloid @xmath0 peptide and the protein @xmath1-synuclein is considered to be a key factor behind the development of alzheimer s and parkinson s diseases , respectively . the polymerization kinetics of these proteins has been shown to consist of nucleation and growth processes and to be strongly accelerated by the presence in solution of pre - existing fibrils @xcite , thereby circumventing the slow primary nucleation of aggregates . it was found that surfaces , such as lipid bilayers @xcite and hydrophobic nanoparticles @xcite can accelerate the nucleation process dramatically . indeed , in the case of @xmath1-synuclein , it was found that in the absence of suitable surfaces , the primary nucleation rate is undetectably slow @xcite . under certain conditions , the surfaces of the aggregates themselves appear to be able to catalyse the formation of new fibrils , leading to autocatalytic behavior and exponential proliferation of the number of aggregates @xcite . this so - called secondary nucleation process is likely to play an important role in the spreading of aggregate pathology in affected brains @xcite , as the transmission of a single aggregate into a healthy cell with a pool of soluble protein might be sufficient for the complete conversion of the soluble protein into aggregates . an intriguing idea is to exploit this observation to screen biological samples based on the presence of very low concentrations of aggregates for pre - clinical diagnosis of neurodegenerative diseases . indeed , this has been achieved in the case of the prion diseases in a methodology called prion misfolding cyclic amplification @xcite , which is based on the amplification of aggregates through repeated cycles of mechanically induced fragmentation and growth . recently , the applicability of this approach to the detection of aggregates formed from the amyloid @xmath0 peptide has been demonstrated @xcite . furthermore , the autocatalytic secondary nucleation of amyloid @xmath0 fibrils has been exploited to demonstrate the presence of aggregates during the lag phase of aggregation @xcite . however , none of these methods currently allow to easily determine the absolute number of aggregates in a given sample . one strategy to address this problem is to divide a given sample into a large number of sub volumes and determine for each of the sub - volumes whether it contains an aggregate or not . due to advances in microfluidic technology and microdroplet fabrication @xcite , it is now possible to monitor protein aggregation in micron - scale samples @xcite , a technique that could be used to design microarrays targeted for protein polymerization assays . to be successful this program needs guidance from theory to quantify possible measuring errors due to false positive and negative detection . current understanding of protein polymerization is based on mean - field reaction kinetics that have proved successful to describe key features of the aggregation process in macroscopic samples @xcite . this theory is , however , designed to treat the system in the infinite volume limit , where the intrinsic stochasticity of the nucleation processes can not manifest itself , so that its applicability to small volume samples is questionable . the importance of noise in protein aggregation was clearly illustrated in ref . @xcite , who proposed and solved the master equation kinetics of a model for polymer elongation and fragmentation , obtaining good agreement with experiments on insulin aggregation @xcite . here we address the problem by numerical simulations of a three dimensional model of diffusion - limited aggregation of linear polymers @xcite , including explicitly auto - catalytic secondary nucleation processes @xcite . a three dimensional model overcomes the limitations posed by both mean - field kinetics @xcite and master equation approaches @xcite , which do not consider diffusion and spatial fluctuations . most practical realizations of protein aggregation reactions are not diffusion limited , due to the slow nature of the aggregation steps , caused by significant free energy barriers @xcite . this leads to the system being well mixed at all times and mean - field theories providing a good description . there are , however , cases both _ in vitro _ ( e. g. when protein concentrations and ionic strengths are high , leading to gel formation @xcite ) and _ in vivo _ ( due to the highly crowded interior of the cell ) , where a realistic description can not be achieved without the explicit consideration of diffusion . here we use our model to study fluctuations in the aggregation process induced by small volumes and to provide predictions for the reliability of a seed detection assay . simulations are performed using a variant of the protein aggregation model described in ref . @xcite where individual protein molecules sit on a three dimensional cubic lattice . the model considers primary nucleation due to monomer - monomer interaction , polymer elongation due to addition of monomers to the polymer endpoints and secondary nucleation processes in which the rate of monomer - monomer interaction is enhanced when the process occurs close to a polymer ( see fig . [ fig:1]a for an illustration ) @xcite . in particular , monomers diffuse with rate @xmath2 and attach to neighboring monomers with rate @xmath3 ( primary nucleation ) but when a monomer is nearest neighbor to a site containing a polymer composed of at least @xmath4 monomers , then the nucleation rate increases to @xmath5 ( secondary nucleation ) . we do not consider polymer fragmentation , since this term is mostly relevant for samples under strong mechanical action @xcite , and some of the most important amyloid - forming proteins have been shown to exhibit aggregation kinetics dominated by secondary nucleation under quiescent conditions @xcite . a monomer can attach to a polymer with rate @xmath6 if it meets its endpoints . polymers move collectively by reptation with a length - dependent rate @xmath7 , where @xmath8 is the number of monomers in the polymer ( see ref . @xcite p. 89 ) , and locally by end rotations , with rate @xmath9 , and kink moves with rate @xmath10 ( for a review of lattice polymer models see @xcite ) . simulations start with a constant number of @xmath11 monomers in a cubic system of size @xmath12 ( with @xmath13 an integer ) where @xmath14 is the typical monomer diameter , with periodic boundary conditions in all directions . we perform numerical simulations using the gillespie monte carlo algorithm @xcite and measure time in units of @xmath15 and rates in units of @xmath16 . we explore the behavior of the model by varying independently both the monomer concentration @xmath17 and the number of monomers @xmath11 at fixed @xmath18 , but also the rate constants . for the simulations results reported in the following , the rates describing polymer motion are chosen to be @xmath19 which is smaller or equal than the diffusion rate of the monomers @xmath2 . as expected , secondary nucleation efficiently decreases the half - time before rapid polymerization . we illustrate this by changing the critical polymer size @xmath4 needed to induce secondary nucleation . we observe that the lower @xmath4 the shorter the half - time ( see fig . [ fig:1](b ) ) . currently , no experimental data exists on the value of @xmath4 , but it can be expected to be of a similar magnitude as the smallest possible amyloid fibril , defined as the smallest structure for which monomer addition becomes independent of the size of the aggregate and an energetic downhill event . the progress of reactions observed experimentally in bulk systems can be well approximated by a mean - field model @xcite , without fragmentation or depolymerization of polymers . such models are in contrast to our three dimensional computational model , which describes also monomer diffusion and polymer motion due to reptation , kink motion and end - rotations , which are not treated by mean - field approximation . despite this , it is still possible to fit polymerization curves resulting from three dimensional simulation results through mean - field theory with effective diffusion dependent parameters . the fact that both experimental and simulated polymerization curves are described by the same mean - field theory ensures that our model is appropriate to describe experiments . in the mean - field model , the evolution of the concentration @xmath20 of polymers of length @xmath21 , where @xmath22 is the nucleation size , is given by @xcite @xmath23 where dots indicate time derivatives and @xmath24 is the monomer concentration . the first term on the right - hand side represents an increase in the concentration of polymers of size @xmath22 due to polymer nucleation by aggregation of @xmath22 monomers with rate constant @xmath25 ; this is a generalized version of the dimer formation with rate constant @xmath3 in the 3d model . the second term represents an increase in the concentration of polymers of size @xmath26 by attachment of a monomer to a polymer of size @xmath27 , with rate constant @xmath28 . the third term is the corresponding loss of concentration of polymers of size @xmath26 when they attach to a monomer . these two terms are the mean - field equivalent of the endpoint attachment of monomers to polymers with rate constant @xmath6 in the 3d model . the final term represents secondary nucleation , which in the mean - field model is described as an increase in concentration of polymers of size @xmath29 ( the secondary nucleus size ) occurring at a rate proportional to the mass of polymers and with a rate constant @xmath30 . by conservation of mass , the evolution of the monomer concentration is @xmath31 the evolution of the number concentration @xmath32 and mass concentration @xmath33 can be found by summing over @xmath26 in . after some algebra , one obtains @xmath34 analytical approximation @xcite of the system of equations gives a solution that depends on two parameters , @xmath35 and @xmath36 . we fit our data with the form given in eq . 1 of ref . @xcite , using a least squares method . each curve is fitted independently . diffusion plays an important role in the aggregation progress , shifting the aggregation curves as shown in fig . [ fig:2]a and fig . [ fig:2]b . for a considerable parameter range , however , the time evolution of the fractional polymer mass can be fitted by mean - field theory ( lines in fig . [ fig:2]a and fig . [ fig:2]b ) with effective parameters that now depend on the diffusion rate @xmath2 ( see fig . [ fig:2]c and [ fig:2]d ) . similarly , mean - field theory describes the density dependence of the aggregation curves as shown in fig . [ fig:2]b . having confirmed that our computational model faithfully reproduces polymerization kinetics in macroscopic samples , we now turn to the main focus of the paper , the study of sample - to - sample fluctuations in small volumes , a feature that can not be studied with mean - field kinetics . when the sample volume is reduced , we observe increasing fluctuations in the aggregation kinetics as shown in figs . [ fig:3]a and [ fig:3]b . these results are summarized in fig [ fig:3]c showing the complementary cumulative distributions of half - times for different monomer numbers @xmath11 and constant monomer concentration @xmath37 where @xmath38 is the probability density function and @xmath39 is defined as the half - time of the polymerization curve ( i.e. the time at which @xmath40 ) . the steepness of the aggregation curves in figs . [ fig:3]a and [ fig:3]b suggests that , for @xmath41 , fluctuations are mostly ruled by the time of the first primary nucleation event , @xmath42 , whose complementary cumulative distribution @xmath43 can be estimated analytically as @xmath44 where @xmath45 is the average number of possible primary nucleation events per unit monomer . we estimate that @xmath46 using a poisson approximation , as we show in details in the following section . note that eq . [ eq : s0 ] displays a size dependence that is reminiscent of extreme value statistics @xmath47 where @xmath48 is a function that does not depend on @xmath11 @xcite . if @xmath49 , the half - time @xmath39 differs from the nucleation time by a weakly fluctuating time @xmath50 . this comes from the observation that , once the first primary nucleation event has happened , the polymerization follows rapidly , thanks to fast growth and secondary nucleation . this yields a weakly fluctuating delay @xmath51 , which in general depends on the number density @xmath18 and on the number of monomers @xmath11 . the distribution and average values of @xmath50 are reported in fig . [ figsalpha ] . the average value of @xmath50 decreases with @xmath18 and displays only a smaller dependence on @xmath11 . the distribution of @xmath50 is always peaked around its average but while at small values of @xmath18 the peak shifts with @xmath11 while the standard deviation remains constant , for higher values of @xmath18 only the standard deviation depends on @xmath11 and the peak position does not change . since the fluctuations in @xmath50 are much smaller than the fluctuations of @xmath42 we can safely assume that @xmath52 for @xmath53 , so the complementary cumulative distribution takes the form @xmath54 the prediction of eqs . [ eq : s0 ] and [ eq : slag ] are in agreement with numerical simulations results for @xmath43 and @xmath55 , respectively ( see figs . [ fig:3]c and [ figs0_32 ] ) . in particular , the behavior of @xmath43 is obtained without any fitting parameters , while @xmath55 only needs the estimate of the single parameter @xmath56 ( additional comparisons for different values of @xmath18 are reported in figs . [ figs0_16 ] and [ figs0_72 ] ) . the corresponding average values of @xmath57 are shown in the inset of fig . [ fig:3]c as a function of @xmath11 ( see also fig . [ figs0_32]b ) . in this section , we provide a detailed derivation of eqs . [ eq : s0 ] and [ eq : slag ] in the limit of relatively large diffusion when the system is well mixed . to this end , we consider a cubic lattice composed by @xmath58 nodes , in which @xmath11 monomers are placed randomly at time @xmath59 . as illustrated in fig . [ fig : lattice ] , each monomer @xmath8 sits near @xmath60 neighboring monomers and @xmath61 neighboring empty sites , where @xmath60 is in general a fluctuating time dependent quantity . in the model , each monomer @xmath8 can either diffuse into one of the @xmath61 empty sites or form a dimer with one of the @xmath60 neighboring monomers . therefore , at any given time @xmath62 the number of possible diffusion events in the system is @xmath63 and the number of possible aggregation events is @xmath64 , where the factor @xmath65 is needed to correct for the double counting of the number of monomer pairs . we can compute the time of first aggregation of @xmath11 monomers using poisson statistics , considering for simplicity the case in which the number of possible aggregation events @xmath66 would not depend on time . in this case , the probability of having an aggregation event within @xmath67 is @xmath68 . we can then divide the time interval @xmath42 in @xmath69 elementary time subintervals @xmath70 , the rate of aggregation event at @xmath42 , i.e. the probability per unit time to have the first dimer formed after a time interval @xmath42 has elapsed , is given by the following expression : @xmath71 as stressed previously , @xmath66 and @xmath72 are in principle fluctuating quantities and therefore eq.[p0_t0 ] is not valid . yet , as shown in fig . [ fig : ergodic ] : @xmath66 and @xmath72 are are both @xmath73 stationary , @xmath74 ergodic , @xmath75 weakly fluctuating and @xmath76 linearly dependent on @xmath11 , on average . hence , the probability @xmath77 for a monomer to form a dimer at @xmath42 can reasonably be approximated by its ensemble average @xmath78 where we have replaced @xmath66 by its average value @xmath79 and defined @xmath80 . from eq.([p_real_norm ] ) , we easily obtain the complementary cumulative distribution function @xmath81 recovering eq . [ eq : s0 ] . to conclude our calculation , we still need to evaluate @xmath82 . to this end , we perform a discrete enumeration of the possible configurations of a single monomer , in the spirit of cluster expansions for percolation models . in particular , the six relevant configurations for a single monomer in contact with other monomers are reported in fig . [ fig : lattice ] . the weight @xmath83 of a configuration in which a monomer has @xmath84 occupied neighbors is assumed to be given by the binomial distribution @xmath85 this single particle picture suggests that the average number of primary nucleation events per monomer @xmath45 , corresponds to the average number of nearest neighbors @xmath86 , divided by a factor 2 since any nucleation event encompasses 2 particles . with a similar reasoning , we estimate @xmath87 , i.e. the average number of empty directions . then , from the binomial distribution ( [ binomial ] ) we get @xmath88 and @xmath89 . using these values in eq . [ eq : s0 ] and eq.([eq : slag ] ) , we obtain agreement with numerical simulations as illustrated in figs . [ figs0_16 ] and [ figs0_72 ] ( panels ( a ) and ( b ) respectively ) for different values of @xmath18 . finally , we calculate the averages of the first aggregation time and the half - time as @xmath90 and @xmath91 , where @xmath92 . the expression for @xmath93 is given by @xmath94 in fig . [ figs0_32]b we show the perfect agreement of the theoretical estimate given by eq.([mean_t0 ] ) with the numerical values of the average time for the first primary nucleation event as a function of @xmath95 , for several densities . notice that no fitting parameters are involved . the average time of @xmath39 follows from @xmath96 : the inset of fig . [ fig:3]c confirms the agreement between the theoretical estimates ( dashed lines ) and the numerical values ( symbols ) . while the fluctuations we observe are intrinsic to the random nature of nucleation events , the ones usually encountered in bulk experiments are likely due to contamination or differences in initial conditions @xcite . in those bulk systems ( @xmath97l and larger ) , the number of protein molecules involved in the aggregation process is extremely large , even at low concentrations , so that we can exclude intrinsic kinetic fluctuations . for instance , a volume of 100@xmath97l at a concentration of 1@xmath97 m still contains @xmath98 monomers , leading to a large number ( hundreds to thousands ) of nucleation events per second for a realistic value of the nucleation rate @xcite . however , if the relevant volumes are made small enough ( pico- to nanolitres ) , the stochastic nature of primary nucleation can be directly observed . this has been exploited by aggregation experiments performed inside single microdroplets , where individual nucleation events could be observed , due to their amplification by secondary processes @xcite . in these experiments , the average half - time is found to scale with volume in a similar manner to what is shown in the inset of fig . [ fig:3 ] , thus in agreement with our simulations . we are now in a position to use our model to design a test _ in silico _ to detect the presence of pre - formed polymers , that act as seeds and nucleation sites for the secondary nucleation process , and that are thus amplified . as illustrated in fig . [ fig:6]a , the test considers a set of small volume samples containing protein solutions with a given concentration at time @xmath99 . the aim of the test is to detect the samples containing at least one seed ( case b in fig . [ fig:6]a ) . in an ideal experiment , the size of the microdroplets would be adjusted such that most droplets contain no seeds , some contain one seed , and the proportion of droplets containing more than one seed is negligible . in practice , these conditions can be easily adjusted experimentally by progressively reducing droplet volumes until only a small fraction of them display aggregates . after a fixed time @xmath100 , one can observe which samples contain macroscopic , detectable amounts of aggregates , enabling exact quantification of the number of aggregates present in the initial sample . ideally the test should be positive only when at least one seed was initially present , but given the large fluctuations intrinsic to the nucleation processes we demonstrated above , as well as the competition with _ de novo _ nucleation , there is a chance for false tests . in particular , a false positive test occurs when an unseeded sample is found to contain aggregates , while a false negative test corresponds to the case in which a seeded sample does not produce detectable amounts of aggregates within the time scale of the experiment . in fig . [ fig:6]b we report the complementary cumulative distribution of aggregation half - times @xmath39 as a function of the primary nucleation rate @xmath3 for samples with or without seeds , in this case a single pre - formed trimer . for small values of @xmath3 , seeded and unseeded samples yield distinct results , as also illustrated by the average half - times reported in fig . [ fig:6]c . as the value of @xmath3 increases , however , the distributions become closer in the two cases . in fig . [ fig:6]d , we quantify the fraction of false positive and false negative tests for two different testing times ( e.g. @xmath101 and @xmath102 ) . as expected , for large @xmath3 errors are very likely and the test would not be reliable . for intermediate values of @xmath3 , one can try to adjust @xmath103 to reduce possible errors with a caveat : decreasing @xmath103 reduces false positive errors , but at the same time increases false negatives ( fig . [ fig:6]d ) . it is , however , possible to optimize @xmath103 so that both types of errors are minimized . in an experimental realisation of such a setup , the most important system parameter that needs to be optimized for any given protein is the ratio of secondary nucleation rate to primary nucleation rate . due to potentially different dependencies of these two rates on the monomer concentration @xcite , ph @xcite and potentially other factors , such as temperature , salt concentration etc . , it is possible to fine tune this ratio and adjust it to a value that allows for an easy discrimination between droplets that do contain a seed aggregate and those that do not . in conclusion , we study protein polymerization in a three dimensional computational model and elucidate the role of protein diffusion in the polymerization process . most theoretical studies of protein aggregation neglect completely the role of diffusion and any other spatial effect . when the polymer diffusion and elongation rates are large enough we recover the standard polymerization curves that can also be obtained from mean - field analytical treatments and that can be used to fit , for example , kinetic data of amyloid @xmath0 aggregation @xcite . it would be interesting to explore if for small diffusion rates and small densities mean - field kinetics would eventually fail to describe the results , but this is a challenging computational task . at low densities , diffusion could play an important since a critical timescale would be set by the time needed by two monomers to meet before aggregating . this time - scale can be estimated considering the time for a monomer to cover a distance @xmath104 , yielding @xmath105 . this timescale is not relevant for our simulations since at the relatively high densities we study a considerable fraction of monomers are close to at least another monomer ( see @xmath66 in fig . [ fig : ergodic]a ) . consequently , the distribution of the first aggregation time does not depend on the diffusion rate @xmath106 ( see eq . [ eq : s0 ] ) . the half - time distribution , however , depends on diffusion even in this regime ( fig . [ fig:2 ] ) . our simulations show intrinsic sample - to - sample fluctuations that become very large in the limit of small volumes and low aggregation rates . we show that the corresponding half - time distributions are described by poisson statistics and display size dependence . as a consequence of this , the average half - times scale as the inverse of the sample volume , in agreement with insulin aggregation experiments performed in microdroplets @xcite and with calculations based on a master - equation approach @xcite . we use this result to design and validate _ in silico _ a pre - clinical screening test based on a subdivision of the macroscopic sample volume that will ultimately allow the determination of the exact number of aggregates that was initially present . this is the first step to develop microarray - based _ in vitro _ tests for early diagnosis of neurodegenerative diseases . gc , zb , atand sz are supported by erc advanced grant 291002 sizeffects . camlp thanks the visiting professor program of aalto university where part of this work was completed . sz acknowledges support from the academy of finland fidipro program , project 13282993 . akb thanks magdalene college , cambridge and the leverhulme trust for support . 27 ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1 '' '' @noop [ 0]secondoftwo sanitize@url [ 0 ] + 12$12 & 12#1212_12%12 @startlink[1 ] @endlink[0 ] @bib@innerbibempty link:\doibase 10.1016/0092 - 8674(93)90635 - 4 [ * * , ( ) ] link:\doibase 10.1073/pnas.1315346111 [ * * , ( ) ] link:\doibase 10.1074/jbc.m114.585703 [ * * , ( ) ] link:\doibase 10.1038/nchembio.1750 [ * * , ( ) ] link:\doibase 10.1021/ja505502e [ * * , ( ) ] link:\doibase 10.1073/pnas.0703306104 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1038/nature14547 [ * * , ( ) ] link:\doibase 10.1038/nprot.2012.067 [ * * , ( ) ] link:\doibase 10.1016/j.celrep.2014.02.031 [ * * , ( ) ] link:\doibase 10.1021/ja408765u [ * * , ( ) ] link:\doibase 10.1002/anie.200906653 [ * * , ( ) ] link:\doibase 10.1073/pnas.1105555108 [ * * , ( ) ] link:\doibase 10.1126/science.1178250 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3608917 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.098101 [ * * , ( ) ] link:\doibase 10.1038/ncomms4620 [ * * , ( ) ] link:\doibase 10.1002/anie.201108040 [ * * , ( ) ] link:\doibase 10.1063/1.3608916 [ * * , ( ) ] , ed.,@noop _ _ ( , ) link:\doibase 10.1016/0021 - 9991(76)90041 - 3 [ * * , ( ) ] @noop _ _ ( , ) @noop _ _ , edited by ( , ) link:\doibase 10.1016/j.ab.2010.02.001 [ * * , ( ) ] link:\doibase 10.1074/jbc.m010907200 [ * * , ( ) ] link:\doibase 10.1042/bse0560011 [ * * , ( ) ] link:\doibase 10.1073/pnas.1401564111 [ * * , ( ) ] a ) schematic of the protein aggregation model describing the main processes involved : primary nucleation occurring with rate @xmath3 ( and correspondingly @xmath107 in mean - field ) , polymer elongation with rate @xmath6 ( @xmath108 in mean - field ) , secondary nucleation with rate @xmath16 ( @xmath109 in mean - field ) and diffusion with rate @xmath2 which is not described by mean - field theory . b ) simulations showing the dependence on the minimal polymer size @xmath4 needed to catalyze secondary nucleation of the aggregation curve , the polymer mass fraction @xmath110 , obtained for @xmath111 , @xmath112 , @xmath113 , @xmath114 and @xmath115 . the dashed line is the curve obtained in the limit @xmath116 , or equivalently in absence of secondary nucleation . ] * protein aggregation depends on monomer diffusion : * a ) simulations showing the diffusion dependence of the aggregation curve , the polymer mass fraction @xmath110 , obtained for @xmath117 , @xmath118 , @xmath119 and @xmath120 . the curves are well fit by mean - field theory ( full lines ) with effective parameters that depend on @xmath2 . b ) simulations of the density dependence of the polymer mass fraction for @xmath119 , @xmath117 , @xmath121 and @xmath122 . fits by mean - field theory are plotted as lines with effective parameters reported in the panels c ) and d ) . time is measured in units of @xmath15 . c)-d):the effective mean - field parameters @xmath123 and @xmath124 obtained by fitting simulations performed for @xmath125 , @xmath118 as a function of the concentration @xmath18 and the diffusion rate @xmath2 . ] half - time sample - to - sample fluctuations are due to extreme value statistics . different replicates of the simulations display wide fluctuations in half - times especially for small numbers of monomers . a ) simulation results obtained for @xmath126 monomers at @xmath127 , @xmath128 , @xmath113 , @xmath114 , @xmath122 . the graph shows that the half - time @xmath39 is very close to the nucleation time @xmath42 at which the curves depart from zero . b ) same as panel a ) but with @xmath129 . c ) the complementary cumulative distributions of half - times obtained from simulations for different values of @xmath11 are in agreement with the theory described in the text . the inset shows that the average half - times for different concentrations @xmath18 as a function of @xmath11 . the general trend is in agreement with experiments @xcite . time is measured in units of @xmath15.,width=499 ] for four different monomer numbers @xmath11 and density @xmath127 . the symbols correspond to the numerical simulations while lines correspond to the theoretical predictions obtained from eq . b ) the average time @xmath130 as a function of @xmath95 for four different number densities . the theoretical predictions ( dashed lines ) are obtained from eq.(13 ) . here @xmath112 and @xmath131 , width=432 ] ( a ) and @xmath55 ( b ) for four different monomer numbers @xmath11 and density @xmath115 . the symbols correspond to the numerical simulations while the lines represent the theoretical predictions obtained from eq . ( 6 ) and eq . ( 7 ) . here @xmath112 and @xmath131 . , width=432 ] ( a ) and @xmath55 ( b ) for four different monomer numbers @xmath11 and density @xmath132 . the symbols correspond to the numerical simulations while he lines represent the theoretical predictions obtained from eq.(6 ) and eq.(7 ) . here @xmath112 and @xmath131 . , width=432 ] , obtained from numerical simulations , as a function of the number density @xmath18 for four different monomer numbers @xmath11 . for any @xmath11 and @xmath18 the averages are calculated over the different numerical simulations outcomes . the distributions of delay times @xmath50 as a function of the number of monomers @xmath11 for b ) @xmath115 and c ) @xmath133 . here @xmath112 and @xmath131.,width=432 ] and @xmath72 , respectively ) are a linear function of the number of monomers @xmath11 . b ) the corresponding frequencies @xmath82 and @xmath136 fluctuate very little in time . here @xmath112 , @xmath131 , @xmath119 and @xmath115 . ] intrinsic fluctuations rule errors in the detection of protein aggregation prone samples . a ) a test to detect seeds for protein aggregation is based on small volume sampling of protein solutions at time @xmath99 and on the assumption that only seeded samples ( e. g. sample b ) would form aggregate at time @xmath103 . b ) simulations allow to compute the distribution of half - times for samples with and without a seed as a function of the rate of primary nucleation @xmath135 . data are obtained sampling over @xmath137 independent realization . c ) average half - time ( @xmath138 standard deviation ) as a function of the rate of primary nucleation @xmath135 . d ) fraction of false positives ( fp ) and false negatives ( fn ) for testing times @xmath102 and @xmath101 . time is measured in units of @xmath15 , @xmath129 . ]	autocatalytic fibril nucleation has recently been proposed to be a determining factor for the spread of neurodegenerative diseases , but the same process could also be exploited to amplify minute quantities of protein aggregates in a diagnostic context . recent advances in microfluidic technology allow analysis of protein aggregation in micron - scale samples potentially enabling such diagnostic approaches , but the theoretical foundations for the analysis and interpretation of such data are so far lacking . here we study computationally the onset of protein aggregation in small volumes and show that the process is ruled by intrinsic fluctuations whose volume dependent distribution we also estimate theoretically . based on these results , we develop a strategy to quantify in silico the statistical errors associated with the detection of aggregate containing samples . our work opens a new perspective on the forecasting of protein aggregation in asymptomatic subjects .
since neutrinos interact only weakly with matter , neutrino observatories provide a powerful tool for exploring the deepest reaches of stars and galaxies . in recent years there has been an upswell in the number of experiments , either running or about to run , that are aimed at detecting ultra - high energy ( uhe ) neutrinos , neutrinos with energies from the multi - tev range to beyond the eev range . their existence is predicted by a number of theoretical models . although the production mechanism can be different - either as a byproduct of creation and decay of pions in the interaction between high energy primary cosmic rays and the cosmic microwave background , as predicted by gzk @xcite , or as products of the same astrophysical sources that generate the cosmic rays observed in the highest energy air showers @xcite -there is wide expectation in the community that neutrinos of galactic or cosmic origin will be detected in the uhe energy range in the near future . many experimental limits on astrophysical neutrino fluxes have already been established . frejus , baikal , amanda and macro @xcite have reported limits on neutrinos from astrophysical sources in the @xmath8 gev range , while fly s eye , agasa , amanda and rice @xcite have given limits in the range @xmath9 gev , and agasa , fly s eye , rice , forte , anita and glue @xcite have reported limits in the range above @xmath10 gev . in the near future , icecube @xcite , anita and auger @xcite can be expected to release stronger results based on more recent data . in the longer term , proposed expansions of icecube , such as aura @xcite , or a salt - based radio telescope like salsa @xcite could afford substantially enhanced sensitivity . further in the future , an orbiting telescope like the proposed euso project @xcite opens the possibility to achieve a huge effective volume . the design of neutrino telescopes depends critically on the estimates of neutrino cross sections in the uhe regime , most of which is far beyond currently available data . moreover , the systematic study of the scattering of such high energy neutrinos with baryonic matter might prove to be an essential instrument to test new physics effects , such as those given by models of low scale gravity ( lsg ) . models of extra dimensions @xcite have lowered the characteristic quantum - gravity scale to energies comparable with the electro - weak scale , enhancing the expected neutrino cross sections at gzk - energies and above , even for conservative parameter choices . efforts at realistic calculations of the uhe neutrino - nucleon cross section have been performed in the framework of qcd and the standard model @xcite , and many speculative calculations based on new physics effects have been presented over the past decade or more . the only detailed calculations of neutrino event rates within the framework of the lsg models that include the neutrino - proton eikonal cross section with lsg graviton exchange , rely upon the saddle point approximation @xcite . in this article we present in detail the calculation for the full eikonal approximation of the lsg neutrino - proton cross section , including a study of the effects of a finite tension on the brane @xcite . we comment on the inapplicability of the saddle point approximation at neutrino energies below @xmath11 gev . in sec . ii we summarize the main procedure of the eikonal approximation ( ea ) in quantum field theory and apply it to the calculation of the desired differential cross sections at the parton level in the arkani - hamed , dimopoulous and dvali ( add ) model of extra dimensions @xcite . we consider the cases of an infinitely thin standard model brane and the one where it has finite thickness . in sec . iii we present the full ea calculation and compare the results with those in the saddle point approximation , and with the standard model cross section . iv is dedicated to the summary of results , comments on the sensitivity of results to choice of gravity scale and brane tension and on the origin of the discrepancy between the saddle point and full ea calculation . we also comment on the impact of our cross section results on neutrino telescope event rate expectations . the appendix presents details of the representation of the eikonal amplitude in terms of meijer g - functions . the eikonal approximation is a technique widely used in quantum mechanics and wave physics to derive an expression for elastic scattering in the limit of large center of mass ( cm ) energies ( see for example reference @xcite ) . in quantum field theory the amplitude can be obtained by summation of all the ladder and cross - ladder diagrams for boson exchange at all order @xcite , as exemplified in fig . 1 . the approximation is valid for center of momentum energies that are high with respect to @xmath12 , the absolute magnitude of the momentum transfer @xmath13 , i.e. @xmath14 in the usual mandelstam variables for the @xmath15 scattering . the high energy scattering angle is small and the four - momenta of the incoming particles are approximately equal to those of the corresponding outgoing particles , @xmath16 and @xmath17 . for any given kind of boson exchanged ( scalar , vector , tensor ) , the vertex and external legs factor is independent from the spin of the incoming particles , as one can neglect their mass and the recoil of the matter field . with these assumptions , @xmath18 and the tree - level amplitude ( fig . 2a ) , or born term for the exchange of one intermediate boson is given by @xmath19 where @xmath20 is the dimension - dependent interaction coupling constant , @xmath21 is the mass of the exchanged particle and @xmath22 is its spin . at one - loop level ( fig . 2b ) , the ea simplifies the matter propagators : @xmath23 where @xmath24 is the mass of the incoming particles . the one - loop eikonal amplitude , obtained when the above propagators are put on shell by the cutkowsky rule , is imaginary . the integral over the transferred four - momentum reduces to an integral over its components in the plane perpendicular to the momenta of the incoming particles , @xmath25 . one gets for the eikonal at one loop : @xmath26\nonumber\\ & = & -2si\int d^{2}b_{\perp}e^{iq_{\perp}\cdot b_{\perp}}\left(-\frac{1}{2}\chi^{2}\right)\label{one_loop}\end{aligned}\ ] ] where the eikonal phase @xmath27 is the fourier transform of the born term . the procedure can be repeated with higher order diagrams @xcite , such that one can sum all the terms in @xmath28 to obtain : @xmath29 it is worth noting at this point of our discussion that the validity of the ea in describing the asymptotic behavior of elastic scattering has been long debated . references @xcite and @xcite argue that the eikonal is not a correct description of the high energy limit in the cases of scalar particle exchange , abelian vector exchange and non - abelian vector exchange . in the cases of vector particle exchange , when we put @xmath30 in eq . ( [ gen_born ] ) , the ea describes accurately the asymptotic behavior of the sum of exchange diagrams , but fails to consider the inelastic diagrams related by unitarity to particle production , of which fig . 2c is an example . as has been shown by @xcite , such diagrams become dominant at eighth order in @xmath20 . on the other hand , the @xmath31 dependence of the graviton - exchange born term in the quantum gravity ea renders the sum of exchange diagrams dominant with respect to the inelastic unitarity diagrams ( @xcite and references therein ) . consistent with the state of present knowledge , we will consider the eikonal a good approximation of the asymptotic behavior for graviton exchange . in models of lsg with large extra dimensions @xcite , graviton exchange ( @xmath32 ) is ultraviolet divergent already at tree - level @xcite . this is due to the presence of an infinite tower of massive kaluza - klein modes over which the amplitude has to be summed . for a number @xmath7 of extra - dimensions we have , approximating the sum by an integral , @xmath33 where @xmath3 is the @xmath34-dimensional plank mass ( where @xmath35 ) and @xmath36 is the surface of a @xmath7-dimensional unit sphere . as the notation indicates , this sum is adopted as the `` born input '' to the eikonal calculation @xcite . in the literature , there are two common solutions to the problem of tree - level divergencies . one can introduce a cut - off of the order of the low gravity scale , as this is the limit below which the effective theory is deemed to be valid @xcite . or , one can consider the physically reasonable case where the brane has a finite tension @xmath6 , corresponding to an extension @xmath37 of the standard model fields along the extra dimensions @xcite . in this case , if the extra - dimensional part @xmath38 of the standard model wave function has a gaussian cut - off of the kind @xmath39 eq . ( [ lsg_born ] ) is modified to @xmath40 the born term in eq . ( [ lsg_born ] ) can be calculated by dimensional regularization @xcite : @xmath41 substitution of the latter into eq . ( [ chi ] ) yields @xmath42 \frac{1}{b^{n}}\equiv\left(\frac{b_{c}}{b}\right)^{n}\label{grw_chi}\ ] ] where we ve followed the commonly simplified notation @xmath43 , @xmath44 . the critical impact parameter @xmath45 separates the region of space where the born term is dominant ( @xmath46 ) from the one where higher order terms in @xmath28 become relevant . expression ( [ grw_chi ] ) is clearly finite for all @xmath47 . thus , it can be seen from eqs . ( [ lsg_born])-([grw_chi ] ) that when the brane is infinitely thin ( @xmath48 ) the ultraviolet divergence gets absorbed by the eikonal phase calculation . each regularization procedure leads to the same finite result @xcite . the eikonal amplitude is then given by @xmath49 with @xmath50 and @xmath51 the bessel function of order zero.as mentioned in passing in @xcite , but not pursued , the @xmath52 can be expressed in terms of meijer g - functions . we implement our calculation explicitly in these terms , providing details in the appendix . the result is @xmath53 where the real and imaginary part are given by @xmath54 and @xmath55 where @xmath56 are the meijer g - functions @xcite . in the limit of large momentum transfer with respect to the inverse impact parameter ( @xmath57 ) , the amplitude of eq . ( [ grw_eik ] ) can be approximated analytically by the steepest descent method ( @xcite ) and it reads @xmath58^{\frac{n+2}{n+1}},\label{sad_point}\ ] ] where @xmath59^{\frac{1}{n+1}},\label{spampl}\ ] ] @xmath60^{\frac{n}{n+1}},\label{spphase}\ ] ] and @xmath45 is given by eq . ( [ grw_chi ] ) . the location of the saddle point , @xmath61 = @xmath45 @xmath62 , should satisfy @xmath63 in order for the saddle point to give the dominant contribution to the amplitude . on the other hand , in the case of a finite tension @xmath6 in the brane , substitution of eq . ( [ born_tens ] ) into eq . ( [ chi ] ) yields @xcite : @xmath64 where @xmath65 is a confluent hypergeometric function of the second kind @xcite . by substitution of eq . ( [ sjod_chi ] ) into eq . ( [ eikonal ] ) the eikonal amplitude can be calculated easily : @xmath66,\label{sjod_eik}\ ] ] where in the last expression we have used the functional identity : @xmath67 with @xmath68 , @xmath69 . we note here that , to our knowledge , there is no saddle point approximation for the extended brane case in the literature . nor we were able to find for eq . ( [ sjod_eik ] ) a simple , useful closed - form like ( [ sad_point])-([spphase ] ) , as the solution requires the numerical treatment of a transcendental equation . the differential cross section at the parton level can now be calculated with the usual substitution @xmath70 , where @xmath71 is the fraction of momentum of the proton carried by the parton and @xmath72 is the inelasticity . we get for the infinitely thin brane : @xmath73 with @xmath74 given by eq . ( [ grw_f]).for the brane @xmath37-thick : @xmath75 with @xmath76.\label{sjod_g}\ ] ] the @xmath77 are the usual parton distribution functions ( pdf ) summed over all the quark and antiquark flavors and the gluons . note here that the parameter @xmath78 of eq . ( [ grw_chi ] ) has to be calculated with respect to the center of mass energy squared of the neutrino - parton system @xmath79 . the total cross section is obtained by integration of eqs . ( [ grw_xsect ] ) and ( [ sjod_xsect ] ) . the @xmath72-integral has been performed with mathematica using the cteq5 pdfs @xcite , in the interval @xmath80 $ ] , where @xmath81 , @xmath82 is the mass of the @xmath83-boson and @xmath84 is the schwarzschild radius for black - hole production in the @xmath7-dimensional theory @xcite . we adopt @xmath85 as the scale parameter in the pdf s . the upper limit of integration is introduced since the eikonal cross section is not applicable for @xmath86 . the lower limit , as in standard model calculations , allows us to be safely inside the perturbative qdc regime . the @xmath71-integral has been performed by dividing the interval @xmath87 $ ] into 38 logarithmic subintervals and performing simpson s rule . this procedure gives an accurate evaluation of the cross section , which is reasonably smooth in @xmath71 . we consider the case of @xmath0 extra - dimensions , which is sufficient to illustrate all our points , since by changing @xmath7 the calculations differ by amounts small compared to changes due to @xmath3 , @xmath6 . in fig . 3 we show the comparison between our calculation of the proton - neutrino cross section in the full ea with the saddle point approximation of the same cross - section given by eqs . ( [ sad_point ] ) , ( [ spampl ] ) and ( [ spphase ] ) ) by the born term for @xmath88 . for a detailed treatment of the born term , see the first paper of ref . the brane is here assumed to be infinitely thin , with the ( 4 + 5)-dimensional planck scale set at @xmath89 tev . in the range @xmath90 gev , where the ea is expected to be valid , the full eikonal cross section is significantly larger than its evaluation at the saddle point . the difference ranges from a factor 2.6 at the low end to about 1.5 at the high end . as discussed in the next section , the conditions for the dominance of the saddle point evaluation are not satisfied until extremely high energies , explaining the slow convergence of the saddle point value to the full eikonal result . we point out here that the calculation of the saddle point approximation at @xmath91 differs from the one at @xmath0 by @xmath92 over the entire energy range , in accordance with the findings of ref . as it will be even clearer in what follows , this is a small amount in comparison to how the other parameters affect the results . , @xmath93 gev , @xmath94 , @xmath0 . dashed light line : saddle point approximation corrected by the born term . solid dark line : standard model neutral current.,width=302,height=226 ] as also shown in fig . 3 , even with the saddle point approximation to the lsg eikonal , the neutral current cross section in this model exceeds the standard model values @xcite at cm energies above the gravity scale , corresponding to roughly @xmath95 gev for @xmath3 = 1 tev @xcite . the dominance is even more striking when the full eikonal is used : at neutrino energies greater than @xmath96 gev new physics effects raise the cross section by almost three orders of magnitude . the presence of a thick brane with a tension @xmath1 tev sharply reduces the cross section , due to the effective gaussian cut - off in the born amplitude of eq . ( [ born_tens ] ) . the lower tension suppresses the eikonal cross section by an order of magnitude at the low energy end and a factor of two at the high end , as shown in fig . 4 . as the tension rises , and the brane gets thinner , the cross section should approach the infinitely thin limit . we already obtain an almost perfect overlapping when the tension is @xmath2 tev . gev for branes with different tension . ( not corrected by the born term . ) solid light : @xmath94 , @xmath0 . dashed light : @xmath97 gev , @xmath0 . solid dark : standard model nc.,width=302,height=226 ] fig . 5 shows the reduction to the cross sections due to a higher planck scale @xmath98 tev . we here display the cases of the infinitely thin brane and of the @xmath1 tev - tension brane . we see that the increase in scale to 2 tev suppresses the cross section an order of magnitude over the entire energy range for the thin brane case ( solid light line ) . the relative suppression is even bigger in the case of a thick brane ( dashed light line ) , where the cross section is lowered to values beneath the standard model calculation at energies less than @xmath10 gev . gev for branes with different tension . ( not corrected by the born term . ) solid light : @xmath94 , @xmath0 . dashed light : @xmath97 gev , @xmath0 . solid dark : standard model nc.,width=302,height=226 ] we have numerically calculated the parton level cross - sections for uhe neutrino - nucleon gravitational scattering in the eikonal approximation , under the assumption that the standard model particles are confined in a 4-dimensional brane embedded in a higher number of compactified extra - dimensions ( add model ) . the analytical calculations are presented in the most general case of @xmath7 extra - dimensions , but the numerical calculations were performed for @xmath0 to illustrate the magnitude of the effects for a value of the extra dimensions that is only weakly constrained by current laboratory and astrophysical data ( @xmath99 ) . lower dimensions are required by astrophysical data to have corresponding gravity scales too high to allow significant effects on the uhe cross sections , whereas @xmath99 are constrained only by collider data , which put lower limits on the gravity scale less than a tev @xcite . the cross sections are only mildly dependent on @xmath7 , at any rate @xcite . previous applications of the add - type models to the uhe neutrino cross section and expectations for event rates relied on the saddle point estimation of the eikonal amplitude , and only in the thin brane limit . our application completes the picture of the impact of low scale gravity on the predictions of high energy neutrino - nucleon cross sections . we reach the same qualitative conclusions as previous studies : the effects are sensitive to the scale , or radius , of the extra dimensions but much less sensitive to the actual number of dimensions . we also find that the brane tension , or thickness , greatly affects the size of lsg influence on uhe neutrino scattering , weakening sharply as brane tension decreases , or equivalently as brane thickness increases . this is in agreement with similar predictions for hadron collider applications @xcite . quantitatively speaking , we have found that the presence of new physics at lsg @xmath93 gev enhances the `` neutral current '' cross section compared to the standard model cross section by three order of magnitude at cm energies of @xmath96 gev and above , and that the full eikonal calculation of the cross section gives values consistently higher than the corresponding saddle point approximation , when the brane is considered to be infinitely thin . the enhancement above the standard model expectation is not nearly as marked in the case of a brane physically extended @xmath100 gev@xmath101 along the extra - dimensions , or in the case where the lsg planck mass is increased by a factor of two . however , we found that calculations of the cross sections performed with brane tensions of 10 tev and above are indistinguishable , indicating that a ratio of tension to gravity scale of 10:1 is effectively `` infinite '' for these purposes . the preceding statement is quite @xmath7-independent as can be seen from the r.h.s . of eq . ( [ sjod_chi ] ) : since in the region of validity of the ea ( @xmath102 ) the impact parameter @xmath103 is of the order of @xmath45 , one can see that when @xmath104 the hypergeometric function @xmath105 tends rapidly to @xmath106 . the eikonal phase is therefore of order one , as expected , independently from the number of extra - dimensions . correspondingly , the approach to the thin brane limit is only weakly dependent on @xmath7 . to understand the difference between the uhe cross section evaluated in the saddle point approximation versus the full eikonal calculation , we recall that the saddle point dominance of the integral in question requires @xmath107 . in our eikonal evaluation , as in those in ref . @xcite , we required @xmath108 . let us check the consistency of these two requirements . defining @xmath109 , we find that @xmath110 as @xmath111 runs from @xmath112 gev to @xmath113 gev for @xmath3 = @xmath114 gev and @xmath0 . the ranges are @xmath115 and @xmath116 for @xmath117 and @xmath91 respectively . moreover , the saddle point value of the impact parameter , @xmath118 is only slightly less than @xmath45 for these parameter choices , failing to satisfy the condition @xmath119 needed for saddle point dominance of the phase integral , eq . ( [ grw_f ] ) . these considerations tell us that , even with the maximal @xmath120 consistent with use of the eikonal , the strong coupling , saddle point conditions are , at best , only marginally satisfied even at the highest , super gzk energies we consider . the sensitivity of the calculation to @xmath6 and @xmath3 values is much more important to the estimate of event rates than the sensitivity to the use of the saddle point approximation , rather than the full ea . since collisions that satisfy the ea conditions tend to be highly elastic , the consequences for event rate estimates based on the model assumptions are not severe , since the fraction of deposited energy is small , and the corresponding events are mixed with highly inelastic events at neutrino energies an order of magnitude or more lower , where the flux is presumably much larger . thus the factor of ten difference in cross section between the @xmath121 1 tev and 2 tev , or between the infinite and finite tension at fixed lsg is , in a sense , part of uncertainty in a `` background '' correction to inelastic processes , such as black hole formation , where the neutrino energy is nearly all converted to visible energy . only if black hole formation is severely suppressed , does the enhanced signal from elastic processes become a sensitive tool to detect lsg . in summary , our conclusions are that , in agreement with other studies in collider settings , the lsg effects are sensitively dependent on the choice of scale parameters - the gravity scale and the brane tension . the eikonal neutrino - nucleon cross section always rises sharply above the standard model versions , but the energy at which the onset occurs and the degree to which the new physics dominates , is strictly dictated by the input scales . in the regime we consider , @xmath122 gev , the saddle point dominance conditions are not well satisfied , and the approximation , though lending itself to simple expressions and intuitive pictures , noticeably underestimates the cross section . having said that , the uncertainty in @xmath3 and @xmath6 values is nevertheless the dominant effect . _ acknowledgements : _ we thank danny marfatia for many helpful conversations on this work . e. s. is supported in part by nsf grant no . d.w.m . receives support from doe grant no . de - fg02 - 04er41308 . the integral of eq . ( [ grw_f ] ) can be expressed as a mellin convolution : @xmath123 where @xmath124 we can calculate the mellin transforms of eqs . ( [ f1 ] ) , ( [ f2 ] ) : @xmath125 @xmath126 upon defining @xmath127 , the integral of eq . ( [ mellin ] ) turns out to be a _ mellin - barnes _ integral : @xmath128 that can be expressed explicitly in terms of meijer g - functions . the real part of eq . ( [ mellin_barnes ] ) is @xmath129 } { \gamma\left(1-\frac{s}{2}\right)}y^{-s}ds\nonumber\\ & = & \frac{1}{2\pi i}\oint\frac{2^{s-1}\pi\gamma\left(\frac{s}{2}\right)\gamma \left(\frac{s-2}{n}\right ) } { n\gamma\left(1-\frac{s}{2}\right)\gamma\left(\frac{n-1}{n}+\frac{s}{2n}\right ) \gamma\left(1-\frac{n-1}{n}-\frac{s}{2n}\right)}y^{-s}ds,\label{refstep1}\end{aligned}\ ] ] where in the last step we have used euler s duplication formula ( @xmath130 $ ] ) . in the standard definition of meijer g - functions @xcite the coefficient multiplying @xmath131 inside the gamma functions has to be @xmath132 . we therefore perform the substitution @xmath133 to get @xmath134^{-s}ds,\label{refstep2}\ ] ] and use some known identities of the gamma function to recast eq . ( [ refstep2 ] ) into @xmath135 the integral is now in the form of a meijer g - function : @xmath136\equiv\\ -2^{-\frac{2}{n}-1}\pi^{\frac{1}{2}}n^{-1}\textrm { } \times g^{n+1,0}_{0,2(n+1)}\left(\frac{y^{2n}}{2^{2n+2}n^{2n}}\left\| \begin{array}{c } \\ 0,\frac{n-2}{2n},\frac{1}{n}, ... ,\frac{n-1}{n},-\frac{1}{n},0,\frac{n-1}{n},\frac{n-2}{n}, ... ,\frac{1}{n } \end{array } \right.\right ) . \label{finalref}\end{gathered}\ ] ] a similar calculation for the imaginary part yields @xmath137\equiv\\ 2^{-\frac{2}{n}-1}\pi^{\frac{1}{2}}n^{-1}\textrm { } \times g^{n+1,0}_{0,2(n+1)}\left(\frac{y^{2n}}{2^{2n+2}n^{2n}}\left\| \begin{array}{c } \\ 0,\frac{1}{n}, ... ,\frac{n-1}{n},\frac{n-1}{n},-\frac{1}{n},0,\frac{n-2}{2n},\frac{n-2}{n}, ... ,\frac{1}{n } \end{array } \right.\right).\label{finalimf}\end{gathered}\ ] ] in eqs . ( [ finalref ] ) and ( [ finalimf ] ) , the first line refers to the mathematica input necessary to perform the calculation , while the second line refers to the standard mathematical notation of eqs . ( [ im_meij ] ) and ( [ re_meij ] ) @xcite . 99 k. greisen , phys . * 16 * , 748 ( 1966 ) . g. t. zatsepin and v. a. kuzmin , jetp lett . * 4 * , 78 ( 1966 ) [ pisma zh . * 4 * , 114 ( 1966 ) ] ; v. s. berezinsky and g. t. zatsepin , phys . * 28b * ( 1969 ) ; sov . . phys . * 11 * , 111 ( 1970 ) . p. biermann and g. sigl , lect . notes phys . * 576 * , 1 ( 2001 ) [ arxiv : astro - ph/0202425 ] . w. rhode _ et al . _ [ frejus collaboration ] , astropart . phys . * 4 * , 217 ( 1996 ) ; v. balkanov _ et al . _ [ baikal collaboration ] , arxiv : astro - ph/0105269 ; j. ahrens _ et al . _ , phys . lett . * 90 * , 251101 ( 2003 ) [ arxiv : astro - ph/0303218 ] ; m. ambrosio _ et al . _ [ macro collaboration ] , astropart . * 19 * , 1 ( 2003 ) [ arxiv : astro - ph/0203181 ] . r. m. baltrusaitis _ et al . _ , phys . d * 31 * , 2192 ( 1985 ) ; s. yoshida _ et al . _ , _ prepared for 27th international cosmic ray conference ( icrc 2001 ) , hamburg , germany , 7 - 15 aug 2001 _ ; s. hundertmark [ amanda collaboration ] , _ prepared for 28th international cosmic ray conferences ( icrc 2003 ) , tsukuba , japan , 31 jul - 7 aug 2003 _ ; i. kravchenko _ et al . _ , _ prepared for 28th international cosmic ray conferences ( icrc 2003 ) , tsukuba , japan , 31 jul - 7 aug 2003_. i. kravchenko _ [ rice collaboration ] , phys . d * 73 * , 082002 ( 2006 ) ; s.w . et al . _ [ anita collaboration , phys . lett . * 96 * , 171101 ( 2006 ) ; n. letinen _ et al . _ [ forte collaboration ] , phys . rev . * d 69 * , 013008 ( 2004 ) ; p. gorham _ et al . _ , [ glue collaboration ] , phys . * 93 * , 041101 ( 2004 ) ; `` limits on the diffuse neutrino flux from the hires data '' , w. deng _ et al _ [ hires collaboration ] , proceedings of the 29th international cosmic ray conference . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * , 263 ( 1998 ) [ arxiv : hep - ph/9803315 ] . l. randall and r. sundrum , phys . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 436 * , 257 ( 1998 ) [ arxiv : hep - ph/9804398 ] . g. m. frichter , d. w. mckay and j. p. ralston , phys . lett . * 74 * , 1508 ( 1995 ) [ erratum - ibid . * 77 * , 4107 ( 1996 ) ] [ arxiv : hep - ph/9409433 ] . r. gandhi , c. quigg , m. h. reno and i. sarcevic , astropart . phys . * 5 * , 81 ( 1996 ) [ arxiv : hep - ph/9512364 ] ; r. gandhi , c. quigg , m. h. reno and i. sarcevic , phys . d * 58 * , 093009 ( 1998 ) [ arxiv : hep - ph/9807264 ] . p. jain , s. kar , d. w. mckay , s. panda and j. p. ralston , phys . d * 66 * , 065018 ( 2002 ) [ arxiv : hep - ph/0205052 ] ; s. hussain and d. w. mckay , phys . d * 69 * , 085004 ( 2004 ) [ arxiv : hep - ph/0310091 ] ; s. hussain and d. w. mckay , phys . b * 634 * , 130 ( 2006 ) [ arxiv : hep - ph/0510083 ] . m. sjodahl and g. gustafson , eur . j. * c53 * , 109 ( 2008 ) [ arxiv : hep - ph/0608080 ] ; l. lonnblad and m. sjodahl , jhep * 0610 * , 088 ( 2006 ) [ arxiv : hep - ph/0608210 ] . sakurai , `` modern quantum mechanics '' , _ menlo park , ca : the benjamin / cummings publishing company ( 1985 ) _ d. kabat and m. ortiz , nucl . b * 388 * , 570 ( 1992 ) [ arxiv : hep - th/9203082 ] . d. kabat , comments nucl . * 20 * , 325 ( 1992 ) [ arxiv : hep - th/9204103 ] . g. f. giudice , r. rattazzi and j. d. wells , nucl . b * 544 * , 3 ( 1999 ) [ arxiv : hep - ph/9811291 ] . t. han , j. d. lykken and r. j. zhang , phys . d * 59 * , 105006 ( 1999 ) [ arxiv : hep - ph/9811350 ] . r. emparan , phys . d * 64 * , 024025 ( 2001 ) [ arxiv : hep - th/0104009 ] . g. f. giudice , r. rattazzi and j. d. wells , nucl . b * 630 * , 293 ( 2002 ) [ arxiv : hep - ph/0112161 ] . r. emparan , m. masip and r. rattazzi , phys . d * 65 * , 064023 ( 2002 ) [ arxiv : hep - ph/0109287 ] .	we calculate low scale gravity effects on the cross section for neutrino - nucleon scattering at center of mass energies up to the greisen - zatsepin - kuzmin ( gzk ) scale , in the eikonal approximation . we compare the cases of an infinitely thin brane embedded in @xmath0 compactified extra - dimensions , and of a brane with a physical tension @xmath1 tev and @xmath2 tev . the extra dimensional planck scale @xmath3 is set at @xmath4 gev and @xmath5 gev . we also compare our calculations with neutral current standard model calculations in the same energy range , and compare the thin brane eikonal cross section to its saddle point approximation . new physics effects enhance the cross section by orders of magnitude on average . they are quite sensitive to @xmath6 and @xmath3 choices , though much less sensitive to @xmath7 .
the measurement of the nuclear structure function in deep inelastic electron - nucleus scattering ( dis ) is the best way to improve our knowledge of the nuclear parton distributions and qcd dynamics in the high energy regime ( see , e.g.@xcite ) . however , after more than 30 years of experimental and theoretical studies , a standard picture of nuclear modifications of structure functions and parton densities has not yet emerged . fixed target dis measurement on nuclei revealed that the ratio of nuclear to nucleon structure functions ( normalized by the atomic mass number ) is significantly different from unity . in particular , these data demonstrate an intricate behavior , with the ratio being less than one at large @xmath3 ( the emc effect ) and at small @xmath3 ( shadowing ) and larger than one for @xmath4 ( antishadowing ) . the existing data were taken at lower energies @xcite and therefore the perturbative qcd regime ( @xmath5 gev@xmath6 ) was explored only for relatively large values of the ( bjorken ) @xmath3 variable ( @xmath7 ) . experimentally , this situation will hopefully change with a future high energy electron - ion collider ( eic ) ( for recent reviews see , e.g. @xcite ) , which is supposed to take data at higher energies and explore the region of small @xmath3 ( @xmath8 ) in the perturbative qcd regime . the theory of nuclear effects in dis is still far from being concluded . the straightforward use of nucleon parton distributions evolved with dglap equations and corrected with a nuclear modification factor determined by fitting the existing data as in refs . @xcite is well justified only in the large @xmath9 region and not too small @xmath3 . moreover , these approaches do not address the fundamental problem of the origin of the nuclear shadowing and can not be extended to small @xmath3 , where we expect to see new interesting physics related to the non - linear aspects of qcd and gluon saturation ( for reviews see ref . currently , there are several phenomenological models which predict different magnitudes for the shadowing in the nuclear structure function based on distinct treatments for the multiple scatterings of the partonic component of the virtual photon , assumed in general to be a quark - antiquark ( @xmath10 ) color dipole . some works @xcite address the origin of the nuclear shadowing through the glauber - gribov formalism @xcite in the totally coherent limit ( @xmath11 , where @xmath12 is coherence length ) , which considers the multiple scattering of the color dipole with a nucleus made of nucleons whose binding energy is neglected . in the high energy limit , the eikonal approximation is assumed , with the dipole keeping a fixed size during the scattering process . in this approach the total photon - nucleus cross section is given by _ ^a = d^2r dz \|(r , z)\|^2 _ da(x , r ) [ sigga ] where @xmath13 is the probability of the photon to split into a @xmath14 pair of size @xmath15 and @xmath16 is the dipole - nucleus cross section , which is expressed as @xcite _ da(x , r ) = d^2b 2 [ sigda ] with @xmath17 being the nuclear thickness function and @xmath18 is the dipole - proton cross section . it must be stressed that once @xmath18 is fixed , the extension to the nuclear case is essentially parameter free in this approach . in the glauber formula ( [ sigda ] ) it is assumed that the dipole undergoes several elastic scatterings on the target . although reasonable and phenomenologically successful this assumption deserves further investigation . this model can be derived in the classical approach of the color glass condensate formalism @xcite . another approach largely used in the literature is based on the connection between nuclear shadowing and the cross section for the diffractive dissociation of the projectile @xcite , which was established long time ago by gribov @xcite . its result can be derived using reggeon calculus @xcite and the abramovsky - gribov - kancheli ( agk ) cutting rules @xcite and is a manifestation of the unitarity . this formalism can be used to calculate directly cross sections of photon - nucleus scattering for the interaction with two nucleons in terms of the diffractive photon - nucleon cross section . in this formalism , the total photon - nucleus cross section is expressed as a series containing the contribution from multiple scatterings ( 1 , 2 , @xmath19 ) : _ ^a = _ ^a^(1 ) + _ ^a^(2 ) + _ ^a^(3 ) + [ eq1 ] with the first term being the one that arises from independent scattering of the photon off @xmath20 nucleons : _ ^ a^(1)=a_^p and the first correction to the non - additivity of cross sections being _ ^ a^(2)=-4a(a-1)d^2b t_a^2(b ) _ m^2_^m^2_dm^2 . t=0 f_a^2(t _ ) [ sigalb ] where @xmath21 is the mass of the diffractively produced system , @xmath22 is the nucleus form factor which takes into account the coherence effects and the differential @xmath23 cross section for diffractive dissociation of the virtual photon appearing in ( [ sigalb ] ) is given by : . _ t=0= x_f^(3)_2(q^2,x _ , ) [ eq3 ] where @xmath24 is the diffractive slope parameter and @xmath25 is the diffractive proton structure function . moreover , @xmath26 , @xmath27 and @xmath28 . the integration limits in @xmath21 are @xmath29 gev@xmath6 , @xmath30 and @xmath31 . a shortcoming of this approach is that the inclusion of the higher order rescatterings is model dependent . this resummation is specially important at small @xmath3 , where multiple scattering is more likely to happen . in general it is assumed that the intermediate states in the rescatterings have the same structure and two resummation schemes are considered : ( a ) _ the schwimmer equation _ @xcite , which sums all fan diagrams with triple pomeron interactions and which is valid for the scattering of a small projectile on a large target . it implies that the photon nucleus cross section is given by : _ ^ * a^s(x , r)=_^p(x , r ) a d^2b [ schwimmer ] and ( b ) _ the eikonal unitarized cross section _ , given by ^e_^a(x , r)=_^p ( x , r ) a d^2b , [ eikonal ] where f(x , q^2)= _ m^2_min^m^2 _ dm^2 . \|_t=0 f^2_a(t _ ) . [ ff ] as shown in @xcite , the eikonal unitarization predicts a larger magnitude for the nuclear shadowing than the schwimmer equation . for models which take into account the possibility of different intermediate states see , e.g. , ref . @xcite . except for the choice of the resummation scheme , the predictions for @xmath32 obtained using ( [ schwimmer ] ) or ( [ eikonal ] ) are parameter free once the diffractive cross section is provided . models based on this non - perturbative regge - gribov framework are quite successful in describing existing data on inclusive and diffractive @xmath1 and @xmath33 scattering @xcite . however , they lack solid theoretical foundations within qcd . it is important to emphasize that some authors @xcite use these models as initial conditions for dglap evolution . the comparison among the predictions of the different models for nuclear shadowing presented in ref . @xcite , including the models discussed above , shows that they coincide within @xmath34 in the region where experimental data exist ( @xmath35 ) but differ strongly for smaller values of @xmath3 , with the difference being almost of a factor 2 at @xmath36 . our goal in this paper is try to reduce the theoretical uncertainty present in these predictions . in particular , differently from previous studies , which consider different inputs in the calculations using the glauber , schwimmer and eikonal approaches , we will consider a unique model for the projectile - nucleon interaction . we will calculate the dipole - nucleon cross section and the diffractive structure function using the dipole picture and the solution of the running coupling balitsky - kovchegov equation @xcite , which is the basic equation of the color glass condensate formalism . recently , this approach was shown to describe quite well the @xmath1 hera data for inclusive and diffractive observables ( see , e.g. refs . following this procedure we are able to estimate the magnitude of the theoretical uncertainty associated to the way the multiple scatterings are considered , reducing the contribution associated to the choice of initial conditions used in the calculations . moreover , we discuss the possibility of discriminating between these unitarization procedures in a future electron - ion collider . this paper is organized as follows . in sec . [ dipole ] we present a brief description of inclusive and diffrative @xmath37 - nucleon processes in the color dipole picture with particular emphasis in the dipole - proton cross section given by the color glass condensate formalism . in section [ results ] we present the predictions of the three unitarization schemes discussed above using as input the cgc results for the dipole - proton interaction and compare them with the existing experimental data . moreover , we present a comparison between the predictions for the kinematical region which will be probed in a future electron - ion collider . finally , in section [ conc ] we summarize our results and present our conclusions . the photon - hadron interaction at high energy ( small @xmath3 ) is usually described in the infinite momentum frame of the hadron in terms of the scattering of the photon off a sea quark , which is typically emitted by the small-@xmath3 gluons in the proton . however , as already mentioned in the introduction , in order to describe inclusive and diffractive interactions and disentangle the small-@xmath3 dynamics of the hadron wavefunction , it is more adequate to consider the photon - hadron scattering in the dipole frame , in which most of the energy is carried by the hadron , while the photon has just enough energy to dissociate into a quark - antiquark pair before the scattering . in this representation the probing projectile fluctuates into a quark - antiquark pair ( a dipole ) with transverse separation @xmath39 long before the interaction , which then scatters off the target @xcite . the main motivation to use this color dipole approach is that it gives a simple unified picture of inclusive and diffractive processes . in particular , in this approach the proton structure function is given in terms of the dipole - proton cross section , @xmath40 , as follows : f_2^p(x , q^2 ) = d^2r dz \|(r , z)\|^2 _ dp(x , r ) [ sigga ] where @xmath13 is the probability of the photon to split into a @xmath14 pair of size @xmath15 . moreover , the total diffractive cross sections take the following form ( see e.g. ref . @xcite ) , @xmath41 where @xmath42 it is assumed that the dependence on the momentum transfer , @xmath43 , factorizes and is given by an exponential with diffractive slope @xmath24 . the diffractive processes can be analysed in more detail by studying the behaviour of the diffractive structure function @xmath44 . following ref . @xcite we assume that the diffractive structure function is given by @xmath45 where the @xmath46 contribution with longitudinal polarization is not present because it has no leading logarithm in @xmath9 . the different contributions can be calculated and for the @xmath47 contributions they read @xcite @xmath48 @xmath49 \phi_{1 } + m_f^{2 } \phi_{0 } \right\ } \label{qqbt}\ ] ] where the lower limit of the integral over @xmath50 is given by @xmath51 , the sum is performed over the quark flavors and @xcite @xmath52 the @xmath53 contribution , within the dipole picture at leading @xmath54 accuracy , is given by @xcite @xmath55 } \label{qqg } \\ & \times & \int_{0}^{(1-z)q^{2}}\mbox{d } k_{t}^{2 } \ln \left(\frac{(1-z)q^{2}}{k_{t}^{2}}\right ) \left [ \int_{0}^{\infty } u \mbox{d}u \ ; \sigma_{dp}(u / k_{t } , x_{i\!\!p } ) k_{2}\left ( \sqrt{\frac{z}{1-z } u^{2}}\right ) j_{2}(u ) \right]^{2}.\nonumber\end{aligned}\ ] ] as pointed in ref . @xcite , at small @xmath56 and low @xmath9 , the leading @xmath57 terms should be resummed and the above expression should be modified . however , as a description with the same quality using the eq . ( [ qqg ] ) is possible by adjusting the coupling @xcite , in what follows we will use this expression for our phenomenological studies . we use the standard notation for the variables @xmath58 and @xmath59 , where @xmath60 the total energy of the @xmath61 system . the main input for the calculations of inclusive and diffractive observables in the dipole picture is @xmath62 which is determined by the qcd dynamics at small @xmath3 . in the eikonal approximation , it is given by : @xmath63 where @xmath64 is the forward scattering amplitude for a dipole with size @xmath65 and impact parameter @xmath66 which can be related to the expectation value of a wilson loop @xcite . it encodes all the information about the hadronic scattering , and thus about the non - linear and quantum effects in the hadron wave function . in general , it is assumed that the impact parameter dependence of @xmath67 can be factorized as @xmath68 , where @xmath69 is the profile function in impact parameter space , which implies @xmath70 . the forward scattering amplitude @xmath71 can be obtained by solving the bk evolution equation @xcite or considering phenomenological qcd inspired models to describe the interaction of the dipole with the target . bk equation is the simplest nonlinear evolution equation for the dipole - hadron scattering amplitude , being actually a mean field version of the first equation of the b - jimwlk hierarchy @xcite . in its linear version , it corresponds to the balitsky - fadin - kuraev - lipatov ( bfkl ) equation @xcite . the solution of the lo bk equation implies that the saturation scale grows much faster with increasing energy ( @xmath72 , with @xmath73 ) than that extracted from phenomenology ( @xmath74 ) . in the last years the next - to - leading order corrections to the bk equation were calculated @xcite through the ressumation of @xmath75 contributions to all orders , where @xmath76 is the number of flavors . thanks to these works it is now possible to estimate the soft gluon emission and running coupling corrections to the evolution kernel . the authors have found out that the dominant contributions come from the running coupling corrections , which allow us to determine the scale of the running coupling in the kernel . the solution of the improved bk equation was studied in detail in ref . the running of the coupling reduces the speed of the evolution to values compatible with experimental data , with the geometric scaling regime being reached only at ultra - high energies . in @xcite a global analysis of the small @xmath3 data for the proton structure function using the improved bk equation was performed ( see also ref . in contrast to the bk equation at leading logarithmic @xmath77 approximation , which fails to describe the hera data , the inclusion of running coupling effects in the evolution renders the bk equation compatible with them ( see also @xcite ) . in what follows we consider the bk predictions for @xmath71 ( from now on called rcbk ) obtained using the gbw @xcite initial condition . . in what follows we shall consider two different nuclei , ca and pb , and use the deuteron ( d ) as a reference to calculate the experimentally measured ratios @xmath78 and @xmath79 . we assume that the diffractive slope parameter is @xmath80 gev@xmath81 and that the nucleus form factor is given by : f_a(t_)=d^2b j_0(b)t_a(b ) , [ eq2 - 1 ] where the thickness function is given in terms of the nuclear density @xmath82 as : @xmath83 with the normalization fixed by @xmath84 . in fig . [ fig1 ] we compare the predictions of the glauber ( solid line ) , schwimmer ( dot - dashed line ) , eikonal ( dashed line ) and double scattering ( dot - dot - dashed line ) models for the ratios with the e665 experimental data at small @xmath3 @xcite . although joined with lines , our results are computed at the same @xmath85 and @xmath86 as the experimental data . our results demonstrate that if we compute the nuclear structure function up to two scatterings , which implies that @xmath87 , we are not able to describe the experimental data . furthermore , since the magnitude of the first correction , @xmath88 , is very large , then there is no hope to estimate the nuclear structure function by just summing a few terms in the multiple scattering series . therefore , a full resummation of the multiple scatterings is necessary , which makes the predictions model dependent . the agreement with the current experimental data at small @xmath3 of the glauber , schwimmer and eikonal models is quite reasonable taking into account that no parameters have been fitted to reproduce the data . this implies that the current data are not able to discriminate between the unitarization schemes . [ cols="^,^ " , ] 0.5 cm the behaviour of the nuclear wave function at high energies provides fundamental information for the determination of the initial conditions in heavy ion collisions and particle production in collisions involving nuclei . one of the main uncertainties is associated to the magnitude of the nuclear shadowing , which comes mainly from the way in which the multiple scattering problem is treated and from the modelling of the projectile - nucleon interaction . since a future eic will probe the shadowing region while keeping sufficiently large @xmath9 , new studies which determine the main sources of uncertainties in the predictions are necessary . in this work we compare three frequently used approaches to estimate the nuclear shadowing in nuclear dis . as in these approaches the nuclear cross section is completely determined once the interaction of the projectile with the nucleon is specified , we considered a single model ( rcbk ) as input of our calculations in order to quantify the theoretical uncertainty which comes from the choice of the unitarization model . in particular , we calculate the nuclear ratio between structure functions considering the glauber , schwimmer and eikonal approaches down to very low-@xmath3 utilizing the rcbk results both for inclusive and diffractive cross sections in @xmath23 scattering . our results demonstrate that the current experimental data at small @xmath3 are described successfully by the three approaches . however , the difference between their predictions becomes large in the kinematical region which will be probed in the future electron - ion colliders . this work was partially financed by the brazilian funding agencies capes , cnpq and fapesp . f. gelis , e. iancu , j. jalilian - marian and r. venugopalan , ann . nucl . part . sci . 60 * , 463 ( 2010);e . iancu and r. venugopalan , arxiv : hep - ph/0303204 ; h. weigert , prog . part . nucl . phys . * 55 * , 461 ( 2005 ) ; j. jalilian - marian and y. v. kovchegov , prog . * 56 * , 104 ( 2006 ) . v. p. goncalves , m. s. kugeratski , m.v.t . machado and f. s. navarra , phys . c * 80 * , 025202 ( 2009 ) ; e. r. cazaroto , f. carvalho , v. p. goncalves , m. s. kugeratski and f. s. navarra , phys . b * 696 * , 473 ( 2011 ) . l. frankfurt , v. guzey , m. mcdermott and m. strikman , jhep * 0202 * , 027 ( 2002 ) ; l. frankfurt , v. guzey and m. strikman , phys . rev . d * 71 * , 054001 ( 2005 ) ; l. frankfurt , v. guzey and m. strikman , phys . b * 586 * , 41 ( 2004 ) . b. z. kopeliovich , j. raufeisen and a. v. tarasov , phys . b * 440 * , 151 ( 1998 ) ; b. z. kopeliovich , j. raufeisen and a. v. tarasov , phys . c * 62 * , 035204 ( 2000 ) ; j. nemchik , phys . c * 68 * , 035206 ( 2003 ) . e. iancu , a. leonidov , l. mclerran , nucl . a * 692 * , 583 ( 2001 ) ; e. ferreiro , e. iancu , a. leonidov , l. mclerran , nucl . a * 703 * , 489 ( 2002 ) ; j. jalilian - marian , a. kovner , l. mclerran and h. weigert , phys . d * 55 * , 5414 ( 1997 ) ; j. jalilian - marian , a. kovner and h. weigert , phys . d * 59 * , 014014 ( 1999 ) , _ ibid . _ * 59 * , 014015 ( 1999 ) , _ ibid . _ * 59 * 034007 ( 1999 ) ; a. kovner , j. guilherme milhano and h. weigert , phys . rev . d * 62 * , 114005 ( 2000 ) ; h. weigert , nucl . phys . * a703 * , 823 ( 2002 ) .	the measurement of the nuclear structure function @xmath0 at the future electron - ion collider ( eic ) will be of great relevance to understand the origin of the nuclear shadowing and to probe gluon saturation effects . currently there are several phenomenological models , based on very distinct approaches , which describe the scarce experimental data quite successfully . one of main uncertainties comes from the schemes used to include the effects associated to the multiple scatterings and to unitarize the cross section . in this paper we compare the predictions of three distinct unitarization schemes of the nuclear structure function which use the same theoretical input to describe the projectile - nucleon interaction . in particular , we consider as input the predictions of the color glass condensate formalism , which reproduce the inclusive and diffractive @xmath1 hera data . our results demonstrate that the experimental analysis of @xmath2 will be able to discriminate between the unitarization schemes .
the advent of collider energies to heavy ion physics is leading to a profound change in the field . the new tools available , specially the access to the large transverse momentum part of the spectrum , allow for an unprecedented characterization of the high density state created in such collisions . accordingly , the traditional goal of producing a _ quark - gluon plasma _ , the deconfined state of quarks and gluons predicted from qcd , has been enlarged to the study of several other mechanisms as thermalization , parton distribution functions at very small-@xmath0 or in medium evolution of parton showers . in general words , a new line is emerging which attempt to study how the collective properties of the fundamental interactions appear . for that end , large energy densities in _ extended _ regions need to be produced in contrast with the traditional direction in high energy physics which attempt to create the largest possible energy scales in well localized spacial regions for new physics to become observable . the relevant questions which can be addressed in experiments of heavy ion collisions can be ( artificially ) classified depending on the time scale of the relevant phenomena : i ) before the collision , the structure of the two lorenz - contracted nuclei is rather different from a typical hadron at the same energy and collective phenomena , nowadays generically known under the name of color glass condensate , appear ; ii ) once the high - density state is created , the relevant question is how thermalization and other collective mechanisms , as a possible hydrodynamical behavior , appear ; iii ) finally , the properties of the eventually equilibrated medium need to be studied by means of some indirect signals . the lhc is in an excellent position to address these questions @xcite . reaching collider energies provides the additional tool of the hard part of the spectrum , characterized by large virtualities , to be used for these studies . a typical hard cross section can be written in the form @xmath1 with a factorization between the short - distance perturbative cross section @xmath2 , computable in powers of @xmath3 and two long - distance terms , the proton / nuclear parton distribution functions ( pdf ) , @xmath4 , encoding the partonic structure of the colliding objects , and the fragmentation functions ( ff ) , @xmath5 , describing the hadronization of the parton @xmath6 into a final hadron @xmath7 . these long - distance terms are non - perturbative objects , but whose evolution can be computed in perturbative qcd . these are precisely the objects which will be modified in the case that the _ extension _ of the colliding system interferes with the dynamics , while the short - distance part is expected to remain unchanged if the virtuality is large enough . this interference between geometry and dynamics makes hard probes perfect tools to characterize the medium properties through the modification of the long - distance terms in ( [ eqhard ] ) . a conceptually simple example is the case of the @xmath8 whose production cross section is , schematically , @xmath9}(x_1,x_2,q^2 ) \langle { \cal o}([c\bar c]\to j/\psi)\rangle\ , , \ ] ] where now @xmath10\to j/\psi)\rangle$ ] describes the hadronization of a @xmath11 pair in a given state ( for example a color octet ) into a final @xmath8 . this long - distance part is expected to be modified in a medium at finite temperature in which the deconfinement avoids the bound states to exist @xcite . however , this modification , being non - perturbative , lacks of good theoretical control and the real situation about the @xmath8-suppression is complicated @xcite . from the computational point of view , a theoretically simpler case involves the modification of the _ evolution _ of both the parton distribution and the fragmentation functions in a dense or finite temperature medium . this needs of large scales @xmath12 in order for the strong coupling to be small and perturbative methods to apply . in general , the presence of a dense medium is translated into non - linear terms in the evolution equations . the study of the modifications of the nuclear partonic distributions is nowadays known under the generic name of _ color glass condensate_. in its original formulation @xcite it provides a general framework for the whole collision , based on an effective theory separating the fast modes in the nuclear wave function from the _ generated _ slow modes , associated to small-@xmath0 gluons . this small-@xmath0 gluons have parametrically large ( @xmath13 ) densities and can be treated as classical fields . the quantum evolution equation of this setup is known and , remarkably , can be written in a rather simple form in the large-@xmath14 limit @xcite . a sophisticated technology has been developed in the last decade in this framework which , in addition to describe the structure of the incoming wave functions , aims to provide the link to the subsequent evolution into a thermal system @xcite . from a phenomenological point of view , the most successful applications of these formalism are the description of the multiplicities measured at rhic @xcite ; the possible presence of the predicted geometric scaling in lepton - hadron data @xcite ; and the suppression of inclusive particles at forward rapidities at rhic @xcite , which has been predicted as a result of small-@xmath0 quantum evolution @xcite . a particularly economic description is presented in fig . [ fig1 ] . here , the saturation scale is obtained from lepton - proton and lepton - nucleus data as @xmath15 with fitted parameters @xmath16 and @xmath17 . [ fig1 ] left shows the quality of the geometric scaling in lepton - proton @xcite and lepton - nucleus @xcite data . assuming the same scaling to hold in aa collisions , the multiplicity in the central rapidity can be written as @xcite @xmath18 where only a total normalization factor @xmath19 is needed once the energy and centrality dependences are fixed by lepton - proton and lepton - nucleus data respectively . [ fig1 ] shows the comparison of this simple formula with available data @xcite . and auau collisions at different centralities @xcite and the corresponding description from eq . ( [ eqmult])@xcite . ] and auau collisions at different centralities @xcite and the corresponding description from eq . ( [ eqmult])@xcite.,scaledwidth=80.0% ] a perturbative modification of the fragmentation of the highly virtual produced partons , has been proposed as a tool to characterize the medium produced in heavy ion collisions @xcite . these effects on high-@xmath20 particles , generically known as _ jet quenching _ , constitute one of the main experimental observations at rhic so far @xcite . in the near future , the lhc will extend the range in transverse momentum for at least one order of magnitude @xcite . in the vacuum , a perturbatively produced high-@xmath20 parton with virtuality @xmath21 develops a parton shower by radiating other partons ( mainly gluons ) with decreasing virtuality . this shower stops when a typical hadronic scale @xmath22gev@xmath23 is reached . the resulting jet is an extended object in which the collinear and soft emitted gluons are emitted inside a cone around the original parton direction . when the high-@xmath20 particle is produced in the medium created after a heavy ion collision , this parton shower is modified . the jets , being extended structures , provide an excellent tool to characterize the medium properties at different scales . at high enough parton energies , the main mechanism driving the modifications of high-@xmath20 evolution is the medium - induced gluon radiation . as in general high - energy processes , the propagation of the partons through the medium can be described in terms of wilson lines averaged in the allowed configuration of a medium . several prescriptions exist for these averages and in the multiple soft scattering limit , the saddle point approximation of the wilson lines define a single parameter of the medium , the transport coefficient , @xmath24 , given by the average transverse momentum squared acquired by the gluon per mean free path . the typical energy spectrum of radiated gluons is softer than in the vacuum although formation time effects provide a typical scale @xmath25 under which the radiation is suppressed . in the same way , the angular distribution is regulated at angles smaller than @xmath26 . as a result of formation time effects ( or coherence effects ) the medium - induced gluon radiation is , hence , finite in the infrared or collinear limits . the two main predictions are the suppression of high-@xmath20 yields due to additional in - medium energy loss and the associated broadening of the jet structures . although the degree of theoretical refinement of the jet quenching formalisms is still not completely satisfactory , it provides a good description of experimental data with the medium - modified fragmentation function computed as @xmath27 here , @xmath28 gives the probability of an additional in - medium energy loss and is normally computed by assuming a simple poisson distribution with the medium - induced gluon radiation as input @xcite . once the geometry of the system is correctly taken into account , a fit to rhic data ( see fig . [ figraa ] ) gives the value of the time - averaged transport coefficient @xmath29 gev@xmath30/fm . this large value and the corresponding uncertainty is a direct consequence of the surface trigger bias effect in inclusive particle suppression measurements @xcite . this is an intrinsic limitation of inclusive measurements on the characterization of the medium and on the study of the dynamics underlying the propagation of highly energetic partons through a dense medium . further constraints can be found by i ) measuring different particles species , and in particular heavy quarks , as the formalism predicts the hierarchy @xmath31 ; ii ) by directly measuring the induced radiation , i.e. by reconstructing the jet structure in a heavy ion collision . new data from rhic on non - photonic electrons @xcite attempt to answer the question on heavy quark in - medium energy loss . these electrons are expected to come from the decays of charm and beauty quarks . the perturbative description of the relative contribution of both quarks to the electron yield is , however , not under good control and the @xmath32 crossing point can be as low as 2 gev or as large as 10 gev when the usual mass and scale uncertainties are taken into account in the approximation . this translates into a large uncertainty in the ratio @xmath33 as can be seen in fig . [ fighq ] @xcite . the description of the experimental data is reasonable within the error bars , although not completely satisfactory . a clear distinction between heavy mass effects in medium - modified gluon radiation seems only possible with a better identification of the @xmath34 and @xmath35 contributions and , ideally , by a direct measurement of the @xmath36 and @xmath37 mesons . the modification of the jet properties is thought to be the most powerful tool to investigate the medium properties as well as the dynamics underlying jet - quenching . from an experimental point of view , the callibration uncertainty of true jets in the large background environment of a heavy ion collision is the main issue to overcome @xcite . the identification of observables with small sensitivity to background subtractions is of primary importance @xcite . from a theoretical point of view , studying medium modification of jet shapes in the most general case is complicated and likely would need a full monte carlo implementation of a in - medium parton shower process , which is not yet available . here , we quote a simplified study @xcite which provides a quite good phenomenological understanding of the surprising non - gaussian shape found in two particle correlations at rhic @xcite . the main observation is that for energies smaller than @xmath38 the typical medium - induced radiation angle reaches its maximum value @xmath39 . in this conditions , the medium - induced gluon radiation spectrum has the simple form @xcite @xmath40 taking a typical value @xmath41 gev@xmath30/fm , eq . ( [ eq : cohspec ] ) is valid for @xmath42 3 gev . compared to the corresponding spectrum in the vacuum , @xmath43 the one in ( [ eq : cohspec ] ) is softer and the typical emission angles larger , producing a broadening of the jet signals . for practical applications , exclusive distributions , giving the probability of one , two ... emissions are needed . how to construct such probabilities , using sudakov form factors @xmath44 , \label{eqsudak}\ ] ] is a well known procedure in the vacuum . in @xcite it is proposed to generalize this procedure by defining eq . ( [ eqsudak ] ) for the medium by just changing @xmath45 to @xmath46 given by eq . ( [ eq : cohspec ] ) . there are convincing motivations to make this _ ansatz _ : ( i ) it provides a clear probabilistic interpretation with the right limit to the most used `` quenching weights '' @xcite when the virtuality is ignored ; ( ii ) the evolution equations in the case of nuclear fragmentation functions in dis can be written as usual dglap equations with medium - modified splitting functions @xmath47 @xcite . in our case ( [ eq : cohspec ] ) would correspond to @xmath48 with the appropriate factors . two extreme cases were studied in @xcite : ( 1 ) one of the particles takes most of the incoming energy @xmath49 and , hence , @xmath50 `` j - configuration '' ; ( 2 ) the two particles share equally the available energy , @xmath51 and @xmath52 `` y - configuration '' . _ case ( 1)_. the splitting probability in the laboratory variables @xmath53 and @xmath54 @xmath55 presents a non - trivial angular dependence , which we stress has been found by the same perturbative mechanism and parameters describing the inclusive particle suppression . the distribution found has two maxima whose positions are determined by : @xmath56 the angular shape in ( [ eq : splitlab ] ) is very similar to the one found experimentally . in @xcite a simple model is proposed to take into account the different smearing effects from the experimental triggering conditions . the results are plotted in fig . [ fig : ndist ] for three different medium lengths and @xmath57 gev . to compare with the measured centrality dependence on the position of the maxima @xcite @xmath58 was taken and plotted in fig . [ fig : ndist ] see @xcite for details . the above explanation for the non - gaussian shapes found experimentally at rhic has the appeal of presenting a unified description of all high-@xmath20 effects in heavy ion collisions measured at rhic . this effects could be further amplified for jets developing in a flowing medium @xcite . however , other mechanisms lead to similar away side correlations . in the most popular one , these angular structures correspond to the shock waves produced by the highly energetic particle traversing the medium @xcite . in this picture , a large amount of the energy lost must be transferred to the medium almost instantaneously and take part in the hydrodynamical evolution . in general the released energy excites both sound and dispersive modes , and only the first ones produce the desired cone signal ( in the dispersive mode , the energy travels basically collinear with the jet ) . the energy deposition needed for the sound modes to become visible in the spectrum has been found to be quite large @xcite . to our knowledge , no attempt has been made so far to describe the centrality dependence of the shape of the azimuthal correlations in this approach . given the fact that the two formalisms rely on completely different hypotheses , finding experimental observables which could distinguish between them is certainly an issue which deserves further investigation . together with the high-@xmath20 , the most relevant data from rhic so far is the measurement of azimuthal anisotropies in the momentum distributions , which is considered as the main check of themalization in the medium . as originally proposed @xcite the hydrodynamical evolution of initial spacial anisotropies lead to momentum anisotropies due to the presence of pressure gradients . although not perfectly , the experimental data from rhic at low and moderate-@xmath20 can be consistently described by considering the formation of a thermalized medium at the very early stages of the collision @xmath59 fm , which evolves as a perfect fluid @xcite i.e. follows a hydrodynamical evolution with negligible viscosity . this observation has lead to the developing paradigm of an ideal fluid being created at rhic @xcite . this state would be a strongly coupled quark gluon plasma , with properties far from those of the asymptotic case of a gas of free quarks and gluons . whether this interpretation is correct or not is a matter of debate at present . very interesting effects appear in the intermediate region of @xmath60 gev / c . the most spectacular of them is the appearance of valence quark number scaling laws for baryons and mesons : ( i ) @xmath61 seems to depend only on the valence number of the produced particle ; ( ii ) the elliptic flow parameter @xmath62 is universal when plotted as @xmath63 , @xmath64 being the number of valence quarks . the most successful model to describe these features is a two component soft+hard model . in this model , the soft spectrum is supposed to come from the recombination of quarks in a medium in thermal equilibrium @xcite . thus for a particle with @xmath65 valence quarks @xmath66 k_n(p_\perp)\;. \label{eq:4qspec}\ ] ] @xmath67 is the transverse area of the partonic medium at freeze - out , @xmath68 the hadronization time , @xmath69 is the transverse mass of the hadron , @xmath70 is the hadron degeneracy factor and @xmath71 is the fugacity factor for a given quark @xmath72 . the corresponding parameters can be found in @xcite . in [ eq:4qspec ] the short - hand notation for a n - quark hadron @xmath73 \ , \label{eq : k4q}\end{aligned}\ ] ] has been introduced with @xmath74 and @xmath75 the corresponding bessel functions . the hard part of the spectrum is basically given by eq . ( [ eqhard ] ) with a simplified energy loss formalism , which constitutes an effective way of parameterizing the more refined one given in section [ sec : jq ] . in @xcite this model has been extended to the case of a 4-quark meson to study the sensitivity of these observables to make a case for the discovery of exotic states in heavy - ion collisions , in particular for the @xmath76 . in fig . [ raa : rhic ] @xmath77 for different mesons and baryons are plotted and compared with experimental data from rhic . the conclusion from this figure is that if recombination is the right mechanism underlying the baryon / meson difference at intermediate @xmath78 , heavy - ion collisions are ideal tools to study the quark content of different resonances and to find a definitive answer to the structure of these exotic states . for @xmath79 , @xmath80 , @xmath76 as a @xmath81 state , @xmath82 , @xmath83 , @xmath76 as a 4-quark state and @xmath84 . data from ref . the field of ultrarelativistic heavy ion collisions has been , arguably , the most active field of research and discoveries in qcd in the last years . it is not unexpected that this will continue in the future , in particular within the unexplored regions accessible at the lhc . this activity is also reflected in the interest shown from other fields , in particular in the relation of experimental observables with string theory computable quantities by means of the ads / cft correspondence . the literature is already very vast , and impossible to quote here all the most relevant contributions ; let just us mention the calculation of the shear viscosity @xcite ; the heavy quark diffusion coefficient @xcite ; the jet quenching transport coefficient @xcite ; the presence of shock waves @xcite ; the relation with hydrodynamics @xcite . cas is supported by the 6th framework programme of the european community under the marie curie contract meif - ct-2005 - 024624 . t. matsui and h. satz , phys . b * 178 * ( 1986 ) 416 . v. n. tram [ phenix collaboration ] , nucl - ex/0606017 . l. d. mclerran and r. venugopalan , phys . d * 49 * ( 1994 ) 2233 ; phys . d * 49 * ( 1994 ) 3352 ; phys . d * 50 * ( 1994 ) 2225 . r. baier , a. h. mueller , d. schiff and d. t. son , phys . b * 502 * ( 2001 ) 51 ; t. lappi and l. mclerran , hep - ph/0602189 ; p. romatschke and r. venugopalan , hep - ph/0605045 . d. kharzeev and m. nardi , phys . b * 507 * ( 2001 ) 121 . n. armesto , c. a. salgado and u. a. wiedemann , phys . * 94 * ( 2005 ) 022002 a. m. stasto , k. golec - biernat and j. kwiecinski , phys . rev . * 86 * ( 2001 ) 596 ; a. m. stasto these proceedings . a. freund _ et al . * 90 * ( 2003 ) 222002 i. arsene _ et al . _ [ brahms collaboration ] , phys . * 93 * ( 2004 ) 242303 . d. kharzeev , e. levin and l. mclerran , phys . b * 561 * ( 2003 ) 93 ; r. baier , a. kovner and u. a. wiedemann , phys . d * 68 * , 054009 ( 2003 ) ; d. kharzeev , y. v. kovchegov and k. tuchin , phys . d * 68 * ( 2003 ) 094013 ; j. l. albacete _ et al . lett . * 92 * , 082001 ( 2004 ) . b. b. back _ et al . _ , nucl . phys . a * 757 * ( 2005 ) 28 . for a recent review see e.g. c. a. salgado , hep - ph/0510062 and references therein . r. baier , y.l . dokshitzer , a.h . mueller and d. schiff , jhep * 0109 * ( 2001 ) 033 c. a. salgado and u. a. wiedemann , phys . d * 68 * , 014008 ( 2003 ) . b. muller , phys . c * 67 * ( 2003 ) 061901 ; k. j. eskola _ et al . a * 747 * ( 2005 ) 511 . a. dainese , c. loizides and g. paic , eur . j. c * 38 * ( 2005 ) 461 ; hep - ph/0511045 . s. s. adler _ et al . _ [ phenix collaboration ] , phys . c * 69 * ( 2004 ) 034910 ; j. adams _ et al . _ [ star collaboration ] , phys . * 91 * ( 2003 ) 172302 ; b. b. back _ et al . _ [ phobos collaboration ] , phys . b * 578 * ( 2004 ) 297 ; i. arsene _ et al . _ [ brahms collaboration ] , phys . lett . * 91 * ( 2003 ) 072305 . m. shimomura [ phenix ] , nucl - ex/0510023 . m. cacciari , p. nason and r. vogt , phys . * 95 * ( 2005 ) 122001 n. armesto , _ et al . b * 637 * ( 2006 ) 362 . s.s . adler _ [ phenix ] nucl - ex/0510047 ; j. bielcik [ star ] nucl - ex/0511005 . c. a. salgado and u. a. wiedemann , phys . * 93 * ( 2004 ) 042301 s. s. adler _ et al . _ [ phenix collaboration ] , nucl - ex/0507004 . n. armesto , c. a. salgado and u. a. wiedemann , phys . * 93 * ( 2004 ) 242301 ; phys . c * 72 * ( 2005 ) 064910 ; h. stoecker , nucl . a * 750 * , 121 ( 2005 ) ; j. casalderrey - solana , e. v. shuryak and d. teaney , hep - ph/0411315 ; j. ruppert and b. muller , phys . b * 618 * ( 2005 ) 123 ; a. d. polosa and c. a. salgado , hep - ph/0607295 . x. n. wang and x. f. guo , nucl . a * 696 * ( 2001 ) 788 . n. grau [ phenix collaboration ] , nucl - ex/0511046 . j. casalderrey - solana , e. v. shuryak and d. teaney , hep - ph/0602183 . j. y. ollitrault , phys . d * 46 * ( 1992 ) 229 . see e.g. k. j. eskola , _ et al _ phys . c * 72 * ( 2005 ) 044904 . see e.g. e. shuryak , hep - ph/0510123 . l. maiani , a. d. polosa , v. riquer and c. a. salgado , hep - ph/0606217 . r. j. fries , _ et al . c * 68 * , 044902 ( 2003 ) . s. salur [ star collaboration ] , nucl - ex/0509036 . g. policastro , d.t . son and a.o . starinets , phys . rev . lett . * 87 * ( 2001 ) 081601 . j. casalderrey - solana and d. teaney , arxiv : hep - ph/0605199 . h. liu , k. rajagopal and u. a. wiedemann , hep - ph/0605178 ; n. armesto , j. d. edelstein and j. mas , arxiv : hep - ph/0606245 .	some of the new developments in the theory of heavy ion collisions are reviewed . much of the last progress have been triggered by the high energies available at rhic . in the near future , the lhc will extend the energy reach in heavy ions by a factor thirty and give access to new qcd regimes characterized by large densities and temperatures and corresponding modified evolution equations .
significant advances in understanding isospin violation in the nuclear force have been made in the past decade . experimental progress ( which is reviewed and summarized in refs.@xcite ) has been supplemented recently by the advent of chiral perturbation theory ( chpt)@xcite . this powerful technique casts the symmetries of qcd into effective interactions of the traditional , low - energy degrees of freedom of nuclear physics ( viz . , nucleons and pions ) . these building blocks ( in the lagrangian ) can then be combined in a systematic way to produce nuclear forces that violate isospin in exactly the same way as in qcd . an important feature of effective field theories is power counting , which is the technique used to organize calculations@xcite . a well - defined ordering of terms in the lagrangian according to scales intrinsic to qcd and nuclei is used to generate all terms of a particular size . terms in the lagrangian are labeled ( in the conventional way , @xmath2 ) by the number of implicit powers ( @xmath3 ) of the inverse of the large - mass scale of qcd , @xmath4 1 gev . when considering processes where the typical momentum is of the order of the pion mass , @xmath5 , amplitudes are expanded in powers of @xmath6 . much work has been done in the past few years regarding the derivation of the isospin - symmetric part of the nuclear potential@xcite . the components of short ( @xmath7 ) range in the potential come from ( renormalized ) contact interactions whose strengths are not determined by symmetry . the components of longer range arise from pion exchanges , and are determined in terms of pion couplings to the nucleon , which are fewer and in many cases determined from processes involving a single nucleon . perhaps the most significant result of this method is the first derivation of a two - pion - range potential consistent with the approximate chiral symmetry of qcd@xcite . this two - pion - exchange potential ( tpep ) has been incorporated in the nijmegen phase - shift analysis@xcite . pion - nucleon couplings determined from two - nucleon data were found to be in good agreement with other determinations based on pion - nucleon scattering . in effective theories , isospin - breaking interactions can be classified@xcite according to whether their origin is the mass difference between @xmath8 and @xmath9 quarks or hard electromagnetic ( em ) interactions at the quark level . the soft em interactions ( such as the coulomb force ) can be constructed in the usual way@xcite . the power counting for isospin - violating interactions was developed by one of us@xcite , and it explains the sizes of the various isospin structures present in the nuclear force . a convenient and universal@xcite classification for nuclear isospin is : class ( i ) - isospin conserving ; class ( ii ) - charge - independence breaking ( cib ) of isospin , but charge symmetric ; class ( iii ) - charge - symmetry breaking ( csb ) of isospin ; class ( iv ) - isospin mixing in the @xmath10 system between @xmath11 and @xmath12 . power counting can be used to demonstrate@xcite that a class ( n ) force appears only at order @xmath13 or higher . we therefore deduce on the basis of qcd a result that was noticed before on an empirical basis : class ( i ) forces are stronger than class ( ii ) , which are stronger than class ( iii ) , which are stronger than class ( iv ) . using this formalism , several of the various isospin components of the nuclear potential have been calculated@xcite . in particular , we have computed the long - range components of the class ( i)@xcite and class ( ii)@xcite two - nucleon potentials , up to order @xmath14 . the class ( iii ) isospin violation is the purview of this work . charge symmetry@xcite can be loosely defined as invariance under turning neutrons into protons and protons into neutrons . it has long been considered a particularly interesting aspect of isospin violation because the difference in quark masses is a source of csb . on the other hand , a significant interaction that violates this symmetry is the electromagnetic interaction , which is large between two protons and very small between two neutrons . moreover , the coulomb interaction between protons is long - ranged , while it has a nuclear range for two neutrons and is therefore indistinguishable from nuclear interactions . there are corrections of order @xmath15 to the static coulomb interaction ( where @xmath16 is the nucleon mass ) , which are part of the breit interaction . although these familiar electromagnetic interactions dominate the csb in nuclei , we will largely ignore them and concentrate on the nuclear mechanisms@xcite . in the nuclear force the comparison that interests us is the difference in forces between two protons and two neutrons , which is restricted to the @xmath12 channel for two nucleons . the longstanding interest in csb@xcite has been highlighted recently by two exciting new experiments . one experiment , carried out at triumf and currently undergoing final stages of analysis , measures the front - back asymmetry of the pion produced in the reaction @xmath17 close to threshold@xcite . another experiment has just been completed at iucf , and measures the near - threshold cross section for the reaction @xmath18@xcite . it has been argued in ref.@xcite that the chiral properties of two different contributions to the nucleon - mass difference can give a relatively large effect in @xmath17 , of opposite sign to more well - known mechanisms . a similar phenomenon might exist in @xmath18@xcite . there is reason to hope that these data will allow a model - independent ( albeit crude ) determination of both quark - mass and electromagnetic components of the nucleon - mass difference . an issue that arises naturally is the role of the nucleon - mass difference in the two - pion - exchange nucleon - nucleon potential . ref.@xcite calculated the crossed - box contribution , and it contains a comprehensive discussion of the older literature@xcite , where aspects of chiral symmetry are not emphasized . more recently , ref.@xcite calculated a csb seagull contribution using the formalism of ref.@xcite . in this paper we discuss the relative sizes of csb interactions in effective field theories , and calculate the two - nucleon potential of two - pion range that arises from the nucleon - mass difference . pieces of this potential have appeared before@xcite . here we complete the calculation of the long - range component of the class ( iii ) potential up to ( and including ) order @xmath14 . in the chiral lagrangian , the largest isospin - violating terms are those that determine the mass differences of nucleons and pions@xcite . in order to compare these and various other contributions , it is useful to estimate relative sizes of em and quark - mass effects . the pion - mass splitting provides a good starting point . the pion - mass splitting ( squared ) , @xmath19 , receives contributions from both the difference between the masses of the @xmath8 and @xmath9 quarks and em interactions . because of the chiral transformation properties of the pion , the contribution from the quark - mass difference is proportional to @xmath20 . the proportionality constant ( @xmath21 ) is on the order of the square of the difference of quark masses divided by their sum ( @xmath22 ) . the pion - mass difference is therefore mostly due to em interactions . it is nominally of order @xmath23 , but the large strength implied by this counting is compensated by the small fine - structure constant ( @xmath24 ) , which reduces the strength by slightly more than two orders of magnitude . the size of the pion - mass splitting is then rather accurately described by @xmath25 , with @xmath26 , the mass of the @xmath27-meson@xcite . this suggests that the relevant dimensionless parameter for the em - induced isospin violation is @xmath28 , which numerically is close to @xmath29 . since the latter would ordinarily correspond in power counting to @xmath14 , we deduce a convenient mnemonic of adding 3 to the order of the em - induced isospin - violating lagrangian when comparing sizes with quark - mass - induced mechanisms . henceforth our power counting for em - induced interactions will contain this additional factor of 3 ( e.g. , @xmath30 as counted above will be counted as @xmath31 below ) . the power counting has been used in ref.@xcite to classify various contributions to the nuclear potential . the leading isospin - breaking interaction appears at @xmath31 , when the pion - mass difference is inserted in the one - pion - exchange potential ( opep ) . it gives a class ( ii ) force , which is of relative strength @xmath32 ( times opep)@xcite . many other class ( ii ) forces appear up to order @xmath14 . there are three em mechanisms of roughly the same relative strength , @xmath33 ( times the usual opep ) produced by @xmath14 terms in the lagrangian or from loops . they are : the two - pion - exchange potential with different charged and neutral pion masses@xcite , cib in the pion - nucleon coupling constant@xcite and consequently in opep , and the @xmath34-exchange potential@xcite . moreover , there are opep corrections that arise from higher - order isospin - conserving pion - nucleon interactions ( such as recoil in the usual pion - nucleon coupling ) , and from second - order effects due to the the pion- and nucleon - mass differences@xcite . here we concentrate on the class ( iii ) mechanisms . it is worthwhile to discuss briefly the various mechanisms that contribute to csb in the @xmath35 system and in the trinucleon system ( @xmath0he @xmath1 @xmath0h ) , their relative sizes , and their relationship in the context of power counting . the bulk of the csb is due to the soft em interactions . the 764 kev binding - energy difference of @xmath0he and @xmath0h is largely accounted for by the 648 kev coulomb energy difference@xcite plus approximately 29 kev from the small breit corrections plus vacuum polarization@xcite . calculations of these contributions appear to be rather robust , and we henceforth ignore them . the remaining mechanisms that generate approximately 87 kev are contained in chpt , and we discuss them below . the most obvious signature of csb due to nuclear forces is the difference in the scattering lengths of two neutrons and two protons , once all em mechanisms are accounted for and the effect of the different nucleon masses on the kinetic energy ( see eqn . ( 1c ) below ) is treated . the resulting `` experimental '' @xmath36 scattering - length difference@xcite of @xmath371.5(5 ) fm is then attributed to csb in the nuclear force . typically one treats only the short - range ( as opposed to pion - range ) csb force , by adjusting the force to produce the desired difference in scattering lengths . this csb modification of the force then produces a contribution to the @xmath0he @xmath1 @xmath0h binding - energy difference of approximately 65(22 ) kev , a number that also appears robust@xcite and can accommodate the 87 kev binding - energy difference missing after soft em processes are taken into account . note that this scattering - length difference is generated by the total csb nuclear force , and does not differentiate between components of different ranges . the mass of a single nucleon can be expressed in terms of isospin operators ( the pauli isospin operator @xmath38 satisfies @xmath39 , where @xmath40 gives + 1/2 for a proton and @xmath371/2 for a neutron ) : @xmath41 where @xmath42 is the average nucleon mass , @xmath43 is the nucleon - mass difference , and @xmath44 is a csb nucleon - mass interaction . the nucleon - mass difference , @xmath45 , receives contributions from both the quark - mass difference , @xmath46 , and em interactions at the quark level , @xmath47 , @xmath48 ( we neglect here mixed contributions of higher order . ) the piece due to the quark - mass difference is the sole csb contribution of order @xmath3=1 , implying that this quantity is proportional to @xmath49 . the parameter @xmath50 plays an important role in the power counting for isospin violation , and would lead to a nucleon - mass difference on the order of 8 mev ( a factor of three higher than most model estimates , and therefore reasonable ) . the corresponding nucleon - mass term from em interactions at the quark level is of order @xmath51 , but is numerically comparable to the quark - mass term ( somewhat smaller and of opposite sign ) because it implicitly contains a power of the fine - structure constant . the largest csb interactions in the lagrangian thus come from the nucleon - mass difference of orders @xmath3= [ 1,2 ] for the two different mechanisms . the effects of the nucleon - mass difference on nuclear amplitudes are somewhat suppressed , however . summing eqn . ( 1a ) over all nucleons produces an overall constant ( determined by the average nucleon mass ) , which can be removed by a shift in the zero of energy , and a term proportional to the @xmath52-component of the total nuclear isospin , @xmath53 . the latter term will contribute to a csb shift in the nuclear kinetic energy ( for simplicity we restrict ourselves to only two nucleons in their center - of - mass ( cm ) frame ) @xmath54 where @xmath55 is the usual cm kinetic energy for two nucleons with equal masses , and @xmath56 for two nucleons . the csb nucleon - mass difference will also lead to a modification of the nuclear potential energy . the csb part of eqn . ( 1a ) ( @xmath57 ) will commute with an isospin - conserving hamiltonian and therefore will play no role to order @xmath45 , except inside certain loops ( viz . , crossed - box and triangle diagrams ) that involve exchanging two charged pions . in the latter case for a process initiated by two identical nucleons of one type , the intermediate state would contain two nucleons of the @xmath58 type . exchanging neutral mesons clearly does not invoke this mechanism , nor does ladder approximation ( sequential exchanges of neutral mesons ) for two protons or two neutrons . because this effect requires a loop , it is of order @xmath14 . thus although nominally of order @xmath31 ( by itself ) , the nucleon - mass difference actually first contributes in the much higher order @xmath14 , which is a reduction in size typical of loop contributions . in addition to these loop insertions , there are associated triangle - graph interactions from seagulls ( discussed below ) . other contributions in the lagrangian begin at order @xmath59 for the quark - mass terms ( i.e. , they are the largest ) and arise from two types of csb nuclear forces : short - range forces ( such as those arising from @xmath60@xcite and @xmath61@xcite mixing ) , and from isospin violation in the pion - nucleon coupling constant ( a model for the latter is provided by @xmath62 mixing@xcite ) . we denote the former by @xmath63 and the latter by @xmath64 . to leading order@xcite for two nucleons one has @xmath65 where @xmath66 is the separation of nucleons 1 and 2 , and @xmath67 and @xmath68 have unit volume integral . the constants @xmath69 and @xmath70 stem from the quark - mass difference and are of order @xmath71 , while @xmath72 and @xmath73 are em corrections of order @xmath74 . this implies a potential strength @xmath75 , where @xmath76 is the expectation value of the short - range interaction between either the @xmath77 pair in @xmath0he or the @xmath78 pair in @xmath0h . a 34-channel calculation using the av18 potential@xcite gives @xmath377.6 mev for the latter , or an estimate of about 45 kev for the contribution of the csb short - range interaction to the @xmath0he @xmath1 @xmath0h binding - energy difference . this interaction has traditionally always been a part of csb studies . the opep contains a csb ( coupling constant ) modification@xcite @xmath79 where @xmath80 is the usual isospin - conserving opep , and we do not write ( higher - order ) recoil corrections explicitly . here @xmath81 is the ( quark - mass induced ) csb pion - nucleon coupling constant and @xmath82 is the ( em - induced ) csb pion - nucleon coupling constant . the size of @xmath83 is constrained only by an upper limit on @xmath84 , arising from the nijmegen phase - shift analysis of @xmath35 data@xcite . a numerical estimate of @xmath64 in the trinucleon system can be obtained from eqn . ( 3 ) , which shows that the contribution to the trinucleon binding - energy difference is approximately @xmath85 , where @xmath86 is the expectation value of the potential energy due to opep between either the @xmath77 pair in @xmath0he or the @xmath78 pair in @xmath0h . a 34-channel faddeev calculation using the av18 potential@xcite produces @xmath371.67 mev for the latter . using the experimental value of @xmath87 from ref.@xcite produces @xmath88 0(50 ) kev of uncertain sign . the dimensional estimate@xcite using @xmath89 produces @xmath88 50 kev , and also of uncertain sign . thus a substantial part of the `` short - range '' contribution to csb ( 65(22 ) kev ) in the trinucleons could come from csb in opep , rather than from a shorter - range interaction . in any event the two mechanisms are predicted by power counting to be the dominant hadronic contributions and to be roughly comparable in size . the next order in chpt for csb is @xmath14 . the modification of the nuclear kinetic energy induced by the different nucleon masses ( eqn . ( 1c ) ) is of this order , and is always taken into account , both in the @xmath35 interaction and in the trinucleon csb , where it contributes a robust 14 kev@xcite . a power - counting estimate of its size can be made by using @xmath90 to estimate the average value of the nucleon momentum in the trinucleons , which produces a value @xmath91 25 kev ( and within a factor of two of the actual value ) . the size of this contribution illustrates the general caveat about power counting : even though this kinetic - energy modification is an order smaller in the lagrangian than the potential couplings ( @xmath14 vs. @xmath59 ) , they can make actual contributions in a @xmath92 that are not very different in size . this contribution added to the soft - em mechanisms discussed earlier leaves about 73 kev to be explained by the various csb nuclear potentials . the other mechanism of order @xmath14 is the two - pion - exchange csb arising from the nucleon - mass difference in certain loops , which we discussed above and will treat below . it depends not simply on the full proton - neutron mass difference , but on @xmath46 and @xmath47 separately . since these two components of the nucleon - mass difference are presently unknown , we can only estimate the impact of this potential on observable quantities . power counting suggests that this contribution is one order smaller than the ( three ) previously discussed mechanisms . we can make a numerical estimate of its contribution to the @xmath0he @xmath1 @xmath0h binding - energy difference using naive dimensional analysis . again assuming that the typical momentum in the nucleus is @xmath93 and that the loop integral gives the usual factor of @xmath94 , we expect in coordinate space @xmath95 , which is more than an order of magnitude smaller than the estimate for @xmath96 of 25 kev . as we will see below , however , the loop integrals in this case only give * one * power of @xmath97 . the csb tpep then likely supplies a larger contribution than expected on the basis of power counting . note that the csb tpep contribution is also a part of the traditional `` short - range '' nuclear csb mechanism , and its effect is therefore included in the 65(22 ) kev short - range part of the trinucleon binding - energy difference . one can of course accommodate all of these short - range mechanisms using models@xcite . alternatively one can hope that a sophisticated partial - wave analysis of nucleon - nucleon scattering , such as that carried out by the nijmegen group@xcite , may be able to disentangle the csb interactions of one - pion range , two - pion range , and short range , each of which contains an unknown parameter that must be fitted to the data . incorporating these forces into their procedure should be straightforward , although there may not be enough sensitivity in the data to distinguish between the three mechanisms . a full analysis of few - nucleon systems within effective field theory@xcite will eventually include all these csb mechanisms . in summary , there are two mechanisms for csb of order @xmath59 at the lagrangian level ( csb short - range forces in eqn . ( 2 ) , and csb opep in eqn . ( 3 ) ) and one of order @xmath14 ( nuclear csb kinetic energy in eqn . ( 1c ) ) that should make roughly comparable contributions in a nucleus . the tpep modified by @xmath46 and @xmath47 , together with the associated csb seagull interactions ( all given in eqns . ( 9 ) below ) , should be somewhat smaller . isospin - conserving two - pion - exchange potentials ( tpeps ) are an old problem with a new twist . in static order ( containing only terms that remain when the nucleon mass , @xmath16 , or the large - mass scale of qcd , @xmath98 , becomes very large ) the diagrams of fig . ( 1 ) ( properly symmetrized ) contribute to the tpep . the vertices and propagators follow from the leading - order lagrangian for pions and nucleons , @xmath99 + n^{\dagger}[i\partial_{0}-\frac{1}{4 { f_{\pi}}^{2 } } { \mbox{\boldmath $ \tau$}}\cdot ( { \mbox{\boldmath $ \pi$}}\times\dot{{\mbox{\boldmath $ \pi$}}})]n + \frac{g_{a}}{2 { f_{\pi } } } n^{\dagger}{\vec{\sigma}}\cdot{\vec{\nabla}}({\mbox{\boldmath $ \tau$}}\cdot { \mbox{\boldmath $ \pi$}})n \ , , \eqno ( 4)\ ] ] where the @xmath100 term is the weinberg - tomozawa ( wt ) interaction@xcite and the @xmath101 term is the usual interaction that depends on the axial - vector coupling constant , @xmath102 , and the pion - decay constant , @xmath103 . terms with additional pions or nucleons are neglected here , as they only contribute to the nuclear force at higher orders . the wt term has a specific normalization ( @xmath104 ) required by the underlying chiral symmetry . the nucleon - mass difference corresponds to lagrangian terms that arise from both quark - mass differences , @xmath46 , and from em interactions at the quark level , @xmath47 . the largest term of the first type is@xcite @xmath105n \ , , \eqno ( 5)\ ] ] and the leading - order term of the em type is@xcite @xmath106n \ , . \eqno ( 6)\ ] ] as with the wt term , the @xmath100 interactions in eqns . ( 5 ) and ( 6 ) are _ required _ by chiral symmetry and have a fixed strength ( @xmath107 ) relative to the mass terms . we again drop terms that involve more pion fields and do not contribute to the nuclear potential in low orders . the csb tpep can be computed in a straightforward way . we simply consider all insertions of the mass and interaction terms above into the diagrams of fig . ( 1 ) , including external lines . alternatively , we implement the @xmath108-@xmath3 mass - difference mechanism by including the simple @xmath108-@xmath3 mass difference ( i.e , the sum of the first terms in each lagrangian in eqns . ( 5 ) and ( 6 ) ) in the initial and final nuclear states , and then compensate for this addition by a subtraction . that is , we write for a single nucleon @xmath109 where @xmath110 is the reference value of @xmath111 for that nucleon . the expectation value of this contribution for all the nucleons will vanish , as does the contribution from uncrossed box diagrams . the crossed - box diagram shown in fig . ( 2a ) and the triangle diagram in fig ( 2b ) can also be easily computed . in addition there are seagull terms involving two pions ( from the second parts of the two lagrangians in eqns . ( 5 ) and ( 6 ) ) , which will generate triangle diagrams ( fig . ( 2c ) ) , and in principle `` football '' diagrams ( fig . ( 2d ) ) . the latter vanish for this problem because of the isospin symmetries of the wt ( antisymmetric ) and csb seagull ( symmetric ) vertices . we find a result that is a pure class ( iii ) potential . our results in momentum space are given below , with both pion momenta @xmath112 and @xmath113 pointing into the seagull vertex . the loop momentum is @xmath114 , and @xmath115 is the momentum transfer . writing @xmath116 and @xmath117 graph ( 2a ) gives @xmath118 while graph ( 2b ) generates @xmath119 and finally graph ( 2c ) produces @xmath120 the integrals in these expressions are divergent , requiring regularization and renormalization by the spin - independent nucleon - nucleon contact terms given in eqn . the loop integration over @xmath114 gives for the non - analytic terms @xmath121 and @xmath122 the long - range part of the potential in configuration space is independent of the regularization procedure . it is simplest to derive it by introducing a cutoff ( `` form '' ) factor @xmath123 for each pion line carrying a momentum @xmath115 in or out of a vertex . the three nucleon - nucleon potentials depicted in figs . ( 2a - c ) are : @xmath124 with @xmath125 and @xmath126 in these potentials we have defined @xmath127 , all derivatives are with respect to @xmath128 , and @xmath129 is the regulated yukawa function @xmath130 where the last form holds only for @xmath131 . note that @xmath132 generates a purely short - range contribution ( i.e. , a @xmath133-function if @xmath134 ) that is indistinguishable from other short - range contributions . it is therefore permissible to ignore such terms and replace @xmath132 by @xmath129 . ( 10 ) , ( 10a ) , and ( 10bc ) form our main result : the model - independent , long - range part of the csb tpep . this result can now be used , for example , as added input to the nijmegen phase - shift analysis@xcite . the contribution from the crossed - box diagram , eqn . ( 9a ) , is proportional to the total @xmath108-@xmath3 mass difference @xmath45 . our results agree with the functional form of ref.@xcite in momentum space . a potential in configuration space was not presented there . the contribution from the the csb seagull , eqn . ( 9c ) , was first calculated in ref.@xcite . the triangle with mass insertion , eqn . ( 9b ) , has not been previously calculated . there are cancellations between the latter two terms , which make them proportional to the quark - mass component @xmath46 of the nucleon - mass difference . while the quantity @xmath45 is a precisely known observable , the quantity @xmath46 is not ( yet ) . note that our expressions ( 8a ) and ( 8bc ) contain only one power of @xmath135 , while usually a loop contributes two powers . ignoring @xmath46 and treating the dimensionless @xmath129 and any of its derivatives as order ( 1 ) , we get @xmath136 , which is about half the estimate for @xmath96 of 25 kev . numerical estimates of the effect of terms ( a ) and ( c ) have already been published@xcite . at present only the effect of the sum of the one - pion , ( the just - calculated ) two - pion , and short - range csb potentials is constrained by experiment , while none of the individual terms are uniquely specified . in summary , we have calculated the additional two - pion - exchange csb nucleon - nucleon force that is the same order as ( and complements ) the two - pion - exchange two - nucleon csb force previously calculated in refs.@xcite . this force has been developed using chpt , and should be somewhat smaller than the other csb interactions that we discussed . we have also discussed the interplay between the csb opep and short - range csb force , both of which are larger than the csb tpep . the observed csb in the @xmath0he @xmath1 @xmath0h system is consistent with modern calculations . no model - independent resolution of the csb nuclear potentials into components of different range has yet been made , because each force of one - pion , two - pion , or short range contains an ( as yet ) undetermined constant . it is hoped that by completing the one - pion - range and two - pion - range parts of these potentials in chpt a phase - shift analysis of nucleon - nucleon scattering data that incorporates this information may be able to differentiate the components of various ranges . in addition to the two - pion - exchange nucleon - nucleon force derived here and in refs.@xcite , there will be two - pion - exchange three - nucleon csb forces of order @xmath14 , which we have not treated herein . we would like to thank rob timmermans and bob wiringa for several very helpful discussions about csb . uvk is grateful to the department of physics at the university of washington for its hospitality , and to riken , brookhaven national laboratory and the u.s . department of energy [ de - ac02 - 98ch10886 ] for providing the facilities essential for the completion of this work . the work of jlf was performed under the auspices of the doe , glp was supported in part by the doe , sac was supported in part by the u.s . national science foundation , and uvk was supported in part by the doe outstanding junior investigator program and the alfred p. sloan foundation . e. m. henley , in _ isospin in nuclear physics _ , d. h. wilkinson , ed . ( north - holland , amsterdam , 1969 ) , p.15 ; e. m. henley and g. a. miller , in _ mesons and nuclei _ , m. rho and g. e. brown , eds . ( north - holland , amsterdam , 1979 ) , vol . i , p. 405 . g. a. miller , b. m. k. nefkens , and i. laus , _ phys . * 194 * , 1 ( 1990 ) ; g. a. miller and w. t. h. van oers , in _ symmetries and fundamental interactions in nuclei _ , w. haxton and e. m. henley , eds . ( world scientific , singapore , 1995 ) . s. a. coon , nucl - th/9903033 , invited talk given at xiii international seminar on high energy physics problems ( ishepp 13 ) , `` relativistic nuclear physics and quantum chromodynamics '' , joint institute for nuclear research , dubna , russia , september , 1996 . s. weinberg , _ physica _ * 96a * , 327 ( 1979 ) ; s. weinberg , _ nucl . phys . _ * b363 * , 3 ( 1991 ) ; _ phys . lett . _ * b251 * , 288 ( 1990 ) ; _ phys . _ * b295 * , 114 ( 1992 ) . u. van kolck , ph . d. thesis , university of texas , 1993 ; u. van kolck , _ few - body systems suppl . _ * 9 * , 444 ( 1995 ) . c. ordez and u. van kolck , _ phys . lett . _ * b291 * , 459 ( 1992 ) ; c. ordez , l. ray , and u. van kolck , _ phys . rev . lett . _ * 72 * , 1982 ( 1994 ) ; u. van kolck , _ phys . c _ * 49 * , 2932 ( 1994 ) ; c. ordez , l. ray , and u. van kolck , _ phys . rev . c _ * 53 * , 2086 ( 1996 ) . j. l. friar and s. a. coon , _ phys . rev . _ * c49 * , 1272 ( 1994 ) ; n. kaiser , r. brockmann , and w. weise , _ nucl . phys . _ * a625 * , 758 ( 1997 ) ; j .- ballot , m. r. robilotta , and c. a. da rocha , _ phys . rev . _ * c57 * , 1574 ( 1998 ) ; e. epelbaoum , w. glckle , and u .- g . meiner , _ nucl . phys . _ * a637 * , 107 ( 1998 ) ; j. l. friar , _ phys . rev . c _ * 60 * , 034002 ( 1999 ) . p. f. bedaque and u. van kolck , _ ann . nucl . part . * 52 * ( 2002 ) . this is a comprehensive current review of the application of chpt to nuclear physics . a. manohar and h. georgi , _ nucl . _ * b234 * , 189 ( 1984 ) ; h. georgi , _ phys . lett . _ * b298 * , 187 ( 1993 ) . j. l. friar , _ few - body systems _ * 22 * , 161 ( 1997 ) . m. c. m. rentmeester , r. g. e. timmermans , j. l. friar , and j. j. de swart , _ phys . lett . _ * 82 * , 4992 ( 1999 ) . r. g. e. timmermans ( private communication ) . the nijmegen group is presently analyzing the effect of chiral two - pion - exchange nuclear forces on @xmath10 scattering . their @xmath77 results are given in ref . r. b. wiringa , v. g. j. stoks , and r. schiavilla , _ phys . rev . c _ * 51 * , 38 ( 1995 ) . u. van kolck , m. c. m. rentmeester , j. l. friar , t. goldman , and j. j. de swart , _ phys . lett . _ * 80 * , 4386 ( 1998 ) . u. van kolck , j. l. friar , and t. goldman , _ phys . _ * b371 * , 169 ( 1996 ) . j. l. friar and u. van kolck , _ phys . c _ * 60 * , 34006 ( 1999 ) . s. a. coon and j. a. niskanen , _ phys . c _ * 53 * , 1154 ( 1996 ) . j. a. niskanen , _ phys . c _ * 65 * , 037001 ( 2002 ) . r. machleidt , _ adv . _ * 19 * , 189 ( 1990 ) ; g. a. miller , _ nucl . phys . _ * a578 * , 345 ( 1990 ) ; _ chinese j. phys . _ * 32 * , 1075 ( 1994 ) . these are fairly recent reviews of charge symmetry . see also refs.@xcite . a. k. opper and e. korkmaz ( spokespersons ) , triumf e-704 proposal . a. d. bacher and e. j. stephenson ( spokespersons ) , iucf ce-82 proposal . u. van kolck , j. a. niskanen , and g. a. miller , _ phys . lett . _ * b493 * , 65 ( 2000 ) . d. o. riska and y. h. chu , _ nucl . _ * a235 * , 499 ( 1974 ) ; j. v. noble , in _ the interaction between medium energy nucleons in nuclei _ ( aip conf 97 ) , h .- o . meyer , ed . ( aip , new york , 1983 ) , p. 83 ; p. g. blunden and m. j. iqbal , _ phys . * b385 * , 25 ( 1996 ) . s. a. coon and m. d. scadron , _ phys . c _ * 26 * , 562 ( 1982 ) . j. l. friar , b. f. gibson , and g. l. payne , _ phys . rev . c _ * 35 * , 1502 ( 1987 ) . r. a. brandenburg , s. a. coon , and p. u. sauer , _ nucl . phys . _ * a294 * , 305 ( 1978 ) . y. wu , s. ishikawa , and t. sasakawa , _ phys . rev . lett . _ * 64 * , 1875 ( 1990 ) ; * 66 ( e ) * , 242 ( 1991 ) . s. c. pieper , v. r. pandharipande , r. b. wiringa , and j. carlson , _ phys . c _ * 64 * , 014001 ( 2001 ) [ see table xi ] ; s. c. pieper and r. b. wiringa , _ ann . nucl . part . sci . _ * 51 * , 53 ( 2001 ) . a. nogga , a. kievsky , h. kamada , w. glckle , l. e. marcucci , s. rosati , and m. viviani , nucl - th/0202037 . r. machleidt and h. mther , _ phys . c _ * 63 * , 034005 ( 2001 ) ; g. q. li and r. machleidt , _ phys . rev . c _ * 58 * , 1393 ( 1998 ) . s. a. coon and r. c. barrett , _ phys . c _ * 36 * , 2189 ( 1987 ) . s. a. coon , b. h. j. mckellar , and v. g. j. stoks , _ phys . * b385 * , 25 ( 1996 ) . e. epelbaum , w. glckle , and u .- g . meiner , _ nucl . phys . _ * a671 * , 295 ( 2000 ) ; d. r. entem and r. machleidt , _ phys . lett . _ * b524 * , 93 ( 2002 ) . e. epelbaum , a. nogga , w. glckle , h. kamada , u .- g . meiner , and h. witaa , _ phys . c _ * 66 * , 064001 ( 2002 ) . s. weinberg , _ phys . * 17 * , 616 ( 1966 ) ; y. tomozawa , _ nuovo cimento a _ * 46 * , 707 ( 1966 ) .	charge - symmetry breaking in the nucleon - nucleon force is investigated within an effective field theory , using a classification of isospin - violating interactions based on power - counting arguments . the relevant charge - symmetry - breaking interactions corresponding to the first two orders in the power counting are discussed , including their effects on the @xmath0he @xmath1 @xmath0h binding - energy difference . the static charge - symmetry - breaking potential linear in the nucleon - mass difference is constructed using chiral perturbation theory . explicit formulae in momentum and configuration spaces are presented . the present work completes previously obtained results . = 0.25 in = 0.25 in = .10 in = 1ex = 2ex = 6ex = 3ex plus 1000pt minus 1000pt # 1 # 1= to + + _ theoretical division , los alamos national laboratory + los alamos , nm 87545 _ + + _ department of physics , university of arizona + tucson , az 85721 _ + and + _ riken - bnl research center , brookhaven national laboratory + upton , ny 11973 _ + + _ department of physics and astronomy , university of iowa + iowa city , ia 52242 _ + + _ department of physics , new mexico state university + las cruces , nm 88003 _ + and + _ division of nuclear physics , u.s . department of energy + sc-23 , germantown building , 1000 independence avenue + washington , d.c . 20585 - 1290 _ +
the origin of strong magnetic fields @xmath2 in some compact stars , called magnetars @xcite , remains an open problem of modern astrophysics . despite the popularity of some models describing the generation of such magnetic fields , which are based on magnetohydrodynamics of stellar plasmas , none of them can satisfactory describe the observational data . these models are reviewed in ref . @xcite . recently , the methods of elementary particle physics , mainly the chiral magnetic effect ( cme ) @xcite , were applied in ref . @xcite to generate toroidal magnetic fields in a neutron star ( ns ) , and in particular to solve the problem of magnetars @xcite . the major motivation to apply cme to produce magnetic fields in ns is that the nonzero chiral imbalance of electrons @xmath3 is created in nonequlibrium urca processes , which are parity violating . it happens since ultrarelativistic left electrons are washed out from the system producing @xmath4 . the nonzero @xmath5 generates the electric current of ultrarelativistic electrons along the magnetic field . this current , in its turn , leads to the magnetic field instability resulting in the growth of a magnetic field . another possibility to utilize the electroweak interaction for the production of the magnetic field instability was proposed in refs . it consists in the fact that the induced anomalous electric current along the magnetic field gets the contribution proportional to the difference of the effective potentials of the effective electroweak interaction of left and right electrons with background matter . thus the electroweak interaction becomes a constant driver of the magnetic field instability . then in refs . @xcite this idea was applied to generate strong large scale magnetic fields in ns due to the electroweak electron - nucleon interaction . the key issue in the application of cme to generate a stellar magnetic field is the presence of left and right charged fermions in a star . strictly speaking , the possibility to separate a fermionic field into left and right chiral projections is only possible if this particle is massless , i.e. when the chiral symmetry in unbroken . despite the typical energy of an electron in the ns matter is much greater than its mass , one can not claim these electrons are chiral particles there . thus we can expect that cme for electrons is unlikely to appear in ns . this claim is also true with respect to the model in refs . note that , for the first time the fact that a nonzero particle mass destroys cme was noticed in ref . recently , in ref . @xcite , this result of ref . @xcite was confirmed in the presence of the electroweak interaction . despite of the above disappointing observation , we can still expect the existence of astrophysical media where the chiral symmetry is unbroken . it is the quark matter in the core of a hybrid star ( hs ) or in a hypothetical quark star ( qs ) . hs is a ns having the quark core . qs is based on the strange matter hypothesis . the properties of these compact stars are reviewed in ref . note that , despite of the sporadic claims of the observations of hs / qs ( see , e.g. , ref . @xcite ) , there is a certain skepticism on the existence of these compact stars . the present work is devoted to the application of the methods of refs . @xcite to describe the magnetic field instability , leading to its growth , in quark matter in hs / qs . in sec . [ sec : bfieldevol ] , we derive the kinetic equations describing the evolution of the magnetic field and chiral imbalances in quark matter . we also formulate the initial conditions corresponding to a typical astrophysical medium . then , in sec . [ sec : res ] , we present the results of the numeric solutions of these kinetic equations . finally , in sec . [ sec : disc ] , we discuss the obtained results and their applicability for modeling magnetic fields in magnetars . the computation of the helicity flip rates of quarks in their mutual collisions is provided in appendix [ sec : helflr ] . in this section we shall derive the equations for the evolution of the spectra of the magnetic helicity density and the magnetic energy density as well chiral imbalances in degenerate matter containing @xmath6 and @xmath7 quarks interacting by the parity violating electroweak forces . let us consider a dense quark matter consisting of @xmath6 and @xmath7 quarks . the density of this matter is supposed to be high enough for the chiral symmetry to be restored . in this case we can take that the quarks are effectively massless . recently , in ref . @xcite , it was shown with help of lattice simulations that the chiral symmetry has a tendency to restore in a quark matter at high density . therefore we can decompose the quark wave functions into left and right chiral components , which evolve independently , and attribute different chemical potentials @xmath8 , where @xmath9 , for each chiral component . generalizing the results of refs . @xcite , we get that , in the external magnetic field @xmath10 , there is the induced electric current @xmath11 where @xmath12 and @xmath13 are the electric charges of quarks , @xmath14 is the elementary charge , @xmath15 is the chiral imbalance , @xmath16 , and @xmath17 are the effective potentials of the electroweak interaction of left and right quarks with background fermions . the potentials @xmath17 were found in ref . @xcite on the basis of the effective lagrangian for the @xmath18 electroweak interaction , @xmath19 where @xmath20 here @xmath21 , @xmath22 , @xmath23 are the dirac matrices , @xmath24 is the fermi constant , @xmath25 is the weinberg parameter , @xmath26 are the number densities of @xmath6 and @xmath7 quarks , and @xmath27 is the element of the cabbibo - kobayashi - maskawa matrix . the matter of the star is supposed to be electrically neutral . thus we should have @xmath28 and @xmath29 , where @xmath30 is the total number density of quarks in the star . @xmath31 assuming that @xmath32 , we get that @xmath33 and @xmath34 . note that in eqs . ( [ eq : jind ] ) and ( [ eq : leff ] ) we do not account for the @xmath35 and @xmath36 interactions . however as shown in ref . @xcite , basing on the direct calculation of the two loops contribution to the photon polarization operator , that such contributions to the induced current in eq . ( [ eq : jind ] ) are vanishing . using eq . ( [ eq : jind ] ) and the results of refs . @xcite , we can obtain the system of kinetic equations for the spectra of the density of the magnetic helicity @xmath37 and of the magnetic energy density @xmath38 , as well as the chiral imbalances @xmath39 and @xmath40 , in the form , @xmath41 + \frac{1}{9 } \left [ \mu_{5d}(t)+v_{5d } \right ] \right\ } \rho_{\mathrm{b}}(k , t ) , \nonumber \\ \frac{\partial\rho_{\mathrm{b}}(k , t)}{\partial t}= & -\frac{2k^{2}}{\sigma_{\mathrm{cond}}}\rho_{\mathrm{b}}(k , t ) + \frac{2\alpha_{\mathrm{em}}}{\pi\sigma_{\mathrm{cond } } } \left\ { \frac{4}{9 } \left [ \mu_{5u}(t)+v_{5u } \right ] + \frac{1}{9 } \left [ \mu_{5d}(t)+v_{5d } \right ] \right\ } k^{2}h(k , t ) , \nonumber \\ \frac{\mathrm{d}\mu_{5u}(t)}{\mathrm{d}t}= & \frac{2\pi\alpha_{\mathrm{em}}}{\mu_{u}^{2}\sigma_{\mathrm{cond } } } \frac{4}{9 } \int\mathrm{d}k \left [ k^{2}h(k , t ) - \frac{4\alpha_{\mathrm{em}}}{\pi } \left\ { \frac{4}{9 } \left [ \mu_{5u}(t)+v_{5u } \right ] + \frac{1}{9 } \left [ \mu_{5d}(t)+v_{5d } \right ] \right\ } \rho_{\mathrm{b}}(k , t ) \right ] \nonumber \\ & - \gamma_{u}\mu_{5u}(t ) , \nonumber \\ \frac{\mathrm{d}\mu_{5d}(t)}{\mathrm{d}t } = & \frac{2\pi\alpha_{\mathrm{em}}}{\mu_{d}^{2}\sigma_{\mathrm{cond } } } \frac{1}{9 } \int\mathrm{d}k\left[k^{2}h(k , t)-\frac{4\alpha_{\mathrm{em}}}{\pi } \left\ { \frac{4}{9 } \left [ \mu_{5u}(t)+v_{5u } \right ] + \frac{1}{9 } \left [ \mu_{5d}(t)+v_{5d } \right ] \right\ } \rho_{\mathrm{b}}(k , t)\right ] \nonumber \\ & - \gamma_{d}\mu_{5d}(t),\end{aligned}\ ] ] where @xmath42 are the rates for the helicity flip in @xmath18 plasma , @xmath43 is the qed fine structure constant , @xmath44 is the electric conductivity of @xmath18 quark matter , and @xmath45 are the mean chemical potentials of @xmath6 and @xmath7 quarks . in the electroneutral @xmath18 plasma , we get that @xmath46 and @xmath47 , where @xmath48 . the functions @xmath37 and @xmath38 in eq . ( [ eq : systgen ] ) are related to the total magnetic helicity @xmath49 and the magnetic field strength by @xmath50 where @xmath51 is the normalization volume . the integration in eq . ( [ eq : hdef ] ) is over all the range of the wave number @xmath52 variation . note that we assume the isotropic spectra in eq . ( [ eq : hdef ] ) . in our model for the magnetic field generation in magnetars , we suggest that background fermions are degenerate . nevertheless there is a nonzero temperature @xmath53 of the quark matter , which is much less than the chemical potentials : @xmath54 . the conductivity of the degenerate quark matter was estimated in ref . @xcite as @xmath55 where @xmath56 is the qcd fine structure constant , @xmath57 is the initial temperature corresponding to the time @xmath58 , when the star is already in a thermal equilibrium . using eq . ( [ eq : sigmaqcd ] ) , we obtain that @xmath59 where we assume that @xmath60 . note that @xmath44 in quark matter is several orders of magnitude less than the conductivity of electrons in the nuclear matter in a neutron star @xcite . the volume density of the internal energy of degenerate background quarks is @xmath61 @xcite , where @xmath62 is the temperature independent part and @xmath63t^{2}/2 $ ] is the temperature correction . in ref . @xcite we suggested that the growth of the magnetic field is powered by the transmission of @xmath64 to the magnetic energy density @xmath65 . the energy conservation law in the magnetized @xmath18 plasma reads @xmath66 @xcite . integrating this expression with the appropriate initial condition one gets @xmath67 t^{2 } + b^{2 } = \left [ \mu_{u}^{2}+\mu_{d}^{2 } \right ] t_{0}^{2},\ ] ] where we assume that initially the thermal energy is greater than the magnetic energy , which is the case for a young pulsar . indeed , if one starts with a seed field @xmath68 , one gets that @xmath69 and @xmath70 . it means that @xmath71 where the equipartition magnetic field reads @xcite @xmath72 t_{0}^{2 } = 1.23\mu_{0}^{2}t_{0}^{2}.\ ] ] note that eq . ( [ eq : magcool ] ) describes the magnetic cooling , i.e. the temperature decreasing because of the magnetic field enhancement . as we will see later , other channels of the star cooling , such as the neutrino emission @xcite , are negligible on the time scale of the magnetic field growth in our model . the dependence of the temperature on the magnetic field is analogous to the quenching of the parameter @xmath73 in eq . ( [ eq : jind ] ) introduced in ref . @xcite ( see also ref . @xcite ) . although we suppose that the chiral symmetry is restored in the star and quarks are effectively massless , there are induced quark masses due to the interaction with dense matter . the effective masses of @xmath6 and @xmath7 quarks were computed in ref . @xcite , @xmath74 note that the effective quark masses in eq . ( [ eq : meff ] ) should be accounted for only in quarks collisions ( see appendix [ sec : helflr ] ) . it implies the transitions between left and right particles in their mutual collisions . the helicity flip rates @xmath42 for each quark types are computed in appendix [ sec : helflr ] , @xmath75 where we use eq . ( [ eq : mu5identpart ] ) . let us introduce the following dimensionless functions : @xmath76 where we assume @xmath77 , @xmath78 , @xmath79 is the star radius , @xmath80 , and @xmath81 is the minimal scale of the magnetic field , which is a free parameter . using the dimensionless parameters , @xmath82 as well as eqs . ( [ eq : sigmat ] ) , ( [ eq : magcool ] ) , and ( [ eq : gammaud ] ) , we can rewrite eq . ( [ eq : systgen ] ) in the form , @xmath83 + \mathcal{m}_{d}(\tau ) + \mathcal{v}_{d } \right ) \mathcal{r}(\kappa,\tau ) \right ] , \nonumber \\ \frac{\partial\mathcal{r}(\kappa,\tau)}{\partial\tau}= & \left ( 1-\frac{b^{2}}{b_{\mathrm{eq}}^{2 } } \right)^{5/6 } \left [ -\kappa^{2}\mathcal{r}(\kappa,\tau ) + 0.22 \left ( 4 \left [ \mathcal{m}_{u}(\tau)+\mathcal{v}_{u } \right ] + \mathcal{m}_{d}(\tau)+\mathcal{v}_{d } \right ) \kappa^{2}\mathcal{h}(\kappa,\tau ) \right ] , \nonumber \\ \frac{\mathrm{d}\mathcal{m}_{u}(\tau)}{\mathrm{d}\tau}= & 1.85 \left ( 1-\frac{b^{2}}{b_{\mathrm{eq}}^{2 } } \right)^{5/6 } \int_{1}^{\kappa_{\mathrm{max}}}\mathrm{d}\kappa \left [ \kappa^{2}\mathcal{h}(\kappa,\tau ) - 0.22 \left ( 4 \left [ \mathcal{m}_{u}(\tau)+\mathcal{v}_{u } \right ] + \mathcal{m}_{d}(\tau)+\mathcal{v}_{d } \right)\mathcal{r}(\kappa,\tau ) \right ] \nonumber \\ & - \mathcal{g}_{u}\mathcal{m}_{u}(\tau ) , \nonumber \\ \frac{\mathrm{d}\mathcal{m}_{d}(\tau)}{\mathrm{d}\tau}= & 0.29 \left ( 1-\frac{b^{2}}{b_{\mathrm{eq}}^{2 } } \right)^{5/6 } \int_{1}^{\kappa_{\mathrm{max}}}\mathrm{d}\kappa \left [ \kappa^{2}\mathcal{h}(\kappa,\tau ) - 0.22 \left ( 4 \left [ \mathcal{m}_{u}(\tau)+\mathcal{v}_{u } \right ] + \mathcal{m}_{d}(\tau)+\mathcal{v}_{d } \right ) \mathcal{r}(\kappa,\tau ) \right ] \nonumber \\ & - \mathcal{g}_{d}\mathcal{m}_{d}(\tau),\end{aligned}\ ] ] where @xmath84 , @xmath85 and @xmath86 are given in eqs . ( [ eq : hdef ] ) and ( [ eq : beq ] ) . whyle solving of eq . ( [ eq : systdmls ] ) numerically , we use the initial kolmogorov spectrum of the magnetic energy density , @xmath87 , where the constant @xmath88 can be obtained by equating the initial magnetic energy density , computed on the basis of eq . ( [ eq : hdef ] ) , to @xmath89 ( see ref . the initial spectrum of the magnetic helicity density is @xmath90 , where the parameter @xmath91 , corresponds to initially nonhelical , @xmath92 , and maximally helical , @xmath93 , fields . in ref . @xcite we found that the evolution of the magnetic field is almost independent on the initial values of the chiral imbalances @xmath94 because of the huge helicity flip rates @xmath42 . therefore we can take almost arbitrary values of @xmath94 only requiring that @xmath95 . in our simulations we shall take that @xmath96 . in this section we present the results of the numerical solution of eq . ( [ eq : systdmls ] ) with the initial conditions corresponding to a quark matter in a compact star . in fig . [ fig : bevol ] we show the amplification of the initial magnetic field @xmath68 by two or three orders of magnitude . this result is obtained by numerically solving eq . ( [ eq : systdmls ] ) with the initial conditions discussed in sec . [ sec : bfieldevol ] . these initial conditions are quite possible in a dense quark matter in a hs / qs . -.7 cm + -.7 cm one can see in fig . [ fig : bevol ] that the magnetic field reaches the saturated strength @xmath97 . this result is analogous to the findings of refs . @xcite . for @xmath98 in figs . [ 1a ] and [ 1b ] , @xmath99 ; and for @xmath100 in figs . [ 1c ] and [ 1d ] , @xmath101 . however , unlike refs . @xcite , @xmath97 in fig . [ fig : bevol ] is defined entirely by @xmath102 . the obtained @xmath97 is close to the magnetic field strength predicted in magnetars @xcite , especially for @xmath100 . the time of the magnetic field growth to @xmath97 is several orders of magnitude shorter than in refs . this fact is due to the smaller value of the electric conductivity @xmath44 in quark matter in eq . ( [ eq : sigmaqcd ] ) compared to @xmath44 for electrons in nuclear matter which we used in refs . @xcite . this fact can be understood with help of the faraday equation , @xmath103 which is equivalent to the first two lines in eq . ( [ eq : systgen ] ) . using eq . one gets that the saturation time @xmath104 , where @xmath105 is the magnetic field scale . it means that the smaller @xmath44 is , the faster the magnetic field growths to @xmath97 . moreover , we can see that short scale magnetic field should reach @xmath97 faster . the later fact , which was also established in refs . @xcite , is confirmed by the comparison of figs . [ 1a ] and [ 1b ] as well as figs . [ 1c ] and [ 1d ] . in our model of the magnetic field generation , the thermal energy of background fermions is converted to the magnetic energy . one can say that a star cools down magnetically . the typical values of @xmath106 are @xmath107 in figs . [ 1a ] and [ 1b ] and @xmath108 in figs . [ 1c ] and [ 1d ] . at such short time scales , other cooling channels , such as that due to the neutrino emission @xcite , do not contribute to the temperature evolution significantly . therefore , unlike refs . @xcite , we omit them in our present simulations . in fig . [ fig : bevol ] we can see that , although the initial magnetic helicity can be different ( see solid and dashed lines there ) , the subsequent evolution of such magnetic fields is almost indistinguishable , especially at @xmath109 . it means that , besides the generation of strong magnetic field , we also generate the magnetic helicity in quark matter . this result is in the agreement with refs . in the present work we have applied the mechanism for the magnetic field generation , proposed in refs . @xcite , to create strong large scale magnetic fields in dense quark matter . this mechanism is based on the magnetic field instability driven a parity violating electroweak interaction between particles in the system . we have established the system of kinetic equations for the spectra of the magnetic helicity density and the magnetic energy density , as well as for the chiral imbalances , and have solved it numerically . although there is a one - to - one correspondence between the mechanisms for the magnetic field generation in refs . @xcite and in the present work , the scenario described here is likely to be more realistic . as mentioned in ref . @xcite the generation of the anomalous current in eq . ( [ eq : jind ] ) is impossible for massive particles . electrons in ns are ultrarelativistic but have a nonzero mass . as found in ref . @xcite , the chiral symmetry can be restored at densities @xmath110 , that is much higher than one can expect in ns . therefore the chiral magnetic effect for electrons as well as the results of refs . @xcite are unlikely to be applied in ns . recently this fact was also mentioned in ref . @xcite . on the contrary , the chiral symmetry was found in ref . @xcite to be restored for lightest @xmath6 and @xmath7 quarks even at densities corresponding to a core of hs or in qs . accounting for the existence of the electroweak parity violating interaction between @xmath6 and @xmath7 quarks , we can conclude that the application of the methods of refs . @xcite to the quark matter in a compact star is quite plausible . we have obtained that , in quark matter , the seed magnetic field @xmath68 , which is typical in a young pulsar , is amplified up to @xmath111 , depending on the initial temperature . such magnetic fields are predicted in magnetars @xcite . therefore hs / qs can become a magnetar . the obtained growth time of the magnetic field to @xmath97 is much less than that in electron - nucleon case studied in refs . it means that , in our model , strong magnetic fields are generated quite rapidly with @xmath112 several hours after a star is in a thermal equilibrium . note that , in the present work , instead of the quenching of the parameter @xmath73 in eq . ( [ eq : jind ] ) suggested in ref . @xcite to avoid the excessive growth of the magnetic field , we used the conservation of the total energy in eq . ( [ eq : magcool ] ) and the dependence of the electric conductivity on the temperature in eq . ( [ eq : sigmat ] ) ; cf . ref . it results in a more explicit saturation of the magnetic field in eq . ( [ eq : systdmls ] ) ; cf . fig . [ fig : bevol ] . despite the plausibility of the results , several important assumptions were made . firstly , while calculating the helicity flip rates in appendix [ sec : helflr ] , we have taken that quarks exchange by plasmons in their scattering . it is , however , known ( see , e.g. , ref . @xcite ) that modified effective interaction potentials can exist in a dense degenerate matter . if one takes into account these interactions it can somehow change the values of @xmath42 . nevertheless , since the present work is a qualitative study of the magnetic field generation in the degenerate quark matter , we shall restrict ourselves to the the plasmon interaction of quarks . secondly , we have considered the simplest case of a compact star consisting of only @xmath6 and @xmath7 quarks . however , strange stars , having a certain fraction of @xmath113 quarks are also actively studied @xcite . the nonzero fraction of @xmath113 quarks , which can not exceed @xmath114 , is also required by the beta equilibrium . nonetheless @xmath113 quarks are unlikely to contribute significantly to the generation of magnetic fields in our model . firstly , the mass of an @xmath113 quark @xmath115 is quite great , i.e. the chiral symmetry will remain broken for these particles . thus , @xmath113 quarks do not contribute to the induced current in eq . ( [ eq : jind ] ) . secondly , even if @xmath113 quarks contribute to the helicity flip rates of @xmath6 and @xmath7 quarks , it will not change the evolution of the magnetic field . indeed , @xmath42 computed in appendix [ sec : helflr ] , is already great enough to wash out the initial chiral imbalances @xmath116 . any bigger contribution to @xmath42 will eliminate @xmath116 faster . however , the growth of the magnetic field is driven by @xmath117 , which is constant , rather than by @xmath118 . summarizing , we have described the generation of strong large scale magnetic fields in dense quark matter driven by the magnetic field instability caused by the electroweak interaction of quarks . the described phenomenon may well exist in the core of hs or in qs . we suggest that the obtained results can have implication to the problem of magnetars since the generated magnetic fields have strength close to that predicted in these highly magnetized compact stars . i am thankful to s.i . blinnikov , a.v . borisov , v.v . braguta , m.i . krivoruchenko , n. leite , a.e . lobanov , b.v . martemyanov , k.a . postnov , v.b . semikoz , g. sigl , m.i . vysotsky , v.i . zakharov , and v.ch . zhukovsky for useful discussions , as well as to the tomsk state university competitiveness improvement program , rfbr ( research project no . 15 - 02 - 00293 ) , and daad ( grant no . 91610946 ) for partial support . in this appendix we shall compute the helicity flip rates of @xmath6 and @xmath7 quarks in their collisions in dense matter as well as derive the kinetic equations for the chiral imbalances . as mentioned in sec . [ sec : bfieldevol ] , quarks acquire effective masses in dense matter . thus the helicity of quarks will change when the particles collide . there are three types of reactions : ( a ) scattering of identical quarks , with helicities of both particles being changed ; ( b ) scattering of different quark flavors , with helicities of both particles being changed ; and ( c ) scattering of different quark flavors , with helicity of only one particle being changed . we shall successively discuss all the cases . quarks are supposed to interact by the plasmon exchange . there are four reactions in this group : @xmath119 , @xmath120 , @xmath121 , and @xmath122 . we study in details only the process @xmath123 , where @xmath124 are the momenta of incoming quarks and @xmath125 are the momenta of outgoing quarks . in this reaction , the number of left particles is decreased by two units and the number of right particles is increased by two units . other reactions in this group can be studied analogously . the matrix element has the form , @xmath126.\ ] ] where @xmath127 and @xmath128 are the mandelstam variables . the square of the matrix element in eq . ( [ eq : mululurur ] ) is @xmath129,\end{aligned}\ ] ] where the density matrices are @xcite @xmath130 \left [ 1+\gamma^{5 } \left ( \gamma\cdot a_{1,2 } \right ) \right ] , \notag \\ \rho'_{1,2 } = & \frac{1}{2 } \left [ \left ( \gamma\cdot p'_{1,2 } \right ) + m_{u } \right ] \left [ 1+\gamma^{5 } \left ( \gamma\cdot a'_{1,2 } \right ) \right].\end{aligned}\ ] ] here @xmath131 is the effective mass given in eq . ( [ eq : meff ] ) and the polarization vectors are @xcite @xmath132 which correspond to left and right particles . choosing the center - of - mass frame of colliding quarks and assuming the elastic scattering , one gets that @xmath133 where @xmath134 is the scattering angle , i.e. the angle between @xmath135 and @xmath136 in the center - of - mass frame . in the same frame one has @xmath137 where @xmath138 is the energy of colliding quarks in the center - of - mass frame . in eq . ( [ eq : tucm ] ) , we also assume that the scattering is elastic . we can express @xmath138 in term of the variables in the laboratory frame , i.e. where the star is at rest , as @xmath139 \right\ } /2 $ ] . since we study the probability in the lowest order in the effective mass and traces in eq . ( [ eq : trululurur ] ) are proportional to @xmath140 , we neglect @xmath131 in eq . ( [ eq : tucm ] ) as well as in the following calculations . finally , eq . ( [ eq : m2ululurur ] ) takes the form @xmath141 \right\ } ^{2}}.\end{aligned}\ ] ] the total probability of the process has the form @xcite , @xmath142 \left [ 1-f \left ( e'_{2}-\mu_{u\mathrm{r } } \right ) \right],\end{aligned}\ ] ] where @xmath143^{-1}$ ] is the fermi - dirac distribution of quarks , @xmath144 is the reciprocal temperature , @xmath145 are the chemical potentials of left and right quarks , and @xmath146 is given in eq . ( [ eq : m2ushort ] ) . here we assume that quarks are degenerate , i.e. @xmath147 , where @xmath148 is the heaviside step function . ( [ eq : wululurur ] ) we introduce the additional factor @xmath149 in the denominator to take into account identical particles in the initial and final states . the direct calculation of the integrals over the phase space in eq . ( [ eq : wululurur ] ) accounting for @xmath146 in eq . ( [ eq : m2ushort ] ) gives @xmath150 analogously we can compute the probabilities of other reactions in this group . the kinetic equations for the evolution of the total number of left and right @xmath6 quarks @xmath151 are @xmath152 accounting for eq . ( [ eq : wululururfin ] ) and the analogous expression for @xmath7 quarks , one gets the evolution of the chiral imbalances @xmath153 in the form , @xmath154 in eq . ( [ eq : mu5identpart ] ) we take into account the relation between the number densities @xmath155 and chemical potentials of left and right quarks @xmath156 + 1 } \approx \frac{\mu_{\mathrm{l , r}}^{3}}{6\pi^{2}},\ ] ] where we assume that massless quarks have only one polarization . in particular we get from eq . ( [ eq : nmurel ] ) that@xmath157 . there are also four reactions : @xmath158 , @xmath159 , @xmath160 , and @xmath161 in this group . let us first study the following process : @xmath162 . the matrix element has the form , @xmath163 instead of using eqs . ( [ eq : densmatr ] ) and ( [ eq:4dpolar ] ) to compute @xmath146 , we can utilize the solution of the dirac equation , corresponding to left and right particles @xmath164 where @xmath165 are the helicity amplitudes which can be found in ref . @xcite . the spinors in eq . are normalized as @xmath166 . analogous spinors are valid for @xmath7 quarks . the direct calculation of @xmath146 with help of eq . ( [ eq : soldireq ] ) gives @xmath167 where we keep the leading term in the effective quark masses and assume the elastic scattering . analogously to eq . one obtains the total probability for the reaction @xmath168 in the form , @xmath169 after the computation of the integrals over the quark momenta in eq . ( [ eq : wuldlurdr ] ) one has @xmath170 where @xmath171 . comparing @xmath172 with , e.g. , @xmath173 in eq . ( [ eq : wululururfin ] ) , one gets that @xmath174 since @xmath175 ( see eq . and ref . it means that the contribution of the reactions in the considered group to the helicity flip rates is negligible . [ [ scattering - of - ud - quarks - only - one - particle - changes - helicity ] ] scattering of @xmath18 quarks : only one particle changes helicity + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one has eight reactions @xmath176 , @xmath177 , @xmath178 , and @xmath179 present in this group . let us first study the process @xmath180 . the matrix element for this reaction is @xmath181 the calculation of @xmath146 can be made with help of eq . ( [ eq : soldireq ] ) . here we present only the final result , @xmath182 } { \left [ 1- \left ( \mathbf{n}_{1}\cdot\mathbf{n}'_{1 } \right ) \right]},\ ] ] where as usual we assume that quarks are ultrarelativistic and the scattering is elastic . the total probability of the process @xmath183 is @xmath184,\end{aligned}\ ] ] where @xmath146 is given in eq . ( [ eq : m2uldlurdl ] ) . note that , in eq . ( [ eq : wuldlurdl ] ) we do not replace the initial and final distributions of @xmath7 quarks with step functions since @xmath7 quark does not change the helicity . the integration over the particles momenta can be made as in ref . @xcite . here we present only the final result , @xmath185 we just mention that , to get eq . ( [ eq : wuldlurdlfin ] ) we have to avoid the infrared divergence . for this purpose we introduce the plasma frequency @xmath186 in the degenerate @xmath18 quark matter @xcite . comparing eq . ( [ eq : wuldlurdlfin ] ) with eq . ( [ eq : wululururfin ] ) one can see that @xmath187 since @xmath188 in the degenerate matter . that is why the reactions in this group can be omitted as well . at the end of this appendix we mention that we do not study the influence of the electroweak interaction between quarks on the helicity flip in quark collisions . the contribution of the electroweak interaction to the scattering probability of electrons off protons was studied in ref . @xcite , where it was found that @xmath189 does not enter to the analog of eq . ( [ eq : mu5identpart ] ) for the evolution of the chiral imbalance @xmath190 . r. turolla , s. zane , and a.l . watts , magnetars : the physics behind observations . a review , rep . 78 ( 2015 ) 116901 , arxiv:1507.02924 . miransky and i.a . shovkovy , quantum field theory in a magnetic field : from quantum chromodynamics to graphene and dirac semimetals , phys . 576 1 ( 2015 ) 1209 , arxiv:1503.00732 . m. dvornikov and v.b . semikoz , magnetic field instability in a neutron star driven by the electroweak electron - nucleon interaction versus the chiral magnetic effect , phys . d 91 ( 2015 ) 061301 , arxiv:1410.6676 . m. dvornikov and v.b . semikoz , generation of the magnetic helicity in a neutron star driven by the electroweak electron - nucleon interaction , j. cosmol . phys . 05 ( 2015 ) 032 , arxiv:1503.04162 . m. dvornikov , role of particle masses in the magnetic field generation driven by the parity violating interaction , phys . b 760 ( 2016 ) 406410 . n. k. glendenning , compact stars : nuclear physics , particle physics , and general relativity , 2nd ed . , springer , new york , 2000 . dai , s.q . wang , j.s . wang , l.j . wang , and y.w . yu , the most luminous supernova asassn-15lh : signature of a newborn rapidly rotating strange quark star , astrophys . j. 817 ( 2016 ) 132137 , arxiv:1508.07745 . if we suppose that @xmath191 in eq . ( [ eq : magcool ] ) , we can rewrite eq . ( [ eq : magcool ] ) as @xmath192 . then , we account for that @xmath193 in a nuclear matter in a neutron star @xcite . the instability of the magnetic field in eq . ( [ eq : systgen ] ) proceeds from the terms containing @xmath194 and @xmath195 . thus , to avoid the excessive growth of the magnetic field , it is sufficient to replace @xmath196 or @xmath197^{-1}$ ] only in these terms @xcite .	we study the generation of strong large scale magnetic fields in dense quark matter . the magnetic field growth is owing to the magnetic field instability driven by the electroweak interaction of quarks . we discuss the situation when the chiral symmetry is unbroken in the degenerate quark matter . in this case we predict the amplification of the seed magnetic field @xmath0 to the strengths @xmath1 . in our analysis we use the typical parameters of the quark matter in the core of a hybrid star or in a quark star . we also discuss the application of the obtained results to describe the magnetic fields generation in magnetars .
precision measurements at the lep collider at cern and other colliders like e.g. slc at slac , tevatron at fermilab and hera at desy have reached a precision which has exceeded all expectations . this is true especially for electron colliders @xcite . at the moment the precision of measurements related to the parameters of the standard model of electroweak interactions reaches values up to @xmath4 @xcite . at future colliders like the lhc or the ilc ( including gigaz ) , further improvements are expected @xcite . compared to this , theoretical predictions accomplish this precision only in very few cases . to check the validity of the standard model and to be able to draw conclusions about `` new physics '' , progress in theoretical methods and their application is necessary @xcite . the complexity of the calculations and the number of the graphs which have to be calculated within perturbation theory grows considerably order by order . at second order only a few observables are calculated @xcite . still there is no method available which allows for the ( semi)automatic calculation of arbitrary processes at this order . much work has been done on nnlo calculations @xcite . while for mixing qcd and electroweak @xmath5 nnlo vertex corrections a few calculations were done several years ago @xcite , for the evaluation of two - loop three - point diagrams in the general mass case only the methods proposed in refs . @xcite have been used successfully in order to calculate electroweak nnlo corrections @xcite . with increasing energy as it will be used at the ilc , radiative corrections will become increasingly important . therefore , there is still need for independent methods to calculate general massive two - loop vertex diagrams . in this paper we present a new algorithm which allows one to calculate nnlo - corrections for general massive planar two - loop vertex functions as they arise in the standard model of electroweak interactions using feynman gauge , e.g. for the effective weak mixing angle @xmath6 @xcite occuring in @xmath7 and for other processes like @xmath8 . even though the calculation of particular physical processes is not subject of this paper , the new algorithm enables one to perform such calculations . for our algorithm we use the parallel / orthogonal space method @xcite which allows one to separate effects coming from inner momenta from those of momenta of the outer particles of the process . the benefit of the parallel / orthogonal space method is the fact that the calculation stays close to the physical process . the introduction of feynman parameters and the application of the wick rotation which estrange the calculation from the physical process for other methods are not necessary for the parallel / orthogonal space method . instead , a decay process can be calculated in the rest frame of the decaying particle . the introduction of gram determinants which might cause artificial divergences is not necessary . the analytical integration leads to still rather simple basic functions like logarithms and dilogarithms . for the remaining numerical integration the integrand can be analyzed at physical thresholds . landau singularities are mirrored directly onto the parameters of the integrands . finally , the method is totally independent from other methods and therefore allows for an independent check . up to now , the parallel / orthogonal space method was used successfully to evaluate the general massive scalar integrals in the case of the planar topology @xcite and in the case of the crossed topology @xcite . however , it was not possible to evaluate two - loop three - point functions which contain loop momenta in the numerator . the method proposed in this paper allows one to calculate also tensor integrals for the planar vertex topologies . we present a tensor reduction which reduces the planar and rotated reduced planar two - loop three - point topologies containing an arbitrary tensor structure to a set of master integrals with strongly restricted numerator structure @xcite . as for most of the master integrals which remain after tensor reduction there are already established methods available @xcite , we do not give results in these cases . the same is valid for particular mass cases @xcite and for the uv - singular parts @xcite . efficient methods for several cases including few masses have been developed in refs . we also do not dwell on the calculation of massless diagrams for which other methods are applied successfully ( see e.g. @xcite ) . as ir divergences occur in this case , there is no systematic method available within the parallel / orthogonal space method because for each special case the subtraction has to be found seperately `` by hand '' @xcite . instead , we concentrate on the most complicated part , namely the calculation of the uv - finite parts of those master integrals which keep the planar or rotated reduced planar topology after tensor reduction . two - loop integration in four - dimensional space - time needs eight integration steps . using the parallel / orthogonal space method , we can do six of them analytically using different techniques . the last two integration steps are done numerically . because most of the integrations can be performed analytically , the landau singularities of the feynman diagrams can be related graphically to the integrands in the remaining two - dimensional integration region . using different algorithms , we demonstrate the numerical stability . the algorithm for the numerical calculation of the last two integrations is developed and its reliability is demonstrated . in addition , further exhaustive numerical tests are done . while for the numerical integration programs like the monte carlo integration routine vegas @xcite and the program divonne of the library cuba @xcite are used , the implementation of the algorithms for the semi - automatical calculation in xloops @xcite is part of the work described in this paper . in the xloops - ginac project developed by the members of the thep working group in mainz @xcite , work on the automatic generation of feynman diagrams and their evaluation with analytical and numerical means is in progress . for this purpose the language for algebraic calculation ginac was developed @xcite . while most of the master two - loop integrals not considered in this paper are already implemented in xloops , the missing planar topologies will be added in the course of the work introduced here . the complete set of algorithms described in this paper is implemented under xloops - ginac . however , the implementation of parts of the algorithms for known topologies as mentioned in sec . [ sumredpro ] still has to be done . because in this spirit the implementation is not yet finished , the calculation of physical processes is not subject of this paper but will be presented in a future publication . the paper is organized as follows . in sec . 2 we introduce the main tools of the analytical calculation associated with the parallel / orthogonal space method . the tensor reduction procedure is explained in detail in sec . 3 for the non - reduced planar two - loop three - point diagram , and modifications in case of the rotated reduced planar diagrams are mentioned and explained . the tensor reduction leads to master integrals which are integrated analytically in sec . 4 up to the last two integrations . in sec . 5 we deal with the analysis of landau singularities and thresholds . 6 is devoted to the numerical integration . in both sections we present examples to show the reliability of our procedure . in sec . 7 we give our conclusions . in appendix a we deal with the integral basis . in this section we provide the reader with the tools necessary for the calculation of massive two - loop tensor vertex integrals . most of the tools were already introduced in the literature @xcite . therefore , we can be brief in presenting these tools . the integrals we have to calculate are determined by the two momenta of the produced particles , @xmath9 and @xmath10 . because the integrals are expressed in terms of covariant quantities , we are free to decompose any loop momentum @xmath11 into two covariant vectors , @xmath12 , where @xmath13 has components in the _ parallel space _ which is the linear span of the external momenta @xmath14 , while @xmath15 is the orthogonal complement with @xmath16 with components in the _ orthogonal space _ @xcite . for three - point functions we can consider the process in the rest frame of the decaying particle . in this frame the two emerging particles are produced back - to - back , defining the @xmath17-axis .- component will be written as the second component of the four - vector in the following . ] in this frame the representation @xmath18 can be used where @xmath19 is the momentum of the decaying particle with @xmath20 and @xmath21 . accordingly , the loop momenta @xmath11 and @xmath22 necessary for the description of the two - loop integral are parametrized by @xcite @xmath23 the two - dimensional vectors @xmath24 and @xmath25 are represented in polar coordinates , using the squared absolute values @xmath26 and @xmath27 and two angles , the angle @xmath28 of @xmath24 with the @xmath29-axis and the relative angle @xmath30 . we can write @xmath31 the choice of parallel and orthogonal subspaces itself is lorentz invariant , so that the calculation can still be done in any lorentz frame . the benefit of the parallel / orthogonal space method ( p / o - space method ) is that the contributions belonging to the orthogonal space can be integrated out and we are left with the contributions in parallel space only . the integration measure is accordingly written as @xmath32 where the trivial integration over @xmath28 has already been performed . using feynman gauge , the denominators of integrals occurring in two - loop vertex calculations contain up to six propagator factors . a typical factor is given by @xmath33 where @xmath34 . in the p / o - space representation this propagator factor reads @xmath35 if we replace @xmath36 , we obtain @xmath37 where @xmath38 no longer appears quadratically . this replacement , also known as _ linearization _ , is allowed because the integral for @xmath39 ranges from @xmath40 to @xmath41 . however , because of the occurrence of the mixing propagator factor @xmath42 the signs for the linearizations in @xmath11 and @xmath22 are coupled . in applications of this linearization we use the sign which is the most appropriate for our aims . the benefit of the linearization is that the integrals over @xmath38 and @xmath43 can then be calculated by using the residue theorem . the quantity @xmath17 occurs only in the just mentioned propagator factor @xmath44 . after linearization this propagator factor can be written as @xmath45 where @xmath46 @xmath44 occurs only once in the denominator ( if at all ) . the resulting integral @xmath47}{(a+i\eta)^2-b^2 } } = \frac\pi{{\sqrt[\star]}{(a+i\eta)^2 - 4st}}\ ] ] can be written in this closed form only if we make a convention different from the usual one @xcite . we stipulate that the cut of the square root is located on the positive real axis instead of the negative one . for this reason we have added the star in @xmath48{(a+i\eta)^2-b^2}$ ] which reminds one of the different cut . the occurrence of the modified square root has consequences for the subsequent integrations because in closing paths for the residue theorem we have to avoid crossing this cut . in order to find conditions for this we split the square root up into a product of two square roots @xmath49}{a\pm b+i\eta}$ ] . if one of the radicands is equal to a positive real number @xmath29 , we are definitely crossing the cut . written in terms of @xmath38 and @xmath43 , we obtain @xmath50 note that the two undetermined signs are not correlated . while the sign between @xmath51 and @xmath52 depends on which of the two square roots is taken , the global sign of the right hand side is due to the sign of the linearization . considering only the linearizations @xmath53 , @xmath54 , the signs of the real and imaginary part of the cut still depend on the sign of @xmath55 . if @xmath56 , the cut is located in the lower complex half plane . the real part starts ( for @xmath57 ) at some finite value and runs to @xmath41 , independent of which square root we selected . in order to avoid the cut , we therefore have to close the integration path for @xmath38 resp . @xmath43 in the upper complex half plane while for @xmath58 we have to close it in the lower half plane . the situation is opposite for the linearization with a minus sign . after having introduced the main tools for the calculation , we can start with the calculation itself . the starting point is the tensor integral @xmath59 where @xmath60 if one is doing standard model calculations using feynman gauge , the powers are restricted by @xmath61 and @xmath62 , as well as by @xmath63 and @xmath64 . nevertheless , the algorithm introduced in this paper works for general powers . in fig . [ momflow ] we show the momentum flow for the diagram in order to define the momenta . for the indices @xmath65 we allow non - negative integer values . in performing the tensor reduction the integrals are simplified to integrals with simpler numerator and/or denominator structure . for one - loop integrals the numerator can always be removed by reduction procedures @xcite . for two - loop integrals this need not be the case . in general there are not enough propagator factors to cancel all components of the numerator that occur . for the genuine planar two - loop three - point function all numerator factors related to the second loop momentum @xmath22 can be cancelled . this is not the case for the reduced topologies , and it is not the case for the first loop momentum @xmath11 . the aim of the tensor reduction in general is to reduce the numerator as far as possible so that the integrations can be performed similar to what is done for a trivial numerator @xcite . the reduction procedure will be explained in the following . the representation of the procedure by diagrams will show the topologies only . in this spirit plus signs like the one occuring in eq . ( [ diagmix0 ] ) have to be understood as sums over diagrams with the same topology but different factors and signs . the first step in the cancellation procedure consist in cancelling powers of the mixed factor @xmath66 . this factor occurs also in the propagator factor @xmath44 . we can write @xmath67 where @xmath68 used iteratively , we obtain @xmath69 this iterative formula is necessary because the propagator factor @xmath44 occurs only once . as applied to the integral , we obtain @xmath70 where we have skipped all the other numerator factors for convenience . the first part is the same planar integral with a different numerator structure . however , the second part no longer contains the mixing propagator factor @xmath44 . instead , the diagram factorizes . in addition , we can conclude that this vanishes if @xmath71 is odd . the reason is that the integration over @xmath17 is given by @xmath72 which vanishes if the integrand is odd . diagramatically we can write @xmath73 after having cancelled the mixing part it is irrelevant with which loop momentum we continue . however , we always start with the orthogonal space components . the factor @xmath75 occurs in the denominator factor @xmath76 . we can write @xmath77 the iterative formula @xmath78 can be used for the integral to obtain @xmath79 schematically this reduction reads @xmath80 in the following only the diagrammatic reduction will be shown . in anticipation of the linearization we replace the numerator factors @xmath38 by @xmath81 , using @xmath82 while the additional factors @xmath83 will be cancelled together with @xmath84 in the next step , the subsequent linearization will replace @xmath85 by @xmath86 . @xmath76 and @xmath87 are the only propagator factors that contain only @xmath11 and not @xmath22 . we can combine these both to obtain @xmath88 an iterative formula can be constructed as before , the reduction reads @xmath89 in using @xmath91 we can cancel the factors @xmath90 , ending up with the reduction scheme @xmath92 different from the situation for the replacement of @xmath38 , we have three propagator factors which do not contain @xmath11 . in this case the factors @xmath43 can be cancelled completely . we use @xmath93 to reduce powers of @xmath43 according to the scheme @xmath94 finally , we can cancel the factors @xmath95 as well , using @xmath96 the reduction scheme reads @xmath97 after performing the reduction procedure , we can collect all the steps into a single one . all diagrams which are still of the ( original ) planar topology after reduction procedure have a numerator factor @xmath85 in different powers . the maximal power , i.e. the maximal value for @xmath28 , is given by @xmath98 . the numerator factors of the diagrams with reduced topology are changed but not easier . nevertheless , because of the different topology , reduction procedures which are known in the literature or which will be explained in the following can be applied . schematically we write @xmath99 denoting the topologies by @xmath100 according to the cancelled propagator factor , we can cite the following references : * _ two - loop three - point functions with two - point subloop _ , as they are given in our case by the topologies @xmath101 and @xmath102 , can be calculated by using dispersion relations @xcite . * the _ factorizing topology @xmath103 _ is calculated as a product of one - loop topologies . this contributions may be calculated using xloops @xcite or similar packages like sanc @xcite . * the _ effective two - loop two - point topology @xmath104 _ can be calculated using xloops @xcite . * what is left are the _ rotated reduced planar topologies _ @xmath105 and @xmath106 . they can be reduced in a similar manner as the original planar topologies . in the following section we will deal with these topologies and the necessary modifications of the reduction procedure . for the rotated reduced planar topologies we assume the same numerator factor as for the non - reduced one , @xmath107 . the procedure is quite the same as in the case discussed before . but because one of the denominator factors @xmath108 and @xmath109 is absent , it is not possible to cancel the factors of @xmath43 as in the previous case . instead , we anticipate the linearization as in the case of @xmath38 ( and with the same sign ) in writing @xmath110 again , while the additional factors @xmath111 will be cancelled together with @xmath112 in the last step , the subsequent linearization will replace @xmath113 by @xmath114 . however , this last step makes the difference between the two topologies . for the topology @xmath105 the denominator factor @xmath108 is absent . however , we can obtain @xmath115 at this point the different signs for the linearization enter the game . while we have designed the term @xmath113 in the way that it can be combined with the factors from the cancellation of @xmath43 , in order to obtain @xmath95 we have to divide by @xmath116 however , if the second produced particle is massless , @xmath117 , this factor will become zero . in order to include also the massless case , we will prefer the linearization @xmath118 for this topology , i.e. the upper sign . the whole reduction procedure for the case of the topology @xmath105 reads @xmath119 all the diagrams on the right hand side except for the first one with maximal powers @xmath98 for @xmath28 and @xmath120 for @xmath121 are again members of the different simpler topology classes which were mentioned in the previous paragraph . looking at the topology @xmath106 , the propagator factor @xmath109 is absent . in this case we can obtain a reduction formula from combining @xmath108 and @xmath122 to obtain @xmath123 in this case the linearization @xmath124 ( lower sign ) is more appropriate . the result of the reduction reads @xmath125 after having reduced the numerator of the integrand to factors @xmath126 in case of the planar two - loop topology and @xmath127 in case of the rotated reduced planar two - loop topologies , we can start to integrate this set of master integrals @xmath128 . before we do so , we have to consider the occurrence of ultraviolet ( uv ) divergences and their treatment in terms of appropriate subtractions . as for the treatment of infrared ( ir ) divergences we refer to ref . @xcite for the scalar case . in the more general case , methods taken from ref . @xcite can be applied . however , in this paper we consider only ir - finite examples . the uv - divergences can be subdivided into three classes : divergences with respect to the two loop momenta and global divergences . it is quite obvious that if we have @xmath129 propagator factors depending only on @xmath11 , @xmath130 propagator factors depending only @xmath22 , @xmath131 propagator factors depending both on @xmath11 and @xmath22 , and a power @xmath132 in the numerator , the corresponding degrees of divergence in @xmath133 space - time dimensions are given by @xmath134 the two - loop integral is divergent in @xmath133 dimensions , if at least one of the degrees @xmath135 , @xmath136 , or @xmath137 is zero or negative . if the degree of divergence is zero , this divergence is called logarithmic divergence . in order to calculate the integral , we first have to regularize it . we use dimensional regularization and write @xmath138 . after that , integrals can be split off into a divergent and a convergent part . while the convergent part can be calculated for @xmath139 , i.e. for @xmath140 space - time dimensions , the integrand of the divergent part is simpler but has the same uv - behaviour . in an appropriate subtraction procedure , therefore , we subtract and add an integrand which is simple enough to be integrated analytically ( at least the singular part ) but has the same singularities as the integrand itself in order to cancel the singularities . this subtraction procedure is found and will be formulated multiplicatively @xcite . the subtracted integrand contains the subtraction factors @xmath141 with @xmath142 where @xmath143 , @xmath144 , @xmath145 , and @xmath146 are _ subtraction masses _ which have to be introduced artificially . the degrees of divergence for this subtracted integrand increase by @xmath147 and @xmath148 , resp . and thus may lead to integrals without singularities for arbitrary values for @xmath149 , @xmath150 , @xmath151 , and @xmath152 . finally , the limit @xmath153 can be performed , leaving the subtracted integral convergent and enabling the semi - analytical calculation . since the subtraction terms always contain a two - point subloop , the singular part can be calculated analytically . for the non - singular part one can use @xmath139 @xcite . in the end the sum of subtracted integral and subtraction terms have to be independent of the subtraction masses . in the following we will deal with the subtracted integrals only . the reduction procedure explained in the previous section leads to the integrals @xmath154 the degrees of divergence can be calculated in four space - time dimensions , @xmath155 there are no subdivergences in @xmath22 , i.e. divergences caused by the integration over the loop momentum @xmath22 . however , for @xmath156 we have to subtract subdivergences in @xmath11 . we choose @xmath157 and multiply the integrand in eq . ( [ inttal0 ] ) by @xmath158 where @xmath159 and @xmath160 are given in eq . ( [ propfac ] ) . the modified propagator factors read @xmath161 note that each of the subtraction factors in eq . ( [ subk12 ] ) improves the degree of divergence at least by @xmath162 because by power counting the denominator is given by @xmath163 and terms of lower order in @xmath11 while the numerator starts with @xmath164 . the masses @xmath165 of the subtraction can be chosen arbitrarily . however , in order to avoid the introduction of spurious ir - singularities and squared propagator factors in the denominator , the subtraction masses are usually chosen to be non - zero and different from the physical ones and from each other @xcite . furthermore , in order not to introduce new thresholds , they should be larger than the masses of the decaying particles . on the other hand , they should not be too large because of the possible occurence of numerical instabilities . after the linearization @xmath166 , in the p / o - space representation the propagator factors are given by @xmath167 in case of the ( non - reduced ) planar diagram where both linearizations are possible , we select the linearizations @xmath166 and @xmath168 for convenience . the propagator factors which are linear expressions in @xmath38 can be written in closed form expressions , @xmath169 because the parameters @xmath170 can take arbitrary values , the denominator of @xmath171 in eq . ( [ subkl ] ) together with the factors @xmath76 and @xmath87 from the integrand consists of different linear factors . integrating over @xmath38 , the integration path @xmath172 $ ] can be closed in the upper or lower complex half plane , and cauchy s theorem can be applied . the decision for one of these paths depends on the sign of the linearization and on the sign of @xmath55 ( cf . ( [ k1l1cut ] ) and its discussion ) . the poles are given by the ( different ) zeros of the propagator factors present in the calculation , the residues are given by @xmath173 = \frac1{p_2}\bigg\|_{p_1=0},\nonumber\\ & & \mbox{res\ } \left[\frac1{p_1p_2}\prod_{i=1}^{j_k } \left(1-\frac{p_1p_2}{p_{1,i}p_{2,i}}\right);p_2=0\right ] = \frac1{p_1}\bigg\|_{p_2=0},\nonumber\\ & & \mbox{res\ } \left[\frac1{p_1p_2}\prod_{i=1}^{j_k } \left(1-\frac{p_1p_2}{p_{1,i}p_{2,i}}\right);p_{1,j}=0\right ] = -\frac1{p_{2,j}}\prod_{\scriptstyle i=1\atop i\ne j}^{j_k } \left(1-\frac{p_1p_2}{p_{1,i}p_{2,i}}\right)\bigg\|_{p_{1,j}=0},\nonumber\\ & & \mbox{res\ } \left[\frac1{p_1p_2}\prod_{i=1}^{j_k } \left(1-\frac{p_1p_2}{p_{1,i}p_{2,i}}\right);p_{2,j}=0\right ] = -\frac1{p_{1,j}}\prod_{\scriptstyle i=1\atop i\ne j}^{j_k } \left(1-\frac{p_1p_2}{p_{1,i}p_{2,i}}\right)\bigg\|_{p_{2,j}=0}.\end{aligned}\ ] ] the condition @xmath174 results in @xmath175 . one obtains @xmath176 in the same way @xmath177 results in @xmath178 . one obtains @xmath179 note that @xmath180 . for the other two types of residues , however , a simplification appears after inserting the poles . because of the fact that for the semi - analytical calculation of the convergent part of the integral we use the limit @xmath181 for all @xmath182 , we obtain @xmath183 in this limit , and on the pole @xmath184 given by @xmath185 , the different propagator factors read @xmath186 obviously , the dependence on @xmath187 appears only in the numerator . we can write @xmath188 and similarly @xmath189 the different poles are given by @xmath190 for @xmath191 ( @xmath192 ) the corresponding pole can be found in the upper complex half plane . for the linearization @xmath166 and @xmath193 , the path has to be closed in the upper half plane . in this case we obtain a non - vanishing residue . for @xmath194 , however , the residue occurs only in case of @xmath195 and has the opposite sign . using @xmath196,\qquad n=0,1,2\ ] ] where @xmath197 is the step function , we can integrate over @xmath38 to obtain @xmath198 where @xmath199 is given e.g. by the inverse of @xmath49}{(a(k_1,l_1)+i\eta)^2 - 4st}$ ] ( cf . ( [ sqrts ] ) ) . for the integration over @xmath43 we have to consider the propagator factors @xmath108 , @xmath109 , and @xmath122 . because the degree of divergence @xmath136 is positive , no subtractions are needed . after the linearization @xmath168 we can write @xmath200 which again can be cast into the closed form expression @xmath201 the procedure which was explained for the integration over @xmath38 works accordingly . the poles of @xmath202 are found at @xmath203 from eq . ( [ k1l1cut ] ) we read off that the cut is located in the same complex half plane as in case of the integration over @xmath38 . therefore , the occurrence and the sign of the residue is determined by @xmath204,\qquad m=4,5,6.\ ] ] using cauchy s theorem , we obtain @xmath205 of course , the same is valid if @xmath38 is replaced by @xmath206 or @xmath207 . usually , @xmath208 and @xmath209 are different for @xmath210 . in this case we can write @xmath211 with @xmath212 we perform a partial fraction decomposition to obtain @xmath213 where @xmath214 are the corresponding coefficients including @xmath215 . a special situation occurs if @xmath216 . because of @xmath217 , in this case we have @xmath218 and the corresponding denominator factor contributes in the form @xmath219 only . however , if @xmath117 , no special case has to be taken . the situation is the opposite . ] the integrations over @xmath38 and @xmath43 and the integration over @xmath17 can be combined in the closed form expression for the convergent part @xmath220 of the master integral @xmath221 in case of @xmath181 , @xmath222}{(a_{n , m}+i\eta)^2 - 4st } } + \sum_{i=1}^{j_k}\sum_{a=0}^{2j_k-2}\frac{{\cal b}^{kl}_{0,m } { \cal n}^k_{n , i , a}s^ads}{{\sqrt[\star]}{(a_{n , i , m}+i\eta)^2 - 4st}}\right)\nonumber\\\end{aligned}\ ] ] where @xmath223\ ] ] and @xmath224 are the coefficients @xmath225 in eq . ( [ defab ] ) with @xmath38 replaced by @xmath226 and @xmath43 replaced by @xmath227 , i.e. @xmath228 and @xmath229 . for the rotated reduced planar topologies we are left with the integrals @xmath230 where in case of the topology @xmath105 ( @xmath231 ) the upper sign , in case of the topology @xmath106 ( @xmath232 ) the lower sign is valid . the degrees of divergence can be calculated to be @xmath233 for @xmath234 the integration over @xmath22 becomes divergent and a subtraction has to be performed in addition to the subtraction in @xmath11 . the subtraction factor is given by @xmath235 where in addition to @xmath236 and @xmath237 we have @xmath238 the same limitations as they were mentioned at the end of sec . [ subnonred ] for the subtraction masses @xmath239 and @xmath240 apply for the masses @xmath241 and @xmath242 as well . the integration over @xmath38 can be done in the same way as for the original ( non - reduced ) planar topology . however , special care has to be taken concerning the different linearization in case of the topology @xmath106 . for the topology @xmath106 we have @xmath243 and the step function term @xmath244 has to be replaced by @xmath245.\ ] ] for the integration over @xmath43 we have to perform a subtraction . after the linearization @xmath246 the propagator factors in the p / o - space representation are given by @xmath247 these propagator factors can be written as @xmath248 in the limit @xmath249 , which will be considered in the following , we see that @xmath250 in case of the linearization @xmath168 the integration path @xmath172 $ ] over @xmath43 can be closed in the upper complex half plane for @xmath56 and in the lower complex half plane for @xmath58 . for the other linearization @xmath251 the situation is just the opposite . because the poles are given by @xmath252 the integration ranges in @xmath253 are constrained by @xmath254&for $ x=4 $ ( topology $ { \cal t}^5$)\cr 2\pi i\left[\theta(k'_0+l'_0)\theta(-\phi_m ) -\theta(-(k'_0+l'_0))\theta(\phi_m)\right]&for $ x=5 $ ( topology $ { \cal t}^4$).\cr}\ ] ] in order to use cauchy s theorem , we calculate the residues @xmath255=\frac1{p_6}\bigg\|_{p_x=0},\nonumber\\[7pt ] & & \mbox{res\ } \left[\frac1{p_xp_6}\prod_{i=1}^{j_l}\left(1-\frac{p_x}{p_{x , i } } \right);p_{x , j}=0\right]=\frac1{p_6}\prod_{\scriptstyle i=1\atop i\ne j } \left(1-\frac{p_x}{p_{x , i}}\right)\bigg\|_{p_{x , j}=0},\nonumber\\ & & \mbox{res\ } \left[\frac1{p_xp_6}\prod_{i=1}^{j_l}\left(1-\frac{p_x}{p_{x , i } } \right);p_6=0\right]=\frac1{p_x}\prod_{i=1}^{j_l}\left(1-\frac{p_x}{p_{x , i } } \right)\bigg\|_{p_6=0}.\end{aligned}\ ] ] for the first residue we obtain @xmath256 for the second one we first calculate @xmath257 therefore , the denominator of this residuum is independent of @xmath253 while the numerator is a power series up to the power @xmath258 . we define the coefficients @xmath259 by @xmath260 the last residue appears to be a combination of a pole in @xmath253 and a power series . however , we can separate these two parts by adding and subtracting an appropriate term . this term is given by @xmath261 if we subtract this term from the result for the last residue , we obtain @xmath262 the difference no longer has a factor @xmath263 in the denominator . instead , we obtain again a power series up to the power @xmath258 . we define the coefficients @xmath264 by @xmath265 having calculated the residues , we can perform the integration over @xmath43 to obtain @xmath266 the integrations over @xmath38 and @xmath43 and the integration over @xmath17 can be combined in the closed form expression for the convergent part @xmath267 of the master integral @xmath268 in case of @xmath269 . one has @xmath270}{(a_{n , m}+i\eta)^2 - 4st}}\nonumber\\&&\qquad\qquad + \sum_{i=1}^{j_k}\sum_{a=0}^{2j_k-2}\frac{{\cal b}^{kl}_{0,m } { \cal n}^k_{n , i , a}s^a{\cal n}^l_m}{(t - t_{x,6}-i\eta)\ { \sqrt[\star]}{(a_{n , i , m}+i\eta)^2 - 4st}}\nonumber\\&&\qquad\qquad + \sum_{j=1}^{j_l}\sum_{b=0}^{j_l-1}\frac{{\cal b}^{kl}_{n,6 } { \cal n}^k_n{\cal n}^l_{m , j , b}t^b}{(s - s_{1,2}-i\eta)\ { \sqrt[\star]}{(a_{n , m , j}+i\eta)^2 - 4st}}\nonumber\\&&\qquad\qquad + \sum_{i=1}^{j_k}\sum_{j=1}^{j_l}\sum_{a=0}^{2j_k-2}\sum_{b=0}^{j_l-1 } \frac{{\cal b}^{kl}_{0,6}{\cal n}^k_{n , i , a}s^a{\cal n}^l_{m , j , b } t^b}{{\sqrt[\star]}{(a_{n , i , m , j}+i\eta)^2 - 4st}}\bigg\}.\kern96pt\end{aligned}\ ] ] for the topology @xmath105 ( @xmath231 ) , @xmath271 is given by eq . ( [ calbkl0 ] ) , i.e. it is the same as for the non - reduced planar topology . for the topology @xmath106 ( @xmath232 ) , however , we obtain @xmath272.\ ] ] the coefficients in the square roots read @xmath228 , @xmath229 , @xmath273 , and @xmath274 . the integrals over @xmath187 and @xmath253 resulting for both the non - reduced and the rotated reduced planar topology are called basic integrals . they are of the types @xmath275}{(r_0-r_ss - r_tt)^2 - 4st}},\nonumber\\ { \cal f}^s_\alpha(r_s , r_t , r_0,t_0)&=&\int_0^\infty \frac{s^\alpha ds\,dt}{(t - t_0-i\eta)\ { \sqrt[\star]}{(r_0-r_ss - r_tt)^2 - 4st}},\nonumber\\ { \cal f}^t_\beta(r_s , r_t , r_0,s_0)&=&\int_0^\infty \frac{t^\beta ds\,dt}{(s - s_0-i\eta)\ { \sqrt[\star]}{(r_0-r_ss - r_tt)^2 - 4st}},\nonumber\\ { \cal f}^{st}_{\alpha,\beta}(r_s , r_t , r_0)&=&\int_0^\infty \frac{s^\alpha t^\beta ds\,dt}{{\sqrt[\star]}{(r_0-r_ss - r_tt)^2 - 4st}}.\end{aligned}\ ] ] all these integrals are divergent except for the first one . however , after the subtraction they occur in sums in which the divergences vanish . while the integral @xmath276 is calculated in refs . @xcite , the other integrals are new . the results can be found in appendix [ appa ] in terms of the parameters @xmath277 , @xmath278 , @xmath279 , @xmath280 , and @xmath281 . in this paragraph we are dealing only with the dependence on the parameters @xmath277 , @xmath278 , and @xmath279 and on the poles found in the residue integration . in case of the integral @xmath276 we obtain @xmath282 where the signs correspond to the two linearizations @xmath283 . note that @xmath284 and @xmath285 in regions where the numerical integration has to be done . for @xmath286 we have to replace @xmath287 by @xmath288 and @xmath289 by @xmath290 . for @xmath291 we have to replace @xmath209 by @xmath292 and @xmath293 by @xmath294 . for @xmath295 , finally , both replacements have to be performed . note that up to the choice for the parameters @xmath280 , @xmath281 , @xmath277 and @xmath278 , @xmath286 and @xmath291 are the same integrals . the imaginary parts in the rational factors in the numerator are used to separate real and imaginary parts according to the sokhotsky plemely relations @xmath296 where `` p '' indicates the principal value integral . we have to consider different cases in order to analyze the analytic behaviour of the square root occurring in the integrand @xcite . the equation @xmath297 parametrizes an ellipse which separates the positive and negative values of the radicand . outside of the ellipse the radicand is positive , while for points inside the ellipse it takes negative values . the ellipse touches the axes at @xmath298 and @xmath299 . for @xmath300 , therefore , the ellipse is located in the third quadrant . but because the integration is performed in the first quadrant ( @xmath301 $ ] , @xmath302 $ ] ) , the square root will give real values and no imaginary part is produced . for @xmath303 , however , the ellipse moves into the integration region and imaginary parts enter the calculations . in order to evaluate the square root uniquely , we also consider the imaginary part of the radicand , caused by the term @xmath304 . it turns out that the sign of the imaginary part is determined by the sign of @xmath305 , the zeros of which are given by the straight line connecting the points @xmath298 and @xmath299 on the axes where the ellipse touches the axes . above this line ( as seen from the origin for @xmath303 ) , the imaginary part is negative , below this line it is positive . the situation is shown in fig . [ ellipse ] . we distinguish the following cases : * outside the ellipse and below the line ( @xmath306 , @xmath307 ) we have @xmath308}{(r_0-r_ss - r_tt)^2 - 4st}=+\sqrt{(r_0-r_ss - r_tt)^2 - 4st}\ ] ] * outside the ellipse and above the line ( @xmath306 , @xmath309 ) we have @xmath308}{(r_0-r_ss - r_tt)^2 - 4st}=-\sqrt{(r_0-r_ss - r_tt)^2 - 4st}\ ] ] * inside the ellipse ( @xmath310 ) we have @xmath308}{(r_0-r_ss - r_tt)^2 - 4st}=i\sqrt{4st-(r_0-r_ss - r_tt)^2}\ ] ] multiloop integrals with many propagator factors are influenced by their singularity structure . the pole structure can be analyzed by using feynman parametrization @xcite . starting from @xmath311 with @xmath312 outer momenta @xmath313 , @xmath314 loop momenta @xmath315 , @xmath316 inner masses @xmath317 , and @xmath318 a linear combination of @xmath313 and @xmath315 , we can apply the feynman parametrization @xmath319 by shifting the inner momenta @xmath315 to @xmath320 by finite amounts ( which are determined by the outer momenta ) , the denominator can be rewritten as @xmath321 where @xmath322 is independent of inner momenta and @xmath323 is a quadratic form in the shifted momenta @xmath320 . it can then be shown @xcite that the denominator vanishes , i.e. the integral has so - called landau singularities , if @xmath324 the landau equations can be used to analyze a given feynman diagram . for two - point functions , singularities occur if the squared outer momentum crosses thresholds or pseudothresholds . for a genuine sunset diagram as shown in fig . [ sunset ] , a two - loop two - point function with three lines of mass values @xmath325 , @xmath326 and @xmath327 connecting the incoming with the outgoing leg , these thresholds and pseudothresholds are given by @xmath328 where the three - particle threshold is located at @xmath329 . in general , the genuine threshold is the one with the highest value . if the squared momentum crosses this threshold , an imaginary part appears . according to the cutkosky rules @xcite , this imaginary part corresponds to the situation where the particles move onto their mass shell . in terms of feynman diagrams , inner lines are cut into outer legs for physical particles . this can be visualized in feynman diagrams by drawing a cutting line . in the diagrams we are working with we expect two- and three - particle thresholds . while the two - particle thresholds are related to the values of @xmath330 and @xmath331 , all the three - particle thresholds are related to the values of @xmath332 , @xmath333 anticipating examples to follow , for the subtraction terms some of the two - particle thresholds vanish while new three - particle thresholds appear . for the diagram shown on the right hand side of fig . [ planard0 ] , additional three - particle thresholds appear for @xmath334 ( @xmath335 ) and @xmath336 ( @xmath337 ) while two - particle thresholds including the masses @xmath325 and @xmath326 no longer occur . as an example we consider the original planar topology in eq . ( [ inttal0 ] ) with @xmath338 . because we are only interested in a qualitative analysis , we use the following fictitious values for masses and outgoing momenta , @xmath339 for the decaying particle we vary the value of @xmath340 in a range between @xmath341 and @xmath342 and plot the real and imaginary part . the result is shown in fig . [ landaus ] . -17pt a three - particle threshold is expected for @xmath343 . at this point we see that the imaginary part starts to differ from zero . the two - particle threshold at @xmath344 is characterized by a sharp peak in the real part , accompanied by an instant decrease of the imaginary part , leading to a vertical tangent to the curve at this point . for the second two - particle threshold in this energy range at @xmath345 , the situation is the opposite . the imaginary part shows a sharp peak while the real part increases with vertical slope . if we consider tensor integrals we have to take subtraction terms into account . as it was shown in the previous section , for the planar two - loop three - point master diagram with power @xmath346 we need a single subtraction . each subtraction replaces the propagator factors by the subtracted ones . in the present case one has @xmath347 ( for higher order subtractions the scaled propagator factors appear in higher powers ) . if we perform the limit @xmath348 for @xmath349 and @xmath350 , the propagator factors loose their dependence on the outer momenta @xmath14 . in this case the momentum scheme given for the original diagram is no longer valid . we first have to perform a partial fraction decomposition for the subtraction term , @xmath351.\ ] ] after that , the lines can be rearranged according to their momenta . the planar diagram is shown in fig . [ planard0 ] together with the subtraction . for convenience , relative factors are not mentioned in the diagrammatic representation . after having performed all other integrations analytically , only the two integrations over @xmath86 and @xmath114 are left . the integration region is given by the non - vanishing of the factors @xmath271 which read @xmath352\ ] ] depending on the linearizations @xmath36 and @xmath283 . if we take into account that @xmath353 the integration regions can be seen to be triangles which are bound by the off - diagonal @xmath354 and the different unequalities coming from @xmath287 and @xmath209 . in case of the index couples @xmath355 the integration region vanishes identically . the non - vanishing integration regions are shown in fig . [ triangles ] in case of the linearizations @xmath166 and @xmath168 . two remarks are in order at this point . * special care has to be taken for the case @xmath356 . in this case , the integration region vanishes only in the limit @xmath269 . however , we have thoroughly checked that in this limit the integrand does not diverge , i.e. the integral does not give a non - vanishing contribution even with a vanishing integration region @xcite . * in case that one of the masses vanishes , integration regions might vanish . in fig . [ triangles ] we see that if @xmath357 , i.e. if @xmath216 , the triangles @xmath358 vanish . in case of the opposite linearization and with @xmath117 , the same holds for @xmath359 instead . however , the integrand does not diverge in these cases . the integrations are performed using numerical routines like the monte carlo integration routine vegas @xcite and the program divonne of the library cuba @xcite . while numerical routines sometimes overestimate the precision of the result @xcite , the use of two different routines enables us to have a measure for the accuracy of the result . in this section we will present a few examples in order to demonstrate the reliability of our method . as a first example we consider the basic integral in eq . ( [ v02int ] ) occuring in a process where the @xmath0 boson couples to a triangle with top quark current ( cf . [ v02dia ] ) . we have started the discussion already in eq . ( [ v02int ] ) and will continue at this point . the masses are chosen to be @xmath360 where we use the values @xmath361 , @xmath362 @xcite , and @xmath363 . the subtraction masses which should have different values are taken as @xmath364 and @xmath365 . because the squared mass of the @xmath0 boson is below all possible thresholds , the imaginary part of this integral vanishes . even though physical masses are chosen in this paragraph , we stress that instead of calculating the whole process we are dealing with the occuring planar basic diagram only . we calculate the integral for @xmath366 points between @xmath367 and @xmath368 for the ( dimensionless ) parameter @xmath369 . as a next step we approximate the points by a low degree polynomial in order to extrapolate to @xmath370 . the result of this extrapolation can then be compared with the result for the integral where we used @xmath370 from the very beginning . for the integration with vegas , 260000 internal points were used , whereas for divonne we use a standard precision of @xmath371 . the results of this fit are shown in tables [ v02veg ] and [ v02div ] . .[v02div]extrapolation of the contributions for @xmath372 using divonne [ cols=">,<,<,<",options="header " , ] the results are shown in fig . [ v03gra ] . again both methods lead to the same result in the limit @xmath373 . after having performed these tests , the semi - analytical evaluation of the partial results from the different master integrals appears to be feasible . this might be done by performing the numerical integrations separately for each master integral and afterwards combining the results . however , it turns out that the more stable method is to first perform the analytical integrations separately and then to do the numerical integration for the sum of all basic integrands relevant for a given process . this is especially true for squared momenta on the threshold or close to it . as an example we use @xmath374 after the tensor reduction explained in sec . 3 , we are left with the original planar topology and the two rotated reduced planar topologies , both with different powers of @xmath85 and @xmath113 , and diagrams of simpler topologies . the master integrals for the non - reduced and rotated reduced planar topologies can be calculated by applying the subtractions explained in sec . 4 . for the convergent part the limit @xmath269 can be performed . all integrations up to the last two can be done analytically . the combination of the different uv - finite contributions with planar and rotated reduced planar topology at this point is called @xmath375 . as explained before , there are two possibilities to proceed . we can combine all analytical results to a single expression and integrate this expression numerically ( combined integration ) , or we can perform the numerical integration of the analytical expression for each of the master integrals separately and add up the results afterwards ( separated integration ) . we will perform both calculations in the following to compare them . for the physical masses and outgoing momenta we again use the values in eq . ( [ standmass ] ) . the square of the decay mass @xmath376 is chosen to be equal to the two - particle threshold @xmath377 . for the subtraction masses we take values with an offset @xmath378 which in this example runs from @xmath379 to @xmath380 , still does not need to be independent of the subtraction masses because only in the sum with the finite parts of the subtraction terms this dependence will vanish . ] @xmath381 subtraction terms and corresponding subtraction masses for the second loop momentum @xmath22 are not necessary , since in this particular case the contributing planar and rotated reduced planar topologies do not contain subdivergences in @xmath22 . the results are shown in fig . [ vexgra ] . the left hand side of fig . [ vexgra ] shows real and imaginary parts for the separated integration , the right hand side shows real and imaginary parts for the combined integration . it is obvious that the combined integration is much more stable . this result is easy to understand . assuming that the relative variances are the same for all @xmath382 partial integrals and for the complete integral , we can use the same number of intersection points in the monte carlo integration for each seperate integral and for the complete integral . in this case , the standard deviation of the seperately integrated and summed integrals is bigger by a factor of @xmath383 than the standard deviation of the complete integral @xcite . using the parallel / orthogonal space method , we have elaborated on a new method to calculate the planar and rotated reduced planar ladder two - loop three - point diagrams in the general massive case . our method does not depend on a specific physical decay process but is suitable to calculate all planar two - loop vertex diagrams which may arise in the standard model of electroweak interactions using feynman gauge . we have presented a tensor reduction which reduces the mentioned topologies with arbitrary tensor structure to a set of master integrals with strongly restricted , well - defined numerator structure . we have developed and implemented an algorithm for the semi - analytical calculation of the uv - finite part of those master integrals which still have the mentioned topologies after tensor reduction . while six of the eight integrations are done analytically , for the remaining two integrations we used different numerical methods . the integration of the uv - finite part of the important master integrals was checked exhaustively in the limit of vanishing subtraction parameters by comparing the semi - analytical calculation with the results of a numerical integration . as the final step of the semi - analytical calculation , the numerical integration turns out to be stable , i.e. one obtains the same result as for the above mentioned limit . this is valid especially in the case where the numerical integration is not done for each master integral separately but for the sum of all analytical results necessary for the given process . for the remaining master integrals and subtraction terms with simpler topologies there exist other methods . in part , these methods are already implemented in xloops . therefore , in writing links to existing codes , these cases can easily be implemented into our calculations . the still missing topologies are the two - loop three - point topologies containing a two - point subloop . these parts still have to be implemented @xcite . after having done this , we will be able to evaluate semi - analytically all ir - finite nnlo corrections containing planar two - loop vertex topologies as they arise in the standard model . finally , because the calculation of the scalar integrals for the non - planar two - loop vertex function @xcite is quite similar to the calculation of the scalar integrals for the planar two - loop vertex function @xcite , our algorithm is expected to be easily extendable to the non - planar two - loop vertex topology . having extended our algorithm in this direction , we are then able to calculate all nnlo vertex corrections as they arise in the standard model . as far as possible , the computer codes written in the course of this paper are developed and implemented for most general decay processes . because of this , we plan to include our calculations in the xloops - project in the near future . we thank h.s . do , j.g . krner , a.v . hameren , m. rogal , k. schilcher , c. schwinn , h. spiesberger , and e. tuiran for helpful discussions . m.k . wants to thank c. bauer and j. vollinga for computer - technical support . s.g.acknowledges support by the dfg as a guest scientist in mainz , by the estonian target financed project no . 0182647s04 , and by the estonian science foundation under grant no . m.k . was supported in parts by the bmbf and by the graduiertenkolleg `` eichtheorien experimentelle tests und theoretische grundlagen '' , mainz university . in this appendix we list the analytical results for the basic integrals in eq . ( [ basicint ] ) in terms of the parameters @xmath277 , @xmath278 , @xmath279 , @xmath280 , and @xmath281 . in order to calculate these integrals , we introduce cutoffs @xmath384 and @xmath385 . the results depending on these cutoffs are extremely long and contain products of two logarithms as well as dilogarithms @xcite . however , one can show that these divergences cancel in the sums in eqs . ( [ k1l1sint0 ] ) and ( [ k1l1sintx ] ) . in order to make the reader familiar with these features , we explain the calculation for the simplest example . the regularized integral is given by @xmath387}{(r_0-r_ss - r_tt+i\eta)^2 - 4st}}}\nonumber\\ & = & -\frac{\lambda_t}{r_s}\ln\left\|\frac{r_sr_0+(2-r_sr_t)\lambda_t -r_s^2\lambda_s - r_s\lambda}{2(r_sr_0+(1-r_sr_t)\lambda_t)}\right\| -\frac{\lambda_s}{r_t}\ln\left\|\frac{r_tr_0+(2-r_sr_t)\lambda_s -r_t^2\lambda_t - r_t\lambda}{2(r_tr_0+(1-r_sr_t)\lambda_s)}\right\| \nonumber\\ & & + \frac{r_0}{1-r_sr_t}\bigg(\ln\left\|\lambda_t+\frac{r_sr_0}{1-r_sr_t}\right\| + \ln\left\|\lambda_s+\frac{r_tr_0}{1-r_sr_t}\right\|+i\pi\theta(r_0 ) \nonumber\\ & & -\ln\left\|r_0(1+r_sr_t ) + ( 1-r_sr_t)(r_s\lambda_s+r_t\lambda_t+\lambda)\right\| -\ln\left\|\frac{r_0}{2(1-r_sr_t)^2}\right\|\bigg)\end{aligned}\ ] ] where @xmath388 what to do with the singularity ? as a first step we can separate the divergent part which depends on @xmath384 , @xmath385 and @xmath389 from the convergent part . but how does the divergent part cancel ? if we look at eq . ( [ k1l1sintx ] ) where @xmath386 appears , we see that it actually appears as part of a sum . if we concentrate on the summation over @xmath382 and consider how @xmath390 and @xmath391 are defined in eqs . ( [ n1jadef ] ) and ( [ n2jadef ] ) , and if we take into account that @xmath392 , we see that @xmath393 . the same is valid for @xmath394 and @xmath395 , i.e.@xmath396 . therefore , an expression which is of relevance for the subtracted integrals is the sum @xmath397 in replacing @xmath384 by @xmath398 and @xmath385 by @xmath399 with @xmath400 with arbitrary but positive and fixed values @xmath401 and @xmath402 we indeed can show that @xmath403 for @xmath404 , therefore , the divergent part will cancel in the difference . we only have to consider the convergent part @xmath405 finally , we see that @xmath406 ( cf . ( [ r0stdef ] ) including remarks given there ) . therefore , we can skip all terms which are linear in @xmath279 . the result which we have to implement is @xmath407 in the following we will only give results which are to be implemented . for the integrals with one index we will show only results for @xmath286 . results for @xmath291 can easily be obtained by replacing @xmath408 and @xmath409 . the integral @xmath386 is used for illustrative reasons only . it does not appear in calculations because the integration region vanishes . the same is valid for the integrals @xmath410 . therefore , it turns out that the integrals @xmath286 and @xmath291 are sufficient for calculations within the standard model of electroweak interactions using feynman gauge . bardin , j. phys . * g29 * ( 2003 ) 75 w. hollik _ et al . _ , acta phys . * b35 * ( 2004 ) 2533 [ lep collaborations ] , `` precision electroweak measurements on the @xmath412 resonance '' , + report no . cern - ph - ep-2005 - 041 , [ arxiv : hep - ex/0509008 ] , + submitted to physics reports j. a. aguilar - saavedra _ et al . _ [ ecfa / desy lc physics working group ] , + `` tesla technical design report part iii : physics at an @xmath413 linear collider '' , + report no . desy - tesla-2001 - 23 [ arxiv : hep - ph/0106315 ] s. heinemeyer and g. weiglein , `` electroweak precision tests with gigaz '' , + published in _ batavia 2000 , physics and experiments with future linear @xmath413 colliders _ , pp . 511514 , report no . lc - th-2001 - 001 [ arxiv : hep - ph/0012364 ] g. weiglein , eur . j. * c33 * ( 2004 ) s630 g. weiglein _ et al . _ [ lhc / lc study group ] , `` physics interplay of the lhc and the ilc '' , report no . slac - pub-10764 [ arxiv : hep - ph/0410364 ] s. heinemeyer , `` loop calculations : summary '' , + report no . cern - ph - th-2004 - 159 [ arxiv : hep - ph/0408269 ] d. kreimer , phys . lett . * b292 * ( 1992 ) 341 n.i . usyukina and a.i . davydychev , phys . * b348 * ( 1995 ) 503 a.i . davydychev and j.b . tausk , nucl . * b465 * ( 1996 ) 507 j. fujimoto , y. shimizu , k. kato and t. kaneko , int . j. mod . * c6 * ( 1995 ) 525 a.a . pivovarov , phys . atom . * 63 * ( 2000 ) 1646 [ yad . fiz . * 63n9 * ( 2000 ) 1734 ] k.g . chetyrkin , r. harlander , j.h . khn and m. steinhauser , + nucl . instrum . meth . * a389 * ( 1997 ) 354 s. groote , j.g . krner and a.a . pivovarov , phys . * b443 * ( 1998 ) 269 ; nucl . * b542 * ( 1999 ) 515 ; eur . j. * c11 * ( 1999 ) 279 ; phys . rev . * d60 * ( 1999 ) 061701 ; phys . rev . * d61 * ( 2000 ) 071501 ; phys . * d65 * ( 2002 ) 036001 ; eur . j. * c36 * ( 2004 ) 471 ; `` on the evaluation of a certain class of feynman diagrams in @xmath29-space : sunrise - type topologies at any loop order '' , report no . mz - th-05 - 08 [ arxiv : hep - ph/0506286 ] ; s. groote and a.a . pivovarov , nucl . * b580 * ( 2000 ) 459 a. ghinculov and y.p . yao , + nucl . * b516 * ( 1998 ) 385 ; phys . * d63 * ( 2001 ) 054510 a.i . davydychev and v.a . smirnov , nucl . instrum . meth . * a502 * ( 2003 ) 621 u. aglietti and r. bonciani , nucl . * b668 * ( 2003 ) 3 ; * b698 * ( 2004 ) 277 w. bernreuther , r. bonciani , t. gehrmann , r. heinesch , t. leineweber , p. mastrolia and e. remiddi , nucl . phys . * b706 * ( 2005 ) 245 ; * b712 * ( 2005 ) 229 r. bonciani , p. mastrolia and e. remiddi , nucl . * b661 * ( 2003 ) 289 + [ erratum * b702 * ( 2004 ) 359 ] ; * b676 * ( 2004 ) 399 ; * b690 * ( 2004 ) 138 ; + r. bonciani , a. ferroglia , p. mastrolia , e. remiddi and j.j . van der bij , + nucl . b701 * ( 2004 ) 121 m. caffo , h. czy and e. remiddi , nucl . * 116 * ( 2003 ) 422 p. mastrolia and e. remiddi , nucl . * 116 * ( 2003 ) 412 s. bauberger , f.a . berends , m. bhm and m. buza , nucl . * b434 * ( 1995 ) 383 ; + s. bauberger , m. bhm , g. weiglein , f.a . berends and m. buza , + nucl . suppl . * 37b * ( 1994 ) 95 p. post , ph.d . thesis , mainz university , 1997 g. passarino , nucl . * b619 * ( 2001 ) 257 a. ferroglia , m. passera , g. passarino and s. uccirati , + nucl . * b680 * ( 2004 ) 199 s. actis , a. ferroglia , g. passarino , m. passera and s. uccirati , + nucl . * b703 * ( 2004 ) 3 w. hollik , u. meier and s. uccirati , `` the effective electroweak mixing angle @xmath414 with two - loop fermionic contributions '' , + report no . mpp-2005 - 77 [ arxiv : hep - ph/0507158 ] m. awramik , m. czakon , a. freitas and g. weiglein , + phys . * 93 * ( 2004 ) 201805 ; nucl . * 135 * ( 2004 ) 119 d. kreimer , phys . atom . * 56 * ( 1993 ) 1546 a. czarnecki , u. kilian and d. kreimer , nucl . * b433 * ( 1995 ) 259 a. frink , u. kilian and d. kreimer , nucl . * b488 * ( 1997 ) 426 d. kreimer , mod . * a9 * ( 1994 ) 1105 a.v . kotikov , phys . * b259 * ( 1991 ) 314 j. fleischer , a.v . kotikov and o.l . veretin , + phys . * b417 * ( 1998 ) 163 ; nucl . * b547 * ( 1999 ) 343 v.a . smirnov and e.r . rakhmetov , + theor . * 120 * ( 1999 ) 870 [ teor . mat . * 120 * ( 1999 ) 64 ] t.g . birthwright , e.w.n . glover and p. marquard , jhep * 0409 * ( 2004 ) 042 j. fleischer , v.a . smirnov , a. frink , j.g . krner , d. kreimer , k. schilcher and j.b . tausk , eur . j. * c2 * ( 1998 ) 747 a. frink , ph.d . thesis , mainz university , 2000 t. ohl , comput . * 120 * ( 1999 ) 13 t. hahn , `` cuba : a library for multidimensional numerical integration '' , + report no . mpp-2004 - 40 [ arxiv : hep - ph/0404043 ] d. kreimer , nucl . instrum . * a389 * ( 1997 ) 323 l. brcher , nucl . instrum . meth . * a389 * ( 1997 ) 327 j. franzkowski , nucl . instrum . * a389 * ( 1997 ) 333 a. frink , j.g . krner and j.b . tausk , `` massive two - loop integrals and higgs physics '' , published in the _ proceedings of the joint particle physics meeting _ , ouranoupolis , greece , may 1997 , pp . 175200 c. bauer and h.s . do , comput . * 144 * ( 2002 ) 154 c. bauer , a. frink and r. kreckel , report no . mz - th/00 - 17 [ arxiv : cs.sc/0004015 ] d. kreimer , z. phys . * c54 * ( 1992 ) 667 d. kreimer , int . j. mod . * a8 * ( 1993 ) 1797 a. frink , diploma thesis , mainz university , 1996 g. passarino and m.j.g . veltman , nucl . * b160 * ( 1979 ) 151 a. andonov _ et al . _ , `` sancscope - v.0.41 '' , arxiv : hep - ph/0411186 r. harlander and m. steinhauser , prog . part . * 43 * ( 1999 ) 167 u. kilian , ph.d . thesis , mainz university , 1996 r.h . kleiss and a. lazopoulos , + `` error in monte carlo , quasi - error in quasi - monte carlo '' , arxiv : hep - ph/0504085 a.f.w . van hameren , ph.d . thesis , + nijmegen university , 2001 [ arxiv : hep - ph/0101094 ] s. eidelman _ et al . _ [ particle data group ] , phys . * b592 * ( 2004 ) 1	using the parallel / orthogonal space method , we calculate the planar two - loop three - point diagram and two rotated reduced planar two - loop three - point diagrams . together with the crossed topology , these diagrams are the most complicated ones in the two - loop corrections necessary , for instance , for the decay of the @xmath0 boson . instead of calculating particular decay processes , we present the new algorithm which allows one to calculate arbitrary nnlo calculations for massive planar two - loop vertex functions in the general mass case . all integration steps up to the last two ones are performed analytically and will be implemented under xloops as part of the mainz xloops - ginac project . the last two integrations are done numerically using methods like vegas and divonne . thresholds originating from landau singularities are found and discussed in detail . in order to demonstrate the numeric stability of our methods we consider particular feynman integrals which contribute to different physical processes . our results can be generalized to the case of the crossed topology . mz - th/05 - 17 + hep - ph/0508173 + august 2005 + * evaluating massive planar + two - loop tensor vertex integrals * + stefan groote@xmath1 and markus m. knodel@xmath2 + @xmath2 institut fr physik der johannes - gutenberg - universitt , + staudinger weg 7 , 55099 mainz , germany + @xmath3 fsika - keemiateaduskond , tartu likool , the 4 , 51010 tartu , estonia +
the event horizon is given by @xmath11 where @xmath12 is chosen so that the area of a sphere at constant @xmath12 equals @xmath13 . there are two roots @xmath14 and @xmath15 . we find that @xmath16 , @xmath17 , @xmath18 . the entropy is @xmath19 , and we set @xmath20 . solving for @xmath5 the fundamental thermodynamical relation is @xmath21 . the temperature is @xmath22 . in its natural coordinates the weinhold metric becomes @xcite @xmath23 introducing new coordinate @xmath24 ; @xmath25 we now find that @xmath26 in these coordinates the ruppeiner metric is given by @xmath27 this metric is flat ! to see this introduce new coordinates @xmath28 , @xmath29 . the ruppeiner metric now takes rindler form @xmath30 , @xmath31 and @xmath32 are rindler coordinates on the forward light cone in minkowski space . in fig . 1 we use the inertial coordinates @xmath33 , @xmath34 . [ cols="^,^ " , ] in higher dimensional spacetimes there can be more than one angular momentum . for the kerr black hole with a single nonzero spin we obtain the event horizon from @xcite @xmath36 the area of the event horizon is given by @xmath37 where @xmath38 is the area of the unit @xmath39-sphere . the adm mass of the hole is defined by @xmath40 . the angular momentum per unit mass is @xmath41 . we have @xmath42 and set @xmath43 and set newton s constant @xmath44 to @xmath45 to obtain the entropy @xmath46 explicit @xmath47 in arbitrary dimension can not be obtained , but we can work via the weinhold metric . the mass is @xmath48 the weinhold metric @xmath49 of the @xmath39-dimensional kerr black hole is found to be flat ! in a diagonal form @xcite : @xmath50 @xmath51 gives rindler coordinates for @xmath52 : @xmath53 . for @xmath52 the extremal limit is @xmath54 , hence @xmath55 is bounded by @xmath56 . in minkowski style coordinates the state space of the kerr black hole is a wedge bounded by @xmath57 . for @xmath58 the extremal limit is given by @xmath59 and @xmath60 where @xmath61 . hence we obtain a wedge with wider opening angle @xmath62 . remarkably , the wedge of the @xmath63 kerr black hole fills the entire light cone . this is because for black holes in @xmath63 there are no extremal limits . kerr black holes shown as a wedge in a flat minkowski space . the slope of the wedge measures approximately 80@xmath64 from the @xmath65-axis . curves of constant entropy give causal structure to the state space of the black hole.,width=226 ] in any dimension , we obtain the ruppeiner metric from the weinhold metric using the conformal relation . it is a curved metric with the curvature scalar @xmath66 in @xmath52 the curvature scalar diverges along the curve @xmath67 . in @xmath58 the curvature is reduced to @xmath68 which diverges in the extremal limit of the @xmath58 kerr black hole . for @xmath63 we have a curvature blow - up but not in the limit of extremal black hole , rather at @xmath69 . this is where emparan and myers @xcite suggest that the kerr black hole becomes unstable and changes its behavior to be like a black membrane . note that in @xmath58 for some values of the parameters , there exist `` black ring '' solutions @xcite whose entropy is larger than that of the black hole studied in our papers . a careful observation of curvature scalar @xmath70 indicates that nothing special happens to the gibbs surface of the kerr black hole _ a la _ myers - perry . we can now state a small theorem , namely : if @xmath71 then the information metric is flat @xcite . the converse does not hold . the most straightforward way to prove the theorem is to change to the new coordinates @xmath72 , and @xmath73 . an explicit calculation shows that @xmath74 this is diagonal if @xmath75 . if we introduce the new coordinate @xmath76 it is also manifestly a flat metric , and it covers a wedge shaped region . given the function @xmath77 we can reparametrize @xmath78 so that we end up with polar coordinates , or rindler coordinates if the metric is lorentzian , and read off the opening angle of the wedge . anyway the small theorem is proved . there is an exception if @xmath79 and @xmath80 , since then the metric is degenerate . we can restate our small theorem in thermodynamical language : let @xmath81 . if @xmath79 the ruppeiner metric is flat . if @xmath82 the weinhold metric is flat . an explicit example : the reissner nordstrm black hole in arbitrary spacetime dimension @xmath39 has the fundamental relation @xmath83 , @xmath84 . the ruppeiner geometry is a timelike wedge in a flat minkowski space , with an opening angle that grows with @xmath39 . it is a black hole if @xmath85 . this theorem is now helping us treating other sorts of black holes as e.g. dilaton black holes , see @xcite and forthcoming papers . for computer algebra calculations we used grtensor for maple and classi . several interesting papers has recently been published , see for instance arcioni & lozano - tellechea @xcite , shen @xcite , quevedo @xcite , mirza & zamani - nasab @xcite and ruppeiner @xcite . * weinhold * + any @xmath86 & kerr & curved & flat + & reissner nordstrm & flat & curved + @xmath52 & kerr - newman & curved & curved + & rn - anti - de sitter & curved & curved + @xmath58 & double - spin kerr & curved & curved + we thank ingemar bengtsson for valuable discussions .	the hessian of the entropy function can be thought of as a metric tensor on state space . in the context of thermodynamical fluctuation theory ruppeiner has argued that the riemannian geometry of this metric gives insight into the underlying statistical mechanical system ; the claim is supported by numerous examples . we study these geometries for some families of black holes and find that the ruppeiner geometry is flat for reissner nordstrm black holes in any dimension , while curvature singularities occur for the kerr black holes . kerr black holes have instead flat weinhold curvature . 28 december 2007 * ruppeiner geometry of black hole thermodynamics * + , * narit pidokrajt * + _ department of physics + stockholm university + se-106 91 stockholm + sweden _ + we are here not studying the spacetimes of black holes as such . instead we examine entropy @xmath0 or mass @xmath1 , where @xmath2 is charge and @xmath3 angular momentum . the ruppeiner metric @xcite is the hessian of negative entropy : @xmath4 while the weinhold metric @xcite is the hessian of @xmath5 : @xmath6 . the two metrics are conformally related @xmath7 where @xmath8 , @xmath9 and @xmath10 is the temperature .
there is much current interest in the use of coherent control in order to generate novel matter - radiation states in cavity qed and atom - optics systems @xcite . in addition , the field of cavity qed has caught the interest of workers in the field of solid - state nanostructures , since effective two - level systems can be fabricated using semiconductor quantum dots , organic molecules and even naturally - occuring biological systems such as the photosynthetic complexes lhi and lhii and in biological imaging setups involving fret ( fluoresence resonance energy transfer ) @xcite . such nanostructure systems could then be embedded in optical cavities or their equivalent , such as in the gap of a photonic band - gap material @xcite . we refer to ref . @xcite for a discussion of the size and energy - gaps of the artificial nanostructure systems which can currently be fabricated experimentally . in a parallel development , phase transitions in quantum systems are currently attracting much attention within the solid - state , atomic and quantum information communities @xcite . most of the focus within the solid - state community has been on phase transitions in electronic systems such as low - dimensional magnets @xcite while in atomic physics there has been much interest in phase transitions in cold atom gases and in atoms coupled to a cavity . in particular , a second - order phase transition , from normal to superradiant , is known to arise in the dicke model which considers @xmath0 two - state atoms ( i.e. ` spins ' or ` qubits ' @xcite ) coupled to an electromagnetic field ( i.e. bosonic cavity mode ) @xcite . the dicke model itself has been studied within the atomic physics community for fifty years , but has recently caught the attention of solid - state physicists working on arrays of quantum dots , josephson junctions , and magnetoplasmas @xcite . its extension to quantum chaos @xcite , quantum information @xcite and other exactly solvable models has also been considered recently @xcite . it has also been conjectured that superradiance could be used as a mechanism for quantum teleportation @xcite . here we extend our discussion in ref . @xcite on the exploration of novel phase transitions in atom - radiation systems exploiting the current levels of experimental expertise in the area of coherent control . the corresponding experimental set - up can be a cavity - qed , atom - optics , or nanostructure - optics system , whose energy gaps and interactions are tailored to be the required generalization of the well - known dicke model @xcite . we show that , according to the values of these control parameters , the phase transitions be driven to become first - order . the well - known dicke model from atom - optics ignores interactions between the constituent two - level systems or ` spins ' @xcite . in atomic systems where each ` spin ' is an atom , this is arguably an acceptable approximation if the atoms are neutral _ and _ the atom - atom separation @xmath1 where @xmath2 is the atomic diameter . however there are several reasons why this approximation is unlikely to be valid in typical solid - state systems . first , the ` spin ' can be represented by any nanostructure ( e.g. quantum dot ) possessing two well - defined energy levels , yet such nanostructures are not typically neutral . hence there will in general be a short - ranged ( due to screening ) electrostatic interaction between neighbouring nanostructures . second , even if each nanostructure is neutral , the typical separation distance @xmath3 between fabricated and self - organised nanostructures is typically the same as the size of the nanostructure itself . hence neutral systems such as excitonic quantum dots will still have a significant interaction between nearest neighbors @xcite . motivated by the experimental relevance of ` spin spin ' interactions , we introduce and analyze a generalised dicke hamiltonian which is relevant to current experimental setups in both the solid - state and atomic communities @xcite . we show that the presence of transverse spin spin coupling terms , leads to novel first - order phase transitions associated with super - radiance in the bosonic cavity field . a technologically important example within the solid - state community would be an array of quantum dots coupled to an optical mode . this mode could arise from an optical cavity , or a defect mode in a photonic band gap material @xcite . however we emphasise that the @xmath0 ` spins ' may correspond to _ any _ two - level system , including superconducting qubits and atoms @xcite . the bosonic field is then any field to which the corresponding spins couple @xcite . apart from the experimental prediction of novel phase transitions , our work also provides an interesting generalisation of the well - known dicke model . the method of solution that we present here is in fact valid for a wider class of hamiltonians incorporating spin spin and spin boson interactions @xcite . we follow the method of wang and hioe @xcite , whose results also proved to be valid for a wider class of dicke hamiltonians . we focus on the simple example of the dicke hamiltonian with an additional spin spin interaction in the @xmath4 direction . @xmath5 following the discussion above , the experimental spin spin interactions are likely to be short - ranged and hence only nearest - neighbor interactions are included in @xmath6 . the operators in eqs . 1 and 2 have their usual , standard meanings . to obtain the thermodynamical properties of the system , we first introduce the glauber coherent states @xmath7 of the field @xcite where @xmath8 , @xmath9 . the coherent states are complete , @xmath10 . in this basis , we may write the canonical partition function as : @xmath11 as in ref . @xcite , we adopt the following assumptions : 1 . @xmath12 and @xmath13 exist as @xmath14 ; 2 . @xmath15 can be interchanged we then find @xmath16 where @xmath17 we first rotate about the @xmath4-axis to give @xmath18 we note here that the resulting hamiltonian is of the type of an ising hamiltonian with a transverse field , and it exhibits a divergence in concurrence at its quantum phase transition ( see , e.g. , @xcite ) . this particular model is instrumental in understanding the nature of coherence in quantum systems . going back to the calculations , we may now diagonalise @xmath19 by performing a jordan - wigner transformation , passing into momentum - space and then performing a bogoliubov transformation ( see , for example , ref . we then have , in terms of momentum - space fermion operators @xmath20 , the diagonalised @xmath19 : @xmath21 with @xmath22 we may then write @xmath23 where @xmath24 from the transformation , we may associate the spin - up state with an empty orbit on the site and a spin - down state with an occupied orbital . using the commutation relations for the @xmath20 and the fact that @xmath25 ( see , for example , ref . @xcite ) , we obtain @xmath26 writing @xmath27 , @xmath28 and integrating out @xmath29 we obtain @xmath30 + \log ( 2 ) \right\ } } \ .\ ] ] we now let @xmath31 . writing @xmath32 as @xmath33 , yields @xmath34 where @xmath35 + \log ( 2 ) \right\ } \end{aligned}\ ] ] and @xmath36 from here on , we omit the @xmath37 term in @xmath38 since it only contributes an overall factor to @xmath39 . laplace s method now tells us that @xmath40 \right\}.\ ] ] denoting @xmath41 $ ] by @xmath42 , we recall that the super - radiant phase corresponds to @xmath42 having its maximum at a non - zero @xmath43 @xcite . if there is no transverse field , i.e. , if @xmath44 , and the temperature is fixed , then the maximum of @xmath42 will split continuously into two maxima symmetric about the origin as @xmath45 increases . hence the process is a continuous phase transition . however the case of non - zero @xmath46 is qualitatively different from @xmath44 . as a result of the frustration induced by the tranverse nearest - neighbour couplings , _ there are regions where the super - radiant phase transition becomes first - order . _ hence _ the system s phase transition can be driven to become first - order by suitable adjustment of the nearest - neighbour couplings_. this phenomenon of first - order phase transitions is revealed by considering the functional shape of @xmath42 , as shown in fig . [ pt ] . figure [ x - lambda ] shows the value of @xmath43 that maximises @xmath42 at fixed @xmath47 and two different values of @xmath46 . from the two lines , we can see that the spin spin coupling actually acts to inhibit the phase transition . as we increase @xmath46 from @xmath48 to @xmath49 we can see that the value to which we have to increase @xmath50 to induce a phase transition is higher . figure [ maximiser ] plots the maximiser of @xmath42 with @xmath50 fixed at a value of 1.3 . for small @xmath46 , the local ( non - zero ) maximum of @xmath42 converges to zero as we increase @xmath47 and the system is no longer super - radiant . this is no longer the case if @xmath46 is increased . in this case , @xmath42 has a global maximum when @xmath47 is small ; however as @xmath47 increases , the non - zero local maxima becomes dominant and as a result a first - order phase transition occurs . we note that the barriers between the wells are infinite in the thermodynamic limit , hence we expect that the sub - radiant state is metastable as @xmath47 increases . this observation also suggests the phenomenon of hysteresis , which awaits experimental validation . in fig . [ op ] we consider the order parameter of the transition , @xmath51 . following the same method as above , we may calculate this to be equivalent to @xmath52 with an additional @xmath53 term that comes from the imaginary part of the coherent states of the radiation field @xcite . we can see from the figure that as we lower @xmath54 we drive the system first through a first order phase transition and then through a continuous phase transition . thus we are able to achieve both a first and second order phase transition by varying the one parameter , @xmath54 . in conclusion , we have shown that the experimentally relevant spin - spin interaction in the dicke model transforms it into an ising - hamiltonian with a photon - field dependent transverse field , which allows for an existence of both first - order and second - order phase transitions as parameters vary . our results highlight the importance of spin - spin coupling terms in spin - boson systems and opens up the possibility of coherently controlling the competition between the sub - radiant and super - radiant states in experimental atom - radiation systems @xcite . 99 t. vorrath and t. brandes , phys . rev . b * 68 * , 035309 ( 2003 ) ; w.a . al - saidi and d. stroud , phys . rev . b * 65 * , 224512 ( 2002 ) ; x. zou , k. pahlke and w. mathis , quant - ph/0201011 ; s. raghavan , h. pu , p. meystre and n.p . bigelow , cond - mat/0010140 ; n. nayak , a.s . majumdar and v. bartzis , j. nonlinear optics * 24 * , 319 ( 2000 ) ; t. brandes , j. inoue and a. shimizu , cond - mat/9908448 and cond - mat/9908447 . n. lambert , c. emary and t. brandes , phys . lett . * 92 * , 073602 ( 2004 ) ; s. schneider and g.j . milburn , quant - ph/0112042 ; g. ramon , c. brif and a. mann , phys . a * 58 * , 2506 ( 1998 ) ; a. messikh , z. ficek and m.r.b . wahiddin , quant - ph/0303100 . e. hagley et al . , phys . rev . lett . * 79 * , 1 ( 1997 ) ; a. rauschenbeutel et al . , science * 288 * , 2024 ( 2000 ) ; a. imamoglu et al . lett . * 83 * , 4204 ( 1999 ) ; s.m . dutra , p.l . knight and h. moya - cessa , phys . a * 49 * , 1993 ( 1994 ) ; y. yamamoto and r. slusher , physics today , june ( 1993 ) , p. 66 ; d.k . young , l. zhang , d.d . awschalom and e.l . hu , phys . rev . b * 66 * , 081307 ( 2002 ) ; g.s . solomon , m. pelton and y. yamamoto , phys . lett . * 86 * , 3903 ( 2001 ) ; b. moller , m.v . artemyev and u. woggon , appl . * 80 * , 3253 ( 2002 ) ; n. f. johnson , j. phys . matter 7 , 965 ( 1995 ) .	we propose the use of coherent control of a multi - qubit cavity qed system in order to explore novel phase transition phenomena in a general class of multi - qubit cavity systems . in addition to atomic systems , the associated super - radiant phase transitions should be observable in a variety of solid - state experimental systems , including the technologically important case of interacting quantum dots coupled to an optical cavity mode .
theoretical ideas about the first , metal - free stars have developed rapidly in recent years , despite the absence of direct observational evidence of their properties . as a specific prediction of the big bang theory , metal - free first stars confront both cosmology and astrophysics generally with important questions : when did the first stars form ? what was their initial mass function ( imf ) ? what were their lifetimes and feedback effects ? where are their remnants ( compact objects and/or metal enrichment ) ? the standard tools of stellar evolution theory have been applied to calculate the evolving structure and radiative spectra of metal - free stars , given their mass ( tumlinson & shull 2000 ; marigo et al . 2000 ; bromm , kudritzki , & loeb 2001 ; schaerer 2002 ; tumlinson , shull , & venkatesan 2003 , hereafter tsv03 ) . many uncertainties remain about the cosmological context and feedback effects of the first stars , the largest of which concern the duration of the metal - free phase and the imf . despite these unknowns , theoretical progress on the first stars has arrived at six points of consensus : * the first stars form from metal - free gas in the first collapsed dark matter halos around @xmath7 ( ricotti , gnedin , & shull 2002a , b ) . * the first stars may be very massive ( @xmath12 @xmath1 ) , because they are restricted during their formation to inefficient cooling by molecular hydrogen below @xmath13 k ( abel , bryan , & norman 2000 ) . * these very massive first stars seed their own halos and possibly enrich nearby ones by releasing metals from pair - instability supernovae ( 140 260 @xmath1 ; heger & woosley 2002 , hereafter hw02 ) . * at critical metallicity @xmath14 , protostellar clouds are able to cool and fragment more efficiently , leading to a `` normal '' imf ( bromm & loeb 2003 ; schneider et al . where this metallicity has not been reached , the imf retains its unusual properties . halo and igm enrichment probably occur inhomogeneously in time and space during the transition to `` normal '' star formation . * the first stars begin , and may complete the reionization of intergalactic h i ( gnedin & ostriker 1997 ) and he ii ( venkatesan , tumlinson , & shull 2003 , hereafter vts03 ) . the total first - stars contribution to the global ionizing photon budget is unknown , but the efficient ionization of vms is thought to be required . there may be partial recombination during the transition phase in metallicity , followed by a second reionization near @xmath15 ( cen 2003a ; wyithe & loeb 2003 ; vts03 ) . * the first - stars epoch ends when all star forming regions have achieved the critical metallicity . this is thought to occur before the final stages of h i reionization at @xmath16 . this consensus is not universally accepted in all details , but it serves as a working hypothesis in the literature . point ( 2 ) , the _ vms hypothesis _ , has appeared often in studies of the first stars concerned with their cosmic origins , their sites of formation , and their importance for reionization . in this paper , we examine this hypothesis with the objective of constructing an imf for the first stars that is consistent with current observational constraints . we meet this goal by examining the arguments for the vms hypothesis and by investigating how its key lines of evidence can also support an imf without vms . our critique of the vms hypothesis addresses consensus points ( 3 ) and ( 5 ) particularly . the empirical basis of point ( 3 ) is the work of oh et al . ( 2001 ) and qian & wasserburg ( 2002 ) , who argued that the relative metal abundances in extremely metal - poor ( emp ) halo stars match the expected signatures from vms ( hw02 ) . their conclusion has since been adopted as positive evidence of vms metal - enrichment and ionizing photon production in the first stellar generation . we argue that this line of evidence is not supported by the broader sets of abundance data presently available . instead , we adopt the view of umeda & nomoto ( 2004 , hereafter un04 ) that these observed abundances are better matched by core - collapse supernovae or `` hypernovae '' from 10 50 @xmath1 progenitors . we then examine point ( 5 ) in light of evolving spectra models for metal - free stars ( tsv03 ) and recent studies of star formation at zero metallicity . vms have been suggested as the most viable cause of the high thomson electron - scattering optical depth @xmath17 derived from the cosmic microwave background ( cmb ) data obtained by the wilkinson microwave anisotropy probe ( _ wmap _ ; spergel et al . 2003 ) . instead , we find that reionization models incorporating an imf that excludes both low - mass stars and vms can meet the reionization constraint . the lower but equally uncertain value @xmath18 found by joint analysis of _ wmap _ and the sdss matter power spectrum by tegmark et al . ( 2004 ) is reproduced by our models , but since it is also easier to accommodate to a standard imf we conclude that the optical depth to reionization is not a decisive constraint on the imf at early times . because the ionizing efficiency of metal - free stars increases more rapidly from 1 50 @xmath1 than it does above this range ( see 4 ) , we are able to construct an imf with ionizing properties similar to those of vms . we conclude , on the basis of these results , that vms are not necessary to meet the existing constraints commonly taken to motivate them . we also do not need to invoke high values for the escape fraction of ionizing radiation or the star formation efficiency in halos ( cf . , oh et al . 2001 ; somerville , bullock & livio 2003 ) . this paper is structured as follows . first , in 2 , we review the theoretical motivation for vms in the first stellar generation and distinguish two versions of the vms hypothesis . in 3 we review the extremely metal - poor halo star abundances and compare specific patterns in the nucleosynthetic evidence to theoretical yields . in 4 we use our results to motivate a range of possible first - stars imfs and derive a timescale for the epoch of metal - free star formation . in 5 we discuss semi - analytic reionization models that compare these different imfs to vms . in 6 we draw general conclusions and discuss some unsolved problems in the study of the first stars . we define vms to include stars with @xmath19 @xmath1 , the lowest mass at which metal - free stars explode as pair - instability supernovae ( pisn ; hw02 ) . this is the only clear mass cutoff higher than the uncertain @xmath20 @xmath1 division between core - collapse sn and complete collapse to black holes ( hw02 ) . the @xmath19 @xmath1 definition cleanly separates two mass regimes with very different nucleosynthetic signatures ( see 3 ) but similar radiative properties . the mass cutoff is rigid for the purposes of definition only ; it may be more loosely defined in reality without interfering with our conclusions , provided it marks a qualitative change in the nucleosynthetic signature of the stars . the theoretical motivation for the vms hypothesis rests primarily on the microscopic processes of radiative cooling in low - density primordial gas . at @xmath21 k , cooling by h and he is negligible and metal - line cooling is dominated by [ c ii ] , [ o i ] , and [ si ii ] fine - structure lines . because metal - free gas is restricted to relatively inefficient cooling by h@xmath22 at @xmath21 k , cooling rates are reduced , and the balance between pressure support and gravitational collapse is shifted to higher temperatures . primordial protostellar objects , being unable to cool or fragment , must therefore be more massive to overcome their elevated gas pressure . the final star should also be more massive if all the cloud can accrete . this simple physical description was rigorously implemented by abel , bryan , & norman ( 2000 ) , who used an adaptive mesh refinement hydrodynamics code to follow the evolution of metal - free gas from cosmological initial conditions down to the formation of a self - gravitating object with @xmath23 @xmath1 . the simulation was halted when the core became optically thick to 2 line emission , which occurred before the object was completely formed . this study implies vms with @xmath0 @xmath1 , if all the gravitationally bound gas can accrete to a single star . recent studies ( omukai & palla 2002 ; tan & mckee 2002 ; bromm & loeb 2004 ) , taking the abel et al . ( 2000 ) result as a starting point for semi - analytic models of primordial star formation , indicate that reality is complicated by stellar feedback on the accreting matter . omukai & palla ( 2002 ) identified a critical accretion rate ( @xmath24 @xmath25 @xmath1 yr@xmath26 ) below which laminar accretion onto a spherically - symmetric protostar can proceed to @xmath12 @xmath1unimpeded by feedback . above this critical rate , the accretion luminosity exceeds the eddington limit and the flow is reversed at 100 300 @xmath1 . models by bromm & loeb ( 2004 ) have found a conservative upper limit of 500 @xmath1 , by setting the time available to accretion to be the stellar lifetime ( @xmath27 yr ) but ignoring feedback . tan & mckee ( 2002 ) consider the effects of disk accretion geometry , rotation , and radiation feedback in limiting the mass of primordial stars . they find that these feedback mechanisms are likely to operate at @xmath28 @xmath1 , perhaps limiting the masses of metal - free stars to this range . however , they later found ( tan & mckee 2004 ) that none of the feedback mechanisms considered can halt the accretion before the star achieves @xmath29 @xmath1 . this important result indicates that both vms and low - mass stars may not form at zero metallicity and suggests that the mass limits of the primordial imf may be quite different from today . in 4 we use this idea and empirical constraints to construct an imf for the first stars . these theoretical studies of primordial protogalaxies and star formation posit that the first generation of stars was partially , and perhaps completely , composed of vms . to capture existing evidential claims , we introduce a distinction between the _ strong _ vms hypothesis ( `` the first generation were exclusively vms '' ) and the _ weak _ vms hypothesis ( `` the first generation included vms in addition to @xmath30 @xmath1 stars '' ) . both versions appear in the literature , although not with these names . the evidence can bear differently on the two versions of the hypothesis , because the ionizing properties and nucleosynthetic yields vary across the mass function . these labels do not connote the relative merit of the two versions ; they suggest only that the `` strong '' hypothesis is more readily falsified than the `` weak '' . the `` weak '' hypothesis is expected to be more difficult to constrain , because it includes low - mass stars and so is more like the present - day circumstances . we evaluate these versions of the hypothesis below . the hw02 stellar evolution and supernova models show that , between 140 260 @xmath1 , zero - metallicity stars are highly susceptible to the pair - production instability after core he depletion . this instability quickly disrupts the star , ejects metals , and leaves no compact remnant . because it is triggered at an unusually early period in the star s evolution , this pair - instability supernova ( pisn ) produces an unusual nucleosynthetic signature . if the vms hypothesis is correct , the signature of pisn could appear in the metal abundances of `` second - generation '' emps in the galactic halo ( see 3 ) . qian & wasserburg ( 2002 ) and oh et al . ( 2001 ) used this idea to argue that these stars justify the vms hypothesis . models with vms have also been proposed to explain the current data on igm reionization , specifically the high electron - scattering optical depth @xmath2 found in the _ wmap _ cmb data . for the simplest possible reionization history ( a step function at fixed @xmath31 ) , this result implies that complete reionization occurred at @xmath32 ( spergel et al . 2003 ; kogut et al . recent work that combines the three - dimensional matter power spectrum , @xmath33 , from the sloan digital sky survey with _ wmap _ data finds that @xmath34 ( tegmark et al . the large uncertainties arise from the fact that the first - year _ wmap _ @xmath2 results were based on temperature - polarization correlations , which are an indirect measure of @xmath2 . additional _ wmap _ data will allow auto - correlation of the polarized cmb maps and should provide a direct and more precise determination of @xmath35 . when joined with spectroscopic observations that indicate a rapidly rising h i gunn - peterson opacity up to @xmath36 6.4 ( becker et al . 2001 ; djorgovski et al . 2001 ) , the _ wmap _ result implies that the igm experienced a relatively complex ionization history , with extended periods of ionization in h i and/or he ii at @xmath36 6 20 ( vts03 ; cen 2003a ; wyithe & loeb 2003 ) . in a study of the ionizing properties of vms , bromm , kudritzki , & loeb ( 2001 ) found vms with @xmath37 @xmath1 to be an order of magnitude more efficient , _ on the zero - age main sequence _ , at generating ionizing photons per unit stellar mass than are stars in a normal imf with @xmath38 @xmath1(tumlinson & shull 2000 ) . vms produce efficient ionization and have short lifetimes , and they are thought to form in the metal - free gas of collapsed halos at high @xmath39 . in light of these features , scenarios involving vms in combination with subsequent stellar populations have been widely proposed as an attractive explanation for the observed ionization and thermal history of the igm ( hui & haiman 2003 ) and the large detected @xmath2 to the cmb ( cen 2003b ; haiman & holder 2003 ; fukugita & kawasaki 2003 ; ciardi , ferrara & white 2003 ; kaplinghat et al . 2003 ) . these `` nucleosynthesis '' and `` reionization '' lines of evidence are often taken to support the vms hypothesis . but are vms required ? are there other viable models for the emp metal abundances and reionization that do not include vms ? we argue that neither line of evidence is compelling . we take up the nucleosynthesis arguments first . many studies of reionization by the first stars ( 2 ) have adopted the vms hypothesis , citing as evidence the apparent match between pisn yields and the metal abundances of emp halo stars ( oh et al . 2001 ; qian & wasserburg 2002 ) . a full review of the literature on metal abundances ( elements li to u ) in emp stars is beyond the scope of this study . the review by mcwilliam ( 1997 ) gives a survey up to that time , and papers that cite it ( available from the astrophysical data service ) provide useful entry points to the more recent literature . we focus here on three clear features in galactic halo emp abundances which are confirmed by multiple studies and which bear directly on the vms hypothesis . these are the specific fe - peak element ratios ( especially zn ) , the widespread presence of @xmath4-process elements , and elevated [ c , n , o / fe ] ratios . we summarize these features here , and in 3.2 we compare them in detail to theoretical models incorporating the vms hypothesis . _ iron - peak elements ( cr zn ) : _ because these elements are produced in explosive events , they are sensitive to the energy , rotation , mass cut , and asymmetry of the supernova and to the pre - sn stellar properties . large samples presented by mcwilliam et al . ( 1995 ) and carretta et al . ( 2002 ) found that the fe - peak ratios to fe qualitatively change their behavior at [ fe / h ] @xmath40 . below this metallicity , [ zn / fe ] and [ co / fe ] increase while [ cr / fe ] and [ mn / fe ] decrease . this trend was interpreted as a change in the nature of the supernovae progenitors , and the increased scatter was attributed to inhomogeneous enrichment . the more recent and higher resolution study by cayrel et al . ( 2003 ) found lower scatter and a smooth rather than abrupt change in the abundance ratios at [ fe / h ] @xmath41 to -4 . in any case , the overall abundance pattern is roughly consistent across the three studies , and it provides a good test for theoretical yields . _ @xmath4-process elements ( a @xmath42 100 ) : _ these elements are thought to be produced by rapid neutron capture in hot , dense , neutron - rich environments during explosive events . although the exact physical sites are still uncertain , the proposed mechanisms are all associated with massive stars in the range @xmath5 @xmath1 ( truran et al . the absolute abundances and relative ratios of @xmath4-process elements are thus sensitive indicators of core - collapse supernova activity . the existing samples of emp stars show widespread evidence of @xmath4-process elements down to [ fe / h ] @xmath43 ( mcwilliam et al . 1995 ; burris et al . 2000 ) . the mean [ @xmath4/fe ] is similar to the solar value at all [ fe / h ] , but with up to 2 dex scatter at [ fe / h ] @xmath43 . the relative abundances ( i.e. , [ eu / ba ] ) are also similar to the solar values . _ primary elements ( c , n , o ) : _ of these direct products of main - sequence stellar nucleosynthesis , c is easily found in emps , but n and o are difficult to measure . many emp stars are c - rich relative to fe . further study of this class of stars may ultimately reveal them to be _ iron - poor _ , rather than _ metal - poor _ , and therefore less chemically primitive than their low [ fe / h ] makes them appear . as pointed out by bromm & loeb ( 2003 ) , c and not fe is the dominant interstellar coolant that may influence the early imf . as shown in tsv03 , c is the nuclear burning catalyst that determines by its absence the unusual behavior of the first stars . these factors favor c instead of fe as the `` reference element '' for finding chemically primitive stars in the local universe . by this measure , these c - enhanced stars may be less primitive than those with higher [ fe / h ] . for now however , we follow the observational studies by assuming they are in same class as the other emps . in this section , we describe the nucleosynthetic evidence for vms as stated in the literature . observational studies of emp metal abundances have regarded the trends in the fe - peak and @xmath4-process elements as the most critical constraints on the nature of the first stars . we examine the model of wasserburg & qian ( 2000 ) and qian & wasserburg ( 2002 ) , which represents only one example from a broad class of possible models that could explain the observed variations . we can not address all the possible variations in such a class , so we use this one because it admits vms objects as a possible source of early enrichment . we later use the phenomenological model of fields , truran , & cowan ( 2002 ) to illustrate why a successful model for the observed trends in @xmath4-process abundances has difficulty incorporating vms . wasserburg & qian ( 2000 ; hereafter wq00 ) examined the behavior of [ @xmath4/fe]with [ fe / h ] in the mcwilliam et al . ( 1995 ) sample and argued that the qualitative change in [ @xmath4/fe ] at [ fe / h ] @xmath44 presented an `` iron conundrum '' ; namely , that the wide dispersion in eu and ba abundances at [ fe / h ] @xmath40 suggested unrelated sources of fe and @xmath4 elements . in response , they proposed a `` prompt '' ( p ) inventory of fe production by an initial population with large fe yields but little or no @xmath4 production . this initial inventory ceased at [ fe / h ] @xmath45 , where the onset of high - frequency ( hf ) events ( sn ii ; @xmath46 yr ) and low - frequency ( lf ) events ( sn ia ; @xmath47 ) initiated a correlation between @xmath4 and fe . in this model , the wide variation in fe for small barium abundance @xmath48(ba ) is attributed to diverse fe yields for p events or to multiple p progenitors . the low @xmath48(ba ) at [ fe / h ] @xmath44 is attributed to small co - production of @xmath4 elements by p events or to dilution of a modest hf contribution . wq00 made no specific claims about the mass or metallicity of the p inventory sources . after establishing the general properties of the p inventory in wq00 , qian & wasserburg ( 2001 ) and qian & wasserburg ( 2002 ; hereafter qw02 ) noted that vms of a few hundred @xmath1 possess the fe - rich , @xmath4-poor yields and short lifetimes ( @xmath49 myr ) necessary to produce the p inventory fe over the short time ( @xmath50 yr ) before the onset of hf events . this vms model readily accommodates the qualitative change in observed abundances at [ fe / h ] @xmath51 , near the @xmath52 threshold where metal - line cooling is expected to change the imf ( bromm & loeb 2003 ) . this model was extended by oh et al . ( 2001 ) to link the vms hypothesis to reionization . they pointed out that the trends in [ fe / h ] and ba also appear in si and ca , elements produced by pisn , but not in c , which is not abundantly produced by pisn . their conclusion has since been adopted as positive evidence that vms dominated metal enrichment from the first stellar generation . on this assumption , oh et al . ( 2001 ) argued that the fe enrichment in the p inventory ( [ fe / h ] @xmath53 ) implies cumulative cosmic star formation activity sufficient to reionize the universe if all star formation is in vms . this result forms the empirical basis of the contention that vms could explain both the observed emp metal abundances and reionization together ( see points 3 and 5 in the introduction ) . however , closer scrutiny shows that the p inventory does not match all the observations . the p - inventory model is incompatible with the strong vms hypothesis , because vms have no significant post - he nuclear burning and therefore produce no @xmath4 elements ( hw02 ) . if vms produce all the fe up to [ fe / h ] @xmath43 , the @xmath4 elements should be absent instead of appearing at [ @xmath4/fe ] @xmath54 as observed . progenitors with @xmath55 @xmath1 are needed in the first generation to provide the observed @xmath4-process elements ( truran et al . 2002 ) , provided that the @xmath4-process mechanisms that operate at finite metallicity can be safely extended to @xmath56 . thus , the strong vms hypothesis is convincingly excluded , and the weak hypothesis is more naturally accommodated by the p - inventory model . the p - inventory model also provides a means for testing the weak vms hypothesis , because it attributes virtually all the fe to vms . to see why , we follow wq00 and estimate the fraction of fe in the p inventory that could be attributed to hf events . if we take the lowest @xmath48(ba ) to represent the contribution of a single hf event , then 40 such events contribute to the scatter in @xmath48(ba ) at [ fe / h ] @xmath57 for a modest increase in [ fe / h ] . this small gain in fe enrichment limits the hf contribution to the p inventory to @xmath58% . another approach to this proof comes from consideration of a model imf . if we assume a salpeter imf with @xmath59 @xmath1 , adopt hw02 yields for vms and a uniform fe yield per star , @xmath60 @xmath1 for @xmath61 @xmath1 , we find that the vms mass range produces @xmath8 2/3 of the total fe yield over time ( more if the imf slope or mass limits are changed to favor high masses ) . thus , any element x with mass yield from vms similar to fe should appear with the appropriate ratio [ x / fe ] in the p inventory . in the top panel of figure 1 we compare the pisn yields ( hw02 ) to the emp metal abundances from three sources ( mcwilliam et al . 1995 ; carretta et al . 2000 ; cayrel et al . we plot these samples separately and mark the theoretical pisn yields for three progenitor mass ranges with filled boxes . yields for unobserved elements are included for completeness and marked with lighter colors . we see that the observed abundances are consistent across the different studies for elements they have in common . clearly the high - mass pisn match well with si , ca ( as pointed out by oh et al . 2001 ) , and to some extent with mn , but they are a poor match to the other elements . they overpredict cr , underpredict v , co , and produce little zn . the vms models are a poor fit to the data , especially for the fe - peak elements co and zn . the separation by mass range emphasizes that enrichment by one or a few stars is a much poorer fit to the observed data than a mass - weighted ( imf - averaged ) yield . some elements are matched best by the lower mass range , while others favor the higher masses ( which have fe yields of up to 40 @xmath1 ) . thus , the data appear to be inconsistent with a model , such as the p inventory , in which these metal - poor stars formed from gas that was enriched only once or a few times by vms precursors . the generally poor match of pisn yields leads us to conclude that vms were probably not the precursors of these emps . the phenomenological model for @xmath4-process patterns by fields , truran , & cowan ( 2002 ) does not make specific reference to vms , but it is a good example of another class of models that can test the vms hypothesis . this study explains the @xmath8 2 dex scatter in [ @xmath4/fe ] by a stochastic treatment of supernova enrichment that defines two types of sn with identical fe mass yield but very different [ @xmath4/fe ] ( class a , high [ @xmath4/fe ] ; class b , low or zero [ @xmath4/fe ] ) . their artificial halo star sample reproduces the observed scatter in [ eu / fe ] below [ fe / h ] @xmath62 when a events comprise @xmath63 5% of all sn . in this model , a events produce all the @xmath4-process elements , and b events make virtually all the fe . the assumed uniform fe mass yield was adopted from sn 1987a ( woosley et al . 1994 ) , but it also matches quite well to the fe yields of vms pisn from 140 160 @xmath1 . the vms pisn represent a good candidate for the fe - producing but @xmath4-process - poor b events , but in fact this model illustrates the difficulty with any model that attempts to attribute the emp fe budget to vms . as shown above ( and in figure 1 ) , the observed abundances of fe - peak elements in emps are not a good match to pisn . although the absolute fe yields from vms in the 140 160 @xmath1 range serve well as b events , stars in this mass range produce negligible zn . the fields et al . ( 2002 ) model is a generalization of the wq00 p - inventory model that accurately describes the observed scatter in [ @xmath4/fe ] at low [ fe / h ] . both models thoroughly contradict the strong vms hypothesis and are inconsistent with the weak version . the fields et al . ( 2002 ) model is successful without including vms , so the general class of models it represents ( including wq00 ) does not provide positive evidence for vms . some of the recent detections of galactic halo stars with [ fe / h ] @xmath64 but enhanced [ c , n , o / fe ] present another puzzle to nucleosynthesis models . the lowest [ fe / h ] star ( he0107 - 5240 , [ fe / h ] @xmath65 ; christlieb et al . 2002 ) exhibits [ c / fe ] = 4.0 and [ n / fe ] = 2.3 ( o was not measured ) , so it may be more accurate to call this star `` iron poor '' than `` metal - poor '' . recent high - resolution analyses of cs 22949 - 037 ( norris et al . 2002 ; depagne et al . 2002 ) and cs 29498 - 043 ( aoki et al . 2003 ) show similar patterns . in their analysis of cs22949 - 037 , norris et al . ( 2002 ) point out a good fit of [ n / fe ] = 2.6 to the rotating zero - metallicity vms models of fryer et al . ( 2001 ) , in which primary n is produced by c and o mixing into the h - rich shell . however , they note that this model overproduces n by a factor of 10 and that the star does not exhibit the exaggerated odd - even nuclei effect of pisn . in a subsequent analysis of [ o / fe ] in this same star , depagne et al . ( 2002 ) argue for a close match to an unpublished model of heger & woosley with @xmath66 @xmath1 . this model ( z35z ) matches the observed abundances of cs22949 - 037 when modified to include nlte effects for al and na and enhanced n production by c and o mixing into the h - rich shell . furthermore , this model produces zn in the correct ratio to fe , while no pisn models can do so . in a study of he0107 - 5240 , limongi et al . ( 2003 ) argue that this star can not be fitted with the yields of a single supernova because the fe - peak elements favor a deep mass cut ( such that relatively more material is ejected ) , while the light elements favor a shallow one ( to produce high [ c / fe ] ) . instead , they obtain a good fit for two sn sources : an earlier 15 @xmath1 sn with a deep mass cut between the ejected material and the remnant that releases the heavier elements , and a later 35 @xmath1 sn with such a shallow mass cut that the supernova nearly fails to eject any metals at all . although it requires two fine - tuned sn models to do so , their model matches he0107 - 5240 without recourse to vms . umeda & nomoto ( 2003 ) also find that the abundance pattern in this star fits nucleosynthesis from metal - free stars with @xmath67 @xmath1 . because @xmath4-process enrichment is common in emp stars , and because pisn experience no @xmath4-process ( hw02 ) , we can rule out the strong vms hypothesis ; clearly 8 40 @xmath1 stars must have existed in the generation that enriched the emp stars . furthermore , because the existing models with plausibly constructed imfs in the `` weak '' vms formulation must attribute all the early fe to vms , they also fail to match the observed abundance patterns of emp stars . thus , we find that the nucleosynthesis constraints do not provide concrete , positive evidence for vms that can not be explained by other means . although ambiguities remain in the interpretation of the cno - rich emps , they clearly do not _ require _ vms and/or pisn . unless additional input physics in pisn models change the odd - even and fe - peak characteristics of their yields , it appears the emp stars do not support the vms hypothesis . umeda & nomoto ( 2002 , 2003 , 2004 ) , and nomoto et al . ( 2003 ) have developed an alternative explanation for emp metal abundances . they argue that energetic supernovae or `` hypernovae '' ( hn ; @xmath68 , where @xmath69 erg ) that appear to be connected with some low-@xmath39 gamma - ray bursts could be relevant for early nucleosynthesis if they exist at low metallicity . empirical estimates based on light curves ( hamuy 2003 ) suggest that these events arise from supernovae with @xmath70 and progenitors of @xmath71 @xmath1 . since they have been found at low redshift , these objects are somewhat less speculative than vms and pisn , which currently have only a theoretical motivation . umeda & nomoto ( 2004 ) argue that the trends in the fe - peak elements in emp stars reflect nucleosynthesis by hn from zero - metallicity progenitors of @xmath72 @xmath1 . in their models , the joint effects of a deeper mass cut between remnant and ejecta , mixing in the ejected material , and the fallback of some of the metals onto the remnant enhance the products of complete si - burning ( zn , co ) relative to the products of incomplete si - burning ( mn , cr ) . a higher explosion energy also increases the dilution mass ( the mass of pristine gas into which the ejected metals are mixed ) . thus , second - generation stars born of their products will possess lower [ fe / h ] than stars enriched by lower - energy sn , with higher [ ( zn , co)/fe ] , and lower [ ( cr , mn ) ] as [ fe / h ] decreases . this detail in the models may need to change in light of the tighter scatter and smoother trends in the emp observations of cayrel et al . however , in general , hn appear to provide a superior match to the emp data . furthermore , umeda & nomoto ( 2004 ) argue that cno - rich emp stars can be explained with variations in explosion energy and the degrees of mixing and fallback in the ejected material . for comparison to the pisn yields , the lower panel of figure 1 shows the hn models compared to the three emp surveys . we display the hn yields for the fe - peak elements cr , mn , co , and zn resulting from explosions with @xmath73 @xmath1 and @xmath74 ( un04 ) . these elements are insensitive to the location of the mass cut in the supernova explosion , but they can vary with the electron mass fraction ( @xmath75 ) in the ejected material , the effects of which we include in the plotted range to show the improved fit ( e.g. , figure 4 of umeda & nomoto 2004 ) . the large explosion energies of hypernovae are needed to reproduce the observed zn . for all other elements , we plot the yields of umeda & nomoto ( 2002 ) , with @xmath76 @xmath1 , @xmath77 , and including mixing and fallback of the ejected material . while the hypernova phenomenon is gaining support from gamma - ray burst observations and theory , the hn explosion energy is still a free parameter in the umeda & nomoto ( 2004 ) models , and it is somewhat degenerate with stellar mass as a determining factor of the metal yields . there is also uncertainty associated with extrapolation from the low-@xmath39 , roughly solar - metallicity conditions to @xmath78 . although the apparent match of hn to the emp metal abundances is encouraging , they can not yet be conclusively identified as the products of the first stars . nevertheless , the possible existence of hn with @xmath79 @xmath1 in the first generation is an important part of constructing a first - stars imf without vms . we have reviewed the available evidence from stellar nucleosynthesis studies that bear on the mass function of the first stars . ( 2001 ) proposed that the correspondence between the hw02 pisn yields and the wq00 prompt inventory implied that vms were the sources of fe , si , and ca at [ fe / h ] @xmath80 . they used this fe abundance to estimate the global ionizing photon budget available to reionize the universe . these conclusions are not supported by the broader sets of elemental abundances . in contrast , we draw the following conclusions from our examination of the nucleosynthetic evidence : * the common presence of @xmath4-process elements in halo emps ( mcwilliam et al . 1995 ; burris et al . 2000 ) suggests progenitors of 8 40 @xmath1 , where the @xmath4-process elements are thought to form ( truran et al . this requirement is incompatible with the strong vms hypothesis and emphasizes the importance of the lower mass limit of the imf . * the relative abundances of fe - peak elements in emps ( especially zn ; cayrel et al . 2003 ) imply type ii supernovae from massive stars , or energetic hypernovae with deep mass cuts and mixing in the ejected material ( umeda & nomoto 2004 ) . these abundances are not well - fitted by the yields of pair - instability supernovae from vms . this is not a refutation of the weak vms hypothesis , because it does not constitute evidence that vms were not present at some level . it does imply that vms are not responsible for the bulk of the fe at [ fe / h ] @xmath81 ( as suggested by wq00 and oh et al . 2001 ) . much depends on the true nature of the galactic halo emp stars . if they do not , in fact , trace the enrichment of the first stellar generation , then we will be left with few , if any , such tracers and none that indicate vms . for example , damped ly@xmath82 systems ( dlas ) in qso spectra are often considered to be candidates for studies of primordial nucleosynthesis , because they have low metallicity and , at high @xmath39 , they may reflect early chemical enrichment . however , the existing literature on dlas ( prochaska & wolfe 2002 and references therein ) shows that they generally have [ fe / h ] @xmath83 , while the galactic halo emps range down to @xmath84 . thus , they may not reflect the same primordial metal enrichment thought to be seen in the emps with [ fe / h ] @xmath85 . furthermore , the selective depletion of fe and other refractory elements onto dust grains in dlas can change the fe - peak abundance ratios in the gas and thus complicate comparisons to theoretical yields and other observations . finally , direct ratios between zn , cr , and mn in dlas match the emp stars that also have [ fe / h]@xmath86 , above the metallicity where the qualitatively distinct behavior of emps appears . while galactic stars and dlas in this higher metallicity regime may derive their metals from similar sources , and may in fact have unusual features , dlas by themselves do not indicate that they have been enriched by pisn . the presence of vms is also not required by either the observed metal ratios in qso broad emission line regions ( venkatesan , schneider & ferrara 2004 ) or by [ si / c ] ratios in high - redshift lyman limit systems ( levshakov et al . 2003 ) . the recent studies ( see 2 ) of reionization by vms have expanded on the reionization thesis offered by oh et al . ( 2001 ) , but they have relied for the most part on that initial study to ground their use of the vms hypothesis in terms of the nucleosynthetic evidence . although it has been evaluated by observers in their presentation of the data , the independent evidence for the vms hypothesis has not been widely examined in the theoretical literature . previous studies have in many cases met the _ wmap _ and gunn - peterson constraints on the h i reionization history of the igm , but they have done so by assuming extreme astrophysical parameters or by assuming the vms hypothesis , which we find to have no firm external justification . we return to the issue of reionization by the first stars in 5 . with the nucleosynthetic results in mind , we can now construct a first - stars imf that is consistent with the observations . to the observational constraints discussed in 3.4 we add the following theoretical motivations for constructing an imf that is deficient in low - mass stars but not skewed to the vms mass range : * the general astrophysical arguments ( cooling , etc . ) used to motivate a `` top - heavy '' imf still apply . apparently forming massive stars at @xmath56 is a fundamentally different process than it is today . while tan & mckee ( 2003 ) point out that formation feedback may limit the mass of metal - free stars to @xmath8 30 @xmath1 , they also suggest that there is no known mechanism for halting the accretion below this range ( @xmath87 @xmath1 ) or fragmenting into smaller clumps . on general grounds , this justifies increasing the lower limit of the imf . * as shown in figure 2 , the ionizing efficiency of metal - free stars increases rapidly from 1 50 @xmath1 , with a plateau at @xmath88 @xmath1 . thus , we can construct an imf that achieves an ionizing efficiency similar to even a `` strong '' vms model without recourse to stars with @xmath19 @xmath1 . we propose a first - stars imf that embodies the observational constraints ( 1 , 2 ) and the theoretical arguments ( 3 , 4 ) . it is consistent with the available nucleosynthetic evidence , and it maximizes the ionizing photon efficiency to meet existing constraints on the reionization without violating those on igm metal enrichment below @xmath89 ( venkatesan & truran 2003 ) . in table 1 we present six new imf cases that meet these criteria . there are four cases with salpeter slope and varying upper and lower mass limits ( b d , g ) , and two cases with varying slope ( e f ) to capture the effects of proportionally more massive star formation in addition to variable mass limits . we also include a pure - vms imf ( a ) for comparison . although it is tempting to change the functional form of the imf from a power law to a more complicated function that favors high - mass stars , we avoid this for two reasons . first , the formation models have not given us specific reason to do so ; they suggest that the mass limits could change but have not yet determined the shape of the distribution . second , a more general function departing from a power law would introduce another parameter , and the imf slope is already poorly constrained . even if one knows the imf slope or mass limits , but not both , the inherent degeneracies in the problem may permanently preclude well - justified constraints on the other . a major part of the imf ( @xmath90 @xmath1 ) leaves little or no nucleosynthetic trace , and so the relative contributions of these stars to the ionizing radiation can not be easily constrained . thus , there remain significant uncertainties and ambiguities in constraining the first - stars imf . figure [ phot ] shows why our proposed imf cases can achieve reionization results similar to vms . here we show the lifetime - integrated ionizing photon production per baryon ( `` ionizing efficiency '' ) in metal - free stars of 5 500 @xmath1 . up to @xmath91 @xmath1 , the models are from tumlinson , shull , & venkatesan ( 2003 ) ; for @xmath12 @xmath1 they are from unpublished models based on the same evolution and stellar atmosphere code , checked for consistency against the results of schaerer ( 2002 ) . the notable features of this plot are the sharp rise to a peak at @xmath92 @xmath1 , followed by a slow decline at higher mass . this effect is a consequence of the stellar lifetimes decreasing more rapidly than the main - sequence ionizing photon production increases . this curve shows that an imf that excludes low mass stars and their poor ionizing efficiency can closely approximate the total ionizing photon production of an imf composed purely of vms . recently , venkatesan & truran ( 2003 ) showed that vms may not be responsible for both reionization and igm metal enrichment , since they produce only about 0.35 ionizing photons per baryon before they cease forming due to metal pollution ( reionization at @xmath93 is thought to require @xmath94 ionizing photons per baryon in the igm , and more are required at higher redshifts ) . this contradiction would be resolved if the first stars were all vms with @xmath95 @xmath1 and reionization was accomplished by stars that emit ionizing radiation but release no metals . however , this way is probably closed because no mechanism has been proposed for forming 300 @xmath1 stars without also forming 200 @xmath1 stars in roughly equal or greater proportion . furthermore , a primordial imf composed exclusively of @xmath95 @xmath1stars would never lead to the observed present - day star formation , as metals would not be released to trigger the transition from the first stars to a local imf ( the enrichment paradox ; schneider et al . . the only way out of this paradox is to form stars below @xmath96 @xmath1 , where metals would be released by pisn and influence later star formation . although this argument could be modified by future results from formation models , this probably means that excluding 140 260 @xmath1 also rules out more massive objects . a key parameter in modeling the radiative and chemical input from the first stars is the duration of the metal - free " phase of star formation . given the complications of an unknown imf , interstellar mixing , and halo merging history , this question is difficult to pose quantitatively . here we make simplifying assumptions to estimate the enrichment time of small halos at @xmath97 to estimate the time during which metal - free gas is available to form stars . as noted previously ( tumlinson , shull , & venkatesan 2003 ; vts03 ) , the duration of the metal - free phase determines , with the imf , the total output of ionizing photons from the first massive stars . as they die , these stars eject heavy elements from supernovae and enrich the surrounding gas within their own halo . these metals may also be transported across intergalactic space to enrich neighboring halos . although these enrichment and transport processes are intrinsically complicated , we can make a few relevant calculations that constrain the duration of this metal - free epoch . our calculations are similar in spirit to those of scannapieco , schneider , & ferrara ( 2003 ) , but we make different assumptions . first , we compute the _ global _ rate of metal enrichment per unit volume , based on an average star formation rate density ( sfr ) , hereafter quoted in standard units of comoving @xmath98 . by finding high - redshift galaxies through b , v , and @xmath99-band dropouts , steidel et al . ( 1999 ) and giavalisco et al . ( 2004 ) estimated a sfr @xmath100 between @xmath101 . from the required rate of ionizing emission to reionize the igm by @xmath102 , miralda - escud ( 2003 ) showed that the ionizing ( comoving ) emissivity can not decline by more than a factor of 1.5 from @xmath103 up to @xmath104 . recent numerical simulations of galaxy formation ( gnedin 2000 ; ricotti , gnedin , & shull 2002a , b , hereafter denoted rgs ) allow us to extrapolate to the earliest epochs of galaxy formation ( @xmath105 ) when star formation begins through h i and h@xmath22 cooling of protogalactic halos , with virial temperatures @xmath106 k to @xmath107 k. rgs found that the average sfr increases monotonically , from @xmath108 to @xmath109 between @xmath110 and 10 , according to the approximate formula : @xmath111 these simulations agree fairly well with the extended press - schechter estimates of scannapieco et al . ( 2003 ) . to estimate the corresponding metal production , we adopt a fiducial sfr = @xmath112 , with standard metal yield , @xmath113 for metal - free stars of @xmath38 @xmath1 . after a time @xmath114 , the average igm metallicity is : @xmath115 the elapsed time between @xmath110 and @xmath116 is @xmath117 yr for @xmath118 km s@xmath26 , @xmath119 , and @xmath120 . the above estimate shows that early star formation can raise the _ average _ igm metallicity to values near the suggested critical threshold , @xmath121 at which metal cooling may produce a transition to a different mode of star formation involving fragmentation into low - mass clumps ( omukai 2000 ; bromm & loeb 2003 ; schneider et al . 2002 ) . complicating this estimate is the likelihood that primordial metal enrichment proceeded inhomogeneously ( scannapieco et al . 2002 , 2003 ) . here , we consider a halo with total mass @xmath122 , in which @xmath123 @xmath1 of baryons are distributed in @xmath124 clumps spread over a halo of virial radius @xmath125 and virial temperature @xmath126 : @xmath127^{1/3 } = ( 165~{\rm pc } ) m_6^{1/3 } \left [ \frac { 20}{(1+z ) } \right ] \\ \nonumber \\ t_{\rm vir } & = & \left [ \frac { g m_h m_h } { 3 k_b r_{\rm vir } } \right ] = ( 1060~{\rm k } ) m_6^{2/3 } \left [ \frac { ( 1+z)}{20 } \right ] \ ; .\end{aligned}\ ] ] we assume cosmological densities , @xmath128 , with @xmath129 and @xmath130 ( for matter and baryons , respectively ; spergel et al . 2003 ) and a virialized density ratio @xmath131 . we set halo radius @xmath132 and divide the matter into @xmath124 clumps , with contrast ratio @xmath133 , where @xmath134 . we next assume that 5% of the halo s baryons undergo a burst of star formation at the center of the halo . as these stars eject their metals , a small fraction , @xmath135 , of these metals are captured by the nearest - neighbor clumps , while most are blown out into the igm . the capture fraction may be estimated from the solid angle of the neighboring clump , of radius @xmath4 , as seen from the halo center , @xmath136 . the mean distance between clumps within the halo is @xmath137 . we also have @xmath138 . after some algebra , we find @xmath139 ( \rho_h / \rho_{\rm cl})^{2/3 } \nonumber \\ & \approx & ( 0.096)(\rho_h/\rho_{\rm cl})^{2/3 } \ ; .\end{aligned}\ ] ] we arrive at the simple result that the fraction of metals captured by the nearest - neighbor clump depends only on the density contrast fraction between the clump and halo . for example , for the @xmath140 halo , with total baryonic mass @xmath141 , we assume @xmath142 clumps , each with @xmath143 in baryons and a clump - interclump contrast ratio , @xmath144 . if 5% of the halo s baryons ( @xmath145 ) undergo a burst of star formation at the halo center , they will produce approximately 83 sne and @xmath146 of heavy elements ( @xmath147 ) . the nearest - neighbor clump will capture a fraction @xmath148 , for an enriched " metallicity of @xmath149 or @xmath150 . thus , the sub - clumps within the halo will be enriched to @xmath151 solar abundances on a timescale of a few myr , the crossing time of the sne ejecta across the halo . a longer , and perhaps more relevant , timescale is that required to transport the heavy elements between halos , across the igm . from recent simulations ( gnedin 2000 ; rgs ) , the co - moving density of @xmath140 star - forming halos at @xmath152 is approximately @xmath153 mpc@xmath154 , corresponding to a mean ( physical ) distance between halos of @xmath155 $ ] . thus , the ejected heavy elements , if they are transported at mean velocity @xmath156 , would not reach neighboring halos for over @xmath157 yrs . these neighboring halos may recede with the local hubble flow , @xmath158 at @xmath110 . the hubble expansion velocity between these halos could be as large as @xmath159^{1/2 } \ ; .\end{aligned}\ ] ] alternatively , this region of space could have turned around from the expansion , depending on the extent and spatial correlations of the mass perturbations . transporting the ejected metals across 30 kpc at @xmath160 is not easy , given the amount of intervening intergalactic matter . a sphere of 1 kpc ( physical distance ) surrounding a halo at @xmath110 contains @xmath161 of baryonic mass , at the mean hydrogen number density , @xmath162 ^ 3 \ ; .\ ] ] any heavy elements ejected in sne explosions or outflows are accompanied by significant energy deposition . they rapidly sweep up many times their own mass , form cosmological blast waves , and eventually develop radiative shells ( ostriker & cowie 1981 ; shull & silk 1981 ) which slow down as they expand . consider the 5% burst of star formation from the previous example . the resulting 83 sne would deposit an energy @xmath163 into the igm , if we assume an energy of @xmath164 erg per sn or hn ( umeda & nomoto 2002 ) . an estimate of the blast wave extent comes from the non - cosmological sedov - taylor formula , @xmath165^{1/5 } \nonumber \\ & \approx & ( 7.35~{\rm kpc } ) \left [ e_{54 } t_8 ^ 2 /\delta \right]^{1/5 } \left [ \frac { 1+z}{20 } \right]^{-3/5 } \ ; .\end{aligned}\ ] ] here , @xmath166 is the igm mass density , @xmath167 is the overdensity relative to the mean , and @xmath168 is the time after the injection of sn energy . in reality , one must take cosmological expansion and igm inhomogeneity into account , as well as the gradual injection of energy into the superbubble " produced by multiple sne over a total time of @xmath169 yr . from the above approximate formulae , we see that the expanding blast wave could take over @xmath157 yr to contaminate neighboring halos . we conclude that the halo - to - halo enrichment timescale is likely to be of order @xmath157 yr . these simple calculations motivate our choices of @xmath170 and @xmath157 yr as the duration of the first stars epoch for primordial star formation . these timescales are in accord with those calculated from numerical simulations of halo self - enrichment by wada & venkatesan ( 2003 ) and of interhalo enrichment by bromm , yoshida , & hernquist ( 2003 ) . the effects of these two timescales on the reionization models are discussed in 5 . in fact , our proposed first - stars imf has another virtue along these lines . if most of the mass in the early imf is formed into stars in the 50 140 @xmath1 mass range , the unique effects of a zero - metallicity imf could persist longer than they would if the imf favored vms . to see why , consider a case where all mass goes into forming stars with either @xmath171 @xmath1 or @xmath95 @xmath1 . since these stars collapse directly to black holes without releasing metals , the metal - free mode of star formation would be self - perpetuating . if this extreme imf is modified to incorporate increasing but still relatively small numbers of sn ii or hn progenitors , then metal enrichment will gradually achieve the global value of * @xmath172 * thought to curtail the unique low - metallicity star formation mechanisms . in a rough sense , this timescale is inversely proportional to the fraction of matter formed into stars that explode and release metals relative to those that do not . thus , a significant advantage of the @xmath173 @xmath1 imf is that the duration of metal - free star formation at early epochs could be longer , owing to the increased stellar lifetimes in this imf relative to vms and to the lower total metal yields . we have derived a first - stars imf that is consistent with the available nucleosynthetic evidence , and we have estimated the time ( @xmath174 yr ) over which metal - free stars can form in small primordial halos . we must now evaluate our proposed imf in terms of reionization models to compare their effect on the igm to that of vms . we hold fixed all the model assumptions except the imf and the metal - enrichment timescale to determine the extent to which the imf matters to the outcome . we adopt the semianalytic stellar reionization model described in vts03 for a concordance @xmath3cdm cosmology . the growth of ionized regions is tracked by a press - schechter formalism in combination with numerical solutions for the growth of individual ionization fronts . for all cases here , the cosmological and astrophysical parameters are the same as in vts03 ; we do this in order to consistently compare the effects of a vms imf relative to those proposed here ( there is one exception , mentioned below ) . specifically , our models are described by the parameter set [ @xmath175 , @xmath176 , @xmath177 , @xmath178 , @xmath179 , @xmath180 , @xmath181 , @xmath182 , @xmath183 , @xmath184 = [ 0.9 , 1.0 , 0.7 , 0.04 , 0.7 , 0.3 , 30 , 0.05 , 0.05 , 0.025 ] , where @xmath175 and @xmath176 are the usual normalization and index of the scalar matter power spectrum , @xmath181 is the space - averaged clumping factor of h ii or he iii ( assumed to be the same here ) , @xmath177 is the hubble constant in units of 100 km s@xmath26 mpc@xmath26 , and @xmath178 , @xmath179 , and @xmath180 represent the cosmological density parameters of baryons , the cosmological constant , and matter respectively . the astrophysical parameters are @xmath182 , the fraction of baryons forming stars in each halo , and @xmath183 and @xmath185 , the escape fractions of h i and he ii ionizing photons from halos , respectively . we allow star formation in all halos of virial temperature @xmath186 10@xmath187 k ( rather than 10@xmath188 k as in vts03 ) , so as to have sufficient baryons in a @xmath189 dark matter halo at @xmath110 for our adopted star formation efficiencies to form the most massive star ( @xmath190 ) in our imf cases . metal - free stars are assumed to form starting from @xmath191 and last in each halo for either of two approximate timescales derived earlier for intrahalo and halo - halo enrichment , @xmath170 yr and @xmath157 yr . subsequent to this , the ionizing spectrum is switched to a representative example of pop ii stars with metallicity @xmath192 ( leitherer et al . reionization is defined as the overlap of individual ionized regions of h ii or he iii as appropriate , when their volume filling factors equal unity ( see vts03 ) . relative to the results of vts03 , reionization in our new models occurs earlier , owing to the lower halo mass threshold and the fixed metal - free star forming phase in each halo before switching to a pop ii ionizing spectrum . vts03 assumed @xmath193 k and shut off metal - free star formation at fixed cosmological ages . thus , in vts03 , the majority of small halos , in which most of the baryons reside until @xmath194 9 when they merge into larger systems , did not experience any metal - free star formation , even if they had not undergone prior star formation or experienced pollution from neighboring halos . our treatment here ensures that each halo experiences at least one episode of metal - free star formation prior to cosmological times of @xmath195 gyr . the assumption of @xmath196 k is somewhat more optimistic from the point of view of early reionization , and relies on the assumption that halos with @xmath197 k can effectively cool with h@xmath22 ( rgs ) . other models ( e.g. , haiman & holder 2003 ) have even more optimistically assumed star formation in halos down to @xmath198 k ; these are not massive enough to form the most massive star in our imf cases ( 1000 @xmath1 ) . the assumption that each new halo forms metal - free stars breaks down at @xmath199 , when the earliest small halos , presumably metal - enriched from their first star formation , begin to coalesce into larger objects , which are presumably never metal - free . our models reionize early enough to avoid this complication ( see below ) . in table 1 , we display the various imfs considered in this work . we include newly calculated cases that are motivated by the nucleosynthetic data ( 4.1 ) and cases that represent a top - heavy imf , either in the form of a higher stellar mass range or of a flatter imf slope , @xmath82 . also shown are the ionizing photon rates , @xmath200 , for h i and he ii from synthetic stellar clusters at a starburst age of 1 myr in units of photons s@xmath201 . we also show the h i and he ii reionization redshifts , @xmath31 , for each case and the total associated @xmath2 ( including the @xmath202% contribution from he ii ) , assuming that the post - recombination igm ionization fraction is about @xmath203 over @xmath36 20 1000 . the equivalent ionizing rates for pop ii are @xmath204 s@xmath26 @xmath1@xmath26 and @xmath205 s@xmath26 @xmath1@xmath26 after the transition from pop iii to pop ii assuming a salpeter imf with 1 - 100 @xmath1 , halos with @xmath206 k , and a pop iii to pop ii transition time of 10@xmath207 yr . ] . we see that an imf derived from the emp nucleosynthetic constraints and lacking in low - mass stars can generate sufficiently early reionization and values of @xmath208 . this is consistent with current data on reionization without resorting to a top - heavy imf ; high ionizing efficiency need not arise from vms only . we also note that the uncertainty on @xmath35 from _ wmap _ arises from a broad probability distribution , and that it is degenerate with @xmath209 . recently , @xmath18 has been determined for _ wmap _ plus the three - dimensional matter power spectrum from sdss ( tegmark et al . our models reproduce the _ wmap _ result at the @xmath210 confidence level and match the tegmark et al . ( 2003 ) value without resort to vms . in fact , the close agreement achieved by our imf suggests that the astrophysical parameters ( star formation efficiency , escape fraction ) are now the limiting factors in achieving good constraints on reionization . we also note that the models from vts03 produced @xmath211 with a more restrictive halo virial temperature cutoff ( @xmath13 k ) and a salpeter imf . the difference between this @xmath212 and model b shows that the more optimistic halo prescription of the present study contributes only a modest amount . an important caveat for the results here is that the igm , once reionized completely in h or he , stays ionized . for h i , if reionization occurs once in the model at some redshift , e.g. , @xmath213 , and if metal - free stars are allowed to keep forming subsequently for @xmath170@xmath157 yr in each new halo , the model igm does not recombine and stays ionized , even though recombinations are explicitly allowed . this result is probably valid for h i in the underdense voids , which represent most of the volume in the igm . in the case of h i in dense regions and he ii in general , this behavior may be suspect , owing to high recombination rates at typical igm densities at @xmath214 5 ( shull et al . however , even if he iii partially recombined in those of our cases in which he ii reionization occurs early , the contribution to @xmath2 would be impacted at less than the 10% level . therefore , our numbers for he ii should be considered only for the purposes of comparison between various imfs . this is especially true at @xmath215 where we have not included the ionizing radiation from qsos . the simple reionization history in our models also does not reproduce the known h i gunn - peterson opacity at @xmath93 or the observed igm temperatures at @xmath216 . these constraints would require additional input into the models , such as the contribution of quasars to the ionizing photon budget , or a more complicated reionization history than we explore here . models that attempt to match all available constraints exist in the literature ( see 2 ) . the additional complications and new parameters would , if they were included in our model , needlessly distract from the goal of directly comparing the different imf cases with the same minimal set of cosmological and astrophysical parameters . we also note that the @xmath217 gunn - peterson constraint can be met by a modest amount of recombination at @xmath218 , perhaps coincident with the expected pop iii to pop ii transition , with only a modest effect on @xmath212 . this is possible because even a trace neutral fraction is sufficient to provide @xmath219 with negligible impact on the ionized fraction that scatters photons from the cmb . because these external constraints can be met with minor adjustments to our model , and because they are incidental to the imf comparisons , we do not include them here . we close by drawing some important conclusions from the foregoing results and discussion . these are : 1 . the vms hypothesis should be distinguished into two versions : the `` strong '' version ( `` the first stars were exclusively vms '' ) and the `` weak '' version ( `` the first stars included vms and stars with @xmath72 @xmath1 '' ) . this distinction is necessary because the radiative and nucleosynthesis properties of metal - free stars vary widely across the mass function . we have demonstrated that these two versions have different testable predictions and can be independently falsified . the existing samples of metal abundances in emp galactic halo stars , which are commonly thought to trace the enrichment products of the first generation of stars , are better matched to the products of type ii sn and hypernovae like those seen at low redshift . specifically , the fe - peak and @xmath4-process abundances in emps are not well - matched by vms . 3 . because they are efficient sources of ionizing photons , vms have been proposed as the ionizing sources for the first stage of the igm reionization and the only means of generating large electron - scattering optical depths seen in cmb maps . by excluding the low - mass end of the imf , as suggested by models of primordial protostars , we have constructed imf cases that are just as efficient as vms at converting baryons to ionizing photons . thus constraining the lower mass limit of the imf is perhaps more important for reionization than a constraint on the upper limit . 4 . direct comparisons of our test imfs in semi - analytic reionization models show that without vms these models can reproduce @xmath35 in the observed range , and similar to vms models with the same astrophysical parameters . we can simultaneously reproduce the nucleosynthesis and reionization constraints with a single imf . we confirm earlier studies ( vts03 , wyithe & loeb 2003 ) which found that the lower end of the _ wmap _ range of @xmath212 can be reproduced without recourse to a nonstandard imf . 5 . with simple metal - transport calculations , we derive a timescale for metal - free star formation of @xmath220 yr . by effectively decoupling the earliest stages of ionization and metal - enrichment , our proposed imf can allow the period of metal - free star formation to persist longer than an imf composed of vms , and perhaps long enough to fully reionize the igm before self - enrichment of halos permanently curtails metal - free star formation . we have shown that the nucleosynthesis constraints exclude the strong vms hypothesis and have difficulty accommodating the weak vms hypothesis . we have also shown that an imf without vms can reproduce @xmath221 if it includes stars with @xmath222 @xmath1and @xmath223 if it does not . rather than provide strong evidence for the vms hypothesis , the existing @xmath35 values instead argue against low - mass stars but provide little constraint on massive stars in the imf . in light of the metal - free star formation models , both the high- and low - mass ends of the stellar mass function are poorly constrained . it is only when the nucleosynthetic information is added that we can say anything about the imf . with these uncertainties in mind , we consider how future developments might bear on the first - stars imf . first , a tighter range of @xmath35 is necessary to constrain the ionizing efficiency of the first stars , and , indirectly , the imf . such an improvement might come with additional _ wmap _ data , including polarization correlations . however , the greatest need is for consistent results from the various combinations of the cmb and additional data ( i.e. , matter power spectrum , ly@xmath82 forest ) that can constrain @xmath2 . uncertainty of @xmath224 on @xmath35 would leave the astrophysical parameters in a reionization model as the limiting uncertainties ( venkatesan 2002 ) . it is important to point out that a low @xmath35 does not , by itself , disfavor the vms hypothesis . because their ionizing efficiency is lower than the peak ( see figure 2 ) , a strong - vms model could give a low @xmath35 with appropriately low star formation efficiency or photon escape fraction ( if reionization is late ) , or with a short time available for metal - free star formation . unless it settles on an extreme high value , @xmath35 is not likely by itself to provide decisive constraints on the imf . this latter point emphasizes the critical importance of robust constraints from nucleosynthesis as the remaining means of discriminating degenerate imf cases ( such as a strong vms model versus a weak vms model without @xmath225 @xmath1 , see table 1 ) . at the present time , emp stars appear to exclude vms . this could change because of developments in the observations ( new emp stars with lower [ fe / h ] ) or refinements in the theory ( reconciliation of pisn to the fe - peak elements would revitalize the weak vms hypothesis ) . a particular concern would be evidence that the sample of emp stars we presently take as the `` second generation '' is not , in fact , the product of the first stars . beyond their chemically primitive atmospheres and ages , there is no hard evidence for this assumption . a new sample of metal - poor stars with [ fe / h ] @xmath226 could change the outcome if they were better matched to the yields of pisn and/or clearly devoid of @xmath4-process elements . there is also room for significant progress on the theoretical front : additional physics such as asymmetry , mass loss , and rotation could be added to the hypernovae and pisn models , perhaps with a corresponding change in the logical basis of the vms hypothesis . the reionization results in table 1 and the ionizing efficiencies in figure 2 suggest another observational discriminant between our proposed first - stars imf and one composed purely of vms . although the h i ionizing efficiency of vms declines from the peak at @xmath227 @xmath1 , the ionizing efficiency for he ii continues to increase into the vms range . this feature of vms has two potential observational consequences . first , vms will be more efficient at producing the he ii 1640 recombination line , which has been suggested as a key direct signature of the first stars ( tumlinson , giroux , & shull 2001 ; schaerer 2002 ) . although the many degeneracies in the age and metallicity dependence of the @xmath2281640 line luminosity could give an ambiguous result , an unusually high value could be uniquely attributed to vms . the second possible signature is more speculative : the he ii gunn - peterson test at high redshift is , in principle , sensitive to the spectrum of the first ionizing sources . a vms imf may reionize he ii in the igm before @xmath102 , while a non - vms imf may or may not ( see table 1 ) . even if both imfs can produce early he ii reionization , they may still be discriminated by a gunn - peterson test . to see why , we recall the ratio @xmath229 defined by fardal , giroux , & shull ( 1998 ) to characterize the ionizing sources of the igm . harder qso - type spectra have @xmath230 , while ordinary stars produce @xmath231 . the increased he ii ionizing efficiency of metal - free stars in a normal imf can produce @xmath232 ( tumlinson & shull 2000 ) . by excluding inefficient low - mass stars , our proposed first - stars imf gives @xmath233 . a pure vms imf could result in @xmath234 . recent work at @xmath235 with the _ far ultraviolet spectroscopic explorer _ ( shull et al . 2004 ; zheng et al . 2004 ) has shown that @xmath236 and @xmath237 can be discriminated . the feasibility of this test would depend largely on the chance combination of a qso with @xmath238 and little intervening absorption of the near - ultraviolet continuum by the common ly@xmath82 forest or lyman limit systems at @xmath239 . although such a coincidence is unlikely , qsos discovered at @xmath238 should be observed in the near uv to qualify them for this test . obtaining adequate s / n to accurately measure the predicted low values of the he ii opacity would likely require a large collecting area in the space ultraviolet , perhaps from a successor to the _ hubble space telescope_. if the practical problems can be overcome , the direct detection of he ii @xmath2281640 and the high-@xmath39 he ii gunn - peterson test provide some means of testing the first - stars imf . finally , we add that , as a theoretical point of interest , the lower mass limit of the imf can be just as important as the better - studied upper limit , because it can determine the total ionizing efficiency of the population . the lowest mass achievable by metal - free stars may depend on the details of primordial cooling , rotation , fragmentation , feedback , and magnetic fields in unknown ways . attention has been focused on modeling the formation of very massive first stars ; our results indicate that potential mechanisms for forming low - mass metal - free stars should be pursued . this is especially true if @xmath240 settles to a high value that requires high ionizing efficiency and the present nucleosynthesis constraints on the strong vms hypothesis are confirmed . our proposed imf can solve some key puzzles in the study of the first stars and reionization . however , there are still significant open questions about the formation of metal - free stars ( especially the feedback mechanisms that influence their final masses ) , and the unknown time over which metal - free star formation persists . furthermore , there may be no high - redshift direct tracers that can independently constrain the proposed imf , especially since its most salient feature is the absence of stars with 1 10 @xmath1 that are not directly observable at high - redshift . until the launch of the _ james webb space telescope _ , which may be able to directly detect the first stars , future work should focus on more sophisticated treatments of the indirect tracers , particularly the emp stars , as sensitive indicators of the first stellar generation . nonetheless , we have shown that vms are not needed for reionization if the imf is suitably constructed , and that such an imf is better justified on the grounds of the nucleosynthetic data than either version of the vms hypothesis . in the future , better understanding of primordial star formation , more refined models of pisn , or a new regime of extremely metal - poor stars with more vms - like metal abundances may change the picture . for now , we conclude only that our proposed imf is better matched to the available constraints . we acknowledge helpful discussions with jim truran and constructive comments from sergei levshakov , avi loeb , and an anonymous referee . a. v. is supported by an nsf astronomy and astrophysics postdoctoral fellowship under award ast-0201670 . j. m. s. acknowledges support at the colorado astrophysical theory program from nasa ltsa grant nag5 - 7262 and nsf grant ast02 - 06042 . abel , t. , bryan , g. , norman , m. l. 2000 , science , 293 , 93 aoki , w. , et al . 2003 , , 576 , l141 becker , r. , et al . 2001 , aj , 122 , 2850 bromm , v. , kudritzki , r .- p . , & loeb , a. 2001 , , 552 , 464 bromm , v. , yoshida , n. , & hernquist , l. 2003 , , 596 , 137 bromm , v. , & loeb , a. 2003 , nature , 425 , 812 bromm , v. , & loeb , a. 2004 , new astronomy , 9 , 353 burris , d. , et al . 2000 , , 544 , 203 cayrel , r. , et al . 2003 , a&a , 416 , 1117 carretta , e. , gratton , r. , cohen , j.g . , beers , t. c. , & christlieb , n. 2002 , , 124 , 481 christlieb , n. , et al . 2002 , nature , 419 , 904 ciardi , b. , ferrara , a. , & white , s. d. m. 2003 , mnras , 344 , 7 cen , r. 2003a , , 591 , 12 cen , r. 2003b , , 591 , l5 depagne , e. , et al . 2002 , a&a , 390 , 187 djorgovski , s. g. , castro , s. m. , stern , d. , & mahabal , a. 2001 , , 560 , l5 fardal , m. , giroux , m. l. , & shull , j. m. 1998 , aj , 115 , 2206 fields , b. d. , truran , j. w. , & cowan , j. j. 2002 , 575 , 845 fryer , c. l. , woosley , s. e. , & heger , a. 2001 , 550 , 372 fukugita , m. , & kawasaki , m. 2003 , mnras , 343 , 25 giavalisco , m. , et al . 2004 , , 600 , l103 gnedin , n. y. , & ostriker , j. p. 1997 , 486 , 581 gnedin , n. y. 2000 , , 535 , 530 haiman , z. , & holder , g. p. 2003 , 595 , 1 hamuy , m. 2003 , , 582 , 905 heger , a. , & woosley , s. e. 2002 , , 567 , 532 ( hw02 ) hui , l. , & haiman , z. 2003 , apj , 596 , 9 kaplinghat , m. , et al . 2003 , apj , 583 , 24 kogut , a. , et al . 2003 , , 148 , 161 leitherer et al . , 1999 , , 123 , 3 levshakov , s. , et al . 2003 , a&a , 397 , 851 limongi , m. , chieffi , a. , & bonifacio , p. 2003 , , 594 , l123 marigo , p. , girardi , l. , chiosi , c. , & wood , p. r. 2001 , a&a , 371 , 152 mcwilliam , a. , preston , g. w. , sneden , c. , & searle , l. 1995 , , 109 , 2757 mcwilliam , a. 1997 , ara&a , 35 , 503 miralda - escud , j. 2003 , , 597 , 66 nomoto , k. , et al . 2003 , in carnegie observatories astrophysics series , vol . 4 : origin and evolution of the elements , ed . a. mcwilliam and m. rauch , 1 norris , j. e. , et al . 2002 , , 569 , l107 oh , s. p. , nollett , k. m. , madau , p. , & wasserburg , g. j. 2001 , , 562 , l1 omukai , k. 2000 , , 534 , 809 omukai , k. , & palla , f. 2002 , apj , 589 , 677 ostriker , j. p. , & cowie , l. l. 1981 , , 243 , l127 prochaska , j. x. , & wolfe , a. m. 2002 , 566 , 68 qian , y .- z . , & wasserburg , g. j. 2002 , , 567 , 515 ( qw02 ) ricotti , m. , gnedin , n. y. , & shull , j. m. 2002a , , 575 , 33 ( rgs ) ricotti , m. , gnedin , n. y. , & shull , j. m. 2002b , , 575 , 49 scannapieco , e. , ferrara , a. , & madau , p. 2002 , , 574 , 590 scannapieco , e. , schneider , r. , & ferrara , a. 2003 , , 589 , 35 schaerer , d. 2002 , a&a , 382 , 28 schneider , r. , ferrara , a. , natarajan , p. , & omukai , k. 2002 , , 571 , 30 shull , j. m. , & silk , j. 1981 , , 249 , 26 shull , j. m. , tumlinson , j. , giroux , m. l. , kriss , g. a. , & reimers , d. 2004 , , 600 , 570 somerville , r. s. , bullock , j. s. , & livio , m. 2003 , apj , 593 , 616 spergel , d. , et al . 2003 , , 148 , 175 steidel , c. c. , adelberger , k. l. , giavalisco , m. , dickinson , m. , & pettini , m. 1999 , , 519 , 1 tan , j. , & mckee , c. m. 2002 , in `` the emergence of cosmic structure '' , proceedings of the aip , 666 , 93 tan , j. , & mckee , c. m. 2004 , astroph/0307414 tegmark , m. 2004 , apj , in press , astro - ph/0310723 truran , j. w. , cowan , j. j. , pilachowski , c. a. , & sneden , c. 2002 , , 114 , 1293 tumlinson , j. , & shull , j. m .. 2000 , , 528 , l65 tumlinson , j. , giroux , m. l. , & shull , j. m. 2001 , , 550 , l1 tumlinson , j. , shull , j. m. , & venkatesan , a. 2003 , , 584 , 608 ( tsv03 ) umeda , h. , & nomoto , k. 2002 , , 565 , 385 umeda , h. , & nomoto , k. 2003 , nature , 422 , 871 umeda , h. , & nomoto , k. 2004 , astro - ph/0308029 ( un04 ) venkatesan , a. 2002 , , 572 , 15 venkatesan , a. , tumlinson , j. , & shull , j. m. 2003 , , 584 , 621 ( vts03 ) venkatesan , a. , & truran , j. w. 2003 , apj , 594 , l1 venkatesan , a. , schneider , r. , & ferrara , a. 2004 , mnras , 349 , l43 wada , k. , & venkatesan , a. 2003 , , 591 , 38 wasserburg , g. j. , & qian , y .- z . 2000 , , 529 , l21 ( wq00 ) woosley , s. e. , et al . 1994 , , 433 , 229 wyithe , s. , & loeb , a. 2003 , , 586 , 693 zheng , w. , et al . 2004 , , 605 , 671	we critique the hypothesis that the first stars were very massive stars ( vms ; @xmath0 @xmath1 ) . we review the two major lines of evidence for the existence of vms : ( 1 ) that the relative metal abundances of extremely metal - poor galactic halo stars show evidence of vms enrichment , and ( 2 ) that the high electron - scattering optical depth ( @xmath2 ) to the cmb found by _ wmap _ requires vms for reionization in a concordance @xmath3cdm cosmology . the yield patterns of vms exploding as pair - instability supernovae are incompatible with the fe - peak and @xmath4-process abundances in halo stars . models including type ii supernovae and/or `` hypernovae '' from zero - metallicity progenitors with @xmath5 @xmath1 can better explain the observed trends . we use the nucleosynthesis results and stellar evolution models to construct an initial mass function ( imf ) for reionization . with a simple metal - transport model , we estimate that halo enrichment curtails metal - free star formation after @xmath6 yr at @xmath7 . because the lifetime - integrated ionizing photon efficiency of metal - free stars peaks at @xmath8 120 @xmath1 and declines at higher mass , an imf with an approximate lower bound at @xmath9 @xmath1 and no vms can maximize the ionizing photon budget and still be consistent with the nucleosynthetic evidence . an imf devoid of low - mass stars is justified independently by models of the formation of primordial stars . using a semi - analytic model for h i and he ii reionization , we find that such an imf can reproduce @xmath10 , consistent with the range from _ wmap _ , without extreme astrophysical assumptions , provided that metal - free star formation persists @xmath11 yr after star formation begins . because stars in the mass range 50 140 @xmath1 are the most efficient sources of ionizing photons , but are expected to collapse to black holes without releasing metals , this imf effectively decouples early metal enrichment and early ionization . such an imf may allow the unique properties of the zero - metallicity imf to persist longer than they would in the pure vms case , and to contribute significantly to the global ionizing photon budget before halo self - enrichment and/or inter - halo metal transport truncates metal - free star formation . we conclude , on the basis of these results , that vms are not necessary to meet the existing constraints commonly taken to motivate them .
support from the dfg under the program sfb 689 is acknowledged . 25ifxundefined [ 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 _ _ , vol . ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , ) @noop * * , ( )	nonequilibrium physics of rare events , or fluctuations , is an unique fingerprint of a given system . by virtue of its very nature , it differs from the nonequilibrium physics enclosed in the system s mean quantities and provides an experimental alternative to conclusively uncover various microscopic structures . here , using a keldysh field theory with fermionic topological constraints , we demonstrate that in noninteracting systems , whose dynamics is driven by majorana states , the effective charge @xmath0 , characterizing the electric current fluctuations , is fractional . this is in contrast to noninteracting dirac systems with the trivial electronic charge , @xmath1 . quite the opposite , in the majorana state we predict two different fractional effective charges at low and high energies , @xmath2 and @xmath3 , which can be accessed at low and high bias voltages , respectively . we show that while the low energy effective charge @xmath2 is sensitive to thermal fluctuations of the current , the high energy effective charge @xmath3 is robust against thermal noise and represents an unique fluctuation signature of majorana fermions in experiments on strongly nonequilibrium systems even at high temperatures . the physics of fluctuation phenomena , or noise , dating back to brownian @xcite motion has received a systematic scientific framework after the einstein s @xcite and smoluchowski s @xcite conceptual theoretical breakthrough proven experimentally by svedberg @xcite and perrin @xcite . spontaneous or excited by external perturbations fluctuations represent an extremely insightful experimental tool known as the fluctuation spectroscopy . this technique is highly sensitive and , as a result , it scans the microscopic structure of a system in question in much more detail than it could be addressed in terms of the expectation values of the system s observables . in equilibrium , nevertheless , kinetics of a given system makes a clever link between random deviations of its physical quantities from expectation values and the expectation values themselves . this link dating back to the nyquist s @xcite and callen s and welton s @xcite fundamental discovery is known as the fluctuation - dissipation theorem @xcite . in nonequilibrium , however , the fluctuation - dissipation theorem breaks and , therefore , for a given system its fluctuation physics becomes independent of the physics which governs behavior of the system s mean quantities . here the piece of information obtained from fluctuation spectroscopy has often no counterparts in equilibrium fluctuations or even in principle can not be obtained in equilibrium as it happens , _ e.g. _ , in the case of the resistance fluctuations in metals known as the modulation noise . importantly , even when a system is only slightly , in a certain sense , perturbed out of its equilibrium state the nonequilibrium noise might be comparable to or , in fact , become stronger than the equilibrium noise . thus , nonequilibrium noise experiments are a reliable and comprehensive method to conclusively reveal the microscopic structure of a system when measurements of its mean quantities are physically inconclusive . this is what currently happens in another fascinating field of modern condensed matter physics dealing with materialization of a particle cloning its own antiparticle . namely , via unpairing majorana @xcite fermions , composing a single dirac fermion , by means of different implementations @xcite of the kitaev s @xcite model it is hoped to be able to detect a single majorana state @xcite . an experimental proof of the theoretical predictions , however , mainly focuses on measurements of mean quantities such as the differential conductance which , as a function of a bias voltage , should exhibit a peak with the maximum equal to one - half of the unitary limit known for dirac fermions . this way of experimental @xcite detecting unpaired majorana fermions is argued to be inconclusive . indeed , such a peak structure of the differential conductance might result , _ e.g. _ , from the kondo effect @xcite taking place in a mesoscopic system asymmetrically coupled to contacts by means of which the bias voltage is applied to this system . this problem is inherent to many transport majorana s experiments dealing with mean quantities . nevertheless , it is possible to get a conclusive signature of majorana fermions from the mean value description if , instead of transport , one resorts to thermodynamic signatures of majorana fermions @xcite . the freedom to involve nonequilibrium noise in the fluctuation spectroscopy of majorana fermions triggers transport experiments on majorana physics to a new azimuth and makes it more interesting . this is because , as mentioned above , fluctuations are usually conclusive on the microscopic structure of a system and at the same time these are transport experiments which are in general simpler than thermodynamic ones . for example , experiments @xcite on nonequilibrium fluctuations in quantum dots have at present reached a remarkable accuracy and significantly extend our understanding of the kondo effect . in particular , these experiments enable one to measure the noise of the electric current and provide an essential characteristic of nonequilibrium fluctuations known as the effective charge , @xmath0 , which is an unique fluctuation fingerprint of a given system . in the present work we explore strongly nonequilibrium fluctuations of the electric current flowing through a noninteracting quantum dot coupled to a topological superconductor supporting at its ends two majorana bound states implemented via the kitaev s chain model . it is well known that in the absence of majorana fermions the effective charge for a noninteracting quantum dot is trivial and identical to the electronic charge , @xmath1 . here we demonstrate that in the presence of majorana fermions 1 ) the effective charge fractionalizes to 2 ) @xmath2 at low energies , to 3 ) @xmath3 at high energies and show that 4 ) even when the low energy effective charge @xmath2 is washed out by , _ e.g. _ , thermal noise , the high energy effective charge @xmath3 is robust and persists up to very high temperatures providing a simple and reliable experimental platform for an unique signature of majorana fermions out of strongly nonequilibrium fluctuations . let us consider a setup similar to the one of ref . @xcite . it represents a noninteracting quantum dot coupled via tunneling interaction to two ( @xmath4 and @xmath5 ) noninteracting contacts . in contrast to the equilibrium setup of ref . @xcite , here the contacts may be used to apply a bias voltage @xmath6 to the quantum dot , @xmath7 , @xmath8 . the quantum dot has a one single particle level which is spin nondegenerate as may be experimentally implemented , _ e.g. _ , by means of the zeeman splitting which also filters out the kondo effect @xcite . note also that below we explore strongly nonequilibrium states which are accessed at very high bias voltage @xmath6 so that the kondo state totally ruined @xcite in any case and , therefore , does not lead to any experimental ambiguity . finally , similar to ref . @xcite , the quantum dot interacts via another tunneling mechanism with a grounded topological superconductor supporting at its ends two majorana bound states . to give the problem a concrete and mathematically convenient treatment we formulate it in terms of the quantum many - particle keldysh lagrangian @xmath9 which constitutes the basis for the keldysh action , @xmath10 . here the momenta @xmath11 and coordinates @xmath12 are the fermionic coherent states of the system and there conjugate partners , respectively . the lagrangian formulation is fully equivalent to the quantum hamiltonian @xmath13 formulation but has a certain technical advantage in calculating strongly nonequilibrium fluctuations of the electric current via the keldysh field integral @xcite which we employ below to obtain the current - current correlation function . for the quantum dot ( @xmath14 , @xmath15 ) , contacts ( @xmath16 , @xmath17 ) and tunneling between them we , respectively , have : @xmath18 where @xmath19 is the quantum dot energy level and the real time @xmath20 runs along the keldysh closed contour @xmath21 , @xmath22 , @xmath23 where below we will assume identical quantum numbers @xmath24 in both @xmath4 and @xmath5 contacts as well as large contacts so that their spectrum @xmath25 is continuous and their density of states @xmath26 is constant in the vicinity of the fermi energy , @xmath27 where below we use for the dirac tunneling the standard assumption that the tunneling matrix elements weakly depend on the contacts quantum numbers , @xmath28 , which allows one to characterize the tunneling coupling by the energy scale @xmath29 or @xmath30 . we will also assume that @xmath31 . finally , the topological ( @xmath32 , @xmath33 ) part of the hamiltonian is given as the sum of the hamiltonians of the topological superconductor and its tunneling interaction with the quantum dot : @xmath34 where @xmath35 is the energy originating from the overlap of the majorana bound states @xmath36 and @xmath37 ( in particular , @xmath38 if there is no overlap as in sufficiently long kitaev s chains ) , @xmath39 where the majorana tunneling entangles the dirac fermions of the quantum dot with only one majorana state , @xmath36 , and is characterized by the energy scale @xmath40 . due to the fundamental property of the majorana fields , @xmath41 , @xmath42 , and the canonic fermionic anticommutation relations the field integral for the keldysh partition function , @xmath43 is a functional integral with the constraints @xmath41 , @xmath44 which might be viewed as fermionic topological constraints . in contrast , if @xmath45 , @xmath46 , then any fermionic topological constraint is absent and , as usual , @xmath47 , @xmath48 . thus , because of the presence of the topological superconductor , here one deals with a fermionic topological field integral and the whole problem represents a sort of a topological field theory . nevertheless , since the hamiltonian @xmath49 is quadratic in all @xmath11 , @xmath12 , this topological field integral may be easily solved exactly as in many standard textbooks @xcite . using an imaginary time field theory , it has been rigorously proven @xcite by entropic reasoning that a macroscopic state of the above setup is majorana dominated at low temperatures . therefore , it must exhibit various fractionalizations @xcite of its observables . here , in particular , we are interested in fractionalizations of the electric current fluctuations . to this end we introduce suitable sources into the keldysh partition function ( [ z_k ] ) turning it into the keldysh generating functional : @xmath50 where @xmath51 is the source field and @xmath52 is the electric current field . the mean current and current - current correlator are then obtained by proper functional differentiations of eq . ( [ j_l ] ) with respect to the source field . here , calculating the current - current correlator , one should remember that due to the topological superconductor ( or fermionic topological constraints present in the field integral ) various anomalous expectation values do not vanish , _ i.e. _ , in general @xmath53 also includes the term @xmath54 . the second derivative of the current noise , @xmath55 , as a function of the second derivative of the mean current times the electronic charge @xmath56 , @xmath57/dv^2 $ ] . the curves are parameterized by the bias voltage @xmath6 which grows in the direction of the arrows from @xmath58 to @xmath59 . for both curves the temperature is the same , @xmath60 . the black curve is for @xmath61 . the red curve is for @xmath62.,width=8 ] the second derivative of the current noise , @xmath55 , as a function of the second derivative of the mean current times the electronic charge @xmath56 , @xmath57/dv^2 $ ] . for all the curves @xmath62 . the temperatures are @xmath60 ( black ) , @xmath63 ( red ) , @xmath64 ( blue ) and @xmath65 ( green ) . the curves are parameterized by the bias voltage @xmath6 which grows in the direction of the arrows from @xmath58 to @xmath59 ( black ) , from @xmath66 to @xmath59 ( red ) , from @xmath67 to @xmath59 ( blue ) and from @xmath68 to @xmath59 ( green).,width=8 ] below we are interested in the so called greater current - current correlator @xmath69 ( as opposed to the lesser one , @xmath70 ) , where @xmath71 is the current fluctuation field and @xmath72 is the mean current . more precisely , we calculate the fourier component @xmath73 as a function of the bias voltage @xmath6 at zero frequency , @xmath74 . the essential and experimentally relevant characteristic of the electric current fluctuations is the effective charge @xmath0 which relates the nonlinear parts of @xmath75 and @xmath76 . in modern experiments @xcite on nonequilibrium current fluctuations in quantum dots one measures the shot noise , @xmath75 , and the mean current , @xmath76 , subtracts the linear parts to get the corresponding nonlinear quantities @xmath77 , @xmath78 and finally obtains the ratio @xmath79 at small bias voltages . in particular , for noninteracting quantum dots one gets the trivial result , @xmath1 . we generalize the above definition of the effective charge in such a way that it reproduces not only the standard definition at small bias voltages but also provides an unique fluctuation fingerprint of a system far from equilibrium where the system s dynamics is highly nonlinear and the expansion in powers of @xmath6 makes no sense at all . to this end we note that at small values of @xmath80 the nonlinear parts of both the shot noise and the mean current are cubic in @xmath6 . therefore , at low voltages the second derivatives @xmath55 and @xmath81 are linear in @xmath6 and thus linearly depend on each other with the ratio @xmath82/\|d^2[ei(v)]/dv^2\|=s^>_k(v)/e\|i_k(v)\|$ ] . therefore , we define the effective charge as : @xmath83}{dv^2}\big\| } , \label{eff_ch}\ ] ] which is applicable when the second derivatives linearly depend on each other . note , that the linear dependence of @xmath55 on @xmath81 does not necessarily imply a linear dependence of these derivatives on the bias voltage . in fact , at large bias voltages expansions in powers of @xmath6 do not exist while @xmath55 may still linearly depend on @xmath81 and , therefore , the effective charge in eq . ( [ eff_ch ] ) makes sense even at extremely high bias voltages . note also that the definition in eq . ( [ eff_ch ] ) is highly consistent because , as shown below , in the absence of majorana fermions it gives for the noninteracting case @xmath1 both at low and very large bias voltages . importantly , this definition is also highly relevant for experiments because each of the second derivatives may be measured with sufficient accuracy already at present . the second derivative of the current noise , @xmath55 , as a function of the second derivative of the mean current times the electronic charge @xmath56 , @xmath57/dv^2 $ ] . for all the curves @xmath61 . the temperatures are @xmath60 ( black ) , @xmath63 ( red ) , @xmath64 ( blue ) and @xmath65 ( green ) . the curves are parameterized by the bias voltage @xmath6 which grows in the direction of the arrows from @xmath58 to @xmath59 for all the curves.,width=8 ] let us consider the situation when @xmath84 . fig . [ figure_1 ] shows @xmath55 as a function of @xmath81 when @xmath85 for two different values of the overlap energy @xmath35 . the black curve is for @xmath61 . in this case the majorana fermions strongly overlap forming a single dirac fermion leading to the current fluctuations with the trivial effective charge equal to the electronic charge both at low ( @xmath86 ) and high ( @xmath87 ) energies . indeed , the curve is linear near the origin both at its starting point and at its ending point with the tangent lines having unit absolute slope resulting in @xmath88 . however , when @xmath62 , majorana bound states overlap weakly and the fluctuation physics is governed by fractional degrees of freedom leading to fractional effective charges at low and high energies . in this case the curve is linear near the origin both at its starting point and at its ending point with the tangent lines having , respectively , absolute slopes equal to 1/2 and 3/2 resulting in @xmath2 at low energies ( @xmath86 ) and @xmath3 at high energies ( @xmath89 ) . at voltages @xmath90 the majorana state is ineffective and the curve acquires the trivial linear character with @xmath1 . strictly speaking , the fractional high energy effective charge @xmath3 is perfectly achieved only at @xmath91 which requires @xmath92 . however , according to our numerical analysis we estimate that for @xmath84 ( and small @xmath35 ) it weakly deviates from the value @xmath93 . namely , @xmath94e$ ] . so that @xmath85 gives @xmath95 , @xmath96 gives @xmath97 and @xmath98 gives @xmath99 . note also that usually the kondo temperature @xmath100 is much smaller than @xmath101 , _ i.e. _ , @xmath102 . therefore , @xmath103 and , as mentioned above , the kondo state is ruined by high bias voltages . as a result , the kondo effect will not create any ambiguity in experimental observations of the fractional high energy effective charge @xmath3 . in fig . [ figure_2 ] we show @xmath55 as a function of @xmath81 for @xmath85 , @xmath62 for different temperatures . since the overlap energy is small , the current fluctuations are essentially governed by the majorana degrees of freedom . here we increase the bias voltage of the starting points of the high temperature curves to stay in the regime @xmath104 in order to avoid high values of the thermal majorana noise which is not in the focus of the present research . at high temperatures ( red , blue and green curves ) the low energy effective charge @xmath2 is completely washed out by thermal fluctuations of the electric current . however , the high energy effective charge @xmath3 is robust against thermal noise and persists up to very high temperatures , @xmath65 ( green curve ) . let us estimate the temperature at which the fractional high energy effective charge @xmath3 might be observed in experiments . if the induced superconducting gap is taken from ref . @xcite , @xmath105 , and @xmath106 , then we obtain @xmath107 which is easily reachable in modern experiments . if the induced superconducting gap is taken from ref . @xcite , @xmath108 , then @xmath109 which is even more reachable . fig . [ figure_3 ] demonstrates @xmath55 as a function of @xmath81 for @xmath85 , @xmath110 . here the majorana state is destroyed by the strong overlap of the majorana bound states and the current fluctuations are governed only by the dirac fermions of the quantum dot . similar to the majorana dominated regime the low energy effective charge is not well defined here too due to thermal fluctuations of the electric current . at the same time the high energy effective charge persists up to very high temperatures , @xmath65 ( green curve ) , and gives the trivial electronic charge @xmath111 as it must be in the absence of majorana fermions . finally , we would like to note that the fractional low energy effective charge @xmath112 characterizes the current fluctuations also in the regime @xmath113 . however , the fractional high energy effective charge @xmath3 does not appear in this regime because of the dominant role of the quantum dot dirac fermions in the current fluctuations already at moderate bias voltages . as a consequence , the high energy fluctuations in this regime are always characterized by the trivial effective charge @xmath111 . in conclusion , we have explored strongly nonequilibrium majorana fluctuations of the electric current . it has been shown that in general these fluctuations are characterized by two fractional effective charges @xmath2 and @xmath3 at low and high energies , respectively . we have demonstrated that the low energy effective charge @xmath2 might be washed out by thermal noise but the high energy effective charge @xmath3 is robust and persists up to very high temperatures . the latter , thus , represents a challenge for modern experiments on noise phenomena in quantum dots since it may become a reliable platform for an unique signature of majorana fermions out of strongly nonequilibrium fluctuations .
probably the most promising corner of string theory to construct models with realistic particle spectra is heterotic m - theory , also known as the horava witten setup @xcite . in it , the spacetime is taken to be minkowski space @xmath3 times a calabi - yau threefold @xmath4 times an interval @xmath5 in the eleventh direction . the two 10-dimensional boundaries each support a @xmath6 gauge theory , one of which should be broken by instantons and/or wilson lines to the standard model gauge group . the other @xmath7 is then hidden and only couples gravitationally to the visible sector . in addition , there can be a number of @xmath8-branes wrapping a curve of the calabi - yau threefold in the interior of the interval . in order to realize weak - scale supersymmetry breaking either the @xmath8-brane or the hidden @xmath7 gauge sector should break supersymmetry and gravitationally mediate the effect to the ( @xmath9 supersymmetric ) visible @xmath10 sector . in order to find a compactification of heterotic m - theory one thus has to specify a calabi - yau threefold @xmath4 and two @xmath6 gauge bundles on it . for our purposes it will be convenient to pick an @xmath11 subgroup in @xmath6 , and use this group embedding to promote @xmath11 bundles ( that is , vector bundles @xmath12 and @xmath13 ) to @xmath6 bundles . the commutant of its holonomy @xmath14 inside the visible @xmath10 is then the unbroken low - energy gauge group in @xmath15 dimensions . here , we are going to be interested in the case where @xmath12 is a rank @xmath15 holomorphic , slope stable ( hence preserving @xmath9 by itself ) vector bundle together with a @xmath16 wilson line . in other words , @xmath17 the unbroken gauge group is then @xmath18 the standard model gauge group times a @xmath19 which constrains proton decay . the ( visible ) matter fields of the effective @xmath15-dimensional theory correspond to the zero modes of the dirac operator @xmath20 on the calabi - yau threefold @xmath4 . they can be computed as the @xmath21 cohomology groups of the vector bundle @xmath12 and associated bundles @xmath22 , @xmath23 , @xmath24 . in @xcite , we constructed a suitable vector bundle yielding precisely the mssm matter spectrum of * three families of quarks and leptons , * one ( up , down ) pair of higgs , and * no other charged fields . one interesting possibility for the hidden sector ( cancelling the heterotic anomaly ) consists of an unbroken @xmath7 plus anti - five branes wrapping a rigid curve in the bulk @xcite . see also @xcite for other heterotic constructions . mathematically , deformations of holomorphic vector bundles are unobstructed ( at a smooth point in the bundle moduli space ) . therefore , there can not be any perturbative superpotential for the bundle moduli fields . however , world - sheet instantons can , in principle , generate a superpotential . the action for a worldsheet instanton wrapped on a holomorphic curve @xmath25 is @xmath26 = \int_c \omega , \ ] ] where @xmath27 is the complexified khler class @xmath28 the instanton - generated superpotential @xmath29 is then the sum over all holomorphic curves and weighted with the pfaffian ( a function of those moduli which determine the bundle restricted to @xmath25 ) : @xmath30 where * @xmath31 are the @xmath32 vector bundle moduli , * @xmath33 are the @xmath34 complex structure moduli , and * @xmath35 are the @xmath36 khler moduli . ignoring any torsion parts in homology and cohomology for the moment , we can generate @xmath37 by the basis two - forms @xmath38 of eq . and , therefore , write the homology class of the curve @xmath25 as @xmath39 = ( n_1,\dots , n_{h^{11 } } ) = \big ( \int_c j_1 , \dots , \int_c j_{h^{11 } } \big ) \in { \ensuremath{{\mathbb{z}}}}^{h^{11}}\simeq h_2(x)\ ] ] usually , we use the fourier - transformed khler moduli @xmath40 and write the instanton action part as @xmath41 unlikely as this may seem , in common constructions , like gauged linear sigma models and monad constructions , the bundle moduli superpotential eq . vanishes . one possible explanation was offered by beasley and witten , who have argued @xcite that the sum over all instantons within a given homology class vanishes under certain circumstances . that is , rewrite eq . as @xmath42 = \vec{n } } \mathop{\text{pf } } ( \phi , c ) \right)}_{=0\text { ? } } \prod q_a^{n_a}\ ] ] then , for bundles coming from the toric ambient space , the inner sum actually does vanish . as we will see later in subsection [ sec : numbers ] , this argument does not apply to our vector bundle . since our underlying calabi - yau threefold @xmath4 is not a toric complete intersection , this should not be too surprising . we now construct the calabi - yau threefold @xmath4 as a free @xmath16-quotient of a simply connected calabi - yau threefold @xmath43 . clearly , then , the fundamental group @xmath44 allows for suitable wilson lines , see subsection [ sec : hsm ] . the covering space @xmath43 is the complete intersection calabi - yau threefold defined by a degree-@xmath45 and @xmath46 polynomial in @xmath47 . discrete symmetries appear for special values of the complex structure moduli . in fact , there are _ two different _ loci in the complex structure moduli space with free @xmath48 group actions , leading to two different @xmath48 quotients . the two quotients have different complex structures , but can not be distinguished at the level of cohomology groups . one of the quotients was used in @xcite to construct a heterotic mssm , and the explicit group action and polynomial equations can be found in @xcite . however , we are now going to consider the other quotient , where one can use toric mirror symmetry . the covering space of the second quotient @xmath49 } \times { \ensuremath{\mathop{\null { \mathbb{p}}}}\nolimits}^1_{[t_0:t_1 ] } \times { \ensuremath{\mathop{\null { \mathbb{p}}}}\nolimits}^2_{[y_0:y_1:y_2]}\ ] ] is the complete intersection ( see also @xcite ) @xmath50 where we labelled three particular complex structure moduli @xmath51 . the significance of these three moduli is that they parametrize @xmath16 symmetric complete intersections . explicitly , the group action is ( @xmath52 ) @xmath53 \mapsto [ x_0:\zeta x_1:\zeta^2 x_2 ] \\ [ t_0:t_1 ] \mapsto [ t_0:t_1 ] ~\text{(no action ) } \\ [ y_0:y_1:y_2 ] \mapsto [ y_0:\zeta y_1:\zeta^2 y_2 ] \end{cases } \\ g_2:&\ ; \begin{cases } [ x_0:x_1:x_2 ] \mapsto [ x_1:x_2:x_0 ] \\ [ t_0:t_1 ] \mapsto [ t_0:t_1 ] ~\text{(no action ) } \\ [ y_0:y_1:y_2 ] \mapsto [ y_1:y_2:y_0 ] . \end{cases } \end{split}\ ] ] since the two equations are each cubic in one @xmath54 , one can easily see that @xmath43 is a double elliptic fibration . hence its euler number must be @xmath55 . a straightforward geometric calculation yields the hodge diamond , which turns out to be @xmath56 note that , as mentioned above , @xmath4 has @xmath57 complex structure parameters . the hurewicz isomorphism tells us that @xmath58 together with the hodge diamond eq . , the universal coefficient theorem and poincar duality , this determines the integral homology up to a finite , abelian group @xmath59 ( `` torsion '' ) . the resulting integral homology groups are given in table [ tab : hxz ] . [ tab : hxz ] @xmath60 the hard part is to compute the torsion part @xmath61 , that is , the torsion homology classes of curves in @xmath4 . clearly , such a torsion homology class can not be represented by a holomorphic curve . however , as we will see , torsion homology classes can be represented by formal differences of holomorphic curves . it turns out that @xmath62 how did we obtain this ? first , note that by definition we have a quotient map @xmath63 which induces the push - forward in homology , @xmath64_{q } & h_2({{\ensuremath{\widetilde{x}}}},{\ensuremath{{\mathbb{z}}}})={\ensuremath{{\mathbb{z}}}}^{19 } \ar[d]_{q_\ast } \\ x & h_2(x,{\ensuremath{{\mathbb{z}}}})={\ensuremath{{\mathbb{z}}}}^3\oplus { \ensuremath{{\mathbb{z}}}}_3 \oplus { \ensuremath{{\mathbb{z}}}}_3 } } \ ] ] the @xmath16 group action on @xmath43 identifies curves @xmath65 the quotient under this equivalence relation is called `` coinvariant homology '' , and denoted as @xmath66 . by explicitly constructing a basis of curves on @xmath4 and identifying the group action on them , one can explicitly determine the @xmath16-group action on @xmath67 in terms of two commuting @xmath68 matrices . we found that @xcite @xmath69 in general , the relationship between coinvariant homology and the homology of the quotient @xmath70 is determined by the cartan - leray spectral sequence @xmath71 without going into details , we point out that the @xmath72 column of the @xmath73 tableau is the coinvariant homology , @xmath74 but , as usual , there are more entries in the spectral sequence , as well as higher differentials and extension ambiguities . however , one can show @xcite that these ultimately do not contribute to @xmath75 . to summarize , the homology classes of curves are given in eq . the torsion curve classes can be understood as coinvariant curve classes on the covering space @xmath43 . having constructed the calabi - yau threefold @xmath4 , we would like to identify the holomorphic curves on it . the well - known trick to count the curves is via the instanton corrections to the yukawa couplings in the standard embedding , @xmath76 the instanton numbers are encoded in the power series @xmath77 , multi - indexed by @xmath78 . fortunately , the @xmath79 power series come from a single prepotential @xmath80 the trilogarithm in the prepotential takes care of multi - covers of a single instanton . written as in eq . , the expansion coefficients @xmath81 are the instanton numbers , and they count are still integers , but may be negative . ] the holomorphic curves in the homology class @xmath82 . after properly distinguishing the torsion curves , the prepotential must be of the form @xmath88 knowing the general form of the series expansion , how do we go about and compute it ? we compared three different approaches @xcite * start on @xmath43 and mod out @xmath16 . + the prepotential on @xmath43 is partially known @xcite . the quotienting by the @xmath16 action essentially amounts to a variable substitution @xcite , and one obtains @xmath89{l } { \ensuremath{\theta_{e_8}}}\big ( q^3;\ , q^2 b_1 ^ 2 b_2 ^ 2,\ , q b_2,\ , 1,\ , b_1 ^ 2 b_2,\ , b_2 ^ 2,\ , 1,\ , b_1 ^ 2,\ , q^{-1 } \big ) \times \\ { \ensuremath{\theta_{e_8}}}\big ( r^3;\ , r^2 b_1 b_2,\ , r b_2 ^ 2,\ , 1,\ , b_1 b_2 ^ 2,\ , b_2,\ , 1,\ , b_1,\ , r^{-1 } \big ) \times \\ p\big ( q^3 \big)^{12 } p\big ( r^3 \big)^{12 } + o(p^2 ) . \end{array } \end{split}\ ] ] * identify and count curves directly on @xmath4 . + by counting sections and singular kodaira fibers in the double torus fibration @xmath4 , one can directly count the curves up to linear order in @xmath90 . the result is @xmath91 setting @xmath92 , the equality of eq . and eq . is a known identity of @xmath6 theta functions @xcite , but we do not know a mathematical proof in general . * toric mirror symmetry . + this approach yields the full power series for the whole prepotential , but only numerical with degree bounded by computer resources . in practice , it is useful to look at ( toric ) @xmath93-quotients of the toric complete intersection @xmath43 @xcite . v. braun , m. kreuzer , b. a. ovrut and e. scheidegger , _ worldsheet instantons , torsion curves , and non - perturbative superpotentials _ b * 649 * , 334 ( 2007 ) [ arxiv : hep - th/0703134 ] . v. braun , m. kreuzer , b. a. ovrut and e. scheidegger , _ worldsheet instantons and torsion curves . part a : direct computation _ , jhep * 0710 * , 022 ( 2007 ) [ arxiv : hep - th/0703182 ] . v. braun , m. kreuzer , b. a. ovrut and e. scheidegger , _ worldsheet instantons and torsion curves , part b : mirror symmetry _ , jhep * 0710 * , 023 ( 2007 ) [ arxiv:0704.0449 [ hep - th ] ] . p. horava and e. witten , _ eleven - dimensional supergravity on a manifold with boundary _ , nucl . b * 475 * ( 1996 ) 94 [ arxiv : hep - th/9603142 ] . a. lukas , b. a. ovrut and d. waldram , _ non - standard embedding and five - branes in heterotic m - theory _ , phys . rev . d * 59 * , 106005 ( 1999 ) [ arxiv : hep - th/9808101 ] . v. braun , y. h. he , b. a. ovrut and t. pantev , _ the exact mssm spectrum from string theory _ , jhep * 0605 * , 043 ( 2006 ) [ arxiv : hep - th/0512177 ] . v. braun , y. h. he and b. a. ovrut , _ stability of the minimal heterotic standard model bundle _ , jhep * 0606 * , 032 ( 2006 ) [ arxiv : hep - th/0602073 ] . v. braun and b. a. ovrut , _ stabilizing moduli with a positive cosmological constant in heterotic m - theory _ , jhep * 0607 * , 035 ( 2006 ) [ arxiv : hep - th/0603088 ] . w. buchmuller , k. hamaguchi , o. lebedev and m. ratz , _ supersymmetric standard model from the heterotic string _ , * 96 * , 121602 ( 2006 ) [ arxiv : hep - ph/0511035 ] . v. bouchard and r. donagi , _ an su(5 ) heterotic standard model _ , phys . b * 633 * , 783 ( 2006 ) [ arxiv : hep - th/0512149 ] . c. beasley and e. witten , _ residues and world - sheet instantons _ , jhep * 0310 * , 065 ( 2003 ) [ arxiv : hep - th/0304115 ] . v. braun , t. brelidze , m. r. douglas and b. a. ovrut , _ calabi - yau metrics for quotients and complete intersections _ , arxiv:0712.3563 [ hep - th ] . p. candelas , x. de la ossa , y. h. he and b. szendroi , _ triadophilia : a special corner in the landscape _ , arxiv:0706.3134 [ hep - th ] . s. hosono , m. saito , and j. stienstra , _ on the mirror symmetry conjecture for schoen s calabi - yau 3-folds _ , arxiv : alg - geom/9709027 . given at taniguchi symposium on integrable systems and algebraic geometry , kyoto , japan , 7 - 11 jul 1997 .	we study aspects of worldsheet instantons relevant to a heterotic standard model . the non - simply connected calabi - yau threefold used admits @xmath0 wilson lines , and a more detailed investigation shows that the homology classes of curves are @xmath1 . we compute the genus-@xmath2 prepotential , this is the first explicit calculation of the gromov - witten invariants of homology classes with torsion ( finite subgroups)@xcite . in particular , some curve classes contain only a single instanton . this ensures that the beasley - witten cancellation of instanton contributions can not happen on this ( non - toric ) calabi - yau threefold .
the author is supported by the epsrc advanced fellowship .	we correct some errors in the two papers published with the above title in class . quant . grav . * 19 * ( 2002 ) . in particular , the correct prescription for computing the probabilities is given , in that appropriate normalization factors are introduced . the resulting computation of the semi - classical limit of probabilities actually becomes much simpler , and no cft analysis is necessary . in spite of some mistakes , the conclusions of these two papers are to a large extent unchanged . in particular , we still get an exponentially small answer @xmath0 for the black hole creation - evaporation probability . let us first emphasise that the prescription given in the two papers @xcite should be viewed as more of an outline of a theory of @xmath1 quantum 3d gravity than an actual proposal for such a theory . the arguments of @xcite do lead to quantitative predictions only in the semi - classical limit of the large radius of curvature , when it does not matter which exactly cft is used . in these note we correct the prescription given in @xcite , and correct the answer for the black hole out of two point particles creation probability . the result is still exponentially small @xmath2 , where @xmath3 are the inverse temperature and mass of the black hole being created . the interpretation of this result is basically unchanged from the one given in @xcite . there many gaps in the arguments of @xcite , some of which have been filled after the papers were written , some are still to be settled . the main open issue is still to find the cft which is relevant for @xmath1 quantum gravity and which is referred to in these papers as a relative of liouville theory . once this cft is found , the prescription for computing amplitudes formulated in these two papers can be used for computing probabilities of physically interesting processes . today , four years after the papers have been written , the outlined programme for the construction of the theory of @xmath1 quantum gravity is closer to its completion . important advances have been made in understanding the chern - simons theory of the lorentz group , see @xcite and references therein . some important progress has also been achieved in understanding the classical theory . thus , the analytic continuation procedure on which the papers @xcite are based was made much more precise in the works of benedetti and bonsante , see @xcite and references therein . more recently , the work @xcite in particular analysed the phase space of @xmath1 gravity , and it was shown that the reduced phase space is the cotangent bundle over the teichmuller space of the spatial slice . the same work also proposed yet another well defined analytic continuation procedure between the lorentzian and euclidean signature negative curvature 3-manifolds . we now turn to corrections for the first paper . * in several places in the paper ( e.g. page 3982 , the paragraph before last on this page , page 3985 , last paragraph of section 3 ) it is incorrectly stated that manifolds containing lines of conical singularities ( point particles ) can be obtained by considering discrete groups , subgroups of @xmath4 , containing elliptic elements . this is in general not true . a group generated , in particular , by elliptic generators , in general fails to be discrete , and the quotient @xmath5 is not a manifold . this happens already in the case of a single elliptic generator , unless the deficit angle is rational . thus , in general , manifolds containing point particles are not quotients of @xmath6 and have to be described and treated differently . a good construction for such manifolds is to consider a hyperbolic manifold , such as e.g. @xmath6 , with a two geodesic half - planes identified , so that a wedge is cut out . the axis of the transformation that maps one half - plane into the other becomes particle s worldline . to summarize , the spaces considered in this paper do exist , they are just not given by quotients anymore . all other arguments of the paper go through unchanged . the description of spaces with particles is more involved than that of non - singular spaces , and is subject of current interest of the mathematical community . we refer the reader e.g. to @xcite , where an extensive treatment of all such issues is given . * the argument given on page 3986 explaining why it is natural for the quantum states to be functionals on the teichmuller space is essentially correct , in that it uses the fact that the reduced phase space of @xmath1 gravity is the cotangent bundle over the teichmuller space . this last fact , however , can be shown much more directly , see @xcite for the proof . * the prescription for extracting probabilities that is given in the paper needs to be modified , as the numbers one gets using the prescription of the paper are not normalised . the modification is straightforward . let us first consider the case of emission of a particle by the btz black hole . the state ( amplitude ) for a black hole with a particle emitted is correctly given by ( 5.1 ) . however , to extract the probability , is not enough to simply square the amplitude . the correct procedure is as follows . let us introduce a state @xmath7 describing the btz black hole with no extra point particle . this state is given by the cft partition function on the cylinder . this is a state in the tensor product of two hilbert spaces , one hilbert space @xmath8 for each asymptotic region . the hilbert space in question is just that of the cft . namely , it is the direct sum @xmath9 of verma modules @xmath10 , where @xmath11 runs over the set of conformal dimensions of primary fields of the theory . thus , the amplitude ( state ) of the btz black hole , as well as the state of the black hole with a point particle emitted , is in @xmath12 . to find the probability of emission we need to find an overlap of these two states . it is easiest to do this by introducing density matrices that refer only to one of the asymptotic regions , and then applying the usual rules of quantum mechanics to these density matrices . thus , for the black hole , we introduce the density matrix @xmath13 by taking the product of @xmath7 with @xmath14 and summing over the degrees of freedom of one of the asymptotic regions . as we have sketched in the paper , such a sum is really a sum over the boundary conditions at that asymptotic region , and its effect is to glue two riemann surfaces together , to produce a `` longer '' cylinder . both ends of this cylinder now correspond to the same asymptotic region : @xmath15 \,\ , z_{\rm cft}\left[\,\ , \lower0.15in\hbox{\epsfig{figure = cylinder - ba.eps , height=0.3 in } } \,\,\right]= z_{\rm cft}\left[\,\ , \lower0.15in\hbox{\epsfig{figure = cylinder - aa.eps , height=0.3 in } } \,\,\right]\ ] ] one applies a similar procedure to the state of the black hole with point particle , to obtain the density matrix as the cft partition function on the `` long '' cylinder with two vertex operators inserted : @xmath16\ ] ] the probability of the emission process is now given by the overlap of these two density matrices , divided by the traces of each of these matrices : @xmath17 this is related to the quantity ( 5.2 ) . the later is , however , not correctly normalised to be interpreted as the probability . the arguments given later in section 5.1 still apply and show that ( [ emission - prob ] ) is qualitatively correct . * the prescription given in section 5.2 should also be modified . the expression 5.11 for the state of ads with two point particles in it is correct . however , to extract the black hole creation probability one should follow a procedure similar to the one above . we again introduce the density matrix of the black hole given by ( [ density - bh ] ) . one can similarly create the density matrix describing the pure state ( 5.11 ) : @xmath18\ ] ] the probability of black hole creation is given by the overlap of the density matrices of the black hole and that of the pair of point particles , properly normalised : @xmath19 the quantity in the numerator is the 4-point function ( 5.12 ) projected on a particular intermediate state ( determined by the black hole size ) . the quantities in the denominator are the cft partition function on the torus , and the partition function on the 4-punctured sphere correspondingly . the arguments of the second paper can still be applied to this expression , and conclusions will be analysed below . here are the corrections for the second paper . * the prescription for computing the black hole creation probability given in the introduction should be replaced by the prescription ( [ creation - prob ] ) above . the prescription ( 1.3 ) is similar , but incorrectly normalised . thus , the derivation of the answer ( 1.4 ) should be modified . the correct semi - classical regime answer can be obtained very easily as follows . in the semi - classical limit the 4-point function is peaked on a particular intermediate state - the one that corresponds to the black hole that would be created in the process . therefore , in the expression for the probability of creating this black hole the numerator of ( [ creation - prob ] ) , which is the 4-point function projected on a particular intermediate state , is approximately equal to the second term in the denominator , which is the full 4-point function . thus , in the semi - classical regime , the creation probability is given by inverse of the black hole partition function : @xmath20 here @xmath21 are the bh partition function , free energy and entropy correspondingly , and @xmath3 are the inverse temperature and mass of the bh being created . this last expression coincides with the one given by ( 1.4 ) . note , however , that it is incorrectly stated in the text that the probability is given by @xmath22 . the correct answer is given by ( [ prob ] ) . the arguments of the remainder of the introduction go through almost unchanged . * the analysis of the rest of the paper is essentially correct , but as is clear from the previous argument , is not necessary in the semi - classical regime . the contents of the sections 2 - 3 are interesting in its own right , as they deal with the classical theory . we believe that the contents of sections 5 - 6 are still of interest , as they demonstrate the problems that will have to be overcome for a full quantum computation of the probability to be performed . the analysis of this sections is also necessary to confirm that the probability is peaked on the black hole that would be created in the corresponding classical process . the analysis of the probability in section 7 needs to be modified in that the normalising factors ( [ creation - prob ] ) should be introduced . for this reason , the answer ( 7.7 ) needs to be modified , with the correct answer given above by ( [ prob ] ) . let us also note that there is a sign mistake in section 7 getting ( 7.7 ) as the value of the 4-point function . the logarithm of such a 4-point function for the case of maximally massive particles is equal ( in the semi - classical regime ) to half of the logarithm of the black hole partition function . thus , the semi - classical limit of the 4-point function for such particles is given by an exponential with a _ positive _ large argument . this answer makes it obvious that normalization factors such as ones in ( [ creation - prob ] ) are necessary to obtain the probability . * the argument in the discussion section , page 4023 , paragraph before last on this page , has to be modified in view of the fact that the correct answer is given by ( [ prob ] ) . thus , the exponentially small probability is not surprising in view of the fact that the probability we got can also be interpreted as the probability of a bh of mass @xmath23 blowing up into two particles . the probability @xmath2 we got is that of a thermal emission of all of @xmath24 mass @xmath23 at temperature @xmath25 , which seems a natural answer .
the r - process , or rapid neutron - capture process , of stellar nucleosynthesis is invoked to explain the production of the stable ( and some long - lived radioactive ) neutron - rich nuclides heavier than iron that are observed in stars of various metallicities , as well as in the solar system ( for a review , see * ? ? ? despite important effort to model potential r - process sites , all the proposed scenarios face serious problems and the site(s ) of the r - process is ( are ) not identified yet . until now , type - ii supernovae or @xmath5-ray bursts models have failed to provide convincing evidence for a successful r - processing that could significantly contribute to the galactic enrichment in r - material @xcite . only magneto - rotational supernova explosions with extremely strong pre - collapse magnetic fields and fast rotation seem to provide favourable conditions for r - processing , but such rare events are not expected to be at the origin of the global galactic enrichment in r - process nuclei @xcite . for this reason , special attention is now being paid to neutron star ( ns ) mergers following the confirmation by hydrodynamic simulations that a significant amount of r - process enriched material , typically about @xmath6 to a few @xmath7@xmath8 , can be ejected @xcite . recent nucleosynthesis calculations by @xcite show that the combined contribution of both the dynamical ( prompt ) ejecta expelled during binary ns or ns - black hole ( bh ) mergers and the neutrino and viscously driven outflows generated during the post - merger remnant evolution of relic bh - torus systems can lead to the production of r - process elements from mass number @xmath9 up to thorium and uranium . the corresponding abundance distribution reproduces the solar distribution extremely well and can also account for the elemental distributions observed in low - metallicity stars @xcite . furthermore , recent studies @xcite have reconsidered the galactic or cosmic chemical evolution of r - process elements in different evolutionary contexts . although they do not converge towards one unique quantitative picture , most of them arrived at the conclusion that double compact star mergers may be the major production sites of r - process elements . despite the recent success of nucleosynthesis studies for ns mergers , the possibility of r - processing in these events is still affected by a variety of uncertainties . in particular , the impact of neutrino interactions is not yet studied and understood in detail , the main reason of which is the not yet manageable computational complexity associated with neutrino transport in a generically three - dimensional ( 3d ) , highly asymmetric environment with nearly relativistic fluid velocities and rapid changes in time . a computationally simple and much more efficient alternative to solving the time - dependent transport equation for neutrino distributions in the six - dimensional phase space is the use of a neutrino leakage scheme , in which the local neutrino net - emission ( i.e. , emission minus absorption ) rate is estimated by a weighted interpolation between the pure emission rate and an optical - depth dependent diffusive loss term ( e.g. , * ? ? ? * ; * ? ? ? however , neutrino absorption can not be straightforwardly included in a self - consistent manner in a leakage scheme because information about the local neutrino densities is missing in such a treatment . it was long believed that neutrino interactions could not , at least not drastically , affect the initial neutron richness of the ejecta . such expectations were based on numerical merger models . the temperatures of the merging nss in these simulations , and therefore the neutrino production rates , remain low until the nss collide with each other , and they rise in an increasing volume only gradually on a time scale of several milliseconds after the first contact of the two nss @xcite . at this time a significant fraction of the ejecta material is already being expelled with nearly relativistic velocities . newtonian merger studies suggested that a large part of the ejecta material even from symmetric mergers ( i.e. , for two equal - mass or close to equal - mass nss ) is thrown out by tidal forces in extended spiral arms forming from matter of the outer faces of the merging objects during their final approach and collision @xcite . by the tidal stretching these arms naturally remain unshocked and thus stay cool . therefore they reach low densities so quickly that electron and positron captures can not become efficient . moreover , the ejecta escape to large radii before the neutrino emission of the compact merger remnant becomes sizable and neutrino absorption can affect the electron fraction significantly . absorption of neutrinos radiated by the massive merger remnant is also diminished because the tidal ejection in newtonian models happens preferentially in the orbital plane ( see e.g. * ? ? ? * ; * ? ? ? * ) while the neutrinos are predominantly emitted perpendicular to this plane @xcite . however , the described situation applies well only for newtonian mergers . the situation is different in the relativistic case . relativistic simulations of symmetric mergers do not show the development of prominent tidal arms , and the ejection of unshocked matter is therefore not important . instead , the collision shock that builds up at the interface of the two nss is typically stronger than in newtonian conditions ( see e.g. * ? ? ? * ) , leading to potentially higher temperatures and higher , faster rising neutrino luminosities @xcite . the ejecta are expelled fairly spherically instead of equatorially @xcite and consist of two main components , namely a first one in which very hot material is squeezed out from the collision interface of the two merging bodies , and a second , slightly delayed one that is expelled in waves from a torus - like belt of matter around the high - density core of the merger remnant . this torus is heated by spiral shocks that are sent out from the aspherical , wobbling , and rotating high - density core and that also lead to outward acceleration of parts of the torus matter @xcite . indeed , recent relativistic ns - ns merger simulations that took into account neutrino emission by means of a leakage scheme and absorption by an additional approximate transport treatment based on a moment formalism @xcite , found that neutrino interactions with free nucleons can significantly increase the electron fraction in the dynamical ejecta for cases in which the collapse of the merger remnant to a black hole is delayed or does not happen . under such conditions nuclei with mass numbers @xmath10 can also be created in the dynamical ejecta in addition to the heavy r - process elements ( @xmath11 ) . weak interaction processes of free nucleons can consequently affect the strength of the r - process and the emerging abundance distribution . the accurate inclusion of neutrino interactions in hydrodynamical simulations remains a highly complex task . this motivated us to conduct a simple , parametric study in order to quantify the potential impact of weak interactions on the electron - fraction evolution in merger ejecta and thus to explore the consequences of charged - current neutrino - nucleon reactions for the nucleosynthesis and possible r - processing in these ejecta . more specifically , we investigate the influence of @xmath0-interactions of electron neutrinos ( @xmath2 ) and electron antineutrinos ( @xmath12 ) with free @xmath13 and @xmath14 and of their inverse reactions , @xmath15 on the @xmath1 distribution and r - process nucleosynthesis at conditions representative of the dynamical ejecta expelled by hydrodynamical forces during ns - ns mergers . these reactions have been neglected in all previous studies of r - process nucleosynthesis for such ejecta except those of @xcite , where their quantitative effects may depend on the adopted equation of state @xcite . the role of weak interactions for the electron fraction and the corresponding implications for r - process nucleosynthesis , however , demand further exploration in more detail , in particular also by basic , parametric modeling , because a multitude of uncertainties will prevent rigorous , self - consistent solutions of the full problem in the near future . such uncertainties are associated with , for example , the extreme complexities of 3d energy - dependent neutrino transport in relativistic environments , with the neutrino opacities of dense , potentially highly magnetized matter , and with neutrino - flavor oscillations at rapidly time - variable , largely aspherical conditions of neutrino emission . the main objective of our present work is a sensitivity study by the use of a parametric approach . it is intended to motivate further explorations of neutrino effects in relativistic ns - ns mergers in more detail and breadth . to this end , we set up a simplified and idealized theoretical framework to test the individual roles of the different weak interaction processes , considering the density and temperature evolution of fluid elements ejected from a prototype hydrodynamical , relativistic ns - ns merger model . for our parameter study we make assumptions about the neutrino emission properties that are guided by data taken from the literature . the electron fractions resulting from the neutrino - processing of the ejecta elements are then used as input for nuclear network calculations , allowing us to immediately link the effects of neutrino interactions to the final heavy - element production . in sect . [ sect_weak ] the employed merger model and our treatment of the weak neutrino reactions with free nucleons are described . the effects of @xmath0-processes on the electron fraction are reported in sect . [ sect_ye ] , and the subsequent r - process nucleosynthesis is analysed in sect . [ sect_rpro ] . conclusions are drawn in sect . [ sect_conc ] . we adopt the density evolution of ejecta fluid elements from a representative ns - ns merger model , namely the symmetric 1.35@xmath8 - 1.35@xmath8 binary model obtained with the temperature - dependent dd2 equation of state @xcite ( essentially identical to the one in * ? ? ? * but with a higher resolution of @xmath1610@xmath17 particles ) . this relativistic hydrodynamic simulation also provides the temperature evolution . we do not apply any temperature post - processing as in @xcite to disentangle temperature jumps in shocks from artificial heating associated with the use of a numerical viscosity in the smoothed - particle hydrodynamics scheme . for each trajectory , we follow the expansion starting at a fiducial density of @xmath18g@xmath19 , where we assume equilibrium to hold between electrons , positrons and neutrinos for the given total lepton number provided by our ns - ns merger model . below @xmath18g@xmath19 , electron , positron and electron neutrino and antineutrino captures are systematically included . reactions of neutrinos on nuclei are , however , neglected . as long as the temperature remains in excess of typically @xmath20 k , the abundance of heavy nuclei is determined by nuclear statistical equilibrium ( nse ) at the given electron fraction , density and temperature . from the density @xmath21 down to the density @xmath22 , at which the temperature reaches 10gk and the full reaction network is initiated , the @xmath0-interactions of free nucleons may affect the electron fraction @xmath1 . if a trajectory stays cooler than 10gk below the density @xmath21 , the network calculation is started at the neutron - drip density @xmath23 , i.e. , @xmath24g@xmath19 . the considered @xmath0-reactions involve free nucleons , whose abundance variations are given by @xmath25 where @xmath26 and @xmath27 . the @xmath28 denote capture rates of species @xmath29 onto free nucleons according to the @xmath0-reactions , eqs . ( [ eq : betareac1],[eq : betareac2 ] ) . the free neutron and proton numbers are related to @xmath1 by @xmath30 where @xmath31 is the molar fraction ( and @xmath32 the mass fraction ) of the nucleus ( @xmath33 ) of atomic mass @xmath34 . assuming that the nse molar fractions of nuclei remain constant over the time step @xmath35 , the time evolution of @xmath1 can be related to the change of the number of free protons , and written as @xmath36 where @xmath37 and @xmath38+\lambda_- \sum_{z\ge 2}zy \ , . \label{eq_ls}\ ] ] if only free neutrons and protons are present , @xmath39 . the impact of heavy nuclei is to increase @xmath40 towards @xmath41 ( for matter made of @xmath42-particles only , @xmath43 ) . the @xmath42-effect @xcite , or more generally , the effect of heavy nuclei in binding neutrons and protons inside nuclei , is known to be responsible for driving @xmath1 towards 0.5 and is included in this term @xmath40 . if we assume that @xmath44 and @xmath40 remain constant over the time step @xmath35 ( or , specifically , that @xmath35 is chosen such that @xmath44 and @xmath40 as well as the abundance of nuclei remain essentially constant during the time step ) , eq . ( [ eq_dye ] ) can be integrated analytically leading to @xmath45 \ , , \label{eq_ye}\ ] ] where @xmath44 and @xmath40 are estimated at time @xmath46 . this equation is used to follow @xmath1 from the initial density @xmath21 down to the density @xmath47 , at which the temperature has dropped to 10 gk ( or , alternatively , to the drip density if temperatures above 10 gk are not reached for the considered trajectory ) . for @xmath48 , @xmath1 reaches the equilibrium value @xmath49 given by @xmath50 the electron ( anti)neutrino capture rates can be written in terms of the average ( anti)neutrino capture cross sections @xmath51 @xcite as @xmath52 here , the local ( anti)neutrino number densities are expressed by the ratios of the global luminosities , @xmath53 , and the mean energies of the radiated neutrinos , @xmath54 , multiplied with the spherical surface @xmath55 that surrounds the central neutrino source at a radial distance @xmath56 ( for every trajectory we adopt the time - dependent radial distance @xmath56 from our hydrodynamical model ) . this @xmath57 dilution of the neutrino flux is a crude approximation and holds , at best , far away from the neutrinosphere , provided the emission is isotropic , i.e. , if directional variations of the neutrino fluxes do not play a role . close to and below the neutrinosphere , however , such a description breaks down but can be justified by the fact that at these locations electron and positron captures dominate and their competition enforces a state of weak equilibrium . at large distances the asymptotic electron fraction is determined by neutrino and antineutrino absorptions , for which reason direction - dependent differences of the neutrino exposure of the ejecta would be important for a detailed discussion of neutrino effects on merger ejecta . nevertheless , despite these shortcomings , we apply the simple ansatz at all radii @xmath56 where @xmath58 in order to discuss basic aspects of the impact of neutrino processes with nucleons in the merger ejecta in a parametric way . similar to eqs . ( [ eq_lnu],[eq_lnub ] ) , the electron and positron capture rates are given in terms of the average electron / positron capture cross sections @xmath59 by @xmath60 where @xmath61 is the speed of light and @xmath62 and @xmath63 the electron and positron densities , as detailed in @xcite . in turn , the average cross sections for electron neutrino and antineutrino captures , as well as those for electron and positron captures , including the weak magnetism and recoil corrections , are taken from @xcite ( see also * ? ? ? while the electron and positron capture rates are temperature- and density - dependent only , the ( anti)neutrino capture rates depend on the ( anti)neutrino luminosities and mean energies , hence require a detailed knowledge of the neutrino properties at each time step . for the present study , we consider representative ( anti)neutrino luminosities and angle - averaged mean energies that are assumed to remain constant in time . for a given mean energy @xmath64 , the electron neutrino temperature @xmath65 is deduced from the relation @xmath66 where @xmath67 is the boltzmann constant and @xmath68 are the fermi integrals @xcite of order @xmath13 for vanishing chemical potential , assuming nondegenerate neutrino spectra . a similar expression holds for the antineutrino temperature . for given ( anti)neutrino luminosities and mean energies this allows us to determine all other moments of the ( anti)neutrino energy spectra , and consequently the corresponding capture cross sections and rates ( eqs . [ eq_lnu],[eq_lnub ] ) . ( black solid line ) as defined by eq . ( [ eq_yeinf_nu ] ) . the black dotted line gives the @xmath69 without weak magnetism and recoil corrections , @xmath70 @xmath71 in eq . ( [ eq_yeinf_nu ] ) . the ( anti)neutrino mean energies are taken consistently from fig . 1 ( lower panel ) of @xcite . asymptotic values of @xmath69 are calculated only for non - negligible neutrino luminosities , @xmath70 for times @xmath72 ms . ] , but for the electron neutrino and antineutrino luminosities from @xcite . the ( anti)neutrino mean energies are taken consistently from fig . 17 ( lower panel ; model bc ) of this reference . ] assuming that electron ( anti)neutrino captures dominate over electron and positron captures and the abundance of heavy nuclei is negligible , the asymptotic value of @xmath1 ( eq . [ eq_yeinf ] ) can be approximated by @xmath73 because the rates can be expressed as @xmath74 ( @xmath75 ) , where @xmath76 and @xmath77 corresponds to the weak magnetism and recoil corrections that can be found in @xcite . the corresponding value of @xmath78 is shown in fig . [ fig_yeinf1 ] for the luminosities and mean energies of @xcite and lies between 0.25 and 0.50 for times @xmath79ms after the merging of the binary nss . these values may , however , still be modified by effects of heavy nuclei . the weak magnetism and recoil corrections on the ( anti)neutrino rates are seen in fig . [ fig_yeinf1 ] to increase the asymptotic value of @xmath78 by up to 20% because they reduce the antineutrino capture cross section and simultaneously increase the neutrino capture cross section @xcite . the neutrino properties , and in particular the antineutrino to neutrino luminosity ratio , are found to vary significantly between different hydrodynamical simulations but also depend on the adopted equation of state @xcite . avoiding the complexity of self - consistent neutrino transport in hydrodynamical simulations , we shall restrict ourselves in our sensitivity study to constant luminosities and mean energies taken at selected times from previous simulations @xcite . we consider first two representative sets of values for electron ( anti)neutrino luminosities and mean energies as obtained by @xcite ( see their fig . 1 ) , namely those corresponding to the instants of 5 and 6ms , i.e. , * case 1 : @xmath80 ms with @xmath81 erg / s ; @xmath82 erg / s ; @xmath83 mev ; @xmath84 mev , * case 2 : @xmath85 ms with @xmath86 erg / s ; @xmath87 erg / s ; @xmath88 mev ; @xmath84 mev . note that case 1 leads to an asymptotic value @xmath89 , whereas case 2 yields @xmath90 ( fig . [ fig_yeinf1 ] ) . in order to test more thoroughly the impact of the neutrino processes on the nucleosynthesis , we also consider here the electron ( anti)neutrino properties calculated in the ns merger simulation of @xcite ( cf model bc in their fig . the corresponding ( anti)neutrino luminosities are shown in fig . [ fig_yeinf2 ] , together with the asymptotic values @xmath78 ( with and without the weak magnetism and recoil corrections ) . overall , lower luminosities are found in this model in comparison with @xcite , but also a relatively higher emission of electron antineutrinos . consequently , lower values of @xmath78 are predicted for this model . in our present sensitivity analysis we also select two cases for the electron ( anti)neutrino luminosities and mean energies from the results of @xcite , namely those corresponding to times of 6 and 10ms ( and hereafter referred to as cases 3 and 4 , respectively ) , i.e. , * case 3 : @xmath85 ms with @xmath91 erg / s ; @xmath92 erg / s ; @xmath93 mev ; @xmath94 mev , * case 4 : @xmath95 ms with @xmath81 erg / s ; @xmath96 erg / s ; @xmath97 mev ; @xmath98 mev . case 3 leads to an asymptotic value @xmath99 , while case 4 yields @xmath100 ( fig . [ fig_yeinf2 ] ) . finally , it should be mentioned that our assumption of time - independent ( anti)neutrino luminosities can be questioned , since the neutrino - ejecta interaction is a highly time - dependent problem , where the relative time between the growth of the neutrino emission and the mass ejection matters . considering constant luminosities is a very crude but simple approximation , which is sufficiently good to demonstrate the impact of neutrino processes on the time evolution and mass distribution of the electron fraction and the corresponding consequences for the r - process . a predictive assessment of neutrino effects on the nucleosynthesis in merger ejecta would also have to take account variations of the neutrino emission with different directions . matter expelled towards the polar directions is exposed to different neutrino conditions than matter that leaves the system along equatorial trajectories . again , a more detailed description of neutrino transport effects is demanded and is beyond the scope of our present parametric study . to illustrate the impact of electron ( anti)neutrino , electron and positron captures on the evolution of @xmath1 , we show in figs . [ fig_traj1 ] and [ fig_traj2 ] the time evolution for two specific trajectories during the expansion from the initial density @xmath18g@xmath19 down to density @xmath47 , where the reaction network calculations are initiated . both trajectories are studied including ( anti)neutrino captures with neutrino properties for case 1 . , between @xmath21 and @xmath47 for trajectory 400720 . _ lower panel _ : analogue for the electron ( anti)neutrino , electron and positron capture rates , and for @xmath40 ( eq . [ eq_ls ] ) and the electron fraction . the ( anti)neutrino properties correspond to case 1 . the late - time increase of @xmath1 is caused by the @xmath42-effect , which does not asymptote to a terminal value until the network is started . @xmath1 continues to evolve subsequently during the nucleosynthesis because of @xmath0-decays . ] , but for trajectory 486857 . ] for trajectory 400720 ( fig . [ fig_traj1 ] ) , neutrino and antineutrino captures on free nucleons with rates up to some @xmath101 lead to a rapid increase of @xmath1 from 0.17 up to 0.31 , corresponding to the asymptotic value of @xmath78 in case 1 . the ( anti)neutrino rates are found to dominate the electron and positron capture rates already at densities slightly below @xmath21 . this is linked to the slow @xmath102 decrease of the ( anti)neutrino fluxes , which is more shallow than the steep temperature dependence ( roughly like @xmath103 ) of the electron and positron capture rates . at @xmath104ms , free neutrons and protons in the expanding matter partially recombine into @xmath42-particles ( @xmath105 ) , thus giving rise to the @xmath42-effect and therefore a further increase of @xmath1 . the rate @xmath40 becomes larger than @xmath106 and approaches @xmath107 of material dominated by @xmath42-particles . for trajectory 486857 ( fig . [ fig_traj2 ] ) , at the initial density @xmath21 , the mass element is characterised by a low temperature @xmath108 k and consequently is composed of heavy nuclei @xmath109 typical of the low values of @xmath110 in the outer ns crust . the fast @xmath111 and ( anti)neutrino capture rates lead to a rapid increase of @xmath1 , but all these rates are comparable and @xmath1 fluctuates wildly . the mass element is then subject to a new compression phase that sets in at @xmath112 ms , and the corresponding high temperatures ( @xmath113 k ) photodissociate the matter into free nucleons . it should be noted that during this recompression episode , the higher densities cause neutrinos to become trapped again , for which reason the application of eqs . ( [ eq_lnu],[eq_lnub ] ) remains problematic and highly schematic . however , at these high - density , high - temperature conditions , electron captures dominate and re - neutronize the material until they are counterbalanced by positron captures , whose rate increases dramatically to achieve weak equilibrium with the electron captures . during the subsequent expansion phase ( at @xmath114ms ) , @xmath1 rises gradually until ( anti)neutrino absorptions take over to push @xmath1 towards its asymptotic value of @xmath115 ( at @xmath1166ms ) . with decreasing temperatures , the @xmath42-effect finally becomes responsible for a late increase of @xmath1 at @xmath117ms . six different cases are studied to estimate the impact of electron and positron captures as well as electron ( anti)neutrino absorption on the ejecta mass distribution as function of @xmath1 . in the first case , weak interactions of free nucleons are not allowed . in the second case , electron and positron captures are turned on , but not ( anti)neutrino absorptions . in the remaining four cases , both electron / positron captures as well as ( anti)neutrino captures are switched on , with the ( anti)neutrino properties being defined by cases 14 . for these six cases , the resulting @xmath1 distributions at density @xmath47 are shown in fig . [ fig_histoye ] . in the first case without weak interactions of free nucleons below @xmath21 ( upper left panel of fig . [ fig_histoye ] ) , the @xmath1 distribution at @xmath47 is identical to the one given at the initial density @xmath21 . this case , however , differs from the standard case we considered previously @xcite in two aspects : first , we start our calculations of weak interactions at density @xmath18g@xmath19 with @xmath0-equilibrium distributions of electrons , positrons and neutrinos . this shifts and broadens the @xmath1 distribution from previous values of @xmath1180.1 to a wider range between 0 and @xmath160.3 . second , no temperature post - processing is performed here . the initial temperatures of the trajectories are significantly higher than those deduced from the post - processing applied in our previous studies . as visible in fig . [ fig_histoye ] , lower left panel , the @xmath1 distribution at density @xmath22 is significantly affected by @xmath119-captures between @xmath21 and @xmath22 , although some low-@xmath1 ( @xmath120 ) ejecta are left . dominant parts of the ejecta are now found at @xmath1 values between 0.3 and 0.4 . when switching on neutrino absorptions , for any of the cases 14 the asymptotic values of @xmath69 are approached , and further enhancement by the @xmath42-effect produces peaks of the mass distributions in a range of @xmath1 values between 0.2 and 0.5 . the peak values depend sensitively on the adopted neutrino properties . cases 1 and 4 lead to rather similar @xmath1 distributions , owing to the fact that the neutrino properties are broadly comparable . these distributions also depend sensitively on the temperature . when the temperature along the trajectory is artificially increased or decreased by factors of 3 , rather different results are obtained , as shown in fig . [ fig_histoye_t ] , using the ( anti)neutrino properties of case 1 . as before we start the expansion evolution at @xmath21 with the @xmath1 mass distribution shown in the upper left panel of fig . [ fig_histoye ] . reduced temperatures diminish the presence of positrons and thus favor electron captures compared to positron captures . moreover , lower temperatures also lead to a faster freeze - out of @xmath121 captures during the ejection of the mass elements . without neutrino and antineutrino absorptions , lower temperatures therefore tend to neutronize the ejecta for @xmath122 and the mass distribution becomes more narrow and is shifted to lower values in the range of @xmath123 ( fig . [ fig_histoye_t ] , upper left panel ) . including neutrino and antineutrino absorptions , reduced temperatures have the opposite effect in pushing the mass distribution to higher values of @xmath1 ( with a peak above 0.4 ) compared to the standard - temperature result for case 1 in fig . [ fig_histoye ] ( upper right panel , with a peak of the distribution between @xmath124 and 0.3 ) . this behavior can be understood by the efficient recombination of free nucleons to @xmath42 particles and heavy nuclei , which strengthens the heavy - nuclei ( @xmath42 ) effect so that the peak of the distribution wanders to @xmath125 . on the other side , increased temperatures reduce the electron degeneracy and thus allow for the presence of higher positron densities , thus enhancing positron captures on neutrons . in addition , @xmath121 captures continue for a longer period of time along the ejecta trajectories . without ( anti)neutrino absorption , these effects shift the @xmath1 mass distribution from the initial one at @xmath21 ( upper left panel of fig . [ fig_histoye ] ) towards higher values of @xmath1 . this shift is stronger for more slowly expanding mass elements and weaker when the expansion is very fast . correspondingly , the mass distribution versus @xmath1 at @xmath22 is very broad and stretches from @xmath160.03 up to a very pronounced maximum close to 0.5 , because the @xmath121 capture equilibrium at high - entropy conditions favors symmetric conditions with respect to neutrons and protons . taking into account ( anti)neutrino absorption prevents this dramatic shift towards @xmath126 , because at large distances neutrino captures dominate @xmath121 absorptions and therefore @xmath1 asymptotes to values around @xmath69 ( @xmath127 for case 1 , lower right panel of fig . [ fig_histoye_t ] ) . since the high temperatures favor nucleons and suppress the early formation of @xmath42 particles and heavier nuclei , the influence of the @xmath42 effect is clearly weaker than in the case of reduced temperatures ( compare lower and upper right panels of fig . [ fig_histoye_t ] ) . as mentioned above , our calculations with varied temperatures are not consistent with the initial @xmath1 distributions used at a density of @xmath21 , because these distributions are calculated for the original ejecta temperatures provided by the hydrodynamic ns - ns merger model . however , fig . [ fig_histoye_t ] demonstrates that asymptotic values determined by neutrino capture equilibrium and influenced by the @xmath42 ( heavy - nuclei ) effect are reached for most of the trajectories . the final @xmath1 values can therefore be expected to mostly have lost the memory of the initial conditions at @xmath21 . correspondingly , when neutrino absorptions are included , the final mass distributions of @xmath1 ( at @xmath22 ) are considerably more narrow than the relatively broad distribution of initial @xmath1 values before the expansion from @xmath21 to @xmath22 ( compare the right panels of fig . [ fig_histoye_t ] with the upper left panel of fig . [ fig_histoye ] ) . at density @xmath47 for @xmath121 captures exclusively at @xmath122 ( left ) and including electron , positron captures as well as ( anti)neutrino captures during the evolution from @xmath21 to @xmath47 ( case 1 ; right ) , when the temperatures are decreased ( upper panels ) or increased ( lower panels ) artificially by factors of 3 . ] at densities @xmath128 , the abundance evolution is followed by a full reaction network ( for more details , see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which in contrast to earlier studies includes now also the weak interactions of free nucleons as detailed in sect . [ sect_weak ] . the final abundance distributions are shown in fig . [ fig_rpro ] for the same six different cases discussed in sect . [ sec : yedistributions ] and displayed in fig . [ fig_histoye ] , namely for a case without weak interactions of free nucleons below @xmath21 , a case including @xmath119-captures but without ( anti)neutrino absorption reactions , and four cases where both @xmath119- and ( anti)neutrino captures are taken into account at @xmath122 ( cases 14 for the @xmath129 properties ) . because of its low-@xmath1 distribution , the case neglecting @xmath0-interactions gives rise to an elemental abundance distribution similar to the one obtained in our previous studies @xcite . it is characterized by the production of essentially only @xmath11 nuclei through several loops of fission recycling . when @xmath121-captures on free nucleons are switched on , some low-@xmath1 ( @xmath120 ) material can still lead to nucleosynthesis with fission recycling and a significant production of the third r - process peak , but the higher-@xmath1 ( 0.30.4 ) matter can now also contribute to the production of @xmath130 nuclei with a strong second @xmath131 peak . when ( anti)neutrino captures are included , too , nucleosynthesis with fission recycling does not take place any longer , but the final abundance distribution still resembles the one in the solar system fairly well . a significant amount of @xmath132 nuclei can now also be produced in addition to the @xmath133 r - nuclei , especially for ( anti)neutrino properties corresponding to case 2 , as shown in fig . [ fig_rpro_nu ] . in this case , important element production around @xmath134ni is obtained , originating from ejected mass elements with @xmath135 . without @xmath0-interactions below @xmath21 , about 90% of the ejected material is found to be r - process rich . if @xmath121-captures on nucleons are switched on , the material is made of 76% r - process nuclei , and when ( anti)neutrino captures are effective , too , we find that 45% of the ejected matter is made of r - nuclei in case 1 , only 1.9% in case 2 , 67% in case 3 and 40% in case 4 . in the last four cases , a significant part of the material is made of @xmath42-particles ( @xmath136% in cases 1 and 4 , 50% in case 2 , and 17% in case 3 ) , and the remaining part consists of @xmath137 nuclei ( see fig . [ fig_rpro_nu ] ) . despite the rather robust production of a solar - like distribution of r - nuclei , the absolute and relative amounts of r - material vary strongly from case to case , sensitively depending on the neutrino exposure of the ejecta as determined by the assumed properties of the neutrino emission . neutrino properties corresponding to case 2 yield a total amount of ejected r - material that is significantly smaller than for the other three sets of neutrino properties . the total , mass - averaged nuclear energy - release rate that is available for heating the ejecta per unit of mass , the average temperature of the ejecta , and the average atomic mass number of the ejected abundance yields are plotted as functions of time for our six studied cases in fig . [ fig_qta ] . after some 10s and up to nearly one day , the energy release rates of all cases are fairly similar except for the treatment of neutrinos and @xmath0-interactions according to case 2 . in this case , the large electron fractions of the ejecta favor the production of light elements and only a small amount ( @xmath138 ) of the ejecta material consists of r - process nuclei ( fig . [ fig_rpro_nu ] ) and is therefore subject to fast @xmath0-decays . at late times , typically after one day , an additional source of energy is found from the @xmath42-decay of long - lived heavy nuclei that can only be significantly produced when no @xmath0-interaction of nucleons are included below @xmath21 . despite the considerable differences of the nuclear energy - release rates at early times , the cooling evolution as measured by the mass - averaged temperatures is rather similar in all cases . the time evolution of the average nuclear mass number of the ejected material clearly shows that only without @xmath0-interactions of free nucleons a significant amount of fissile nuclei can be produced with @xmath139 reaching values up to 170 , which is slightly below the value of about 200 obtained when the @xmath1 distribution of the inner crust of a cold ns is considered as initial state ( see , in particular , * ? ? ? the sequence of decreasing values of @xmath139 follows basically the hierarchy of increasing values of the mean electron fraction , namely @xmath140 when @xmath0-interactions are ignored , @xmath141 when only @xmath119-captures are taken into account , and @xmath142 , 0.34 , 0 . 35 , 0.45 in the cases 3 , 1 , 4 and 2 , respectively . the only ( slight ) inversion is obtained for the calculation with only @xmath119-captures , because in this case the @xmath1 distribution is very wide ( fig . [ fig_histoye ] , lower left panel ) and the significant amounts of low-@xmath1 ( @xmath143 ) material contribute to the production of heavy nuclei which increase the average nuclear mass number @xmath139 . ns merger model , including electron , positron as well as ( anti)neutrino captures for cases 1 and 2 of the neutrino properties . ] ( red ) , mass number @xmath144 ( blue ) , and temperature @xmath145 ( black ) , all mass - averaged over the ejecta , for the 1.351.35@xmath8 ns merger . the solid lines correspond to the case without @xmath0-interactions of free nucleons at @xmath122 , the dotted lines to the case where only @xmath121-captures are taken into account , the dashed lines to the case where @xmath121 and ( anti)neutrino captures for neutrino properties according to case 1 are included , the dash - dot lines for neutrinos according to case 2 , the long - dash - dot lines correspond to case 3 , and the double - dash - dot lines to case 4 . ] in this paper we reported the results of a parametric study to investigate how @xmath0-interactions of free nucleons can affect the @xmath1 evolution and mass distribution in ns merger ejecta and the corresponding nucleosynthesis . to this end we used the temperature - density trajectories of a large set of mass elements representing the @xmath146 of matter ejected in a relativistic merger simulation of a symmetric 1.35@xmath8 - 1.35@xmath8 ns binary with the non - zero temperature dd2 nuclear equation of state . using the total lepton number provided by our ns - ns merger simulation , we assume matter to be in @xmath0 equilibrium at the temperature of the hydrodynamical model and a fiducial density of @xmath147g@xmath19 . these conditions define the starting points of our post - processing of the composition histories of the considered ejecta elements , for which we take into account electron and positron captures as well as electron neutrino and antineutrino absorption on free neutrons and protons . below a density @xmath22 , where the temperature has decreased to @xmath148k ( or the neutron drip density , if matter below @xmath21 remains cooler than @xmath148k ) the full network calculation is applied instead of nuclear statistical equilibrium . our description of weak interactions of free nucleons includes weak magnetism and recoil corrections according to @xcite and @xcite . avoiding the complications of treating neutrino transport , we simply use exemplary data from publications for prescribing the neutrino luminosities and mean energies needed to compute the neutrino absorption rates in our parametric approach ( `` cases 14 '' ) . this elementary prescription also accounts for the still large uncertainties of the model predictions for the neutrino emission and its directional asymmetries in the generically three - dimensional merger scenario . our modeling strategy follows the spirit of previous , numerous parametric investigations of nucleosynthesis in the neutrino - driven wind of newly formed neutron stars in supernovae ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) and of accretion tori around black holes as remnants of compact object mergers ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in particular , we track in detail the charged - current @xmath0-interactions of neutrinos with free nucleons , which determine the evolution of the electron fraction outside of the neutrino - trapping regime , and employ a full set of trajectories that characterizes the conditions in dynamical ejecta from the merging phase of a representative binary neutron star . these conditions differ from proto - neutron star and accretion - torus winds not only concerning the range of entropies . the ejecta , especially , possess much faster expansion velocities , which for the bulk of the matter can be 2550% of the speed of light , for some fraction of the ejecta even faster , whereas neutrino - driven proto - neutron star winds have typical velocities of 37% of the speed of light ( e.g. , * ? ? ? * ) , neutrino - driven winds from massive neutron stars as relics of ns mergers may achieve expansion velocities up to 810% of the speed of light @xcite , and the main mass of neutrino - driven outflows from bh - accretion tori can reach 1020% of the speed of light @xcite . the correspondingly shorter expansion time scales of dynamical merger ejecta can enable a strong r - process even for moderately low @xmath1 . our calculations confirm recent results of @xcite that solar - like r - process abundances are produced in dynamical ns merger ejecta even when @xmath0-reactions of free nucleons are taken into account and lead to a significant increase of the average electron fraction . in contrast to previous works , where charged - current neutrino - nucleon interactions were ignored , however , also nuclei with mass numbers @xmath10 are ejected in larger amounts . in detail , our results can be summarized by the following points : * ignoring @xmath0-interactions at densities @xmath149 , we confirm our previous results of @xcite that almost exclusively r - nuclei in the regime @xmath150 are produced , in spite of a moderate increase of the average ejecta @xmath1 associated with the assumption of @xmath0-equilibrium at density @xmath21 instead of our previous use of electron fractions of cold neutron star crust matter . * positron and electron - neutrino captures , enhanced by weak magnetism corrections ( which increase the absorption cross section of @xmath2 and reduce that of @xmath3 ) and supported by the @xmath42 effect , lead to a shift of the average ejecta @xmath1 towards higher values when matter expands downwards from an initial density of @xmath18g@xmath19 . * captures of @xmath121 cause a wide spread of the @xmath1 mass distribution , reaching from values of @xmath151 up to 0.40.5 . this reflects the wide range of thermodynamic conditions of the ejecta at @xmath21 with ejecta trajectories that describe cool as well as hot conditions . absorption processes of @xmath2 and @xmath3 take over when the temperature of the expanding matter has fallen to low values where @xmath121 absorptions become ineffective . these ( anti)neutrino captures tend to push @xmath1 towards the asymptotic value @xmath69 for @xmath2-@xmath3 capture equilibrium . the corresponding @xmath1 mass distributions ( at @xmath22 , where the full network calculation was started ) are rather narrow in all cases , with a spread of @xmath1 values of only 0.10.15 . the mean values of the distributions , however , vary between @xmath160.25 and @xmath160.45 , depending on the relative size of the @xmath2 and @xmath3 luminosities . the latter are highly uncertain and variable and are sensitive to the binary - parameter dependent merger dynamics , the nuclear equation of state , and the still not well determined directional asymmetries of the neutrino emission . * in the presence of @xmath0-interactions of free nucleons the temperature also plays an important role for the @xmath1 evolution of the ejecta . it is another aspect of the hydrodynamical merger models that depends on the system properties and the detailed ejection dynamics ( which differ between different ejecta components ) and can be numerically problematic , because resolution and numerical / artificial viscosity can have an influence on the accuracy of the determination of the thermal conditions . lower temperatures reduce the effects of @xmath121 captures but increase the importance of the @xmath42 effect ( thus pushing the average @xmath1 closer to 0.5 ) , whereas higher temperatures enhance @xmath121 captures but nevertheless have little influence on the final @xmath1 distribution at @xmath22 , for which the asymptotic value of @xmath2-@xmath3 capture equilibrium is more relevant than the initially fast @xmath121 captures . * in all investigated model cases , the production of heavy r - process matter with strong second and third abundance peaks and a near - solar distribution in the rare - earth region is a robust outcome . however , in contrast to previous results where weak interactions of free nucleons were ignored , also considerable amounts of matter are synthesized to @xmath10 nuclei . the strength of the production of @xmath152100 material is sensitive to the neutrino - emission properties that determine the ( anti)neutrino absorption . in extreme cases where @xmath1 gets close to 0.5 , significant amounts of iron - group nuclei ( @xmath15360 ) can be ejected . the relative fraction of heavy r - process matter ( from the second peak upward ) in the ejecta therefore varies dramatically between the different investigated cases of neutrino - emission conditions and spans a range from more than 75% down to just 2% . we emphasize that our parametric approach is highly simplified and ignores important neutrino - transport effects like the exact spectral distribution of the neutrino fluxes , the direction dependence of the neutrino emission and corresponding precise radial dilution function , and the time evolution of the neutrino emission relative to the ejection time of the matter . nevertheless , our results , in support of those of @xcite and @xcite , have important consequences for the further exploration of the nucleosynthesis connected to compact binary mergers and the discussion of astrophysical implications . in fact , they demand a major revision of the current picture of r - process production in such events . in view of our results it is obvious that a proper treatment of the neutrino physics , in particular of the neutrino irradiation of the ejected material , is essential for making quantitative predictions of the elemental yields and especially of the total mass of r - process material that is thrown out by the dynamical merger ejecta . approximations like the ones used in our study can be satisfactorily removed only when ultimately detailed , three - dimensional neutrino transport is consistently included in the hydrodynamical simulations . our study suggests that the relative contributions of matter with @xmath152 , @xmath154 and @xmath150 are likely to depend strongly on the binary properties and even the direction of mass ejection , in addition to the equation of state dependence that was found for the electron - fraction distribution in the recent work of @xcite . different from expectations so far , this means that symmetric or nearly symmetric binary ns mergers could exhibit a significantly different ejecta composition than highly asymmetric ns - ns mergers and ns - bh mergers , in which the ns is disrupted before it can be swallowed by the bh . in the last two cases the lower - mass component develops an extended tidal tail , from which considerable amounts of cold , unshocked matter can be centrifugally ejected before the neutrino luminosities rise high and thus before neutrino exposure of these ejecta plays an important role . in such a situation the ejecta will not only be expelled highly anisotropically but will also carry a far dominant fraction of the mass in the form of @xmath150 nuclei as predicted in previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in contrast , in symmetric or nearly symmetric ns mergers the contribution of @xmath155 material will be higher . if the merger remnant collapses to a bh on a millisecond time scale , neutrino exposure of the ejecta may be avoided , but the broad @xmath1 distribution caused by @xmath121 captures will allow for the production of @xmath133 nuclei with a strong second @xmath131 peak . if the merger remnant remains transiently or permanently stable , neutrino exposure of the dynamically expelled matter becomes important , enabling a higher production of @xmath152 species . for extreme cases of a luminous @xmath2 flux , the ejecta might then even be dominated by iron - group nuclei including radioactive nickel isotopes . in particular , however , the exact composition and the relative fraction of high - mass and low - mass species could depend on the direction of the mass ejection . if most of the neutrino flux is emitted to the polar directions due to the rotational deformation of the merger remnant , matter expelled near the equatorial plane will receive less neutrino exposure in addition to its potentially faster escape . this will allow for more neutron - rich conditions close to the equator whereas polar ejecta may contain more proton - rich contributions . future , more complete merger models with neutrino transport will have to clarify these possibilities . since the photon opacity , @xmath156 , of the expanding gaseous ejecta is strongly dependent on the presence of high - opacity , complex ions ( the lanthanides ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the relative contribution of trans - iron elements to the ejecta will have a severe impact on the peak luminosity , @xmath157 , peak time , @xmath158 , and the effective peak temperature , @xmath159 , of the electromagnetic transient that is expected from the radioactively heated , dynamical ejecta cloud ( macronova or kilonova , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the current picture of merger - type and remnant - type dependent redder ( near - infrared ) or bluer emission @xcite and in particular of the envisioned late - time infra - red radiation component from the dynamical ejecta @xcite might require revision or extension in view of our results . finally , it is evident that the strong impact of neutrinos on the neutron - to - proton ratio and the nuclear composition of the dynamic merger ejecta must move neutrino oscillations into the focus of interest . it will be necessary to study the effects of collective neutrino oscillations ( see , e.g. , * ? ? ? * for a review ) with similiar intensity as this subject currently receives in the context of the neutrino emission from newly born neutron stars in supernovae . sg acknowledges financial support from fnrs ( belgium ) . at garching , this research was supported by the max - planck / princeton center for plasma physics ( mppc ) and by the deutsche forschungsgemeinschaft through the cluster of excellence exc 153 `` origin and structure of the universe '' ( http://www.universe-cluster.de ) . ab is a marie curie intra - european fellow within the 7th european community framework programme ( ief 331873 ) . we are also grateful for computing time at the rechenzentrum garching ( rzg ) . arcones a. , janka h .- , scheck l. , 2007 , a&a , 467 , 1227 arnould m. , goriely s. , takahashi k. , 2007 , phys . rep . , 450 , 97 barnes j. , kasen d. , 2013 , apj , 775 , 18 bauswein a. , goriely s. , janka h .- , 2013 , apj , 773 , 78 burrows a. , 2013 , reviews of modern physics , 85 , 245 caballero o.l . , mclaughlin g.c . , surman r. , 2012 , apj , 745 , 170 l. , ott c. d. , burrows a. , rosswog s. , livne e. , 2009 , apj , 690 , 1681 duan h. , fuller g.m . , qian y .- z . , 2010 , arnps 60 , 569 freiburghaus c. , rosswog s. , thielemann f .- k . , 1999 , apj , 525 , l121 goriely s. , bauswein a. , janka h .- , 2011 , apjl , 738 , l32 goriely s. , sida j .- , lematre j .- f . , panebianco s. , dubray n. , hilaire s. , bauswein a. , janka h .- , 2013 , phys . lett . , 111 , 242502 horowitz c.j . , li g. , 1999 , phys . rev . lett . , 82 , 5198 k. , kiuchi k. , kyutoku k. , okawa h. , sekiguchi y .- i . , shibata m. , taniguchi k. , 2013 , phys . d , 87 , 024001 janka h .- , 2012 , ann . science , 62 , 407 just o. , bauswein a. , ardevol pulpillo r. , goriely s. , janka h .- , 2015 , mnras , 448 , 541 kasen d. , badnell n.r . , barnes j. , 2013 , apj , 774 , 25 kasen d. , fernndez r. , metzger b.d . , arxiv:1411.3726 ( astro - ph ) komiya y. , et al . , 2014 , apj , 783 , 132 korobkin o. , rosswog s. , arcones a. , winteler c. , 2012 , mnras , 426 , 1940 kulkarni s.r . , 2005 , arxiv : astro - ph/0510256 li l .- x . , paczyski b. , 1998 , apjl , 507 , l59 matteucci f. , et al . , 2014 , mnras , 438 , 2177 mclaughlin g.c . , fuller g.m . , wilson j.r . , 1999 , apj , 472 , 440 mennekens n. , vanbeveren d. , 2014 , a&a , 564 , a134 metzger b.d . , martinez - pinedo g. , darbha s. , et al . , 2010 , mnras , 406 , 2650 metzger b.d . , fernndez r. , 2014 , mnras , 441 , 3444 meyer b.s . , mclaughlin g.c . , fuller g.m.,1998 , phys . c , 58 , 3696 nishimura n. , takiwaki t. , thielemann f .- k . , 2015 , arxiv:1501.06567 perego a. , rosswog s. , cabezn r. m. , korobkin o. , kppeli r. , arcones a. , liebendrfer m. , 2014 , mnras , 443 , 3134 pllumbi , e. , et al . , 2014 , arxiv1406.2596 roberts l.f . , kasen d. , lee w.h . , ramirez - ruiz e. , 2011 , apjl , 736 , l21 roederer i.u . , 2011 , apjl , 732 , l17 roederer i.u . , et al . , 2012 , apjs , 203 , 27 s. , liebendrfer m. , 2003 , mnras , 342 , 673 rosswog s. , liebendrfer m. , thielemann f .- k . , davies m.b . , benz w. , piran t. , 1999 , a&a , 341 , 499 m. , janka h .- t . , 1999 , a&a , 344 , 573 ruffert m. , janka h .- th . , 2001 , a&a , 380 , 544 m. , janka h .- t . , schaefer g. , 1996 , a&a , 311 , 532 sekiguchi y. , kiuchi k. , kyutoku k. , shibata m. , 2015 , arxiv:1502.06660 s. , cooke r. , ramirez - ruiz e. , madau p. , mayer l. , guedes j. , 2014 , arxiv:1407.3796 surman r. , mclaughlin , g.c . , 2005 , apj , 618 , 397 surman r. , mclaughlin , g.c . , ruffert m. , janka h .- , hix w.r . , 2008 , apjl , 679 , l117 takahashi k. , el eid m.f . , hillebrandt w. , 1978 , a&a , 67 , 185 tanaka m. , hotokezaka k. , 2013 , apj , 775 , 113 tanaka m. , hotokezaka k. , kyutoku k. , et al . , 2014 , apj , 780 , 31 typel s. , rpke g. , klhn t. , blaschke d. , wolter h.h . , 2010 , phys . c , 81 , 015803 f. , quataert e. , hopkins p. f. , keres d. , faucher - giguere c .- a . , 2015 , mnras , 447 , 140 vangioni e. , goriely s. , daigne f. , franois p. , belczynski k. , arxiv : astro - ph/1409.2462 wanajo s. , janka h .- th . , & mller , b. , 2011 , apjl , 726 , l15 wanajo s. , janka h .- th . , 2012 , apj , 746 , 180 wanajo s. , et al . , 2014 , apj , 789 , l39 wehmeyer b. , m. pignatari m. , thielemann f .- k . , 2015 , arxiv : 1501.07749 winteler c. , kppeli r. , perego a. , et al . , 2012 , apjl , 750 , l22	we investigate @xmath0-interactions of free nucleons and their impact on the electron fraction ( @xmath1 ) and r - process nucleosynthesis in ejecta characteristic of binary neutron star mergers ( bnsms ) . for that we employ trajectories from a relativistic bnsm model to represent the density - temperature evolutions in our parametric study . in the high - density environment , positron captures decrease the neutron richness at the high temperatures predicted by the hydrodynamic simulation . circumventing the complexities of modelling three - dimensional neutrino transport , ( anti)neutrino captures are parameterized in terms of prescribed neutrino luminosities and mean energies , guided by published results and assumed as constant in time . depending sensitively on the adopted @xmath2-@xmath3 luminosity ratio , neutrino processes increase @xmath1 to values between 0.25 and 0.40 , still allowing for a successful r - process compatible with the observed solar abundance distribution and a significant fraction of the ejecta consisting of r - process nuclei . if the @xmath2 luminosities and mean energies are relatively large compared to the @xmath3 properties , the mean @xmath1 might reach values @xmath40.40 so that neutrino captures seriously compromise the success of the r - process . in this case , the r - abundances remain compatible with the solar distribution , but the total amount of ejected r - material is reduced to a few percent , because the production of iron - peak elements is favored . proper neutrino physics , in particular also neutrino absorption , have to be included in bnsm simulations before final conclusions can be drawn concerning r - processing in this environment and concerning observational consequences like kilonovae , whose peak brightness and color temperature are sensitive to the composition - dependent opacity of the ejecta . [ firstpage ] nuclear reactions , nucleosynthesis , abundances neutrinos stars : neutron dense matter
electricity can not be stored efficiently in large quantities , therefore it is critical to ensure that the amount generated at a given time is sufficient to meet the load plus the losses while not exceeding this amount significantly . predictive methods for accurately forecasting the demand of electricity have thus become important tools that guide planning and operation of utility companies . while electric load forecasting is a well - established , several decades old research area in engineering , new modeling problems keep appearing as technological and legislative transformations affect the power industry . with the advent of smart grids and meters , larger and richer sources of data are becoming available , making it possible to build more sophisticated models that enable more accurate billing of electricity and dynamic pricing . a variety of tools from time series analysis , statistics , and more recently machine learning , have been employed for electricity load forecasting . for an overview on the vast body of available literature on the subject , we refer the reader to the recent book by @xcite . classical techniques include linear and non - linear regression models estimated by means of variants of least squares fitting , and various types of armax models expressing the forecast as a function of previously observed values of the load and possibly other weather or social variables . techniques inspired by artificial intelligence research such as expert systems , fuzzy logic , and neural networks have also been applied to load forecasting . in particular , black - box models based on neural networks have been extensively analyzed , see the influential review by @xcite . in recent years , generalized additive models ( gam ) @xcite have established themselves as state of the art tools for electricity load forecasting @xcite , due to the existence of efficient and scalable training algorithms and the interpretability of the model , which allows to clearly visualize the effect of individual variables on the load by means of simple longitudinal plots . meanwhile , kernel methods have been employed with great success in the last decade . already back in 2001 , a kernel - based support vector regression ( svr ) approach was employed to win a competition on electricity load forecasting @xcite organized by eunite ( european network on intelligent technologies for smart adaptive systems ) . later on , various types of kernel - based regularization methods and support vector machines have been applied to predict the demand of electricity , see for instance @xcite . most research articles on electricity load forecasting focus on predicting a single time series representing the electricity load aggregated over a large number of nodes of the electricity network . due to aggregation , such time series exhibit high regularity and are therefore significantly easier to forecast than load profiles at lower levels of the network . nevertheless , making forecasts of the loads at lower levels is becoming increasingly feasible due to the availability of rich smart meter datasets , therefore the problem is attracting considerable interest in the industry . forecasting electricity demand at low levels of the network ( such as the demand of an individual household ) presents several challenges . first of all , it involves analyzing a much larger number of time series , calling for scalable techniques that can handle a very large amount of measurements . in addition , demand profiles at lower levels of the electricity network are much less regular and thus harder to predict . to tackle these challenges , recent works have investigated the use of clustering techniques for automatically aggregating multiple load time series , reporting improved predictive performance at aggregated level @xcite . in this paper , we study the problem of electricity load forecasting at low network level , and we suggest to solve it by means of kernel - based multi - task learning techniques that can discover and take advantage of the relationships between multiple profiles . kernel based multi - task learning has been studied in a variety of papers @xcite while , in recent years , the problem of learning and exploiting relationships between multiple tasks is a topic that is attracting considerable attention in the machine learning literature @xcite . herein , we develop and compare a variety of kernel - based models for long - term electricity demand forecasting in multiple nodes , with the goal of identifying the best way to capture the complex seasonal effects that characterize such demand patterns . we design kernels specifically tailored to capture the seasonal effects present in electricity load data . by doing so , we expose the performance limits of the very popular additive models , showing that they are often outperformed by multiplicative kernel models . we formulate the problem of forecasting the demand in multiple nodes of the network as a multi - task learning problem , illustrating the usefulness of jointly learning and exploiting similarities between multiple load profiles . finally , we show how recently developed multi - task learning techniques can be used to gain insights and interpretability on real demand data , while achieving state of the art predictive performance . our experimental analysis is based on data provided by the irish commission for energy regulation ( cer ) . electric load forecasting aims at predicting the future load in one or multiple nodes of an electricity network . depending on how far ahead in time the forecast is required , the corresponding estimation problem exhibits different characteristics , and influence decisions of significantly different nature . it is therefore common to classify forecasting problems in three categories : short - range forecasting ( several minutes up to one week ahead ) , medium - range forecasting ( up to 10 years ahead ) , and long - range forecasting ( as far as several decades ahead ) , see @xcite for a more comprehensive discussion . forecasting models are built starting from datasets containing one or multiple time series , each of them representing the load measurement at a specific location and level in the network , ranging from highly aggregated loads in the transmission network down to the distribution network and to the loads of individual users . missing measurements and different sampling rates contribute to make these data noisy and challenging to analyze . moreover , defective meters at a low level in the network are hard to detect , and faulty meters can report wrong measurements before being replaced . being mostly driven by human activity , a variety of temporal patterns can be observed in the load data . the top panel of fig . [ fig : patterns ] shows a typical profile for aggregated electricity load over several years ( data source : french rseau de transport delectricit ) , from which a clear yearly seasonal pattern can be observed , with higher demand in winter and lower demand in the summer . a closer look at this data also reveals typical weekly ( fig . [ fig : patterns ] , middle panel ) and daily ( fig . [ fig : patterns ] , bottom panel ) profiles . correctly capturing these seasonal patterns is an crucial aspect of the problem , which can be dealt with by properly extracting and utilizing temporal and calendar features . the type of day of the week can be also taken into account : the bottom panel of fig . [ fig : patterns ] shows a specific week where all days have a similar profile but a difference between week days and weekend can be clearly noticed . forecasting is particularly challenging on public holidays , and different public holidays may exhibit significantly different load profiles . [ fig : holidays ] illustrates the difficulty of fitting and forecasting the demand in correspondence of public holidays and special events : without including specific information in the model , the prediction error can be particularly high in corrispondence with such events . for more details about cer data.,scaledwidth=100.0% ] a variety of additional features can be typically extracted from the data or obtained from other sources and utilized to forecast the electricity demand . for instance , the electricity consumption is affected by weather conditions ( particularly due to heating and air - conditioning ) , therefore variables such as temperature , humidity and irradiance are often taken into account by forecasting models . economic indicators such as gross domestic product can be used to model trends in long - range scenarios . finally , short term forecasting models are typically based on time series techniques , where auto - regressive lagged values of the load itself are incorporated in the model and used to track short - range trends and deviations from stationarity . in summary , typical forecasts of the electricity demand may depend on a variety of features that include time and calendar variables , weather and economic conditions , previously observed values of the load , and information about the node of the network where the forecast is required . a general model that takes into account the previously discussed features takes the form @xmath0 where the dependent variables are the following : * @xmath1 is the time of day expressed in hours , * @xmath2 is the day of the year , * @xmath3 is the type of day , e.g. monday to sunday , weekday / weekend , holiday , * @xmath4 is a real vector containing lagged values of the measured electric load , * @xmath5 is a real vector containing measurements of lagged values of exogenous variables other than the load ( such as temperature ) , * @xmath6 is the node i d in the electricity network , e.g. meter / customer / zone i d , * @xmath7 is a vector of features describing the node type , e.g. customer / zone type . in section [ sec : lfbymtl ] , we analyze one of the many possible multi - task learning problems that naturally appear within this framework , namely the problem of simultaneously predicting the load in each node of the network . this amount to disaggregate the overall dataset over the multiple smart meters and treat each node @xmath6 of the network as a different learning task . in the next section , we briefly recall the standard setup of multi - task regression and review the techniques that will be employed in section [ sec : lfbymtl ] to solve the problem sketched above . in the following , we focus on multi - variate ( multi - task ) regression problems where the goal is to learn multiple functions @xmath8 from multiple datasets of pairs @xmath9 . here , @xmath10 is a set of input features , @xmath11 denotes the number of tasks , and @xmath12 the number of examples for the @xmath13-th task . letting @xmath14 denote the vector - valued function with components @xmath15 , we are going to search @xmath16 by minimizing the following regularization functional @xmath17 where @xmath18 is a regularization parameter , and @xmath19 is a reproducing kernel hilbert space ( rkhs ) of vector - valued functions with ( matrix - valued ) kernel @xmath20 here , @xmath21 is a positive semidefinite kernel called _ input kernel _ , and the square matrix @xmath22 is the _ output kernel _ matrix whose entries @xmath23 express the similarity between the tasks ( output components ) @xmath13 and @xmath6 . in view of the _ representer theorem _ , there exist functions @xmath24 minimizing @xmath25 in the form : @xmath26 we refer to @xcite for more details about rkhs of vector - valued functions and the corresponding representer theorem . expression shows that inter - task transfer is possible only when off - diagonal elements of the output kernel matrix are different from zero . indeed , by choosing @xmath28 equal to the identity matrix all the tasks are learned independently by solving a standard kernel regularized least squares problem @xmath29 where @xmath30 is the rkhs of scalar functions with kernel @xmath31 . this single - task baseline is referred to as _ independent _ kernel ridge regression . in this subsection , we review a kernel - based multi - task regression approach called low - rank output kernel learning ( okl ) , recently developed in @xcite . in such approach , the functions @xmath15 and the output kernel @xmath28 are jointly optimized by solving the following problem @xmath32 where @xmath33 is the cone of positive semidefinite matrices with rank less than or equal to @xmath34 . instead of imposing a low - rank constraint or regularizing the trace of the output kernel , other type of regularizers could be tried , see e.g. @xcite . the low - rank approach has the advantage of allowing us to tightly control the memory required to store the models . the representer theorem still applies to the inner minimization problem of . by plugging the expression into , one obtains a functional that is convex quadratic with respect to both the coefficients @xmath35 and @xmath28 . although the resulting problem is not jointly convex , the alternating minimization procedure described in @xcite can be applied to obtain a minimizer . an important aspect of the method is that , by selecting the rank parameter @xmath34 , is it possible to control the overall number of parameters of the model , as well as the memory requirements and the computation time to obtain a solution . more specifically , letting @xmath36 , one can show that the solution can be rewritten as @xmath37 where @xmath38 , @xmath39 , @xmath40 , and the coefficients @xmath41 form a low - rank factor of @xmath28 . it is therefore sufficient to store and optimize @xmath42 parameters , which can be much smaller than @xmath43 . in this section , we focus on predicting the demand of electricity in multiple nodes of an electricity network , a multi - task learning problem where each task corresponds to one of the measured smart meters in the network . .number of meters and sparsity for each customer group in the irish cer dataset [ cols="<,>,>",options="header " , ] to analyze our long - term forecasting approach , we adopt data provided by the the irish commission for energy regulation ( cer ) , containing electric load measurements from 6435 smart meters , half - hourly sampled from july 14 , 2009 to december 31 , 2010 ( 536 days ) . these meters include residential customers and small - to - medium industrial sites . we consider a mid - term test scenario where the goal is to forecast the load in multiple nodes over a time horizon of 171 days , using one year of measurements to build the model . due to the long forecasting horizon , dynamic features are not available and are therefore dropped from the general model in eq . . such model does not rely on recent measurements of the load , therefore it is able to make predictions over an arbitrarily long horizon . the load forecast for the @xmath6-th node is thus simply given by @xmath44 , where @xmath45 are the multiple functions to be learned , taking into account time and calendar features . from the original cer dataset , several pre - processing steps were performed . the day of the year and time of the day were extracted from the five - digit timestamps . in this dataset , the time of day is an non - zero integer indexing the number of half - hours , and therefore it should be normally in the set @xmath46 . two meters containing time of days higher than @xmath47 half - hours were discarded , as it was unclear how to interpret these measurements . the dataset also contains days with 46 and 50 measurements and time of days up to 50 . these inconsistencies are caused by the start and the end of daylight saving time ( dst ) and are easily fixable . when dst starts in ireland , the 1 am to 2 am hour get skipped , and half - hourly time of day indices should be @xmath48 . when dst starts in ireland , the 1 am to 2 am hour `` happens twice '' , and half - hourly time of day indices should be @xmath49 , instead of @xmath50 as found in the dataset . we then downsampled each time - series from half - hourly sampling to 3-hour sampling , by averaging available measurements for each time slot of 3 hours ( [ 12 am , 3 am ) , [ 3 am , 6 am ) , etc ) and a total of 8 measurements per day . our final dataset contains @xmath51 smart meters sampled over @xmath52 time slots . characteristics of such pre - processed dataset are summarized in tab . [ tab : groups ] . one year ( @xmath53 downsampled observations ) was used for training and validation , and the remaining @xmath54 observations were used for testing . in order to perform tuning of the regularization parameter , we extracted a validation set containing a subset of the original non - test data , obtained by randomly choosing @xmath55 time samples , equal for all the meters . forecasting performance can be evaluated for each time slot @xmath56 and any arbitrary group of meters @xmath57 . for this purpose , let @xmath58 denote the subset of @xmath57 for which measurements are available in the @xmath59-th time slot . we define two different metrics , namely the aggregated mean absolute percentage error ( @xmath60 ) and the normalized mean absolute error ( @xmath61 ) : @xmath62 @xmath63 measures the relative absolute percentage error incurred when forecasting the aggregated demand using the sum of the forecasts . on the other hand , @xmath64 is the sum of the forecasting errors over individual tasks , relative to the naive baseline of predicting @xmath65 for all @xmath66 . since the demand values @xmath67 are always non - negative , the two metrics are undefined only for those groups on meters for which the cumulative demand in the @xmath59-th time slot is identically zero , or no measurements are available for any of the meters . we compute the average and standard deviation of these two metrics over all the time slots in the test period . one advantage of the long - term forecasting approach is that it allows to naturally incorporate and handle time series with missing observations , without resorting to inputing techniques or discarding data . in the following , we analyze a variety of kernel based models to solve this multi - task regression problem . first of all we introduce kernels based on the time / calendar features of model , @xmath68 where @xmath69 is a change of variable that yields @xmath70-periodic kernels over the square @xmath71 ^ 2 $ ] . by observing that the fourier transform of @xmath72 is non - negative , it can be easily shown that periodized kernels such as @xmath73 and @xmath74 are positive semidefinite , see @xcite . in our experiment , @xmath75 and @xmath76 were respectively set to 4 hours and 120 days . in order to define @xmath77 , we combine these three kernels to define a variety of models : * additive models @xmath78 * semi - additive models @xmath79 * multiplicative models @xmath80 first of all , we have trained independent kernel ridge regression models ( see sec . [ sec : krr ] ) for each measured smart meter using all the kernels from to . we compare these models against a multi - task learning approach that simultaneously performs estimates for all the meters , and also allows us to exploit the available meter grouping information in the dataset . specifically , we have trained two separate multi - task output kernel learning ( okl ) models ( see sec . [ sec : okl ] ) : the first is trained over all the residential meters ( the union of meters labeled `` residential '' and `` others '' ) , and the second over industrial ( sme ) meters . the maximum rank constraint for the first model was set to @xmath81 to obtain a compact model that fits into memory , while the okl model for sme meters was trained with full rank @xmath82 . we refer the reader to @xcite for a discussion on the effect of this parameter . both okl models utilize the multiplicative input kernel , as it proves to be the most effective at capturing the seasonal effects . figure [ fig : predictions ] illustrates the challenges of long - term forecasting at low network level versus forecasting aggregated demands . we analyze the measured load and the corresponding forecast over a window of 5 weeks within the test period . in the top panel , one can see the aggregated load and the corresponding forecast obtained as the sum of all disaggregated forecasts obtained using model . the kernel - based forecasts are rather accurate overall , only slighly estimating the total load during the christmas week , a particularly problematic period to predict . in the middle panel , the measured load for a single sme meter is compared with the corresponding forecast . the varying demand profiles of different days of the week are captured rather well by the model . again , there is a larger error over the christmas week , caused by a sudden drop of the demand to a low baseline value ( probably due to interruption of business activities followed by a slow resumption in the subsequent days ) . this leads the models to over - estimate the load , though the model learned with a multi - task approach is less affected . finally , the bottom panel shows the electricity demand of a residential customer , characterized by rapid variations with sharp consumption peaks and irregular patterns , that make the forecast even more difficult . @llrrrrr@ & & & & + & & mean & std & & mean & std + additive models & @xmath83 & 0.4829 & 0.0834 & & 8.0384 & 7.2644 + & @xmath84 & 0.4897 & 0.0846 & & 8.0435 & 6.9948 + + semi - additive models & @xmath85 & 0.4546 & 0.0682 & & 7.1413 & 6.6078 + & @xmath86 & 0.4507 & 0.0649 & & 7.2921 & 6.5868 + + multiplicative models & @xmath87 & 0.4663 & 0.0771 & & 5.5284 & 5.4252 + & @xmath88 & 0.4237 & 0.0528 & & 4.2917 & 4.3055 + + multi - task okl & @xmath88 & * 0.4226 * & * 0.0487 * & & * 4.0222 * & * 4.0541 * + @llrrrrr@ & & & & + & & mean & std & & mean & std + additive models & @xmath83 & 0.5114 & 0.0929 & & 13.1083 & 9.9519 + & @xmath84 & 0.5157 & 0.0941 & & 13.0127 & 10.2067 + + semi - additive models & @xmath85 & 0.5005 & 0.0900 & & 10.7139 & 9.9966 + & @xmath86 & 0.4977 & 0.0846 & & 10.8144 & 9.6131 + + multiplicative models & @xmath87 & 0.5058 & 0.0844 & & 8.4289 & 7.1833 + & @xmath88 & 0.4776 & 0.0721 & & 5.2692 & 5.6305 + + multi - task okl & @xmath88 & * 0.4711 * & * 0.0716 * & & * 4.9166 * & * 5.2588 * + @llrrrrr@ & & & & + & & mean & std & & mean & std + additive models & @xmath83 & 0.4517 & 0.2204 & & 15.0748 & 23.1005 + & @xmath84 & 0.4590 & 0.1769 & & 16.6668 & 17.0665 + + semi - additive models & @xmath85 & 0.3704 & 0.1238 & & 8.1880 & 10.8081 + & @xmath86 & 0.3646 & 0.1305 & & 8.1422 & 11.7269 + + multiplicative models & @xmath87 & 0.4006 & 0.1923 & & 12.5359 & 20.4140 + & @xmath88 & * 0.3127 * & 0.1105 & & 5.9289 & 10.3842 + + multi - task okl & @xmath88 & 0.3194 & * 0.0725 * & & * 5.2940 * & * 5.8227 * + @llrrrrr@ & & & & + & & mean & std & & mean & std + additive models & @xmath83 & 0.4948 & 0.0779 & & 7.4672 & 6.9361 + & @xmath84 & 0.5001 & 0.0804 & & 7.4141 & 6.4660 + + semi - additive models & @xmath85 & 0.4659 & 0.0638 & & 6.4749 & 5.9244 + & @xmath86 & 0.4601 & 0.0605 & & 6.6273 & 5.9189 + + multiplicative models & @xmath87 & 0.4801 & 0.0710 & & 5.4504 & 5.5997 + & @xmath88 & 0.4370 & 0.0517 & & 4.0279 & 4.2522 + + multi - task okl & @xmath88 & * 0.4361 * & 0.0490 & & * 3.7450 * & * 3.9503 * + table [ tab : cerresults_overall ] reports the performance of all methods over the full set of 6433 smart meters , while tables [ tab : cerresults_res ] , [ tab : cerresults_sme ] and [ tab : cerresults_others ] report disaggregated performance measures over each group from tab . [ tab : groups ] . we start by analyzing the performance of the additive models , which are probably the most widely adopted in the literature . by comparing the performance of models and , we can observe that adding a constant bias specific to the type of the day of week ( kernel @xmath89 ) does not necessarily improve the accuracy of the model . the overall @xmath61 and @xmath60 are in fact higher for model , see tab . [ tab : cerresults_overall ] . semi - additive models where the type of day of the week is utilized to switch between different profiles yields a significant improvement in performance . the two semi - additive models and achieve similar performance over the groups residential and others ( see respectively tab . [ tab : cerresults_res ] and tab . [ tab : cerresults_others ] ) . however , for the sme customer group , model is better in terms of both @xmath61 and @xmath60 ( see tab . [ tab : cerresults_sme ] ) . in previous works such as @xcite , semi - additive models of the form have been proposed to switch between different daily patterns , depending on the type of day . interestingly , our results show that in certain situations , such as when modeling industrial customers , it is even better to switch the overall sum of the daily pattern and the yearly pattern . we took a step even further by utilizing fully multiplicative models and . the multiplicative model pools over different days of the week , while learns independent models for each day . while the former is not always better than the semi - additive models , the latter significantly outperforms them . we can conclude that a multiplicative kernel structure is the best at capturing yearly , weekly and daily seasonal effect , both overall and for each customer group . such conclusion is aligned with recent results presented in @xcite , where tensor product basis functions were utilized to capture weekly and yearly seasonalities in the simpler context of load forecasting for a single highly aggregated time series . a further performance improvement can be obtained by utilizing a multi - task learning approach , where correlation between electricity demand behavior of multiple customers is learned and exploited . as the tables show , the multi - task okl approach provides the lowest mean @xmath61 over all the meters performance , residential and others , and second lowest mean @xmath61 for sme ( only @xmath90 higher than the lowest for this group ) . the multi - task learning approach also provides the lowest mean aggregated @xmath60 , overall and for each customer groups . finally , the multi - task approach is more robust , as the temporal standard deviation is the lowest for both @xmath61 and @xmath60 . again , such robustness can be observed overall the customers as well as for each customer group . in particular , it is worth mentioning a @xmath91 improvement of the standard deviation of the @xmath60 for sme meters , compared to the best single task model that uses the multiplicative model . in addition to improving forecasting accuracy , the low - rank multi - task learning model is significantly more compact in terms of number of parameters . for all the single - task methods ( with additive , semi - additive and multiplicative kernels ) , the number of parameters is equal to the overall number of training observations @xmath92 . in our experiment , this amounts to about 13 million parameters ( precisely @xmath93 parameters ) . the low - rank output kernel learning method models each prediction function @xmath24 as a linear combination of @xmath34 latent functions , shared by all tasks ( see sec . [ sec : okl ] ) . these @xmath34 functions @xmath94 can be seen as typical load profiles . as a consequence , only @xmath95 parameters are required to learn the prediction functions for all smart meters . in our experiment , this gave a total of about 3 million parameters ( precisely @xmath96 ) thus producing a model that is about @xmath97 times more compact , in addition to being more accurate . our analysis shows that kernel - based multi - task learning is effective for the resolution of electric load forecasting problems . focusing on the challenging problem of forecasting the electric load of individual customers , we designed kernels that take into account relevant multiple seasonality patterns . we demonstrated the clear benefits of multiplicative kernel models over additive or semi - additive models . our results suggest a new modeling direction , as opposed to the ( generalized ) additive models , widely employed in the energy community . we illustrated further performance gain made possible by using a multi - task learning approach over a large number of single - tasks baselines . while recent studies reported @xmath60 around @xmath98 for the short term forecasting of an aggregated signal of a few thousands of smart meters e.g. @xcite , our method achieves a @xmath60 of @xmath99 on a long term forecasting scenario , which is a much harder problem as auto - regressive terms are not available . the ideas and results presented in this paper open a wide range of considerations . first of all , they suggest that electricity demand data can be used as natural test benchmarks for multi - task learning methods . in addition , these problems motivate developing new techniques that allow to incorporate more complex task relationships structures taking into account , for instance , topological and physical constraints from the electricity network . the development of online methods that can automatically discover relationships between multiple tasks seems to be particularly important for short - term load forecasting scenarios . finally , combining online multi - task learning methods with topological network constraints would allow to start tackling very complex scenarios such as forecasting on a full electricity network with dynamic reconfigurations . c. alzate and m. sinn . improved electricity load forecasting via kernel spectral clustering of smart meters . in _ 2013 ieee 13th international conference on data mining ( icdm ) _ , pages 943948 . ieee , 2013 . c. alzate , m. espinoza , b. de moor , and j. ak suykens . identifying customer profiles in power load time series using spectral clustering . in _ artificial neural networks icann 2009 _ , pages 315324 . springer , 2009 . a. ba , m. sinn , y. goude , and p. pompey . adaptive learning of smoothing functions : application to electricity load forecasting . in _ advances in neural information processing systems 25 ( nips 2012 ) _ , pages 25192527 , 2012 . f. dinuzzo , c. s. ong , p. gehler , and g. pillonetto . learning output kernels with block coordinate descent . in _ proceedings of the 28th annual international conference on machine learning _ , bellevue , wa , usa , 2011 . f. dinuzzo , c.s . ong , and k. fukumizu . output kernel learning methods . in m ; argyriou a suykens , j ; signoretto , editor , _ regularization , optimization , kernels , and support vector machines,_. crc press , 2014 . e. elattar , j. goulermas , and h. wu . electric load forecasting based on locally weighted support vector regression . _ systems , man , and cybernetics , part c : applications and reviews , ieee transactions on _ , 400 ( 4):0 438447 , 2010 . s. humeau , t. k. wijaya , m. vasirani , and k. aberer . electricity load forecasting for residential customers : exploiting aggregation and correlation between households . in _ sustainable internet and ict for sustainability ( sustainit ) , 2013 _ , pages 16 . ieee , 2013 . a. saha , p. rai , h. daum iii , and s. venkatasubramanian . online learning of multiple tasks and their relationships . in _ international conference on artificial intelligence and statistics ( aistats ) _ , ft . lauderdale , florida , 2011 . y. zhang and d .- y . yeung . a convex formulation for learning task relationships in multi - task learning . in _ proceedings of the 26th conference on uncertainty in artificial intelligence ( uai ) _ , pages 733442 , catalina island , ca , usa , 2010 .	we explore the application of kernel - based multi - task learning techniques to forecast the demand of electricity in multiple nodes of a distribution network . we show that recently developed output kernel learning techniques are particularly well suited to solve this problem , as they allow to flexibly model the complex seasonal effects that characterize electricity demand data , while learning and exploiting correlations between multiple demand profiles . we also demonstrate that kernels with a multiplicative structure yield superior predictive performance with respect to the widely adopted ( generalized ) additive models . our study is based on residential and industrial smart meter data provided by the irish commission for energy regulation ( cer ) .
it is known that a connection exists between the thermodynamics and gravity . this connection was first investigated in the works of bardeen , carter and hawking @xcite . in this work , it was suggested that an analogy exists between the laws of thermodynamics and gravitational physics . however , the existence of this connection was only established after the discovery of hawking radiation @xcite . the hawking radiation comes from black holes , as black holes behave as hot bodies with a temperature proportional to the surface gravity of the black hole . the black holes also have an entropy associated with them , and this entropy is proportional to the area of the horizon @xcite . thus , the black hole physics establishes a connection between the laws of thermodynamics and gravitational physics . motivated by the relation between thermodynamics and gravity , it has been proposed that gravity is actually only an emergent thermodynamic phenomena . this formalism in which gravity is described as a emergent phenomena is called the jacobson formalism @xcite . in this formalism , the einstein field equation can be derived from the first law of thermodynamics @xmath6 . thus , in this formalism einstein field equation is viewed as an equation of state for the spacetime . it is possible to derive the friedmann equations from the clausius relation @xcite . in this analysis , the entropy was taken to be proportional to the apparent horizon of frw universe . thus , if @xmath7 is the area of the apparent horizon with temperature @xmath8 , then the entropy @xmath9 can be written as @xmath10 it may be noted that even though the entropy is proportional to the area of the apparent horizon , the temperature is not proportional to the surface gravity @xmath11 . in fact , it is possible to express the surface gravity @xmath11 as @xmath12 it has been possible to use the misner - sharp energy relation to express the dynamical friedman equation in the form of the first law of thermodynamics @xcite . however , for consistency of the results an extra work term had to be added to the original expression for the first law of thermodynamics . it has been demonstrated that the work density @xmath13 can be expressed in terms of the energy density of the universe @xmath14 and the pressure of matter in the universe @xmath15 as @xmath16 using this work function , the dynamical friedman equations could be expressed as @xmath17 it may be noted that the hawking radiation is connected to the uncertainty principle in quantum mechanics @xcite . this connection is established by modeling the black hole as a @xmath18-dimensional sphere whose radius is related to the schwarzschild radius . the emitted particles from the black hole obey the uncertainty principle , as the hawking radiation is a purely quantum mechanical process . this fact can be used to derive the thermodynamical properties of the black hole @xcite . it is known that the temperature of a black hole increases as the size of the black hole decreases . this temperature tends to infinity as the mass size of the black hole approaches zero , and this in turn leads to a catastrophic evaporation of the black hole . this is because in string theory it is not possible to probe the spacetime below string length scale , and so string theory comes equipped with a minimum measurable length scale which is the string length scale . furthermore , the energy required to probe the spacetime below planck length is more than the energy required to form a mini black hole in that region of spacetime . thus , there are strong indications that there exists a minimum measurable length scale for the spacetime @xcite . in fact , the existence of a minimum measurable length scale is universal feature of almost all approaches to quantum gravity @xcite . the existence of the minimum measurable length scale is not consistent with the usual uncertainty principle . this is because according to the usual uncertainty principle , it is possible to measure the length to an arbitrary accuracy , if the momentum is not measured . however , it is possible to generalize the usual uncertainty principal to a generalized uncertainty principle ( gup ) , such that this new uncertainty principle is consistent with the existence of a minimum measurable length . as the hawking radiation depends on the uncertainty principle , the modification of the usual uncertainty principle to the generalized uncertainty principle , also modifies the thermodynamics of the black holes . it has been demonstrated that there is no catastrophic evaporation of the black hole in this modified thermodynamics @xcite . this is because a black hole remnant forms in the gup deformed thermodynamics . thus , the gup prevents the black hole from evaporating completely , just like the standard uncertainty principle prevents the hydrogen atom from collapsing . in fact , the usual relation between the entropy and the area of a black hole gets significantly modified due to the gup @xcite . as the friedmann equations are viewed as thermodynamical relations in the jacobson formalism , the modification of the thermodynamics by the gup will also modify the friedmann equations in the jacobson formalism . such modification to the friedmann equations has been recently studied @xcite . in this analysis , the relation between the area of the apparent horizon and entropy was used for calculating the modification to the friedman equations from the gup . it has been demonstrated that in these gup modified friedman equations , there exists a maximum energy density close to planck density . thus , the maximum energy density is reached in a finite time . this means that in this model there is an absence of the cosmological evolution beyond this point from a spacetime prospective . in fact , it was demonstrated that the maximum energy density and a general nonsingular evolution is independent of the equation of state and the spacial curvature @xmath19 . this state of maximum energy density was reached in a finite time @xcite . so , in this model , the big big bang singularity is not accessible , and the energy density in the spacetime can not be extended beyond planck density . this bound on the energy density of spacetime occurs because of the gup . as the friedmann equations get modified due to the gup , we expect the inflationary cosmology also to get modified due to the gup in the thermodynamical approach . so , in this work we will study the inflationary cosmology using the jacobian formalism . we will also study the effect of gup deformed thermodynamics on the inflationary cosmology it is possible to study the inflationary cosmology by using a potential to derive the properties of the inflationary cosmology . thus , a specific form of the potential is chosen , and the model of th inflationary cosmology depends on the details of the potential chosen @xcite . however , it is also possible to use the hamilton - jacobi formalism to model the inflationary universe @xcite . in this formalism , the hubble parameter is expressed in term of a scalar field . it is possible to deducing a form of the potential in the the hamilton - jacobi formalism . furthermore , it is also possible to obtain an exact solution for the scalar field in this approach . thus , we will use the hamilton - jacobi formalism in this work . the paper is organized as follows , in sec . [ ah ] , we review the derivation of modified friedmann equations due to gup @xcite in the context of the first law of thermodynamics . in sec . [ raychud ] , we discuss direct implications of the modified friedmann equations on inflation . finally , in sec . [ t ] , we summarize our results . we also discuss some possible extensions of this present work in this last section . in this section , we briefly review the derivation of friedman equations from the first law of thermodynamics relation with the apparent horizon of frw universe with assuming that the entropy is proportional to the area of the apparent horizon @xcite . we then review briefly the calculations of the impact of gup on friedman equations through this thermodynamic approach @xcite . the @xmath20-dimensional frw universe is represented by the following spacetime metric : @xmath21 where @xmath22 , @xmath23 is the metric of @xmath24-dimensional sphere , @xmath25 and the spatial curvature constant @xmath19 takes the values @xmath26 for a flat , closed and open universe , respectively . the apparent horizon is calculated from the relation @xmath27 . so , its radius can be written as @xcite : @xmath28 where @xmath29 is the hubble parameter . we assume that the frw universe is occupied with a perfect fluid . now we can express the energy - momentum tensor @xmath30 where @xmath31 is the four velocity of the fluid . the energy conservation law ( @xmath32 ) leads to the continuity equation @xmath33 motivated by the arguments of @xcite , the work density @xmath13 can be defined as follows @xmath34 where @xmath35 is the projection of the energy - momentum tensor @xmath36 in the normal direction . for our case , the work function is simplified to be @xmath37 now , we briefly review the derivation of einstein equations in the thermodynamic approach @xcite . thus , we will start by writing the first law of thermodynamics as @xmath38 utilizing eq . ( [ 1st ] ) , the infinitesimal change in the total energy ( @xmath39 ) could be calculated as following @xmath40 now the following result can be obtained , @xmath41 here we have used the continuity equation ( [ continuity ] ) and eq . ( [ denergy ] ) . for the work term @xmath42 , we obtain the following expression @xmath43 for the term @xmath44 , we should use definition of hawking temperature of eq . ( [ hawking ] ) as well as the entropy - area law . the hawking temperature can be obtained from the surface gravity as @xmath45 here the surface gravity is calculated as @xmath46 we also use the entropy - area law of eq . ( [ entropy ] ) , and the expression for the area for a @xmath18-dimensional sphere @xmath47 , to obtain the following expression for the temperature @xmath48 \left(\frac{n(n-1)\omega_n}{4 g } \tilde{r}_a^{n-2}\right ) \label{tds } \end{aligned}\ ] ] now by using eqs . ( [ denergy1 ] , [ dwork ] and [ tds ] ) , and the first law of thermodynamics given by eq . ( [ 1st ] ) , we obtain @xmath49 we also use eq . ( [ app - radius ] ) to obtain @xmath50 @xcite . thus , the eq . ( [ 1frw ] ) produces the dynamical friedman equation . @xmath51 furthermore , using the continuity equation ( [ continuity ] ) and integrating eq . ( [ doth ] ) , one obtains the following result , @xmath52 this is the friedmann equation for @xmath20-dimensional frw universe . now , we review briefly the modification of the friedman equations by the gup . thus , we will first review the gup @xcite , and its effect on the area - entropy law @xcite . the existence of a minimum measurable length originates as an intriguing prediction of various frameworks of quantum gravity such as string theory @xcite and black hole physics @xcite . this implies a direct modification of the standard uncertainty principle@xcite . thus , according to gup , the uncertainty in the length is modified to @xmath53 , \label{gup } \end{aligned}\ ] ] where @xmath54 is the planck length and @xmath55 is a dimensionless constant which is generated from some quantum gravitational effect . the new correction term in eq . ( [ gup ] ) becomes effective when the momentum and length scales are of order the planck mass and of the planck length , respectively . it has been demonstrated that the eq.([gup ] ) , implies the existence of minimal measurable length scale , @xmath56 as the black hole thermodynamics is related to the uncertainty principle , the gup modifies the thermodynamics of the black hole . this modification of the thermodynamics of a black hole also modifies the bekenstein - hawking entropy of a black hole @xcite ) . thus , gup modifies the relation between the entropy of a black hole and the area of its horizon , @xcite ) @xmath57}. \label{exact } \end{aligned}\ ] ] where @xmath58 , so the expression in eq . ( [ exact ] ) can be integrated to obtain the gup modified entropy - area law @xcite ) . @xmath59 + s_0 \label{exactentropy } \end{aligned}\ ] ] where @xmath60 is an integration constant . thus , eq . ( [ exactentropy ] ) is the modified bekenstein - hawking entropy . this modification has occurred due to the existence of a minimum measurable length scale or the gup . now in the thermodynamical approach the friedmann equation can be calculated from the entropy of the apparent horizon of frw universe . as the entropy - area law is modified because of the gup , we can expect that the friedmann equations will also be modified because of gup . in fact , it is possible to calculate the exact form of the gup deformed friedmann equations @xcite , @xmath61 , \label{mfra}\\ - 4 \pi g(\rho+p)&=&\left(\dot{h}-\frac{k}{a^2}\right ) \frac{\alpha}{8\ , \pi}\frac{(h^2+\frac{k}{a^2 } ) } { \left[1-\left(1-\frac{\alpha}{4\,\pi}\left(h^2+\frac{k}{a^2}\right)\right)^{\frac{1}{2}}\right]}.\label{mfrb}\end{aligned}\ ] ] here the gup modified friedmann equations @xcite have been calculated by using eq . ( [ mfra ] ) and eq . ( [ mfrb ] ) , in the the thermodynamical approach @xcite . these consequences of this modified friedmann equations have been studied @xcite . it has been demonstrated that these equations lead to bounded energy density @xmath14 as the inequality @xmath62 must be satisfied . this is because if this inequality is not satisfied , the density will become complex . by studying solutions to the modified raychaudhuri equation @xcite , using phase space method @xcite , it was found that the big bang singularity is not accessible in this description . this was because the spacetime itself can not be extended beyond planck density as a result of the gup modified thermodynamic approach to gravity . [ inflationgup ] now we can analyze the inflationary cosmology in the presence of a minimum measurable length . in this regards , the hamilton - jacobi approach is picked out as an appropriate formalism for studying inflation and its interesting prediction , quantum perturbations . we set @xmath63 in our case for studying inflation . considering the density given by @xmath64 , the modified friedman equation will be as follows : @xmath65 = \frac{8 \pi g}{3 } \left(\frac{\dot{\phi}^2}{2 } + v ( \phi)\right)\ ] ] note that , from the friedmann equation ( [ fri ] ) , it is realized that the expression @xmath66 must be always smaller than unity . in the rest of the work , we take into account this and expand the term up to the third order . + the hamilton - jacobi formalism is utilized as a strong tool to provide numerous inflationary models with exactly known analytic solutions for the background expansion@xcite . in this formalism , instead of potential , the hubble parameter is introduced as a function of scalar field @xmath67 . consequently , the time derivative of the hubble parameter is reexpressed ass @xmath68 , where @xmath69 denotes derivative with respect to scalar field . from the second friedmann equation ( [ mfrb ] ) , we have the time derivative of the scalar field as @xmath70 substituting the relation in the first friedmann equation comes to the hamilton - jacobi equation @xmath71 = \frac{8 \pi g}{3 } \left(\frac{1}{2(4 \pi g)^2 } \frac{{h'}^2 } { ( 1 + \frac{h}{4})^2 } + v ( \phi)\right)\ ] ] where the predicted potential of the model could easily be estimated , and the general behavior of the potential is investigated easily . + the most interesting aspect of the inflationary scenario is that the scenario predicts quantum perturbation in the very early times of the universe evolution . there are three types of perturbation known as scalar , vector , and tensor perturbations , and the most important ones are scalar and tensor perturbations . scalar fluctuations become seeds for cosmic microwave background ( cmb ) anisotropies or for large scale structure ( lss ) formation . therefore , by measuring the spectra of the cmb anisotropies and density distribution , the corresponding primordial perturbations could be determined . employing the amplitude of scalar perturbations from @xcite , the scalar spectra index is read as @xcite @xmath72 where @xmath73 and @xmath74 are the first and second slow - rolling parameters , given by @xcite @xmath75 besides scalar fluctuations , the inflationary scenario predicts tensor fluctuations , which is known as a gravitational wave , too . the produced tensor fluctuations induce a curved polarization in the cmb radiation and increase the overall amplitude of their anisotropies at a large scale . the physics of the early universe could be specified by fitting the analytical results the of cmb and density spectra to corresponding observational data . by contrast to scalar and vector perturbations , energy - momentum perturbations have no role in tensor perturbation equation . after doing some algebraic analysis , and obtaining the amplitude of tensor perturbations @xcite , the tensor spectra index is derived as @xcite @xmath76 the imprint of tensor fluctuations on the cmb bring this idea to indirectly determine its contribution to power spectra by measuring cmb polarization @xcite . such a contribution could be expressed by the @xmath77 quantity , which is known as tensor - to - scalar ratio and represents the relative amplitude of tensor - to - scalar fluctuations , @xmath78 . therefore , constraining @xmath77 is one of the main goals of the modern cmb survey , and is given by @xmath79 inflation lasts until the slow roll parameter @xmath73 approaches unity . then , the final value of scalar field could be read from eq.([srp ] ) . in order to estimate the field value in the beginning of inflation , the common approach is to employ the number of e - folds equation . the number of e - folds , indicated by @xmath80 , is expressed as following @xmath81 where the subscripts `` @xmath82 '' and `` @xmath83 '' respectively denote the initial and end of inflation . integrating the equation for specific hubble parameter , and using @xmath84 , the initial value of scalar field could be calculated , which will be done in the following subsection . + one of the absorbing aspect of the hamilton - jacobi formalism is that it allows to simply consider the attractor behavior of the model . the common approach is to suppose a homogeneous perturbation for hubble parameter as @xmath85 @xcite , and insert it in the hamiltonian - jacobi equation up to the first order . if the expression @xmath86 approaches to zero with increasing time , the attractor condition could be satisfied . in the following lines , the calculation is explained step by step for more clarity + @xmath87 thus , we can write @xmath88 now we can write @xmath89 where @xmath90 so , we can write @xmath91 to the first order , we can write @xmath92\end{aligned}\ ] ] thus , we can write @xmath93 \delta h = -\frac{{h_0'}}{a_1}\delta h ' \frac{1}{12 \pi g } \end{aligned}\ ] ] and @xmath94\end{aligned}\ ] ] now we have @xmath95 the attractor behavior is satisfied if the above expression goes to zero by passing time . + the general form of the potential was introduced in eq.([h - j ] ) . in order to study the behavior of the potential during the inflationary times , the potential could be drew in term of scalar field . rewriting the hamilton - jacobi equation ( [ h - j ] ) , the potential could be expressed by @xmath96 - \frac{1}{2 ( 4 \pi g)^2}\frac{{h'}^2}{(1 + \frac{\alpha}{16 \pi } h^2)^2}\ ] ] this a the general form of the potential in term of the hubble parameter . to investigate the behavior of the potential , we need to introduce a function for the hubble parameter in term on scalar field , that is what we are going to do in next subsection . + to go further , and investigate the result in more clear detail , it is necessary to propose a specific function for the hubble parameter in term of scalar field . in this regards , we assume that there is @xmath97 , in which @xmath98 is a constant . substituting the definition in the above equations , the situation will be studied in more detail . + inflation ends when the slow rolling parameter @xmath73 approaches unity . the general form of the equations were acquired in the previous section . in this section , we are going straight to the result for three different case , and consider the general behavior of tensor - to - scalar ratio , potential , and equation of state parameter @xmath99 ( it is assumed that there is an equation of state as @xmath100 ) . + the tensor - to - scalar , the potential , and the equation of state parameter respectively have been plotted for @xmath101 , @xmath102 , and three different values of number of e - folds as : @xmath103(solid line ) , @xmath104(dashed line ) and @xmath105(dotted - dashed line ) . in this case , we see that all three ones are same as each others . + the tensor - to - scalar , the potential , and the equation of state parameter respectively have been plotted for @xmath106 , @xmath107 , and three different values of number of e - folds as : @xmath108(solid line ) , @xmath109(dashed line ) and @xmath110(dotted - dashed line ) . in this case , we see that the plots , except the potentials , are same as each others . + the tensor - to - scalar , the potential , and the equation of state parameter respectively have been plotted for @xmath111 , @xmath107 , and three different values of number of e - folds as : @xmath112(solid line ) , @xmath113(dashed line ) and @xmath114(dotted - dashed line ) . in this case , we see that all three ones are same as each others . + in all above three cases , we observe the same behavior for @xmath77 , @xmath115 , and @xmath99 . the result displays that , the predicted values for tensor - to - scalar ratio is in good agreement with the latest observational data of planck as @xmath116 . in addition , the general behavior of the potential sounds to be suitable . at the beginning of inflation , the scalar field ( inflaton ) stands on top of the potential , and it rolls down to the minimum with increasing time , or with reducing scalar field . then , it is realized that the inflation in this work could be counted as a member of _ `` large field model '' _ class of inflation . the third part of figures depicts the equation of state parameter in term of scalar field . it is exhibited that in the initial times of inflation , the parameter is near to @xmath117 . then , the universe undergoes a quasi - de sitter expansion for the universe ; a result that was expected . finally , some parameters , such as amplitude of scalar perturbation , scalar spectra index , and tensor spectra index , have been estimated and presented in table.[t01 ] , [ t02 ] , and [ t03 ] for each case . the model prediction for the parameters are consistence with the planck data , where @xmath118 and @xmath119 . it states that the model could be taken into account as a proper model for studying inflation . + beside these , integrating @xmath120 during the inflation times interval shows that the expression is equal to a large negative value . consequently , from eq.([attractor ] ) , one could realized that the parameter @xmath121 approaches zero with increasing time , and the attractor behavior could be satisfied by the model . in this paper , we have analyzed the effect of having a minimum measurable length scale on inflationary cosmology using inflationary cosmology in the jacobian approach . this approach is motivated by the relation between gravity and thermodynamics where einstein field equations are the equation of state for the geometry of spacetime . so , in this paper , the friedmann equations are viewed as the clausius relation . then we analyze the modification of the friedmann equations because of the existence of a minimum measurable length . this is done by analyzing the effect of the existence of a minimum measurable length on the entropy of the cosmological horizon . finally , these modified friedmann equations are used for calculating the modifications to the inflationary cosmology . this analysis is done using the hamiltonian - jacobi approach . we thus explicitly calculate the effect of having a minimum measurable length scale on inflationary cosmology . it may be noted that property of the spectrum of large - scale magnetic fields which is generated due to the breaking of the conformal invariance has been studied in the context of inflationary cosmology @xcite . in this analysis , it has been demonstrated that the spectrum of the magnetic fields should not be perfectly scale - invariant . in fact , it was observed that this spectrum should be slightly red so that the amplitude of large - scale magnetic fields can be stronger than a certain value . in this analysis , it was assumed that the absence of amplification occurs due to the late - time action of some dynamo mechanism . it would be interesting to repeat this analysis by assuming the existence of a minimum measurable length scale . the inflation has also been studied using a non - canonical lagrangian @xcite-@xcite . in this case , the modification to the kinetic term is modified . this modification to the kinetic term depends only on the fields and not the derivatives of the fields . the non - canonical inflation has also been studied in the context of string - inspired inflation models @xcite the dbi action has also been used to study such models of inflation @xcite-@xcite . it has been demonstrated that the standard hubble slow roll expansion to the non - canonical case can be generalized @xcite . this generalization corresponds to the derivation of the expressions for observables in terms of the generalized slow roll parameters . it would be interesting to analyze the models with non - canonical kinetic term in the thermodynamic approach and also analyze the modification to these models that can occur because of the existence of a minimum measurable length scale . + utilizing hamilton - jacobi formalism in the introduced generalized friedmann equation allows one to properly study inflation epoch of the universe evolution . the calculated result demonstrates that the model could suitably describe inflation , and it sounds that the result is in consistence with the latest observational data of planck . the tensor - to - scalar ratio is plotted versus @xmath122 for three different cases , and in all of them the parameter @xmath77 stands in the predicted bound by planck as @xmath116 . the estimated values for @xmath122 and @xmath123 indicate that these parameters are almost about @xmath124 and @xmath125 ; which are in perfect agreement with planck data , where @xmath119 and @xmath126 . considering the predicted potential behavior exhibits that the proposed model stands in _ large field class _ of inflationary models , so that the scalar field leaves the top of the potential at the beginning of inflation and slowly rolls down to the minimum . in addition , behavior of equation of state parameter @xmath99 is in agreement with whole work . it expresses a quasi - de sitter expansion for the universe , in which at the initial of inflation , the parameter is so close to @xmath117 . finally , investigation of attractor behavior for the model comes to another pleasant outcome . the homogenous perturbation @xmath121 approaches zero with increasing time for all three cases describing that the model could satisfy the attractor behavior . b. majumder , phys . b * 703 * , 402 ( 2011 ) . [ arxiv:1106.0715 [ gr - qc ] ] ; a. f. ali , jhep * 1209 * , 067 ( 2012 ) [ arxiv:1208.6584 [ hep - th ] ] ; a. f. ali , h. nafie and m. shalaby , europhys . lett . * 100 * , 20004 ( 2012 ) . m. maggiore , m. maggiore , phys . b * 304 * , 65 ( 1993 ) [ arxiv : hep - th/9301067 ] ; m. maggiore , phys . d * 49 * , 5182 ( 1994 ) [ arxiv : hep - th/9305163 ] ; m. maggiore , phys . b * 319 * , 83 ( 1993 ) [ arxiv : hep - th/9309034 ] ; l. j. garay , int . j. mod . a * 10 * , 145 ( 1995 ) [ arxiv : gr - qc/9403008 ] ; s. hossenfelder , m. bleicher , s. hofmann , j. ruppert , s. scherer and h. stoecker , phys . b * 575 * , 85 ( 2003 ) [ arxiv : hep - th/0305262 ] ; c. bambi and f. r. urban , class . quant . grav . * 25 * , 095006 ( 2008 ) [ arxiv:0709.1965 [ gr - qc ] ] ; a. f. ali , s. das and e. c. vagenas , phys . b * 678 * , 497 ( 2009 ) [ arxiv:0906.5396 [ hep - th ] ] . s. v. ketov and t. terada , jhep * 1412 * , 062 ( 2014 ) [ arxiv:1408.6524 [ hep - th ] ] ; w. buchmuller , e. dudas , l. heurtier , a. westphal , c. wieck and m. w. winkler , jhep * 1504 * , 058 ( 2015 ) . [ arxiv:1501.05812 [ hep - th ] ] ; b. mcinnes , nucl . b * 894 * , 553 ( 2015 ) [ arxiv:1501.01759 [ hep - th ] ] ; g. barenboim and w. i. park , phys . d * 91 * , 063511 ( 2015 ) [ arxiv:1501.00484 [ hep - ph ] ] . a. aghamohammadi , a. mohammadi , t. golanbari and k. saaidi , phys . d * 90 * , 084028 ( 2014 ) [ arxiv:1502.07578 [ gr - qc ] ] ; j. m. cline and l. hoi , jcap * 0606 * 007 ( 2006 ) [ astro - ph/0603403 ] ; s. chongchitnan and g. efstathiou , jcap * 0701 * , 011 ( 2007 ) [ astro - ph/0611818 ] ; m. spalinski , jcap * 0704 * , 018 ( 2007 ) [ hep - th/0702118 ] ; r. bean , d. j. h. chung and g. geshnizjani , phys . d * 78 * , 023517 ( 2008 ) [ arxiv:0801.0742 [ astro - ph ] ] . r. b. mann and s. n. solodukhin , phys . d * 55 * , 3622 ( 1997 ) [ hep - th/9609085 ] ; r. k. kaul and p. majumdar , phys . lett . * 84 * , 5255 ( 2000 ) [ gr - qc/0002040 ] ; s. das , p. majumdar and r. k. bhaduri , class . * 19 * , 2355 ( 2002 ) [ hep - th/0111001 ] . g. amelino - camelia , m. arzano and a. procaccini , phys . rev . d 70 , 107501 ( 2004 ) [ gr - qc/0405084 ] ; e.m . lifshitz , l.p . pitaevskii and v.b . berestetskii , landau - lifshitz course of theoretical physics , volume 4 : quantum electrodynamics , ( reed educational and professional publishing , 1982 ) . k. bamba , phys . d * 75 * , 083516 ( 2007 ) . [ arxiv:0703647[astro - ph ] ] c. armendariz - picon , t. damour and v. f. mukhanov , phys . b * 458 * , 209 ( 1999 ) [ arxiv:9904075[hep - th ] ] . j. garriga and v. f. mukhanov , phys . b * 458 * , 219 ( 1999 ) [ arxiv:9904176[hep - th ] ] . s. kachru , r. kallosh , a. linde , j. m. maldacena , l. mcallister and s. p. trivedi , jcap * 0310 * , 013 ( 2003 ) [ arxiv:0308055[hep - th ] ] . e. silverstein and d. tong , phys . d * 70 * , 103505 ( 2004 ) [ arxiv:0310221[hep - th ] ] . m. alishahiha , e. silverstein and d. tong , phys . d * 70 * , 123505 ( 2004 ) [ arxiv:0404084[hep - th ] ] . g. barenboim and w. h. kinney , jcap * 0703 * , 014 ( 2007 ) [ arxiv:0701343[astro - ph ] ] .	in this paper , we will study the effect of having a minimum measurable length on inflationary cosmology . we will analyse the inflationary cosmology in the jacobson approach . in this approach , gravity is viewed as an emergent thermodynamical phenomena . we will demonstrate that the existence of a minimum measurable length will modify the friedmann equations in the jacobson approach . we will use this modified friedmann equation to analyse the effect of minimum measurable length scale on inflationary cosmology . this analysis will be performed using the hamiltonian - jacobi approach . we compare our results to recent data , and find that our model may agree with the recent data . * * a. mohammadi@xmath0 , ahmed farag ali@xmath1 , t. golanbari @xmath2 , + a. aghamohammadi@xmath3 , kh . saaidi @xmath4 , mir faizal@xmath5 + + + + + + + +
a decade ago , the ads / cft correspondencemaldacena:1997re , witten:1998qj , gubser:1998bc was conjectured . since then , it has provided us with new insights into gauge theories in a strong coupling regime . there has been substantial progress in studying strong coupling gauge theories by applying this technique . in particular , polchinski and strassler @xcite studied the deep inelastic scattering ( dis ) of hadrons by using the correspondence in forward compton scattering where the usual structure functions @xmath0 and @xmath1 are calculated for both spinless and spin-@xmath2 hadrons . an infrared cutoff in the fifth dimension at @xmath3 , which breaks the conformal symmetry in cft , is introduced to give rise to confinement . in addition , this cutoff scale also provides a mass scale for the hadrons with @xmath4 . this particular model with the hard cutoff in the infrared is then called the hard wall model . there have been a lot of interesting studies@xcite and further developments@xcite along this direction . these studies of dis and structure functions have provided an interesting picture of hadrons at the strong coupling limit . one finds that there is no large-@xmath5 parton which carries a finite amount of longitudinal momentum of the target hadron and most of the longitudinal momentum is distributed in the partons sitting in the extremely low-@xmath5 region . furthermore , dis , in terms of dipole ( quark - antiquark pair with an open string attached ) scattering on a hadron / nucleus target , is discussed in refs . in particular , the ads / cft computation of deep inelastic scattering off the finite temperature plasma has been recently studied in refs . @xcite , which also indicate that most of the constituents in the strongly coupled plasma are located in the very low @xmath5 region . based on hard scatterings in ads / cft , the saturation picture is then developed in refs . hatta:2007he , dominguez:2008vd . recently , we extended the above dis calculation in ads / cft to the case of polarized dis in ref . @xcite and obtained the spin - dependent structure functions @xmath6 and @xmath7 for a spin-@xmath2 hadron . in ref . @xcite , the spin decomposition of spin-@xmath2 hadrons at large coupling limit was analyzed . these studies show that the spin of hadrons at large thooft coupling seems solely due to orbital angular momentum . suggested by ji@xcite in qcd , the deeply virtual compton scattering ( dvcs ) can be used as the gateway to the total quark and gluon contributions to the spin of the proton . the dvcs amplitude contains the important information about the total spin of constituents in the hadron which is encoded in the generalized parton distributions . in addition , one can access the orbital angular momentum of quarks and gluons by subtracting the spin contributions . therefore , in order to further investigate the spin and the orbital angular momentum , we need to study the nonforward compton scattering and extract the so - called generalized parton distributionsmueller:1998fv , ji:1996ek , ji:1996 nm , radyushkin:1996nd , radyushkin:1997ki , ji:2004gf , belitsky:2005qn , goeke:2001tz , diehl:2003ny , brodsky:2000xy . as a first step , we study the nonforward compton scattering , in particular , the dvcs , in the ads / cft correspondence . we will leave the extraction of generalized parton distributions for future study . the objective of this paper is to study the nonforward compton scattering . we use dilatons ( scalar ) or dilatinos ( fermion ) as the targets which are scattered by a virtual gauge boson field ( a physical virtual photon , so to speak ) . we are particularly interested in the dvcs since it is widely studied in qcd and of great importance in spin physics . the final states contain a photon , which is real in dvcs , and the dilaton or dilatino target . this calculation is equivalent to a four - point function calculation that involves two currents @xmath8 and two targets ( dilaton or dilatino ) in the ads / cft . four - point functions of the dilaton and axion are studied in refs . liu:1998ty , freedman:1998bj , dhoker:1999pj . recently , four - point functions of @xmath9-currents are discussed in ads supergravitybartels:2009sc . to be more specific , we compute the supergravity graphs corresponding to compton scattering in cft . for the case of the dilaton , we have calculated the _ s_-channel and _ u_-channel graphs where all the kaluza - klein excitations of intermediate states are summed over . we have also included a four - point contact term and shown that the sum of these three graphs are gauge invariant , namely , satisfying the current conservation constraint . for the case of the dilatino , we compute the contributions from the _ s_-channel and _ u_-channel graphs explicitly . assuming that protons can be identified as supergravity modes of the dilatino , we compare the calculated dvcs cross section to the low - energy h1 and zeus data , and find that our calculation is consistent with the data . we also study the real compton scattering and find that the scattering amplitudes are parametrically the same as the one found in qed . we notice that there is an additional _ t_-channel graviton exchange contribution , which is supposed to be dominant at high energy . we provide a heuristic derivation for the computation of _ graviton exchange at the end of this paper . since the graviton exchange only couples to the energy momentum tensor of the target , we find the results for the dilaton and dilatino are quite similar . the rest of the paper is organized in the following manner : in sec.[scalar ] , we consider the usual _ s_-channel and _ u_-channel exchanges as well as the four - point interaction contributions for a dilaton target . in sec.[fermion ] , we evaluate the supergravity graphs of the _ s_-channel and _ u_-channel exchanges for dilatino targets . t_-channel exchange of the graviton for both dilaton and dilatino targets is discussed in sec.[gravitonex ] . we summarize and present some more discussion in sec.[conclusion ] . in this section , we will formulate the compton scattering on a scalar target in the cft . first , let us consider the free scalar field and derive its bulk to bulk propagator in @xmath10 space . scalars in cft correspond to supergravity modes of dilatons . the metric in @xmath11 space can be written as @xmath12where @xmath13 is the mostly plus flat space metric . we assume that the initial / final dilaton wave function can be written as @xmath14where , for simplicity , we have neglected the dependence on the coordinates @xmath15 . the equation of motion of the dilaton in ads space can be written as@xmath16with the fifth dimension mass @xmath17 . it reduces to @xmath18where @xmath19 . the solution to the above equation is given by @xmath20where @xmath21 . with the consideration of boundary conditions for normalizable modes , we set @xmath22 and choose only the @xmath23 part . in order to calculate the compton scattering amplitude , we also need the bulk to bulk propagator of dilatons in @xmath24 space , which satisfies @xmath25 g(x , z;y , z^{\prime } ) = z^{5}\delta ( z - z^{\prime } ) \delta ^{(4)}(x - y ) . \label{green}\]]in @xmath10 space , from coordinate @xmath26 to @xmath27 , the propagator can be written as @xcite , @xmath28where @xmath5 and @xmath29 are four - dimensional coordinates . one can easily check the above expression satisfies eq . ( [ green ] ) by noticing that @xmath30 z^{2}j_{\delta -2}(\omega z)=-\omega ^{2}z^{2}j_{\delta -2}(\omega z)\]]and @xmath31 note that @xmath32 corresponds to the timelike case , and @xmath33 corresponds to spacelike case . when the particle is at rest , we shall have @xmath34 . for the case of the _ s_-channel compton scattering , we find that @xmath35 . the dilaton propagator in the fifth dimension is then defined as @xmath36with @xmath37 for the _ s_-channel scattering and @xmath38 for the _ u_-channel scattering . following the prescription in ref . @xcite , the incident / outgoing current is chosen to be the @xmath9 current which couples to the hadron as an isometry of @xmath39 with the killing vector @xmath40 . this is equivalent to saying that the current couples to the target with a conserved charge @xmath41 . according to the ads / cft correspondence , the current excites a nonnormalizable mode of a kaluza - klein gauge field on the minkowski boundary of the @xmath42 space @xmath43this gauge field fluctuation @xmath44 ( a physical virtual photon , so to speak ) can be viewed as a vector boson field which couples to the @xmath45 current @xmath46 on the minkowski boundary , and then propagates into the bulk as gravitational waves , and eventually interacts with the supergravity modes of the dilaton or dilatino . the gauge field satisfies maxwell s equation in the bulk , @xmath47 which can be explicitly written as @xmath48 = 0,\]]where @xmath49 , @xmath50 , ... are indices on @xmath42 . with the boundary condition @xmath51and the lorentz - like gauge @xmath52the maxwell equation can be written as @xmath53the solutions to the above equations are given by , @xmath54where @xmath55 and @xmath56 are both modified bessel functions and @xmath57 is the polarization vector . note that @xmath58 for the spacelike current in the mostly plus metric signature . now we are ready to consider the compton scattering on a dilaton target . the relevant supergravity graphs are shown in figs . [ dvcs][dvcs3 ] , corresponding to @xmath59-channel exchange , @xmath60-channel exchange and four - point interaction , respectively . the @xmath61-channel graviton exchange illustrated in fig . [ dvcs4 ] will be considered in sec.gravitonex . the standard compton amplitude is defined as @xmath62 we first start with the calculation of the _ s_-channel amplitude as illustrated in fig . the bulk to bulk propagator from @xmath63 to @xmath64 represents high excited states of the dilaton target after absorbing the incoming virtual photon . this is different from the dis calculation where the propagator is always put on shell . the relevant supergravity interaction of the dilaton and photon is @xmath65the dilaton is taken to be in a charge eigenstate with the charge @xmath66 under the @xmath67 symmetry . this yields @xmath68 . using the dilaton wave functions along with the bulk to bulk propagator and above interaction vertex , it is straightforward to write down the _ s_-channel amplitude @xmath69where each integration over the @xmath39 space gives a factor of @xmath70 and the green s function @xmath71 picks up a factor of @xmath72 when we take the @xmath39 space into account . ( one can actually set @xmath73 in the calculation since the final result should not depend on the curvature @xmath74 . ) the initial and final wave functions of the dilaton are @xmath75respectively . the boundary condition is set as @xmath76 in the fifth dimension to mimic confinement . in addition , this cutoff scale also yields a mass scale for the hadrons with @xmath4deteramond:2005su , brodsky:2006uqa . it is straightforward to integrate out @xmath5 and @xmath29 , and then find @xmath77the _ s_-channel amplitude @xmath78 can be cast into @xmath79where @xmath80@xmath81 k_{0}(q^{\prime } z^{\prime } ) , \end{aligned}\]]@xmath82 k_{0}(qz)g(z , z^{\prime } ) q^{\prime } j_{\delta -2}(mz^{\prime } ) k_{1}(q^{\prime } z^{\prime } ) , \end{aligned}\]]and @xmath83 k_{0}(q^{\prime } z^{\prime } ) \nonumber \\ & & -\int dzdz^{\prime } k_{0}(qz)\frac{1}{z^{2}}\partial _ { z}\left [ z^{2}j_{\delta -2}(mz)\right ] \partial _ { z^{\prime } } g(z , z^{\prime } ) j_{\delta -2}(mz^{\prime } ) k_{0}(q^{\prime } z^{\prime } ) \nonumber \\ & & + \int dzdz^{\prime } k_{0}(qz)\frac{1}{z^{2}}\partial _ { z}\left [ z^{2}j_{\delta -2}(mz)\right ] g(z , z^{\prime } ) \frac{1}{z^{\prime 2}}\partial _ { z^{\prime } } \left [ z^{\prime 2}j_{\delta -2}(mz^{\prime } ) \right ] k_{0}(q^{\prime } z^{\prime } ) .\end{aligned}\ ] ] using the eq . ( [ norm ] ) and the recurrence relations @xmath84 -j_{\nu } ( \beta z)\frac{d}{dz}\left [ z\frac{d}{dz}j_{\nu } ( \alpha z)\right ] = ( \alpha ^{2}-\beta ^{2})zj_{\nu } ( \alpha z)j_{\nu } ( \beta z),\end{aligned}\]]one obtains @xmath85 \right\ } k_{0}(q^{\prime } z^{\prime } ) \nonumber \\ & = & \frac{m^{2}+s}{q^{\prime 2}}\int dz^{\prime } q^{\prime } g(z , z^{\prime } ) j_{\delta -2}(mz^{\prime } ) k_{1}(q^{\prime } z^{\prime } ) + z^{3}k_{1}(q^{\prime } z)j_{\delta -2}(mz)/q^{\prime } , \end{aligned}\]]where @xmath86 . eventually , one can simplify and cast these @xmath87 amplitudes into @xmath88and @xmath89where @xmath90 in the end , we obtain the _ s_-channel amplitude @xmath91 \mathcal{c}_{1},\end{aligned}\]]where we have defined that @xmath92 similarly , the amplitude from the _ u_-channel diagram , which is illustrated in fig . [ dvcs2 ] , is found to be @xmath93 \mathcal{c}_{1},\end{aligned}\]]where @xmath94and @xmath95 . note that actually the _ u_-channel amplitude and _ s_-channel amplitude are related by interchanging @xmath96 and @xmath97 according to the crossing symmetry . last but not least , we need to calculate the four - point interaction contribution , namely , the contact term as depicted in fig . [ dvcs3 ] . the four - point interaction amplitude is @xmath98then it is easy to obtain @xmath99 summing over all the contributions from those supergravity graphs , we arrive at the gauge invariant results @xmath100 \nonumber \\ & & + 2\mathcal{a}\left ( \eta _ { \mu \nu } q^{\prime } q - q_{\mu } ^{\prime } q_{\nu } ^{\prime } \frac{q}{q^{\prime } } -q_{\mu } q_{\nu } \frac{q^{\prime } } { q}% + q_{\mu } q_{\nu } ^{\prime } \frac{q\cdot q^{\prime } } { q^{\prime } q}\right ) \mathcal{c}_{1}.\end{aligned}\]]the overall constant is found to be @xmath101 . one can easily check the conservation of the current @xmath102 it is well - known that the dis amplitude and structure functions can be extracted from the imaginary part of forward compton scattering amplitude according to the optical theorem . therefore , one can easily obtain the structure functions @xmath0 and @xmath1 by taking the forward limit and the imaginary part of @xmath103 . the results agree with those found in ref . @xcite in the large-@xmath5 region . the same conclusion can be also drawn for a dilatino target . in the case of real compton scattering , it follows that @xmath104 since the incoming and outgoing photons are taken to be real photons . since a real photon only has 2 degrees of freedom , one can choose a polarization vector @xmath105 which is orthogonal to @xmath106 and @xmath107 in a special gauge . similarly we can also choose to have @xmath108 and @xmath109 for the outgoing photon in the target rest frame . therefore , the amplitude of real compton scattering can be reduced to @xmath110with @xmath111 > 0 $ ] when @xmath112 . this eventually leads to the unpolarized compton cross section after averaging over @xmath113 and summing over @xmath114 which gives @xmath115 . thus eventually one gets @xmath116where @xmath117 is the angle between the incoming photon and outgoing photon . this result ) will still hold while the constant @xmath118 may change since it depends on the details of the ir cutoff . ] is identical to the cross section found in scalar electrodynamics except for the overall constant . and the calculation of _ t_-channel graviton exchange amplitude is provided in sec . [ gravitonex ] . ] in the case of deeply virtual compton scattering , only the final state photon is real . therefore , we set @xmath119 but assume @xmath120 is large as compared to @xmath121 . hereafter in this subsection , we work in a limit that @xmath122 and @xmath123 but with @xmath124 fixed . in this case , the outgoing real photon has 2 degrees of freedom , while the incoming virtual photon has 3 . it is usually convenient to choose to stay in the target rest frame ( the lab frame ) . in this frame , we can choose to have @xmath125 and @xmath126 as well as @xmath127 . this yields @xmath128where@xmath129and@xmath130where we approximately write @xmath131 , which is an exact result if @xmath123 . one can approximately write @xmath132 . to obtain the cross section , let us work out the kinematics first . we analyze the deeply virtual compton scattering in the lab frame , in which the initial scalar is at rest . thus , @xmath133according to the conservation of energy momentum , it is straightforward to see that @xmath134 in addition , we know that @xmath135which lead us to @xmath136we write the polarization vectors explicitly in the following way:@xmath137where all the transverse polarization vectors are normalized to @xmath138 . therefore , one can easily average over @xmath113 and sum over @xmath139 , which yields @xmath140with the above polarization vectors , we find @xmath141 \left [ \varepsilon ^{\prime } \left ( q^{\prime } , \lambda ^{\prime } \right ) \cdot p^{\prime } \right ] \left [ \varepsilon \left ( q,\lambda \right ) \cdot \varepsilon ^{\prime } \left ( q^{\prime } , \lambda ^{\prime } \right ) \right ] = -mq_{0}\frac{q_{3}^{2}}{q^{2}}\sin ^{2}\theta .\end{aligned}\]]therefore , the averaged amplitude can be written as @xmath142 \nonumber \\ & & + 8\mathcal{q}^{4}\frac{c_{i}^{2}c_{f}^{2}}{z_{0}^{4}}\mathcal{c}_{1\text{% dvcs}}^{2}\left [ \frac{4m^{2}q_{3}^{2}}{\left ( m^{2}+(p+q)^{2}\right ) ^{2}}+% \frac{4mq_{0}}{m^{2}+(p+q)^{2}}\right ] \frac{q_{3}^{2}}{q^{2}}\sin ^{2}\theta .\end{aligned}\]]the phase space integral in this frame is @xmath143plugging everything into the cross - section formula , one finds @xmath144 , \end{aligned}\]]with @xmath145 \\ & = & \frac{q^{2}}{4m^{2}\left ( 1-x\right ) ^{2}}\left ( 1+\frac{4m^{2}}{q^{2}}% x\right ) ^{2}.\end{aligned}\]]when @xmath146 , one finds @xmath147 . in this section , we formulate the compton scattering on a fermion target by evaluating the _ s_-channel and _ u_-channel supergravity diagrams . spin-@xmath2 hadrons correspond to supergravity modes of the dilatino which satisfy the dirac equation in @xmath10 space . in the conformal region the dilatino field can be written as @xmath148where @xmath149 is an @xmath150 spinor on @xmath10 and @xmath151 is an @xmath152 spinor on @xmath39 . the wave function @xmath153 satisfies a five - dimensional dirac equation in ads space . let us first derive the five - dimensional dirac equation in the following . since the dirac gamma matrices are defined in the flat spacetime , we need to use vielbeins to convert them to curved spacetime and make the product @xmath154 lorentz invariant . a convenient choice of such vielbeins is given by @xmath155where @xmath156 . the levi - civita connection , which is also known as the christoffel connection , is given by @xmath157here we use @xmath158 to denote indices in flat space , and @xmath159 to denote indices in curved space ( @xmath10 space ) . in addition , the greek indices @xmath160 are defined in minkowski space . from the metric , one knows @xmath161it is then straightforward to work out the levi - civita connection in @xmath162 space @xmath163from the above vielbeins and levi - civita connections , we can compute the spin connection which is defined as @xmath164the only nonvanishing spin connections are@xmath165using the above results , the operator @xmath154 can be cast into @xmath166with @xmath167$].-matrices according to the dirac algebra in the mostly plus metric signature @xmath168 . the @xmath169 that we used differ from the conventional gamma matrices by a factor of @xmath170 . ] the free dilatino field in @xmath10 space satisfies the dirac equation @xcite @xmath171its normalizable solution is given by @xmath172 u_{\sigma } \]]with @xmath173therefore , we choose the initial and final wave functions of the dilatino to be @xmath174 u_{i\sigma } \nonumber \\ \overline{\psi } _ { f}^{\ast } ( y^{\prime } , z^{\prime } ) & = & \frac{c_{f}^{\prime } } { z_{0}r^{9/2}}e^{-ip^{\prime } \cdot y^{\prime } } \overline{u}_{f\sigma ^{\prime } } z^{\prime 5/2}\left [ j_{mr-1/2}(mz^{\prime } ) p_{-}+j_{mr+1/2}(mz^{\prime } ) p_{+}\right ] , \end{aligned}\]]with a chosen boundary condition @xmath175 . furthermore , we can now consider the bulk to bulk propagator of dilatino . it satisfies the following differential equation @xmath176using the identity @xmath177 \right\ } \nonumber \\ & = & \frac{z}{r}\left\ { e^{ip\cdot y}z^{\frac{5}{2}}\left [ j_{mr-1/2}(\omega z)p_{-}+j_{mr+1/2}(\omega z)p_{+}\right ] \right\ } \left ( ik\hspace{-6pt}% \slash-\omega \right ) , \label{greenfunction}\end{aligned}\]]it is to straightforward to verify that the solution of eq . ( greenfunction ) is given by @xmath178 \nonumber \\ & & \times \frac{-ik\hspace{-6pt}\slash+\omega } { \omega ^{2}+k^{2}-i\epsilon } % z^{\prime 5/2}\left [ j_{mr-1/2}(\omega z^{\prime } ) p_{-}+j_{mr+1/2}(\omega z^{\prime } ) p_{+}\right ] .\end{aligned}\]]following the usual convention , we start with the final state fermion and end with the initial state fermion when we write the amplitude . thus the following green s function is used hereafter : @xmath179 \nonumber \\ & & \times \frac{ik\hspace{-6pt}\slash+\omega } { \omega ^{2}+k^{2}-i\epsilon } % z^{5/2}\left [ j_{mr-1/2}(\omega z)p_{-}+j_{mr+1/2}(\omega z)p_{+}\right ] .\end{aligned}\ ] ] the interaction vertex between the kaluza - klein gauge field and dilatino is given by@xmath180where @xmath113 stands for the full 10-dimensional wave function of the dilatino state . in arriving at the above expression , we have used the fact that @xmath181 and the integration over @xmath39 gives @xmath70 . now we are ready to calculate the compton scattering amplitude of the dilatino from gauge / string duality . in this section , we consider the following two diagrams , namely , the _ s_-channel and _ u_-channel graphs . the _ s_-channel is illustrated in fig . here we use the bulk to bulk propagator from @xmath182 to @xmath63 , which is different from the deeply inelastic scattering where the propagator is always taken to be on shell . using the dilatino bulk to bulk propagator and above interaction vertex , we can write _ s_-channel amplitude @xmath78 as @xmath183where @xmath184 \gamma _ { \mu } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\gamma _ { \nu } \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p),\end{aligned}\]]@xmath185 \gamma _ { \mu } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\gamma ^{5}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p),\end{aligned}\]]@xmath186 \gamma ^{5 } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\gamma _ { \nu } \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p),\end{aligned}\]]@xmath187 \gamma ^{5 } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\gamma ^{5}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p),\end{aligned}\]]with @xmath188 , and find that the compton amplitude is independent of @xmath74 . ] and @xmath189 \nonumber \\ & & \times \frac{ik\hspace{-6pt}\slash+\omega } { \omega ^{2}+k^{2}-i\epsilon } % z^{5/2}\left [ j_{mr-1/2}(\omega z)p_{-}+j_{mr+1/2}(\omega z)p_{+}\right ] .\end{aligned}\ ] ] using eq . ( [ kr ] ) and @xmath190 , along with the recurrence relations @xmath191we can have @xmath192 \nonumber \\ & & \times \gamma ^{5}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & = & \int dzz^{2}qk_{1}(qz)(ik\hspace{-6pt}\slash+\omega ) \left [ j_{mr-1/2}(\omega z)p_{-}+j_{mr+1/2}(\omega z)p_{+}\right ] \frac{iq\hspace{% -6pt}\slash}{q^{2 } } \nonumber \\ & & \times \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & + \int dzz^{2}qk_{1}(qz)\frac{\omega ^{2}+k^{2}}{q^{2}}\left [ j_{mr-1/2}(\omega z)p_{+}+j_{mr+1/2}(\omega z)p_{-}\right ] \nonumber \\ & & \times \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p).\end{aligned}\]]similarly , we can also find @xmath193 \nonumber \\ & & \times \gamma ^{5}\left [ j_{mr-1/2}(\omega z)p_{+}+j_{mr+1/2}(\omega z)p_{-}\right ] ( ik\hspace{-6pt}\slash+\omega ) \nonumber \\ & = & -\int dzz^{2}q^{\prime } k_{1}(q^{\prime } z)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ j_{mr-1/2}(mz)p_{-}+j_{mr+1/2}(mz)p_{+}\right ] \nonumber \\ & & \times \frac{iq^{\prime } \hspace{-8pt}\slash}{q^{\prime } { } ^{2}}\left [ j_{mr-1/2}(\omega z)p_{+}+j_{mr+1/2}(\omega z)p_{-}\right ] ( ik\hspace{-6pt}% \slash+\omega ) \nonumber \\ & & -\int dzz^{2}q^{\prime } k_{1}(q^{\prime } z)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ j_{mr-1/2}(mz)p_{-}+j_{mr+1/2}(mz)p_{+}\right ] \nonumber \\ & & \times \left [ j_{mr-1/2}(\omega z)p_{-}+j_{mr+1/2}(\omega z)p_{+}\right ] \frac{\omega ^{2}+k^{2}}{q^{\prime } { } ^{2}}.\end{aligned}\ ] ] this can simplify the compton scattering amplitude a lot , and lead to @xmath194 \gamma _ { \mu } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\frac{q_{\nu } q\hspace{-5pt}\slash}{q^{2}}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \mu } p_{+}u_{\sigma } ( p)\frac{q_{\nu } } { q}\int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \mu } p_{-}u_{\sigma } ( p)\frac{q_{\nu } } { q}\int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr+1/2}^{2}(mz ) , \\ t_{3\mu \nu } ^{s } & = & -\mathcal{a}^{\prime } \int dz\int dz^{\prime } z^{-1/2}z^{\prime } { } ^{-1/2}q^{\prime } k_{1}(q^{\prime } z^{\prime } ) qk_{1}(qz ) \nonumber \\ & & \times \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ j_{mr-1/2}(mz^{% \prime } ) p_{-}+j_{mr+1/2}(mz^{\prime } ) p_{+}\right ] \frac{q_{\mu } ^{\prime } q^{\prime } \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2 } } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\gamma _ { \nu } \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \nu } p_{+}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } } { q^{\prime } } \int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \nu } p_{-}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } } { q^{\prime } } \int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr+1/2}^{2}(mz)\end{aligned}\]]and @xmath195 \frac{q_{\mu } ^{\prime } q^{\prime } \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2 } } \nonumber \\ & & \times g(z^{\prime } , z;p+q)\frac{q_{\nu } q\hspace{-6.5pt}\slash}{q^{2}}% \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) q\hspace{% -6.5pt}\slash p_{+}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { q^{\prime } q}\int dzz^{3}k_{1}(q^{\prime } z)k_{1}(qz)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) q\hspace{% -6.5pt}\slash p_{-}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { q^{\prime } q}\int dzz^{3}k_{1}(q^{\prime } z)k_{1}(qz)j_{mr+1/2}^{2}(mz ) \nonumber \\ & & -\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma ^{5}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { q^{\prime } q}\int dzz^{3}q^{\prime } k_{0}(q^{\prime } z)k_{1}(qz)j_{mr-1/2}(mz)j_{mr+1/2}(mz).\end{aligned}\ ] ] using the same trick , it is then straightforward to calculate the _ u_-channel graph as illustrated in fig . ( [ dvcs2 ] ) and obtain @xmath196where @xmath197 \gamma _ { \nu } \nonumber \\ & & \times g(z^{\prime } , z;p - q^{\prime } ) \gamma _ { \mu } \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) , \label{tu1}\end{aligned}\ ] ] @xmath198 \gamma _ { \nu } \nonumber \\ & & \times g(z^{\prime } , z;p - q^{\prime } ) \frac{q_{\mu } ^{\prime } q^{\prime } % \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2}}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \nu } p_{+}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } } { q^{\prime } } \int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \nu } p_{-}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } } { q^{\prime } } \int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr+1/2}^{2}(mz ) , \label{tu2 } \\ t_{3\mu \nu } ^{u } & = & -\mathcal{a}^{\prime } \int dz\int dz^{\prime } z^{-1/2}z^{\prime } { } ^{-1/2}qk_{1}(qz^{\prime } ) q^{\prime } k_{1}(q^{\prime } z ) \nonumber \\ & & \times \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ j_{mr-1/2}(mz^{% \prime } ) p_{-}+j_{mr+1/2}(mz^{\prime } ) p_{+}\right ] \frac{q_{\nu } q\hspace{% -7pt}\slash}{q^{2 } } \nonumber \\ & & \times g(z^{\prime } , z;p - q^{\prime } ) \gamma _ { \mu } \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \mu } p_{+}u_{\sigma } ( p)\frac{q_{\nu } } { q}\int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & + i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \mu } p_{-}u_{\sigma } ( p)\frac{q_{\nu } } { q}\int dzz^{3}k_{1}(qz)k_{1}(q^{\prime } z)j_{mr+1/2}^{2}(mz ) \label{tu3}\end{aligned}\ ] ] and @xmath199 \frac{q_{\nu } q\hspace{% -6.5pt}\slash}{q^{2 } } \nonumber \\ & & \times g(z^{\prime } , z;p - q^{\prime } ) \frac{q_{\mu } ^{\prime } q^{\prime } % \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2}}\left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) q^{\prime } % \hspace{-7.5pt}\slash p_{+}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { % q^{\prime } q}\int dzz^{3}k_{1}(q^{\prime } z)k_{1}(qz)j_{mr-1/2}^{2}(mz ) \nonumber \\ & & -i\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) q^{\prime } % \hspace{-7.5pt}\slash p_{-}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { % q^{\prime } q}\int dzz^{3}k_{1}(q^{\prime } z)k_{1}(qz)j_{mr+1/2}^{2}(mz ) \nonumber \\ & & -\mathcal{a}^{\prime } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma ^{5}u_{\sigma } ( p)\frac{q_{\mu } ^{\prime } q_{\nu } } { q^{\prime } } \int dzz^{3}k_{0}(qz)k_{1}(q^{\prime } z)j_{mr-1/2}(mz)j_{mr+1/2}(mz)\end{aligned}\ ] ] after some more algebra , one can find the total contribution is given by @xmath200 \left ( \gamma _ { \mu } -% \frac{q_{\mu } ^{\prime } q^{\prime } \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2}}% \right ) \nonumber \\ & & \times g(z^{\prime } , z;p+q)\left ( \gamma _ { \nu } -\frac{q_{\nu } q\hspace{% -6.5pt}\slash}{q^{2}}\right ) \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}% \right ] u_{\sigma } ( p ) \nonumber \\ & & + \mathcal{a}^{\prime } \int dz\int dz^{\prime } z^{-1/2}z^{\prime } { } ^{-1/2}qk_{1}(qz^{\prime } ) q^{\prime } k_{1}(q^{\prime } z ) \nonumber \\ & & \times \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ j_{mr-1/2}(mz^{% \prime } ) p_{-}+j_{mr+1/2}(mz^{\prime } ) p_{+}\right ] \left ( \gamma _ { \nu } -% \frac{q_{\nu } q\hspace{-6.5pt}\slash}{q^{2}}\right ) \nonumber \\ & & \times g(z^{\prime } , z;p - q^{\prime } ) \left ( \gamma _ { \mu } -\frac{q_{\mu } ^{\prime } q^{\prime } \hspace{-7.5pt}\slash}{q^{\prime } { } ^{2}}\right ) \left [ j_{mr-1/2}(mz)p_{+}+j_{mr+1/2}(mz)p_{-}\right ] u_{\sigma } ( p).\end{aligned}\]]it is obvious that the current conservation , namely , @xmath201 and @xmath202 , is satisfied as a result of cancellation between _ s_-channel and _ u_-channel amplitudes due to gauge invariance . in addition , eq . ( [ total1 ] ) can be recast into a simple form @xmath203 \nonumber \label{total2 } \\ & & \times \gamma ^{\alpha } \left [ i(p\hspace{-6pt}\slash+q\hspace{-6pt}\slash% ) + \omega \right ] \gamma ^{\beta } \left [ \mathcal{c}_{1}(q,\omega ) p_{+}+% \mathcal{c}_{2}(q,\omega ) p_{-}\right ] u_{\sigma } ( p ) \nonumber \\ & & -\mathcal{a}^{\prime } e_{\mu \beta } ^{l}(q^{\prime } ) e_{\nu \alpha } ^{l}(q)\int_{0}^{\infty } d\omega \frac{\omega } { \omega ^{2}+(p - q^{\prime } ) ^{2}-i\epsilon } \bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ \mathcal{c}% _ { 1}(q,\omega ) p_{-}+\mathcal{c}_{2}(q,\omega ) p_{+}\right ] \nonumber \\ & & \times \gamma ^{\alpha } \left [ i(p\hspace{-6pt}\slash - q^{\prime } \hspace{% -7pt}\slash)+\omega \right ] \gamma ^{\beta } \left [ \mathcal{c}_{1}(q^{\prime } , \omega ) p_{+}+\mathcal{c}_{2}(q^{\prime } , \omega ) p_{-}\right ] u_{\sigma } ( p),\end{aligned}\]]where @xmath204 in the case of real compton scattering , both initial and final state photons are real . in the limit @xmath205 , choosing the boundary condition @xmath206 with @xmath207 , we can have @xmath208 here we would like to work in a limit that @xmath122 and @xmath209 but with @xmath124 fixed . if we take the limit @xmath209 , we can obtain @xmath210setting both @xmath205 and @xmath119 , the real compton scattering amplitude is found to be @xmath211 in such approximation , the real compton scattering amplitude found above is parametrically the same as the compton scattering amplitude of the fundamental particle of spin @xmath212 . in the case of dvcs , the incoming photon is virtual , and the outgoing photon is assumed to be real . therefore , we set @xmath119 while @xmath213 , and still let @xmath123 but with @xmath124 kept fixed . thus , the scattering amplitude in eq . ( [ total2 ] ) is found to be @xmath214 u_{\sigma } ( p ) \nonumber \label{trcs1 } \\ & & -\mathcal{a}^{\prime } e_{\nu \alpha } ^{l}(q)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ \mathcal{c}_{1}(q , m)p_{-}+\mathcal{c}_{2}(q , m)p_{+}% \right ] \gamma ^{\alpha } \frac{i(p\hspace{-6pt}\slash - q^{\prime } \hspace{-7pt% } \slash)+m}{m^{2}+(p - q^{\prime } ) ^{2}}\gamma _ { \mu } u_{\sigma } ( p),\end{aligned}\]]where @xmath215to calculate the dvcs cross section , we define @xmath216 . the amplitude @xmath217 can be written as @xmath218 u_{\sigma } ( p ) \nonumber \\ & & -\frac{\mathcal{q}^{2}c_{i}^{\prime } c_{f}^{\prime } } { z_{0}^{2}}e_{\nu \alpha } ^{l}(q)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \left [ \mathcal{c}% _ { 1}(q , m)p_{-}+\mathcal{c}_{2}(q , m)p_{+}\right ] \gamma ^{\alpha } \frac{i(p% \hspace{-6pt}\slash - q^{\prime } \hspace{-7pt}\slash)+m}{m^{2}+(p - q^{\prime } ) ^{2}}\gamma _ { \mu } u_{\sigma } ( p),\end{aligned}\]]where one can see @xmath217 is dimensionless with our normalization . one can further simplify the amplitude as @xmath219 \nonumber \\ & & + \mathcal{d}_{2}e_{\nu \alpha } ^{l}(q)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma _ { \mu } \frac{i(p\hspace{-6pt}\slash+q\hspace{-6.5pt}% \slash)+m}{m^{2}+(p+q)^{2}}\gamma ^{\alpha } \gamma ^{5}u_{\sigma } ( p ) \nonumber \\ & & -\mathcal{d}_{2}e_{\nu \alpha } ^{l}(q)\bar{u}_{\sigma ^{\prime } } ( p^{\prime } ) \gamma ^{5}\gamma ^{\alpha } \frac{i(p\hspace{-6pt}\slash% -q^{\prime } \hspace{-7.5pt}\slash)+m}{m^{2}+(p - q^{\prime } ) ^{2}}\gamma _ { \mu } u_{\sigma } ( p),\end{aligned}\]]with @xmath220 and @xmath221 . before we evaluate the dvcs cross section , we need to write down the polarization of real and virtual photons in the lab frame . first for a real photon , we suppose the momentum and the polarization of the photon can be written as @xmath222according to the current conservation@xmath223one obtains @xmath224it is straightforward to show that @xmath225therefore , we see that @xmath226 . now for a virtual photon , we find that the momentum and the polarization are @xmath227where @xmath228 . according to the current conservation@xmath229one gets @xmath230it is easy to see that @xmath231and @xmath232therefore , for a virtual photon , we know @xmath233 using the results from the last subsection , we can find the following identity @xmath234together with the definition @xmath235after some tedious but straightforward computation , we find @xmath236with @xmath237 \\ a_{us}^{\left ( 1\right ) } & = & \frac{2}{\tilde{u}\tilde{s}}\left [ m^{2}(4m^{2}-% \tilde{s}-\tilde{u})-t(t+\tilde{s}+\tilde{u})\right ] \nonumber \\ & & + \frac{3q^{2}}{q_{3}^{2}\tilde{u}\tilde{s}}\left [ ( t+2m^{2})(4m^{2}+\tilde{% u}+t)-\tilde{u}^{2}+2m^{2}\tilde{s}\right ] , \end{aligned}\]]and @xmath238 \right . \nonumber \\ & & -\tilde{u}^{2}(4m^{4}-4m^{2}\tilde{s}-2m^{2}t-2m^{2}\tilde{u}-\tilde{s}% \tilde{u } ) \nonumber \\ & & \left . -\tilde{s}^{2}(4m^{4}-4m^{2}\tilde{u}-2m^{2}t-2m^{2}\tilde{s}-% \tilde{s}\tilde{u})\right\ } \nonumber \\ & & + \frac{3q^{2}}{q_{3}^{2}\tilde{s}^{2}\tilde{u}^{2}}\left ( 4m^{2}-\tilde{s}% + t+\tilde{u}-\frac{\tilde{s}\tilde{u}}{2m^{2}}\right ) \left [ m^{2}(\tilde{s}+% \tilde{u})^{2}+t\tilde{s}\tilde{u}\right ] \\ & = & -\frac{1}{\tilde{s}^{2}\tilde{u}^{2}}\left\ { \left ( \tilde{s}^{2}+\tilde{u% } ^{2}\right ) \tilde{s}\tilde{u}+2\left ( q^{2}-2m^{2}\right ) \left [ m^{2}\left ( \tilde{s}+\tilde{u}\right ) ^{2}+\tilde{u}\tilde{s}t\right ] \right\ } \nonumber \\ & & + \frac{3q^{2}}{q_{3}^{2}\tilde{s}^{2}\tilde{u}^{2}}\left ( 4m^{2}-\tilde{s}% + t+\tilde{u}-\frac{\tilde{s}\tilde{u}}{2m^{2}}\right ) \left [ m^{2}(\tilde{s}+% \tilde{u})^{2}+t\tilde{s}\tilde{u}\right ] , \end{aligned}\]]and @xmath239 \nonumber \\ & & -\frac{3}{8q^{2}q_{3}^{2}}\left [ 4m^{2}q^{2}+(q^{2}-\tilde{s})^{2}\right ] ( -2+\frac{t}{m^{2 } } ) \\ a_{uu}^{\left ( 2\right ) } & = & -\frac{1}{2q^{2}\tilde{u}^{2}}\left [ 2m^{2}(8m^{2}q^{2}+2q^{4}-4q^{2}\tilde{u}-\tilde{u}^{2})-\tilde{u}^{2}(% \tilde{s}+\tilde{u})+q^{2}\tilde{u}(\tilde{u}+2\tilde{s})\right ] \nonumber \\ & & -\frac{3}{8q^{2}q_{3}^{2}\tilde{u}^{2}}\left [ 2(4m^{2}q^{2}\tilde{s}% ^{2}+8m^{2}q^{2}\tilde{s}\tilde{u}+4q^{4}\tilde{s}\tilde{u}+q^{4}\tilde{u}% ^{2}-2q^{2}\tilde{s}^{2}\tilde{u}-\tilde{s}^{2}\tilde{u}^{2})\right . \nonumber \\ & & \left . \hspace{2cm}+\frac{t}{m^{2}}(q^{2}+\tilde{s})^{2}\tilde{u}^{2}% \right ] \\ a_{us}^{\left ( 2\right ) } & = & -\frac{1}{q^{2}\tilde{u}\tilde{s}}\left [ 4m^{2}q^{2}(4m^{2}+2q^{2}+t)-2m^{2}(\tilde{s}^{2}+\tilde{u}^{2}+\tilde{s}% \tilde{u})+(2q^{4}-\tilde{s}\tilde{u})t\right ] \nonumber \\ & & -\frac{3}{4q^{2}q_{3}^{2}\tilde{u}}\left [ 2(4m^{2}q^{2}\tilde{u}+3q^{4}% \tilde{u}+q^{2}\tilde{s}^{2}-2q^{2}\tilde{s}\tilde{u}-q^{2}\tilde{u}^{2}-% \tilde{s}^{3}-\tilde{s}^{2}\tilde{u}-\tilde{s}\tilde{u}^{2})\right . \nonumber \\ & & \left . \hspace{2cm}+\frac{t}{m^{2}}(q^{4}-\tilde{s}^{2})\tilde{u}\right]\end{aligned}\]]and @xmath240 \nonumber \\ & & -\frac{3}{2q^{2}q_{3}^{2}\tilde{u}^{2}}\left [ ( 2m^{2}q^{2}-\tilde{s}\tilde{u% } ) ( \tilde{s}+\tilde{u})^{2}+2q^{4}\tilde{u}(\tilde{s}+\tilde{u})+q^{2}\tilde{% u}^{2}t(2+q^{2}/m^{2})\right ] .\end{aligned}\]]the mandelstam variables used above are defined as follows @xmath241 using the lab frame we define in eqs.([frame1 ] ) and eq.([frame2 ] ) , along with the phase space integral @xmath242we can write cross - section formula as @xmath243 . \label{fdvcs}\end{aligned}\ ] ] ) cross section as a function of @xmath244 for @xmath245 and @xmath246 . the data points are taken from refs . the solid line represents the dilatino dvcs cross section in eq . ( [ fdvcs ] ) after integrating over @xmath61 . we set @xmath247 , the mass @xmath248 to be the proton mass and @xmath249 . we have used one of the data points to fix the overall constant . , width=566 ] the above equation is one of our major results in this paper . assuming the qcd coupling is large during the interaction due to confinement , we can compare this result to the hera data . we plot the integrated cross section as a function of @xmath244 in fig . [ sigma ] . as shown in fig . [ sigma ] , we compare the dvcs cross section in eq . ( [ fdvcs ] ) after integrating over @xmath61 ( @xmath250 ) to the low - energy h1 and the zeus data@xcite and find that our result is consistent with the data . equation ( [ fdvcs ] ) only includes the _ s_-channel and _ u_-channel graphs . there should be the _ t_-channel graviton exchange contribution which is discussed in the next section in the plot . however , we argue that the _ s_-channel and _ u_-channel contributions are dominant in the low - energy region , whereas the _ t_-channel graviton exchange dominates in the high - energy limit . in addition , technically , it is hard to fix the normalization of the _ t_-channel graph as compared to the _ s_-channel and _ u_-channel graphs . it is obvious that the _ s_-channel and _ u_-channel contributions are proportional to @xmath251 , with @xmath41 being the charge of the targets . on the other hand , the _ t_-channel graviton exchange only couples to the energy momentum tensors of the target hadrons , and it does not depend on the quantum number of the target . in qcd , the situation is quite similar . in the low - energy region , where valence quark contribution dominates , the scattering amplitude certainly is proportional to @xmath252 , where @xmath253 is the proton charge . however , in the high - energy region , where the pomeron exchange starts to dominate the amplitude , we find the parton distribution becomes universal and independent of the quantum number of the target hadron . therefore , assuming the _ t_-channel graviton exchange graphs are small in the low - energy region , we neglect this contribution when we compare the calculated cross section to the low - energy h1 and the zeus data@xcite . for @xmath254 and @xmath246 . the solid line represents the dilaton dvcs cross section calculated in sec . [ scalar ] . we set @xmath255 and use the same normalization and parameters as the dilatino in the plot.,width=566 ] as a comparison , we also plot the integrated cross section for a scalar target , which is calculated in sec . [ scalar ] , as function of @xmath244 in fig . [ sigma2 ] . we use the same parameter and overall constant as in fig . [ sigma ] . numerically , there is little difference between these two plots . the exact evaluation of the _ t_-channel graviton exchange graph as illustrated in fig . [ dvcs4 ] is not an easy task . here we provide a heuristic derivation at the large-@xmath59 limit where the _ t_-channel graviton exchange contribution dominates , with @xmath59 being the center of mass energy . therefore , hereafter we neglect the mass terms in the energy momentum tensors and only keep the leading order terms . for simplicity , we also drop the @xmath39 dependence of the wave functions , because it only contributes an overall constant . it is well - known that the graviton couples to the energy momentum tensor . the _ t_-channel graviton exchange contribution does not depend on the charge of the target hadron ( @xmath41 ) , which is uniquely different from the _ s_-channel and _ u_-channel graphs contributions computed above . technically , it is hard to normalize the _ t_-channel graviton contribution as compared to the _ s_-channel and _ u_-channel graphs contributions . we assume that the _ t_-channel contribution is small at low energy . however , it is the dominant contribution at high energy due to its energy dependence . in this section , we set the curvature of the ads space @xmath256 , since the final results do not depend on @xmath74 . the _ t_-channel graviton exchange amplitude between the dilaton target and the photon can be written in the following : @xmath257where @xmath258 and @xmath259 are the polarization vectors and @xmath260 is the gravitational coupling to the stress - energy tensor . @xmath261 is the graviton propagator which can be written aspolchinski:1998rq @xmath262with @xmath263 . strictly speaking , one should use the full graviton propagator derived in @xmath42 space as in ref . however , in the high - energy limit , one can still obtain the correct expression using the above graviton propagator ( see , e.g. , in refs . the stress - energy tensor @xmath264 for a massless dilaton field is @xmath265we put @xmath266 as the initial dilaton wave function and @xmath267 as the final dilaton wave function . the stress - energy tensor @xmath268 for a kaluza - klein gauge field can be written as @xmath269with @xmath270the indices @xmath271 denote the four dimensions in minkowski spacetime , and the rest of the indices @xmath272 are defined in @xmath10 space . it is easy to work out the tensor contractions and obtain @xmath273this expression agrees with ref . @xcite in the forward limit . at high energy@xcite , the first term in the above effective action dominates , because it gives rise to terms like @xmath274 , which is large as compared to @xmath120 and @xmath121 . therefore , the graviton exchange amplitude for the compton scattering can be written as @xmath275it is straightforward to find that @xmath276following the derivation in ref.@xcite , we get @xmath277 q^{2}q^{\prime 2}\mathcal{t}_{1}^{g } \nonumber \\ & & + \mathcal{a}^{\prime \prime } \left [ \begin{array}{c } \frac{q^{\prime } q}{4xx^{\prime } } \left ( \eta _ { \mu \nu } + \frac{2x}{q^{2}}% p_{\mu } q_{\nu } + \frac{2x^{\prime } } { q^{\prime 2}}q_{\mu } ^{\prime } p_{\nu } ^{\prime } + \frac{4xx^{\prime } } { q^{2}q^{\prime 2}}q\cdot q^{\prime } p_{\mu } p_{\nu } ^{\prime } \right ) \\ + \frac{q^{\prime } q}{4\tilde{x}\tilde{x}^{\prime } } \left ( \eta _ { \mu \nu } -% \frac{2\tilde{x}}{q^{2}}p_{\mu } ^{\prime } q_{\nu } -\frac{2\tilde{x}^{\prime } % } { q^{\prime 2}}q_{\mu } ^{\prime } p_{\nu } + \frac{4\tilde{x}\tilde{x}^{\prime } % } { q^{2}q^{\prime 2}}q\cdot q^{\prime } p_{\mu } ^{\prime } p_{\nu } \right ) \end{array}% \right ] q^{2}q^{\prime 2}\mathcal{t}_{2}^{g},\end{aligned}\]]with @xmath278 , @xmath279 , @xmath280 , @xmath281 , @xmath282 and @xmath283with @xmath284 and @xmath285 . here we have neglected the contributions of @xmath286 and @xmath287 since they are subleading at the high - energy limit . the contour @xmath288 runs parallel to the imaginary axis from @xmath289 to @xmath290 with @xmath291 . it is interesting to note that @xmath292 is always conserved , namely , @xmath293 and @xmath294 , thanks to its coupling to the photon energy momentum tensor and the equation of motion . assuming that the interaction is local , one obtains@xmath295 in the forward compton scattering limit , our result agrees with the one found in refs.@xcite . if the energy gets so high that @xmath296 becomes of order @xmath138 , the imaginary part of the graviton exchange amplitude can no longer be assumed local due to diffusion@xcite . because graviton has a spin of 2 , the forward scattering amplitude is proportional to @xmath297 up to some curvature corrections . for a dilatino target , we use the following dilatino energy momentum tensor @xmath298from the above calculation for the dilaton , one can easily see that the leading contribution at high energy comes from @xmath299thus the _ t_-channel graviton exchange amplitude for the dilatino is @xmath300 q^{2}q^{\prime 2}\mathcal{t}_{1}^{g\psi } \nonumber \\ & & + \mathcal{a}^{\prime \prime } \left [ \begin{array}{c } \frac{q^{\prime } q}{4xx^{\prime } } \left ( \eta _ { \mu \nu } + \frac{2x}{q^{2}}% p_{\mu } q_{\nu } + \frac{2x^{\prime } } { q^{\prime 2}}q_{\mu } ^{\prime } p_{\nu } ^{\prime } + \frac{4xx^{\prime } } { q^{2}q^{\prime 2}}q\cdot q^{\prime } p_{\mu } p_{\nu } ^{\prime } \right ) \\ + \frac{q^{\prime } q}{4\tilde{x}\tilde{x}^{\prime } } \left ( \eta _ { \mu \nu } -% \frac{2\tilde{x}}{q^{2}}p_{\mu } ^{\prime } q_{\nu } -\frac{2\tilde{x}^{\prime } % } { q^{\prime 2}}q_{\mu } ^{\prime } p_{\nu } + \frac{4\tilde{x}\tilde{x}^{\prime } % } { q^{2}q^{\prime 2}}q\cdot q^{\prime } p_{\mu } ^{\prime } p_{\nu } \right)% \end{array}% \right ] q^{2}q^{\prime 2}\mathcal{t}_{2}^{g\psi } , \end{aligned}\]]with @xmath301the evaluation of the above compton scattering amplitude is quite complicated . nevertheless , we can approximately find the corresponding dvcs cross section for both dilaton and dilatino targets at the @xmath302 limit is @xmath303where the factor @xmath304 comes from the phase space integral and @xmath305 arises due to the graviton exchange together with its curvature correction . the energy enhancement in eq . ( gravs ) indicates that the _ t_-channel graviton exchange should dominate the total scattering amplitude in the high - energy limit . in this paper , we calculate the supergravity graphs corresponding to compton scatterings for both dilaton and dilatino targets in the cft . for a dilaton target , we have calculated the _ s_-channel , _ u_-channel graphs where all the kaluza - klein excitations of intermediate states are summed over , and the four - point contact term . we sum all these three graphs and compute the usual real compton scattering amplitude and dvcs cross section . for a dilatino target , we compute the contributions from the _ s_-channel and _ u_-channel graphs explicitly . we find that the real compton scattering amplitude , up to an overall constant , is identical to the amplitude found in scalar qed . the dvcs cross section is also discussed afterwards and compared to the low - energy h1 and zeus data . the curve is consistent with the experimental data . the end of this paper is devoted to discussing the _ t_-channel graviton exchange contribution , which is dominant contribution in the high - energy limit . we have provided a heuristic derivation for the computation of the _ t_-channel graviton exchange . it will be interesting to continue this study and extract the generalized parton distributions as well as the information about the total spin of constituents in a hadron . this would help us to understand the role of the orbital angular momentum and the spin sum rule at strongly coupled regime . we acknowledge inspiring discussions with e. iancu , y. hatta , c. marquet , a. h. mueller and f. yuan . j. g. acknowledges financial support by the china postdoctoral science foundation funded project under contract no . . b. x. is supported by the director , office of energy research , office of high energy and nuclear physics , divisions of nuclear physics , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 . 99 j. m. maldacena , adv . theor . math . * 2 * , 231 ( 1998 ) [ int . j. theor . phys.38 , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] . e. witten , adv . * 2 * , 253 ( 1998 ) [ arxiv : hep - th/9802150 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , phys . b * 428 * , 105 ( 1998 ) [ arxiv : hep - th/9802109 ] . j. polchinski and m. j. strassler , phys . lett . * 88 * , 031601 ( 2002 ) [ arxiv : hep - th/0109174 ] . j. polchinski and m. j. strassler , jhep * 0305 * , 012 ( 2003 ) [ arxiv : hep - th/0209211 ] . r. c. brower and c. i. tan , nucl . b * 662 * , 393 ( 2003 ) [ arxiv : hep - th/0207144 ] . h. boschi - filho and n. r. f. braga , phys . b * 560 * , 232 ( 2003 ) [ arxiv : hep - th/0207071 ] . r. c. brower , m. j. strassler and c. i. tan , jhep * 0903 * , 092 ( 2009 ) [ arxiv:0710.4378 [ hep - th ] ] . r. c. brower , j. polchinski , m. j. strassler and c. i. tan , jhep * 0712 * , 005 ( 2007 ) [ arxiv : hep - th/0603115 ] . c. a. ballon bayona , h. boschi - filho and n. r. f. braga , jhep * 0809 * , 114 ( 2008 ) [ arxiv:0807.1917 [ hep - th ] ] . c. a. ballon bayona , h. boschi - filho and n. r. f. braga , jhep * 0810 * , 088 ( 2008 ) [ arxiv:0712.3530 [ hep - th ] ] . c. a. ballon bayona , h. boschi - filho and n. r. f. braga , jhep * 0803 * , 064 ( 2008 ) [ arxiv:0711.0221 [ hep - th ] ] . b. pire , c. roiesnel , l. szymanowski and s. wallon , phys . b * 670 * , 84 ( 2008 ) [ arxiv:0805.4346 [ hep - ph ] ] . l. cornalba and m. s. costa , phys . rev . d * 78 * , 096010 ( 2008 ) [ arxiv:0804.1562 [ hep - ph ] ] . j. l. albacete , y. v. kovchegov and a. taliotis , jhep * 0807 * , 074 ( 2008 ) [ arxiv:0806.1484 [ hep - th ] ] . e. levin , j. miller , b. z. kopeliovich and i. schmidt , jhep * 0902 * , 048 ( 2009 ) [ arxiv:0811.3586 [ hep - ph ] ] . y. v. kovchegov , z. lu and a. h. rezaeian , phys . d * 80 * , 074023 ( 2009 ) [ arxiv:0906.4197 [ hep - ph ] ] . y. hatta , e. iancu and a. h. mueller , jhep * 0801 * , 063 ( 2008 ) [ arxiv:0710.5297 [ hep - th ] ] . a. h. mueller , a. i. shoshi and b. w. xiao , nucl . a * 822 * , 20 ( 2009 ) [ arxiv:0812.2897 [ hep - th ] ] . e. avsar , e. iancu , l. mclerran and d. n. triantafyllopoulos , jhep * 0911 * , 105 ( 2009 ) [ arxiv:0907.4604 [ hep - th ] ] . f. dominguez , arxiv:0912.1641 [ hep - th ] . c. a. b. bayona , h. boschi - filho and n. r. f. braga , arxiv:0912.0231 [ hep - th ] . e. iancu and a. h. mueller , arxiv:0912.2238 [ hep - th ] . y. hatta , e. iancu and a. h. mueller , jhep * 0801 * , 026 ( 2008 ) [ arxiv:0710.2148 [ hep - th ] ] . f. dominguez , c. marquet , a. h. mueller , b. wu and b. w. xiao , nucl . a * 811 * , 197 ( 2008 ) [ arxiv:0803.3234 [ nucl - th ] ] . j. h. gao and b. w. xiao , phys . d * 80 * , 015025 ( 2009 ) [ arxiv:0904.2870 [ hep - ph ] ] . y. hatta , t. ueda and b. w. xiao , jhep * 0908 * , 007 ( 2009 ) [ arxiv:0905.2493 [ hep - ph ] ] . x. d. ji , phys . * 78 * , 610 ( 1997 ) [ arxiv : hep - ph/9603249 ] . d. mueller , d. robaschik , b. geyer , f. m. dittes and j. horejsi , fortsch . phys . * 42 * , 101 ( 1994 ) [ arxiv : hep - ph/9812448 ] . x. d. ji , phys . d * 55 * , 7114 ( 1997 ) [ arxiv : hep - ph/9609381 ] . a. v. radyushkin , phys . lett . b * 380 * , 417 ( 1996 ) [ arxiv : hep - ph/9604317 ] . a. v. radyushkin , phys . d * 56 * , 5524 ( 1997 ) [ arxiv : hep - ph/9704207 ] . x. ji , ann . nucl . part . sci . * 54 * , 413 ( 2004 ) . a. v. belitsky and a. v. radyushkin , phys . rept . * 418 * , 1 ( 2005 ) [ arxiv : hep - ph/0504030 ] . k. goeke , m. v. polyakov and m. vanderhaeghen , prog . part . phys . * 47 * , 401 ( 2001 ) [ arxiv : hep - ph/0106012 ] . m. diehl , phys . * 388 * , 41 ( 2003 ) [ arxiv : hep - ph/0307382 ] . s. j. brodsky , m. diehl and d. s. hwang , nucl . b * 596 * , 99 ( 2001 ) [ arxiv : hep - ph/0009254 ] . h. liu and a. a. tseytlin , phys . d * 59 * , 086002 ( 1999 ) [ arxiv : hep - th/9807097 ] . d. z. freedman , s. d. mathur , a. matusis and l. rastelli , phys . b * 452 * , 61 ( 1999 ) [ arxiv : hep - th/9808006 ] . e. dhoker , d. z. freedman , s. d. mathur , a. matusis and l. rastelli , nucl . b * 562 * , 353 ( 1999 ) [ arxiv : hep - th/9903196 ] . j. bartels , j. kotanski , a. m. mischler and v. schomerus , arxiv:0908.2301 [ hep - th ] . g. f. de teramond and s. j. brodsky , phys . * 94 * , 201601 ( 2005 ) [ arxiv : hep - th/0501022 ] . s. j. brodsky and g. f. de teramond , phys . lett . * 96 * , 201601 ( 2006 ) [ arxiv : hep - ph/0602252 ] . m. henningson and k. sfetsos , phys . b * 431 * , 63 ( 1998 ) [ arxiv : hep - th/9803251 ] . w. mueck and k. s. viswanathan , phys . d * 58 * , 041901 ( 1998 ) [ arxiv : hep - th/9804035 ] . a. aktas _ et al . _ [ h1 collaboration ] , eur . j. c * 44 * , 1 ( 2005 ) [ arxiv : hep - ex/0505061 ] . s. chekanov _ et al . _ [ zeus collaboration ] , phys . b * 573 * , 46 ( 2003 ) [ arxiv : hep - ex/0305028 ] . f. d. aaron _ et al . _ [ h1 collaboration ] , phys . b * 659 * , 796 ( 2008 ) [ arxiv:0709.4114 [ hep - ex ] ] . j. polchinski , _ string theory : an introduction to the bosonic string _ ( cambridge university press , cambridge , england , 1998),vol . 1 , p. 402 .	we study the nonforward compton scattering , in particular , the deeply virtual compton scattering from ads / cft . we first calculate the contributions from the _ s_-channel and _ u_-channel supergravity diagrams as well as the four - point interaction diagram which correspond to the compton scatterings on a dilaton target in cft . furthermore , we study the compton scattering on a dilatino target . assuming that protons can be identified as supergravity modes of the dilatino , we compare the calculated deeply virtual compton scattering cross section to the low - energy experimental data from the h1 and zeus collaborations and find good agreement . we also discuss the _ t_-channel graviton exchange contribution and show that it should be dominant in the high - energy limit .
spintronics has attracted attention because it is expected to realize new magnetic memories and computing devices by use of electronic spins.@xcite a necessity of spintronics is to understand the nature of a spin current , a flow of spin angular momentum . the concept of a spin current stems from spin - polarized transport in ferromagnetic metals . ferromagnetic metals conduct a spin - dependent electric current , a charge current accompanied by a spin current . this current induces interesting phenomena such as giant magnetoresistance@xcite and current - induced magnetization reversal@xcite . the complete separation into spin and charge currents leads to the concept of a _ pure _ spin current , a flow of electronic spins with _ no _ charge currents . spin pumping@xcite is a versatile method for generating a pure spin current . this method uses magnetization precession in a ferromagnet to inject a spin current into an adjacent metal . ferromagnet ( f)/normal metal ( nm ) bilayer films have offered themselves as the benchmark for the validity of spin pumping . experimental and theoretical studies have concentrated on this structure and contributed to a solid understanding of spin pumping . spin - current injection enhances magnetization relaxation in f and broadens ferromagnetic - resonance ( fmr ) line width . the injection is electrically detectable in nm as well ; spin - orbit interaction in nm can convert an injected spin current into a charge current . this conversion is known as the inverse spin hall effect@xcite ( ishe ) and induces an electric field @xmath2 that satisfies the relation@xcite @xmath3 here , @xmath4 is the spatial direction of a spin current and @xmath5 is the spin - polarization vector . spin pumping in f / nm / f trilayer films will be of further interest since new collective excitation can be induced by magnetic couplings between the ferromagnets . in fact , such excitation was verified by fmr measurements.@xcite this paper presents an experimental investigation on spin pumping in a @xmath6 trilayer film by means of ishe . dc voltage is found to be enhanced when both the ferromagnets are excited simultaneously . the enhanced voltage is unconventional in the sense that it can not be explained by spin - pumping theory of f / nm bilayer films . control experiments demonstrates that the unconventional voltage comes from spin pumping at the py / pt interface . we discuss the origin of the unconventional voltage and narrow down contributing factors on an experimental basis . figure 1(a ) shows a schematic illustration of a @xmath7 trilayer film . the @xmath8 ( pt ) layer was 1 mm wide , 3 mm long and 50 nm ( 5 nm ) thick . the @xmath9 layer was 1 mm wide , 2 mm long and 10 nm thick . we grew the @xmath8 layer on a @xmath10 ( ggg ) substrate by a metal organic decomposition method@xcite ; we then sputtered the pt layer on the yig layer and finally deposited the py layer by the electron - beam evaporation in a high vacuum . we attached two electrodes to the ends of the pt layer to measure dc voltage . the film was placed at the center of a @xmath11 microwave cavity at which a microwave magnetic field ( electric field ) maximizes ( minimizes ) itself . a frequency of the microwave was 9.44 ghz . we applied a static magnetic field perpendicular to the microwave magnetic field . figure 1(b ) defines the out - of - plane magnetic - field angle @xmath12 and the magnetization angle @xmath13 ( @xmath14 ) of yig ( py ) . spin pumping injects a spin current into the pt layer , leaving the spin - polarization vector @xmath15 parallel to the precession axis on temporal average [ see fig . 1(c)].@xcite ishe in the pt layer converts the injected spin current into dc voltage . dc voltage and an fmr spectrum were measured at room temperature with changing static - field magnitude or direction . figure 2(a ) shows spectra of the microwave absorption signal at @xmath16 = @xmath1790 deg . ( i.e. , with in - plane fields ) . we see fmr of py ( yig ) near @xmath18 = 120 ( 300 ) mt . correlating with fmr , dc voltage appears as shown in fig . remember the direction of relevant vectors : @xmath19 , @xmath20 , and @xmath21 [ see fig . 1(c ) ] . here , @xmath22 , @xmath23 , and @xmath24 represent unit vectors in each direction . the voltage spectra are consistent with eq . ( [ ishe ] ) . at @xmath16 = 90 deg . , voltage is negative at fmr of py while it is positive at fmr of yig . this opposite polarity is due to the opposite directions of spin - current injection ( @xmath25 , @xmath20 , and thus @xmath26 ) . moreover , when the field direction is reversed , the voltage sign reverses at each fmr . this reversal is due to the opposite directions of spin polarization ( @xmath27 , @xmath28 , and thus @xmath26 ) . these results all support that the voltage is dominated by ishe . the replacement of pt with cu reveals irrelevance of galvanomagnetic effects such as the anomalous hall effect and the planar hall effect@xcite . the replacement greatly reduces voltage . as a result , the voltage with cu can not reproduce the voltage with pt , even considering a conductance difference between pt and cu . this reduction is consistent with the fact that ishe is negligibly small in cu owing to weak spin - orbit interaction . figure 2(c ) shows the @xmath16 dependence of a resonance field @xmath29 . the result is consistent with previous studies of bilayer films.@xcite in terms of @xmath29 , we observed little difference between the trilayer film and its corresponding yig / pt and py / pt bilayer films . different saturation magnetization between yig and py results in the different @xmath16 dependence of @xmath29 via a demagnetization field along the @xmath30-axis [ see fig . 1(b)].@xcite this enables us to control magnetization dynamics in each layer . at @xmath31 deg , resonance fields are different between @xmath8 and @xmath9 . thus , while one magnetic layer is on resonance with large precessional amplitude , the other is off resonance with small precessional amplitude . around @xmath32 deg . , on the other hand , resonance fields are nearly identical : magnetization precesses simultaneously in the two ferromagnets . figure 3(a ) shows the magnetic - field dependence of voltage at various values of @xmath12 . we first confine ourselves to voltage at fmr of py . voltage monotonically decreases in magnitude as @xmath12 approaches zero and changes its sign across @xmath33 deg .. this behavior is consistent with eq . ( [ ishe ] ) because @xmath34 holds [ see fig . 1(b ) ] . to fit the voltage spectra near fmr of py , we used a sum of symmetric and asymmetric lorentzian functions@xcite @xmath35 here , @xmath36 is the full width at half maximum and @xmath37 ( @xmath38 ) is the magnitude of a symmetric ( asymmetric ) component of voltage . @xmath37 corresponds to ishe and @xmath38 to the anomalous hall effect in this study . 3(b ) bottom shows that the @xmath16 dependence of @xmath37 is roughly represented by a sine curve and reveals no difference between the yig / pt / py trilayer film and a py / pt bilayer film . we next focus on voltage at fmr of yig for @xmath39 deg . 3(a ) shows that as @xmath12 decreases from 90 deg . , voltage magnitude decreases and almost vanishes at 30 deg .. notable is that voltage of the opposite sign reappears at 20 deg .. this reversal of sign can not be explained by spin pumping in a yig / pt bilayer film . application of eq . ( 1 ) to the bilayer film yields voltage of a constant sign for @xmath39 deg . because @xmath40 holds . an unconventional feature is not limited to the sign of voltage : the upper inset in fig . 3(a ) clearly shows enhancement of voltage near @xmath41 deg , where the resonance fields of yig and py are equal . anomalies in sign and magnitude are summarized in fig . 3(b ) and ( c ) . at this stage , we can not conclude that the unconventional voltage originates in spin pumping at the yig / pt interface . we also need to consider spin pumping at the py / pt interface ; yig and py can affect mutual dynamics , or spin pumping via magnetic interaction . we note that the observed unconventional voltage can not be attributed to heating effects due to microwave absorption . the microwave absorption at fmr heats the magnetic layer and may cause a heat current flowing from the magnetic layer to the pt layer . eventually dc voltage is induced by thermoelectric effects such as the ordinary nernst effect , the anomalous nernst effect and the longitudinal spin seebeck effect@xcite ( lsse ) . the dc voltage is proportional to @xmath42 ( ordinary nernst effect ) and @xmath43 ( anomalous nernst effect and lsse ) , where @xmath44 denotes temperature . the magnitude of @xmath45 is proportional to the microwave absorption at fmr @xmath46 ; thus the dc voltage is proportional to @xmath47 ( ordinary nernst effect ) and @xmath48 ( anomalous nernst effect and lsse ) . since @xmath46 decreases as @xmath12 approaches zero , the heating effects can not reproduce the observed @xmath12 dependence of voltage . voltage anomalies appearing when @xmath49 suggest that magnetization dynamics , or spin pumping is affected by a magnetic coupling between the ferromagnets . one of the possibilities is a dynamic exchange coupling@xcite , a coupling via spin currents mediated by conduction electrons in an intermediate nm layer . this coupling persists on the spin - diffusion length scale in the nm layer and may exist in the yig / pt / py trilayer film , because pt has a spin diffusion length of around 10 nm . to examine a dynamic exchange coupling , we prepared a yig/@xmath50 ( 20 nm thick)/pt ( 5 nm thick)/py ( 10 nm thick ) quadrilayer film . although a spin current is interrupted across the yig / pt interface , voltage anomalies persist near @xmath51 deg .. moreover , the quadrilayer film reveals the similar @xmath16 dependence of @xmath37 at fmr of yig as observed in the trilayer film [ see fig . the persistence and the angular dependence both suggest that a dynamic exchange coupling is irrelevant to the trilayer film . the voltage anomalies in the yig/@xmath50/pt / py quadrilayer film suggest that unconventional voltage originates in spin pumping at the py / pt interface . to reinforce this argument , we prepared a yig / pt ( 5 nm thick)/@xmath50 ( 20 nm thick)/py ( 10 nm thick ) quadrilayer film , where the @xmath50 layer suppresses a dynamic exchange coupling by interrupting a spin current across the py / pt interface . the unconventional voltage is found to disappear [ see fig . this result further proves that unconventional voltage originates in spin pumping at the py / pt interface . we need to explain at least the following four features of the unconventional voltage : ( feature 1 ) the unconventional voltage persists even when the ferromagnets are separated by up to about 20 nm ; ( feature 2 ) the voltage originates in spin pumping at the py / pt interface ; ( feature 3 ) the voltage appears when both the ferromagnets are excited simultaneously ; ( feature 4 ) the voltage takes extrema at @xmath12 where yig and py are equal in @xmath29 . the long - range nature ( feature 1 ) suggests that a dipolar coupling may be relevant to the unconventional voltage . as shown below , however , this coupling alone is unlikely to yield our results consistently . magnetization dynamics in the yig layer yields a time - dependent dipole field@xcite . a sum of the time - dependent dipole field and a microwave field act on magnetization in the py layer . the increased driving field may enhance the magnetization dynamics in the py layer . note that yig and py are greatly different in volume , the saturation magnetization , and the damping constant . these differences may enhance spin pumping of py while having little influence on that of yig . on the other hand , a static dipole coupling should lead to a resonance - field shift in the yig / pt / py trilayer film . this shift will become prominent especially when yig and py are excited simultaneously.@xcite at least we did not observe resonance - field shifts greater than an error range of 10 mt . it is unreasonable to infer that a dynamic dipole field enhances magnetization dynamics while a static counterpart has little influence on resonance fields . the unconventional voltage will need contributing factors other than a dipolar coupling . we investigated spin pumping in a yig / pt / py trilayer film by means of ishe . unconventional voltage appears when both the ferromagnets are excited simultaneously by a microwave . the voltage is attributed to spin pumping at the py / pt interface . the voltage persists when a dynamic exchange coupling is suppressed . several features are pointed out to characterize the unconventional voltage . the voltage persists with the ferromagnets separated by up to 20 nm ( long - range feature ) . the voltage magnitude increases when yig and py are equal in @xmath29 ( dynamic feature ) . considering these features , a time - dependent dipolar coupling was proposed to explain the origin of the unconventional voltage . however , our experiments show that a static dipolar coupling does not shift resonance fields . this result appears inconsistent with the proposed mechanism . the origin of the unconventional voltage requires further elucidation . this work was supported by crest - jst `` creaction of nanosystems with novel functions through process integration '' , a grant - in - aid for scientific research ( a ) ( 24244051 ) from mext , japan , the murata science foundation , the funding program for next generation world - leading researchers , the asahi glass foundation , the noguchi institute , and the mitsubishi foundation . is a static magnetic field . ( b ) definitions of the magnetic - field angle @xmath12 and the magnetization angle @xmath52 . @xmath53 is a static magnetic field and @xmath54 is magnetization . ( c ) a schematic illustration of spin pumping at the py / pt ( left ) and yig / pt interfaces ( right ) . @xmath54 denotes magnetization ; @xmath4 the spatial direction of a spin current ; @xmath15 the polarization vector ; @xmath2 an ishe - induced electric field . ] dependence of the microwave abosrption signal @xmath55 . @xmath29 is a resonance field . ( b ) the in - plane field @xmath56 dependence of dc voltage @xmath57 for the yig / pt / py trilayer film ( red and blue circles ) and the yig / cu / py trilayer film ( black circles ) . the inset shows the @xmath56 dependence of @xmath57 near fmr of yig . ( c ) the field - angle @xmath12 dependence of the resonance field @xmath29 . the filled circles represent data for the yig / pt / py trilayer film ; the open triangles for the yig / pt and py / pt bilayer films . data are blue for yig and red for py . ] dependence of dc voltage @xmath57 for the yig / pt / py trilayer film at magnetic - field angles @xmath12 between @xmath58 and @xmath59 deg . the blue ( red ) dashed line represents the resonance field of yig ( py ) . orange ( black ) circles indicate unconventional ( conventional ) voltage at fmr of yig . the upper inset shows the @xmath56 dependence of @xmath57 near fmr of yig . the lower inset shows experimental data ( open circles ) and fitting ( dashed lines ) at @xmath60 deg . red and blue dashed lines are a symmetric lorenzian function and the green dashed line is an asymmetric lorenzian function . ( b ) the magnetic - field angle @xmath12 dependence of the resonance - field difference @xmath61 . @xmath61 is defined as @xmath62 . orange circles indicate where yig and py are equal in @xmath29 . ( c ) the magnetic - field angle @xmath12 dependence of the symmetric component of dc voltage @xmath37 at fmr of yig . @xmath63 is the normalized dc voltage . circles represent data for the yig / pt / py trilayer film ; triangles for the yig / pt bilayer film . orange circles indicate where yig and py are equal in @xmath64 . ( d ) the magnetic - field angle @xmath12 dependence of the symmetric component of dc voltage @xmath37 at fmr of py . @xmath63 is the normalized dc voltage . circles represent data for the yig / pt / py trilayer film ; triangles for the yig / pt bilayer film . ] dependence of the symmetric component of dc voltage @xmath37 at fmr of yig . the black circles represent data for the yig/@xmath65/pt / py quadrilayer film ; the dark brown circles for the yig / pt/@xmath65/py quadrilayer films ; the sky blue circles for the yig / pt / py trilayer film . the inset shows the magnetic field @xmath56 dependence of @xmath37 near fmr of yig . @xmath12 is such that yig and py are equal in the resonance field @xmath29 . the red ( blue ) arrow stresses a positive ( negative ) sign of @xmath37 . ( b ) the magnetic - field angle @xmath12 dependence of the symmetric component of dc voltage @xmath37 at fmr of py . the black circles represent data for the yig/@xmath65/pt / py quadrilayer film ; the dark brown circles for the yig / pt/@xmath65/py quadrilayer films ; the light pink circles for the yig / pt / py trilayer film . ]	we study spin pumping in a @xmath0(yig)/pt/@xmath1(py ) trilayer film by means of the inverse spin hall effect ( ishe ) . when the ferromagnets are not excited simultaneously by a microwave , ishe - induced voltage is of the opposite sign at each ferromagnetic resonance ( fmr ) . the opposite sign is consistent with spin pumping of bilayer films . on the other hand , the voltage is of the same sign at each fmr when both the ferromagnets are excited simultaneously . futhermore , the voltage greatly increases in magnitude . the observed voltage is unconventional ; neither its sign nor magnitude can be expected from spin pumping of bilayer films . control experiments show that the unconventional voltage is dominantly induced by spin pumping at the py / pt interface . interaction between yig and py layers is a possible origin of the unconventional voltage .
the standard model of cosmology predicts a structure formation scenario driven by cold dark matter ( e.g. , * ? ? ? * ) , where galaxies form from molecular gas cooling within growing dark matter haloes . hence , understanding the correlation between different properties of the dark matter haloes is imperative to build up a comprehensive picture of galaxy evolution . many authors have explored the correlation between dark - halo properties , such as mass , spin and shape , both in low- ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and high - redshift ( e.g. , * ? ? ? * ; * ? ? ? * ) regimes . estimating the strength of these correlations is critical to support semi - analytical and halo occupation models , which assume the mass as determinant factor of the halo properties ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? nevertheless , alternative approaches , based on principal components analysis ( pca ) , found that concentration is a key parameter , contrary to what expected before @xcite , and stressed the need for further investigations . pca belongs to a family of techniques ideal to explore high - dimensional data . the method consists in projecting the data into a low - dimensional form , retaining as much information as possible ( e.g. , * ? ? ? hence , pca emerges as a natural technique to investigate correlation and temporal evolution of halo properties . because of its versatility , pca has been applied to a broad range of astronomical studies , such as stellar , galaxy and quasar spectra ( e.g. , * ? ? ? * ; * ? ? ? * ) , galaxy properties @xcite , hubble parameter and cosmic star formation ( sf ) reconstruction ( e.g. , * ? ? ? * ; * ? ? ? * ) , and supernova ( sn ) photometric classification @xcite . despite its generality , pca is not the only way to handle huge data sets , and the growth in complexity of scientific experimental data makes the ability to extract newsworthy and meaningful information an endeavor per se . the yearning for novel methodologies of data - intensive science gave rise to the so - called fourth research paradigm ( e.g. , * ? ? ? data mining methods have been used in many areas of knowledge such as genetics ( e.g. , * ? ? ? * ) and financial marketing decisions ( e.g. , * ? ? ? * ) , and their importance for astronomy has been recently highlighted as well ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? likewise observations , cosmological simulations are continuously increasing in complexity , lessening the distance between observed and synthetic data ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? none the less , the application of data - mining to cosmological simulations remains a _ terra incognita_. in this work , we investigate the statistical properties of baryons inside high - redshift haloes , including detailed chemistry , gas physics and stellar feedback . we make use of robust pca ( rpca ) and maximal information coefficient ( mic ) to study a set of various halo parameters . rpca represents a generalization of the standard pca , whose advantage is its resilience to outliers and skewed data , while mic is expected to be the correlation analysis of the 21st century @xcite , in particular due to mic ability in quantifying general associations between variables . therefore , this project represents the first application of mic to @xmath5-body / hydro simulations , and the first use of pca to explore the low - mass end of the halo mass function and the birth of the first galaxies . the outline of this paper is as follows . in section [ sec_sim ] , we describe the cosmological simulations and their outcomes . in section [ sec_statistics ] , we describe the statistical methods . in section [ sec_results ] , we present our analysis and main results . finally , in section [ sec_conclusions ] , we present our conclusions . we analyzed the results of a cosmological @xmath5-body , hydrodynamical , chemistry simulation based on @xcite ( see also @xcite ) , that was run by means of a modified version of the smoothed - particle hydrodynamics code gadget2 @xcite . the modifications include relevant chemical network to self - consistently follow the evolution of e@xmath6 , h , h@xmath7 , h@xmath6 , he , he@xmath8 , he@xmath9 , h@xmath10 , h@xmath11 , d , d@xmath7 , hd , heh@xmath7 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , ultraviolet background radiation , metal pollution according to proper stellar yields ( he , c , o , si , fe , mg , s , etc . ) , lifetimes and stellar population for population iii ( pop iii ) and population ii / i ( pop ii / i ) regimes @xcite , radiative gas cooling from molecular , resonant and fine - structure transitions ( e.g. * ? ? ? * and references therein ) and stellar feedback @xcite . the transition from the pop iii to the pop ii / i regime is determined by the value of the gas metallicity ( @xmath12 ) compared to the critical value @xmath13 ( e.g. , * ? ? ? * ; * ? ? ? * ) , assumed to be @xmath14 . the cosmic field is sampled at redshift @xmath15 , adopting standard cosmological parameters : @xmath16 km / s / mpc and @xmath17 . we considered snapshots in the range @xmath18 , within a cubic volume of comoving side 0.7 mpc , and @xmath19 particles per gas and dark - matter species corresponding to particle masses of 42 and @xmath20 , respectively . the identification of the simulated objects is done by applying a friends - of - friends ( fof ) technique with linking length equal to 20 per cent the mean interparticle separation and substructures are identified by using a subfind algorithm @xcite , which discriminates among bound and non - bound particles . the halo characteristics , such as position , velocity , dark matter and baryonic properties , are computed and stored at each redshift . the simulation outcomes investigated here consist of seven parameters : dark - matter mass ( @xmath0 ) , gas mass ( @xmath1 ) , stellar mass ( @xmath2 ) , star formation rate ( sfr ) , @xmath12 , gas temperature ( _ t _ ) , and gas molecular fraction @xmath21 . we refer the reader to previous works , where more details and additional analyses about halo spin and shape distribution @xcite , feedback mechanisms @xcite , primordial streaming motions @xcite , non - standard cosmologies @xcite , high-@xmath22 luminosity function @xcite , early gamma ray bursts- @xcite and sne - host properties @xcite , ly@xmath23 emitters @xcite and damped ly@xmath23 ( dla ) system chemical content @xcite are presented and discussed . the total dataset is composed by a few thousands haloes at very high redshift , @xmath24 , and reaches about 25000 primordial objects at @xmath25 . in order to avoid numerical artifacts , created by a poor number of gas particles @xcite , we selected only those structures in which the gas content is resolved with at least @xmath26 gas particles . this usually corresponds to selecting only objects with a total number of particles of at least @xmath27 . the remaining data are therefore composed of @xmath281680 haloes in the whole redshift range , of which @xmath29 are at @xmath30 . [ fig : variables_evol ] shows the probability distribution function ( pdf ) for the seven halo parameters : @xmath31 , sfr , @xmath32 , @xmath33 and @xmath12 at each redshift . they are portrayed by a violin plot . each violin centre represents the median of the distribution , while the shape , its mirrored pdf . a visual inspection in fig . [ fig : variables_evol ] indicates the first stages of significant sf activity around @xmath34 , giving rise to a subsequent boost in metal enrichment at @xmath35 , and a similar growth of @xmath36 in the same redshift range . just after this episode , we can see the rapid spread in the @xmath33 variance , peaking few orders of magnitude above average . the masses of the haloes range between @xmath37 and @xmath38 . typical temperatures range from 500 to @xmath39 k , where @xmath40 shapes the thermal conditions of early objects . hotter temperatures are due to the thermal effects of sn explosions that heat and enrich the gas in nearby smaller haloes . ( red ) , @xmath41 ( blue ) , @xmath42 ( green ) , _ sfr _ ( magenta ) , t ( orange ) , @xmath43 ( cyan ) and z ( khaki ) . they are portrayed by a violin plot . each violin centre represents the median of the distribution , while the shape , its mirrored pdf . ] the ultimate goal of pca is to reduce the dimensionality of a multivariate data , while explaining the data variance with as few principal components ( pcs ) as possible . pca belongs to a class of projection - pursuit ( pp ; e.g. , * ? ? ? * ) methods , whose aim is to detect structures in multidimensional data by projecting them into a lower - dimensional subspace ( lds ) . the lds is selected by maximizing a projection index ( pi ) , where pi represents an _ interesting feature _ in the data ( trends , clusters , hyper - surfaces , anomalies , etc . ) . the particular case where variance ( @xmath44 ) is taken as a pi leads to the classical version of pca ) , with the resulting eigenvectors corresponding to pcs and the resulting eigenvalues to the variance _ explained _ by the pcs . + the eigenvector corresponding to the largest eigenvalue gives the direction of greatest variance ( pc1 ) , the second largest eigenvalue gives the direction of the next highest variance ( pc2 ) , and so on . since covariance matrices are symmetric positive semidefinite , the eigenbasis is orthonormal ( spectral theorem ) . ] . given @xmath45 measurements @xmath46 , all of them column vectors of dimension @xmath47 , the first pc is obtained by finding a unit vector @xmath48 which maximizes the variance of the data projected on it : @xmath49 where @xmath50 is the transpose operation and @xmath51 is the direction of the first pc is the set of values of @xmath52 for which the function @xmath53 attains its largest value . ] . once we have computed the @xmath54th pc , the direction of the @xmath55th component , for @xmath56 , is given by @xmath57 where the condition of each pc to be orthogonal to all previous ones , ensures a new uncorrelated basis . in spite of these attractive properties , pca has some critical drawbacks as the sensitivity to outliers ( e.g. , * ? ? ? * ) , and inability to deal with missing data ( e.g. , * ? ? ? in order to overcome this limitation , several robust versions were created based on the pp principle . instead of taking the variance as a pi in equation ( [ eq : pc1 ] ) , a robust measure of variance is taken . hereafter , we will refer the standard variance as @xmath58 and robust variance as @xmath59 . two common measures of robust variance @xcite are the median absolute deviation ( mad ; e.g. , * ? ? ? * ) , @xmath60 and the first quartile of the pairwise differences between all data points ( _ q _ ; e.g. , * ? ? ? * ) , @xmath61 where @xmath62 is a given univariate dataset and the square of mad or _ q _ gives a robust variance . hereafter all calculations of the pcs are performed using the grid search base algorithm @xcite with mad , but using _ q _ has no influence on our results . also note that before applying the pca , we standardize the halo properties by subtracting the mean and dividing by the standard deviation . therefore we are formally using the correlation matrix that can be seen as the covariance matrix of standardized variables . the maximal information - based non - parametric exploration ( mine ) statistics represent a novel family of techniques to identify and characterize general relationships in data sets @xcite . mine introduce mic as a new measure of dependence between two variables , which possesses two desired properties for data exploration : ( i ) generality , the ability to capture a broad range of associations and functional relationships ) measures the strength of monotonicity between paired data . ] ; ( ii ) equitability , the ability to give similar scores to equally noisy relationships of different types . a lack of equitability introduces a strong bias and entire classes of relationships may be missed @xcite . ] . mic measures the strength of general associations , based on the mutual information ( @xmath63 ) between two random variables @xmath64 and @xmath65 : @xmath66 where @xmath67 and @xmath68 are the marginal pdfs of @xmath64 and @xmath65 , and @xmath69 is the joint pdf . + consider d a finite set of ordered pairs , @xmath70 , partitioned into a @xmath52-by-@xmath71 grid of variable size , @xmath72 , such that there are @xmath52-bins spanning @xmath73 and @xmath71-bins covering @xmath74 , respectively . + the pdf of a particular grid cell is proportional to the number of data points inside that cell . we can define a characteristic matrix @xmath75 of a set @xmath76 as @xmath77 + representing the highest normalized mutual informations of @xmath76 . the mic of a set @xmath76 is then defined as @xmath78 representing the maximum value of @xmath79 subject to @xmath80 , where the function @xmath81 was empirically determined by @xcite exponent value represents a compromise since high values of @xmath82 lead to non - zero scores even for random data , as each point gets its own cell , while low values only probe simple patterns . ] . hereafter we discuss the relations between halo properties and their relative importance . our matrix is composed by 1680 haloes , spanning the redshift range @xmath83 , with @xmath28 200 ( 30 ) haloes at @xmath22 = 9 ( 19 ) , each halo containing at least @xmath27 particles . each row of the matrix represents a halo and each column represents one of the halo properties . pca probes the entire matrix at once . on the other hand , mic is a pair - variable comparison , therefore requiring @xmath84 operations , with @xmath5 being the number of halo properties . it is worth to highlight here that each approach has its own advantages and disadvantages . pca is suitable for high - dimensional data , when a pair comparison becomes unfeasible , however the method only searches for linear relationships . mic , instead , finds general associations in data structures , but may be impractical to deal with a large amount of parameters . in order to better understand the pros and cons of using rpca , we first start the analysis with the standard pca . fig . [ fig : pca_evol ] shows the contribution of the first three pcs to @xmath85 , as a function of redshift . three pcs account for more than @xmath86 per cent of @xmath85 at any redshift , while two pcs explain more than @xmath87 per cent except at @xmath88 , when the contribution drops to @xmath89 per cent . the sharp variation of the pcs around @xmath90 acts as a smoking gun for a global cosmological event . indeed , this is a direct consequence of first sf episodes and the interplay between chemical and mechanical feedback from the first stars , that takes place around @xmath91 ( e.g. , * ? ? ? * ; * ? ? ? as molecules are produced over time , they lead to gas collapse , stellar formation and metal pollution , with consequent back reaction on the thermal behavior of the surrounding gas ( see e.g. , * ? ? ? * ; * ? ? ? this redshift range represents an epoch of fast and turbulent growth of the metal filling factor , from @xmath92 at @xmath93 to @xmath94 at @xmath88 ( see fig . 1 from * ? ? ? at the beginning , only the gas at high densities is affected by metal enrichment , due to sf concentration in these regions . as sf and metal spreading proceed , the surrounding lower density environments are affected as well . sne heat high - density gas within star - forming sites and , consequently , hot low - density gas is ejected from star - forming regions by sn winds . the contribution of each pc dramatically changes if we use rpca instead . the clearest advantage is the amount of variance explained by each component ( hereafter , when necessary to avoid ambiguity , the pcs from rpca analysis will be referred as rpcs ) . rpc1 accounts for no less than @xmath95 per cent of the @xmath59 anytime , whilst two rpcs account for more than @xmath96 per cent . moreover , the rpc2 contribution mostly stands out between at @xmath97 and @xmath98 . albeit contributing differently to the total variance , the general behavior of pc1 and pc2 is similar to the rpc1 and rpc2 , as well as the physical interpretation . but rpca assigns less weight to the baryonic properties , suggesting the halo mass as the most significant factor . this difference occurs because even a small fraction of large errors can cause arbitrary corruption in pca s estimate . for instance , pca is more sensitive to rapid variations of the halo chemical properties , having a steeper reaction in their first pcs . thus , as expected rpca surpass pca in their ultimate goal : reduce the system dimensionality . nevertheless , the greatest power to synthesize information carries the assumption that outliers are caused by corrupted data , which is not always the case . this potential drawback will be better understood looking at the contribution of each variables to the @xmath55-th pcs as discussed in the following . [ fig : pc_bar ] shows the relative contribution of each parameter to the first three pcs ( rpcs ) on the left ( right ) side . for the pca case , @xmath0 and @xmath1 dominate pc1 at @xmath99 ( no less than @xmath100 62 per cent ) , followed by a smaller contribution of sfr and @xmath33 nevertheless , as gas collapses into potential wells , the relative contribution from @xmath1 increases , surpassing @xmath0 at @xmath101 . the dominant contribution of @xmath12 and @xmath33 to pc1 at @xmath102 indicates a critical epoch for the cosmic chemical enrichment ( see also discussion above ) , triggered by a rapid variation of @xmath33 , followed by a wide metal pollution at @xmath103 . after a decline in the chemical enrichment rate , a second peak in @xmath12 occurs at @xmath104 . this self - regulated , oscillatory behavior is caused by the simultaneous coexistence of cold pristine - gas inflows and hot metal enriched outflows that create hydro instabilities and turbulent patterns with reynolds numbers @xmath105 ( see e.g. fig . 2 from * ? ? ? . finally at @xmath30 , @xmath0 and @xmath1 have become almost subdominant , since pc1 is mainly led by @xmath32 and @xmath12 , as a result of the ongoing cosmic heating from sf and thermal feedback . the dominance by @xmath32 to pc1 at this redshift occurs due to the presence of some small ( see fig . [ fig : variables_evol ] ) , high - temperature objects , whose properties are contaminated by hot enriched material at @xmath106 k. an inspection of pc2 reveals the _ supporting roles _ during the galaxy formation process . the pc1 peak in @xmath12 at redshift 13 is preceded by a strong contribution of _ sfr _ and halo masses to pc2 , while the second pc1 peak in @xmath12 , around @xmath107 , is anticipated by an increasing contribution to pc2 from the formed stars , which later explode as sne and start the metal enrichment of the universe . the first rise of pc2 at @xmath108 , dominated by sfr , occurs because the protogalaxies at this epoch are experiencing the first bursts of sf . nevertheless , not all of them have necessarily formed stars already . whilst the second peak is composed of a more balanced contribution from sfr and @xmath36 . the oscillatory behavior might be caused by the competitive effects of different feedback mechanisms : the gas undergoing sf is heated by sn explosions and it is inhibited to continuously form stars ( mostly in smaller structures that suffer significantly gas evaporation processes ) ; while shock compressions and spreading of metals in the medium enhance gas cooling capabilities and consequently induce more sf . the former preferentially occurs in bigger objects that can keep and re - process their metals because of the deeper potential wells . pc3 is nearly negligible in the whole redshift range aside @xmath109 , where @xmath110 dominates the general behavior . this epoch is preceded by a significant contribution from @xmath36 at @xmath111 . a comparison with fig . @xmath112 reveals that this behavior coincides with a growth in the @xmath33 variance at the same redshift . this indicates a transition in the regular trend of increasing @xmath33 with increasing mass at @xmath113 , when initial collapse phases boost @xmath33 up to @xmath114 . this rapid growth of @xmath33 preferentially occurs in galaxies of @xmath115 , that are forming their first stars and have not been previously affected by feedback mechanisms . at @xmath116 , feedback effects from pop iii forming galaxies become responsible for increasing the variance of @xmath33 by several orders of magnitude , either by dissociating molecules , or by partially enhancing their formation by shocks and gas compression ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? looking the rpca , the rpc1 is dominated by halo masses during all cosmic evolution ( no less than 68 per cent ) , with other baryonic properties relegated to rpcs of higher orders . some caution is needed to interpret these results . the higher level of compressibility presented by rpca is a direct consequence of attributing a smaller weight to rare events . therefore , if one intends to describe all haloes properties using the fewest parameters possible , rpca appears to succeed , since it states that as a first approximation , the total halo mass is the main factor to describe all other properties . the mass determines the potential well and consequently the ability of the halo to form stars , retain the metals , etc , therefore roughly dictating the baryonic dynamics at a first sight . since rpca ascribes a lower weight to the tails of each parameter distribution , the physical interpretation may become less evident for the highest rpcs . however , we can still see the importance of @xmath12 , @xmath33 and sfr , with the difference that now they are considered second order effects , hence starting to be dominant from the rpc2 forward . to better understand these differences between rpca and pca we discuss the strength with which each variable is related to one another as follows . [ fig : scatter ] shows how the seven halo properties correlate to each other . the main diagonal of fig . [ fig : scatter ] shows the density distribution of each variable at different redshifts ( a zoomed version of half - violin presented in fig . [ fig : variables_evol ] ) . the majority of the parameters have a well behaved distribution , with small variations in its shape during the cosmic evolution , while quantities related to the stellar feedback ( @xmath117 ) have their distribution shaped during the transition from a regime without sf activity at @xmath118 to the burst of sfr around @xmath116 . the lower triangular part of the panel shows scatter plots for each variable combination colored accordingly to their redshift . [ fig : scatter ] fig . [ fig : mine2 ] shows mic and @xmath119 for each combination of parameters as a function of redshift , because of the high number of zeros in the matrix makes the correlation measurements unreliable . ] . at high redshift , due to the poor statistics ( less than 30 haloes at @xmath120 , with a considerably amount of null parameters ) , most variables are uncorrelated , receiving a low score by both @xmath119 and mic . as expected @xmath1 , @xmath0 and @xmath32 are strongly correlated , receiving the highest values . this is consistent with the fact that pc1 dominates at @xmath121 and is basically dictated by @xmath122 and @xmath123 . the result suggests that at higher redshifts , haloes are much simpler objects and their properties are basically controlled by their masses . comparing with fig . [ fig : pc_bar ] , it seems that the correlation between halo mass and @xmath32 shows a better agreement with rpca , which makes of @xmath32 a factor almost as important as @xmath1 and @xmath0 in the determination of rpc1 . + the molecular content , which is directly dependent on the local gas density and _ t _ , shows a correlation with @xmath12 that increases at lower redshifts until @xmath124 . this trend is in agreement with the dominance of @xmath33 and @xmath12 on pc1 and rpc2 at @xmath125 , caused by the increase in the contribution of the _ sfr _ to pc2 and rpc2 at earlier redshifts . + at @xmath126 , @xmath33 keeps a regular trend of increasing with halo mass . nevertheless , the _ sf _ activity at @xmath127 leads to a dispersion of @xmath33 followed by a metal enrichment process , as discussed in section [ sec_results ] . also @xmath1 shows a stronger correlation with @xmath33 than with other quantities like sfr and @xmath12 , which indicates the crucial role of @xmath33 to initiate sf and consequent metal pollution from pop iii and pop ii / i regimes in primordial galaxies . comparing with fig . [ fig : pc_bar ] , we see that rpca better apprehends this effect . at high redshift , with the exception of @xmath128 , where the peak in rpc2 is caused by the first stages of metal enrichment ( fig . [ fig : variables_evol ] ) , @xmath33 maintains a dominant contribution to rpc2 , together with halo mass . + the correlation between sfr with @xmath123 and @xmath122 is roughly linear , increasing at later times . this may be explained by the wider spread of sfr in low massive haloes at @xmath129 , which is caused by gas evaporation processes due to sn explosions , in contrast with later structures that have a more sustained sf activity . albeit both pca and rpca are sensitive to this effect , rpca ascribes a lower weight to the sfr than to @xmath33 , in accordance to the correlation analysis . a surprising disagreement between mic and @xmath130 appears when comparing @xmath12 , @xmath2 and sfr . @xmath119 suggests a nearly perfect correlation between @xmath12 and @xmath2 , while mic found no significant association at the highest redshifts . this highlights the robustness of mic with skewed and sparse data . in this redshift range , @xmath131 , there are very few haloes with non - null @xmath12 and @xmath2 values ( fig . [ fig : variables_evol ] ) . therefore , the high @xmath119 score for these two quantities is misleading , as confirmed by a visual inspection of their corresponding distributions ( figs . [ fig : variables_evol ] and [ fig : scatter ] ) . the same argument holds for the comparison between @xmath12-sfr , and @xmath2-sfr . during the course of cosmic evolution though , the correlations between the properties of the haloes tighten and both @xmath119 and mic converge for most of them at @xmath132 ( with @xmath119 slightly overestimating the strength of correlation compared to mic ) , as shown in fig . [ fig : mine2 ] . correlation is shown on the left scale with orange bars , while the mic is represented by green bars on the right . ] we investigate the redshift evolution of the gas properties of primordial galaxies using rpca and mic statistics making a comprehensive comparison with standard approaches . this is the first attempt to probe the baryon properties of early mini - haloes and the effects of feedback processes by means of a highly solid statistical approach . we explore the correlation of different baryonic properties as expected from numerical @xmath5-body , hydrodynamical , chemistry simulations including gas molecular and atomic cooling , sf , stellar evolution , metal spreading and feedback effects . the wide range of redshifts analyzed here ( @xmath18 ) allowed us to perform an unprecedented study of the temporal evolution of the pc contribution to the total variance of the halo properties . the standard pca needs two pcs to explain more than 92 per cent of the data variance ( in the greater part of redshifts studied here ) with pc1 dropping below 50 per cent at lower redshifts . the first rpc from rpca analysis explains no less than 84 per cent of all data variance anytime , with two first rpcs explaining more than 95 per cent of the total robust variance . first sf episodes and feedback mechanisms cause a drop of pc1 at @xmath133 , when a sharp variation in the pcs behavior marks the onset of cosmic metal enrichment . at @xmath99 the halo properties are basically dictated by the halo mass . among the advantages in using rpca is the possibility to increase the capability to reduce the dimensionality of the original dataset , although at the cost to be less sensitive to rare events that may be physically relevant . since rpca ranks the contribution of variables to the rpcs in better agreement with their levels of correlation . it seems to be in better agreement with our independent mic and @xmath130 correlation analysis . an inspection in the first and second pcs reveals some interesting facts . the pc1 peak in @xmath12 at redshift 13 is preceded by a strong contribution of sfr and halo masses to pc2 . while the second pc1 peak in @xmath12 , around @xmath107 , is anticipated by an increasing contribution to pc2 by the formed stars , which later explode as sne and enrich the universe . this indicates the importance of stellar evolution in shaping baryon properties in primordial haloes . a similar trend holds for rpca although attenuated by the smoothing effect created by the use of robust statistics . it is important to note , however , that the relatively small number of haloes studied here might lessen the robustness of our results at very high redshifts . therefore , future investigations of similar techniques into larger simulations boxes is highly recommended . overall @xmath119 agrees reasonably with mic , but mic seems to be more robust to study highly sparse data regimes ( like at early epochs ) . all gas properties , aside @xmath1 , @xmath0 and @xmath32 , are weakly correlated at high redshift . nevertheless , due to the interplay between chemical and mechanical feedback from the ongoing stellar formation and the consequent back reaction on the thermal behavior of the surrounding medium , baryonic quantities start to present a moderate to high level of correlation as redshift decreases . in particular , @xmath33 shows the highest level of correlation with @xmath1 , followed by @xmath32 , sfr , @xmath36 and @xmath12 respectively . in general , structure formation processes depend not only on the dark matter halo properties , but also on the local thermodynamical state of the gas , which is , in turn , affected by cooling , _ sf _ and feedback . our analysis suggests that all the gaseous properties have a stronger correlation with @xmath1 than with @xmath0 , while @xmath1 has a deeper correlation with @xmath3 than with _ z _ or sfr . the relevance of the molecular content for the baryon properties represents the physical origin of gas collapse and concentration , crucial to initiate sf and consequent metal pollution from pop iii and pop ii / i regimes in primordial galaxies . + this work represents a leap forward in the statistical analysis of @xmath5-body / hydro simulations , performed by means of rpca and mic into a cosmological context . we therefore stress that the use of dimensionality reduction algorithms and mutual information based techniques in numerical simulations might be a precious instrument for future investigations , thanks to their potential to unveil non - trivial relationships , which may go undetected by standard methods . u.m . has received funding from the european union seventh framework programme ( fp7/2007 - 2013 ) under grant agreement n.267251 . for the bibliographic research we made use of the nasa astrophysics data system archive . we thank the referee by the very useful comments , helping to enhance the article presentation .	we present a novel approach , based on robust principal components analysis ( rpca ) and maximal information coefficient ( mic ) , to study the redshift dependence of halo baryonic properties . our data are composed of a set of different physical quantities for primordial minihaloes : dark - matter mass ( @xmath0 ) , gas mass ( @xmath1 ) , stellar mass ( @xmath2 ) , molecular fraction ( @xmath3 ) , metallicity ( _ z _ ) , star formation rate ( sfr ) and temperature . we find that @xmath0 and @xmath1 are dominant factors for variance , particularly at high redshift . nonetheless , with the emergence of the first stars and subsequent feedback mechanisms , @xmath3 , sfr and _ z _ start to have a more dominant role . standard pca gives three principal components ( pcs ) capable to explain more than 97 per cent of the data variance at any redshift ( two pcs usually accounting for no less than 92 per cent ) , whilst the first pc from the rpca analysis explains no less than 84 per cent of the total variance in the entire redshift range ( with two pcs explaining @xmath4 per cent anytime ) . our analysis also suggests that all the gaseous properties have a stronger correlation with @xmath1 than with @xmath0 , while @xmath1 has a deeper correlation with @xmath3 than with _ z _ or sfr . this indicates the crucial role of gas molecular content to initiate star formation and consequent metal pollution from population iii and population ii / i regimes in primordial galaxies . finally , a comparison between mic and spearman correlation coefficient shows that the former is a more reliable indicator when halo properties are weakly correlated . [ firstpage ] cosmology : large - scale structure of universe , early universe ; methods : statistical , n - body simulations
the measurements of cosmic microwave background ( cmb ) temperature anisotropies and large - scale structures have significantly improved in accuracy over the last decade @xcite . in particular , the recently released planck data @xcite showed that the primordial power spectrum of curvature perturbations is slightly red - tilted from the exact scale - invariance . this is consistent with the theoretical prediction of standard slow - roll inflation driven by a nearly flat potential of a scalar field @xmath7 ( called `` inflaton '' ) @xcite . while the wmap and planck data support the inflationary scenario overall , there are some anomalies in the data @xcite which are difficult to be addressed in the context of single - field slow - roll inflation . one of them is broken rotational invariance of the cmb perturbations @xcite . the power spectrum of curvature perturbations @xmath8 with broken statistical isotropy can be expressed in the form @xcite _ ( k)= p_^(0 ) ( k)(1+g_^2 _ k , v ) , [ anispe ] where @xmath9 is the comoving wave number , @xmath10 is the isotropic power spectrum , @xmath2 characterizes the deviation from the isotropy , @xmath11 is a privileged direction close to the ecliptic poles , and @xmath12 is the angle between @xmath9 and @xmath11 . from the wmap data , groeneboom _ et al . _ @xcite obtained the bound @xmath13 with the exclusion of @xmath14 at @xmath15 by including the cmb multipoles up to @xmath16 . there is still a possibility that some systematic effect such as the asymmetry of the instrument beam accounts for broken rotational invariance @xcite . ( 68%cl ) by using the planck data . this limit was derived after eliminating the asymmetry of the planck beam and the galactic foreground emission . while the isotropic power spectrum is consistent with the planck data , the anisotropy of the order of @xmath17 is not excluded yet . in this paper we allow for the possibility that @xmath6 can be of the order of 0.1 without using the new bound of kim and komatsu . ] if broken statistical isotropy is really present for primordial perturbations , we need to go beyond the slow - roll single - field inflationary scenario to explain the origin of statistical anisotropies @xcite . for the models in which the inflaton @xmath7 has a coupling with a vector kinetic term @xmath0 , there exists an attractor - like solution along which an anisotropic vector hair survives even during inflation @xcite ( see refs . @xcite for early related works and refs . @xcite for rich phenomenologies of anisotropic inflation ) . even if the background energy density of the vector field is suppressed relative to that of the inflaton , the isotropic power spectrum @xmath10 is modified to have the form ( [ anispe ] ) with a negative anisotropic parameter @xmath2 @xcite . moreover the non - linear estimator @xmath5 of scalar non - gaussianities can be as large as the order of 10 for the squeezed shape averaged over all directions of a wave number @xmath18 with @xmath19 @xcite ( see also refs . @xcite for the large non - gaussianities generated by vector fields ) . recently , the present authors showed that anisotropic inflation can be also realized for the model in which the inflaton couples to a two - form field with the kinetic term @xmath1 @xcite ( see ref . @xcite for an early proposal ) . in this case , the anisotropic power spectrum is given by eq . ( [ anispe ] ) with @xmath20 . since the sign of @xmath2 is opposite to that in the vector model , we can observationally distinguish between the two anisotropic inflationary scenarios . in ref . @xcite the bispectrum and trispectrum of curvature perturbations have been also evaluated in the two - form model by using the interacting hamiltonian picture @xcite . it was shown that , in the strict squeezed limit , the non - linear estimator @xmath5 vanishes and that , in the equilateral and enfolded limits , @xmath5 can be larger than the order of 10 . in this paper we place observational constraints on two anisotropic inflation models based on the vector and the two - form fields in the light of the recent planck data @xcite . we first derive the anisotropic power spectra of gravitational waves to evaluate the tensor - to - scalar ratio @xmath3 correctly . using the observational bounds of @xmath3 as well as the scalar spectral index @xmath4 constrained by the joint data analysis of planck and other measurements @xcite , we test for several representative models such as chaotic and natural inflation in the presence of anisotropic corrections to the scalar and tensor power spectra . we also compute the cross correlations between curvature perturbations and gravitational waves . while the cross correlation survives in the vector model , it vanishes in the two - form model . this property is useful to distinguish between the two anisotropic inflationary scenarios from the correlation between the observed temperature perturbation and the b - mode polarization ( tb correlation ) @xcite . we also revisit the estimation of the anisotropic scalar non - gaussianities for several different shapes of momentum dependence ( local , equilateral , and enfolded shapes ) in both the vector and the two - form models . in fact , we show that the local non - linear estimators in the strict squeezed limit ( @xmath21 with @xmath22 , @xmath23 ) vanish in both models . however , if we average over all directions of a wave number @xmath18 with @xmath19 for nearly squeezed shapes @xcite , we have non - zero values of @xmath5 for @xmath24 . taking this prescription , we show that the local non - linear estimator @xmath25 in the two - form model is smaller than that in the vector model by one order of magnitude for the same order of @xmath6 . this paper is organized as follows . in sec . [ backsec ] we review the background dynamics of anisotropic inflation for both the vector and the two - form models . in sec . [ perksec ] we derive anisotropic corrections to the scalar / tensor power spectra and also evaluate the cross correlation between curvature perturbations and gravitational waves . in sec . [ obssec ] we put constraints on several different inflaton potentials by using the 68%cl and 95%cl observational contours in the @xmath26 plane . in sec . [ nonsec ] we compare two anisotropic inflationary models from the non - linear estimator @xmath5 of scalar non - gaussianities . [ consec ] is devoted to conclusions . in this section we briefly review the background dynamics of anisotropic inflation for two classes of models in which the inflaton field @xmath7 couples to ( i ) a vector field @xmath27 @xcite and ( ii ) a two - form field @xmath28 @xcite . for suitable choices of couplings , the energy densities of the fields @xmath27 and @xmath28 can survive even during inflation . for details , we refer the reader to the review articles @xcite . let us first discuss the vector model given by the action @xmath30\ , , \label{eq : action}\end{aligned}\ ] ] where @xmath31 is the determinant of the metric @xmath32 , @xmath33 is the reduced planck mass , and @xmath34 is the scalar curvature . @xmath35 and @xmath36 are a potential and a kinetic function of the inflaton @xmath7 , respectively . the field strength of the vector field is characterized by @xmath37 now , we consider cosmological solutions in this system . without loosing the generality , one can take @xmath38-axis for the direction of the vector field . using the gauge invariance , we can express the vector field as @xmath39 where @xmath40 is a function of the cosmic time @xmath41 . even if initial inhomogeneities and isotropies in @xmath42 are present , it was shown that only the background field @xmath40 survives during anisotropic inflation in the bianchi type i universe @xcite . the same conclusion also applies to more general anisotropic backgrounds @xcite . there remains the rotational symmetry in the @xmath43 plane . hence , we can take the metric ansatz @xmath44 \ , \label{anisotropic - metric}\ ] ] where @xmath45 and @xmath46 are the isotropic scale factor and the spatial shear , respectively . it is easy to solve the equation of motion for @xmath42 as @xmath47 where @xmath48 is a constant of integration and an overdot denotes a derivative with respect to the cosmic time @xmath41 . the friedmann equation and the inflaton equation of motion are given , respectively , by @xmath49 \ , , \label{eq : hamiltonian - vector } \\ \ddot{\phi } & = & -3\dot{\alpha}\dot{\phi}-v_{,\phi } + p_a^2f(\phi)^{-3 } f_{,\phi}e^{-4\alpha -4\sigma}\ , , \label{inflatoneq}\end{aligned}\ ] ] where @xmath50 is the hubble expansion rate , and @xmath51 , @xmath52 . we define the energy density of the vector field as @xmath53 in order to sustain inflation , the potential energy @xmath35 of the inflaton needs to dominate over @xmath54 and @xmath55 . since the shear term @xmath56 should be suppressed relative to @xmath57 , the friedmann equation ( [ eq : hamiltonian - vector ] ) reads @xmath58 if @xmath59 is a rapidly decreasing function in time , it happens that @xmath55 does not decay . in particular , for the coupling @xmath60 the energy density ( [ rhoa ] ) stays nearly constant ( under the approximation @xmath61 ) . in this case , neglecting the contribution of the vector field on the r.h.s . ( [ inflatoneq ] ) , the field @xmath7 satisfies the slow - roll equation of motion @xmath62 . combining this equation with eq . ( [ friap ] ) , it follows that @xmath63 . then , the critical coupling ( [ fphi1 ] ) can be expressed as @xmath64 the shear term obeys the equation of motion @xmath65 if an anisotropy converges to a nearly constant value @xmath66 , then the quantity @xmath67 reduces to @xmath68 where we used eq . ( [ friap ] ) . in the following , let us estimate the ratio @xmath67 during anisotropic inflation . ignoring the @xmath69 term in eq . ( [ inflatoneq ] ) and combining it with eq . ( [ friap ] ) , it follows that @xmath70 where we used eq . ( [ fphive ] ) . neglecting the variation of @xmath71 relative to that of @xmath72 , we can integrate eq . ( [ dphial ] ) to give @xmath73 where @xmath74 is an integration constant . substituting eq . ( [ intvec ] ) into eq . ( [ rhoa ] ) , we have @xmath75 where @xmath76 , and we used the fact that @xmath77 under the slow - roll approximation . provided that @xmath78 , the ratio @xmath79 is nearly constant . as we will see in sec . [ perksec ] , the quantity @xmath79 is related to the anisotropic parameter @xmath2 appearing in the power spectrum of curvature perturbations . from eqs . ( [ sigvector ] ) and ( [ ra ] ) we obtain @xmath80 which means that the anisotropy survives during inflation for @xmath78 . the above discussion can be generalized to the coupling of the form @xmath81 where @xmath82 is a constant parameter . if the condition @xmath83 is satisfied , the energy density of the vector field grows as @xmath84 during the slow - roll phase of the inflaton . eventually , the vector field becomes relevant to the inflaton dynamics governed by eq . ( [ inflatoneq ] ) . however , when the third term on the r.h.s . of eq . ( [ inflatoneq ] ) dominates over the second term , the inflaton does not roll down , which makes @xmath85 decrease . hence the condition @xmath86 is always satisfied . in this way , there appears an attractor where inflation continues even when the vector field affects the inflaton dynamics . for the coupling ( [ geneco ] ) the shear to the hubble expansion rate approaches the value @xcite @xmath87 thus , inflation is slightly anisotropic and the energy density of the vector field never decays for the coupling ( [ geneco ] ) with @xmath88 . from eqs . ( [ sigvector ] ) and ( [ sighvector ] ) it follows that the ratio @xmath79 defined in eq . ( [ ra ] ) is nearly constant , i.e. , @xmath89 . in the attractor regime where the vector field contributes to the dynamics of the system there is the relation @xmath90 @xcite , in which case the coupling ( [ geneco ] ) evolves as @xmath91 , i.e. , the same as eq . ( [ fphi1 ] ) . in sec . [ perksec ] we shall use this property for the evaluation of two - point correlation functions of primordial perturbations . in the presence of a two - form field coupled to the inflaton @xcite , anisotropic hair can survive during inflation as in the vector model . in this case , the action reads @xmath93\ , , \label{oriaction}\ ] ] where the field strength @xmath94 is related to a two - form field @xmath28 , as @xmath95 without loss of generality , one can take the @xmath43 plane in the direction of the two - form field . then we can express @xmath28 in the form @xmath96 where @xmath97 is a function with respect to @xmath41 . since there exists a rotational symmetry in the @xmath43 plane , the metric can be parametrized by the form ( [ anisotropic - metric ] ) . the equation of motion for the two - form field @xmath98 is easily solved as @xmath99 where @xmath100 is a constant of integration . the hubble parameter @xmath101 and the inflaton @xmath7 obeys the equations of motion @xmath102\ , , \label{eq : hamiltonian}\\ \ddot{\phi } & = & -3\dot{\alpha}\dot{\phi}-v_{,\phi } + p_b^2f(\phi)^{-3 } f_{,\phi}e^{-2\alpha + 4\sigma}\ , , \label{inflatoneq2}\end{aligned}\ ] ] which are analogous to eqs . ( [ eq : hamiltonian - vector ] ) and ( [ inflatoneq ] ) with the difference of the exponential factors . defining the energy density of the two - form field as @xmath103 it follows that @xmath104 stays nearly constant for the coupling @xmath105 in the slow - roll regime of the inflaton there is the relation @xmath106 , so that the coupling ( [ couplingtwo ] ) can be expressed as @xmath107 since the shear @xmath108 satisfies the equation of motion @xmath109 the ratio @xmath67 should converge to @xmath110 where eq . ( [ friap ] ) is used . the sign of @xmath67 is opposite to that of the vector model . during anisotropic inflation the ratio ( [ sigtwo ] ) reads @xcite @xmath111 where @xmath74 is a constant . from eqs . ( [ sigtwo ] ) and ( [ sigtwo2 ] ) we have @xmath112 which is nearly constant for @xmath78 . the ratio ( [ rb ] ) appears in the anisotropic scalar power spectrum . we can generalize the coupling ( [ coutwo ] ) to the form @xmath113 where @xmath82 is a constant . for the super - critical case characterized by @xmath114 there is an attractor solution along which the ratio @xmath67 approaches the value @xcite @xmath115 whose sign is opposite to eq . ( [ sighvector ] ) . in this regime there is the relation @xmath116 , so that the coupling ( [ coutwo2 ] ) evolves as eq . ( [ couplingtwo ] ) . note that the ratio @xmath117 is nearly constant , i.e. , @xmath118 . the anisotropy induced by the two - form field is the prolate - type , in contrast to the vector field which induces the oblate - type anisotropy . this difference comes from the fact the vector @xmath27 extending to the @xmath38-direction speeds down the expansion in that direction , while the two - form field @xmath28 extending in the @xmath43 plane speeds down the expansion in the @xmath43-direction . we note that the couplings ( [ fphive ] ) and ( [ coutwo ] ) , which give rise to anisotropic inflation , are present for any slow - roll inflaton potentials . for the exponential potential @xmath119 the couplings @xmath36 are of the exponential forms @xmath120 @xcite , as they often appear as a dilatonic coupling in string theory . for some inflaton potentials the functions @xmath36 may not be so natural , but there is a possibility that such couplings can be motivated by future development of string theory or supergravity . it is worth mentioning that power - law kinetically driven anisotropic inflation ( k - inflation @xcite ) can be generally realized for the exponential couplings @xmath120 @xcite . in order to study observational signatures of anisotropic inflation , we need to know the two - point correlation functions of curvature perturbations and gravitational waves as well as their cross correlations . for the vector model the power spectrum of curvature perturbations was derived in refs . @xcite , whereas the anisotropic contribution to gravitational waves in the same model was discussed in refs . @xcite . for the two - form field model the present authors obtained the anisotropic scalar power spectrum @xcite , but the tensor power spectrum has not been derived yet . we also note that in the vector model the correlation between the temperature perturbation and the b - mode polarization was studied in ref . @xcite , but the cross - correlation between scalar and tensor perturbations in the two - form model has not been studied . here we provide all the formulas of these observables convenient to confront with observations . since the anisotropy of the expansion rate needs to be sufficiently small for the compatibility with observations , it is a good approximation to neglect the effect of the anisotropic expansion for the derivation of the perturbation equations @xcite . the effect of the anisotropy appears in the interacting hamiltonians between vector / two - form fields and scalar / tensor perturbations , by which the scalar / tensor power spectra are modified . then , we consider a general perturbed metric with four scalar functions @xmath121 and the tensor perturbation @xmath122 about the flat friedmann - lematle - robertson - walker ( flrw ) background @xcite : @xmath123dx^i dx^j \right\}\ , , \label{permet}\ ] ] where @xmath124 is the conformal time . after the end of inflation , the coupling @xmath36 approaches a constant because the inflaton stabilizes at the potential minimum . in this case vector perturbations decay after inflation as in the standard scenario , so we neglect its contribution to the cmb observables relative to those of scalar and metric perturbations . we introduce the gauge - invariant comoving curvature perturbation @xcite ( see also refs . @xcite ) : @xmath125 where @xmath126 is the perturbation of the inflaton @xmath7 . in the following we choose the spatially flat gauge ( @xmath127 ) , in which case @xmath128 . the curvature perturbation can be expressed in terms of the fourier components with the comoving wave number @xmath9 , as @xmath129 where the annihilation and creation operators @xmath130 and @xmath131 satisfy the commutation relation @xmath132=\delta^{(3 ) } ( { \bm k}-{\bm k}')$ ] . we define the scalar power spectrum @xmath133 in terms of the two - point correlation function of @xmath8 , as @xmath134 we decompose @xmath8 into the isotropic field @xmath135 and the contribution @xmath136 coming from the anisotropic fields , as @xmath137 in what follows we shall focus on the couplings ( [ fphive ] ) and ( [ coutwo ] ) , i.e. , @xmath138 . the situation is similar for the general couplings ( [ geneco ] ) and ( [ coutwo2 ] ) with @xmath82 close to 1 . then we can employ the usual slow - roll relations @xmath139 and @xmath140 , so that @xmath141 . the solution to the fourier mode @xmath142 , which recovers the bunch - davies vacuum state for the field perturbation @xmath126 in the asymptotic past ( @xmath143 ) , is given by @xcite @xmath144 the power spectrum can be written as the sum of the two contributions from @xmath135 and @xmath136 , as @xmath145 . using the solution ( [ backsol ] ) long time after the hubble radius crossing ( @xmath146 ) , the isotropic power spectrum of @xmath8 is given by @xmath147 in secs . [ pervec ] and [ pertwo ] we shall evaluate the anisotropic corrections to @xmath148 in both the vector and the two - form field models . for the tensor perturbation @xmath122 we impose the traceless and transverse conditions @xmath149 , as usual . the second - order action for @xmath122 reads @xmath150\,,\ ] ] where the prime denotes the differentiation with respect to @xmath151 . we have two physical degrees of freedom for @xmath122 which can be characterized by the symmetric polarization tensors @xmath152 satisfying @xmath153 where @xmath154 represent the polarizations . it is convenient to adopt the normalization @xmath155 where @xmath156 represents a complex conjugate . remark that the following relation holds : @xmath157 now , it is straightforward to quantize tensor perturbations . the mode expansion can be written as @xcite @xmath158e_{ij}^{(s)}({\bm k})\ , , \label{hijde}\ ] ] where the creation and annihilation operators are normalized as @xmath159 = \delta_{ss ' } \delta^{(3 ) } ( \bm{k}-\bm{k } ' ) $ ] . we define the tensor power spectrum @xmath160 , as @xmath161 when we study the polarization of tensor perturbations , we can take both the vectors @xmath162 and @xmath163 lying on the ( @xmath164)-plane without lose of generality ( because of the momentum conservation @xmath165 ) . in this case we can take @xmath166 where @xmath167 represents the angle between @xmath162 and @xmath38-axis . for @xmath168 , i.e. , @xmath169 , the polarization tensors @xmath170 satisfying the relations ( [ pore1])-([basis_prop2 ] ) are @xmath171 to obtain the polarization for @xmath172 , we need to rotate the above one by @xmath167 as @xmath173 we write the fourier mode @xmath174 in eq . ( [ hijde ] ) , as @xmath175 using eq . ( [ eij ] ) , it follows that @xmath176 which will be used for the evaluation of the interacting hamiltonians between gravitational waves and vector / two - form fields . we decompose the tensor perturbation @xmath177 into the isotropic field @xmath178 and the anisotropic contribution @xmath179 . the isotropic mode function @xmath180 obeys the following evolution equation @xmath181 where the canonical commutation relation leads to the normalization condition @xmath182 once a set of mode functions satisfying this normalization is specified , the corresponding fock vacuum is determined by @xmath183 . the mode function in a de sitter background is given by @xmath184 , that is @xmath185 using this solution and ( [ hijbmk ] ) , ( [ delh ] ) long after the hubble radius crossing , the isotropic power spectrum defined by ( [ tensorspe ] ) reads @xmath186 in the following we evaluate the anisotropic corrections to @xmath187 in both the vector and the two - form models . in doing so , it is convenient to notice the following commutation relations @xmath188 \simeq -i \frac{h^2}{6\epsilon m_{\rm pl}^2 } ( \tau^3-\tau{'^3 } ) \delta^{(3 ) } ( { \bm k}+{\bm k}')\ , , \label{commutator1 } \\ & & [ \hat{h}_s^{(0)}({\bm k } , \tau ) , \hat{h}_s^{(0)}({\bm k}^{\prime } , \tau^{\prime } ) ] \simeq -i\frac{4h^2}{3m_{\rm pl}^2}(\tau^3-{\tau^{\prime}}^3)\delta^{(3 ) } ( { \bm k } + { \bm k}^{\prime})\ , , \label{commutator3 } \end{aligned}\ ] ] which can be derived by employing the solutions ( [ backsol ] ) and ( [ hsso ] ) in the super - hubble regime ( @xmath189 ) . for the model described by the action ( [ eq : action ] ) we decompose the vector field @xmath27 into the fourier components by choosing the coulomb gauge : @xmath191 { \epsilon}^{(\lambda)}_i ( { \bm k})\,,\ ] ] where @xmath192 is the background component , and @xmath193 ( @xmath194 ) are polarization vectors satisfying the relations @xmath195 with the previous parametrization @xmath196 , an explicit representation is given by @xmath197 the rescaled field @xmath198 obeys the equation of motion @xmath199 for the coupling @xmath59 given by eq . ( [ fphi1 ] ) , we have @xmath200 on the de sitter background ( @xmath201 ) and hence @xmath202 . in this case the resulting vector field perturbation is scale - invariant . the solution to eq . ( [ vlam ] ) , which recovers the bunch - davies vacuum in the asymptotic past , is @xmath203 it is convenient to define the electric components @xmath204 where @xmath205 and @xmath206 correspond to the background and the perturbed values . the next step is to derive anisotropic contributions @xmath207 and @xmath208 to the isotropic scalar and tensor power spectra ( [ powerli ] ) and ( [ pten ] ) . the tree - level interacting lagrangian is @xmath209 where in the first line @xmath210 represents the background value and after the second line we picked up the second - order perturbation terms and dropped the symbol @xmath210 . we also used @xmath211 , which follows from the relation @xmath128 and the slow - roll conditions . we decompose @xmath212 into the fourier components , as @xmath213 using the solution ( [ alam ] ) in the super - hubble regime @xmath189 , the mode function @xmath214 is given by @xmath215 { \epsilon}^{(\lambda)}_i ( { \bm k})\,.\ ] ] the contributions to the interacting hamiltonian @xmath216 , which come from the four interacting lagrangians in eq . ( [ lint_vec ] ) , are given , respectively , by @xmath217 where @xmath218 is the angle between the wave number @xmath9 and the @xmath38-axis . in deriving the above hamiltonians , we used eqs . ( [ zeta0def ] ) , ( [ hijde ] ) , ( [ delh ] ) , ( [ delei ] ) , and replaced @xmath219 and @xmath220 for the isotropic perturbations @xmath221 and @xmath222 , respectively . the two - point correction of scalar perturbations following from the interacting hamiltonian ( [ hzeta ] ) to the isotropic power spectrum @xmath148 reads @xmath223 , h_\zeta ( \tau_2 ) \right ] \|0 \rangle \nonumber\\ & = & \frac{4e_{x}^2}{9\epsilon^2 m_{\rm pl}^4 h^4 } \prod_{i=1}^2 \int_{-1/k_i}^{\tau } \frac{d\tau_i}{\tau_i^4 } \left ( \tau^3 - \tau_i^3 \right ) \langle 0\|\delta { \cal e}_{x}({\bm k}_1 , \tau_1 ) \delta { { \cal e}}_{x } ( { \bm k}_2 , \tau_2 ) \|0 \rangle \nonumber\\ & = & \frac{2\pi^2}{k_1 ^ 3 } \delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 ) \frac{e_{x}^2n_k^2 \sin^2 \theta_{{\bm k}_1 , { \bm x}}}{\pi^2 \epsilon^2 m_{\rm pl}^4}\ , , \label{powerspe}\end{aligned}\ ] ] where we used the relation ( [ commutator1 ] ) . in the first line of eq . ( [ powerspe ] ) the two integrals have been evaluated in the super - horizon regime characterized by @xmath224 , that is , @xmath225 with @xmath226 . the choice of this contour is based upon the standard vacuum in the interacting field theory , that is , the change @xmath227 for large @xmath228 in the exponent @xmath229 in the mode function ( [ alam ] ) @xcite . this means that the oscillating term in the sub - horizon regime is exponentially suppressed , so that the main contribution to the integral ( [ powerspe ] ) comes from the super - horizon mode ( @xmath224 ) @xcite . in fact , the direct computation of the oscillating contributions to the integrals appearing in the correlation functions shows that the prescription mentioned above leads to the similar results to those derived by the regularization ( time averaging ) of the oscillating terms ( see appendix b of ref . @xcite ) . in the second line of eq . ( [ powerspe ] ) the upper bound @xmath230 of the second integral has been replaced by @xmath151 by dividing the factor 2 ! because of the symmetry of the integrand . we also used the property @xmath231 in the regime @xmath232 , where @xmath233 is the number of e - foldings before the end of inflation at which the modes with the wave number @xmath234 left the hubble radius . since @xmath235 , it follows that @xmath236 . thus , the total scalar power spectrum in the vector model is given by @xcite @xmath237 where , in the second equality , we used the relation @xmath238 and the definition @xmath79 given in eq . ( [ ra ] ) . for the parametrization ( [ anispe ] ) , this result corresponds to a negative anisotropic parameter @xmath239 in order to avoid that the anisotropic contribution does not exceed the isotropic spectrum , we demand the condition @xmath240 . from the wmap data there is the bound @xmath13 @xcite . under the condition @xmath240 it follows that @xmath241 for @xmath242 . since @xmath72 in eq . ( [ ra ] ) corresponds to the number of e - foldings from the onset of inflation , we have @xmath243constant for @xmath244 . we recall that , for the coupling ( [ geneco ] ) , the quantity @xmath79 is also constant . thus , the scalar spectral index reads @xmath245 where @xmath246 . the momentum vector @xmath247 does not necessarily need to lie on the @xmath248-plane , but it is generally given by @xmath249 , where @xmath250 and @xmath251 . it then follows that @xmath252 . the average of @xmath253 integrated over all the angles of @xmath254 and @xmath255 is @xmath256 using this property , the scalar spectral index ( [ ns_vec0 ] ) reads @xmath257 the anisotropic corrections to the two - point isotropic correlation of tensor perturbations @xmath258 @xmath259 are @xmath260 , h_{b } ( \tau_2 ) \right ] \|0 \rangle \nonumber \\ & = & \frac{4e_x^2}{m_{\rm pl}^4}\frac{\delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 ) } { k_1 ^ 3 } n_k^2 \sin^2\theta_{{\bm k}_1,{\bm x } } \ , , \label{hh_vec}\end{aligned}\ ] ] where all possible combinations of interacting hamiltonians @xmath261 and @xmath262 are taken , and @xmath263 , h_{h3 } ( \tau_2 ) \right ] \|0 \rangle \nonumber \\ & = & \frac{4e_x^2}{m_{\rm pl}^4}\frac{\delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 ) } { k_1 ^ 3 } n_k^2 \sin^2\theta_{{\bm k}_1,{\bm x } } \ , . \label{hh_vec2}\end{aligned}\ ] ] hence we obtain the total correction @xmath264 therefore , the total tensor power spectrum is given by @xmath265 the tensor - to - scalar ratio can be evaluated by using the anisotropic parameter ( [ g_vec ] ) as @xmath266 taking the same average over angles as ( [ sinave ] ) , it follows that @xmath267 from eq . ( [ ph_vec ] ) the tensor spectral index reads @xmath268 where we neglected the anisotropic contributions because they are second order in slow - roll parameters . the cross correlation between curvature perturbations and the plus mode of gravitational waves is given by @xmath269 , h_{a } ( \tau_2 ) \right ] \|0 \rangle + \text{perm . } \nonumber \\ & = & -\frac{4e_x^2}{\sqrt{2}\epsilon m_{\rm pl}^4}\frac{\delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 ) } { k_1 ^ 3 } n_k^2 \sin^2\theta_{{\bm k}_1,{\bm x } } \ , , \label{zeta_h}\end{aligned}\ ] ] where `` perm . '' represents the terms obtained by the permutations of @xmath270 and @xmath271 . similarly we have @xmath272 we define the cross power spectrum @xmath273 by @xmath274 . while there is no cross correlation without anisotropic interactions , it remains for the model ( [ eq : action ] ) as @xmath275 this gives rise to the non - vanishing tb cross power spectrum of cmb anisotropies @xcite . let us proceed to the two - form model given by the action ( [ oriaction ] ) . employing the gauge conditions @xmath276 , we have only one degree of freedom for the two - form field . the mode expansion can be expressed as @xmath277\epsilon_{ij}({\bm k } ) \,,\ ] ] where @xmath278 , @xmath279 are annihilation and creation operators , @xmath280 is the background value with only two non - zero components @xmath281 , and @xmath282 is the polarization tensor satisfying the following relations by @xmath283 to obtain the normalization @xmath284 . we also multiplied the complex factor @xmath285 to ensure the relation @xmath286 . ] @xmath287 the tensor @xmath288 looks similar to @xmath289 satisfying the relations ( [ pore1])-([basis_prop2 ] ) , but the difference is that the former is anti - symmetric while the latter is symmetric in the @xmath248 plane , @xmath290 can be expressed as @xmath291 ] . for the coupling ( [ couplingtwo ] ) , the field @xmath292 satisfies the following equation of motion on the de sitter background : @xmath293 in which case the scale - invariant spectrum follows . we can deduce the mode functions as @xmath294 for convenience we introduce the following variables @xmath295 then the tree - level interacting lagrangian is given by @xmath296 where , in the second line , we picked up the second - order terms of perturbations . decomposing the perturbation @xmath297 into the fourier modes , as @xmath298 the solution in the super - hubble regime ( @xmath189 ) is given by @xmath299 \epsilon_{ij}({\bm k})\,.\ ] ] the interacting hamiltonians @xmath300 can be written as the sum of the contributions from the five terms in eq . ( [ lint_two ] ) . they are given , respectively , by @xmath301 for the wave number @xmath302 lying on the @xmath248 plane the component @xmath303 is 0 , see eq . ( [ epij ] ) . hence the interacting hamiltonian @xmath304 associated with the tensor mode @xmath305 vanishes . the anisotropic contribution @xmath207 to @xmath148 can be evaluated from the interacting hamiltonian @xmath306 . the total scalar power spectrum has been derived in ref . @xcite , as @xmath307 where , in the second equality , we used @xmath308 and the definition @xmath117 given in eq . ( [ rb ] ) . comparing eq . ( [ ps_two ] ) with eq . ( [ anispe ] ) , the anisotropic parameter can be expressed as @xmath309 which is positive unlike the vector model . since @xmath117 is nearly constant for @xmath78 , we obtain the scalar spectral index @xmath310 in the last equality we used the fact that the average of @xmath311 integrated over all the angles is 1/3 . summing up all possible combinations of the interacting hamiltonians @xmath261 , @xmath262 , and @xmath312 , the anisotropic correction from the @xmath313 mode to the tensor power spectrum is and @xmath314 appearing in eqs . ( [ htwo1])-([htwo3 ] ) have the basis components @xmath315 and @xmath316 , respectively , see eq . ( [ epij ] ) . it then follows that @xmath317 . hence the sum of interacting hamiltonians between the two - form field and tensor perturbations vanishes . ] @xmath318 , h_{b } ( \tau_2 ) \right ] \|0 \rangle \nonumber \\ & = & \frac{4e_{yz}^2}{m_{\rm pl}^4}\frac{\delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 ) } { k_1 ^ 3 } n_k^2 \left(\sin^4\theta\cos^2\theta+\cos^6\theta+\cos^2\theta+2\sin^2\theta\cos^4\theta-2\sin^2\theta\cos^2\theta-2\cos^4\theta\right ) \nonumber \\ & = & 0 \ , . \label{hh_two}\end{aligned}\ ] ] we also have @xmath319 because of the property @xmath320 . since @xmath321 , it follows that @xmath322 . thus , the tensor - to - scalar ratio is given by @xmath323 where , in the last equality , we have taken the average over all the angles . the tensor spectral index reads @xmath324 similarly , the cross correlations between scalar and tensor perturbations are computed to give @xmath325 which mean that the cross power spectrum @xmath326 is 0 . this is an interesting property by which the two - form model can be distinguished from the vector model . [ obssec ] in this section , we place observational constraints on each anisotropic inflationary model with concrete inflaton potentials . using the cosmological monte carlo ( cosmomc ) code @xcite , we carry out the likelihood analysis with the latest planck data @xcite combined with the wmap large - angle polarization ( wp ) @xcite , baryon acoustic oscillations ( bao ) @xcite , and act / spt temperature data of high multipoles ( high-@xmath327 ) @xcite . the flat @xmath328cdm model is assumed with @xmath329 relativistic degrees of freedom and with the instant reionization . we also set the runnings of the scalar and tensor spectral indices to be 0 . the pivot wave number is chosen to be @xmath330 mpc@xmath331 . we confirmed that the different choices of @xmath332 such as @xmath333 mpc@xmath331 give practically identical likelihood results . from eqs . ( [ r_vec ] ) , ( [ nt_vec ] ) , and ( [ r_two ] ) , ( [ nt_two ] ) , the consistency relations in the two anisotropic models are given by @xmath334 the presence of anisotropic interactions modifies the standard consistency relation @xmath335 . if @xmath336 and @xmath337 , then we have @xmath338 for the vector model and @xmath339 for the two - form model , respectively . we have run the cosmomc code by using these consistency relations and found that the likelihood contours are very similar to those derived with the relation @xmath335 . therefore , we plot observational contours obtained by varying the three inflationary observables @xmath340 , @xmath341 , and @xmath342 with the consistency relation @xmath343 . in the vector model , anisotropic interactions lead to the enhancement of the scalar power spectrum @xmath133 on larger scales because the amplitude ( [ ps_vec ] ) increases for larger @xmath344 . as a result , the spectral index @xmath4 gets smaller for any inflaton potentials . the power spectrum of gravitational waves is also enhanced in the presence of anisotropic sources , but its effect is small compared to that on @xmath133 . hence the tensor - to - scalar ratio gets smaller irrespective of the inflaton potentials . we recall that the decreases of @xmath4 and @xmath3 are controlled by the negative anisotropic parameter @xmath2 given in eq . ( [ g_vec ] ) . in the two - form model , anisotropic interactions also lead to the decrease of @xmath4 and @xmath3 with the positive parameter @xmath2 given in eq . ( [ g_two ] ) . since the level of the enhancement of @xmath133 is different from that of the vector model , the observables @xmath4 and @xmath3 exhibit some difference between the two anisotropic inflation models . in the following we study the vector and two - form models separately for several different inflaton potentials . since we are considering the case @xmath138 , we can employ the standard slow - roll equations ( [ friap ] ) and @xmath345 at the background level . under this approximation , the number of e - foldings is given by @xmath346 where @xmath347 is the value of @xmath7 at the end of inflation determined by the condition @xmath348 . for the comparison of the inflationary observables with the cmb data , we fix @xmath349 . let us study observational constraints on the vector model . first , we consider chaotic inflation characterized by the power - law potential @xcite @xmath350 where @xmath351 and @xmath352 are positive constants . in this case , we have that @xmath353 and @xmath354 . the field value at the end of inflation can be estimated as @xmath355 . from eq . ( [ efold ] ) the number of e - foldings @xmath344 is related to the field @xmath7 , as @xmath356 . then the observables ( [ ns_vec ] ) and ( [ r_vec ] ) , which are averaged over all the angles , reduce to @xmath357 g _ } { n_k(n+4n_k)(3 - 2g_)}\ , , \qquad r= \frac{8n[6(n+4n_k)-ng_]}{(n+4n_k)^2(3 - 2g_)}\ , . \label{chaoticob}\ ] ] in the left panel of fig . [ fig1 ] we plot the theoretical values of @xmath4 and @xmath3 for the anisotropic parameter @xmath2 ranging in the region @xmath358 with @xmath349 . the self - coupling potential @xmath359 is outside the 95%confidence level ( cl ) observational boundaries even in the presence of anisotropic interactions . the quadratic potential @xmath360 is inside the 95%cl boundaries , but it is still outside the 68%cl contours . when @xmath14 , the linear potential @xmath361 , which appears in the axion monodromy scenario @xcite , is outside the 95%cl boundary constrained by the planck+wp+bao+high-@xmath327 data , but the vector anisotropy with @xmath362 allows the model to be inside the 68 % cl contour . a similar property also holds for another axion monodromy potential @xmath363 @xcite , but larger values of @xmath6 are required for the compatibility with the data . 2-dimensional observational bounds in the @xmath26 plane with the number of e - foldings @xmath364 and the pivot wave number @xmath330 mpc@xmath331 . the bold solid curves represent the 68%cl ( inside ) and 95%cl ( outside ) boundaries derived by the joint data analysis of the planck+wp+bao+high-@xmath327 data , whereas the thick dashed curves correspond to the contours constrained by the planck+wp+bao data . in both cases the consistency relation @xmath343 is used . we consider several different inflaton potentials in the vector model , i.e. , ( i ) @xmath365 ( @xmath366 ) ( left ) , ( ii ) @xmath367 $ ] ( right ) , and ( iii ) @xmath368 ( right ) , with @xmath2 ranging @xmath358 . the thin dotted curves correspond to the anisotropic parameters @xmath369 , respectively . the thin line labelled as `` @xmath370 inflation '' shows the theoretical plots predicted by the potential ( [ pot_s ] ) . in the presence of the vector field , both @xmath4 and @xmath3 get smaller.,title="fig:",width=336,height=316 ] 2-dimensional observational bounds in the @xmath26 plane with the number of e - foldings @xmath364 and the pivot wave number @xmath330 mpc@xmath331 . the bold solid curves represent the 68%cl ( inside ) and 95%cl ( outside ) boundaries derived by the joint data analysis of the planck+wp+bao+high-@xmath327 data , whereas the thick dashed curves correspond to the contours constrained by the planck+wp+bao data . in both cases the consistency relation @xmath343 is used . we consider several different inflaton potentials in the vector model , i.e. , ( i ) @xmath365 ( @xmath366 ) ( left ) , ( ii ) @xmath367 $ ] ( right ) , and ( iii ) @xmath368 ( right ) , with @xmath2 ranging @xmath358 . the thin dotted curves correspond to the anisotropic parameters @xmath369 , respectively . the thin line labelled as `` @xmath370 inflation '' shows the theoretical plots predicted by the potential ( [ pot_s ] ) . in the presence of the vector field , both @xmath4 and @xmath3 get smaller.,title="fig:",width=336,height=316 ] we also study natural inflation characterized by the potential @xcite @xmath371 \ , , \label{nat_pot}\ ] ] where @xmath328 and @xmath372 are constants having a dimension of mass . the relation between @xmath344 and @xmath7 is given by @xmath373 $ ] , where @xmath347 is known by solving the equation @xmath374 = 2(f / m_{\rm pl})^2 $ ] . the observables ( [ ns_vec ] ) and ( [ r_vec ] ) read @xmath375 -\{2n_k m_{\rm pl}^2[3-\cos(\phi / f)]+4f^2[1+\cos(\phi / f ) ] \ } g _ } { [ 1+\cos(\phi / f)]f^2n_k(3 - 2 g _ * ) } \ , , \label{ns_nat}\\ r&=&\frac{2\{12f^2[1+\cos(\phi / f)]-m_{\rm pl}^2[1-\cos(\phi / f)]g_\ } [ 1-\cos(\phi / f)]m_{\rm pl}^2}{[1+\cos(\phi / f)]^2f^4(3 - 2 g _ ) } \ , . \label{r_nat}\end{aligned}\ ] ] for a given value of @xmath372 we can numerically identify the field value @xmath7 corresponding to @xmath349 . then , we evaluate @xmath4 and @xmath3 according to the formulas ( [ ns_nat ] ) and ( [ r_nat ] ) . in the limit that @xmath376 , these observables approach the values ( [ chaoticob ] ) of chaotic inflation with @xmath377 . in the right panel of fig . [ fig1 ] , we show the theoretical values of @xmath4 and @xmath3 for different values of @xmath372 and @xmath2 . for smaller @xmath372 and larger @xmath6 , both @xmath4 and @xmath3 get smaller . when @xmath14 , the mass scale @xmath372 is constrained to be @xmath378 ( 68%cl ) from the planck+wp+bao+high-@xmath327 data @xcite . for larger @xmath6 , the allowed parameter space inside the 68%cl contours tends to be narrower . in particular , if @xmath379 , then the model is outside the 68%cl boundary constrained by the planck+wp+bao+high-@xmath327 data . thus , in natural inflation , the presence of anisotropic interactions leads to the deviation from the observationally favored region . let us also discuss the inflaton potential of the form @xmath380 where @xmath381 is a constant having a dimension of mass . this potential arises in the starobinsky s model @xmath382 @xcite after a conformal transformation to the einstein frame with the field definition @xmath383 $ ] @xcite . recently there have been numerous attempts to construct the potential ( [ pot_s ] ) in the context of supergravity and quantum gravity @xcite . in the regime @xmath384 , the number of e - foldings is related to the inflaton , as @xmath385 @xcite . the slow - roll parameters are approximately given by @xmath386 and @xmath387 , which means that @xmath388 is much smaller than @xmath389 . therefore , the observables ( [ ns_vec ] ) and ( [ r_vec ] ) reduce to @xmath390 when @xmath14 we have @xmath391 and @xmath392 , which correspond to the values in the starobinsky s model @xcite . the anisotropic interactions lead to the decrease of @xmath4 , but still the model is well inside the 68%cl contour , see the right panel of fig . [ fig1 ] . we also study hybrid inflation characterized by the potential @xmath393 @xcite , where @xmath328 and @xmath394 are constants . when @xmath14 , this model gives rise to a blue - tilted spectrum ( @xmath395 ) . in the presence of anisotropic interactions it is possible to have a red - tilted spectrum , but we find that @xmath4 is larger than 0.99 for @xmath396 . hence the model is still outside the 95%cl region . the same observational contours as those given in fig . [ fig1 ] ( denoted as thick solid / dashed curves ) . the thin solid and dotted lines show the theoretical predictions of the two - form model with @xmath349 for three inflaton potentials ( [ chao_pot ] ) , ( [ nat_pot ] ) , and ( [ pot_s ] ) . we choose five different values of the anisotropic parameter : @xmath397 . for larger @xmath2 , both @xmath4 and @xmath3 get smaller.,title="fig:",width=336,height=316 ] the same observational contours as those given in fig . [ fig1 ] ( denoted as thick solid / dashed curves ) . the thin solid and dotted lines show the theoretical predictions of the two - form model with @xmath349 for three inflaton potentials ( [ chao_pot ] ) , ( [ nat_pot ] ) , and ( [ pot_s ] ) . we choose five different values of the anisotropic parameter : @xmath397 . for larger @xmath2 , both @xmath4 and @xmath3 get smaller.,title="fig:",width=336,height=316 ] in the two - form model the scalar spectral index and the tensor - to - scalar ratio , which are averaged over all the angles , are given by eqs . ( [ ns_two ] ) and ( [ r_two ] ) , respectively . for chaotic inflation characterized by the potential ( [ chao_pot ] ) , these observables reduce to @xmath398 g _ * } { n_k(n+4n_k)(3+g_)}\ , , \qquad r= \frac{48n}{(n+4n_k)(3+g_)}\,.\ ] ] in the left panel of fig . [ fig2 ] the theoretical predictions of chaotic inflation are shown for @xmath2 ranging in the region @xmath399 . for larger @xmath2 both @xmath4 and @xmath3 decrease , but the quadratic potential is outside the 68%cl region . in the presence of anisotropic interactions the potentials with @xmath400 and @xmath401 enter the 95%cl boundaries , but still these models are outside the 68%cl region constrained by the planck+wp+bao+high-@xmath327 data . this shows that , for the same value of @xmath6 , the power - law potentials with @xmath402 in the two - form model are more difficult to be compatible with the data relative to the same potentials in the vector model . for natural inflation given by the potential ( [ nat_pot ] ) , the observables ( [ ns_two ] ) and ( [ r_two ] ) read @xmath403 + \ { n_k m_{\rm pl}^2[3- \cos(\phi / f ) ] + 2f^2[1+\cos(\phi / f ) ] \}g _ * } { [ 1+\cos(\phi / f)]f^2n_k ( 3+g _ * ) } \ , , \\ r & = & \frac{24m_{\rm pl}^2 [ 1- \cos(\phi / f ) ] } { f^2(3+g_)[1 + \cos(\phi / f)]}\,,\end{aligned}\ ] ] which decrease for larger @xmath2 and smaller @xmath372 . the difference from the vector model is that , for the same value of @xmath6 , the allowed region of the two - form model inside the 68%cl observational contours is wider . when @xmath337 , for example , we find that the mass scale @xmath372 is constrained to be @xmath404 ( 68%cl ) in the two - form model . as long as @xmath2 is smaller than @xmath405 , there are some allowed values of @xmath372 compatible with the observational data at 68%cl . for the potential ( [ pot_s ] ) we have @xmath406 as we see in the right panel of fig . [ fig2 ] , this potential is well within the 68%cl region . we also find that hybrid inflation with the potential @xmath393 is outside the 95%cl boundaries for @xmath407 . in summary , the observables @xmath4 and @xmath3 in the two - form model exhibit similar properties as those discussed in the vector model , but there are some potentials which can be inside and outside the 68%cl region depending on the values of @xmath2 . the precise observational bounds of @xmath2 ( in particular the signs of @xmath2 ) will be able to distinguish the two anisotropic inflationary models further . in this section , we estimate anisotropic contributions to the statistical non - gaussianities of curvature perturbations . the three - point correlation functions of @xmath8 have been already derived both in the vector model @xcite and in the two - form model @xcite . in the vector model , bartolo _ et al . _ @xcite considered the squeezed shape in which the angles @xmath408 and @xmath409 between the momentum vectors satisfying the relation @xmath410 are not necessarily close to @xmath411 and they took the average over angles to evaluate the local non - linear estimator @xmath25 . in the two - form model , the present authors @xcite took the strict squeezed limit , @xmath412 , @xmath413 , @xmath414 , and showed that @xmath25 vanishes . we clarify the difference between these two approaches and estimate the local non - linear estimators in both analyses . in the two - form model , the non - linear estimators have been also computed for other shapes of non - gaussianities such as the equilateral and enfolded ones @xcite . since the similar analysis has not been yet done in the vector model , we evaluate @xmath5 in the equilateral and enfolded limits . in order to calculate the three - point correlation function of curvature perturbations in the vector model described by the action ( [ eq : action ] ) , we need to derive the hamiltonian @xmath415 following from the third - order interacting lagrangian @xmath416 , in addition to @xmath306 given in eq . ( [ hzeta ] ) . on using eqs . ( [ zeta0def ] ) and ( [ delei ] ) , this interacting hamiltonian is given by @xmath417 then we can evaluate the three - point correlation of @xmath8 , as @xcite @xmath418 , h_\zeta ( \tau_2 ) \right ] h_\zeta ( \tau_3 ) \right ] \|0 \rangle + \text{2 perm . } \nonumber\\ & = & \frac{4e_{x}^2}{27\epsilon^3 m_{\rm pl}^6 h^6 } \prod_{i=1}^3 \int_{-1/k_i}^{\tau } \frac{d\tau_i}{\tau_i^4 } \left ( \tau^3 - \tau_i^3 \right ) \nonumber \\ & & \times \int \ , \frac{d^3p}{(2\pi)^{3/2 } } \langle 0\|\delta { \cal e}_{x}({\bm k}_1 , \tau_1 ) \delta { { \cal e}}_{x}({\bm k}_2 , \tau_2 ) \delta { \cal e}_{i}({\bm p } , \tau_3 ) \delta { { \cal e}}_{j}({\bm k}_3-{\bm p } , \tau_3 ) \|0 \rangle + \text{2 perm . } \nonumber\\ & \simeq & 288\sqrt{2}\pi^{5/2 } \frac{e_{x}^2}{\epsilon v } n_{k_1 } n_{k_2 } n_{k_3 } ( { \cal p}_{\zeta}^{(0)})^2 \delta^{(3 ) } ( { \bm k}_1 + { \bm k}_2 + { \bm k}_3 ) \nonumber \\ & & \times \left[\frac{1}{k_1 ^ 3 k_2 ^ 3 } ( 1- \cos^2 \theta_{{\bm k}_1 , { \bm x } } -\cos^2\theta_{{\bm k}_2 , { \bm x}}+ \cos\theta_{{\bm k}_1 , { \bm x}}\cos\theta_{{\bm k}_2 , { \bm x}}\cos\theta_{{\bm k}_1 , { \bm k}_2 } ) + { \rm 2~perm.}\right ] . \label{bispe}\end{aligned}\ ] ] the total anisotropic bispectrum @xmath419 is defined by @xmath420 . we introduce the non - linear estimator @xmath5 in the form @xmath421 by which @xmath5 for the vector model can be derived as @xcite @xmath422 \nonumber \\ \hspace{-0.7 cm } & \simeq & 60\left(\frac{-g_}{0.1}\right)\left(\frac{n_k}{60}\right)\frac{1}{k_1 ^ 3+k_2 ^ 3+k_3 ^ 3 } [ k_3 ^ 3 ( 1- \cos^2 \theta_{{\bm k}_1 , { \bm x } } -\cos^2\theta_{{\bm k}_2 , { \bm x}}+ \cos\theta_{{\bm k}_1 , { \bm x}}\cos\theta_{{\bm k}_2 , { \bm x}}\cos\theta_{{\bm k}_1 , { \bm k}_2})+{\rm 2~perm . } ] , \label{fnl_vec}\end{aligned}\ ] ] where , in the first line , we used @xmath238 and the quantity @xmath79 defined in eq . ( [ ra ] ) . in the second line we employed the approximations @xmath423 , @xmath424 , and the definition of @xmath2 given in eq . ( [ g_vec ] ) . let us first take the same approach as that used in ref . @xcite for the two - form model . it is convenient to introduce the following parameters @xmath425 if we fix @xmath426 and define the angles @xmath427 and @xmath428 , it follows that @xmath429 . \label{fnl_vec2}\end{aligned}\ ] ] the angle @xmath430 is in the range @xmath431 ( i.e. , @xmath432 ) . in the strict squeezed limit ( @xmath433 and @xmath434 ) , the local estimator @xmath25 vanishes for any values of @xmath435 . this limit corresponds to the case in which the angles @xmath436 and @xmath437 approach @xmath411 . bartolo _ et al . _ @xcite estimated @xmath25 in the following way . for the incomplete squeezed shape the angles @xmath409 and @xmath438 are not necessarily close to @xmath439 . in other words , unless we take the strict squeezed limit @xmath440 , we can consider any angle between @xmath18 and @xmath441 . averaging over @xmath5 in all the directions along the same line as eq . ( [ sinave ] ) , the local estimator for the squeezed shape ( @xmath442 , @xmath443 , and @xmath444 ) reads @xmath445}{4/9 } \ , , \label{fnl_vec_local}\ ] ] where we used the fact that the average value of the function in the last square bracket integrated over all the angles is @xmath446 . in this analysis the local non - linear estimator does not vanish and it can be as large as the order of @xmath447 for @xmath448 . the planck group placed the bound on the local non - linear estimator to be @xmath449 ( 68%cl ) @xcite , but we can not literally use this bound to constrain the anisotropic inflationary models . as we have seen above , the local non - gaussianities depend on what kind of squeezed shapes to be taken in the data analysis . it is interesting that the strict or incomplete squeezed shapes do matter to estimate the level of non - gaussianities correctly . the detailed data analysis based on different squeezed shapes is beyond the scope of our paper . using the result ( [ fnl_vec2 ] ) , we can calculate the non - linear estimator for the shapes other than the squeezed one . in the equilateral ( @xmath450 ) and the enfolded ( @xmath451 ) limits , we have @xmath452 unlike the squeezed case , these estimators do not depend on how we take the limits of equilateral and enfolded shapes . the equilateral non - linear estimator is independent of the angle @xmath435 and it is typically of the order of @xmath453 for @xmath454 . meanwhile , @xmath455 depends on @xmath435 and it has a maximum at @xmath456 . the maximum value of @xmath455 can be as large as @xmath457 for @xmath458 , which is an interesting signature of the vector model . in the two - form model the hamiltonian following from the third - order interacting lagrangian @xmath459 is given by @xmath460 the three - point correlation function of @xmath8 can be computed by using the interacting hamiltonians ( [ hzeta2 ] ) and ( [ hzeta2_two ] ) in the way similar to the derivation of eq . ( [ bispe ] ) . in ref . @xcite this was already derived as @xmath461 . \label{bispe_two}\ ] ] from the definition ( [ def_fnl ] ) of @xmath5 , it follows that @xmath462 \ , , \label{fnl_two}\end{aligned}\ ] ] where we used the anisotropic parameter @xmath2 given in eq . ( [ g_two ] ) with @xmath463 . for the squeezed shape , if we take the average over angles as the derivation of eq . ( [ fnl_vec_local ] ) , the local non - linear parameter reads @xmath464}{1/9 } \ , , \label{fnl_two_local}\ ] ] where the value @xmath465 is the total spatial average of the function in the last square bracket . therefore , @xmath466 dose not vanish in this analysis . we note that @xmath466 in this case is smaller than that in the vector model by one order of magnitude , see eq . ( [ fnl_vec_local ] ) . thus the local non - gaussianities averaged over all the directions can allow us to discriminate between the vector and the two - form models . on the other hand , in the strict squeezed limit characterized by @xmath412 , @xmath467 , and @xmath468 , the non - linear estimator ( [ fnl_two ] ) vanishes , @xmath469 @xcite . the non - linear parameters in the equilateral and the enfolded limits were already evaluated in ref . @xcite by considering a configuration with @xmath470 and @xmath427 , @xmath428 . they are given , respectively , by @xmath471 the equilateral non - linear estimator is independent of @xmath435 , but the enfolded one depends on @xmath435 . unlike the vector model , @xmath455 has a maximum at @xmath472 or @xmath473 . both @xmath474 and @xmath455 are about half times smaller than those in the vector model . in general , for the same values of @xmath6 , the two - form model is easier to satisfy observational bounds of non - gaussianities relative to the vector model . we have studied differences between two anisotropic inflationary scenarios paying particular attention to their observational signatures . in the presence of a vector or a two - form field coupled to the inflaton , there exist background solutions along which the anisotropic hairs survive during inflation . in the vector model the anisotropic shear @xmath66 divided by the hubble rate @xmath57 is given by eq . ( [ sigh1 ] ) for the coupling ( [ fphive ] ) , while in the two - form model it is given by eq . ( [ sigtwo2 ] ) for the coupling ( [ coutwo ] ) . the opposite signs of @xmath67 between the two models reflect the fact that the types of anisotropies are different ( either oblate or prolate ) . in the presence of anisotropic interactions , we have computed the power spectra of scalar / tensor perturbations and the resulting spectral indices convenient to confront with the cmb observations . the different types of anisotropies affect the sign of @xmath2 appearing in the scalar power spectrum defined by eq . ( [ anispe ] ) , i.e. , @xmath475 in the vector model and @xmath20 in the two - form model . the anisotropic contributions to the isotropic tensor power spectrum are suppressed relative to those to the isotropic scalar one . in particular , in the two - form model , we showed that anisotropic interactions do not give rise to any corrections to the isotropic tensor power spectrum . we have computed the two - point cross correlations between curvature perturbations and gravitational waves . while the cross correlation remains in the vector model , we find that it vanishes in the two - form model . the latter property may reflect the fact that , unlike the vector field , the two - form field can be mapped to the form of a scalar field . the different signatures of the two anisotropic models will be useful to discriminate between those models in future observations of the tb power spectrum . in the light of the recent planck data , we have placed observational constraints on several different inflaton potentials from the information of the scalar spectral index @xmath4 and the tensor - to - scalar ratio @xmath3 . in both the vector and two - form models anisotropic interactions lead to the enhancement of the scalar power spectrum on larger scales , by which both @xmath4 and @xmath3 decrease for any inflaton potentials . in the vector model , we found that the potentials @xmath361 and @xmath363 show the better compatibility with the data for larger @xmath6 and that the potential of natural inflation is outside the 68%cl region constrained by the planck+wp+bao+high-@xmath327 data for @xmath476 . in the two - form model , the level of the decreases of @xmath4 and @xmath3 is less significant relative to the vector model for the same values of @xmath6 . the potential ( [ pot_s ] ) , which originates from the starobinsky s @xmath477 model , is well inside the 68%cl region even in the presence of anisotropic interactions with @xmath478 . we also compared the three - point correlation functions of curvature perturbations between the two anisotropic inflationary scenarios . if the strict squeezed limit characterized by @xmath412 , @xmath413 , and @xmath414 is taken , the local non - linear estimators @xmath25 vanish in both the vector and the two - form models . however , for the squeezed shape where the angles @xmath408 and @xmath409 are not necessarily close to @xmath411 , the non - linear estimator averaged over all the directions is given by eq . ( [ fnl_vec_local ] ) in the vector model and by eq . ( [ fnl_two_local ] ) in the two - form model . the former is larger than the latter by one order of magnitude for the same order of the anisotropic parameter @xmath2 . we also evaluated the non - linear estimators in the equilateral and enfolded limits and found that @xmath5 in the vector model is about twice larger than that in the two - form model for the same values of @xmath6 . in general , the two - form model is easier to satisfy observational bounds of non - gaussianities relative to the vector model . we have thus shown that the two anisotropic inflationary scenarios can be distinguished from each other by evaluating several cmb observables . in particular , the precise measurements of @xmath2 as well as the tb correlation will clarify which anisotropic model is favored over the other . this work is supported by the grant - in - aid for scientific research fund of the ministry of education , science and culture of japan ( nos . 23@xmath4796781 , 25400251 , and 24540286 ) , the grant - in - aid for scientific research on innovative area ( no . 21111006 ) . v. f. mukhanov and g. v. chibisov , jetp lett . * 33 * , 532 ( 1981 ) ; + a. h. guth and s. y. pi , phys . * 49 * ( 1982 ) 1110 ; + s. w. hawking , phys . lett . b * 115 * , 295 ( 1982 ) ; + a. a. starobinsky , phys . b * 117 * ( 1982 ) 175 ; + j. m. bardeen , p. j. steinhardt and m. s. turner , phys . d * 28 * , 679 ( 1983 ) . s. yokoyama and j. soda , jcap * 0808 * , 005 ( 2008 ) [ arxiv:0805.4265 [ astro - ph ] ] ; + k. dimopoulos , m. karciauskas , d. h. lyth and y. rodriguez , jcap * 0905 * , 013 ( 2009 ) [ arxiv:0809.1055 [ astro - ph ] ] ; + m. karciauskas , k. dimopoulos and d. h. lyth , phys . d * 80 * ( 2009 ) 023509 [ arxiv:0812.0264 [ astro - ph ] ] ; + c. a. valenzuela - toledo , y. rodriguez and d. h. lyth , phys . d * 80 * , 103519 ( 2009 ) [ arxiv:0909.4064 [ astro-ph.co ] ] ; + s. kanno , j. soda and m. a. watanabe , jcap * 0912 * , 009 ( 2009 ) [ arxiv:0908.3509 [ astro-ph.co ] ] ; + c. a. valenzuela - toledo and y. rodriguez , phys . b * 685 * , 120 ( 2010 ) [ arxiv:0910.4208 [ astro-ph.co ] ] ; + n. bartolo , e. dimastrogiovanni , s. matarrese and a. riotto , jcap * 0910 * , 015 ( 2009 ) [ arxiv:0906.4944 [ astro-ph.co ] ] ; + n. bartolo , e. dimastrogiovanni , s. matarrese and a. riotto , jcap * 0911 * , 028 ( 2009 ) [ arxiv:0909.5621 [ astro-ph.co ] ] ; + e. dimastrogiovanni , n. bartolo , s. matarrese and a. riotto , adv . astron . * 2010 * , 752670 ( 2010 ) [ arxiv:1001.4049 [ astro-ph.co ] ] ; + j. m. wagstaff and k. dimopoulos , phys . d * 83 * , 023523 ( 2011 ) [ arxiv:1011.2517 [ hep - ph ] ] . p. v. moniz and j. ward , class . * 27 * , 235009 ( 2010 ) [ arxiv:1007.3299 [ gr - qc ] ] ; + k. murata and j. soda , jcap * 1106 * , 037 ( 2011 ) [ arxiv:1103.6164 [ hep - th ] ] ; + t. q. do , w. f. kao and i. -c . lin , phys . d * 83 * , 123002 ( 2011 ) ; + t. q. do and w. f. kao , phys . d * 84 * , 123009 ( 2011 ) ; + r. emami , h. firouzjahi , s. m. sadegh movahed and m. zarei , jcap * 1102 * , 005 ( 2011 ) [ arxiv:1010.5495 [ astro-ph.co ] ] ; + m. karciauskas , jcap * 1201 * , 014 ( 2012 ) [ arxiv:1104.3629 [ astro-ph.co ] ] ; + m. shiraishi and s. yokoyama , prog . * 126 * , 923 ( 2011 ) [ arxiv:1107.0682 [ astro-ph.co ] ] ; + s. bhowmick and s. mukherji , mod . a * 27 * , 1250009 ( 2012 ) [ arxiv:1105.4455 [ hep - th ] ] ; + r. emami and h. firouzjahi , jcap * 1201 * , 022 ( 2012 ) [ arxiv:1111.1919 [ astro-ph.co ] ] ; + k. yamamoto , m. a. watanabe and j. soda , class . * 29 * , 145008 ( 2012 ) [ arxiv:1201.5309 [ hep - th ] ] ; + k. dimopoulos , int . j. mod . d * 21 * , 1250023 ( 2012 ) [ erratum - ibid . d * 21 * , 1292003 ( 2012 ) ] [ arxiv:1107.2779 [ hep - ph ] ] ; + a. a. abolhasani , r. emami , j. t. firouzjaee and h. firouzjahi , arxiv:1302.6986 [ astro-ph.co ] ; + d. h. lyth and m. karciauskas , arxiv:1302.7304 [ astro-ph.co ] ; + s. baghram , m. h. namjoo and h. firouzjahi , arxiv:1303.4368 [ astro-ph.co ] ; + n. bartolo , s. matarrese , m. peloso and a. ricciardone , arxiv:1306.4160 [ astro-ph.co ] . a. e. gumrukcuoglu , b. himmetoglu and m. peloso , phys . d * 81 * , 063528 ( 2010 ) [ arxiv:1001.4088 [ astro-ph.co ] ] ; + t. r. dulaney and m. i. gresham , phys . d * 81 * , 103532 ( 2010 ) [ arxiv:1001.2301 [ astro-ph.co ] ] . n. barnaby and m. peloso , phys . lett . * 106 * , 181301 ( 2011 ) [ arxiv:1011.1500 [ hep - ph ] ] ; + n. barnaby , r. namba and m. peloso , jcap * 1104 * , 009 ( 2011 ) [ arxiv:1102.4333 [ astro-ph.co ] ] ; + n. barnaby , e. pajer and m. peloso , phys . d * 85 * , 023525 ( 2012 ) [ arxiv:1110.3327 [ astro-ph.co ] ] ; + n. barnaby , r. namba and m. peloso , phys . d * 85 * , 123523 ( 2012 ) [ arxiv:1202.1469 [ astro-ph.co ] ] . s. v. ketov and a. a. starobinsky , phys . d * 83 * , 063512 ( 2011 ) ; + s. v. ketov and s. tsujikawa , phys . d * 86 * , 023529 ( 2012 ) [ arxiv:1205.2918 [ hep - th ] ] ; + j. ellis , d. v. nanopoulos and k. a. olive , arxiv:1305.1247 [ hep - th ] ; + r. kallosh and a. linde , jcap * 1306 * , 028 ( 2013 ) [ arxiv:1306.3214 [ hep - th ] ] ; + f. farakos , a. kehagias and a. riotto , arxiv:1307.1137 [ hep - th ] ; + w. buchmuller , v. domcke and k. kamada , arxiv:1306.3471 [ hep - th ] ; + f. briscese , l. modesto and s. tsujikawa , arxiv:1308.1413 [ hep - th ] .	we study observational signatures of two classes of anisotropic inflationary models in which an inflaton field couples to ( i ) a vector kinetic term @xmath0 and ( ii ) a two - form kinetic term @xmath1 . we compute the corrections from the anisotropic sources to the power spectrum of gravitational waves as well as the two - point cross correlation between scalar and tensor perturbations . the signs of the anisotropic parameter @xmath2 are different depending on the vector and the two - form models , but the statistical anisotropies generally lead to a suppressed tensor - to - scalar ratio @xmath3 and a smaller scalar spectral index @xmath4 in both models . in the light of the recent planck bounds of @xmath4 and @xmath3 , we place observational constraints on several different inflaton potentials such as those in chaotic and natural inflation in the presence of anisotropic interactions . in the two - form model we also find that there is no cross correlation between scalar and tensor perturbations , while in the vector model the cross correlation does not vanish . the non - linear estimator @xmath5 of scalar non - gaussianities in the two - form model is generally smaller than that in the vector model for the same orders of @xmath6 , so that the former is easier to be compatible with observational bounds of non - gaussianities than the latter .
nearby galaxies are unique laboratories for investigating the processes that drive galaxy evolution . the information obtained from the photometric and spectroscopic analyses of individual stars can be employed to probe the history of nearby galaxies to a level of detail that is not possible for distant systems , where stellar content studies are limited to information gleaned from the analysis of integrated light . star counts also provide a means of tracing structures to lower densities than is possible with integrated light , thereby allowing diffuse tidal debris trails and low surface brightness companions , both of which provide clues into the past history of a galaxy , to be identified and characterized . the majority of large galaxies within @xmath7 mpc are disk - dominated , and studies of their resolved stellar contents reveal considerable diversity in recent star - forming histories . galaxies such as m31 ( williams 2002 ) and ngc 5102 ( davidge 2008a ; 2010 ) have depressed star formation rates ( sfrs ) at the present day when compared with @xmath8 gyr in the past . in contrast , a number of nearby galaxies , such as m82 , ngc 5253 , ngc 253 , and m83 , have elevated levels of central star - forming activity . that a number of nearby galaxies are experiencing elevated levels of star - forming activity indicates that this is not a rare phenomenon in the local universe , and starbursts account for at least 20% of all recent local star formation ( brinchmann et al . 2004 ; lee et al . 2009 ) . there is a compelling association between elevated sfrs and galaxy - galaxy interactions ( e.g. larson & tinsley 1978 ; iono et al . 2004 ; kewley et al . this suggests that tidal effects are the primary cause of starbursts , at least among massive galaxies , although interactions with intergalactic material can also trigger elevated star - forming activity ( e.g. kapferer et al . 2009 ) . it is thus worth noting that some of the nearest starburst or post - starburst galaxies do not have obvious companions or are not located in environments that are thought to contain a dense intergalactic medium , leaving no obvious trigger for the starburst activity . two such examples are ngc 5102 and ngc 253 . the former galaxy was investigated by davidge ( 2008c ) , while the latter is the subject of this paper . ngc 253 is the largest member of the filamentary collection of galaxies that is classically referred to as the sculptor group . with a crossing time of 6 7 gyr ( karachensev 2005 ) , the group is not dynamically relaxed , and interactions between its dominant members are likely rare . still , ngc 253 has not evolved in isolation . the spin of the ngc 253 disk is consistent with torquing from ngc 247 ( whiting 1999 ) , which at a projected distance of @xmath9 kpc from ngc 253 is its closest large companion . extragalactic hi , which is a possible signature of an interaction that affected the disk of one or both galaxies , has not been detected in the vicinity of ngc 253 and ngc 247 ( putman et al . 2003 ) , although neutral hydrogen may be ionized by background radiation at large distances from galaxies ( e.g. westmeier , bruns , & kerp 2008 ) . despite its isolated location , kinematic studies suggest that ngc 253 was involved in a recent merger . anantharamaiah & goss ( 1996 ) and prada et al . ( 1998 ) find evidence of two kinematically distinct systems near the center of ngc 253 . one system has an axis of rotation that differs from that of the galaxy disk , while the other appears to be counter - rotating with respect to the galaxy disk . das et al . ( 2001 ) model the kinematics of material near the center of ngc 253 and detect twisting in the velocity field @xmath10 arcsec from the nucleus . all three studies interpret their results as kinematic signatures of a merger . das et al . ( 2001 ) estimate that the merger may have involved the accretion of @xmath11 m@xmath12 of material @xmath13 years in the past , and they point out that this may be the remnants of what was initially a much larger object . much of the literature on ngc 253 deals with the central regions of the galaxy and the superwind . rieke & low ( 1975 ) first noted that the photometric properties of the central regions of the galaxy were consistent with a transient level of elevated star formation , and the presence of a large central young population was confirmed with the discovery that the near - infrared light is dominated by red supergiants ( rsgs ; rieke et al . rieke et al . ( 1988 ) suggest that ngc 253 may be in an earlier phase of starburst activity than m82 , thereby making it a potentially important object for understanding the early phases of starburst evolution . the central velocity field defined by co emission in ngc 253 is indicative of a bar , which is likely a conduit for material that feeds the central starburst ( paglione et al . the present - day sfr of ngc 253 is 5 m@xmath12 year@xmath14 ( melo et al . 2002 ) , and the sne rate is 0.2 year@xmath14 ( pietsch et al . 2001 ) . engelbracht et al . ( 1998 ) estimate that this level of nuclear star formation has been on - going for 20 30 myr , although the sfr may have declined during the past 5 myr . starburst activity has almost certainly occured for a much longer time than calculated by engelbracht et al . ( 1998 ) , as star - forming activity in starbursts appears to be variable in both time and location ( e.g. forster schreiber et al . 2003 ; mcquinn et al . 2009 ) . ngc 253 is interacting with its immediate surroundings . the extraplanar regions near the minor axis of ngc 253 are populated by a mix of hot gas , cool gas , and dust , and heckman et al . ( 1990 ) estimate that 1.2 m@xmath12 year@xmath14 is ejected from the central regions . as in the m82 outflow , x - ray and h@xmath15 emission in the ngc 253 outflow occurs in close proximity . the x - ray and h@xmath15 emission likely originates in a boundary environment where the superwind interacts with a dense cloud of cool material ( e.g. mccarthy et al . 1987 ; strickland et al . the cool material is presumably tidal in origin , although it could also be remnant natal gas . there is prominent extraplanar uv emission , and hoopes et al . ( 2005 ) suggest that this is light from the starburst that is reflected by dust . beck et al . ( 1982 ) and malin & hadley ( 1997 ) detect diffuse extraplanar light from ngc 253 at visible wavelengths , and some of this light may be emission from the outflow . gas and dust in ngc 253 are centrally concentrated . interstellar material accounts for 7% of the mass near the center of ngc 253 , but only 4% of the mass at larger radii ( scoville et al . the radial distribution of molecular material in ngc 253 is a factor of 2 more concentrated than in other nearby galaxies ( sorai et al . 2000 ) , and there is also a central hi concentration ( combes et al . 1977 ) . a large reservoir of molecular material is found in a star - forming ring that is 2 kpc from the nucleus ( sorai et al . 2000 ) , and the spiral structure of ngc 253 appears to eminate from this structure ( hoopes et al . 1996 ) . while most spiral galaxies have hi envelopes that extend well past the optical limits of the disk , puche et al . ( 1991 ) and boomsma et al . ( 1995 ) find that the hi disk in ngc 253 extends to only @xmath16 the holmberg radius . in addition , the hi distribution in the disk is not symmetric , in the sense that hi in the south west quadrant extends to larger radii than in the north east quadrant ( boomsma et al . 2005 ) . as noted by boomsma et al . ( 2005 ) , the lack of hi emission at large radii may not signal an absence of gas in the outer disk ; rather , any hi may be ionized if the disk is warped at large radii , such that the gas is exposed to radiation from the nucleus . tidal effects could also induce a truncated hi distribution , and it is intriguing that the closest large neighbor of ngc 253 , ngc 247 , also has a relatively compact hi distribution ( carignan & puche 1990 ) . the elevated levels of star - forming activity in ngc 253 have not been restricted to the central regions of the galaxy . the ngc 253 disk contains numerous filamentary and bubble - like structures that are signatures of a high sfr , and such structures may provide outlets for interstellar material to escape from the disk ( sofue et al . in fact , boomsma et al . ( 2005 ) detect diffuse extraplanar hi in the eastern half of ngc 253 . the hi is detected out to distances of 12 kpc from the disk plane , and the extraplanar component accounts for 3% of the total hi mass of ngc 253 . as one of the nearest starburst galaxies , ngc 253 is a prime target for stellar content studies , although only a modest number of investigations of its resolved stellar content have been published to date . the luminosity distribution of x - ray sources in the ngc 253 disk is similar to that in m33 and m31 , and this is suggestive of a common source population , with the caveat that the overlap between the ngc 253 and local group lfs is modest ( vogler & pietsch 1999 ) . davidge et al . ( 1991 ) discuss deep multicolor observations of a field along the north east major axis of the disk . a distance modulus of 27 was computed based on the properties of the detected rsgs and blue supergiants ( bsgs ) . the photometric properties of these stars change with location , suggesting that there are substantive spatial variations in extinction across the disk . karachentsev et al . ( 2003 ) discuss deep images of an area of the ngc 253 disk that is close to the davidge et al . ( 1991 ) field . the cmd in figure 2 of karachentsev et al . ( 2003 ) is dominated by a broad rgb that is capped by a richly populated agb . karachentsev et al . ( 2003 ) compute a distance modulus of 27.98 from the rgb - tip , and this distance modulus is adopted for the present study . there is also a bright main sequence in the karachentsev et al . cmd , which has a width of @xmath17 mag in @xmath18 . this dispersion exemplifies the extent of differential reddening in the ngc 253 disk . ngc 253 was imaged as part of the angst survey ( dalcanton et al . 2009 ) . in accordance with the survey guidelines , the fields observed in ngc 253 were restricted to the major axis , and cover roughly one half of the disk . while there is a single major axis field at moderately large distance from the galaxy center , the angst data do not sample the extraplanar regions of the galaxy . the main sequence in some of the angst cmds of ngc 253 has a width of 1 - 1.5 mag in f475w f814w , highlighting the influence of differential reddening in photometric studies of the galaxy at visible wavelengths . in fact , an inspection of the cmds in figures 9 22 of dalcanton et al . ( 2009 ) indicates that the extent of differential extinction in ngc 253 exceeds that in other galaxies in the angst sample . as with the disk , there have been only a modest number of studies of resolved stars in the extraplanar regions of ngc 253 . in the first deep ccd study of resolved stars in ngc 253 , davidge & pritchet ( 1990 ) observed a field along the north west arm of the minor axis . the completeness - corrected cmd shown in their figure 8 includes the upper portions of the rgb , and there is a break near @xmath19 in their cmd that corresponds to the rgb - tip found in subsequent studies . assuming a stellar content like that in the outer regions of m31 , a distance modulus of 26.8 was computed from the brightest star . this places ngc 253 much closer than more recent distance estimates ( e.g. karachentsev et al . 2003 ; mouhcine et al . 2005 ) - a result that may be explained if this part of ngc 253 contains a young population ( e.g. comeron et al . 2001 , and this study ) , rendering the assumption of a stellar content like that in the outer regions of m31 invalid . comeron et al . ( 2001 ) detect bright main sequence stars along the south east minor axis of ngc 253 out to distances of 15 kpc from the disk plane . using stellar lifetime arguments , comeron et al . ( 2001 ) suggest that the stars that are furthest from the disk plane must have formed in the outflow ; a similar conclusion was reached independently for young stars in the extraplanar regions of m82 ( davidge 2008c ) . mouhcine ( 2005 ) discusses deeper observations of a portion of the area surveyed by comeron et al . a population of rgb stars was detected , from which a distance modulus of 27.6 was deduced based on the rgb - tip brightness . with the caveat that young and intermediate age stars may bias the distance estimate and skew the shape of the rgb , the photometrically derived metallicity distribution function ( mdf ) of rgb stars in this field peaks near [ m / h ] = 0.7 , with a metal - poor tail that extends to 2 ( mouhcine 2006 ) . such an mdf is not characteristic of a classical old , metal - poor halo . rather , this mdf is reminsiscent of that seen throughout much of m31 ( e.g. bellazzini et al . 2003 ; koch et al . 2008 ; richardson et al . 2009 ; tanaka et al . 2010 ) , which is a galaxy that has experienced interactions with its satellites ( e.g. ibata et al . 2007 ) . dust has a major influence on the appearance of ngc 253 ( sofue et al . 1994 ) , and at least 90% of the h@xmath15 flux from the galaxy is obscured ( hoopes et al . 1996 ) . the dust is widespread , and the disk contains 94% of the total cool dust supply ( melo et al . the emission from cool dust has a scale length that is similar to ( melo et al . 2002 ) or possibly even larger than ( radovich et al 2001 ) that of the stellar light . the high dust content of the ngc 253 disk provides a compelling motivation to survey its stellar content at wavelengths longward of @xmath20 m , where the impact of extinction is greatly reduced with respect to visible wavelengths . the impact of line blanketing on the photometric properties of the reddest stars , such as those that are evolving near the asymptotic giant branch ( agb ) tip , is also greatly reduced in the infrared when compared with the visible . this enhances the utility of these objects as probes of stellar content , as it increases the contrast between the brightest red stars and the underlying blue body of the disk , thereby reducing the affect of crowding . in the present paper , near - infrared observations of three @xmath21 arcmin fields , recorded with the cfht wircam , are used to map the projected distribution of the brightest red stars in the disk and extraplanar regions of ngc 253 . the data , the basic reduction procedures , and the photometric measurements are described in 2 , while the spatial distribution of various stellar types are described in 3 . the properties of the brightest red stars in the disk plane and extrapalanar regions are examined in 4 and 5 , respectively . a summary and discussion of the results follows in 6 . the discovery of an extended halo around ngc 253 at ultraviolet ( uv ) wavelengths , which is distinct from the starburst outflow and appears to be due to an intermediate age extraplanar component ( 5 ) , is reported in the appendix . the data were recorded with wircam ( puget et al . 2004 ) on the canada - france - hawaii telescope ( cfht ) as part of the 2007b observing queue . the wircam detector is a mosaic of four @xmath22 hgcdte arrays that are deployed in a @xmath23 format . there is a 45 arcsec gap between detectors , and each pixel subtends @xmath24 arcsec@xmath25 . thus , a single exposure covers @xmath26 arcmin@xmath25 . three fields , that sample the major axis of ngc 253 at both ends of the disk ( fields 1 and 2 ) and the south east segment of the minor axis ( field 3 ) , were observed through @xmath27 and @xmath28 filters . the central co - ordinates of these fields are listed in table 1 . the locations of the fields on the sky are indicated in figure 1 . the positioning of fields 1 and 2 avoids the innermost regions of the galaxy , where resolving stars is problematic ; hence , the main body of ngc 253 is not imaged in its entirety . a five point dither pattern , defining a @xmath29 arcsec square on the sky with a central point , was employed to facilitate the identification and the suppression of bad pixels and cosmic rays . a dither pattern with a substantially larger throw could have been used to fill the gaps between detectors . however , it was decided not to dither over large angular scales since a goal of this study is to assess the distribution of stars in the outer regions of ngc 253 , and large dither offsets would lower the maximum distance that would be sampled with full data coverage . while the gaps between detectors are cosmetically unappealing , they do not hinder the ability to chart large scale trends in the stellar distribution . four 20 sec exposures were obtained per filter at each dither position . the dither pattern was repeated 7 times in @xmath27 and 14 times in @xmath28 for each field . the total exposure time per field is thus 2800 sec in @xmath27 and 5600 sec in @xmath28 . the data were reduced using standard techniques for near - infrared imaging . the initial processing , which was done with the iiwi pipeline at the cfht , consisted of dark subtraction and flat - fielding . a calibration frame to remove interference fringes and the thermal signatures of warm objects along the light path was constructed by combining the iiwi - processed images of the three fields . a mean sky level was subtracted from each image prior to combination , and the portions of fields 1 and 2 that have high stellar density were masked to prevent introducing artifacts into the final calibration frame . the calibration frames constructed in this manner were subtracted from the flat - fielded data , and the results were spatially registered and stacked to obtain the final datasets for photometric analysis . stars in the final images have fwhm @xmath30 arcsec . the photometric measurements were made with the point spread function ( psf ) fitting routine allstar ( stetson & harris 1988 ) . the source catalogs , initial magnitudes , and psfs that are used by allstar were obtained from routines in daophot ( stetson 1987 ) . the criteria for selecting psf stars were a stellar appearance , a brightness that was well above the faint limit of the data , and a lack of bright neighbors . each psf is the result of combining roughly 50 stars . faint sources close to the psf stars were removed iteratively . the photometry was calibrated using standard star observations that were recorded over the same time period as the ngc 253 observations , and the results were checked against measurements in the 2mass point source catalogue . the mean difference between the wircam and 2mass measurements is 0.008 magnitudes in @xmath31 , with a standard deviation of @xmath32 magnitudes . the photometric catalogue was filtered to remove objects with psf - fit errors , @xmath33 , @xmath34 magnitude . this defines the faint limit of the photometry , rejecting objects for which photometry is problematic owing to a low s / n ratio . objects that departed from the dominant relation between @xmath33 and magnitude were also removed , and these tend to be sources that are either moderately extended , in very crowd environments , associated with diffraction spikes , or cosmetic defects . finally , objects with a daophot round parameter that is markedly larger than the norm , which tend to be either galaxies or blended stars , were also removed from the photometric catalogue . the majority of sources that were removed from the photometric catalogue were identified with the two @xmath33 criteria . an example of the relation used to cull objects based on @xmath33 , and the type of objects that are removed or retained , is shown in figure 2 for sources in field 3 . the dashed line in figure 2 is the curve that was used to reject sources ; objects that fall above this curve were deleted from the photometric catalogue . the two highlighted objects with @xmath33 values that fall above the dashed line in figure 2 are clearly galaxies , while the two highlighted objects that fall below this line have a stellar appearance . the culling function that was applied is not overly zealous , and a more draconian cut could have been applied that runs much closer to the stellar locus near @xmath35 . however , it was decided to err on the side of caution , as there is not a clear - cut division between stars and galaxies . extended objects that remain after this initial culling are accounted for in a statistic fashion , using source counts made at large distances from ngc 253 . artificial star experiments were run to assess sample completeness and estimate the uncertainties in the photometeric measurements that are due to photon noise and crowding . artificial stars were assigned @xmath36 , which is the approximate color of the bright end of the agb in ngc 253 . like actual stars in ngc 253 , an artificial star was only considered to be detected if it was recovered in both filters . the artificial stars were subject to the @xmath33 and round culling criteria described earlier in this . the uncertainties in the photometric measurements increase rapidly near the magnitude at which the sample is 50% complete . the 50% completeness magnitude varies across the field , in the sense that it becomes brighter towards progressively more crowded environments . the stellar catalogues are 50% complete near @xmath37 in the disk 10 kpc from the galaxy center , and near @xmath38 in the least crowded portions of all three fields . these magnitudes are consistent with the general appearance of the luminosity functions ( lfs ) discussed in sections 4 and 5 . many nearby galaxies that have superficially mundane morphologies , are actually embedded in highly complex stellar distributions , with sub - structures that are relics of past interactions . as the nearest large external spiral galaxy , m31 has been the target of numerous studies ( e.g. tanaka et al . 2010 ; ibata et al . 2007 , and references therein ) . during the past few years efforts to identify and characterize low surface brightness structures have been extended to more distant galaxies , including m81 ( davidge 2008d ; mouhcine & ibata 2009 ) , m82 ( davidge 2008c ) , ngc 5102 ( davidge 2010 ) , ngc 891 ( mouhcine et al . 2010 ) , ngc 5907 ( martinez - delgado et al . 2008 ) , and ngc 4013 ( martinez - delgado et al . 2009 ) . these studies reveal that tidal structures are a common feature among nearby spiral galaxies , as might be anticipated given the angular momentum content of their disks ( e.g. hammer et al . 2007 ) . there are indications that ngc 253 may also have extraplanar stellar sub - structures . beck et al . ( 1982 ) and malin & hadley ( 1997 ) detect diffuse extraplanar structures in deep photographic images of ngc 253 . while the nature of these structures is not clear , they should be readily apparent in images recorded with digital detectors . moreover , the brightest stars in these structures should also be detectable during good seeing conditions . in this section , the overall stellar distribution in ngc 253 is discussed , with emphasis on identifying features and trends that will be discussed in greater detail in subsequent sections . samples of m giants , c stars , and rsgs have been defined based on location in the @xmath39 cmd . the criteria used here are indicated in the right hand panel of figure 6 ( 4 ) , and figure 3 shows the spatial distribution of each stellar type . a large fraction of the m giant and c star samples at large distances from the center of ngc 253 are actually background galaxies . m giants far outnumber c stars and rsgs , and the distribution of m giants in figure 3 is shown with an expanded scale to facilitate the identification of subtle features . two distinct cross patterns , made up of columns and rows in which stars are abscent , are evident in the upper left hand detector of field 1 . these are centered on bright stars that were targeted by the wircam image stabilization unit , which repeatedly reads the signal from a small region centered on selected stars to fine - tune guiding . while the use of bright stars as guide beacons negates the use of the columns and rows that converge on them for science , the rest of the frame is not affected . while all of the stars considered here are in short - lived stages of evolution , and hence are relatively rare , several structures can still be identified in the stellar distributions . the disk of ngc 253 is the most prominent structure in the distribution of all three stellar types . the cleanest disk distribution is defined by rsgs ; not only are the ages of rsgs such that they are largely confined to the disk plane , but contamination from foreground stars and background galaxies is also modest in the part of the cmds that contains rsgs . there is a clear concentration of rsgs in the north east part of the disk , indicating that this area had a higher sfr than elsewhere in the disk during the past few tens of myr . the bright m giants in the disk define a ring - like structure with dimensions that are similar to the rsg distribution . the central ` hole ' in the m giant distribution is due to crowding in the dense inner disk , which hinders the detection of individual stars . crowding is less of an issue for rsgs in this part of ngc 253 , as these stars are significantly brighter than m giants . while rsgs are confined to a tight disk - like distribution , m giants are found over a larger area . the outer boundary of the m giant distribution is relatively diffuse when compared with the rsg distribution , and the m giants also define structures that may be signatures of an interaction with a companion . the most prominent such structure is the spur in the m giant distribution that extends to the south east of the disk in field 1 . this feature is also evident in the deep images discussed by beck et al . ( 1982 ) and malin & hadley ( 1997 ) . this is an area of h@xmath15 emission , but there is no obvious uv emission , suggesting that there is not a large concentration of young stars ( hoopes et al . 2005 ) . figure 2 of boomsma et al . ( 2005 ) shows a nose in the outer contour of the hi distribution in this part of ngc 253 , suggesting that neutral gas may also be present . this spur is not evident in the c star and rsg distributions , although this may be due to small number statistics . structures similar to this can result from the disruption of a satellite system ( e.g. figure 2 of fardal et al . if ngc 253 was involved in such an event then deeper imaging surveys may find a stream that defines the debris trail left by the companion galaxy . bright m giants can be traced to larger distances along the major axis than rsgs , and this suggests that the most recent episodes of star formation have been confined to small and intermediate radii in the disk . this is contrary to what is seen in the outer regions of many other nearby galaxies , such as m83 ( davidge 2010 ) , where the luminosity - weighted age drops with increasing radius . an absence of gas , and hence young stars , in the outer disk may be a consequence of tidal interactions , as gas is funnelled towards the center of the dominant galaxy and stars are scattered outwards ( e.g. hopkins et al . 2009 ) . the distribution of m giants in the disk of ngc 253 is lop - sided , in the sense that the density of m giants in the outer disk is higher in the north west quadrant than in the south east quadrant . the lop - sided distribution of young stars is also evident in the distribution of uv light ( 6.1.2 ) . zaritsky & rix ( 1997 ) investigate the incidence of such asymmetries in nearby galaxies , and find that they are not rare . they also find that lop - sided galaxies tend to have higher sfrs than symmetric galaxies , and attribute the morphological asymmetry to the accretion of a companion . with the caveat that there are differences in the number of sources , the c star and m giant distributions share many similarities . one exception is a possible concentration of c stars near the minor axis of ngc 253 that is much less obvious in the m giant distribution . the number density of c stars in this part of ngc 253 exceeds that near the edges of the wircam survey , where contaminating objects presumably dominate , at roughly the @xmath40 level . while the presence of stars near the minor axis may suggest an association with the outflow from the central starburst , such a connection is far from ironclad . at present , outflow emission on the south east side of the galaxy is offset from the minor axis ( figure 1 of hoopes et al . perhaps the most compelling reason not to associate the c stars with the outflow is that c stars have ages that are in excess of 0.2 gyr . based on dynamical arguments and the physical extent of the region affected by the outflow , pietsch et al . ( 2000 ) estimate that the outflow has an age of at least 25 myr , which is one to two orders of magnitude shorter than the time frame over which c stars form . another challenge is to explain why the c stars have not dissipated away from the minor axis into the extraplanar field . a long - lived sequence of c stars may result if for example they are part of a coherent stream on a circumpolar orbit . if so , then c stars will also be seen along the north west segment of the minor axis , which was not observed with wircam . while a highly inclined orbital geometry of the supposed companion might seem contrived , the kinematics of material near the center of ngc 253 is consistent with the accretion of an object that was on such an orbit ( anantharamaiah & goss 1996 ; prada et al . 1998 ) . with the exception of foreground stars and background galaxies , the wircam fields contain a mix of sources that belong to the disk and the extraplanar regions of ngc 253 . in the present work , the sorting of the sources into disk and extraplanar components is done using distance from the principle axis of the galaxy , which is a reasonable starting point for investigating large - scale spatial trends . such a classification is possible given that the disk of ngc 253 is viewed almost edge - on . to be sure , sorting objects into broad ` disk ' and ` extraplanar ' components is simplistic , and does not address the presence of physically distinct structures . the extraplanar region defined here almost certainly contains stars that belong to stucturally distinct components , such as a thick disk , a classical halo , tidal features , and possibly stellar ensembles that are associated with the nuclear outflow . this being said , since the wircam observations are restricted to the brightest evolved stars in ngc 253 then the analysis is biased against the detection of structural components that are dominated by old objects , such as the classical halo , and there is a bias towards structures that are populated by objects with ages @xmath41 gyr . indeed , the @xmath31 lfs of agb stars in the extraplanar regions of fields 1 and 2 are very similar ( 5 ) . this result , coupled with large scale distribution of these objects , which is also explored in 5 , indicates that at least from a stellar population perspective the wircam data probe what appears to be a single extraplanar component . deeper photometric studies will allow for a more complete identification and characterization of the other components that make up the disk and extraplanar regions of ngc 253 . the main body of ngc 253 in moderately deep exposures ( e.g. the dss blue image shown in figure 1 ) has a width of @xmath42 arcsec ( 2.3 kpc at the assumed distance of ngc 253 ) , and this is consistent with the distribution of rsgs and m giants in figure 3 . for the purpose of investigating stellar content trends in the disk plane , stars that fall within a 120 arcsec wide strip along the major axis are assumed to be in the disk , and the photometric properties of these objects are discussed in 4 . stars that fall outside of the disk strip , including all of those in field 3 and in the diffuse disk / extraplanar boundary zone , are assumed to be extraplanar , and the photometric properties of these objects are discussed in 5 . the term ` extraplanar ' was intentionally selected to describe these objects because it is very general , and lacks conotations as to the origins of these objects . referring to these as ` halo ' stars might imply to some readers that they are old and metal - poor , which is not the case ( 5 ) . there is cross - contamination between the disk and extraplanar samples . the portions of the extraplanar region that are closest to the disk strip undoubtedly contain some stars that belong to the classical disk , while the disk strip is contaminated by extraplanar stars ; such contamination of the disk sample will be most significant for regions of the disk strip that are at large distances from the galaxy center , and it is demonstrated in 4 that this contamination is modest . the @xmath39 cmds of sources in the disk plane of fields 1 and 2 are shown in figures 4 and 5 . each cmd contains objects in a 4 kpc wide annulus in the plane of the disk , with the inclination of the disk to the line of sight assumed to be 76.1 degrees . the distance listed in each panel , r@xmath43 , refers to the midpoint of the annulus , as measured along the semi - major axis . the panels labelled ` sum ' show the composite cmd of sources in all the distance intervals in that row , and these are shown to illustrate the differences between the objects that are present at r@xmath44 kpc and at r@xmath45 kpc . there is substantial contamination at intermediate and larger r@xmath43 from objects that do not belong to ngc 253 . indeed , the number of sources arcmin@xmath46 is constant at large r@xmath43 , as expected if the number counts are dominated by foreground stars and background galaxies , which should have an approximately uniform distribution when averaged over moderately large angular scales . foreground stars populate the sequence with @xmath47 near @xmath48 , while background galaxies , which are the dominant source of contamination at the magnitudes covered by ngc 253 stars , populate the diffuse , red plume of objects with @xmath49 . comparisons with isochrones ( 4.2 ) indicate that the majority of sources in the r@xmath44 kpc cmds are oxygen - rich agb stars , and these form a sequence with @xmath49 that peaks near @xmath50 ( m@xmath51 ) . moderately bright objects with @xmath52 are seen to the right of the oxygen - rich agb sequence . when compared with source counts at larger r@xmath43 , overdensities of objects with @xmath52 are clearly evident redward of the m giant agb sequence with @xmath31 between 19 and 20 in the 10 kpc ( field 1 ) and 14 kpc ( field 2 ) cmds , and these red objects are c stars . the vast majority of objects above the agb - tip in the r@xmath53 kpc cmds are rsgs . these stars formed during the past few tens of millions of years and are the brightest signatures of recent star formation in the near - infrared . there are almost no rsgs in the 14 kpc cmds , indicating that recent star formation in the ngc 253 disk has largely been restricted to r@xmath53 kpc . there are differences between the field 1 and 2 cmds . the most conspicuous difference occurs at the bright end of the cmds , where there is a well - defined rsg sequence in the r@xmath54 kpc field 2 cmd , but many fewer rsgs , defining a less coherent sequence , in the r@xmath54 kpc field 1 cmd . the number of rsgs in the r@xmath55 kpc cmds of fields 1 and 2 are similar , suggesting that at smaller radii the rsg population is more uniformly distributed . the red stellar content resolved with wircam thus indicates that stars that formed during the past @xmath3 gyr are not uniformly distributed throughout the outer disk of ngc 253 , and that there are large areas where the recent sfr has evidently been much higher than elsewhere in the disk . a more subtle difference between field 1 and 2 is evident between @xmath56 and @xmath57 in the intermediate radii cmds . there is a sudden drop in the prominence of the agb sequence in this magnitude range between 14 and 18 kpc in field 1 . in contrast , the radial decrease in agb stars in the field 2 cmds is more gradual , with agb stars apparent out to r@xmath58 kpc in field 2 . these data thus suggest that the agb content of the ngc 253 disk is azimuthally asymmetric , and this is investigated further in 4.3 . the only published spectroscopic study of hii regions in ngc 253 was conducted by webster & smith ( 1983 ) , and their measurements indicate that [ o / h ] in ngc 253 is 0.4 dex higher than in the lmc . thus , comparisons with isochrones in the present work are confined to solar metallicities , and the r@xmath54 kpc cmds of fields 1 and 2 are compared with solar metallicity isochrones from girardi et al . ( 2002 ) in figure 6 . the error bars in the right hand panel show the @xmath59 scatter in the photometry predicted from the artificial star experiments . a distance modulus of 28.0 , as measured by karachentsev et al . ( 2003 ) from the rgb - tip , has been assumed , along with a line of sight foreground extinction a@xmath60 ( burstein & heiles 1982 ) . this extinction is lower than that measured by schlegel et al . ( 1998 ) , which may be skewed by cool dust in ngc 253 ( e.g. davidge 2008c ) . the models from which the isochrones were constructed assume oxygen - rich atmospheres , and so follow the evolution of m giants , but not c stars . the isochrones indicate that the m giants with @xmath4 between 1 and 1.5 and m@xmath61 and @xmath62 ( i.e. @xmath31 between 20.5 and 19.5 ) have ages @xmath63 a few gyr . as for the oldest stars in ngc 253 , the wircam data sample the agb component and - at least in in low density areas - stars that are evolving at the upper end of the red giant branch ( rgb ) . the peak brightness of the agb concentration is consistent with an age @xmath3 gyr , indicating that the disk of ngc 253 has been actively forming stars throughout the recent past . the 10 myr and 30 myr isochrones define a near - vertical rsg sequence near the bright end of the cmds . there is a rich rsg sequence in the field 2 cmd , and the range of m@xmath64 occupied by these rsgs suggests that they formed @xmath65 myr in the past . rsgs do not form in systems with ages that are much less than 10 myr , and so it is worth noting that the peak of the rsg sequence in field 2 matches that of the 10 myr isochrones to within a few tenths of a magnitude . the rsg locus in field 2 is offset by 0.2 magnitudes in @xmath4 from the 10 myr isochrone . such a difference in color could indicate that the rsgs have a metallicity that is higher than solar . however , this is not consistent with the abundances measured by webster & smith ( 1983 ) from hii regions . the red colors of the rsgs could also indicate uncertainties in the model colors , although m supergiants in the galaxy and the magellanic clouds have @xmath66 ( e.g. elias , frogel , & humphreys 1985 ) , which is consistent with the isochrone predictions . a more likely explanation for the difference in color between the models and rsgs is that there is substantial reddening within ngc 253 . the reader may recall that only foreground extinction was removed from the cmds in figure 6 , and there is evidence for considerable amounts of dust in the ngc 253 disk ( e.g. sofue et al . in fact , internal extinction provides a numerically plausible explanation of the color offset between the rsgs and isochrones in figure 6 . pierce & tully ( 1992 ) apply the method described by tully & fouque ( 1985 ) to compute an internal extinction of a@xmath67 magnitudes for ngc 253 . this statistically - based extinction estimate corresponds to @xmath68 magnitude , and thus accounts for one half of the color difference seen in figure 6 . this is almost certainly a lower limit to the actual internal extinction of the brightest rsgs as young star - forming regions typically have higher levels of extinction than the main body of the disk , and this could amount to many tenths of a magnitude in a@xmath69 ( e.g. zaritsky 1999 ) . therefore , internal extinction is a credible explanation for much ( and possibly all ) of the difference in color between the isochrones and the observed rsg sequence . the areas of the cmd that contain the samples of m giants , c stars , and rsgs that were used to construct figure 3 are indicated in the right hand panel of figure 6 . the rsg area follows the portion of the field 2 r@xmath54 kpc cmd that contains these objects , with the lower magnitude boundary corresponding to the approximate peak brightness of m giant agb stars in ngc 253 . the lower magnitude boundary for m giants at m@xmath70 is a compromise between the growing number of these objects towards fainter m@xmath64 and the need to reduce contamination from background galaxies . this magnitude also corresponds approximately to the onset of thermally pulsing ( tp ) agb evolution , and so most of the m giants plotted in figure 3 are evolving on the tp - agb . the red boundary for m giants is set at @xmath71 , which is the approximate color in the lmc @xmath39 cmd at which a distinct c star plume becomes evident ( nikolaev & weinberg 2002 ) , while the blue boundary of the m giant region at @xmath72 follows the blue envelope of the agb plume . the upper and lower magnitude boundaries of the c star region in figure 6 bracket the magnitude range of cool c stars in the lmc , as defined by the c star plume in the @xmath39 cmd of that galaxy ( e.g. nikolaev & weinberg 2002 ) . the blue limit at @xmath71 separates m giants and c stars in the lmc , while a red color limit of @xmath73 was also imposed . the c star area in figure 6 thus excludes ` warm ' c stars , which have @xmath4 colors that overlap with those of m giants . lacking information on the depths of c - based molecules in the spectra of individual stars , the identification of warm c stars is problematic . however , warm c stars have intrinsically fainter m@xmath64 than their cool brethren . indeed , the @xmath74band brightness of c stars in the lmc ( figure 2 of battinelli et al . 2007 ) drops from m@xmath70 at @xmath75 to m@xmath76 at @xmath77 , which places these stars below the faint limit of the wircam observations . the omission of warm c stars thus does not have a major impact on the luminous end of the c star lf . the lfs of stars with @xmath4 between 0.8 and 1.6 in the disk planes of fields 1 and 2 are compared in figure 7 . while this color range was selected primarily to accomodate m giants , rsgs are also included at the bright end of the lfs . the lfs in figure 7 were corrected for foreground stars and background galaxies by subtracting source counts from the r@xmath78 and 34 kpc annuli . the relation between the peak m@xmath64 of the agb and age predicted from the solar metallicity models of girardi et al . ( 2002 ) is shown at the top of figure 7 . the field 1 and 2 m giant lfs at r@xmath55 and 10 kpc have comparable numbers of stars at m@xmath79 and 8 . given that the age calibration suggests that agb stars with m@xmath79 formed @xmath80 gyr in the past , then it appears that stars that formed during intermediate epochs are uniformly distributed throughout the inner regions of the ngc 253 disk . however , there are obvious differences between the field 1 and 2 r@xmath55 and 10kpc lfs at the bright end , where rsgs dominate the number counts . the number of rsgs at r@xmath54 kpc in field 1 is almost an order of magnitude lower than at the same r@xmath43 in field 2 , confirming the impression obtained from the cmds and figure 3 that the youngest stars in the disk of ngc 253 are not distributed symmetrically throughout the disk . evidence for asymmetries in the ngc 253 disk is also found at larger r@xmath43 . the number densities of m giants at r@xmath81 and 18 kpc in field 2 exceeds those in field 1 by 0.4 0.6 dex . the offset between the field 1 and 2 lfs is roughly constant over the range of m@xmath64 probed with these data , indicating that the difference in number counts is not due solely to the youngest stars . this indicates that while the agb population in the _ inner _ disk of ngc 253 is distributed symmetrically , the number density of agb stars in the _ disk is lop - sided , in the sense of significantly higher densities in field 2 . with the exception of the bright end of the field 2 lf , the shape of the disk plane lfs do not change with radius . neglecting the systematic differences in the number density of stars at r@xmath81 and 18 kpc , then within the ( large ) uncertainties at the bright end of the lfs with r@xmath82 kpc the overall shapes of the field 1 and 2 lfs are similar . the calibration at the top of figure 7 suggests that the brightest m giants in the r@xmath81 and 18 kpc annuli of both fields have ages @xmath83 gyr . the r@xmath58 and 26 kpc lfs were combined to boost the s / n ratio in these regions , and significant numbers of stars are seen in the resulting composite lf . the majority of stars detected in the r@xmath58 26 kpc interval have m@xmath84 , and this is consistent with them being upper rgb stars in moderately metal - poor ( [ m / h ] @xmath85 ) old populations , although they may also be intermediate age agb stars . the agreement between the number counts in the field 1 and field 2 @xmath86 kpc lfs indicates that the lop - sided mass distribution may not extend to the outermost regions of the disk . because they are in a short - lived stage of evolution , and as such are relatively rare , the detection of the most evolved agb stars becomes progressively more problematic with increasing r@xmath43 . this introduces a radial bias in the age estimates , in the sense of overestimating the age of the youngest population as one goes to larger r@xmath43 . if the r@xmath81 kpc lfs are scaled to match the numbers of stars with m@xmath87 at r@xmath58 and 26 kpc , then it is evident that bright agb stars would not be expected in statistically significant numbers . therefore , the possibility that the disk of ngc 253 at very large r@xmath43 may contain stars that formed within the past few 100 myr can not be excluded with the wircam data . the distribution of bright m giants in the disk plane is examined further in figure 8 , where n@xmath88 is the number of stars per arcmin@xmath89 with @xmath31 between 19.5 and 20.5 ( i.e. m@xmath64 between 8.5 and 7.5 ) and @xmath4 between 0.8 and 1.6 . the difference in the number density of m giants in field 1 and 2 , which was noted previously , is clearly evident , and the number counts in field 2 exceed those in field 1 by @xmath2 dex at all r@xmath43 . in addition , the field 2 number counts define an exponential profile that continues to r@xmath90 kpc ( @xmath91 disk scale lengths ) . this is in marked contrast to the field 1 number counts , which follow only an ill - defined exponential , with a plateau in the number counts at r@xmath92 kpc . in 5 it is demonstrated that there is a diffuse extraplanar component , and some of the stars at large r@xmath43 in the disk sample may be stars that are not in the disk plane . however , the majority of stars detected at this large r@xmath43 in the disk sample likely are not extraplanar . indeed , source counts discussed in 5 indicate that the number density of extraplanar stars in field 2 is @xmath93 dex lower than in the field 2 disk plane at roughly the same distance from the galaxy center , and an even greater difference in number density occurs when comparing with the disk and extraplanar lfs of field 1 . thus , the highest stellar densities at large distances from the center of ngc 253 are found in the disk plane . in summary , the wircam data indicate that the disk of ngc 253 extends out to distances of at least 22 26 kpc from the galaxy center . c stars provide additional insights into the star - forming history of ngc 253 . while sharing a common physical characteristic an atmosphere in which [ c / o ] @xmath94 , which results in an absorption spectrum that is dominated by c - based molecules c stars are a heterogeneous group of objects from a stellar content perspective , as they can have ages that range from 0.2 to 3.2 gyr . a further complication is that the fraction of the time during tp - agb evolution that a star spends as a c star depends on metallicity , since as one moves to higher metallicities then more c must be dredged up to bind atmospheric o. the fraction of the tp - agb phase that is spent as a c star thus drops as metallicity increases . the statistics of c stars in the lmc suggest that stars with z = 0.008 that formed 0.45 2.5 gyr in the past will spend at least 50% of their tp - agb evolution as a c star . for comparison , stars with solar metallicity in the same age range will typically spend only @xmath95 of their tp - agb evolution as c stars ( e.g. maraston 2005 ) . bolometric magnitudes for c stars were computed using @xmath74band bolometric corrections from bessell & wood ( 1984 ) for galactic and lmc stars , and the bolometric lfs of c stars in fields 1 and 2 are compared in figure 9 . the relation between peak agb m@xmath96 and age that is predicted by the solar metallicity models of girardi et al . ( 2002 ) is shown at the top of figure 9 . while the models assume o - rich atmospheres , the intrinsic luminosities of o - rich and c - rich stars with the same age and mass should be comparable . due to small number statistics , the brightest c stars give only a lower limit to the age of a system , and stars with m@xmath96 that correspond to that of the agb peak for a given age in the girardi et al . ( 2002 ) calibration must be that age or younger . the c star lfs of disk stars in fields 1 and 2 at r@xmath55 are in excellent agreement , echoing the agreement between the m giant lfs at m@xmath97 noted previously . such a uniform distribution among stars with ages 1 3 gyr is perhaps not surprising , as stars with this age have been in place for many disk rotations , and thus might be expected to be well - mixed throughout the ngc 253 disk . this being said , the shape of the disk c star lf changes with r@xmath43 in both fields . the r@xmath53 kpc c star lfs are relatively flat or climb to higher numbers at the bright end . however , when compared with the lfs at smaller r@xmath43 , the r@xmath81 kpc lfs are deficient in objects at the bright end , while the field 2 lf at r@xmath98 kpc has a distinct downward tilt towards brighter m@xmath96 . this change in lf characteristics likely signals an age gradient in the ngc 253 disk , in the sense that the mean age of c stars increases towards larger r@xmath43 . with the caveat that the measurement of the agb - tip is problematic when only small stellar samples are available , the brightest m giants in the field 1 and 2 @xmath86 kpc lfs have ages @xmath99 gyr if they are evolving near the agb - tip . there is a systematically higher density of c stars in field 2 than in field 1 at r@xmath100 kpc , and this difference grows with increasing r@xmath43 . at r@xmath81 kpc the number density of c stars is 0.5 dex higher in field 2 than in field 1 , which is in line with the difference between the m giant lfs of these fields at the same r@xmath43 . at r@xmath101 kpc c stars have all but disappeared in field 1 , whereas significant numbers are seen in field 2 . m giants are seen in significant numbers out to r@xmath102 kpc in field 1 , with number densities that are comparable to those in field 2 at large r@xmath43 ( see figure 7 ) . if the stellar content in field 1 is like that in field 2 then c stars should be seen at large radii in field 1 , but none are detected . why do c stars disappear in field 1 at large r@xmath43 ? two general explanations for the change in c star counts are considered here : star - forming history and metallicity . the discussion starts with the assumption that the c stars formed _ in situ _ in the ngc 253 disk , although the prospect of them forming in a companion is discussed at the end of the section . assuming a constant sfr and a moderately high metallicity for ngc 253 , then the majority of c stars formed @xmath103 gyr in the past ( figure 12 of maraston 2005 ) . a dearth of c stars at r@xmath101 kpc in field 1 could then result if the eastern half of ngc 253 is deficient in stars that formed 1 2 gyr in the past when compared with the western half of the disk . if this was the case then m giants in field 1 at r@xmath98 kpc would have formed predominantly within the past 1 gyr , and the m giant locus in the outer regions of field 1 would then have a bluer ridgeline color than in field 2 . in fact , the m giant locus in field 1 _ is _ matched approximately by the 0.1 gyr isochrone in figure 6 , while in field 2 the m giant locus roughly coincides with the 1 gyr isochrone . alternatively , such a difference in color could also be due to systematically higher reddening in field 2 when compared with field 1 . metallicity is another quantity that plays a key role in defining c star statistics . however , metallicity is a cumulative quantity that is the result of gyr of evolution . to explain the difference in c star numbers noted above , the star - forming history of field 1 would have had to have differed from that of field 2 for a long period of time if the disk of ngc 253 has evolved as a closed system . this seems unlikely given that mixing will occur throughout the disk during gyr timescales . it is not known if the c stars in the outer regions of ngc 253 formed in ngc 253 , or in an object that was subsequently accreted by ngc 253 . if the latter is correct , then the bulk of the satellite was presumably accreted into field 2 and has not yet been mixed throughout the galaxy ( i.e. the object was accreted within the last gyr ) . the difference in c star content between fields 1 and 2 could then be due to either star - forming history and/or metallicity . metallicity becomes a much more viable explanation for the dominant cause of c star differences if the stars at large r@xmath43 in field 2 were either accreted from a companion galaxy or formed from gas that was accreted from another galaxy , as the chemical evolution of that system was almost certainly very different from that of the ngc 253 disk . the @xmath39 cmds of sources in field 3 are shown in figure 10 , while the cmds of the extraplanar objects in fields 1 and 2 are shown in figures 11 and 12 . as in figures 4 and 5 , the cmds include sources in 4 kpc wide intervals . however , unlike figures 4 and 5 , the midpoint radius of each annulus shown in figures 10 12 is the radial distance from the center of the galaxy measured on the plane of the sky , @xmath104 , as opposed to distance measured in the plane of the disk . composite cmds of all sources in the cmds in each row are also shown . the general appearance of the cmds in figures 10 12 changes gradually with @xmath104 , transitioning from cmds that are dominated by ngc 253 stars to those that are dominated by foreground stars and background galaxies . the cmds with @xmath105 kpc have a prominent plume at @xmath49 and with @xmath106 that is due to m giants , and a modest concentrations of m giants is seen in the cmds out to @xmath107 kpc . there is also a spray of objects with @xmath52 , that is most obvious in the @xmath108 kpc cmd , and many of these are cool c stars . the cmds in figures 10 12 thus indicate that an intermediate age extraplanar component can be traced out to large @xmath104 . in addition , that the extraplanar component extends from field 3 into fields 1 and 2 indicates that intermediate age extraplanar stars are not restricted to the minor axis of ngc 253 , but likely surround the galaxy . while all sources in field 3 have been classified as extraplanar , there is some ambiguity in the physical location within ngc 253 of stars in field 3 . the orientation of ngc 253 on the sky is such that disk stars huddle along a relatively narrow strip , and the amount of disk contamination in the extraplanar regions drops rapidly with increasing z@xmath109 , as the density of disk stars drops and progressively larger extraplanar volumes are sampled . still , some of the bright agb stars in the fields with the smallest @xmath104 in field 3 probably belong to the disk . if the classical disk of ngc 253 extends out to r@xmath110 kpc , as suggested by star counts in fields 1 and 2 , then some objects that have @xmath111 kpc near the minor axis in field 3 may belong to the classical disk , as opposed to being extraplanar , but these will be modest in number given the density of objects in this part of teh disk ( figure 8) . the cmds of sources in field 3 with very different @xmath104 are compared with z = 0.019 isochrones from girardi et al . ( 2002 ) in figure 13 . as with the comparisons in figure 6 , a correction was applied only for foreground extinction , and not internal extinction . given the orientation of ngc 253 on the sky , a small number of sources in the @xmath112 kpc cmd may belong to the classical disk . the cmd of sources with z@xmath113 kpc is also shown to allow the reader to assess by eye the contamination from foreground stars and background galaxies in the @xmath112 kpc cmd . the @xmath112 kpc and @xmath114 kpc intervals sample similar total areas on the sky . when compared with the @xmath114 kpc cmd , the @xmath112 kpc cmd contains a clear excess population of objects with m@xmath115 and @xmath4 between 1 and 1.5 . the 1 gyr isochrone passes through the middle of this population of objects , which are m giants in ngc 253 . the peak of the m giant sequence in the @xmath112 kpc interval indicates that as in the disk of ngc 253 the extraplanar regions of ngc 253 contain large numbers of stars that formed at least 1 gyr in the past . a well - defined rsg sequence is not seen in the @xmath112 kpc cmd , and so there are no signs of very recent star formation in the wircam data at these @xmath104 . the @xmath31 luminosity functions ( lfs ) of extraplanar objects with @xmath4 between 0.8 and 1.6 , which is the color range containing bright m giants , are shown in figure 14 . the field 1 , 2 , and 3 lfs are plotted with solid , dashed , and dotted lines , respectively . as in 4 , the lfs have been corrected statistically for contamination from galactic foreground stars and background galaxies by subtracting source counts at large offsets from the galaxy center . that sources are seen in field 3 in significant numbers out to @xmath116 kpc indicates that an extraplanar component is present , and extends to large distances off of the disk plane , in agreement with the visual impression obtained from the cmds . in addition , there is considerable field - to - field variation in the number density of extraplanar stars at the same z@xmath109 . the spatial distribution of extraplanar stars is skewed to higher values near the disk plane , with the highest concentration of objects in field 2 . such field - to - field differences are suggestive of either a lumpy stellar distribution or a departure from the circular symmetry that is assumed to compute @xmath104 . the number densities of stars with m@xmath117 in field 2 tend to be a few tenths of a dex higher than in field 1 at all r@xmath43 , while the number density of sources in field 3 is roughly an order of magnitude lower than in fields 1 and 2 . the agreement between the source densities in fields 1 and 2 suggests that localized density variations may account for differences in number counts of no more than a few tenths of a dex when averaged over kpc scales . as for a non - cicular projected distribution , the tendency for higher number densities to occur in fields 1 and 2 at a given @xmath104 is indicative of an elliptical distribution , in broad agreement with the overall visual distribution of stars in figure 3 . that the number density of stars with @xmath31 between 20 and 21 at @xmath116 kpc in field 3 matches the number density of stars in the same magnitude range at @xmath118 kpc in fields 1 and 2 suggests that the projected ellipticity of the extraplanar component is @xmath119 . the difference in number densities of extraplanar sources between fields 1 and 2 mirrors the distribution of disk stars , hinting that the disk plane and extraplanar populations identified in this paper are not completely de - coupled , and might have a common heritage . the degree of similarity between the lfs of stars in the disk plane and the extraplapanar regions is investigated in figure 15 , where the extraplanar lfs of fields 1 and 2 are compared with the r@xmath81 kpc disk plane lfs of m giants in the same fields . the disk plane lfs have been scaled to match the mean number counts in the extraplanar lfs at m@xmath120 , and 8.5 . the higher stellar density in the disk elevates the completeness limit in the disk plane lfs with respect to that in the extraplanar regions ( 2 ) . field - to - field differences in completeness notwithstanding , the disk plane and extraplanar lfs have similar shapes at the bright end , signalling similar star - forming histories over a large fraction of the age of the universe . the agreement between the disk and extraplanar lfs in figure 15 is consistent with a disk origin for the extraplanar stars . if this is correct then the disk must have been disrupted not too far in the past , given the similarities between the disk and extraplanar lfs at the bright end . the spatial distribution of extraplanar m giants in ngc 253 provides further insights into their nature and origins . the number density of agb stars with @xmath31 between 19.5 and 20.5 ( i.e. m@xmath64 between 8.5 and 7.5 ) and @xmath4 between 0.8 and 1.6 is shown in figure 16 as functions of z@xmath109 ( top panel ) and z@xmath121 ( lower panel ) . a linear relation in the top panel is indicative of an exponential distribution , as is associated with rotationally - supported disks , while a linear relation in the lower panel is indicative of a distribution that is associated with a classical non - rotating virialized system . there are only subtle differences between the distribution of points in the two panels of figure 16 . while a linear relation with a small amount of scatter can be fit to the points in the top panel , a reasonable linear fit can also be made to the points in the lower panel , albeit with larger scatter than in the top panel . these data thus lack the statistical power to characterize unambiguously the spatial distribution of the brightest m giants in this part of ngc 253 . the structural characteristics of the brightest m giants notwithstanding , the absence of marked discontinuities in the distribution of the brightest m giants suggests that these stars belong to a single population of objects that is well - mixed throughout the extraplanar regions of ngc 253 . a population of objects with @xmath52 is seen throughout the extraplanar regions , and some fraction of these are luminous cool c stars . the m@xmath96 lfs of these objects , constructed following the procedures described in 4 , including a statistical correction for contamination from foreground and background sources , are shown in figure 17 . as with the m giant lfs , field 2 tends to have the highest density of objects . with the caveat that the uncertainties in the field 1 and field 3 lfs tend to be substantial at most @xmath104 , the density of objects in field 3 tends to be comparable to that in field 1 . the c stars thus appear to have a more circular distribution on the sky about the center of ngc 253 than the luminous m giants ; a survey for c stars along the north west minor axis will provide additional information to test the veracity of this result . the c star lfs of all three fields have similar shapes , that appear not to change with @xmath104 . the age calibration at the top of figure 17 suggests that the youngest c stars throughout much of the extraplanar regions have ages @xmath41 gyr , further supporting the notion that the regions surrounding ngc 253 contain stars that formed during cosmologically - recent epochs . the relation between peak agb brightness and age that is shown at the top of figure 14 indicates that subtantial numbers of stars with ages @xmath122 gyr are present throughout the extragalactic regions out to projected distances from the galaxy center that are comparable to that of the physical extent of the disk along the major axis . the ages of the youngest stars , which could be gauged from the peak brightness of the agb , is of obvious interest , but this is a difficult measurement to make . small number statistics are an issue at @xmath123 at almost all @xmath104 , although there are significant numbers of stars with magnitudes that are consistent with an agb - tip age of 0.5 gyr at small @xmath104 . while providing information about the age of the youngest agb stars , the agb - tip has limitations as a probe of stellar content . the use of the agb - tip as the sole age diagnostic is inefficient , in the sense that other age - related information is ignored . in addition , determining the magnitude of the agb - tip can be affected by small number statistics , photometric variability , and the potential for obscuration by thick circumstellar disks , which introduces a potential bias against the detection of the intrinsically most luminous stars in surveys at visible and near - infrared wavelengths . while not providing a pancea for overcoming the shortcomings associated with the agb - tip , the lf of agb stars offers additional information about the mix of stellar ages within a system . the lf also has the merit of using all stars in a sample that are brighter than some magnitude cut - off , as opposed to only those that are the very brightest . in this study , insights into the star - forming history of the extraplanar regions of ngc 253 are obtained by comparing the observed lfs with model lfs that were constructed from routines in starfish ( harris & zaritsky 2001 ) . the goal of this effort is not to model specific features in the observed lfs , but rather to generate lfs that will serve as interpretive guides . the simplest star - forming history is a single episode of star formation that forms stars having the same metallicity - a so - called simple stellar population ( ssp ) . if the bright extraplanar stars in ngc 253 formed _ in situ _ as the result of a single event then they might be expected to have only a modest dispersion in age and metallicity , and hence the lf of these stars could be approximated by that of a ssp . model lfs for ssps generated from the solar metallicity girardi et al . ( 2002 ) isochrones are compared in figure 18 with the mean lf of sources in the @xmath124 and 14 kpc intervals of field 3 . a salpeter mass function has been assumed , although the model lfs in the magnitude range considered here are not sensitive to the properties of the mass function . the field 3 lf is steeper than the ssp model lfs . this departure from flatness is most noticeable when m@xmath125 , and indicates that the agb stars in the extraplanar regions of ngc 253 have a range of ages . thus , the extraplanar stars are not part of an ssp that formed within , say , the past @xmath8 gyr . another interesting reference model is that of a single - metallicity system that has experienced a continuous star - forming history with a constant sfr . this scenario is highly idealized and does not account for chemical evolution . still , the star - forming activity of spiral galaxy disks and dwarf irregular galaxies with m@xmath5 , which is the luminosity range in which stochastic flucuations in star - forming history are greatly reduced ( e.g. weisz et al . 2008 ) , is more - or - less continuous . the model labelled ` constant sfr ' in figure 18 shows the lf of a solar metallicity system that has experienced a constant sfr over the age range 0.2 10 gyr . this model predicts decreasing star numbers towards more luminous m@xmath64 , and the slope of the model lf is roughly consistent with that of the @xmath126 kpc field 3 lf . the agreement between the observations and the constant sfr model lf is not perfect , and it should be recalled that no effort has been made to finesse the star - forming history to match the observations . for example , the break in the @xmath126 kpc lf near m@xmath127 could be the result of a drop in the sfr @xmath128 gyr in the past . other astrophysically relevant improvements would be to include ( 1 ) a large departure from a constant sfr at very early epochs to account for the elevated sfr that is associated with the initial stages of galaxy assembly , and ( 2 ) the effects of chemical enrichment . however , neither of these additions to the models would change the basic conclusion that can be reached from the comparisons in figure 18 : the extraplanar regions of ngc 253 contain stars that span a range of ages . including a sfr spike at early epochs would only impact the two faintest bins plotted in figure 18 . as for the inclusion of chemical enrichment , the lf is dominated by stars with ages @xmath41 gyr . therefore , only very steep age - metallicity relations would have a significant impact on the models , and a steep age - metallicity relation is not consistent with observations of nearby spiral galaxies ( e.g. magrini et al . 2009 ) . the potential for fine - tuning the model lfs aside , it is clear that the lf of extraplanar stars is consistent with them not being part of a coeval system ; rather , the extraplanar regions of ngc 253 are populated by stars that formed over a protracted period of time . deep @xmath27 and @xmath31 images obtained with the cfht wircam have been used to explore the stellar content in the outer regions of the nearby starburst galaxy ngc 253 . this is the first panoramic stellar survey of this galaxy . photometrically - selected samples of rsgs , m giants and c stars are used to trace the spatial distribution of stars with young and intermediate ages . stars are assigned to the disk or the extraplanar regions based on their projected location on the sky ( 1 ) to assess population characteristics , and ( 2 ) to investigate spatial trends and search for structures . while there is inevitable contamination between the disk and extraplanar samples , this is only an issue at modest offsets from the disk plane , near the boundary separating the disk and extraplanar regions . the brightest disk stars in the near - infrared are rsgs , and comparisons with isochrones suggest that the youngest of these have ages @xmath7 myr in ngc 253 . rsgs are highly effective probes of recent star formation by virtue of their intrinsic brightness , and the distribution of rsgs in the ngc 253 disk is lop - sided , with many more in the north east quadrant of the galaxy than in the south west quadrant . unfortunately , small number statistics restrict the usefullness of rsgs as stellar content probes in low - density environments . m giants that are evolving on the agb are more plentiful in composite stellar systems than rsgs , and hence provide a better means of tracing diffuse structures . the photometric depth of the wircam data is such that m giants with ages that span the lifetime of the galaxy are present in these data , although the vast majority of m giants detected here probably have ages @xmath129 gyr . while m giants have the merit of being comparatively easy to detect because of their intrinsic luminosity , the ability to use them to identify individual star - forming episodes diminishes towards progressively older ages . this is because the rate of change of agb - tip brightness with time decreases with increasing age , making efforts to isolate individual large - scale star - forming events that occured more than a few gyr in the past difficult . an added complication is that a significant fraction of stars near the agb - tip are lpvs , and photometric variability blurs information in the lfs . because of their comparatively high numbers , m giants have been adopted as the main tracer of structure in this paper . c stars provide insights into the star - forming history of a system that are complementary to those provided by m giants . whereas the brightest m giants have ages @xmath6 gyr , the majority of c stars have ages 1 3 gyr ( e.g. maraston 2005 ) . that c stars are seen over the same spatial intervals as m giants throughout the disk and extraplanar regions of ngc 253 provides direct evidence that the disk and extraplanar regions of this galaxy contain stars that span a wide range of ages . the extent to which ngc 253 can serve as a template for understanding more distant starburst systems depends on its nature prior to the events that sparked the current starburst if ngc 253 has not had a ` typical ' history then it may not be a good comparison object for probing ` typical ' starburst activity . in fact , the properties of the globular cluster system hint that ngc 253 may not be a textbook spiral galaxy in terms of its past evolution . expanding upon the sample of clusters identified by beasley & sharples ( 2000 ) , olsen et al . ( 2004 ) used deep spectra to identify ngc 253 globular clusters and study their kinematics . olsen et al . ( 2004 ) find that the globular clusters system of ngc 253 is remarkable in two respects . first , it appears to be rotationally supported , with a characteristic scale length that is comparable to that of the present - day hi disk , although there are admittedly large uncertainties in this quantity . while ngc 253 is not unique in having a rotationally - supported globular cluster system , it is the best characterized such system ( olsen 2010 , private communication ) . second , the specific frequency of globular clusters in ngc 253 is more than an order of magnitude lower than in other sculptor group spiral galaxies . that the characteristic size of the star - forming disk of ngc 253 has evidently not changed since early epochs is surprising given the evidence for inside - out disk growth in spiral galaxies ( e.g. trujillo & pohlen 2005 ; colavitti et al . if disks form from the inside - out then globular clusters that formed early - on should have a more compact distribution than the present day stellar disk ( at least assuming that the stellar disk of ngc 253 has not been tidally pruned , which appears not to be the case given the large physical extent of the stellar disk defined by m giants see below ) . a core assumption is that all of the clusters formed as part of ngc 253 , and were not accreted by an interaction with a companion . if ngc 253 did accrete clusters then the companion must have been on a coplanar orbit in order to retain a disk - like cluster distribution in the final merged system . in any event , cluster accretion seems unlikely given the low globular cluster specific frequency of ngc 253 . the presence of a spatially - extended disk - like globular cluster system is also surprising given that major mergers can have a highly disruptive effect on the properties of stars in disks prior to the merger . the vast majority of nearby spirals appear to have experienced such an event during the past few gyr ( hammer et al . 2007 ) . hopkins et al . ( 2009 ) investigate the impact of major ( mass ratios @xmath130 ) mergers on gas - rich disks , and find that while the initial disk of the more massive galaxy may be obliterated , it re - forms on a cosmologically short time scale . existing stars are scattered out of the plane of the primary galaxy , and these subsequently virialize to form a pressure - supported system . an old globular cluster system in a disk galaxy that has experienced a major merger should then be pressure - supported , rather than distributed in a disk . the disk - like globular cluster distribution in ngc 253 is counter to this expectation , unless it experienced a major merger with a companion with an orbit that was coplanar with the ngc 253 disk . such a merger geometry is conducive to producing starburst activity and maintaining asymmetries in the disk ( such as those discussed in 4 ) if the companion orbit is prograde ( hopkins et al . 2009 ) . the sculptor group is dynamically unevolved ( karachentev 2005 ) , and ngc 253 may simply be one of the @xmath131 of nearby spirals that have avoided a major merger ( hammer et al . 2007 ) . major mergers involving gas - rich systems produce environments that are conducive to globular cluster formation ( e.g. ashman & zepf 1992 ) . therefore , an uneventful interaction history could explain the low specific frequency of globular clusters in ngc 253 . if ngc 253 has not experienced a major merger for a large fraction of the hubble time then it might have a relatively compact stellar disk when compared with other nearby galaxies , due to a relative deficiency in angular momentum ( hammer et al . m81 has an m@xmath64 that is comparable to that of ngc 253 , and the disk scale length of ngc 253 ( 1.7 kpc ; forbes & depoy 1992 ) is comparable to that measured for m81 from the @xmath74band surface brightness profiles given by jarrett et al . the wircam data also indicate that the physical extent of the ngc 253 disk is not compact . agb stars in the disk are traced out to at least 22 kpc along the major axis ( 4 ) , which corresponds to 13 scale lengths . given the diffuse appearance of the outer disk in figure 3 , it is possible that the stars seen at large r@xmath43 may not be in the classical disk ; rather , they could be the result of tidal interactions . still , that stars are detected out to this distance in both fields 1 and 2 with comparable number densities argues that the majority of these stars are part of a dynamically mature component , which we associate with the disk . the extent of the stellar disk in ngc 253 is thus comparable to the distances out to which stars have been resolved in ngc 247 ( davidge 2006 ) and m81 ( davidge 2009 ) , and exceeds the stellar disk of ngc 2403 ( davidge 2007 ) . we conclude that the stellar disk of ngc 253 is not compact when compared with other nearby galaxies . if , as the other evidence suggests , ngc 253 avoided mergers up to the recent epochs , then ngc 253 may have received angular momentum from torques exerted by ngc 247 ( whiting 1999 ) . in fact , the disk of ngc 253 is likely even more extended than indicated by agb stars . the holmberg radius of ngc 253 is 17.6 arcmin ( puche et al . 1991 ) , or 21 kpc . bright agb stars are relatively rare , and deeper images that sample stars on the upper portions of the rgb , will probably trace the stellar disk of ngc 253 out to even larger radii . blue main sequence stars may also provide a powerful means of plumbing the outer limits of the ngc 253 disk . the main sequence phase of evolution has a much longer duration than the agb phase , so that main sequence stars will occur in much larger numbers than agb stars . moreover , the use of blue stars also results in greater contrast with respect to background galaxies , the majority of which have red colors ( e.g. davidge 2008d ) . the spatial distribution of recent star formation can be used to probe the recent history of the galaxy . starbursts are most commonly associated with galaxy - galaxy interactions and so the disk of ngc 253 might be expected to harbour signatures of a tidal encounter . in fact , evidence for such an interaction is evident in the sub - structure close to the disk plane ( figure 3 ) . if there was an interaction then it evidently produced elevated levels of star formation not only in the inner regions of the galaxy , but also in the disk . indeed , the high density of rsgs in the north east portion of ngc 253 indicates that this was an area of very active star formation within the past few tens of myr . this interpretation is supported by the presence of nuv and fuv emission ( hoopes et al . 2005 ) and intense h@xmath15 emission ( hoopes et al . 1996 ) in this part of the galaxy . the bright h@xmath15 emission indicates that the elevated star - forming activity heralded by the rsgs almost certainly continues up to the current time . while the south west part of the disk is not devoid of h@xmath15 emission and uv - bright objects , the sources there are fainter and less numerous than at the other end of the major axis . the north east portion of the ngc 253 thus appears to have had the highest levels of star formation in the main body of the disk during at least the past few tens of myr . this part of the galaxy also shows an enhancement in the number of bright agb stars with respect to the south east arm of the major axis ( 4 ) , suggesting that excess star formation in this area has occured for hundreds of gyr . while the orbital properties of the supposed interacting galaxy are not known , there are clues to possible trajectories . the distribution of stars in figure 3 is suggestive of an encounter involving a companion with an orbit that was close to the ngc 253 disk plane , and this is consistent with the flattened distribution of extraplanar m giants measured in 5 . dynamically de - coupled material near the galaxy center ( anantharmaiah & goss 1996 ; prada et al . 1998 ) suggests that the encounter may also have involved the central regions of the galaxy . the range of interaction geometries investigated by hopkins et al . ( 2009 ) indicate that the most pronounced asymmetries in the disk result from prograde encounters in which the companion orbit is close to that of the disk plane of the larger galaxy . the lop - sided nature of the ngc 253 disk thus suggests that the galaxy that interacted with ngc 253 was probably not on a retrograde orbit . we are not aware of published studies that investigate an interaction between ngc 253 and a possible companion . however , interactions between m31 , which has an integrated brightness that is similar to that of ngc 253 and is known to be surrounded by tidal features , and a companion have been examined . there is some similarity between the general appearance of m31 with its tidal structures in figure 1 of fardal et al . ( 2007 ) and ngc 253 in figure 3 of the current paper . fardal et al . ( 2007 ; 2008 ) investigate the interaction between a single satellite and the m31 disk . the main emphasis of these studies was to model the tidal features that have been found near m31 , and the orbital pericenter of the companion is a few kpc ( fardal et al . in addition to producing the m31 giant stellar stream , these models produce structures that are close to the observed disk plane . while originating in the satellite , the stars in these structures have a spatial distribution that is defined largely by the m31 gavitational potential . if the extraplanar population in ngc 253 is analogous to that in m31 then the fardal et al . ( 2007 ; 2008 ) simulations indicate that the extraplanar regions of ngc 253 are populated predominantly by stars that originated in a companion galaxy . an interaction with a satellite will also affect the stellar distribution in the m31 disk . block et al . ( 2006 ) consider a companion ( presumably m32 ) that passes through the center of m31 , and that study provides insights into signatures in the disk that such an encounter might produce . the interaction in that model produces a density wave that propogates outwards from the point of impact . this wave spurs star formation as it moves along , and localized regions of star formation in ring- or arm - like structures result ( figure 2 of block et al . 2006 ) . is there evidence for a collisionally - induced propagating density wave in ngc 253 ? galex uv images are examined in an effort to answer this question . the nuv and fuv morphology of ngc 253 , corrected to show the galaxy as it would appear if viewed face - on , is shown in figure 19 . the de - projection involved stretching the image along the minor axis to simulate tilting the line of sight , and then applying a deconvolution filter to correct for distortions introduced by this process . the overall distribution of star - forming regions is not affected by the deconvolution process . no attempt was made to correct for the inherent thickness of the ngc 253 disk , and emission from the outflow blurs the right and left hand edges of the disk in the de - projected images . the position angle and ellipticity measured from 2mass images ( jarrett et al . 2003 ) were adopted , and residual oblateness in the de - projected images suggests that the ellipticity was underestimated by a small amount . the side lobes that bracket bright point sources are artifacts of the deconvolution process . the de - projected @xmath31-band 2mass image of ngc 253 is also shown in figure 19 . this is an interesting comparison dataset as near - infrared light traces mass , whereas uv light mainly traces young stars ; hence , differences between the near - infrared and uv images are expected . the barred morphology of the central regions of the galaxy is clearly evident in the @xmath31 image , as are the two spiral arms that emerge from the ends of the bar . the high dust content of ngc 253 almost certainly affects the appearance of the galaxy in the uv , and the bar that is prominent in the near - infrared is invisible in the uv . dust in the outflow likely produces the streaks in the right half of the southern disk of the uv images as well as the lack of uv emission immediately to the left of the nucleus . the uv sources loosely follow the spiral structure that is conspicuous in the near - infrared . the agreement between the near - infrared arms and uv sources in the northern half of the galaxy is reasonable ; while there is some uv emission along the spiral arm , the brightest sources of uv emission fall along the leading edge of the spiral arm , as expected if the leading edge of the density wave compresses the ism and triggers star formation . there is similar agreement between uv emission and spiral structure in the southern half of the disk . with the caveat that there is substantial extinction throughout the ngc 253 disk , there is no evidence for ring - like distribution of star - forming regions . if a galaxy passed through the ngc 253 disk then it did not leave a signature like that predicted by the encounter modelled by block et al . this might indicate that the companion was not as massive as m32 and/or that the passage had an oblique collision geometry . we close the discussion of the disk by noting that the north east disk of ngc 253 bisects the extraplanar hi found by boomsma et al . ( 2005 ) , suggesting a causal relationship between the elevated sfrs and the gas emission . the high levels of recent star formation in the north east portion of the ngc 253 disk may have an impact on the circumgalactic environment through the formation of chimneys , which are conduits for channeling gas out of the disk plane ( e.g. norman & ikeuchi 1989 ) . significant mass outflow rates do not require large sfrs ( e.g. ceverino & klypin 2009 ) . melioli et al . ( 2009 ) model the impact on the extraplanar environment of sne from a star - forming complex in a milky way - like system . their simulations indicate that the moderately compact concentration of star - forming regions produces a vertical plume of gas that diffuses only modest distances from the chimney axis after one disk rotation ( @xmath132 myr in their model ) . gas is ejected @xmath133 kpc above the disk plane in these models , whereas the extraplanar hi near ngc 253 is traced out to 12 kpc ( boomsma et al . figure 2 of boomsma et al . ( 2005 ) shows extended spurs in the extraplanar hi emission , especially the column that extends to the north west of the disk plane , and this is consistent with the morphology of the chimney outflow predicted by melioli et al . ( 2009 ) . while the arguements presented here are not ironclad , feedback from a ` boiling disk ' ( sofue et al . 1994 ) , of which the concentration of rsgs in the north east disk are the most conspicuous manifestation at near - infrared wavelengths , is a viable mechanism for producing the extraplanar hi in ngc 253 . a diffuse extraplanar stellar component , which comparisons with model lfs in 5 indicate contains stars that formed over gyr or longer timescales , is seen in all three wircam fields . the distribution on the sky of the extraplanar component appears to be flattened , with an ellipticity near 0.4 . it can be anticipated that stars belonging to this extraplanar component should also be found to the north of the galaxy disk . the extraplanar regions of ngc 253 contain clues about the events that triggered the starburst . if the starburst was triggered by a tidal encounter with another galaxy then detritus from the interaction may lurk in the extraplanar regions , and evidence for such material is seen in figure 3 . while the extraplanar stars are widespread , girdling at least the southern part of the galaxy , there are minor field - to - field differences in extraplanar stellar density , indicating that the extraplanar stars are not yet well mixed throughout the halo . smaller scale structures , such as streams and shells , may become better characterised with deeper images that sample larger numbers of objects , such as stars on the rgb . in a related vein , studies of the present - day extraplanar component can also be used to predict the future appearance of the ngc 253 ` halo ' ; for example , will an observer in the distant future find that the halo of ngc 253 contains a complex mix of stellar components , the origins of which can only be identified through extremely deep photometric studies ? to the extent that the simulations run by fardal et al . ( 2007 ; 2008 ) can be applied to ngc 253 , then any tidal streams may have lower surface brightnesses than the morass of stars that surround the ngc 253 disk . selection effects may also come into play , as higher surface brightness tidal structures tend to be associated with material accreted from more metal - rich companions that have been recently accreted ( gilbert et al . hence , the prominence of stellar tidal streams provides clues about the nature of the companion and the time of the encounter . the extraplanar regions of ngc 253 contain a complex mix of stellar populations . ngc 253 is almost certainly surrounded by an old , classical stellar halo , although the spatial distribution and specific frequency of globular clusters hints that the classical halo may be even more poorly populated than the classical halos around other galaxies , and have a highly flattened morphology . comeron et al . ( 2001 ) find stars with young ages near the minor axis of ngc 253 , that presumably are associated with the outflow from the nuclear starburst . the first stars that formed in the outflow ( i.e. at least a few tens of myr in the past ; pietsch et al . 2000 ) may have diffused away from their places of birth by now , and the dominant gravitation influence of ngc 253 will pull the young stars towards the disk plane , where they may ultimately form a more compact , kinematically relaxed distribution . a fraction of stars that formed in the outflow may also obtain relatively high velocities ( e.g. stone 1991 ) , allowing them to move large distances from their places of birth . the diffuse extraplanar component contains stars that are far too young to belong to a classical halo , while the wide age span ( @xmath134 gyr ) deduced from the m giant lf is not consistent with formation in the outflow , which has probably been in place for ( at most ) only a few 100 myr ( i.e. since the interaction with a companion ) . the uniform mix of the extraplanar population over large spatial scales , coupled with a lf that is indicative of an extended star - forming history that is typical of disk systems , is consistent with the stars having formed in an environment other than the extraplanar regions of ngc 253 . while it is argued here that the stars did not form in the extraplanar regions , pockets of star formation in such environments are not without precedent ( e.g. tullmann et al . 2003 ) , and _ in situ _ formation has been suggested as a possible mechanism for the formation of young stellar ensembles near m82 ( davidge 2008e ) . two possible sources of origin for the diffusely distributed extraplanar stars are the disk of ngc 253 or a gas - rich companion galaxy with m@xmath5 . an interaction between ngc 253 and another galaxy could perturb the orbits of stars in the ngc 253 disk , and produce an extraplanar component . as for an origin in a dwarf galaxy , ngc 253 does not have an entourage of companions , and so the companion in question must have been completely destroyed or absorbed by ngc 253 . the chemical composition of stars in the extraplanar regions of ngc 253 will provide obvious clues into their origins . if these stars have near - solar metallicities than this would be indicative of having formed in the disk , whereas a metallicity that is , for example , one - tenth solar would favor formation in a lower luminosity dwarf galaxy . the photometric properties of rgb stars are one means of estimating metallicities , albeit for stars that formed at least a few gyr in the past . the deep cmd discussed by mouchine ( 2006 ) suggests that the majority of stars in the extraplanar regions near the minor axis have a mean [ m / h ] @xmath135 , ostensibly suggesting an origin in a dwarf companion . however , this mean metallicity assumes that the stars are all ` old ' . if the extraplanar regions are populated by stars that formed over a range of ages , then a large fraction may have ages @xmath136 gyr , and the metallicities of these objects will be underestimated if they are assumed to be old . the combination of visible and near - infrared data would provide considerable wavelength leverage for measuring improved metallicities . still , the ability to measure the characteristic metallicity of rgb stars may be complicated by the extensive dust content of the galaxy . differential reddening causes conspicuous blurring of the cmds of ngc 253 in the disk ( e.g. the discussion of the karachentsev et al . 2003 and dalcanton et al . 2009 cmds in 1 of this paper ) . there is also evidence for substantial quantities of dust in the extraplanar regions of the galaxy ( kaneda et al . 2009 ) , which probably was ejected by winds from star - forming regions . the broadening of sequences on the cmds of ngc 253 fields caused by differential extinction may be mistaken for a mixture of ages and/or metallicities . moderate resolution specroscopy provides a means of measuring metallicities that is less susceptible to differential extinction . given the low temperatures of the brightest stars in the extraplanar regions of ngc 253 , coupled with the need for high angular resolutions to resolve sources at small offsets from the disk plane , near - infrared spectra are of obvious interest , and the feasibility of obtaining such data is briefly discussed here . the near - infrared spectral region contains a number of metallicity - sensitive features that can be examined with spectral resolutions of @xmath137 , and the calibration of these has been investigated by frogel et al . the brightest m giants in the extraplanar regions of ngc 253 have @xmath138 , and experience with spectrographs on 8 10 meter telescopes indicates that total on - source exposure times of hundreds of hours would be required to obtain a spectrum with a s / n = 100 of a @xmath50 source with a resolution @xmath139 . however , the exposure time is markedly lower if the observations are conducted with a 30 meter telescope working at the diffraction limit , due to the d@xmath140 advantage . indeed , a simple scaling of the exposure times required for 8 meter telescopes suggests that spectra with a suitable s / n ratio could be obtained with exposure times of only a few hours with a 30 metre telescope . thus , while obtaining spectra of the brightest red stars is not feasible with 8 10 metre telescopes , this is a project that is well within the grasp of a 30 meter telescope . ngc 247 is the nearest large galaxy to ngc 253 . could ngc 253 and ngc 247 have interacted in the not too distant past ? an interaction between ngc 247 and ngc 253 seems unlikely at first blush , as ngc 247 does not show classical signatures of an interaction , such as a central starburst or extended tidal arms . surveys also have not found hi streams near ngc 247 or ngc 253 ( haynes & roberts 1979 ; putman et al . 2003 ) that could be part of a debris trail . still , the disks of these galaxies are oriented in a manner that is consistent with mutual tidal torquing ( whiting 1999 ) . in addition , ngc 247 like ngc 253 has a truncated hi disk ( carignan & puche 1990 ) , a tendency for mean age to increase with radius in the outer disk ( davidge 2006 ) , and what is perhaps most suggestive of a recent encounter an intermediate age extraplanar population ( davidge 2006 ) . if ngc 247 and ngc 253 had a close encounter then tidal fragments may have been removed from one or both galaxies . while there are a number of nearby dwarf irregular and dwarf spheroidal galaxies , these are located to the north west of ngc 253 and are clustered near ngc 247 ( e.g. figure 3 of karachentsev et al . could some of the actively star - forming dwarf galaxies near ngc 247 be tidal dwarfs ? tidal dwarfs have characteristics that are distinct from those of classical galaxies that formed within primordial dark matter halos . the gas and stars that tidal forces displace from a galaxy quickly develop different kinematic properties , as gas is a collisional system that can dissipate energy , while stellar ensembles are collisionless systems . gas may thus form concentrated pockets where star formation may subsequently occur . because tidal dwarfs are not in dark matter dominated halos , they have structural properties that are distinct from those of classical galaxies , and may dissipate on timescales that are significantly less than the hubbble time . in addition , they will not have an underlying concentrated substrate of older stars , of which rgb stars are among the most prominent signature . finally , if the tidal gas came from a massive galaxy then the stars in a tidal dwarf will have a metallicity that differs from that expected from the [ m / h ] m@xmath141 relation defined by classical galaxies . there are two galaxies close to ngc 247 on the sky that have conspicuous blue populations : ddo 6 and ddo 226 . karachentsev et al . ( 2003 ) present @xmath142 cmds of both galaxies that are based on moderately - deep wfpc2 observations . the cmd of ddo 6 has a prominent rgb , indicating that it contains stars with ages of at least a few gyr . a subsequent investigation of the star - forming history of ddo 6 by weisz et al . ( 2006 ) suggests that the majority of stars in ddo 6 formed 12 14 gyr in the past , and that star formation was only rejuventated within the past 0.5 gyr . thus , based on the presence of a large old population , ddo 6 is not a tidal dwarf . the situation might be different for ddo 226 , the cmd of which lacks a prominent rgb . there _ is _ a concentration of red stars in the cmd with @xmath143 , but these do not form a well - defined sequence . working on the assumption that the red stars are on the rgb , karachentsev et al . ( 2003 ) assign ddo 226 a distance modulus of 28.5 , placing it @xmath8 mpc beyond ngc 247 and ngc 253 . deeper images will help establish the nature of ddo 226 . more specifically , the detection of a well - defined rgb would indicate that ddo 6 is not a tidal fragment . the manner and timescale in which a galaxy and its companions evolve depends on initial conditions , such as the orbital characteristics of satellites , and environment , including the distances to the nearest large galaxies . while no two galaxies have the same initial conditions , and hence do not have the same evolutionary histories , the study of resolved stars can be used to identify nearby galaxies that may have experienced similar events , but are in different stages of their evolution and as such may appear to be unrelated . identifying such relationships will lead to a better understanding of galaxy evolution , as it will then be possible to directly probe the influence of for example local environment . m31 is potentially an interesting comparison object for ngc 253 . not only is m31 sufficiently close that it can be studied in considerable detail , but ngc 253 and m31 have similar m@xmath141 and m@xmath64 . the distribution of stars outside of the disk plane in m31 ( e.g. figure 1 of fardal et al . 2007 ) is also reminiscent of that seen in ngc 253 . still , the validity of pairing m31 and ngc 253 may seem dubious , given the obvious differences in overall morphology and environment . indeed , ngc 253 lacks the large bulge that dominates the central regions of m31 , while m31 is not presently experiencing a starburst . ngc 253 also lacks an entourage of companions , whereas m31 has many accompanying galaxies . the extraplanar structure around m31 indicates that it has had an extremely eventful past , and the number of companions in the past was probably even greater than at present . evidence for this comes in the form of conspicuous streams and shells , which are the remnants of one or more satellites that have all but disintegrated ( e.g. fardal et al . 2007 , 2008 ; ibata et al . 2007 ; koch et al . 2008 ) ; tanaka et al . ( 2010 ) distinguish 16 substructures in the outer regions of m31 . the globular cluster system of m31 is also highly inhomogeneous , suggesting a diversity in origins ( beasley et al . 2005 ) . in addition to the well - documented structures in the outer regions of m31 , there is also evidence of a diffuse extraplanar stellar component . some of these stars belong to a classical metal - poor halo ( e.g. chapman et al . 2006 ; kalirai et al . 2006 ; ibata et al . 2007 ; koch et al . 2008 ) . however , the majority of stars in all but the outermost fields have characteristic ages and metallicities that differ substantially from those associated with a classical halo ( e.g. bellazzini et al . 2003 ; brown et al . 2007 ; richardson et al . 2009 ; tanaka et al . 2010 ) . in addition to classical halo stars and tidal debris , brown et al . ( 2007 ) argue that the extraplanar regions of m31 also contain stars that formed in the disk of m31 , and were displaced during intermediate epochs in a merger event . evidence has been presented in this paper that a similar event may have happened in ngc 253 , probably within the past few 100 myr . if ngc 253 evolves in isolation in the future and does not experience another interaction , which seems likely given its isolation , then the diffuse extraplanar component will evolve passively , slowly fading and becoming less conspicuous ; the differences between the old populations that have been in place since the initial formation of ngc 253 and the diffuse extraplanar component that formed recently will thus become less obvious , as the main sequence turn - off of the most recently formed stars dims . the end result will be a mix of old and intermediate age stars smilar to what is seen in the outer regions of m31 today . despite the obvious differences in the evolution of ngc 253 and m31 , the present day ngc 253 may provide a glimpse of a younger m31 . the wircam observations have revealed extraplanar agb stars with ages of a few hundred myr throughout the southern half of ngc 253 . there should then be a diffuse population of main sequence turn - off ( msto ) stars with masses @xmath144 m@xmath12 . while such blue stars are too faint to be detected with wircam , they are relatively bright at near - ultraviolet ( nuv ) and far - ultraviolet ( fuv ) wavelengths , and if they are present in sufficient numbers then they will produce a diffuse uv halo surrounding ngc 253 . given that hotter , younger main sequence stars should be confined to the thin disk of ngc 253 , then a uv color gradient off of the disk plane might also be expected . archival galex nuv and fuv images of ngc 253 , which are those described by hoopes et al . ( 2005 ) , were used to search for emission from a diffuse population of moderately bright msto stars . a @xmath145 kpc wide region on either side of the minor axis of ngc 253 was excluded from the analysis , to avoid confusion with sources of emission associated with the outflow . to facilitate the analysis , the uv emission from intermediate age stars was assumed to be axially symmetric about both the major and minor axes . this assumption boosts the s / n ratio , and is reasonable given that bright extraplanar agb stars permeate the wircam fields , and so likely surround the galaxy . nuv and fuv light profiles were obtained by averaging the data in the direction parallel to the disk plane at distances between 5.5 and 14 kpc from the minor axis . nuv and fuv light above the background level is detected out to @xmath146 kpc , in good agreement with the spatial distribution of m giants in field 3 . in addition , while the fuv nuv color is constant between 0 and 10 kpc from the major axis , it becomes @xmath147 magnitudes redder between @xmath124 and 15 kpc , signalling a change in the sed of the dominant sources of emission . the galex data thus indicate that there is diffuse extraplanar uv emission around ngc 253 , as predicted from the star counts at near - infrared wavelengths . this emission is not associated with the outflow ( e.g. light from the nucleus that is scattered by outflowing dust ) , as a @xmath145 kpc ` exclusion zone ' was employed on either side of the minor axis when examining the galex data . the spatial extent of the uv emission off of the disk plane is consistent with that determined from resolved agb stars . that the fuv nuv color within 10 kpc of the disk plane is constant is consistent with a disk - like stellar mix , again in agreement with the study of resolved stars , which indicate that the extraplanar agb lf is similar to that of disk stars . the wircam and galex data allow predictions to be made about the extraplanar regions of ngc 253 that can be tested with deeper observations . imaging studies of the extraplanar regions of ngc 253 at visible wavelengths should reveal a population of blue msto stars with @xmath148 , that collectively are the source of the diffuse uv emission . two predictions can also be made about the distribution of these objects . first , they will be seen out to @xmath149 kpc above the disk plane , based on the detection of bright agb stars and uv emission out to this same distance . second , the uv color profile suggests that the luminosity - weighted age changes @xmath7 kpc from the major axis , in the sense that a slightly older population dominates when @xmath150 kpc than at smaller distances . there should then be a change in the brightness of the msto near @xmath151 kpc .	near - infrared images obtained with wircam on the canada - france - hawaii telescope are used to investigate the recent history of the nearby sculptor group spiral ngc 253 , which is one of the nearest starburst galaxies . bright agb stars are traced out to projected distances of @xmath0 26 kpc ( @xmath1 disk scale lengths ) along the major axis . the distribution of stars in the disk is lop - sided , in the sense that the projected density of agb stars in the north east portion of the disk between 10 and 20 kpc from the galaxy center is @xmath2 dex higher than on the opposite side of the galaxy . a large population of red supergiants is also found in the north east portion of the disk and , with the exception of the central 2 kpc , this area appears to have been the site of the highest levels of star - forming activity in the galaxy during the past @xmath3 gyr . it is argued that such high levels of localized star formation may have produced a fountain that ejected material from the disk , and the extraplanar hi detected by boomsma et al . ( 2005 ) may be one manifestation of such activity . diffuse stellar structures are found in the periphery of the disk , and the most prominent of these is to the south and east of the galaxy . bright agb stars , including cool c stars that are identified based on their @xmath4 colors , are detected out to 15 kpc above the disk plane , and these are part of a diffusely distributed , flattened extraplanar component . comparisons between observed and model luminosity functions suggest that the extraplanar regions contain stars that formed throughout much of the age of the universe . additional evidence of a diffuse , extraplanar stellar component that contains moderately young stars comes from archival galex images . it is suggested that the disk of ngc 253 was disrupted by a tidal encounter with a now defunct companion . this encounter introduced asymmetries that remain to this day , and the projected distribution of stars in and around ngc 253 suggests that the companion had an orbit that was prograde and moderately inclined to the ngc 253 disk . the star - forming history of the extraplanar stars suggests that they either originated in the ngc 253 disk , or in a gas - rich companion . in the latter case the companion must have had an initial m@xmath5 in order to produce the more - or - less continuous star - forming history that is suggested by the stellar content . the ages of the youngest extraplanar stars suggests that the event that produced the extraplanar population , and presumably induced the starburst , occured within the past @xmath6 gyr . = 1.0 cm
the cold dark matter(cdm ) scenario@xcite has been the standard theory of the formation and evolution of the structures in the universe . in this scenario , galaxies and clusters of galaxies are formed bottom - up . it has been remarkably successful in explaining the large scale structures @xcite and numerous observational results ( e.g. @xcite ) . until recently , however , the small - scale structures of cdm , like subhalos in a galaxy - sized halo , could not be studied by numerical simulation because of the lack of the computational power . recent improvement of computational powers made it possible to study not only dark matter halos , but subhalos in a parent halo . usually these subhalos are interpreted as corresponding to galaxies in cluster - sized halos and satellite dwarf galaxies in galaxy - sized halos . @xcite and ( 1999 , hereafter m99 ) reported that about 1000 subhalos formed in a simulated galaxy - sized halo . the number distribution of subhalos as the function of the circular velocity normalized by that of the parent halo turned out to be remarkably similar for galaxy - sized and cluster - sized halos . this result , of course , is the direct outcome of the scale - free nature of the gravity and almost power - law amplitude of the density fluctuation . this similarity , which looks quite natural from the theoretical side , showed serious discrepancy with observation . the number of subhalos in the simulated cdm halo agreed well with the observation of the cluster of galaxies , while the simulation of a galaxy - sized halo predicted too many dwarf - sized halos . this discrepancy is now known as the `` dwarf galaxy problem '' , or `` missing - dwarf problem '' , and has been one of the main topics in the study of galaxy formation and structure formation in the universe . solutions so far proposed for this problem can be classified into the following two categories : * dark matter is different from the cdm model in small scales . * only a small fraction of subhalos is observed as dwarf galaxies . many modified dark matter models had been proposed . self interacting dark matter@xcite is a cdm with small self - interaction cross section , which dumps small - scale fluctuation . warm dark matter@xcite has a power spectrum which has cut off at small scale . some simulations with these dark matter models have been done . ( e.g. @xcite , @xcite ) if we assume that the standard cdm is correct and consider the possibility that not all subhalos are observed as dwarf galaxies , the question is which subhalos are observed as dwarf galaxies . @xcite argued that observed dwarf galaxies are presently most massive subhalos , by comparing numerical result with observation of velocity dispersion profile of dwarf galaxies . on the other hand , ( 2004 , hereafter k04 ) tracked each subhalo s evolution and showed many low - mass subhalos were massive in the past . additionally , they introduced the model of star formation condition , and claimed that the number and distribution of satellite dsph galaxies was in agreement with that of such previously massive subhalos . before jumping to such a solution , however , we should understand what is really the problem . in previous studies , in order to determine the number of subhalos with circular velocity @xmath5 , the detection limit was set to a few tens of particles . it is not clear whether or not we can rely on the result obtained with such a small number of particles . both softening parameters and the two - body relaxation might significantly affect mass and density profiles of subhalos ( e.g. @xcite , @xcite ) . @xcite showed that the low mass end of the circular velocity distribution of subhalos is affected by numerical resolution by comparing the results of two cluster - scale simulations with different resolutions . same tendency can be seen in the more recent studies ( e.g. @xcite , @xcite and @xcite ) . this means that the agreement in a cluster scale described in m99 might be a numerical artifact . in the present paper , as a first step to understand the formation and evolution of subhalos , we investigated the reliability of size distribution of subhalos in simulation results . we performed large - scale @xmath0-body simulations of formation of cdm halos . we simulated both galaxy - sized and cluster - sized halos . we carried out @xmath0-body simulations of nine halos of different masses and resolutions and analyzed the distribution of subhalos . from the results of these simulations , we estimated the effect of the mass resolution on the distribution of subhalos . when the number of particles in one subhalo is small , both the mass and the circular velocity of that halo are affected , resulting in the deviation from simple power - law dependence . we formulated the reliability criterion for mass and circular velocity . we then compared the velocity distribution function obtained by @xmath0-body simulations with that of observed galaxies in the virgo cluster . we found that the observational result is in good agreement with our medium - resolution result , which is affected by small-@xmath0 effect around @xmath6 . in other words , our high - resolution simulation predicts the overabundance of moderate - sized galaxies . previously the mass function of galaxies in a cluster has been believed to be well reproduced by cdm simulations ( e.g. m99 ) . our result shows that the agreement , in the range of @xmath7 , is due to the lack of mass resolution in previous simulations . the discrepancy between the simulated halos and observed galaxies extends to higher mass than in the galaxy scale . this is simply because there was no satellite of that mass in galaxy - scale calculations . our result indicates that an analogue to the `` missing dwarf problem '' exists also in clusters of galaxies . therefore , the `` solutions '' for the missing dwarf problem should give a distribution consistent with observations in both the galaxy scale and the cluster scale . the contents of this paper is as follows . in section 2 , we describe the models and method used in our simulations . we give detailed explanation of the subhalo detection method we developed . in section 3 , we discuss the reliability limit of the number count of subhalos . finally we summarize our result and discuss possible solutions to the missing dwarf problem in section 4 . we adopted the standard cdm model(scdm ) , so that we can directly compare our result with that of m99 who adopted scdm . cosmological parameters are @xmath8 , @xmath9 , @xmath10 , @xmath11 . to generate the initial condition , we used the standard re - simulation method@xcite . first we simulated 15 comoving @xmath12 radius using @xmath13 particles(@xmath14 ) and 150 comoving @xmath12 radius using @xmath13 particles ( @xmath15 ) to select candidates for galactic and cluster halos from the distribution of particles at @xmath16 . to select halo candidates , we used the standard friends of friends method ( fof , @xcite ) with the linking length which is 0.2 times the mean interparticle distance . then we selected particles inside the radius four times larger than radius enclosing particles detected by fof as the region of high resolution calculation , and traced these particles back to the initial condition . we replaced these particles with high resolution particles and re - ran the simulation . in this way , the external tidal force from outside the high resolution region was correctly taken into account . to generate initial conditions , we used the grafic2 package @xcite . the calculation was done on an ibm pseries690 of the data reservoir project of the university of tokyo , and on a workstation with amd opteron 242 processors and 16 gb memory . we prepared initial conditions of one region with three different resolutions for both galaxy - scale and cluster - scale runs . these runs are named c1-h , c1-m , and c1-l and g1-h , g1-m , g1-l , for cluster scales and galaxy scales . here , postfixes h , l and m denotes high , medium and low - resolution runs , respectively . we also selected two regions for additional cluster - scale runs with medium resolution . these runs are named c2-m and c3-m . the initial redshift @xmath17 and particle mass @xmath18 are listed in table [ tab : param ] . [ cols="^,^,^,^,^,^",options="header " , ] the time integration was done in physical time and physical coordinates . in high-@xmath19 regime , we varied time step @xmath20 and the softening length @xmath21 to reduce the time integration error . we adopted the procedure similar to that used in @xcite . we used the following formulation @xmath22 where @xmath23 is the redshift at which we switch from @xmath19-dependent timestep and softening to constant timestep and softening . the constant values are @xmath24 and @xmath25 . these equations imply @xmath26 and @xmath27 for @xmath28 , where @xmath29 is the mean density , @xmath30 is the scale factor . the initial redshift(@xmath17 ) and @xmath23 correspond to the standard deviation of density fluctuation @xmath31 is 0.2 and 1.0 , respectively . we chose @xmath24 so that the energy conservation is better than @xmath32 when calculations are done only for the high resolution particles . we chose @xmath25 so that it is sufficiently small compared to the size of subhalos . for runs with different @xmath0 , we varied @xmath25 in proportion to @xmath33 . these values are summarized in table [ tab : param ] . for actual time integration , we used a parallel implementation of barnes - hut tree algorithm @xcite on grape-6@xcite and grape-6a @xcite . the opening angle was @xmath34 for all run . time integration was done using the standard leapfrog . table [ tab : prop ] lists virial radii(@xmath35 ) , masses ( @xmath36 ) and numbers of the particles ( @xmath37 ) inside @xmath35 , and circular velocities(@xmath2 ) of the main halo . here , @xmath35 is the radius in which the average density is 200 times the critical density . figure [ fig : image ] shows snapshots of a galaxy and a cluster halos from high resolution runs . ( 0.45,0.45)figure1a.eps ( 0.45,0.45)figure1b.eps + ( 0.45,0.45)figure1c.eps ( 0.45,0.45)figure1d.eps to detect subhalos in simulated galaxy halos , we developed a subhalo detection program based on the hierarchical friends of friends ( hfof ) method @xcite . the hfof method is an extension of the standard fof . the fof algorithm with the linking length @xmath38 detects the regions with the density higher than @xmath39 . in the hfof method , fof is repeatedly applied to the regions found by fof , with decreasing linking length , so that we can identify all density peaks . initially , at level 1 , we use the linking length @xmath40 here @xmath41 is the critical density of the universe and defined as @xmath42 . for halos detected at level @xmath43 , we apply the fof method at level @xmath44 with linking length @xmath45 i.e. , we shrink the linking length by @xmath46 at each level . this factor must be small enough so that the result does not depend on its value . when fof of level @xmath44 is applied to a halo found in level @xmath43 , there are three possibilities . the first one is that a single halo is again found , but with a smaller number of particles . second is that nothing is found , or there are too few particles in the detected halo . the third possibility is that multiple halos are found . when one or more halos are found , we continue to level @xmath47 . when the halo vanished at level @xmath44 , we regard that a single halo is detected at level @xmath43 . the lower limit for the number of particles in a subhalo is 10 . any method for subhalo detection based on the isodensity contour has a problem that many groups found are not gravitationally bound . to select gravitationally bound groups , we employed the evaporative method @xcite . using this method , we can discard unbound groups . in addition , we can remove unbound particles from bound groups . since the iterative calculation of gravitational potential necessary for this method is computationally expensive , we used grape-6 to calculate the potential energy in order to accelerate calculation . we applied this method to all subhalo candidates , even when the number of particles is very large . thus , for all halos , we can accurately determine the bound mass . to summarize , our halo finding algorithm is following . first , the hfof method is applied to all particles to construct density hierarchy . then we apply evaporative method to each of detected density peaks ( at the highest level for which that halo is found ) . if the peak is rejected by the evaporative method , that peak is regarded as not detected and we apply the same procedure to its parent halo ( halo in one level lower ) . if the parent halo of one halo contain no other halo , we also regard that it is detected only at level @xmath43 . we apply this procedure to all hierarchies of subhalos , and regard all remained halos as detected . total numbers of detected subhalos @xmath48 are listed in table [ tab : prop ] . right panels of figure [ fig : image ] show the detected subhalos . ( 0.45,0.45)figure2a.eps ( 0.45,0.45)figure2b.eps + ( 0.45,0.45)figure2c.eps ( 0.45,0.45)figure2d.eps ( 0.45,0.45)figure3a.eps ( 0.45,0.45)figure3b.eps + ( 0.45,0.45)figure3c.eps ( 0.45,0.45)figure3d.eps top two panels of figure [ fig : num_vc ] show the cumulative number @xmath49 of subhalos as the function of the circular velocity @xmath2 for g1 and c1 halos . in both cases , @xmath1 is proportional to @xmath50 for large subhalos ( higher @xmath2 ) , but for small subhalos the number of subhalos drops off the power - law line . if we investigate the curves carefully , we can see that @xmath1 drops off from the power law in two stages . for example , the curve for g1-h starts to deviate from the power - law line at @xmath51 , and then show the second leveling off at @xmath52 . all other curves show similar two - stage behavior . to see this tendency more clearly , in bottom panels of figure [ fig : num_vc ] we plot the cumulative number normalized by @xmath50 . in g1 runs , runs h and m show narrow flat region , where the relation @xmath53 holds . the l ( low - resolution ) run shows hardly any flat region . similar tendency is visible for c1 runs , though somewhat less clear . in these panels , it is clear not only the second leveling - offs but also the first deviations are dependent on the resolution . in other words , first deviations are numerical artifact , and simulation results are reliable only for the subhalos larger than these first bendings . this result indicates that the region where the numerical result is reliable is quite narrow . for g1-h run , the numerical result is reliable only for @xmath54 , for which only @xmath55 halos exist . figure [ fig : num_mass ] shows the same cumulative number of subhalos , but now as the function of bound mass of halos @xmath56 . we can see that the result is quite similar to that of figure [ fig : num_vc ] , except that the second leveling - off is not visible . the reason why the leveling - off is visible for velocity and not for mass is simple . since we have posed minimum number of particles for subhalos ( @xmath57 ) , there is no subhalo with @xmath58 , which is the reason why the curve terminate suddenly for @xmath59 plot . in the case of @xmath49 , halos with @xmath60 but with small @xmath2 were counted , but halos with @xmath61 are not counted no matter how high its @xmath2 is . thus , @xmath1 levels off gradually . the bottom two panels of figure [ fig : num_mass ] show the cumulative number normalized by @xmath62 . again , the bending is clearly visible . the `` flat '' region is not very clear , perhaps simply due to small number statistics . for example , if we interpret the result of run c1-m naively , we might conclude that normalized plot in the mass range @xmath63 to @xmath64 is not flat . however , since there are only 20 subhalos with mass larger than @xmath64 , this result is not statistically significant . for the small - mass region where runs with different mass resolutions show different results , the statistical noise is small because the number of halos is large . at the first sight the result in this region might look reliable , simply because of this low noise . however , clear difference between runs with different resolutions indicate that result in this region is numerical artifact and is not reliable at all . from figure [ fig : num_mass ] we can estimate the minimum number of particles in the subhalos above which the number count is reliable . for run c1-h , the mass at which the normalized cumulative count bends off from the flat line is between @xmath65 and @xmath66 , which correspond to the number of particles in the subhalos between 100 and 200 . for other five runs , bending points are all similar , between 100 and 200 . so we can conclude that the cumulative count is only reliable for subhalos with number of particles larger than 200 . in order to see this tendency more clearly , in figure [ fig : num_pnum ] we show the cumulative number of subhalos plotted as functions of the number of particles in subhalos . thin solid curves are fitting function of the form @xmath67 where @xmath68 and @xmath69 are fitting parameters . for all curves , we use @xmath70 . the other parameter @xmath68 was chosen by eye . we can see that for all runs formula ( [ eq : ncfitt ] ) gives excellent fit . @xcite performed simulations similar to what we performed , and measured the cumulative number of subhalos as functions of circular velocity and mass . they have not reported this bending at @xmath71 . for velocity distribution , they did not see this bending simply because their cutoff number of particles for subhalos is much bigger than what we used . as the result , the second leveling off occurs almost simultaneously as the first bending . for mass distribution function , their figure 11 shows the same bending behavior as we see in figure [ fig : num_pnum ] , though the tendency is less clear simply because of the way the data are plotted , and they did not discuss the shape of the mass distribution function . qualitatively same tendencies were reported by @xcite in galaxy scale and @xcite in cluster scale . ( 0.9,0.9)figure4.eps for our largest calculation with 4 m particles in the virial radius of the parent halo , we found that the distribution of subhalos is reliable only up to 100 - 200 most massive subhalos . for smaller subhalos , the dependence of the cumulative number on both the circular velocity and mass becomes significantly shallower than the simple power - law . thus , we should assume that the number of subhalos formed in real cdm cosmology would show this simple power - law behavior down to mass or circular velocity much lower than the limit for which the current simulations can give reliable result . the classic work by m99 used around 1 m particles for the parent halo , and compared the distribution of around 1,000 subhalos . on the other hand , we can conclude that their result is actually reliable only for around 20 most massive halos , and the number of smaller subhalos was significantly underestimated . figure [ fig : num_vel_c123 ] shows the cumulative number of subhalos for runs c2-h , m and l , as well as that estimated for the virgo cluster in the way same as in m99 . we can see that run c2-m , which used the number of particles similar to that in m99 , shows the best agreement with the observation . the result of run c2-h shows the slope steeper than the observation starting at @xmath72 . the actual number count for run c2-h becomes higher than observation only at @xmath73 or around , because the number of massive halos in run c2-h is slightly smaller than the observational value , while c2-m shows better fit for @xmath74 . this difference between c2-h and c2-m is purely due to the fluctuation in the initial model . the high - wave number initial fluctuations we added for re - simulation are not exactly the same for these runs . as we have seen in figure [ fig : num_vc ] , the number count for this run drops off from the power - law line also around @xmath75 . thus , the actual number of galaxies observed in the virgo cluster goes even lower than the simulation which gives the number lower than the power - law . for reference , we drew the power - law curve in figure [ fig : num_vel_c123 ] . the curve implies that , ideally , for @xmath76 , the difference is already significant , and for 50km / s , there is more than a factor of 10 difference . ( 0.9,0.9)figure5.eps in figure [ fig : vel_mass ] the mass of each subhalo is plotted against its circular velocity for runs g1-h and c1-h . figure [ fig : vel_mass ] shows that the correlation between @xmath56 and @xmath2 is fairly tight , and well fitted by @xmath78 , for both runs g1 and c1 . this tight relation , however , becomes somewhat loose for the velocity smaller than 20 km / s in the case of the galaxy scale g1 and than 100 km / s in the case of the cluster scale c1 . we show the @xmath79 unbiased variances in figure [ fig : vel_mass ] , from which we can sort of see that the distribution of circular velocity for a given mass range is wider for lower mass . to see this tendency more clearly , in figure [ fig : vel_disp_ave ] we plot the unbiased variance of the circular velocity as the function of halo size . here , we used the number of particles in subhalos as the indicator of the size of the halo . we can see that the variance shows essentially the same behavior for three runs with different mass resolutions , at least for cluster - scale runs c1-x . this behavior indicates that the increase of the variance in low - mass halos is not the physical reality but the numerical artifact . the real variance is probably around 15% , independent of the mass of subhalos . the galaxy - scale runs show a similar tendency . the number of particles below which the variance becomes large is around 200 , the same as the number of particles below which the number count becomes unreliable . thus , these deviations are probably driven by the same mechanism . ( 0.45,0.45)figure6a.eps ( 0.45,0.45)figure6b.eps + ( 0.45,0.45)figure7a.eps ( 0.45,0.45)figure7b.eps + ( 0.45,0.45)figure7c.eps ( 0.45,0.45)figure7d.eps we carried out large scale @xmath0-body simulations of formation of cdm halos . we simulated the formation of both a galactic size halo and a cluster size halo . we determined the masses and circular velocities of subhalos for both cases . we found that the cumulative distribution of subhalos , both as a function of the circular velocity and as that of the mass , is affected by mass resolution of the simulation . in all runs , the number of subhalos @xmath80 with more than 200 particles is well expressed by a single power - law of @xmath81 . for the subhalos with less than 200 particles , the number of subhalos becomes smaller than this power - law line . we investigated the distribution of subhalos in the plane of their masses and circular velocities , and found that the there is a tight correlation with @xmath82 . the tightness of the correlation again depends on the number of particles in subhalos , and for subhalos with less than 200 particles the correlation becomes weaker . from runs g1-x , the number of subhalos with @xmath83 are about 30 for run g1-l , 200 for run g1-h . so roughly speaking , our result implies that the cumulative distribution of subhalos obtained by @xmath0-body simulation is reliable only for @xmath84 most massive halos , where @xmath0 is the number of particles in the virial radius of the parent halo . thus , results obtained with , for example , 1 m particles is okay for only the first 50 subhalos . this result means that the good agreement of the observation and simulation result obtained with around 1 m particles is actually a numerical artifact , and there is a serious discrepancy between the observed number of galaxies in clusters like the virgo cluster and the number of subhalos obtained by @xmath0-body simulation . ideally the number of subhalos with , say , circular velocity of 100km / s is factor two or three times bigger than the observed count , and for 50km / s this factor is more than 10 . most `` solutions '' so far proposed for the dwarf galaxy problem were aimed at reducing the number of dwarfs with @xmath85 or less . ( @xcite , @xcite , @xcite , k04 ) however , most of them do not work for halos with mass more than @xmath86 . one serious theoretical problem with the subhalos with mass @xmath86 is that when they were first formed they were generally much more massive . ( @xcite , k04 ) most of their mass have been stripped out through the tidal interaction with the parent halo . in other words , they were initially almost as massive as the halo of our galaxy . thus , a `` solution '' which can account for the discrepancy between the observed number of galaxies in clusters and the subhalos in a cdm halo need to be able to reduce the number of halos with initial virial mass as large as @xmath87 by a factor of two or so . we thank toshiyuki fukushige and hideki yahagi for helpful discussions . the simulation was partly done using ibm pseries 690 of the data reservoir project of the university of tokyo . this research is partially supported by the special coordination fund for promoting science and technology ( grape - dr project ) , and a grant for the 21st century coe program of the university of tokyo , `` quantum extreme systems and their symmetries '' ministry of education , culture , sports , science and technology , japan . , v. , white , s. d. m. , jenkins , a. , frenk , c. s. , yoshida , n. , gao , l. , navarro , j. , thacker , r. , croton , d. , helly , j. , peacock , j. a. , cole , s. , thomas , p. , couchman , h. , evrard , a. , colberg , j. , & pearce , f. 2005 , , 435 , 629	we investigated the formation and evolution of cdm subhalos in galaxy - sized and cluster - sized halos by means of @xmath0-body simulations . our aim is to make clear what the `` dwarf galaxy problem '' is . it has been argued that the number of subhalos in simulated galaxy - sized halos is too large compared to the observed number of dwarfs in the local group , while that in cluster - sized halos is consistent with observed number of galaxies in clusters such as the virgo cluster . we simulated nine halos with several different mass resolutions and physical scales . we found that the dependence of the cumulative number of subhalos @xmath1 on their maximum circular velocity @xmath2 is given by @xmath3 , down to the reliability limit , independent of the mass of the main halo . this implies that simulations for cluster - sized halos give too many halos with @xmath4 or less . previous comparisons of cluster - sized halos gave much smaller number of subhalos in this regime simply because of their limited resolution . our result implies that any theory which attempts to resolve the missing dwarf problem should also explain the discrepancy of the simulation and observation in cluster - sized halos .
the connection between the mass of a central supermassive black hole ( hereafter mbh ) and the properties of the host galaxy ( see , e.g. , the @xmath5 relation ; ferrarese & merritt 2000 , gebhardt et al . 2000 ) , together with the recognition that a mbh resides at the center of every massive spheroid in the local universe ( e.g. , ferrarese & ford 2005 ) , has led to the idea that the formation and evolution of the mbh and the host galaxy are intimately related ( e.g. , springel , di matteo & hernquist 2005 , sazonov et al . 2005 , hopkins et al . 2006a ) . in this picture , feedback from the mbh has been advocated to regulate starformation at early epochs , producing the observed galaxy mass function and colors ( e.g. , croton et al . 2006 ) ; in later epochs , feedback is required to heat the galactic flows solving the `` cooling flow '' problem ( e.g. , omma et al . 2004 ; ostriker & ciotti 2005 , 2007 ; churazov et al . 2005 , mcnamara et al . 2005 ) . although feedback is a promising mechanism for solving many observational puzzles , our understanding of how accretion and feedback work , and of their evolution over cosmic time , is still uncomplete . it has been suggested that after the bright agn phase accretion switches to a more radiatively quiet or inefficient mode ( e.g. , churazov et al . 2005 , hopkins et al . 2006b ) possibly associated with radio jets or outflows that may carry out most of the accretion power ( di matteo et al . 2003 , pellegrini et al . 2003 , falcke et al . 2004 , allen et al . indeed , the nuclei of most of the nearby spheroids are either radiatively quiescent or exhibit low levels of activity ( e.g. , ho 2005 , pellegrini 2005a ) . typically , these nuclei are not associated with detectable radio sources , but the absence of radio emission in itself does not exclude nuclear activity , since frustrated or not collimated nuclear outflows would be hard to detect ( nagar et al . also , intermittent agn activity is suggested by the x - ray and radio morphology of nearby clusters of galaxies ( e.g. , fabian et al . 2006 ) and galaxies ( e.g. , machacek et al . 2006 ) , and may originate in an accretion flow that switches between different accretion modes ( janiuk et al . 2004 ) . in nearby spheroids we can attempt to constrain observationally the modalities of accretion ( and feedback ) through detailed studies of the stellar population , the ism and the current star formation rate for the central @xmath6one hundred parsec region , especially if the agn emission does not dominate over the other nuclear components . ngc 821 , an isolated elliptical galaxy at a distance of 24.1 mpc ( table [ mainlog ] ) , is an ideal target for studying how these phenomena have evolved in the local universe . there is compelling evidence of the presence of a central mbh in this galaxy , from resolved dynamical studies ( gebhardt et al . ngc821 has a very regular and smooth optical appearance ( lauer et al . 2005 ) and the old and metal rich stellar population typical of elliptical galaxies ( proctor et al . 2005 ) ; no cold ( hi ) or dusty ism ( sarzi et al . 2006 , ravindranath et al . 2001 ) have been observed in it , restricting the potential reservoir for accretion . the mbh of ngc821 has a mass of @xmath7 ( table [ mainlog ] ) and an eddington luminosity of @xmath8 . therefore , if radiatively efficient accretion were taking place , this galaxy should be a luminous agn . instead , this mbh is extremely quiescent . very low limits on optical emission lines ( h@xmath9 , h@xmath10 , or [ oiii ] ) are reported for the nucleus of ngc821 ( ho 2002 ; ho et al . 2003 ; sarzi et al . 2006 ) and the limits on nuclear radio emission are also stringent : @xmath111.5 mjy at 15 ghz ( with 150 mas resolution ; nagar et al . 2005 ) and @xmath110.5 mjy at 5 ghz ( with 5@xmath12 resolution ; wrobel & heeschen 1991 ) . thanks to a deep @xmath0 pointing for a total exposure of 230 ksec ( discussed in pellegrini et al . 2007 , hereafter p07 ) , a source was detected for the first time at the nucleus of ngc821 ( called s2 ) . s2 is extended ( tab . [ tabspec ] ) , but it has a flat ( @xmath13 ) unabsorbed x - ray spectrum , excluding a central concentration of hot , optically thin ism . its 210 kev luminosity is @xmath14 erg s@xmath15 . an upper limit of 2.8@xmath16 erg s@xmath15 was placed on a possible pointlike 0.38 kev emission associated with the mbh ; this is one of the smallest values obtained with @xmath0 for galactic nuclei ( pellegrini 2005b ) . in addition to the nucleus , also the hot ism that could provide a source of fuel for the mbh can be measured in the x - rays ( e.g. , loewenstein et al . 2001 ; soria et al . our deep , sub - arcsecond @xmath0 pointing revealed diffuse emission in the central part of ngc 821 , but also led to the detection of a population of x - ray binaries which can account for most if not all of the diffuse emission , so that we could place a very stringent upper limit on the contribution of a hot gaseous component ( p07 ) . this deep pointing highlighted the presence of three other sources around s2 , of luminosity and spectral shape consistent with those of low mass x - ray binaries ; only one is consistent with being pointlike , though , the other two ( s1 and s4 ) are extended ( tab . [ s1-s4 ] ) , and could be due to the superposition of few point sources and/or truly diffuse emission . in particular , the morphology of s1 resembles a jet - like feature ( see also fabbiano et al . 2004 ) . in this paper we report the results of an observational campaign aimed at complementing the @xmath0 results , to better constrain the nature of the processes taking place in the vicinity of the mbh in ngc821 . new data presented in this paper include sensitive vla and @xmath3 irac observations ; we also analyzed archival @xmath17 @xmath18 @xmath19 ( @xmath4 ) observations . with this multi - wavelength data set we investigate : circumnuclear starformation , which may result from accretion ( tan & blackman 2005 ) ; radiatively inefficient accretion , possibly coupled to a compact or resolved nuclear jet ( e.g. , narayan 2005 ) ; and highly obscured scaled - down agn emission . the paper is organized as follows : in section [ endistr ] we summarize the main @xmath0 results concerning the nuclear sources , we present the results of the new vla and @xmath3 observations , of the archival @xmath4 optical data , and previous nicmos results ; in section [ quie ] we discuss possible scenarios for accretion around this mbh ; in section [ concl ] we summarize our conclusions . the main outcomes of the deep @xmath0 pointing concerning the properties of the nuclear source s2 are summarized in tab . [ tabspec ] . we also show the source spectrum ( fig . [ s2spec ] ) and the results of a spectral fit including fluorescent line emission from cold iron at 6.4 kev ( tab . [ tabspec ] ) , which produced only an upper limit on the equivalent width of the line . tab . [ s1-s4 ] summarizes the properties of the other two extended sources ( s1 and s4 ) detected by @xmath0 in the central galactic region . the position and extent of the sources s1 , s2 and s4 are also shown in fig . [ radioim ] . below we present new vla observations in sect . [ vla ] ; the analysis of ir data from a @xmath3 program to observe a sample of radiatively quiescent mbhs ( pi : fabbiano ) in sect . [ spiz ] , together with a previous nicmos measurement ; the reanalysis of archival @xmath4 wfpc2 data taken with the f555w and f814w filters , and of our previous r - band int images in sect . table [ tablesed ] summarizes the results and fig . [ sed ] shows them . we observed ngc821 with the vla of the nrao at 1.4 ghz in the most extended ( a ) configuration on 2004 december 31 , and at 4.8 and 8.4 ghz in the second most compact ( c ) configuration on 2005 july 21 . we used a 200 mhz instantaneous continuum bandwidth , observing 3c48 for flux density calibration and j0203 + 1134 for phase calibration , using standard techniques and the aips data reduction package . ngc821 lies @xmath20 from the phase calibrator . empirical limits on systematic position uncertainties were estimated from snapshots images of three quasars ( j0204 + 1514 , j0149 + 0556 , and j0242 + 1101 ) that lie within @xmath21 of the phase calibrator . the flux density scale was accurate to @xmath22 and absolute position measurements were accurate to @xmath23 . we mapped the entire primary beam of the vla with 3-d deconvolution to obtain noise levels ( @xmath24 ) of 20 @xmath2jy , 20 @xmath2jy , and 17 @xmath2jy at 1.4 , 4.8 , and 8.4 ghz respectively , for effective total integrations of 2.02.2 hours at each frequency . we adopted variance weighting of ( _ u , v_)-data with some downweighting of short baselines ( aips parameter robust=0 ) . beam half - power sizes are given in tab . [ tablesed ] . we detected a radio counterpart to source s2 at @xmath25 , @xmath26 at 1.4 ghz ( fig . [ radioim ] , tab . [ tabspec ] ; sect . 4.2 of p07 discusses the coincidence of the position of s2 with the optical center in @xmath4 wfpc2 images ) , with a flux density of 127 @xmath2jy ( 6@xmath27 ) . emission was marginally detected at 8.4 ghz ( 71 @xmath2jy ; 4.2@xmath27 ) , peaking at the same position to within the measurement uncertainty . the noise level at 8.4 ghz was about one and a half order of magnitude lower than the earlier upper limit obtained at 15 ghz by nagar et al . a marginal detection at 4.8 ghz was also possible ( 80@xmath2jy ; 4.0@xmath27 ) ; in this case the prospective peak lies @xmath28 ( or @xmath29 beamwidth ) to the south . the three flux densities correspond to a power law with spectral index @xmath30 ( @xmath31 ) . the estimated spectral index may be viewed with some caution because it depends on measurements made with very different beamwidths . however , we have studied images at 1.4 ghz made with different weighting schemes ( i.e. , natural and uniform ) and we did not find a significant change in peak signal to noise ratio at the location of s2 , for beam sizes of @xmath32 to @xmath33 . at this and other frequencies , there is no indication that we might be partially resolving the emission for s2 ; however , multi - configuration data would be needed to properly address this question . no sources were detected at the location of the other @xmath0 sources s1 , s3 and s4 in the central galactic region ( sect . 4 in p07 ; fig . [ radioim ] ) . therefore , we can place @xmath34 upper limits on their emissions of 80@xmath2jy , 80@xmath2jy and 68@xmath2jy respectively at 1.4 , 4.8 and 8.4 ghz . ngc821 was observed with the infrared array camera ( irac , fazio et al . 2004 ) on board the @xmath3 space telescope ( werner et al . 2004 ) on august 21 , 2005 for a total of 725.5 s ( program i d 20371 ) . after the standard @xmath3 irac processingv1.pdf ] , the four irac channels resulted in images at 3.6 , 4.5 , 5.8 and 8 @xmath35 ; the central region of ngc821 was detected in each of them . we based our analysis on the irac instrument performance as given in the infrared array camera data handbook . using ds9 and @xmath36 , we extracted counts from circular regions of approximately 4 square arcseconds area ( @xmath37 radius ) centered on the peak of the emission . in every case the position of this peak , with the @xmath3 wcs , falls in - between the optical center and the @xmath0 position for s2 ( see tab . [ tabspec ] ) ; both of these are well within the count extraction radius . this position ( @xmath3 wcs ) is at ra=@xmath38 @xmath39@xmath40 - 0.1truecm.11 , dec=@xmath41 @xmath42 @xmath43 , although given the pixellation of the data , the peak count pixel may be slightly displaced . the background was estimated both from a surrounding annulus , to attempt a rough subtraction of stellar light from the center - most emission , and from an off - source circle . the fluxes were all normalized to an area of 4 square arcseconds ( that of circles of @xmath44 radius ) . the results are shown in table [ spizflu ] , for both choices of the background . we also list in tab . [ tablesed ] the upper limit on any unresolved nuclear emission in the h - band ( centered at 1.6@xmath45 ) derived from nicmos data ( ravindranath et al . this limit refers to the nic2 camera and corresponds to an intrinsic gaussian of fwhm @xmath46 pixel . we carried out aperture photometry on the archival @xmath4 wfpc2 images taken with the f555w and f814w filters ( lauer et al . 2005 ) , following standard procedures described in the wfpc2 handbookhand/wfpc2@[email protected] . ] . the instrumental f555w and f814w magnitudes were converted to standard v and i magnitudes using the most updated versioncalib/. ] of the color coefficients provided by a. dolphin . in order to compare the optical emission with that estimated from @xmath3 irac observations , it was calculated for a circle of @xmath44 radius , the same area used to extract the @xmath3 fluxes ( sect . [ spiz ] ) . the optical brightness inside this area is @xmath48 mag , and @xmath49 , after correcting for a galactic extinction @xmath50 ( schlegel et al . the r - band magnitude from the same area was also calculated , by using int images ( graham et al . 2001 ) and the corresponding surface brightness profile derived by soria et al . ( 2006b ) ; it results into @xmath51 in the cousins system . in order to constrain the emission coming from the mbh , from the deconvolved f555w and v i profiles derived by lauer et al . ( 2005 ) we also calculated the v and i luminosities within a circle of @xmath52 radius ( the innermost radius at which deconvolution can give accurate results according to lauer et al . ) . these luminosities are upper limits to the mbh emission , since both our analysis and that of lauer et al . ( 2005 ) reveal the absence of an optical `` nucleus '' , that is a compact light source rising above the surface brightness profile at small radii when extrapolated inward . in general these nuclei , which have been found in a large fraction of early - type galaxies from @xmath4 images ( see also ravindranath et al . 2001 ) , are bluer than the background starlight and could be nuclear star clusters or low luminosity agns . the deep @xmath0 image of ngc821 revealed an extended , hard source at the position of the galactic center ( p07 ) ; our associated vla observations have led to the discovery of a 1.4 ghz source at the same position ( sect . [ vla ] ) ; pointlike nuclear emission is not detected in the optical and the ir ( sects . [ spiz ] and [ opt ] ) . in the 0.38 kev band , the @xmath53 upper limit to any pointlike emission associated with the mbh is @xmath54 , and the @xmath55 at 1.4 ghz is just @xmath56 erg s@xmath15 . therefore the mbh of ngc821 is one of the radiatively quietest mbh known , with @xmath57 . most mbhs are radiatively quiescent in the local universe , or show very low activity levels ( sect . [ intro ] ) , therefore our deep look at the nucleus of ngc821 has a bearing on a very common state of mbhs . in the following we use our observational results to constrain the process of accretion in this nucleus . given the general lack of fuel available for accretion observed at all wavelengths , including the x - rays ( sect . [ intro ] ) , is a truly dead mbh what we should expect at the center of ngc821 ? in fact , an aging stellar population continuously returns gas to the ism , via its stellar mass losses ( ciotti et al . 1991 , david et al . 1991 ) , and the circumnuclear region should be replenished with this fuel . indeed , hydrodynamical simulations specific for ngc821 ( p07 ) demonstrate that this fuel should be present . while the bulk of the hot gas is expelled out in a wind ( consistent with the lack of detection of hot gas ) , the stellar mass losses are accreting within a very small region of @xmath58 pc from the nucleus ; at the innermost radius resolved by the simulations ( 10 pc ) , the mass accretion rate is @xmath59few@xmath60 yr@xmath15 . notwithstanding the limitations of the simulations , and the uncertainties associated with the observables used as input parameters , the presence of this small inflowing region was considered a robust result ( see p07 ) ; similarly , a range of values for @xmath61 yr@xmath15 was considered reliable ( with the caveat that the true , final accretion rate on the mbh could be somewhat higher ; p07 ) . @xmath62 should produce a luminosity @xmath63(14)@xmath64 erg s@xmath15 if , closer in to the mbh , it ends up in a standard accretion disc with a radiative efficiency @xmath65 as assumed for classical agns ( shakura & sunyaev 1973 ) . this would be a highly sub - eddington agn , though , since @xmath66few@xmath67 . for a standard spectral energy distribution ( sed ) of agns ( elvis et al . 1994 ) , the upper limit on a pointlike nuclear x - ray emission ( tab . [ tabspec ] ) corresponds to a bolometric upper limit of @xmath68 erg s@xmath15 , that is already 30 to 100 times less than @xmath69 . below we explore scenarios for the evolution of the accreting material , that would explain why the nucleus is not observed at a luminosity as high as @xmath69 . we consider , in turn , the presence of angular momentum in the accreting material ( sect . [ ang ] ) , a reduction of the accretion rate due to starformation in a circumnuclear disc ( sect . [ bc03 ] ) , accretion at low radiative efficiency ( sect . [ effic ] ) , the presence of a jet / outflow , whose kinetic power possibly accounts for a fraction of @xmath70 ( sect . [ aneta ] ) , a transition in the accretion mode due to disk instabilities ( sect . [ instab ] ) and finally obscuration of the nuclear radiation produced by a scaled - down agn ( sect . [ osc ] ) . note that the mass accretion rate of galactic nuclei has been often estimated ( e.g. , di matteo et al . 2003 ) assuming spherically symmetric accretion from a nonrotating polytropic gas with given density and temperature at infinity ( bondi 1952 ) . moscibrodzka ( 2006 ) calculated model spectra emitted by such a bondi flow for a sample of low luminosity agns , including ngc821 . the adopted value of gas density was larger than constrained now thanks to the deep @xmath0 pointing ( p07 ) ; even so , the predicted x - ray emission was @xmath71 ergs s@xmath15 , that is @xmath72 orders of magnitude below the limit on pointlike nuclear emission ( tab . [ tabspec ] ) . however , the bondi accretion is a severe mathematical approximation of the accretion process ( see also p07 ) ; for example , some initial angular momentum of the flow captured by the mbh , and the inclusion of viscosity , may result in a higher x - ray luminosity . radiation models applied to global mhd simulations of low angular momentum accretion flows are being developed and will be applied to this nucleus ( moscibrodzka et al . , in preparation ; see also balbus & hawley 2002 ) . the inclusion of even slow rotational motion of the gas at large radii can significantly reduce the rate at which mass is captured into the accretion flow , compared to the non - rotating case ( proga & begelman 2003 ) . the motivation for the proga & begelman s study was to make less severe the problem that mbhs in the local universe are much dimmer than expected for a mass accretion rate estimated at large radii via the bondi ( 1952 ) formula ( sect . [ dead ] ) and ending up in a standard disc ( e.g. , fabian & canizares 1988 , pellegrini 2005a ) . while the bondi formula assumes spherically symmetric accretion from a nonrotating gas under the sole influence of the central gravity , proga & begelman ( 2003 ) studied the evolution of the flow with a small angular momentum , including regions beyond the domination of the mbh gravity . they found that the mass supply rate to the mbh can be smaller by up to @xmath73 orders of magnitude with respect to the nonrotating case . in ngc821 the stellar kinematics becomes disc - like within the central @xmath74 ( scorza & bender 1995 , emsellem et al . 2004 ) , therefore the stellar mass losses close to the galactic center are likely to have angular momentum , which would lead to a reduction of @xmath62 . whether the expected luminosity could become consistent with the observed upper limits is however more questionable , since it is not known what is the final , net effect of the inclusion of angular momentum at large radii on the resulting luminosity of the flow closer to the mbh . it is probably not plausible that gas inflowing at the rate shown by the simulations ( that is equal to @xmath62 of sect . [ dead ] at present and was larger in the past , for a total integrated mass of few@xmath75 over the past @xmath76 gyrs ; p07 ) continues to accumulate for many gyrs in the nuclear region without accreting and without becoming observable . however , in order for the gas to be able to accrete , angular momentum must be transferred outward through viscosity , and this may actually increase the luminosity of the flow . mhd simulations with rotation and cooling by radiation included are needed to really state what happens to the luminosity of the flow ( see also sect . [ dead ] ) . condensation into stars in a gravitationally unstable disc can prevent a significant fraction of the accreting gas from reaching the mbh ( tan & blackman 2005 ) , resulting into a much lower true accretion rate ; this starformation should produce h@xmath9 and ir emission . as reported in sect . [ intro ] , the optical spectrum of ngc821 shows only absorption lines ; the @xmath53 upper limit on its h@xmath9 luminosity is @xmath77 erg s@xmath15 ( ho et al . 2003 , rescaled for the distance in tab . [ mainlog ] ) . from the kennicutt ( 1998 ) relation between starformation rate and @xmath78 , the resulting starformation rate is @xmath79 yr@xmath15 , much larger than @xmath62 given by the simulations ( sect . [ dead ] ) . also the observed luminosity in the irac bands ( tab . [ tablesed ] , fig . [ sed ] ) is at least one order of magnitude higher than expected in the full ir band from starformation at the @xmath62 rate ( kennicutt 1998 ) . however , the aperture used to derive the irac luminosities includes also emission from the normal , old stellar population , that has to be accounted for . the analysis of the sed of ngc821 ( fig . [ sed ] ) shows that it can be explained just by the normal , old galactic stellar population , without requiring ongoing starformation . to check this , we used the matched aperture photometry ( i.e. , all for @xmath44 aperture radius ) from the @xmath3 to the v filter data points ( tab . [ tablesed ] ) and compared it with sed templates for stellar populations of different ages and metallicities , for the kroupa ( 2001 ) imf ( bruzual & charlot 2003 ) ; we also allowed for an exponentially declining starformation rate with time scale @xmath80 . the fitting procedure described by bolzonella et al . ( 2003 ) , available on the web at http://webast.ast.obs-mip.fr/hyperz and built by bolzonella et al . ( 2003 ) . ] was applied . the observed data points are very well fitted by an age of 11.0 gyrs and a metallicity of 0.25 solar , with a short time scale ( @xmath81 gyr ) and without current star formation ( solid line in fig . [ sed ] ) . these properties match closely those obtained from spatially resolved optical spectroscopy for the bulk of the stellar population ( proctor et al . the magnitudes and fluxes derived here have been corrected for extinction ( sect . [ endistr ] ) ; to check whether a residual extinction is left in the data , we allowed for free extinction in the fit , but the best fit extinction turned out to be consistent with zero . we can conclude that the optical and mid - ir emission within a central projected circle of @xmath44 radius is dominated by the normal stellar population of the galaxy . from their analysis based on lick absorption - line indices , proctor et al . ( 2005 ) also found at the very center of ngc821 ( within a central radius of @xmath82 ) an emission weighted abundance of @xmath83 times solar , and evidence for a burst of starformation between 1 and 4 gyrs ago , that involved @xmath84% of the galaxy s mass . therefore we fitted the observed data points also with a metallicity of 2.5 solar ( the largest value available for the templates ) ; this resulted in an age of @xmath85 gyr , but the fit was not as good as the previous one ( dotted line in fig . [ sed ] ) . the optical points considered here contain less information than the indices used by proctor et al . ( 2005 ) ; it is also likely that the true best fit sed is a combination of templates with different ages and abundances , but a fit with composite seds is beyond the scope of this work . for the purposes of the present investigation , the conclusion here is that current starformation is not required to explain the sed ; however , @xmath62 is so low that the available optical and ir data can not constrain whether a part ( or even the total ) of it goes into star formation . it has been suggested that high - luminosity and low - luminosity agns host two different accretion modes : a radiatively efficient disk - dominated one and a radiatively inefficient one , which is expected to be coupled to strong outflows ( see narayan 2005 for a recent review ) . the sed of a few low - luminosity agns indeed has been modelled with truncated optically thick disks with inner radiatively inefficient flows ( e.g. , m81 and ngc4579 , quataert et al . 1999 ; ngc3998 , ptak et al . 2004 ; ngc1097 , nemmen et al . 2006 ) . in the radiatively inefficient mode , the flow produces the x - ray emission , while the radio emission comes from the base of a nuclear jet , if present ( e.g. , falcke et al . 2004 , krding et al . in fact , observations seem to suggest that low - luminosity agns are `` radio - loud '' ( ho 2002 , terashima & wilson 2003 , wu & cao 2005 ) ; moreover , often their radio power and sometimes also their radio spectrum require a separate , compact jet component ( anderson et al . 2004 ; nagar et al . 2005 ) . on the theoretical side , mhd simulations show that at low accretion rates winds and jets naturally develop ( e.g. , stone et al . 1999 , hawley & balbus 2002 , igumenshchev et al . 2003 , de villiers et al . 2003 ) , since the magnetic and kinetic energy release in the flow are enough to support an outflow or a jet ( see also blandford & begelman 2004 ) . in the context of low radiative efficiency accretion , the relationship between the black hole mass , the core radio luminosity at 5 ghz and the 210 kev emission has been investigated for a large sample of accretion - powered sources , going from x - ray binaries to mostly low - luminosity agns by merloni et al . they found the sources to lie on a plane described by log @xmath86 log @xmath87 0.78 log @xmath887.33 ( with a scatter of @xmath89 ) , which was attributed to the presence of a radiatively inefficient accretion flow @xmath90 jet system , that can exist with similar properties at different scales . low radiative efficiency accretion is compatible with the very low value of @xmath57 for the mbh of ngc821 ( sect . [ quie ] ) , and was already considered as a viable solution for its low emission by pellegrini ( 2005a ) and soria et al . ( 2006a , b ) . given our new radio observation , we examine here the position of this mbh with respect to the plane of merloni et al . the 210 kev upper limit on nuclear pointlike emission is @xmath91 erg s@xmath15 ( tab . [ tabspec ] ) , and the nuclear 5 ghz luminosity is @xmath92 erg s@xmath15 ( a @xmath93 detection , though ; sect . [ vla ] ) ; the resulting location is shown in fig . if the 5 ghz detection is real , and the nuclear x - ray emission is smaller by a factor of @xmath94 than the upper limit , then ngc821 is consistent with the plane , and it may host a radiatively inefficient accretion flow @xmath90 jet system . if instead the x - ray emission is close to that of the whole s2 source , or even higher ( see sect . [ osc ] ) , then this nucleus lies well below the predictions of the plane . in this case , its sed could be consistent with that of a classical , bright agn ( see sect . [ osc ] below ) . the sample of merloni et al . includes mostly objects of relatively high @xmath95 . below a critical @xmath96 , the jet emission can become dominant also in the x - ray domain , and the radio x - ray correlation should turn steeper ( yuan & cui 2005 ) . the slope of the fundamental plane should then be different from when the x - ray emission is dominated by the accretion flow ( see also wang et al . 2006 ) , and the fundamental plane in this regime becomes log @xmath97 log @xmath87 0.25 [email protected] ( with @xmath99 at 8.5 ghz , yuan & cui 2005 ) . this relation predicts a lower radio luminosity for given x - ray emission and therefore the nucleus of ngc821 is located close to such a plane ( fig . [ fp ] ) provided that the radio emission is close to the @xmath100 detection at 8.4 ghz and the 210 kev emission close to the @xmath53 upper limit . in conclusion , a radiatively inefficient flow coupled to a nuclear jet is a viable scenario for the accretion process . also the observed spectral shape of the nuclear source in the radio is consistent with this interpretation . the flux density of a jet with optically thick core / nozzle emission is approximately flat ( @xmath101 , with @xmath102 ) up to high radio frequencies ( zensus 1997 , falcke et al . 2004 ) . within the uncertainties due to the marginal detections at 4.8 and 8.4 ghz and the unmatched beamwidths at the different frequencies ( sect . [ vla ] ) , the radio spectral shape of the nuclear source is relatively flat , and is therefore consistent with being produced by the base of a jet . this jet could be confined within the inner few arcsecs ( see tab . [ sed ] for the beam sizes ) , because it is disrupted or frustrated within it ; nuclei with parsec - scale jets have already been found in elliptical galaxies with the vlbi ( e.g. , nagar et al . 2005 ) . in the next section we examine whether a jet / outflow can be present on a larger scale ( outside s2 ) . the elongated x - ray sources s1 and s4 in the central galactic region ( p07 , fig . [ radioim ] ) may be due to an outflow or a jet ; in the radio , only s2 is detected , but at the location of sources s1 and s4 we could place @xmath34 upper limits ( sect . [ vla ] ) . we use here the observed x - ray to radio flux ratios for typical jets to investigate whether s1 and s4 can be due to a jet ( note however that the properties of jet emission depend on the internal jet structure as well as the environment in which the jet propagates ) . kataoka & stawarz ( 2005 ) compiled a sample of x - ray jets with @xmath0 data , observed outside the host galaxy of quasars and agn . the x - ray luminosity usually dominates over the radio for the jet / knots emission , while for lobes and hot spots the radio luminosity is comparable to or exceeds the x - rays . in any case , in the kataoka & stawarz ( 2005 ) sample the x - ray to radio l@xmath103/l@xmath104 monochromatic luminosities ratio does not exceed 100 . this same ratio is @xmath105 for the source s1 , and @xmath106 for s4 , assuming that the entire x - ray luminosity of each source is due to a jet , and the @xmath34 upper limits at 5 ghz ( sect . . however , if the observed x - ray luminosity is contributed by several blended knots then the limits quoted above are not meaningful . the case for unresolved knots in sources s1 and s4 is better investigated with more nearby objects . one example of a nearby jet moving within the interstellar medium of a galaxy is that of centaurus a , with many resolved x - ray knots ( hardcastle et al . 2003 , kataoka et al . cen a is at 3.4 mpc distance , so that @xmath82 is equivalent to 17 pc . the size of source s1 ( tab . [ s1-s4 ] ) corresponds to @xmath107 pc , and it is feasible for a jet to create many knots on that distance , given the large number of knots observed in the cen a jet . also , for a few knots in the cen a jet the x - ray to radio luminosity ratio exceeds 100 , with one case of @xmath108 ( hardcastle et al . the lowest detected 5 ghz luminosity of a radio knot in the cen a jet is 1.4@xmath109 erg s@xmath15 , which is below our @xmath34 upper limit of 2.7@xmath109 erg s@xmath15 at 5 ghz . the available radio and x - ray data , although very deep , do not provide therefore a definite conclusion on the jet presence outside the s2 region . note that our current data do not resolve the nuclear s2 source , whose x - ray and radio luminosities are measured for a region of @xmath110 pc radius , and could also be related to a small jet . however , the radio spectrum of s2 ( @xmath111 , sect . [ vla ] ) does not have the typical slope observed for extragalactic jets or hot spots ( @xmath112 , kataoka & stawarz 2005 ) , but is instead more similar to that of radio - loud agns ( elvis et al . 1994 ) , and therefore more likely produced by the base of a jet . if present , the launch of a nuclear jet / outflow could account for a non - negligible fraction ( of the order of a few percent ) of the energy associated with the rest mass of the material being accreted ( i.e. , of @xmath70 ) , as shown to be the case for m87 ( di matteo et al . 2003 ) , ic1459 ( fabbiano et al . 2003 ) and ic4296 ( pellegrini et al . 2003 ) and recently in a few other hot gas rich sources ( allen et al . this may be a characteristic of the late stages of galaxy evolution , when accretion onto the mbh falls below the eddington rate ( e.g. , churazov et al . unfortunately , radio synchrotron emission traces poorly the true jet power ( owen et al . 2000 , brzan et al . 2004 ) , with the ratio of the mechanical ( kinetic ) luminosity to the 1.4 ghz synchrotron luminosity ranging between a few and roughly a few thousand for luminous radio sources , and up to several thousand for weaker sources . notwithstanding these uncertainties , we can estimate that the jet kinetic power could account for just a few@xmath113 of @xmath70 , for the nuclear radio source coincident with s2 and detected at 1.4 ghz , and at most this fraction for the s1 and s4 sources with upper limits at 1.4 ghz . an injection of energy from the nucleus may have heated the surroundings , so that accretion has been recently stopped or lowered to values below the @xmath62 estimated in the numerical simulations of p07 that do not include feedback ( di matteo et al . 2003 ; ciotti & ostriker 2007 ) . the mode of accretion may switch to a quiescent state due to the change in internal structure of the accretion flow produced by disk instabilities ( siemiginowska et al . 1996 , janiuk et al . 2004 ) , with the timescale in the low quiescent state longer than that in the active state . then the accretion flow can be described by a hybrid model , where an outer torus or cold disk accumulates accreting matter at the steady accretion rate , while the inner torus / disk has a low accretion rate and is radiatively inefficient . the transition radius between the two regions is set by the hydrogen ionization instability and depends on the accretion conditions onto the outer torus . for the mbh mass of ngc821 ( tab . [ mainlog ] ) the gravitational radius is located at @xmath114 cm . assuming an accretion rate of @xmath115 yr@xmath15 ( sect . [ dead ] ) , the location of the ionization zone is at @xmath116 ( janiuk et al . 2004 ) . in the hybrid model we expect the outer disk to be cold and to emit a thermal spectrum with a total disk luminosity for these parameters of about @xmath117 ergs sec@xmath15 , that will be radiated mainly in the optical and near ir bands . this value is well within the observed limits ( tab . [ tablesed ] and fig . [ sed ] ) . the x - ray luminosity will come from the inner quasi - spherical accretion flow and depend on the density and temperature of the plasma ; its value needs to be self - consistently calculated , which is beyond the scope of this work and will be addressed by moscibrodzka et al . ( in preparation ) . we examine here the possibility that the nucleus of ngc821 is a normal agn working at a very sub - eddington rate of @xmath118few@xmath67 , with @xmath119 erg s@xmath15 estimated for a standard disc as described in sect . [ dead ] , and test this hypothesis using the whole observed sed ( fig . [ sed ] ) . @xmath69 is 30100 times larger than the upper limit on @xmath120 of pointlike nuclear emission ( for a bolometric correction of @xmath121 , elvis et al . we consider then the possibility that the x - ray radiation is heavily absorbed , so that the intrinsic @xmath120 is comparable to @xmath69 . a compton thin agn with @xmath122 @xmath123 ( e.g. , bassani et al . 1999 ) can be excluded , since the x - ray spectrum of s2 does not show the characteristic cut - off at low energies ( tab . [ tabspec ] ) . a more interesting possibility is that s2 is a compton thick agn , where the direct nuclear emission is not detectable below @xmath76 kev because of an absorber of very large column density ( @xmath124 @xmath123 ; e.g. , guainazzi et al . 2005 ) . however , some few percent of the intrinsic emission is often scattered and reprocessed off some compton thick mirror visible both to us and from the central continuum source , and may emerge below 10 kev with a flatter photon index than the intrinsic x - ray continuum ( i.e. , with @xmath125 ) , as for example in the circinus galaxy ( matt et al . the resulting x - ray spectrum also shows a very strong ( equivalent width @xmath126 kev ) iron @xmath127 fluorescent line ( as in ngc1068 , levenson et al . s2 has quite a flat spectral shape ( @xmath13 ) , reminiscent of what is found for compton - thick sources ; a 6.4 kev iron line is not seen in its x - ray spectrum , but only with a @xmath24 upper limit on its equivalent width of 2 kev ( tab . [ tabspec ] ) . note also that extended emission has been observed associated with some obscured seyfert nuclei , up to sizes of @xmath85 kpc ( elvis et al . 1990 , ogle et al . 2000 ) , and is understood as emission from gas photoionized by the central agn , or thermal emission from a hot collisionally ionized plasma that forms the intercloud medium . therefore , this x - ray faint mbh may be an extreme scaled - down version of such obscured agns , and the extended s2 source could be nuclear flux scattered and fluorescing off cold gas . evidence for seyfert nuclei ( including compton - thick ones ) being scaled - down versions of more luminous agns down to 210 kev luminosities of @xmath128 erg s@xmath15 has been provided recently ( panessa et al . 2006 , 2007 ) , based on the significance of the correlation between @xmath129 and l(h@xmath9 ) down to such low x - ray luminosity values . this implies a similar proportion of x - ray and uv ionizing radiation in high and low luminosity nuclei , so that the shape of their seds may be similar in these spectral regions . if we assume that a typical 1% of the intrinsic agn flux is scattered into our line of sight and observed as source s2 with a l(210 kev)@xmath130 erg s@xmath15 ( tab . [ tabspec ] ) , the intrinsic @xmath131 erg s@xmath15 for a standard agn sed ( elvis et al . 1994 ) , a luminosity that is consistent with @xmath69 . @xmath120 could be even higher , since some nearby agn show @xmath132% of their x - ray flux scattered , perhaps due to the lack of an appropriately positioned mirror ( ngc4051 , uttley et al . an intrinsic 210 kev luminosity as large as assumed here would place the ngc821 nucleus way off the fundamental plane of sect . [ effic ] ( increasing its x - axis value of 1.4 in fig . [ fp ] ) , consistent with the hypothesis of a standard disc rather than a radiatively inefficient accretion flow . how does this hypothesis of the nucleus of ngc821 being a scaled - down , highly obscured agn compare with the available observational constraints ? the intrinsic sed would have the typical agn shape of elvis et al . ( 1994 ) shown in fig . [ sed ] , rescaled up by a factor of @xmath133 for a 1% reflectance ( since the average agn sed in fig . [ sed ] is normalized to the upper limit on pointlike x - ray emission in tab . [ tabspec ] , that is @xmath73 times lower than the whole x - ray luminosity of s2 ) . the upper limits in the v and i bands allow for this shift , but the nicmos f160w band limit lies just a factor @xmath76 above both the radio - quiet and radio - loud seds before rescaling . even though the dispersion in the seds of agn is rather large ( even an order of magnitude in the x - ray to optical - uv ratio , risaliti & elvis 2004 ) , scaling up the optical ir sed of ngc821 by only a factor of @xmath76 and the x - rays by a factor of @xmath110 would produce an @xmath134 index of @xmath135 , that is definitely too low with respect to what typically observed ( risaliti & elvis 2004 ) . therefore , a hidden agn may have a @xmath136% reflectance , rather than 1% as assumed above , and then @xmath137 erg s@xmath15 . this @xmath120 is a factor of @xmath73 lower than @xmath69 , which could be explained by a similarly lower value of the radiative efficiency ( with respect to the standard value of 0.1 used to estimate @xmath69 in sect . [ dead ] ) and/or of @xmath62 ( due , e.g. , to angular momentum in the flow before being captured by the mbh , sect . [ ang ] ) . with a 20% reflectance , and the h@xmath9 upper limit of sect . [ bc03 ] , this nucleus would lie close to the l(210 kev)l@xmath138 correlation ( panessa et al . 2006 ) , at higher x - ray luminosities with respect to the best fit line but within the observed scatter . another observational constraint comes from the fact that thermal reradiation of an obscured continuum should appear as dust emission in the ir . the irac data in fig . [ sed ] fit very well to the sed of a stellar population ( sect . [ bc03 ] ) , so that any warm / hot dust must be negligible . the @xmath139 flux ( tab . [ tablesed ] ) corresponds to @xmath140 erg s@xmath15 , and any fraction due to dust must be much less than this . a conservative limit would be that the dust flux is less than the difference of the fluxes at @xmath139 and @xmath141 , otherwise the sed would depart from the galaxy template . that difference ( 0.5 mjy , from tab . [ tablesed ] ) corresponds to a luminosity limit of @xmath142 erg s@xmath15 , that is a factor of a few less than @xmath69 . the prevalence of a non - nuclear origin for the @xmath3 irac measurements , that refer to an area of 4 square arcseconds , is indicated also by a color analysis . @xmath3 irac colors have been shown to provide a robust technique for identifying agns , i.e. , inactive galaxies can be separated from agns in a [ 3.6][4.5 ] versus [ 5.8][8.0 ] plot ( stern et al . 2005 ) . we calculated such colors for the nucleus of ngc821 , using fluxes derived with a local background in order to better single out the nuclear emission ( tab . [ spizflu ] ) , and found that it falls well within the region of normal galaxies . this confirms that stars within the central @xmath44 radius dominate over the ir nuclear emission ( sect . [ bc03 ] ) . longer wavelength data are crucial to definitely rule out the possibility of heavily obscured emission . unfortunately , ngc821 has not been detected by @xmath143 ( knapp et al . 1989 ) detection of 0.5 jy at 100@xmath45 could be placed [ j. knapp 1994 , private communication to ned ( http://nedwww.ipac.caltech.edu/ ) ] , corresponding to @xmath144 erg s@xmath15 . ] . @xmath3 mips imaging has been scheduled and , possibly together with deeper @xmath4/nicmos f160w images , should be useful to definitely test the hypothesis of a hidden agn . ngc821 is an extreme example of a quiescent mbh , for which deep @xmath0 and vla observations revealed a very sub - eddington nuclear source for the first time . since ngc821 is relatively nearby , it is a good testcase to study how accretion proceeds in the very low luminosity galactic nuclei that are the vast majority in the local universe ( sect . [ intro ] ) . our multiwavelength analysis has shown that : 1 . a radio counterpart to the @xmath0 nuclear source s2 is detected at 1.4 ghz , with a flux density of 127 @xmath2jy . a source is also marginally ( at the @xmath145 level ) detected at 4.8 and 8.4 ghz . within current uncertainties , the radio spectral shape of the nuclear source is relatively flat . upper limits can be placed at the positions of the other elongated @xmath0 sources in the central galactic region ( s1 and s4 ) . 2 . the central @xmath37 radius region is also detected with @xmath3 irac , with the emission peak coincident with the position of the galactic center . archival @xmath4 images taken with wfpc2/f555w , wfpc2/f814w and nicmos ( h - band ) do not show a pointlike source at the galactic center . a dead mbh could be expected from the lack of detection of fuel at all wavelengths . however , the stellar mass losses in the circumnuclear region should produce a luminosity of @xmath146few@xmath64 erg s@xmath15 , if they end in a standard disc ( such a scaled - down agn would work at @xmath147few@xmath67 ) . 4 . disc - like stellar kinematics in the central galactic region , with the consequent angular momentum of the accretion flow at large radii , may account for a reduction of the mass accretion rate ; mhd simulations with rotation and cooling by radiation included are needed to estimate the final luminosity of the flow . starformation in the accreting material could also reduce the actual fuel supply to the mbh . the mid - ir to v - band photometric data for the central @xmath44 radius region , though , are very well fitted by the spectral energy distribution of an old and metal rich stellar population . the upper limit on pointlike 210 kev nuclear emission , together with the 5 ghz ( @xmath93 ) and 8.4 ghz ( @xmath100 ) detections , and the mbh mass known from @xmath4 data , are consistent with the predictions for radiatively inefficient accretion coupled to a compact , nuclear jet . the relatively flat radio spectral shape of the nuclear source is also consistent with being produced by the base of a jet . the extended @xmath0 sources s1 and s4 could be due to several knots in a jet , given their x - ray to radio luminosity ratios not too far from those observed for the knots in the jet of cen a ; more sensitive radio observations are needed to draw final conclusions on this possibility . if these sources or the central radio source are jet - like , their mechanical energy would be a very small fraction ( a few@xmath113 ) of the energy associated with the rest mass of the material being accreted . the nucleus of ngc821 could be a standard agn working at @xmath118few@xmath148 but heavily obscured as in a compton - thick source , possibly also extended as s2 is . the nicmos upper limit on the nuclear emission constrains the reflectance in the x - rays , but does not rule out this possibility . thermal re - radiation by dust is expected in the obscured scenario : the possible presence of warm / hot dust is already constrained by the @xmath3 mid - ir data , but far - ir observations are needed to assess the emission from cold dust . in conclusion , the deep study of the nucleus of ngc821 has shown that mbhs in the local universe are still emitting , though at an extremely low level , even in low @xmath149 ellipticals , that are expected to be very poor of hot gas . this agrees with the findings of hydrodynamical simulations that stellar mass losses in the circumnuclear region can fuel the mbh , even in low @xmath149 galaxies . the present multiwavelength investigation leaves open the possibilities that the final stages of accretion are radiatively inefficient , or that a standard disc in a compton - thick scaled - down agn is present ( provided that the radiative efficiency and/or the mass accretion rate are reduced ) . deeper nicmos and radio observations , together with far - ir data , are expected to solve the puzzle of the nature of the accretion process in this intriguing nucleus . we thank m. bolzonella for having provided the fitting seds for sect . s.p . acknowledges financial support from asi ( agenzia spaziale italiana ) contract i/023/05/0 . partial support for this work was provided by the nasa @xmath0 guest observer grant go5 - 6110x and by the @xmath0 x - ray center nasa contract nas8 - 39073 and by the spitzer cycle-2 program 20371 . the data analysis was supported by the cxc ciao software and caldb . we have used nasa ned and ads facilities , and have extracted archival data from the hubble space telescope archive . allen , s. w. , dunn , r. j. h. , fabian , a. c. , taylor , g. b. , reynolds , c. s. 2006 , mnras , in press anderson , j. m. , ulvestad , j. s. , ho , l. c. 2004 , apj 603 , 42 balbus , s.a . , hawley , j.f . 2002 , apj 573 , 749 bassani , l. , dadina , m. , maiolino , r. , salvati , m. , risaliti , g. , della ceca , r. , matt , g. , zamorani , g. 1999 , apjs 121 , 473 brzan , l. , rafferty , d.a . , mcnamara , b.r . , wise , m.w . , nulsen , p.e.j . 2004 , apj 607 , 800 blandford , r.d . , begelman , m.c . 2004 , mnras 349 , 68 bolzonella , m. , miralles , j.m . , pell , m. 2003 , a&a 363 , 476 bondi , h. 1952 , mnras , 112 , 195 bruzual , g. , charlot , s. 2003 , mnras 344 , 1000 churazov , e. , sazonov , s. , sunyaev , r. , forman , w. , jones , c. , bhringer , h. 2005 , mnras 363 , l91 ciotti l. , dercole a. , pellegrini s. , renzini a. 1991 , apj 376 , 380 ciotti , l. , & ostriker , j.p . 2001 , apj , 551 , 131 ciotti , l. , & ostriker , j.p . 2007 , in press on apj ( astro - ph/0703057 ) croton , d.j . , springel , v. , white , s.d.m . , 2006 , mnras 365 , 11 david , l.p . , forman , w. , jones , c. 1991 , apj 369 , 121 de vaucouleurs g. , de vaucouleurs a. , corwin jr . , buta r.j . , paturel g. , fouque p. , 1991 , third reference catalogue of bright galaxies , ( new york : springer verlag ) ( rc3 ) de villiers , j .- p . , hawley , j. f. , & krolik , j. h. 2003 , , 599 , 1238 dickey , j. m. , lockman , f. j. 1990 , ara&a 28 , 215 di matteo , t. , allen , s.w . , fabian , a.c . , wilson , a.s . , & young , a.j . 2003 , apj , 582 , 133 elvis , m. , fassnacht , c. , wilson , a. s. , briel , u. 1990 , apj 361 , 459 elvis m. , wilkes , b.j . , mcdowell , j.c . , et al . , 1994 , apjs 95 , 1 elvis , m. , risaliti , g. 2004 , _ supermassive black holes in the distant universe _ , amy j. barger ed . , kluwer academic publishers , dordrecht , the netherlands , p.187 emsellem , e. , cappellari , m. , peletier , r.f . , et al . 2004 , mnras 352 , 721 fabbiano , g. , elvis , m. , markoff , s. , siemiginowska , a. , pellegrini , s. , zezas , a. , nicastro , f. , trinchieri , g. , & mcdowell , j. 2003 , apj 588 , 175 fabbiano , g. , baldi , a. , pellegrini , s. , siemiginowska , a. , elvis , m. , zezas , a. , & mcdowell , j. 2004 , , 616 , 730 fabian , a.c . , & canizares , c.r . 1988 , nature , 333 , 829 fabian , a. c. , sanders , j. s. , taylor , g. b. , allen , s. w. , crawford , c. s. , johnstone , r. m. , iwasawa , k. 2006 , mnras 366 , 417 falcke , h. , koerding , e. , markoff , s. 2004 , a&a 414 , 895 fazio , g. g. , hora , j. l. , allen , l. e. , et al . 2004 , apjs 154 , 10 ferrarese , l. , merritt , d. 2000 , apj 539 , l9 ferrarese , l. , ford , h. 2005 , space science reviews 116 , 523 gebhardt , k. , bender , r. , bower , g. , et al . 2000 , apj 539 , l13 gebhardt , k. , richstone , d. , tremaine , s. , et al . 2003 , , 583 , 92 graham , a. w. , erwin , p. , caon , n. , & trujillo , i. 2001 , apj , 563 , l11 guainazzi , m. , fabian , a.c . , iwasawa , k. , matt , g. , fiore , f. 2005 , mnras 356 , 295 hardcastle , m. j. , worrall , d. m. , kraft , r. p. , forman , w. r. , jones , c. , & murray , s. s. 2003 , 593 , 169 hawley , j. f. , & balbus , s. a. 2002 , , 573 , 738 ho , l.c . 2002 , apj 564 , 120 ho , l. c. , filippenko , a. v. , sargent , w. l. w. 2003 , apj 583 , 159 ho , l.c . 2005 , ap&ss 300 , 219 hopkins , p.f . , hernquist , l. , cox , t. j. , di matteo , t. , robertson , b. , springel , v. 2006a , apjs 163 , 1 hopkins , p.f . , narayan , r. , hernquist , l. 2006b , apj 643 , 641 janiuk , a. , czerny , b. , siemiginowska , a. , & szczerba , r. 2004 , apj 602 , 595 igumenshchev , i.v . , narayan , r. , abramowicz , m.a . 2003 , apj 592 , 2042 kataoka , j. , stawarz , . 2005 , , 622 , 797 kataoka , j. , stawarz , . , aharonian , f. , takahara , f. , ostrowski , m. , edwards , p. g. 2006 , , 641 , 158 kennicutt , r.c . 1998 , ara&a 36 , 189 knapp , g. r. , guhathakurta , p. , kim , d .- w . , jura , m. a. 1989 , apjs 70 , 329 krding , e. g. , jester , s. , fender , r. 2006 , mnras 372 , 1366 kroupa , p. 2001 , mnras 322 , 231 lauer t.r . , et al . 2005 , aj 129 , 2138 loewenstein , m. , mushotzky , r.f . , angelini , l. , arnaud , k.a . , & quataert , e. 2001 , , 555 , l21 levenson , n. a. , heckman , t. m. , krolik , j. h. , weaver , k. a. , zycki , p. t. 2006 , apj 648 , 111 machacek , m. , nulsen , p. e. j. , jones , c. , forman , w. r. 2006 , apj 648 , 947 matt , g. , guainazzi , m. , maiolino , r. et al . 1999 , a & a 341 , l39 mcnamara , b. r. , nulsen , p. e. j. , wise , m. w. , rafferty , d. a. , carilli , c. , sarazin , c. l. , blanton , e. l. 2005 , nature 433 , 45 merloni , a. , heinz , s. , di matteo , t. 2003 , mnras 345 , 1057 moscibrodzka , m. 2006 , a&a 450 , 93 nagar , a. , falcke , h. , wilson , a.s . 2005 , a&a 435 , 521 narayan , r. 2005 , ap&ss 300 , 177 nemmen , r.s . , storchi - bergmann , t. , yuan , f. , eracleous , m. , terashima , y. , wilson , a.s . 2006 , 643 , 652 ogle , p. m. , marshall , h. l. , lee , j. c. , canizares , c. r. 2000 , apj 545 , 81 omma , h. , binney , j. , bryan , g. , & slyz , a. 2004 , , 348 , 1105 ostriker , j.p . , ciotti , l. 2005 , rspta 363 , 667 owen , f. n. , eilek , j. a. , kassim , n. e. 2000 , apj 543 , 611 panessa , f. , bassani , l. , cappi , m. , dadina , m. , barcons , x. , carrera , f. j. , ho , l. c. , iwasawa , k. 2006 , a&a 455 , 173 panessa , f. , barcons , x. , bassani , l. , cappi , m. , carrera , f. j. , ho , l. c. , pellegrini , s. 2007 , a&a in press pellegrini s. , venturi , t. , comastri , a. , fabbiano , g. , fiore , f. , vignali , c. , morganti , r. , trinchieri , g. 2003 , apj 585 , 677 pellegrini , s. 2005a , , 624 , 155 pellegrini , s. 2005b , mnras 364 , 169 pellegrini , s. , baldi , a. , kim , d.w . , fabbiano , g. , soria , r. , siemiginowska , a. , elvis , m. 2007 , submitted to apj ( p07 ) pinkney , j. , gebhardt , k. , bender , r. , et al . 2003 , apj 596 , 903 proctor , r. n. , forbes , d. a. , forestell , a. , gebhardt , k. 2005 , mnras 362 , 857 proga , d. , begelman , m.c . 2003 , apj 592 , 767 ptak , a. , terashima , y. , ho , l. c. , quataert , e. 2004 , apj 606 , 173 quataert , e. , di matteo , t. , narayan , r. , ho , l.c . 1999 , 525 , l89 ravindranath , s. , ho , l.c . , peng , c.y . , filippenko , a.v . , sargent , w.l.w . 2001 , aj 122 , 653 sarzi , m. , falcon - barroso , j. , davies , r.l . , et al . 2006 , mnras 366 , 1151 sazonov , s.yu . , ostriker , j.p . , ciotti , l. , sunyaev , r.a . 2005 , , 358 , 168 schlegel et al . 1998 , apjs 500 , 525 scorza , c. , bender , r. 1995 , a&a 293 , 20 shakura , n.j . , & sunyaev , r.a . 1973 , a&a , 24 , 337 siemiginowska , a. , czerny , b. , kostyunin , v. 1996 , apj 458 , 491 soria , r. , fabbiano , g. , graham , a. , baldi , a. , elvis , m. , jerjen , h. , pellegrini , s. , siemiginowska , a. 2006a , apj 640 , 126 soria , r. , graham , a. , fabbiano , g. , baldi , a. , elvis , m. , jerjen , h. , pellegrini , s. , siemiginowska , a. 2006b , apj 640 , 143 springel , v. , di matteo , t. , hernquist , l. 2005 , apj 620 , l79 stern , d. , eisenhardt , p. , gorjian , v. , et al . 2005 , apj 631 , 163 stone , j.m . , pringle , j.e . , begelman , m.c . 1999 , mnras 310 , 1002 tan , j. , blackman , e. 2005 , mnras 362 , 983 terashima , y. , wilson , a.s . 2003 , apj 583 , 145 tonry , j.l . , dressler , a. , blakeslee , j.p . , et al . 2001 , , 546 , 681 uttley et al . 2003 , apj 595 , 656 wang , r. , wu , x .- b . , kong , m .- z . 2006 , apj 645 , 890 werner , m. w. , roellig , t. l. , low , f. j. , et al . 2004 , apjs 154 , 1 wrobel , j. m. , heeschen , d. s. 1991 , aj 101 , 148 wu , q. , cao , x. 2005 , apj 621 , 130 yuan , f. , cui , w. 2005 , apj 629 , 408 zensus , j.a . 1997 , ara&a 35 , 607 lccc position ( j2000 ) & ra & dec & ref . + @xmath4 ( galactic center ) & -1.truecm @xmath38 @xmath39 @xmath40 - 0.1truecm.13 & @xmath41 @xmath42 @xmath152 & p07 + @xmath0 & -1.truecm @xmath38 @xmath39 @xmath40 - 0.1truecm.10 & @xmath41 @xmath42 @xmath153 & p07 + vla & -1.truecm @xmath38 @xmath39 @xmath154 & @xmath155 @xmath156 @xmath157 & sect . [ vla ] + @xmath3 & -1.truecm @xmath38 @xmath39 @xmath40 - 0.1truecm.11 & @xmath41 @xmath42 @xmath43 & sect . [ spiz ] + size & @xmath158 & + spectral analysis of @xmath0 data : & & + net counts & 246@xmath159 & + model @xmath160 : & & + n@xmath161 ( 10@xmath162 @xmath123 ) & @xmath163 & + @xmath164 & @xmath165 & + @xmath166dof & 11.5/9 & + l(0.38 kev)/@xmath167 erg s@xmath15 & 6.0 & + l(210 kev)/@xmath167 erg s@xmath15 & 3.8 & + model @xmath168 : & & + n@xmath161 ( 10@xmath162 @xmath123 ) & @xmath163 & + @xmath164 & @xmath169 & + ew ( kev ) & 0.9 ( @xmath170 ) & + @xmath166dof & 11.2/8 & + l(0.38 kev)/@xmath167 erg s@xmath15 & 6.3 & + l(210 kev)/@xmath167 erg s@xmath15 & 5.0 & + ccrcc 1.4 ghz & @xmath174jy & @xmath56 & @xmath175 & vla ( sect . [ vla ] ) + 4.8 ghz & ( @xmath176jy ) & ( @xmath92 ) & @xmath177 & @xmath12 + 8.4 ghz & ( @xmath178jy ) & ( @xmath179 ) & @xmath180 & @xmath12 + 15 ghz & @xmath181 mjy & @xmath182 & r=0.15 circle & vla ( @xmath183 limit ; nagar et al . 2005 ) + 8 @xmath45 & 1.3 @xmath184 0.2 mjy & @xmath140 & r=@xmath185 circle & @xmath3 irac ( sect . [ spiz ] ) + 5.8 @xmath45 & 1.8 @xmath184 0.2 mjy & @xmath186 & @xmath12 & @xmath12 + 4.5 @xmath45 & 3.1 @xmath184 0.4 mjy & @xmath187 & @xmath12 & @xmath12 + 3.6 @xmath45 & 5.6 @xmath184 0.5 mjy & @xmath188 & @xmath12 & @xmath12 + 1.6 @xmath45 & @xmath189 mjy & @xmath190 & r=@xmath191 circle & nicmos ( ravindranath et al . 2001 ) + i ( @xmath192 ) & 7.45@xmath193 mjy & @xmath194 & r=@xmath185 circle & wfpc2 f814w ( sect . [ opt ] ) + @xmath12 & @xmath195 mjy & @xmath196 & r=@xmath197 circle & @xmath12 + r ( @xmath198 ) & 5.47@xmath199 mjy & @xmath200 & r=@xmath185 circle & int ( sect . [ opt ] ) + v ( @xmath201 ) & 3.54@xmath202 mjy & @xmath203 & @xmath12 & wfpc2 f555w ( sect . [ opt ] ) + @xmath12 & @xmath204 mjy & @xmath205 & r=@xmath197 circle & @xmath12 + @xmath206 hz & @xmath207ph.@xmath123s@xmath15kev@xmath15 & @xmath208 & @xmath209 & @xmath0 acis - s @xmath171 ( sect . [ endistr ] ) + @xmath210 hz & @xmath211ph.@xmath123s@xmath15kev@xmath15 & @xmath212 & @xmath12 & @xmath12 + @xmath213 hz & @xmath214ph.@xmath123s@xmath15kev@xmath15 & @xmath215 & @xmath12 & @xmath12 +	the relatively nearby ( distance=24.1 mpc ) elliptical galaxy ngc821 hosts an extreme example of a quiescent central massive black hole , for which deep @xmath0 observations revealed a nuclear source for the first time , with @xmath1 . we present here a multiwavelength study of this nucleus , including vla observations that detect a radio counterpart to the @xmath0 nuclear source at 1.4 ghz , with a flux density of 127 @xmath2jy and possibly a flat spectral shape ; we also consider new @xmath3 irac observations and archival @xmath4 images . with these data we discuss possible scenarios for the accretion modalities of the sole material that seems available for fuelling , i.e. , the stellar mass losses steadily replenishing the circumnuclear region . the final stages of accretion could be radiatively inefficient and coupled to a compact nuclear jet / outfow . the stellar mass losses could instead end up in a standard disc only if a compton - thick agn is present . two extended sources detected by @xmath0 close to the nucleus could be due to several unresolved knots in a jet . if a jet is present , though , its kinetic energy would be only a very small fraction of the energy associated with the rest mass of the material being accreted . starformation close to the nucleus is not shown by the available data . deeper nicmos , radio and far - ir observations are expected to further constrain the accretion process in this intriguing nucleus .
in recent years the idea that the ground state of nuclear matter at high density contains a bose condensate of kaons has received considerable attention . the existence of this novel state of matter is suggested by the attractive _ s_-wave kaon - nucleon interactions found in low energy effective chiral lagrangians , which , by construction , preserve the symmetries of qcd . this is supported experimentally by kaonic atom data which indicate @xcite a strongly attractive @xmath1 optical potential . attractive kaon - nucleon interactions imply that in dense matter , such as that in the interior of neutron stars , the effective kaon mass is lower than in free space . it could thus be energetically favorable for leptons in the matter to convert to kaons via strangeness changing interactions , in which case the lowest energy kaon mode becomes macroscopically populated , forming a condensate . since the original suggestion by kaplan and nelson@xcite , studies of kaon condensation and kaon propagation in the nuclear medium can broadly be divided into three categories : studies of the basic interactions , consequences for heavy - ion collisions and astrophysical implications . a number of studies have been aimed at improving the description of kaon - nucleon interactions over that of the original model . brown , lee , rho and thorsson@xcite included all terms to next - to - leading order ( @xmath2 ) in chiral perturbation theory , and fitted the expansion parameters to @xmath3 scattering lengths . lee , brown , min and rho@xcite included all terms to next - to - next - to - leading order ( @xmath4 ) in chiral perturbation theory and also fitted kaonic atom data . have also included the @xmath5 ( which dominates the low - energy @xmath6 interaction ) , four - baryon interactions , and pauli blocking . other noteworthy studies are those of ref.@xcite , concerning the effect of kaon - nucleon correlations , and the work of the kyoto group ( see e.g. ref.@xcite ) . waas , kaiser , and weise@xcite have recently investigated kaon modes in matter using a multichannel k - matrix analysis . for our purposes , it is worth mentioning that the majority of these more refined investigations have shown that the basic mechanism for kaon condensation , namely the existence of an attractive interaction , is robust . as regards heavy ion collisions , strangeness is conserved on the relevant timescales and a kaon condensate is not likely to form . however , modification of the kaon modes following from the chiral lagrangian could have consequences in kaon flow@xcite , for subthreshold production of kaons@xcite , or for the @xmath7-meson peak in dilepton spectra@xcite . another major direction that investigations have taken is to study the astrophysical consequences of the formation of a kaon condensate ( the effects appear to be reduced if hyperons are also present @xcite , a possibility that we do not consider here ) . the presence of such a condensate considerably enhances the proton concentration which leads to a softer equation of state ( eos ) and a lower maximum mass for the cold neutron star . the actual value of this maximum mass is dependent on the nuclear eos , but for the non - relativistic eos employed here it is quite close @xcite to the lower limit of @xmath8 set by psr 1913 + 16 @xcite . an interesting effect occurs when the neutron star has just formed since electron neutrinos are trapped for about 1015 s ( also the entropy / baryon is @xmath9 , but this plays a lesser role ) . the changes in the stellar composition induced by trapping increase the threshold for kaon condensation . this stiffens the eos and allows a maximum mass which is @xmath10 larger than in the absence of trapped neutrinos . thus there is the possibility of metastable neutron stars which become black holes after about 1015 s when the trapped neutrinos have left @xcite . this could account for the fact that fewer pulsars are observed in supernova remnants than expected from conventional scenarios @xcite . as regards dynamical simulations @xcite of the delayed collapse of hot neutron stars into black holes , baumgarte , shapiro and teukolsky@xcite have recently examined the effect of kaon condensates . here the cold eos is supplemented with simple assumptions about corrections for finite temperature . one - loop calculations to date are valid only up to the density at which the condensate forms where the condensate amplitude is vanishingly small . in order to go beyond threshold we analyze the fluctuations around the kaon condensate to one loop order and include the full non - linearities of the lagrangian . this is of general theoretical interest , as well as in the particular context of neutron stars . while we shall restrict our numerical analysis here to the zero - point contribution at zero temperature , the formalism also provides a consistent treatment of the thermal contributions due to kaons . these would be needed in studies of newborn neutron stars . in the work which follows , we shall make the following approximations . nucleons are treated non - relativistically , whereas a more complete treatment would involve a relativistic treatment to one loop order . in the meson sector , we restrict the discussion to kaonic degrees of freedom . we thus omit from the discussion fluctuations in the pion and eta sectors . physically one might argue that ignoring non - kaonic pseudoscalar degrees of freedom is justified since one expects that the corresponding modes are less modified in nuclear matter than those of kaons . the methods developed here could be used to include the full octet of pseudoscalar fluctuations , but this would further complicate the present analysis . lastly , we have sought to treat the chiral lagrangian at the most simple level starting from the lagrangian of kaplan and nelson @xcite , although one would wish to include the additional terms of refs . @xcite at a later stage . the paper is organized as follows . in sec . ii , we present the basic formalism , including the evaluation of the kaon partition function , a discussion of divergences in the zero - point energy , the nucleon hamiltonian , and the equilibrium conditions . in sec . iii we compare numerical results in neutron star matter with and without zero - point fluctuations . we discuss the critical density for kaon condensation , the composition of the star and the eos . following the conclusions in sec . iv are three appendices . appendix a concerns the cancellation of divergences dependent on the chemical potential in a toy model , while appendix b discusses the case of the actual chiral lagrangian . in appendix c we derive the critical density from the kaon propagator to one - loop order , and show that it is the same as that obtained by expanding the zero - point energy in terms of the kaon fields . our principal interest is the evaluation of the impact of the kaon zero - point energy contribution on the structure of neutron stars . in subsec . a we obtain the partition function for the chiral lagrangian of kaplan and nelson @xcite to one loop order and in subsec . b we discuss the elimination of the divergences . for the nucleons alone we use a simple non - relativistic hamiltonian which should be sufficient for present purposes , as outlined in subsec . c. the stellar equilibrium conditions that need to be satisfied are given in subsec . d. the @xmath11 chiral lagrangian involving the pseudoscalar meson octet and baryon octet is@xcite _ k & = & l^(1)+l^(2)+l^(0)_ct + l^(3)_ct+l^(4)_ct + l^(1 ) & = & i tr \|b v b + itr\|bv^[v_,b ] + 2 d tr \|b s^\ { a _ , b } + 2 f tr l^(2 ) & = & f^2 tr_u^u+ ctrm_q(u+u^-2 ) + a_1tr\|b(m_q+h.c.)b + & & + a_2tr\|bb(m_q+h.c . ) + a_3\{tr\|bb } \{tr(m_qu+h.c . ) } + l^(c , d , f ) + l^(2)_ct . [ pje0 ] the bracketed superscripts denote the order in the low momentum expansion defining the effective field theory . the various one - loop counterterms , with subscripts _ ct _ , are discussed in appendices a and b ; here in the main text we will simply put them all together in a single @xmath12 . the additional terms denoted by @xmath13 are given in refs . @xcite and those with coefficients @xmath14 can contribute to @xmath15-wave kaon - nucleon interactions . in the interests of simplicity we set @xmath16 and so use as a starting point the lagrangian of kaplan and nelson@xcite . the lagrangian ( [ pje0 ] ) is derived in the limit of heavy baryons , where @xmath17 is the velocity four - vector of the baryons , and @xmath18 is the spin operator ( @xmath19,@xmath20 ) . in eq . ( [ pje0 ] ) , @xmath21 is the non - linear field involving the pseudoscalar meson octet , from which we retain only the @xmath22 contributions @xmath23 and from the baryon octet we retain just the nucleon contributions @xmath24 consequently the @xmath25-wave interactions in @xmath26 with coefficients @xmath27 and @xmath28 do not enter ( @xmath29 ) . in eq . ( [ pje0 ] ) , the quark mass matrix @xmath30 ; _ i.e. _ , only the mass of the strange quark is taken to be non - zero . for the mesonic vector current , @xmath31 , only the time component survives in an infinite system with @xmath32 $ ] . also , the pion decay constant @xmath33 mev , and @xmath34 , and @xmath35 are constants . after some algebra , the relevant part of @xmath36 can be written _ k&=&(1+)_k^+^k^- + + & & + 2 + l_ct . [ lkaon ] here , the kaon mass is given by @xmath37 , and we have employed the following definitions : & & ^2=k^+k^-;=()^2;2=()^2 + & & b=(1+x);c=(a_1x+a_2 + 2a_3 ) . [ defs ] expectation values of nucleon operators are evaluated in the mean - field approximation . the total nucleon number density is @xmath38 and the proton fraction is @xmath39 . note that since we shall use a non - relativistic model for the nucleons we have not distinguished here between the scalar density and the number density in the expression for @xmath40 . the nucleon kinetic energy in @xmath26 will be considered in subsection c and so is not included in eq . ( [ lkaon ] ) . we have not included in eq . ( [ lkaon ] ) terms which simply give a constant shift to the baryon masses ; they indicate that @xmath41 mev and @xmath42 mev , using the hyperon - nucleon mass differences . the remaining constant @xmath43 is not accurately known , and we shall use values in the range @xmath44 to @xmath45 mev corresponding to 0 to 20% strangeness content for the proton . the corresponding range for the kaon - nucleon sigma term , @xmath46 is 167520 mev . some guidance is provided by recent lattice gauge simulations @xcite , which find that the strange quark condensate in the nucleon is large , _ i.e. _ , @xmath47 . from the relation @xmath48 and using @xmath49 mev , we obtain @xmath50 mev , which is in the middle of our range of values . in order to determine the kaon partition function we generalize the finite temperature procedure outlined by kapusta @xcite and benson , bernstein and dodelson @xcite . first , by studying the invariance of the lagrangian under the transformation @xmath51 , the conserved current density can be identified . the zero component , _ i.e. _ the charge density , is @xmath52 next , we transform to real fields @xmath53 and @xmath54 , @xmath55 the lagrangian becomes _ k&=&a__1^_1 + b__2^_2+c__1^_2 + & & + 2+l_ct , [ lphi ] where a=;b=;c=. [ abc ] since , to our order of approximation , the counterterm lagrangian , @xmath12 , is to be evaluated with constant mean fields it plays no role in determining the conjugate momenta . these are _ 1&=&a_0_1+c_0_2 - 2 b_2 + _ 2&=&b_0_2+c_0_1 + 2 b_1 . the hamiltonian density is @xmath56 , and the partition function of the grand canonical ensemble can then be written as the functional integral @xmath57[d\pi_2]\int_{periodic}[d\phi_1][d\phi_2]\nonumber\\ & & \quad\times\exp\left\{\int\limits_0^{\beta}d\tau\int d^3x\left ( i\pi_1\frac{\partial\phi_1}{\partial\tau}+ i\pi_2\frac{\partial\phi_2}{\partial\tau}-{\cal h}_k(\phi_i,\pi_i ) + \mu j_0(\phi_i,\pi_i)\right)\right\}\;.\label{zint}\end{aligned}\ ] ] here @xmath0 is the chemical potential associated with the conserved charge density , @xmath58 is the inverse temperature and the fields obey periodic boundary conditions in the imaginary time @xmath59 , namely @xmath60 . the gaussian integral over momenta in eq . ( [ zint ] ) can be performed yielding z_k&=&n^e^-s + s&=&_0^dd^3x\ { a()^2 + b()^2 + 2c + & & + a(_1)^2+b(_2)^2 + 2c_1_2 + 2i(+2 b)_2_1 + & & + ( _ 1 ^ 2+_2 ^ 2)-2l_ct}. [ action ] the dependence of the action ( [ action ] ) upon the chemical potential @xmath0 can also be obtained by gauging the fields as discussed in appendices a and b. next the fields are decomposed into constant condensate parts and fluctuations according to @xmath61 here @xmath62 denotes the volume of the system . in the fourier expansion of the fluctuating part @xmath63 and the matsubara frequency @xmath64 . since we wish to go to one loop order , @xmath12 is evaluated with constant condensate fields and we write the contribution as a potential @xmath65 . the remainder of @xmath66 must be expanded to second order in the fluctuations . the partition function can then be written in the form z_k = ne^-v(v_tree+v_ct)_n , [ d_1,n ( ) ] [ d_2,n ( ) ] ^e^-s_loop . [ zone ] introducing the definitions & & ( ) = 2 ( m_k^2 + c - 2 b ) f^2 ^2- ^2 f^2 ^2 + & & ( ) = - = 4 b f^2 ^2 + f^2 ^2 , the tree potential is v_tree= ( ) . [ vtree ] the one loop contribution to the action then becomes s_loop&=&_n,(_1,-n(- ) , _ 2,-n(- ) ) ( ) , + ^-2f^2 d_11 & = & af^2 ( _ n^2 + * p^2 ) - _ n ( ) 2 + ^2+ + ^-2f^2 d_22 & = & bf^2 ( _ n^2 + p^2 ) + _ n ( ) 2 + ^2 + + ^-2f^2 d_12 & = & cf^2 ( _ n^2 + p^2 ) + 2 _ n ( + ^2 ( ) ) + ( - ) + ^-2f^2 d_21 & = & cf^2 ( _ n^2 + p^2 ) -2 _ n ( + ^2 ( ) ) + ( - ) , [ dmatrix ] with a prime denoting a partial derivative with respect to @xmath67 . here the quantities @xmath68 and @xmath69 of eq . ( [ abc ] ) are to be evaluated with the condensate fields , giving a=1+(-1)^2;b=1+(-1)^2;c=(1- ) , and @xmath70 . the latter replacement should also be made in the measure of the integral ( [ zone ] ) , to the order considered . then the integration can be carried out and the grand potential @xmath71 is given by = -z_k = v_tree+v_ct + _ n , . [ grand1 ] evaluating the determinant yields & = & ^4 + & & ^2 ( _ n^2 + e_+^2 ) ^2 ( _ n^2 + e_-^2 ) , [ detd ] where & & c_1 = ( ) , c_2 = , c_3 = [ cs ] + & & e_^2 ( ) = p^2 + c_1 + c_2 + c_3 . [ disp ] notice that there is no longer any dependence on the condensate phase @xmath72 . the sum over the matsubara frequencies in eq . ( [ grand1 ] ) can then be evaluated in the usual way @xcite to yield & & _ k ^-1=v_tree+v_ct+v_zp+v_thermal + & & v_zp = [ vzp ] + & & v_thermal = ^-1 ( 1 - e^-e_+())(1 - e^-e_- ( ) ) . [ vthermal ] we are interested in zero temperature here , so @xmath73 may be dropped . the zero - point energy @xmath74 , or equivalently , the one - kaon - loop contribution to the effective potential@xcite , can be represented as an infinite series of diagrams , some of which are shown in fig . vertices are given by the lagrangian ( [ lkaon ] ) . external legs represent nucleon or kaon mean fields . it is interesting to note the chiral order of the diagrams contributing to the effective potential . consider the _ scattering amplitudes _ corresponding to the diagrams of fig . 1 , where the external lines are on their respective vacuum mass - shells . the chiral order , @xmath75 , of a given scattering amplitude is given by the weinberg@xcite counting rule , according to which an amplitude involving @xmath76 external nucleon lines and @xmath77 external kaon lines is proportional to @xmath78 , where @xmath79 is a characteristic small momentum scale involved in the process and = 2 + 2l - e_n + _ i ( d_i + n_i -2 ) , [ count ] where @xmath80 is the number of loops , the sum is over all vertices @xmath81 , @xmath82 the number of derivatives that act on the @xmath81th vertex and @xmath83 the number of nucleon lines attached to the @xmath81th vertex . in this case 1a gives @xmath84 . furthermore , it easy to convince oneself that adding a vertex from @xmath26 , corresponding to the @xmath85 term in eq . ( [ lkaon ] ) , will decrease @xmath75 by 1 . on the other hand , adding a vertex from @xmath86 , corresponding to the kinetic , mass , or @xmath40 terms in eq . ( [ lkaon ] ) leaves @xmath75 invariant . the fact that these scattering amplitudes give @xmath87 due to the @xmath26 interaction might appear to be grounds for concern . however , exactly the same apparent counting problem arises when one considers the computation of the tree - level in - medium kaon propagator . the correct in - medium propagator is indeed obtained by summing an infinite series@xcite in diagrams that here correspond to amplitudes with decreasing powers of @xmath75 . the evaluation of the effective potential is entirely analogous . we conclude that while eq . ( [ count ] ) provides an effective way to organize terms in order of importance for a given scattering process , it can not be naively applied in the same way to the effective potential in the case at hand . as shown in appendices a and b , @xmath65 cancels the ultraviolet divergences in @xmath74 . for our purposes the zero - point energy is most conveniently written v_zp = ( ^2 + c_1 + c_2 + c_3 + ) ^12 . [ zpdiv ] we regularize the infinities that occur here by introducing an ultraviolet cutoff @xmath88 and requiring @xmath89 . ( dimensional regularization is discussed in appendix b. ) the counterterm potential @xmath65 following from @xmath12 ( see appendix b ) is v_ct = - ( d_4+f_4 ) - ( c_1 + 2 c_2 + 2 c_3 ) ( d_2+f_2 ) + ( d_0+f_0 ) . [ ct ] here we have introduced finite terms @xmath90 , @xmath91 and @xmath92 as well as the quartically , quadratically and logarithmically divergent integrals d_4 & = & _ k k = [ quartdiv ] + d_2 & = & _ k = [ quaddiv ] + d_0 & = & _ k = ( ) . [ logdiv ] here @xmath93 is an arbitrary scale introduced so that the argument of the logarithm is dimensionless . note that the quadratically and logarithmically divergent pieces of @xmath74 in ( [ zpdiv ] ) and @xmath65 in ( [ ct ] ) are individually dependent upon the chemical potential @xmath0 . this dependency on the chemical potential @xmath0 arises from the functions of the kaon fields multiplying the kinetic terms in eq . ( [ lkaon ] ) . however , @xmath0-dependent divergences in the terms @xmath74 and @xmath65 cancel , as do the @xmath0-independent divergences of the vacuum theory . the cancellation of @xmath0-dependent divergences follows from a careful treatment of the counterterm lagrangian which regulates the vacuum theory . further discussion of this point is given in appendices a and b. here we note that @xmath0 may be considered a constant external vector potential @xmath94 , with vanishing spatial components . the general statement being made is that all green functions with constant external @xmath94 fields are finite . it is easy to check this for simple examples related to our theory , such as scalar electrodynamics . the fact that it also holds for a complicated nonlinear theory expanded about an arbitrary point @xmath53 , @xmath54 in function space is almost certainly a consequence of the ward - takahashi identities associated with the u(1 ) symmetry transformation preceding eq . ( 7 ) . in order to determine the finite contributions @xmath95 we apply three conditions in the vacuum . first @xmath71 , or equivalently the pressure , should be zero in the vacuum and , since there is no condensate , the loop contribution must vanish . secondly , we require the zero - momentum propagator to be of the usual form @xmath96 , see appendix c. thirdly we require that the four - point vertex @xmath97 of eq . ( c3 ) be unmodified in vacuum . these conditions imply & & v_zp+v_ct\|_vacuum=0 + & & ( v_zp+v_ct)\|_vacuum=0 + & & ( v_zp+v_ct)\|_vacuum=0 . expanding to fourth order in @xmath67 , one finds & & f_4=;f_2=- + & & f_0=(+- ) , where using our three momentum cutoff , the constant @xmath98 ; interestingly a closely similar value for the constant is obtained with dimensional regularization , namely @xmath99 ( here @xmath100 represents the digamma function @xcite ) . notice that the arbitary scale @xmath93 only occurs in the combination @xmath101 where it cancels out . the disappearance of the arbitary scale is one justification for introducing the finite counterterms , we discuss another below . in evaluating the grand potential for the general case numerical integration of eq . ( [ zpdiv ] ) for @xmath74 is required . however below threshold when @xmath102 ( or just at threshold when @xmath67 is vanishingly small ) the expressions simplify greatly and an analytic form can be obtained . explicitly we find @xmath103 , @xmath104 , and eigenvalues e _ = ( b+ ) ^2=b^2 + m_k^2 + c = 0 . [ epm ] notice that @xmath105 is simply the tree level energy of a @xmath22 meson . in this limit the one loop energy becomes v_zp + v_ct & = & - ( d_4+f_4 ) - ^2(d_2+f_2 ) + ^4 ( d_0+f_0 ) + & = & - = 0 , [ vzp0 ] which vanishes in the vacuum since @xmath106 there . to describe the nuclear kinetic energy and interactions we employ the non - relativistic approach suggested by prakash , ainsworth and lattimer @xcite , which has been rather widely employed , _ e.g. _ refs . @xcite . here the hamiltonian density is written _ n= e_f^0 u^n_0 + v(u ) + n(1 - 2x)^2s(u ) , [ hnuc ] where nuclear saturation density is denoted by @xmath107 ( @xmath108 ) and the corresponding fermi momentum and energy are @xmath109 and @xmath110 , respectively . the potential contribution to the energy density of symmetric nuclear matter is denoted by @xmath111 , where @xmath112 and it is given by v(u)= au^2n_0 + + 3un_0_i=1,2c_i()^3 ( - ) . here @xmath113 is the fermi momentum which is related to @xmath114 by @xmath115 . the remaining quantities are parameters which are given in refs . we choose the set which yields a compression modulus @xmath116 mev for equilibrium nuclear matter since this is in the middle of the generally accepted range of @xmath117 mev . the remaining term in eq . ( [ hnuc ] ) is the symmetry energy , _ i.e. _ it describes the change in the energy density when the proton fraction @xmath39 differs from @xmath118 . again following ref . @xcite we take s(u)=(2 ^ -1 ) e_f^0(u^-f(u))+s_0f(u ) , where @xmath119 mev is the bulk symmetry energy parameter . since neutron star properties are not strongly dependent on the form of @xmath120 , we choose the simplest of the suggested parametrizations namely @xmath121 . in terms of the hamiltonian density , the grand potential for nucleons is = h_n-_n n + xn , where @xmath122 and the chemical potential @xmath123 is the fermi energy for neutrons . we shall study cold , catalyzed neutron stars which are in chemical equilibrium under beta decay processes . the process @xmath124 in equilibrium establishes the relation _ n-_p = _ e , [ beta ] since the neutrinos have left the star and their chemical potential is therefore zero . at densities where @xmath0 exceeds the muon mass @xmath125 , muons can be formed by @xmath126 , hence the muon chemical potential is @xmath127 . negatively charged kaons can be formed in the process @xmath128 when @xmath129 becomes equal to the energy of the lowest eigenstate of a @xmath1 in matter . chemical equilibrium requires that @xmath130 , as we have implicitly assumed in the text . to complete our grand potential we need the lepton component for which it is sufficient to use the standard non - interacting form : = -+(\|\|-m_)\ { - } , where @xmath131 is the heaviside function ( @xmath132 if @xmath133 , @xmath134 if @xmath135 ) and the quantity @xmath136 . the total grand potential is then _ total = _ k + _ the values of the chemical potential @xmath0 , the proton fraction @xmath137 and the condensate amplitude @xmath67 are determined by extremizing @xmath138 . extremizing with respect to @xmath0 is equivalent to the requirement of charge neutrality . it yields 0=-n_k+n_p - n_l [ mumin ] , where @xmath139 and @xmath140 are the number densities for negatively charged kaons and leptons respectively . the kaon charge density is n_k = f^2(4b^2+^2 ) - . [ nk ] notice that the fluctuations carry no net charge in the absence of a condensate , _ i.e. _ @xmath141 for @xmath102 . here , and in the other cases , the zero point contribution requires a numerical integration and it is straightforward , though tedious , to evaluate the integrand . the lepton charge density is n_l=+ ( \|\|-m _ ) . next extremizing with respect to @xmath137 gives 0=n(2a_1m_s-)^2 + + n-4n(1 - 2x)s(u ) [ xmin ] . finally extremizing with respect to @xmath67 yields 0=f^2(m_k^2+c-2b-^2 ) + [ thmin ] . below the threshold for kaon condensation @xmath102 and this solution is always possible since the expression is an odd function of @xmath67 . above threshold the solution @xmath142 minimizes the grand potential . at threshold , where @xmath67 is vanishingly small , it is possible to obtain an analytic expression by retaining terms through order @xmath143 in @xmath71 . the extremization condition becomes [ uc ] 0&= & ( m_k^2+c)(1+)-2b(1 - 2)-^2(1 + ) [ uca ] & = & ( ^2-m_k^2+^2 ) . [ ucb ] this may alternatively be obtained by computing the inverse propagator , @xmath144 , to one loop order , see appendix c. notice that , by construction , the inverse propagator is @xmath145 in vacuum since @xmath146 there . in fact for any density @xmath147 . that this is necessary may be seen as follows . the arguments of the inner square root in eq . ( [ zpdiv ] ) is positive for all @xmath148 , only if @xmath149 _ or _ both @xmath150 and @xmath151 if @xmath152 . however in our calculations we have found that going even slightly above threshold implies @xmath153 , so that it is the latter condition that must be satisfied . now the above equation ( [ uca ] ) can be put in the form @xmath154 . since @xmath155 ( sec.iii ) , and @xmath85 and @xmath0 are positive , we must have @xmath156 . it is interesting to note that in the absence of finite counterterms the requirement @xmath156 can only be achieved in a very limited region of parameter space . having satisfied the conditions ( [ mumin ] ) , ( [ xmin ] ) and ( [ thmin ] ) , the energy density can be obtained from = + _ n n . the chemical potential @xmath123 may be obtained by functional differentiation of the energy density with respect to the neutron density , _ viz . _ @xmath157 , whence the pressure can be calculated from @xmath158 . equivalently one can write @xmath159 . in this section , we give numerical estimates for the effect of the loop corrections upon the critical density for kaon condensation and upon the properties of neutron star matter both above and below the condensate threshold . as follows from eqs . ( [ uc ] ) and ( c1 ) , the critical density for kaon condensation is the density at which the lepton chemical potential @xmath0 ( an increasing function of density ) , and the position of the zero - momentum @xmath1 pole of the kaon propagator in normal nuclear matter ( a decreasing function of density ) become equal . the chemical potential @xmath0 and the proton fraction @xmath137 , prior to condensation , are obtained from eqs . ( [ mumin ] ) and ( [ xmin ] ) by setting @xmath102 . the pole position of the kaon propagator is a solution to the quadratic equation @xmath160 , where @xmath161 is given in eq . ( c4 ) . in table i , we list the critical density for kaon condensation in units of nuclear matter density , @xmath162 , for three representative values of the parameter @xmath43 . in column 2 , @xmath163 is the critical density obtained at tree level@xcite , corresponding to setting @xmath164 in eq . ( [ uc ] ) and omitting the @xmath165 contributions from eqs . ( [ mumin ] ) and ( [ xmin ] ) . in column 3 , the loop contribution is included and the corresponding critical density is denoted by @xmath166 . we see that @xmath167 and , while the magnitude of the threshold shift increases for larger values of @xmath168 , it is always small . this due to the fact that in eq . ( [ uc ] ) @xmath169 which is small in comparison to unity . it is rather easy to show that @xmath167 . the tree level inverse propagator @xmath170 at threshold , while with loops included one has @xmath171 . since @xmath172 is a monotonic function of density and we have pointed out that @xmath156 , it follows that @xmath167 . the change in the threshold due to loops , small though it is , is dominated by the @xmath173 term this is shown in column 4 of table i for which we set @xmath174 in the loop expressions . in this case the threshold is essentially unchanged from tree level . the effect of the @xmath40 term can become quite critical at densities above threshold and in eq . ( [ defs ] ) the scalar density @xmath175 has been replaced by the baryon density @xmath176 . in ref.@xcite it was argued that in the heavy baryon limit , and to the chiral order considered , no distinction should be made between the operators @xmath177 and @xmath178 in the lagrangian ( [ pje0 ] ) . corrections to this would , in principle , require other effects of similar order for consistency . while @xmath179 for @xmath180 , at high density @xmath181 becomes less than @xmath182 . the results of ref . @xcite indicate that this can be significant , in particular it leads to an increase in @xmath183 . since our results are sensitive to this question , in some of our calculations we shall replace the baryon density @xmath182 in the quantity @xmath40 of eq . ( [ defs ] ) by a phenomenological form of @xmath181 . based on a simple parameterization of the effective nucleon mass obtained by serot and walecka@xcite , we use n_s = n ( 1 - ) [ ns ] for @xmath184 and set @xmath185 for @xmath186 . this is extremely crude , but it will give some qualitative feel for the effect . the results for the critical density with this approach , labelled @xmath187 , are shown in columns 5 and 6 of table i. the shift in the critical density between tree and tree + loop levels is slightly smaller , but basically similar to that obtained before . the overall increase in the critical density is expected , but is somewhat larger than models which treat nucleons relativistically @xcite would suggest . even below threshold there is a kaon zero - point contribution for normal neutron star matter . firstly there is a shift in the minimum due to the zero point contribution to the equilibrium eq . ( [ xmin ] ) at @xmath102 . secondly there is the zero point energy ( [ vzp0 ] ) itself . the shift in the proton fraction @xmath137 and the lepton chemical potential @xmath0 is negligible . table ii shows that @xmath188 , the zero - point energy density is approximately 13% of the total . thus the matter properties are only affected to a small degree so that it is not necessary to refit the parameters of the nuclear hamiltonian . above the critical density the properties of matter are obtained by solving eqs . ( [ mumin ] ) , ( [ xmin ] ) , and ( [ thmin ] ) numerically . in practice , one uses a three - momentum cutoff @xmath88 to evaluate eq . ( [ zpdiv ] ) and derivatives thereof . the cutoff used was @xmath189 mev / c , and results were found to be extremely stable with respect to variation in @xmath88 . for example taking @xmath190 mev / c , the largest change in any of the tabulated values was 0.3% . in table iii , we list the properties of normal neutron star matter ( column 2 ) and matter containing a kaon condensate at tree level ( columns 37 ) and at one - loop level ( columns 813 ) for @xmath191 mev . here the kaon fraction is defined as @xmath192 in terms of the kaon density @xmath139 of eq . ( [ nk ] ) , and the lepton fraction @xmath193 may be obtained from @xmath194 . the energy density is also listed . we have not felt it worthwhile to tabulate the remaining thermodynamic variable , the pressure , in view of the small size of the loop effects . for @xmath195 the parameter @xmath196 becomes negative which precludes real solutions to the zero point energies ; we discuss this further below . in table iv , we give the corresponding quantities for @xmath197 mev , in which case @xmath196 becomes negative for @xmath198 . this difficulty does not arise if we employ the phenomenological scaling given by eq . ( [ ns ] ) and the results are given for @xmath197 mev in table v. if one simply sets @xmath174 in the loop calculations the results shown in fig . 2 are obtained . in this figure @xmath199 represents the energy density gained by the formation of a condensate and is the difference between the total energy density and the tree level energy density for normal nuclear matter . we first note that loop effects for the case @xmath191 mev are essentially negligible . although they increase in magnitude for larger values of @xmath168 , only small modifications of the tree level results are found . table iv shows that just above threshold the tree + loop calculation gives values of @xmath67 and @xmath137 which are greater than , and @xmath0 which are less than , the corresponding tree level results . this naturally follows from the fact that @xmath166 is less than @xmath200 since , once a condensate is present , @xmath67 rises rapidly and @xmath137 is enhanced , while @xmath0 is reduced at high densities , this ordering is reversed . near the limiting density , @xmath201 , @xmath67 typically decreases by @xmath202% while @xmath137 decreases somewhat less , @xmath203% , and @xmath0 increases @xmath204% . the zero - point energy density is of the order @xmath205 and can decrease the tree level result significantly at high density when the latter can be small . for a given density the effects are smaller if the scalar density of eq . ( [ ns ] ) is employed as shown in table v. on the other hand , setting @xmath174 altogether as in fig . 2 yields larger effects at very high density . for @xmath206 mev , the above trends are also present , and are enhanced , as expected . the coefficient @xmath196 goes to zero for @xmath207 , at which point @xmath208 and @xmath209 . as mentioned above , @xmath196 often vanishes at some density above the critical density . the condition @xmath210 implies @xmath211 , where @xmath212 is the tree level energy density function of eq . ( [ vtree ] ) . the significance of this condition is best illustrated by the simple @xmath213 invariant lagrangian @xmath214 , where @xmath215 . the one - loop effective potential@xcite is @xmath216 , where the trace is over momentum . now take @xmath217 , with @xmath218 . for @xmath219 , @xmath220 is real for all @xmath7 . on the other hand , for @xmath221 , the `` mexican hat '' potential , the one - loop effective potential is complex in the region @xmath222 which signals a physical instability . weinberg and wu @xcite show that the imaginary part can be interpreted as a decay rate per unit volume of a well - defined state . the connection between the instability and the chaotic dynamics of a classical field theory has recently been discussed by matinyan and mller @xcite , see also references therein . unfortunately no solution is known to this general problem . our case is entirely analogous . at threshold @xmath223 corresponds to @xmath219 . as the density is increased , the tree level potential eventually corresponds to @xmath221 . this effect is more pronounced for larger values of @xmath168 . since the effective potential is by definition real , this shows that some , as yet unknown , technique is needed in these regions of parameter space . the two other cases we have considered , excluding @xmath40 from the loop contribution and employing the softer density dependence of @xmath40 with the parametrized scalar density @xmath181 alleviate this problem . we have derived the zero - point energy due to fluctuations in kaonic modes in neutron star matter containing a kaon condensate . to our knowledge , this calculation is the first in which the nonlinearities of the chiral lagrangian have been taken into account at the one loop level for densities _ above _ the critical density for kaon condensation . the zero - point energy was derived through the thermodynamic partition function by considering fluctuations around an arbitrary condensate . we have demonstrated by gauging the counterterms of the vacuum theory that no divergences are present when matter is introduced via the charge chemical potential . in the usual way the total thermodynamic potential , including the classical contribution , is extremized so that the minimum energy ground state is found for locally charge neutral and beta - equilibrated matter . the critical density for kaon condensation was found to be reduced by fluctuations , but only by less than 1% for the range of parameters considered . at higher densities the loop effects were negligible for @xmath191 mev which corresponds to zero strangeness content for the nucleon . the effects were slightly increased with larger values of @xmath168 , corresponding to larger value of the kaon nucleon sigma term @xmath173 , but they remained small in comparison to the dominant tree level contributions . the size of the loop effects relative to the tree contributions is set , at least in the threshold region , by the parameter @xmath224 which is less than @xmath225 . in some cases it was not possible to obtain solutions up to arbitrary density because the effective potential became complex this is a well - known and unsolved problem in one loop calculations with spontaneous symmetry breaking . the fact that changes induced by the zero - point motion are generally small leads one first to conclude that no major revision is needed of earlier investigations in which such fluctuations have not been taken into account . second , the loop expansion appears to be convergent where it is well defined . the formalism presented here yielded the contribution to the grand potential arising from thermal excitations of the fluctuation modes . these effects would have to be included at finite temperatures and would be involved in the extremization conditions . such an analysis is needed for a consistent study of newborn neutron stars . extension of the present work to include the full octet of pseudoscalar fluctuations , and to include additional terms in the chiral expansion could also be contemplated . we wish to thank p. b. arnold , a. fayyazuddin , u. van kolck , and l. yaffe for helpful discussions , and d. b. kaplan , m. rho , r. venugopalan and a. wirzba for valuable comments on the manuscript . we are grateful to m. prakash for stimulating our interest in this problem and for careful reading of the manuscript . this work was supported in part by the u. s. department of energy under grant de - fg06 - 88er40427 . in this appendix , we illustrate how divergences dependent on the charge chemical potential @xmath0 are cancelled in a toy model which contains two real fields , @xmath53and @xmath54 . the model we consider is = l_+ l_ct _ = , [ gtoytree ] and @xmath226 is an arbitrary analytic function of @xmath227 with @xmath228 . it is useful to discuss this model to one loop order since it has a simpler structure than the chiral lagrangian , but the @xmath0-dependent divergences are removed in a similar way . we first derive @xmath12 , and then demonstrate that on introducing an external chemical potential @xmath0 , the effective potential ( or equivalently , the zero - point energy ) is free of divergences . finite counterterms are not relevant in the present context and will be ignored . expanding the function @xmath226 to all orders in the fields gives the the standard kinetic energy term plus an infinite set of interaction terms involving an even number of fields and two derivatives . from these interaction vertices , we may compute one - particle irreducible @xmath229-point green functions to one - loop order . the counterterm lagrangian @xmath12 is defined in such a way as to minimally cancel divergences . consider first quadratic divergences , proportional to @xmath230 ( defined in eq . ( [ quaddiv ] ) ) . the 4-point vertex contributes a quadratic divergence to the 2-point function proportional to @xmath230 and the external momenta . the corresponding counterterm lagrangian is proportional to @xmath231 $ ] . the 6-point vertex contributes to the 4-point function so as to require a counterterm lagrangian proportional to @xmath232 $ ] . continuing this sequence , we find that the contribution of quadratic divergences to the counterterm lagrangian is @xmath233 $ ] , where primes denote derivatives with respect to @xmath7 . now consider logarithmic divergences , proportional to @xmath234 ( see eq . ( [ logdiv ] ) ) . green functions containing such divergences typically contain two vertices , and evaluating them requires involved combinatorics . it is simpler instead to derive the counterterm lagrangian using the effective potential ( see e.g. ref.@xcite ) . one thus avoids combinatorics , and the method is more easily applied to lagrangians more complicated than eq . ( [ gtoytree ] ) . let _ 1 _ 1 + _ 1 , _ 2 _ 2 + _ 2 , [ phiexpand ] where @xmath53 and @xmath54 are arbitrary fields which , for the present , may vary in spacetime . expanding to quadratic order in the fields @xmath235 and @xmath236 leads to _ _ + ( cc _ 1 & _ 2 ) d ( c _ 1 + _ 2 ) , where the matrix of second order variations is given by d & = & - + l - m , + l_&= & 2 ( cc _ 1 _ _ 1 & _ 1 _ _ 2 + _ 2 _ _ 1 & _ 2 _ _ 2 ) , + m & = & - ( cc _ 1 ^ 2 + _ 2 ^ 2 & _ 1 _ 2 ( - ) + _ 1 _ 2 ( - ) & _ 2 ^ 2 + _ 1 ^ 2 ) . integrating out the fields @xmath235 and @xmath236 gives the unrenormalized effective potential to one - loop order - d & = & - i -i ( - ) + & & + _ k=1^ ( [ l - l + 2 m + ( l ) ] ) ^k , [ effpotgamma ] where , as before , the trace is taken both over spacetime and charge . the factor @xmath237 is cancelled by a corresponding contribution from the measure , and will not be considered further . in the last expression , we have symmetrized the derivative contribution and separated the mass - like contributions involving @xmath238 $ ] . we shall see that only the mass - like contributions are important for our discussion . the trace over spacetime is evaluated by inserting complete sets of four - momentum eigenstates . a momentum expansion is performed on the resulting integrals , leading to a factorization into ultraviolet divergent loop integrals and integrals over functions of the fields . to minimally cancel the divergences in the effective potential we need the following counterterm lagrangian _ ct = l_ct^quartic + l_ct^mass - like + l_ct^derivative , where the contribution from the quartic divergence ( [ quartdiv ] ) is d^4x_ct^quartic = -i ( - ) = d^4x d_4 [ toyquart ] . the contribution from mass - like terms is d^4x _ ct^mass - like & = & _ k=1 ^ 2 ( [ 2 m + ( l ) ] ) ^k + & = & d^4x - d^4x [ 2 m + ( l)]^2 + & = & - d^4x \ { ( + ) -^2 } + & & - d^4x \{^2 + & & - + & & + ( ) ^2 } . [ masslike ] for the contribution from derivatives one finds ( see also refs . @xcite ) d^4x _ ct^derivative & = & _ k=2 ^ 4 ( ( l - l ) ) ^k + & = & - d^4x \ { ( _ l_- _ l_)^2 - ( _ l_- _ l _ ) [ l _ , l _ ] + & & + l_^2 l_^2 + ( l_l_)^2 } - d^4x l_^2 . [ dervs ] this concludes the derivation of @xmath12 , to one - loop order . having derived a one - loop finite vacuum theory , we are now in a position to study the theory in the presence of a finite chemical potential @xmath0 , at finite inverse temperature @xmath239 . as before , the euclidean time is @xmath240 , and the action @xmath241 under the transformation to euclidean space . the chemical potential acts as an external vector field with spatial components zero . it enters by `` gauging '' the fields @xmath53 and @xmath54 via the transformation i i - _ 2 ; i i + _ 1 . [ imtrans ] the resulting action is given by s & = & s_+ s_ct , where [ sgammamu ] + s_&= & _ 0^dd^3x _ , s_ct = _ 0^dd^3x _ ct + l_ct & = & - d_4 - \ { ( + ) l_+ } + \ { ( l_)^2 + } , [ toylct ] + l _ & = & ( _ _ 1 ) ^2 + ( _ _ 2 ) ^2 + ( _ 1 ) ^2 + ( _ 2 ) ^2 + 2 i ( _ 2 _ _ 1 - _ 1 _ _ 2 ) - ^2 ^2 . ( [ dervs ] ) is invariant under the transformation ( [ imtrans ] ) and is therefore not included above . we have now derived the action to one - loop order , for arbitrary fields @xmath53 and @xmath54 in the presence of a chemical potential @xmath0 . we then take the final step , and evaluate the one - loop effective potential for fields @xmath53 and @xmath54 which are _ constant _ in spacetime ( see eq . ( [ decomp ] ) ) so that their derivatives vanish . expanding eq . ( [ sgammamu ] ) to quadratic order in the fields and integrating out the fluctuating field gives the action s & = & s_tree + s_loop + s_ct , + s_tree & = & - _ 0^dd^3x ^2 ^2 , + s_loop & = & - d , [ sloop ] where the matrix of second order variations about the constant fields , @xmath27 , is given by d_11 & = & ( _ n^2 + p^2 ) - 2 _ n _ 1 _ 2 - ^2 - ^2 ( _ 2 ^ 2 + 4 _ 1 ^ 2 ) - ^2 _ 1 ^ 2 + d_22 & = & ( _ n^2 + p^2 ) + 2 _ n _ 1 _ 2 - ^2 - ^2 ( _ 1 ^ 2 + 4 _ 2 ^ 2 ) - ^2 _ 2 ^ 2 + d_12 & = & 2 _ n ( + _ 1 ^ 2 ) - ^2 _ 1 _ 2 ( + 3 ) + d_21 & = & - 2 _ n ( + _ 2 ^ 2 ) - ^2 _ 1 _ 2 ( + 3 ) , and @xmath242 is the matsubara frequency . the determinant is given by eq . ( [ detd ] ) , with an additional factor of @xmath243 on the left , and with c_1 = ( + ) , c_2 = - ( 2 + 4 + ^2 ) , c_3 = - ( 2 + ) . as before , it is convenient to factorize the determinant , leading to dispersion relations for two modes given by eq . ( [ disp ] ) . ( [ sloop ] ) then separates into a zero - point and a thermal contribution s_loop = _ 0^dd^3x ( v_zp + v_thermal ) , with @xmath74 and @xmath73 given by eqs . ( [ vzp ] ) and ( [ vthermal ] ) respectively . the counterterm action obtained from eqs . ( [ toyquart ] ) , ( [ masslike ] ) and ( [ dervs ] ) after gauging and setting @xmath244 to be constant can be written s_ct = _ 0^dd^3x v_ct , where @xmath65 is given by the divergent parts of eq . ( [ ct ] ) , _ i.e. _ the terms involving @xmath234 , @xmath230 and @xmath245 . as noted in sec . iib , the @xmath0 dependent divergences in @xmath74 are exactly cancelled by those in @xmath65 . in this appendix , we give details of the cancellation of quadratic divergences dependent on the chemical potential @xmath0 for the chiral lagrangian , eq . ( [ lphi ] ) . ( the cancellation of quartic divergences needs no discussion since it is straightforward . ) we follow the procedure given in appendix a. it is first convenient to break the lagrangian down as follows : _ k & = & l_h+ l_v + l_u + l_ct + l_h & = & a ( _ _ 1)^2 + b ( _ _ 2)^2 + c _ _ 1 _ _ 2 + l_v & = & b _ 12 ( _ 1_2 ) + l_u & = & - 2 f^2 ( m_k^2+c ) ^2- u ( ) . note that in addition to having more terms than the toy model of appendix a , the kinetic term , @xmath246 , is nondiagonal here . expanding as in eq . ( [ phiexpand ] ) , the matrix @xmath27 of second order variations of the fields @xmath247 is given by d = -t + l - m ; t = ( cc a & c + c & b ) . the matrix @xmath80 can be written l^&= & l_h_1^+ l_h_2^+ l_v_1^+ l_v_2^ + l_h_1^&= & 2 ( cc _ 1 ^_1 & _ 1 ^_2 + _ 2 ^_1 & _ 2 ^_2 ) ; l_h_2^= - 2 ( cc _ 1 ^_1 & _ 1 ^_2 + _ 2 ^_1 & _ 2 ^_2 ) + l_v_1^&= & ^0 2 b _ 12 ( cc 0 & 1 + -1 & 0 ) ; l_v_2^= ^0 2 b ( cc - _ 1 _ 2 & _ 1 ^ 2 + - _ 2 ^ 2 & _ 1 _ 2 ) . [ lterms ] the matrix @xmath248 takes the form m & = & m_h_1 + m_h_2 + m_v_1 + m_v_2 + m_u + m_h_1 & = & - ( cc _ 1 ^ 2 + _ 2 ^ 2 & _ 1 _ 2 ( - ) + _ 1 _ 2 ( - ) & _ 2 ^ 2 + _ 1 ^ 2 ) + m_h_2 & = & ( cc _ 1 _ 1 & _ 1 _ 2 + _ 2 _ 1 & _ 2 _ 2 ) + m_v_1 & = & 2 b ( cc - _ 1 ^0 _ 2 & - _ 2 ^0 _ 2 + _ 1 ^0 _ 1 & _ 2 ^0 _ 1 ) + m_v_2 & = & - b ( _ 1_2 ) ( cc _ 1 ^ 2 _ 12 + _ 12 & _ 1 _ 2 ( _ 12 - _ 12 ) + _ 1 _ 2 ( _ 12 - _ 12 ) & _ 2 ^ 2 _ 12 + _ 12 ) + m_u & = & ( cc _ 1 ^ 2 u + _ 2 ^ 2 & _ 1 _ 2 ( u - ) + _ 1 _ 2 ( u - ) & _ 2 ^ 2 u + _ 1 ^ 2 ) . [ mterms ] as in appendix a primes denote derivatives with respect to @xmath7 . it may be checked that by gauging @xmath27 according to eq . ( [ imtrans ] ) and then taking the fields @xmath244 to be constant , one recovers eq . ( [ dmatrix ] ) . it is now straightforward , but tedious , to compute the the quadratically divergent contributions to the counterterm lagrangian by evaluating the terms of the form ( t^-1 m ) - ( t^-1 l^)^2 . omitting terms that are invariant under the transformation ( [ imtrans ] ) , we find the following contributions to @xmath249 : ( t^-1 m_h_1 ) & = & - ( + ) [ mh1 ] + ( t^-1 m_h_2 ) & = & ( _ 1 _ 1 + _ 2 _ 2 ) [ mh2 ] + ( t^-1 m_v_1 ) & = & - b ( _ 1_2 ) [ mv1 ] + ( t^-1 m_v_2 ) & = & - b ( _ 12 + ) ( _ 1_2 ) [ mv2 ] + ( t^-1 m_u ) & = & ( + u ) [ mu ] + - ( t^-1 l_v^)^2 & = & b^2 ( _ 12 + _ 12 ) [ lv ] + - ( t^-1 [ l_h_1^+ l_h_2^ ] ) ^2 & = & - ( - ) ^2 ( _ 1 ^_1 + _ 2 ^_2 ) ^2 [ lhsq ] + - ( t^-1 l_v^t^-1 [ l_h_1^+ l_h_2^ ] ) & = & b ( - ) ( _ 1_2 ) [ lvlh ] . following appendix a , we use the transformation ( [ imtrans ] ) on @xmath12 and find that the quadratically divergent contribution to @xmath65 is v_ct^quad & = & d_2 \ { ^2 ^2 ( + ) + ^2 ( 1 - ) + b _ 12 + b ^2 ( + ) + & & + b ( 1 - - ) - b^2 ( _ 12 + _ 12 ) - ( u + ) } . it may be checked that this agrees with the divergent parts of eq . ( [ ct ] ) using the explicit form of the coefficients @xmath250 , @xmath251 and @xmath196 : c_1 & = & 2 ^2 ( + ) + 2 b ( 2 _ 12 + + _ 12 ) + 2 b^2 ( _ 12 + _ 12 ) + c_2 & = & - ^2 ( 2 + 4 + ^2 ) - b ( 2 _ 12 + 4 _ 12 + ^2_12 ) + u + c_3 & = & - ^2 ( 2 + ) - b ( 2_12 + _ 12 ) + . we now discuss the chiral ordering of the above counterterms in the lagrangian ( [ pje0 ] ) . to do so , it helpful to evaluate the divergent integrals of eqs . ( [ quartdiv ] ) , ( [ quaddiv ] ) and ( [ logdiv ] ) using dimensional regularization . this gives d_4 & = & = - ( + + ( 3 ) ) + d_2 & = & = - ( + + ( 2 ) ) + d_0 & = & = ( + + ( 1 ) ) . [ dimregds ] here , we have included @xmath252 explicitly in the kaon propagator . a factor of @xmath253 has been subsumed in @xmath93 which is introduced so that the dimension of the integral is independent of the dimensionality of spacetime . the parameter @xmath254 , where @xmath182 is the dimensionality of the integral , and @xmath255 is the digamma function @xcite . in minimal subtraction the integrals of eq . ( [ dimregds ] ) are regulated by removing the @xmath256 terms . it is then evident that in the chiral expansion @xmath245 , @xmath230 and @xmath234 are @xmath257 , @xmath2 and @xmath258 respectively . to evaluate the chiral order of counterterms , we promote the nucleon fields in @xmath85 to dynamical fields . we may then determine the chiral order of counterterms from weinberg s counting rule ( [ count ] ) , either directly from the tree - level counterterms or from the representation of these terms as one - loop diagrams . counterterms ( [ mv1 ] ) , ( [ mv2 ] ) and ( [ lvlh ] ) contribute to @xmath259 . expanding to lowest order in the fields @xmath53 and @xmath54 , these terms renormalize the kaon - nucleon vector interaction coupling constant . counterterms ( [ mh1 ] ) , ( [ mh2 ] ) and ( [ lhsq ] ) contribute to @xmath260 . to lowest order in the fields , they contribute to wave function renormalization . ( [ mu ] ) is also @xmath257 , but to lowest order in the fields gives an overall irrelevant constant and a mass renormalization . ( [ lv ] ) contributes to @xmath261 and yields a four - nucleon interaction to lowest order in the fields . in this appendix , we have not discussed logarithmic divergences . we believe the cancellation of @xmath0 dependent logarithmic divergences will proceed as in the toy model of appendix a. however we have not derived the full counterterm lagrangian proportional to @xmath234 since this , while in principle straightforward , would involve prohibitively complicated algebra . the counterterms will have a chiral order which ranges from 0 to 4 as can be seen from eqs . ( [ masslike ] ) and ( [ dervs ] ) in conjunction with eqs . ( [ lterms ] ) and ( [ mterms ] ) . in conclusion we note that although we have referred to chiral counting throughout this manuscript , our lagrangian should be regarded as a `` u(1 ) '' or `` o(2 ) '' analogue of the chiral expansion . in this appendix , we derive the critical density from the kaon 2-point function to one - loop order and show that it is equal to that obtained from expanding the zero point energy . close to threshold , the grand potential may be expanded in terms of the condensate as follows = - d^-1 ( ) f^2 ^ 2 + o(^4 ) [ threshold ] where @xmath27 is the zero - momentum kaon propagator , relevant for _ s_-wave condensation , and @xmath0 is the lepton chemical potential prior to kaon condensation . the threshold for kaon condensation is therefore given by @xmath262 , _ i.e. _ the density at which the energy of negatively charged kaons equals the lepton chemical potential . from the chiral lagrangian , eq . ( [ lkaon ] ) , we may read off the interactions . the 2-point interactions are _ u2 = - ( m_k^2+c ) k^+k^- , l_v2 = -i b k^+k^- . [ tps ] here we treat the standard mass term as an interaction and take the kaon propagator to be massless . the 4-point interactions are @xcite _ h4 & = & ( k^+k^- ) ^2 , l_v4 = - ( k^+k^- ) k^+k^- + l_u4 & = & ( k^+k^- ) ^2 . we compute the kaon 2-point function with external four - momentum @xmath263 . the interactions ( [ tps ] ) contribute at tree - level . to one - loop order , we have the diagrams of fig . 3 . the contributions of figs . 3(a ) , 3(b ) and 3(c ) are @xmath264 , @xmath265 and @xmath266 , respectively . in the last two diagrams , the internal lines contain the interactions given in eq . ( [ tps ] ) , which are denoted by @xmath252 and @xmath85 , respectively . 3(d ) contributes @xmath267 , and fig . 3(e ) contributes @xmath268 . in addition , we must include all possible insertions of the interactions ( [ tps ] ) on internal lines . including the counterterm contribution , the result is d^-1 ( ) & = & ^2 + 2 b - m_k^2 -c + ( ^2 - 4b - m_k^2-c ) + & = & - ( - 2 ( d_2+f_2 ) + 2 ^2 ( d_0+f_0 ) ) [ dinverse ] eq . ( [ dinverse ] ) agrees with the expansion of @xmath165 given in eq . ( [ uc ] ) , as required , and in vacuum we obtain the free inverse propagator @xmath269 . note that figs . 3(a ) , ( c ) , and the contribution to ( d ) proportional to @xmath270 , do not involve nucleon fields and are mass and wavefunction renormalization diagrams in the vacuum . these diagrams are @xmath257 . if we promote the nucleon fields to dynamical ones , we generate nucleon legs on figs . 3(b ) and ( e ) , corresponding to figs . 1(g ) and 1(f ) respectively , and find that they are @xmath4 . these diagrams correspond to figs . 1(a ) and 1(b ) of ref.@xcite respectively ; since only nucleon and kaon fields are used in the present calculation , the remaining diagrams of ref.@xcite do not enter . ellis , r. knorren and m. prakash , phys . b349 * , 11 ( 1995 ) ; r. knorren , m. prakash and p.j . ellis , phys . rev . * c52 * , 3470 ( 1995 ) ; j. schaffner and i.n . mishustin , phys . rev . * c53 * , 1416 ( 1996 ) ; j. schaffner , j. bondorf and i.n . mishustin , strangeness 96 ( budapest , hungary , 1996 ) heavy ion physics , to be pub . , nucl - th/9607019 . m. prakash , i. bombaci , m. prakash , p. j. ellis , j. m. lattimer and r. knorren , nucl - th/9603042 , phys . ( 1996 ) in press ; p. j. ellis , j. m. lattimer and m. prakash , comments on nucl . and part . * 22 * , 63 ( 1996 ) ; m. prakash , s. reddy , j.m . lattimer and p.j . ellis , strangeness 96 ( budapest , hungary , 1996 ) heavy ion physics , to be pub . , nucl - th/9607031 . . values of the critical density for kaon condensation , @xmath162 , in units of equilibrium nuclear matter density , for various choices of the parameter @xmath43 . superscript @xmath271 denotes the tree level contribution , and @xmath272 denotes the loop contribution . the scalar kaon - nucleon interaction @xmath40 is defined in eq . ( 5 ) , while @xmath187 is the corresponding quantity with the density @xmath182 replaced by the scalar density @xmath181 given in eq.(50 ) [ cols="<,>,>,^ , > , > " , ]	we investigate the contribution of zero - point motion , arising from fluctuations in kaon modes , to the ground state properties of neutron star matter containing a bose condensate of kaons . the zero - point energy is derived via the thermodynamic partition function , by integrating out fluctuations for an arbitrary value of the condensate field . it is shown that the vacuum counterterms of the chiral lagrangian ensure the cancellation of divergences dependent on @xmath0 , the charge chemical potential , which may be regarded as an external vector potential . the total grand potential , consisting of the tree - level potential , the zero - point contribution , and the counterterm potential , is extremized to yield a locally charge neutral , beta - equilibrated and minimum energy ground state . in some regions of parameter space we encounter the well - known problem of a complex effective potential . where the potential is real and solutions can be obtained , the contributions from fluctuations are found to be small in comparison with tree - level contributions . 2 _
the rapidly expanding field of network analysis , reviewed in @xcite , has provided examples of networks exhibiting accelerating " network growth , where total link number grows faster than linearly with network size @xcite . for instance , the internet and world wide web appear to grow by adding links more quickly than sites @xcite though the relative change over time is small @xcite . similarly , both the number of links per substrate in the metabolic networks of organisms @xcite and the average number of links per scientist in collaboration networks increases linearly over time @xcite , while also languages appear to evolve via accelerated growth @xcite . these studies have motivated examinations of stochastic accelerating networks where the number of new links added is probabilistic ( though integral and greater than one ) causing network transitions from scale - free to exponential statistics @xcite . these previous studies of deterministic or stochastic networks have typically considered networks where every node has at least one connection which constrains the rate of acceleration that can be considered . in particular , if each new node added to a network of @xmath0 nodes is accompanied by @xmath1 new links with acceleration parameter @xmath2 , then ensuring that the network is less than fully connected constrains acceleration parameters to the range @xmath3 @xcite . equivalent limits were considered in ref . however , real networks can contain a majority of nodes that are entirely unconnected many computers are unconnected or only intermittently connected to the internet and many people are unconnected nodes in social or sexual contacts networks . modelling networks where a substantial proportion of the nodes are unconnected requires a probabilistic approach in which the number of links added with each new node scales as @xmath4 for some probability parameter @xmath5 . the introduction of a probabilistic envelope lifts constraints on the acceleration parameter allowing networks displaying deceleration @xmath6 , no acceleration @xmath7 , acceleration @xmath3 , and hyper - acceleration @xmath8 . hyper - accelerating probabilistic networks with @xmath9 and @xmath8 are not merely of theoretical interest , and appear for instance in prokaryote gene regulatory networks where many gene nodes are unregulated and merely constitutively or stochastically expressed @xcite . prokaryote gene regulatory networks are sparsely connected ( @xmath10 ) and display hyper - acceleration @xmath11 as established by independent comparative genomics analyses @xcite . this is likely due to their reliance on sequence homology interactions between protein transcription factors and specific promoter binding sequences sites @xcite . in these regulatory networks , outbound regulatory links are preferentially attached to existing regulator nodes as gene duplication events contribute to up to 75% of each of new genes @xcite and regulatory transcription factors @xcite . in contrast , inbound links to regulated nodes are randomly formed due to the random drift of gene promotor sequences , although their subsequent fixation is determined by selection . the high acceleration parameter @xmath11 ensures that these prokaryote gene regulatory are size and complexity constrained at predicted limits which closely match genomic size limits observed in the evolutionary record @xcite . accelerating networks are more prevalent and important in society and in biology than is commonly realized . in fact , any network in which the dynamical state of any node depends on the immediately preceding state of ( ideally ) every other node is accelerating and will perhaps display either structural transitions from randomly connected , to scale free statistics , to densely connected and perhaps finally to fully connected statistics , or explicit size and complexity constraints @xcite . this paper generalizes previous analysis to consider probabilistic accelerating networks in decelerating regimes with @xmath6 , accelerating regimes with @xmath3 , and hyper - accelerating regimes where @xmath12 . we will show that hyper - accelerating networks under preferential attachment of new connections to existing nodes can display transitions in their connectivity statistics from scale - free to exponential statistics dependent on the acceleration parameter @xmath13 , the probability parameter @xmath14 , and network size @xmath0 . these transitions occur at the point where the accelerating growth in connection number can no longer be sustained introducing exponential cut - offs in the connectivity distributions . these cutoffs can be generated by different mechanisms and appear in growing networks subject to strong aging effects @xcite , in finite sized networks @xcite , and in information filtered networks @xcite . to complete our examination of network transitions in accelerating , hyper - accelerating and decelerating probabilistic networks , we first define our growing network models using undirected links in section [ sect_prob_accelerating_networks ] . this definition allows an immediate rough quantification of the probability of an accelerating network forming a single giant connected component or of undergoing a transition to a fully connected regime with network growth . in section [ sect_pref_attachment ] we examine probabilistic networks growing through the preferential attachment of new links to existing connected network nodes , while in section [ sect_random ] we examine growing probabilistic networks featuring the random attachment of new links to all existing nodes . we consider networks growing through the sequential addition of numbered nodes @xmath15 for @xmath16 where at network size @xmath17 , node @xmath18 ( @xmath19 ) has @xmath20 undirected links . the addition of node @xmath15 and its @xmath21 links will increase the probable number of links attached to existing nodes @xmath18 for @xmath22 so @xmath23 . we assume new links are added only with each arriving node ( so no new links are added between established nodes ) , and that the average number of new links attached to node @xmath15 on its entry to the network is a function of network size @xmath24 where the probability constant @xmath14 satisfies @xmath5 . this average is then a decreasing function of network size for @xmath6 , constant for @xmath7 , and an increasing function of network size for @xmath25 . the total expected number of links @xmath26 in a probabilistic network of large size @xmath0 is then @xmath27 consequently , the expected number of connections is @xmath28 , and the average connectivity per node is @xmath29 . in later numerical simulations , we compare the statistics of networks of different sizes which is possible provided they possess the same average connectivity . we achieve this for accelerating networks with @xmath2 by choosing different probability parameters @xmath14 for each different sized network according to @xmath30 this allows us to later consider different networks with identical average connectivity @xmath31 , where networks have either a critical size of @xmath32 nodes connected by @xmath33 links with @xmath34 , or a critical size of @xmath35 nodes connected by @xmath36 links with @xmath37 . because link formation is probabilistic , the total number of links attached to node @xmath15 on entry to the network , denoted @xmath38 say , ranges between @xmath39 and @xmath17 , so @xmath40 with average @xmath41 . as each link either forms or does not form , we model the link formation process using a binomial distribution . for the binomial distribution , the average @xmath42 equates to the product of the maximum number of links @xmath17 and the link formation probability . conversely , the size dependent link formation probability at network size @xmath17 equates to @xmath43 . consequently , the probability distribution that node @xmath15 forms exactly @xmath38 links on entry to the network is @xmath44^{k - j}.\ ] ] that is , when @xmath7 the latest node @xmath15 can potentially form links with every one of the @xmath17 existing nodes ( counting itself ) with a network size dependent probability @xmath45 to give an average number of new links of @xmath14 connections with each new node . when @xmath11 , the latest node @xmath15 can potentially form links with every other node with constant probability @xmath14 to give an average number of new links of @xmath46 with each new node , and so on . the connection probability is constrained to be less than unity , @xmath47 , or equivalently , the average number of links added to node @xmath15 is required to be less than @xmath17 , @xmath48 . for constant @xmath14 , hyper - accelerating networks with @xmath12 will eventually violate these constraints with network growth which effectively imposes size and structural constraints on these networks . there are two types of network connectivity transitions which have been of interest in the literature . the first occurs when a growing network changes from sparse to dense connectivity as this change necessarily implies that the final connectivity statistics will become exponential in nature . that is , while a sparsely connected network can possess either scale - free or exponential statistics , a densely connected network must exhibit exponential statistics ( as multiple connections are not allowed between any two given nodes ) . this transition occurs for undirected links when approximately @xmath49 . the second transition of interest occurs when the connected nodes coalesce to form a single giant island of interconnected nodes . to borrow results from random graph theory @xcite ( which have been roughly validated for accelerating biological networks @xcite ) , the giant island is expected to form when approximately @xmath50 . we now investigate the dependence of these transitions on the acceleration parameter @xmath13 . it is evident that decelerating networks with @xmath6 are very sparsely connected so neither of the dense connectivity or giant island transitions will occur . hereinafter , we only consider accelerating networks with @xmath2 . for such networks , the dense connectivity and giant island transitions will occur respectively at the points @xmath51 this establishes that , for instance , non - accelerating networks with @xmath7 can never undergo a transition to a densely connected regime for any choice of the probability parameter @xmath14 for large network sizes @xmath0 , though these networks can form giant islands with network growth as long as @xmath52 . this inability to become densely connected remains for all accelerating regimes @xmath53 , but dense connectivity can emerge in the hyper - acceleration regime . at this point with @xmath11 , growing networks can undergo a transition to the densely connected regime when the probability parameter is large , @xmath54 . for yet higher acceleration rates @xmath12 , all growing networks undergo a transition to the dense connectivity regime and display single giant islands , and these transitions occur at network sizes determined by the probability parameter @xmath14 . in the following sections , we determine the connectivity statistics for accelerating and hyper - accelerating networks under both preferential and random attachment of new links to established nodes . we now turn to consider accelerating and hyper - accelerating networks ( @xmath2 ) where new links are principally preferentially attached to existing connected nodes . the final generated link distribution will depend on the balance of different countervailing trends , namely the respective weights given to preferential versus random attachment , and the magnitude of the acceleration parameter which influences both the rate at which links are attached to newer nodes and the rate at which while older nodes accumulate links . we now examine these trends in detail . on entry into the network , node @xmath15 establishes an average of @xmath55 links with existing nodes @xmath56 . we suppose that these links either are preferentially attached to node @xmath57 with probability proportional to @xmath57 s current connectivity @xmath58 , or are randomly attached to @xmath57 with probability proportional to @xmath59 , a random connection parameter . using the continuous approximation @xcite , the rate of growth in link number for node @xmath57 is @xmath60 here , the rate of growth in the connectivity @xmath58 is determined by the number of new links added with node @xmath15 ( i.e. @xmath21 ) which can be either preferentially attached to the existing nodes @xmath57 with probability proportional to that nodes current connectivity ( i.e. @xmath58 ) , or randomly attached if @xmath61 allowing even initially unconnected nodes to establish connections . ( the case @xmath62 ensures that a node that receives zero links on entry to the network remains unconnected for all time . ) the denominator of eq . [ eq_continuum_links ] is a probability weighting to ensure normalization and is roughly equal to the total number of connections ( @xmath63 ) for all nodes at network size @xmath17 . following @xcite , we can evaluate the denominator using the identity @xmath64 which can be evaluated using eq . [ eq_continuum_links ] to give @xmath65 noting @xmath66 , this can be integrated to determine the denominator of eq . [ eq_continuum_links ] to be @xmath67 the first term on the rhs is the total number of connections at network size @xmath17 and is in agreement with eq . [ eq_net_parameters ] . substituting this relation back into eq . [ eq_continuum_links ] gives @xmath68 the connectivity statistics can readily be obtained by integrating eq . [ eq_partial_ljk ] with the initial conditions @xmath69 at time @xmath38 to obtain @xmath70 differentiation of these curves with respect to network size @xmath0 satisfies eq . [ eq_continuum_links ] , while integrating these link numbers over all node numbers @xmath38 ( numerically for @xmath25 ) gives the required total number of links as in eq . [ eq_net_parameters ] . further , differentiating the linkage distribution @xmath71 with respect to the age index @xmath38 establishes that average node connectivity is monotonically decreasing with increasing node age @xmath38 for @xmath72 , flat for @xmath11 , and monotonically increasing for @xmath12 . in other words , irrespective of the choice of the @xmath59 parameter , slowly accelerating networks with @xmath72 have the majority of their connections in the oldest portions of their networks , while hyper - accelerating networks with @xmath12 place the majority of the connections in the youngest portions of their networks . the transition point occurs at @xmath11 when connectivity is uniformly distributed over the network irrespective of node age . the further analysis of this age dependent connectivity distribution is heavily dependent on the setting of the random choice parameter @xmath59 . self - evidently , setting @xmath59 small ensures that preferential attachment processes dominate while the choice of large @xmath59 ensures that random attachment dominates . appendix [ app_pref_random_attach ] uses a taylor series expansion to demonstrate that preferential attachment dominates hyper - accelerating networks only when @xmath59 is so small as to be essentially zero , and hereinafter , we analyze preferential attachment using the setting @xmath62 . when at network size @xmath73 . ( a ) the average connectivity distribution @xmath71 as a function of node age @xmath38 for various values of the acceleration parameter @xmath13 . connectivity is monotonically decreasing for @xmath3 , flat for @xmath11 , and monotonically increasing for @xmath12 . ( b ) example simulated networks for various acceleration parameters @xmath13 with nodes listed sequentially left to right from @xmath74 to @xmath75 with the connectivity of each node indicated by vertical lines . the horizontal dashed line indicates an average connectivity of two.__,title="fig : " ] when at network size @xmath73 . ( a ) the average connectivity distribution @xmath71 as a function of node age @xmath38 for various values of the acceleration parameter @xmath13 . connectivity is monotonically decreasing for @xmath3 , flat for @xmath11 , and monotonically increasing for @xmath12 . ( b ) example simulated networks for various acceleration parameters @xmath13 with nodes listed sequentially left to right from @xmath74 to @xmath75 with the connectivity of each node indicated by vertical lines . the horizontal dashed line indicates an average connectivity of two.__,title="fig : " ] for convenience , we note that the choice @xmath62 sets the connectivity versus age distribution to be @xmath76 for arbitrary @xmath13 . the choice @xmath7 duplicates existing results found for growing networks which add a constant number of links with each new node subject to preferential attachment @xcite , while additionally , choosing @xmath77 duplicates the deterministic results of ref . @xcite in the regimes @xmath3 considered in that paper . this purely preferentially attached distribution is shown in fig . [ f_ljn_dist ] along with example simulation runs at different acceleration parameters @xmath13 . ( the floating free end of the first link to the first connected node is randomly attached to any of the existing nodes as they all possess zero links reducing preferential attachment to random attachment . ) part ( a ) of this figure shows the average connectivity @xmath71 of node @xmath57 which is monotonically decreasing with respect to node age for @xmath3 and monotonically increasing when @xmath12 . these different trends depend on whether the accumulative effects of preferential attachment outweighs the accelerating number of added links or not . the setting @xmath11 ensures the average number of links per node is independent of node number as preferential attachment exactly cancels the bias in initial link number towards later nodes . part ( b ) of this figure shows the actual connectivity of node @xmath57 ( rather than the average connectivity ) in example simulated networks , making it evident that actual node connectivity is generally a monotonically decreasing function of node age . why this is so is discussed later . the @xmath71 distribution contains information about both node connectivity and node age and so approximates network statistics ( simulated or observed ) when all of this information is available . however , it is usually the case that node age information is unavailable necessitating calculation of connectivity distributions that are not conditioned on node age . this effectively requires binning together all nodes irrespective of their age to obtain a final link distribution . following ref . @xcite , the usual continuum approach @xcite must be modified when non - monotonically decreasing connectivity statistics are present . in particular , the final connectivity distribution is obtained via @xmath78,\end{aligned}\ ] ] where @xmath79 is the solution of the equation @xmath80 . the top line is used when all nodes possess the same average link number while the second line is applicable with the plus ( negative ) sign when the average numbers of links per node is monotonically increasing ( decreasing ) with node number . non - monotonic cases require alternate approaches . we now use eq . [ eq_links_l_jn ] to calculate the age independent final link distribution @xmath81 relevant when age information is unavailable . in the case @xmath11 we have @xmath82 , meaning the connectivity @xmath17 is independent of the node age @xmath38 , so the delta - function of eq . [ eq_final_link_dist ] immediately integrates to give @xmath83 . for other cases , we have the constraint @xmath84 the age constraint @xmath85 translates into the connectivity constraints @xmath86 $ ] for @xmath3 , and @xmath87 $ ] for @xmath12 . via eq . [ eq_final_link_dist ] , the final continuous connection distribution under preferential attachment is then @xmath88 each of these separate distributions is normalized over @xmath89 $ ] via @xmath90 where @xmath91 for @xmath3 , @xmath92 $ ] for @xmath11 , and @xmath93 $ ] for @xmath12 . the average connectivity per node @xmath94 as required over these same ranges . again , the choice @xmath77 and @xmath3 duplicates the deterministic results of ref . @xcite . as a function of acceleration parameter @xmath13 . the log plot means that straight lines indicate power - law distributions . the connectivity distributions are monotonically decreasing for @xmath3 over the range @xmath91 , a delta function for @xmath11 , and over the range @xmath93 $ ] , are monotonically increasing for @xmath95 , flat for @xmath96 and monotonically decreasing for @xmath97 . networks in this figure possess an average connectivity per node of @xmath98 at a network size of @xmath99 . _ ] the resulting age - independent connectivity distributions @xmath81 are shown in fig . [ f_ckn_alpha ] and are monotonically decreasing for acceleration parameters @xmath3 , monotonically increasing for @xmath95 , uniformly flat for @xmath96 , and monotonically decreasing for @xmath97 . in all cases except @xmath11 , the age independent distribution @xmath81 is a power law proportional to @xmath100 with @xmath101 . the exponent here can take the values @xmath102 for @xmath103 and @xmath104 for @xmath105 , so no value of the acceleration parameter permits exponents in the range @xmath106 $ ] . the continuous connectivity distributions @xmath81 are typically defined over very limited ranges @xmath107 $ ] so it can often be the case that every node of the network is predicted to have fractional connectivity less than unity . of course , it is impossible to observe nodes with fractional connectivities every node either has zero connections or one connection , or two connections , and so on . when most connected nodes have fractional connectivity , it is necessary to convert the unobservable continuous distribution @xmath81 to an observable discrete distribution , which we denote @xmath108 , giving the proportion of network nodes possessing an integral number of @xmath17 links for @xmath92 $ ] . for networks with connectivity distributions which are long tailed , the continuous distribution @xmath81 closely approximates the discrete distribution @xmath108 and this step is often not required . in appendix [ app_exp_cutoff ] , we detail how to calculate the observed integral connectivity distribution @xmath108 generated from a continuous power - law connectivity distribution @xmath81 defined over a finite range @xmath107 $ ] . it is generally the case that finite ranges impose exponential cutoffs on scale - free continuous distributions @xmath81 to give discrete distribution @xmath108 possessing exponential statistics . a number of alternative mechanisms can impose an exponential cutoff on a power - law connectivity distribution @xcite , and we now add hyper - acceleration to this list . as a function of acceleration parameter @xmath13 under preferential attachment . it is evident that while the unobservable continuous distributions @xmath81 appear to be distinguishable , the observable discrete distributions are virtually indistinguishable at low connectivity numbers . _ ] appendix [ app_exp_cutoff ] specifies a numerical algorithm for calculating the observable discrete distribution @xmath108 defined by the unobservable continuous distribution @xmath81 . the calculated discrete distributions equivalent to the continuous distributions of fig . [ f_ckn_alpha ] appear in fig . [ f_ck_discrete_pref ] demonstrating that the generated discrete distributions are virtually indistinguishable over wide ranges of the acceleration parameter in marked contrast to the very different shapes of the continuous distributions @xmath81 at different acceleration parameters . this is due to the exponential cutoffs imposed when the continuous distributions are defined only over finite ranges . all of the calculations involved in the numerical algorithm of appendix [ app_exp_cutoff ] can in principle be done by hand as we now illustrate using distributions @xmath81 where the boundary point @xmath109 lies within the range of support @xmath110 for low acceleration regimes @xmath3 , but outside the range of support @xmath111 for high acceleration regimes @xmath112 . using appendix [ app_exp_cutoff ] , this choice gives @xmath113^{\frac{2}{1-\alpha}}- \left[\frac{pn^\alpha}{k+1/2}\right]^{\frac{2}{1-\alpha } } , & 0<\alpha<1 , k\in[2,\infty ) , \\ & \\ 0 , & \alpha=1 , k\in[0,n ] , \\ & \\ 0 , & \alpha>1 . \\ \end{array } \right.\ ] ] all of these distributions then have nonzero support to the left of the boundary @xmath109 and so can be partitioned to give @xmath114 nodes labelled @xmath115 each with average connectivity @xmath116 equal to @xmath117 these values then feed directly into eq . [ eq_total_c_k ] to give the total discrete distribution @xmath108 . to further illustrate this last stage of the calculation , consider the case @xmath11 where we have @xmath118 and @xmath119 for all nodes @xmath120 , and noting the absence of any contribution from the right of @xmath109 so @xmath121 , we have @xmath122 for @xmath92 $ ] . as mentioned above , [ f_ljn_dist ] makes it apparent that the actual connectivity of connected nodes generally decreases with node age even though average node connectivity can be monotonically increasing or decreasing . the reason is that the @xmath71 curves average connectivity over all nodes rather than only over connected nodes . ( recall that @xmath62 means that nodes that are unconnected on entry to the network forever remain unconnected . ) the probability that node @xmath57 is connected is straightforwardly @xmath123 using eq . [ eq_p_j_k_dist ] or @xmath124 hence , the average link number per node at node @xmath57 ( eq . [ eq_links_l_jn ] ) equates to the product of the average number of links per connected node at node @xmath57 , denoted @xmath125 , and the probability that node @xmath57 is connected , or @xmath126 . by definition then , we have @xmath127 giving @xmath128 consequently , the average number of connections of connected nodes @xmath125 is always a monotonically decreasing function of node age @xmath38 . ( an equivalent derivation appeared in ref . @xcite . ) it is difficult to further analyze the connectivity distribution for actually connected nodes as the distribution of eq . [ eq_actual_connectivity ] is only incompletely sampled node @xmath57 is connected only with probability @xmath129 . [ f_k_constant ] shows age independent connectivity statistics for simulated networks at a snapshot size of @xmath99 nodes and with different acceleration and probability parameters @xmath130 chosen to generate networks over a range of average connectivities @xmath131 . to do this , we choose different acceleration parameters in the set @xmath132 with probability parameters set according to eq . [ eq_bar_p ] . the resulting slopes of the power law connectivity distributions @xmath100 averaged over the corresponding acceleration parameters varies over the range @xmath133 . this regime was forbidden prior to taking account of the exponential cutoffs imposed by the finite ranges . ( the choice @xmath134 generating @xmath135 is roughly equivalent to the accelerating simulations of prokaryote gene regulatory networks @xcite . ) additional network growth would see these scale - free distributions undergoing a full transition to exponential statistics . for connected nodes under preferential attachment for various values of the average network connectivity @xmath136 at a snapshot network size of @xmath99 . the curves naturally group together according to average network connectivity , and are characterized by a power law @xmath100 with slope @xmath137 as shown . further network growth would eventually convert these scale - free distributions to exponential distributions . for average connectivity @xmath138 , we show curves for @xmath139 ; for @xmath140 , @xmath141 ; for @xmath98 , @xmath142 ; for @xmath143 , @xmath144 . _ ] finally , we briefly consider the size of the largest connected island in the artificial case of purely preferentially attached networks where initially unconnected nodes remain always unconnected . in this particular case , preferential attachment ensures that every connected node belongs to the same interconnected largest island whose size then equates to the number of connected nodes . as a node is connected only if it gains some connections on entry to the network with probability @xmath145 from eq . [ eq_p_j_k_dist ] , the size @xmath146 of the largest island is @xmath147.\ ] ] for small @xmath14 and small network sizes @xmath0 , this equates to the number of links @xmath148 as each connected node likely has only one link in sparsely connected networks . in this regime , the largest island grows at an acceleration rate determined by the acceleration parameter @xmath13 . however , outside this regime , the growth of the largest island eventually saturates when almost every new node is connected ( i.e. when @xmath149 ) meaning the largest island grows proportionally to network size @xmath0 . this occurs in the limit @xmath150 or @xmath0 large giving @xmath151 . should a network absolutely need to maintain an accelerated growth in the size of its largest island , as conjectured for the regulatory gene network of prokaryotes due to competitive pressures , then the point at which the largest island makes a transition from accelerated to linear growth will represent an upper network size limit @xcite . we now turn to consider accelerating and hyper - accelerating networks ( @xmath2 ) where new links are randomly distributed over all existing nodes . the final generated link distribution will depend on the balance of two countervailing trends . the first sees newer nodes attracting more links due to acceleration which confers an increasing average number of links on younger nodes @xmath15 , while in contrast , older nodes have a longer time to acquire links from the newer nodes . under random attachment , the rate of growth of the link number for node @xmath57 when the network has grown to size @xmath17 is given under the continuous approximation @xcite as @xmath152 where @xmath55 . here , the rate of growth in the connectivity @xmath58 is determined by the number of new links added with node @xmath15 ( @xmath21 ) which are equally proportioned over the @xmath17 existing nodes . the resulting connectivity statistics can readily be obtained through integrating this equation with the initial conditions @xmath69 at time @xmath38 to obtain @xmath153 , & \alpha=0 , \\ & \\ \frac{p}{\alpha } \left [ n^\alpha + ( \alpha-1 ) j^\alpha \right ] , & \alpha>0 . \\ \end{array } \right.\ ] ] these results satisfy eq . [ eq_continuum_links_rand ] , while integration of the link numbers over all node numbers @xmath38 gives the required total number of links as in eq . [ eq_net_parameters ] . ( as established in appendix [ app_pref_random_attach ] , these random attachment equations are exactly reproduced by applying a taylor series expansion of the general equations eq . [ eq_links_l_jn_beta ] about the point @xmath154 with the retention only of terms linear in @xmath155 . ) in passing , we present the exact distribution for links under random attachment . first , note that node @xmath15 initially receives an average of @xmath156 links where @xmath157 $ ] with each link formed with probability @xmath158 , and subsequently receives an average of @xmath159 links from itself where @xmath160 $ ] with each link formed with probability @xmath158 . the arrival of node @xmath161 is accompanied by an additional @xmath162 new links of which an average @xmath163 attach to node @xmath15 where @xmath164 $ ] with each link formed with probability @xmath165 , and so on until it receives an average of @xmath166 links from node @xmath75 with @xmath167 $ ] with each link formed with probability @xmath168 . the joint probability that node @xmath15 obtains @xmath169 connections is then @xmath170^j \left[1-pk^{\alpha-1}\right]^{k - j } \times \nonumber \\ & & \hspace{-2 cm } \times \prod_{n = k}^n { n \choose j_n } \left[pn^{\alpha-2}\right]^{j_n } \left[1-pn^{\alpha-2 } \right]^{n - j_n}. \nonumber\end{aligned}\ ] ] the average number of inbound links for node @xmath15 is then @xmath171 which integrates to give eq . [ eq_links_l_jn_rand ] as found by the continuum approach . unfortunately , further work with this more exact multivariate distribution is generally intractable . under random attachment as a function of node age @xmath38 for various values of the acceleration parameter @xmath13 . connectivity is monotonically decreasing for @xmath3 , flat for @xmath11 , and monotonically increasing for @xmath12 . ( b ) example simulated networks for various acceleration parameters @xmath13 with nodes listed sequentially left to right from @xmath74 to @xmath75 with the connectivity of each node indicated by vertical lines . the horizontal dashed line indicates an average connectivity of two . _ , title="fig : " ] under random attachment as a function of node age @xmath38 for various values of the acceleration parameter @xmath13 . connectivity is monotonically decreasing for @xmath3 , flat for @xmath11 , and monotonically increasing for @xmath12 . ( b ) example simulated networks for various acceleration parameters @xmath13 with nodes listed sequentially left to right from @xmath74 to @xmath75 with the connectivity of each node indicated by vertical lines . the horizontal dashed line indicates an average connectivity of two . _ , title="fig : " ] the resulting connectivity @xmath71 verses node age @xmath38 distributions are shown in fig . [ f_ljn_dist_rand ] along with example simulation runs at different acceleration parameters @xmath13 . this figure makes clear that the connectivity distribution @xmath71 is monotonically decreasing with @xmath38 for @xmath172 so younger nodes are more heavily connected than older nodes on average , and monotonically increasing for @xmath12 so younger nodes are less heavily connected than older nodes on average . it is only on the boundary between these two regions at @xmath11 that the connectivity distribution is flat so average node connectivity is independent of node age . as previously , the @xmath71 distribution contains information about both node connectivity and node age and so approximates network statistics ( simulated or observed ) when all of this information is available . we now calculate the age independent final link distribution @xmath81 as in eq . [ eq_final_link_dist ] . in the case @xmath11 we have @xmath82 , meaning the connectivity @xmath17 is independent of the node age @xmath38 , so the delta - function of eq . [ eq_final_link_dist ] again integrates to give @xmath173 . for other cases , we have @xmath174^{1/\alpha } , & 0<\alpha , \alpha\neq 1 . \\ \end{array } \right.\ ] ] the age constraint @xmath85 translates into the connectivity constraints @xmath175 $ ] for @xmath7 , @xmath176 $ ] for @xmath177 , @xmath178 for @xmath11 , and @xmath179 $ ] for @xmath12 . via eq . [ eq_final_link_dist ] , these relations determine the predicted age independent final link distribution @xmath81 under random attachment to be @xmath180 each of these separate distributions is normalized over @xmath181 via @xmath90 where @xmath182 for @xmath7 , @xmath176 $ ] for @xmath177 , @xmath178 for @xmath11 , and @xmath179 $ ] for @xmath12 . the average connectivity per node @xmath94 as required over these same ranges . as shown in fig . [ f_ckn_alpha_rand ] , the predicted age independent continuous connectivity distribution @xmath81 is monotonically decreasing as a function of @xmath17 for both @xmath53 and @xmath12 , and a delta - function for @xmath11 . as a function of acceleration parameter @xmath13 for randomly connected networks . the distributions are monotonically decreasing for @xmath183 over the respective ranges @xmath182 for @xmath7 , @xmath176 $ ] for @xmath177 , and @xmath179 $ ] for @xmath12 . _ ] as a function of acceleration parameter @xmath13 for random attachment . it is evident that while the unobservable continuous distributions @xmath81 are distinguishable , the observable discrete distributions are virtually indistinguishable . we also show the actual points for different network simulations of @xmath99 nodes with average connectivity @xmath31 with @xmath7 ( triangles ) , @xmath11 ( diamonds ) , @xmath96 ( circles ) , and @xmath184 ( squares ) . _ _ ] the continuous connectivity distributions @xmath81 are again typically defined over very limited ranges @xmath107 $ ] for acceleration parameters greater than zero so many nodes possess fractional connectivity less than unity . this again requires that the continuous distribution @xmath81 be converted into an equivalent discrete distribution @xmath108 following the methods of appendix [ app_exp_cutoff ] . [ f_discrete_rand ] shows the results of taking full account of the finite ranges of the continuous distributions @xmath81 which impose exponential cutoffs on the generated discrete distribution @xmath108 . it is clear that the exponential cutoffs render the observable discrete distributions for different acceleration parameters essentially indistinguishable . this result is borne out by simulations of networks with @xmath99 nodes with average connectivity @xmath31 for different acceleration parameters ( also shown in fig . [ f_discrete_rand ] ) . again , we illustrate the numerical algorithm of appendix [ app_exp_cutoff ] via specific examples for randomly attached networks . consider continuous distributions @xmath81 where the boundary point @xmath109 lies within the range of support for the distribution for all acceleration parameters except @xmath11 . that is , we assume @xmath185 for @xmath3 and @xmath186 for @xmath12 , while for @xmath11 we assume @xmath187 . these illustrative choices then give @xmath188^{\frac{1}{\alpha}}- \left[\frac{1-\frac{\alpha}{pn^\alpha}(k+\frac{1}{2})}{1-\alpha}\right]^{\frac{1}{\alpha } } , & 0<\alpha<1 , \\ & \\ 0 , & \alpha=1 , \\ & \\ \left[\frac{\frac{\alpha}{pn^\alpha}(k+\frac{1}{2})-1}{\alpha-1}\right]^{\frac{1}{\alpha}}- \left[\frac{\frac{\alpha}{pn^\alpha}(k-\frac{1}{2})-1}{\alpha-1}\right]^{\frac{1}{\alpha } } , & \alpha>1 . \\ \end{array } \right.\ ] ] in this example , the allowed ranges for the connectivity @xmath17 are @xmath189 for @xmath7 , @xmath190 $ ] for @xmath177 , and @xmath191 $ ] for @xmath12 . our example distributions have nonzero support to the left of the boundary point @xmath109 and so can be partitioned to give @xmath114 nodes labelled @xmath115 each with average connectivity @xmath116 equal to @xmath192 , & \alpha=0 , \\ & \\ \frac{pn^\alpha}{\alpha } \left [ 1 - ( 1-\alpha ) \left(1-\frac{j}{n } \right)^\alpha \right ] , & 0<\alpha<1 , \\ & \\ pn , & \alpha=1 , \\ & \\ \frac{pn^\alpha}{\alpha } \left [ 1 + ( \alpha-1 ) \left(\frac{j}{n } \right)^\alpha \right ] , & \alpha>1 . \\ \end{array } \right.\ ] ] these values then feed directly into eq . [ eq_total_c_k ] to give the total discrete distribution @xmath108 . to illustrate this last stage of the calculation , consider the case @xmath11 where we have @xmath118 and @xmath119 for all nodes @xmath120 , and noting the absence of any contribution from the right of @xmath109 so @xmath121 , we have @xmath122 for @xmath92 $ ] . expressed as a proportion of growing network size for various acceleration parameters @xmath13 . each network has average connectivity @xmath31 at a critical network size @xmath32 nodes , and in every case , we assume that the average size of smaller islands is @xmath193 . _ ] we now turn to establish how the formation of a single giant connected component depends on the acceleration parameter @xmath13 . the number of nodes in the largest island @xmath146 will be a function of network size and will grow as new nodes and new connections are added to the network . the largest island will grow by one when the new node @xmath15 forms one link with probability @xmath194 which attaches to the largest island with probability @xmath195 , and will grow by the average size of all smaller islands @xmath196 when node @xmath15 forms two links with probability @xmath197 either of which connects to the largest island with probability @xmath195 while the remaining link connects to an external island with probability @xmath198 . altogether , the rate of growth of the largest island goes as @xmath199 using eq . [ eq_p_j_k_dist ] and noting the restriction @xmath200 . numerical simulations ( and some analytically tractable solutions ) indicate that the first term here is negligible compared to the second , and hereinafter we ignore this first term . for initial conditions , we have @xmath201 at node @xmath202 where the first link likely appears , see appendix [ app_exp_cutoff ] . [ f_islands ] shows the predicted growth of the largest island compared to simulated networks for different acceleration parmeters @xmath13 , where all networks possess an average connectivity per node of @xmath203 at a critical network size of @xmath73 , and we assume an average size of smaller islands of @xmath193 . there is an evident close connection between theoretically predicted curves and observed statistics . we have examined the network structural transitions displayed by accelerating and hyper - accelerating probabilistic networks motivated by the observation that important biological networks such as prokaryote gene regulatory networks are both hyper - accelerating and size limited due to network structural transitions . we examined accelerating growing networks of nodes connected by undirected links which were probabilistically added with each new node and either preferentially or randomly attached to existing nodes . the addition of a probabilistic envelope allowing the number of new links added with each new node to be an integer greater than or equal to zero , allowed us to extend network theory to model sparsely connected networks where the majority of nodes are entirely unconnected ( e.g. prokaryote gene regulatory networks ) . consequently , the probabilistic envelope also allowed us to lift existing constraints on the modelling of accelerating networks allowing us to treat decelerating , accelerating , and hyper - accelerating networks . these latter two classes of networks were shown to be subject to transitions in which either a single giant connected component forms or the network condenses into a fully connected state with exponential statistics . we were able to roughly locate these transitions as a function of the acceleration and probability parameters and the network size . mean field approximations were compared to network simulations over a wide range of parameters and shown to be consistent . the present paper uses mean field network theory to model rapidly accelerating networks consisting of many unconnected nodes allowing the examination of the statistical transitions generated under accelerated growth . as such , these extended models will be useful in treating models of , for instance , accelerating biological regulatory networks mainly consisting of unconnected nodes and displaying network transitions which limit size and complexity . such models are required to explain the observed evolutionary record of prokaryote gene regulatory networks . in analyzing eq . [ eq_links_l_jn_beta ] , it is first necessary to determine appropriate limits on @xmath59 giving access to regimes where preferential attachment dominates random attachment for accelerating networks . appropriate limits on @xmath59 can be obtained by performing a taylor expansion of eq . [ eq_links_l_jn_beta ] about the point @xmath154 while retaining only terms linear in @xmath155 . using @xmath7 in eq . [ eq_links_l_jn_beta ] , we have @xmath204.\ ] ] noting @xmath205=a^x[1+(1+x)\ln a]$ ] , the leading terms of a taylor expansion immediately reproduce the @xmath7 result for the random attachment model of eq . [ eq_links_l_jn_rand ] . similarly , using @xmath25 in eq . [ eq_links_l_jn_beta ] gives @xmath206 \left [ \frac{\frac{2}{\alpha+1}x^\alpha+1}{\frac{2}{\alpha+1}x \left(\frac{j}{n } \right)^\alpha+1 } \right]^{\frac{\alpha+1}{2\alpha } } - 1 \right\}.\ ] ] a straightforward differentiation then gives a taylor expansion whose leading terms exactly equal the @xmath25 result for the random attachment model of eq . [ eq_links_l_jn_rand ] . hence , random attachment entirely dominates when @xmath207 or equivalently , when @xmath208 for values of @xmath209 , preferential attachment will influence the final distribution . for sufficiently small @xmath59 , preferential attachment will dominate ( rather than merely contribute ) and this latter boundary can be located as follows . the first connected node , denoted @xmath202 ( i.e. the @xmath210 node ) , likely appears when the cumulative average number of added initial links sums roughly to unity , @xmath211 now the floating end of the new link attached to node @xmath202 can be either randomly attached to one of the nodes @xmath74 to @xmath202 with probability proportional to @xmath212 , or preferentially attached to node @xmath202 with probability proportional to @xmath213 in the preferential attachment regime . for preferential attachment to dominate , we require @xmath214 , or equivalently , @xmath215^{\frac{1}{\alpha+1 } } \ ; = \ ; \left [ \frac{\langle k\rangle}{2 n_c^\alpha}\right]^{\frac{1}{\alpha+1}}.\ ] ] for the typical hyper - accelerating networks considered here , this last constraint ensures that preferential attachment dominates only when @xmath59 is so small as to be essentially zero , and in this paper , we analyze preferential attachment using the setting @xmath62 . defined over the finite range @xmath216 $ ] generates an observable discrete distribution @xmath108 for @xmath217 . the @xmath81 distribution lying to the left of the boundary @xmath218 represents network nodes with fractional average connectivity distributions , and is partitioned into regions each of area @xmath219 with boundaries @xmath220 for @xmath221 . the @xmath81 distribution lying to the right of the boundary @xmath218 represents network nodes with integral average connectivity , and is partitioned into regions as shown . an arbitrary continuous network connectivity distribution @xmath81 defined over a finite range @xmath216 $ ] is essentially unobservable , and in actuality , generates an equivalent discrete connectivity distribution @xmath108 with @xmath92 $ ] which can be compared to observed statistics . here , we show how to calculate the observable discrete distribution from the theoretically predicted but unobservable continuous distribution . the first step is to partition the continuous distribution @xmath81 into two parts at an arbitrarily chosen boundary point @xmath109 of order unity . we choose @xmath218 as shown in fig . [ f_cont_discrete ] . then , that proportion of the continuous distribution @xmath81 lying to the right of @xmath109 can be considered to contribute to the long tail of the observed discrete distribution @xmath108 in the normal way . that is , the proportion of the continuous distribution in the region @xmath222 for integral @xmath17 equal to @xmath223 can be entirely assigned to the discrete distribution bin @xmath108 . the total proportion of the continuous distribution @xmath81 so assigned is @xmath224 and this proportion of the distribution describes the connectivity of a total of @xmath225 nodes . when this proportion is close to unity as usually applies for long tailed distributions , nothing further need be done and the distribution bins @xmath226 then equate to the predicted discrete observable distribution . however , when @xmath227 is significantly less than unity as applies for hyper - accelerating networks , it is necessary to assign the remaining proportion of the continuous distribution lying to the left of the chosen value @xmath109 to the discrete distribution bins @xmath108 . the proportion of the continuous distribution remaining to be assigned is @xmath228 with this proportion of the distribution describing the connectivity of a total of @xmath114 nodes . the best way to understand how this remaining assignment is done is through a successive partitioning of the usual normalization constraint @xmath229 here , the continuous distribution to the left of the point @xmath230 is partitioned into @xmath114 regions with respective boundaries @xmath220 for @xmath231 $ ] and @xmath232 while as noted above , that to the right of @xmath230 is partitioned into potentially an infinite number of regions with boundaries @xmath222 for integral @xmath233 . the second contribution to the discrete bin @xmath108 comes from the @xmath114 regions of the continuous distribution lying to the left of @xmath109 . we would like each of these @xmath114 regions to correspond to a single node of the network with non - integral average connectivity , and achieve this by choosing the regional boundaries @xmath234 such that @xmath235.\ ] ] here , each region defines an area of exactly @xmath219 so @xmath38 regions have a combined area of @xmath236 , see fig . [ f_cont_discrete ] . the average connectivity of the node corresponding to the @xmath237 region is approximately @xmath238 and is typically non - integral . ( other alternatives such as @xmath239 might be chosen but the difference is negligible for large @xmath0 . ) as any given node can , in practise , only possess an integral number of connections , these non - integral average connectivities require that the link numbers for these nodes be assigned probabilistically over the range @xmath92 $ ] so as to give the required average . this is achieved by assigning an integral number @xmath240 of links to this region with @xmath241 $ ] according to the probability distribution @xmath242 with probabilities @xmath243 chosen to satisfy @xmath244 to ensure that this distribution has average @xmath245 . we note that later results will not be overly dependent on whether the @xmath246 distribution chosen here is binomial in form , or some other reasonable distribution . summing over all the @xmath114 regions to the left of @xmath109 , the proportion of nodes possessing an integral number of @xmath247 links is then @xmath248 for @xmath92 $ ] and zero otherwise . this is normalized according to @xmath249 . consequently , the total proportion of the network possessing a discrete number of @xmath17 links is expected to be @xmath250 as required , this distribution is normalized and has average @xmath251 . this gives the discrete probability that a node has @xmath17 links for @xmath252 . it is also common to calculate the related probability that connected nodes have @xmath17 links and this is given by @xmath253 for @xmath254 . in summary , for the general case of arbitrary continuous distributions @xmath81 defined over the range @xmath107 $ ] , the predicted discrete distribution is @xmath255 where @xmath256 , @xmath257 , @xmath234 is chosen to satisfy @xmath258 , and the partition point @xmath109 is arbitrarily chosen to be close to unity . m. faloutsos , p. faloutsos , and c. faloutsos . power - law relationships of the internet topology . in l. chapin , j. p. g. sterbenz , g. parulkar , and j. s. turner , editors , _ proceedings of the conference on applications , technologies , architectures , and protocols for computer communication _ , pages 251262 , new york , 1999 . acm press . m. j. gagen and j. s. mattick . inherent size constraints on prokaryote gene networks due to accelerating " growth . 2005 . accepted by _ theory in bioscience_. ( arxive : q - bio.mn/0312021 , http://arxiv.org/abs/q-bio.mn/0312021 ) . l. j. croft , m. j. lercher , m. j. gagen , and j. s. mattick . is prokaryotic complexity limited by accelerated growth in regulatory overhead ? arxive : q - bio.mn/0311021 . see http://arxiv.org/abs/q-bio.mn/0311021 , 2003 . h. tefani and v. zlati . preferential attachment with information filtering : node degree probability distribution properties . arxive : cond - mat/0404495 . see http://arxiv.org/abs/cond-mat/0404495 .	many growing networks possess accelerating statistics where the number of links added with each new node is an increasing function of network size so the total number of links increases faster than linearly with network size . in particular , biological networks can display a quadratic growth in regulator number with genome size even while remaining sparsely connected . these features are mutually incompatible in standard treatments of network theory which typically require that every new network node possesses at least one connection . to model sparsely connected networks , we generalize existing approaches and add each new node with a probabilistic number of links to generate either accelerating , hyper - accelerating , or even decelerating network statistics in different regimes . under preferential attachment for example , slowly accelerating networks display stationary scale - free statistics relatively independent of network size while more rapidly accelerating networks display a transition from scale - free to exponential statistics with network growth . such transitions explain , for instance , the evolutionary record of single - celled organisms which display strict size and complexity limits .
many recent research programs in galaxy evolution trace changes in correlations between fundamental galaxy properties as a function of cosmic epoch . one such approach compares the number density and luminosity function of galaxies at earlier times ( lilly 1995 ; bershady 1997 ; sawicki , lin , & yee 1997 ; lin 1999 ) to the luminosity function in the nearby universe ( zucca 1997 ; marzke 1998 ; norberg 2002 ) . evidence suggests an increase in the number density of small , blue galaxies at earlier times with only a small amount of passive fading among the more luminous , redder galaxies . other investigations compare the relation between rotation velocity and luminosity ( t - f relation ; tully & fisher 1977 ) locally with that observed in more distant disk galaxies ( e.g. , forbes 1995 ; vogt 1996 , 1997 ; simard & pritchet 1998 ) . most results indicate that galaxies of a given rotational amplitude appear 0.2 - 1.0 mag brighter at @xmath13 , although bershady ( 1998 ) argue that the variation is significant for only the bluest galaxies , and kannappan , fabricant & franx ( 2002 ) show that tully - fisher residuals are strongly correlated with galaxy color . simard ( 1999 ) find no evidence for evolution of the size - magnitude relation among 190 field galaxies out to @xmath14 in the groth strip ( groth 1994 ) . they show that selection effects favor high surface brightness galaxies in existing surveys . these biases may mimic the appearance of luminosity evolution and/or surface brightness evolution reported in many studies ( schade 1996a , 1996b ; roche 1998 ; lilly 1998 ) . yet another line of investigation compares the fundamental plane correlations for early - type galaxies as a function of epoch . recent results suggest a decrease in the mean mass - to - light ratio with lookback time for elliptical galaxies ( van dokkum & franx 1996 ; kelson 1997 ) . in paper x of this series , i m ( 2001 ) found a brightening of 1 - 2 mag for early - type @xmath15 galaxies out to @xmath13 , and in paper ix of this series , gebhardt find an increase in the surface brightness by a similar amount . in this paper we explore the evolution of the correlation between metallicity , @xmath16 , and total luminosity , i.e. , the luminosity - metallicity ( l - z ) relation to which nearly all types of _ local _ galaxies conform . in early - type systems , absorption line indices provide a measure of the mean stellar iron or magnesium abundances ( faber 1973 ; brodie & huchra 1991 ; trager 1999 ) while in spiral ( zaritsky , kennicutt , & huchra 1994 ) and dwarf systems ( lequeux 1979 ; french 1980 ; skillman , kennicutt , & hodge 1989 ; richer & mccall 1995 ) the region oxygen abundance relative to hydrogen is the basis of the metallicity measurement . the luminosity - metallicity relation ( l - z ) can be used as a sensitive probe and consistency check of galaxy evolution . the metallicity of a galaxy can only increase monotonically with time ( unless large - scale infall of primordial gas is invoked ) , while the luminosity may increase or decrease depending on the instantaneous star formation rate . metallicity is less sensitive to variations due to transient star formation events in a galaxy s history . most models and observations of galaxies at earlier epochs predict higher star formation rates and a larger fraction of blue , star - forming galaxies in the past ( e.g. , madau 1996 ; lilly 1996 ; somerville & primack 1999 ) . thus , a high or intermediate - redshift galaxy sample ought to be systematically displaced from the local sample in the luminosity - metallicity plane if individual galaxies participate in these cosmic evolution processes . however , if local effects such as the gravitational potential and `` feedback '' from winds and supernovae regulate the star formation and chemical enrichment process , then the l - z relation might be independent of cosmic epoch . semi - analytic models of somerville & primack ( 1999 ) , for example , show very little evolution of the l - z relation with redshift . one goal of this paper is to provide data capable of testing these alternative hypotheses . at high redshift , absorption line surveys have been tracing the chemical evolution of the lyman - alpha forest and damped lyman - alpha systems for more than a decade ( bergeron & stasiska 1986 ; sargent 1988 ; steidel 1990 ; pettini 1997 ; lu , sargent , & barlow 1997 ; prochaska & wolf 1999 ) , while several studies have recently begun to trace the cosmic chemical evolution of _ individual galaxies _ with known morphological and photometric properties . kobulnicky & zaritsky ( 1999 ) measured region oxygen and nitrogen abundances for a sample of 14 compact star - forming galaxies with kinematically narrow emission lines in the range @xmath17 . their sample was found to conform to the local l - z relation , within observational uncertainties . carollo & lilly ( 2001 ) studied 13 star - forming galaxies at @xmath18 from the canada - france redshift survey ( cfrs ; lilly 1995 ) and found no significant evidence for evolution in the l - z relationship out to @xmath19 . however , at larger redshifts of @xmath20 , both kobulnicky & koo ( 2000 ) and pettini ( 2001 ) found that lyman break galaxies with metallicities between @xmath21 are 2 - 4 magnitudes more luminous than local galaxies of similar metallicity . this deviation from the local l - z relation demonstrates that the luminosity - to - metal ratio varies throughout a galaxy s lifetime and is a potentially powerful diagnostic of its evolutionary state . the deep ( deep extragalactic evolutionary probe ; vogt 2003 ; paper i ) team has been assembling keck spectroscopic data on galaxies in the groth strip survey ( groth 1994 ) in order to study the evolution of field galaxies . previous papers in this series include a study of the evolution of the fundamental plane ( fp ) of early - type galaxies ( paper ix ; gebhardt 2003 ) , the evolution of the luminosity function of e / s0 galaxies ( paper x ; i m 2001 ) , and the structural parameters of groth strip galaxies ( simard 2002 ; paper ii ) . in this paper , we explore the chemical properties of 64 star - forming emission - line field galaxies observed as part of the deep groth strip survey ( dgss ) . we measure the interstellar medium oxygen abundances using the the ratio of strong [ o ii ] , [ o iii ] , and h@xmath22 emission line equivalent widths ( e.g. , pagel 1979 as extended by kobulnicky & phillips 2003 ; hereafter kp03 ) . we analyze the degree of metal enrichment as a function of redshift , luminosity , and other fundamental parameters . throughout , we adopt a cosmology with @xmath23=70 mpc@xmath24 , @xmath25 , and @xmath26 . in order to establish whether dgss galaxies systematically differ from nearby galaxies in their chemical or luminous properties , we require a suitable sample of local galaxies for comparison . for a comparable sample of local star - forming galaxies , we compiled three sets of emission - line galaxy spectra from the literature . the first set , consisting of 16 objects , comes from the 55-object spatially - integrated spectroscopic galaxy atlas of kennicutt ( 1992b ) supplemented by 6 objects from kobulnicky , kennicutt , & pizagno ( 1999 ) . we refer to this as the k92 local sample . for each galaxy with detectable ( s / n@xmath278:1 ) [ o ii]@xmath283727 , [ o iii]@xmath284959 , [ o iii]@xmath285007 , and h@xmath22 emission , we measured the emission - line fluxes and equivalent widths in the same manner as dgss galaxies described below . as a second local sample , we selected the compilation of nearby field galaxies from jansen ( 2000a , b ; nfgs ) with integrated spectra , adopting their published emission - line fluxes and equivalent widths . the nearby field galaxy survey is a collection of 200 local ( @xmath29 ) morphologically diverse galaxies selected from the cfa i redshift catalog ( huchra 1983 ) which has the virtue of being nearly complete to the photographic limits of the survey . see jansen ( 2000a ) for a discussion of the differences between the nfgs and k92 samples . briefly , the k92 objects have a higher fraction of star - forming galaxies ( objects with strong emission lines ) compared to the nfgs sample . as a third local sample , we chose the emission - line selected galaxies from the kitt peak national observatory spectroscopic survey ( kiss ; salzer 2000 ) . kiss is a large - area objective prism survey of local ( @xmath30 ) galaxies selected solely on the basis of strong @xmath31 emission lines . because of this selection criterion , the kiss galaxies are preferentially gas - rich star - forming systems . kiss spectra were obtained using a fiber - fed spectrograph with 3 apertures , so , unlike the first two samples , the spectra are not spatially integrated ( i.e. , do not include the entire galaxy ) . the deep groth strip survey ( dgss ) consists of keck spectroscopy over the wavelength range @xmath324400 9500 obtained with the low resolution imaging spectrometer ( lris ; oke 1995 ) . the sample is magnitude limited at @xmath33 on the vega system . spectra have typical resolutions of @xmath3234 . integration times ranged from 3000 s to 18,000 s with a mean of around 6000 s. the spatial window of the spectral extraction was typically 1.5 . a full description appears in vogt ( paper i in this series ; 2003 ) . we searched the spectroscopic database for galaxies in the groth strip survey with nebular emission lines suitable for chemical analysis . only galaxies where it was possible to measure all of the requisite [ o ii]@xmath34 , h@xmath22 , [ o iii]@xmath284959 , and [ o iii]@xmath285007 lines were retained . these criteria necessarily exclude objects at redshifts of @xmath35 since the requisite [ o ii]@xmath283727 line falls below the blue limit of the spectroscopic setup . likewise , objects with redshifts @xmath36 are excluded because the [ o iii]@xmath285007 line falls beyond the red wavelength limit of the survey . as of february 2002 , there were 693 objects with keck spectra with identified redshifts in the deep groth strip survey . of these 693 objects , 398 unique objects have spectroscopic redshifts within the nominal limits @xmath0 . the usable sample is further reduced because 23 candidates were positioned on the slitmask such that the [ o iii ] lines fell off the red end of the spectral coverage . an additional 49 objects were rejected because their position on the slitmask caused the [ o ii]@xmath283727 line to fall off the blue end of the spectral coverage . furthermore , atmospheric @xmath37 absorption troughs between 6865 6920 ( the `` b band '' ) and between 7585 @xmath38 ( the `` a band '' ) prohibit accurate measurement of emission lines for objects in particular redshifts ranges . we removed 14 objects from the sample in the redshift interval @xmath39 where the @xmath40 line falls in the b band . we removed 8 additional objects in the redshift interval @xmath41 where @xmath40 falls in the a band . we removed 24 objects in the redshift interval @xmath42 which places both [ o iii ] @xmath284959 and [ o iii ] @xmath285007 in the a band . the b band is sufficiently narrow that either [ o iii ] @xmath284959 or [ o iii ] @xmath285007 is always available for measurement . following this selection process , 276 objects remained . an additional 210 objects were removed from the sample because the @xmath40 emission line was absent or too weak ( s / n@xmath438:1 ) for reliable chemical determinations ( kobulnicky , kennicutt , & pizagno 1999 for a discussion of errors and uncertainties ) . the spectra of objects rejected due to a weak @xmath40 line are usually dominated by stellar continuum rather than nebular emission from star - forming regions . most local early - type spirals and elliptical galaxies share these spectral characteristics . for these objects , h@xmath22 is seen in absorption against the stellar spectrum of the galaxy . thus early type galaxies with older stellar populations are preferentially rejected in favor of late type galaxies with larger star formation rates . there should , however , be no metallicity bias introduced by rejecting these 210 objects since we will compare the dgss sample with a local sample selected in the same manner : on the basis of @xmath40 , [ o iii ] @xmath44 and [ o ii ] @xmath34 emission lines measured with a signal - to - noise of 8:1 or better . objects with strong @xmath40 but immeasurably weak [ o ii ] or [ o iii ] lines present a data selection conundrum . in principle , such objects should be included in the sample to avoid introducing a metallicity bias , but it is not possible to compute metallicities if the oxygen lines are not detected with n / n of 8:1 or better . only 3 such objects were found in the database . for these objects , it appears that poor sky subtraction caused the [ o iii ] features to be immeasurably weak . intrinsically - weak oxygen lines may be caused by either extremely high ( @xmath45 ) or extremely low ( @xmath46 ) metal content . in the latter case , oxygen lines are weak because of the lack of @xmath47 and @xmath48 ions . however , in the local universe , galaxies with extremely low intrinsic abundances are under - luminous ( @xmath49 ) , and no such faint galaxies are included in our sample . in the high - metallicity case , efficient cooling decreases the mean collisional excitation level , reducing the [ o iii ] line strengths . however , the balmer lines and [ o ii ] lines are not strongly affected by reductions in electron temperature . it is unlikely that our rejection criteria bias the sample by preferentially excluding either metal - poor or metal - rich systems . of the original 693 objects , 66 galaxies remain . these objects appear in table 1 , along with their equatorial coordinates , redshift , absolute blue magnitude , restframe @xmath50 color in the johnson vega systems , half - light radius @xmath51 , and bulge fraction @xmath52 derived from model fitting routines ( simard 1999 , 2002 ) . the restframe ( @xmath53 and ( @xmath54 colors used in this work were calculated following the procedure of lilly ( 1995 ) , by interpolating the measured v - i color of dgss galaxies over a subset of kinney ( 1996 ) spectra , which were then used to synthesize the different restframe colors . figure 1 shows their redshifted spectra with major emission lines identified next to the f814w hst greyscale images . a cursory glance at the images reveals an assortment of galaxy types , from small compact objects to spiral disks and apparent mergers - in - progress . in order to assess whether the 66 selected objects are representative of the 398 galaxies with spectra in the @xmath55 redshift range , figure [ hist ] shows histograms of their morphological and photometric properties . the six panels show the redshift distribution @xmath56 , the absolute b magnitude @xmath57 , the rest - frame color @xmath50 , the half - light radius @xmath51 , the bulge fraction @xmath52 , and the total asymmetry index , @xmath58 . examination of figure [ hist ] reveals that the 66 galaxies selected for chemical analysis are representative of the entire deep sample in terms of their luminosities , redshift distributions , sizes , and bulge fractions , but they are preferentially bluer and more asymmetric than the sample as a whole . this disproportionate fraction of asymmetric galaxies with strong emission lines may be understood as either 1 ) mergers which trigger star formation and produce regions , or 2 ) isolated galaxies dominated by star - forming regions which give rise to an asymmetric morphology . the possible systematic effects introduced by the selection criteria will be discussed further below . spectra taken on different observing runs with different slitmasks enable us to combine multiple spectra for most dgss galaxies . after wavelength calibrating each spectrum , we combined the available 2-dimensional spectra to produce a master spectrum with higher signal - to - noise . the spectra are not flux calibrated . we manually measured equivalent widths of the emission lines present in each spectrum with the iraf splot routine using gaussian fits . the [ o ii ] doublet , which is visibly resolved , was fit with two gaussian components and the sum recorded . table 1 lists the equivalent widths and measurement uncertainties for each line . where [ o iii ] @xmath284959 was below our nominal s / n threshold of 8:1 , we calculated the ew based on the strength of [ o iii ] @xmath285007 assuming an intrinsic ratio of 3:1 . the reported equivalent widths are corrected to the rest frame using @xmath59 associated uncertainties are computed taking into account both the uncertainty on the line strength and the continuum level placement using @xmath60 where @xmath61 , @xmath62 , @xmath63 , and @xmath64 are the line and continuum levels in photons and their associated @xmath65 uncertainties . we determine @xmath64 manually by fitting the baseline regions surrounding each emission line multiple times . we adopt @xmath66 where 12 is the number of pixels summed in a given emission line for this resolution , and rms is the root - mean - squared variations in an adjacent offline region of the spectrum . using this empirical approach , the stated uncertainties implicitly include _ internal _ errors from one emission line relative to another due to poisson noise , sky background , sky subtraction , readnoise , and flatfielding . in nearly all cases , the continuum can be fit along a substantial baseline region , so that @xmath67 . there are , however , additional uncertainties on the _ absolute _ values of the equivalent widths due to uncertainties on the continuum placement , particularly for the [ o ii ] @xmath283727 line , which are difficult to include in the error budget . for several objects , the @xmath40 emission line falls on both the red and the blue halves of the spectrum so that two independent measurements of @xmath68 are available . in these five cases , the two measurements differ by up to 20% . this difference is understandable since the red and blue spectra were taken at different times during a night , separated by a periods of slitmask realignment and seeing variations . small positional changes of the slitmask may alter the region of the galaxy observed on the red versus blue spectra . thus , the absolute values of the equivalent widths may be uncertain by 20% . however , since we are primarily interested in the oxygen abundances derived from the ratios of these equivalent widths , the impact on the error budget is less severe when all emission lines are measured on the same exposure . this is rarely the case . [ o iii ] and @xmath40 generally fall on the red spectrum and [ o ii ] on the blue spectrum . to account for differences in sky subtraction and slitmask placement on the red versus blue spectra , we therefore have assigned an uncertainty of 20% to the ew of [ o ii ] @xmath283727 . this error is not reflected in column 13 of table 1 , but has been included in the oxygen abundance computations that follow . in the analysis that follows , accurate assessment of chemical abundances in the warm ionized medium requires that the observed emission lines arise in h ii regions powered by photoionization from massive stars . non - thermal sources such as active galactic nuclei ( agn ) often produce emission - line spectra that superficially resemble those of star - forming regions . agn must be identified as such because blindly applying emission - line metallicity diagnostics calibrated from h ii region photoionization models will produce erroneous metallicities . traditionally , agn can be distinguished from starbursts on the basis of distinctive [ n ii]/h@xmath69 , [ s ii]/h@xmath69 , [ o i]/h@xmath69 , [ o iii]/h@xmath22 , and [ o ii]/[o iii ] line ratios ( see heckman 1980 ; baldwin , phillips , & terlevich 1981 ; veilleux & osterbrock 1987 ) . however , some or most of these diagnostic lines are unobservable in the dgss and in the current generation of ground - based optical surveys at increasingly higher redshifts . with these limitations in mind , rola , terlevich , & terlevich ( 1997 ) have explored using ratios of @xmath70}$ ] and @xmath68 as substitute diagnostics for identifying agn . they distinguish agn from normal star forming galaxies partly on the basis of large @xmath71 \lambda3868 $ ] to [ o ii ] @xmath34 ratios . two dgss objects , 092 - 7832 and 203 - 3109 have @xmath72\lambda3826}/ew_{[o~ii]\lambda3727 } > 0.4 $ ] , compared to @xmath72\lambda3826}/ew_{[o~ii]\lambda3727}\sim0.1 - 0.2 $ ] for normal star - forming objects ( e.g. , kennicutt 1992 ) . osterbrock ( 1989 ) also notes the presence of enhanced @xmath71\lambda3868 $ ] as a common signature in agn and liners . on this basis , we remove these two @xmath73 objects from further analysis . following rola , terlevich , & terlevich ( 1997 ) we also explored the diagnostic utility of enhanced @xmath74\lambda3727$]/@xmath40 ratios as indicators of agn activity . as shown by their figure 5 , there is considerable overlap between bona - fide agn and normal star - forming galaxies . among the 66 dgss galaxies , there are 8 objects with @xmath70}$]/@xmath75 which might suggest the presence of liner or agn activity . none of these galaxies have measured @xmath76\lambda6584/h\alpha$ ] ratios , so more traditional diagnostics are not available . however , we do not see enhanced @xmath71\lambda3826$]/@xmath74\lambda3727 $ ] ratios in these galaxies , nor is there any evidence for broad linewidths or large @xmath77\lambda5007/h\beta$ ] ratios that are also typical signatures of agn . lacking any additional evidence to the contrary , we proceed under the assumption that the 64 remaining galaxies in our sample are all dominated by emission lines from star formation regions . figure [ select ] shows in greater detail the distribution of magnitude , color , restframe @xmath68 and @xmath40 luminosity for the 64 remaining galaxies compared to the original set of 276 dgss objects with emission lines in the range @xmath0 . symbols distinguish objects by redshift : `` low '' ( 17 objects ; @xmath78 ) , `` intermediate '' ( 21 objects ; @xmath79 ) , and `` high '' ( 26 objects ; @xmath80 ) . the ratio of comoving volumes in these three redshift bins is approximately 1:3:4.6 . filled symbols denote galaxies selected for chemical analysis , while open symbols denote the entire set of galaxies in each redshift interval . points with error bars denote the means and dispersions of each sample . the lower row shows @xmath50 color versus @xmath57 . objects in the lowest redshift bin exhibit a correlation between luminosity and @xmath50 color , while the samples in the latter two redshift bins do not . note also that the distribution of colors among the dgss galaxies is bimodal , especially in the higher two redshift bins . a similar bimodality is seen in the @xmath81 combo-17 survey ( bell 2003 ) . the dgss galaxies selected for analysis are distributed evenly within the bluer grouping of galaxies . the two highest redshift bins contain a higher fraction of luminous blue objects which are absent in the low - redshift bin . in the high - redshift bin , the magnitude - limited nature of the sample becomes obvious , as there are no galaxies fainter than @xmath82 . in all redshift bins , the selected galaxies are preferentially those with the highest @xmath68 and the bluest colors . this effect is most significant in the highest redshift bin . the correlation between blue magnitude and @xmath40 luminosity in all three redshift bins reflects the fact that the @xmath40 luminosities were computed from @xmath57 and @xmath68 . no extinction corrections have been applied to the data . the distribution of equivalent widths indicates that the star formation per unit luminosity is roughly similar for galaxies in the two highest redshift intervals , but that the lowest redshift bin has fewer objects with large ew . this is most likely a volume effect since there will be fewer extreme starburst objects in the smaller volume between @xmath4 than @xmath83 or @xmath84 . oxygen is the most easily measured metal in regions due to its strong emission lines from multiple ionization species in the optical portion of the spectrum . measurement of the oxygen abundance relative to hydrogen is based upon the intensity ratio of the collisionally excited [ o ii]@xmath283727 and [ o iii]@xmath284959,5007 lines relative to balmer series recombination lines ( e.g. , h@xmath22 ) using standard analysis techniques ( see osterbrock 1989 ) . even when the physical conditions of the ionized gas such as electron temperature and density can not be measured , the ratio of strong forbidden oxygen emission lines can still provide a measure of the overall metallicity of the gas ( the so - called strong line @xmath85 ratio method ; pagel 1979 ; kobulnicky , kennicutt , & pizagno 1999 , hereafter kkp ) . however , in the dgss and other large spectroscopic surveys , relative emission - line intensities are often not measured . only equivalent widths are available . in a companion paper to this one , kobulnicky & phillips ( 2003 ) demonstrate that the ratio of emission - line equivalent widths , @xmath86 , is a quantity comparable to @xmath85 and is suitable for measuring oxygen abundances . we adopt the @xmath85 calibration of mcgaugh ( 1991 ) relating the ratio of [ o iii ] @xmath87 4959 , 5007 , [ o ii ] @xmath283727 , and @xmath40 to the oxygen abundance relative to hydrogen , o / h . although multiple prescriptions have been proposed in the literature , ( as reviewed in kkp and in figure [ r23oh ] ) , the exact choice is unimportant here since we are only interested in _ relative _ abundances between local and distant galaxy samples analyzed in the same manner . furthermore , the sample includes only luminous galaxies ( @xmath88 , with three exceptions ) so that , based on local analogs , they all are expected fall on the upper ( metal - rich ) branch of the empirical strong - line calibrations . the nine dgss galaxies with measurable [ n ii ] and h@xmath69 lines all have [ n ii]/@xmath89 , confirming that these objects belong on the metal - rich branch of the calibration.8.4 have [ n ii]/h@xmath69 ratios @xmath270.1 . adopting the less - plausible hypothesis that the dgss galaxies lie on the metal - poor branch of the double - valued strong - line @xmath85@xmath90 relation would require all the galaxies to have extremely low oxygen abundances , @xmath91 . given that direct measurements of electron temperatures and oxygen abundances in similarly luminous @xmath92 objects rule out the lower branch possibility ( kobulnicky & zaritsky 1999 ) we believe this upper - branch assumption to be generally valid . ] to ensure that the nfgs and kiss comparison objects are on the metal - rich branch , both local samples have been culled to contain only objects with @xmath93\lambda6584}/f_{h\alpha}>0.15 $ ] . in this paper , we compute oxygen abundances adopting the analytical expressions of mcgaugh ( 1991 , 1998 as expressed in kkp ) which are based on fits to photoionization models for the metal - rich ( upper ) branch of the @xmath85o / h relation . in terms of the reddening corrected line intensities , this relation is @xmath94 where @xmath95\lambda4959 } + i_{[o~iii ] \lambda 5007 } } \over{i_{[o~ii]\lambda3727}}}\biggr),\ ] ] and @xmath96 \lambda3727 } + i_{[o~iii ] \lambda4959 } + i_{[o~iii ] \lambda 5007}}\over{i_{h\beta}}}\biggr).\ ] ] figure [ r23oh ] shows graphically the adopted relation between @xmath85 , @xmath97 and oxygen abundance . alternative relations from the literature are shown for comparison . a star marks the orion nebula value ( based on data of walter , dufour , & hester 1992 ) which is in excellent agreement with the most recent solar oxygen abundance measurement of @xmath98 ( prieto , lambert , & asplund 2001).8.7 based on the new solar oxygen abundance determination of prieto , lambert , & asplund ( 2001 ) . this is 0.1 - 0.2 dex lower than the oft - adopted anders & grevesse ( 1989 ) value , but resolves the discrepancy between solar and orion nebula oxygen abundances . for the orion nebula , @xmath99 , @xmath100 ( walter , dufour , & hester 1992 ) , which , given the calibration cited ( equation 3 ) , leads to @xmath101 , in excellent agreement with the prieto , lambert , & asplund ( 2001 ) measurement . ] kp03 show that strong line equivalent width ratios are tightly correlated with flux ratios and can be used to obtain a metallicity - sensitive parameter akin to @xmath85 for a wide range of galaxy colors and emission line properties . in this paper we follow the prescriptions of kp03 . using equivalent widths , equations 5 and 6 become @xmath102\lambda4959 } + ew_{[o~iii ] \lambda 5007 } } \over{ew_{[o~ii]\lambda3727}}}\biggr),\ ] ] and @xmath103 \lambda3727 } + ew_{[o~iii ] \lambda4959 } + ew_{[o~iii ] \lambda 5007}}\over{ew_{h\beta}}}\biggr).\ ] ] as noted in kp03 , the use of equivalent width ratios , rather than line flux ratios , has the particular advantage of being less sensitive to the ( unknown ) amount of extinction within the host galaxy , at least if the reddening toward the gas and stars is similar . the range of emission line equivalent widths and galaxy colors among the dgss sample studied here falls within the range where @xmath86 is a good surrogate for @xmath85 . the appendix shows histograms of dgss galaxy colors and equivalent widths compared to the local galaxy samples used by kp03 to validate the @xmath86 approach . following kp03 , we perform a rough correction for stellar absorption by adding a constant 2 to the equivalent widths of @xmath40 in table 1 . this correction has the effect of decreasing @xmath86 and increasing the computed oxygen abundance . the impact of this correction is negligible in galaxies with large @xmath68 but raises o / h by as much as 0.08 dex for galaxies with the smallest @xmath68 . table 1 records the oxygen abundances , 12+log(o / h ) , derived from our tabulated emission - line equivalent widths using the @xmath86 prescription above . the galaxies range between 12+log(o / h)=8.4 and 12+log(o / h)=9.0 , with typical _ random _ measurement uncertainties of 0.03 to 0.12 dex . an additional uncertainty of @xmath320.15 dex in o / h , representing uncertainties in the photoionization models and ionization parameter corrections for empirical strong - line calibration , should be added in quadrature to the tabulated measurement errors . in order to look for differences between local and dgss galaxy samples that could potentially bias our conclusions , figure [ zmboh ] shows the relationship between redshift , @xmath57 , @xmath50 color , @xmath68 , @xmath104 , and oxygen abundance for the 64 objects in our sample compared to the three local galaxy samples . solid symbols distinguish dgss objects by redshift : `` low '' ( 17 objects ; @xmath78 ) , `` intermediate '' ( 21 objects ; @xmath79 ) , and `` high '' ( 26 objects ; @xmath80 ) . dots , crosses , and 3-pointed triangles designate the kiss , k92 , and nfgs local galaxy comparison samples . here , the nearby comparison samples have been culled using the same emission - line selection criteria as the dgss galaxies . the redshift luminosity panel of figure [ zmboh ] shows a correlation . as expected for a flux - limited survey , more high luminosity objects and fewer low - luminosity objects are detected in the highest redshift bin compared to the lowest redshift bin . the dgss galaxies are bluer than the mean nfgs galaxy but are consistent with the bluest local galaxies . the dgss galaxies also have , on average , larger emission line equivalent widths than any of the local samples . the lower left panel compares the oxygen abundance versus redshift . the oxygen abundances of galaxies within each redshift bin correlate strongly with blue luminosity and less strongly with @xmath40 luminosity . there is no significant correlation between color and metallicity or between equivalent width and metallicity . the redshift - metallicity ( lower right ) panel reveals that the zero point of the l - z correlation is displaced toward higher luminosities for larger redshifts : dgss galaxies lie predominantly on the upper envelope of the local galaxy samples . while most of the dgss sample galaxies are consistent with a luminosity - metallicity trend , the two least luminous objects , 092 - 1375 and 172 - 1242 with @xmath105 , lie well off the correlation if the empirical line - strength to - metallicity conversion is blindly applied . based on studies of other galaxies of similar luminosity ( @xmath106 ; skillman , kennicutt , & hodge 1989 ) , we suspect that these two galaxies do not belong on the the upper branch of the @xmath85o / h relation . they lie in the `` turn - around '' region between the upper and lower branches where the strong - line calibration is particularly uncertain . if these objects lie on the lower branch of the @xmath85 calibration , then they have metallicities near 12+log(o / h)=8.1 . unfortunately neither of these objects have @xmath31 or [ n ii ] detections , so there is no way to way to discriminate between the two metallicity branches . due to their metallicity ambiguity , we remove these two galaxies from any further analysis . we turn now to a detailed comparison of the luminous and chemical properties of the remaining dgss sample to local galaxies and other intermediate - redshift galaxies . we use subsets of the same three sets of local galaxies , constrained now also to be brighter than @xmath107 to better match our sample , and to have @xmath108\lambda 6584 } / f_{h\alpha } > 0.15 $ ] in order to assure they lie on the metal - rich branch of the @xmath85 relation . the k92 , nfgs and kiss samples yielded 22 , 36 and 80 galaxies , respectively , meeting these additional cuts . figure [ lz ] shows the best - fit luminosity - metallicity relation for dgss galaxies compared to the three local galaxy samples . filled symbols denote dgss galaxies in the redshift ranges @xmath109 , @xmath110 , and @xmath2 as in figure [ zmboh ] . crosses , skeletal triangles and dots denote the k92 , the nfgs and the kiss local samples , respectively . the eight open squares show the best - measured objects in the @xmath111 field galaxy study by carollo & lilly ( 2001 ; cl01 ) . , @xmath112 ( @xmath113 ) cosmology adopted in that study . oxygen abundances are computed using the published @xmath85 and @xmath97 values from that work . ] open triangles show the @xmath114 objects from kobulnicky & zaritsky ( 1999 ; kz99 ) which meet the same emission line selection criteria as the dgss . open stars are the high - redshift ( @xmath20 ) galaxies from kobulnicky & koo ( 2000 ; kk00 ) and pettini ( 2001 ; pe01 ) . lines represent linear least squares fits to each sample , taking into account the uncertainties on both o / h and @xmath57 . uncertainties of @xmath115 mag and @xmath116 dex . for the dgss galaxies we adopt @xmath115 mag and we add 0.1 dex in quadrature with the @xmath117 listed in table 1 to account the systematic uncertainties plus measurement errors . ] becuase errors on @xmath57 are usually smaller than the errors on o / h , this fit approaches a fit of metallicity on luminosity . in the upper left panel of local galaxies the long - dashed line marks the fit to the combination of all three local samples ( k92+nfgs+kiss ) while the heavy dashed line shows the fit to just the k92+nfgs data . formal results from these fits and uncertainties appear in table 2 . the local galaxy samples were fit both collectively and then individually because they exhibit different distributions . the kiss galaxies lie preferentially to the right ( metal - rich and underluminous ) side of the best - fit l - z relation for the combined k92+nfgs+kiss sample . one possible explanation for this offset is that kiss spectra were obtained with 2 - 3fibers or slits centered on the nucleus of each galaxy . spectra dominated by emission from the inner several kpc will preferentially sample the metal - rich nuclei of galaxies and bias the kiss sample toward higher metallicities . for this reason , we use only the fit to the k92+nfgs samples ( heavy dashed line in figure [ lz ] ) for comparison to the dgss galaxies . figure [ lz ] shows a representative error bar in the lower left panel indicating the typical statistical uncertainties of @xmath118 mag and @xmath119 dex . it is worth emphasizing here that the oxygen abundances for all samples represented in figure [ lz ] have been computed in an identical manner from measured emission line equivalent widths . no corrections for internal reddening have been made . we have adopted the recommendation of kp03 and increased all local and dgss galaxy @xmath40 equivalent widths for 2 of stellar balmer absorption . the absolute b magnitudes have been computed ( or converted from published values ) assuming the @xmath120 , @xmath121 , @xmath26 cosmology . k - corrections have been applied to the high - redshift samples based on multi - band rest - frame b - band and uv photometry . the various samples represented should be directly comparable , although they might not be completely representative of the entire population of galaxies at that redshift . figure [ lz ] reveals a striking evolution in the slope and zero point of the best fit luminosity - metallicity correlation with redshift . this offset between local and dgss samples is similar even if other fitting schemes are adopted ( e.g. , unweighted fits , or o / h errors only ) . although considerable scatter exists within a given sub - sample , _ the more distant galaxies are markedly brighter at a fixed oxygen abundance . _ alternatively , galaxies of a given luminosity are systematically more metal - poor at higher redshifts . the best fit relation for the local k92+nfgs samples is @xmath122 . for the @xmath123 dgss galaxies , we find @xmath124 . note that these slopes differ considerably from other published l - z relations for local galaxies . melbourne & salzer ( 2002 ) , for example , find @xmath125 . some of this difference is due to the different metallicity calibrations used in that work compared to our study . the majority of the difference is due to the restricted range of galaxy luminosities and metallicities in the present dgss sample . melbourne & salzer ( 2002 ) included all kiss galaxies on both halves of the doubled - valued @xmath85 to o / h relation , down to magnitudes of @xmath126 . they find evidence for a change in slope of the l - z relation at low luminosities which alters the slope of their overall fit compared to that for our more restricted luminosity range . more important than the formal zero point differences of the fits are the luminosity and metallicity offsets of the l - z relation near the midpoints of the sample distributions . these midpoints are approximately at a metallicity of @xmath127 and a magnitude of @xmath128 . the low - redshift dgss ( @xmath129 ) subsample is as much as 1.6 magnitudes brighter and/or 0.10 dex more metal - poor than the mean of the local sample . the intermediate- and high - redshift dgss subsamples are 2.0 and 2.4 magnitudes brighter than the local samples of similar luminosity and/or as much as 0.15 dex more metal - poor . high - redshift @xmath130 galaxies show a larger offset of at least 3 magnitudes compared to the local sample , although the uncertainties on o / h are large for those objects . in the highest redshift dgss bin , there is a statistically significant change in the best - fit slope of the l - z relation in the sense that the least luminous galaxies at @xmath7 are more offset from the local l - z relation compared to their more luminous counterparts at the same redshift . figure [ lz ] shows that there is considerable dispersion among the local samples , such that some of the brightest , most metal - rich local galaxies overlap the @xmath7 l - z relation . at luminosities fainter than @xmath131 , however , local galaxies are all 1 - 3 magnitudes fainter , for a given metallicity , compared to the @xmath7 objects . the change in slope of the l - z relation at @xmath7 is in the sense that the low - luminosity dgss galaxies in this redshift interval show greater offsets from the local samples than the high - luminosity galaxies . the lack of dgss galaxies in the lower right corner of the plot ( low luminosity but high metallicity ) represents a real offset from the local relation and is not due any selection effect we can identify . galaxies that might occupy this locus near @xmath131 , @xmath132 would have measurable emission lines and line ratios , @xmath85 , identical to the more luminous galaxies with @xmath133 . although the ratios of [ o iii ] and [ o ii ] line strengths to h@xmath22 line strengths drop with increasing metallicity , the sample selection process would not preferentially exclude objects with low [ o iii]/h@xmath22 ( high metallicity ) in favor of those with high [ o iii]/h@xmath22 ( low metallicity ) . if such low - luminosity , high - metallicity objects existed , they would have been detected and included in the analysis . in our selection of objects from the dgss database , we found only 3 objects ( at @xmath134 ) with measurable @xmath40 but immeasurably weak [ o iii ] ( s / n@xmath135 ) . in all those cases , poor sky subtraction is clearly the cause of the weak [ o iii ] . furthermore , there is no correlation between @xmath68 and o / h ( see figure [ zmboh ] ) , as might be expected if selection effects had preferentially removed objects with weak line strengths from the sample or if the corrections for balmer absorption dominated the errors in a systematic way . although other selection effects in the local or distant samples may give the appearance of luminosity - metallicity evolution from @xmath136 to the present , we proceed in the next section to show that we can rule out several obvious selection biases . selection biases in the local and distant galaxy samples could produce the signature of an evolving l - z relation seen in figure [ lz ] . for instance , comparing local galaxies having moderate @xmath68 to the rare dgss galaxies with strong starbursts and uncommonly large @xmath68 values could lead to misleading conclusions about the nature of distant versus local populations . we investigated sources of selection bias by examining the residuals in the l - z relation as a function of other fundamental galaxy parameters . using unweighted linear least - squares fits to the kiss , nfgs , and dgss samples separately , we computed residuals for each sample and searched within each sample for correlations with galaxy color , @xmath40 luminosity , size , and equivalent width . only @xmath68 is marginally correlated with l - z residuals , and only for the dgss galaxies . figure [ lzresidew ] plots galaxy @xmath68 versus luminosity residuals , @xmath137 , from the above best - fit linear relation for each subgroup in the l - z plane . there is no correlation for the kiss sample or for the nfgs sample . we find a small trend for the highest - redshift dgss galaxies which is driven by the handful of galaxies with extreme values of @xmath68 , but the correlation coefficient indicates an equal or greater correlation could be achieved in a random sampling of points at least 20% of the time . _ the lack of a significant correlation between magnitude residuals and @xmath68 is evidence against the hypothesis that extreme starbursts , which temporarily increase the emission line strength and overall luminosity , are a source of dispersion in the l - z relation . _ hence , the larger emission line equivalent widths in the dgss samples compared to the local samples do not bias the distant galaxies toward larger luminosities and are not responsible for the observed l - z relation offsets . further evidence for this conclusion comes from figure [ lzlim ] , which shows only those dgss and local kiss , nfgs , and k92 galaxies with the restricted ew criterion @xmath138 . best fit lines to each sample show that the basic conclusion remains the same : distant dgss galaxies are 1 - 2 magnitudes more luminous than local galaxies of comparable metallicity . this result does not appear to be due to some kind of selection effect since the dgss sample analyzed here spans the entire range of luminosities from the entire dgss sample in each redshift interval , i.e. , we are not simply selecting the brightest objects from each redshift interval , as shown by figure [ select ] . figure [ select ] hints at another kind of selection effect which may lead to erroneous conclusions regarding differences between dgss and local galaxy samples . all of the dgss galaxies have very blue colors with @xmath139 , while only a small fraction of local galaxies are as blue . the dgss samples are likely to contain a larger fraction of extreme star - forming galaxies with younger light - weighted stellar populations . as a second means of examining selection / population differences between the local and dgss samples , we have constructed a third version of figure [ lz ] using only local galaxies with the bluest colors . figure [ lzlimboth ] shows only those dgss and local k92 , kiss , and nfgs galaxies with @xmath139 _ and _ @xmath138 . although the number of nfgs and kiss data points are fewer , best fit lines to each sample and the basic conclusions remain the same . another way to look at this result is to say that the @xmath7 galaxies are 10% to 80% ( 0.04 to 0.25 dex , depending on @xmath57 ) more metal - poor than local galaxies of comparable luminosity . selection effects such as differences in color distributions or ew distributions between the local and dgss samples do not appear to be responsible for the observed variation of the l - z relation with epoch . one straightforward interpretation of figure [ lz ] , figure [ lzlim ] , and figure [ lzlimboth ] is to say that emission line galaxies _ have undergone considerable luminosity evolution , @xmath140 mag , from @xmath141 ( 7 gyr ago ) to the present and as much as @xmath142 mag since @xmath130 ( 11 gyr ago ) . _ this amount of brightening is consistent with , or greater than than , the interpretation of tully - fisher evolution studies out to @xmath13 ( e.g. , forbes , 1995 ; vogt 1997 ; simard & pritchet 1998 ; ziegler 2002 ) . an alternative way of understanding our new result is to say that emission - line field galaxies have experienced _ a measurable degree of metal enrichment ( up to a factor of 2.0 for the least luminous objects ) at constant luminosity _ since @xmath141 . the arrows in figure [ lzlimboth ] indicate evolution in the l - z plane caused by constant star formation , passive evolution , metal - poor gas inflow , and star formation bursts and/or galaxy mergers . constant star formation drives galaxies mainly to the right in figure [ lzlimboth ] ( i.e. , metal enrichment at constant luminosity ) while passive evolution moves galaxies downward ( i.e. , fading at constant metallicity ) . some combination of these two processes are probably responsible for evolving the @xmath143 galaxies into the region occupied by today s @xmath10 galaxies , provided that the evolved descendants of the dgss sample can be found among the local samples . given that many of the dgss galaxies appear to be disklike systems with active star formation , it is reasonable to suppose that a significant fraction of them are still disklike star - forming systems at the present epoch . the apparent evolution in the luminosity - metallicity relation seen in figures [ lz ] and [ lzlimboth ] stands in contrast to the conclusions of kobulnicky & zaritsky ( 1999 ) and carollo & lilly ( 2001 ) , who found little or no evolution in the redshift ranges @xmath144 and @xmath145 , respectively . open squares in figure [ lz ] show that the cl01 objects fall at the high - luminosity end of the luminosity - metallicity correlation where the local and `` distant '' samples overlap . a comparison of the cl01 objects to the local samples and to the brightest dgss galaxies shows that they have luminosities and metallicities similar to our most luminous objects and are consistent with the l - z relation at that epoch . there is very little difference between the l - z relations at @xmath7 and @xmath10 at the luminous end . hence , the conclusion of cl01 , namely that their galaxies lie near the l - z relation for local galaxies , is appropriate given the restricted magnitude range of that sample . the new conclusion based on figure [ lz ] is made possible by the addition of larger numbers of lower - luminosity galaxies which greatly extend the luminosity and metallicity baselines . although the kobulnicky & zaritsky ( 1999 ) sample included a broad range of luminosities , the least luminous objects were at the lowest redshifts where offsets from local galaxies are least pronounced . furthermore , kobulnicky & zaritsky ( 1999 ) used a local comparison sample that included only a restricted subset of galaxies from the kennicutt ( 1992a , b ) spectral atlas . objects from this sample show increased scatter toward low metallicities at fixed luminosity , due mostly to the errors introduced by variable , unknown amounts of stellar balmer absorption . the increased scatter in the local sample masked the trend which has now become apparent in figure [ lz ] . the global shift and slope change of the l - z relation between @xmath10 and @xmath136 supports the conclusions of kobulnicky & koo ( 2000 ) , pettini ( 2001 ) and mehlert ( 2002 ) at higher redshifts , who found galaxies to be overly luminous for their metallicity at @xmath146 . how do our results compare to conclusions about the evolution of the luminosity function ( lf ) ? lin ( 1999 ) use early cnoc2 data on @xmath147 galaxies over the redshift range @xmath148 to conclude that , among late - type galaxies ( the ones most comparable to the dgss objects selected here ) , changes in the lf are most consistent with _ density _ evolution , and little or no _ luminosity evolution_. however , cohen ( 2002 ) , based on studies in the hubble deep field north vicinity over the redshift range @xmath149 , concluded that emission - line objects show _ moderate luminosity evolution _ and _ little density evolution_. interpreting figure 5 of lin ( 1999 ) as luminosity evolution suggests that , between @xmath150 and @xmath151 , late - type galaxies with @xmath131 fade by @xmath1521 mag , while galaxies with @xmath153 fade by @xmath154 mag . more recent results of wolf ( 2002 ; combo-17 survey ) show a decrease in the luminosity of @xmath155 galaxies from z@xmath3 to the present by as much as 2 magnitudes , dpending on spectral type . given the uncertainties in the distant luminosity functions , this amount of fading is consistent with the amount of fading required for the @xmath7 dgss galaxies to match the local l - z relation in figure [ lz ] . however , this consistency does not prove the fading model , and other possibilities are discussed below . the addition of chemical information on galaxies at earlier epochs provides a new type of constraint on theories of galaxy formation and evolution . if local effects such as the gravitational potential and `` feedback '' from supernova - driven winds are the dominant regulatory mechanisms for star formation and chemical enrichment , then the l - z relation might be nearly independent of cosmic epoch . the semi - analytic models of kauffman ( 1996 ) and somerville & primack ( 1999 ) show little or no evolution in the l - z relation with epoch , although uncertainties in the model - dependent prescriptions for stellar feedback and galactic winds may have large impacts on the computed chemical and luminous properties . based on current understanding that the volume - averaged star formation rate was higher in the past ( pei & fall 1995 ; madau 1996 ) , and that the overall metallicity in the universe at earlier times was correspondingly lower , we might expect galaxies to be both brighter and more metal - poor in the past . a high- or intermediate - redshift galaxy sample ought , therefore , to be systematically displaced from the local sample in the luminosity - metallicity ( l - z ) plane _ if individual galaxies reflect these cosmic evolution processes_. the key questions are two - fold : which dominates the chemical evolution or luminosity evolution , and how do those rates differ , if they do , in low - luminosity galaxies versus high - luminosity galaxies ? comparison of the data from figure [ lz ] to some simple galaxy evolution models can help distinguish among these possibilities . in a simple evolutionary model , galaxies begin as parcels of gas which form stars and produce metals as the gas fraction decreases from 100% to 0% . the b - band luminosity is related to the star formation history , which may vary as a function of time but is generally taken to be proportional to the mass or surface density of the remaining gas ( e.g. , kennicutt 1998 ) . for a galaxy that evolves as a `` closed box '' , converting gas to stars with a fixed initial mass function and chemical yield , the metallicity is determined by a single parameter : the gas mass fraction , @xmath156 . the metallicity , @xmath16 , is the ratio of mass in elements heavier than he to the total mass and is given by @xmath157 where @xmath158 is the `` yield '' as a mass fraction . a typical total metal yield for a salpeter imf integrated over 0.2100 @xmath159 is @xmath160 by mass ( i.e. , 2/3 the solar metallicity of 0.018 ; see pagel 1997 , chapter 8) . a total oxygen yield for the same imf would be @xmath161 . effective yields in many local galaxies range from solar to factors of several lower ( kennicutt & skillman 2001 ; garnett 2002 ) , implying either that the nucleosynthesis prescriptions ( e.g. , woosley & weaver 1995 ) are not sufficiently precise , that assumptions about the form of the initial mass function are incorrect , or that metal loss is a significant factor in the evolution of galaxies . garnett ( 2002 ) finds that oxygen yields of @xmath162 to @xmath163 among local irregular and spiral galaxies are correlated with galaxy mass , suggesting an increasing amount of metal loss among less massive galaxies . high - velocity winds capable of producing mass loss are observed in local starburst galaxies ( e.g. , heckman 2000 ) and in high - redshift lyman break galaxies ( pettini 2001 ) , but the actual amount of mass ejected from galaxies is difficult to estimate . simulations indicate that winds may be metal - enriched , such that metals are lost from galaxies more easily than gas of ambient composition ( vader 1987 ; de young & gallagher 1990 ; maclow & ferrara 1999 ) . recent chandra observations of a local dwarf starburst galaxy , ngc 1569 , provide the first direct evidence for metal - enhanced outflows ( martin , kobulnicky , & heckman 2002 ) . thus , the closed - box models are probably not appropriate , and more realistic models including selective metal loss are required . in order to compare the luminous and chemical evolution of galaxies in the l - z plane to theoretical expectations , we ran a series of pgase2 ( fioc & rocca-@xmath164volmerange 1999 ) models . pgase2 is a galaxy evolution code which allows the user to specify a range of input parameters . the critical input parameters include : * @xmath165 : the total mass of gas available to form the galaxy . we explored a range of masses , and we present two representative masses of @xmath166 and @xmath167 . no dark matter is included . * @xmath168 : the timescale on which the galaxy is assembled . we assume that galaxies are built by continuous infall of primordial - composition gas with an infall rate that declines over an exponential timescale , @xmath168 , as implemented in pgase2 , + @xmath169 + we chose two representative gas infall timescales of @xmath170 gyr and @xmath1715 gyr . * @xmath172 : the form of the stellar initial mass function . we adopt the salpeter value , @xmath173 , between 0.1 and 120 ( following baldry 2002 ) . * @xmath158 : the chemical yields from nucleosynthesis . we assign woosley & weaver 1995 b - series models for massive stars . the effective metal yield of these models is @xmath174 . the pgase2 models treat a galaxy as a single zone with uniform chemical composition . this oversimplification is adequate to achieve a basic understanding of how fundamental parameters like star formation rate and gas supply affect gross galaxy properties like luminosity and metallicity , but it is insufficient for a detailed analysis of large galaxies with a range of internal metallicities . * @xmath175 : the form of the star formation rate . we adopt a physically motivated prescription where the star formation rate is proportional to the mass of accumulated gas ( or the gas density ; schmidt 1959 ; kennicutt 1998 ) . * @xmath176 : extinction due to dust . we included an inclination - averaged extinction prescription as implemented in pgase2 , but extinction changes the model b magnitudes by only 0.2 mag . extinction is of minor importance in the present comparison . * @xmath177 : the time after the big bang that a galaxy begins to assemble . while not explicitly part of the pgase2 models , this parameter is implicitly required to match the ages of pgase2 galaxies to a particular redshift and lookback time for our adopted cosmology . we initially assume that galaxies begin to form 1 gyr after the big bang so that local @xmath10 galaxies have approximate ages of 12 gyr ( in a 13.5 gyr old universe ) and dgss galaxies in our high - redshift bin at @xmath7 have ages of 5.76.5 gyr ( 6.85.8 gyr lookback times ) . for convenience , table 3 lists the lookback time and age of the universe at a given redshift for our adopted cosmology . we show later that allowing a range of @xmath177 provides a natural explanation for some of the features seen in the l - z relations . our goal in the pgase2 modeling was to reproduce the qualitative behavior of figure [ lz ] where , 1 ) galaxies at @xmath178 are displaced toward brighter magnitudes and/or lower metallicities relative to their @xmath10 counterparts , and 2 ) dgss galaxies at @xmath7 exhibit a change in slope compared to more local l - z relations . a significant constraint on the models is the position of @xmath179 lyman break galaxies , which lie 2 - 4 magnitudes brighter ( 0.5 dex more metal - poor ) than the local l - z relation . if the present - day emission - line samples are at all representative of the evolved descendants of the @xmath7 and @xmath179 galaxies , ) that qualitatively resemble disks , and their bulge - to - disk ratios are @xmath180 . even if the current levels of star formation decline , their evolutionary descendants are likely to be systems with substantial disk components thus , it is plausible to think that the descendants of the @xmath7 dgss galaxies are disk - like star - forming galaxies today , similar to the k92 , kiss , and nfgs objects plotted in figure [ lz].the evolutionary relationship between dgss and lyman break galaxies is less clear . ] then a successful model must reproduce the rapid rise in metallicity and luminosity in the first 2 - 3 gyr ( as evidenced by the lyman break galaxies ) , and it must reproduce the lack of significant l - z evolution during the past 2 - 3 gyr ( as evidenced by the similarity of the @xmath181 galaxies to the present - day l - z relation ) . this means that the rate of evolution of the l - z relation must be rapid during the first few gyr of a galaxy s formation , and it must then drop dramatically as a galaxy approaches the present - day l - z relation . furthermore , large luminous galaxies should slow their l - z evolution more than less luminous galaxies . a slowing of the luminous and/or chemical evolution of a galaxy must physically correspond either 1 ) to cessation of star formation and ultimately to depletion of the gas available to fuel star formation , or 2 ) to continued inflow of metal - poor gas which fuels star formation and dilutes the composition of the ism so that a galaxy remains fixed in l - z space . figure [ lzevp2 ] shows evolutionary tracks of galaxies in the l - z plane based on the infall pgase2 models . the warm ism metallicity by mass , @xmath16 , is marked along the upper @xmath182-axis along with the corresponding closed - box gas mass fraction given by the models ( equivalent to the gas mass fractions predicted by equation [ mu ] using a total yield of @xmath158=0.016 ) as a comparison . pentagons in figure [ lzevp2 ] mark the model galaxies at ages of 1 , 2 , 4 , 8 , and 12 gyr ( from left to right ) . the numbers in parentheses next to each model point show the lookback time in gyr assuming that the galaxy begins its assembly 1 gyr after the big bang . dashed lines denote models with @xmath170 gyr and solid lines denote models with @xmath183 gyr . the upper set of curves shows galaxies with @xmath184 while the lower curves show galaxies with @xmath185 ( not including dark matter ) . solid lines mark the best - fit l - z relations from figure [ lz ] . these models produce the rapid rise in metallicity in the first few gyr . these model masses are in reasonable agreement with observational estimates which place lower limits on the _ dynamical _ masses of @xmath179 lyman break galaxies at @xmath186 ( kobulnicky & koo 2000 ; pettini 2001 . ] and indicate star formation rates of @xmath3250 /@xmath187 . however , these models over - enrich the ism at late times after 8 gyr compared to the data . they predict gas mass fractions that are as low as 1% and metallicities as high as 3 times solar ( z=0.054 ) after 12 gyr , at odds with the observed properties of nearby galaxy samples . we can achieve better agreement between the models and data by varying several crucial model parameters . * _ varying the sfr _ : lowering the model star formation rates would delay the chemical enrichment until later times and lower the luminosity at any given time . effectively , this change produces a wholesale shift in the model curves to the left and downward in figure [ lzevp2 ] . however , lowering the sfr would also lower the predicted emission line equivalent widths which may already be too low in the models compared to the data ( mean of @xmath188 after 8 gyr versus 15 - 20 in the data ) . thus , changing the sfr does not improve the model agreement with the data . * _ varying the chemical yield , @xmath158 _ : as an experiment to produce models that better resemble the data , we introduced a metal yield efficiency parameter , @xmath189 , which reduces the effective yield of the models . reducing the chemical yields simply moves the model curves _ en masse _ to the left without changing the spacing of the model points ( i.e. , without changing the @xmath190 over a fixed time interval ) . physically , @xmath189 may represent the fraction of metals retained in galaxies during episodes of wind - driven metal loss , or it may represent a correction to the theoretical yields from stellar populations . figure [ lzevp ] shows the same models and data as figure [ lzevp2 ] , except that the metal yield has been reduced to 1/2 ( @xmath191 for @xmath192 ) .. ] the @xmath184 model with @xmath170 gyr is now in better agreement with the @xmath7 and @xmath10 data at ages of 8 - 12 gyr . however , the same model for @xmath185 becomes too metal - rich at ages of 8 - 12 gyr compared to the @xmath10 data . * _ varying @xmath168 _ : increasing the infall timescale from 1 gyr to 5 gyr , as shown in figure [ lzevp ] , alters a galaxy s evolution in several respects . it significantly decreases the luminosity during the first few gyr , while producing a slight drop in the gas - phase metallicity at each timestep . longer @xmath168 also decreases the change in metallicity in each timestep , i.e. , decreases the @xmath190 over a fixed time interval , most noticeably at late epochs . in the context of the cold dark matter paradigm , the objects with the largest initial overdensities form most rapidly ( e.g. , blumenthal 1984 ; navarro , frenk , & white , 1995 ) . whether these first objects grow to become massive galaxies depends on the environment in which they are conceived . initial galaxy cores which collapse in the vicinity of larger concentrations of baryonic matter grow to become the massive galaxies at the centers of today s galaxy clusters . objects that form in regions of lower cosmological overdensity would accrete gas on longer timescales and , if isolated from external influences , would proceed more slowly through the star formation and chemical evolution process . a longer formation timescale for lower mass field galaxies observed in the dgss at @xmath7 can plausibly explain their observed evolution in the l - z relation . figure [ lzevp ] demonstrates that @xmath183 gyr provides better agreement than the @xmath170 gyr to the @xmath7 and @xmath10 galaxies . however , we stress that a single set of model parameters can not reproduce either the l - z relation itself at _ any _ epoch or the variation in slope of the l - z relation with epoch . the l - z relation predicted by a single set of model parameters at any epoch would be a vertical line in figure [ lzevp ] ! * _ varying @xmath193 _ : delaying the epoch at which model galaxies begin to assemble simply relabels the ages of models . for example , setting @xmath194 gyr instead of @xmath195 gyr renumbers the ages of the model points in figure [ lzevp ] from 1,2,4,8,12 gyr to 0,1,3,7,11 gyr . for a given lookback time , the galaxies are younger . the effects of decreasing @xmath158 , increasing @xmath168 , and increasing @xmath177 are , to some degree , degenerate . they all serve to move model points toward the left ( metal - poor ) side of figure [ lzevp ] , and we can not , here , discriminate between their relative importance . however , within the context of these simple models , _ no single parameter can solely explain both the slope of the l - z relation and the change in slope with lookback time_. these observed qualities can only be produced in the context of the pgase2 models if at least _ two _ of the following conditions obtain : 1 ) the chemical yield of low - luminosity galaxies are reduced relative to high - luminosity galaxies ( e.g. , garnett 2002 ) , 2 ) @xmath168 is larger for low - luminosity galaxies relative to high - luminosity galaxies , 3 ) @xmath193 is larger for low - luminosity galaxies relative to high - luminosity galaxies ( e.g. , as in the delayed formation scenarios for dwarfs , babul & rees 1992 ) . such a scenario , whereby galaxies assembly timescales or formation epochs vary with environment , is similar to that proposed by sandage , freeman , & stokes ( 1970 ) to explain the differences along the hubble sequence . there is considerable observational support for the idea that lossy galactic winds remove metals and reduce the effective chemical yields from star formation ( e.g. , martin , kobulnicky & heckman 2002 ; garnett 2002 ) . winds are more effective in low - mass galaxies with lower gravitational potentials . the overall slope of the l - z relation may plausibly be explained by a decreasing effective yield in low - luminosity ( low - mass ) galaxies . some combination of longer @xmath168 , and/or later @xmath177 for low - mass galaxies may then be invoked to explain the change in slope of the l - z relation which preferentially displaces @xmath7 low - luminosity galaxies more dramatically from the local l - z relation . figure [ lzevp4 ] uses the same pgase2 models to explore how one possible combination of reduced metal yields in low - mass galaxies relative to larger galaxies combined with a later formation epoch for low - luminosity galaxies , can come close to reproducing both the slope of the l - z relation _ and _ its zero point and slope variations with redshift . solid lines show the 5 gyr infall models from previous figures . as in figure [ lzevp ] , we use @xmath192 for the @xmath167 galaxy but here we choose @xmath196 for the @xmath166 galaxy . this variation in yield is well within the range found by garnett ( 2002 ) . for the @xmath167 galaxy we adopt a formation epoch @xmath195 gyr while for the @xmath166 galaxy we choose a formation epoch @xmath197 gyr . the numbers next to each model point denote the lookback time and the numbers in parentheses denote galaxy age . note how the slope of the l - z relation is in good agreement with a line connecting the model points at the present epoch , but it is in good rather poor with the slope of a line connecting the model points at a lookback time of 7 gyr ( roughly @xmath7 ) . moreover , the dgss galaxies at @xmath7 are , as a whole , more metal - rich than the model points at 7 gyr lookback times , indicating that these particular pgase2 models fail to reproduce ( i.e. , they overpredict ) the magnitude of zero - point evolution in the l - z relation . nevertheless , this combination of later formation epoch and reduced yield for low - mass galaxies does a reasonable job of reproducing the sense of the zero point and slope evolution in the l - z plane with redshift , especialyl for the low - luminosity galaxies . the largest discrepancies are for luminous galaxies whose gas - phase metallicity evolves less than the pgase2 models predict . as noted , the single - zone pgase2 models in the above figures fail to reproduce one important feature of the l - z relation data , namely , the observed lack of evolution for luminous galaxies between @xmath7 and @xmath10 . all of the models in figures [ lzevp2],[lzevp ] , and [ lzevp4 ] predict an increase of 0.2 0.4 dex in o / h between lookback times of @xmath198 gyr and the present . yet , the linear fits to the local and the @xmath7 dgss data indicate get little ( @xmath199 dex ) change in o / h for the most luminous galaxies ( @xmath200 ) . the only ways to decrease the @xmath201 per timestep withinthe framework of these single - zone models are 1 ) to drastically reduce the star formation rate , i.e. , to slow the conversion of gas into stars and heavy elements , or 2 ) to increase the infall timescale beyond the 5 gyr models . however , such a decrease in star formation rate or increase in @xmath168 would produce model galaxies which are dramatically more metal - poor and gas - rich at a given age compared to the present models , and would be in poor agreement between the models and the data . we can find no reasonable range of pgase2 model parameters which reproduce the similarity of luminous local galaxies to @xmath7 luminous dgss galaxies . the resolution to this crisis may lie in the considerable internal complexity of large spiral galaxies ( e.g. , radial chemical gradients , spatially variable extinction and star formation rates ) , which single - zone model like pgase2 do not include . more sophisticated models which treat galaxies as a collection of annuli with differing gaseous and chemical properties , for example , might more accurately reproduce the observed characteristics of the large galaxies seen locally and in the dgss . the lack of l - z evolution among luminous galaxies in figure [ lzevp ] might be explained if star formation and chemical enrichment in these objects proceeds radially from the inside - out . if the star formation in local galaxies is dominated by activity at a larger radius than the star formation activity in @xmath7 dgss galaxies , then the luminosity - weighted dgss spectra will preferentially reflect the chemical composition at different ( i.e. , smaller ) galactocentric radii in dgss galaxies than in local galaxies . although local luminous spiral galaxies may , on whole , be substantially more metal - rich today than at @xmath7 , their emission line luminosities today would be dominated by star formation in comparably - enriched regions , but at larger radii , compared to the dgss galaxies . observations of the mean star formation radius within disk galaxies as a function of redshift could test this hypothesis . regardless of whether this particular hypothesis is true , additional information about the spatial distribution of star forming regions in distant objects ( e.g. , using the hubble space telescope ) , will lead to a greater understanding of the luminous and chemical histories of large galaxies . one additional free parameter not incorporated into the pgase2 models is the role of starbursts on timescales of a few myr that occur in galaxies that evolve on long ( gyr ) timescales according to the model prescriptions above . studies of local galaxies show that their star formation histories are not smooth but instead are punctuated by periods of enhanced activity . such short - term enhancements in the star formation rate of galaxies otherwise evolving smoothly can elevate the luminosity and emission line equivalent widths by factors of several . the observed emission line equivalent widths in dgss galaxies average 10 - 20 , but some exceed 50 ! these characteristics would be better fit by superimposing a burst of star formation of @xmath202 on the model galaxies . the impact of instantaneous bursts on the evolution in the l - z plane would be to create spikes along the model tracks toward higher luminosities . the amplitude of the spikes would scale with the amplitude of the burst , and could conceivably be up to several magnitudes and would persist up to a few 100 myr . given the morphological evidence for ongoing mergers among the dgss sample , evidenced by inspection of figure 1 , it seems possible that short - term enhancements in the star formation rate are responsible for the more extreme luminosities and emission line equivalent widths among the sample . examination of the pgase2 models used here shows that as the @xmath68 declines from 50 to 10 , the b - band luminosity drops by 0.7 mag . if we were to try to correct the magnitudes of the dgss samples by some factor to account for their larger equivalent widths compared to the local samples , this factor would be less than 0.7 mag , not enough to account fully for the larger 1 - 2 mag offsets observed in the l - z relations this is further evidence against the possibility that the offset of the l - z relation at @xmath203 is merely caused by starbursts with temporarily elevated luminosities and equivalent widths , although this may be part of the answer . observations of star - forming galaxies from the deep extragalactic evolutionary probe ( deep ) survey of groth strip galaxies in the redshift range @xmath0 show a correlation between b - band luminosity and ism oxygen abundance , like galaxies in the local universe . both the slope and zero - point of the l - z relation evolve with redshift . dgss galaxies are 13 mag more luminous at @xmath7 than @xmath10 galaxies of similar metallicity . said another way , galaxies of comparable luminosity are 0.1 - 0.3 dex more metal - rich at @xmath10 compared to @xmath7 . the change in slope of the l - z relation is most pronounced for dgss galaxies at @xmath7 and indicates that the apparent chemo - luminous evolution is more dramatic for the least luminous galaxies . we have ruled out several selection effects and biases in the galaxy samples which could produce this result . evolution of galaxies in the l - z plane over the redshift range @xmath204 to @xmath10 appears to result from a combination of fading and chemical enrichment for plausible single - zone galaxy evolution models . within the context of these models , the slope of the l - z relation and the observed evolution of _ slope and zero - point _ with redshift imply that that at least two of the following are true for star - forming galaxies out to @xmath205 : 1 ) low - luminosity galaxies have lower effective chemical yields than massive galaxies , 2 ) low - luminosity galaxies assemble on longer timescales than massive galaxies , 3 ) low - mass galaxies began the assembly process at a later epoch than massive galaxies . comparison of observed ism metallicities to simple galaxy evolution models suggests that loss of up to 50% of the oxygen is necessary to avoid over - enriching the galaxies at the observed redshifts . while the models do a reasonable job of reproducing the evolution of the less luminous galaxies in our sample ( @xmath11 ) , the models do a poor job of reproducing the relative lack of chemo - luminous evolution seen among the luminous galaxies in our sample ( @xmath12 ) , suggesting a breakdown of the single - zone approximation . we predict that dgss galaxies have significantly higher ( 20%-60% ) gas mass fractions than comparably luminous local galaxies . such a systematic difference in the mean gas mass fractions of distant galaxies could , in principle , be observed with instruments such as the proposed radio - wave square kilometer array . we thank john salzer for insightful conversations and for providing the kiss data in electronic form , shiela kannappan for a helpful discussion about the nfgs , matt bershady and james larkin for scientific inspiration . detailed suggestions by a very perceptive referee greatly improved this manuscript . h. a. k was supported by nasa through grant # hf-01090.01 - 97a awarded by the space telescope science institute which is operated by the association of universities for research in astronomy , inc . for nasa under contract nas 5 - 26555 and by nasa through nra-00 - 01-ltsa-052 . this work was also made possible by nsf grants ast95 - 29098 and ast00 - 71198 and nasa / hst grants ar-07532.01 , ar-06402.01 , and ar-05801.01 . kobulnicky & phillips ( 2003 ) showed that the quantity @xmath86 is a good surrogate for the metallicity indicator @xmath85 over a wide range of galaxy colors and emission line properties . however , the correlation between @xmath86 and @xmath85 is tightest for galaxies with strong emission lines ( e.g. , @xmath206 ) . furthermore , the residuals from the 1:1 correspondence show a slight correlation with galaxy color , the residuals being smallest for the bluest galaxies . figure [ histcompare ] reproduces part of figure 5 from kp03 and shows the distribution of @xmath86 , @xmath207 , @xmath208}$ ] , @xmath209}/ew_{[o~ii]})$ ] , and @xmath210 , ( histograms ) of 66 dgss galaxies compared to the residuals from the @xmath86 vs. @xmath85 relation . this figure demonstrates that the dgss galaxies studied here lie in regimes where the @xmath86 vs. @xmath85 correlation is strong and well - behaved . for these galaxies , @xmath86 is a good surrogate for @xmath85 . anders , e. , & grevesse , n. 1989 , gecoa , 53 , 197 babul , a. , & rees , m. j. 1992 , mnras 255 , 346 baldry , i. k. 2002 , apj , 569 , 582 baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , pasp , 93 , 5 bell , e. f. , wolf , c. , meisenheimer , k. , rix , hans - walter , borch , a. , dye . s. , kleinheinrich , m. , & mcintosh , d. 2003 , apj , in press bergeron , j. , & stasinska , g. 1986 , a&a , 169 , 1 bershady , m. a. , haynes , m. p. , giovanelli , r. , & andersen , d. r. 1988 , in galaxy dynamics , eds . d. r. merritt , m. valluri , & j. a. sellwood ( asp conf series ) blumenthal , g. m. , faber , s. m. , primack , j. r. , & rees , m. j. 1984 , nature , 311 , 517 brodie , j. p. , & huchra , j. p. 1991 , apj , 379 , 157 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , apj , 429 , 582 carollo , c. , m. & lilly , s. j. 2001 , apj , 548 , l153 ( cl01 ) cohen , j. g. 2002 , apj , 567 , 672 de young , d. s. , & gallagher , j. s. iii 1990 , apj , 356 , l15 djorgovski , s. , & davis , m. 1987 , apj , 313 , 59 dopita , m. a. , & evans , i. n. 1986 , apj , 307 , 431 dressler , a. , lynden - bell , d. , burstein , d. , davies , r. l. , faber , s. m. , terlevich , r. j. , & wegner , g. 1987 , apj , 313 , 42 edmunds , m. g. & pagel , b. e. j. 1984 , mnras , 211 , 507 faber , s. m. 1973 , apj , 179 , 423 fioc , m. & rocca-@xmath164volmerange , b. 1999 , astro - ph/9912179 forbes , d. a. , phillips , a. c. , koo , d. c. , & illingworth , g. d. 1995 , apj , 462 , 89 french , h. b. , 1980 , apj , 240 , 41 garnett , d. r. 2002 , apj , 581 , 1019 gebhardt , k. , faber , s. m. , koo , d. c. , i m , m. , simard , l. , illingworth , g. d. , phillips , a. c. , sarajedini , v. l. , vogt , n. p. , & willmer , c. n. a. 2003 , apj , in press groth , e. j. , kristian , j. a. , lynds , r. , oneil , e. j. jr . , balsano , r. , & rhodes , j. 1994 , baas , 185 heckman , t. m. 1980 , a&a , 87 , 152 heckman , t. m. , lehnert , m d. , strickland , d. k. , & armus , l. 2000 , apjs , 129 , 493 huchra , j. p. , davis , m. , latham , d. , & tonry , j. 1983 , apjs , 52 , 89 i m , m. , simard , l. , faber , s. m. , koo , d. c. , gebhardt , k. , willmer , c. n. a. , phillips , a. c. , illingworth , g. d. , vogt , n. p. , & sarajedini , v. l. 2002 , apj , 571 , 136 jansen , r. a. , franx , m. , fabricant , d. , & caldwell , n. 2000a , apjs , 126 , 271 ( nfgs ) jansen , r. a. , fabricant , d. , franx , m. , & caldwell , n. 2000b , apjs , 126 , 331 kannappan , s. j. , fabricant , d. g. , & franx , m. 2002 , aj , 123 , 2358 kauffman , g. 1996 , mnras , 281 , 475 kelson , d. d. , van dokkum , p. g. , franx , m. , illingworth , g. , & fabricant , d. 1997 , apj , 478 , l13 kennicutt , r. c. jr . 1983 , apj , 272 , 54 kennicutt , r. c. jr . 1992a , apjs , 79 , 255 kennicutt , r. c. jr . 1992b , apj , 388 , 310 kennicutt , r. c. jr . 1998 , apj , 498 , 541 kennicutt , r. c. jr . & skillman , e. d. 2001 , aj , 121 , 1461 kewley l. j. , & dopita , m. a. 2002 , apjs , 142 , 35 kinney , a. l. , calzetti , d. , bohlin , r. c. , mcquade , k. , storchi - bergmann , t. , & schmitt , h. r. 1996 , apj , 467 , 38 kobulnicky , h. a. , kennicutt , r. c. , & pizagno , j. 1998 , apj , 514 , 544 kobulnicky , h. a. & koo , d. c. 2000 , apj , 545 , 712 ( kk00 ) kobulnicky , h. a. & phillips , a. c. 2003 , apj , 000 ( kp03 ) kobulnicky , h. a. & zaritsky , d. 1999 , apj , 511 , 118 ( kz99 ) lequeux , j. , peimbert , m. , rayo , j. f. , serrano , a. , & torres peimbert , s. 1979 , a&a , 80 , 155 lin , h. , yee , h. k. , carlberg , r. g. , morris , s. l. , sawicki , m. , patton , d. r. , wirth , g. , & shepard , c. w. 1999 , apj , 518 , 533 lilly , s. j. , le fvre , o. , hammer , f. , & crampton , d. 1996 , apj , 460 , l1 lilly , s. j. , tresse , l. , hammer , f. , crampton , d. , & le fvre , o. 1995 , apj , 455 , 108 lilly , s. j. , 1998 , apj , 500 , 75 lu , l. , sargent , w. l. w. , barlow , t. .a . , 1997 , apj , 484 , 131 maclow , m. m , & ferrara , a. 1999 , apj , 513 , 142 madau , p. , ferguson , h. , dickinson , m. , giavalisco , m. , steidel , c. , & fruchter , a. 1996 , mnras , 283 , 1388 marzke , r. o. , da costa , l. n. , pellegrini , p. s. , willmer , c. n. a. , geller , m. j. 1998 , apj , 503 , 617 martin , c. l. , kobulnicky , h. a. , & heckman , t. m. 2002 , apj , 574 , 663 mccall , m. l. , rybski , p. m. , & shields , g. a. 1985 , apjs , 57 , 1 ( mrs ) mcgaugh , s. 1991 , apj , 380 , 140 mcgaugh , s. 1998 , private communication mehlert , d. 2002 , a&a , 393 , 809 melbourne , j. , & salzer , j. j. 2002 , aj , 123 , 2302 navarro , j. p. , frenk , c. s. , & white , s. d. m. 1995 , mnras , 275 , 561 norberg , p. , ( the 2dfgrs team ) , 2002 , mnras , 336 , 907 oke , j. b. , 1995 , pasp , 107 , 375 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei , university science books : mill valley ca pagel , b. e. j. edmunds , m. g. , blackwell , d. e. , chun , m. s. , & smith , g. 1979 , mnras , 189 , 95 pagel , b. e. j. 1997 , `` nucleosynthesis and chemical evolution of galaxies '' , cambridge university press , cambridge , p. 217 pei , y. c. , & fall , s. m. 1995 , apj , 454 , 69 pettini , m. , shapley , a. e. , steidel , c. c. , cuby , j .- g . , dickinson , m. , moorwood , a. f. m. , adelberger , k. l. , & giavalisco , m. 2001 , apj , 554 , 981 ( pe01 ) pettini , m. , smith , l. j. , king , d. l. , hunstead , r. w. 1997 , apj , 486 , 665 press , w. h. , flannery , b. p. , teukolsky , s. a. , & vettering , w. t. 2002 , numerical recipes in c++ , cambridge university press , cambridge , p 671 prieto , c. a. , lambert , d. l. , & asplund , m. 2001 , apj , 556 l63 prochaska , j. x. & wolf , a. m. 1999 , apjs , 121,369 richer , m. g. , & mccall , m. l. 1995 , apj , 445 , 642 rola , c. s. , terlevich , e. , & terlevich , r. j. 1997 , mnras , 289 , 419 roche , n. , ratnatunga , k. , griffiths , r. e. , i m , m. , & naim , a. 1998 , mnras , 293 , 157 salzer , j. j. , gronwall , c. , lipovetsky , v. a. ; kniazev , a. , moody , j. w , , boroson , t. a. , thuan , t. x. , izotov , y. i. , herrero , j. l. , frattare , l. m. 2000 , aj , 120 , 80 ( kiss ) sandage , a. , freeman , k. c. , & stokes , n. r. 1970 , apj , 160 , 831 sarajedini , v. l. , & the deep team , 2003 , in prep . ( paper xiii ) . sargent , w. l. w. , steidel , c. c. , & boksenberg , a. 1988 , apjs , 68 , 539 sawicki , m. j. , lin , h. , & yee , h. k. 1997 , aj , 113 , 1 schade , d. , carlberg , r. g. , yee , h. k. , & lopez - cruz , o. 1996a , apj , 464 , 63 schade , d. , carlberg , r. g. , yee , h. k. , & lopez - cruz , o. 1996b , apj , 465 , 103 schmidt , m. 1959 , apj , 129 . 243 simard , l. , & pritchet , c. j. 1998 , apj , 505 , 96 simard , l. , koo , d. c. , faber , s. m. , sarajedini , v. l. , vogt , n. p. , phillips , a. c. , gebhardt , k. , illingworth , g. d. , & wu , k. l. 1999 , apj , 1999 , 519 , 563 simard , l. , willmer , c. n. a. , vogt , n. p. , sarajedini , v. l. , phillips , a. c. , koo , d. c. , i m , m. , illingworth , g. d. , gebhardt , k. , & faber , s. m. 2002 , apjs , 142 , 1 ( paper ii ) skillman , e. d. , kennicutt , r. c. , & hodge , p. 1989 , apj , 347 , 875 somerville , r. s. , & primack , j. r. 1999 , mnras , 310 , 1087 somerville , r. s. , primack , j. r. , & faber , s. m. 2001 , mnras , 320 , 504 steidel , c. c. 1990 , apjs , 74 , 37 trager , s. c. , worthey , g. , faber , s. m. , burstein , d. , & gonzalez , j. j. 1998 , apjs , 116 , 1 tully , r. b. , & fisher , j. r. 1977 , a&a , 54 , 661 vader , p. 1987 , apj , 317 , 128 van dokkum , p. g. , & franx , m. 1996 , mnras , 281 , 985 veilleux , s. , & osterbrock , d. e. 1987 , apjs , 63 , 295 vogt , n. , forbes , d. , phillips , a. c. , gronwall , c. , faber , s. m. , illingworth , g. d. , & koo , d. c. 1996 , apj , 465 , l15 vogt , n. , phillips , a. c. , faber , s. m. , gallego , j. , gronwall , c. , guzman , r. , illingworth , g. , koo , d. c. , & lowenthal , j. d. 1997 , 479 , l121 vogt , n. 2003 , apj , in prep , ( paper i ) walter , d. k. , dufour , r. j. , & hester , j. j. 1992 , apj , 397 , 196 wolf , c. , meisenheimer , k. , rix , h .- w . , borch , a. , dye , s. , & kleinheinrich , m. 2002 , a&a , 401 , 17 woosley , s. e. & weaver , t. a. 1995 , apjs , 101 , 181 zaritsky , d. , kennicutt , r. c. , & huchra , j. p. 1994 , apj , 420 , 87 ziegler , b. l. , bhm , a. , fricke , k. j. , jger , k. , nicklas , h. , bender , r. , drory , n. , gabasch , a. , saglia , r. p. , seitz , s. , heidt , j. , hehlert , d. , mllenhoff , c. , noll , s. , & sutorius , e. 2002 , apj , 564 , l69 zucca , e. eso slice project team 1997 , a&a , 326 , 477 rrrrcccccccrrrrrrrl [ srctable.tab ] 1 & 073 - 2658 & 14:17:47.597 & 52:29:03.36 & 0.2863 & 21.13 & -18.9 & 0.34 & 2.88 & 0.28 & 0.04 & 0.10 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.7 & a , b + 2 & 072 - 2273 & 14:17:38.928 & 52:29:34.72 & 0.2888 & 22.03 & -18.0 & 0.36 & 2.36 & 0.00 & 0.03 & 0.05 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.3 & a + 3 & 072 - 4040 & 14:17:37.838 & 52:28:58.44 & 0.2908 & 20.01 & -19.6 & 0.70 & 9.42 & 0.00 & 0.15 & 0.06 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.0 & a + 4 & 172 - 1242 & 14:16:35.570 & 52:17:27.91 & 0.3288 & 24.22 & -16.4 & 0.17 & 1.37 & 0.10 & -.02 & 0.01 & [email protected] & [email protected] & [email protected] & 121@xmath2114 & [email protected] & 0.1 & a , c + 5 & 092 - 1375 & 14:17:26.699 & 52:27:19.48 & 0.3315 & 23.84 & -16.6 & 0.39 & 1.54 & 0.15 & 0.06 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.1 & a , c + 6 & 134 - 2322 & 14:17:05.547 & 52:21:18.17 & 0.3319 & 21.28 & -18.3 & 1.02 & 2.39 & 0.34 & 0.05 & 0.15 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.7 & a + 7 & 142 - 4644 & 14:16:51.341 & 52:20:52.38 & 0.3322 & 21.52 & -18.8 & 0.47 & 3.62 & 0.00 & 0.03 & 0.01 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.2 & a + 8 & 223 - 2132 & 14:16:07.432 & 52:11:25.04 & 0.3359 & 23.16 & -17.2 & 0.48 & 1.16 & 0.04 & -.01 & 0.00 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.1 & a + 9 & 124 - 3864 & 14:17:14.790 & 52:21:50.38 & 0.3564 & 21.55 & -18.8 & 0.56 & 3.02 & 0.00 & 0.05 & 0.07 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.5 & a + 10 & 173 - 5210 & 14:16:36.664 & 52:17:39.89 & 0.3570 & 20.03 & -20.4 & 0.54 & 4.51 & 0.05 & 0.07 & 0.06 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.7 & a + 11 & 103 - 6061 & 14:17:27.434 & 52:26:08.93 & 0.3631 & 21.88 & -18.6 & 0.51 & 0.93 & 0.49 & 0.03 & 0.01 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.6 & a + 12 & 203 - 4851 & 14:16:21.680 & 52:14:14.78 & 0.3639 & 22.58 & -18.1 & 0.36 & 1.18 & 0.70 & 0.02 & 0.01 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.3 & a + 13 & 213 - 6640 & 14:16:13.664 & 52:13:20.44 & 0.3656 & 20.22 & -20.3 & 0.50 & 4.62 & 0.03 & 0.06 & 0.05 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.9 & a , b + 14 & 313 - 3557 & 14:15:11.917 & 52:01:10.88 & 0.3757 & 20.56 & -20.2 & 0.42 & 2.35 & 0.25 & 0.06 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.8 & a + 15 & 304 - 4028 & 14:15:17.787 & 52:01:23.63 & 0.3760 & 22.78 & -17.9 & 0.41 & 1.38 & 0.02 & 0.03 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.7 & a + 16 & 183 - 6415 & 14:16:30.453 & 52:16:42.93 & 0.3865 & 21.00 & -19.6 & 0.55 & 7.06 & 0.00 & 0.14 & 0.07 & 109@xmath21150 . & [email protected] & [email protected] & [email protected] & [email protected] & 3.6 & a , b + 17 & 193 - 5319 & 14:16:24.702 & 52:15:23.83 & 0.3929 & 20.62 & -20.1 & 0.46 & 4.45 & 0.03 & 0.06 & 0.07 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.8 & a + 18 & 092 - 1962 & 14:17:26.421 & 52:27:06.04 & 0.4261 & 19.49 & -21.5 & 0.40 & 6.35 & 0.03 & 0.20 & 0.15 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.3 & a + 19 & 304 - 3354 & 14:15:17.610 & 52:00:56.86 & 0.4263 & 20.57 & -20.6 & 0.29 & 1.90 & 0.51 & 0.09 & 0.12 & 113@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 12.7 & a + 20 & 223 - 6341 & 14:16:07.359 & 52:12:07.83 & 0.4263 & 21.94 & -19.2 & 0.29 & 2.71 & 0.06 & 0.06 & 0.07 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.5 & a + 21 & 292 - 3870 & 14:15:14.882 & 52:03:47.19 & 0.4266 & 20.90 & -20.2 & 0.36 & 2.98 & 0.00 & 0.19 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 4.2 & a + 22 & 093 - 6667 & 14:17:34.454 & 52:27:25.05 & 0.4306 & 23.08 & -18.0 & 0.34 & 3.07 & 0.01 & -.00 & 0.01 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.5 & a + 23 & 164 - 3515 & 14:16:47.287 & 52:17:57.62 & 0.4320 & 21.74 & -19.3 & 0.42 & 3.23 & 0.01 & 0.02 & 0.02 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.2 & a + 24 & 164 - 2417 & 14:16:46.114 & 52:17:53.61 & 0.4323 & 21.30 & -19.8 & 0.38 & 3.86 & 0.08 & 0.08 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.4 & a + 25 & 074 - 4757 & 14:17:48.165 & 52:27:48.29 & 0.4323 & 21.49 & -19.6 & 0.40 & 4.82 & 0.03 & 0.15 & 0.09 & 103@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 3.1 & a + 26 & 094 - 1054 & 14:17:31.025 & 52:25:24.32 & 0.4331 & 19.66 & -21.3 & 0.54 & 5.37 & 0.10 & 0.09 & 0.11 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.8 & a + 27 & 313 - 7545 & 14:15:09.713 & 52:01:46.44 & 0.4503 & 22.70 & -18.4 & 0.42 & 2.54 & 0.01 & 0.01 & 0.01 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.4 & a , b + 28 & 183 - 4770 & 14:16:36.632 & 52:16:38.91 & 0.4691 & 21.38 & -19.9 & 0.41 & 1.85 & 0.56 & 0.05 & 0.06 & 121@xmath2111 & [email protected] & [email protected] & 126@xmath2113 & [email protected] & 7.7 & a + 29 & 223 - 7714 & 14:16:04.136 & 52:12:15.01 & 0.4711 & 21.72 & -19.4 & 0.53 & 2.12 & 0.31 & 0.02 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.9 & a + 30 & 062 - 1570 & 14:17:46.321 & 52:30:43.30 & 0.4770 & 21.38 & -19.9 & 0.45 & 2.49 & 0.34 & 0.05 & 0.06 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.2 & a + 31 & 162 - 5547 & 14:16:37.359 & 52:18:33.42 & 0.4776 & 21.24 & -20.0 & 0.48 & 2.03 & 0.03 & 0.06 & 0.05 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 4.5 & a + 32 & 083 - 6536 & 14:17:37.724 & 52:28:28.13 & 0.4798 & 23.75 & -17.6 & 0.45 & 0.44 & 0.26 & 0.03 & -.02 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.2 & a + 33 & 212 - 2260 & 14:16:08.191 & 52:13:03.10 & 0.4832 & 21.43 & -19.9 & 0.42 & 2.77 & 0.14 & 0.08 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.9 & a + 34 & 303 - 1256 & 14:15:18.750 & 52:01:58.57 & 0.4850 & 22.09 & -19.3 & 0.41 & 5.05 & 0.52 & 0.05 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.0 & a + 35 & 303 - 3546 & 14:15:17.146 & 52:02:18.14 & 0.4850 & 20.52 & -20.7 & 0.51 & 4.96 & 0.10 & 0.04 & 0.06 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 4.6 & a + 36 & 183 - 1153 & 14:16:35.713 & 52:16:00.42 & 0.5070 & 20.74 & -20.8 & 0.39 & 5.96 & 0.42 & 0.09 & 0.13 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.9 & a , b + 37 & 262 - 5149 & 14:15:33.122 & 52:06:56.39 & 0.5084 & 23.14 & -18.3 & 0.46 & 2.64 & 0.08 & 0.02 & -.03 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.8 & a + 38 & 114 - 3114 & 14:17:19.113 & 52:23:47.26 & 0.5470 & 21.59 & -20.1 & 0.46 & 3.43 & 0.03 & 0.03 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.1 & a + 39 & 283 - 3961 & 14:15:31.425 & 52:04:46.65 & 0.5586 & 23.57 & -18.1 & 0.51 & 1.07 & 0.63 & 0.04 & 0.03 & [email protected] & [email protected] & [email protected] & 239@xmath21111 & [email protected] & 1.8 & a + 40 & 082 - 1064 & 14:17:33.800 & 52:28:19.15 & 0.6039 & 23.26 & -18.7 & 0.41 & 1.36 & 0.18 & -.01 & 0.06 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.4 & a + 41 & 282 - 2474 & 14:15:22.654 & 52:05:04.57 & 0.6315 & 21.52 & -20.6 & 0.46 & 2.35 & 0.30 & 0.03 & 0.05 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.6 & a + 42 & 282 - 3252 & 14:15:22.320 & 52:04:41.62 & 0.6317 & 22.33 & -19.8 & 0.34 & 3.09 & 0.01 & 0.08 & 0.07 & 118@xmath2111 & [email protected] & [email protected] & 141@xmath2112 & [email protected] & 4.7 & a + 43 & 212 - 6648 & 14:16:03.907 & 52:12:41.35 & 0.6371 & 23.66 & -18.5 & 0.36 & 1.92 & 0.02 & 0.05 & 0.09 & 117@xmath2112 & [email protected] & 152@xmath2114 & 560@xmath21110 & [email protected] & 3.1 & a + 44 & 093 - 6526 & 14:17:30.101 & 52:27:15.58 & 0.6431 & 23.45 & -18.7 & 0.38 & 1.67 & 0.37 & 0.01 & 0.04 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.7 & a + 45 & 063 - 5323 & 14:17:49.728 & 52:30:31.88 & 0.6437 & 21.27 & -20.9 & 0.32 & 4.29 & 0.00 & 0.19 & 0.14 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 7.1 & a + 46 & 293 - 4412 & 14:15:19.779 & 52:03:30.50 & 0.6462 & 22.90 & -19.3 & 0.30 & 6.21 & 0.00 & 0.04 & 0.05 & 175@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 2.8 & a + 57 & 262 - 3751 & 14:15:34.530 & 52:07:00.73 & 0.6497 & 20.74 & -21.3 & 0.50 & 4.68 & 0.10 & 0.18 & 0.14 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.7 & a + 48 & 284 - 3046 & 14:15:30.071 & 52:03:24.75 & 0.6506 & 23.35 & -18.8 & 0.46 & 1.08 & 0.77 & 0.00 & 0.02 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.1 & a + 49 & 172 - 5435 & 14:16:31.329 & 52:17:11.67 & 0.6605 & 22.55 & -19.6 & 0.48 & 3.61 & 0.01 & 0.08 & 0.03 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.3 & a + 50 & 152 - 1633 & 14:16:48.315 & 52:19:38.62 & 0.6736 & 21.80 & -20.5 & 0.44 & 6.99 & 0.02 & 0.06 & 0.09 & 119@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 7.2 & a + 51 & 063 - 7209 & 14:17:47.868 & 52:30:47.38 & 0.6760 & 21.01 & -21.3 & 0.33 & 3.73 & 0.49 & 0.09 & 0.17 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 17.2 & a + 52 & 164 - 3859 & 14:16:48.501 & 52:17:15.55 & 0.6828 & 22.42 & -19.9 & 0.47 & 4.62 & 0.02 & 0.03 & 0.05 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.3 & a + 53 & 292 - 4369 & 14:15:14.367 & 52:03:45.35 & 0.6926 & 20.61 & -21.7 & 0.47 & 3.41 & 0.58 & 0.07 & 0.09 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 19.6 & a + 54 & 282 - 6050 & 14:15:19.468 & 52:04:32.81 & 0.6946 & 22.70 & -19.7 & 0.39 & 2.50 & 0.01 & 0.08 & 0.12 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 4.3 & a + 55 & 153 - 3721 & 14:16:51.198 & 52:19:47.44 & 0.7043 & 21.07 & -21.4 & 0.32 & 3.19 & 0.25 & 0.10 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 16.9 & a + 56 & 292 - 7343 & 14:15:11.836 & 52:03:14.00 & 0.7450 & 23.56 & -19.0 & 0.29 & 1.64 & 0.02 & 0.03 & 0.07 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.4 & a + 57 & 084 - 6809 & 14:17:42.554 & 52:27:29.20 & 0.7467 & 21.92 & -20.6 & 0.46 & 6.54 & 0.09 & 0.10 & 0.12 & 121@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 8.0 & a + 58 & 182 - 7536 & 14:16:22.681 & 52:15:59.33 & 0.7505 & 22.73 & -19.8 & 0.40 & 3.86 & 0.04 & 0.07 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 2.4 & a + 59 & 292 - 6724 & 14:15:12.920 & 52:02:56.62 & 0.7659 & 21.93 & -20.7 & 0.47 & 3.37 & 0.09 & 0.06 & 0.10 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 3.7 & a + 60 & 134 - 0967 & 14:17:05.035 & 52:20:31.42 & 0.7885 & 22.83 & -19.9 & 0.26 & 3.76 & 0.00 & 0.14 & 0.07 & 117@xmath2112 & [email protected] & [email protected] & 225@xmath2113 & [email protected] & 6.9 & a + 61 & 142 - 4838 & 14:16:51.266 & 52:20:45.96 & 0.8077 & 20.18 & -22.6 & 0.28 & 5.17 & 0.25 & 0.16 & 0.09 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 18.5 & a + 62 & 174 - 3527 & 14:16:40.994 & 52:16:35.72 & 0.8096 & 21.59 & -21.2 & 0.46 & 3.19 & 0.00 & 0.10 & 0.18 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 13.2 & a + 63 & 152 - 3226 & 14:16:46.841 & 52:19:28.70 & 0.8127 & 22.52 & -20.3 & 0.34 & 4.42 & 0.42 & 0.05 & 0.10 & 109@xmath2112 & [email protected] & [email protected] & [email protected] & [email protected] & 5.4 & a + 64 & 153 - 6078 & 14:16:56.673 & 52:20:22.44 & 0.8128 & 20.98 & -21.8 & 0.48 & 6.87 & 0.06 & 0.14 & 0.15 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 58.3 & a + 65 & 092 - 7832 & 14:17:20.880 & 52:26:23.36 & 0.6824 & 23.57 & -18.7 & 0.29 & 1.43 & 0.54 & 0.00 & 0.10 & [email protected] & [email protected] & 117@xmath2114 & 237@xmath2114 & [email protected] & 8.2 & d + 66 & 203 - 3109 & 14:16:17.618 & 52:13:49.43 & 0.6848 & 22.41 & -19.9 & 0.34 & 5.60 & 0.02 & 0.10 & 0.08 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 4.9 & d + + + + + + + lccrrrr [ fittable.tab ] k92+nfgs+kiss & @xmath430.09 & 133 & -0.063 ( 0.007 ) & 7.60 & + 1.0 & + 0.07 + k92+nfgs & @xmath430.09 & 54 & -0.052 ( 0.009 ) & 7.74 & 0.0 & 0.0 + dgss & 0.29 - 0.40 & 15 & -0.056 ( 0.024 ) & 7.56 & -1.6 & -0.10 + dgss+kz99 & 0.40 - 0.60 & 24 & -0.082 ( 0.014 ) & 7.02 & -2.0 & -0.12 + dgss+cl01 & 0.60 - 0.82 & 33 & -0.128 ( 0.012 ) & 6.04 & -2.4 & -0.15 + lrr [ costable.tab ] 0.1 & 1.30 & 12.15 + 0.2 & 2.43 & 11.02 + 0.4 & 4.28 & 9.17 + 0.6 & 5.71 & 7.74 + 0.8 & 6.82 & 6.62 + 1.6 & 9.49 & 3.95 + 3.2 & 11.50 & 1.95 + 6.4 & 12.61 & 0.83 + 10 & 13.0 & 0.45 +	using spectroscopic data from the deep extragalactic evolutionary probe ( deep ) groth strip survey ( dgss ) , we analyze the gas - phase oxygen abundances in the warm ionized medium for 64 emission - line field galaxies in the redshift range @xmath0 . these galaxies comprise a small subset selected from among 693 objects in the dgss . they are chosen for chemical analysis by virtue of having the strongest emission lines . oxygen abundances relative to hydrogen range between @xmath1 with typical internal plus systematic measurement uncertainties of 0.17 dex . the 64 dgss galaxies collectively exhibit an increase in metallicity with b - band luminosity , i.e. , an l - z relation like that seen among local galaxies . using the dgss sample and local galaxy samples for comparison , we searched for a `` second parameter '' which might explain some of the dispersion seen in the l - z relation . parameters such as galaxy color , emission line equivalent width , and effective radius were explored but found to be uncorrelated with residuals from the the mean l - z relation . subsets of dgss galaxies binned by redshift also exhibit l - z correlations with slopes and zero - points that evolve smoothly with redshift . dgss galaxies in the highest redshift bin ( @xmath2 ) are brighter , on average , by @xmath3 mag at fixed metallicity compared to the lowest dgss redshift bin ( @xmath4 ) and brighter by up to @xmath5 mag compared to local ( @xmath6 ) emission - line field galaxies . alternatively , dgss galaxies in the highest redshift bin ( @xmath2 ) are , on average , 40% ( 0.15 dex ) more metal - poor at fixed luminosity compared to local ( @xmath6 ) emission - line field galaxies . for @xmath7 galaxies , the offset from the local l - z relation is greatest for objects at the low - luminosity ( @xmath8 ) end of the sample and vanishingly small for objects at the high - luminosity end of the sample ( @xmath9 ) . we compare these data to simple single - zone exponential - infall pgase2 models which follow the chemical and luminous evolution of galaxies from formation to @xmath10 . a narrow range of model parameters can qualitatively produce the slope of the l - z relation and the observed evolution of slope and zero - point with redshift when at least two of the following are true : 1 ) low - mass galaxies have lower effective chemical yields than massive galaxies , 2 ) low - mass galaxies assemble on longer timescales than massive galaxies , 3 ) low - mass galaxies began the assembly process at a later epoch than massive galaxies . the single - zone models do a reasonable job of reproducing the observed evolution for the low - luminosity galaxies ( @xmath11 ) in our sample , but fail to predict the relative lack of evolution in the l - z plane observed for the most luminous galaxies ( @xmath12 ) . more realistic multi - zone models will be required to explain the chemo - luminous evolution of large galaxies .
the formation of galaxies and their systems ( ` haloes ' ) is known to be more intricate than its simplified rendering in the context of spherical collapse models . the process is characterised by gradual growth of the system mass and evolution of its morphological properties through multiple merging events , evidence for which comes from observations and hydrodynamical simulations . in clusters , mergers affect also the evolution of intracluster ( ic ) gas density and temperature and their spatial profiles . on the theoretical side , studies of halo mergers and related issues began with the works of bond et al . ( 1991 ) and lacey & cole ( 1993 ) , who developed the theory of _ excursion sets _ in the context of structure formation . this approach was originally devised by bond et al . in order to address the `` cloud - in - cloud '' problem , who showed that the press & schechter ( 1974 ) mass function , including the `` fudge factor '' of 2 , could be derived under certain assumptions . they also used their formalism to derive expressions for merger probabilities . lacey & cole ( hereafter lc ) used the excursion set formalism ( esf ) to extract such related quantities as halo merger rates , halo survival times , and halo formation times . the nfw concentration parameter ( navarro , frenk , & white 1995 ) , which characterises dark matter ( dm ) distribution in a halo is known from n - body simulations to be correlated with its formation time ( e.g. jing 2000 ; bullock et al . 2001 ; zhao et al . 2003 ) due to the fact that haloes which form earlier are likely to have more condensed cores , reflecting the higher background density of the universe . we use this inferred correlation , and the probability distribution function ( pdf ) of halo formation times , to construct pdfs of halo concentrations . this provides a convenient framework for a comparison of theoretically predicted values of the concentration parameter to observational results . our calculations of the formation time and concentration parameter pdfs are carried out in a standard @xmath0cdm cosmology , adopting the set of cosmological parameters,(@xmath3 , extracted from the 3-year wmap+wl data ( spergel et al . these parameters are selected for the sake of consistency with results obtained in other works , including calculations of the pdf of einstein radii . specifically , we show that the observed high einstein radius in the lensing cluster a1689 , whose value is roughly in the range @xmath4 ( e.g. , broadhurst & barkanna 2008 ) , is very improbable in the standard @xmath0cdm cosmology , but much more probable in a cosmological model characterised by an early component of dark energy , provided that its virial radius is larger ( within a plausible range of variance ) than what is predicted in the simple spherical collapse scenario . this paper is arranged as follows : in 2 we briefly detail the derivation of the pdfs of halo formation time and concentration parameter ; results are provided in 3 . the case of the lensing cluster a1689 is explored in 4 , followed by a discussion , 5 . in the lc formalism the cumulative distribution function ( cdf ) of halo formation times , which describes the probability that a halo of mass @xmath5 had a parent with mass in the range [ @xmath6 at redshift @xmath7 , with @xmath8 denoting the fraction of mass assembled through mergers by redshift @xmath7 is , @xmath9 here @xmath10 and @xmath11 denote the mass variance and critical density for spherical collapse , respectively , with @xmath11 extrapolated to @xmath12 , and @xmath13 is the conditional probability that a halo with mass @xmath5 at time @xmath14 had a progenitor of mass @xmath15 at time @xmath16 . put in the language of the esf , this is the probability that a random walk trajectory which reached the @xmath17 barrier at mass scale @xmath5 had traversed for the first time the @xmath18 barrier at mass scale @xmath15 . this probability is the key result from the esf which is used to determine the halo mass function and quantities related to its hierarchical evolution , such as the pdf of formation times . for primordial gaussian density fields , @xmath19ds_1.\ ] ] this analytical result derived from the esf is an outcome of the fact that smoothed gaussian fields still obey a gaussian distribution , and inherently includes the `` fudge factor '' of 2 which press & schechter have `` artificially '' introduced into their mass function so as to ensure that all the mass is included in haloes . the pdf of halo formation times can be now obtained by differentiating equation ( [ eq : cdf ] ) with respect to the redshift : @xmath20 the definition of halo formation time is somewhat arbitrary ; haloes are commonly considered to have formed when a fraction @xmath8 of their total mass was assembled , usually with either @xmath21 or @xmath22 . it is important to make the distinction between the observation and formation times : the former corresponds to the redshift at which the halo is observed to be ; the formation redshift is distributed in the range @xmath23 , and its pdf must satisfy @xmath24 the concentration parameter is defined as @xmath25 , where @xmath26 is the virial radius of the halo and @xmath27 is a characteristic inner radius , which roughly marks the transition radius from a @xmath28 to @xmath29 behaviour of the nfw density profile ( navarro , frenk , & white 1995 ) . results of n - body simulations suggest that haloes formed at higher redshifts tend to be associated with higher concentration parameters . this has been explained as an outcome of the formation of higher density cores when the background density was higher . this association directly implies also a correlation between the concentration parameter and formation time , evidence for which has been seen in n - body simulations of dm haloes ( e.g. zhao et al . 2003 , wechsler et al . 2002 ) . an analytic expression describing the correlation between the concentration parameter and formation time can be used in order to derive a pdf of halo concentrations , yielding the probability of finding a given concentration parameter for a given mass , or the cpdf of finding a concentration higher than a given value . such an expression has been deduced ( wechsler et al . 2002 ) from a statistical sample of haloes identified in n - body simulations , @xmath30 where @xmath31 , and @xmath32 , @xmath33 are the scale factors at the time of observation and formation , respectively , a relation which ( according to wechsler et al . ) fits well in the entire mass range explored in their simulations . note that these authors have defined the characteristic formation time as the epoch in which the mass accretion rate falls below some specified value . this differs from the formation time definition in the lc formalism , where it is defined as the time at which a constant fraction @xmath8 of the halo mass has been accumulated . to assess the impact of the different definition of the formation time , wechsler et al . repeated the analysis for merger trees produced by the esf and compared the resulting halo distribution as function of the formation epoch @xmath33 . their results suggest that the esf formalism predicts formation times higher ( i.e. later ) by a factor 1.25 . therefore , in order to use the scaling described in equation ( [ eq : cvir ] ) we should first divide the the esf - based formation times indicated by this factor : @xmath34 where @xmath35 is the formation redshift . using relation ( [ eq : cvir ] ) we finally obtain @xmath36 where @xmath37 . a pdf of the halo concentration parameter can now be derived by using the usual probability distribution transformation law , @xmath38 , such that @xmath39 whereas the corresponding cpdf - i.e. the probability that a halo has a concentration parameter higher than a given value - is calculated as @xmath40 . pdf of formation times of haloes with various masses at ( observation ) redshift @xmath41 were calculated using equation ( [ eq : pz ] ) , and are illustrated in fig . ( [ fig : ftdist ] ) . haloes were defined to have formed once they assembled a fraction of either @xmath21 ( left - hand panel ) , or @xmath22 ( right - hand panel ) of their current masses . in both panels of fig . ( [ fig : ftdist ] ) blue , red , black and orange curves correspond to mass scales of @xmath42 , and @xmath43 @xmath44 , respectively . the merger picture describes the formation of haloes as an ongoing process of mutual collapse of subhaloes . since low - mass subhaloes are more abundant than high - mass subhaloes , haloes of lower masses are more likely to form at higher redshifts . these gradually merge with other subhaloes to form increasingly larger structures , as can be clearly seen in fig . ( [ fig : ftdist ] ) , where low - mass haloes ( @xmath45 ) form relatively early and at high abundances . the enhancement of the formation time probability and the consequent higher abundances . of subhaloes of this mass scale , imply higher likelihood for a merger to occur and higher - mass objects to form . the latter become progressively more common , and their active formation time - defined as the epoch at which the formation time pdf attains a maximum - peaks at a lower redshift with respect to the active formation epoch of their progenitors . high abundance of the lowest - mass subhaloes enhances the merger probabilities ; as a result these subhaloes grow less common with time . consequently , the pdf of their formation time is reduced with decreasing redshift , and is eventually dominated by the pdf of their higher - mass products . the process goes on , with ever - increasing mass objects becoming more abundant , accompanied by a reduction of the pdf for lower - mass haloes . this is apparent in either left - hand panel of the figure , where the pdf of formation times peak at lower redshifts with increasing mass . the fall of the curves from the active formation time towards lower redshifts can thus be explained by the fact that low - mass subhaloes hierarchically merge to yield higher - mass subhaloes , thereby diminishing the probability of low - mass objects to form at low redshifts . the corresponding progression in the high - mass halo range can be attributed to the fact that merger rates are substantially reduced with increasing mass . this , too , is in accord with expectations from the esf , as was demonstrated by , e.g. , lc . the earlier formation times implied by the @xmath21 case with respect to the @xmath22 case are clearly reflected in the plots ; the active formation time lies at manifestly higher redshifts in the @xmath21 case . whereas it peaks at approximately @xmath46 for haloes of @xmath47 @xmath44 for @xmath22 , the corresponding formation times with @xmath21 are @xmath48 , i.e. , at least twice as high . it can also be seen that the pdfs for the @xmath21 case have more pronounced peaks and fall more steeply on either side of the peak than for @xmath22 , where the curves are decidedly flatter at lower redshifts than for @xmath21 . this clearly is due to the fact that haloes can assemble half of their present mass at considerably earlier stages of their evolution than it takes to assemble 3/4 of their present mass . thus , their formation times peak at higher redshifts , and the likelihood of merging and forming subhaloes of higher masses increases at higher redshifts as well , resulting in lower probabilities for low - mass haloes to form at low redshifts . we have repeated the calculations for an observation redshift of @xmath49 and @xmath21 , results for which are illustrated in fig . ( [ fig : ftdistrat ] ) . obviously , the pdfs shift to the right , towards higher redshifts , but retain their general shape . results of the pdf and cdpf of halo concentrations are presented in fig . ( [ fig : cp ] ) for observation redshifts of @xmath41 and @xmath49 . for @xmath41 , haloes with @xmath50 @xmath44 peak at concentrations @xmath51 , respectively . the corresponding peak concentrations for @xmath49 are @xmath52 , i.e. somewhat lower , since the higher observation redshift limits the redshift range available for halo formation . the cpdfs plotted in the right - hand panels of fig . ( [ fig : cp ] ) provide a convenient means of determining the probability of finding a concentration higher than a given value . for example , at @xmath41 the probability of finding haloes of @xmath50 @xmath44 with concentrations higher than @xmath53 are @xmath54 . for @xmath49 the corresponding values are @xmath55 . clearly , the probability for a high value of @xmath56 in a @xmath57 cluster is quite low in the standard @xmath0cdm cosmology . recently , broadhurst & barkanna ( 2007 , hereafter bb ) have reported the results of a comparative study of einstein radii calculated for a large sample of simulated clusters of masses @xmath58 , and those observed in the three massive clusters a1689 , c10024 - 17 , and a1703 . while the simulations yield einstein radii in the range @xmath59 , the measured values for these clusters are @xmath60 , and @xmath61 , respectively , a result which is seen by bb to constitute a challenge to the standard @xmath0cdm model . since we know all relevant properties of a1689 , namely its virial mass , virial radius , concentration parameter ( assuming an nfw dm profile ) , redshift , and the lensed source redshift , it is possible to construct a pdf of einstein radii for an a1689-like cluster . the equation governing the relation between the concentration parameter and the einstein radius is , assuming an nfw profile ( bb ) , @xmath62 where @xmath63 , @xmath64 , @xmath65 , and @xmath66 are the virial radius , critical density at redshift @xmath67 , overdensity at virialisation , and critical surface density , respectively . all these quantities are specified by bb , with the exception of the virial radius , for which we adopt the theoretical value found from the relation @xmath68 , resulting in @xmath69 , for @xmath70 . also , @xmath71 , where @xmath72 is the einstein radius , and @xmath73 the solution of eq . ( [ eq : gx ] ) provides the einstein radius as a function of @xmath56 , from which it is trivial to derive the angular einstein radius , @xmath74 , the ratio between the ( physical ) einstein radius and the angular diameter distance to the lens , the measured redshift of a1689 , @xmath75 . it remains to determine the pdf of the angular einstein radius . this can be accomplished using once more the transformation law of probability distribution functions : @xmath76 from which we obtain @xmath77 the first derivative is calculated analytically using eq . ( [ eq : cv2 ] ) , whereas a numerical calculation based on the @xmath78 relation in eq . ( [ eq : gx ] ) is used in order to infer the second derivative . results for the pdf and cdf of the angular einstein radius for an a1689-like cluster are presented in fig . ( [ fig : er ] ) . the results generated within the framework of the standard @xmath0cdm are represented by the black continuous curve ; the shaded areas represent the @xmath79 uncertainties in the cosmological parameters @xmath80 , @xmath81 , and @xmath82 . as is clear from the figure , large einstein radii ( @xmath83 ) have extremely low probabilities in the standard @xmath0cdm universe . in fact , the cumulative probability for an a1689-like cluster to induce an angular einstein radius larger than @xmath84 amounts to @xmath85 in this model ; the corresponding probability for the @xmath86 level increases to @xmath87 . non - standard cosmological models characterised by earlier evolution of the large scale structure may resolve the apparent discrepancy between measured and predicted values of @xmath72 , by virtue of earlier halo formation times , reflected in higher central halo concentrations and , therefore , higher probabilities for large einstein radii . for example , models based on positively skewed primordial density fluctuations lead to earlier growth of the large scale structure . early dark energy ( hereafter ede ) models provide another alternative for inducing earlier formation times by virtue of the modified cosmic dynamics implied by a non - negligible de component in the early universe . to explore this possibility , we have repeated our calculations for a specific ede model , with an early quintessence density parameter @xmath88 and equation of state coefficient @xmath89 at @xmath12 . these parameters are consistent with recent wmap results ( doran & robbers , 2006 ) . the other cosmological parameters were not modified , with the exception of @xmath90 , a value obtained by normalizing the cumulative halo mass function to the same number of halos generated in the standard @xmath0cdm model . complete details concerning the calculation of the relevant large scale quantities within the framework of ede models can be found in , e.g. , bartelmann , doran , & wetterich ( 2006 ) and sadeh , rephaeli & silk ( 2007 ) . the latter work was motivated by the possible need to have a higher level of cmb anisotropy induced by the sunyaev - zeldovich effect than is predicted in the standard @xmath0cdm model . results of the pdf and cdf of the einstein radius for a1698 are shown by the continuous orange curve of fig . ( [ fig : er ] ) . note that for the ede model the theoretically inferred virial radius reduces slightly to @xmath91 . the probability for an a1689-like cluster to produce an einstein radius larger than @xmath84 in this model is @xmath92 , i.e. , approximately two orders of magnitude higher than the corresponding probability in the @xmath0cdm model . the @xmath86 uncertainty ( which is not shown in the plot ) in the cosmological parameters would lead to even higher probabilities than those indicated by the orange curve of the ede model . however , these probabilities are still very low . the observationally inferred high concentration parameters associated with several massive clusters have been claimed to constitute a major challenge to the currently favoured standard cosmological model . such high concentrations are expected in haloes formed at relatively high redshifts , reflecting the high universal background density at that time . as einstein radii in lensing haloes increase monotonically with their concentration parameters , observed high radii also indicate earlier formation times . massive clusters form preferentially at low redshifts , as attested by results of both n - body simulations and semi - analytic calculations based on the excursion set theory . our assessment of the apparent discrepancy between theoretical predictions and measurements of the concentration parameter and einstein radii of high - mass clusters has been based on the assumption of nfw mass profiles . this is motivated by the explicit use of this profile in the analysis of n - body simulations ( wechsler et al . 2002 , neto et al . 2007 ) , which were used to derive the scaling relation between @xmath56 and @xmath93 equation ( [ eq : cv2 ] ) in these works , and had to be adopted in our analysis for consistency . the nfw profile is not the only profile that provides an acceptable fit to simulated haloes . it is quite possible that most of the discrepancy explored here stems from the insistence on using the nfw profile . even so , the fact that this profile is used extensively in the description of galaxies and clusters provides sufficient motivation to focus on its implications in the context of our work here . if indeed the apparent discrepancy is considered a serious difficulty for the standard model , then it is only reasonable to consider close alternatives to the standard @xmath0cdm model . earlier formation times are naturally predicted in non - standard cosmologies , such as those characterised by non - gaussian , positively skewed primordial density fluctuations , and early dark energy models . we have shown in this work that cosmological models incorporating an early de component may induce higher probabilities of finding large einstein radii in a1689 than in the standard @xmath0cdm model , although perhaps not sufficiently high to remove the apparent conflict between theory and observations . the calculation of the number of clusters whose masses and redshifts are comparable to or larger than those of a1689 , and which are expected to induce einstein radii larger than a given value , is not simple since this would require an integration over the formation time pdfs of all of these masses and redshifts . in addition to this technical issue , one would also need to draw an arbitrary source redshift so as to be able to calculate the corresponding angular diameter distances needed in eq . ( [ eq : gx ] ) . however , we can provide a crude estimate of the upper limit to this number by integrating the relevant ps mass functions - shown in fig . ( [ fig : mf ] ) for both models at observation redshifts @xmath41 and @xmath94 - over the mass range [ @xmath95 . with our specific choice of cosmological parameters , the standard @xmath0cdm and ede models described in this work predict @xmath96 ( @xmath97 for the @xmath86 level of @xmath82 - which agrees better with the corresponding figure inferred from large scale studies ) , and @xmath98 cumulative cluster counts with @xmath99 and @xmath100 , respectively . obviously , not all of these clusters have einstein radii larger than @xmath84 , as these depend also on the redshifts of the lensed objects . we can readily set an upper limit on the number of clusters with @xmath101 , whose mass and redshift are within these ranges , by multiplying the cumulative numbers by the probabilities of finding an a1689-like cluster with @xmath101 . as specified in the previous section , these probabilities are @xmath102 ( or @xmath103 at the @xmath86 level ) , and @xmath104 in the standard @xmath0cdm and ede models , respectively . thus , the estimated numbers of clusters are @xmath105 ( @xmath106 at the @xmath86 level ) and @xmath107 in the @xmath0cdm and ede models , respectively . it is important to realize that while our analysis of the pdf of the concentration parameter and einstein radius poses an appreciable difficulty for the standard large scale model , it can not be construed as insurmountable evidence against the validity of the standard @xmath0cdm model . in fact , there is a significant level of uncertainty in the relation between the formation time and concentration parameter , an uncertainty which we have not yet taken explicitly into account . many factors contribute to this uncertainty , including the reported variance in the scaling , as shown by wechsler et al . to be at the level of @xmath108 , such that , for example , a halo with mass @xmath109 would have a concentration @xmath110 , representing only the @xmath111 uncertainty level . the actual uncertainty may even be appreciably higher , due to the fact that the baryonic component was ignored in the cited n - body simulations , and due to the large degree or arbitrariness in the definition of the formation time . the higher than predicted value of @xmath112 is reflected more directly in the discrepancy between the theoretically predicted and measured values of @xmath63 . the solutions of eq . ( [ eq : gx ] ) , which yield the einstein radius as a function of concentration parameter , are very sensitive to the virial radius of the cluster . we have used the most probable theoretically - inferred values of @xmath63 in our calculations , @xmath113 and @xmath114 in the standard @xmath0cdm and ede models , respectively . but these values too could be uncertain ; we explore here the impact of a variance of @xmath2 in @xmath63 that would presumably reflect the intricacies of the process of hierarchical clustering through mergers and non - spherical collapse , which was shown ( e.g. del popolo 2002 ) to yield significantly lower virialisation overdensities than predicted in the spherical collapse scenario . for example , an isolated prolate spheroid of axial ratio @xmath115 , which is the mean ratio found in an n - body simulation in which 878 clusters were identified ( hennawi et al . 2007 ) , was shown by del popolo to have a virialisation overdensity of @xmath116 , lower by more than a factor @xmath117 than the corresponding quantity for a spehrical virialised region . ( although carried out in the context of an einstein de sitter universe , the reported results are not likely to change qualitatively in a @xmath0cdm model . ) for a given mass and redshift , @xmath68 , the lower overdensities obviously imply larger virial radii . a larger virial radius is also in accord with the results of a recent detailed , model - independent analysis of a1689 by lemze et al . ( 2008 ) , who ( deduced the mass profile and ) determined @xmath118 from joint fitting to extensive x - ray and lensing measurements . with @xmath70 we have @xmath119 , which is indeed higher by @xmath2 than the theoretically predicted value for this cluster . this has a very significant impact on the pdf and cpdf of the einstein radius , as demonstrated in fig . ( [ fig : er2 ] ) . with this larger virial radius , the probability for an a1689-like cluster to have @xmath120 is @xmath121 ( or @xmath122 at the @xmath86 level ) in the standard @xmath0cdm model , and @xmath123 in the ede model . these large differences originate in the larger einstein radii predicted by eq . ( [ eq : gx ] ) for the same concentration parameter , and give rise to substantially higher upper limits on the number of clusters expected to be detected with @xmath101 , @xmath124 ( at the @xmath86 level ) , and as many as @xmath125 in the standard @xmath0cdm and ede models , respectively . finally , we note that the above considerations indicating significantly higher halo concentrations and larger einstein radii in clusters may have significant ramifications for the non - standard , ede model : had the simulations been carried out within the framework of a specific ede model , the formation time - concentration scaling would have probably predicted higher concentrations , by virtue of the earlier formation of structure . here , too , the probabilities of finding haloes with high concentrations and einstein radii would be markedly higher , perhaps even prohibitively higher , so much so that this could turn out to be a significant constraint on the parameters of this model . useful discussions with dr . tom broadhurst are gratefully acknowledged . this work was supported by a grant from the israel science foundation . [ b]artelmann m. , doran m. , wetterich c. , 2006 , , 454 , 27 [ b]ond j.r . , cole s. , efstathiou g. , & kaiser n. , 1991 , , 379 , 440 [ b]roadhurst t.j . , barkanna r. , 2008 , preprint ( astro - ph/0801.1875 ) [ b]ullock j.s . , kolatt t.s . , sigad y. , somerville r.s . , kravtsov a.v . , klypin a.a . , primack j.r . , dekel a. , 2001 , , 321 , 559 [ d]el popoplo a. , 2002 , , 387 , 759 [ d]oran m. , robbers g. , 2006 , , 6 , 26 [ j]ing y.p . , 2000 , 535 , 30 [ l]acey c. , cole s. , 1993 , , 262 , 627 [ l]emze d. , barkana r. , broadhurst t.j . , rephaeli y. , 2007 , preprint ( astro - ph/0711.3908 ) [ n]avarro j.f . , frenk c.a . , & white s.d.m , 1995 , , 275 , 720 [ n]eto a.f . , gao l. , bett p. , cole s. , navarro j. , frenk c.s . , white s.d.m , springel v. , jenkins a. , 2007 , , 381 , 1450 [ p]ress w.h . & schechter p. , 1974 , , 187 , 425 [ s]adeh s. , rephaei y. , silk j. , 2007 , , 380 , 637 [ s]pergel d.n . 2007 , , 170 , 377 [ w]echsler r.h . , bullock j.s . , primack j.r . , kravtsov a.v . , & dekel a. , 2002 , , 568 , 52 [ z]hao d.h . , mo h.j . , ping y.p . , & brner g. , 2003 , , 339 , 12	we provide a quantitative assessment of the probability distribution function of the concentration parameter of galaxy clusters . we do so by using the probability distribution function of halo formation times , calculated by means of the excursion set formalism , and a formation redshift - concentration scaling derived from results of n - body simulations . our results suggest that the observed high concentrations of several clusters are quite unlikely in the standard @xmath0cdm cosmological model , but that due to various inherent uncertainties , the statistical range of the predicted distribution may be significantly wider than commonly acknowledged . in addition , the probability distribution function of the einstein radius of a1689 is evaluated , confirming that the observed value of @xmath1 is very improbable in the currently favoured cosmological model . if , however , a variance of @xmath2 in the theoretically predicted value of the virial radius is assumed , then the discrepancy is much weaker . the measurement of similarly large einstein radii in several other clusters would pose a difficulty to the standard model . if so , earlier formation of the large scale structure would be required , in accord with predictions of some quintessence models . we have indeed verified that in a viable early dark energy model large einstein radii are predicted in as many as a few tens of high - mass clusters . [ firstpage ] cosmology : large - scale structure of universe gravitational lensing galaxies : clusters : general
the history of the cosmological reionisation is one of the open problems of modern cosmology . questions like what causes the reionisation and how it proceeds are intimately related to the evolution of matter density fluctuations and to the formation of the first structures @xcite . the lack of observational data makes it quite difficult to answer these questions today . up to now , the available probes of the reionisation epoch are just a few , e.g. the gunn - peterson test @xcite and the cosmic microwave background ( cmb ) radiation , in particular its polarisation @xcite . the observation of the gunn - peterson effect , which probes the amount of neutral hydrogen , suggests that full reionisation was accomplished by @xmath1 @xcite . the cmb polarisation , in turn , measures the optical depth of the thomson scattering from the last scattering surface to the present epoch . the wmap 3 years data provide an optical depth @xmath2 @xcite suggesting that reionisation began around @xmath3 . in addition to those two probes , the observation of fluctuations of the 21 cm line background is expected to be one of the most promising methods to study reionisation @xcite . the 21 cm line corresponds to the energy of the hyperfine spin flip transition of neutral hydrogen atoms . after decoupling of the baryon temperature from the cmb temperature around @xmath4 , neutral hydrogen atoms absorb or emit a 21 cm line , depending on their temperature . since the 21 cm line is redshifted by the cosmological expansion , we can obtain redshift slices of the universe by choosing the frequency of observation . it is thus possible to follow the evolution of the intergalactic medium before and during reionisation directly . the 21 cm line observations have recently gained a lot of attention @xcite and several projects are designed for measuring the fluctuating line background ( e.g. mwa , lofar , ska ) . in the near future , we will therefore have different direct and complementary probes of reionisation . we thus expect that cross - correlations between them will provide even more information than their respective auto - correlations . this is the case especially for the cross - correlation between the cmb temperature anisotropies and the 21 cm line fluctuations on large scales ( @xmath5 * ? ? ? * ; * ? ? ? * ) and on small scales ( @xmath6 * ? ? ? * ; * ? ? ? * ; * ? ? ? since the observations of the 21 cm lines are expected to be dominated by foregrounds and noise @xcite , calculating the cross - correlation is indeed a good way to extract information on reionisation . in particular on large scales , the 21 cm line fluctuations are correlated with the cmb doppler temperature anisotropies which are due to the motion of ionised baryons . @xcite showed that the maximum amplitude of the cross - correlation is reached at the redshift when the ionised fraction is one half , which potentially provides a way to constrain double reionisation models ( e.g. * ? ? ? * ) . in the present study , we examine the cross - correlation between the cmb @xmath0-mode polarisation ( a more effective probe than the temperature anisotropies ) and the fluctuations of the 21 cm line background on large scales . measuring @xmath0-mode polarisation is one of the present challenging observations for cmb experiments . however , projects like planck and cmb - pol will provide precise cmb polarisation maps in the near future . similarly , cosmological observation of 21 cm line background is still belongs to the future , but lofar and mwa are being completed and will soon provide us with maps of neutral hydrogen at high redshifts . hence , it is useful to investigate what information on reionisation we can extract from the cross - correlation of those maps . indeed , we show here that the cross - correlations are a potentially powerful tool to constrain the history of cosmic reionisation . this article is organised as follows . in sec . ii , we provide the analytic formula of the cross - correlation between the cmb @xmath0-mode polarisation and the fluctuations of the 21 cm line background . in sec . iii , we show and discuss the results of the cross - correlation . vi is devoted to the conclusions . throughout the paper , we use wmap 3-year values for the cosmological parameters , i.e. @xmath7 , @xmath8k , @xmath9 and @xmath10 @xcite for a flat cosmology . we also set the speed of light to @xmath11 . the cmb polarisation is best described by the stokes parameters , @xmath12 and @xmath13 . since the stokes parameters have spin 2 , they can be decomposed onto the plane waves and spin-2 spherical harmonics @xmath14 @xcite , @xmath15 here , @xmath16 , @xmath17 , @xmath18 correspond to the scalar , vector and tensor types . in this paper , we only consider @xmath16 modes , because they dominate other polarisation modes at the scales we are interested in . the cmb polarisation components , @xmath19 and @xmath20 , are obtained from the boltzmann equations for cmb . according to those , @xmath20 = 0 for all @xmath21s and the evolution of @xmath19 is given by the following integral solution , @xmath22 where @xmath23 is the @xmath16 source term due to thomson scattering , @xmath24 is the differential optical depth for thomson scattering in conformal time , given by @xmath25 with the electron number density @xmath26 , the cross section of thomson scattering @xmath27 and the scale factor @xmath28 normalised to the present epoch , and @xmath29 is the optical depth between the conformal times @xmath30 and @xmath31 . the source term @xmath23 is related to the initial gravitational potential @xmath32 via the transfer function @xmath33 , @xmath34 we provide the detailed calculation of the transfer function in the appendix . we also define the @xmath0-modes in the direction @xmath35 on the sky by @xmath36 the observed brightness temperature of the 21 cm lines in a direction @xmath35 and at a frequency @xmath37 is given as in @xcite by @xmath38 here , the conformal time @xmath39 refers to the redshift @xmath40 that satisfies @xmath41 with @xmath42 being the frequency corresponding to the 21 cm wavelength . the optical depth for the 21 cm line absorption is @xmath43 ^{1/2 } , \label{eq : tau21}\ ] ] where @xmath44 is the fraction of neutral hydrogen , and @xmath45 is the spin temperature given by the ratio of the number density of hydrogen atoms in the excited state to that of hydrogens in the ground state . equations ( [ eq:21cmline ] ) and ( [ eq : tau21 ] ) show that the observed brightness temperature of the 21 cm lines will reflect the baryon density fluctuations @xmath46 and the fluctuations of the ionised fraction @xmath47 . we can rewrite eq . ( [ eq:21cmline ] ) in the linear approximation for both @xmath46 and @xmath48 as @xmath49(1+\delta_b ) t_{0 } , \label{eq:21cmlinefl}\ ] ] where we @xmath50 is the ionised fraction and @xmath51 ^{1/2 } { \rm mk}.\ ] ] here , we assume that @xmath52 is much larger than the cmb temperature which is valid soon after the beginning of the reionisation @xcite . after performing the fourier transform of eq . ( [ eq:21cmlinefl ] ) , we obtain the 21 cm line fluctuations , @xmath53(1+\delta_b ) t_{0 } \nonumber \\ & = & t_{0 } \int \frac{dk ^3}{(2 \pi)^3 } \left [ ( 1- \bar x_e ) \delta_{\rm b } - \bar x_e \delta_x \right ] \exp{[- i { \bm k } \cdot ( ( \eta -\eta_{\rm obs } ) \hat { \bm n})]}. \label{eq : fluc-21}\end{aligned}\ ] ] therefore , the 21 cm line fluctuation map at the frequency @xmath37 can be described by @xmath54 j _ \ell(k ( \eta_0-\eta_{\rm obs } ) ) y_\ell ^0 ( \hat{\bm n } ) . \label{eq:21cm - map}\ ] ] where we applied rayleigh s formula to eq . ( [ eq : fluc-21 ] ) , @xmath55 where @xmath56 is measured in the coordinate system where @xmath57 is @xmath58 . in eq . ( [ eq:21cm - map ] ) , we included the factor @xmath59 to account for the enhancement of the fluctuation amplitude due to the redshift distortion ( kaiser effect @xcite ) on the 21 cm line fluctuations where @xmath60 and @xmath61 with @xmath62 the linear growth factor of baryon fluctuations @xcite . the angular power spectrum is defined as the average of the spherical harmonic coefficients @xmath63 over the @xmath64 @xmath65-values , @xmath66 where the @xmath63 are defined for an arbitrary sky map @xmath67 as @xmath68 from eqs . ( [ eq : emode - map ] ) and ( [ eq:21cm - map ] ) , we can thus derive the angular power spectrum of the cross - correlation between @xmath0-modes and 21 cm line fluctuations @xmath69 \right \rangle } { ( 2 \ell + 1 ) } j_\ell ( k ( \eta_0 -\eta_\nu ) ) \nonumber \\ & = & -\frac{3}{\pi } t_0 \sqrt{\frac{(\ell+2)!}{(\ell-2 ) ! } } \int { dk } \int d \eta k^2 \dot \tau e^{-\tau } d_e ( k , \eta ) \left [ \frac{4}{3 } ( 1- \bar x_e ) p_{\phi \delta_{\rm b}}- \bar x_e p_{\phi \delta_x } \right ] \frac{j_\ell ( k ( \eta_0 -\eta_{\rm obs } ) ) j_\ell ( k ( \eta_0 -\eta))}{(k ( \eta_0 -\eta))^2 } , \label{eq : cross - cl}\end{aligned}\ ] ] where @xmath70 and @xmath71 are the power spectra of the cross - correlation between the gravitational potential and the baryon density fluctuations , and between the gravitational potential and the fluctuations of the ionised fraction , respectively . in the second line , we use @xmath72 for the matter dominated epoch . the cosmological linear perturbation theory provides the relation between the baryon density fluctuations and the initial gravitational potential ( e.g. * ? ? ? * ) , @xmath73 where @xmath74 is the transfer function of baryons . the power spectrum @xmath70 can thus be written in terms of the initial power spectrum of the gravitational potential @xmath75 as @xmath76 although the reionisation process is not well - known , we can reasonably expect that there is a relation between the fluctuations of the ionised fraction and that of the matter density . ionising sources are formed in dense regions . they ionise the surrounding medium with an efficiency that depends on the density of the medium . this translates into two possible cases : one where ionised fluctuations and matter over - densities coincide , and one where ionised fluctuations and matter density are anti - biaised ( e.g. * ? ? ? following @xcite , we assume that the fluctuations of the ionised rate are associated with the matter density contrast using the press - schechter description @xcite . as a result , the power spectrum @xmath77 is given by @xmath78 d_{\rm m}(k , \eta ) k^2 p_{\phi},\ ] ] where @xmath79 is the transfer function of matter ( both dark and baryonic ) , @xmath80 is the average bias of dark matter halos more massive than @xmath81 , the minimum mass of the source of ionising photons @xmath82 where @xmath83 is the variance of the density fluctuations smoothed with a top - hat filter of the scale corresponding to a mass @xmath84 , and @xmath85 is the fraction of matter collapsed into halos with @xmath86 . in this paper , we choose @xmath81 such that the halo virial temperature is @xmath87 k. this choice corresponds to the assumption that the ionising sources form in dark matter halos where the gas cools efficiently via atomic cooling . the parameter @xmath88 describes the reionisation regime we are interested in . for @xmath89 , we are in the `` photon - counting limit '' case where recombinations are not important and where the progress of the reionisation depends on the number of ionising photons only . the over - dense regions contain more collapsed objects which are sources of ionising photons . therefore , in this case , ionisation in over - dense regions is easier than in under - dense regions . on the contrary , @xmath90 indicates the strmgren limit " case where ionisation is balanced by recombination . although the over - dense regions contain more sources of ionising photons , the recombination rate in over - dense regions is higher than in under - dense regions . hence , over - dense regions in the @xmath90 case have a lower ionised fraction than in the @xmath89 case ( for details , see * ? ? ? * ) . finally , in order to calculate eq . ( [ eq : cross - cl ] ) , we need the evolution of the mean neutral hydrogen fraction for which we use a simple parameterisation based on two key quantities , the reionisation epoch ( defined as the redshift at which the ionised fraction equals 0.5 ) , @xmath91 , and the duration of reionisation , @xmath92 , @xmath93}. \label{eq : xh}\ ] ] in the following we will vary both @xmath91 and @xmath92 , and investigate their effect on the cross - correlation signal . as one can see in eq . ( [ eq : cross - cl ] ) , the angular power spectrum @xmath94 of the cross - correlation is a function of several key parameters that encode the prominent details of the reionisation history . consequently , and as we show here , the global shape of the @xmath94 as well as the amplitude and redshift dependence of its first peak are features that can be traced back to the reionisation characteristics . first it is worth noting that because of the double integration of the spherical bessel functions in eq . ( [ eq : cross - cl ] ) , the angular power spectrum reflects the source term @xmath95 , namely the quadrupole term of the cmb temperature anisotropy , at @xmath96 where @xmath97 satisfies @xmath98 . accordingly , the angular power spectrum of the cross - correlation exhibits its first peak at a multipole @xmath99 which corresponds to the angular separation of the quadrupole at @xmath96 . the oscillations at higher @xmath21s ( @xmath100 ) are due to the free streaming of the quadrupole at redshifts higher than @xmath40 . we first investigate the effect of the duration of reionisation @xmath92 . we present in the left panel of fig . [ fig : clre10 ] three cases : one corresponding to a instantaneous reionisation @xmath101 and two with longer durations ( @xmath102 and 2 ) . we compute the angular spectra of the cross - correlation between 21 cm line fluctuations and the cmb @xmath0-mode fluctuations for those different durations but for the same given epoch of reionisation , @xmath103 , an observing redshift , @xmath104 , and in the case of the `` photon - counting limit '' ( @xmath89 ) . the difference in the duration of reionisation induces both a minor and a major effects on the spectra . the minor effect is the observed shift of the peak position to higher @xmath21s with increasing durations . since a longer duration increases the contribution of the visibility function @xmath105 in eq . ( [ eq : cross - cl ] ) , which we plot in the right panel of fig . [ fig : clre10 ] , the contributions from redshifts larger than @xmath96 accumulate in the calculation of the integration . however , spherical bessel functions have a steep cutoff after @xmath39 so that the contributions from epochs earlier than @xmath39 are negligible . therefore , additional effective contributions come from only the former epoch @xmath106 , and make the peak shift toward higher @xmath21 modes . the major effect of @xmath92 exhibited in the plot ( left panel of fig . [ fig : clre10 ] ) is the damping at high @xmath21s with increasing durations of reionisation . this occurs again due to the increase of the amplitude of the visibility function at @xmath107 . at high @xmath21s , the source term @xmath23 exhibits rapid oscillations . since the tail of the visibility function is larger , the range of the integral is larger , and it thus encompasses more oscillations . as a result , many more cancellations occur which damp the overall power . we note that the damping depends on the duration of the reionisation which is an advantage of observing the cross - correlation signal . as a matter of fact , it is difficult to see differences among cmb polarisation spectra for different durations , because the thomson optical depth is not strongly sensitive to this parameter . however the plots in the left panel of fig . [ fig : clre10 ] , clearly show that the damping of angular spectrum of the cross - correlation is sensitive to the duration of reionisation making it a potential tool for constraining this parameter of the reionisation history . cc -mode polarisation and 21 cm line fluctuations . we set @xmath108 , @xmath109 and @xmath89 for all curves . the dotted line represents the angular spectrum for @xmath110 . the solid and dashed lines are for @xmath102 and @xmath111 , respectively . for comparison the thin solid line , shows the angular power spectrum of the cmb @xmath0-modes for @xmath108 and @xmath102 . the right panel is the evolution of the visibility function @xmath105 in the same reionisation model as the left panel . the dotted , solid and dashed lines are for @xmath110 , @xmath102 and @xmath111 , respectively.,height=188 ] -mode polarisation and 21 cm line fluctuations . we set @xmath108 , @xmath109 and @xmath89 for all curves . the dotted line represents the angular spectrum for @xmath110 . the solid and dashed lines are for @xmath102 and @xmath111 , respectively . for comparison the thin solid line , shows the angular power spectrum of the cmb @xmath0-modes for @xmath108 and @xmath102 . the right panel is the evolution of the visibility function @xmath105 in the same reionisation model as the left panel . the dotted , solid and dashed lines are for @xmath110 , @xmath102 and @xmath111 , respectively.,height=188 ] -mode polarisation and 21 cm line fluctuations . the dotted line is for @xmath112 . the solid and dashed lines are for @xmath113 and @xmath114 , respectively . we use @xmath115 , @xmath116 and @xmath89 for all curves.,height=188 ] we now vary the reionisation redshift @xmath91 keeping the other parameters fixed , and setting @xmath109 , @xmath115 and @xmath89 . we find that increasing @xmath91 shifts the position of the first peak to small scales ( fig . [ fig : differentrecross ] ) . the reason is the same as the one invoked for the shifts observed in the left panel of fig . [ fig : clre10 ] . when reionisation happens early , the integration of the @xmath0-mode polarisation in eq . ( [ eq : cross - cl ] ) takes into account more contribution from higher redshifts . this induces a shift of the peak position to higher @xmath21 modes . figure [ fig : peakzzre10 ] shows the evolution of the first peak amplitude as a function of @xmath96 , the redshift of the 21 cm line emission , for the three above - mentioned durations of reionisation @xmath92 . here , we set @xmath117 and @xmath89 . as shown by @xcite using eq . ( [ eq : xh ] ) , the amplitude of the cross - correlation of the cmb doppler temperature anisotropies and the 21 cm line fluctuations reaches its maximum value at @xmath118 where the ionised fraction is about half . the same is true in the case of the cross - correlation of 21 cm line fluctuations and @xmath0-mode polarisation . the reason for this is that the cmb polarisation is produced in the ionised regions whereas 21 cm line fluctuations are associated with neutral regions . we note that the signal in fig . [ fig : peakzzre10 ] is smaller at lower redshifts than at higher redshifts . this is due to the fact that the 21 cm signal vanishes once reionisation is completed , i.e. at redshifts smaller than @xmath119 . for higher redshifts ( @xmath120 ) , the low scattered quadrupole component of the @xmath0-modes is suppressed like in the lower redshifts part , but the integration in the @xmath0-modes picks up contributions from the reionisation epoch and the @xmath0-modes are correlated with a larger contribution of the 21 cm signal . as a result , the amplitude of the cross - correlation does not decline as steeply as for @xmath121 . the difference between the curves is easily interpreted . in the @xmath111 case , full reionisation takes longer to complete than in the @xmath102 case . therefore , neutral regions persist longer , broadening the skewed bell - shape of the curve centered on @xmath122 . the slight difference in the peak value is due to the cumulative effect of the integral in eq . ( [ eq : cross - cl ] ) . this is particularly well illustrated by the case of `` instantaneous reionisation '' ( @xmath110 ) . it is worth noting that our assumption , @xmath123 , is formally not valid before the beginning of reionisation . in order to evaluate the cross - correlation signal correctly , we should rather take into account the full evolution of @xmath124 with redshift in the calculation of the 21 cm line fluctuations . however , by the time the ionisation fraction has risen above the per cent level , efficient compton and x - ray heating will guarantee that the spin temperature be well above @xmath125 ( e.g. * ? ? ? the dotted , solid and dashed lines indicate the amplitudes for @xmath110 , @xmath102 and @xmath111 , respectively . for all plots , we set @xmath126 and @xmath89.,height=188 ] the evolution of the first peak amplitude for different @xmath91 is shown in fig . [ fig : peakzzred ] . as expected , the amplitude reaches its maximum value at each @xmath91 . since the 21 cm optical depth is higher at high @xmath96 , the amplitude of the 21 cm line fluctuations at higher redshifts is higher than that at lower redshifts . therefore , and as exhibited by the comparison of the amplitudes for the different @xmath91 , the maximum value of the first peak cross - correlation amplitude is larger in the early reionisation case than in the late reionisation case . this again is the same evolution that shows up in the power spectrum of the cross - correlation between the cmb temperature anisotropy and the 21 cm line fluctuations ( cf . fig . 2 of * ? ? ? * ) . the evolution , with the observed redshift , of the shape and the amplitude of the first peak of cross - correlation between the 21 cm line fluctuations and @xmath0-mode fluctuations should allow us to constrain both the duration and redshift of reionisation . the dotted , solid and dashed lines represent the evolution for @xmath127 , @xmath118 and @xmath128 , respectively . in this figure , we assume @xmath115 and @xmath90.,height=188 ] ultimately , the parameter @xmath88 is not independent of the epoch @xmath119 of reionisation and its duration @xmath92 . however , making that dependency explicit would require the very detailed description of the reionisation processes that we are still lacking today . we can nevertheless illustrate how the cross - correlation power spectrum changes for different global reionisation scenarios . the left panel in fig . [ fig : peakzzreionize ] shows the variation of the evolution of the first peak amplitude with the parameter @xmath88 . there is no difference of the peak position in both spectra for the `` photon counting limit''@xmath89 and the `` strmgren limit '' @xmath90 but the amplitudes are different . when @xmath88 increases , recombinations become more and more important . accordingly , over - dense regions in the case of @xmath89 have higher ionisation fraction than in the case of @xmath90 as mentioned in sec . [ subsec : cross ] . since the fluctuations of the neutral fraction are defined by @xmath129 , the absolute value of the 21 cm line fluctuations for @xmath89 is higher than for @xmath90 . therefore , the absolute value of the amplitude of the cmb @xmath0-mode anisotropies and the 21 cm line fluctuations correlation for @xmath89 is higher than for @xmath90 as shown in the left panel of fig . [ fig : peakzzreionize ] . finally , we also calculate the cross - correlation between the 21 cm line fluctuations and the @xmath0-mode cmb anisotropies in the case of a double reionisation . for illustrative purposes , we chose a simple parametrisation of the mean ionised fraction @xmath130 } + a \exp \left[-\frac{(z - z_{\rm first})^2 } { \delta z_{\rm first}^2}\right],\ ] ] which is the simple reionisation case formula with @xmath131 and @xmath115 on top of which we added a gaussian with @xmath132 , @xmath133 , and @xmath134 that encodes a first passage through an eighty per cent reionised universe at a redshift of twelve . these parameters have been adjusted so as to reproduce the same value of the optical depth ( @xmath135 ) as the single reionisation model shown in the left panel of fig . [ fig : peakzzreionize ] . the redshift evolution of @xmath136 in this toy model is represented by the dashed line in the right panel of fig . [ fig : peakzzreionize ] ( read right axis for values of @xmath137 ) . in the same figure , the solid line shows the corresponding evolution of the first peak of the cross - correlation ( left axis ) . every time the ionised fraction takes the value 0.5 there is a peak in the cross - correlation signal . in the case of our model for double reionisation , we observe three peaks , i.e. the ionisation fraction crosses the 0.5 line thrice . however , the two peaks at @xmath138 and @xmath139 are very close to each other . if two consecutive crossings of the 0.5 line occur very soon after each other , the corresponding peaks in the cross - correlation signal will be essentially indistinguishable . similarly to what has been found by @xcite , this behaviour is a very interesting property as it implies that we can use the `` trajectory '' of the evolution of the peak amplitude to constrain the reionisation model . the solid and dashed lines are for @xmath89 and @xmath90 , respectively . we set @xmath140 and @xmath141 in all plots . the right panel shows the evolution of the first peak amplitude in the case of a double reionisation model . the solid line ( left axis ) is for the first peak amplitude at given redshift . the dotted line ( right axis ) indicates the ionisation fraction in the double reionisation model.,height=188 ] . the solid and dashed lines are for @xmath89 and @xmath90 , respectively . we set @xmath140 and @xmath141 in all plots . the right panel shows the evolution of the first peak amplitude in the case of a double reionisation model . the solid line ( left axis ) is for the first peak amplitude at given redshift . the dotted line ( right axis ) indicates the ionisation fraction in the double reionisation model.,height=188 ] we have examined the cross - correlation between the @xmath0-mode cmb polarisation and the fluctuations of the 21 cm line background . in particular , we calculated the angular power spectrum of the cross - correlation on large scales using a simple parametrisation of reionisation , focusing explicitly on the redshift @xmath91 of its occurrence and on its duration @xmath92 . we also considered the case of double reionisation for which we obtained specific signatures . the angular power spectrum of the cross - correlation traces the quadrupole component of the cmb temperature anisotropies at the redshift where the 21 cm line is observed . both the amplitude and the position of the first peak depend on the reionisation epoch , as well as on the redshift @xmath96 probed by the 21 cm observations . we showed that the signal peaks when @xmath96 matches @xmath91 , in the very same way as does the cross - correlation between the cmb doppler temperature anisotropies and the fluctuations of the 21 cm line background ( see * ? ? ? * ) , that is when the 21 cm observations probe the epoch at which the ionised fraction becomes one half . when probing higher redshifts , the polarisation and 21 cm cross - correlation signal decreases slowly , whereas it falls off rapidly when lower redshifts @xmath96 are considered . the location of the first peak in the ( @xmath96 @xmath142 ) plane is essentially independent of @xmath92 . therefore , tracking the evolution of the first peak amplitude by scanning the sky in @xmath96 bins would allow to determine @xmath91 . moreover , the width of the bell - shaped curve describing the evolution with @xmath96 of the @xmath94 first peak depends on @xmath143 . the duration of the reionisation is thus also accessible through this observable quantity . we also showed that , in addition to the amplitude and the position of the first peak , the duration of reionisation induces a significant damping of the oscillations on small scales . this strong dependence on the duration shows specifically in the cross - correlation power spectrum as it does not arise in the auto - correlations power spectra . these significant features , which are the oscillations and the maximum at the redshift when the ionised fraction are one half , is explained by eq . ( [ eq : cross - cl ] ) which is an exact analytic expression . they are robust and independent of the details of the model . assumptions are only introduced to relate the gravitational potential to the baryon density contrast and to the ionised fraction in eq . ( [ eq : cross - cl ] ) . for the @xmath144 relation , we assume linear perturbation theory which is valid on large scales ( @xmath145 ) and at high redshift ( @xmath146 ) . for the @xmath147-@xmath148 relation , we need to assume a reionisation model . we choose a simple model based on the halo model . in our redshift range when the reionised fraction is around one half , as shown in many numerical studies ( e.g. , @xcite ) , the typical size of ionised bubbles is smaller than the instrumental resolution . in this case , we can assume that the ionised fraction can be set by the number of ionising sources ( for a detailed discussion see the appendix in @xcite ) . numerical simulations that provide detailed descriptions of the reionisation history can test those assumptions . in the present study , we assumed the observation of the 21 cm line ideal . in practice , it amounts to consider that the frequency window function is the delta function @xmath149 , implying that observations are done at the specified frequency @xmath150 only . of course , for real observations , the finite instrumental frequency width has to be taken into account . for too broad window functions , this is susceptible to induce an additional damping of the higher @xmath21 oscillations . we thus checked this effect in the case of the lofar interferometer that has a frequency width of 4 mhz . as a result , we did not obtain any additional significant damping of the oscillations for @xmath151 . both cmb and 21 cm signals suffer from the foreground contaminations . some of them will be correlated like the synchrotron emission from our galaxy . recent studies have shown that the cleaning of foreground were under control in both cmb and 21 cm observations @xcite . in addition , the multi frequency information on the cmb will allow the cross - correlation signal to be even less sensitive to the correlated foregrounds . finally , due to the cosmic variance that limits cmb measurements on large scales , it may actually be difficult to pin - point the first peak with great accuracy . moreover , ground - based telescopes designed for observations of the 21 cm line background can not cover the whole sky . those two facts are not important limitations for our study , since the damping of oscillations in the cross - correlation power spectrum on intermediate scales ( @xmath152 ) , which are well enough sampled , depends strongly on the reionisation evolution . actually , even if the reionisation duration is so long that we can not detect the oscillations due to the strong damping , it is still possible to obtain a lower limit for the duration . we will study this in more detail with numerical simulations in the future . 99 p. , furlanetto s. , 2008 , mnras , 384 , 291 m. a. , komatsu e. , dor o. , shapiro p. r. , 2006 , apj , 647 , 840 r. , loeb a. , 2001 , phys . , 349 , 125 a. j. , nusser a. , sugiyama n. , lacey c. g. , 2001 , mnras , 320 , 153 s. , ali s. s. , 2004 , mnras , 352 , 142 j. , et al . , 2006 , astro - ph/0604101 r. , 2003 , apj , 591 , 12 b. , madau p. , 2003 , apj , 596 , 1 a. , 2004 , phys . d , 70 , 063509 a. , 2006 , physical review letters , 97 , 261301 x. , et al . , 2006 , aj , 132 , 117 g. , 1958 , proc . iree aust . , 46 , 240 s. r. , oh s. p. , pierpaoli e. , 2006 , phys . d , 74 , 103502 j. e. , peterson b. a. , 1965 , apj , 142 , 1633 w. , sugiyama n. , 1995 , apj , 444 , 489 w. , white m. , 1997 , phys . d , 56 , 596 , i. t. , mellema , g. , pen , u .- l . , merz , h. , shapiro , p. r. , alvarez , m. a. , 2006 mnras , 369 , 1625 v. , et al . , 2008 , arxiv 0804.1130 n. , 1987 , mnras , 227 , 1 h. , sasaki m. , 1984 , progress of theoretical physics supplement , 78 , 1 a. , zaldarriaga m. , 2004 , physical review letters , 92 , 211301 p. , meiksin a. , rees m. j. , 1997 , apj , 475 , 429 l. , et al . , 2007 , apjs , 170 , 335 w.h . , schechter p. , 1974 , apj , 187 , 425 j. r. , furlanetto s. r. , 2007 , mnras , 376 , 1680 r. , ciardi b. , ferrara a. , baccigalupi c. , 2005 , mnras , 360 , 1063 a. , cooray a. , silk j. i. , 2007 , mnras , 377 , 168 d. n. , et al . , 2007 , apjs , 170 , 377 h. , sugiyama n. , 2006 , mnras , 372 , 1060 p. , madau p. , meiksin a. , rees m. j. , 2000 , apj , 528 , 597 s. , 1952 , aj , 57 , 31 m. , 1997 , phys . d , 55 , 1822 the source term of @xmath0-mode polarisation in eq . ( [ eq : emode - inte ] ) @xcite , @xmath153 where @xmath154 is the quadrupole moment of the cmb temperature anisotropies . the cmb temperature anisotropies are evaluated by the following integral solutions obtained from the boltzmann equations , @xmath155 where the @xmath156 are expressed in terms of the spherical bessel functions @xmath157 , @xmath158 . \label{eq : def - jl}\ ] ] and @xmath159 are source terms written as @xmath160 , \label{eq : source - temp}\ ] ] where @xmath161 is the velocity of the baryon fluid , @xmath162 and @xmath163 are the scalar - type metric perturbations in the conformal newtonian gauge . the cosmological linear perturbation theory provides the evolution of @xmath162 and @xmath163 and the initial condition of @xmath164 in the matter dominated epoch as @xcite @xmath165 where @xmath166 is the initial metric perturbation . before recombination , photons and baryons are tightly coupled by compton scattering . hence , the tight - coupling approximation is a strong tool for solving eqs . ( [ eq : emode - inte ] ) and ( [ eq : temp - inte ] ) . @xcite provided a useful analytic expression of the monopole component in this approximation , @xmath167 \cos k r_{\rm s } ( \eta ) -\frac{k}{\sqrt{3 } } \int ^\eta _ 0 d \eta ' [ \phi ( \eta')-\psi ( \eta ' ) ] \sin[k r_{\rm s}(\eta)-k r_{\rm s}(\eta ' ) ] , \label{eq : acoustic - tight}\ ] ] where @xmath168 is the integration of the sound speed @xmath169 , @xmath170 the dipole and quadrupole components , and the baryon velocity are written as @xmath171 accordingly , the quadrupole component of @xmath0-modes is given by @xmath172 after recombination , the tight - coupling approximation is no longer valid . using eqs . ( [ eq : metric - ini])-([eq : emode - tight ] ) as initial conditions , we solve eqs . ( [ eq : emode - inte ] ) and ( [ eq : temp - inte ] ) numerically in order to obtain the transfer function @xmath173 in eq . ( [ eq : def - transemode ] ) .	the cosmic microwave background ( cmb ) polarisation and the 21 cm line fluctuations are powerful probes of cosmological reionisation . we study how the cross - correlation between the cmb polarisation ( @xmath0-modes ) and the 21 cm line fluctuations can be used to gain further understanding of the reionisation history , within the framework of inhomogeneous reionisation . since the @xmath0-mode polarisation reflects the amplitude of the quadrupole component of the cmb temperature fluctuations , the angular power spectrum of the cross - correlation exhibits oscillations at all multipoles . the first peak of the power spectrum appears at the scale corresponding to the quadrupole at the redshift that is probed by the 21 cm line fluctuations . the peak reaches its maximum value in redshift when the average ionisation fraction of the universe is about half . on the other hand , on small scales , there is a damping that depends on the duration of reionisation . thus , the cross - correlation between the cmb polarisation and the 21 cm line fluctuations has the potential to constrain accurately the epoch and the duration of reionisation . cosmology : theory large - scale structure of universe
two of the most fundamental and intriguing questions in astronomy are how galaxies form and how they evolve with time . in order to answer these questions , we must be able to compare and contrast the properties of galaxies at different redshifts . whereas we have a wealth of information about galaxies at z @xmath14 1.0 , our knowledge of the properties of galaxies at higher redshifts is limited . however , with the advent of the hubble deep field - north ( hdf - n ) project @xcite , we can probe to fainter surface brightness limits and smaller angular scales than before , the sizes , shapes , and colors of distant galaxies which will ultimately yield important clues to understanding their structure , formation , and subsequent evolution . the near - infrared data from nicmos combined with the optical data from wfpc-2 give us the unique opportunity to compare the surface brightness properties of galaxies at the same rest - frame wavelengths over a range of redshifts allowing us to better understand galaxy formation and evolution from an observational standpoint . one observational test of galaxy evolution is the study of color profiles which will give us an idea of the distribution of stellar populations in the galaxies and whether this distribution changes with time . due to limits in resolution , the study of the color profiles of galaxies have been predominantly restricted to the low and intermediate redshift regimes . however , the deep , high resolution , multi - wavelength data from the hdf - n probed in this study provides us with the rare opportunity to study the color profiles of galaxies at higher redshifts than those studied in the past . previous studies of early - type galaxies at z @xmath14 1.0 have shown that they tend to have redder colors in their central regions and gradually become bluer outwards @xcite . this trend in the color profiles can be explained by either a stellar age or metallicity gradient which become degenerate at @xmath15 @xcite . most studies concur that models involving the metallicity gradient best reproduce the color gradients seen in early - type galaxies at z @xmath14 1.0 . on the other hand , the gradients observed in late - type galaxies may be due to both age and metallicity effects , i.e. , the central parts of the galaxies in general have older stars and higher metallicity than the outer parts making them have redder colors towards the centers . in their study of the near - ir and optical color profiles of 86 face - on disk dominated galaxies , @xcite concluded that their color gradients were best reproduced by models involving both stellar age and metallicity gradients . the existence or lack of dust in galaxies can also complicate matters and must be addressed when interpreting color profiles since , in theory , dust may also be responsible for the central reddening in galaxies if we assume that dust generally tends to be more concentrated in the center and consequently would produce more extinction there @xcite . a study of the color profiles and surface brightness profiles of galaxies spanning a wide range of redshifts will help us place constraints on galaxy formation and evolution by enabling us to compare what we learn from observations with what we predict from theoretical models . currently , the hierarchical structure formation model @xcite represents the popular framework for how structure formed and evolved in the universe . this model assumes that the universe is dominated by nonbaryonic dark matter which only interacts with visible matter through its gravitational influence and ultimately determines where galaxies will form . it predicts that the gravitational perturbations in the early universe will cause the smallest mass fluctuations to collapse first and then to subsequently merge into progressively larger structures until they form the mature galaxies we observe today . @xcite ( hereafter bcfl ) analyzed the properties of the high redshift lyman break galaxies in the context of this model . in their semi - analytic treatment of galaxy formation in hierarchical clustering theories , they generated mock catalogs of the high redshift lyman break galaxies ( lbgs ) using the color criterion imposed by @xcite and modeled the growth of dark matter haloes by the accretion of matter through mergers taking into account the cooling of gas into stars and feedback . @xcite ( hereafter spf ) also applied semi - analytic models of galaxy formation within the hierarchical clustering framework to analyze the population of lyman break galaxies at high redshift . since they conveniently address the properties of lyman break galaxies ( the sample of significant interest to us ) within the current popular framework for galaxy formation and evolution , we will interpret our results within the context of the bcfl and spf models . it is important to study the properties of the galaxies at different redshifts at the same rest - frame wavelengths because some galaxies might experience what is called `` morphological k - corrections '' . this is a phenomena which occurs when galaxies at different redshifts look different when viewed through the same passband since the passband would represent different rest - frame wavelengths . for example , the @xmath2 image for a galaxy at redshift @xmath16 would reflect the properties of the galaxy at the rest optical wavelengths whose light would include both young and old stellar populations , whereas for a galaxy at redshift @xmath17 , the @xmath2 image would represent the rest - uv wavelengths of the galaxy which is dominated by the younger , bluer stellar populations . thus , if we want to compare the properties of the same stellar populations , we need to choose our passbands such that they reflect the same rest wavelengths at all redshifts . the original hdf - n data was a director s discretionary program on hst designed to image a field at high galactic latitude using the wfpc-2 camera in four passbands spanning the wavelengths @xmath18 @xmath19 m @xcite . at low redshifts , these images were sufficient to study the galaxies in the rest - frame optical wavelengths , however at z @xmath20 1 , the wfpc-2 images could only be used to study the rest ultraviolet properties of the galaxies since their rest optical wavelengths were shifted into the near - ir . near - ir data of galaxies in the hdf - n were available from ground - based observations @xcite . however , since the ground - based data were taken at much lower resolution ( @xmath21 1 @xmath22 ) and reached shallower depths than did the wfpc-2 images , it was very difficult to compare the properties of high redshift galaxies with their lower redshift counterparts at the same rest - frame wavelengths . to address this problem , @xcite , embarked on a general observer s ( go ) program 7817 to image the entire wfpc-2 region of the hdf - n in the near - ir using the nicmos camera 3 f110w and f160w filters . the data reached an average depth of @xmath23 in f110w and f160w with a signal - to - noise ratio of 10 in 0.@xmath227 diameter aperture and has a psf of @xmath24@xmath25 after the dithered data was combined by drizzling . to avoid the morphological k - correction problem , we study the intermediate redshift galaxies in the hdf - n using the wfpc-2 images taken in f450w ( @xmath26 ) , f606w ( @xmath27 ) and f814w ( @xmath2 ) and the high redshift galaxies using wfpc-2 f814w ( @xmath2 ) and the nicmos cam3 images taken in f110w ( @xmath28 ) and f160w ( @xmath5 ) in order to probe their rest - frame near - uv and optical properties . since we will be comparing the color profiles of galaxies at a range of redshifts , it is important to establish what we mean by intermediate and high redshift galaxies . we will define _ intermediate redshift galaxies as those having redshifts between 0.5 @xmath14 z @xmath14 1.2 and _ high redshift galaxies as those having redshifts between 2.0 @xmath14 z @xmath14 3.5 . we do not include galaxies with 1.2 @xmath14 z @xmath14 2.0 in our sample because at these redshifts , there are no prominent optical features to constrain the redshifts with the high precision required in our study . our sample consists of 33 intermediate redshift galaxies with @xmath2 @xmath3 25 identified by spectroscopic redshifts @xcite and 50 high redshift galaxies with @xmath5 @xmath3 27 , of which 25 were identified by spectroscopic redshifts @xcite and 25 by photometric redshifts @xcite . in our paper , we assume a cosomology where @xmath29 and @xmath30 . _ _ we performed detailed surface photometry on our sample of galaxies using the ellipse task in the iraf / stsdas package . this task fits elliptical isophotes to the galaxies using the iterative scheme described in @xcite . we entered inital guesses for the ellipse center ( _ x , y ) , ellipticity ( _ e ) , and position angle ( @xmath31 ) and allowed the task to update these parameters following the iteration scheme . the errors in the azimuthally averaged intensity are obtained from the rms scatter of the intensity measurements along each fitted isophote . in order to generate color profiles , it was important that we kept the ellipses fixed between the different filters since positional displacement of the isophotes in the different filters may cause the color distribution of the galaxy to be artificially asymmetric . thus , we combined the images taken in different passbands , fitted ellipses to the combined image , and then used the set of ellipses generated from the combined image as the input ellipses for no - fit mode on each individual image . in this way , we made sure that the surface brightness profiles from the different bands were generated using the same fitted ellipses . it is also important that the psfs between the different filters match . this was accomplished by degrading the @xmath26 , @xmath27 , @xmath2 , and @xmath28 images to match the resolution of the @xmath5 image . the psf matching was verified by looking for color gradients in stellar profiles . many of the galaxies , especially at high redshifts , possessed nearby companions . in order to perform the fits only on the galaxy of interest , we masked out the companion . we then converted all fluxes to ab magnitudes . _ _ from the elliptical isophotes , we produced azimuthally averaged radial surface brightness profiles . for our intermediate redshift sample , we produced surface brightness profiles of the images taken in @xmath2 , while for the high redshift sample , we produced surface brightness profiles for the images taken in @xmath5 . at the respective redshifts of the galaxies , these images represent approximately their rest - frame b wavelengths . we analyzed the resulting intensity distribution using the stsdas task nfit1d to perform @xmath32 fitting to the surface brightness profiles . normally we would fit the profiles with the sum of an @xmath13 bulge @xcite and an exponential disk @xcite profile . however , for most of the galaxies , especially at high redshifts , the scale lengths of the galaxies were so small and severely affected by psf smearing that it was difficult to obtain reliable fits using these profiles . thus , we decided to fit the radial light profiles with a generalized exponential @xcite : @xmath33\ ] ] where _ r(e ) is the half - light radius , _ s(e ) is the flux at _ r(e ) , @xmath34 is the surface brightness at r , and _ n is the srsic index which tells us about the degree of flattening in the profile ( @xmath35 recovers the exponential disk profile and @xmath36 recovers the @xmath13 law ) . @xcite have applied an analytic approach to study the effects of the psf on the srsic profiles . they found that , of the free parameters in the srsic profile , the one that is affected most by the psf is @xmath37 which is smaller in the observed profile than in the actual profile . furthermore , the higher the original value of @xmath37 is , the more it will be affected by the psf . at the faint magnitudes and small scale lengths of typical high redshift galaxies in the hdf - n , we will show that the psf can potentially flatten both an exponential and an @xmath13 profile into a srsic profile with _ n @xmath3 1 . it then becomes challenging to distinguish between galaxies having exponential profiles from those having de vaucouleurs profiles . the next section explains our attempts to understand the effects of the psf on the surface brightness profiles in order to more accurately classify morphologies . _ _ _ _ _ in order to account for the effects of the psf on the data , we created model galaxies and convolved them with the psf to match the resolution of the data . to reproduce the surface brightness profiles of the intermediate redshift sample , we created 10 galaxies with exponential disk profiles ( @xmath38 ) and 10 with @xmath13 ( @xmath39 ) bulge profiles with varying half - light radii and @xmath2 magnitudes = 21 , 22 , 23 , and 24 ( this spans the range of magnitudes in our intermediate redshift sample ) for a total of 80 galaxies . since the wfpc-2 and nicmos images have all been degraded to match the resolution of the @xmath5 passband , we convolved the model galaxies with an isolated , unsaturated star taken from this passband . we performed the convolution using the iraf task fconvolve which takes the fourier transform of the model galaxy and the psf ( normalized to conserve flux ) , multiplies them together , and then takes the inverse transform of the product to produce the final convolved galaxy . we then added the model galaxies onto blank areas of the sky in the hdf @xmath2 image . by doing so , we include the contribution of the background noise which exists in the image . however , we have chosen to neglect the poisson noise from the galaxies themselves since it will be negligible compared to the background noise . to simulate our high redshift sample , we repeated the process but added the model galaxies onto blank areas of the hdf @xmath5 image with @xmath5 magnitudes = 23 and 24 . we then generated surface brightness profiles and fit srsic profiles to the convolved disk and spheroid galaxies in order to determine the degree of flattening due to the psf . tables 1 and 2 lists the half - light radii and the srsic indices of the convolved model galaxies with a range of magnitudes in the @xmath2 and @xmath5 passbands respectively . from tables 1 and 2 , we can see that after convolution with the psf , the profiles of all the galaxies are significantly flattened , i.e. their srsic index @xmath37 decreased significantly from the original profile ( recall that @xmath35 for disks and @xmath36 for spheroids ) . in fact , table 1 shows that after convolution with the psf , the srsic indices of the mock intermediate redshift galaxies originally possessing an @xmath13 profile can drop all the way to a value of 1 which is indicative of an exponential profile . table 2 shows an even more drastic drop in srsic indices of the mock high redshift galaxies after they had been convolved with the psf of the hdf . at these small scale lengths and faint magnitudes typical of the high redshift galaxies in our sample , the srsic indices of the model spheroids drop to 1 and below rendering them almost indistinguishable from the model disks . furthermore , the smaller the galaxy is , the more the value of @xmath37 decreases for both the disk and bulge models . consequently it would be very difficult to determine from the convolved profile whether it originally possessed an exponential or an @xmath13 profile . fortunately , trends exist which allow us to reliably distinguish between the two types of galaxies even after they have been significantly flattened by the psf . we notice that at a given radius and magnitude both profiles are severely smoothed out , but that @xmath37 for the spheroid is always higher than @xmath37 for the disk , i.e. the disk has a more flattened profile . for example , at @xmath2 = 21 and half - light radius of approximately 0.@xmath2284 , @xmath37 for the spheroid is 1.736 whereas @xmath37 for the disk is 0.789 ( table 1 ) which is significantly lower . this trend occurs at all radii and magnitudes . thus , although the surface brightness profiles of the galaxies in the hdf - n are significantly affected by the psf , we can still broadly classify galaxies morphologically by comparing the value of @xmath37 and the half - light radii of these galaxies at a given magnitude with those of our model galaxies . however , caution must be taken when using this method to classify the morphologies of galaxies with half - light radii less than @xmath40 since it would be pushing the resolution limit of the instrument and , consequently , the results would not be reliable . fortunately , there are no galaxies in our sample where this is the case . we must also note that in our high redshift sample , we can only reliably classify the morphologies of galaxies with magnitudes less than @xmath5 = 24.5 since , at fainter magnitudes , the srsic indices of a disk and spheroid at a given radius is within the range of the scatter . we must keep in mind also that our galaxy classification scheme is not intended to be used to classify galaxies on a case by case basis , but rather to obtain statistics about the galaxy population as a whole . we determined the color profiles of the galaxies by subtracting the surface brightness profile of one passband from the surface brightness profile of the adjacent passband . for our intermediate redshift sample , we generated @xmath41 color profiles , and for our high redshift samples , we produced @xmath42 color profiles . we then applied k - corrections described in the following section to transform them to their rest - frame @xmath0 colors where uv @xmath21 2192 and u @xmath21 2943 . the k - corrections applied are minimal since we have chosen to work with passbands which closely map to their rest - frame wavelengths . we then performed a weighted least - squares fit to the profile . we started the fit 0.14 out from the center of the profile since the color differences near the center might reflect slight differences in the psf between the two passbands rather than indicate true color changes in the galaxy . from the fits , we were able to obtain the value of the color gradient ( @xmath7/@xmath8log(r ) ) where @xmath0 is the rest - frame color and r is the radius in kpc . the main source of error in this procedure is the uncertainty in the determination of the local sky background values . the global sky background subtraction might be adequate for the larger and brighter galaxies in our intermediate redshift sample , but for the smaller and fainter galaxies in our high redshift sample , there could be residuals in the background which could significantly change the value and even the sign of the color gradient if the effect is huge . we determined the value of the local sky background by extending the elliptical isophotes out past five scale lengths in 0.12 increments taking care to mask out nearby objects . we then adopted the value of our background as the typical value of the isophotes located far enough from the object such that the flux level ceased to decrease steadily , but instead fluctuated around some value . from this method , we found that in general there had been an overestimation in the global background value in the reduced images used for this study . when we observe two galaxies at different distances with the same intrinsic brightness through the same passband , we will measure two different magnitudes not only because of the inverse square law , but also because their spectral energy distributions ( sed ) will be shifted towards longer wavelengths and also `` stretched '' by a factor of @xmath43 due to the expansion of the universe . in order to correct for this effect and to convert the color profiles to their rest - frame values , we need to apply k - corrections to our results . we use the definition of the k - correction from @xcite : @xmath44\ ] ] where @xmath45 is the sensitivity function of the detector i , @xmath46 is the observed energy flux density , @xmath47 is the energy flux density of the galaxy at rest , @xmath48 is the bandwidth of the detector _ i , and @xmath49 is the k - correction of the galaxy observed through detector _ i at redshift z. the first term of the correction is due to the `` stretching '' of the passband whereas the second term represents the shifting of the sed to longer wavelengths . _ _ since k - corrections do not currently exist for the nicmos passbands , we generated our own set by using the seds from @xcite . they used the stardust model to produce seds for 17 galaxies ranging from local spirals , starbursts , luminous infrared galaxies , and ultraluminous infrared galaxies ( ulirgs ) extending from the far - uv to sub - mm wavelengths . their seds also included the effects of internal extinction and emission of dust in the galaxies . in the process of analyzing the color profiles of the galaxies , we also studied their surface brightness profiles . tables 3 and 4 list the statistics for the 33 galaxies in the intermediate redshift and 50 galaxies in the high redshift regime respectively . column ( 4 ) of tables 3 and 4 lists the srsic indices in the @xmath2 passband for galaxies at intermediate redshifts and the @xmath5 passband for galaxies at high redshifts . these passbands translate to approximately the rest - frame b band for all the galaxies . ideally galaxies which possess exponential profiles would have a srsic index of 1 whereas galaxies which possess a de vaucouleurs profile would have a srsic index of 4 . thus , from a naive inspection of the srsic indices of the galaxies at high redshifts , we would assume that only one ( i d 4 - 555.11 with @xmath37 = 1.467 ) has a steeper profile than an exponential . however , we have demonstrated in section 3.3 that when we classify morphologies based on the srsic indices , it is critical to take into account the role of the psf since in many cases , especially at high redshifts , the psf can smooth what was originally an @xmath13 profile to one with @xmath37 @xmath3 1 . the 5th column lists our profile classification for each galaxy ( after taking into account psf effects ) based on the galaxy models discussed in section 3.3 where `` dev '' represents a de vaucouleurs @xmath13 profile , `` exp '' represents an exponential profile , `` int '' represents an intermediate profile with 1 @xmath14 @xmath37 @xmath14 4 ( these galaxies most likely possess both a bulge and disk component ) , and `` o '' represents `` others '' for galaxies showing irregular structure which could not be fit with any profile . in our attempt to classify the morphologies of the galaxies , we do not automatically assume that objects which possesss exponential profiles are disks while objects which possess de vaucouleurs profiles are spheroids since this is not necessarily always the case because many dwarf spheroids have been shown to possess exponential profiles @xcite . we naively lump into one category ( `` o '' ) those galaxies which have irregular profiles and do not follow an exponential or @xmath13 law . @xcite explain that there are actually two types of `` irregular '' galaxies . one type is the traditional irregular which is the late - type system classified in the hubble scheme whereas the other type possesses an irregular profile as a result of galaxy - galaxy interactions and mergers . however , it is not within the scope of this paper to differentiate between the two types . thus , it is evident that our classification scheme represents an oversimplification designed to obtain a sense of the overall trends in the morphologies of the galaxies and should not be taken as a robust classification scheme on the individual level . figures @xmath50 show the surface brightness profiles and the fits to the profiles of all the galaxies in our sample . the williams et al . ( 1996 ) identification numbers and srsic indices are included in the plots . figures 4 and 5 represent plots of the half - light radii versus the srsic indices ( @xmath51 ) of the model disks and spheroids after they had been convolved with the hdf psf in the @xmath2 ( with @xmath2 = 21 , 22 , 23 , and 24 ) and @xmath5 ( with @xmath5 = 23 and 24 ) passband respectively . again , for the high redshift sample , the results from the galaxy models are only reliable for @xmath5 @xmath14 24.5 since for fainter magnitudes , it is difficult to clearly distinguish the locus of srsic indices for the model spheroids from the locus of srsic indices for the model disks . the triangles represent the values for the model spheroids and the asterisks represent the values for the model disks . the dashed and dotted lines represent weighted least - squares fit to the model spheroids and model disks respectively , whereas the solid lines represent the 1@xmath52 error in the fits . for all @xmath2 magnitudes and for @xmath5 @xmath14 24.5 , we have also plotted the values of the galaxies in our intermediate and high redshift sample ( represented by squares ) to attempt to classify their morphologies by observing where they fall on the diagram . we do not include in these plots those galaxies which could not be fit by the srsic law since their profiles were too irregular . these galaxies were , however , still included in our statistics and were labeled as `` o '' for `` others '' . at intermediate redshifts , the morphologies are better constrained since the galaxies are brighter and the distinction between the model spheroids and disks is more apparent . as a further check on the accuracy of our classification system , we can utilize the images we have of the intermediate redshift galaxies since , for many , they are large enough and bright enough to classify their morphologies by eye . with the @xmath51 plots , we have attempted to classify the morphologies of all the galaxies in the intermediate redshift bin and those with @xmath5 @xmath14 24.5 in the high redshift bin , although it should be kept in mind that the process is rather crude since it relies mainly on the galaxy models which should be regarded as highly speculative at small scale lengths and faint magnitudes . the galaxies whose srsic indices fall above the locus for the model spheroids were classified as posessing @xmath13 profiles since this would indicate that they possess profiles that are steeper than our models , whereas those whose srsic indices fall below the locus for the model disks are classified as possessing exponential profiles since this would indicate they possess profiles that are shallower than our models . for the galaxies which have srsic indices in between the locus of our model spheroids and disks , we placed an acceptance limit of 5@xmath52 . if the values of the srsic indices were less than 5@xmath52 away from the locus of spheroids , we classified them as possessing @xmath13 profiles , if they were less than 5@xmath52 from the locus of the disks , we classified them as possessing exponential profiles , and if they were more than 5@xmath52 away from either loci , we classified them as possessing intermediate profiles . there were a few galaxies which possessed srsic indices with less than 5@xmath52 deviations from both loci . in these cases , we classified them according to the magnitude of their deviation . if the deviation from the locus of model spheroids was less than it was from the locus of model disks , then we classified those galaxies as possessing @xmath13 profiles . if however , the converse was true , we classified those galaxies as possessing exponential profiles . but here is where we introduce some uncertainty in our classification scheme . since these galaxies have such small deviations from both loci , even though we associate these galaxies with a certain profile , it does not mean that it can not possess the other types of profiles . for example , i d 3 - 404 was classified as possessing an exponential profile since its srsic index was only 2.1@xmath52 away from the locus of the model disks , however , there is still a good chance that it could possess an @xmath13 profile since it was only 3.3@xmath52 away from the locus of model spheroids . or it could also possibly possess an intermediate profile . there are only two galaxies in the intermediate redshift sample ( ids 2 - 353 and 3 - 404 ) and one in the high redshift sample ( i d 2 - 321 ) which have less than 5@xmath52 deviations from both loci . if we include these uncertainties , it will at most increase the poissonian errors by 1 percent . thus , the profile classification is very robust , and the uncertainties in the classification system is dominated by the poissonian error . table 5 summarizes our best effort to classify the morphologies of the galaxies based on the combination of the results from the fits to the surface brightness profiles of the galaxies in the two redshift bins , the plots of the galaxy models , and visual inspection . quoted are the percentages found for each profile with their poissonian error . the number of galaxies possessing that profile is in parenthesis . in our intermediate redshift sample , out of 32 galaxies , we classified 28 @xmath11 9% as possessing @xmath13 profiles , 53 @xmath11 13% as possessing exponential profiles , 13 @xmath11 6% as intermediates , and 6 @xmath11 3% as `` others '' since they possessed irregular substructure . the galaxy that we did not include in the statistics is i d 3 - 610 in which both an @xmath13 and and exponential component could be resolved . the srsic index for the inner region of this galaxy is 2.32 while for the outer , disk part it is 0.826 . this is the only galaxy large enough in our sample to distinguish the two components . in our high redshift sample only 20 galaxies had @xmath5 @xmath53 . out of these galaxies , we classified 60 @xmath11 15% as possessing @xmath13 profiles , 20 @xmath11 10% as possessing exponential profiles , and 20 @xmath11 10% as `` others '' . in the high redshift bin there are no galaxies which we classified as intermediates . this does not mean that no intermediates exist at these redshifts , but rather that since the separation between the locus of model spheroids and model disks is very small , there is not a clearly defined region for the intermediate cases . we must stress that we are limited by small number statistics , especially at high redshifts since we are only able to classify the morphologies of 20 out of our original 50 galaxies . nevertheless , we can still comment on the general trends seen . from table 5 , we see that , at higher look - back time , the fraction of the galaxies possessing @xmath13 profiles increased by 32 @xmath11 18% whereas the fraction of exponential - type profiles decreased by 33 @xmath11 16% . these differences are marginally significant since they represent @xmath21 2@xmath52 changes in both cases . our results differ from what has been previously published @xcite . by using a profile decomposition method on 522 galaxies down to @xmath54 @xmath3 26 , @xcite found that the majority of hdf - n galaxies are disk dominated with only 8% having dominant bulge fractions . they interpreted their results as measuring a decrease in the number of bulge dominated galaxies as a function of look - back time . given the high degree of difficulty in the classification of the morphologies of these small faint galaxies , which are severely affected by the psf and the limitation of small number statistics , it is not surprising that at this point , we are not able to reach a general consensus . at high redshifts , we also see a statistically insignificant increase of 14 @xmath11 10% in the fraction of galaxies which we have classified as `` others '' . this is along the lines of what was observed in previous studies of galaxies in the hdf - n which have shown that by @xmath55 @xmath20 24 , there are few galaxies possessing traditional hubble sequence morphologies . instead they tend to be more `` peculiar '' in the sense that they may represent mergers in progress @xcite . although we detect changes in morphology between the two redshift bins , these changes may not be solely attributed to evolution . in section 6.0 we discuss the possibility that the trends in morphology may be influenced by observational selection effects . we generated ( @xmath56 ) and ( @xmath57 ) color profiles for the 33 intermediate redshift and 50 high - redshift sample respectively and applied k - corrections in order to determine if these galaxies exhibited significant color changes ( i.e. @xmath58 @xmath7/ @xmath8log(r ) @xmath58 @xmath59 0.2 mag dex@xmath12 ) in their rest - frame @xmath0 profiles ( figures @xmath60 ) . we can assume that since we are observing the color profiles for the rest - frame @xmath0 , we will be probing primarily the light from the young stars . thus , the sign of the gradient will give us an idea of the locations of the star forming regions if the color gradient is caused by an age gradient alone . if the galaxy possesses a _ positive gradient , this would indicate a blueing towards the center , which might be explained by star formation that is centrally concentrated . if the galaxy possesses a _ negative gradient , this would point to a blueing with increasing radius which would imply an older stellar population in the center . if , however , we observe no significant gradient in the galaxy , we can assume that the star formation regions are uniformly dispersed throughout the galaxy . _ _ due to the @xmath61 fall off in the surface brightness , the color profiles of our high redshift sample do not extend to the low surface brightness levels of our intermediate redshift sample . for example , in our high redshift galaxies , we might reach depths of as much as 28 @xmath62 , but due to the fall off in surface brightness , this would translate to a depth of only @xmath21 24.5 @xmath63 in our intermediate redshift sample . in order to consistently compare the results for our two redshift bins , we must sample over approximately the same surface brightness range , thus , we must apply a cutoff in surface brightness in our intermediate redshift sample . applying a cutoff in the radial range is approximately the same as applying the cutoff in surface brightness . thus , we have chosen to fit the color profiles of the intermediate redshift sample only out to a radius of 4 kpc . this radius cutoff represents approximately twice the mean half - light radius of our high redshift sample which was determined by summing up all the light within approximately 28 @xmath62 . column ( 8) in tables 3 and 4 lists the color gradients from the weighted least - squares fit to the color profiles and their associated errors for our intermediate redshift and high redshift samples , respectively . the errors quoted include the uncertainties from the fit and a conservative upper limit in the uncertainty from the systematic background subtraction ( @xmath21 @xmath11 0.037 mag / dex ) . from table 3 , column ( 8) we notice that 12 ( 36% ) of the galaxies in our intermediate redshift sample possess positive gradients , but only two of them have gradients greater than 0.2 mag dex@xmath12 which indicates that they may have a strong central concentration of star formation perhaps indicative of a burst . although 64% of the galaxies at intermediate redshifts possess negative gradients , only 39% of them possess gradients less than -0.20 mag dex@xmath12 . furthermore , the mean of the color gradients of all the galaxies at intermediate redshifts is -0.091 @xmath11 0.007 mag dex@xmath12 which is relatively shallow . thus , although the negative color gradients indicates that , in general , the colors of these galaxies are bluer in the outer parts of the galaxies , since the mean of the gradient is not too steep , the color changes are gradual across the galaxies . column ( 8) in table 4 lists the rest - frame @xmath0 color gradients and the associated errors for our high redshift sample determined using all the isophotes which reach up to depths @xmath21 28 @xmath62 . we notice that the color gradients of the majority of the high - redshift galaxies ( 72% ) are positive with a mean of 0.272 @xmath11 0.007 mag dex@xmath12 . out of the 36 galaxies that have positive gradients , 83% of them have gradients greater than 0.200 mag dex@xmath12 indicating that there are significant color changes across these galaxies . figure 9 represents a plot of the redshift versus the color gradients of the galaxies in our two redshift bins . when we compare the rest - frame @xmath0 color gradients for our intermediate redshift sample with our high redshift sample , we detect different trends in the two redshift bins . in our intermediate redshift sample , we saw a slight trend for a blueing with increasing radius ( negative color gradients ) , implying that the star formation regions may be located generally in the outer disks of the galaxies or that the older stellar populations may be located towards the center . but given that the amplitude of the gradients were not too significant , we conclude that in general the color changes are gradual across the galaxies . however , in our high redshift sample , the majority of the galaxies possessed significant positive gradients in their color profiles . on average , these color gradients were 0.363 @xmath11 0.010 mag dex@xmath12 higher than their lower redshift counterparts . if these color gradients are due primarily to an age gradient , this would mean that these galaxies have bluer colors towards the center implying that their star formation regions may be centrally concentrated , i.e. there are more young stars at the centers than in the outer parts of the galaxies . numerous analysis show that both present - day disks and spheroids have metallicity gradients with higher metallicities toward the galaxy centers . such a metal gradient would only strengthen the age gradient needed to make the centers bluer . thus , from our analysis , we conclude that there is evolution in the color profiles of galaxies in the hdf - n . when we look further back in time , the star formation regions of the galaxies are more centrally concentrated . since it is beyond the scope of this paper to generate models to interpret our results , in the next section , we discuss modeling of color gradients taken from the literature in order to infer the dominant processes responsible for producing the trends seen in the color gradients . not much is currently known about the ultraviolet color gradients of galaxies in the local universe . the few studies of this kind were conducted using the information from the images taken with the ultraviolet imaging telescope during the astro-1 and 2 missions @xcite . from the astro-1 mission @xcite analized the ( 152 - 249 ) color profiles of two sb bulges and two e galaxies which were generated using the images taken with the 152 nm and 249 nm filters . they concluded that all but m32 exhibited prominent positive gradients in their profiles . in their study of the fuv(1500)-b color profiles for eight early - type galaxies , @xcite found , again , that all but m32 exhibited large positive color gradients , with differences in color greater than 1.0 mag over the entire profile . these large ultraviolet color gradients are reflective of the `` uv upturn '' or `` uvx '' population which has been observed in the far - uv ( 1200 @xmath3 @xmath64 @xmath3 2000 ) imaging of local spiral bulges and early - type galaxies ( @xcite . they propose that a combination of an increase in metallicity combined with the existence of older stellar populations may contribute in producing these color gradients . the trends in the far - uv color gradients of these galaxies are in the opposite sense to what we observe in the rest - frame near - uv for the majority of the intermediate redshift galaxies in our hdf - n sample and local galaxies which have been observed in the optical and near - ir . however , since the study of the far - uv color profiles of galaxies were conducted at shorter wavelengths , and given that the sample sizes were much smaller and limited only to early - type systems and spiral bulges , it is difficult to compare it with our study . thus , more observations of local galaxies in the ultraviolet are needed in order to fairly compare with our sample of galaxies at intermediate and high redshifts . in our intermediate redshift sample of galaxies , we find that most of the galaxies ( 6 out of 9 ) which we classified as having @xmath13 profiles possess negative rest - frame @xmath65 color gradients indicating reddening towards the center of the profile . this is consistent with the rest - frame optical color profiles observed in the past . in general early - type galaxies have been found to exhibit negative gradients , i.e. they have redder colors in the center and gradually become bluer outwards @xcite . in their study of the ( @xmath66 ) color profiles of 10 elliptical galaxies in the hdf - n from @xmath67 , @xcite observed the same trends . they found that all but two ( one of them being galaxy 2 - 251 in our sample ) exhibited redder colors towards the centers . at @xmath15 , both an age and a metallicity gradient in the stellar populations can be responsible for producing the negative color gradients observed in the galaxies , i.e. the galaxy colors are redder in the center than in the outer parts because they either have older stellar populations or higher metallicities there . this age - metallicity degeneracy makes it difficult to determine the cause of the color gradient . in order to break the degeneracy , they generated two model gradients which reproduce the typical color gradients of elliptical galaxies at @xmath15 , one caused by a metallicity gradient of old stars , the other from an age gradient of stars with the same metallicity . they then evolved these models to the redshifts of their observed galaxy sample . by comparing the model gradients with the observed gradients , they concluded that the model involving the metallicity gradient agreed well with observations at all redshifts whereas the model involving the age gradient was only in good agreement up to a redshift of z @xmath21 0.3 . we also have ( @xmath66 ) color profiles for the galaxies in our intermediate redshift sample which reveal the same trends in the color gradients ( figure 10 ) . we again observe that over half ( 5 out of 9 ) of the galaxies possessing @xmath13 profiles exhibit negative color gradients . thus , we may interpret the central reddening in the color profiles of the elliptical galaxies in our intermediate redshift sample in the hdf - n as being due primarily to a metallicity gradient ; the centers of the galaxies have higher metallicities than in the outer parts . we observe the same negative trend in the color gradients for the majority of the intermediate redshift galaxies ( 14 out of 23 ) which we classified as peculiars / irregulars or having exponential profiles . in the study of the near - ir and optical color profiles of 86 face - on disk dominated galaxies , @xcite concluded that their negative color gradients were best reproduced by models involving both stellar age and metallicity gradients . therefore , we interpret the gradients seen in the late - type galaxies as arising as a result of both age and metallicity effects ; the central parts of the galaxies in general have older stars and higher metallicity than the outer parts making them have redder colors towards the centers . whereas the trends seen in the color profiles of our intermediate redshift sample are well understood , we can not say the same for our high redshift sample . since this is the first study of its kind to be conducted on galaxies with z @xmath59 2.0 , there are no other high redshift observations of color profiles to compare our results with . thus , we must attempt to account for the positive gradients seen in our high redshift sample by understanding what is responsible for producing similar trends in the color profiles seen at intermediate and low redshifts . @xcite generated rest - frame @xmath68 color profiles for five compact narrow emission line galaxies ( cnelgs ) which are low mass ( @xmath69 @xmath70 ) starburst systems seen at @xmath71 and are similar to nearby hii galaxies . they noticed that although the gradients were not significant , there was a slight trend for the bluest color to occur in the central regions . on average , they found that the color inside the half - light radius is @xmath72 mag bluer than outside the half - light radius . @xcite futhermore describes the cnelgs as having either compact cores , fans and tails , or several small compact regions . these morphologies are along the lines of what we see in our high redshift sample . they also determined half - light radii to span from @xmath73 kpc ( @xmath74 , @xmath75 ) which match the range we see in our high redshift sample . given that the morphologies , half - light radii , and color profiles of the cnelgs are similar to what we observe in the high redshift sample , it is not unreasonable to believe that cnelgs could be the low redshift , lower - luminosity analogs of the higher redshift lyman break galaxies . we can then interpret the positive gradients seen in our high redshift sample as being caused by centrally concentrated starbursts . the high redshift galaxies have been shown to possess modest to high star formation rates ( @xmath76 @xmath77 @xmath70/yr , @xmath78=0.5 ) @xcite which further enforces the idea that the positive gradients seen in our sample could be caused by centrally concentrated starbursts . starbursts can be triggered by several mechanisms including mergers and interactions @xcite , bar instabilities @xcite , and sne and stellar winds , e.g. @xcite . since the volume of the high redshift universe is smaller than it is now , mergers have been implied to be the cause of the enhanced star formation , however , this topic is still under debate . in section 7 , we argue that the starbursts which may be responsible for causing the central blue colors of our high redshift sample are most likely ignited through mergers and interactions . when we discuss the trends seen in the color profiles , we must take into account the uncertainities involved . these uncertainties include the determination of the sky background , the correlation of the pixels , and the role of dust . while statistical errors dominating the inner parts of the color profiles and psf effects contribute to the uncertainty in the least - squares fit , the systematic errors due to sky subtraction dominate in the outer parts of the profile . in order to assess how our color gradients are affected by sky subtraction , we chose 20 high redshift galaxies possessing positive color gradients and for each we obtained statistics on 11x11 pixel wide grids in six blank areas of the sky around the object in the @xmath2 and the @xmath28 passbands . we generated a histogram of the 726 total pixels in the six grids to determine the mean and the standard deviation of the mean ( @xmath52 ) . for each galaxy , we then determined both a lower and an upper limit of the sky values in @xmath2 and @xmath28 by subtracting 3@xmath52 from the means to obtain their lower limits ( i- , j- ) and adding 3@xmath52 to the means to obtain their upper limits ( i+,j+ ) . lastly , we generated their ( @xmath57 ) color profiles and obtained their color gradients for each of the five possible combinations of sky values subtracted off ( ( 1 ) mean in @xmath2 , mean in @xmath28 ( 2 ) i- , j- ( 3 ) i+,j+ ( 4 ) i-,j+ ( 5 ) i+ , j- ) . the mean in the color gradients for the five different cases respectively are 0.3722 , 0.425 , 0.321 , 0.249 , and 0.503 which represent a 3@xmath52 deviation of @xmath21 0.112 mag / dex making the typical uncertainty in the estimation of the sky background to be @xmath21 0.037 mag / dex . as we can see , the mean value of the color gradients remain positive and greater than 0.200 mag / dex in all cases . thus , we find that the typical gradients seen in the high redshift galaxies ( @xmath21 0.26 mag / dex ) are significant enough , that even with the uncertainties involved in the estimation of the sky background , our main conclusion that these galaxies have centrally concentrated star formation still holds . another potential source of concern is the correlation of the pixels . since the images have been drizzled , convolved to recover approximately the same resolution as the @xmath5 image , and fitted with ellipses at 0.1 pixel increments , the errors that we obtain from the photometry will be correlated due to the oversampled pixels . thus , when we performed a weighted least - squares fit to the color profiles , the error bars used in the determination of the weights were correlated . consequently , the parameters that we obtain from the fit , i.e. the color gradient in this case , may be off from what would be measured if the errors had been independent . to understand how the correlated errors in our photometry affect the least - squares fitting to a straight line , we obtained rms images ( m. dickinson 2001 , private communication ) in the passbands of interest . these images quantify the uncertainty of the noise in the background ( sky + readout noise + dark current ) before any processes were performed to cause interpixel correlations . we re - analyzed ten random galaxies from the intermediate and high redshift sample using the rms images . with the same set of ellipses that we used in the surface photometry , we replaced the correlated errors with the `` true '' uncorrelated errors for each galaxy . in doing so , we have chosen to neglect the poisson noise from the galaxies themselves since they are so faint that it would be insignificant compared to the background noise . from this analysis , we find that even after taking into account interpixel correlations , the sign of the color gradients for the galaxies does not change . the high redshift galaxies still possess predominantly positive color gradients whereas the intermediate redshift galaxies possess predominantly negative color gradients . however , the rest - frame @xmath0 color profiles may be affected by the presence of dust . in theory , dust can cause the color profiles of galaxies to be redder towards the center if we assume that the dust would be more concentrated at the center and would produce more extinction there @xcite . @xcite conducted monte carlo simulations of light rays traveling through a dusty medium in order to understand to what extent dust can affect color gradients in late - type systems . they concluded that although a fraction of the color gradients could be attributed to dust , an additional explanation is needed to account for the total reddening in the system . @xcite produced color profiles extincted by dust for a set of model ellipticals from the u to the k passband and compared them with 52 early - type galaxies from the literature . they concluded that if dust is the primary cause of color gradients then the dust must be spatially extended ( @xmath79 @xmath80 ) implying high dust masses comparable to the dust masses inferred by the iras data . thus , the presence of dust may be partly responsible for producing the color gradients seen in our intermediate redshift sample for both early - type and late - type systems , however , given that the total mass of the dust and its spatial distribution are not well constrained , we can not say , for now , to what degree dust affects the color distribution in the galaxies . on the other hand , even though dust might play a role in our high redshift sample , there is no reason why its presence would cause the color profile to be bluer towards the center . as previously mentioned , dust would tend to cause extinction and , thus , reddening towards the center , not the outer parts of the galaxies . in fact , if dust is found to cause extinction in the centers of these galaxies , then the color gradients seen in our high redshift sample would be even greater . thus , although dust may play a role in generating the color gradients of our intermediate redshift sample , it can not explain the trends seen in our high redshift sample . anytime we compare observations of galaxies in two different redshift ranges , we must take into account the effect of observational selection . at high redshifts , we will be able to detect and study only those galaxies with the highest luminosities and surface brightnesses , thus , the trends that we see in the surface brightness and color profiles may potentially be affected by this bias . we have tried to minimize some selection effects by observing the galaxies at different redshifts in the same rest - frame wavelengths . this guarantees that at all redshifts , we will be probing the same population of stars . furthermore , at high redshifts we sample not only galaxies with spectroscopic redshifts , but also galaxies with photometric redshifts which are , in general , fainter than those galaxies that can be detected spectroscopically . thus , we are sampling the fainter population at high redshifts and are not only just picking out the uv luminous galaxies . @xcite performed simulations on galaxies in the hdf - n in order to understand the selection effects that may arise when studying galaxies at different redshifts . she takes the galaxies with @xmath81 in the f814w band and simulates their appearance at @xmath82 and @xmath83 by using the corresponding images in f606w and f450w filters respectively to minimize k - corrections . she finds that at @xmath82 only 33% of the original objects can still be detected and by @xmath83 , only 3% of them remain . to further understand the sampling of the galaxy population as a function of redshift , she defines a concentration versus asymmetry index ( @xmath84 ) . she plots the @xmath85 relation for galaxies in the hdf separating them into five redshift bins with redshifts ranging from z @xmath14 0.6 to redshifts z @xmath20 3.0 . from these plots , she concludes that whereas the range of @xmath86 and @xmath87 values remain constant for z @xmath3 1.2 , at higher redshifts , the number of high @xmath87 objects seem to decrease while the number of high @xmath86 objects seem to increase . since high concentration is associated with early - type systems such as ellipticals and early - type spirals and high asymmetry is associated with merging systems , it seems from her plots that at z @xmath20 1.2 , there is a decrease in the early - type systems and an increase in merging systems . however , she shows in her work , that these trends can naturally be explained by selection effects . in her simulations she has shown that , by @xmath83 most of these galaxies with high concentration have faded so much that they `` disappear '' into the noise and can no longer be detected . from wu s ( 1999 ) discussion it is quite possible that the trends we see in the surface brightness profiles may be influenced by these selection effects . in our work with the surface brightness profiles , if we had not corrected for the psf effect , we would have seen the same decrease in early - type systems . if we just inspect the srsic indices of the galaxies and ignore the effect of psf smearing , we would naively classify 5 out of a total of 32 ( 16 @xmath11 7% ) of the galaxies in the intermediate sample and only 1 out of the 20 ( 5 @xmath11 5% ) galaxies with h @xmath14 24.5 in the high redshift sample as possessing an @xmath13 profile . thus , it would seem that the number of early - type systems had decreased with redshift . however , since we have corrected for the effect of the psf , we recover the number of galaxies with @xmath13 profiles ( i.e. 28 @xmath11 9% for the intermediate redshift sample and 60 @xmath11 15% in the high redshift sample ) . thus , by correcting for the psf , we take care of one selection effect . however at z @xmath20 2 , we are still sampling only the tip of the surface brightness function where asymmetries are more evident . consequently , it is not surprising that we would find a slightly higher fraction of `` irregular '' galaxies in the high redshift sample , 20 @xmath11 10% as opposed to 6 @xmath11 3% in the intermediate redshift sample . although the selection effects may be responsible for some of the trends seen in the morphological classification of galaxies , they can not explain the trends that exist in the color profiles . _ they can not explain why at high redshifts , we detect a significant population of galaxies possessing large positive color gradients but at lower redshifts , they seem to `` disappear '' . _ in order to fully comprehend galaxy formation and evolution , it is important to compare observations with theoretical models . however , only in the last decade has it been possible to study galaxy formation observationally . previously , we were only able to predict how galaxies form and evolve by using n - body simulations and semi - analytical modeling of dark matter halos , which demonstrated that structure builds up into larger and larger units through continual accretion and merger @xcite . these models gave support to the `` hierarchical clustering '' scenario pioneered by @xcite which is a less restricted version of the standard cold dark matter ( cdm ) model since it allows for the assumption of different cosmologies . in the last several years , much advancement has been made in the observational regime since many have used the lyman break technique @xcite to reliably detect galaxies at high redshift in large numbers . with the detection of numerous high redshift galaxies at high resolution and sensitivities , it is now feasible to compare observations with theoretical predictions . from our surface brightness analysis of the galaxies in the hdf - n , we find that as we look further back in time , the star formation regions are more centrally concentrated and their morphologies are more irregular , deviating from the traditional hubble sequence . although our results may be affected by uncertainties and selection effects , we can compare to models to test for consistency and determine what may be the driving force behind the trends seen if they are at the very least attributed in some way to galaxy evolution . we have discussed the possibility of mergers and interactions as the cause of the trends seen . in order to understand how our results fit in with the overall picture of galaxy formation and evolution , we compare our observations of galaxies in the hdf - n with the predictions from the semianalytic model of galaxy formation described by bcfl and spf which focuses particularly on the high redshift lyman break galaxies . in their model , bcfl used a monte carlo simulation to generate binary merger trees which essentially describes the merging histories for dark matter halos with predetermined final masses . they generated the mock high redshift progenitors by applying the selection criteria for lyman break candidates described in @xcite and quantified the amount of cold gas available by measuring the rate at which cold gas in thin shells can cool to form stars . they chose values for the star formation timescale , the feedback parameter , the shape of the initial mass function ( imf ) , the overall luminosity normalization , and the merger timescale for galaxies that would best reproduce the local b and k band luminosity functions . lastly , they fixed the threshold mass for a galaxy merger causing a disk to become a spheroid to match the relative population of ellipticals , s0 s , and spirals locally . they produced their model for three different cosmologies : the standard cdm model ( @xmath88 , @xmath89 , @xmath90 ) , the flat cdm model ( @xmath91 , @xmath92 , @xmath93 , @xmath94 ) , and the open cdm model ( @xmath95 , @xmath93 , @xmath96 ) . from their simulations , bcfl show that present day galaxies that make up the bright end of the luminosity function have at least one lyman break progenitor at @xmath97 furthermore , depending on their merging histories , these lyman break progenitors can end up as a l @xmath20 l * galaxy with any morphological type . if they experienced major mergers at recent epochs they may evolve to become ellipticals and s0s . if they grew by the quiescent accretion of cooling gas , the gas might form a disk around a bulge that was created by an earlier merger event . and if they experienced minor mergers which are too weak to destroy the disk , they might accumulate more stars in the central bulge . they expect the lyman break galaxies to be rotating disks with typical half - light radii @xmath98 and predict that the stars of the progenitors will congregate primarily in the central regions . bcfl additionally conclude that the star formation of galaxies at all redshifts occurs quiescently with sfrs that never obtain a value higher than a few solar masses per year . their conclusion matches the observations of lbgs only if internal dust extinction is ignored . however , in reality , the effects of dust is probably non - negligible although how much of a role it plays is still presently uncertain . @xcite measured the extinction to be a factor of @xmath21 3 at 1500 whereas @xcite and @xcite report values closer to @xmath99 . now the general consensus is to use an extinction correction of @xmath21 5 @xcite . if the central blueing seen in the color profiles of our high redshift sample is caused by a central starburst , which seems likely , then bcfl s assumption of quiescent star formation in an extended disk does not agree with our observations . on the other hand , the results of the model from spf would agree well with what we observe in the color profiles and with what we currently know about the sfrs of lbgs . instead of undergoing quiescent star formation , they favor the idea that lbgs are experiencing what they term as `` collisional starbursts '' which describe starbursts that are induced by galaxy - galaxy mergers . spf explain that they obtain different results from bcfl because of the different assumptions made in the two models . the assumptions made in bcfl for supernovae feedback , gas cooling , and types of mergers were such that , taken in combination , would act to suppress star formation at early epochs ( please refer to bcfl and spf for the specifics on the differences in the two models ) . spf use a semi - analytic modeling of galaxies in the cdm hierarchical clustering framework adopting the @xmath100 cdm cosmology ( @xmath101 , @xmath102 , @xmath103 , @xmath104 ) to understand the nature of lyman break galaxies . their models take into account the effects of the cooling rate of gas , supernovae feedback , galaxy - galaxy mergers , and star formation and subsequent evolution . they address the effects of dust extinction by following the recipe described in wang & heckman ( 1996 ) for nearby starbursts which returns the optical depth for a given extinguished uv luminosity , intrinsic uv luminosity , and face - on optical depth . they generate monte - carlo based merger trees in which each branch in the tree is associated with a merger . within this framework , they considered three different models to represent the variety of star formation events . the `` collisional starburst '' model is one in which star formation occurs quiescently with additional star formation occurring in bursts . the `` constant efficiency quiescent '' model only involves quiescent star formation with no bursts . and the `` accelerated quiescent '' model involves quiescent star formation but with the star formation rate increasing with look - back time . spf conclude that among the three models , the `` collisional starburst '' model best reproduces the observations of the lbgs including comoving number density , star formation rates , internal velocity dispersions and the observed luminosity function , global star formation rate density , and cold gas and metallicity abundances in the universe . thus , they advocate that star formation occurred at a higher rate in the high redshift universe and that mergers were responsible for increasing the star formation rate by triggering gas inflow to the central regions . the positive color gradients we see at high redshifts would then be associated with nuclear starbursts induced by mergers and/or interactions , which could also cause star formation to `` migrate '' over time from the central regions to the spiral arms where we see most of the star formation today . that is , when galaxies are first formed , the stars would tend to be born in the dense central regions . then some galaxies may undergo mergers , which funnel gas to the central regions causing a burst of star formation . after awhile , the gas will be depleted and dispersed such that during the late stages of merger , the central light profiles will evolve to follow the @xmath13 law @xcite . during the late stages of the merger , the presence of a central bulge might suppress the inflow of gas and consequently reduce the star formation efficiency @xcite . then if the galaxy experiences further interactions with another galaxy , material can rain in on the outer parts causing the gas to collapse to form stars around the spheroid . therefore , if the hierarchical structure formation scenario represents an accurate prediction of structure formation , then at redshifts between 2.0 @xmath14 z @xmath14 3.5 , we may be witnessing the epoch at which galaxies are experiencing mergers and their star formation is enhanced and centrally concentrated . from our surface brightness and color analysis of galaxies at a range of redshifts probed at the same rest - frame wavelengths in the hubble deep field - north , we conclude : 1 ) . the color profiles of galaxies reveal that at earlier epochs , there are fewer galaxies with old red stellar populations in their centers and most galaxies have centrally concentrated star formation . as in local samples of galaxies , our intermediate redshift galaxies ( 0.5 @xmath14 z @xmath14 1.2 ) have redder central regions due to a combination of age , metallicity and dust gradients . since age gradients are the only viable explanation for the central blueing of our high redshift sample , we conclude that the majority of high redshift galaxies contain centrally concentrated starbursts . 2 ) . from the galaxy models , we have demonstrated the importance of taking into account the psf effects before attempting to classify morphologies . after correcting for the psf effects in the surface brightness profiles , we find that the population of galaxies possessing @xmath13 profiles have slightly increased with look - back time while those with exponential profiles have slightly decreased . our results also suggest a statistically insignificant increase in the fraction of galaxies with peculiar / irregular structure at higher redshifts . the results should be viewed with a note of caution , however , due to the small sample size and large uncertainties in the galaxy models especially at small scale lengths and faint magnitudes . if , as predicted by hierarchical galaxy formation models , mergers and interactions played an important role in the lifetime of these galaxies , such processes would be responsible for the positive color gradients indicative of centrally concentrated star formation seen in the high redshift galaxies . spf show that their `` collisional starburst '' model , in which bursts are triggered by mergers , best reproduces the observed properties of lyman break galaxies and of the universe in general . such centrally condensed nuclear star bursts are consistent with our observed trends in the color gradients . a good way to test whether the `` collisional starburst '' model holds is by applying the color - asymmetry diagram developed by @xcite to distinguish between starbursts driven by mergers and interactions from starbursts which were ignited by some other method . since the hubble deep field represents only a small volume of space , we must be cautious in generalizing our results to the universe as a whole . we need to obtain more observations of high redshift galaxies to measure critical quantities such as their virial masses and estimation of dust content . we also need to better understand how and to what degree observational selection comes into play when we compare galaxies at different redshifts . given that we are now detecting fainter galaxies at higher redshifts and that instruments with multi - object spectroscopic capabilities in the near - ir and 8-m class telescopes are now readily available , much information will undoubtedly be unraveled in the near future concerning the nature of lbgs which will allow us to formulate a more robust theory on the formation and evolution of galaxies . we would like to thank r. guzmn for his much appreciated suggestions during the finishing stages of the paper , m. dickinson for providing extremely informative answers to questions about the data , k. wu for her support and useful discussion on selection effects , devriendt et al . for kindly emailing us their seds , r. pina and j. radomski for their help on idl , and e. mckenzie for a useful discussion on k - corrections . funds for this research were made possible by the career grant ast-9875448 and nasa grant `` nicmos map of the hdf '' go-07817.05 - 96a . p. moth also acknowledges funding from the florida space grant fellowship . abraham , r. g. , tanvir , n. r. , santiago , b. x. , ellis , r. s. , glazebrook , k. , & van den bergh , s. 1996 , , 279 , 47 barnes , j. e. 1992 , , 393 , 484 barnes , j. , & hernquist , l. 1996 , , 471 , 115 baugh , c. m. , cole , s. , frenk , c. s. , & lacey , c. g. 1998 , , 498 , 504 budavri , t. , szalay , a. s. , connolly , a. j. , csabai , i. , & dickinson , m. 2000 , , 120 , 1588 bunker , a. 1999 , , astro - ph/9907196 byun , y. i. , freeman , k. c. , & kylafis , n. d. 1994 , , 432 , 114 code , a. d. , & welch , g. a. 1979 , , 228 , 95 cohen , j. g. , cowie , l. l. , hogg , d. w. , songaila , a. , blandford , r. , hu , e. m. , & shopbell , p. 1996 , , 471 , l5 cole , s. , aragon - salamanca , a. , frenk , c. s. , navarro , j. f. , & zepf , s. e. 1994 , 271 , 781 conselice , c. j. , bershady m. a. , & gallagher , j. s. , iii 2000 a & a , 354 , 21 de boer , k. s. 1982 , a & as , 50 , 247 de jong , r. s. 1996 , a & a , 313 , 377 de vaucouleurs , g. 1948 , anap , 11 , 247 devriendt , j. e. g. , guiderdoni , b. , & sadat , r. 1999 a & a , 350 , 381 dickinson , m. , 1999 , in _ after the dark ages : when galaxies were young , _ eds . s. holt & e. smith , ( aip ) , 122 dickinson , m. et al . 2000 , , 531 , 624 evans , r. 1994 , , 266 , 511 franx , m. , illingworth , g. , & heckman , t. 1989 , , 98 , 538 freeman , k. c. 1970 , 160 , 811 guzmn , r. , jangren , a. , koo , d. c. , bershady , m. , & simard , l. 1998 , , 495 , l13 heckman , t. m. , armus , l. , & miley , g. k. 1990 , , 74 , 833 hogg , d. w. , neugebauer , g. , armus , l. , matthews , k. , pahre , m. a. , soifer , b. t. , & weinberger , a. j. 1997 , , 113 , 2338 jedrzejewski , r. i. 1987 , , 226 , 747 kauffmann , g. , nusser , a. , & steinmetz , m. 1997 , , 286 , 795 koo , d. c. , bershady , m. a. , wirth , g. d. , stanford , s. a. , & majewski , s. r. 1994 , , 427 , l9 koo , d. c. , guzmn , r. , faber , s. m. , illingworth g. d. , bershady , m. a. , kron , r. g. , & takamiya , m. 1995 , 440 , l49 lowenthal et al . 1997 , , 481 , 673 meurer , g. , heckman , t. , & calzetti , d. 1999 , , 521 , 64 meurer , g. , heckman , t. , lehnert , m. , leitherer , c. , & lowenthal , j. 1997 , , 114 , 54 marleau , f. r. & simard , l. 1998 , , 507 , 585 mihos , j. & hernquist , l. 1994 , , 425 , 13 hernquist , l. & mihos , j. 1995 , , 448 , 41 mihos , j. & hernquist , l. 1996 , , 464 , 641 mihos , c. j. , richstone , d. o. , & bothun , g. 1992 , , 400 , 153 negroponte , j. & white , s. d. m. 1983 , , 205 , 1009 oconnell et al . 1992 , , 395 , l45 ohl et al . 1998 , , 505 , l11 oke , j. b. & sandage , a. 1968 , , 154 , 210 peebles , p. j. e. 1970 , 75 , 13 peletier , r. f. , valentijn , e. a. , & jameson , r. f. 1990 , a & a , 233 , 62 pettini , m. , king d. , smith , l. , & hunstead , r. 1997 , , 478 , 536 pettini , m. , kellogg , m. , steidel , c. c. , dickinson , m. , adelberger , k. l. , & giavalisco , m. 1998 , , 508 , 539 roukema , b. f. , quinn , p.j . , peterson b. a. , & rocca - volmerange , b. 1997 , , 292 , 835 sawicki , m. , & yee , h. 1998 , , 115 , 1329 shlosman , i. , begelman , m. c. , & frank , j. 1990 , nature , 345 , 679 silva , d. r. & elston , r. 1994 , , 428 , 511 somerville , r. , primack , j. r. , & faber , s. m. 2001 , , 320 , 504 srsic , j. l. 1968 , atlas de galaxias australes ( cordoba : observatorio astronomica ) steidel , c. , adelberger , k. , dickinson , m. , giavalisco , m. , pettini , m. , & kellogg , m. 1998 , , 492 , 428 steidel , c. c. & hamilton , d. 1993 , , 105 , 2017 steidel , c. c. , pettini , m. , & hamilton , d. 1995 , , 110 , 2519 steidel , c. c. , giavalisco , m. , dickinson , m. , & adelberger , k. l. 1996 , , 112 , 352 tamura , n. & ohta , k. 2000 , , 120 , 533 tamura , n. , kobayashi , c. , arimoto , n. , kodama , t. , & ohta , k. 2000 , , 119 , 2134 thompson , r. i. , storrie - lombardi , l. j. , weymann , r. j. , rieke , m. j. , schneider , g. , stobie , e. , & lytle , d. 1999 , , 117 , 17 trujillo , i. , aguerri , j. a. l. , cepa , j. , & gutirrez , c. m. 2001 , , 321 , 269 vader , j. p. , vigroux , l. , lachieze - rey , m. , & souviron , j. 1988 , a & a , 203 , 217 wang , b. & heckman , t. 1996 , , 457 , 645 white , s. d. & frenk , c. s. 1991 , , 379 , 52 williams et al . 1996 , , 112 , 1335 wise , m. w. & silva , d. r. 1996 , , 461 , 155 wu , k. 1999 , ph.d . thesis , university of california , santa cruz zepf , s. e. , moustakas , l. a. , & davis , m. 1997 , , 474 , l1	we fit elliptical isophotes to the hubble deep field - north wfpc-2 and nicmos data to study the rest - frame @xmath0 color profiles and rest - frame b surface brightness profiles of 33 intermediate redshift galaxies ( @xmath1 ) with @xmath2 @xmath3 25 and 50 high redshift galaxies ( @xmath4 ) with @xmath5 @xmath3 27 . from the weighted least - squares fit to the color profiles we find that , at intermediate redshifts , the galaxies possess negative color gradients ( @xmath6 @xmath7/@xmath8log(r ) @xmath9= @xmath10 @xmath11 0.007 mag dex@xmath12 ) indicating a reddening towards the center of the profile similar to local samples whereas , at high redshifts , the galaxies possess positive color gradients ( @xmath6 @xmath7/@xmath8log(r ) @xmath9= 0.272 @xmath11 0.007 mag dex@xmath12 ) indicating that star formation is more centrally concentrated . although the presence of dust can cause some reddening to occur towards the centers of the profiles seen at intermediate redshifts , it can not explain the strong central blueing of light seen at high redshifts . thus , we are witnessing a population of galaxies with strong positive color gradients at high redshifts which do not seem to exist in large numbers at lower redshifts . this indicates that star formation is more centrally concentrated in the distant galaxy sample which differs from the prevalent mode of extended disk star formation that we observe in the local universe . additionally , we find that it is critical to correct for psf effects when evaluating the surface brightness profiles since at small scale lengths and faint magnitudes , an @xmath13 profile can be smoothed out substantially to become consistent with an exponential profile . after correcting for psf effects , we find that at higher look - back time , the fraction of galaxies possessing exponential profiles have slightly decreased while the fraction of galaxies possessing @xmath13 profiles have slightly increased . our results also suggest a statistically insignificant increase in the fraction of peculiar / irregular type galaxies . we compare our results with recent semi - analytical models which treat galaxy formation and evolution following the cold dark matter hierarchical framework .
the optimistic situation during the years before the crisis in the second half of 2008 had trended from simple , marked to marked liquid products to progressively more complex and illiquid products which were marked to model . although the degree of exoticity of products increased , they were still priced under considerably simple modeling assumptions . this led to a lot of reasonable but already limited variations of pricing models which in the very end were priced under refined assumptions of simple models such as black - scholes , hull - white , gaussian copula and so on . wrong use of these simple models was one of the reasons which led to the 2008 crisis . after the crisis the trend has evolved toward less exotic and more liquid products valued under more complex pricing assumptions . however , there is still a big concern about model risk and an increasing trend toward provision or reserve calculation as an internal policy of the financial institution . markets such as foreign exchange continue trading liquidly first generation barrier products ( barrier call / put , touch and no - touch options ) which still exhibit a considerable model risk uncertainty . this means that although products are getting simpler , model risk estimation will be from now on a big concern anyway . before the crisis , this first generation of fx exotic products were valued with the volga - vanna ( vv ) model ( see @xcite , @xcite , @xcite and @xcite ) . this is a heuristic analytical model based on the black scholes price with a smile correction given by the price difference of a portfolio of three vanilla options valued with smile and at - the - money volatilities . this portfolio of vanilla options is calculated to match the vega ( @xmath0 ) , vanna ( @xmath1 ) and volga ( @xmath2 ) of the barrier black - scholes price ( @xmath3 ) with at - the - money volatility . it has been justified ( see @xcite ) that this model provides accurate results for non - path - dependent options ( e.g. quanto options ) . however , nothing could be concluded for path - dependent options such as barriers . a provision calculation philosophy should cover the expected hedging loss plus some of the uncertainty of that loss . some characteristics of a good provision policy is the fact that they should be transparent and easy to compute ( so that both front office and risk department can calculate them easily ) . they should be dynamic ( change through time ) and stable with a smooth decreasing evolution as expiry approaches ( the closer to expiry the less uncertainty and therefore the less model risk provision ) . from a management perspective , provisions should balance risk mitigation with business limitation . this means that they can provide the means to approve marketing derivative products to customers using limited models with controllable risk . for instance , if a deal can be valued with a slow but accurate model but the number of deals is huge , product marketing could be approved with a simpler less accurate model if the appropriate provision is applied . in addition , provisions introduce an incentive to improve models . a less accurate model may be charged a high provision so that when it gets improved , the provision is reduced . model risk can arise out of many different sources of uncertainty : unrealistic model hypothesis , inaccurate calibration , different calibration strategies , different sets of calibration products or externally input parameters which are not calibrated to market ( e.g. correlations , dividends , mean reversion ) . therefore , model risk could be estimated on a deal - by - deal basis by calculating sets of prices varying these sources of uncertainty . for example , a set of prices can be calculated varying one or more of these sources of uncertainty : model hypothesis ( e.g. change mean reversion , correlations or evaluate same deal with different models ) , calibration sets , calibration strategies or externally - input parameters not calibrated to market . this set of prices gives an interval which could be interpreted as a bid - offer spread ( this bid - offer spread would be theoretically composed by the set of arbitrage - free prices according to the arbitrage free measures given by all possible underlying hypotheses or `` models '' ) . the absolute value of the difference between the highest price ( offer side ) and the lowest price ( bid side ) would give an estimate for model risk for a single deal . the key question now would be which parameters to change and how much they should be changed . in addition , this strategy is usually carried out on a deal - by - deal basis and portfolio diversification is not considered . furthermore , each product may have its own way of estimating model risk . in this context , the provision would be a fraction or multiplier of this model risk quantity . through this approach , the senior management can clearly define their risk profile . this `` absolute '' way of measuring model risk is formalized and analyzed in a systematic way by @xcite . in this reference , a given product is valued under different arbitrage - free pricing measures which must satisfy certain properties . each measure gives a different price and all the prices are confined in an interval ( a bid - offer spread ) . the model risk measure is then defined as the difference between the offer and bid prices . this can be regarded as an `` absolute '' model risk measure , because any pricing hypothesis can virtually be considered and the interval can get progressively broader as long as more pricing measures are incorporated . in practice , only a limited number of measures can be considered ( e.g. changing unobservable market parameters , calibration sets or hypotheses of the evolution of the underlying risk factors ) . this limited number of pricing measures allow estimating model risk . the portfolio effect can be considered when the whole portfolio is priced with each measure and model risk is calculated from the bid - offer spread obtained for the whole portfolio . this paper does not address this `` absolute '' model risk approach . instead , it is concerned with a relative model risk measure of a particular product priced with a given model with respect to a reference model under which the market dynamics are assumed to be driven . the relative measure is defined by the expected shortfall at a given significance level of the hedging strategy of the product priced according to its model , assuming that the market is driven by the reference model . the advantage of the expected shortfall measure is the fact that it is coherent ( see @xcite ) . this means that it satisfies a number of properties which provide a framework to deal with model risk in a systematic way . this relative measure provides a homogeneous framework to compare products priced with different models under the same reference . the products chosen are double - no - touch and fader options , the models under analysis are volga - vanna and black - scholes and the reference model is heston . the ideal situation would be that the reference model is chosen to be a market proxy . the original motivation of this work was to have a reasonable criterion to decide from a list of models which one was the most appropriate to price and manage a product . if the risk department is worried because many double - no - touch deals are closed what would be your answer if your boss asks you : what would you use to manage these positions ? a heuristic model such as the volga - vanna or the well - known black - scholes model but without skew pricing hypothesis ? getting back to basics , simulating the hedging strategy seemed a reasonable way to answer this question . considering hedging strategies to contrast valuation models is not new ( see for instance @xcite , @xcite and @xcite ) . in this context , the main goal of this paper is to estimate relative model risk in a way which could be consistent and robust , general purpose and applicable to a portfolio of deals . the approach followed by this paper is to simulate the hedging strategy . therefore , only analytical models can be considered in practice as the computational cost is very high . that s why this study was carried out with the help of a grid of computers . two assumptions are considered for the hedging strategy . the first one assumes that the market volatility surface is driven by heston s model ( see @xcite ) calibrated to market at a given time horizon . the second one is the fact that hedging inaccuracies from using actual sensitivities provided by the pricing model or using several vanillas to hedge the movements of the volatility surface are eliminated . this is possible because market dynamics are known and the hedging ratios are calculated to hedge the two known factor which drive the market . this methodology can be applied to a single deal or a portfolio of them . every model is evaluated under the same framework and therefore , different models and products can be quantitatively compared with each other . a reasonable provision would be equal to this relative model risk measure . this method will be applied to a single deal ( double - no - touch option ) and a portfolio of fader options with the aid of a grid of computers . the first one will be valued under black scholes and volga - vanna hypothesis and the second one only using volga - vanna . under the black - scholes @xcite framework , i.e. the asset follows a lognormal distribution on the real world measure with drift @xmath4 , there exists a clear mapping between the real world probability @xmath5 and the risk neutral probability @xmath6 used for pricing and hedging . this is also true under dupire s framework @xcite . apart from these two models it is difficult to find other with such direct relationship . for example , merton @xcite jump diffusion model assumes that jumps are part of the firm idiosyncratic risk . since the market portfolio cancels all firm idiosyncratic risk by diversification , according to capital asset pricing model @xcite , no risk premium should be taken into account . heston @xcite model needs to assume that investors have a very particular utility function to derive the partial differential equation of the option price and find a correspondance between pricing measures @xmath5 and @xmath6 . beyond these ideal cases authors have become more and more aware that a perfect hedging might not be possible either because the market is incomplete ( a unique risk neutral probability does not exist ) or because for practical reasons traders might want to use a simpler and faster model for pricing and hedging . the former case has been extensively studied by scheweizer @xcite , contributing to the mean - variance hedging technique and zhou @xcite , embedding the problem into a `` markowitz '' type problem . the latter case has been treated by corielli @xcite using @xmath7 energy inequities for parabolic equations which allow estimating expected pricing and hedging errors between the `` true '' model and the one used by the trading desks . our work could be regarded as a practitioner approach to this latter type of problem . although heston s model has been used as `` real world probability '' it could be argued that as long as the risk department is able to provide a coherent evolution of the underlying and the vanilla options , the methodology would still hold . nevertheless , the presented methodology is not directly meant to solve the mean - variance hedging problem rather than to study for a given model developed by a front office team , how well will it behave as market conditions are stressed . the paper starts with the formulation of the hedging strategy in section [ sec : formulation ] . section [ sec : dnt ] applies the methodology to double - no - touch options valued with black scholes and volga - vanna models . section [ sec : modelriskmeasure ] defines the relative model risk measure which will allow comparing different models in the following sections . section [ sec : dnt ] applies the methodology to double - no - touch options valued with black - scholes and volga - vanna models . it is concluded that double - no - touch products have high model risk and hedging with volga - vanna is better than black - scholes . section [ sec : faders ] applies the methodology to a portfolio of forward fader options using only the volga - vanna model . it is concluded that model risk is irrelevant for this product . finally , section [ sec : conclusions ] ends with some conclusions . this section presents the formulation of the hedging strategy which will be used to compare different products evaluated with different models . subsection [ sec : marketsimulation ] shows the hypothesis assumed for the market and how to calculate the hedge ratios . subsection [ sec : implementationhedging ] explains the implementation of the hedging strategy through an intuitive derivation . the rigorous derivation of the hedging strategy is presented in subsection [ sec : rigorousderivationhedging ] justifying why model risk leads to a drift in the profit or loss of the hedging strategy and quantifying it on average . the main and hardest assumption of this hedging strategy is that the market is driven by heston s process given by equation ( [ eq : marketprocess ] ) , where @xmath8 is the underlying spot price at time @xmath9 , @xmath10 and @xmath11 are the time - dependent domestic and foreign interest rates , @xmath12 is the variance process , @xmath13 is the long - term variance , @xmath14 and @xmath15 are correlated brownian motions with correlation @xmath16 ( @xmath17 ) and @xmath18 is the volatility of the variance process . @xmath19 this model has two risk factors : the evolution of the underlying @xmath8 and the variance @xmath12 . therefore , on a given date , the set of model parameters ( @xmath20 , @xmath13 , @xmath18 and @xmath16 ) and these two factors completely define the market . this means that the spot and forward volatility surface ( out of vanilla and forward start call and put options ) are completely defined ( see @xcite and @xcite for a semi - analytical implementation of heston s pricing for spot and forward start vanilla options ) . this assumption that the market is driven by heston s process might not be realistic ( even if it is calibrated to market for a given horizon ) . however , it will be seen in the following sections that in spite of this fact , this hypothesis provides a framework in which different models can be compared . under this assumption , the evolution of the market is obtained by simulating the factors @xmath8 and @xmath12 at each point of time using a set of model parameter which were obtained calibrating the model to real market on a particular date . from these factors , the implied volatility surface is calculated for each day and path simulated by this hypothetical market . the hedging strategy is built in order to neutralize the uncertainty of these two factors . the hedging products used are the underlying and an at - the - money 6 month vanilla call option . thus , the sensitivities of the premium of the product being hedged and the hedging products ( the 6 month vanilla option ) are calculated with respect to @xmath8 and @xmath12 as indicated by equation ( [ eq : greekdefinition ] ) , where @xmath21 and @xmath22 . @xmath23 the premium @xmath24 is calculated by tweaking @xmath12 , rebuilding the whole volatility surface with an initial variance @xmath25 and repricing the product with the rebuilt volatility surface ( for the 6 month vanilla it is only necessary to change the initial variance leaving heston s parameters constant ) . the same should be done for @xmath26 . however , as the fx volatility surface is expressed in terms of delta ( rather than strike ) , the volatility surface does not change when @xmath8 is varied . it is important to stress that @xmath27 is the standard fx delta ( moving the spot without changing the implied volatility surface ) . however , it would not be the standard delta in an equity context ( the surface should be rebuilt changing @xmath8 ) . concerning @xmath28 , it is clearly not the standard vega ( premium sensitivity under a parallel movement of the volatility surface ) . these sensitivities are `` external '' in the sense that they are not calculated by the pricing model itself but by the hedging algorithm to properly hedge the two factors driving the hypothetical market . in this context , the sensitivities given by the model are ignored , only the premium is used . @xmath29 if vega sensitivities provided by the model were used to compute @xmath28 , equation ( [ eq : vega_model ] ) should be used instead of equation ( [ eq : greekdefinition ] ) , where @xmath30 is the actual vega returned by the model at different maturity and delta buckets and @xmath31 is the sensitivity of heston s implied volatility of each bucket with respect to the change of the initial variance @xmath12 . under heston s model , implied volatility movements with respect to @xmath12 have only term structure ( all delta buckets move parallel to each other ) . this means that to hedge the risk factor @xmath12 , only the term structure of vega returned by the model would be needed ( the sum of equation ( [ eq : vega_model ] ) would be only across maturities ) . if the market were represented by a mixed volatility model ( see @xcite ) , the movements of the implied volatility surface would depend on both the spot level and the initial variance and the implied volatility sensitivities with respect to the initial variance would not be equal across delta buckets . in addition , a vanna contribution @xmath32 should be added . the reason to calculate @xmath28 according to equation ( [ eq : greekdefinition ] ) instead of equation ( [ eq : vega_model ] ) is to avoid the hedging noise of the prediction error of the premium change obtained out of model sensitivities for the given movement of the implied volatility surface . however , comparing the final distribution of the hedging cost when @xmath28 is calculated according to equations ( [ eq : greekdefinition ] ) and ( [ eq : vega_model ] ) would provide a quantitative measure of the quality of the sensitivities provided by the model and their capability of predicting premium change ( this would be very useful for model validation purposes ) . see that a regular trader would only know @xmath30 ( the regular vega sensitivity provided by the model used to manage the position ) but not @xmath31 because the dynamics of the market are unknown for the trader . however , a regular trader would not use a single option ( e.g. 6 month at - the - money ) as the study presented here , but most likely three vanillas ( to account for the level , steepness and convexity of the volatility surface ) for a number of maturities . in the study presented in this paper , it is possible to hedge all movements of the volatility surface with a single vanilla because the dynamics of the market are known and therefore the movements of the volatility surface with respect to the market driving factors ( @xmath8 and @xmath12 ) . in a more realistic situation , the dynamics of the market would be unknown but several vanilla options would be used for hedging leading to a similar result . the more realistic approach is not followed here because the hedging would be worse and the purpose of this paper is to provide an estimation of the model risk premium , filtering out the imperfections of hedging with several vanillas and using the sensitivities provided by the model . comparing model risk when these imperfections are not filtered out is also very interesting but will be left for a future research . as the foreign exchange implied volatility surface is expressed in terms of delta ( and not strike ) , it is build through an iterative method . this method starts with an initial term structure of implied volatility given by the solution of the deterministic part of the variance process ( @xmath33 ) . the fx volatility surface conventions are converted into the simple black scholes setting ( spot delta , premium not included in delta , atm straddle , call and put on foreign currency ) . strikes are calculated out of the surface delta buckets and the initial term structure of implied volatility . from the strikes , vanilla call option prices are calculated according to heston s model with the given set of parameters and their implied volatilities are calculated thereafter . black scholes call deltas are then calculated out of the strikes and implied volatilities . they will not coincide with the delta buckets of the surface expressed as call option deltas . therefore , the implied volatilities of the delta buckets are interpolated from the just calculated delta versus implied volatility points and the process starts again . strikes are calculated with the new volatilities , then prices , implied volatilities and deltas which get compared with those of the buckets . the process repeats until the deltas obtained in the iteration are close enough to the delta buckets of the surface . this section presents the implementation of the hedging strategy using an intuitive derivation . the rigorous justification for continuous time will be provided in subsection [ sec : rigorousderivationhedging ] to interpret the findings of sections [ sec : dnt ] and [ sec : faders ] . consider now a portfolio @xmath34 composed by the portfolio of options to hedge @xmath35 ( this is not cash but positions on option contracts ) valued with the model under test and the hedging portfolio @xmath36 as indicated by equation ( [ eq : pitotal ] ) . equation ( [ eq : pihedgebf ] ) shows that the hedging portfolio is composed by a cash amount @xmath37 in domestic currency which will get funded , an amount of underlying @xmath38 and @xmath39 units of the 6 month vanilla call option whose price is denoted by @xmath40 . see that the underlying security is not just @xmath8 but @xmath41 , where @xmath42 is the process for the evolution of the foreign bank account given by equation ( [ eq : bt_f ] ) and @xmath11 is the foreign risk free rate ( an equivalent `` bank account '' could be defined for dividend payments ) . this underlying is the real traded security ( the foreign bank account @xmath42 expressed in domestic units ) and that is why is is used as the underlying security . @xmath43 @xmath44 @xmath45 consider a set of instants of time @xmath46 $ ] up to a final horizon @xmath47 ( intervals are usually daily or even more frequent ) where the hedging ratios @xmath48 and @xmath49 will change . assuming no dependence of the product and the hedging porfolio with respect to time ( time is frozen ) the condition to derive @xmath38 and @xmath39 would be given by equation ( [ eq : dpitot_eq0 ] ) . namely , there is neither gain nor loss in each time interval . @xmath50 equations ( [ eq : dpi ] ) and ( [ eq : dpihedge ] ) show an approximation of the change of @xmath35 and @xmath51 , where @xmath52 and @xmath53 are the deltas of the portfolio @xmath35 and @xmath40 , and @xmath54 and @xmath55 are their sensitivities with respect to the variance process @xmath12 . these sensitivities are calculated according to equation ( [ eq : greekdefinition ] ) . replacing equations ( [ eq : dpi ] ) and ( [ eq : dpihedge ] ) in ( [ eq : dpitot_eq0 ] ) yields the values of @xmath38 and @xmath39 given by equation ( [ eq : alphabeta ] ) . @xmath56 @xmath57 @xmath58 see that the sensitivities @xmath59 and @xmath60 of equation ( [ eq : dpi ] ) are calculated using the model under analysis , whereas the sensitivities @xmath61 and @xmath62 are calculated with a premium according to the model assumed for the market . equation ( [ eq : hhedge ] ) shows the value of the hedging position @xmath63 at time @xmath64 with hedge ratios calculated at time @xmath65 . @xmath66 the steps of the hedging strategy start with a given portfolio of deals or a new deal which is bought or sold . at a given point of time , the hedging strategy starts with the price @xmath67 of a deal or portfolio . this price can be calculated with a range of models ( local volatility , volga - vanna , black scholes , heston and so on ) . this price is positive for a net long position and negative for a short position . the hedging portfolio would then start with a cash position , @xmath68 , and no position in underlying or options ( @xmath69 and @xmath70 ) . if @xmath67 is positive , this means that a long position is held and it has to be financed by borrowing money at the inter - bank prevailing rate . if @xmath67 is negative , a net short position would be considered instead and the equivalent amount of money would be received and it could be lent at the prevailing inter - bank lending rate should be deposited into the collateral account of the counterparty and the collateral prevailing rate ( e.g. eonia ) would be received on a daily basis . ] . ignoring transaction costs , the implemented hedging routine process would be given by the following steps : * sell the hedging position @xmath71 bought in the previous instant ( the first time there is no position ) . * buy a new hedging position with the updated hedge ratios . * between selling and buying the hedging portfolio , management of coupon payments , strike fixings , barrier touching , cancellations , expiration of options and so on should be carried out ( incoming payments are positive in the cash account @xmath37 and outgoing ones are negative ) . * add interest earned ( if cash amount is positive ) or paid ( when it is negative ) on the cash amount in domestic currency and add foreign interest earned on the the underlying position in foreign currency . the last step accounts the financing costs ( depending whether the balance is positive or negative or there is a collateral agreement ) . in an equity context , continuous dividend payments should be deposited into the cash account , causing a falling drift of the underlying process . @xmath72 @xmath73 @xmath74 equation ( [ eq : pihedge_t0 ] ) shows the resulting hedging portfolio on the first date ( the amount in parenthesis is held in cash ) . see that the only thing which has been done is to buy the new hedge position @xmath75 and its price , @xmath76 , has been subtracted from the cash account . equation ( [ eq : pihedge_t1 ] ) shows the hedging portfolio at @xmath77 which has been held one period and therefore the cash position gets accrued by @xmath78 , where @xmath79 is the bond value on @xmath65 of a bond expiring on @xmath64 . see that the hedging position is sold at the beginning of the second period with the positive income @xmath80 of the hedging position of the previous instant @xmath81 ( valued at @xmath77 ) and the negative outcome , @xmath82 , to buy the new hedging position @xmath83 . finally , equation ( [ eq : pihedge_tn ] ) shows the hedging portfolio for a generic instant @xmath47 . the summation term shows the cumulative profit or loss of holding the hedging position from one instant to the following . the negative term @xmath84 is the cash outflow to buy the hedging position @xmath85 at time @xmath47 . the construction of the evolution of the hedging portfolio of equation ( [ eq : pihedge_tn ] ) helps understanding how the implementation is carried out in discrete time . although the derivation of the hedging ratios is intuitive but not rigorous , section [ sec : rigorousderivationhedging ] shows that the hedge ratios are indeed those obtained in equation ( [ eq : alphabeta ] ) . in addition , this section will also show that when the hedge ratios are calculated with a model whose pricing assumptions do not reflect the market driving factors , a consistent profit or loss leakage will be seen on a daily basis in the cash account of the hedging portfolio . the formerly described hedging strategy is computationally expensive even if the underlying portfolio of options is priced with analytical models . this is so because front office pricing models require the underlying spot , interest curve and a detailed volatility surface as inputs . therefore , at each point in time it is necessary to calculate the complete volatility surface consistent to heston s parameters for the hypothetical market from the simulated @xmath8 and @xmath12 . @xmath86 \label{eq : request}\ ] ] to perform this task a computer grid was set up . on the client side a set of paths ( @xmath8 , @xmath12 ) was generated . each one of these paths together with the portfolio of options under management , @xmath35 , would become a request message as indicated by equation ( [ eq : request ] ) . @xmath87 \label{eq : result}\ ] ] on the server side , each @xmath88 path ( the calculation of the volatility surface together with the hedging strategy ) is executed in a single server computer out of the total number of servers of the grid . the result message would contain the evolution of the profit and loss @xmath34 , the value of the whole portfolio @xmath35 , @xmath89 and @xmath90 over time as showed by equation ( [ eq : result ] ) . consider the change of value of the portfolio @xmath34 given by equation ( [ eq : ddiscpi ] ) which is composed by the options to hedge @xmath35 and the hedging portfolio @xmath51 given by equation ( [ eq : pihedgebf ] ) . @xmath91 when the whole randomness has been hedged out , the portfolio , @xmath34 should earn the risk free rate ( in this example the domestic rate @xmath92 ) as indicated by equation ( [ eq : ddiscpi ] ) . @xmath93 @xmath94 equation ( [ eq : df_ito ] ) shows the dynamics of any pricing model , @xmath95 , whose inputs are the spot price and the volatility surface given by the model assumed for the market ( the function @xmath95 only depends on the factors @xmath8 and @xmath12 ) . the dynamics are obtained by applying it s formula to the function @xmath95 . the quadratic variations corresponding to the model assumed for the market are given by equation ( [ eq : quadraticvar ] ) . @xmath96 @xmath97 @xmath98 equations ( [ eq : dct ] ) and ( [ eq : dpi ] ) show the dynamics of the option to hedge , @xmath35 , and the 6 month call option used for hedging , @xmath40 . these expressions are obtained replacing the quadratic variations ( [ eq : quadraticvar ] ) and the market dynamics ( [ eq : marketprocess ] ) in equation ( [ eq : df_ito ] ) for @xmath99 and @xmath100 . the coefficients @xmath101 , @xmath52 , @xmath102 and @xmath54 are obtained using equation ( [ eq : greekdefinition ] ) and @xmath103 given by equation ( [ eq : loperator ] ) is the differential operator ( infinitesimal generator ) of heston s process assumed for the market . @xmath104 + s_t db_t^f } \right ) + \beta _ t dc_t \\ \begin{array}{{20}c } { } & { } \\ \end{array } = \left ( { r_t^d b_t + \alpha _ t r_t^d s_t b_t^f } \right)dt + \alpha _ t \sqrt { v_t } s_t b_t^f dw_t + \beta _ t dc_t \\ \end{array } \label{eq : dpihedge}\ ] ] @xmath105 equation ( [ eq : dpihedge ] ) shows the dynamics of the hedging portfolio , where @xmath106 is given by equation ( [ eq : bt_f ] ) and it is assumed that @xmath107 . replacing equations ( [ eq : dct ] ) , ( [ eq : dpi ] ) , ( [ eq : dpihedge ] ) and ( [ eq : pihedgebf ] ) in equation ( [ eq : ddiscpi ] ) and moving the right hand side of equation ( [ eq : ddiscpi ] ) to the left , yields equation ( [ eq : dpi_dpihedge ] ) . this equation represents the change of value of the total portfolio @xmath108 through time , taking out the domestic risk free return on the total portfolio . this change should be ideally equal to zero according to equation ( [ eq : ddiscpi ] ) . the right hand side shows two stochastic terms multiplying @xmath14 and @xmath15 . the hedge ratios @xmath38 and @xmath39 should then be chosen to cancel these two terms to remove the uncertainty or randomness of the process . see that the hedge ratios which cancel these two terms are equal to those obtained in section [ sec : implementationhedging ] and given by equation ( [ eq : alphabeta ] ) . the expression @xmath109 is heston s partial differential equation ( see @xcite ) which should be equal to zero for a derivative following the dynamics of the market . see that for the case @xmath100 , the equation is satisfied and @xmath110 because the vanilla option used for hedging is valued according to the assumed market . @xmath111 replacing @xmath112 in equation ( [ eq : dpi_dpihedge ] ) and eliminating the terms equal to zero yields equation ( [ eq : dpitot ] ) . if the premium @xmath35 were calculated with the same model assumed for the market , the right hand side of equation ( [ eq : dpitot ] ) would be equal to zero and the change of the total portfolio , @xmath108 , minus the interest rate earnings would be zero ( the desired result ) . this means that when the pricing model is not the same model assumed for the market dynamics , the right hand side of equation ( [ eq : dpitot ] ) will be different from zero and the evolution of @xmath108 will have a drift . for negative drifts , @xmath108 will show a consistent loss through time . this is what happens in the example of section [ sec : dnt ] . section [ sec : faders ] shows the opposite case ( the drift is positive ) where a consistent gain is achieved through time . @xmath113 consider now equation ( [ eq : dpitot_tn ] ) , where @xmath114 represents the final position of a portfolio of contracts ( and not cash ) and @xmath115 is the hedging portfolio whose initial value is equal to @xmath67 valued with the model under test . taking expectations , @xmath116 $ ] , conditioned on the information at @xmath81 with respect to the risk free measure of the process assumed for the market and considering @xmath117 as numeraire ( at @xmath47 it is equal to 1 ) yiels equation ( [ eq : exppi_tn ] ) . this expression gives the expected value of @xmath118 or what would be lost or earned on average when the model under test is different from the market model . the first term is directly the price of the portfolio @xmath114 according to the market model ( the true value ) . @xmath119 = p_{t_0 , t_n } ^d { \bf e}_{t_0 } ^{mkt } \left [ { \pi _ { t_n } } \right ] + p_{t_0 , t_n } ^d { \bf e}_{t_0 } ^{mkt } \left [ { \pi _ { t_n } ^{hedge } } \right ] \\ \begin{array}{{20}c } { } & { } & { } & { } & { } & { } \\ \end{array } = \pi _ { t_0 } ^{mkt } - p_{t_0 , t_n } ^d \frac{{\pi _ { t_0 } } } { { p_{t_0 , t_n } ^d } } = \pi _ { t_0 } ^{mkt } - \pi _ { t_0 } \\ \end{array } \label{eq : exppi_tn}\ ] ] equation ( [ eq : exphedge ] ) details the calculation of the expectation of the hedging strategy . this expression is obtained applying the tower law ( or iterated expectations ) to equation ( [ eq : pihedge_tn ] ) ( see that the cash amount @xmath120 cancels the cash value of the position @xmath121 on underlying and @xmath122 on vanilla call ) . the expectation on the right hand side is equal to zero , because @xmath123 over the numeraire @xmath124 is a martingale as indicated by equation ( [ eq : exph_pd ] ) , because @xmath123 is composed of two traded assets : the underlying @xmath125 and the vanilla call option @xmath40 . equation ( [ eq : exps_pd ] ) shows the martingale condition for the underlying and the vanilla call . @xmath126 = \frac { { - \pi _ { t_0 } } } { { p_{t_0 , t_n } ^d } } + \overbrace { { \bf e}_{t_0 } ^{mkt } \left [ { \sum\limits_{i = 1}^n { \left ( { { \bf e}_{t_{i - 1 } } ^{mkt } \left [ { \frac{{{\rm h}_{t_i } ^{t_{i - 1 } } } } { { p_{t_i , t_n } ^d } } } \right ] - \frac{{{\rm h}_{t_{i - 1 } } ^{t_{i - 1 } } } } { { p_{t_{i - 1 } , t_n } ^d } } } \right ) } } \right]}^{{\rm equal \ ; to \ ; } 0 } \label{eq : exphedge}\ ] ] @xmath127 = { \bf e}_{t_{i - 1 } } ^{mkt } \left [ { \left ( { \alpha _ { t_{i - 1 } } \frac{s_{t_i } b^f_{t_i}}{{p_{t_i , t_n } ^d } } + \beta _ { t_{i-1 } } \frac{c_{t_i } } { { p_{t_i , t_n } ^d } } } \right ) } \right ] = \frac{{{\rm h}_{t_{i - 1 } } ^{t_{i - 1 } } } } { { p_{t_{i - 1 } , t_n } ^d } } \label{eq : exph_pd}\ ] ] @xmath128 = \frac{s_{t_{i - 1 } } b^f_{t_{i-1}}}{{p_{t_{i - 1 } , t_n } ^d } } } & { } & { { \bf e}_{t_{i - 1 } } ^{mkt } \left [ { \frac{{c_{t_i } } } { { p_{t_i , t_n } ^d } } } \right ] = \frac{{c_{t_{i - 1 } } } } { { p_{t_{i - 1 } , t_n } ^d } } } \\ \end{array } \label{eq : exps_pd}\ ] ] the conclusion of equation ( [ eq : exppi_tn ] ) can not but be overstressed : the average profit or loss of the hedging strategy is the difference of the market price minus the price of the model under test . if the price of the contract was bought at a lower price than the market , a profit should be realized on average . on the other hand , if the position was sold for a lower price than the market , some loss must be realized on average ( when a short position is considered , @xmath67 and @xmath129 are negative ) . this conclusion will be verified in section [ sec : dnt ] for the first case study . see that this conclusion does not depend on the hedging ratios @xmath38 and @xmath39 ( they could be even random ) because the hedging strategy is self - financing ( the hedging ratios are fixed along each hedging period ) . this means that the impact of the hedging ratios are not on the average profit or loss of the hedging strategy , but its uncertainty ( e.g. the standard deviation ) as will be seen in the coming sections . therefore , hedge ratios given by a good model close to the market will reduce the uncertainty of the profit or loss and hedge ratios given by another worse model will lead to more dispersion in the distribution of the hedging profit or loss . when a deal is closed , the average hedging loss can be easily estimated through fair value adjustment ( fva ) by comparing the price of the model used to manage the position ( the model which gives the hedging ratios ) with another better pricing model ( perhaps too slow for management ) which reflects better the hypothesis of the market . this provision needs to be calculated and accounted for only once when the operation is closed . however , the standard deviation of the hedging loss is considerably more difficult to estimate and may have a bigger impact , depending on the quality of the hedging ratios given by the model . the following sections will estimate this standard deviation of the hedging error . the relative model risk of a product priced according to a given model with respect to another reference model is defined by the expected shortfall at a given significance level of the profit and loss given by the hedging strategy of the product priced with its model under the assumption that the market is driven by the dynamics of the reference model . equation ( [ eq : mr_es ] ) presents the definition of expected shortfall in a formal way , where @xmath130 is the profit and loss of the hedging strategy of product @xmath131 , priced with model @xmath132 with respect to the market reference model @xmath133 , @xmath134 is the significance level for which the expected shortfall is calculated and @xmath135 is the quantile function given by equation ( [ eq : x_alpha ] ) . @xmath136 - \frac{{x_\alpha } } { \alpha } \left ( { \alpha - p\left [ { x_{p_i , m_j } ^{m_{ref } } \le x_\alpha } \right ] } \right ) \label{eq : mr_es}\ ] ] @xmath137 in simpler words , the expected shortfall for @xmath138 is the expectation of the hedging profit and loss of the worst 35% paths . see that the expected shortfall is defined as a positive number when losses ( negative values of @xmath130 ) occur . @xmath139 @xmath140 @xmath141 @xmath142 the main advantage of this measure is the fact that it is coherent ( see @xcite ) . this means that some properties are satisfied which allow dealing with model risk in a more systematic way . these properties are monotonicity given by equation ( [ eq : prop1 ] ) , sub - additivity ( this property is not satisfied by the measure value - at - risk and considers that diversification can only reduce risk ) by ( [ eq : prop2 ] ) , positive homogeneity by ( [ eq : prop3 ] ) and translation invariance by ( [ eq : prop4 ] ) . this section applies the hedging methodology of section [ sec : formulation ] to estimate the model risk of a double - no - touch ( `` dnt '' ) foreign exchange option when the initial valuation and the management throughout the life of the option is carried out using two models with different valuation hypothesis : the volga - vanna ( `` vv '' ) and black scholes ( `` bs '' ) models . a number of conclusions will be reached concerning model risk ( what happens when a limited model is used to quote and manage a deal ) and provision calculation . the selected framework is very appropriate to compare different models because the market dynamics are known ( heston s model ) and therefore the correct price of the product . the comparison among models is homogeneous and provides an objective criterion to judge and measure which model is better . the deal under test is a double - no - touch option maturing 1 year and 17 days from present time . the underlying is the eur / usd foreign exchange rate . this contract pays the notional in usd at maturity provided that the underlying does not breach two barrier levels ( 1.2130 and 1.3622 ) . if any of the barriers get touched , the product cancels and pays out zero . figure [ fig : calibratedvolsurf ] shows the volatility surface of the eur / usd generated by heston s model with parameters calibrated to market on august 22nd , 2006 for a 1 year horizon . the market is skewed favoring out - of - the - money calls ( they have higher volatility ) . the spot price is 1.2812 and interest rates have been set to zero ( both domestic and foreign ) . figure [ fig : dntspot_vol ] shows the evolution of the underlying @xmath8 and volatility ( @xmath147 ) for 120 simulated paths according to heston s model with the parameters indicated in figure [ fig : calibratedvolsurf ] . when the barriers get touched , the path ends ( the deal disappears ) and the last live spot or volatility point is maintained up to the end of the contract drawing a straight line . the few straight lines which are not too close to the barrier levels are due to daily big movement events ( up to 3 standard deviation ) . ) and volatility ( @xmath147 ) paths simulated according to heston s model . straight lines show a barrier breach on the following date.,title="fig:",scaledwidth=45.0% ] ) and volatility ( @xmath147 ) paths simulated according to heston s model . straight lines show a barrier breach on the following date.,title="fig:",scaledwidth=45.0% ] carr p. , sun j. , a new approach for option pricing under stochastic volatility , _ review of derivatives research _ , vol . 10 , no . 2 , 87 - 150 , may 2007 . available at : http://www.math.nyu.edu/research/carrp/research . elices a. , `` models with time - dependent parameters using transform methods : application to heston s model '' , _ arxiv:0708.2020v2 [ q-fin.pr]_ , october 2010 . available at `` http://arxiv.org/abs/0708.2020 '' .	this paper introduces a relative model risk measure of a product priced with a given model , with respect to another reference model for which the market is assumed to be driven . this measure allows comparing products valued with different models ( pricing hypothesis ) under a homogeneous framework which allows concluding which model is the closest to the reference . the relative model risk measure is defined as the expected shortfall of the hedging strategy at a given time horizon for a chosen significance level . the reference model has been chosen to be heston calibrated to market for a given time horizon ( this reference model should be chosen to be a market proxy ) . the method is applied to estimate and compare this relative model risk measure under volga - vanna and black - scholes models for double - no - touch options and a portfolio of forward fader options .
scrambling describes a property of the dynamics of closed quantum systems , in which initially localized information spreads out over the whole system , thereby becoming inaccessible locally . the concept of scrambling originates from the study of black holes in quantum gravity @xcite . if information escapes from a black hole , the thermal nature of the hawking radiation @xcite indicates that the state of any matter and information falling into the black hole has been scrambled and so gets lost from the perspective of an external observer . in particular , the `` fast scrambling conjecture '' @xcite states that the fastest scramblers take time logarithmic in the system size to scramble information , and that black holes are the fastest scramblers . scrambling and similar notions play important roles in other areas of physics as well . for example , scrambling is closely related to many - body localization and thermalization ( see @xcite for a recent review ) : quantum systems that exhibit localization clearly do not scramble or thermalize , since information of local initial conditions fails to spread , and so remains accessible to local measurements . by contrast , a many - body system that undergoes scrambling moves to states that appear random with respect to local measurement : here , the notion of scrambling can be seen as a form of thermalization at infinite temperature . quantum chaos is also a close relative of scrambling . under chaotic dynamics , initially local operators grow to overlap with the whole system ( the butterfly effect ) . that is , chaotic quantum systems are scramblers @xcite . in particular , the behaviors of the so - called out - of - time - order ( oto ) correlators can probe the growth of local perturbations @xcite . their role as diagnostics of chaos has led to the active application of oto correlators to the study of scrambling @xcite and many - body localization @xcite . this work is mainly motivated by two key features of scrambling . first , scrambling of quantum information and the growth of entanglement go hand in hand : information initially present in local perturbations ends up encoded in global entanglement so that it becomes irretrievable by simple measurements . entanglement captures the nonclassical essence of scrambling , and leads to information - theoretic measures of scrambling such as the entanglement entropy . second , scrambling can be achieved by a sufficiently random dynamics . the main idea of the foundational paper of this field @xcite is to use random dynamics to simulate scrambling behaviors of black holes . that is , scrambling is also related to the generation of randomness . the goal of this paper is to connect entanglement entropy , given by generalized entropies of different orders , with degrees of randomness , characterized by designs . we establish a strong connection between rnyi entanglement entropies and the degree of randomness induced by designs of the same order , in both the random unitary and random state settings . ( we note that a recent paper @xcite establishes a related connection between @xmath1-point oto correlators and @xmath0-designs via frame potentials . however , the average oto correlators that @xcite mainly studies may not directly correspond to entropies in our setting . ) generalized entropies of order @xmath0 raise the density matrix to the @xmath0-th power . the higher the order of the generalized entropy , the more sensitive that entropy is to nonuniformity in the spectrum of the density matrix . an @xmath0-design is an ensemble of states or unitaries whose first @xmath0 moments are indistinguishable from completely random states or unitaries ( drawn uniformly from the haar measure ) . the higher the order of the design , the better it emulates the full randomness of the haar distribution . our main result is that @xmath0-designs induce almost maximal rnyi-@xmath0 entanglement entropies . our analysis strengthens known results relating entanglement entropy and randomness , such as page s theorem @xcite for random states and hosur - qi - roberts - yoshida @xcite for random unitaries . these results allow us to elucidate the tradeoffs involved in creating and characterizing higher degrees of randomness . on the one hand , a completely random unitary channel ( drawn uniformly from the haar measure ) certainly scrambles information , but it requires exponentially many local gates and random bits to generate @xcite . in fact , 2-designs , which can be efficiently implemented @xcite , can already achieve information scrambling . so , for example , simple quantum circuits can create 2-designs with almost maximal von neumann entanglement entropy . on the other hand , generalized entanglement entropies of higher orders are more sensitive to nonuniformity ( such as sharp peaks ) in the spectrum of the reduced density matrix , i.e. the entanglement spectrum @xcite , which can be invisible to the ordinary ( von neumann ) entanglement entropy . pseudorandom unitaries may typically exhibit almost maximal von neumann entanglement entropy , but do not necessarily maximize high - order rnyi entanglement entropies . our results reveal a fine - grained hierarchy of complexities between information and haar scrambling . this hierarchy is defined relative to the moments of the haar measure , and can be probed by generalized entanglement entropies . to summarize , we aim at deriving tight relations between entropy and randomness . to do so , we calculate the generalized entanglement entropies averaged over state and unitary designs . now we introduce our approaches and results more specifically . to study scrambling properties of unitary channels , we map them to a dual state via choi isomorphism , and study the entanglement properties of this dual state . as in @xcite , we partition the input register of the choi state into two parts , @xmath2 and @xmath3 , and the output register into @xmath4 and @xmath5 . our results rely on the calculation of average @xmath6 , the defining element of order-@xmath0 entanglement entropies between @xmath7 and @xmath8 of the choi state . we employ tools from weingarten calculus to explicitly compute the haar integrals of @xmath6 for all @xmath0 in both the asymptotic and nonasymptotic regimes , which are by definition equal to the average over unitary @xmath0-designs . we are able to use these results to lower bound the regular rnyi entanglement entropies , as they are convex in @xmath6 . the key conclusion is that the rnyi-@xmath0 entanglement entropies averaged over unitary @xmath0-designs are almost maximal . in other words , a random unitary sampled from a unitary @xmath0-design typically exhibits nearly maximal rnyi-@xmath0 entanglement entropies . these results indicate that the rnyi-@xmath0 entanglement entropies can diagnose the randomness complexity of unitary @xmath0-designs . we also study the @xmath9 limit of rnyi , the min entanglement entropy , in particular . we show that unitary designs of an order that scales logarithmically in the dimension of the unitary achieve almost maximal average min entanglement entropy , which implies that they are already indistinguishable from haar random by the entanglement spectrum alone . higher moments make no difference in the entanglement spectrum . then we consider the more straightforward and well - known problem of entanglement in random states . previous results in this setting , e.g. page s theorem , are also not tight in a similar sense . we obtain analogous results as in the random unitary setting . most importantly , we show that ( projective ) @xmath0-designs exhibit almost maximal rnyi-@xmath0 entanglement entropies , which can be regarded as a tight version of page s theorem . as in the case of unitary designs , state designs of logarithmic order already maximize min entanglement entropy . moreover , we are able to show that there exists a projective 2-design such that all higher rnyi entanglement entropies are bounded away from maximum . the existence of such a 2-design establishes a clear separation between the entropic complexity of 2-designs and those of higher orders . our results reveal intrinsic connections between the order of generalized entropies and the moment of randomness . we also include several results related to e.g. rnyi entropies , designs , and weingarten calculus , which may be of independent interest . the paper is organized as follows . in sec . [ sec : prelim ] , we formally define the central concepts of this paper : the generalized quantum entropies and designs . sections [ sec : randu ] and [ sec : rands ] contain analysis of the choi model of random unitary and the random state settings respectively . finally we provide some concluding remarks about our results and future directions in sec . [ sec : dis ] . peripheral results and technical details are included in the appendix . see , e.g. , @xcite for a comprehensive introduction of standard and soft notations of asymptotics ( e.g. big - o and soft big - o ) that will be used throughout this paper . the theme of this paper is to establish connections between generalized quantum entropies and designs , which we shall formally introduce in this section . some parametrized generalizations of the shannon and von neumann entropy , most importantly the rnyi and tsallis entropies , are found to be useful in both classical and quantum regimes . we will find that rnyi entropies are more suitable for characterizing scrambling than tsallis entropies . here we focus on entropies defined on a quantum state @xmath10 living in a finite - dimensional hilbert space . a unified definition of generalized quantum entropies is given in @xcite : the quantum unified @xmath11-entropy of a state @xmath12 is defined as @xmath13.\ ] ] the two parameters @xmath0 and @xmath14 are respectively referred to as the order and the family of an entropy in this paper . the @xmath15 element plays a key role in this paper . entropies specified by a certain order @xmath0 are collectively called @xmath0 entropies . the @xmath16 limit gives the von neumann entropy . by fixing @xmath14 , one obtains a family of entropies parametrized by order @xmath0 . we define the following function to be the characteristic function of an entropy : @xmath17 which is obtained by treating @xmath15 as the argument @xmath18 . the convexity of this characteristic function is important to many of our results . the most representative families are rnyi ( the limiting case @xmath19 ) and tsallis ( @xmath20 ) . however , @xmath21 entropies ( e.g. tsallis ) have some undesirable features for our purposes . most importantly , the maximal value is dependent on @xmath0 , thus it does not make much sense to directly compare @xmath21 entropy of different orders @xmath0 , which is key to our results . moreover , they are not even additive on maximally mixed states , so on runs into trouble when deriving e.g. tripartite information from entanglement entropy . in addition , notice that for @xmath22 ( excluding tsallis ) , the characteristic function is concave in the relevant @xmath23 regime , which prevents us from using jensen s inequality to establish several crucial bounds . these problems will be explained in more detail later . in contrast , rnyi entropies do not have these problems . in this work , we mainly focus on this well - behaved family : the quantum rnyi-@xmath24 entropy of a state @xmath12 is defined as @xmath25 for @xmath26 , @xmath27 is singular and defined by taking a limit . @xmath28 is the max / hartley entropy ; @xmath29 is just the von neumann entropy . the @xmath30 limit , the min entropy , plays a special role in our study : the quantum min entropy of a state @xmath12 is defined as @xmath31 where @xmath32 denotes the operator norm of @xmath10 , @xmath33 is the largest eigenvalue of @xmath10 . other rnyi entropies are well - defined by eq . ( [ renyidef ] ) . the @xmath34 case @xmath35 , often called the rnyi-2 or collision entropy , is also a widely used and highly relevant quantity . in the context of scrambling , a key result of @xcite is that the rnyi-2 entanglement entropy is directly related to the 4-point oto correlators , thus can probe chaos . ref . @xcite also discusses scrambling in a wess - zumino - witten model using this quantity . also notice that @xmath36 is directly related to the quantum purity @xmath37 ( recall that less pure subsystems dictate entanglement ) , and is so frequently employed in the study of entanglement @xcite . it can be directly verified that rnyi entropies are additive and have the same maximal value @xmath38 for @xmath38 qubits . an important feature of rnyi entropy for analyzing degree of scrambling is that the values of rnyi entropy of different orders can in principle be well gapped : consequently , the behavior of states and unitaries at different orders of rnyi entropy can be used to distinguish between different degrees of scrambling . here we give a simple example . consider a density operator in @xmath39-dimenisonal hilebert space has one large eigenvalue @xmath40 , and the rest of the spectrum is uniform / degenerate . that is , the spectrum reads @xmath41 then the min entropy only cares about the largest eigenvalue by definition : @xmath42 which is @xmath43 ( linear in the number of qubits ) away from maximum . however , rnyi-2 is insensitive to this single peak : @xmath44 which is almost maximal ( with a residual constant ) . this establishes a clear separation between the low and high ends of rnyi entropies . in fact , @xmath45 produces @xmath43 gaps between all finite orders : @xmath46 for @xmath47 . so the slope decreases with @xmath0 . indeed , it equals 1 for @xmath34 , and approaches @xmath48 in the @xmath9 limit . the intuition is simply that promoting the power of eigenvalues essentially amplifies the nonuniformity of the spectrum . similar separations will be constructed in the random state setting . in the appendix , we derive more properties of rnyi entropies , including inequalities relating different orders of rnyi entropies ( appendix [ app : renyiineq ] ) , and a weaker form of subadditivity ( appendix [ app : renyisub ] ) . fig . [ fig : entropy ] summarizes the important generalized entropies in the relevant regime @xmath49 . -entropies , @xmath49 . italicized names refer to the whole line . ] in general , the notion of design is introduced to characterize distributions that can produce the behaviors of certain moments of the uniform distribution , and so can be considered as good approximations to complete randomness . the idea applies to both quantum states and unitary channels , as we shall formally introduce now . complex projective designs are distributions of vectors on the complex unit sphere that are good approximations to the uniform distribution ( pseudorandom ) , in the sense that they match the uniform distribution up to certain moments @xcite . they are of interest in many research areas , such as approximation theory , experimental designs , signal processing , and quantum information . there are many equivalent definitions of exact designs ( see @xcite ) . here we mention a few that are directly relevant to the current study . thr canonical definition is based on polynomials of vector entries . define @xmath51 as the space of polynomials homogeneous of degree @xmath50 both in the coordinates of vectors in @xmath52 and in their complex conjugates . an ensemble @xmath53 of pure state vectors in dimension @xmath39 is a ( complex projective ) _ @xmath50-design _ if @xmath54 where the integral is taken with respect to the ( normalized ) uniform measure on the complex unit sphere in @xmath52 . the second definition , based on the frame operator , will be useful in the error analysis . let @xmath55 be the @xmath50-partite symmetric subspace of @xmath56 with corresponding projector @xmath57}$ ] . the dimension of @xmath55 reads @xmath58}=\binom{d+t-1}{t}.\ ] ] the @xmath50-th frame operator of @xmath53 is defined as @xmath59}\operatorname{\mathbb{e}}_\nu ( { { { { \|\psi\rangle}}\!{{\langle\psi\|}}}})^{\otimes t},\ ] ] and the @xmath50-th frame potential is @xmath60 the ensemble @xmath53 is a @xmath50-design if and only if @xmath61}$ ] or , equivalently , if @xmath62}$ ] @xcite . in analogy to complex projective designs , unitary designs are distributions on the unitary group that are good approximations to haar - random unitaries , in the sense that they match the haar measure up to certain moments @xcite . they also play key roles in many research areas , such as randomized benchmarking , data hiding , and decoupling . as in the case of state designs there are also many equivalent definitions of exact unitary designs ( see @xcite ) . similarly , we formally define unitary designs by polynomials and frame operators / potentials . let @xmath63 be the space of polynomials homogeneous of degree @xmath50 both in the matrix elements of @xmath64 and in their complex conjugates . an ensemble @xmath53 of unitary operators in dimension @xmath39 is a _ unitary @xmath50-design _ if @xmath65 where the integral is taken over the normalized haar measure on @xmath66 . the @xmath50-th frame operator of @xmath53 is defined as @xmath67,\ ] ] and the @xmath50-th frame potential is @xmath68 the ensemble @xmath53 is a unitary @xmath50-design if and only if @xmath69 , where @xmath70 is the @xmath50th frame operator of the unitary group @xmath71 with haar measure @xcite . in addition , @xmath72 and the lower bound is saturated if and only if @xmath53 is a unitary @xmath50-design @xcite . when @xmath73 , which is the case we are mostly interested in , @xmath74 we first study the scrambling properties of random unitary channels . as suggested by @xcite , we employ the choi isomorphism to map a unitary channel to a dual state , and study scrambling by the entanglement properties of this state . in this section , we first briefly introduce the choi state model , and then present several calculations of averaged generalized entanglement entropies . the results lead to an entropic notion of scrambling complexities , which we shall discuss in depth . ref . @xcite proposed that one can use the negativity of the tripartite information associated with the choi state of a unitary channel to probe scrambling . the negativity of the tripartite information is a measure of global entanglement that quantifies the degree to which local information in the input to the channel becomes non - local in the output . here we introduce the definitions and motivations of this formalism to set the stage . the choi isomorphism ( more generally , the channel - state duality ) is widely used in quantum information theory to study quantum channels as states . it says that a unitary operator @xmath75 acting on a @xmath39-dimensional hilbert space @xmath76 is dual to the pure state @xmath77 which is called the choi state of @xmath75 . now consider arbitrary bipartitions of the input register into @xmath2 and @xmath3 , and the output register into @xmath4 and @xmath5 . let @xmath78 be the dimensions of subregions @xmath79 respectively ( @xmath80 ) . one expects that , in a scrambling system , any measurement on local regions of the output can not reveal much information about local perturbations applied to the input . in other words , the mutual information between local regions of the input and output @xmath81 and @xmath82 should be small . this suggests that the negative tripartite information @xmath83 can diagnose scrambling , since it essentially measures the amount of information of @xmath2 hidden nonlocally over the whole output register . here @xmath84 is the mutual information , which measures the total correlation between @xmath2 and @xmath4 . since the input and output are maximally mixed due to unitarity , the four subregions are all maximally mixed . for example , here @xmath81 is reduced to @xmath85 , so we only need to analyze the entanglement entropy @xmath86 . note that @xmath87 can be reduced to the conditional mutual information @xmath88 @xcite , which is a quantity of great interest in quantum information theory . the haar - averaged ( completely random ) values of the terms in the von neumann @xmath89 was computed in @xcite , as a baseline for scrambling . however , it is clear that a pseudorandom ensemble ( such as a 2-design ) can already reach these roof values @xcite , which indicates that there is a hierarchy of fine - grained complexities of scrambling , corresponding to different degrees of randomness . this section aims at formalizing this observation in the same choi state model . we rewrite the information - theoretic quantities in terms of generalized quantum entropies ( such as rnyi and tsallis ) of different orders , to study the connections between these entropies and designs . by using individual indices for different subregions , we rewrite the choi state in eq . ( [ choi ] ) as @xmath90 where @xmath91 are respectively indices for @xmath79 . the corresponding density operator is then @xmath92 by tracing out @xmath8 , we obtain the reduced density operator of @xmath7 : @xmath93 the entropy of @xmath94 measures the entanglement between @xmath7 and @xmath8 . in order to compute the generalized @xmath0 entanglement entropies , we need to raise @xmath94 to the power @xmath0 : @xmath95 thus @xmath96 this result can also take more concise operator forms : @xmath97 where @xmath98 where @xmath99 denotes partial transpose on even parties . notice that @xmath100 so @xmath101 is unitary . other density operators can be derived in a similar way . again note that the input and output are maximally entangled due to unitarity , so all four subregions are maximally mixed . we first employ tools from random matrix theory and in particular weingarten calculus to study haar integrals of the trace term in generalized entropies . it is known that the haar - averaged value of each monomial of degree @xmath0 can be written in the following form @xcite : @xmath102 where @xmath103 is the symmetric group of @xmath0 symbols , and @xmath104 are called weingarten functions of @xmath105 . here @xmath106 means @xmath45 is a partition of @xmath0 , @xmath107 is the corresponding character of @xmath103 , and @xmath108 is the corresponding schur function / polynomial . notice that @xmath109 is simply the dimension of the irrep of @xmath105 associated with @xmath45 . the weingarten function can be derived by various tools in representation theory , such as schur - weyl duality @xcite and jucys - murphy elements @xcite . therefore , we obtain the following general result : @xmath110 where @xmath111 is the number of disjoint cycles associated with @xmath112 , and @xmath113 is the 1-shift ( canonical full cycle ) . one can easily recover the @xmath114 results given in @xcite from eq . ( [ general ] ) as follows . the weingarten functions for @xmath115 are @xmath116 there are 4 terms corresponding to 2 different weingarten functions : [ cols="^,^,^,^,^,^,^",options="header " , ] [ tab:2 ] plugging them into eq . ( [ general ] ) yields @xmath117 which confirms eq . ( 66 ) of @xcite . a series of results of @xcite such as an @xmath118 gap between haar and maximal rnyi-2 entanglement entropy are obtained based on this formula . more generally , we have @xmath119 where @xmath120 is the product of canonical full cycles on each of the @xmath14 blocks with @xmath0 symbols . we now analyze the asymptotic behaviors of generalized entropies in the @xmath121 limit to provide a big picture . later we shall introduce some explicit bounds that hold for general @xmath39 . to simplify the analysis , we further require that @xmath122 here . that is , all partitions are equal . this does not affect the main idea . [ [ trace ] ] trace + + + + + we first introduce a series of useful combinatorics lemmas , which play critical roles in the behavior of generalized entanglement entropies ( in particular rnyi ) . these results are known in the context of random matrix theory . we refer to appendix a of @xcite ( c.f . references therein ) for a good summary of related results . for intuition , we still include our independent proof by induction for the key result lemma [ sumcycle ] in appendix [ app : lem ] . [ sumcycle ] @xmath123 for all @xmath124 . this result can be obtained by combining lemmas a.1 and a.4 of @xcite . see appendix [ app : lem ] for our proof . [ g ] let @xmath125 be the number of @xmath124 that saturates the inequality in lemma [ sumcycle ] . then @xmath126 , i.e. , the @xmath0-th catalan number . this result follows from lemmas a.4 and a.5 of @xcite . such permutations lie on the geodesic from identity to @xmath127 . it guarantees that the gap between haar and maximal rnyi entropies is independent of the system size @xmath38 , as will become clear shortly . we note that catalan numbers frequently occur in counting problems . the first few catalan numbers are @xmath128 . some useful bounds on the catalan numbers are derived in appendix [ app : cat ] . [ cyclecor ] @xmath129 for all @xmath130 . the number of @xmath130 that saturates the inequality @xmath131 . we also need the large @xmath39 asymptotic behaviors of the weingarten function : [ asym ] given @xmath124 with cycle decomposition @xmath132 . let @xmath133 be the minimal number of factors needed to write @xmath112 as a product of transpositions . the mbius function of @xmath112 is defined by @xmath134 where @xmath135 is the @xmath38-th catalan number ( defined in lemma [ g ] ) . in some literature , @xmath136 here is written as @xmath137 where the @xmath138 actually means the length of the cycle . to avoid confusions , we stick to the number of transpositions notation . then , in the large @xmath39 limit , the weingarten function has the asymptotic behavior @xmath139 [ wgcor ] we mainly need to distinguish the following two cases : * @xmath140 : @xmath141 and @xmath142 , thus @xmath143 ; * @xmath144 : @xmath145 , thus @xmath146 . some bounds on the mbius function are derived in appendix [ app : moeb ] . now we are equipped to derive the asymptotic behaviors of @xmath147 : [ thm : asym ] for equal partitions ( @xmath122 ) , in the large @xmath39 limit , @xmath148 starting from eq . ( [ general ] ) : @xmath149 here , the second line follows from the equal bipartition assumption , the third line follows from lemma [ asym ] and corollary [ wgcor ] , and the fourth line follows from lemmas [ sumcycle ] , [ g ] and some simple scaling analysis . similarly , the asymptotic behavior of @xmath150 is @xmath151 by corollary [ cyclecor ] . [ [ s0-entropies ] ] @xmath21 entropies + + + + + + + + + + + + + + + + + + + + + + + + + + + the calculations of @xmath21 entropies ( e.g. tsallis ) are straightforward , since the term @xmath152 linearly appears in the definition . by theorem [ thm : asym ] , for positive integers @xmath153 : @xmath154 notice that the maximum value of @xmath155 for a @xmath39-dimensional state is ( when it is the maximally mixed state @xmath156 ) @xmath157 so we see a gap between the haar - averaged and maximum value of @xmath158 : @xmath159 which is vanishingly small in @xmath39 . as briefly mentioned , @xmath21 entropies are not ideal for our study for several reasons , which we elaborate here : 1 . the definitions of mutual information and tripartite information in terms of @xmath21 entropies might not make much sense , since they are not even additive on product states . recall that all partitions are in the maximally mixed state @xmath160 . however , the mutual information @xmath161 is not directly given by @xmath162 . define @xmath163 then @xmath164 which is dominated by the irrelevant @xmath165 ( @xmath166 is vanishingly small ) . the definition of entropic scrambling complexities and several other arguments rely on comparing generalized entropies of different orders @xmath0 . but we can see from eq . ( [ eq : smax ] ) that the maximally mixed values for @xmath21 entropies vary with @xmath0 , thus it also does not make much sense to do so . recall that the characteristic function for @xmath167 entropies are concave ( tsallis is linear ) , in contrast to rnyi . although theorem [ thm : asym ] enables us to directly calculate the haar - averaged @xmath167 entanglement entropies , we can not use jensen s inequality to lower bound their design - averaged values . therefore , we are not going to devote much attention to @xmath21 entropies in the rest of this work . [ [ rnyi - entropy ] ] rnyi entropy + + + + + + + + + + + + + now we analyze the more interesting case of rnyi , the @xmath19 limit . compared to @xmath21 entropies , the calculations of rnyi are trickier because of the logarithm , which nevertheless leads to some nice features such as additivity and constant roof value . we are able to obtain the following result : for equal partitions , in the large @xmath39 limit , @xmath168 by definition , @xmath169 where @xmath170 is the characteristic function for rnyi . since @xmath171 when @xmath23 , @xmath172 is convex . so @xmath173 by jensen s inequality . we note that this jensen s lower bound due to convexity ( @xmath174 ) will be repeatedly used to establish bounds for rnyi entropies . then according to eq . ( [ expsumcycle ] ) , @xmath175 notice that lemma [ sumcycle ] already guarantees that the leading correction term ( the second term ) is independent of @xmath39 asymptotically . further notice that @xmath176 is asymptotically linear in @xmath0 , and satisfies @xmath177 so the second term does not grow with @xmath0 as well . in summary , in the large @xmath178 limit , @xmath179 so the gap between the haar - averaged and maximum value of @xmath180 ( the `` residual entropy '' ) is @xmath181 that is , the average rnyi entanglement entropies over the haar measure is only bounded by a constant from the maximal . the result implies that a random unitary sampled from haar typically has almost maximal rnyi entanglement entropies for all partitions : by markov s inequality , for any @xmath182 , @xmath183 \leq \frac{o(1)}{\delta}.\ ] ] that is , the probability for @xmath184 to be bounded away from the maximum is vanishingly small . now we consider the rnyi mutual information and tripartite information based on the entanglement entropy results . first , we can directly obtain @xmath185 which is equal to @xmath186 by additivity . the results holds similarly for @xmath187 . that is , the rnyi mutual information between any two local regions of the input and output is vanishingly small compared to the system size . on the other hand , for any partition size , notice that @xmath188 second line : the whole choi state is pure so @xmath189 ; third line : the three partitions involved are maximally mixed ; fourth line : @xmath190 . under the equal partition assumption , @xmath191 this is consistent with the fact that all information of @xmath2 is kept in the whole output @xmath192 because of unitarity . as a result : @xmath193 by plugging in all relevant terms . so the rnyi tripartite information of haar scrambling is indeed close to maximum . however , we would like to mention that the rnyi-@xmath0 entropy is not subadditive except for @xmath194 , thus @xmath195 is not necessarily nonnegative . a weaker form of subadditivity of rnyi entropies is given in appendix [ app : renyisub ] . here we prove some explicit bounds on the haar - averaged trace , rnyi entropies , and in particular the min entropy , in the nonasymptotic regime . detailed definitions and lemmas can be found in the appendices . these results provide further quantitative evidence in finite dimensions that all entanglement entropies are typically almost maximal , and also help us derive further results such as nontrivial moments . [ [ trace - and - rnyi ] ] trace and rnyi + + + + + + + + + + + + + + + [ lem : sumbound ] suppose @xmath196 , and @xmath197 . then @xmath198 where @xmath199 $ ] . define @xmath200 as the number of permutations in @xmath103 with genus @xmath201 , that is , @xmath202 note that @xmath203 is the catalan number @xmath204 by lemma [ g ] . then @xmath205 according to lemma [ lem : ngpermutationt ] in appendix [ app : genus ] . as a consequence of this inequality and the assumption @xmath197 , @xmath206\leq { \mathrm{cat}}_\alpha d_ad_b^{\alpha}\left[1 + \frac{2}{3}\sum_{\delta=1}^{\infty } q^{\delta}\right ] = { \mathrm{cat}}_\alpha d_ad_b^{\alpha}\left[1 + \frac{2q}{3(1-q)}\right ] \nonumber\\ = & h(q){\mathrm{cat}}_\alpha d_ad_b^{\alpha}\leq \frac{4^\alpha h(q)}{\sqrt{\pi}\alpha^{3/2}}d_ad_b^\alpha . \end{aligned}\ ] ] here the last inequality follows from lemma [ lem : catalanbound ] , which sets an upper bound on the catalan numbers . suppose @xmath207 . then @xmath208}{\alpha-1 } , \label{finited2}\end{aligned}\ ] ] where @xmath209 . @xmath210/2 } { \mathrm{wg}}(d,\sigma\gamma^{-1})=\sum_{\zeta\in s_\alpha}\left [ \sum _ { \gamma\in s_\alpha}d^{[\xi(\zeta\gamma\tau)+\xi(\zeta\gamma)+\xi(\gamma\tau)+\xi(\gamma)]/2 } { \mathrm{wg}}(d,\zeta)\right]\nonumber\\ \leq & \sum_{\zeta\in a_\alpha}\left [ \sum _ { \gamma\in s_\alpha}d^{[\xi(\zeta\gamma\tau)+\xi(\zeta\gamma)+\xi(\gamma\tau)+\xi(\gamma)]/2 } { \mathrm{wg}}(d,\zeta)\right]\leq \sum_{\zeta \in a_\alpha } \sum _ { \gamma\in s_\alpha } d^{\xi(\gamma\tau)+\xi(\gamma ) } { \mathrm{wg}}(d,\zeta)\nonumber\\ \leq & { \mathrm{cat}}_\alpha d^{\alpha+1}\left(1+\frac{2q}{3(1-q)}\right ) \sum_{\zeta \in a_\alpha } { \mathrm{wg}}(d,\zeta)\leq \frac{a_\alpha{\mathrm{cat}}_\alpha d}{8}\left(1+\frac{2q}{3(1-q)}\right)\left(7+\cosh\frac{2\alpha(\alpha-1)}{d}\right),\label{cosh}\end{aligned}\ ] ] where @xmath211 is the set of even permutations , i.e. the alternating group . the first inequality follows from the fact that @xmath212 is negative when @xmath213 is an odd permutation ; the second inequality follows from the cauchy - schwarz inequality , note that @xmath214 ; the third inequality follows from lemma [ lem : sumbound ] , and the last inequality follows from lemma [ lem : wgsumbound ] . by plugging eq . ( [ cosh ] ) into eq . ( [ general ] ) , we immediately obtain the trace result eq . ( [ finited1 ] ) . the rnyi result eq . ( [ finited2 ] ) then follows from jensen s inequality . we see that the leading terms indeed match the asymptotic results . the overall observation is similar : the rnyi entanglement entropies is very likely to be almost maximal , for large enough @xmath39 . for intuition , we compute @xmath215 for @xmath216 based on explicit formula for weingarten functions , which also turn out to be useful later . when @xmath115 , @xmath116 when @xmath217 , @xmath218 therefore , @xmath219 [ [ min - entropy ] ] min entropy + + + + + + + + + + + the results so far only directly apply to positive integers @xmath0 . the min entanglement entropy , which corresponds to the special limit @xmath9 , plays crucial roles in the discussions of scrambling complexities . now we examine the min entanglement entropy . in particular . [ thm : aveminentropy ] @xmath220 where @xmath221 . . then we numerically obtain that @xmath223 now suppose @xmath224 . let @xmath225 . then @xmath226 so that @xmath227 consequently , @xmath228 \frac{4^\alpha d^{1-\alpha}}{\alpha^{3/2 } } \leq \frac { 4^\alpha d^{1-\alpha}}{\alpha^{3/2}},\ ] ] and thus @xmath229 the proof of eq . is completed by observing that @xmath230 when @xmath231 and @xmath232 when @xmath233 . . then follows from the convexity of @xmath234 . we note that slightly lower @xmath235 can in principle be obtained by computing to higher orders in eq . ( [ 7 ] ) , which is nevertheless not important for the main idea . as @xmath39 gets large , @xmath235 approaches the limit 4 , and @xmath236 approaches the limit 2 . as an implication of lemma [ lem : entropygap ] , theorem [ thm : aveminentropy ] with @xmath39 replaced by @xmath237 also holds when the four subregions have different dimensions , as long as @xmath238 and @xmath239 . the same remark also applies to theorem [ thm : aveminentropylog ] below . now we state a key observation : the haar integral of @xmath6 , the defining term of @xmath0 entropies , only uses the first @xmath0 moments of the full haar measure . equivalently , pseudorandom unitary @xmath0-designs are already indistinguishable from completely random by @xmath6 . more explicitly , let @xmath240 be an @xmath0-design ensemble . then we have @xmath241\ ] ] by definition . therefore , all haar integrals of @xmath6 from the last part ( eqs . ( [ general2 ] ) and ( [ expsumcycle ] ) ) directly carry over to @xmath24-designs . this observation implies that @xmath0 entropies can generically diagnose whether a scrambler is locally indistinguishable from random dynamics as powerful as @xmath0-designs . the haar - averaged tsallis-@xmath0 entropies ( @xmath20 ) are exactly saturated by @xmath0-designs since tsallis entropy is linear in @xmath15 . however , as mentioned , we can not make analogous arguments for @xmath167 : the exact saturation requires @xmath242 and the haar integral to commute asymptotically , which is not known to hold ; and the above jensen s bound does not apply since @xmath243 becomes concave . in contrast , the rnyi entropies can be lower bounded because of convexity . due to the importance of rnyi entropies , we state the results as a theorem : @xmath244 \geq f^{(\alpha)}_r\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\right ) = \frac{1}{1-\alpha}\log\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\right).\ ] ] in particular , in the large @xmath39 limit , @xmath244 \geq \log d - o(1).\ ] ] @xmath244 = \mathbb{e}_{\nu_\alpha}\left[f_r^{(\alpha)}\left(\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\right)\right]\geq f^{(\alpha)}_r(\mathbb{e}_{\nu_\alpha}[\mathrm{tr}\{\rho_{ac}^\alpha\}])=f^{(\alpha)}_r\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\right),\ ] ] where the inequality follows from jensen , and the last equality follows from the fact that @xmath240 is an @xmath0-design . by plugging in eq . ( [ expsumcycle11 ] ) , we directly obtain the asymptotic result . the theorem enables us to use the haar integrals of traces to lower bound the design - averaged rnyi entanglement entropies in all dimensions . asymptotically , the @xmath118 upper bound on the residual rnyi-@xmath0 entropy still holds . so we can conclude that tsallis and rnyi-@xmath0 entanglement entropies are very likely to be almost maximal when sampling from unitary @xmath0-designs , as well as haar . the min entanglement entropy of designs will be explicitly analyzed later . for rigorousness , we need to analyze the robustness / sensitivity of the entropic properties of designs . indeed , one also expects ensembles that well approximate @xmath0-designs to have close - to - haar @xmath0 entropies . the error analysis will be useful for e.g. relating scrambling complexities to circuit depth . here we relate typical approximation measures of designs to rnyi entropies . first , the canonical definition of designs by polynomials leads to the following direct measure of distance to designs : ensemble @xmath53 is an @xmath245-m - approximate unitary @xmath50-design ( `` m '' represents monomial ) if @xmath246\right\| \leq \epsilon , ~~\forall q^{(k)},k\leq t.\ ] ] where @xmath247 is a monomial of degree @xmath248 both in the entries of @xmath75 and in their complex conjugates . note that the bound is on each monomial with unit constant factor , otherwise the difference can be arbitrarily amplified by including more terms or change the constant . let @xmath249 be an @xmath245-m - approximate unitary @xmath0-design . then @xmath250&\leq\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}+d^\alpha\epsilon,\\ \mathbb{e}_{\omega_\alpha}\left[s_r^{(\alpha)}\left(\rho_{ac}\right)\right ] & \geq\frac{1}{1-\alpha}\log\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}+d^{\alpha}\epsilon\right)\end{aligned}\ ] ] in the large @xmath39 limit , @xmath251 \geq \log{d } - o(1 ) - \frac{1}{(\alpha-1){\mathrm{cat}}_\alpha\ln 2}d^{2\alpha-1}\epsilon\left(1+o\left(d^{-1}\right)\right).\ ] ] @xmath252-\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\leq\left\|\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}-\mathbb{e}_{\omega_\alpha}[\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}]\right\|\leq \frac{1}{d^\alpha}d^{2\alpha}\epsilon = d^{\alpha}\epsilon\ ] ] by triangle inequality , since @xmath253 is the sum of @xmath254 monomials according to eq . ( [ tr ] ) . then @xmath255 \geq f_r^{(\alpha)}\left(\mathbb{e}_{\omega_\alpha}\left[\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}\right]\right ) \nonumber\\= & \frac{1}{1-\alpha}\log\mathbb{e}_{\omega_\alpha}[\mathrm{tr}\{\rho_{ac}^\alpha\}]\geq\frac{1}{1-\alpha}\log\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}+d^{\alpha}\epsilon\right),\end{aligned}\ ] ] where the first inequality follows from jensen , and the second inequality follows from eq . ( [ eq : trerror ] ) and that @xmath234 is monotone decreasing . we can then use the @xmath256 results to analyze the perturbation . most importantly , in the large @xmath39 limit , @xmath257 = -\frac{1}{1-\alpha}\log\left(1+\frac{d^{\alpha}\epsilon}{\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}}\right ) \nonumber \\ \leq&\frac{1}{\alpha-1}\log\left(1+\frac{1}{{\mathrm{cat}}_\alpha}d^{2\alpha-1}\epsilon\left(1+o\left(d^{-1}\right)\right)\right ) \leq\frac{1}{(\alpha-1){\mathrm{cat}}_\alpha\ln 2}d^{2\alpha-1}\epsilon\left(1+o\left(d^{-1}\right)\right ) , \label{eq : renyierror}\end{aligned}\ ] ] where the first inequality follows from eq . ( [ expsumcycle11 ] ) and the following analysis , and the second inequality follows from a logarithm inequality @xmath258 when @xmath259 . so we conclude that the error in @xmath184 at most scales as @xmath260 . recall the other definition of exact designs by frame operators . the deviation of an ensemble from a unitary @xmath50-design can also be quantified by a suitable norm of the deviation operator @xmath261 the operator norm and trace norm of @xmath262 are two common figures of merit . the latter choice is more convenient for the current study : ensemble @xmath53 is a @xmath45-fo - approximate unitary @xmath50-design ( fo represents frame operator ) if @xmath263 we note that this definition is very similar to the quantum tensor product expander ( tpe ) @xcite . tpes conventionally use the operator norm , and the deviation operators relate to each other by partial transposes ( like operators @xmath264 in eqs . ( [ x],[y ] ) ) . here we can directly use the operator form of local density operators derived earlier to do an error analysis of fo - approximate designs . let @xmath265 be a @xmath45-fo - approximate unitary @xmath0-design . we define @xmath266 , and explicit write out @xmath267 : @xmath268 - \int{\rm d}u(u\otimes u^\dagger)^{\otimes\alpha},\\ \delta_\alpha(\omega_\alpha ) & = \mathbb{e}_{\omega_\alpha}[u^{\otimes\alpha}\otimes { u^\dagger}^{\otimes\alpha } ] - \int{\rm d}uu^{\otimes\alpha}\otimes { u^\dagger}^{\otimes\alpha}.\end{aligned}\ ] ] let @xmath249 be a @xmath45-fo - approximate unitary @xmath0-design . then @xmath250&\leq\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}+\frac{1}{d^{\alpha}}\lambda,\\ \mathbb{e}_{\omega_\alpha}[s_r^{(\alpha)}\left(\rho_{ac}\right ) ] & \geq \frac{1}{1-\alpha}\log\left(\int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\}+\frac{1}{d^{\alpha}}\lambda\right).\end{aligned}\ ] ] @xmath269 - \int{\rm d}u\mathrm{tr}\left\{\rho_{ac}^\alpha\right\ } = \frac{1}{d^{\alpha}}\mathrm{tr}\{\tilde\delta_\alpha(\omega_\alpha ) y_\alpha\ } \leq \frac{1}{d^{\alpha}}{\left\lvert\tilde\delta_\alpha(\omega_\alpha)\right\rvert}_1{\left\lverty_\alpha\right\rvert}\nonumber \\= & \frac{1}{d^{\alpha}}{\left\lvert\tilde\delta_\alpha(\omega_\alpha)\right\rvert}_1 = \frac{1}{d^{\alpha}}{\left\lvert\delta_\alpha(\omega_\alpha)\right\rvert}_1\leq\frac{1}{d^{\alpha}}\lambda , \ ] ] where the first inequality follows from hlder s inequality , and the second line follows from the unitarity of @xmath270 . we note that the essential difference between the different definitions of approximate designs is just in the norm used to measure the distance @xcite . letting @xmath271 recovers equivalent definitions of exact designs . as motivated in the introduction , we expect that there is a hierarchy of scrambling complexities that lie in between information scramblers and haar - random unitaries , with different levels of the hierarchy indexed by the order of design needed to mimic the scrambler . then our results of design - averaged rnyi entanglement entropies imply that we can use the generic maximality of rnyi-@xmath0 entanglement entropy as i ) a necessary indicator of the resemblance to an @xmath0-design , and ii ) a diagnostic of the scrambling complexity of an @xmath0-design , or @xmath0-scrambling . for example , if a supposedly random unitary dynamics does not produce nearly maximal rnyi-@xmath0 entanglement entropy in all valid partitions , as @xmath0-designs must do , then it is simply not close to any @xmath0-design . the maximality of rnyi entanglement entropies may not be relevant for probing designs at the global level , but it can probe the typical behaviors of entanglement between local regions which mimic designs . recall that rnyi entropy is monotone nonincreasing in the order , and all orders share the same roof value . so @xmath0-scrambling necessarily implies @xmath272-scrambling , for @xmath273 . here we consider @xmath275 more carefully , since the corresponding rnyi entropies are defined only by taking a limit , and they respectively correspond to the weakest and strongest entropic scrambling complexities . recall that @xmath16 gives the von neumann entropy , which probes information / page scrambling . first notice that 1-designs do not necessarily scramble quantum information . for example , the ensemble of tensor product of pauli operators acting on each qubit @xmath276 forms a 1-design @xcite . however , this pauli ensemble clearly does not scramble since it can not create interaction / entanglement among qubits ( so local operators do not grow ) . so any entanglement entropy will be zero . on the other hand , 2-designs are sufficient to maximize rnyi-2 entropies , which lower bounds the corresponding von neumann entropies . it is shown in @xcite that there actually exists an arbitrarily large gap between them . so one can also say that information scrambling is strictly weaker than 2-scrambling . therefore , the minimal information scrambling corresponds to a scrambling complexity in between 1- and 2-designs . another special case is @xmath9 , which leads to the min entropy @xmath277 . almost maximal min entanglement entropy directly indicates that the spectrum of the reduced density operators ( the entanglement spectrum ) is almost uniform . as the example of @xmath278 ( in section [ entropies ] ) shows , the min entropy is extremely sensitive to even one single peak in the entanglement spectrum , since it only cares about the largest eigenvalue . so it is in a sense the `` harshest '' entropy measure . the min entanglement entropy is then the strongest entropic diagnostic of scrambling : if it is always almost maximal , then the entanglement spectrum can not accommodate any peak , and we simply can not distinguish the scrambler from completely random by looking at the spectrum of any reduced density matrix alone . we call this situation max scrambling . we shall arrive at formal arguments that designs are distinguishable from haar only up to a finite order ( in both the random unitaries and random states settings ) for finite dimensions , by studying the min entanglement entropy . given the definition of max scrambling by the min entropy , one may wonder if the full haar measure is required to achieve this strongest form of entropic scrambling . here we answer this question in the negative : for a given dimension , only finite number of moments ( which scales logarithmically in the dimension ) are needed , which we call nontrivial moments . from the proof of theorem [ thm : aveminentropy ] , we can directly see that the same result holds if the average is taken over an @xmath0-design with @xmath279 . the conclusion is clear from the proof when @xmath224 . when @xmath280 , @xmath281 , so the conclusion also follows from the proof . the conclusion is obvious when @xmath282 . it remains to consider the case @xmath283 , which means @xmath284 . therefore , eq . applies , so that @xmath285 therefore , eqs . and still hold if the average is taken over an @xmath0-design with @xmath279 . we can further show that , in fact , @xmath286-design is sufficient to achieve nearly maximal min entropy : [ thm : aveminentropylog ] let @xmath240 be a unitary @xmath0-design , where @xmath287 and @xmath288 ; then @xmath289 in particular , if @xmath290 , then @xmath291 if @xmath292 , then one can show that eq . holds as in the proof of theorem [ thm : aveminentropy ] even without additional restrictions . therefore , @xmath293\right)^{1/\alpha}\leq 4 \bigl(\frac{d}{\alpha^{3/2}}\bigr)^{1/\alpha}\leq 4 d^{1/\alpha}\leq 4 d^{a/\log d}=2^{2+a},\ ] ] which confirms eq . and implies eq . . now suppose @xmath294 . if @xmath295 , then an @xmath0-design is also a @xmath296-design , so and hold according to theorem [ thm : aveminentropy ] . otherwise , @xmath297 , and the two equations follow from and with @xmath298 . the same conclusion also holds when @xmath299 . a further question then arises as to whether the entropic scrambling complexities form a strict hierarchy , i.e. , whether different complexities are gapped . of course , by the nontrivial moments result just presented , we already know that @xmath300 complexities are not well separated . a straightforward definition of a separation between @xmath0- and @xmath272-scrambling ( @xmath301 ) is the following : there exist scramblers such that the associated rnyi-@xmath272 entropies are always near maximal , but some rnyi-@xmath0 entropies can be bounded away from maximal . such separations are in principle possible according to the properties of rnyi entropies . we tried several approaches to establish general separations in the choi model , with limited success . in particular , we attempted to generalize the partially scrambling unitary model @xcite , and attempted to extend the gap results in the random state setting ( next section ) to random unitaries . the partially scrambling unitary model is used in @xcite to prove a large separation between von neumann and rnyi-2 tripartite information in the choi state setting . by contrast , as we analyze in appendix [ app : partial ] , this model is not likely to provide similar separations among generalized entropies . the analysis nevertheless reveals a rather interesting tradeoff between sensitivity and robustness between rnyi and @xmath21 entropies . however , we are able to establish gaps using projective designs in the random state setting ( see next section ) , but the results can not be directly generalized to unitary designs at the moment . the reasons will be explained in more detail later . we leave the gap problem in the choi model open . via certain models of random dynamics , we can relate entropic scrambling complexities to traditional complexities such as time and circuit depth . for example , it is shown in @xcite that @xmath302)$ ] haar - random local gates are sufficient to form an @xmath245-approximate @xmath0-design of @xmath38 qubits . by the error analysis result , one can easily see that the minimum number of gates / circuit depth needed to maximize rnyi-@xmath0 entropies scales polynomially in @xmath0 and @xmath38 : let @xmath303 so that @xmath304 , then the number of gates scales as @xmath305 , but meanwhile the error term in eq . ( [ eq : renyierror ] ) is vanishingly small , which indicates that such circuit is a good @xmath0-scrambler . that is , the scrambling complexity and the random circuit complexity ( minimum number of gates ) can be polynomially related . we note that the @xmath305 scaling can be improved to @xmath306 for @xmath307 by a recent result @xcite . from a more physical point of view , one may be interested in the minimum time it takes for a physical scrambler , say , a hamiltonian evolution of a quantum many - body system , to achieve certain scrambling complexities , which generalizes the fast scrambling conjecture . @xcite introduces the notion of design hamiltonian , and conjectures that the shortest time for local time - independent hamiltonians to achieve approximate @xmath0-designs scales roughly as @xmath308 with unknown dependence in @xmath245 . note that the error dependence will be crucial in translating it to the language of scrambling complexities : an @xmath309 dependence is sufficient to dominate @xmath310 . suppose we conjecture that the minimum @xmath0-scrambling time scales roughly as @xmath308 as well . then based on the nontrivial moment result : the minimum time for a physical system to max - scramble scales as @xmath311 . the max scrambling conjecture contrasts with the minimum page scrambling time of @xmath312 . the previous section focused on choi states , which encodes the intrinsic properties of corresponding unitary channels . we showed that random unitaries of different degrees are associated with near - maximal generalized entanglement entropies of the corresponding orders . here we consider a more straightforward problem entanglement entropy of random states to strengthen the connections between generalized entropies and designs . in this setting , we obtain analogous main results that designs maximize corresponding rnyi entanglement entropies , and that there are logarithmic nontrivial moments . we are also able to obtain solid results on the gap problem . we shall follow similar steps as in the random unitary setting but be more concise . the original problem and related results have been playing central roles in the study of e.g. black holes and quantum information theory for a long time . the famous page s theorem roughly says that the average entropy of small subsystems of a bipartite state is nearly maximal ; or in other words , a randomly sampled state is very likely to be maximally entangled . similar observations were even earlier made by lubkin @xcite and lloyd and pagels @xcite . in particular , @xcite derived the distribution of the local eigenvalues of a random state , which may imply this result . also see e.g. @xcite for further studies of this phenomenon . consider a bipartite system with hilbert space @xmath313 , where @xmath314 have dimensions @xmath315 , respectively , assuming @xmath197 . we use @xmath316 to denote the average over states drawn uniformly from the unit sphere in @xmath317 . the page conjecture @xcite , proved in @xcite , states that the average entanglement entropy of each reduced state is given by @xmath318 the gap between the average entropy and the maximum @xmath319 is bounded by the dimension - independent constant @xmath320 . in this section we shall strengthen this result by proving the gap between the average rnyi @xmath0-entropy of each reduced state and the maximum @xmath319 is also bounded by a constant that is independent of the dimensions @xmath315 and the parameter @xmath0 . suppose @xmath321 is drawn uniformly from the unit sphere in @xmath317 . to establish a lower bound for the average rnyi @xmath0-entropy , we shall derive an analytical formula for the average of the @xmath0-moment @xmath322 , where @xmath323 is the reduced state of @xmath321 for system @xmath2 . expand @xmath324 in the standard product basis @xmath325 , where @xmath326 label the basis elements for @xmath327 , and @xmath328 label the basis elements for @xmath329 . then @xmath330 the general result of the haar - averaged trace is : @xmath331}}\sum_{\sigma\in s_\alpha}d_a^{\xi(\sigma\tau)}d_b^{\xi(\sigma)},\ ] ] where @xmath332}=\binom{d_ad_b+\alpha-1}{\alpha}=\frac{d_ad_b(d_ad_b+1)\cdots ( d_ad_b+\alpha-1)}{\alpha ! } \ ] ] is the dimension of the symmetric subspace of @xmath333 . by eq . ( [ eq : rhoa ] ) , @xmath334,\ ] ] where @xmath335 therefore , @xmath336}}{\operatorname{tr}}\{p_{[\alpha]}q_\alpha\},\ ] ] where @xmath337}$ ] is the projector onto the symmetric subspace of @xmath333 , and @xmath338}$ ] is its dimension . recall that the symmetric group @xmath103 acts on @xmath333 by permuting the tensor factors , and @xmath337}$ ] can be expressed as follows @xmath339}=\frac{1}{\alpha ! } \sum_{\sigma\in s_\alpha } u_\sigma,\ ] ] where @xmath340 is the unitary operator associated with the permutation @xmath112 . simple analysis shows that @xmath341 consequently , @xmath342}}\sum_{\sigma\in s_\alpha}d_a^{\xi(\sigma\tau)}d_b^{\xi(\sigma)}.\ ] ] we noticed that similar results have been derived and rederived several times @xcite . compared to known approaches , our approach seems simpler ; in addition , it admits easy generalization to states drawn from ( approximate ) complex projective designs , which is not obvious for other approaches we are aware of . to get an intuitive understanding of eq . , it is worth taking a closer look at several concrete examples . when @xmath34 , we reproduce a formula derived by lubkin @xcite : @xmath343 from this equation we can derive a nearly - tight lower bound for the average rnyi 2-entropy , @xmath344 when @xmath345 , the averages of the first few moments are given by @xmath346 which imply that @xmath347 note that the gap of each rnyi entropy from the maximum is tied with the corresponding catalan number . this is not a coincidence . when @xmath345 is large ( @xmath348 ) , the formula for @xmath349 can be simplified : @xmath350 @xmath351}}\sum_{\sigma\in s_\alpha}d_a^{\xi(\sigma\tau)+\xi(\sigma)}=\frac{{\mathrm{cat}}_\alpha d_a^{\alpha+1 } + o(d_a^{\alpha-1})}{d_a^{2\alpha}+o(d_a^{2\alpha-2})}={\mathrm{cat}}_\alpha d_a^{-\alpha+1}+o\bigl(d_a^{-(\alpha+1)}\bigr).\ ] ] consequently , @xmath352 this equation suggests that the gap between the average rnyi @xmath0-entropy and the maximum @xmath319 is bounded by a constant when @xmath345 is sufficiently large . it turns out that this conclusion actually holds without restrictions on @xmath315 : [ lem : alphamoment ] let @xmath353 $ ] . then @xmath354 according to lemma [ lem : sumbound ] , @xmath355}}\sum_{\sigma\in s_\alpha}d_a^{\xi(\sigma\tau)}d_b^{\xi(\sigma)}\leq \frac{h(q){\mathrm{cat}}_\alpha d_ad_b^{\alpha}}{d_a^\alpha d_b^\alpha } \leq h(q){\mathrm{cat}}_\alpha d_a^{1-\alpha } \leq \frac{4^\alpha h(q)}{\sqrt{\pi}\alpha^{3/2}}d_a^{1-\alpha},\ ] ] which in turn implies that @xmath356 [ thm : are ] @xmath357 recall that rnyi @xmath0-entropy is nonincreasing with @xmath0 , so to establish the theorem , it suffices to prove the lower bound for the min entropy , which corresponds to the limit @xmath358 . then @xmath359 here the second inequality follows from lemma [ lem : avenormpower ] below . [ lem : avenormpower ] @xmath360 the conclusion is obvious when @xmath361 . when @xmath362 , note that @xmath363 is nondecreasing with @xmath0 for @xmath364 , so it suffices to prove the lemma in the case @xmath365 . then @xmath366 and @xmath367 . according to lemma [ lem : alphamoment ] , @xmath3684^\alpha d_a^{-\alpha}\nonumber\\ & \leq \frac{3-q}{3(1-q)\sqrt{32\pi q}}4^\alpha d_a^{-\alpha } \leq 4^\alpha d_a^{-\alpha},\end{aligned}\ ] ] which implies the lemma . here the last inequality follows from the observation that @xmath369<1 $ ] for @xmath370 . this fact can be verified immediately if we notice that the derivative @xmath371 has a unique zero at @xmath372 in the interval @xmath373 and that @xmath374 is monotonically decreasing for @xmath375 and monotonically increasing for @xmath376 . when @xmath377 , theorem [ thm : are ] can be improved as follows : [ thm : are2 ] for all @xmath378 , @xmath379 where @xmath380 if @xmath317 is real and @xmath381 if @xmath317 is complex . this theorem follows from lemma [ lem : averootnorm ] below . we believe that the constant @xmath382 in theorem [ thm : are2 ] and lemma [ lem : averootnorm ] can be set to one in both real and complex cases . we note that hayden and winter had a similar result @xcite , but they are not so explicit about the constant and the dimensions for which their result is applicable . the following lemma is proved in appendix [ app : lem ] . [ lem : averootnorm ] @xmath383 where @xmath380 if @xmath317 is real and @xmath381 if @xmath317 is complex . recall page s theorem , which states that haar - averaged von neumann entanglement entropies of small subsystems are nearly maximal . this theorem is not tight from the perspectives of both entropy and randomness : the haar - averaged higher rnyi entanglement entropies are generically close to maximum as well , and the complete randomness is an overkill to maximize the entanglement entropies . our results imply that page s theorem can be strengthend from both sides . similar to the random unitary setting , since @xmath384 only uses @xmath0 moments of the haar measure , all bounds on @xmath384 and @xmath385 from the last part still hold if the average is over @xmath0-designs . so we have a tight page s theorem for each order @xmath0 : let @xmath240 be an @xmath0-design . then @xmath386 in particular , when @xmath345 is large , @xmath387 here we show that the average rnyi @xmath0-entanglement entropy of quantum states drawn from an approximate @xmath0-design with sufficient accuracy is also close to the maximum . a concrete example is provided based on typical clifford orbits . given an ensemble @xmath53 of quantum states , define @xmath388}\operatorname{\mathbb{e}}_\nu ( { { { { \|\psi\rangle}}\!{{\langle\psi\|}}}})^{\otimes t}-p_{[\alpha]}.\ ] ] ensemble @xmath53 is an @xmath245-approximate @xmath0-design if @xmath389 [ lem : appdesignentropy ] let @xmath265 be an @xmath245-approximate @xmath0-design with @xmath390 . then @xmath391}},\\ \operatorname{\mathbb{e}}_{\omega_\alpha } s_r^{(\alpha)}(\rho_a)&\geq \frac{1}{1-\alpha}\log\left ( \operatorname{\mathbb{e}}{\operatorname{tr}}\{\rho_a^\alpha\}+\frac{\epsilon}{d_{[\alpha]}}\right).\end{aligned}\ ] ] according to the same argument that leads to eq . , @xmath392}}{\operatorname{tr}}\left\{(p_{[\alpha]}+\delta_\alpha(\nu))q_\alpha\right\},\nonumber\\ & = \operatorname{\mathbb{e}}{\operatorname{tr}}\{\rho_a^\alpha\}+\frac{1}{d_{[\alpha]}}{\operatorname{tr}}\left\{\delta_\alpha(\nu)q_\alpha\right\}\leq \operatorname{\mathbb{e}}{\operatorname{tr}}\{\rho_a^\alpha\}+\frac{1}{d_{[\alpha]}}\\|\delta_\alpha(\nu)\\|_1 \\|q_\alpha\\|\nonumber\\ & \leq \operatorname{\mathbb{e}}{\operatorname{tr}}\{\rho_a^\alpha\}+\frac{\epsilon}{d_{[\alpha]}},\end{aligned}\ ] ] where the last inequality follows from the assumption @xmath393 and the fact that @xmath394 , since @xmath395 is unitary . we see that @xmath396 as long as @xmath397}=o(d_a^{1-\alpha})$ ] . this leads to an entropic notion of designs : an ensemble of states @xmath53 has an @xmath0-design complexity if the average rnyi-@xmath0 entanglement entropies are nearly maximal . for convenience , we call them page complexities at the moment . as an application of lemma [ lem : appdesignentropy ] , let us consider the average rnyi entanglement entropy of clifford orbits for a multiqubit system . for simplicity we assume @xmath398 , so that @xmath399 recall that the clifford group is a unitary 3-design @xcite , so any orbit of the clifford group forms a 3-design . consequently , the average rnyi @xmath0-entanglement entropy for @xmath400 of any clifford orbit is close to the maximum , @xmath401 for any @xmath402 , where @xmath403 denotes the clifford orbit generated from @xmath402 . however , the clifford group is not a four design , and clifford orbits are in general not four designs @xcite . if @xmath402 is a stabilizer state , then @xmath404 according to @xcite . in this case the bounds for the fourth moment and rnyi 4-entropy provided by lemma [ lem : appdesignentropy ] is not very informative , note that @xmath405 and @xmath406}\approx ( d_ad_b)^4/24=d_a^8/24 $ ] . for a typical clifford orbit , by contrast , @xmath407 is much smaller @xcite . now lemma [ lem : appdesignentropy ] implies that @xmath408}}\approx 14 d_a^{-3 } + 24d_a^{-6}\approx \operatorname{\mathbb{e}}{\operatorname{tr}}\{\rho_a^4\}.\end{aligned}\ ] ] therefore , eq . also holds for typical clifford orbits when @xmath409 . again , the min entanglement entropy corresponds to the strongest entropic design complexity : if the average min entanglement entropies are always close to maximal , then we simply can not distinguish the ensemble from completely random by the entanglement spectrum . the following theorem says that designs of order @xmath410 have almost maximally uniform entanglement spectrum : [ thm : moment ] suppose @xmath324 is drawn from an @xmath0-design in a bipartite hilbert space @xmath313 of dimension @xmath411 , where @xmath412 with @xmath413 . let @xmath323 be the reduced state of subsystem a. then @xmath414 in particular , @xmath415 and @xmath416 if @xmath417 . according to lemma [ lem : alphamoment ] , @xmath418 where the first inequality follows from the fact that @xmath419 and @xmath420 given that @xmath421 by assumption . consequently @xmath422^{1/\alpha}\leq d_a^{1/\alpha}\frac{4}{d_a}\leq d_a^{a/ \log d_a}\frac{4}{d_a}= \frac { 2^{2+a}}{d_a},\\ \operatorname{\mathbb{e}}s_{\min}(\rho_a)&\geq -\log \operatorname{\mathbb{e}}\\|\rho_a\\|\geq -\log \frac{2^{2+a}}{d_a}\geq \log d_a-2-a.\end{aligned}\ ] ] in the case , @xmath298 and @xmath417 , the inequality @xmath423 holds automatically ; therefore , @xmath415 and @xmath416 . in this random state setting , we are able to establish a clear gap between the second and @xmath0-th entropic design complexities with all @xmath424 . in particular , we shall construct a family of 2-designs , the gap of whose average rnyi-@xmath0 entanglement entropies from the maximum is unbounded for @xmath425 . our construction is based on the orbits of a special subgroup of the unitary group on @xmath313 . as mentioned before , any orbit of a unitary @xmath426-design is a complex projective @xmath426-design . what is interesting , our construction of projective 2-designs does not require unitary @xmath426-designs . in this way , we also provide a novel recipe for constructing projective 2-designs , which is particularly useful when the dimension is not a prime power . consider the group @xmath427 , where @xmath428 are the unitary groups on @xmath314 , respectively . it is irreducible , but does not form a 2-design . simple analysis shows that @xmath429 has four irreducible components on @xmath430 , with dimensions @xmath431 , respectively . the symmetric subspace of @xmath430 contains two irreducible components with dimensions @xmath432 and @xmath433 . by a similar continuity argument as employed in @xcite , there must exist an orbit of @xmath429 that forms a 2-design . let @xmath434 be a fiducial vector of a 2-design with reduced state @xmath323 for subsystem a. then @xmath435 is necessarily equal to the average over the uniform ensemble , that is , @xmath436 it turns out that this condition is also sufficient . to see this , note that the condition must be invariant under local unitary transformations and thus only depends on a symmetric polynomial of the eigenvalues of @xmath323 of degree 2 , which is necessarily a function of @xmath435 given the normalization condition @xmath437 . it is worth to point out that the same conclusion also holds if @xmath428 are replaced by groups that form unitary 2-designs on @xmath314 , respectively . next we study rnyi entanglement entropies of 2-designs constructed from orbits of @xmath429 . note that eq . holds if @xmath323 has the following spectrum @xmath438 if @xmath439 , then @xmath440 therefore , @xmath441 , and the gap of all rnyi entropies from the maximum is bounded . we are mostly interested in the case in which the ratio @xmath442 is bounded by a constant , say @xmath443 . then @xmath444 consequently , @xmath445 as @xmath446 increases , the gap of @xmath447 from the maximum is unbounded whenever @xmath47 . we note that such construction can not be directly generalized to establish gaps in the choi setting . as mentioned , any orbit of a unitary @xmath50-design is a complex projective @xmath50-design , but to construct a projective @xmath50-design , a unitary @xmath50-design is not required . here the complex projective 2-design is constructed using a group that is a tensor product . however , such a group can never be a unitary 2-design . also , in the choi setting , four parties are involved , and it is not easy to ensure unitarity using the idea for constructing projective designs . new approaches are necessary for such a construction . this paper explores scrambling by establishing connections between entanglement entropies and degree of randomness . we find that there is a hierarchy of degrees of randomness that lies between the scrambling of information ( page scrambling ) and complete randomness ( haar randomness ) . our results directly relate the order of generalized entropies and designs : we show that @xmath0 designs yield almost maximal renji @xmath0 entropies . moreover , we show by min entanglement entropy that designs of order logarithmic in the dimension already exhibit almost maximally uniform entanglement spectrum , and are indistinguishable from haar by spectral properties alone . there are several open problems , especially in the choi setting for unitary designs . for example , we are not yet able to give a construction that opens a gap between scrambling at order @xmath0 and scrambling at higher order , when @xmath448 in the choi setting . that is , the gaps between the @xmath449 scrambling complexities have not been proven . although we exhibit such gaps for projective 2-designs in the random state setting , similar techniques can not be directly generalized to unitaries , as explained earlier . moreover , due to the lack of subadditivity , we know that the negative tripartite information in terms of rnyi entropies are not necessarily nonnegative . it is worth considering related problems such as when it can be negative , and further about the meanings of such derived quantities . this work focuses mostly on the intrinsic mathematical properties of unitary channels and states , instead of on physical dynamics . the relations of our results to dynamics such as generalized fast scrambling remains speculative . it would be interesting to further explore the dynamics of e.g. black holes or quantum many - body systems , and aspects of fast scrambling , under our framework . we refer to @xcite for a set of interesting results along this direction . we note that a recent paper @xcite concerns a similar gap between maximum entropy and maximum complexity of quantum dynamics . here the complexity roughly means the computational / gate complexity , which is rather difficult to rigorously analyze . it would also be interesting to further study the connections to their framework . we thank dawei ding , alan guth , aram harrow , guang hao low , yoshifumi nakata , kevin thompson , and quntao zhuang for discussions related to this work . special thank goes to aram harrow for bringing several helpful references to our attention . zwl and sl are supported by afosr and aro . eyz is supported by the national science foundation under grant contract number ccf-1525130 . hz is supported by the excellence initiative of the german federal and state governments ( zuk 81 ) and the dfg . research at mit ctp is supported by doe . first , we present a series of inequalities relating rnyi entropies of different orders . it is well known that the rnyi entropy is monotonic nonincreasing with the parameter @xmath0 , that is @xmath450 whenever @xmath451 . on the other hand , @xmath452 can also be used to construct a lower bound for @xmath453 when @xmath454 as shown below , @xmath455 in particular , this equation yields a lower bound for the min entropy @xmath456 when @xmath457 , we have @xmath458 so the difference between @xmath452 and @xmath459 is less than 1 . when @xmath460 , we have @xmath461 , so the difference between @xmath462 and @xmath452 is upper bounded by @xmath463 . next we derive another lower bound for @xmath453 in terms of @xmath452 and the min entropy in the case @xmath454 . the following equation @xmath464 implies that @xmath465.\ ] ] in particular , any rnyi @xmath466-entropy with @xmath467 is lower bounded by a convex combination of rnyi @xmath426-entropy and the min entropy , @xmath468.\ ] ] it is known that rnyi @xmath0-entropy is not subadditive except for the special case @xmath194 . the following lemma yields a weaker form of subadditivity : [ lem : entropygap ] let @xmath469 be any bipartite state on the product hilbert space @xmath470 with dimension @xmath411 . let @xmath471 be the two reduced states . then @xmath472 the first inequality in lemma [ lem : entropygap ] means that the spectrum of @xmath469 majorizes that of @xmath473 . the second and third inequalities are immediate consequences of the first one , which are are equivalent . the second one can be seen as a weaker form of subadditivity , while the third one means that the gap of rnyi entropy of a joint state from the maximum is no smaller than the corresponding gap for each reduced state . let @xmath474 for @xmath475 be an orthonormal basis for @xmath329 and @xmath476 be the corresponding projectors . let @xmath477 where @xmath478 are subnormalized states that sum up to @xmath323 . define @xmath479 where the addition in the indices is modulo @xmath480 ; note that @xmath481 . then all @xmath482 have the same spectrum , which is majorized by @xmath469 , that is , @xmath483 . consequently , @xmath484 since rnyi @xmath0-entropy is schur concave for @xmath485 , it follows that @xmath486 which confirms the second inequality in lemma [ lem : entropygap ] and implies the third inequality . we include here an intuitive proof of lemma [ sumcycle ] by induction . the intuition is that any element @xmath487 can be viewed as a local deformation of some element @xmath488 , such that @xmath489 can only increase by at most 1 . we formalize the argument below . suppose the statement is true for @xmath490 . now for some @xmath491 , look at element @xmath492 . there are two possibilities : 1 . @xmath492 appears in a cycle of length 1 : @xmath493= k+1 $ ] . so @xmath494 , where @xmath495 . write @xmath496 , where @xmath497 . then @xmath498 , with @xmath499 . compare the action of @xmath500 and @xmath501 on various elements . the only differences are : @xmath502=\sigma_-\tau_-[k ] $ ] , @xmath503=k+1 $ ] , @xmath504=\sigma_-\tau_-[k ] $ ] . it simply increases the length of a cycle in @xmath500 by 1 , and does nothing to other cycles . so @xmath505 . from the induction hypothesis , @xmath506 , so @xmath507 . @xmath508 appears in a cycle of length @xmath509 : @xmath510=k+1 $ ] , @xmath493=b $ ] for some elements @xmath511 . define @xmath512 by @xmath513=\sigma[i ] $ ] for @xmath514 and @xmath515=b $ ] . now compare the action of @xmath516 and @xmath517 on various elements . depending on the value of @xmath518 , there are two cases : 1 . the differences are : @xmath520=b$ ] and @xmath521=\sigma[1]$ ] , but @xmath522=k+1 $ ] , @xmath523=b$ ] and @xmath524=\sigma[1]$ ] . they act identically on all other elements . there are two possible effects : 1 . in @xmath516 , @xmath525 and @xmath526\ } $ ] belong to the same cycle . then @xmath527 breaks this cycle into two disjoint ones involving @xmath528\ } $ ] and @xmath529 respectively . so @xmath530 ; 2 . in @xmath516 , @xmath525 and @xmath526\ } $ ] belong to two disjoint cycles . then @xmath527 glues these two cycles together into one . so @xmath531 . 2 . @xmath532 . then @xmath533 and @xmath527 act identically on @xmath534 and in addition @xmath504=k+1 $ ] . so @xmath530 . + in conclusion , @xmath535 can only increase by 1 or decrease by 1 as compared to @xmath536 , so @xmath537 in either case . lastly , consider @xmath538 . the only element of @xmath539 is @xmath540 , and @xmath541 , thus the statement trivially holds . this completes our proof . it is well known that the catalan number @xmath542 $ ] is approximated by @xmath543 when @xmath248 is large . to make this statement more precise , here provide both lower and upper bounds for @xmath544 . [ lem : catalanbound ] the catalan number @xmath544 satisfies @xmath545 where @xmath248 is not necessarily an integer . the basis of our proof is the following stirling approximation formula @xmath546 as an implication , @xmath547 here the second inequality follows from the inequality @xmath548 note that the left hand side is monotonically decreasing with @xmath248 and approaches @xmath549 in the limit @xmath550 . on the other hand , @xmath551 here the last inequality follows from the inequality @xmath552 to confirm this claim , we shall prove the equivalent inequality @xmath553 < 1.\ ] ] the first and second derivatives of @xmath554 read @xmath555 since @xmath556 is positive , @xmath557 is monotonically decreasing , which implies that @xmath558 given that @xmath559 . consequently , @xmath554 is monotonically increasing , which confirms our claim @xmath560 given that @xmath561 . the following two corollaries are easy consequences of lemma [ lem : catalanbound ] , though it is straightforward to prove them directly . [ cor : catalanmono ] @xmath562 for any positive integer @xmath248 . the corollary holds for @xmath563 by direct calculation . when @xmath564 , lemma [ lem : catalanbound ] implies that @xmath565 which confirms the corollary . [ cor : catalansupmul ] @xmath566 for arbitrary positive integers @xmath567 . the corollary holds when @xmath568 or @xmath538 according to corollary [ cor : catalanmono ] , given that @xmath569 . when @xmath570 , lemma [ lem : catalanbound ] implies that @xmath571 recall the definition of the mbius function , @xmath572 [ lem : mobiusbound ] @xmath573 the lower bound is saturated iff @xmath112 is the identity or a product of disjoint transpositions . the upper bound @xmath574 is saturated iff @xmath112 is a cycle of length @xmath575 . the lemma holds when @xmath112 is the identity . otherwise , suppose @xmath112 has disjoint cycle decomposition @xmath576 , where @xmath577 for @xmath578 are nontrivial cycles . then @xmath579 given that @xmath580 for all @xmath581 . the inequality is saturated iff @xmath582 for all @xmath581 , that is , @xmath112 is a product of disjoint transpositions . on the other hand , @xmath583 where the two inequalities follow from corollary [ cor : catalansupmul ] and lemma [ lem : catalanbound ] , respectively . the first inequality is saturated when @xmath538 , but is strict whenever @xmath584 . so the upper bound @xmath574 is saturated iff @xmath112 is a cycle of length @xmath575 . the following theorem is reproduced from @xcite , [ thm : wgboundcm ] when @xmath585 , any @xmath586 satisfies @xmath587 the following lemma is an immediate consequence of theorem [ thm : wgboundcm ] and lemma [ lem : mobiusbound ] . [ lem : wgbound ] when @xmath585 , any @xmath586 satisfies @xmath588 where @xmath589 is defined in theorem [ thm : wgboundcm ] . [ lem : wgsumbound ] suppose @xmath585 ; then @xmath590.\ ] ] according to lemma [ lem : wgbound ] , @xmath591\nonumber\\ & = \frac{7a_k}{8 } + \frac{a_k}{16}\bigl(\frac{4}{d}\bigr)^{k}\left[\prod_{j=0}^{k-1}\bigl(\frac{d}{4}+j\bigr ) + \prod_{j=0}^{k-1}\bigl(\frac{d}{4}-j\bigr ) \right]\nonumber\\ & = \frac{7a_k}{8 } + \frac{a_k}{16}\left[\prod_{j=0}^{k-1}\bigl(1+\frac{4j}{d}\bigr ) + \prod_{j=0}^{k-1}\bigl(1-\frac{4j}{d}\bigr ) \right]\nonumber\\ & \leq\frac{7a_k}{8 } + \frac{a_k}{16}\left[\prod_{j=0}^{k-1}{\mathrm{e}}^{4j / d } + \prod_{j=0}^{k-1}{\mathrm{e}}^{-4j / d}\bigr ) \right ] = \frac{7a_k}{8 } + \frac{a_k}{16}\left[{\mathrm{e}}^{\sum_{j=0}^{k-1}4j / d } + { \mathrm{e}}^{-\sum_{j=0}^{k-1}4j / d}\bigr ) \right]\nonumber\\ & = \frac{7a_k}{8 } + \frac{a_k}{16}\left[{\mathrm{e}}^{2k(k-1)/d } + { \mathrm{e}}^{-2k(k-1)/d}\right]=\frac{a_k}{8}\left[7+\cosh\frac{2k(k-1)}{d}\right ] . \end{aligned}\ ] ] in this appendix , we provide an easy - to - use upper bound for the number of permutations with a given genus ( lemma [ lem : ngpermutationt ] below ) , which plays a crucial role in understanding rnyi entanglement entropies of haar random states as well as states drawn from designs . the basis of our endeavor is the following theorem due to goupil and schaeffer @xcite . [ thm : ngpermutationgs ] the number of permutations in the symmetric group @xmath592 with genus @xmath593 is given by @xmath594 where @xmath595 , @xmath596 , @xmath597 for @xmath598 , and @xmath599 here the summation runes over all partition @xmath600 of @xmath593 , the expression @xmath601 means that @xmath600 has @xmath602 parts equal to @xmath581 , and @xmath603 denotes the number of parts of @xmath600 . in addition , we need two auxiliary lemmas . [ lem : partitions ] @xmath604 for all @xmath605 . by definition , the lemma holds when @xmath606 , or @xmath598 and @xmath607 . now suppose @xmath608 ; then @xmath609\nonumber\\ & \leq \sum_{\substack{\gamma\vdash g,\;\ell(\gamma)=\ell \\ \gamma=1^{c_1}2^{c_2 } g^{c_g}}}\left [ \frac{1}{\prod_{j=1}^gc_j!j^{c_j } } \prod_{j=1}^g \bigl(\frac{1}{2}\bigr)^{c_j}\right]=\sum_{\substack{\gamma\vdash g,\;\ell(\gamma)=\ell \\ \gamma=1^{c_1}2^{c_2 } g^{c_g}}}\left [ \frac{1}{\prod_{j=1}^gc_j!j^{c_j } } 2^{-\sum_{j=1}^g c_j}\right]\nonumber\\ & = \sum_{\substack{\gamma\vdash g,\;\ell(\gamma)=\ell \\ \gamma=1^{c_1}2^{c_2 } g^{c_g } } } \frac{1}{\prod_{j=1}^gc_j!j^{c_j } } 2^{-\ell(\gamma ) } = 2^{-\ell}\sum_{\substack{\gamma\vdash g,\;\ell(\gamma)=\ell \\ \gamma=1^{c_1}2^{c_2 } g^{c_g } } } \frac{1}{\prod_{j=1}^gc_j!j^{c_j } } \leq 2^{-\ell}.\end{aligned}\ ] ] here the last inequality can be derived as follows . note that @xmath610 is the order of the centralizer in @xmath611 of each element in the conjugacy class labeled by the partition @xmath600 . therefore , @xmath612 is the number of elements in this conjugacy class , so that @xmath613 which amounts to the identity @xmath614 as an immediate consequence , @xmath615 [ lem : binomratio ] suppose @xmath616 are nonnegative integers satisfying @xmath617 , @xmath618 , and @xmath619 . then @xmath620 straightforward calculation shows that @xmath621 so without loss of generality , we may assume that @xmath622 . then @xmath623 [ n(n-1)\cdots ( n+j - k+1)]}\nonumber\\ & = \frac{2^kn(n-\frac{1}{2})\cdots ( n-\frac{k}{2}+\frac{1}{2})}{[n(n-1 ) \cdots n - j+1 ] [ n(n-1)\cdots ( n+j - k+1 ) ] } = 2^kf,\end{aligned}\ ] ] where @xmath624 the square of @xmath625 can be bounded from below as follows , @xmath626 therefore @xmath627 , from which the lemma follows . [ lem : ngpermutationt ] @xmath628 recall that @xmath629 $ ] . the values of @xmath630 can be computed explicitly according to theorem [ thm : ngpermutationgs ] , with the result @xmath631 the coefficients @xmath632 necessary for deriving this result are given by @xmath633 as a consequence , @xmath634 therefore , lemma [ lem : ngpermutationt ] holds when @xmath635 . now suppose @xmath636 , so that @xmath637 . according to theorem [ thm : ngpermutationgs ] , we have @xmath638 here the first inequality follows from lemma [ lem : binomratio ] , and the last one from lemma [ lem : partitions ] and the fact that @xmath597 for @xmath639 . the fraction at the end of the above equation is no larger than 1 given that @xmath636 . therefore , @xmath640}{\frac{n}{4}-1}\frac{[\left(\frac{n}{4}\right)^{g_2 + 1}-1]}{\frac{n}{4}-1 } \nonumber\\ & \leq \frac{(n+1)_{2g}}{2^{4 g } } \frac{1}{(\frac{n}{4}-1)^2 } \sum_{g_1+g_2=g } \left(\frac{n}{4}\right)^{g+2}=\frac{(n+1)_{2g}}{2^{4 g } } \frac{(g+1)\left(\frac{n}{4}\right)^{g+2}}{(\frac{n}{4}-1)^2 } \nonumber\\ & = \frac{(g+1)n^{g+2}(n+1)_{2g}}{2^{6g}(n-4)^2 } \leq \frac{(g+1)n^{3g-3 } ( n+1)(n-1)(n-2)(n-3)(n-4)}{2^{6g}(n-4)^2}\nonumber\\ & \leq \frac{(g+1)n^{3g}}{2^{6g}}.\end{aligned}\ ] ] this result confirms the first inequality in lemma [ lem : ngpermutationt ] in the remaining case @xmath641 , which in turn implies the second inequality in the lemma . here we analyze the partially scrambling unitary model proposed in @xcite , which can lead to a large separation between von neumann and rnyi-2 entanglement entropies and tripartite information in the choi state setting . more explicitly , let @xmath642 be a unitary that perfectly scrambles on almost the whole space besides a small subspace . then , on the one hand , @xmath642 still has nearly maximal @xmath87 due to continuity ; while on the other hand , @xmath643 can be gapped from maximum by @xmath644 . however , we find that this model is not likely to provide clear separations between rnyi entropies of order @xmath449 . the generalized partially scrambling unitary is defined as follows . given @xmath0 , define @xmath645 where @xmath646 is @xmath0-scrambling , and @xmath647 controls the size of this @xmath0-scrambling subspace ( labeled by subscript @xmath648 ) . then the choi state of @xmath642 is @xmath649 the question is whether there exists some @xmath5 that can lead to separations between higher rnyi entropies associated with this choi state , say @xmath0 and @xmath272 , @xmath650 . to establish such separations , we need to show a large ( @xmath644 ) gap between rnyi-@xmath272 entropies and the maximum for some small @xmath5 , as well as upper bound the difference between rnyi-@xmath0 entropies and the maximum by continuity . the gap side can work out by directly generalizing the corresponding calculation in @xcite : let @xmath651 . then @xmath652 as long as @xmath466 is a positive constant . however , we find that the continuity bound for unified entropies can only give trivial results on the continuity side : let @xmath10 and @xmath653 be density operators in hilbert space of dimension @xmath39 . denote @xmath654 . for @xmath23 and @xmath655 : @xmath656,\ ] ] where @xmath657 for @xmath658 , and @xmath659 for @xmath660 . @xmath661 denotes the @xmath0 binary entropy . it can be seen that this generalized fannes bound for rnyi grows with the dimension @xmath39 for @xmath23 , which indicates that even a tiny non - scrambling subspace may perturb the rnyi entropies drastically . indeed , some simple scaling analysis can confirm that this bound is trivial even for rnyi-2 . notice that @xmath662 . then it must hold that @xmath663 so that @xmath664 . this gives @xmath665 , which has no overlap with the @xmath666 solution on the gap side when @xmath448 . equivalently , by plugging in @xmath666 we can solve that the desired separation can exist when @xmath667 . summarizing , in order to have a nontrivial bound on rnyi entropies @xmath5 needs to be @xmath668 , which is meaningless . this is hardly surprising : one expects that rnyi entropies are very sensitive , especially in the near - maximum regime , due to the logarithm . in fact , we are able to obtain a large gap on the @xmath272 side basically because of such exponential sensitivity . suppose we consider @xmath21 entropies instead . then the continuity bound is strong since @xmath669 , but it becomes hard to find a gap on the other side . there is a fundamental tradeoff between sensitivity and robustness in these unified entropies . in conclusion , we believe that partially scrambling unitaries are not likely to produce clear separations between generalized entropies in the choi model . to prove lemma [ lem : averootnorm ] , we need to introduce several auxiliary concepts and lemmas . an @xmath670 matrix @xmath429 is a ( standard ) gaussian random matrix if the entries of @xmath429 are i.i.d . standard gaussian random variables ( with mean 0 and variance 1 ) . it is a complex gaussian random matrix if its real part and imaginary part are independent gaussian random matrices . usually this lemma is stated without the intermediate term , as it appears in @xcite . however , the first inequality is essential to achieve our goal . fortunately , this inequality is already implied by the proof in @xcite . note that @xmath672 is the average norm of a vector composed of @xmath673 iid standard gaussian random variables , while @xmath674 is the root mean square norm . this observation implies the second inequality in the lemma , which is nearly tight when @xmath675 are large . it is well known that @xmath680 considered as a unit vector in @xmath313 is distributed uniformly . in addition , the spectrum of @xmath680 is independent of the frobenius norm @xmath681 . therefore , @xmath682^a \operatorname{\mathbb{e}}\left\\|\frac{g}{\\|g\\|_2}\right\\|^{2a}= \operatorname{\mathbb{e}}[{\operatorname{tr}}\{gg^\dag\}]^a \operatorname{\mathbb{e}}\\|\rho_a\\|^a=\frac{2^a\gamma(k+a)}{\gamma(k)}\operatorname{\mathbb{e}}\\|\rho_a\\|^a,\ ] ] from which the lemma follows . here the last equality in the above equation follows from the fact that @xmath683 obeys @xmath684-distribution with @xmath685-degrees of freedom and pdf . @xmath686 which satisfies @xmath687 according to lemmas [ lem : gaussianuniform ] and [ lem : gaussianan ] , in the real case , we have @xmath688 where @xmath689 , and the last inequality follows from the fact that @xmath690 is monotonic increasing with @xmath673 for @xmath691 . this conclusion is intuitive if we observe that @xmath690 is equal to the ratio of the mean length over the root mean square length of a standard gaussian random vector with @xmath673 components . to derive an analytical proof , we can compute the log - derivative of @xmath690 with respect to @xmath673 , note that the definition of @xmath690 can be extended to positive real numbers . straightforward calculations shows that @xmath692\geq \frac{1}{4}\left[\psi^{(0)}\bigl(\frac{m+2}{2}\bigr)-\psi^{(0)}\bigl(\frac{m}{2}\bigr)-\frac{2}{m}\right]=0.\ ] ] here @xmath693 denotes the digamma function ( instead of a ket ) , the inequality follows from the concavity of @xmath693 , and the last equality follows from the identity @xmath694 . 67ifxundefined [ 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 http://stacks.iop.org/1126-6708/2007/i=09/a=120 [ * * , ( ) ] link:\doibase 10.1088/1126 - 6708/2008/10/065 [ * * , ( ) ] , @noop ( ) , @noop * * , ( ) http://projecteuclid.org/euclid.cmp/1103899181 [ * * , ( ) ] link:\doibase 10.1103/physrevd.13.191 [ * * , ( ) ] link:\doibase 10.1146/annurev - conmatphys-031214 - 014726 [ * * , ( ) ] link:\doibase 10.1007/jhep02(2016)004 [ * * , ( ) ] link:\doibase 10.1007/jhep03(2014)067 [ * * , ( ) ] link:\doibase 10.1007/jhep12(2014)046 [ * * , ( ) ] link:\doibase 10.1007/jhep03(2015)051 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.131603 [ * * , ( ) ] link:\doibase 10.1007/jhep08(2016)106 [ * * , ( ) ] @noop ( ) , link:\doibase 10.1007/jhep05(2015)132 [ * * , ( ) ] @noop ( ) , @noop ( ) , link:\doibase 10.1002/andp.201600318 [ ( ) ] @noop ( ) , link:\doibase 10.1002/andp.201600332 [ ( ) ] , link:\doibase 10.1103/physrevlett.71.1291 [ * * , ( ) ] @noop ( ) , @noop ( ) , link:\doibase 10.1103/physrevlett.116.170502 [ * * , ( ) ] @noop ( ) , link:\doibase 10.1103/physrevlett.101.010504 [ * * , ( ) ] link:\doibase 10.1103/physreva.93.012347 [ * * , ( ) ] link:\doibase 10.1063/1.2165794 [ * * , ( ) ] link:\doibase 10.1007/s10955 - 011 - 0231-x [ * * , ( ) ] link:\doibase 10.1103/revmodphys.81.865 [ * * , ( ) ] http://dx.doi.org/10.1038/nature15750 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop ( ) , @noop * * , ( ) _ _ , @noop , ( ) , @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop ( ) , link:\doibase 10.1063/1.523807 [ * * , ( ) ] link:\doibase 10.1155/s107379280320917x [ * * , ( ) ] link:\doibase 10.1063/1.3251304 [ * * , ( ) ] link:\doibase 10.1007/s11005 - 009 - 0365 - 9 [ * * , ( ) ] link:\doibase 10.1142/s2010326316500118 [ * * , ( ) ] link:\doibase 10.1007/s00220 - 006 - 1554 - 3 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.116.170502 [ * * , ( ) ] http://dl.acm.org/citation.cfm?id=2011781.2011790 [ * * , ( ) ] link:\doibase 10.1063/1.523763 [ * * , ( ) ] link:\doibase 10.1016/0003 - 4916(88)90094 - 2 [ * * , ( ) ] link:\doibase 10.1007/s00220 - 006 - 1535 - 6 [ * * , ( ) ] link:\doibase 10.1103/physreva.74.062314 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.72.1148 [ * * , ( ) ] link:\doibase 10.1103/physreve.52.5653 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.1 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physreve.65.046131 [ * * , ( ) ] link:\doibase 10.1214/10-aap722 [ * * , ( ) ] http://dx.doi.org/10.1007/s00220-008-0624-0 [ * * , ( ) ] http://arxiv.org/abs/1510.02619 [ `` , '' ] ( ) , , http://arxiv.org/abs/1510.02769 [ `` , '' ] ( ) , , http://arxiv.org/abs/1510.02767 [ `` , '' ] ( ) , , @noop ( ) , http://arxiv.org/abs/1701.04493 [ `` , '' ] ( ) , , \doibase http://dx.doi.org/10.1006/eujc.1998.0215 [ * * , ( ) ] `` , '' in @noop _ _ , ( , )	scrambling is a process by which the state of a quantum system is effectively randomized . scrambling exhibits different complexities depending on the degree of randomness it produces . for example , the complete randomization of a pure quantum state ( haar scrambling ) implies the inability to retrieve information of the initial state by measuring only parts of the system ( page / information scrambling ) , but the converse is not necessarily the case . here , we formally relate scrambling complexities to the degree of randomness , by studying the behaviors of generalized entanglement entropies in particular rnyi entropies and their relationship to designs , ensembles of states or unitaries that match the completely random states or unitaries ( drawn from the haar measure ) up to certain moments . the main result is that the rnyi-@xmath0 entanglement entropies , averaged over @xmath0-designs , are almost maximal . the result generalizes page s theorem for the von neumann entropies of small subsystems of random states . for designs of low orders , the average rnyi entanglement entropies can be non - maximal : we exhibit a projective 2-design such that all higher order rnyi entanglement entropies are bounded away from the maximum . however , we show that the rnyi entanglement entropies of all orders are almost maximal for state or unitary designs of order logarithmic in the dimension of the system . that is , such designs are indistinguishable from haar - random by the entanglement spectrum . our results establish a formal correspondence between generalized entropies and designs of the same order .
measuring the surface composition of kuiper belt objects ( kbos ) may provide clues about the composition and environment of the primordial solar nebula as well as the important evolutionary processes occurring in the outer solar system over the last 4.5 gyr . unfortunately , there are only a handful of kbos that are known to exhibit ice absorption bands in their spectra . h@xmath2o - ice bands are seen in spectra of 1996 to66 @xcite , varuna @xcite , quaoar @xcite , and orcus @xcite . ch@xmath0-ice bands are seen in spectra of pluto @xcite , neptune s satellite triton , which may be a captured kbo @xcite , eris @xcite , and 2005 fy9 @xcite . there are perhaps a dozen objects that exhibit spectra with no ice absorption bands @xcite the recent discovery of extraordinarily bright and large kbos opens up a new opportunity for the physical and chemical studies of kbo surfaces with high signal precision optical reflectance spectroscopy . bright objects make it possible to either carry out in - depth physical modeling of ice and mineralogical absorption bands or set stringent upper limits on the presence of such bands . here we present high signal precision optical reflectance spectra of kbos 2005 fy9 and 2003 el61 . both objects are candidates for membership in the newly defined class `` dwarf planet . '' the diameter of 2005 fy9 is estimated at 1600 km @xcite while 2003 el61 appears to be a highly elongated ellipsoid with axes of 1950 @xmath3 2500 km @xcite . for comparison , pluto has a diameter of 2350 km . 2005 fy9 and 2003 el61 are among `` scattered - near '' objects ( i.e. non - resonant , non - planet - crossing objects with tisserand parameters less than 3 , relative to neptune ) in the deep ecliptic survey classification system @xcite . 2005 fy9 has a perihelion distance , q , of 38.6 au , an aphelion distance , q , of 52 au , a semi - major axis , a , of 45.3 au , an inclination angle to the ecliptic , i , of 29@xmath4 , and an eccentricity of 0.15 . 2003 el61 has q @xmath5 35.1 au , q @xmath5 51 au , a @xmath5 43.1 au , i @xmath5 28@xmath4 , and e @xmath5 0.18 for comparison , pluto has q @xmath5 29.7 au , q @xmath5 49 au , a @xmath5 39.5 au , i @xmath5 17@xmath4 , and e @xmath5 0.25 . at the time of observation , 2005 fy9 and 2003 el61 were both near aphelion . we obtained spectra of 2005 fy9 and 2003 el61 on 2006 march 5 ut with the 6.5-meter mmt telescope on mt . hopkins , az , the red channel spectrograph , and a 1200 @xmath3 800 ccd . a 150 g mm@xmath6 grating and a 1 arc sec slit width provided wavelength coverage of 5000 to 9500 in first order , a dispersion of 6.38 pixel@xmath6 , and a fwhm resolution of 20.0 . a lp495 blocking filter eliminated contamination from the second order . there were high , thin cirrus clouds through most of the night and the seeing was @xmath7 0.8 arc sec . the kbos were placed at the center of the slit and tracked at kbo rates . we used the image reduction and analysis facility ( iraf ) and standard procedures @xcite to calibrate and extract one - dimensional spectra from the two - dimensional spectral images . specifically , the electronic bias of each image was removed by subtracting its overscan as well as a bias picture . pixel to pixel sensitivity variations were removed from each image by dividing by a normalized twilight flatfield image . extraction of one - dimensional spectra from the images was done with the apall task in iraf . henear spectra were used to correct for flexure and obtain an accurate wavelength calibration . our wavelengths are accurate to @xmath7 one - tenth of a pixel or @xmath7 0.7 . we removed telluric bands and fraunhofer lines from the kbo spectra by observing the solar analog hd 112257 @xcite at airmasses very close to 2005 fy9 and 2003 el61 , and then then dividing the kbo spectra by the normalized solar analog spectra . typically , the airmass difference between the kbos and the solar analog was @xmath8 0.05 . the 2005 fy9 and 2003 el61 spectra in the analysis below are an average of four 10-minute and five 10-minute exposures , respectively . we plot our reflectance spectrum of 2005 fy9 in figure 1 ( bottom ) . the tick marks correspond to previously reported absorption maxima of ch@xmath0-ice ( table 1 ) . the middle and top spectra in figure 1 are of 2005 fy9 @xcite and pluto @xcite . all spectra are normalized to 1 at 6500 . for the purpose of comparison , the @xcite 2005 fy9 and pluto spectra are offset by 0.4 and 0.8 , respectively . clearly , 2005 fy9 exhibits much deeper ch@xmath0-ice bands than pluto . our spectrum differs from the @xcite spectrum in two ways . first , our spectrum exhibits additional subtle absorption bands at 5400 , 5800 , and 6000 . these bands have not been seen before in astronomical or laboratory spectra of ch@xmath0-ice . because the bands are close to the wavelengths of gas phase ch@xmath0 bands and because there are strong ch@xmath0-ice bands at longer wavelengths in the spectrum of 2005 fy9 , we think the new bands are due to ch@xmath0-ice . in addition , it appears the band at 6200 is ch@xmath0-ice as well . it is in the @xcite spectrum , but they do not identify it as ch@xmath0-ice . second , we find a continuum slope between 5500 and 6500 of 10%/1000 while @xcite find a slope of 13%/1000 , with both spectra normalized to 1 at 6500 . we used hd 112257 , a g2v star , for our solar analog star whereas @xcite used bs 4486 , a g0v star , for their solar analog star . an examination of the v@xmath9r colors for g0v and g2v stars suggests the slope difference is due to the use of different solar analog stars . in order to place constraints on the ch@xmath0 grain size as well as the presence of any additional ice components on the surface of 2005 fy9 , we calculated model ch@xmath0-ice spectra . our models use laboratory optical constants for pure ch@xmath0-ice at 30 k @xcite and arbitrary optical constants that absorb more at blue than red wavelengths , and thereby reproduce the observed reddish slope for the 2005 fy9 continuum . the ice and reddening agent were mixed in the model at the molecular level ( i.e. by means of a weighted average of the ice and reddening agent optical constants ) . we used hapke theory @xcite to transform the optical constants into reflectance spectra . we used hapke parameters h @xmath100.1 , b@xmath11 @xmath5 0.8 , @xmath12 @xmath5 30@xmath4 , and p(g ) @xmath5 a two component henyey - greenstein function with 80% in the forward scattering lobe and 20 % in the back scattering lobe and both lobes having asymmetry parameter 0.63 . these values are comparable to numbers used previously to model the surface of pluto @xcite . figure 2 compares the spectrum of 2005 fy9 ( black line ) and a hapke model with a grain size of 0.55 cm ( red line ) . although the model fits the core of the 8897 band , it has too little absorption at all the other bands . next , we fit the core of the weaker 7296 band with a 1.8 cm grain size . then , we fit the core of the even weaker 8000 complex with a 2.5 cm grain size . these models illustrate a trend , weaker bands ( which see deeper into the surface ) aways seem to need bigger grain sizes , or equivalently , greater optical path lengths in ch@xmath0-ice . from the above discussion , it is clear a single grain size will not fit the observations . in figure 3 , we present a hapke model that uses two grain sizes , 6 cm ( 97 % by volume ) and 0.1 cm ( 0.03 % by volume ) , in an intimate mixture . the inability of the two - grain size model to fit the ch@xmath0-ice band at wavelengths less than 7000 in figure 3 is due to the lack of laboratory optical constants . we note that our two - grain size model is not a unique solution . although this model fits the data , other models ( such as courser grains underneath a coating of finer grains ) would probably fit the data just as well . in addition , it is important to realize that grain sizes larger than 1 cm are probably a measure of the spacing of fractures or voids in the solid surface rather than indicating a surface covered by golf - ball sized particles . although we can fit the shape and depth of the absorptions , there is a small but significant difference between our models and the telescope data . specifically , the maxima of the ch@xmath0 absorption bands in the spectrum of 2005 fy9 reside blueward of the maxima in the pure ch@xmath0 hapke models . in figure 4 , we present a small portion of figure 3 centered on the 7296 band to illustrate the shift . such a shift is important because laboratory experiments show that the 8897 , 8968 , and 9019 bands in spectra of pure ch@xmath0-ice are shifted blueward by 27 in a ch@xmath0/n@xmath2 @xmath5 0.8 % mixture @xcite . another experiment finds the 8897 band shifts @xmath7 17 in a ch@xmath0/n@xmath2 @xmath5 20 % ice mixture @xcite . a cross correlation experiment provides a way to quantify the apparent shift in figure 3 . specifically , we wrote a fortran program to shift the model spectrum from @xmath9 25 to @xmath1325 in 1 steps . for each shift , the program finds the difference between the data and model , i.e. it calculates @xmath14 where @xmath15 and @xmath16 represent the ordinate values of the data and the model spectra at wavelength i. then , the program sums the squares of the difference over all n wavelength points between 7000 and 9300 . in other words , we calculate @xmath17 in figure 5 , we present a plot of the sums of the squares of the differences , i.e. @xmath18 , as function of shift . we find a well defined minimum at 3.25 . what is the uncertainty in the 3.25 measurement ? we note that henear spectra enable us to calibrate the wavelengths in the 2005 fy9 spectrum to an uncertainty of @xmath7 1/10 of a pixel or @xmath7 0.7 . specifically , we find the average difference between the centroids of airglow lines in our data and the corresponding laboratory values is 0.7 . a larger source of uncertainty comes from the noise in the spectra and the broadness of the cross correlation minimum in figure 5 . propagation of the noise in the data and model spectra for the 3 shift through equation ( 1 ) gives an ordinate error bar for the minimum point ( 3.25 , 2.218 ) in figure 5 of 0.042 . as a result , the uncertainty in our shift measurement is defined by @xmath18 @xmath19 2.218 @xmath13 0.042 @xmath5 2.260 ( i.e. the section of the curve below the dashed line in figure 5 ) , corresponding to shifts between 1 and 5.5 . in short , we find the ch@xmath0-ice bands in the spectrum of 2005 fy9 are blueshifted relative to the model hapke spectrum by 3.25 @xmath1 2.25 . @xcite gave blueshifts of 2 - 6 for ch@xmath0 bands in their table 1 , but they questioned the reality of the shifts because they are so small . our independent measurement of shifts with similar magnitude bolsters the case for the reality of the small shifts . in fact , when we take the @xcite data and do a cross correlation experiment between our spectrum and their spectrum , we find no measurable shift between the spectra ; but when we perform a cross correlation experiment between their spectrum and the hapke model , we find their ch@xmath0-ice bands are blue shifted 2 relative to the model . our spectrum and the @xcite spectrum are consistent . we plot our reflectance spectrum of 2003 el61 in figure 6 . the continuum has a slope of essentially zero , consistent with b@xmath9v @xmath5 0.63 @xmath1 0.03 and v@xmath9r @xmath5 0.34 @xmath1 0.02 @xcite . evidently , tholins are not present on the surface of 2003 el61 at an abundance seen on 2005 fy9 . ( recall that both objects were observed on the same night and compared against the same solar analog spectrum . ) we find no evidence of ch@xmath0-ice bands , but we can set an upper limit on the thickness of a global glaze of ch@xmath0-ice by applying beer s law , @xmath20 to the continuum near the 8897 band . in the above equation , t is the thickness of the glaze , a@xmath21 are the lambert absorption coefficients for ch@xmath0-ice @xcite , the factor of two comes about because light passes through the glaze once on the way in and once on the way back out , ( i / i@xmath11)@xmath21 is the ratio of light leaving the glaze to the light incident on the glaze . for ( i / i@xmath11)@xmath22 @xmath5 0.97 , we find t @xmath19 0.3 mm . h@xmath2o - ice bands are seen in near - infrared spectra of 2003 el61 @xcite . therefore , it is reasonable to look for evidence of solid - state photolytic or radiolytic chemistry @xcite by seeing if o@xmath2-ice is present on the surface of 2003 el61 . in figure 6 , we superimpose the 5773 and 6275 o@xmath2-ice bands in spectra of ganymede @xcite on top of our 2003 el61 spectrum . there is a tantalizing dip at 5773 in the spectrum of 2003 el61 , but no dip at the weaker 6275 band . the continuum signal precision at the position of these bands is about 150 and so if the 5773 band is real , it would take a continuum signal precision of about 300 to detect the 6275 band . we estimate an upper limit to the thickness of a global glaze of o@xmath2-ice by assuming the possible feature at 5773 is real and then applying beer s law as we did for ch@xmath0-ice . for ( i / i@xmath11)@xmath23 @xmath5 0.99 and a@xmath24 @xmath5 5.5 cm@xmath6 , we find t @xmath19 0.01 mm . our value for a@xmath24 comes from an integrated absorption coefficient of 1.1 @xmath3 10@xmath26 @xmath27 and a band fwhm of @xmath7 200 cm@xmath6 @xcite . our spectrum of 2005 fy9 exhibits strong ch@xmath0-ice bands in agreement with the spectrum of @xcite . from a comparison of our spectrum and a hapke model , we find the 2005 fy9 ch@xmath0-ice bands are shifted 3.25 @xmath1 2.25 blueward relative to the positions of pure ch@xmath0-ice bands . the shift could be due to the presence of another ice - component , possibly n@xmath2-ice , co - ice , or ar . future higher resolution spectra of the individual ch@xmath0 bands should determine if the weaker bands which penetrate deeper into the surface exhibit different shifts than the stronger bands which do not penetrate as deep into the surface . such observations could provide a technique to measure the ch@xmath0 concentration relative to the other ice component as a function of depth below the surface of 2005 fy9 . in addition , it is highly desirable to obtain a measure of the shift as a function of rotational longitude on 2005 fy9 for a particular ch@xmath0-ice band . such a technique could provide a measure of inhomogeneity on the surface of 2005 fy9 . @xcite made such measurements for the 8897 band on pluto and found that the blueshift varied from 0 to 10 over the surface . on pluto , it appears n@xmath2 is sufficiently mobile to form textures that allow incoming photons to travel several centimeters through the very transparent , polycrystalline n@xmath2-ice before the photons are scattered off inclusions or grain boundaries . the long pathlengths make it possible for a small amount of ch@xmath0 relative to n@xmath2 to give deep ch@xmath0 absorption bands . the mobility of co is only a little less than n@xmath2 , and the mobility of ar is only a little less than co. therefore , it is possible that co - ice or ar - ice could provide the host matrix for long optical pathlengths . it is also possible the host matrix is a combination of n@xmath2 , co , and ar . a suite of laboratory experiments that measure the shifts of ch@xmath0 absorption maxima for different concentrations of ch@xmath0 relative to n@xmath2 , co , and ar would be highly valuable for the future interpretation of observational data . our spectrum of 2003 el61 is remarkably featureless . there is a hint of a band at 5773 possibly due to o@xmath2-ice . future work will attempt to double the signal precision of the continuum to at least 300 , which is the level that appears to be required for any possible detection of the weaker o@xmath2-ice band at 6275 . the detection of o@xmath2-ice in significant quantities on kbos could be of considerable interest in the distant future as a source of spacecraft fuel . sct , wr , and km gratefully acknowledge support from nasa planetary astronomy grant nng06g138 g to northern arizona university and the university of oklahoma . wmg gratefully acknowledges support from nasa planetary geology and geophysics grant nng04 g172 g to lowell observatory . gjc gratefully acknowledges support from fordham university . we thank steward observatory for the allocation of telescope time on the mmt . we thank dr . j. licandro for providing us with his spectrum of 2005 fy9 . we thank an anonymous referee for a careful review of our manuscript . barkume , k. m. , brown , m. e. , and schaller , e. l. 2006 , , 640 , l87 . brown , m. e. , barkume , k. m. , blake , g. a. , schaller , e. l. , rabinowitz , d. l. , roe , h. g. , & trujillo , c. a. 2006 , , submitted . brown , m. e. , trujillo , c. a. , & rabinowitz , d. l. 2005 , , 635 , l97 brown , r. h. , cruikshank , d. p. , & pendleton , y. 1999 , , 519 , l101 calvin , w. m. , & spencer , j. r. 1997 , icarus , 130 , 505 . cruikshank , d. p. , pilcher , c. b. , & morrison , d. 1976 , science , 194 , 835 cruikshank , d. p. , roush , t. l. , owen , t. c. , geballe , t. r. , de bergh , c. , schmitt , b. , brown , r. h. and bartholomew , m. j. 1993 , science , 261 , 742 delitsky , m. l. , & lane , a. l. 1997 , j. geophys . res . , 102 , 16385 doressoundiram , a. , tozzi , g. p. , barucci , m. a. , boehnhardt , h. , fornasier , s. & romon j. 2003 , , 125 , 2721 elliot , j. l. , kern , s. d. , clancy , k. b. , gulbis , a. a. s. , millis , r. l. , buie , m. w. , wasserman , l. h. , chiang , e. i. , jordan , a. b. , trilling , d. e. , and meech , k. j. 2005 , , 129 , 1117 fink , u. , smith , b. a. , johnson , j. r. , reitsema , h. j. , benner , d. c. , & westphal , j. a. 1980 , icarus , 44 , 62 fornasier , s. , dotto , e. , barucci , m. a. , & barbieri , c. 2004a , a&a , 422 , 43 fornasier , s. , doressoundiram , a. , tozzi , g. p. , barucci , m. a. , boehnhardt , h. , de bergh , c. , delsanti , a. , davies , j. , & dotto , e. 2004b , a&a , 421 , 353 grundy , w. m. , & buie , m. w. 2001 , icarus , 153 , 248 grundy , w. m. , & fink , u. 1993 , pluto & charon conference , flagstaff , az , july 6 - 9 , 1993 grundy , w. m. & fink , u. 1996 , icarus , 124 , 329 . grundy , w. m. , schmitt , b. , and quirico , e. 2002 , icarus , 155 , 486 . hapke , b. 1981 , j. geophys . res . , 86 , 4571 . hapke , b. 1993 , combined theory of reflectance and emittance spectroscopy ( new york : cambridge univ . press ) hardorp , j. 1982 , a&a , 105 , 120 jewitt , d. c. , & luu , j. 2004 , nature , 432 , 731 johnson , r. e. 1990 , energetic charged - particle interactions with atmospheres and surfaces . ( new york : springer - verlag ) johnson , r. e. , & quickenden , t. i. 1997 , j. geophys . res . , 102 , 10985 landau , a. , allin , e. j. , & welsh , h. l. 1962 , spectrochimica acta , 18 , 1 . licandro , j. , oliva , e. , & di martino , m. 2001 , a&a , 373 , l29 licandro , j. , pinilla - alonso , n. , pedani , m. , oliva , e. , tozzi , g. p. , and grundy , w. m. 2006 , a&a , 445 , l35 . luu , j. & jewitt , d. 1996 , , 112 , 2310 massey , p. , valdes , f. , & barnes , j. 1992 , in a user s guide to reducing slit spectra with iraf quirico , e. , & schmitt , b. 1997 , icarus , 127 , 354 rabinowitz , d. l. , barkume , k. , brown , m. e. , roe , h. , schwartz , m. , tourtellotte , s. , and trujillo , c. 2006 , , 639 , 1238 . -ice bands , see table 1 . the middle spectrum is also of 2005 fy9 @xcite . the top spectrum is of pluto @xcite . all spectra are normalized to 1 at 6500 . the @xcite and pluto spectra are offset by 0.4 and 0.8 , respectively . clearly , 2005 fy9 has deeper ch@xmath0-ice bands than pluto . previously unreported ch@xmath0-ice bands appear near 5400 , 5800 , and 6200 in our spectrum of 2005 fy9 . ] -ice grain size of 0.55 cm ( red line ) . we assume an albedo at 6500 of 0.8 . the model fits the core of the 8897 band , but has too little absorption at all other bands . there are no laboratory absorption coefficients to fit ch@xmath0 bands with @xmath28 @xmath19 7000 . ] -ice grain sizes , 6 cm and 0.1 cm ( red line ) . we assume an albedo at 6500 of 0.8 . grains sizes larger than a cm probably provide a measure of the spacing between fractures or voids in the surface rather than indicating a surface covered with golf - ball sized particles . there are no laboratory absorption coefficients to fit ch@xmath0 bands with @xmath28 @xmath19 7000 ] -ice band in 2005 fy9 ( black line ) is blueshifted relative to the minimum of the pure ch@xmath0-ice hapke model ( red line ) . the other bands in 2005 fy9 show a similar shift . the shift may be due to the presence of another ice component on the surface of 2005 fy9 . ] -ice bands at 5773 and 6275 ( red line ) . there is a tantalizing dip in the spectrum of 2003 el61 at the position of the 5773 band ; however , a much higher continuum signal precision is necessary to test for the presence of the weaker 6275 band . ] lcc 3@xmath29 @xmath13 4@xmath30 & 7296 & 13706 + 3@xmath31 @xmath13 3@xmath30 & 7862 & 12719 + 3@xmath29 @xmath13 3@xmath32 & 7993 & 12511 + 4@xmath31 & 8415 & 11884 + @xmath29 @xmath13 3@xmath31 & 8442 & 11846 + 3@xmath31 @xmath13 2@xmath30 & 8691 & 11506 + 2@xmath29 @xmath13 @xmath31 @xmath13 2@xmath30 & 8897 & 11240 + 3@xmath29 @xmath13 2@xmath30 & 8968 & 11151 + 2@xmath31 @xmath13 4@xmath30 & 9019 & 11088 +	we present high signal precision optical reflectance spectra of the large kuiper belt objects 2005 fy9 and 2003 el61 . the spectrum of 2005 fy9 exhibits strong ch@xmath0-ice bands . a comparison between the spectrum and a hapke model indicates the ch@xmath0 bands are shifted 3.25 @xmath1 2.25 relative to pure ch@xmath0-ice , suggesting the presence of another ice component on the surface of 2005 fy9 , possibly n@xmath2-ice , co - ice , or ar . the spectrum of 2003 el61 is remarkably featureless . there is a hint of an o@xmath2-ice band at 5773 ; however , this feature needs to be confirmed by future spectroscopic observations of 2003 el61 with a higher continuum signal precision , sufficient to detect a second weaker o@xmath2-ice band at 6275 .
the interaction of acoustic or electromagnetic waves with structured , periodic materials is often complicated by the fact that the scattering geometry involves domains where multiple media meet at a single point . examples include the design of diffraction gratings , the development of high efficiency solar cells , and non - destructive optical inspection in semiconductor manufacturing ( metrology ) @xcite . the geometry of a typical scattering problem is shown in fig . [ fig - geom1 ] . , that for the trapezoidal - shaped scatterers is @xmath0 and that of the two layers beneath are @xmath1 and @xmath2 , respectively . we assume that the lowest interface ( here between the @xmath1 and @xmath2 layers ) is located at @xmath3 and that the maximum height of the scatterers is at @xmath4 . we also assume that the unit cell is centered at @xmath5 . the bottom layer is assumed to be infinite in extent.,width=384 ] for the sake of concreteness , we will assume throughout this paper that the governing equations are the maxwell equations in two dimensions ( here , the @xmath6 plane ) . we also assume the incident wave is in tm - polarization @xcite and that each of the constituent materials is locally isotropic with constant permittivity @xmath7 and permeability @xmath8 . in this case , the maxwell equations are well - known to take the simpler form @xmath9 with @xmath10 here , @xmath11 , where we have assumed a time - dependence of @xmath12 with @xmath13 the frequency of interest . using the language of scattering theory , we let @xmath14 where @xmath15 is a known incoming field , @xmath16 and @xmath17 is the unknown scattered field . at material interfaces , @xmath18 = 0 \quad & \rightarrow\quad \left[u \right ] = -\left[u^{{\mbox{\tiny in}}}\right ] \\ \label{dielecbc2 } \left[\frac{1}{\mu } \frac{\partial e}{\partial \nu } \right ] = 0 \quad & \rightarrow\quad \left[\frac{1}{\mu } \frac{\partial u}{\partial \nu } \right ] = -\left[\frac{1}{\mu } \frac{\partial u^{{\mbox{\tiny in}}}}{\partial \nu}\right ] \ , , \end{aligned}\ ] ] where @xmath19 denotes the normal direction and @xmath20 $ ] denotes the jump in the quantity @xmath21 across an interface . for simplicity , we will assume @xmath22 and @xmath7 is distinct in each domain . the essential difficulties that we wish to address are manifested in that setting , so we ignore other variants of the scattering problem without loss of generality . scattering problems of the type illustrated in fig . [ fig - geom1 ] are often called _ quasi - periodic _ since the obstacles are arrayed periodically , but the incoming , scattered and total field experience a phase change in traversing the unit cell : @xmath23 where @xmath24 . ( in this convention , normal incidence corresponds to @xmath25 . ) in the @xmath26-direction , to obtain a well - posed problem , the scattered field @xmath27 must satisfy a somewhat involved radiation condition @xcite - namely that it takes the form of rayleigh bloch expansions @xmath28 assuming , as in fig . [ fig - geom1 ] , that the lowest interface lies at @xmath3 and that @xmath29 is the maximum extent of the scatterers . in this formula , @xmath30 , in order to satisfy the quasi - periodicity condition . letting @xmath31 enforces that the expansion satisfy the homogeneous helmholtz equation in the upper half - space , while letting @xmath32 enforces that the expansion satisfy the homogeneous helmholtz equation in the lower half - space with wavenumber @xmath33 ( @xmath2 in fig . [ fig - geom1 ] ) . above the scatterers in the unit cell ( @xmath34 ) , note that if @xmath35 , then @xmath36 is real and the waves in the rayleigh - bloch expansion ( [ e : rbu ] ) are _ propagating _ modes . if @xmath37 , then @xmath36 is imaginary and the corresponding modes are called _ evanescent_. they do not contribute to the far field . the complex coefficients @xmath38 for propagating modes in the rayleigh - bloch expansion are known as the _ bragg diffraction amplitudes at the grating orders_. for each fixed @xmath39 and @xmath40 , there is a discrete set of frequencies @xmath41 for which some @xmath36 may vanish , at which point the rayleigh - bloch mode is constant in the @xmath26-direction . such modes are called _ wood s anomalies_. ( there is also a discrete set of frequencies where the solution is nonunique , due to guided modes which propagate along the grating . the latter are , in a certain sense , nonphysical and we refer the interested reader to @xcite for further discussion . ) in this paper , we present an integral equation method and a corresponding fast direct solver for scattering problems of the type discussed above . we make use of the _ quasi - periodic green s function _ , which requires only a discretization of the dielectric interfaces within the unit cell . in a recent paper , gillman and barnett @xcite address the same problem using a slightly different formulation with a different approach to imposing quasi - periodicity . we will discuss the relative advantages of the two approaches in section [ sec : conclusions ] . a classical approach to the calculation of quasi - periodic scattering is based on using the green s function that satisfies the desired conditions ( [ qpcond ] ) , ( [ e : rbu ] ) , and ( [ e : rbd ] ) @xcite . this is accomplished by constructing a one - dimensional array of suitably phased " copies of the free - space green s function for the helmholtz equation with wavenumber @xmath42 . more precisely , the quasi - periodic green s function is defined by @xmath43 where @xmath44 is the outgoing hankel function of order zero . it is clear that the sum formally satisfies the condition ( [ qpcond ] ) . the rayleigh - bloch conditions ( [ e : rbu ] ) , ( [ e : rbd ] ) follow from fourier analysis and the fact that @xmath44 itself satisfies the sommerfeld radiation condition . unfortunately , the series in ( [ e : gqp ] ) is only conditionally convergent for real @xmath42 . to obtain a physically meaningful limit , one adds a small amount of dissipation ( @xmath45 ) and considers @xmath46 . ( see @xcite for a more detailed discussion ) . we define the `` near field '' of the quasi - periodic green s function by @xmath47 } e^{im\alpha d } h_0^{(1)}(k\|{{\mathbf x}}-(md,0)\| ) \label{e : gqpn}\ ] ] and the `` smooth '' part of the quasi - periodic green s function by @xmath48 } } e^{im\alpha d } h_0^{(1)}(k\|{{\mathbf x}}-(md,0)\| ) \ , . \label{e : gqpr}\ ] ] the latter is a smooth solution to the helmholtz equation within the unit cell centered at the origin ( see fig . [ fig - geom1 ] ) and can be expanded in a bessel series @xmath49 in the low frequency regime , where the unit cell is on the order of a few wavelengths or smaller , the bessel series converges rapidly so long as the @xmath26-component of the target point @xmath50 is less than @xmath40 . for larger values of @xmath26 it is more convenient to switch representations and use the rayleigh - bloch expansion ( [ e : rbu ] ) directly . an analytic formula for the coefficients @xmath51 of the bessel expansion ( [ jexp ] ) can be obtained from the graf addition theorem ( * ? ? ? 9.1.79 ) : @xmath52 } } e^{im\alpha d } h_n(k\|md\| ) ( -1)^{n \cdot \rm{signum}(m)}\ ] ] these coefficients are known as _ lattice sums _ and depend only on the parameters @xmath53 . most numerical schemes for the rapid evaluation of the quasi - periodic green s function are based on the evaluation of @xmath54 } e^{im\alpha d } h_0^{(1)}(k\|{{\mathbf x}}-(md,0)\| ) + \sum_{n=-\infty}^{\infty } s_n j_n(k\|{{\mathbf x}}\|),\ ] ] combining ( [ e : gqpn ] ) and ( [ jexp ] ) . there is a substantial literature on efficient methods for computing the lattice sums themselves ( see , for example , @xcite ) . in this paper we use a scheme based on asymptotic analysis and the euler - maclaurin formula @xcite . since there are a number of effective schemes for this step , we omit further discussion except to note that 1 . the quasi - periodic green s function fails to exist at wood s anomalies 2 . if the scattering structure in the unit cell has a high aspect ratio @xmath55 , then the lattice sum approach is inconvenient because more images need to be added to @xmath56 in order to ensure convergence of the bessel expansion for @xmath57 . we refer the reader to @xcite for a method capable of handling both these difficulties . here , we assume that @xmath58 is well - defined and that the aspect ratio @xmath59 is less than or equal to 1 . in the absence of triple - points , a number of groups have developed high - order accurate integral equation methods for scattering from periodic structures ( see , for example , @xcite ) . for this , suppose that we have a single scatterer @xmath60 in the unit cell , with helmholtz parameter @xmath0 and boundary @xmath61 . in the context of fig . [ fig - geom1 ] , this would correspond to an absence of the layered substrate ( that is , @xmath62 ) , with @xmath60 an isolated trapezoidal - shaped scatterer . one can then use the representation @xmath63({{\mathbf x } } ) + { { { \cal d}}}^{k_1}[\gamma,\mu]({{\mathbf x } } ) & \mbox{in } \omega\\ { { { \cal s}}}^{k_0}_{{\mbox{\tiny qp}}}[\gamma,\sigma]({{\mathbf x } } ) + { { { \cal d}}}^{k_0}_{{\mbox{\tiny qp}}}[\gamma,\mu]({{\mathbf x } } ) & \mbox{in } u\setminus\overline{\omega } \end{array}\right . \label{e : qprep}\ ] ] where @xmath64 and @xmath65 denote the usual single and double layer operators @xcite @xmath66({{\mathbf x } } ) & = & \int_\gamma g^{k}({{\mathbf x}}-{{\mathbf y } } ) \sigma({{\mathbf y } } ) ds_{{\mathbf y}}\label{e : s } \\ { { { \cal d}}}^{k}[\gamma,\mu]({{\mathbf x } } ) & = & \int_\gamma \frac{\partial g^{k}}{\partial \nu_{{\mathbf y}}}({{\mathbf x}}-{{\mathbf y } } ) \mu({{\mathbf y } } ) ds_{{\mathbf y}}\ , . \label{e : d}\end{aligned}\ ] ] with @xmath67 . the quasi - periodic layer potentials @xmath68 and @xmath69 are simply defined by replacing the free - space green s function @xmath70 with @xmath71 . here @xmath72 indicates that we are integrating in arclength on @xmath61 , and @xmath73 denotes the outward normal at @xmath74 . we will also need the normal derivatives of @xmath75 and @xmath76 at a point @xmath77 , defined by @xmath78 the periodic versions @xmath79 and @xmath80 are defined in the same manner . note that by construction , the governing helmholtz equation is satisfied in each domain . note also that we have only chosen to use the quasi - periodic layer potentials in the exterior domain @xmath81 . in the context of fig . [ fig - geom1 ] , we will use the quasi - periodic layer potentials for the @xmath82 , @xmath1 and @xmath2 domain and the standard layer potentials for the @xmath0 domain . @xmath83 is weakly singular as @xmath84 , and the integral is well - defined . for @xmath85 and @xmath86 , the limiting value depends on the side of @xmath61 from which @xmath50 approaches the curve . for @xmath87 , we assume both are defined in the principal value sense . the operator @xmath88 is _ hypersingular _ and unbounded as a map from the space of smooth functions on @xmath61 to itself . it should be interpreted in the hadamard finite part sense . substituting the representation ( [ e : qprep ] ) into the interface conditions ( [ dielecbc1 ] ) , ( [ dielecbc2 ] ) and taking the appropriate limits yields the system of integral equations [ grpb ] @xmath89 + ( { { \cal d}_{{\mbox{\tiny qp}}}}^{k_0}(\gamma_,\mu ) - { { \cal d}}^{k_1}(\gamma,\mu))[{{\mathbf x } } ] = - \left[u^{{\mbox{\tiny in}}}({{\mathbf x } } ) \right ] \label{ib1}\\ -\sigma({{\mathbf x } } ) & + ( { { \cal n}_{{\mbox{\tiny qp}}}}^{k_{0}}(\gamma,\sigma ) - { { \cal n}}^{k_{1}}(\gamma,\sigma))[{{\mathbf x } } ] + ( { { \cal t}_{{\mbox{\tiny qp}}}}^{k_{0}}(\gamma,\mu ) - { { \cal t}}^{k_{1}}(\gamma,\mu))[{{\mathbf x } } ] = - \left [ \frac{\partial u^{{\mbox{\tiny in}}}}{\partial \nu}({{\mathbf x } } ) \right ] \label{ib2}\end{aligned}\ ] ] for the unknowns @xmath90 $ ] . a critical feature of the system ( [ ib1 ] ) , ( [ ib2 ] ) is that , while @xmath91 itself is hypersingular , only the _ difference _ of hypersingular kernels appears in the equations . all the operators appearing above are compact on smooth domains and we have a system of fredholm equations of the second kind , for which the formal theory is classical @xcite and the solution is unique . the cancellation of hypersingular terms in this manner was introduced in electromagnetics by mller @xcite , and in the scalar case by kress , rokhlin , haider , shipman and venakides @xcite . for smooth domains , the issue of quadrature has been satisfactorily resolved , so that high order accuracy is straightforward to achieve @xcite . the generalized gaussian quadrature method of @xcite , for example , permits the use of composite quadrature rules that take into account the singularity of the green s function and can be stored in tables that do not depend on the curve geometry . assuming the boundary component @xmath61 is subdivided into @xmath92 curved panels with @xmath42 points on each panel , these rules achieve @xmath42-th order accuracy . more precisely , each integral operator @xmath93 is replaced by a sum of the form @xmath94 where @xmath95 is the @xmath96-th gauss - legendre node on panel @xmath97 , @xmath98 is the @xmath99-th gauss - legendre node on panel @xmath100 , @xmath101 is a quadrature weight and @xmath102 is a quadrature kernel " . for nonadjacent panels , @xmath102 is simply the original kernel @xmath103 . for the interaction of a panel with itself or its two nearest neighbors , the quadrature kernel is produced by a somewhat involved interpolation scheme according to the generalized gaussian quadrature formalism @xcite . from a linear algebra perspective , generalized gaussian quadrature can be viewed as producing a block tridiagonal matrix ( with block size @xmath104 ) of interactions of each panel with itself and its two neighbors . these are computed directly . all other block matrix interactions are obtained using standard gauss - legendre weights @xmath105 scaled to the dimensions of the @xmath100-th source panel . this structure of the far - field interactions permits the straightforward use of fast multipole acceleration and the hierarchical direct solver of @xcite . in domains with corners , but not multi material junctions , exponentially adaptive grids maintain high - order accuracy ( see , for example @xcite ) . in the simplest version , one can first divide the boundary into equal size subintervals and employ a @xmath42-th order generalized gaussian quadrature rule on each . for each segment that impinges on a corner point , one can further subdivide it using a dyadically refined mesh , creating @xmath106 additional subintervals , where @xmath107 is a specified numerical precision . if the same @xmath42-th order rule is used for each refined subinterval , it is straightforward to show that the resulting rule has a net error of the order @xmath108 . the need for dyadic refinement comes from the fact that the densities @xmath109 or @xmath8 may develop singularities at the corner points and the refinement yields a high order piecewise polynomial approximation of the density . for @xmath110 and @xmath111 , the net corner error is about @xmath112 while for @xmath113 , it is about @xmath114 ( see fig . [ figskel ] for an illustration ) . in recent work , @xcite have shown that one can dramatically reduce the number of degrees of freedom in the vicinity of the corner by the use of compression , as well . we have not used such optimization here . it is now appreciated ( see , for example , @xcite ) that the condition number of a properly discretized system of equations is very well controlled . following discretization , we use bremer s approach @xcite here , which involves setting the discrete variables to be @xmath115 and @xmath116 , rather than the density values @xmath117 and @xmath118 themselves . this ensures that that spectrum of the discrete system approximates the spectrum of the continuous integral equation in @xmath119 . the formal analysis is somewhat involved , since operators that are compact on smooth domains are only bounded ( but not compact ) on domains with corners . we refer the reader to @xcite for details . in the case of multiple subdomains , a natural approach would be to represent the field in each subdomain @xmath120 with helmholtz coefficient @xmath121 in terms of layer potentials on the boundary @xmath122 of @xmath120 . that is , in subdomain @xmath120 , we would represent the solution as @xmath123({{\mathbf x } } ) + { { { \cal d}}}^{k_i}[\gamma_i,\mu]({{\mathbf x } } ) \label{e : qprepi}\ ] ] with @xmath64 and @xmath65 replaced by their quasi - periodic counterparts for subdomains that extend across the unit cell ( the @xmath82 , @xmath1 , and @xmath2 domains in fig . [ fig - geom1 ] ) . in doing so , it turns out that the analog of equations ( [ ib1],[ib2 ] ) fails to converge in the presence of multi - material junctions . the reason for this is simple , and analyzed in @xcite . consider the interface condition ( [ ib2 ] ) for @xmath124 lying on the segment @xmath125 in fig . [ figskel ] . restricting our attention just to the segments impinging on the corner point @xmath126 , we have @xmath127 + { { \cal n}_{{\mbox{\tiny qp}}}}^{k_{0}}(\overline{ae},\sigma,{\mathbf x } ) - { { \cal n}}^{k_{1}}(\overline{ad},\sigma,{\mathbf x } ) \ + \ \dots \hspace{.3 in } \nonumber \\ + \left [ { { \cal t}_{{\mbox{\tiny qp}}}}^{k_{0}}(\overline{ab},\mu,{\mathbf x } ) - { { \cal t}}^{k_{1}}(\overline{ab},\mu,{\mathbf x } ) \right ] + { { \cal t}_{{\mbox{\tiny qp}}}}^{k_{0}}(\overline{ae},\mu,{\mathbf x } ) - { { \cal t}}^{k_{1}}(\overline{ad},\mu,{\mathbf x } ) \ + \ \dots = - \left [ \frac{\partial u^{{\mbox{\tiny in}}}}{\partial \nu}({{\mathbf x } } ) \right ] \label{ib2a}\end{aligned}\ ] ] note that both the terms @xmath128 and @xmath129 involve hypersingular contributions at the junction @xmath126 without forming part of a difference kernel . this destroys the high - order accuracy of the scheme . by using a global integral representation , it was shown in @xcite that high - order accuracy can be restored . that is , instead of ( [ e : qprepi ] ) , we let @xmath130({{\mathbf x } } ) + { { { \cal d}}}^{k_i}[\gamma,\mu]({{\mathbf x } } ) \label{e : qprepall}\ ] ] and apply the continuity conditions . for @xmath131 lying on an interface between subdomains with helmholtz coefficients @xmath121 and @xmath132 , we have [ grpc ] @xmath133 \label{ic1}\\ - \sigma({\mathbf x } ) & + { { \cal n}}^{k_{i}}(\gamma,\sigma,{\mathbf x } ) - { { \cal n}}^{k_{j}}(\gamma,\sigma,{\mathbf x } ) + { { \cal t}}^{k_{i}}(\gamma,\mu,{\mathbf x } ) - { { \cal t}}^{k_{j}}(\gamma,\mu,{\mathbf x } ) = - \left [ \frac{\partial \phi^{{\mbox{\tiny in}}}}{\partial \nu}({\mathbf x } ) \right ] \label{ic2}\end{aligned}\ ] ] as above , the operators @xmath64 , @xmath65 , @xmath134 , @xmath91 are replaced by their quasi - periodic counterparts for subdomains that extend across the unit cell ( the @xmath82 , @xmath1 , and @xmath2 domains in fig . [ fig - geom1 ] ) . the global representation ( [ e : qprepall ] ) is non - physical " in the sense that the field in a given subdomain is determined , in part , by layer potential components that are not actually part of the subdomain s boundary . by doing so , however , we remove all hypersingular terms from the integral equation . only difference kernels appear in the final linear system . one could improve efficiency somewhat , while achieving similar results , by supplementing the representation ( [ e : qprepi ] ) only by the boundary segments that actually impinge on a multi - material junction . we use the fully global representation in our experiments here for the sake of simplicity . for related approaches addressed at solving problems with multi - material junctions , see @xcite . given a well - conditioned and high order discretization , large scale scattering problems in singular geometries can be solved by using fast multipole - accelerated iterative solution methods such as gmres @xcite . while these are asymptotically optimal schemes , one is often interested in modeling the interaction of a given physical structure ( such as the geometry in fig . [ fig - geom1 ] ) with a large number of incoming fields . this requires the solution of an integral equation with multiple right - hand sides , and standard iterative methods do not take maximal advantage of this fact . direct solvers , on the other hand , first construct a factorization of the system matrix , then solve against each right - hand side using that factorization at a cost that is typically much lower . in the last decade , specialized versions have been created which are particularly suited to the integral equation environment . this is an active area of research and we do not seek to review the literature , except to note selected important developments in the case of hierarchically semiseparable matrices @xcite , @xmath135-matrices @xcite , and hierarchically block separable matrices @xcite . we provide a brief description of the approach , following the presentation of @xcite . let @xmath136 be the matrix discretization of an integral equation such as ( [ grpc ] ) , and let its indices @xmath137 be ordered hierarchically according to a quadtree on the unit cell . this can be done by first enclosing the set of all associated points within a sufficiently large box . if the box contains more than a specified number of points , it is subdivided into four quadrants and its points distributed accordingly between them . this procedure is repeated for each new box added , terminating only when all boxes contain @xmath138 points . the boxes that are not subdivided are called _ leaf boxes_. for simplicity , we assume that all leaf boxes live on the same level of the tree , but this restriction can easily be relaxed . start at the bottom of the tree and consider the partitioning induced by the leaves . let @xmath100 be the number of leaf boxes and assume that each has @xmath139 points so that @xmath140 . then @xmath141 has the block form @xmath142 for @xmath143 . we now use the _ interpolative decomposition _ ( i d ) @xcite to skeletonize @xmath141 . the i d is a matrix factorization that rewrites a given low - rank matrix in terms of a subset of its rows or columns , called _ skeletons_. in the integral equation setting , the off - diagonal block rows @xmath144 \label{eq : block - row}\ ] ] are low - rank due to the smoothness of the green s function , and the same is true of the off - diagonal block columns . thus , it can be shown @xcite that the i d enables a representation of the form @xmath145 for each off - diagonal block , where @xmath146 , @xmath147 , and @xmath148 is a submatrix of @xmath149 , with @xmath150 . the matrix can then be written as @xmath151 where @xmath152 \in \mathbb{c}^{n \times n } \ , , \ ] ] @xmath153 \in \mathbb{c}^{n \times k } , \qquad { \mathsf{r}}= \left [ \begin{array}{ccc } { \mathsf{r}}_{1 } & & 0 \\ & \ddots\\ 0 & & { \mathsf{r}}_{p } \end{array } \right ] \in \mathbb{c}^{k \times n}\ ] ] are block diagonal with @xmath154 , and @xmath155 \in \mathbb{c}^{k \times k}\ ] ] is dense with zero diagonal blocks . [ rmk : proxy ] the efficient calculation of the interpolation matrices @xmath156 and @xmath157 , and the associated skeleton indices , in ( [ eq : block - separable ] ) is somewhat subtle . briefly , it involves separating out neighboring and far - field interactions and representing the latter via free - space interactions with a local `` proxy '' surface . this is justified by the observation that any well - separated interaction governed by a homogeneous partial differential equation ( here , the helmholtz equation ) can be induced by sources / targets on the proxy surface , each of which is expressed in terms of the free - space kernel . for details , see @xcite . in this paper , for a box of scaled size @xmath158 , we use the circle of radius @xmath158 about the box center as its proxy surface . note that all neighbors are defined relative to the periodicity of the unit cell . now consider the linear system @xmath159 . one way to solve it is to construct @xmath160 directly from ( [ eq : compressed - representation ] ) using a variant of the sherman - morrison - woodbury formula . this approach is taken in @xcite . here , we follow the strategy of @xcite instead and let @xmath161 and @xmath162 to obtain the equivalent _ sparse _ system @xmath163 \left [ \begin{array}{c } { \mathsf{x}}\\ { \mathsf{y}}\\ { \mathsf{z}}\end{array } \right ] = \left [ \begin{array}{c } { \mathsf{b}}\\ 0\\ 0 \end{array } \right ] . \label{eq : sparse - embedding}\ ] ] this can be solved efficiently using any standard sparse direct solver and may provide better stability . in this paper , we use the open - source software package umfpack @xcite . since @xmath164 is a submatrix of @xmath141 ( up to diagonal modifications ) , @xmath164 can itself be expressed in the form ( [ eq : compressed - representation ] ) by moving up one level in the tree and regrouping appropriately . this leads to a multilevel decomposition @xmath165 where the superscript indexes the tree level @xmath166 with @xmath167 denoting the root . we call this process _ recursive skeletonization_. the analogue of ( [ eq : sparse - embedding ] ) is @xmath168 \left [ \begin{array}{c } { \mathsf{x}}^{(\lambda)}\\ { \mathsf{y}}^{(\lambda)}\\ \vdots\\ { \mathsf{x}}^{(1)}\\ { \mathsf{y}}^{(1)}\\ { \mathsf{x}}^{(0 ) } \end{array } \right ] = \left [ \begin{array}{c } { \mathsf{b}}\\ 0\\ \vdots\\ 0\\ 0\\ 0 \end{array } \right ] , \label{eq : multilevel - embedding}\ ] ] corresponding to expanding @xmath164 out in the same way . it can be shown that the solver requires @xmath169 work when the unit cell is a moderate number of wavelengths in size . we refer the reader to @xcite for further discussion . for our present purposes , we simply note that the output of the fast direct solver is a _ compressed _ representation of the inverse which is computed in two steps : 1 . a recursive skeletonization procedure to obtain the compressed forward operator ( [ eq : multilevel - representation ] ) ; and 2 . a factorization of the sparse matrix embedding in ( [ eq : multilevel - embedding ] ) . both steps have the same asymptotic complexity , but the constant for compression is typically far larger . after the inverse has been computed , it can be applied to each right - hand side as needed at a much lower cost . the i d can be constructed to any specified relative precision @xmath107 . this is an input parameter to recursive skeletonization and hence to the direct solver . it can be shown that if ( [ eq : multilevel - representation ] ) has relative error @xmath170 , as is often the case numerically , then the algorithm produces a solution with relative error @xmath171 , where @xmath172 is the condition number of @xmath141 . in particular , if @xmath173 , as for the integral equation ( [ grpc ] ) , then the error is @xmath170 . [ rmk : in - out ] although we have assumed in the discussion above that each block at the same level has the same size , this is in no way essential to the algorithm . in fact , our code uses separate `` incoming '' ( row ) and `` outgoing '' ( column ) skeletons for each box . this enables some additional optimization , which , for the present case , can be especially pronounced . this is because while each point receives incoming interactions from only the two wavenumbers on either side of the segment to which it belongs , it sends outgoing interactions consisting of all wavenumbers in the problem . for example , for a point on the segment @xmath125 in fig . [ figskel ] , it receives at wavenumbers @xmath174 and @xmath175 but sends at wavenumbers @xmath174 , @xmath175 , and @xmath176 . therefore , the outgoing skeleton dimension is typically larger , and the amount by which it is larger increases with the total number of wavenumbers / domains . the fast direct solver of the previous subsection allows the robust and accurate solution of @xmath177 where we have made explicit the dependence of the integral equation ( [ grpc ] ) on the incident angle @xmath178 . in the present setting , we are interested in solving ( [ eq : system - angle ] ) for many @xmath178 . this is not a situation that the direct solver can easily handle since @xmath179 is not fixed . in this subsection , we describe a modified strategy for computing a compressed representation ( [ eq : multilevel - representation ] ) of @xmath179 such that it can be rapidly updated to yield a compressed representation of @xmath180 for any @xmath181 _ without having to re - skeletonize_. since skeletonization is typically the most expensive step , this can offer significant computational savings . the sparse matrix in ( [ eq : multilevel - embedding ] ) must still be updated and re - factored , but the relative cost of this is small . to see why such a uniform skeletonization might be possible , consider any finite truncation of the periodic geometry so that it consists merely of a very large array of many , many scatterers . then the governing integral equation is specified in terms of the _ free - space _ green s function so that @xmath141 is independent of @xmath178 . the only angle dependence comes from the incoming data @xmath182 . therefore , only one skeletonized representation of @xmath141 is needed for all @xmath178 . the same is true of any finite approximation to the periodic problem . we now make this intuition precise by considering all interactions , say , incoming on a given box . this is given by the off - diagonal block row ( [ eq : block - row ] ) and can be decomposed as @xmath183 in terms of the near- and far - field contributions , respectively , to the quasi - periodic green s function @xmath184 following section [ s : gqp ] . clearly , an interpolation basis for both terms together provides an interpolation basis for the sum , so @xmath185 can be skeletonized by applying the i d to the rows of the matrix @xmath186.\ ] ] since @xmath57 consists only of well - separated interactions , by remark [ rmk : proxy ] , @xmath187 can be replaced by a matrix @xmath188 corresponding to free - space interactions with a proxy surface . in linear algebraic terms , this means that @xmath189 can be written as @xmath190 for some matrix @xmath191 . hence , @xmath192 \left [ \begin{array}{cc } { \mathsf{i}}\\ & { \mathsf{t}}_{i \leftarrow } ( \theta ) \end{array } \right ] , \label{eq : a - proxy}\ ] ] so @xmath193 can be skeletonized by applying the i d to just the left matrix on the right - hand side . observe that the angular dependence of the far field has been eliminated . to eliminate the angular dependence of the near field , we can similarly expand @xmath56 in terms of a @xmath178-independent basis . this can be done using the functions @xmath194 corresponding to interactions with the self- , left- , and right - images , respectively , with corresponding matrices @xmath195 , @xmath196 , and @xmath197 . then , from ( [ e : gqpn ] ) , @xmath198 \left [ \begin{array}{c } { \mathsf{i}}\\ e^{-i \alpha d } { \mathsf{i}}\\ e^{i \alpha d } { \mathsf{i}}\end{array } \right ] , \label{eq : a - near}\ ] ] where , recall , @xmath199 , so @xmath185 can be skeletonized by applying the i d to @xmath200 , \label{eq : id - matrix}\ ] ] which we note has no angular dependence . thus , the interpolation matrices and skeleton indices resulting from compressing ( [ eq : id - matrix ] ) are valid for all @xmath178 . the same approach can be used for outgoing interactions and for interactions at each wavenumber . the result is a modified compressed representation @xmath201 where only the @xmath202 depend on @xmath178 . therefore , to obtain a compressed representation of @xmath180 for any other @xmath181 , it suffices to perform the update @xmath203 for each @xmath96 . this , in general , consists only of generating a very small subset of entries of the new matrix and requires @xmath169 work with a small constant . in summary , the full algorithm for analyzing multiple incident angles with fast updating is : 1 . compress the matrix @xmath179 for some initial @xmath178 by representing interactions with an angle - independent basis such as ( [ eq : id - matrix ] ) . this is an expensive recursive skeletonization that only needs to be performed once . 2 . embed the resulting decomposition ( [ eq : angle - representation ] ) into the sparse matrix of ( [ eq : multilevel - embedding ] ) and solve . 3 . for each new angle @xmath181 , update the compressed representation ( [ eq : angle - representation ] ) via @xmath204 . repeat step 2 . in our tests , we have often found it unnecessary to decompose @xmath205 as in ( [ eq : a - near ] ) . instead , we apply the i d to the left matrix on the right - hand side of ( [ eq : a - proxy ] ) , which depends on @xmath178 but seems to yield results that recover angle independence numerically . this optimization can reduce the constant associated with skeletonization by about a factor of @xmath206 . the algorithm presented above has been implemented in fortran . each boundary segment ( in the piecewise smooth boundary ) is first divided into @xmath207 equal subintervals . the first and last intervals are then further subdivided with dyadic refinement toward the corner using @xmath208 subintervals each . thus , the total number of intervals on each smooth component of the boundary ( each side ) is @xmath209 and the number of points is @xmath210 . we use the 8th order generalized gaussian quadrature rule of @xcite for logarithmic singularities and solve the integral equations ( [ grpc ] ) using recursive skeletonization @xcite with a tolerance of @xmath211 . all timing listed below are for a laptop with a 1.7ghz intel core i5 processor . points each and @xmath206 degrees of freedom per point , resulting in a complex matrix of size @xmath212 . after skeletonization , there are only @xmath213 points remaining at the coarsest level of the recursion . [ figskel],width=432 ] * example 1 : * we set @xmath214 , with @xmath7 chosen so that the helmholtz coefficient in the upper half - space , the trapezoidal scatterer , and the substrate are @xmath215 , @xmath216 , and @xmath217 , respectively . the incident angle is @xmath218 . the original matrix of dimension @xmath212 is compressed to one of dimension @xmath219 . the incoming and outgoing skeleton dimensions are slightly different as explained in remark [ rmk : in - out ] and computed as part of the recursion . the time for compression in our current implementation was 290 secs . ( while generating the necessary matrix entries required 1219 secs . ) . given the compressed representation , the solution time was 2.46 secs . the resulting accuracy was approximately @xmath220 . we plot the real part of the total field in fig . [ figs1 ] . in fig . [ figskel ] , we plot both the original set of discretization points and the skeletons that remain at the coarsest level of the recursion . incidence impinges on a periodic structure . the unit cell and its first neighbor are shown , with a zoom in the region of the triple - point.,width=432 ] given our integral representation of the scattered field , it is straightforward to compute the coefficients @xmath38 in ( [ e : rbu ] ) or ( [ e : rbd ] ) - the bragg diffraction amplitudes at the grating orders . for an incident field @xmath16 we simply let @xmath221 denote some height above the scatterers and rewrite ( [ e : rbu ] ) in the form @xmath222 where @xmath223 . thus , the @xmath224 can be computed using fourier analysis : @xmath225 the accurate calculation of @xmath38 from this formula depends on ensuring that the discretization in @xmath226 is sufficiently fine to resolve the integrand . in the near field ( when @xmath227 is small ) , the evanescent modes , corresponding to large @xmath139 , are still present in @xmath228 requiring a large number of points to avoid aliasing errors . by making @xmath227 sufficiently large , the evanescent modes are suppressed . and a mesh can be used that resolves only the propagating modes - that is , values of @xmath139 for which @xmath229 . * example 2 : * we now consider a scattering problem with a two - layered substrate ( fig . [ fig2layer ] ) . we again set @xmath214 and choose @xmath7 so that the helmholtz coefficient in the upper half - space , the trapezoidal scatterer , and the two substrate layers are @xmath215 , @xmath216 , @xmath217 and @xmath208 , respectively . we first set up the scattering problem for an angle of incidence of @xmath218 . the original matrix is of dimension @xmath230 , which is compressed to one of dimension @xmath231 . the time for compression was 416.6 secs . ( and for generating the matrix entries , 1762.3 secs . ) . the time for inversion was 2.9 secs . the relative error in the solution ( compared with standard lu factorization ) was @xmath232 . we then changed the angle of incidence to @xmath233 and used the updating method of section [ multangle ] . the time for updating the compressed forward operator was 68.4 secs.and the relative error in the solution was @xmath234 . in this problem , there are six propagating modes , with directions indicated in fig . [ scatmodes2 ] . incidence impinges on a periodic structure with a two - layer substrate . the unit cell and its first neighbor are shown , with a zoom in the region of the triple - point . [ fig2layer],width=432 ] . the length of the arrows in the various diffraction directions indicate the relative magnitude of the amplitudes @xmath38 . approximately @xmath235 of the energy is scattered upward . [ scatmodes2],width=336 ] * example 3 : * the complexity of the scattering pattern can be quite striking . in fig . [ hemifig ] is shown the scattering pattern from a semicircular scatterer with an angle of incidence of @xmath218 . we set @xmath214 and choose @xmath7 so that the helmholtz coefficient in the upper half - space , the semicircular scatterer and the substrate layer are @xmath236 , @xmath237 , and @xmath238 , respectively . there are 19 radiation modes at this angle of incidence . , @xmath239 , @xmath240 and @xmath241 . only the scatterer in the unit cell is plotted . the length of the arrows in the various diffraction directions indicate the relative amplitudes @xmath38 . approximately @xmath242 of the energy is scattered upward . [ hemifig],width=336 ] * examples 4 , 5 : * in our final examples , we compute the diffraction pattern across all angles of incidence from @xmath243 to @xmath244 for the scattering geometries depicted in examples 1 and 3 , except that for the trapezoidal - shaped scatterer , we increased @xmath41 by a factor of 3 , so that @xmath245 instead of 10 . for the semicircular scatterer , we decreased @xmath41 by a factor of 3 , so that @xmath246 instead of 30 . on the left - hand side of figs . [ trapscan ] and [ hemiscan ] are plotted the diffraction orders as a function of incident angle . that is , for each incident angle @xmath178 , the intersection of the indicated vertical line with the various curves are the _ bragg angles _ @xmath247 according to formula ( [ e : rbu ] ) , where @xmath36 and @xmath248 are chosen to enforce both quasiperiodicity and the helmholtz equation . the number of intersections of each vertical line on these left - hand plots defines the precise number of modes for a given angle of incidence . it is easy to see that each of the curves on the left - hand plots traverses the _ incident angle_-_scattered angle _ plane continuously ( until it disappears ) , so that we may enumerate the modes unambiguously from the lower left corner to the upper right corner . the labels ( 10 " , 19 " , 28 " ) in fig . [ trapscan ] are drawn on the 10th , 19th , and 28th such curve . the labels ( 4 " , 7 " , 10 " ) in fig . [ hemiscan ] are drawn on the 4th , 7th , and 10th such curve . on the right - hand side of figs . [ trapscan ] and [ hemiscan ] are plotted the fraction of energy radiated into each mode . the @xmath249th curve in the right - hand plots show the total energy scattered in modes @xmath158 through @xmath249 . thus , the separation between curves corresponding to modes @xmath249 and @xmath250 shows the fraction of energy radiated in the @xmath249th mode . highlighted in gray are the energies scattered in the 10th , 19th , and 28th modes in fig . [ trapscan ] and in the 4th , 7th , and 10th modes in fig . [ hemiscan ] . note that the strength can change quite abruptly when the incident angle is changed only slightly . . ( see the text for a discussion of the plots . ) [ trapscan],width=480 ] . ( see the text for a discussion of the plots . ) [ hemiscan],width=480 ] we have described an integral equation method for quasi - periodic scattering from layered materials with grating - like structures on the top " surface . it combines ( 1 ) the use of the quasi - periodic green s function , ( 2 ) the modified kress / mller / rokhlin integral equation for multi - material junctions @xcite , ( 3 ) the use of exponential refinement near geometric singularities , @xcite , and ( 4 ) the fast direct solver of @xcite . since the quasi - periodic green s function changes with each incident angle , there is a global change to the system matrix with each new illumination . we have shown , however , that the _ difference _ between green s functions at different angles of incidence is ( hierarchically ) smooth so that the compressed representation of the system matrix can be rapidly updated . in recent work , gillman and barnett @xcite developed an alternative fast direct solver based on using the free - space green s function with auxilliary variables to impose quasi - periodicity . in that formulation , the bulk of the matrix is left unchanged for different illuminations . we suspect that the relative advantages of the two approaches will depend on the aspect ratio of the unit cell , the spatial dimension ( 2d vs. 3d scattering ) and detailed implementation issues . both approaches have asymptotically optimal complexity for unit cells that are a modest number of wavelengths in size . this work was supported by the applied mathematical sciences program of the u.s . department of energy under contract defgo288er25053 and by the air force office of scientific research under nsseff program award fa9550 - 10 - 1 - 0180 . klh was also supported in part by the national science foundation under grants dge-0333389 and dms-1203554 . jyl was also supported in part by the priority research centers program ( 2009 - 0093827 ) and the basic science research program ( 2012 - 002298 ) through the national research foundation ( nrf ) of korea . ( 2012 ) , on the nystrm discretization of integral equations on planar curves with corners , _ appl . comput . harm . _ , * 32 * , 4564 . ( 2011 ) , universal quadratures for boundary integral equations on two - dimensional domains with corners , _ j. comput . _ , * 229 * , 82598280 . ( 2011 ) , integral formulation of the second kind for multi - subdomain scattering " , proc . 10th int . conf . on math . aspects of waves ( waves 2011 ) , pacific institute for the mathematical sciences ( to appear , http://www.pims.math.ca/ ) . ( 2012 ) , multiple traces boundary integral formulation for helmholtz transmission problems " , _ adv . comput . math . _ , * 37 * , 3991 . ( 2012 ) , a fast direct solver for structured linear systems by recursive skeletonization " , _ siam j. sci . _ , * 35 * , a2507a2532 .	we present a new integral equation method for the calculation of two - dimensional scattering from periodic structures involving _ triple - points _ ( multiple materials meeting at a single point ) . the combination of a robust and high - order accurate integral representation and a fast direct solver permits the efficient simulation of scattering from fixed structures at multiple angles of incidence . we demonstrate the performance of the scheme with several numerical examples . acoustic scattering , electromagnetic scattering , triple junctions , multiple material interfaces , boundary integral equations , fast direct solvers 43.20.-f , 41.20.jb , 81.05.xj , 02.60.-x , 02.30.rz 65r20 , 78a45 , 31a10 , 35j05
spontaneous emission is a well established fundamental phenomenon @xcite . it occurs due to interaction of excited emitters with the vacuum modes of the environmental electromagnetic field reservoir . useful applications of spontaneous radiation control may arise , for instance , in higher frequency coherent light generation @xcite or spontaneous parametric down conversion processes @xcite . on the other side , spontaneous emission often plays a negative role in quantum processing of information @xcite . therefore , it is not surprising that a significant amount of work is carried out regarding its control . particularly , earlier approaches to influence the spontaneous emission were by using optical cavities @xcite . a modern and more advanced version of those ideas consists in using photonic crystals environments where photon forbidden bands occur leading to spontaneous emission inhibition or localization @xcite . infrequent application to a two - level atom of microwave pulses @xcite or sequence of pulses @xcite , or rather intense low - frequency coherent fields @xcite ( see also @xcite ) lead to spontaneous emission control as well . quenching of spontaneous emission occurs as well via involving quantum interference effects between various decaying pathways which are dependent on mutual orientation of corresponding transition dipoles @xcite . furthermore , the lamb shift of laser - dressed atomic states and quantum interferences due to energy shifts and their effect on spontaneous emission were investigated too , in refs . one can also control the spontaneous emission by periodically shifting the atomic transition frequency from the atom - cavity resonance @xcite or via coupling a single state to a continuum of many states @xcite . remarkably , periodically perturbed atomic transitions lead to a number of fascinating effects such as induced transparency or extreme ultra - short pulses , respectively @xcite . ( color online ) schematic diagram of a two - level emitter with modulated transition frequency : ( a ) without and ( b ) with showing the involved quantum coherences among specific transition pathways . due to modulation , multiple induced decaying channels , @xmath0 , interfere such that spontaneous emission is slowed down . @xmath1 is the single - qubit spontaneous decay rate in the absence of modulation.,width=283 ] here , we demonstrate suppression of spontaneous decay of a two - level system ( qubit ) that is embedded in a broadband electromagnetic field reservoir and is subjected to an intense , time - dependent , frequency modulation driving force . the suppression is a direct consequence of quantum interference effects induced by the modulation . a frequency shift to the transition frequency is induced as well . furthermore , the absolute phase of the modulation can be a convenient tool to control these processes . coherent modulation of the transition frequency leads to appearance of new decay channels that may interfere destructively contributing to spontaneous emission inhibition ( see fig . [ fig1 ] ) . this occurs when the frequency of modulation is of the order of the bare - state qubit s decay rate or less . the spontaneous emission is described by an exponential decaying law with a time - dependent decay rate and exhibiting plateaus with a very slow decoherence rate . the quantum decoherence due to spontaneous emission can be further minimized via stronger frequency modulation depths . moreover , the induced time - dependent frequency shift depends on external control parameters , such as the applied intensity and the external field amplitude absolute phase , and can be influenced accordingly . it vanishes , however , at resonance and in the absence of quantum interference effects due to frequency modulation processes . in a free - space setup , the spontaneous emission inhibition is less probable via transition frequency modulation of an excited two - level emitter . this deviation to @xcite arises because our treatment is classical and especially limited to moderately intense modulating fields . our system can be implemented , for example , via off - resonant laser driving of a two - level emitter @xcite . one can apply a laser field with a high non - resonant frequency @xmath2 and possessing a periodically modulated amplitude of the field strength @xmath3 with @xmath4 to a two - level atom of frequency @xmath5 . then a modulated shift @xmath6 of the transition frequency is achieved via the quadratic stark effect , i.e. , @xmath6=@xmath7 , where @xmath8 is the modulation amplitude . additional systems can be molecules or quantum dots , even those possessing permanent dipoles @xcite . when pumped with an intense low - frequency coherent field , the amplitude of the frequency modulation will be proportional to the magnitude of the permanent dipole multiplied by the external field strength . an alternative scheme can be as well a two - level quantum dot embedded in a broadband microcavity and interacting with a surface acoustic wave coherently modulating its transition frequency @xcite . superconducting qubits with periodically perturbed transition frequencies and weakly coupled with a quantum lc circuit or a nanomechanical resonator are suitable candidates , too @xcite . the article is organized as follows . in sec . ii we describe the analytical approach and the system of interest , while in sec . iii we analyse the obtained results . the summary given in sec . iv is followed by two appendixes . the hamiltonian , @xmath9 , describing the system of interest can be represented via ( see @xcite or for a detailed derivation for the analogous case with usual vacuum appendix a ) : @xmath10 where the first term , @xmath11 , characterizes the free hamiltonian of the electromagnetic field ( emf ) as well as of the qubit subsystem with modulated transition frequency whereas the second one , i.e. @xmath12 , accounts for the interaction of the two - level qubit with the vacuum modes of the environmental electromagnetic field reservoir . here , @xmath13 , @xmath14^{\dagger}$ ] and @xmath15 are the well - known quasispin operators obeying the commutation relations : @xmath16=2s_{z}$ ] and @xmath17=\pm s^{\pm}$ ] . the creation @xmath18 and annihilation @xmath19 electromagnetic field operators satisfy the commutation relations : @xmath20=\delta_{kk'}$ ] and @xmath21=[a^{\dagger}_{k},a^{\dagger}_{k'}]=0 $ ] . further , @xmath5 is the qubit s transition frequency @xmath22 ( see fig . [ fig1 ] ) in the absence of classical modulation , while @xmath8 is the modulation amplitude with frequency @xmath23 and phase @xmath24 . the two - level emitter possessing the transition dipole moment @xmath25 couples with the vacuum modes via the coupling constant @xmath26 . in the following , we perform a unitary transformation : @xmath27 with @xmath28 and arrive at the hamiltonian : @xmath29 where @xmath30 while @xmath31 is the corresponding ordinary bessel function . here , we used the expansion via the @xmath32th - order bessel function of the first kind , i.e. , @xmath33 as well as the notation : @xmath34 , and dropped the tilde afterwards . in the weak qubit - environment coupling limit , one can obtain the master equation describing the quantum dynamics of any atomic operator @xmath35 . for this , we use the standard elimination procedure of the electromagnetic field operators from the heisenberg equation : @xmath36\rangle , \label{eqq}\end{aligned}\ ] ] where the notation @xmath37 indicates averaging over the initial state of both the qubit and the surrounding electromagnetic field bath @xcite . as an environmental electromagnetic field reservoir , we consider a broadband optical cavity possessing the frequency @xmath38 , qubit - cavity coupling being @xmath39 , and a cavity leaking constant denoted by @xmath40 ( the free - space situation is described in appendix a ) . thus , the heisenberg equations for the field operators are : @xmath41 with @xmath42^{\dagger}$ ] and @xmath43 . its formal solution in the weak - coupling limit is @xmath44+@xmath45 , where @xmath46 while @xmath47 in the markov approximation we have @xmath48 . then the integral @xmath49 inserting this expression in eq . ( [ as ] ) and keeping only the slower contributions , that is , we are interested in frequency modulation regimes slower than the cavity decay rate , i.e. @xmath50 , one arrives at : @xmath51 then , one can write down the master equation for an arbitrary mean - value of a qubit operator @xmath35 that can be obtained after introducing eq . ( [ heff ] ) in the corresponding heisenberg equation , i.e. eq . ( [ eqq ] ) : @xmath52\rangle e^{-im(\omega t + \phi ) } \nonumber \\ & + & \langle [ q , s^{+}]a\rangle e^{im(\omega t + \phi)}\ } , \label{mq1}\end{aligned}\ ] ] where an overdot denotes differentiation with respect to time . introducing eq . ( [ af ] ) in the master equation ( [ mq1 ] ) and taking into account that @xmath53 and @xmath54 one obtains : @xmath55\rangle - \gamma(t)\{\langle s^{+}[s^{-},q]\rangle \nonumber \\ & + & \langle [ q , s^{+}]s^{-}\rangle \}. \label{mq}\end{aligned}\ ] ] the operator form of eq . ( [ mq ] ) looks standard , i.e. of lindblad form @xcite , with , however , time - dependent coefficients , namely : @xmath56 } , \nonumber \\ \gamma(t ) & = & \sum^{\infty}_{\{m , n \ } = -\infty}\bar \gamma_{n}j_{m}(\chi ) j_{n}(\chi)\cos{[(n - m)(\omega t + \phi ) ] } , \nonumber \\ \label{kf}\end{aligned}\ ] ] with @xmath57 here , @xmath58 and @xmath59 describe the time - dependent frequency shift and spontaneous decay process , respectively , while @xmath60 is the near - resonance single - qubit spontaneous decay rate without frequency modulation , i.e. when @xmath61 . in the numerical simulations we truncate the summation range @xmath62 to @xmath63 . this is justified , as @xmath64 $ ] for near qubit - cavity resonance . concretely , @xmath65 is chosen such that the results converge , i.e. remain unchanged if one further increases @xmath65 . note that this is not the case for vacuum free - space setups ( see appendix a ) . furthermore , in order to avoid unphysical results @xcite , @xmath65 should be the same for both indices and , also , eq . ( [ mq ] ) should be independent on exchange of indices , i.e. @xmath66 . the population quantum dynamics of an initially excited two - state radiator can be easily obtained from eq . ( [ mq ] ) , namely : @xmath67 } - 1/2 , \label{sz}\end{aligned}\ ] ] with a generalized spontaneous decay rate given by : @xmath68 one can see here that the qubit inversion obeys a modified exponential decay law with a time - dependent decay rate . in the absence of frequency modulation , i.e. @xmath61 , one recovers the standard exponential law near qubit - cavity resonance @xcite : @xmath69 } - 1/2 . \label{sz0}\end{aligned}\ ] ] thus , the periodical modulation of the qubit s transition frequency modifies the spontaneous decay . in the following section , we shall describe the quantum dynamics of an excited two - level emitter with modulated transition frequency . we proceed to investigate the qubit s dynamics based on eqs . ( [ af]-[sz ] ) . one can observe from eq . ( [ af ] ) that the atomic dipole may oscillate at frequencies @xmath70 , where @xmath71 is an arbitrary integer number including zero . this means that photons at these frequencies are generated that can lead to interference effects . indeed , inspecting eq . ( [ mq ] ) , one can realize that the two - level emitter with the frequency modulation is reduced to an equivalent system containing multiple excited dressed levels , @xmath72 @xmath73 , decaying to the ground state ( see fig . [ fig1 ] ) . when the dressed - state splitting is of the order of the cavity mediated radiator s decay rate @xmath1 then quantum interferences occur among various transition decay paths . for instance , in fig . ( [ fig1]a ) the two decay channels @xmath74 interfere leading to appearance of quantum coherences schematically shown in fig . ( [ fig1]b ) . technically , due to the bessel function property , @xmath75 , some of the terms from expressions ( [ kf ] ) cancel each other while others add up . to illustrate this we chose the simplest case @xmath76 and @xmath77 , and then the expression for @xmath59 takes the form : @xmath78}\bigr\}/(\kappa^{2}+\omega^{2 } ) . \label{gmt}\end{aligned}\ ] ] the first three terms from eq . ( [ gmt ] ) describe the spontaneous emission processes on the induced transitions : @xmath79 and @xmath80 , respectively ( see fig . [ fig1]a ) . the last term in eq . ( [ gmt ] ) takes into account the cross - correlations among the spontaneously decaying channels ( see fig . [ fig1]b , where @xmath71 denotes a particular sub - level ) : @xmath81 and @xmath82 or vice versa ( and , hence , a pre - factor of @xmath83 there ) , i.e. characterizes quantum decay interference effects @xcite . on the other hand , the cross - decaying correlations among the transition paths : @xmath84 and @xmath85 reciprocally cancel each other obviously , this illustration scheme can be extended to @xmath86 ( see fig . [ fig1]b showing the induced coherences for @xmath87 ) . these processes together with frequency modulation dressing of the multiple decaying rates will lead to a slowing down of the spontaneous emission processes that are also absolute phase dependent . notice that for @xmath88 the @xmath71th- order bessel function of the first kind can be represented as @xcite : @xmath89 the dependence @xmath90 will also explain the quenching of the spontaneous decay processes for larger modulation depths . this tendency persists even in the absence of quantum coherences due to cross - damping effects . if @xmath91 the spontaneous decay rate tends to even lower values due to the prefactor @xmath92 $ ] . figure ( [ fig2 ] ) shows the population kinetics of an excited two - level emitter given by eq . ( [ sz ] ) for some parameters of interest . particularly , the short - dashed line depicts the typical exponential spontaneous decay low in the absence of frequency modulation which is characterized by eq . ( [ sz0 ] ) . the solid and long - dashed curves describe the spontaneous decay processes when the transition frequency is modulated with a modulation depth @xmath93 and a phase @xmath94 or @xmath95 , respectively . the phase - dependence is a clear evidence of quantum interference effects . this occurs for stronger modulation depths @xmath96 and when the frequency of modulation @xmath23 is comparable with or less than the single - qubit decay rate @xmath1 , and @xmath97 . for the sake of comparison , the dotted curve characterizes the spontaneous emission behaviour without taking into account the quantum coherences due to cross - damping effects ( see fig . [ fig1]a ) , that is , in eq . ( [ sz0 ] ) we have taken @xmath98 instead of @xmath1 . thus , concluding , the cross - damping effects ( i.e. the terms with @xmath99 in eq . [ kf ] and eq . [ sz ] ; see , also , fig . [ fig1]b ) contribute considerably to the final spontaneous decay processes ( compare solid , dotted and the long - dashed curves in fig . [ fig2 ] , respectively ) . the population behaviors shown in fig . ( [ fig2 ] ) are in accordance with the time - dependence form of @xmath59 given in eq . ( [ kf ] ) . the almost decoherence - free plateaus observed in fig . ( [ fig2 ] ) correspond to lower values of @xmath59 ( see fig . [ fig2 g ] ) . the inhibition of the spontaneous decay can be further improved by increasing the modulation depth @xmath96 . therefore , in fig . ( [ fig3 ] ) , we have fixed the evolution time at @xmath100 and changed the modulation depth @xmath96 accordingly . at lower modulation amplitudes , or in its absence , the qubit is in the ground state at this evolution stage . as it was already mentioned , stronger modulation depths contribute to a further slowing of the quantum decoherence . the reason is the interplay between interference effects among multiple decay channels described above and the frequency modulation dressing of the corresponding decay rates ( see eq . [ bsf ] ) . note however , that the opposite case , i.e. @xmath101 , does not show any time- or phase - dependence in the parameters entering in eq . ( [ mq ] ) or eq . ( [ sz ] ) and , consequently , no quantum interference effects among the different transition pathways occur . this situation was nicely investigated in refs . @xcite , respectively . we shall further focus on discussions around the frequency shift due to periodical modulation of the transition frequency . therefore , the frequency shift @xmath58 , given in eq . ( [ kf ] ) , is plotted in fig . ( [ fig4 ] ) for particular parameters . here , again , one can observe phase - dependent behaviors due to induced quantum coherences . particularly , when @xmath76 and @xmath77 , we have from eqs . ( [ kf],[kdg ] ) : @xmath102 where @xmath103 . one can observe here that the frequency shift is due to cross - correlations among the transition channels : @xmath84 and @xmath85 , respectively , i.e. opposite to spontaneous emission contributions where these processes cancel out . evidently , these discussions can be generalized for @xmath86 . notice that this frequency shift vanishes in the absence of transition frequency modulation at resonance , i.e. @xmath61 , or when @xmath101 at @xmath77 . for an experimental realization of the proposed scheme we need moderate modulation depths . this can be achieved , for instance , in molecular or quantum dot systems possessing a permanent dipole @xmath104 as it was also mentioned in the article . for @xmath105 and @xmath106 being the amplitude strength of the applied low - frequency coherent field , one can obtain the necessary modulation depth @xmath107 that is smaller than the transition frequency of the two - level qubit while @xmath88 . in asymmetrical quantum dot systems the permanent dipole is proportional to the size of the quantum dot and this can be used in engineering of the required model @xcite . certain molecules possess this property too , i.e. @xmath105 @xcite . finally , while we have considered a broad - band cavity environmental reservoir , the multiple induced decay interference approach developed may also be applied for quantized vacuum modes of free - space in a related setup @xcite , for instance . there , applying a quantized and sufficiently strong low frequency field beyond applicability here , to a two - level atom including far - off resonant states in free - space , one can induce quantum interferences among few - photon induced transitions . those process s scaling show an interplay between different relevant detunings and applied intensity strengths such that one can stop at a particular @xmath65- photon process @xcite ( see appendix b ) . summarizing , we have demonstrated how quantum decay interference phenomena induced among multiple decaying channels occurring due to moderately intense transition frequency modulation of a two - level emitter embedded in a broadband electromagnetic field reservoir together with frequency modulation dressing of the spontaneous decay rates can suppress quantum dissipations due to spontaneous emission . particularly , phase - dependent low decoherence plateaus appear in such a process . furthermore , a generalized cross - correlated frequency shift to the two - level qubit s transition frequency is induced here as well . we acknowledge the financial support by the german federal ministry of education and research , grant no . 01dk13015 , and academy of sciences of moldova , grant no . 13.820.05.07/gf . furthermore , we benefited from useful discussions with joerg evers particularly on the employed weak - field limit as compared to @xcite . m.m . is grateful to the theory division of the max - planck - institute for nuclear physics from heidelberg , germany , for kind hospitality . in this appendix , we shall focus on the spontaneous decay of an excited two - level emitter with modulated transition frequency via usual vacuum modes of the emf reservoir . the purpose is to show that an unshaped vacuum is not sufficient for the spontaneous emission suppression investigated here . for convenience we derive the modulation hamiltonian entering in eq . ( [ hh ] ) . for this , one considers that a moderately strong low - frequency coherent field is applied to a two - state atom being initially in its excited state . the hamiltonian of this process is : @xmath108 here , @xmath5 is the transition frequency among the states @xmath109 while @xmath110 is the corresponding rabi frequency with @xmath23 being the frequency of the external applied low - frequency coherent field . the atom - vacuum coupling strength is : @xmath111 , where @xmath112 is quantized volume while @xmath113 is the polarization vector with @xmath114 . we apply a unitary transformation to the hamiltonian ( [ hp ] ) , i.e. @xmath115 , where @xmath116}$ ] . this transformation is useful as it allows to represent the hamiltonian via @xmath117photon processes involved , namely : @xmath118 where @xmath119 . it is easy to observe that the even - photon processes with @xmath120 correspond to modulation of the transition frequency , while the odd - photon processes , i.e. @xmath121 lead to induced transitions among the involved energy levels . for our purposes one requires @xmath122 as well as @xmath123 . under these restrictions , i.e. taking into account that : @xmath124^{4}\biggr\},\end{aligned}\ ] ] the working hamiltonian is : @xmath125 here , @xmath126 , @xmath127 and @xmath128 . thus , we have considered that the frequency modulation takes place via two- and four - photon processes , simultaneously , while higher - photon effects are negligible because one can always select a system with @xmath129 , and for an even @xmath71 with @xmath130 . furthermore , the induced transitions through odd - photon processes among the involved energy levels do not occur because of the off - resonance and , therefore , are ignored here ( i.e. the second line of the hamiltonian [ hpf ] ) . the master equation describing the spontaneous decay of a two - level radiator with modulated transition frequency in free - space according to the hamiltonian ( [ hpt ] ) and in the born - markov , dipole and rotating wave approximations has the form : @xmath131\rangle + h.c . \nonumber \\ \label{mqq}\end{aligned}\ ] ] where @xmath132 with @xmath133 here , @xmath134 is the cauchy principal value , while @xmath135 is the free - space single - atom spontaneous decay rate in the absence of frequency modulation @xcite . in the following , we consider that @xmath88 while @xmath136 , i.e. the frequency modulation via simultaneous four - photon processes are negligible this can be achieved in the weak field regime , for example , when @xmath137 and @xmath138 , that is , one has @xmath139 while @xmath140 . for this reason we do not expect to obtain the quantum interference effects without a cavity like in @xcite . thus , the motivation to keep the modulation of the transition frequency due to a simultaneous four - photon process in the hamiltonian ( [ hpt ] ) as well as in the master equation ( [ mqq ] ) , i.e. the term proportional to @xmath141 , was to show that spontaneous decay via the induced absorption / emission of four photons is taking place also due to transition frequency modulation via a two - photon process described by @xmath8 ( for instance when @xmath142 ) . this latter process is more probable than the corresponding one due to @xmath141 . figure ( [ fig5 ] ) shows the spontaneous decay law of an excited atom in free space , i.e. @xmath143}-1/2 $ ] where @xmath144 with @xmath145 , when @xmath146 and @xmath147 , respectively , and for some particular parameters of interest . while the spontaneous decay is clearly slowed down for @xmath148 one can not predict what is the particular value of @xmath149 which would be realized in a real experiment . furthermore , larger values of @xmath149 with a fixed @xmath150 do not lead to spontaneous emission inhibition . thus , it is unlikely that the spontaneous emission will be inhibited in free - space via transition frequency modulation of an excited two - level emitter unless and until we are not able to select a particular @xmath149-photon process . in ref . @xcite a related mechanism was discussed where an effective two level system was driven with a quantum field with frequency @xmath23 of the order of the linewidth of the considered transition . since in reality the low - frequency field will also couple off - resonantly to all transitions in the atom , one can justify a hamilton operator with extra interference terms as indicated in detail in the ref . @xcite with the consequence of possible spontaneous emission elimination . the physics behind atom pumping with a classical low frequency field in the weak field approximation or a quantized low frequency field consists in the fact that in the first case only modulation of the transition frequency occurs which does not affect the spontaneous emission in free space , while in the second case additional transitions are induced that may interfere leading to modification of the spontaneous decay . particularly , in @xcite a quantized low frequency strong field was interacting with a two - level atom being initially in its excited state @xmath151 . in order to adequately describe the system the far - off resonant coupling to the other states was included as well . in what follows , we recall in a somewhat different simplified way how the additional terms due to off - resonant coupling may appear in the interaction hamiltonian . if as an example one considers a two - level atom with one extra far off - resonant state @xmath152 and with its energy levels from down to up denoted as @xmath153 and @xmath152 , respectively , then the quantized low frequency applied field will induce additional transitions via the path @xmath154 here @xmath155 are the corresponding coefficients due to strong external pumping and obtained after elimination of the far - laying excited level @xmath152 , while @xmath156 is the photon creation ( annihilation ) operator for the external low frequency quantized intense field . one can show ( see the last entry of @xcite ) that the hamiltonian ( [ b1 ] ) does not affect the transition @xmath157 when it is dipole allowed . therefore , the spontaneous emission modification scheme described in this part applies to dipole - forbidden atomic transitions . since in a three - level atomic system one transition should be dipole - forbidden it results that the decay on @xmath157 atomic transition may indeed be dipole - forbidden . the excited state population will be described by the following equation : @xmath158 here , @xmath159 is the two - photon spontaneous decay rate on transition @xmath157 , while @xmath160 is the corresponding decay involving additional upper states . one can observe that the strong quantized field indeed modifies the upper state population . notice that a strong classical low frequency laser field will only modulate the frequency on a dipole - allowed transition and , correspondingly , the spontaneous emission will not be modified in free space ( see also appendix a ) . for dipole - forbidden @xmath161 transitions the off - resonant coupling to another state @xmath152 will also modulate the transition frequency as well as induce transitions to the ground state @xmath162 based on a hamiltonian of the next form : @xmath163 the spontaneous decay rate on the @xmath157 atomic transition may be @xmath164 . therefore , the classical off - resonant coupling to another state may not change significantly the decay rate because @xmath165 . in a two - level atomic system with two extra far off - resonant states , the spontaneous emission can be modified due to an applied strong quantized low frequency field even on a dipole - allowed @xmath166 transition in accordance with the results given in ref . @xcite . for instance , one of the terms in the total hamiltonian describing the decay via the path @xmath167 after the elimination of the additional higher energy levels is : @xmath168 such terms are responsible for the spontaneous decay modification in free - space through a quantized and strong low frequency applied field . on the other side , in an idealised situation of ignoring the presence of any other far - off resonant states in the atomic system and when neglecting strong field terms of the low frequency field , the presence of such interference terms could not be reproduced via a classical pumping field . zhu , m. o. scully , phys . lett . * 76 * , 388 ( 1996 ) ; m. a. g. martinez , p. r. herczfeld , c. samuels , l. m. narducci , c. h. keitel , phys . a * 55 * , 4483 ( 1997 ) ; h. lee , p. polynkin , m. o. scully , s .- y . zhu , phys . a * 55 * , 4454 ( 1997 ) ; e. paspalakis , p. l. knight , phys . lett . * 81 * , 293 ( 1998 ) ; c. h. keitel , phys . rev . lett . * 83 * , 1307 ( 1999 ) . y. v. radeonychev , m. d. tokman , a. g. litvak , o. kocharovskaya , phys . lett . * 96 * , 093602 ( 2006 ) ; y. v. radeonychev , v. a. polovinkin , o. kocharovskaya , _ ibid . _ * 105 * , 183902 ( 2010 ) ; v. a. antonov , y. v. radeonychev , o. kocharovskaya , _ ibid . _ * 110 * , 213903 ( 2013 ) .	the resonant quantum dynamics of an excited two - level emitter is investigated via classical modulation of its transition frequency while simultaneously the radiator interacts with a broadband electromagnetic field reservoir . the frequency of modulation is selected to be of the order of the bare - state spontaneous decay rate . in this way , one can induce quantum interference effects and , consequently , quantum coherences among multiple decaying transition pathways . depending on the modulation depth and its absolute phase , both the spontaneous emission and the frequency shift may be conveniently modified and controlled .
let @xmath6 be a smooth , bounded simply connected open set in the plane ( the exact smoothness which is required will be made precise in section [ prel ] ) and let us denote by @xmath7 its boundary . following l. euler , we define its elastic energy as @xmath8 where @xmath9 is the curvature abscissa and @xmath10 is the curvature . we will denote by @xmath11 the area of @xmath6 and @xmath12 its perimeter . the aim of this paper is to prove the following isoperimetric inequality . [ bhe05 ] for any bounded , smooth , simply connected open set @xmath13 @xmath14 where equality holds only for the disc . in other words , using the behavior of the elastic energy on rescaling , we get that for every @xmath15 , the disc is the unique solution for the minimization problem @xmath16 more precisely , if we perform any scaling of ratio @xmath17 , we have @xmath18 and @xmath19 . therefore , it is classical to prove that the following three minimization problems are equivalent ( in the sense that any solution of one gives a solution of the others after a suitable scaling ) : ( i ) : : @xmath20 ( ii ) : : @xmath21 ( iii ) : : @xmath22 let us make some comments . for a detailed bibliography on closed elasticae , we refer to the classical @xcite or the more recent @xcite . inequality was already known for convex domains . indeed , by a famous inequality due to m. gage @xcite , for any bounded convex domain @xmath23 with equality for the disc . therefore , @xmath24 the first inequality being the classical isoperimetric inequality , and the second the gage inequality . if the convexity hypothesis is dropped , then the gage inequality is false ( as shown by the counter - example of figure [ dumbell ] ) . the simply connectedness assumption is necessary . indeed , if we take as a domain @xmath6 the ring @xmath25 we get @xmath26 , while @xmath27 showing that @xmath28 when @xmath29 . in the same way , the boundedness assumption is also necessary . let us consider the following unbounded domain , subgraph of a gaussian function , but with finite area and elastic energy : @xmath30 we have @xmath31 while @xmath32 and we see that @xmath33 as @xmath34 . this shows that the assumptions in theorem [ bhe05 ] can not be weakened . the proof of theorem [ bhe05 ] is a classical variational proof ( existence , regularity , analysis of the optimality conditions ) , but the existence part is by no means easy since we need a control on the perimeter of a minimizing sequence . the boundedness constraints on @xmath35 and @xmath11 do not ensure that the perimeter is uniformly bounded , as shown by a counter - example like a dumbell , see figure [ dumbell ] . one has to be particularly careful that a minimizing sequence may a priori have a diameter going to infinity . the key point of our strategy is to analyze first the minimization of the elastic energy of drops enclosing a fixed area , i.e. closed loops without self - intersection points , which are smooth except one point , where the tangents are opposite . the result for drops will straight forward imply the conclusion of theorem [ bhe05 ] , since we prove that the optimal loop is smooth and satisfies optimality conditions over the full boundary . a direct observation proves that the curve is a circle . here is our plan . * we solve the minimization problem @xmath36 which is equivalent to . * we fix some radius @xmath37 and replace problem by @xmath38 where @xmath39 is the ball centered at @xmath40 of radius @xmath41 . we prove that every simply connected domain in @xmath39 satisfies @xmath42 . this is a key point for proving existence . * in order to be able to exploit the optimality conditions we have to deal with self - intersection points and with the points where the optimal set is touching the boundary of the ball @xmath39 . for this reason , we analyze the problem @xmath43 we refer to section [ drops ] for a precise definition of drops . we prove that an optimal drop does not have self - intersection points and , if @xmath41 is large enough , it does not touch the boundary of the ball @xmath39 ( up to a translation , inside the ball ) . henceforth , optimality conditions allow us to exhibit precisely the optimal drop and to evaluate its energy . this drop turns out to be _ unique_. * we come back to problem and prove that a limit of minimizing sequence can not have self - intersection points and can not touch the boundary of @xmath39 , provided that @xmath41 is large enough . consequently , optimality conditions can be written on all its boundary . the elimination of self - intersection points relies on the previous result on drops , since the presence of at least one such point would make the energy not smaller than the double of the energy of the optimal drop , which turns out to be larger than the energy of a disc of radius @xmath44 . as optimality conditions can be written on the full loop , we prove that there are only four shapes ( all of them having a very simple description ) which can be minimizers . an easy comparison argument , leads to the optimality of the ball . all curves @xmath45{\rightarrow}{{\mathbb r}}^2 $ ] are parametrized by the arc - length . we denote @xmath46 the angle of the tangent to @xmath0 with respect to the axis @xmath47 . the curvature of @xmath0 at the point @xmath48 will be denoted @xmath2 and it is equal to @xmath49 . since we shall work with curves with finite elastic energy , the function @xmath46 belongs to the sobolev space @xmath50 . using the embedding @xmath51 $ ] , for any @xmath52 , the function @xmath46 is , in particular , continuous . all curves we work in this paper have finite elastic energy @xmath53 } \|\theta ' ( s)\|^2 ds<+\infty .\ ] ] [ bh04 ] let @xmath54 and @xmath45 { \rightarrow}{{\mathbb r}}^2 $ ] be a curve parametrized by the arc length such that @xmath55 . there exists @xmath56 such that for every @xmath57 $ ] the curve is a graph in a local system of coordinates with a first axis aligned on @xmath58 , on a segment @xmath59 $ ] , of a function @xmath60 { \rightarrow}{{\mathbb r}}$ ] with @xmath61 , @xmath62 and such that @xmath63 . let us fix @xmath64 and consider the smallest @xmath65 such that @xmath66 . if @xmath67 does not exist , then the conclusion follows directly . we reproduce the curve @xmath68}$ ] eight times , taking successively a reflection with respect to the line passing trough the point @xmath69 and orthogonal to the tangent at @xmath69 , then the same procedure for the image of @xmath70 and a last reflection in order to close the loop . } $ ] and the ( rescaled ) loop built from the curve , title="fig:",width=226]}$ ] and the ( rescaled ) loop built from the curve , title="fig:",width=226 ] let us denote by @xmath71 the curve which is the boundary of the convex envelope of the loop . from @xcite , we have @xmath72 but @xmath73 } } \|\theta'\|^2 ds \ge \int_c k^2 ds\ ] ] thus @xmath74 hence @xmath75 denoting @xmath76 , we conclude the proof . the assertion of this lemma is of course available on backwards , so that the curve is locally a graph in a neighborhood of each point , over an interval of controlled length . [ bh02 ] let @xmath45 { \rightarrow}{{\mathbb r}}^2 $ ] be a curve parameterized by the arc length such that @xmath77 . if @xmath78 is given and @xmath79 are such that @xmath80 then @xmath81 } \|\theta'\|^2ds \ge \frac{{\varepsilon}^2}{l}.\ ] ] as @xmath82 } \|\theta'\|^2ds < + \infty $ ] , we write @xmath83 } \|\theta'\|^2 ) ^\frac12\ ] ] which gives the result . the idea coming out of the lemma is that if there is an @xmath84-variation of the angle , the elastic energy on that section of the curve is at least a constant times @xmath85 , the constant depending on the global length of the curve . let @xmath39 be a ball of radius @xmath41 . [ bhe20 ] let @xmath45 { \rightarrow}{{\mathbb r}}^2 $ ] be a smooth loop parameterized by the arc length such that @xmath77 and @xmath86 ) { \subseteq}b_r$ ] . then @xmath87 denoting @xmath88 , we have @xmath89 but @xmath90 . therefore , the conclusion of the lemma follows from the cauchy - schwarz inequality @xmath91 assume that a simply connected open set @xmath6 is bounded by a loop @xmath0 which does not have self intersections on @xmath92 . we shall call this piece of curve , a free branch . let us first give the optimality conditions satisfied by any free branch of an optimal domain . on such a free branch , we can perform any ( small and compactly supported , smooth ) perturbation . [ opti ] let @xmath0 be any free branch of a minimizer @xmath6 of the energy @xmath93 . then @xmath94 is @xmath95 on @xmath0 and satisfies : 1 . @xmath96 2 . @xmath97 , for some constant @xmath71 3 . @xmath98 , such that @xmath99 : @xmath100 , for some constant @xmath71 4 . @xmath98 , such that @xmath99 : @xmath101 where @xmath102 is a normal vector to @xmath0 . the point @xmath103 in ( b3 ) , ( b4 ) is the same ( see the proof below ) . the constant @xmath71 in ( b2 ) , ( b3 ) is also the same . to see that , take a point @xmath104 on @xmath0 where the curvature @xmath10 is maximum . if this point does not exist , just extend the curve with the same ode . then , according to ( b3 ) , @xmath105 is also maximum and the normal derivative of the boundary at this point is @xmath106 . therefore ( b4 ) yields @xmath107 and plugging into ( b3 ) gives ( b2 ) , because @xmath108 at this point , with the same constant . the @xmath95 regularity of @xmath2 ( and @xmath109 ) comes from a bootstrap argument and equation ( [ opti2 ] ) below . the first condition ( b1 ) comes from the classical _ shape derivative _ of the elastic energy ( under small perturbation of the boundary driven by some smooth vector field @xmath110 ) , see @xcite for more details on the shape derivative . following e.g. the appendix in @xcite , we see that it is given by @xmath111 while the derivative of the area is classically @xmath112 condition ( b1 ) follows since the derivative of @xmath113 must vanish for any @xmath114 . we obtain condition ( b2 ) multiplying ( b1 ) by @xmath115 and integrating . to get condition ( b3 ) , we use another expression of the elastic energy and the area . namely , with the parametrization with the angle @xmath46 we have ( see @xcite for more details ) : @xmath116 where @xmath117 is the triangle @xmath118 . we note @xmath119 for @xmath12 . thus we are led to minimize the sum @xmath120 with the following constraints ( the starting and the ending point of the branch @xmath0 are fixed ) . @xmath121 the derivative of @xmath122 is ( for a perturbation @xmath123 compactly supported ) @xmath124 while the derivative of @xmath125 is given by @xmath126 using ( [ opti0 ] ) and fubini , we can write @xmath127 therefore , the optimality condition for the constrained problem reads : there exists lagrange multipliers @xmath128 such that , for any @xmath123 : @xmath129 which implies ( thanks to @xmath130 ) @xmath131 by integration , we get ( b3 ) setting @xmath132 . at last , differentiating twice ( b3 ) we get @xmath133 ( where @xmath134 is the tangent vector ) and @xmath135 . using ( b1 ) we see that @xmath136 , so ( b4 ) holds where @xmath137 . since @xmath10 is a solution of the ode ( b1 ) , and therefore can be written with elliptic functions , it can only vanish on isolated points and thus ( b4 ) holds everywhere by continuity of both members . in the following lemma , we assume that the simply connected open set @xmath6 is a minimizer of the energy @xmath93 . [ be03 ] any free branch of a minimizer @xmath6 has a length @xmath119 uniformly bounded by @xmath138 we work with a free branch of @xmath0 on @xmath139 and use the optimality conditions above . we also know that the elastic energy of this branch is less than the total energy of the best disk @xmath140 , so that @xmath141 we consider two cases . assume first that @xmath142 on this branch ( @xmath71 is defined above in ( b2 ) , ( b3 ) ) . then we know from ( ode3 ) in the appendix that @xmath143 then , from ( b3 ) @xmath144 hence the arc is contained in the disc centered at @xmath103 with radius @xmath145 . on the other hand , if we put the origin at @xmath103 @xmath146 but @xmath147 and @xmath148 while by cauchy - schwarz and @xmath149 therefore , @xmath119 satisfies @xmath150 which implies ( as soon as @xmath142 ) @xmath151 second case : @xmath152 for this branch . in this case we have from ( ode3 ) in the appendix @xmath153 @xmath154 we decompose the interval @xmath155 in three parts ( some could be empty ) , @xmath156 where @xmath157 @xmath158 @xmath159 and we are going to prove that the length of each part is uniformly bounded , by a controlled constant . first of all , we have seen that the integral of @xmath160 on a period satisfies ( see ( ode4 ) in the appendix ) @xmath161 following , this implies that we can not have more than @xmath162 periods on each free branch . we begin with @xmath163 . obviously @xmath164 therefore @xmath165 , we consider one of its connected components , say @xmath166 . since @xmath167 and @xmath168 , we can not have any local minimum or local maximum of @xmath10 in @xmath169 according to ( ode2 ) form the appendix . therefore , @xmath10 is either increasing from @xmath170 to @xmath171 , or decreasing from @xmath170 to @xmath171 . moreover , there are at most @xmath172 such connected components because there are at most @xmath162 periods of @xmath10 . let us consider the case of @xmath10 increasing from @xmath170 to @xmath171 , the other one being similar . we have @xmath173 . by ( b1 ) , @xmath174 on @xmath166 , so that @xmath10 is convex . therefore @xmath175 which implies @xmath176 now @xmath177 and @xmath178 thus @xmath179 or @xmath180 . since , there are at most @xmath172 such intervals , we have @xmath181 . the set @xmath182 is not empty only when @xmath183 and @xmath184 . the set @xmath182 is composed of connected components @xmath185 $ ] such that @xmath186 or is included in such connected components . since we want to estimate from above the length of @xmath182 , it suffices to look for the length of these connected components . there are at most @xmath162 of these ( identical ) components and @xmath187 by symmetry . now , the elastic energy of such a component satisfies @xmath188 we denote @xmath189 the length of this component . by convexity , on @xmath190 we have @xmath191 thus @xmath192 now , for @xmath152 we have ( see ( ode3 ) in the appendix ) @xmath193 and @xmath194 . therefore @xmath195 and the total length of @xmath196 in this section we prove the existence of a best _ drop _ minimizing the sum of the elastic energy and the area enclosed . we introduce the class of admissible _ jordan drops _ consisting of simply connected open sets @xmath6 bounded by a jordan curve @xmath0 of finite length , which satisfies @xmath197 where @xmath198 is the length of @xmath0 . a drop will be denoted @xmath199 , @xmath4 being the open set enclosed by the jordan curve @xmath0 ( all jordan curves are oriented in the positive sense ) . for some @xmath37 , we consider the problem @xmath200 note that by a similar argument as in lemma [ bhe20 ] , the length of jordan drop @xmath0 can not exceed @xmath201 . indeed , the same argument works for the drop , if the singularity lies at the origin , we have @xmath202 since the diameter of the drop is less than @xmath203 . here is the main result . [ bh03 ] problem has at least one solution . with no assumptions on the radius @xmath41 , it could be possible that the optimal drop @xmath204 touches the boundary of the ball but it may not have self intersections . for simplicity of the notation , the ball @xmath39 will be denoted @xmath140 and the area of @xmath6 will be denoted by @xmath205 . we start with the following . let @xmath204 be a drop contained in @xmath140 . if for some @xmath78 there exists @xmath206 with @xmath207 then there exists a new drop @xmath208 in @xmath140 such that @xmath209 [ proof of the lemma ] assume @xmath9 and @xmath17 satisfy the hypotheses . then , from continuity of @xmath46 , there exists @xmath210 such that @xmath211 moreover , there exists @xmath212 such that @xmath213 and for every @xmath214 @xmath215 indeed , we define @xmath216 and @xmath217 then we notice that the curve @xmath218}$ ] is a graph in the direction @xmath219 , otherwise it would contradict the choice of @xmath220 and @xmath221 . , width=302 ] setting the orientation of the curve in the trigonometric sense , we are in configuration similar to figure [ bh07 ] . } $ ] in the direction @xmath219,title="fig:",width=302 ] } $ ] in the direction @xmath219,title="fig:",width=302 ] using the graph property , we can translate continuously the piece of the curve @xmath222}$ ] in a parallel way in the direction @xmath219 until this piece touches again @xmath0 . we denote @xmath223 $ ] and @xmath224{\setminus}[s',t']$ ] the couples of touching points . we denote @xmath67 , respectively @xmath225 , the minimal and maximal values of @xmath226 . then , one of the curves starting with @xmath225 and ending in @xmath227 , or starting in @xmath228 and ending in @xmath67 is a drop . precisely , it is the one which does not contain the point @xmath229 . without loosing generality we can assume it is the curve @xmath230 and rename the point @xmath231 and denote this curve @xmath232 . we notice that @xmath233 can not touch any the piece of curve @xmath234}$ ] . if there would be a contact point , this contact is generated by the translation of @xmath235}$ ] and has to be precisely @xmath236 . but in this case , @xmath237 lies in the interval @xmath238 $ ] , so the curve starting at @xmath237 and ending at @xmath239 is a drop , which does not touch the piece of curve @xmath240}$ ] . in this way , we built a new drop @xmath208 , which encloses a domain contained in @xmath6 and , in view of lemma [ bh02 ] has an elastic energy smaller by at least an increment of @xmath241 . [ proof of the theorem [ bh03 ] ( continuation ) ] coming back to the proof of theorem [ bh03 ] , les us consider @xmath242 be a minimizing sequence of drops . we may assume that @xmath243 , @xmath244 and @xmath245 are convergent . assume that for every @xmath246 we have @xmath247 . in order to work on a fixed sobolev space @xmath248 , we assume that @xmath249 is formally extended by the constant @xmath250 on @xmath251 $ ] . up to a subsequence , we can assume that @xmath249 converges uniformly on @xmath252 $ ] to some function @xmath46 . we define the limit curve @xmath0 in the following way : @xmath253 and @xmath254{\rightarrow}{{\mathbb r}}^2 $ ] , @xmath255 , where @xmath256 . let us fix @xmath78 . then , from the previous lemma , for every @xmath257 and @xmath246 large enough we have @xmath258 indeed , otherwise we would replace @xmath259 by @xmath260 decreasing the energy by a fixed increment @xmath261 , where @xmath262 is a bound of the lengths . this is in contradiction with the minimality of the sequence . in particular , passing to the limit we get that for every @xmath78 and for every @xmath257 @xmath263 since @xmath84 is arbitrary , we get @xmath264 from the compactness of the class of closed subsets of @xmath265 endowed with the hausdorff metric , and the embedding of @xmath248 into @xmath266 $ ] , we may assume that for some open set @xmath267 @xmath268 and the convergence of @xmath249 leads to @xmath269 ) { \stackrel}{h}{{\longrightarrow } } { \gamma}([0,l_{\gamma}]).\ ] ] we refer to @xcite or @xcite for precise properties of the hausdorff convergence . we know that in general @xmath270 , so that @xmath271 . nevertheless , in our situation the perimeters being uniformly bounded , we get @xmath272 in @xmath273 . moreover , @xmath274)$ ] and @xmath6 is simply connected ( i.e. any loop contained in @xmath6 is homotopic to a point in @xmath6 ) , but not necessarily connected . the curve @xmath0 is possibly self - intersecting , but not crossing , i.e. at every self - interesting point , the tangent line is the same , while looking locally around the point , the pieces of curve passing through it are ( in view of lemma [ bh04 ] ) graphs of functions . from the simple connectedness hypothesis , these functions are necessarily ordered . from lemma [ bh02 ] and the fact that the elastic energy is finite , the number of pieces of curve passing through the touching point is uniformly finite . the situation displayed in figure [ fig6c ] may occur . we shall prove that @xmath0 can not have self intersection points , other that the type above . the key ingredients are the local representation of the curve as a graph and inequality . we shall analyze the different contact types between two pieces of @xmath0 . since the curves are graphs on an interval @xmath275 $ ] , and the representing functions are ordered , we shall look to the orientation of each piece of curve . * case 1 . opposite orientation , not disconnecting . * two branches of @xmath0 touching at some point @xmath276 , are represented as graphs of the functions @xmath277 , on @xmath275 $ ] . we assume that @xmath278 and choose the couple @xmath279 such that for some @xmath78 we have @xmath280 otherwise we change the contact point . this inequality would imply the existence of points @xmath281 and @xmath282 such that @xmath283 , which is in contradiction with , so that this situation can not occur . * case 2 . contact of two branches of the same orientation . * from the simple connectedness , this situation implies that the touching point @xmath48 belongs to at least three branches , in particular between the graphs of @xmath284 and @xmath285 , there is a graph corresponding to piece of curve with opposite orientation . there are two possibilities : either this new contact corresponds to a point @xmath286 or to @xmath287 . the first situation is in fact the case 1 between the contact points @xmath9 and @xmath221 . the second situation leads also to case 1 , but for the contact points @xmath220 and @xmath221 , so we conclude that the second case can not hold . * opposite orientation , disconnecting . * this is the only remaining possibility for self - intersections . there may be several contact points , but every contact point is simple , otherwise we would fall in case 2 . so let us denote @xmath288 the couple of parameters corresponding to the contact points . because of the simple connectedness and of the absence of contact poins as in cases 1 and 2 , we have that if @xmath289 then @xmath290 . consequently , we can identify the contact point @xmath291 such that between @xmath239 and @xmath237 there is no other contact , by setting @xmath292 and @xmath293 . of course , @xmath294 and @xmath237 can not collapse . indeed , in view of lemma [ bh02 ] applied to @xmath295}$ ] if collapse occurs then the elastic energy would blow up . so @xmath296}$ ] is a jordan curve for which all the area enclosed is part of @xmath4 , otherwise , because of the simple connectedness , a branch of the curve must pass through the contact point , bringing it to the case 2 . so @xmath296}$ ] is a drop , with lower elastic energy than @xmath0 and enclosing a surface less than or equal to @xmath297 . this means that @xmath296}$ ] is a solution for problem . [ bhe02 ] there exists @xmath298 , such that if the radius @xmath41 of the ball @xmath39 in theorem [ bh03 ] satisfies @xmath299 , then there exists a translation of the optimal drop which does not touch the boundary of @xmath39 . the proof relies on lemma [ be03 ] . let @xmath300 ( the value of @xmath298 will be precised at the end of the proof ) . assume that @xmath301 is an optimal drop for problem which touches the boundary , such that there is no translation moving the drop at positive distance from the boundary . this means that the touching points between @xmath302 and @xmath39 are distributed in such a way that they do not fit in an arc of length less than @xmath303 . using lemma [ be03 ] , if @xmath41 is large , e.g. @xmath304 , then the center of @xmath39 has to be inside @xmath305 together with a disc of radius @xmath306 . indeed , the longest piece of curve between two contact points or between a contact point and the singularity has a length less than @xmath307 . consequently , the energy of @xmath301 is larger than the area of the disc of radius @xmath306 : @xmath308 , in contradiction with its optimality . taking @xmath309 , the lemma is proved . [ bhe01 ] there exists a unique optimal drop @xmath310 which minimizes the energy @xmath311 among all drops in @xmath3 . this one is fully characterized by the optimality conditions ( b1)-(b2 ) with a unique constant @xmath71 which can be determined . + moreover @xmath312.\ ] ] figure [ the optimal drop ] gives the representation of the optimal drop . the proof of existence follows from theorem [ bh03 ] and lemma [ bhe02 ] . the optimality conditions ( b1)-(b4 ) can be written on the whole @xmath0 ( except at the singularity ) according to theorem [ opti ] . we start for @xmath313 at the origin which is the singular point with an horizontal tangent ( @xmath314 ) . by ( b4 ) and starshaped property , the point @xmath103 is necessarily on the @xmath315-axis , the curvature @xmath2 is negative for @xmath316 small and @xmath317 when @xmath318 . the function @xmath2 is periodic but we will prove below ( see the end of the proof ) that we have only one period for the optimal drop and the curve is symmetric around the @xmath315-axis . therefore to characterize the optimal drop , we can proceed in the following way : for any constant @xmath183 , we solve the ode @xmath319 which has a unique solution . let us denote by @xmath320 the value where @xmath10 is maximum with @xmath321 ( respectively @xmath322 and @xmath323 for the minimum ) . the point @xmath104 of abscissa @xmath320 is necessarily on the @xmath315-axis and its tangent is vertical . thus , we look for the value of @xmath71 for which @xmath324 . we claim that conversely , if we find a value of @xmath71 for which @xmath325 , then we have found the optimal drop . indeed , since it satisfies the optimality conditions , it suffices to check that the curve we obtain by @xmath326 and @xmath327 with @xmath328 is an admissible drop . since @xmath104 is the point where the curvature is maximum , according to ( b3 ) , it is the point on @xmath0 which is the farthest to @xmath103 . but since the tangent is vertical at this point it is necessarily on the @xmath315-axis : @xmath329 and the total length of the curve is @xmath330 . now , since @xmath10 is symmetric with respect to @xmath320 ( see ( ode1 ) in the appendix ) , @xmath331 which provides after integration : @xmath332 . this identity gives @xmath333 and @xmath334 which shows that the curve @xmath0 is a drop . thus to prove uniqueness of the optimal drop , we need to prove that we can find only one @xmath183 for which @xmath335 . let us write @xmath336 where we used the symmetry of @xmath10 with respect to @xmath322 , see ( ode1 ) . this symmetry also shows that @xmath337 . we are going to prove uniqueness of @xmath71 ( and therefore of the optimal drop ) by proving that the function @xmath338 is strictly decreasing . let us perform the change of variable @xmath339 in each above integral . it comes , using ( b2 ) to express @xmath115 : @xmath340 now to compute the derivative of the first integral @xmath341 with respect to @xmath71 , we make the change of variable @xmath342 , it comes @xmath343 we compute the derivative of @xmath344 using @xmath345 ( see ( ode3 ) in the appendix ) and an easy computation gives @xmath346 which is clearly negative . in the same way , we get for the second integral @xmath347 : @xmath348 which is also negative , proving the uniqueness of a solution @xmath71 for the equation @xmath349 . let us remark that a simple computation yields @xmath350 while the limit of @xmath351 when @xmath71 goes to @xmath352 is @xmath353 confirming that there exists a solution to our problem . let us estimate from below the energy of the optimal drop . denote by @xmath354 the first positive zero of @xmath10 , we recall that @xmath322 is the first minimum of @xmath10 and @xmath355 , @xmath320 the first maximum of @xmath10 and @xmath356 . from ( b2 ) @xmath357 and @xmath358 are the real roots of the polynomial ( which is concave ) @xmath359 the maximum of @xmath360 is at @xmath361 and @xmath362 . we have @xmath363 ( @xmath357 can not be nonnegative , otherwise the set @xmath305 would be convex ) . moreover , when @xmath71 increases , @xmath364 is increasing while @xmath365 is decreasing ( with increasing absolute value @xmath366 ) , because we translate the curve @xmath367 up ) . if we denote @xmath368 and @xmath369 the sum and the product of those two roots , classical elimination and relation between roots provide @xmath370 while the two complex roots @xmath371 satisfy @xmath372 , @xmath373 . since @xmath374 and @xmath183 , the last equation gives @xmath375 . let us come back to the computation of the elastic energy of the optimal drop @xmath301 @xmath376 now @xmath377 and @xmath10 is increasing from @xmath322 to @xmath320 ( since @xmath115 can only vanish at zeroes of @xmath378 , which only correspond to maxima @xmath358 and minima @xmath357 ) . we perform the change of variable @xmath379 on this interval @xmath380 . therefore @xmath381 we want to find a lower bound of this integral . for this purpose , we write ( following ) @xmath382 now , the parabola @xmath383 is symmetric with respect to @xmath384 , an since @xmath385 , the maximum of @xmath386 on the interval @xmath387 $ ] is equal to @xmath388 where we have used for the last equality . thus @xmath389 this last integral can be computed explicitly and gives @xmath390 we have @xmath391 and gives @xmath392 it remains to get a bound for the quantity @xmath393 which depends only on @xmath71 . we discuss two cases . * case a. * if @xmath394 , @xmath395 . both mappings @xmath396 , @xmath397 are increasing , thus @xmath398 . we study @xmath399 @xmath400 we get @xmath401 while from @xmath402 and @xmath403 , we get @xmath404 it follows that @xmath405 * case b. * in the case @xmath406 , we use @xmath407 , @xmath408 and @xmath409 to get @xmath410 so in any case , @xmath411 . it follows from that @xmath412 now , integrating ( b4 ) on the curve , we get @xmath413 . therefore @xmath414 let us now conclude by proving that the optimal drop has only one period of the function @xmath2 . the estimate ( [ bhe16 ] ) we get is actually true on any possible period . therefore , if we have a solution @xmath415 with at least two periods , we would have @xmath416 , therefore like in ( [ bhe17 ] ) its total energy would satisfy @xmath417 . now , proceeding in a similar way as we did for the estimate from below , we can get ( details omitted ) an estimate from above for an optimal drop with only one period which is @xmath418 ( the exact value is @xmath419 ) therefore , any critical point with more than one period can not be optimal . with the notation settled in sections [ prel ] and [ drops ] we return to problem , and write @xmath420 first of all we recall that among all circles , the optimal one has the radius @xmath421 . let us consider @xmath422 and solve the problem @xmath423 using the same arguments as in section [ drops ] , a minimizing sequence will converge to a couple @xmath424 . two possibilities occur . assume first that there are self intersections . in this case the limiting couple @xmath424 contains at least two drops , as in case 3 of theorem [ bh03 ] . following theorem [ bhe01 ] , this configuration can not be optimal since the energy of @xmath6 is larger than the double of the optimal energy of a drop , so it is excluded . the second situation is that @xmath424 does not have self - intersections . since the radius is large enough , for a suitable translation the loop does not touch the boundary of the ball , as in lemma [ bhe02 ] . moreover , in this case the optimality conditions @xmath425 can be written on the full boundary . this condition recalls the result of ben andrews ( see theorem 1.5 in @xcite ) which , under the hypothesis of positive curvature would allow directly to conclude that the curve is a circle with a radius equal to @xmath426 ( by direct computation ) . as the curvature is not known to be positive , for our conclusion we shall use again the optimality conditions . actually , andrews s result does not hold true for non convex curves . indeed , figure [ 23fig ] ( which has been obtained using the optimality conditions ) shows a curve which satisfies @xmath425 on the whole boundary . if the curvature is not constant , we can assert that @xmath6 is star shaped and the structure of @xmath0 is a union of periods consisting of two branches @xmath427 and @xmath428 , where @xmath429{\rightarrow}{{\mathbb r}}^2 $ ] is a branch of the curve with increasing curvature such that @xmath430 and @xmath431{\rightarrow}{{\mathbb r}}^2 $ ] is a congruent branch with decreasing curvature from @xmath358 to @xmath357 . following and ( ode4 ) , @xmath0 consists of one , two or three periods @xmath432 ( as explained in the proof of theorem [ bhe01 ] . from the optimality conditions @xmath433 one can eliminate any of those three configurations , since their energy is much larger than the one of the ball . indeed , in the case of two or three periods , a couple @xmath432 has a cap @xmath434 , where @xmath435 is chosen such that @xmath436 is orthogonal to the segment @xmath437 . as on figure [ 23fig ] , we can cut and reflect along the line @xmath438 to get a new domain which smaller area and smaller elastic energy . if there is only one period , the argument is similar since we can center - symmetrize the branch from @xmath439 to @xmath440 , where @xmath435 is chosen such that the normal at @xmath439 is parallel to the segment @xmath437 ( see figure [ 1fig ] ) . both previous constructions are admissible as a consequence of the optimality conditions @xmath433 . in this section , we give several properties of the following ode in nonstandard form @xmath441 where @xmath442 is a constant . this ode is issued from the optimality conditions on a free branch of a minimizer for our problem , see theorem [ opti ] . we also refer the reader to reference @xcite for related analysis . 1 . the solution of the ode is periodic ( the period is denoted by @xmath117 ) , symmetric with respect to its minimum or maximum . [ ode2 ] the only local minima ( maxima ) are actually global minima ( maxima , respectively ) and correspond to @xmath447 ( @xmath448 , respectively ) , and @xmath10 is monotone between these two values . [ ode3 ] the mapping @xmath449 is increasing and its range is from @xmath450 to @xmath352 , while the mapping @xmath451 is decreasing and its range is from @xmath452 to @xmath450 . moreover , @xmath453 when @xmath183 , @xmath454 , @xmath455 , @xmath456 . as well , @xmath457 for @xmath458 . the integral @xmath459 on one period is estimated from below @xmath460 the proof of ( ode4 ) : we have already proved this inequality in section [ drops ] , when @xmath467 . it remains the case @xmath468 . in this case , we have @xmath469 and @xmath470 , so @xmath471 , and the result follows in the same way . this work has been initiated during a stay of the two authors in the isaac newton institute , cambridge in march 2014 during the programme `` free boundary problems and related topics '' . the authors are very grateful to the institute for the very good and stimulating atmosphere here . the results of this paper have been announced in several conferences ( e.g. petropolis in august 2014 , linz in october 2014 ) . while completing this manuscript , we learnt from bernd kawohl that he , with vincenzo ferone and carlo nitsch , proved our theorem [ bhe05 ] by a different way . their manuscript is available on arxiv , see reference @xcite .	for a smooth curve @xmath0 , we define its elastic energy as @xmath1 where @xmath2 is the curvature . the main purpose of the paper is to prove that among all smooth , simply connected , bounded open sets of prescribed area in @xmath3 , the disc has the boundary with the least elastic energy . in other words , for any bounded simply connected domain @xmath4 , the following isoperimetric inequality holds : @xmath5 . the analysis relies on the minimization of the elastic energy of drops enclosing a prescribed area , for which we give as well an analytic answer .
it is well established that the bright - rimmed globules found in regions are often sites of star formation . reipurth ( 1983 ) first showed that these objects do form stars and subsequent work based on iras data by sugitani et al . ( 1991 ) confirmed that they are indeed active `` stellar factories '' which produce intermediate - mass ( herbig aebe ) stars . these condensations are local clumps which emerge from the expanding nebula or form from the fragmentation of the dense molecular layer surrounding the ionized gas . the various theoretical works ( bertoldi , 1989 ; bertold & mckee , 1990 ; lefloch & lazareff , 1994 ) and and the numerous observational studies on bright - rimmed globules ( see e.g. cernicharo et al . 1992 ; lefloch & lazareff , 1995 ) have enabled to draw the following evolutionary picture , summarized in fig . 1 . as a globule of neutral gas is exposed to the ionizing field of the exciting star(s ) of the nebula ( region i in fig . 1 ) , an ionization front ( if ) forms at the surface of the condensation . for standard ionization conditions , the pressure of the surface ionized gas is much higher than in the nebular gas and in the molecular globule . as a consequence , the photoionized gas expands into the hii region , inducing the formation of a photoionized envelope around the globule . it is the ionization front and the photoionized envelope which are detected in the optical as a bright rim ( region ii ) . the incident fuv field drives the formation of a photon - dominated region ( hereafter pdr , region iii ) while a shock front , driven by the surface overpressure , propagates towards the dense molecular core ( region iv ) . observational evidence of this mechanism , also called radiatively - driven implosion ( or rdi ) , has been reported in a few objects by cernicharo et al . ( 1992 ) and lefloch & lazareff ( 1995 ) . progressively , a dense core forms behind the surface , and the globule adjusts its internal structure to balance the pressure of the ionized gas , while the bulk of its mass is photoevaporated . eventually , the globule reaches a quasi steady - state , the `` cometary phase '' , in which the shock front has disappeared . the globule now consists of a small dense `` head '' prolonged by a long tail of diffuse gas . it has long been suggested that the shock front inside the globule could trigger the star formation inside the bright - rimmed globules ( see e.g. reipurth 1983 ; lefloch et al . these objects appear therefore as ideal laboratories to test the scenarios of star formation triggered by an external compression wave . most of the studies led until now were focused onto the molecular core of bright - rimmed globules ( see e.g. lefloch et al . therefore , the physical conditions reigning in the pdr and in the shocked molecular gas are not well characterized and the impact of photoionization on the gravitational collapse is therefore difficult to evaluate . moreover , all the bright - rimmed condensations studied until now are found in relatively old regions , with ages of a few myr . because the condensations are usually found at rather large distances from the ionizing stars , they experience a reduced uv field and it is difficult to discriminate between a star formation induced by radiatively - driven implosion and a spontaneously evolving globule which has already started to form stars by the time it is hit by the ionization front . this is why we have started a systematic multiwavelength study of a young hii region : the trifid nebula . it appears as a small dusty nebula of @xmath12 diameter at an heliocentric distance of @xmath13 ( lynds et al . 1985 ) , with a dynamical age of @xmath14 . the nebula is excited by the o star hd 164492a . in a preliminary work ( cernicharo et al . 1998 , cl98 ) , we reported on the mapping of the thermal dust emission of the nebula at millimeter wavelengths . this mapping revealed the presence of several protostellar cores ( dubbed tc1 to tc4 ) in the shell of dense molecular gas surrounding the region . it was not clear however if the birth of the protostars was triggered or not . for this reason , we undertook a more detailed analysis of the protostellar cores . in a subsequent paper ( lefloch & cernicharo , 2000 ) , we reported on the continuum and molecular line emission around the two most massive protostellar cores : tc3-tc4 . their masses are high , between 60 and @xmath15 . they harbour a class i source and one of the few high - mass class 0 candidates known until now . comparison of their properties with the models of elmegreen & lada ( 1977 ) and whitworth et al . ( 1994 ) allowed to conclude that the formation of tc4 had probably been triggered in the fragmentation of the dense shell surrounding the ionized gas . the molecular properties of tc3 and tc4 are similar to those of the protostellar cores discovered in orion , though at an earlier , `` pre - orion '' , evolutionary stage . we report here on the tc2 , protostellar core , which is associated with a bright - rimmed globule on the southern border of the trifid . tc2 appears to be in a more advanced stage of photoionization than tc3 - 4 : unlike tc3 - 4 which are still embedded cores , tc2 has already emerged from the diffuse molecular gas layers and it exhibits the optical bright rim and the cometary shape typical of photoionized globules . as discussed in this work , tc2 is exposed to a rather strong ionizing field , which drives a shock into the condensation , in agreement with the evolutionary scheme presented above . like tc3 and tc4 , tc2 displays signs of protostellar activity . cl98 reported the presence of the herbig - haro jet hh399 coming out of the head of the globule and propagating into the ionized nebula . it is the best example of globules undergoing at the same time strong photoevaporation and active star formation . it offers a good opportunity to study the quantify the relation both phenomenons hold to each other , and better constrain the role that radiatively - driven implosion could play in the star formation process . in this purpose , we have led a detailed study of the structure and the physical conditions of the globule ( density , temperature , velocity ) . because tc2 lies almost in the plane of the sky , it provides also a good opportunity to study the structure of a typical low - density photon - dominated region , and to confront its properties against the existing models . this work provides the first comprehensive study of the whole gas structure of a bright - rimmed globule , from the ionized surface layers to the cold dense molecular core . the paper is organized as follows . we first derive the structure of the globule : the hii region ( sect . 3 ) ; the bright rim and the photo - evaporated envelope ( sect . 4 ) ; the pdr ( sect . 5 ) ; the dust continuum emission ( sect . 6 ) and the molecular core ( sect . 7 ) . we then discuss the observational evidences of radiatively - driven implosion in tc2 and the implications on the past history of the globule ( sect . 8) . in the following section , we first study the star forming conditions in tc2 and attempt to characterize the protostellar source and the outflowing material , before studying the impact of photo - ionization both on the protostar and the evolution of the globule ( sect the conclusions are presented in sect . the determination of the physical conditions in the various regions in tc2 requires a compared study of the spectral line and continuum emission mapped at various wavelengths . the photoionized envelope at the surface of tc2 was characterized from observing the radio free - free emission with the very large array ( vla ) , following the same approach as lefloch et al . additional constraints on the envelope and the surrounding nebular gas in the trifid were brought by the fine - structure atomic and ionic lines detected in the mid - ir and fir with the infrared space observatory ( iso ; kessler et al . the conditions in the molecular core were derived from the emission of the cold dust and of various molecular tracers at millimeter wavelengths with the iram 30 m telescope . the physical conditions in the pdr were determined mainly from the observations of the fir @xmath4}}$ ] lines at @xmath5 and @xmath6 and the @xmath7}}$ ] line at @xmath8 . the dust properties were derived from the observation of the continuum emission between 5 and @xmath16 with the instruments onboard iso . in all maps presented here the coordinates of all the maps are given in arcsec offsets with respect to the position of hd 164492a : @xmath17 @xmath18 , @xmath19 . the observational data set is summarized in table 1 ; the data described below are shown in figures 2 to 9 and 11 to 16 . the very large array observations at 3.6 cm were made with the array in its highest angular resolution a configuration in 1998 march 13 . we used 1328 + 307 as absolute amplitude calibrator and 1748 - 253 as the phase calibrator . a bootstrapped flux density of [email protected] jy was obtained for 1748 - 253 . the observations were made in both circular polarizations with an effective bandwidth of 100 mhz . the data were edited and calibrated following the standard vla procedures and using the software package aips . the final map at 3.6 cm was made with the aips task imagr , the robust parameter set to 5 , and a gaussian taper of 100 k@xmath21 to the ( u , v ) data to lower the angular resolution . the field of view is approximately @xmath22 ; the synthesized beam has a size of @xmath23 ( hpfw ) and makes a position angle ( p.a . ) of @xmath24 . observations of the 1.25 mm continuum emission in the trifid were carried out in march 1996 and march 1997 at the iram 30m - telescope ( pico veleta , spain ) using the mpifr 19-channel bolometer array . the resolution of the telescope is @xmath25 . the final map was obtained by combining several individual fields centered on the brightest condensations of the nebula . each field was scanned in the horizontal direction by moving the telescope at a speed of @xmath26 per sec ; subsequent scans are displaced by @xmath26 . we used a chopping secondary at 2hz with a throw of 30 to @xmath27 depending on the structure of the region to be mapped . calibration was checked against mars . the weather conditions were good and rather stable during the two observing sessions . the opacity was monitored every hour on average and we found typical zenith opacities between 0.1 and 0.35 . pointing was checked every hour as the source transits at low elevation at pico veleta and corrections were always found to be lower than @xmath28 . the final rms is @xmath29 beam . hence , it is sensitive enough to detect at the @xmath30 level protostellar condensations of @xmath31 at 20 k in one @xmath25 telescope beam ( @xmath32 at the distance of the trifid ) . in order to outline the weak extended dust components in the nebula , we degraded the angular resolution of our map down to @xmath33 ( 0.13 at the distance of the trifid ) by convolving the emission with a gaussian of 11 hpfw . the resulting rms in the map is @xmath34 . all the results quoted in this paper are based on the non - degraded map with 0.09 resolution . we observed in july 1996 and july 1997 the trifid nebula in the millimeter lines of sio , , and cs , with the iram 30m - telescope . the lines were observed with a spatial sampling of @xmath33 : the data are almost nyquist - sampled at 3 mm . we used an autocorrelator as spectrometer , which provided a velocity resolution of @xmath35 in all three bands . the rejections of the receivers were always higher than 10db , and checked against w51d ( mauersberger et al . the trifid was mapped at full sampling in the co @xmath36 , @xmath37 , @xmath38 @xmath37 and @xmath39 @xmath37 lines with the iram 30 m telescope in july 1996 . the autocorrelator provided a kinematic resolution of @xmath40 for all the transitions . additionnal observations of the co transition were carried out at the cso during various observing runs between 1998 and 1999 . the receiver was connected to an aos which provided a kinematic resolution of @xmath41 . all the observed lines , their frequency , the telescope beamwidth and the main - beam efficiencies are summarized in table 2 . we observed the trifid nebula with isocam ( cesarsky et al . 1996 ) onboard iso . an image of the whole nebula was obtained using the broad band lw10 filter ( @xmath42 , @xmath43 ) with a pixel size of 3 . the mid - infrared emission around tc2 was observed in the circular variable filter mode ( cvf ) between 5 and @xmath44 . the pixel size is @xmath45 and the spectral resolution is 40 . the @xmath46 pixel detector covered a total field of view of @xmath47 centered on the globule at @xmath17 @xmath48 , @xmath49 . the size of the point spread function varies between @xmath50 at @xmath51 and @xmath52 at @xmath44 . the data were reduced with the slice software and following the method of miville - deschenes et al . ( 2000 ) . the emission between 2 and @xmath53 was observed with the sws spectrometer onboard iso ( de graauw et al . , 1996 ) in the sw01 mode ( @xmath54 grating scan ) the spectral resolution was 300 . a spectrum was taken right on the head of tc2 ; two additional spectra were taken at a position shifted by + 20 and - 20 in declination , in order to measure the emission towards the bright rim and the hii region and towards the main body of the globule respectively . the positions are marked in fig . we note that there is some overlap between the three beams . in particular the beam centered @xmath55 north of the globule encompasses the northern part of the bright rim . the sws data consists of an `` up '' scan , towards decreasing wavelengths , and a `` down '' scan towards increasing wavelengths . the lines identified and their flux ( obtained after averaging the `` up '' and `` down '' scans ) are given in table 3 . the calibration accuracy varies from @xmath56 at @xmath57 to @xmath58 at @xmath53 ( sws iso handbook , leech et al . because of the high noise , the statistical errors in the line fluxes are _ a priori _ not negligible in front of the systematic errors in the flux calibration . therefore , we have compared the emission in the `` up '' and `` down '' scans in order to estimate the actual uncertainty in the line fluxes . for all the lines detected , _ but the @xmath59}}$ ] line at @xmath60 _ , a very good agreement was found between both measurements . on the contrary , very large variations , up to a factor of 4 , were observed in the flux of the @xmath59}}$ ] line . the emission between 45 and @xmath16 was observed towards the globule with the lws spectrometer onboard iso ( clegg et al . the size of the lws beam is known to vary between @xmath61 and @xmath62 hpfw depending on the detector s band . this effect was taken into account to estimate the line fluxes : the detector beam sizes were taken from the iso handbook for the lws spectrometer ( gry et al . , 2001 ) . the large - scale emission of the nebula was estimated by observing a nearby reference position @xmath63 away of tc2 , at position @xmath64 , @xmath65 ( see fig . 2 ) . the data was reduced using the off the line package ( olp ) version 7 and with the isap package . the fluxes measured are given following the standard flux calibration , assuming a point - like source . the uncertainty in the absolute flux calibration quoted for lws is @xmath66 . however , as discussed in the text , part of the emission in the lws beams arises from the nebular gas , i.e. very extended emission which fills the beam . therefore , we adopt in this case a somewhat more conservative number of @xmath67 . the identified lines and the fluxes are listed in table [ flux_lws ] . the spectrum of the globule was obtained after subtracting the emission of the reference position ( off source ) to the emission measured on source . we proceed first with the analysis of the @xmath68 region as tc2 is embedded in the ionized gas and an estimation of the physical conditions of the ionized bright rim ( see next section ) requires a knowledge of the contribution to the atomic fine structure lines from the hii region . we show in fig . 6 the various atomic and fine - structure atomic lines detected with the lws and sws spectrometers . many of the identified lines exhibit weak variations between the on and off positions . the best example is provided by @xmath7}}$ ] at @xmath8 . on the other hand , the largest variations are observed for the @xmath4}}$ ] lines at 63.3 and @xmath69 . in particular , almost no emission at all is detected in the @xmath6 line at the reference position the @xmath70}}$ ] @xmath71 and @xmath72}}$ ] @xmath73 lines are useful tools to diagnose the electron density in the ionized gas . the ( @xmath70}}$ ] @xmath74 ) and ( @xmath70}}~52$]/@xmath72}}~57 $ ] ) ratios have the advantage of being rather insensitive to the temperature , and mainly trace the electron density @xmath75 in the ionized gas . like for the sws data , the situation is made complicated by the fact that the lws beam encompasses regions with very different physical conditions . in order to gain more insight on the background emission from the @xmath68 region and the pdr associated with the parent molecular cloud , we first consider the reference position . the lws flux scale is based on a point source calibration . in the case of extended source emission , some correcting factors have to be applied to derive the correct fluxes ( see paragraph 4.9.3 in the iso handbook for the lws spectrometer , gry et al . 2001 ) . since the emission at the reference position comes from the @xmath68 region and fills the whole lws beam , the fluxes have been calibrated according to the iso manual prescription . we then obtain @xmath76}}~52/88)= 0.69 $ ] . this corresponds to a mean density @xmath77 , which is consistent with the previous determination from @xmath78}}$ ] . in this density range , the emissivity of the @xmath79 line is : @xmath80 ( dinerstein et al . , 1985 ) . if we assume a standard abundance @xmath81}}/[h^{+}]= 2\times 10^{-4}$ ] , we derive an estimate of the emissivity @xmath82 and the size of the emitting region : @xmath83 . in the density range measured in the region , the emissivity ratio ( @xmath72}}~57/ { \hbox{[o{\small iii}]}}~52 $ ] ) is predicted to be close to 1.6 ( see e.g. lester et al . 1987 ) , whereas the measured line ratio is only 0.88 . since both lines trace physically similar regions ( their critical densities are @xmath84 and @xmath85 respectively ) this discrepancy mainly reflects the different relative elemental abundances . we obtain an elemental abundance ratio ( @xmath72}}/ { \hbox{[o{\small iii}]}})\simeq 0.5 $ ] . this ratio agrees with other determinations in a sample of regions by lester et al . ( 1983 ) who obtained 0.25 - 0.43 . in the sws spectra , the line @xmath86}}$ ] @xmath87 is detected at the three observed positions : towards the head , the bright rim , and the main body of the globule . the fluxes exhibit only weak variations between the three positions . this suggests that the contribution of the @xmath68 region , which fills the sws beams , dominates the emission from the globule . the sws fluxes of the @xmath59}}$ ] @xmath88 line suffer large statistical errors which casts some doubt on the variations measured with the sws : whereas the fluxes in the northen and central positions are similar ( better than 30% ) , there seems to be hardly any emission detected in the southern position . similarly to the @xmath86}}$ ] line , the @xmath78}}$ ] lines at 18.7 and @xmath89 exhibit only weak variations between the three positions observed with sws . in order to determine the origin of the @xmath78}}$ ] emission , we first estimate the electron density from the @xmath78}}$ ] @xmath90 ratio . in the case of extended sources , a correcting factor has to be applied to the sws flux obtained with the reduction pipeline in order to derive the correct source flux ( salama 2000 ) . it is difficult to estimate the interstellar absorption on the line of sight . since the emission arises from the region close to the _ optical _ bright rim , we believe the interstellar extinction should be rather low and we will neglect the latter in what follows . we estimate an emissivity ratio @xmath91 for the northern spectrum . this ratio allows only to set an upper limit on the electron density ( rubin 1989 ) : @xmath92 . if the contribution of the bright rim were dominant , one would obtain a much larger emissivity ratio , of @xmath93 , as expected for the electron densities obtained with the vla ( @xmath94 ) . this agrees with previous determinations of the _ nebular _ gas density by lynds et al . ( 1985 ) and rosado et al . ( 1999 ) from the optical [ sii ] @xmath95 lines and with our determination from the atomic fine structure lines detected with the lws ( see above ) . we conclude that the main contribution to the @xmath78}}$ ] flux arises from the @xmath68 region . simpson et al . ( 1995 ) showed that it is possible to use the ionization fraction ratios of heavy elements such as s , o or n in order to constrain the effective stellar temperature @xmath96 by comparison with models of @xmath68 regions . in particular rubin ( 1985 ) calculated the integrated fluxes of numerous ionic lines of such heavy elements for a wide range of conditions including the neutral gas density and the effective stellar temperature . these models predict that the ionization fraction ratio @xmath97 spans 3 orders of magnitude when the effective temperature of the exciting star ranges from 3 to @xmath98 ( see e.g. fig . 8 in simpson et al . 1995 ) . the method to derive the ionization fraction ratio is fully discussed in simpson et al . we first estimate the ionic ratio @xmath99 from the @xmath70}}$ ] lines at 52 and @xmath100 and the @xmath78}}$ ] @xmath101 line . the emissivities for @xmath70}}$ ] and @xmath78}}$ ] are taken from dinerstein et al . ( 1985 ) and rubin et al . ( 1994 ) respectively . we adopted @xmath102 as elemental abundance , which is the value found in the orion nebula ( rubin et al . we obtain @xmath103 and 0.09 towards the off and on position respectively . the variations of @xmath97 versus the effective stellar temperature @xmath104 for ionizing luminosities in the range @xmath105 , indicates @xmath106 ( fig . 8 in simpson et al . 1995 ) . an effective temperature of 38000 k , typical of an o7 v star , requires a ratio @xmath97 of a few , i.e. 10 to 20 times larger . there is a priori a restriction in directly applying the models of rubin since the density of the trifid ( 100 ) is somewhat less than the density considered in the calculations . however , in the range of temperatures and ly - continuum luminosities considered , the ratio of the ionic lines is almost independent of the density in the range @xmath107 ( rubin , 1994 ) . therefore , the effective stellar temperature of the central star exciting the trifid is @xmath108 . interestingly , this value is typical of an o7.5 iii star , a spectral type proposed by walborn ( 1973 ) for hd 164492a , and not an o7 v star ( the classification proposed by lynds et al . 1985 ) . in both cases , the physical properties of the exciting star are very similar , and , in particular the number of lyman - continuum photons varies by @xmath109 , reaching @xmath110 for an 07.5 iii . a simple estimate of the fuv field intensity can be obtained for spherical , ionization - bounded , regions excited by standard o stars and in pressure equilibrium with the parent cloud ( spitzer , 1978 ) . the fuv field intensity @xmath111 is related to the hydrogen nuclei density @xmath112 via @xmath113 ( spitzer , 1978 ) . here , @xmath114 in an region in pressure equilibrium with the ambient cloud , @xmath115 is the mean electron density in the @xmath68 region ( chaisson & willson , 1975 ) and @xmath111 is measured in units of the average interstellar field flux of @xmath116 ( habing , 1968 ) . one derives a fuv field @xmath117 at the border of the region . a more accurate value is obtained by considering the infrared dust continuum emission . if one makes the simple hypothesis that in the pdr the dust is heated only by the incident fuv field , then , at first order , the dust temperature is related to @xmath111 via @xmath118 ( hollenbach , tielens & takahashi , 1991 ) . the temperature estimated for the warm dust layer in tc2 ( @xmath119 ; see sect . 6 ) yields a fuv field intensity @xmath120 . this value is in good agreement with the ( crude ) previous estimate . it is also consistent with the @xmath121}}~145/63\mu m)$ ] and @xmath121}}~63/{\hbox{[c{\small ii}]}}~158\mu m)$ ] line ratios measured towards tc2 , as we discuss in the following paragraphs . we will adopt @xmath122 in the rest of the article . the bright - rimmed globule tc2 is located on the southern border of the region , at approximately @xmath123 of the exciting star of the nebula : hd 164492a . an optical @xmath124 image of the nebula ( taken from cl98 ) , is shown fig . 2 . the tc2 region is marked by a white rectangle . the globule exhibits the typical bright rim of that class of photoionized condensations ( see also the magnified view in fig . 4 ) ; it indicates that the condensation lies almost in the plane of the sky , coplanar with hd 164492a , on the southern border of the nebula . hence , the projected distance must be close to the physical distance between both objects . we determined the mean radius @xmath125 of tc2 , based on the optical image : @xmath126 . some weak extended [ sii ] emission is detected south of the bright rim ( fig . 2 ) . the brightness is much higher at the surface of the globule than in the dust lanes of the nebula . this indicates the presence of ionized gas between the globule and the observer and gives direct evidence of the physical association between tc2 and the trifid . the kinematical analysis of the molecular gas shows that the globule belongs to a large - scale feature connected to the dust lanes on the front side , and does not lie on the back side of the nebula ( lefloch et al . therefore tc2 is `` bathing '' in the nebular ionized gas . the peak of brightness is found , as expected , in the direction of the ionizing star of the nebula hd 164492a . however , the major axis of the globule does not point towards the star but in the north direction . this difference in the orientations might reflect some inhomogeneties in the gas distribution or the influence of other stars which formed at the same time as hd 164492a : the energy released would have affected the expansion of the region . indeed , lefloch et al . ( 2001 ) have reported the presence of several young high- and intermediate - mass stars in the region , some of which have been identified as strong x - ray emitters ( rho et al . 2001 ) . on the left side of the bright rim , close to the offset position @xmath127 , the hh 399 jet powered by tc2 ( figs . 4 - 5b ) is detected as a conspicuous bright filament which seems to penetrate the surface and makes a p.a . of @xmath128 with respect to north . at the basis of the jet , a `` dark filament '' is detected in absorption against the bright rim and emerges from the globule . this filament is probably related to some inhomogeneities or some instabilities at the surface of the globule . depending on the spectral type adopted ( see sect . 3.2 ) the lyman - continuum ( ly - c ) luminosity of the star lies in the range @xmath129 . moreover , a fraction of the ionizing photons is consumed in radiative recombinations within the ionized gas or absorbed by the dust between the exciting star and the globule . the dusty aspect of the nebula in the optical images suggests that the absorption of the ionizing radiation might play a non - negligible role . cl98 presented a map of the free - free emission at 20 cm , obtained with the vla at @xmath130 resolution . we show in fig . 5a an excerpt of this map , centered on the globule . the map reveals mainly the large - scale emission from the nebula . the photoionized layer is also detected as a region of maximum flux on the western side of the globule . however , the angular resolution is not high enough to discriminate the emission of the ionization front and the ionized gas close to the surface from the nebular emission , associated with lower - density gas . in this section , we first evaluate the electron density and the intensity of the ionizing field at the surface of the globule from the free - free emission observed at the vla . we then use the various fine - structure ionic lines detected with iso to determine the geometry of the illumination with respect to the globule . in order to characterize the ionizing conditions at the surface of tc2 ( ionizing field intensity , electron density ) , we have studied the emission of the bright rim with the vla at high angular resolution ( @xmath131 ) . the map of the 3.6 cm continuum emission is shown in fig . the bright rim is clearly detected ; it is marginally resolved in its transverse direction with a thickness @xmath132 . integrating over the bright rim , we derive a total flux of @xmath133 . the photo - ionization of the surface layers induces the formation of a shell of ionized gas around the globule , which is detected in the optical as the bright rim . the radiative recombinations which take place in the shell shield the molecular material of the globule from the incoming ionizing flux . previous theoretical work ( bertoldi , 1989 ; lefloch & lazareff 1994 ) showed that the ionized shell can be modeled as a shell of uniform density @xmath75 and thickness @xmath134 : @xmath75 is the ionized gas density at the surface of the globule , @xmath125 is the radius of the globule , essentially the radius of the cross - section to the ionizing field , and @xmath135 is a geometrical factor of the order 0.1 - 0.2 . for standard ionization conditions most of the impinging ionizing photons are consumed in the shell . the free - free radiation flux across the ionized shell detected in an area of size @xmath136 can be expressed as a function of the electronic density and temperature , and @xmath136 ( lefloch et al . 1997 ) : @xmath137 we assume in this formula that the size of the emitting region along the line of sight is of the same order as the thickness of the ionized shell , i.e. we neglect any inclination effect . we adopted the mean electron temperature in the nebula determined by chaisson & willson ( 1975 ) from centimeter continuum measurements at @xmath138 resolution : @xmath139 . we measure a total flux of @xmath140 in an area of 74 arcsec@xmath141 , defined by the contour level at @xmath142 . assuming that the bright rim radial size is similiar to the thickness of the ionized shell , i.e. , @xmath143 . we derive an average electron density @xmath144 over the bright rim . at the brightness peak , the 3.6 cm flux is @xmath145 , hence the electron density reaches a local maximum of @xmath146 over the telescope beam . the effective flux of ly - c photons ionizing fresh material is @xmath147 ( @xmath148 is the ionized gas sound speed ) . as expected , this is much less than the flux of photons consumed in the photoionized layer , which is equal to @xmath149 ( @xmath150 is the recombination coefficient on hydrogen levels @xmath151 ) . taking into account the attenuation of the radiation between the star and the globule by the ionized gas , this value is consistent with the ( absorption free ) value derived from the ly - c luminosity of the exciting star , @xmath152 at the projected distance of tc2 . we detected with the sws spectrometer the following ionic and fine - structure atomic lines : @xmath86}}$ ] @xmath87 , @xmath59}}$ ] @xmath153 , @xmath78}}$ ] @xmath154 , @xmath78}}$ ] @xmath155 and @xmath156}}$ ] @xmath157 . the spectra are displayed in fig . the fluxes of the lines identified are given in table [ flux_sws ] . a priori , several regions with very different physical conditions contribute to the emission detected : the @xmath68 region ( see sect . 3 ) , the bright rim and the pdr ( in the case of @xmath156}}$ ] ) . since the @xmath156}}$ ] @xmath157 is believed to arise from the pdr , it is discussed below in sect . the other lines have been discussed in sect . 3 and we concluded that within the sws beam and the observational uncertainties the emission is dominated by the hii region . in order to precise the spatial distribution of the @xmath86}}$ ] and @xmath59}}$ ] lines , we use the cvf data , obtained with a much higher angular resolution ( @xmath158 ) . moreover , the cvf data benefits a better snr because of the low spectral resolution ( @xmath159 ) . in order to determine the emission from the globule itself , we have subtracted the contribution of the region , estimated from a nearby reference position ( position e , fig . the emission from the globule , once the contribution of the nebular gas was removed , is displayed in fig . the @xmath86}}$ ] emission is mostly detected along the border of the globule , at the surface , slightly shifted outside with respect to the emission of the pdr . this spatial shift is shown in fig . 7 where the contours of the pah @xmath9 band are superposed on the @xmath86}}$ ] emission map . the emission delineates a thin layer of @xmath160 . it can be seen that _ there is no @xmath86}}$ ] emission detected over the body of the globule_. a spectrum at position f in the main body of tc2 ( fig . 8f ) and confirms the absence of emission longwards of @xmath161 . therefore , _ the front side of the globule is hardly illuminated by the ly - c photons from the exciting star . in particular , the main body of the globule is not photoionized_. some weak @xmath86}}$ ] emission is detected all around the globule , especially in the northerwestern direction , i.e. towards the ionizing stars . we suggest that it could trace the freely expanding layers of the photoionized envelope around tc2 . this gas is expected to leave the surface of tc2 with a velocity close to the ionized gas sound speed , @xmath162 ( ll94 ) . the detection of high gas velocities , from spectroscopic measurements , and the study of their spatial distribution would allow to confirm the association of this component with the photoevaporated envelope of the globule . the distribution of the @xmath59}}$ ] @xmath163 emission was obtained following the same procedure , and is displayed in fig . 7 . the @xmath59}}$ ] map looks much noisier than @xmath86}}$ ] . again , there is no emission detected towards the body of the globule . the @xmath59}}$ ] emission is limited to the northwestern quadrant , between the globule and the ionizing stars . it overlaps with the outer part of the @xmath86}}$ ] flux distribution , at the surface of the globule , and in the extended component which possibly traces the photoevaporated envelope . this is consistent with the high ionization potential of the @xmath59}}$ ] line ( 41 ev ) with respect to to the @xmath86}}$ ] line ( 22 ev ) . additional lines were detected in the far - infrared with the lws . the emission from tc2 was estimated by subtracting the emission of the reference position ( see sect . 3 ) to the spectrum towards the globule . in this case , no correcting factor was applied to the fluxes since tc2 fills only half the lws beam ( fig . 4 ) . we observe a marked increase of the density around the globule . the ratio ( @xmath70}}~52/88)$ ] ratio yields @xmath164 . a more pronounced increase is observed using the ( @xmath72}}~57/{\hbox{[o{\small iii}]}}~52 $ ] ) ratio : @xmath165 , when adopting an abundance ratio of 0.5 . one can not exclude that the contribution of the region along the line of sight of tc2 is somewhat underestimated by the procedure descrived in sect . 3 . on the other hand , an additional factor can account for the observed increase in density : the photoionized envelope of the globule . as discussed in sect . 4.1 , the latter can be modeled as a layer of effective thickness @xmath166 and density @xmath167 , which expands almost freely around the globule . the emission of this layer could account for the observed increase in density towards tc2 and for the discrepancy between the density estimates from the ( @xmath70}}~52/88)$ ] and ( @xmath72}}~57/{\hbox{[o{\small iii}]}}~52 $ ] ) ratios . this is because the @xmath70}}$ ] @xmath79 and @xmath72}}$ ] @xmath73 lines are more sensitive to dense material than the @xmath70}}$ ] @xmath100 line ( their critical densities are similar to the ionization front density whereas it is only @xmath168 for @xmath70}}$ ] @xmath100 ) . indeed , a simple modelling with two layers of density @xmath169 and @xmath170 , thickness @xmath171 and @xmath172 respectively can account for the emission observed . the contribution of the dense layer is almost unnoticed at @xmath100 whereas it constitutes @xmath173 of the observed flux for the @xmath174 and @xmath175 lines respectively . the analysis of the various fine structure atomic lines detected with the sws and the lws spectrometers confirms that the bright - rimmed globule is bathing in the nebular gas of density @xmath176 . for most of the fine - structure atomic lines , the nebular gas dominates the contribution of the photoionized layer surrounding the globule . the photoionized gas envelope is detected at the vla ; it is indirectly detected in the @xmath72}}$ ] @xmath73 and @xmath70}}$ ] @xmath175 lines . the surface density is @xmath177 and rises up to @xmath178 . the photoionized envelope of tc2 is also detected in the @xmath59}}$ ] and @xmath86}}$ ] lines . the mapping of these lines shows that the globule is illuminated on the rear side . in this section we characterize the properties of the pdr : geometry , hydrogen density and gas column density . we first study the geometry of the pdr from the emission of the unidentified infrared bands ( uibs ) at 6.2 , 7.7 , 8.6 and @xmath9 as observed with isocam . the uibs are a good tracer of the strong uv field region associated with a pdr . although the exact chemical composition of these bands is still not known , the best candidates appear to be the polycyclic aromatic hydrocarbons , or pahs ( puget & lger 1989 ) . we will use indistinctly the former or the latter denomination in what follows . the physical conditions in the pdr are quantified from the emission of dust and of the mid - infrared lines detected with lws and sws : @xmath156}}$ ] @xmath179 , @xmath4}}$ ] @xmath180 , @xmath4}}$ ] @xmath181 and @xmath182}}$ ] @xmath183 . the @xmath4}}$ ] and @xmath7}}$ ] lines are especially interesting tools in the study of pdrs since they can be used to probe the excitation conditions in the gas . the line fluxes are compared with the recent models of kaufman et al . ( 1999 ; hereafter k99 ) and wolfire , tielens & hollenbach ( 1990 ) . these models extend the work by tielens & hollenbach ( 1985 ) , hollenbach , takahashi & tielens ( 1991 ) the main difference is that the model of k99 use very recent grain photoelectric heating rates , which include treaments of small grains and large molecules ( pahs ) . we estimated that such a contribution might be important in the case of tc2 since our cvf data show that the infrared emission shortwards of @xmath184 is dominated by the pahs emission bands . as can be seen in fig . 8 , the uibs ( or pahs ) bands at 6.2 , 7.7 , 8.6 and 11.3@xmath185 dominates the the spectral emission in the mid - infrared range @xmath186 , as observed with the cvf . one notes also in the spectra a strong 12.7@xmath187 m line that is a combination of the [ neii ] line and of a pah band , and the presence of a continuum emission at wavelength greater than 13@xmath187 m , also well detected in the sws spectra ( see sect . 6.2 ) , and the presence of the [ neiii ] 15.5 @xmath187 m line . to study the spatial distribution of the various features in the cvf spectra , we have made a spectral decomposition using a gaussian function for the [ neii ] line , lorentzian lines for the pahs bands ( boulanger et al . 1998 ) and a grey body for the continuum emission . the resulting maps are shown in fig . all the uibs are unambiguously detected and have the same spatial distribution , following closely the border of the globule . we note that there is no spatial segregation between the pahs bands probably as a consequence of the remote distance to the trifid nebula . the emission is shifted by @xmath188 towards the interior of the globule with respect to the photoionized region , as traced by the @xmath189}}$ ] line ( fig . the emission shows a local minimum in the center of the globule . the flux distribution is the brightest in the pdr and reaches its maximum close to the offset position @xmath190 . the pah emitting region in the pdr is mostly unresolved at all wavelengths ; this prevents any detailed physical analysis of the latter . the dark filament absorbs a fraction of the infrared radiation coming from the pdr , causing the presence of a local minimum in the brightness distribution . because of the weakness of the radiation field , almost no emission at all is detected over the body of the globule . only a weak feature is detected around @xmath191 ( fig . some weak emission is also detected outside the globule , behind the ionization front ( fig . we now compare the relative variations of the pah bands measured at a few positions over the globule and the @xmath68 region . in order to increase the snr , the emission has been averaged over four pixels at each position . the resulting spectra are displayed in fig . the contribution of the nebula , as estimated from the reference position ( fig . 8e ) has been subtracted to the spectra in order to outline the emission of the globule . the almost flat spectrum obtained in the core tc2 , especially the absence of ionic lines either in absorption or in emission , shows that the nebular gas emission was removed properly . it is also consistent with tc2 being heated on the rear side . we concentrate on the 7.7 and 11.3@xmath187 m as they have the highest snr . the bands have a ratio 7.7/11.3@xmath192 in the pdr . this ratio is @xmath1931.2 at the reference position ( panel e ) in the region and it reaches @xmath194 in the direction of the ionizing star . part of this variation of the 7.7/11.3 ratio is probably related to variations in the charge of the pahs . the current models and laboratory experiments on the pahs emission show that the relative intensity of the various bands are strongly sensitive to the ionization state . in particular the intensities of the c - c stretching modes ( 6.2 and @xmath195 ) and the c - h in - plane bending mode ( @xmath196 ) are generally stronger in pah cations than in pah neutrals by a factor of 10 ( see e.g. joblin et al . 1996 ) , whereas the intensity of the c - h out - of - plane bending mode ( @xmath9 ) seems to decrease with the ionization . in the pdr of the globule , the models of bakes & tielens ( 1998 ) and dartois & dhendecourt ( 1997 ) predict a very small fraction of pah cations ( 0.01 or less ) . in the photoionized gas around tc2 on the contrary , the spectra look very similar to those of a mix of pah cations superposed to a weak continuum ( see fig . 3d in allamandola , hudgins & sandford , 1999 ) . therefore the decrease of the @xmath9 band can be understood as the dehydrogenation and photoionization of the smallest pahs exposed to the strong uv field of the region . a similar effect has been reported in the region m17 by verstraete et al . ( 1996 ) and crete et al . ( 1999 ) and in ngc 1333 by joblin et al . ( 1996 ) . however , we observe a marked increase of the 7.7/11.3 ratio , above 2 , inside the globule . this value is comparable to that observed in the ionized gas , whereas the ionizing field at the surface of the globule , as traced by the @xmath86}}$ ] and the @xmath59}}$ ] lines , is much lower . hence , the relative variation of the 7.7/11.3 ratio can not be directly accounted for by standard models of pah excitation . as we discuss below , _ the pah emission observed towards the main body of the globule actually comes from the pdr located on the rear side _ and has suffered strong absorption from the globule material ( the millimeter dust map indicates an average @xmath197 over the body of the globule ) . the absorption by the silicates is so large that hardly any radiation escapes from the globule longwards of @xmath198 . this effect is illustrated by the cvf spectrum taken at position f , in the main body of tc2 . almost no emission at all is detected in the spectral window , apart from a weak feature coinciding with the pah band at @xmath199 . it is interesting to compare the emission maps of the globule at @xmath199 and in the @xmath86}}$ ] line at @xmath200 . in both maps , there is no emission in the northeastern region . this is the region where the reference spectrum was taken ( position e , fig . the border of the globule looks very bright at both wavelengths . whereas the main body of tc2 appears void of @xmath86}}$ ] emission , as mentionned before , it is still radiating some flux at @xmath199 ( see fig . cernicharo et al . ( 2000 ) showed that at 5.3 , 6.6 and @xmath201 there is a narrow window in which the absorption by the ices and the silicates is low enough to let the radiation escape even from the deeply embedded cores of class 0 protostars . in order to test our hypothesis , we have calculated the spectrum of the radiation emerging at the front side of the globule , assuming the emitting region ( the pdr ) to be located at the rear surface of the globule . the extinction through the globule was computed from an empirical absorption profile constructed by cernicharo et al . ( 2000 ) and scaled with the visual extinction inside tc2 . this empirical profile is based on the cvf spectrum of the class i protostar vla4 in the l1641 molecular cloud in orion . it takes into account the contribution of silicates , the ices of methanol @xmath202 , water @xmath203 and carbon dioxide @xmath204 . as noted by cernicharo et al . ( 2000 ) , the spectrum of vla4 is very similar to the spectra observed towards deeply embedded objects in massive star - forming regions . we also note that vla4 and tc2 are in a similar evolutionary stage . hence , we believe that despite some possible abundance variations between both protostellar cores , the vla4 absorption profile should represent a good approximation to the absorption profile in tc2 . we have applied this absorption profile to the spectrum of the bright pdr ( position d ) for a visual extinction @xmath197 , a value derived from the 1.3 mm continuum map , and similar to the average extinction over the globule . we show the calculated spectrum in fig . the @xmath199 band intensity is @xmath205 mjy / sr whereas the intensity of the @xmath9 band and the @xmath86}}$ ] line is now only @xmath206 mjy / sr . the synthetic spectrum of the pdr appears very similar to the spectrum in the body of the globule at position f. the apparent strong enhancement of the @xmath199 band with respect to the @xmath207 band is mainly a selective absorption effect of material located on the rear side of the globule . to summarize , the pah bands at 6.2 , 7.7 , 8.6 and @xmath9 are unambigusouly detected in tc2 . the variations of their relative intensities are in agreement with the models of pah excitation . they show evidence for the main body of tc2 being illuminated on the rear side and not on the front side . this is in agreement with the analysis of the @xmath86}}$ ] and @xmath59}}$ ] lines . as a consequence , the line intensities can be obtained , at first order , from the integrated fluxes by dividing by the area of the globule encompassed by the telescope beam . the lws spectra of the @xmath7}}$ ] and @xmath4}}$ ] lines are shown in fig . the emission from the globule was obtained by subtracting a reference position to the spectra taken towards tc2 ( see sect . the uncertainties in the flux of the @xmath4}}$ ] lines are rather weak since the emission at the reference position appears much weaker than towards the globule . on the contrary , the variations of the @xmath7}}$ ] line are much weaker and the contribution of tc2 amounts to @xmath208 of the total flux ( see table [ flux_lws ] ) . the procedure applied to estimate the emission from the globule ignores a possible contribution of the photoionized envelope surrounding tc2 to the observed flux . indeed , the analysis of the far - infrared @xmath78}}$ ] and @xmath70}}$ ] lines carried out in sect . 3.1 suggested that this region was contributing to the flux detected . the @xmath7}}$ ] line is one of the main tools used to probe the physical conditions in pdrs as it provides a direct estimate of the mass and column density of pdr gas . it is therefore important to estimate as accurately as possible the @xmath7}}$ ] flux coming from the pdr itself , and leave aside any other contribution to the flux measured . we consider the 20 cm free - free emission map of the trifid presented in cl98 ( fig . these data were obtained at a resolution of @xmath130 . this allows to trace the extended emission from the main body of the globule , unlike the high - angular observations presented in sect . we apply the method described by heiles ( 1994 ) , who showed that intensity @xmath209 of the @xmath7}}$ ] line scales with the brightness temperature of the free - free emission in low - density region and we derive the relation : @xmath210 where @xmath209 is expressed in units of @xmath211 and @xmath212 is the ionic abundance @xmath213 in units of @xmath214 . the exact ionic abundance of @xmath215 depends on various factors : density , effective stellar temperature . based on the models of rubin ( 1985 ) , we estimate @xmath212 to range between 0.5 and 0.8 . the free - fre emission map shows some extended emission in addition to the bright rim . the emission peaks at @xmath216 in the bright rim . from the distribution of the contour levels around the globule , we find that the average intensity in the surrounding nebular gas is @xmath217 . this implies a peak flux of @xmath218 in the bright rim . following the same method as in sect . 3.1 , we infer a local density @xmath219 , in good agreement with our measurement at higher angular resolution . after subtracting the contribution of the nebular gas and integrating the emission over the globule , we obtain a mean brightness temperature @xmath220 for the residual emission . since the contribution of the nebular gas was removed , this residual is the emission of the photoionized envelope and the ionization front , i.e. the bright rim . hence , the intensity of the @xmath7}}$ ] line @xmath221 . this value represents approximately @xmath58 of the flux detected with the lws towards the globule . after correcting for the contribution of the surface ionized layers , we obtain that the flux of the @xmath7}}$ ] line in the pdr gas is @xmath222 . in the optically thin limit , the mass of atomic gas can be easily obtained from the observed @xmath7}}$ ] line flux , using the analytic formula derived by wolfire et al . ( 1990 ) : @xmath223 \left(\frac{f_{\rm cii}}{10^{-17}w{\hbox{\kern 0.20em cm$^{-2}$}}}\right ) \nonumber \\ & & \times \left(\frac{10^{-21}{\hbox{\kern 0.20em erg}}{\hbox{\kern 0.20em cm$^{-2}$}}{\hbox{\kern 0.20em s$^{-1}$}}{\hbox{\kern 0.20em sr$^{-1}$}}\rm atom^{-1}}{\lambda(\rm cii)}\right ) { \hbox{\kern 0.20em $ m_\odot$}}\nonumber\end{aligned}\ ] ] where @xmath224 is the cooling rate in the @xmath7}}$ ] @xmath225 line per carbon atom and per steradian , @xmath226 is the abundance of ionized carbon per hydrogen atom . here , we adopt a carbon abundance @xmath227 ( see k99 ) . the atomic gas temperature in the pdr of tc2 is close to 300 k ( see next section ) . for hydrogen densities of @xmath228 , comparable to that in the pdr gas , the cooling rate @xmath224 is then @xmath229 . the mass of atomic gas in the pdr is therefore @xmath230 . from the fraction of the globule s area that fills the lws beam , we derive the average hydrogen column density of pdr gas : @xmath231 . we present below the physical conditions derived from the analysis of the far - infrared lines and based on the model of k99 . we find a good agreement between this model and the observational data . the physical parameters of the globule are summarized in table 5 . based on the fuv field intensity @xmath3 and using the model of k99 , we find that the gas is heated by photo - electrons to a temperature of @xmath232 at the surface of the pdr , a value which actually depends rather little on the actual density in the pdr ( see their fig . 2 ) . this value is also a good estimate of the actual temperature in the pdr . the hydrogen density in the pdr was estimated from a large - velocity gradient analysis of the @xmath4}}$ ] lines . this approach offers the advantage of being independent of the oxygen elemental abundance . the size of the emitting region in the lws beam is taken to be @xmath233 . we took @xmath234 as linewidth , based on the observations of the millimeter lines of @xmath235 and cs in the head of the globule ( sect . the core density derived from the millimeter dust thermal emission provides an upper limit to the density in the pdr : @xmath236 . hence , in order to account for the observed @xmath4}}$ ] fluxes at @xmath5 and @xmath6 , the gas temperature has to be larger than @xmath237 . the best match is obtained for a column density @xmath238 ( fig . 10 ) . at a temperature of @xmath239 , as derived above for the pdr , the molecular hydrogen density is @xmath240 . adopting an oxygen elemental abundance @xmath241/[h]= 3.0\times 10^{-4}$ ] , the corresponding gas column density in the pdr is @xmath242 . there is a possible bias in this analysis as the globule is illuminated under some inclination angle while it is viewed face - on by us . this anisotropy in the illumination drives heterogeneous conditions across the globule ( i.e. in the plane of the sky ) . as a consequence , the filling factor of the @xmath4}}$ ] @xmath180 line , more easily excited , is likely to be larger than that of the @xmath6 line . this effect was not taken into account in the present calculation and results in an underestimate of the @xmath4}}$ ] column density whereas the hydrogen density is overestimated . the @xmath121}}~63/{\hbox{[c{\small ii}]}}~158)$ ] and @xmath121}}145/63)$ ] line ratios provide another method to derive the parameters of the pdr . these ratios are 0.066 and 4.2 respectively ; using figs . 4 - 5 in k99 , we obtain direct estimates of the fuv field intensity @xmath122 and of the hydrogen nuclei density @xmath243 . this estimate of @xmath111 is in good agreement with the previous , independent , determination , based on the far - infrared dust continuum . in the range of values taken by the @xmath121}}~63/{\hbox{[c{\small ii}]}}~158)$ ] ratio , the ratio is not very sensitive to the fuv field intensity @xmath111 . hence , an overestimate of the @xmath7}}$ ] line flux would imply that the actual pdr density is somewhat larger . an upper limit of 10 on this ratio yields a hydrogen _ nuclei _ density of @xmath244 . we determine the electron density in the pdr from the photoelectric heating efficiency @xmath245 , which is governed by the factor @xmath246 ( see e.g. bakes & tielens 1994 ) . the integrated intensity of all the fir lines is well approximated by the sum of the @xmath4}}$ ] @xmath5 and @xmath182}}~158\mu m$ ] intensities , which amounts to @xmath247 . we note that since the cooling of the pdr gas is dominated by the @xmath4}}$ ] @xmath5 line , the uncertainties in the @xmath7}}$ ] line flux do not affect significantly the total cooling nor the conditions of thermal balance . we estimate the integrated fir continuum intensity radiated by the pdr from the warm component determined in the spectral energy distribution : @xmath248 . on the other hand , the fuv photon heating amounts to @xmath249 . this value represents a lower limit to the total heating since the contribution of the photons outside the fuv band has been neglected here and could represent a substantial fraction ( up to 1 ) of the fuv heating ( tielens and hollenbach , 1985 ; wolfire et al . the agreement between the fir continuum radiation and the uv photon heating is therefore satisfactory and supports our view that the warm dust layer detected with iso / lws and sws ( see sect . 6 ) is indeed tracing the pdr at the surface of the globule . this yields a photoelectric efficiency @xmath250 . from the variations of @xmath245 with the factor @xmath246 ( see fig . 21 of bakes & tielens 1994 ) , we estimate @xmath251 , hence the electron density in the pdr : @xmath252 . in the pdr , the ionization fraction is determined by the photoionization of carbon . adopting a carbon abundance of @xmath253 , we obtain an estimate of the hydrogen density @xmath254 . the emission of the @xmath156}}$ ] @xmath157 line can be produced in the pdr of molecular clouds which are illuminated by strong fuv fields ( tielens & hollenbach 1985 ) and also in @xmath68 regions ( rubin , 1985 ) . in the case of orion , as prototype of massive star forming region , walmsley et al . ( 1999 ) showed that the @xmath156}}$ ] emission comes mainly from the pdr and not from the @xmath68 region or the ionization front region . we try to determine the origin of the emission observed towards the globule . it is difficult to draw any definite conclusion because of the lack of angular resolution , hence of information about the spatial distribution of the emission . also , the noise in this band of the sws is known to be high , so that the actual variations between the three positions might be somewhat more pronounced . an important constraint comes from the measurements of the thermal dust emission at 1.3 mm ; they indicate an average visual extinction @xmath197 across the globule ( see sect . the values of the interstellar extinction tabulated by mathis ( 1990 ) allow to estimate the opacity of the globule at @xmath255 : @xmath256 . therefore _ the globule is mainly transparent at this wavelength_. for the physical conditions encountered in the pdr of tc2 , the model of wolfire , tielens & hollenbach ( 1990 ) predicts a typical value of 0.015 for the ( @xmath156}}$ ] 34.8/@xmath4}}$ ] 63 ) ratio whereas the observed ratio is 0.5 . assuming all the emission arises from the pdr , this implies a silicon gas phase abundance about 30 times higher than that assumed in their computation , corresponding to @xmath257/[h]\simeq 2.3\times10^{-5}$ ] , i.e. 0.6 times the solar elemental silicon abundance ( @xmath258 ) . this is unrealistically high as observations in other @xmath68 regions like orion indicate that about @xmath259 of silicon is tied up in dust grains ( walmsely et al . 1999 ) , a value also similar to that found in diffuse interstellar clouds , therefore , the pdr of tc2 alone can not account for the measured @xmath156}}$ ] flux . indeed , the sws spectra show that the @xmath156}}$ ] flux at the northern position is only @xmath260 less than at the southern and central positions , whereas the globule fills only one third of the sws beam . since the globule is transparent to the @xmath156}}$ ] radiation and it is illuminated mostly on its rear side , one would rather expect a flux @xmath261 one third of that detected in the central beam , i.e. @xmath262 . this is much less than the flux detected . this reinforces our previous conclusion about the origin of the @xmath156}}$ ] line . assuming the flux variation between the central and northern position is due to the variation of pdr filling factor in the sws beam , one can derive an estimate of the relative contribution of the pdr and the @xmath68 region . the emitted flux is then @xmath263 and @xmath264 for the pdr and the @xmath68 respectively ; the emission of the pdr is somewhat dominated by the contribution of the @xmath68 in the sws beam . the model of wolfire , tielens & hollenbach ( 1990 ) is then consistent with a gas phase abundance of @xmath265 , i.e. @xmath266 the solar abundance . this value is much more compatible with the determinations obtained in other @xmath68 regions . nevetherless , we stress again that new observations with a better snr are required to determine more precisely the silicon abundance in the pdr and the @xmath68 region . in this paragraph , we analyse the continuum emission detected with the sws in the range @xmath267 . the full spectra obtained at the three positions are shown in fig . the continuum emission becomes significant longwards of @xmath268 in all the positions . we have fit the continuum using a black - body modified by a power law @xmath269 . a satisfactory fit was obtained for a dust temperature of 46 k , and a dust spectral index @xmath270 , typical of the values observed in the mid - infrared ( hildebrand , 1983 ) . here , we assume that the mid - ir flux originates from the pdr and fills the sws beam . following the reddening law determined by lynds et al . ( 1985 ) for the trifid , we estimate a hydrogen column density @xmath271 . adopting the standard reddening law yields @xmath272 . these values are only indicative since they rely on the geometry assumed for the emitting region . reasonable fits can be obtained with the temperature in the range 43 - 48k , and hydrogen column densities of @xmath273 . the same procedure has been applied to the other positions observed with the sws . at the northern position , the pdr of the globule fills only partially the sws beam . we have estimated a size of @xmath274 for the region encompassed by the beam . the other parameters for the warm layers were left identical otherwise . at the southern position , the continuum emission could be fit with similar parameters and assuming a smaller column density of warm dust ( @xmath275 ) at about the same temperature . the fits are shown superposed on the spectra in fig . they succeed rather well in reproducing the continuum flux longwards of @xmath276 . it is therefore possible to explain the observed continuum emission by a warm dust layer at about @xmath277 and a column density @xmath278 . as we show below , such parameters also allow to account for the continuum emission detected at longer wavelengths . comparison between the fit of the warm dust component and the sws data shows actually the presence of a residual flux as a flat continuum between 10 and @xmath276 . the continuum emission shortwards of @xmath276 is almost identical at the northern and central positions . it is somewhat weaker in the south . the continuum emission is maximum between 15 and @xmath276 . our cvf map of the continuum emission in the range @xmath279 brings some more information on the distribution and the nature of this hot dust component . the continuum emission at 15@xmath187 m revealed from our spectral decomposition is shown in fig . the map shows a good spatial correlation with the @xmath195 band and the photoionized envelope . the continuum is brightest in the pdr . it is shifted outwards by @xmath280 and gets stronger as one moves outside of the globule , in the photoevaporated gas . there is almost no emission from the body of the globule . we detect a higher continuum level in the western side of the globule with respect to the northern one . this difference might result from a higher heating efficiency on the western side because the uv photons arrive almost normal to the surface . the rise of the continuum could also be related to an increase of the gas density like what was observed in other pdrs ( abergel et al . 2002 ) and in cirrus clouds ( miville - deschnes et al . 2002 ) . within the present understanding of the nature of the mid - infrared emitters , it is impossible to determine the origin of the increase of the continuum emission in denser regions but one plausible explanation could be related to a change of the dust size distribution . there is also a rather good spatial correlation between the 15 @xmath187 m continuum and the distribution of the ionizing gas , as traced by the @xmath86}}$ ] line . the correlation of this emission with the photoionized envelope suggests that the emission detected probably comes from _ very small grains _ heated by the strong uv field . the thermal emission of the cold dust was observed with the mpifr 19 channel bolometer array at the iram 30 m telescope ( fig . we detected some weak emission , at typical fluxes of @xmath281 , which spatially coincides with the region of high obscuration in the optical image , south of the ionization front . the millimeter continuum emission is sharply limited by the ionization front , and the brightness contours are closely spaced between the bright rim and the emission peak . the globule peaks at @xmath282 at the offset position ( @xmath283,@xmath284 ) , only @xmath33 behind the ionization front . from the contour at half power , we estimate a size ( beam - deconvolved ) of @xmath285 ( @xmath286 by @xmath287 ) for the core . assuming an average dust temperature @xmath288 , a spectral index @xmath289 and an absorption coefficient @xmath290 , we derive a hydrogen column density @xmath291 at the flux peak . integrating over the contour at @xmath292 ( hpfw ) , we find a total flux of @xmath293 and estimate a core mass @xmath294 . this implies a typical density @xmath295 for the core . the spectral energy distribution of the globule between 45 and @xmath16 was obtained by subtracting the emission of the reference position to the lws spectrum taken towards tc2 . the size of the emitting region encompassed by the lws beam is @xmath296 , based on the isocam @xmath297 image and the cold dust millimeter emission ( fig . we measured the 1.3 mm continuum filling the lws beam solid angle in order to constrain the fit to the cold dust component . it was found to be @xmath298 jy . the spectral energy distribution is shown in fig . 12 . the emission could be satisfactorily fit by a two - component model : a cold core of column density @xmath299 at a temperature @xmath300 , with a dust spectral index @xmath301 , surrounded by a warm layer at a temperature @xmath302 , with a hydrogen column density @xmath303 , and a dust spectral index @xmath270 . the gas column density of the warm layer is not very high as it corresponds to @xmath304 . we note that it is very similar to the column density of the pdr gas traced by the @xmath7}}$ ] @xmath8 and the @xmath4}}$ ] lines . the warm layer detected with the lws also accounts for the mid - infrared continuum emission detected with the sws longwards of @xmath305 . integrating under the fit of the warm dust component , and correcting for dilution in the lws beam , we obtain the infrared intensity radiated by the warm layer : @xmath306 . the temperature and the infrared luminosity of the warm layer are those expected for a pdr exposed to a fuv field @xmath122 . this is consistent with the intensity of the far - infrared @xmath4}}$ ] and @xmath7}}$ ] lines as observed with lws ( see above sect . the whole set of observational data leads us to the conclusion that _ the warm dust component detected with the lws is actually tracing the pdr of the globule_. the average hydrogen column density is @xmath307 . with a typical hydrogen nuclei density @xmath243 , as estimated from the far - infrared line ratios ( sect . 5.3 ) , we estimate the thickness of the pdr @xmath308 . the cold dust component revealed in the spectral energy distribution is therefore tracing the innermost part of the globule ( the core ) which is protected from the external uv radiation field . hence , it seems a good reasonable approximation to adopt 22 k as temperature of the dust core traced by the 1.3 mm continuum emission in order to the estimate the mass of the core . this temperature is similar to those observed in the molecular cloud surrounding the nebula ( lefloch & cernicharo , 2000 ) . a more accurate determination of the mass and column density of the core and the globule would require a better knowledge of the temperature profile . this is not allowed by the low angular resolution of the lws observations , which averages the emission over a much too large region ( @xmath309 ) . we define the total mass @xmath310 of the dust condensation by integrating over the flux contour at @xmath311 beam ; this area of mean size @xmath312 contains a total flux @xmath313 . adopting a uniform dust temperature @xmath300 , we obtain the average gas column density over the globule : @xmath314 , the total mass of the globule @xmath315 , and the average density in the globule @xmath316 . hence , we find that it is only a small mass fraction of the whole globule ( @xmath317 ) which lies in the atomic surface layers , exposed to the strong fuv field of the stars exciting the trifid . the ratio of the atomic and molecular gas masses is much lower in tc2 than in the omc 1 in orion and in the galactic center , where it is about @xmath318 . on the contrary , the ratio takes a very similar value to that in the young massive star - forming region w49n . this region is one of the youngest massive star - forming place in the galaxy , where several recently born o stars have just started to excavate the parent cloud ( vastel et al . the low atomic to molecular gas ratio measured in tc2 is therefore consistent with a very early evolutionary age for the globule . this point is more thoroughly addressed in sect . the molecular content of the bright - rimmed globule was first observed in the co lines . the line profiles are complex and exhibit several components between -50 and @xmath319 that correspond to various physical regions in the molecular cloud containing the hii region ( lefloch et al . because of the large extent of the co emission in the nebula , antenna temperatures are a better approximation than main - beam brightness temperatures to the brightness of the co lines . we adopted the same approximation for the @xmath38 @xmath37 transition . on the contrary , we assumed main - beam brightness temperatures for @xmath39 . the spatial distribution of the kinetic components shows that the emission of tc2 peaks at @xmath320 ( fig . the distribution of the emission follows closely the bright rim and drops abruptly beyond the ionization front . the antenna temperatures of the co transitions are rather uniform over the globule , with values of @xmath321 k for the and transitions and 15 k for @xmath322 one . we have tried to constrain the temperature and density distribution in the globule by modelling with a radiative transfer code the excitation of the co lines . a priori two zones are contributing to the emission : the molecular core and the pdr . the large ratio between the and the lines is very difficult to explain . observations of the high - density gas show that the velocity shift between the quiescent core and the surface layers , which are accelerated by radiatively - driven implosion , is small , of the order of @xmath323 ( see sect . this is consistent with the symmetric profiles observed for all the molecular transitions ( fig . this velocity shift is small enough that the pdr and the dense core are still radiatively well coupled . therefore , most of the radiation coming from the pdr on the rear side of the globule is absorbed by the dense core . as a consequence , its contribution to the and lines is almost negligible and we detect mainly the emission of the cold core and the front side layers . we infer a kinetic temperature of about 30 - 35 k for the gas , somewhat higher than the dust temperature . the discrepancy could be due to the heating of the front side by optical photons , since the globule is immersed in ionized gas . the data show widespread emission over all the globule , with typical brightness ( antenna ) temperatures of @xmath324 and 12.5 k for the and transitions respectively ( fig . the has a main - beam brightness temperature of @xmath325 . from the ratio of the @xmath326 @xmath37 brightness temperatures (= 5.7 ) and assuming a standard relative abundance of 8 , we derive the @xmath38 line opacity @xmath327 . we note that the opacity derived is fully consistent with the value of the ratio @xmath328 @xmath37 brightness temperatures . an opacity @xmath329 implies a ratio of 2 whereas the actual value is 2.3 . the mean density in the globule is high enough that the @xmath37 line of co and its isotopes are thermalized . we have used the opacity and the brightness of the @xmath38 @xmath37 line to constrain the kinetic temperature and the gas column density . the best match is obtained for a gas temperature @xmath330 and a column density @xmath331 . adopting a standard abundance @xmath332/[{\hbox{h${}_2$}}]=1.6\times 10^{-6}$ ] , we infer the total gas column density @xmath333 good agreement with the value obtained from the dust measurements , and the average gas density @xmath334 , after dividing the gas column density by the mean globule radius . all the molecular lines observed towards the center of the globule peak at @xmath335 without any significant variation between the tracers ( fig . the optically thick @xmath336 line has a distribution similar to that of co ( fig . it was detected all over the globule , with typical main - beam brightness temperatures of 2 - 4 k. we also observed the isotopic @xmath337 line but failed to detect it . the distribution of the dense gas was mapped in the cs and lines . the cs emission is less extended than @xmath235 , as it appears to trace the region associated with the cold dust core ( fig . the cs lines are bright with main - beam temperatures of the order of 2 k , up to @xmath338 ( 4.3 k and 4.0 k for the @xmath36 and the @xmath322 lines respectively ) at the position @xmath339 . the cs @xmath36 transition traces a region approximately circular with a typical size of @xmath340 ( hpfw ) , marginally resolved by the telescope beam . the higher angular resolution of cs data reveals two gas components : a weak , extended , component which overlaps very well with the globule , as traced in @xmath341 and the millimeter continuum , and a strongly peaked condensation in the center of the globule ( fig . 14 ) : the central cs core . the core is slightly elongated in the north - south direction . its dimensions ( beam - deconvolved ) are @xmath342 by @xmath343 , very similar to the size of the the cs @xmath36 emitting region . hence both lines are probing the same region . as can be seen in fig . 15 , there is no evidence of a velocity shift in the cs @xmath36 emission peak along the major axis . the same results holds for the the transition and the @xmath235 line . this indicates that the molecular gas of the globule , in particular the dense core , is in a quiescent stage . in order to determine the physical conditions in the dense core , we carried out an lvg analysis on the and lines assuming a gas kinetic temperature of @xmath344 , suggested by the millimeter continuum observations . three positions were studied : the center of the core and two other positions offset by @xmath33 . the velocity width was estimated from a gaussian fit to the line profiles : @xmath345 . in the central position we find a density @xmath346 and a column density @xmath347 . the line opacities are relatively low , with @xmath348 and @xmath349 . similar densities are obtained around the center ( @xmath350 ) and at the brightness peak , @xmath33 south of the center . at this position , we find @xmath351 . hence , from the core density and the beam size , we find a core mass @xmath352 , in good agreement with the dust estimate . adopting a typical abundance @xmath353/[{\hbox{h${}_2$}}]= 5\times 10^{-10}$ ] , we estimate the size of the cs emitting region @xmath354 . this value is consistent with the the size hpfw derived from the maps of velocity - integrated emission at 3 and 2 mm . the picture coming out of the molecular line and thermal dust continuum analysis is that the globule consists of a cold central core of density @xmath355 surrounded by an envelope slightly warmer in the outer layers , at a temperature @xmath356 , and density of about @xmath357 . we have recalled briefly in the introduction the overall evolution of a photoionized globule . the time evolution of a photoionized globule , the density and the velocity fields , were studied numerically and presented in ll94 ( see in particular their fig . 4 ) under a wide range of ionization conditions , including those typical of bright - rimmed globules . the evolution of the photoionized globule is determined by the two following parameters : + a ) @xmath358 : the ratio of the impinging photons consumed to balance recombinations to those used to ionize neutral material . for tc2 , we obtain @xmath359 . hence , most of the ly - c photons are consumed in the ionized gas layer surrounding the neutral condensation ( see also sect . + b ) @xmath360 : the ratio of the ionized gas density with respect to the neutral gas density . this factor is directly related to the overpressure exerted by the ionized gas , and the intensity of the shock driven in the condensation . the mean globule density , ahead of the pdr , is the most difficult term to estimate . it is probably a good approximation to assume a density comprised between the pdr density ( @xmath361 ) and the molecular gas traced by @xmath38 ( @xmath362 ) . hence , we estimate that @xmath363 lies in the range @xmath364 . the values of @xmath363 and @xmath365 found for tc2 are typical of a bright - rimmed globule . following the convention defined by ll94 , this corresponds to region iv in the @xmath366 plane ( see their fig . 3 ) . in this case , the whole evolution is governed by the propagation a d - critical i - front preceded by a shock front . in their study , ll94 showed that the morphology , the density and velocity structure of a photoionized globule mainly depend on the duration of the ionization , i.e. the time elapsed since the illumination began . it does not depend critically on the `` real '' values of @xmath363 and @xmath365 . a direct estimate of the total ( ram + thermal ) external pressure at the surface of the cloud yields : @xmath367 . we can estimate the inner pressure from the kinematic motions as measured by the the linewidth of the and cs transitions : @xmath368 . we find a kinetic pressure @xmath369 . this is 20 times weaker than the outer pressure . therefore , the inner pressure can not sustain the globule against the overpressure of the ionized material . ll94 showed that in the course of its evolution , the pressure of the ionized gas increases as the radius of the globule decreases ( sect . 5.1 in ll94 ) . in other words , the overpressure at the surface of the globule was less in the past . figure 15 shows the emission of the dense molecular gas in the @xmath336 and cs @xmath36 lines along a cut in declination across the border of the globule . the @xmath336 line is optically thick and is therefore well suited to trace the motions in the surface layers of the globule . next to the @xmath370 emission main peak at @xmath371 , we detect a second component at `` blue '' velocities shifted by @xmath372 . this feature appears as bright as the main body gas emission . it is unambiguously detected in the cs @xmath36 line . it is detected only along the border of the globule , @xmath33 north with respect to the center , and _ not inside _ the main body . it disappears again @xmath373 north of the center , at the tip of the globule . the weakness of the line at these declinations is due to the small beam filling factor as the major fraction of the beam solid angle points towards the ionized gas in the region . the low snr of the spectrum prevents from leading any quantitative analysis . this component is not related with prostostellar activity . first , the blueshifted component is detected _ only _ at the border of the globule . there is no evidence of a blueshifted component towards the dust emission peak @xmath33 south ( more than @xmath374 au away from the border of the globule ) where the protostar is expected to be found ( see sect . second , the blueshifted component is kinematically separated from the main body gas emission and it does not exhibit the typical `` wing '' profile of molecular outflows . third , there is no evidence for an additional , redshifted , component which would trace the other wing of the molecular outflow . on the contrary , the profile of this secondary kinematic component is much more suggestive of a shock propagating into the globule from the _ rear _ side . the detection of this secondary kinematic component at the border of the globule , well separated from the main body gas emission , is typical of a globule in the early collapse phase and testifies that a shock is propagating into the globule from the surface . this is in agreement with the mid - ir analysis which showed that the globule is photoionized on the rear side . _ this secondary molecular component is the kinematical signature of the pdr_. it is completely unresolved by the 30 m telescope . however , since the far - infrared observations indicate a typical size of @xmath375 for the pdr ( @xmath376 at the distance of the trifid ) , it could be resolved out by millimeter interferometers . we compare the physical properties of tc2 with the numerical modelling of a photoionized globule presented in ll94 . the simulation was done for ionization parameters similar to tc2 : @xmath377 and @xmath378 ( @xmath379 and @xmath380 are the initial values of the parameters @xmath363 and @xmath365 ) . it is therefore possible to relate the properties of the _ observed _ globule to the simulated cloud via the simple scaling ( ll94 ) : @xmath381 , @xmath382 , @xmath383 . as mentioned above , tc2 is undergoing the collapse phase , when a shock front propagates into the neutral gas . based on a morphological comparison with the simulated globule ( figure 4 in ll94 ) , we find the best match for an age comprised between @xmath384 and @xmath385 in the computation . at @xmath386 ( fig . 4b ) , a dense core has formed below the surface of the globule , which still exhibits a `` barnacle '' shape . later on , at @xmath385 , ( fig . 4c ) , the bulk of the material has collapsed onto the main axis . the globule has now adopted an elongated shape . the best morphological match is obtained for an intermediate time @xmath387 . the numerical modelling indicates that the amplitude of the secondary kinematic component at such early photoionization stages is typically @xmath388 ( see their fig . this velocity shift could be smaller depending on the inclination angle of the globule surface with respect to the line of sight . the @xmath372 difference observed in tc2 between both kinematic components is therefore fully compatible with the numerical results . at that stage , the simulated globule has lost about @xmath389 of its initial mass ( see fig . 8 in ll94 ) . applying this result to tc2 , it means that the initial mass of the globule was @xmath390 . since the initial mass of the _ simulated _ globule is @xmath391 , we can derive the scaling factor relating the properties of the observed globule to the model : @xmath392 . the radius of the _ scaled _ ( simulated ) globule is then @xmath393 cm whereas we measured @xmath394 cm . we note however that the density is assumed to be uniform in the simulated globule , whereas a core - envelope density structure , as in tc2 , would lead to a smaller radius for the same ionization conditions . the duration of the photoionization , scaled from the simulations , is @xmath395 . this compares very well with the kinematical age of the trifid , as measured from the luminosity of the exciting star and the expansion of the nebula ( lefloch & cernicharo , 2000 ) . it is therefore very likely that the globule was very early exposed to the ionizing radiation , as soon as the o star turned on . the first evidences of ongoing star formation in tc2 were presented by cl98 , based on a @xmath396}}$ ] image of the hh399 jet running out of the bright rim of tc2 . the jet has been studied in the optical by rosado et al . ( 1999 ) and cl98 . these observations revealed the fragmented structure of the jet and could identify several `` knots '' from the head to the base of the jet , a - g , a being the most remote knot ( see fig . our vla observations detect only the external part of the hh 399 jet . we could detect three `` clumps '' which coincide with the knots dubbed a , c and d by rosado et al . the clumps are unresolved in the transverse direction by our observations . the flux peaks are rather similar ( between 70 and @xmath397 ) . we take @xmath398 ( @xmath399 ) as an upper limit to the diameter of the jet , and adopting a temperature between 5000 and @xmath400 we derive average electronic densities of @xmath401 in the clumps . our estimate is in agreement with rosado et al . ( 1999 ) for knot a. there are larger discrepancies for knots c and d , which the authors estimate to have densities of @xmath402 and @xmath403 respectively . this could be due to clumpiness in the jet structure . this can be also explained by our assumption on the geometry of the jet , apart from the uncertainties in the 3.6 cm flux ( the noise is @xmath404 ) . for instance , assuming a jet diameter ten times less would yield an electron density of @xmath405 , closer to the estimates derived at optical wavelengths . as soon as the jet runs out of the pdr , it is exposed to the ionizing radiation of the exciting star . integrating over the contour at @xmath142 , we obtain a total flux @xmath406 for the jet , and an average emissivity @xmath407 . the recombination rate per surface unit of the jet is @xmath408 . this is only half the ionizing flux impinging on the globule , as estimated in sect . 3.1 . hence , there are enough ly - c photons to maintain the jet fully ionized once it propagates out of the globule . hh399 is one of the finest examples of photoionized jets reported so far ( see also reipurth et al . 1998 ) . we searched for some molecular outflow emission which would be the counterpart to the optical jet but did not find any evidence of such a phenomenon . this is probably because of the unfavourable orientation of the jet , very close to the plane of the sky ( rosado et al . 1999 estimate an inclination lower than @xmath409 ) . we also searched for sio emission as it is a tracer of young protostellar outflows , associated with class 0 and/or massive sources ( bachiller , 1996 ; lefloch et al . again , this search yielded only negative results . this suggests that the powering source has probably reached the class i stage , characterized by a less active accretion phase than the class 0 , and has a typical evolutionary age of a few @xmath10 . if this age is correct , it means that star formation began approximatly at the time when the ionizing star hd 164492a turned on and ignited the nebula . the observation of the molecular outflow emission and the overlap between the wings usually provides a good determination of the position of the driving source . since this method can not be applied here , in the absence of molecular outflow , the only indications on the source location are provided by the vla observations . we have detected two small components inside the molecular core of tc2 ( fig . remarkably , they are both aligned with the jet propagation axis . both components are unresolved and are only marginally detected with fluxes of 50 and @xmath410 respectively at a level of @xmath30 and @xmath411 respectively . it is not clear whether these components trace some material associated with the counterjet or some material in the protostellar environment of the driving source itself . however , since they are located close to the millimeter dust emission peak we favor the latter hypothesis . in the mid - infrared , as observed with isocam in the broad band filters , there is no evidence of point - source emission in the globule . this indicates that the powering source of hh399 is a much less luminous and massive object than the protostar detected in southwestern molecular cloud ( lefloch & cernicharo 2000 ) . a close inspection of the continuum emission between 13 and @xmath412 in the globule however reveals a local maximum of compact , unresolved , emission at @xmath413 , @xmath414 . this source is spatially separated from the pdr , though close to it ( @xmath415 ) . it coincides with the vla sources to better than @xmath416 , actually falling between them . altogether , these observations give strong support for a physical association between the vla component(s ) and the driving source of hh399 jet . the bolometric luminosity of the protostellar source is difficult to estimate since the pdr contributes to the luminosity of the globule too . we estimate the protostellar luminosity by taking into account only the cold dust component estimated from the fit of the sed : @xmath417 . an upper limit is obtained by integrating under the spectral energy distribution : @xmath418 . this value is typical of intermediate - mass protostars but we stress again that the actual luminosity could be lower . the radiation field has a deep impact on the herbig - haro jet as soon as it escapes the globule . however , it is less clear how much the star forming process itself is perturbed . in the present stage , the size of the pdr represents approximately one third of the globule s radius . it is the low - density envelope , i.e. the mass reservoir , which is affected by the fuv field , and not the dense core where the material is being accreted . an important paramater in the evolution of the globule is its mass - loss rate , which is determined by the ionizing flux and the radius of the globule ( see ll94 ) : @xmath419 for tc2 , we estimate a present mass - loss rate of @xmath420 . from comparison with numerical simulations , we found that the globule lost about @xmath391 since the ionization began . the globule has evaporated half the mass of its envelope until now and its expected lifetime is therefore @xmath421 . this is a lower estimate as the mass - loss rate increases with radius . on the other hand the active accretion phase is known to last a few @xmath10 ( andr et al . therefore , the life expectancy of the globule is long enough to allow the protostellar objects to complete smoothly the accretion phase and turn into stars before the globule wholly evaporates . tc2 is going to develop progressively a low density tail , turning into a so - called cometary globule . we speculate that within a few @xmath10 the situation should be very similar to the gas fingers observed in the eagle nebula ( white et al . 1999 ) , where the strong uv field is evaporating the ultimate dust and gas layers of the stellar nests , unveiling the star(s ) formed in the early stages of the nebula . we have carried out a multiwavelength study of bright - rimmed globule tc2 in the trifid nebula . the globule lies almost in the plane of the sky , immersed in the low - density gas of the @xmath68 region ( @xmath422 ) . it is illuminated by the o star hd 164492a , mainly on the rear side . the globule consists of a very dense core of cold gas and dust ( t= 22 k , @xmath423 ) , of small dimensions surrounded by a lower - density envelope ( @xmath424 ) . its mass ( @xmath425 ) is typical of bright - rimmed globules , equally distributed between the core and the envelope ( 27 and @xmath426 respectively ) . the ionization conditions at the surface of the globule were determined from vla and iso ( lws and sws ) observations . the impinging ionizing flux is @xmath427 ; it creates an ionization front and a photoevaporated envelope of density @xmath428 at the surface of the globule . the pdr is traced by the emission of the pahs bands at 6.2 , 7.7 , 8.6 and @xmath9 . the observed variations of the relative intensities can be accounted for by the change in the excitation conditions . we find that the intensity of the @xmath9 band drops outside of the pdr in the photoionized layers , as a consequence of the ionization of the pahs . the @xmath195 band is still detected in the photoionized envelope of the globule , though at a weaker level . despite the high visual extinction of the globule , the @xmath195 pah band , excited in the pdr on the rear side is detected in the body of the globule thanks to a minimum in the absorption of the ices and the dust in this wavelength range . millimeter line observations reveal the kinematical signature of the pdr which precedes the ionization front as a shock moving into the globule . the relative projected velocity is weak , @xmath429 . comparison with models of photoionized globules ( ll94 ) indicates that tc2 has been exposed to the ionizing radiation for @xmath430 , almost as soon as the exciting star of the nebula turned on . the structure of the photon - dominated layer has been derived from the iso far - infrared line and continuum observations . the agreement between all the data and the model of k99 is good . the intensity of the radiation field at the surface is @xmath3 . the pdr has a typical column density of @xmath431 and a molecular hydrogen density @xmath432 . the fuv field heats the dust to a temperature @xmath433 whereas the average gas temperature is found close to @xmath239 . the emission of the @xmath156}}$ ] @xmath157 line in the pdr can be accounted for by assuming a silicon gas phase abundance of @xmath265 , i.e. @xmath434 the solar value . the globule is currently undergoing star formation . our observations show that the globule is forming a low- or intermediate - mass star of luminosity @xmath435 . the protostar powers a herbig - haro jet , which appears fully ionized outside of the globule . no molecular counterpart ( outflow ) to the optical jet has been found . in particular , no sio emission was found . this implies that the source is already in an intermediate stage between class 0 and class i , or even a full class i member . the evolutionary age of the source is therefore typically a few @xmath10 , which suggests that star formation in tc2 probably started in the large burst which accompanied the birth of hd 164492a . as a result of the photoevaporation of the surface layers , the globule has evacuated about half the mass in its envelope until now but we have not found any evidence that the birth process itself has been perturbed . on the contrary , the photoevaporation rate is low enough to leave ample time for protostars to reach safely the ultimate stages of star formation . we acknowledge spanish dges for this supporting research under grants pb96 - 0883 and esp98 - 1351e . we thank j.r . pardo for help with the observations at the cso . lfr is grateful to conacyt , mexico , for its support . this research made use of simbad . abergel , a. , bernard , j.p . , boulanger , f. , et al . , 2002 , , in press allamandola , l.j . , hudgins , d.m . , sandford , s.a . , 1999 , , 511 , l115 andr , p. , ward - thompson , d. , barsony , m. , 2000 , in protostar & planets iv , bachiller , r. , 1996 , ann . astrophys . , 34 , 111 bakes , e.l.o . , tielens , a.g.g.m . , 1994 , , 427 , 822 bakes , e.l.o . , tielens , a.g.g.m . , 1998 , , 499 , 258 bertoldi , f. , 1989 , , 346 , 735 bertoldi , f. , mckee , c.f . , 1990 , , 354 , 529 boulanger , f. , boissel , p. , cesarsky , d. , ryter , c. , 1998 , , 339 , 194 cernicharo , j. , bachiller , r. , duvert , g. , gonzalez - alfonso , e. , gmez - gonzalez , j. , 1992 , , 261 , 589 cernicharo , j. , lefloch , b. , garca - lopez , r.j . , & esteban , c. , 1997 , in `` low mass star formation from infall to outflow '' , pg 8 , iau182 , ed . f. malbet & a. castets . cernicharo , j. , lefloch , b. , cox , p. , cesarsky , d. , esteban , c. , yusef - zadeh , f. , mndez , d.i . , acosta - pulido , j. , garca - lopez , r.j . , heras , a. , 1998 , science , 282 , 462 ( cl98 ) cernicharo , j. , noriega - crespo , a. , cesarsky , d. , lefloch , b. , gonzalez - alfonso , e. , najarro , f. , dartois , e. , cabrit , s. , 2000 , science , 288 , 649 cesarsky , c.j . , abergel , a. , agnese , p. , 1996 , , 315 , l32 chaisson , e.j . , willson , r.f . , 1975 , , 199 , 647 clegg , p.e . , ade , p.a.r . , armand , c. , et al . , 1996 , , 315 , l38 crete , e. , giard , m. , joblin , m. , vauglin , i. , leger , a. , rouan , d. , 1999 , , 352 , 277 de graauw , t. , haser , l.n . , beintema , d.a . , et al . , 1996 , , 315 , l49 dartois , e. , dhendecourt , l. , 1997 , , 323 , 534 dinerstein , h.l . , lester , d.f . , werner , m.w . , 1985 , , 291 , 561 elmegreen , b.g . , lada , c.j . , 1977 , , 214 , 725 gry , c. , swinyard , b. , harwood , a. , et al . , 2001 , in `` lws - the long - wavelength spectrometer '' , iso handbook volume iv ( lws ) , sai/99 - 077/dc , eds . t.g . mller & j. blommaert habing , h.j . , 1968 , bull . netherlands , 19 , 421 heiles , c. , 1994 , , 436 , 720 hildebrand , r.h . , 1983 , qjras , 24 , 267 hollenbach , d.j . , takahashi , t. , tielens , a.g.g.m . , 1991 , , 377 , 192 joblin , c. , tielens , a.g.g.m . , geballe , t.r . , wooden , d.h . , 1996 , , 460 , l119 kaufman , m.k . , wolfire , m.g . , hollenbach , d.j . , luhmna , m.l . , 1999 , , 527 , 795 ( k99 ) kessler , m.f . , steinz , j.a . , anderegg , m.e . , 1996 , , 315 , l27 leech , k. , de graauw , t. , van den ancker , m. , et al . , 2001 , in `` sws - the short - wavelength spectrometer '' , iso handbook volume vi ( sws ) , sai/2000 - 008/dc , eds , k. leech , t.g . mller & j. blommaert lefloch , b. , lazareff , b. , 1994 , , 289 , 559 ( ll94 ) lefloch , b. , lazareff , b. , 1995 , , 301 , 522 lefloch , b. , lazareff , b. , castets , a. , 1997 , , 324 , 249 lefloch , b. , castets , a. , cernicharo , j. , loinard , l. , 1998 , , 1998 , 504 , l109 lefloch , b. , cernicharo , j. , 2000 , , 545 , 340 lefloch , b. , cernicharo , j. , cesarsky , d. , demyk , k. , miville - deschnes , m.a . , and rodriguez , l.f . , 2001 , , 368 , l13 lefloch , b. , pardo , j.r . , cernicharo , j. , 2002 , in prep . lester , d.f . , dinerstein , h.l . , werner , m.w . , watson , d.m . , genzel , r.l . , 1983 , , 271 , 618 lester , d.f . , dinerstein , h.l . , werner , m.w . , watson , d.m . , genzel , r. , and storey , j.w.v . , 1987 , , 320 , 573 lynds , b. , canzian , b.j . , and oneil , e.j . jr . , 1985 , , 288 , 164 mathis , j.s . , 1990 , , 28 , 37 mauersberger , r. , gulin , m. , martin - pintado , j. , thum c. , cernicharo , j. , hein , h. , navarro , s. , 1989 , supp . series , 79 , 217 miville - deschnes , m.a , boulanger , f. , abergel , a. , bernard , j.p . , 2000 , a&as , 146 , 519 miville - deschnes , m.a , boulanger , f. , joncas , g. , falgarone , e. , 2002 , , 381 , 209 panagia , n. , 1973 , aj , 78 , 929 puget , j.l . , lger , a. , 1989 , ann.rev . , 27 , 161 reipurth , b. , 1983 , , 117 , 183 reipurth , b. , bally , b. , fesen , r.a . , devine , d. , 1998 , nature , 396 , 343 rho , j. , corcoran , m.f . , chu , y.h . , reach , w.t . , 2001 , , 562 , 446 rosado , m. , esteban . c. , lefloch , b. , cernicharo , j. , garca - lpez , r.j . , 1999 , aj , 118 , 2962 rubin , r.h . , 1985 , supp . 57 , 349 rubin , r.h . , 1989 , supp . 69 , 897 rubin , r.h . , simpson , j.p . , haas , m. , erickson , e.f.,1991 , , 374 , 564 rubin , r.h . , simpson , j.p . , lord , s.d . , colgan , s.w.j . , erickson , e.f . , haas , m.r . , 1994 , , 420 , 772 salama , a. , 2000 , in `` iso beyond point sources : studies of extended infrared emission '' , madrid , spain , esasp-455 simpson , j.p . , colgan , s.w.j . , rubin , r.h . , erickson , e.f . , haas , m.r . , 1995 , , 444 , 721 spitzer , l. , 1978 , _ physical processes in the interstellar medium _ , ( new - york : wiley ) , 107 sugitani , k. , fukui , y. , ogura , k. , 1991 , supp , 77 , 59 tielens , a.g.g.m . , hollenbach , d. , 1985 , , 291 , 722 vastel , c. , spaans , m. , ceccarelli , c. , tielens , a.g.g.m . , caux , e. , 2001 , , 376 , 1064 verstraete , l. , puget , j.l . , falgarone , e. , drapatz , s. , wright , c.m . , timmermann , r. , 1996 , , 315 , 337 walborn , n.r . , 1973 , , 78 , 1067 walmsley , c.m . , pineau des forts , g. , flower , d.r . , 1999 , , 342 , 542 white , g.j . , nelson , r.p . , holland , w.s . , et al . , 1999 , , 342 , 233 whitworth , a.p . , bhattal , a.s . , chapman , s.j . , 1994 , mnras , 268 , 291 wolfire , m.g . , hollenbach , d. , tielens , a.g.g.m . , 1989 , , 344 , 770 wolfire , m.g . , tielens , a.g.g.m . , hollenbach , d. , 1990 , , 358 , 116 fig . 1 schematic drawing of the tc2 region : ( i ) the hii region between the exciting star ( hd 164492a ) and the bright - rimmed globule ; ( ii ) the ionization front ( if ) and the expanding envelope of photoionized gas around the globule ; ( iii ) the photon - dominated region ( pdr ) and the shock ahead of the if ; ( iv ) the molecular core of tc2 . [ fig1 ] fig . 2 [ @xmath124 ] image of the trifid nebula observed with the not telescope ( cl98 ) . the location of bright - rimmed globule tc2 is marked by a white rectangle . the positions observed with iso / lws and the telescope beams are indicated by circles . [ fig2 ] fig . 3 thermal dust emission of tc2 observed at 1.25 mm with the mpifr 19 channel bolometer array ( thin white contours ) superposed on the optical @xmath437}}$ ] emission of the region . contours range from 10% to 90% of the peak flux ( @xmath438 beam ) . in thick white contours , we have superposed the mid - ir emission , as observed with the isocam lw10 filter ( @xmath439 ) . first contour and interval are 160 and 20 mjy / sr respectively . in thin black contours is superposed the 3.6 cm free - free emission observed at the vla . contours are 0.4 , 0.8 , 1.1 , 1.5 , 1.9 mjy / beam . fig . 4 magnified view of tc2 as observed in the [ sii ] lines @xmath440 . the white circle marks the tc2 region encompassed with the lws beam . the rectangles mark the regions observed with iso / sws , and delineate the aperture of the detector ( band 4 ) . the center of the sws fields are marked by stars . [ fig3 ] fig . 5 ( a ) contour map of the free - free emission at 20 cm obtained with the vla ( cl98 ) , superposed on the optical [ sii ] image of the globule . first contour and contour interval are @xmath441 and @xmath442 jy / beam respectively . + b ) contour map of the free - free emission at 3.6 cm obtained with the vla , superposed on an [ sii ] image of the head of bright - rimmed globule tc2 ( greyscale ) . the first contour and contour interval are @xmath443 and @xmath444 jy / beam respectively . [ fig5 ] fig . 7 mid - infrared emission as detected with the cvf in the pah bands at 6.2 , 7.7 , 8.6 and @xmath9 ( greyscale and contours ) , in the @xmath86}}$ ] @xmath445 ( with the pah @xmath200 band ) and the @xmath59}}$ ] @xmath446 line and in the continuum between 13 and @xmath412 ( greyscale ) . we have superposed ( contours ) the pah @xmath9 band emission on the panels of the @xmath445 , @xmath59}}$ ] , and @xmath447 emissions . the location of the spectra shown in fig . 9 is marked with white squares . [ fig6 ] fig . 8 cvf emission in the range @xmath448 at a few positions towards tc2 . the locations are marked in the lower right panel of fig . ( a ) _ and _ ( b ) _ : photoionized region ( positions a and b respectively ) . _ ( c ) _ and _ ( d ) _ : photon - dominated region ( at c and d resp . ) ; _ ( e ) _ : reference position ( position e ) ; _ ( f ) _ : core of tc2 ( position f ) . in panels _ a)-d ) _ , we have superposed ( thin contour ) the spectrum of the reference position scaled to match the intensity of the @xmath195 band . [ fig8 ] fig . 10 large - velocity gradient analysis of the @xmath4}}$ ] lines observed with the lws . isoflux contours in the temperature - density diagram for the @xmath4}}$ ] @xmath5 ( solid ) and @xmath6 ( dashed ) . for three oxygen column densities . for each line , we have also indicated the upper and lower limits at @xmath67 . [ fig10 ] fig . 12 spectral energy distribution of tc2 after subtracting the extended emission of the nebula . we have indicated ( square ) the @xmath449 flux ( integrated over the iso / lws beam ) . we show in dashed the fits to the warm and cold dust component . the fit to the whole emission is indicated by the solid line . [ fig12 ] fig . 13 montage of the molecular lines detected in the core of tc2 . the line fluxes are given in the antenna temperature scale . the dashed line indicates the systemic velocity of the dense core @xmath335 . [ fig13 ] fig . 14 contour maps of the velocity - integrated emission @xmath336 , cs @xmath36 , @xmath450 , @xmath451 , cs @xmath322 and @xmath451 . the emission was integrated between 5 and @xmath452 . contours range 0.1 , 0.2 ... to 0.9 times the emission peak . the contours are superposed on a 1.3 mm continuum emission map of the globule . [ fig14 ] fig . 15 molecular gas emission in the cs @xmath36 and @xmath336 lines in a cut across the photon - dominated region in north - south direction . the dashed lines indicate the systemic velocity of the main body gas in tc2 ( @xmath335 ) and of the second kinematic component detected in the pdr ( @xmath453 ) . [ fig15 ] i ( @xmath68 region ) , & vla & free - free radiation & 3.6 cm , 20 cm & 5a , 5b , 16 + ii ( bright rim ) & not & @xmath124 , @xmath396}}$ ] & optical & 2 , 4 + & sws , cvf , lws & atomic lines : & @xmath454 & 6,7 + & & @xmath86}}$ ] , @xmath59}}$ ] , @xmath78}}$ ] , @xmath72}}$ ] , & & + & & @xmath70}}$ ] , @xmath156}}$ ] , @xmath7}}$ ] & & + & sws & continuum ( dust ) & @xmath455 & 11 + & & & & + iii ( pdr ) & isocam & lw10 & @xmath439 & 3 + & cvf & pah bands & @xmath186 & 7,8 + & sws , lws & continuum ( dust ) & @xmath454 & 11 , 12 + & sws , lws & atomic lines : & @xmath454 & 6,7 + & & @xmath4}}$ ] , @xmath7}}$ ] , @xmath156}}$ ] & & + & & & & + iv ( molecular core ) & lws & continuum ( dust ) & @xmath456 & 12 + & iram-30 m & continuum ( dust ) & 1.25 mm & 3 + & & co , @xmath38 , @xmath39 , & 1 - 3 mm & 13 , 14 , 15 + & & @xmath235 , sio , cs & & + & cso & co & 0.8 mm & 13 + llrcr + & line & frequency & beamwidth & @xmath457 + & & ghz & @xmath458 & + @xmath459 & @xmath37 & 86.75429 & 28 & 0.77 + sio & @xmath36 & 86.84700 & 28 & 0.77 + @xmath235 & @xmath37 & 89.18852 & 27 & 0.75 + cs & @xmath36 & 97.98097 & 24 & 0.71 + @xmath39 & @xmath37 & 109.78218 & 22 & 0.68 + @xmath38 & @xmath37 & 110.20135 & 22 & 0.68 + co & @xmath37 & 115.27120 & 21 & 0.67 + sio & @xmath322 & 130.26870 & 18 & 0.58 + cs & @xmath322 & 146.96905 & 16 & 0.53 + sio & @xmath460 & 217.10494 & 11 & 0.42 + co & @xmath36 & 230.53800 & 10 & 0.39 + co & @xmath322 & 345.79599 & 22 & 0.75 + lcccrrc + line & band & @xmath461 & aperture & @xmath462 & @xmath463 & @xmath464 + & & @xmath465 & @xmath466 & @xmath467 & @xmath467 & @xmath467 + @xmath86}}$ ] & 3a & 12.8 & @xmath468 & @xmath469 & @xmath470 & ( @xmath471 + @xmath59}}$](@xmath472 ) & 3a & 15.5 & @xmath468 & @xmath473 & @xmath470 & _ + @xmath78}}$ ] & 3c & 18.7 & @xmath468 & @xmath474 & @xmath474 & @xmath470 + @xmath78}}$ ] & 4 & 33.5 & @xmath475 & @xmath476 & @xmath477 & @xmath478 + @xmath156}}$ ] & 4 & 34.8 & @xmath475 & @xmath479 & @xmath479 & @xmath480 + lrccccc + line & band & beam size & @xmath461 & @xmath481 & @xmath482 & tc2 ( `` on '' - `` off '' ) + & & ( ) & @xmath465 & @xmath467 & @xmath467 & @xmath467 + @xmath70}}$ ] & sw2 & 84 & 51.7 & @xmath483 & ( @xmath484 & 4.6 ( -18 ) + @xmath72}}$ ] & sw2 & 84 & 57.2 & @xmath485 & ( @xmath486 & 2.7 ( -18 ) + @xmath4}}$ ] & sw3 & 86 & 63.3 & @xmath487 & @xmath488 & 7.2 ( -18 ) + @xmath70}}$ ] & sw5 & 80 & 88.4 & @xmath489 & @xmath490 & 5.0 ( -18 ) + @xmath491}}$ ] & lw2 & 68 & 122.0 & @xmath492 & @xmath493 & 6.3 ( -19 ) + @xmath4}}$ ] & lw3 & 70 & 145.5 & @xmath494 & @xmath495 & 4.3 ( -19 ) + @xmath7}}$ ] & lw4 & 68 & 157.8 & @xmath487 & @xmath496 & 2.5 ( -18 ) + ll + fuv radiation field ( @xmath111 ) & 1000 + infrared luminosity & @xmath497 + total mass ( @xmath472 ) & @xmath498 + mean radius @xmath125 & @xmath499 + mean @xmath500 column density & @xmath314 + & + core & + dimensions ( @xmath501 cm ) & @xmath502 + dust temperature & 22 k + maximum @xmath500 column density ( @xmath472 ) & @xmath503 + mean @xmath500 column density ( @xmath472 ) & @xmath504 + mass & @xmath1 + @xmath500 density & @xmath505 + central source luminosity & @xmath506 + & + molecular envelope & + @xmath500 density ( @xmath38 ) & @xmath507 + gas temperature & @xmath508 + & + photon - dominated region & + thickness & @xmath509 + dust temperature & 46 k + atomic gas temperature ( oi ) & 300 k + gas column density ( @xmath472 ) & @xmath510 + oxygen column density & @xmath511 + oxygen abundance & @xmath512 + carbon column density & @xmath513 + carbon abundance & @xmath253 + density ( @xmath500 ) & @xmath514 + atomic gas mass ( cii ) & @xmath515 + velocity field & radiatively - driven implosion + & + photoionized envelope ( bright rim ) & + thickness & @xmath516 + electron density & @xmath517 +	we have carried out a comprehensive multiwavelength study of bright - rimmed globule tc2 in the trifid nebula , using the iram 30 m telescope , the vla centimeter array and the infrared space observatory ( iso ) . tc2 is one of the very few globules to exhibit signs of active ongoing star formation while being photoevaporated by the ly - c flux of the exciting star of the nebula ( @xmath0 ) . the globule consists of a cold dense core of mass @xmath1 surrounded by a lower density envelope of molecular gas . the impinging ly - c photons induce the propagation of an ionization front into the globule . the evaporation of the ionized gas forms a thin layer of density @xmath2 around the globule , which could be mapped with the vla . the globule is illuminated mainly on its rear side , by a fuv field of intensity @xmath3 . it creates a photon - dominated region ( pdr ) below the surface , which was mapped and characterized with the isocam circular variable filter and the short wavelength spectrometer ( sws ) onboard the infrared space observatory . the physical conditions derived from the analysis of the far - infrared lines @xmath4}}$ ] @xmath5 , @xmath6 and @xmath7}}$ ] @xmath8 , and the continuum emission are in good agreement with some recent pdr models . the emission of the pahs band at 6.2 , 7.7 , 8.6 and @xmath9 is detected over the whole globule . the relative intensity variations observed across the globule , in the pdr and the photoionized envelope , are consistent with the changes in the ionization fraction . in the head of tc2 , we find a second kinematic component which is the signature of the radiatively - driven collapse undergone by the globule . this component indicates that the pdr propagates at low velocity inside the body of tc2 . the molecular emission suggests that the star formation process was probably initiated a few @xmath10 ago , in the large burst which led to the formation of the nebula . the globule has already evaporated half the mass of its envelope . however , the ionization timescale of the globule is long enough ( @xmath11 ) to let the newly born object(s ) reach smoothly the ultimate stages of protostellar evolution . the impact of photoionization on the star formation process appears limited .
one of the great mysteries of the universe today is the nature of dark energy which drives the late time cosmic acceleration . this has been confirmed by number of observational results including type - ia supernovae @xcite , cosmic microwave background radiation ( cmbr ) ( komatsu et al . ( 2011 ) ) as well as the latest surveys of the large scale structure ( eisenstein et al . ( 2005 ) ) . there has been wide range of contrasting proposals to explain this late time cosmic acceleration . most popular of these is to include a dark candidate with a large negative pressure ( also known as dark energy ) in the energy budget of the universe ( bean et al . ( 2005 ) , copeland et al . ( 2006 ) , li et al . ( 2011 ) , padmanabhan(2003 ) , peebles & ratra ( 2003 ) , sahni & starobinsky ( 2000 ) ) . the simplest candidate for this dark energy is a cosmological constant @xmath0 . however , the well - known fine tuning and coincidence problems render @xmath0 a rather unattractive proposal from a theoretical point of view . going beyond @xmath0 , scalar field models with generic features are perhaps the simplest alternatives to a cosmological constant . a large class of scalar field models including quintessence ( ratra & peebles ( 1988 ) , caldwell et al . ( 1998 ) , liddle & scherrer ( 1999 ) , steinhardt et al . ( 1999 ) ) , tachyon ( abramo & finelli ( 2003 ) , bagla et al . ( 2003 ) , aguirregabiria & lazkoz ( 2004 ) , copeland et al . ( 2005 ) ) phantom ( caldwell ( 2002 ) ) and k - essence ( armendariz et al . ( 2001 ) , scherrer ( 2004 ) , sen ( 2006 ) , vikman ( 2005 ) ) have been thoroughly investigated in recent years to explain the late time acceleration of the universe . the advantage of these scalar field models is that not only can they alleviate the fine tuning and coincidence problem , they can also mimick a cosmological constant at present epoch . these models have equation of state as a function of redshifts . this feature is important to distinguish these scalar field models from cosmological constant as large amount observational data from higher redshifts are coming in or expected to come in near future . the simplest scalar field is the one with linear potential having a canonical kinteic energy . in this case , the scalar field is frozen initially due to large hubble damping and behaves like a cosmological constant with @xmath2 . as the universe expands and hubble damping decreases , the field starts rolling and the equation of state starts deviating away from @xmath3 . but because the potential has no minima , the model gives a collapsing universe in future which results in a finite history for our universe ( kratochvil et al . ( 2004 ) ) . more complicated scalar field models can be divided in to two broad classes : the fast roll and slow roll models , also termed as freezing / tracking and thawing models ( caldwell & linder ( 2005 ) ) . the fast roll models have steep potentials allowing the scalar field to mimick the background matter / radiation , and to remain subdominant for most of the history of the universe . only at late times , the field becomes dominant and starts behaving like a component with negative pressure driving the acceleration of the universe . on the other hand , slow roll or thawing models are similar to the inflaton field , where the potential is sufficiently flat . due to this , kinetic energy of the field is much smaller than the potential energy and field is initially frozen due to large hubble damping in the early universe . that is why initially @xmath4 , and the energy density of the scalar field is nearly constant and have a negligible contribution to the total energy budget of the universe . but as the background radiation / matter gets diluted due to the expansion of the universe , the background energy density becomes comparable to the scalar field energy density . at this moment , the scalar field starts thawing and slowly evolves away from the frozen @xmath4 state towards a higher value of @xmath5 . however , in this case some degree of fine tuning is needed in order to get a suitable late time evolution . one of the early studies for thawing models was by alam et al . ( alam et al . ( 2003 ) ) where they studied the decaying dark energy models which have thawing type cosmological evolution . recently , scherrer and sen ( scherrer & sen ( 2008a))have studied the thawing scalar field models with a nearly flat potential satisfying the slow roll conditions . they have shown that under such conditions , thawing models with variety of potentials evolve in a similar fashion and results in a unique analytic expression for the equation of state @xmath6 . this result was later extended to the case of phantom models ( scherrer & sen ( 2008b ) ) and tachyon scalar field models ( ali et al . ( 2009 ) ) . in another work , thawing models ( for both canonical and non canonical scalar field ) have been studied without the assumption of the slow roll condition ( sen et al . ( 2010 ) ) . in a more recent work , dutta and scherrer ( dutta & scherrer ( 2011))have studied the slow roll freezing scalar field models and derived an analytical expression for the equation of state @xmath5 as a function of the energy density of the scalar field @xmath7 . there is another elegant way to explain the present cosmic acceleration . this is achieved through the large scale modification of the gravitational sector of the theory . these models include dgp model ( dvali et al . ( 2000 ) ) , cardassian model ( freese & lewis ( 2002 ) ) , modified gcg model ( barreiro & sen ( 2004 ) ) , f(r ) theories ( starobinsky ( 2007 ) , hu & sawicki ( 2007),amendola et al . ( 2007 ) ) , galileon models ( nicolis et al . ( 2009 ) , chow & khoury ( 2009 ) ) as well as models having non local gravitational interactions ( deser & woodard ( 2007 ) ) . with current obsevational data , the concordance @xmath0cdm indeed appears to provide a very good fit to the present observations . hence , for any given dark energy model ( different from @xmath0cdm ) , it is natural to ask how to distinguish it from @xmath0cdm . keeping this in mind , different diagnostic measures have been proposed . @xmath8 diagnostic and the statefinders are two elegant examples for such diagnostic measures . recently , another interesting diagnostic measure , ` _ statefinder hierarchy _ , @xmath9 ' ( with @xmath10 ) , has been introduced by arabsalmani and sahni ( arabsalmani & sahni ( 2011 ) ) . it was shown that while all the @xmath9 s remain fixed at unity during the entire evolution of the universe for concordance @xmath0cdm , they are in general time dependent for dark energy models . hence they can be used not only to compare different dark energy models with @xmath0cdm but also to compare different theoretical approaches to explain present cosmic acceleration . in the present investigation , we consider thawing dark energy models involving canonical scalar fields . we do not assume the slow roll conditions on the potential of the field . we first use the statefinder diagnostic @xmath9 to compare different thawing models with other dark energy models as well as with modified gravity proposal such as dgp model . next , we use the latest observational results involving background cosmology to put constraints on the deviations of the thawing models from the @xmath0cdm behaviour . we also forecast constraints from future observations . we assume that dark energy is decsribed by a minimally coupled scalar field @xmath11 with canonical kinetic energy , governed by the equation of motion @xmath12 where @xmath13 is the hubble parameter determined by the einstein equation @xmath14 we work in the natural unit @xmath15 . here , @xmath16 is the non - relativistic matter ( pressure @xmath17 ) content of the universe and the @xmath18 is the energy density of the scalar field . we assume a spatially flat universe , so that @xmath19 . with this , one can form an autonomous system using equations ( [ sf ] ) and ( [ hub ] ) ( see scherrer & sen ( 2008a ) ) for detail derivation ) : @xmath20 here @xmath21 ( @xmath22 being the equation of state for the scalar field ) , and @xmath23 versus @xmath24 . larger value of @xmath25 corresponds to the distant past ( @xmath26 ) whereas , the smaller value @xmath27 corresponds to the future . @xmath28 for this plot . the different * solid lines * are ( right to left ) : thawing quintessence model with @xmath29 , @xmath30 , @xmath31 , @xmath32 , @xmath33 respectively . the * dashed lines * corresponds to cg ( 1 ) , gcg ( 2 ) , dgp ( 3 ) , cpl ( 4 ) , @xmath34 ( 5 ) . the region between two vertical solid lines roughly corresponds the present epoch.,width=302 ] .,width=302 ] given the initial conditions for @xmath35 , @xmath7 and @xmath36 , one can solve the above system for any potential @xmath37 ( the information about the form of the potential is encoded in @xmath38 ) . as our goal is to study the thawing models , the scalar field is initially frozen due to large hubble damping ( i.e @xmath39 ) . hence @xmath40 initially which gives @xmath41 as initial condition . we shall not assume slow - roll conditions for the potential @xmath37 . hence , the initial value @xmath42 is a parameter in our model . this initial value of @xmath42 determines the deviation of @xmath22 from the initial @xmath43 frozen state as the universe evolves . for smaller @xmath42 , the deviation is negligibly small , and the scalar field behaves as a cosmological constant at all time . for larger values of @xmath42 , @xmath22 deviates substantially from @xmath44 as the universe evolves . in general , the contribution of scalar field to the total energy density is negligible in the early universe . nevertheless one has to fine tune the initial value of @xmath7 in order to have its correct contribution at present . hence @xmath45 is related to the @xmath46 . with this we evolve our system from the intial era @xmath47 till any redshift we want . we consider various power - law potentials for which @xmath38 is a constant . in recent past , clemson and liddle have considered thawing models using a multiparameter extension of the exponential potential ( clemson & liddle ( 2009 ) ) . one can taylor expand the scale factor of the universe around the present era ( @xmath48 ) as follows : @xmath49^n , \label{state}\ ] ] where , @xmath50 it is easy to check that @xmath51 is the deceleration parameter . similarly @xmath52 is related to the statefinder @xmath53 or jerk @xmath54 and @xmath55 is related to the snap @xmath56 and so on . using the equations ( [ state ] ) and ( [ alpha ] ) , one can now define the statefinder hierarchy @xmath9 ( arabsalmani & sahni ( 2011 ) ) as : @xmath57 it is remarkable to see that for @xmath0cdm , @xmath58 during the entire history of the universe . in the rest of this section we use different combinations of @xmath9 , to study the evolution of thawing models and compare them with other popular dark energy and modified gravity models.we consider the following models : * dark energy with a constant equation of state , @xmath59 . * dark energy with a cpl parametrized equation of state ( chevallier & polarski ( 2001 ) , linder ( 2003 ) ) , @xmath60 , where @xmath61 and @xmath62 are constants . * dark energy with a generalized chaplygin gas ( bento et al . ( 2002 ) ) type equation of state , @xmath63 , where @xmath64 and @xmath65 are constant . @xmath66 gives the standard chaplygin gas . * dvali , gabadadze and porrati ( dgp ) model ( dvali et al . ( 2000 ) ) . .,width=302 ] in fig.1 , we show the behaviours of the models in the @xmath67 phase plane . the figure shows that in this phase plane which involves the second derivative of the scale factor , the models are highly degenerate around the present value of @xmath68 . as we go to the higher order @xmath9 s , one can remove these degeneracies among various models . this is revealed in the subsequent figures , fig . 2 and fig . 3 where we show the phase portrait in the @xmath69 and @xmath70 planes respectively . 4 we show the phase portrait in @xmath71 plane where again the degeneracies among different models are broken . we should comment on @xmath72 in particular . from fig . 2 and fig . 4 , one can see that @xmath72 can be used to break the degeneracies among the thawing models themselves as its present day value varies substantially for different thawing potentials . similarly , @xmath73 lets us differentiate between thawing and other models . we should mention that although we assume @xmath74 for these figures , the conclusions remain same for other values of @xmath1 . . dashed lines are same as fig 1 . the solid lines ( top to bottom ) : thawing quintessence model with @xmath29 , @xmath30 , @xmath31 , @xmath32 , @xmath33 respectively . @xmath74 for this plot . the point ( 1,1 ) is @xmath75cdm.,width=302 ] plane at redshifts @xmath76 _ _ ( solid)__and @xmath77 _ ( dashed)_.the dots _ ( black ) _ represents the potentials @xmath78 ( from top to bottom , for each z)respectively.,width=302 ] with future data , one can probe the background expansion at higher redshifts thereby probing the higher derivatives of the scale factor . hence statefinder hierarchy will be quite useful to distinguish various dark energy models . we further study the @xmath79 phase plane for different thawing dark energy models . here ` prime ' denotes the differentiation with respect to @xmath80 . as discussed in the previous section , for smaller values of @xmath42 , all the thawing models behave similar to the cosmological constant . so to study the @xmath79 phase plane , we assume @xmath81 for which the thawing models behave significantly different from cosmological constant . linder and caldwell have previously showed that the thawing models are constrained to lie between @xmath82 region in @xmath79 phase plane . in fig . 5 and fig . 6 , we show different thawing models in the @xmath79 phase plane at different redshifts and compare them with the constraint proposed by linder and caldwell ( caldwell & linder ( 2005 ) ) ( lc ) . it can be seen that for higher redshifts , all the thawing models are inside the lc bound . this is obvious as all the thawing models behave very close to cosmological constant for larger redshifts and hence satisfy the lc bound . but at the present epoch ( z=0 ) , for smaller values of @xmath1 , thawing models with certain potentials ( as shown in figure 5 ) are outside the lc bound . but as one increases the value of @xmath1 ( see figure 6 ) , they tend to satisfy the lc bound . hence thawing models which deviate significantly from the @xmath0cdm behaviour at smaller redshift tend to satisfy the lc bound for higher values of @xmath1 whereas for lower values of @xmath1 , some of the potentials do not satisfy the lc bound . , in this section , we constrain the parameters in the thawing scalar field models with the assumption of a flat universe by using the latest observational data including the type-1a supernovae union2 compilation ( amanullah et al . ( 2010 ) ) , the bao ( eisenstein et al . ( 2005 ) ) measurement from the sdss ( percival et al . ( 2007 ) , percival et al . ( 2010 ) ) , the cmbr measurement given by wmap7 ( komatsu et al . ( 2011 ) ) observations , the h(z ) data from hst key project ( freedman et al . ( 2001 ) ) and a simulated dataset ( holsclaw et al . ( 2010 ) ) based on the upcoming jdem sn - survey containing around 2300 type-1a supernovae . we consider the supernovae type ia observation ( amanullah et al . ( 2010 ) which is one of the direct probes for the late time acceleration . it measures the apparent brightness of the supernovae as observed by us which is related to the luminosity distance @xmath83 defined as @xmath84 with this , we construct the distance modulus @xmath85 which is experimentally measured @xmath86 where m and m are the apparent and absolute magnitudes of the supernovae which are logarithmic measure of flux and luminosity respectively . and 2@xmath87 confidence contours in @xmath36i - @xmath1 plane for @xmath88.the shaded regions are constraints from sn+bao data while the thick lines are constraints from sn+bao+cmb+h(z ) data respectively . the dashed lines are for the simulated jdem data . the two vertical lines represent the wmap7 bound on @xmath1,width=302 ] .,width=302 ] another observational probe that has been widely used in recent times to constrain dark energy models is related to the data from the bao distance measurements obtained at @xmath89 and @xmath90 from the joint analysis of the 2dfgrs and sdss data ( eisenstein et al . ( 2005 ) , percival et al . ( 2007 ) , percival et al . ( 2010 ) ) . in this case , one needs to calculate the parameter @xmath91 which is related to the angular diameter distance as follows @xmath92 ^{1/3}\ ] ] for bao measurements we calculate the ratio @xmath93 this ratio is a relatively model independent quantity and has a value @xmath94 . the cmb is sensitive to the distance to the decoupling epoch via the locations of peaks and troughs of the acoustic oscillations . we employ the `` wmap distance priors '' given by the five - year wmap observations.this includes the``acoustic scale '' @xmath95,the `` shift parameter '' r and the redshift of the decoupling epoch of photons @xmath96 the acoustic scale @xmath95 describes the distance ratio @xmath97 @xmath98 where @xmath99 factor arises because @xmath100 is the proper angular diameter distance , where @xmath101 is the comoving sound horizon at @xmath96 we use the fitting function of @xmath96 proposed by hu and sugiyama ( hu & sugiyama ( 1996 ) ) @xmath102[1+g_{1}(\omega_{m}h^2)^{g_{2}}]\ ] ] @xmath103 the shift parameter r ( komatsu et al . ( 2011 ) ) is responsible for the distance ratio @xmath104 given by @xmath105 the constraint on the shift parameter @xmath106 from the wmap observations is quoted as @xmath107 . we use new determinations of the cosmic expansion history from red - envelope galaxies . stern et al . ( stern et al . ( 2010 ) ) have obtained a high - quality spectra with the keck - lris spectrograph of red - envelope galaxies in 24 galaxy clusters in the redshift range @xmath108 . they complemented these keck spectra with high - quality , publicly available archival spectra from the spices and vvds surveys.with this , they presented 12 measurements of the hubble parameter h(z ) at different redshift . the measurement at z = 0 is from hst key project @xcite . .,width=302 ] .,width=302 ] to look at what can be achieved in future , we also use a simulated dataset ( holsclaw et al . ( 2010 ) ) based on the upcoming jdem sn - survey containing around 2300 sne . these are distributed over a redshift range from @xmath109 to @xmath110 . we assume a simplified error model where the errors are same for all supernovae and independent of redshift . we assume a statistical error of @xmath111 magnitude , as expected from jdem - like future surveys ( aldering et al . ( 2004 ) ) . we use a @xmath0cdm model with @xmath112 km / s / mpc and @xmath113 to simulate the data . + thawing models are described by three parameters , namely @xmath42 , @xmath114 and @xmath115 . as mentioned in the previous sections , @xmath42 describes the deviation of the thawing model from @xmath0cdm ( @xmath116 represents the @xmath0cdm model ) , and @xmath38 encodes the information about the potential . hence constraining these two parameters essentially constrain the deviation of different thawing models from @xmath0cdm . in figures 7 , 8 , 9 and 10 , we show the constrained region in @xmath117 plane for different potentials . in doing this , we use the data coming from different recent observations e.g snia , bao , cmb as well as h(z ) measurements . we also use the simulated data for the upcoming sn - surveys like jdem to constrain the parameter space . in these figures , we also show the wmap7 bound on @xmath1 . from the figures it is evident that when we consider the current data , there is always an upper bound on @xmath42 . this upper bound is most prominent for the linear potential . hence this gives strong constraint on deviation from the @xmath0cdm behaviour for the linear potential . for exponential or inverse squared potential , this deviation is not strongly constrained and @xmath42 as large as 1 is allowed even at @xmath118 confidence level . moreover if one takes into account the wmap7 bound on @xmath1 , current data always allow the @xmath0cdm behaviour ( @xmath119 ) for all the potentials . but this changes as one does similar analysis with the simulated data for future sn - surveys like jdem ( holsclaw ( 2010 ) ) . with this , considering the wmap7 bound on @xmath1 , although @xmath0cdm ( @xmath119 ) is allowed for some values of @xmath1 , for other values of @xmath1 , @xmath42 has a lower bound different from zero . this clearly rules out @xmath0cdm even at @xmath118 confidence level . for example , consider the constraints obtained from simulated jdem dataset in case of linear potential ( fig.7 ) . the contours are shown by the dashed line . for @xmath120 ( which is allowed by the wmap7 constraint on @xmath1 ) , the lower bound on @xmath42 is 0.75 at @xmath118 confidence level . similarly , in case of an exponential potential ( fig.10 ) , the corresponding lower bound on @xmath42 for the same @xmath1 is 0.85 at @xmath118 confidence level . this rules out the @xmath0cdm even at @xmath118 confidence level . for all the potentials that we considered , one can rule out @xmath0cdm at @xmath118 confidence level for some values of @xmath1 using simulated data for future - sn survey like jdem . although we study a generic class of dark energy , e.g the thawing models , this shows the advantage of the large scale surveys like jdem to distinguish different dark energy models from @xmath0cdm which is difficult with the presently available data . next we study how effective is the simulated data in distinguishing different dark energy models using the _ statefinder hierarchies_. in section 2 , we already mentioned that @xmath72 is the most promising parameter to distinguish different dark energy models . keeping this in mind , we calculate the bound on the present day value of @xmath72 , using the simulated jdem data for different values of @xmath1 . in figure 11 , the regions inside the horizontal lines show this bound at @xmath118 confidence level . this bound on @xmath72 is valid for all thawing models with power law potentials and is independent of the power law form . as one can see , the bound becomes narrower as one increases the @xmath1 . moreover , it is clear that as one decreases the value of @xmath1 , e.g. @xmath121 , @xmath122 is outside the @xmath118 allowed region ruling out the @xmath0cdm . but for higher values of @xmath1 , @xmath0cdm is allowed . this is consisnent with our earlier conclusions from fig 7 - fig 10 . in the same figure , we also show the behaviour of @xmath123 as a function of @xmath42 for different potentials . for @xmath121 , although @xmath0cdm is ruled out , but all the thawing models more or less are inside the allowed region of @xmath72 making them difficult to distinguish . as one increases the @xmath1 , models with higher @xmath42 start going out of the allowed region of @xmath72 . this helps to distinguish different thawing models with higher @xmath42 . we should emphasize the fact that for smaller @xmath42 , all the thawing models have exactly similar behaviour which is same as @xmath0cdm . it is only for higher @xmath42 , the different thawing models behaves differently . we look at observational constraints on the thawing scalar field models which have been proposed to explain late time cosmic acceleration . we first study the degeneracies of different thawing models with other popular dark energy models using the recently proposed ` statefinder hierarchy @xmath9 ' . we show that as one starts probing the higher order @xmath9 , different thawing models not only become distinguishable among themselves , they also behave significantly different from other dark energy models including @xmath0cdm . with expected future data at significantly higher redshifts , this can be a smoking gun to distinguish different dark energy models . we also study the behaviour of different thawing models in the @xmath79 phase plane and compare those with the previously obtained lc bound . using various observational data related to background evolution of the universe , we constrain the deviation of the thawing model from @xmath0cdm . with current precision level of the observational data ( at least for the background evolution ) , the models can not be distinguished from the @xmath0cdm . but with simulated data for large future surveys like jdem , we show that for some values of @xmath1 , the data may distinguish thawing models with @xmath0cdm . moreover , with future data , the @xmath72 parameter may be quite useful in distinguishing different thawing models among themselves for certain plausible values of @xmath1 . as a possible extension of this work , it will be really interesting to do a similar study with freezing type of scalar fields . as we mention in the introduction , dutta and scherrer ( dutta & scherrer ( 2011 ) ) have recently studied the slow - roll freezing scalar field models . in such a setup , one can calculate the ` statefinder hierarchy ' and compare them with those for the thawing models . with future high precision data , such study can be useful to check which kind of scalar field model is actually preferred by the data . this will be our future goal . aas acknowledges the financial support provided by the serc , dst , govt . of india , through major research project grant ( grant no : sr / s2/hep-43(2009 ) ) and the hospitality provided by iucaa , pune where part of this work has been done . gg acknowledges the senior research fellowship provided by csir , govt . of india and the hospitality provided by tifr , mumbai where part of this work has been done . aas and gg gratefully acknowledge the suggestions and comments from v. sahni during the earlier stages of the work .	we look at observational constraints on the thawing class of scalar field models proposed to explain the late time acceleration of the universe . using the recently introduced ` statefinder hierarchy ' , we compare these thawing class of models with other widely studied dark energy ( and modified gravity ) models to check the underlying parameter degeneracies . we put constraints on the deviations of these thawing models from the canonical @xmath0cdm model using a large class of observational data , e.g , the supernova type ia data , the bao data , the cmb data and data from the measurements of the hubble parameter using red - envelope galaxies . we also forecast constraints using a simulated dataset for the future jdem sne survey . our study shows that , although with current data it is difficult to distinguish different thawing models from @xmath0cdm , a future jdem like mission would be able tell apart thawing models from @xmath0cdm for currently acceptable values of @xmath1 . cosmology : dark energy , thawing model , scalar fields .
psr j2032 + 4127 is a radio - loud gamma - ray - emitting pulsar discovered by the fermi large area telescope ( fermi - lat ) and it is a young pulsar with a spin period @xmath1ms , spin - down power @xmath2 and spin - down age @xmath3kyr ( abdo et al . this pulsar had been regarded as isolated pulsar ( camilo et al . 2009 ) because of the lack of apparent variation in its rotation caused by reasonable orbital motion . lyne et al . ( 2015 ) found the variation of the observed pulsar rotation rate , which is consistent with the doppler effect of the orbital motion around the high - mass be star mt91 213 . their study suggests that the orbital period is very long , @xmath4yr , and the orbit is extremely elongated with an eccentricity @xmath5 . ho et al . ( 2016 ) refined the orbit parameters as @xmath6years and @xmath7 ( see figure [ orbit ] ) . psr j2032 + 4127 is a member of a class of rare - type pulsar that is orbiting around a high - mass b star . the next periastron passage will occur in late 2017 or in early 2018 . the binary nature of psr j2032 + 4127 is similar to that of psr b1259 - 63 , which is a well known radio pulsar that is orbiting around the high - mass b star , ls2883 , with @xmath8years and @xmath9 ( johnston et al . 1992 , 1999 , 2005 ) . psr b1259 - 63/ls2883 is known to be a source of non - pulsed and non - thermal emissions in the x - ray / gev / tev bands ( aharonian et al . 2005 ; abdo et al . 2011 ; tam et al . 2011 , 2015 ) , and it is classified as a so - called gamma - ray binary , which comprises a compact object ( neutron star or black hole ) and a high - mass ob star ( see dubus 2013 ) . multi - wavelength observations have confirmed several tev gamma - ray binaries , namely , psr b1259 - 63/ls 2883 , ls5039 ( aharonian et al . 2006 ) , ls i+61@xmath10 303 ( albert et al . 2006 ) , 1fgl j1018.6 - 5856 ( ackermann et al . 2012 ) , and h.e.s.s . j0632 + 057 ( hinton et al . 2009 ) . psr b1259 - 63/ls2883 is the only binary system for which the compact object has been confirmed to be a young pulsar . psr j2032 + 4127/mt91 213 is a candidate for the next gamma - ray binary , for which the compact object is definitely a young pulsar . the origin of the high - energy tev emissions from the psr b1259 - 63/ls 2883 system is likely related to the interaction of the pulsar wind of psr b1259 - 63 and the be wind / disk of ls 2883 ( tavani & arons 1997 ; dubus 2006 , 2013 ) . their interaction results in the formation of a shock , where the ram pressures of the pulsar wind and of the stellar wind / disk are in balance ( figure [ system ] ) . the electrons / positrons in the pulsar wind are accelerated at the shock , and produce nonthermal emissions in the radio to tev gamma - ray bands . the standard scenario assumes the synchrotron radiation and the inverse - compton scattering ( ics ) process produces radio / x - ray emission and tev gamma - ray emission , respectively . we can expect that psr j2032 + 4127/mt91 213 is also a source of x - rays / tev gamma - rays arising from the same mechanisms . x - ray / tev emissions from the gamma - ray binaries have been observed to exhibit temporal variations in their emission . various models have been suggested to explain the energy - dependent orbital modulations : for the pulsar model of the gamma - ray binaries , for example , the doppler boosting effect due to the finite velocity of the shocked pulsar wind ( dubus et al . 2010 ; kong et al . 2012 ; takata et al 2014a ) , evolution of the energy spectra of relativistic electrons of the shocked pulsar wind under the different energy - loss rates ( khangulyan et al . 2007 ; takahashi et al . 2009 ; zabalza et al . 2013 ) , and the interaction between the pulsar wind and the be disk ( sierpowska - bartosika & bednarek 2008 ; takata et al . 2012 ) , have been suggested . in this paper , we will consider the orbital variation of the emission from psr j2032 + 4127/mt91 213 as a result of the radial evolution of the magnetization parameter ( takata & taam 2009 ; kong et al . 2011 ) or the variation of the momentum ratio of the two winds . the gev emission from psr j2032 + 4127/mt91 213 is modulating with the spin period of the pulsar , and hence is dominated bythe magnetospheric process . the gev emission from the gamma - ray binaries has been observed by the fermi - lat . for psr b1259 - 63/ls 2883 , flare - like gev emissions have been observed after the second be disk passage of the pulsar ( abdo et al . , 2011 ; tam et al . 2011 , 2015 ) . although the inverse - compton scattering process with anisotropic soft - photon field from the disk / companion star ( e.g khangulyan et al . 2011 , 2014 ; van soelen et al . 2012 ; dubus & cerutti 2013 ) or synchrotron radiation process of the relativistic electrons / positrons ( chernyakova et al . 2015 ; xing et al . 2016 ) have been considered , the origin of the emission is not yet understood . the gamma - binary ls 5039 , whose orbital period is @xmath11days , is a candidate for which the compact object is a young pulsar . the gev emission from this system has been observed over the whole orbital phase by the fermi - lat ( abdo et al . 2009 ) . because their spectral shape measured by fermi resembles those of the gamma - ray emitting pulsars , it has been suggested that ls 5039 includes a young pulsar . the orbital modulating gev emission confirmed by fermi , however , suggests the emission process in the intra - binary space . the orbital modulations of gev and x - ray / tev emissions from ls 5039 show anti - correlation : gev flux ( or x / tev fluxes ) becomes maximum around the superior conjunction ( hereafter supc ) ( or inferior conjunction ( infc ) ) and becomes minimum around the infc ( or supc ) . within the framework of the shock emission due to the interaction between the pulsar wind and stellar wind , the observed modulation of the x - ray emissions from gamma - ray binary ls 5039 has been interpreted as the doppler boosting effect caused by the finite velocity of the shocked pulsar wind ( dubus et al . 2010 ; takata et al . 2014a ) , while the strong absorption of the tev photons by the stellar photon field will suppress the observed emissions at around supc ( dubus et al . the gev emission may be as a result of the inverse - compton scattering ( hereafter ics ) process of the cold - relativistic pulsar wind ( sierpowska - bartosik & torres 2008 ; kapala et al . 2010 ; torres 2011 ; takata et al . 2014a ) or of the shock - accelerated particles ( yamaguchi & takahara ; zabalza et al . the ics process between the relativistic particles and the stellar photons causes a flux peak in the observed modulation at around the supc ( if there is no absorption ) , where the scattering process of the relativistic particles that propagate to the earth occurs the head - on , and causes a flux minimum around the infc , where the scattering is a tail - on collision process . in this paper , we will estimate the contribution of the inverse - compton scattering process between the cold - relativist pulsar wind and the stellar photon , since the gamma - ray binary will provide a unique laboratory to investigate the properties of the cold - relativistic pulsar wind at the au distance scale from the pulsar ( cerutti et al . 2008 ) . the bolometric luminosity of the high - mass companion in the gamma - ray binaries is of the order @xmath12 . this intense stellar photon field leads to not only observable tev emissions through the ics , but also substantial @xmath13 absorption of tev gamma - rays . because the stellar radiation field is anisotropic in the emission region , the emissivity of the ics and optical depth of the photon - photon pair - creation process integrated along the line of sight depend on the orbital phase . such anisotropic processes will cause the observed orbital modulation of tev gamma - rays ( e.g. kirk et al . 1999 ; bednarek 2000 ; takata et al . the secondary pairs created by the primary tev gamma - rays are also a source of very high - energy photons through the ics process . a pair - creation cascade process may develop in the stellar wind region and may cause the observed tev emissions from the close binary systems ( bednarek 1997 , 2007 ; sierpowska & bednarek 2005 ; sierpowska - bartosik & torres 2007 , 2008 ; yamaguchi & takahara 2010 ; cerutti et al . 2010 ) and a periastron passage of the long - orbit binaries ( sierpowska - bartosik & bednarek 2008 ) . the anti - correlation between the gev and tev light curves of ls 5039 has been interpreted by using this cascade model ( bednarek 2007 ) in the x - ray bands , the observed temporal variations along the orbit will be mainly caused by the variations of the physical conditions of the pulsar wind at the shock . with the extremely elongated orbit of psr b1259 - 63 and psr j2032 + 4127 , the distance of the shock ( @xmath14 ) from the pulsar varies by about a factor of ten along the orbit ( figure [ shock ] ) . hence we can in principle use gamma - ray binaries to probe the physical properties of the pulsar wind ( e.g. magnetization parameter and bulk lorentz factor ) as a function of radial distance . for example , the pulsar wind magnetization parameter ( kennel & coroniti 1984 ) can be determined as @xmath15 where , @xmath16 , @xmath17 and @xmath18 are the magnetic field , number density and lorentz factor of the cold - relativistic pulsar wind , respectively , just before the shock . we can probe how the magnetization parameter changes with the radial distance from the pulsar by examining the observed temporal variations ( takata & taam 2009 ; kong et al . 2011 ) . since psr j2032 + 4127 is now approaching to the periastron passage that will occur in late 2017 or early 2018 , and since the separation between psr j2032 + 4127 and mt91 213 changes by about a factor of ten during @xmath191year and @xmath20year from the periastron ( figures [ orbit ] and [ shock ] ) , we can expect that the change of the shock distance from the pulsar will cause a large variation in the observed emission from this system , as indicated by the results of the recent x - ray observations done by the swift ( section [ observations ] , see also ho et al . 2016 ) . moreover , the ics process of the cold - relativistic pulsar wind before the shock may boost the stellar photons from optical to gamma - ray bands . the next periastron passage may provide u with s a unique opportunity to constrain the lorentz factor @xmath18 of the pulsar wind . this paper is organized as follows . in section [ observations ] , we will analyze the data taken from the swift and fermi - lat . we will confirm that the x - ray emission observed in 2016 is about factor of ten larger than that from previous observations , while the gev flux does not show any significant change . in section [ model ] , we describe our emission models from the intra - binary shock and cold - relativistic pulsar wind . in section [ result ] , we will discuss the high - energy emission within the framework of the pulsar wind / be wind , and probe the evolution of the magnetization parameter with a radial distance by fitting the swift data . in section [ discussion ] , we will discuss the emission as a result of the pulsar wind / be disk interaction model . we will also discuss the possibility of the formation of an accretion disk around the pulsar as a consequence of the pulsar / be disk interaction and its expected emission in the high - energy bands . [ tabel ] [ cols="^,^,^,^,^,^,^,^",options="header " , ] recently , ho et al . ( 2016 ) reported a rapid increase in the x - ray flux observed by swift , as the pulsar approached to the periastron . in this paper , we analyze the fermi - lat data as well as swift data . the swift - x - ray telescope ( xrt ) light curve was obtained using products extracted from the xrt repository http://www.swift.ac.uk/user_objects/ ( see also evans et al . 2007 , 2009 ) . we used a conversion factor of 1.27@xmath2110@xmath22 to convert the count rates into unabsorbed fluxes , where an absorbed power - law with @xmath232.37 and @xmath24 , based on total galactic hi column density in the direction toward the pulsar , is assumed ( kalberla et al . 2005 ; dickey & lockman , 1990 ) . since ho et al . ( 2016 ) adopted the @xmath25 based on the optical extinction of the companion , our unabsorbed x - ray flux is slightly higher than that presented by those authors . this difference does not affect the conclusions of the model calculations discussed in this paper . due to small photon statistics , we did not include observations shorter than 10 ks before 2014 . an observation taken in 2006 had an instrumental issue and was not included as well . an x - ray light curve is shown in fig . [ x - ray_lc ] . following up on the initial flux increase observed in early 2016 , we asked for additional dedicated xrt too observations since march 2016 . three to four observations were grouped together for better signal - to - noise ratios . it is clear that the trend in 2016 is the flux increase in x - rays . we analyzed the gamma - ray long - term light curve of psr j2032 + 4127 using the data from the _ fermi_-lat . photon events with energy ranging from 100 mev to 500 gev and time ranged from 2008 - 08 - 04 to 2016 - 08 - 02 were selected . the event class is `` pass 8 source '' and the corresponding instrumental response function is `` p8r2_source_v6 '' . the region of interest ( roi ) is a @xmath26 square centered at the epoch j2000 position of the source : @xmath27 . to avoid contamination from the earth s albedo , we excluded the time intervals when the roi was observed at a zenith angle greater than @xmath28 . the data reduction in this study was performed using the _ fermi _ science tools package version v10r0p5 . we first modeled the average emission from the background sources over the whole time span . the _ gtlike _ tool was used to perform a maximum binned likelihood analysis . the source model includes all 3fgl catalog sources ( gll_psc_v16.fit ) ( acero et al . 2015 ) that are within @xmath29 from the center of the roi , the galactic diffuse emission ( gll_iem_v06 ) and the isotropic diffuse emission ( iso_p8r2_source_v6_v06 ) , available from the _ fermi _ science support center ( fssc ) . the spectral parameters for sources that are non - variable and located @xmath30 away from the center are fixed to their catalog values . four extended sources within the region , gamma cygni , cygnus cocoon , hb 21 , and cygnus loop , were modeled by the extended source templates provided by the fssc . psr j2032 + 4127 is named as 3fgl j2032.2 + 4126 in the 3fgl catalog and is modeled by a power - law with simple exponential cutoff : @xmath31 where @xmath32 is the normalization constant , @xmath33 is the scale factor of energy , @xmath34 is the spectral power - law index and @xmath35 is the cut - off energy . from the binned likelihood analysis , the best - fit parameters of psr j2032 + 4127 are @xmath36 , @xmath37 and @xmath38mev . after fixing the spectral indices of all sources to the global fit and only leaving the normalization parameters free , the source model is then used to obtain the long - term light curve . the whole data set is then binned into eight bins , each bin is 365days . the energy range of the photons is further limited to 100 mev-100 gev . local fit is performed by the _ gtlike _ tool ( binned likelihood analysis ) to obtain the energy flux of psr j2032 + 4127 in each segment . figure [ 3fgl ] shows the long term evolution of the flux measured by the fermi . no clear indication of the change in the flux can be seen and the average flux over eight - year observations is @xmath39 , which will be dominated by the pulsed emissions from the magnetosphere . in the 1990s , the cgro / egret observed the region of psr j2032 + 4127 for a couple of times , including a more intensive observation during 1994 may to july . there is one source in the revised egret catalog ( casandjian & grenier , 2008 ) , egr j2033 + 4117 , whose position is 0.28 deg away from 3fgl j2032.2 + 4126 ( which has been identified as h psr j2032 + 4127 ) . the 95% error radius of egr j2033 + 4117 is 0.22 degree . due to the slight offset , we cannnot associate the two gamma - ray sources for certain . it is still possible to constrain the gamma - ray emission from psr j2032 + 4127 using these egret data . the photon flux of egr j2033 + 4117 is steady over the several measurements , and is consistent with ( 5 - 6)@xmath40ph @xmath41 s@xmath42 ( above 100 mev ) for all viewing periods . due to the poorer angular resolution of egret compared to lat and allowing for plausible contamination from nearby sources like cyg x-3 , this flux level may be regarded as an upper limit of any source at the position of psr j2032 + 4127 over the 1990s , and is above the average lat flux @xmath431.4@xmath40 ph @xmath41 s@xmath42 in our analysis . the origin of the high - energy tev emission from the psr b1259 - 63/ls 2883 system is likely related to the interaction of the pulsar wind of psr b1259 - 63 and the be wind / disk of ls 2883 ( tavani & arons 1997 ) . their interaction results in the formation of a shock , where the pulsar wind ram pressure and stellar wind / disk ram pressure are in balance . we apply this scenario to the psr j2032 + 4129/mt91 213 binary . in this section we describe our calculation methods . we define the capital letter @xmath44 and small letter @xmath45 as the distances from the b star and from the pulsar , respectively . the relativistic pulsar wind , which may have a bulk lorentz factor of @xmath46 , interacts with the stellar wind and forms a cone - shaped shock separating between the pulsar and the companion star . the radial distance to and its opening angle of the cone - shaped shock are determined by the ratio of the two winds ( e.g. canto et al . 1996 and references therein ) , @xmath47 where we assume that the pulsar wind carries the spin down - power @xmath48 of the pulsar , @xmath49 and @xmath50 is the mass loss rate and wind velocity of the companion star , respectively . the wind velocity of the stellar wind at a distance @xmath44 from the stellar surface may be expressed by ( waters et al . 1988 ) @xmath51 where @xmath52 km with @xmath53k being the temperature of the b star . the terminal velocity is estimated from @xmath54 with @xmath55 and @xmath56 being the mass and radius of the b star , respectively . since the shock is far away from the surface of the b star , we may assume @xmath57 . with @xmath58 of this pulsar ( lyne et al . 2015 ) and a typical mass loss rate @xmath59 of an o - type or b - type star ( e.g. snow 1981 ; smith 2006 ) , the momentum ratio of this system will be @xmath60 indicating the stellar wind is stronger than the pulsar wind with a reasonable mass loss rate of the stellar wind . the distance to the apex of the shock cone , @xmath14 , and the opening angle of the shock , @xmath61 , measured by the pulsar are calculated from ( canto et al . 1996 ) @xmath62 and @xmath63 respectively , where @xmath64 represents the orbital phase and @xmath65 is the separation between two stars . figure [ shock ] summarizes the evolution of the distance ( left panel ) and the magnetic field strength ( right panel ) at the shock apex along the orbit between -4000days and + 4000days from the periastron ( we did not plot the results for the other orbital phase , since we are interested in the emission processes around the periastron ) . in the figure , we used @xmath66 to calculate the magnetic field strength . as the figure shows , the shock distance and the magnetic field at the shock change by more than factor of ten along the orbit . since the pulsar orbit velocity ( @xmath67 ) is at least about a factor of ten slower than the stellar wind velocity ( @xmath68 ) , we assume that the shape of the shock cone is axially symmetric about the axis connecting the two stars . ) ] for the pulsar / be disk interaction , if the pulsar interacts with the be disk at the given orbit phase ( x - axis ) . the disk structure is calculated with @xmath69 , @xmath70 , @xmath71 and circular velocity @xmath72 . the relative velocity is assumed and is calculated from @xmath73 . the solid , dashed and dotted lines show the ratio of the shock distance and disk scale height at the pulsar for the base density of @xmath74 , @xmath75 and @xmath76 , respectively . the dashed - dotted line shows the ratio of the scale height and separation , @xmath77 . a cavity of the pulsar wind around the pulsar may be formed when @xmath78 is less than unity . , scaledwidth=50.0% ] . the magnetization parameter is @xmath79 ( @xmath80 g ) . the solid , dashed and dotted lines are time scale of the adiabatic cooling , synchrotron cooling and ics cooling , respectively . , scaledwidth=50.0% ] since the be - type star , mt91 213 , forms a dense equatorial disk outflow , the pulsar may interact with the be disk at the periastron passage , as in the case with the gamma - ray binary psr b1259 - 63/ls2883 system . the decretion disk model has been explored to describe the structure of the be disk ( lee et al . 1991 ; carciofi & bjorkman 2006 ; okazaki et al . 2011 ) , and it implies the disk mass density ( @xmath81 ) and the scale height ( @xmath82 ) are described by @xmath83 and @xmath84 respectively , where @xmath85 and @xmath86 ( @xmath69 and @xmath70 for an isothermal disk ) . in addition , we apply @xmath87 and the base density @xmath88 . takata et al . ( 2012 ) studied the orbital modulations of the x - ray / tev emission from psr b1259 - 63/ls2883 system . they argued that with the larger base density ( @xmath89 ) for the be disk , the pulsar wind creates a cavity in the disk gas and this causes a significant increase in the conversion efficiency from pulsar spin - down power to the shock - accelerated particle energy . this explains the double - peak structure of the x - ray light curves of the gamma - ray binary psr b1258 - 63/ls 2883 . the pulsar / be disk interaction with a smaller base density ( @xmath90 ) causes no cavity in the pulsar wind and hence no enhancement in the emission . with the elongated orbit and the long orbital period of the psr j2032 + 4127 , a strong pulsar / be disk interaction will be possible at only periastron passage . the radius of the shock from the pulsar may be obtained from a pressure balance condition , @xmath91 where @xmath92 is the relative velocity between the pulsar and the disk rotation . if the pulsar interacts with the be disk at the periastron , where the separation of the two stars is @xmath93au , the mass density and scale height of the disk at the pulsar are @xmath94 and @xmath95 , where we used @xmath69 , @xmath70 , @xmath71 and @xmath96 . the orbital velocity of the pulsar and circular velocity of the disk are @xmath97 and @xmath98 , respectively , where we used @xmath99 . the radial velocity of the disk matter , @xmath100 with @xmath101 being the sound speed ( okazaki et al . 2011 ) , is much slower than the pulsar motion and the disk s circular motion . hence , by assuming the relative velocity @xmath102 , the shock distance calculated from the equation ( [ shockdisk ] ) is @xmath103 at the periastron . a pulsar / be disk interaction may cause a cavity in the pulsar wind around the pulsar , provided that @xmath104 , which yields @xmath105 . figure [ bow ] summarizes the ratio of the shock distance given by equation ( [ shockdisk ] ) and the scale height of the disk , provided that the pulsar interacts with the disk at the given orbital phase in the horizontal axis . the figure summarizes the shock distance for the epoch between -100days and + 100days from the periastron . the different lines show the case for the base density of @xmath106 ( solid line ) , @xmath107 ( dashed line ) and @xmath108 ( dotted line ) , respectively . we find from the figure that a cavity would form ( @xmath109 ) at the periastron passage if @xmath110 . if the base density is less than @xmath111 , on the other hand , the pulsar wind will strip off an outer part of the be disk , truncating it at a radius smaller than the pulsar orbit , and the pulsar wind / be disk interactions will not affect to the observed emission . we calculate the emissions from the shock due to the interaction between the pulsar wind and stellar wind , as follows . the magnetic field at the shock radius but before the shock is calculated from @xmath112 where we defined @xmath113 . at the shock , the kinetic energy of the pulsar wind is converted into the internal energy of the shocked pulsar wind . applying a jump condition of a perpendicular mhd shock , we calculate the velocity @xmath114 , the magnetic field @xmath115 , and the gas pressure @xmath116 of the shocked pulsar wind at the shock ( kennel & coroniti 1984 ) . for the particle kinetic energy dominating the un - shocked flow , that is , for the low @xmath117 regime , we obtain @xmath118 @xmath119 and @xmath120 we assume that the post - shock pulsar wind flows along the shock surface . along the downstream flow , we assume a conservation of the magnetic flux @xmath121 . for the spherical symmetric flow , the bulk velocity decreases with the distance from the shock ( kennel & coroniti 1984 ) . however , the numerical simulations imply that the post - shock bulk flow for the binary systems does not simply decrease but increases with the distance from the shock , because of a rapid expansion of the flow in the downstream region ( bogovalov et al . furthermore , the high - energy emissions take place in the vicinity of the shock radius . in this study , therefore , we assume the velocity field of the post - shock pulsar wind with @xmath122constant . we assume that the electrons and positrons in the pulsar wind are accelerated by the shock and form a power law distribution over several decays in energy ; @xmath123 where we assume that the minimum lorentz factor of the accelerated particles is equal to the average lorentz factor of the particles forming the cold - relativistic pulsar wind ( section [ icpw ] ) . we assume the maximum lorentz factor by balancing between the acceleration time scale @xmath124 and the synchrotron loss time scale @xmath125 , yielding @xmath126^{1/2}$ ] , where @xmath127 represents the efficiency of the acceleration , which will be @xmath128 . the efficiency of the acceleration in the gamma - ray binaries has been explained from the observed spectra of very high - energy emissions ; for example , a high efficiency is required to explain the tev emissions of ls 5039 ( zabalza et al . 2013 ) . for psr j2032 + 4127/mt91 213 , it is difficult to discuss the efficiency , since tev emission from this source has not yet been detected . in the calculations , we found that the predicted x - ray flux and 0.1 - 10tev flux are insensitive to the efficiency if @xmath129 . in this study , therefore , we present the results with @xmath130 . future tev observations at the periastron passage may provide an addition information constraining the efficiency . for the injected particles at the shock , we assume @xmath131 for the power - law index of the distribution . since the particle energy density @xmath132 is related with the pressure @xmath116 as @xmath133 , the normalization factor @xmath134 of equation ( [ inidis ] ) is calculated from the relation @xmath135 . we solve the evolution of the lorentz factor after the shock with a simple one - dimensional treatment , that is , @xmath136.\ ] ] we assume the adiabatic loss given by @xmath137 where @xmath138 is the particle number density . the synchrotron loss is given as @xmath139 we calculate the ics energy loss rate using @xmath140 where @xmath141 is the stellar photon field distribution and @xmath142 is the cross - section with the isotropic photon field . for the soft - photon field , we consider the blackbody radiation from the b star and apply the planck function with the temperature @xmath143k , which is the case for companion star of psr b1259 - 63 ( negueruela et al . , 2011 ) . figure [ cool ] summarizes the time scale of the cooling at the periastron and at the apex of the shock cone . the tev gamma - ray produced by the ics process may be converted into an electron - positron pair through the photon - photon pair - creation process with the stellar photon . the cross section of this process is calculated from @xmath144,\ ] ] where @xmath145 where @xmath146 is the collision angle between the soft photon and gamma - ray , and it is a function of position . the optical depth is calculated from @xmath147 , where @xmath148 represents the energy distribution of the stellar radiation number density ( see table 1 for the parameters of the companion star ) . figure [ tau ] summarizes the optical depth of the high - energy photons emitted from the shock apex ( @xmath149 ) ; different lines in the figure represent difference system inclination angles @xmath150 and different positions of the pulsars along the orbit . we can find in figure [ tau ] that the optical depth of the @xmath151tev photons emitted around the periastron and the supc exceeds unity , indicating the tev photons emitted during the periastron passage are significantly absorbed by the stellar photons . and @xmath152 , respectively . the dotted line is the optical depth at the periastron with @xmath153 . the anomaly of the direction of earth is @xmath154 . , scaledwidth=50.0% ] the pulsar binaries will be a laboratory to search for the evidence of the high - energy gamma - rays produced by the ics process of the cold - relativistic pulsar wind ( cerutti et al . 2008 ) . in this study , we assume the pulsar wind is isotropic and is formed near the light cylinder of the pulsar , which is @xmath155 cm for psr j2032 + 4127 . since there is a theoretical uncertainty in the particle distribution of the cold - relativistic pulsar wind , we explore the process with a relativistic maxwell distribution of the form , @xmath156 which provides an average lorentz factor of @xmath157 . the normalization is calculated from the condition that @xmath158 due to the unknown energy conversion process from the magnetic energy to the particle energy , the magnetization parameter will evolve with the radial distance from the pulsar . from the energy conservation , the average lorentz factor of the cold - relativistic pulsar wind just before the shock may be described as @xmath159 $ ] , which provides @xmath160 in the limit of @xmath161 and @xmath162 , where @xmath163 and @xmath164 are the lorentz factor and the magnetization parameter at the light cylinder , respectively . the pairs created inside the light cylinder lose their momentum perpendicular to the magnetic field line via synchrotron radiation , and they will eventually escape from the light cylinder with a lorentz factor of @xmath165 . hence , the magnetization parameter at the light cylinder may be estimated from @xmath166 where @xmath167 is the multiplicity and we applied @xmath168s and @xmath169 g of psr j2032 + 4127 . from the observations of the pulsar wind nebulae , the multiplicity is expected to be @xmath170 for crab - like young pulsars ( tanaka & takahara 2010 ) and @xmath171 for the vela - like pulsars ( sefako & de jager , 2003 ) . theoretical studies have implied the multiplicity @xmath172 ( hibschman & arons 2001 ; timokhin & harding 2015 ) . hence we assume the typical value of the lorentz factor of the pulsar wind is in the range of @xmath173 . the power per unit energy per unit solid angle of the ics process with an anisotropic soft - photon field may be calculated from @xmath174 where @xmath175 is the solid angle of the sky covered by the b star measured from the emission region , @xmath176 with @xmath177 ( or @xmath178 ) describing the angle between the direction of the particle motion and the propagating direction of the scattered photons ( or background photons ) . in addition , @xmath179 is the planck function of the stellar photon field , and @xmath180 is the differential klein - nishina cross section . figure [ cpw ] shows the expected light curves ( left panel ) and spectra at the periastron ( right panel ) for the different system inclination angles relative to the earth viewing angle ( @xmath150 ) and the average lorentz factor of the pulsar wind @xmath18 , respectively . as the left panel indicates , the calculated emissions for a larger viewing angle are suppressed at around @xmath181days from the periastron . this is because the infc occurs at @xmath182days from the periastron ( figure [ system ] ) , and because at around the infc the ics process between a soft photon from the b star and a pulsar wind particle moving toward the earth is a tail - on collision process , which reduces the ics efficiency . since the supc is @xmath183 days after periastron , the calculated flux measured from the earth becomes maximum around , but after , periastron . as the right panel shows , the energy flux in the gev energy bands is @xmath184 for @xmath185 , which is close to the observed flux level ( @xmath186 ) of the pulsed emission from the fermi - lat . hence , we expect that if @xmath185 , the fermi - lat will observe an increase in the gev flux during the next periastron passage . in the current calculation , we applied the system parameters suggested by ho et al . ( 2016 ) , namely , the orbital period of @xmath187days , and the anomaly of the direction of the earth @xmath188 ( figure [ system ] ) . in addition , we apply @xmath189kpc ( lyne et al . 2015 ) and mjd 58069 as the periastron . we have not present the results of the calculations for the orbital phase @xmath190 days and @xmath191 days from the periastron , since we are interested in the emission processes around the periastron . we would like to note that within the current uncertainties of the orbit parameters ( @xmath192 , @xmath193(25years , 0.93 ) in lyne et al . ( 2015 ) and ( 50years , 0.96 ) given by ho et al . ( 2016 ) , the main conclusions of the model results are not affected , since the system sizes predicted by two authors are very similar to each other . as described in the previous section , the swift observed a rapid increase in the emissions in 0.3 - 10kev energy bands after @xmath43mjd56250 ( see figure [ x - ray_lc ] ) . these x - ray emissions are probably caused as a result of the pulsar wind / stellar wind interaction , since the pulsar / be disk interaction will be important only at around the periastron ( say , between -100 days and + 100 days ) , as discussed in the section [ disk ] . in this section , therefore , we will discuss the x - ray and tev emissions from the shock caused by the pulsar wind and stellar wind interaction , and will examine the pulsar wind properties by comparing the model results with the observed x - ray light curve . first , we consider the case where the magnetization parameter is constant with the radial distance from the pulsar , namely , the magnetization parameter at the shock does not depend on the shock distance . we also ignore the doppler boosting effect caused by the finite velocity of the shocked pulsar wind , which will be discussed in section [ dop ] . figure [ light ] summarizes the calculated flux and orbital modulation in the 0.3 - 10kev energy bands ( left panel ) and in the @xmath195tev energy bands ( right panel ) ; the solid , dashed , and dotted lines are for @xmath196 , 0.1 , and 0.01 , respectively . in the calculation , we assumed @xmath149 ( that is , @xmath197 ) , and @xmath198 . in the left panel of the figure , the results of the swift observations are also displayed . we find in figure [ light ] that the calculated light curve with constant magnetization parameter predicts an increase in the x - ray flux slower than that of the swift observations after @xmath199 days from the periastron . the synchrotron luminosity is roughly proportional to @xmath200 for a constant magnetization parameter . as figure [ shock ] shows , the shock distance decreases by about a factor of @xmath201 during @xmath202days and @xmath203days , and therefore , the calculated flux slightly increases during that epoch , which can not explain the increase in the flux ( a factor of @xmath204 ) measured by the swift . figure [ depend ] shows how the calculated flux depends on the model parameters , namely , the magnetization parameter ( left panel ) and the average lorentz factor of the un - shocked pulsar wind @xmath18 ( right panel ) . as seen in the left panel , the calculated x - ray flux ( solid line ) reaches its maximum value at the magnetization parameter @xmath205 . this is because the internal energy ( @xmath206 ) of the post - shock flow at the shock given by the jump condition decreases as the assumed magnetization parameter increases ( e.g. figure 4 of kennel & coronit 1984 ) , while the magnetic field at the shock ( @xmath115 ) increases as the magnetization parameter increases . the former tends to decrease the calculated x - ray flux , while the latter increases it . these two effects compensate each other and the calculated x - ray flux becomes maximum at the magnetization parameter @xmath205 . we find therefore that it is difficult to explain the observed x - ray flux of swift after @xmath203days with using the magnetization parameter @xmath207 . by comparing the calculated flux and swift observations , therefore , we conclude that the magnetization parameters at the shock is of order of @xmath205 at @xmath203days . it is possible that the lorentz factor @xmath18 of the cold - relativistic pulsar wind evolves with the shock distance from the pulsar . in the right panel of figure [ light ] , however , we can see that the calculated fluxes in x - ray / tev energy bands are less dependent on the minimum lorentz factor of the shocked particles at the shock ( @xmath18 ) . in the calculated tev light curve ( right panel in figure [ light ] ) , we can see an asymmetry relative to the periastron . since the tev photons from the shock are produced by the ics process , the asymmetry is introduced by the dependence of the collision angle between the stellar photons and the shocked pulsar wind particles that emit the photons toward the earth , as in the case with the orbital modulation of the ics of the cold - relativistic pulsar wind ( figure [ cpw ] ) . for the tev photons , the optical depth of the pair - creation process depends on the orbital phase , and it becomes the maximum at the supc for the photons traveling toward the earth . this effect can be seen as a rapid drop in the calculated flux around @xmath183days . the calculated tev emission from the shock depends on the earth viewing angle , as summarized in figure [ tevlight ] . in the previous sections , we ignored the effect of the motion of the post - shocked pulsar wind . in such a case , as figure [ light ] indicates , the orbital modulation of the x - ray emission is symmetric relative to the periastron . asymmetry in the x - ray light curve as well as in the tev light curve will be introduced by the finite velocity of the shocked pulsar wind . the finite velocity of the shocked pulsar wind will cause the doppler boosting effect , which enhances or suppresses the observed emissions . it has been suggested that the observed modulation of the x - ray emissions from gamma - ray binary ls 5039 is a result of the doppler boosting effect ( dubus et al . 2010 ; takata et al 2014a ) . the doppler factor is calculated from @xmath208 where @xmath209 is the lorentz factor of the bulk motion of the post - shocked flow , and @xmath210 is the angle of the flow direction and the line of sight . we assume that the velocity of the post - shocked pulsar wind flow is constant . figure [ light - d ] summarizes the doppler boosting effect with the different velocities of the post - shocked flow , @xmath211 ( solid lines ) given by the jump condition , @xmath212 ( dashed line ) in the limit of @xmath213 and @xmath214 ( dotted line ) . we can see in the figure that the doppler boosting effect enhances the emissions before the periastron , while suppressing after it . with @xmath215 , namely for the stellar wind stronger than the pulsar wind , the shock - cone wraps around the pulsar . around the infc ( @xmath182days ) , the post - shocked pulsar wind moves toward the earth and hence enhances the emissions . these are suppressed at around the supc ( @xmath183days ) where the post - shocked pulsar wind moves away from the earth . however , the doppler boosting effect will not be the reason for the rapid increase in the x - ray emissions observed by the swift . it has been considered that mass loss from the companion star will be driven by the radiation pressure , and the rapidly rotating main - sequence star , such that the angular velocity of the star rotation is close to its keplerian angular velocity , and produces a stellar wind enhanced in the polar region ( georgy et al . 2011 and reference therein ) . it is also considered that anisotropy of the pulsar wind explains the torus - like and jet - like structures of the pulsar wind nebulae , and theoretical models suggest that angular distribution of the pulsar wind energy is proportional to @xmath216 , where @xmath217 is the angle measured from the spin axis ( bogovalov and khangoulian 2002 ) . the anisotropy of the two winds could change the momentum ratio @xmath218 at the their interaction region along the orbit . besides the large structure of the anisotropy , an irregularity of the stellar wind from the high - mass star could be formed . it has been discussed that a radiatively non - stationary acceleration process produces the clumping ( small - scale density inhomogeneity ) of the stellar wind from the high - mass star ( runacres & owocki 2002 : owocki & cohen 2006 ) . it was analyzed that the typical size of a clump at the stellar surface is @xmath219 , and may linearly expand with the radial distance . a larger structure of clumps could be formed in the wind due to difference production mechanisms , magnetic field inhomogeneities , and star s rotation / pulsation ( see bosch - ramon 2013 ) . although the clumps fill only a small fraction of the volume of the wind region , it is thought that they carry most of mass ejected from the star , and have a mass density larger than the average density of the wind . studies with hydrodynamic simulations have shown that the clumping could develop in size and mass density due to merging between dense shells as traveling from the star , although it was also pointed out that a break - up of the dense shell as a consequence of the rayleigh - taylor or thin - shell instability limits the clump growth ( bozzo et al . 2016 , and references therein ) . if a clumpy wind is formed in a high - mass binary system hosting a compact object , it is expected that the irregularity of the mass density in the stellar wind will cause a temporal evolution of the momentum ratio ( @xmath218 ) , yielding the temporal variations in the shock emissions ( owocki et al . 2009 ; bosch - ramon 2013 ; de la cita et al . 2016 ) . the calculated x - ray fluxes depend on the wind momentum ratio , which determines the location of the shock . figure [ eta1 ] shows the calculated light curves for @xmath220 ( solid lines ) , 0.085 ( dashed lines ) and 0.0085 ( dotted lines ) , respectively , with the parameters of @xmath79 and @xmath198 . as we can see in figure [ eta1 ] , the calculated x - ray flux increases with decreasing the momentum ratio , because the shock distance from the pulsar decreases and hence the magnetic field at the shock increases with decreasing the momentum ratio . as the figure shows , a decrease in the momentum ratio ( that is , the stellar wind becomes stronger relative to the pulsar wind ) by a factor of ten can give an increase in the observed x - ray fluxes during @xmath199days to @xmath203days from the periastron . such momentum evolution with time could be caused by the anisotropy of the stellar wind / pulsar wind . a shorter time scale variability will be also possible as a consequence of the pulsar wind / clumpy wind interaction . the observed flux ratio between the x - ray and tev energy bands may provide an additional information for the magnetization parameter and momentum ratio of the pulsar wind to the stellar wind . as we expect , the flux ratio of the x - ray and tev does not greatly depend on @xmath18 , but it is sensitive to the energy densities of the magnetic field and soft - photon field at the emission region , which mainly depends on the magnetization parameter and the momentum ratio of the two winds . figure [ depend1 ] presents the dependence of @xmath221 on the magnetization parameter and the wind momentum ratio . if the future observations could measure the flux ratio in x - ray and tev energy bands , we could discuss the momentum ratio of the two winds ; for example , if the future observations provide @xmath222 at the periastron , the current model predicts that the magnetization @xmath2231 if the momentum ratio @xmath224 . .,scaledwidth=50.0% ] as seen in figures [ light ] , [ light - d ] and [ eta1 ] , the calculated x - ray light curve can not explain the swift observations after @xmath199days , if the momentum ratio does not significantly change along the orbit . the radial - dependent magnetization parameter has been discussed in the previous studies to explain the orbital modulations of the gamma - ray binaries ( takata & taam 2009 ; kong et al . 2011 ; takata et al . 2014a ) . in this paper , we explore the radial dependence with a function form of @xmath225 the left panel in figure [ fit ] shows the results of the fitting for the swift x - ray data with different power law indices ; @xmath226 for solid line , 2 for dashed line and 3 for dotted line . other parameters are @xmath227/yr ) and @xmath198 . in addition , we applied @xmath211 taking into account the doppler boosting effect discussed in section [ dop ] . to explain the observed flux level from swift , we normalized the magnetization parameter so as to be @xmath228 at @xmath229au for each power law index . figure [ sigma ] shows the evolution of the fitting magnetization parameters at the apex of the shock cone as a function of the orbital phase . we can see in the figure [ fit ] that a faster evolution ( larger @xmath230 ) of the magnetization parameter with the radial distance reproduces a x - ray light curve being more consistent with the whole swift observations after @xmath231days . as the figure shows , the current model predicts that the peaks of the flux in the light curve occurs at a day before the periastron and the flux then rapidly decreases during @xmath232days and @xmath183days by about one order of magnitude . this is because ( 1 ) the assumed magnetization parameter determined by the relation , @xmath233 , exceeds unity for @xmath234 , as figure [ sigma ] shows , and ( 2 ) the doppler boosting suppresses the emissions around the supc . this feature can be tested by the future observations . , @xmath79 and @xmath235.,scaledwidth=50.0% ] gamma - ray binaries have also been detected by the fermi - lat and the emission will originate from the magnetospheric processes and/or the intra - binary processes . for psr j2032 + 4127 , the pulsed gev emission due to the pulsar spin has been measured by the fermi - lat with an energy flux @xmath236 ( section [ observations ] ) . the high - energy tail of the synchrotron spectrum and the low energy tail of the ics process of the intra - binary shock discussed in the previous sections could contribute to the energy bands ( 0.1 - 300gev ) of the fermi - lat . in addition to the shock emissions , the ics process of the cold - relativistic pulsar wind with @xmath185 produces gev gamma - rays ( figure [ cpw ] ) . figure [ gevlight ] compares the observed flux level of the pulsed emissions with the predicted flux levels of the shocked emissions ( solid line ) and of the ics process of the cold - relativistic pulsar wind ( dashed line ) . in the calculation , we assume @xmath237 for the average lorentz factor of the particles of the cold - relativistic pulsar wind , and @xmath235 for the inclination of the system . as we see in figure [ gevlight ] , the emissions from the intra - binary space are comparable to the magnetospheric emissions around the periastron . with the current parameters , therefore , we expect that the fermi - lat measures the orbital modulation of the gev flux during the periastron passage that will occur in late 2017 or in early 2018 . in figure [ spectrum ] , we present the model broadband spectra averaged over @xmath238days to @xmath239days ; the model parameters are @xmath240 in equation ( [ sr ] ) , @xmath241 and @xmath211 , which reproduce the observed x - ray light curves . to see the dependence of the lorentz factor @xmath18 , we calculated the spectra for @xmath242 ( right panel ) and @xmath243 ( left panel ) , respectively . the solid line shows the calculated spectrum combining the synchrotron and ics process at the shock , and the dashed line represents the spectrum of the ics process of the cold - relativistic pulsar wind . the current model predicts that the magnetospheric emissions dominate in the spectral energy distribution averaged over the periastron passage . if the minimum lorentz factor of the shocked pulsar wind particles is @xmath244 , the tev spectrum will show a turnover at around @xmath245 and the x - ray spectrum has a break at around 10kev . during the periastron passage , the tev photons from the shocked pulsar wind may be converted into pairs by the photon - photon pair creation process with the stellar photon fields ; in particular , the tev photons emitted toward the b star will be totally absorbed . these secondary pairs created in the stellar wind side will also emit the non - thermal photons via the synchrotron and ics processes , and may initiate a further pair - creation cascade . if these pairs are isotropized , they will emit the photons propagating toward the observer , even though the primary tev photons do not do this ( bednarek 1997 , 2007 ; sierpowska & bednarek 2005 , 2008 ; sierpowska - bartosik & torres 2007 , 2008 ; yamaguchi & takahara 2010 ; cerutti et al . 2010 ) . the emissions from the secondary pairs produced in the stellar wind side will be dominated by the ics process . the ratio of the radiation power between the synchrotron radiation and ics processes will be described by the energy density ratio of the magnetic field to the stellar photon . the ics process dominates the synchrotron radiation process at the radial distance from the star of @xmath246 where @xmath247 represents the stellar magnetic field at the stellar surface . in addition , we assumed @xmath248 as the radial evolution of the stellar magnetic field . it is known that some high - mass main - sequence stars have a surface dipole magnetic field of order of @xmath249 g ( walder et al . 2012 ) . with a typical value of @xmath250 , we can see that the synchrotron radiation of the created pairs in the stellar wind side can be important only at very near the stellar surface and its contribution to th observed x - ray emissions will be negligible . by tracing the primary tev gamma - rays from all emission regions considered in the calculation box , we calculated the ics process of the secondary pairs . we assumed that the pairs produced in the stellar wind side are quickly isotropized , and we calculated the emission process with a constant photon field during the crossing time , @xmath251 , where @xmath252 represents the distance of the pair - creation position from the stellar surface . figure [ cascade ] compares the spectra , which are measured on earth of the ics photons from the shock region ( solid line ) and from the secondary pairs ( dashed line ) at the periastron , where we assumed that all photons from the secondary pairs can escape from the pair - creation process . a further pair - creation cascade process will make the spectra softer . as the figure shows , the integrated flux from the secondary pairs is much less than the primary emissions , and therefore we expect that the contribution of the emissions from higher - order pairs is much less than that from the shock emissions . the emissions from the secondary pairs could contribute to the 0.1 - 1tev bands , where the shock emissions around the periastron are significantly absorbed . the interaction of psr j203 + 4127 and the be disk of mt91 213 may enhance the high - energy emissions , as in the case with the psr b1259 - 69/ls2883 system showing an increase in the x - ray / tev fluxes at the pulsar / be disk interaction phase ( chernyakova et al . 2006 ) . as we discussed in section [ disk ] , we expect that the pulsar / be disk interaction in this system affects the observed emissions , provided that the base density is @xmath253 and the interaction occurs at the periastron passage , say between @xmath238days and @xmath239days from the periastron ( figure [ bow ] ) . since we do not know the geometry ( e.g. inclination relative to the orbital plane ) of the disk , we can not predict when the pulsar interacts with it . we expect however that the pulsar / be disk interaction will occur at least once during the periastron passage during @xmath238days and @xmath239days , since the true anomaly @xmath254 corresponds to @xmath255days from the periastron . if the base density of the be disk is close to @xmath89 , an interaction between pulsar and be disk at the periastron passage will form a cavity in the pulsar wind , and most of the pulsar wind ( @xmath256 sr in solid angle ) will be stopped at the radial distance from the pulsar given by equation ( [ shockdisk ] ) and figure [ bow ] . in figure [ diske ] , we show an example of the calculated light curve for the pulsar / be disk interactions and compare the estimated x - ray / tev fluxes with those for the pulsar / be wind interactions . for the calculation , we assumed the base density @xmath257 , the magnetization @xmath79 and the momentum ratio @xmath149 for the pulsar wind / stellar wind interaction . figure [ diske ] indicates that the pulsar / be disk interaction predicts x - ray fluxes several factors larger than those due to the pulsar / be wind interaction , if the magnetization parameter does not evolve with the radial distance . within the framework of the current calculation , the shock distance from the pulsar is closer , and the magnetic field at the shock is stronger for the pulsar / be disk interaction , in which @xmath258 at periastron , than the pulsar / be wind interaction , for which @xmath259 with @xmath149 . as a result , the synchrotron power is stronger for the pulsar / be disk interaction than the pulsar / be wind interaction . this result suggests that if the base density of the be disk is @xmath89 , the interaction of pulsar / be disk may give rise to local maximum in the orbital modulation of the x - ray emissions . as the right panel in figure [ diske ] shows , the calculated tev flux for the pulsar / be disk interaction is lower than that for the pulsar / be wind interaction at around the periastron . this is because the synchrotron cooling time scale for tev leptons becomes shorter than the ics cooling and adiabatic cooling time scales at the shock for the pulsar / be disk interaction . at the periastron , for example , the shock distance at the apex for the pulsar / be wind interaction is @xmath260 , for which the inverse - compton cooling and adiabatic cooling time scales of tev electrons are shorter than the synchrotron cooling time scale ( figure [ cool ] ) . for pulsar / be disk interaction , the shock distance becomes @xmath261 , which increases the magnetic field at the shock by a factor of @xmath262 , and hence decreases the synchrotron cooling time scale by about @xmath263 . as a result , the ics cooling scale , which is less sensitive to the shock distance , becomes longer than the synchrotron time scale , and the ics emissivity is suppressed . as we discussed in section [ disk ] , if the base density of the disk is high enough , the pulsar wind will be confined within a small region by the be disk matter . when the radius of the cavity estimated in figure [ bow ] is less than the radius below which the kinetic energy of the disk gas is less than the gravitational potential energy of the pulsar , the disk matter may be gravitationally captured by the pulsar , and results in the formation of a disk around the pulsar . the capture radius measured from the pulsar may be estimated as @xmath264 where @xmath265 is the mass of the pulsar and @xmath92 is the relative velocity of the pulsar with respect to the disk matter . near the periastron , @xmath93au , the orbital velocity of the pulsar is @xmath266 . the velocities of the kepler motion and the radial velocity of the be disk at the pulsar position are of the order of @xmath267 and @xmath268 , respectively ( okazaki et al . 2011 ) . hence , we expect that the relative velocity is of order of @xmath269 , although this depends on the disk geometry and the rotation direction . as figure [ bow ] shows , at near the periastron , where the separation is @xmath93au , the estimated shock distance due to the pulsar / be disk interaction is of order of @xmath270 au for @xmath271 , suggesting the disk matter could be captured by the pulsar because @xmath272 . the accretion rate may be estimated as @xmath273 where @xmath274 is the geometrical factor and we used the parameter at @xmath275au ( typical separation during the periastron passage ) with @xmath276 . we note that for the stellar wind , @xmath277 , the capture radius becomes @xmath278au , which is usually smaller than the shock radius due to the pulsar wind and stellar wind interaction ( figure [ shock ] ) . hence the stellar wind will not be captured by the pulsar . there will be an relative angular momentum of the captured matter with respect to the pulsar , indicating the captured matter forms a circular orbit around the pulsar at a distance ( @xmath279 ) where the relative angular momentum per unit mass is equal to the angular momentum of the kepler orbit around the pulsar , @xmath280 . the angular frequency of the orbital motion of the pulsar is @xmath281 . the angular frequency of the circular orbit of the be disk around the b - star is also @xmath282 , because @xmath283 . roughly speaking , therefore , the angular frequency associated with the relative angular momentum will be of the order of @xmath284 , although again it depends on the disk geometry and rotation direction . the circularization radius of the captured matter may be estimated as ( frank et al . 2002 ) @xmath285 if we apply the standard shakura - sunyaev disk model , the disk matter will move inward with a dynamical time scale @xmath286 , where @xmath287 is the viscous parameter in units of 0.1 , suggesting the accreting matter will take several weeks to reach the pulsar after the capture event . if the dynamical time scale ( @xmath288 ) is longer than the pulsar s disk crossing time scale , which may be of the order of @xmath289 , the accretion disk may not develop around the pulsar . we note however that the circularization radius given by equation ( [ circ ] ) is sensitive to the relative velocity ; with @xmath290 cm , for example , @xmath291 cm , which is inside of the light cylinder . moreover , we do not know the geometry of the disk . if the disk plane is parallel to the orbital plane , the pulsar / be disk continues to interact during the periastron passage . hence it is possible that the accretion disk develops with a radial length @xmath292 around the pulsar at the periastron passage . if the accretion disk forms around the pulsar , the disk supplies uv photons in the pulsar wind region . figure [ bb ] summarizes the radiation power ( @xmath293 ) for the accretion disk ( solid line ) , stellar surface ( dashed line ) , and be disk ( dotted line ) . for the accretion disk , we assume an accretion rate @xmath294 and the disk extends from the light cylinder @xmath295 of the pulsar to @xmath296 . for the stellar emission , we assume the planck function with @xmath297k and @xmath96 . for the be disk emissions , we assumed the disk temperature @xmath298 after the calculation of carciofi & bjorkman ( 2006 ) . we ignore the effects of the emission and absorption lines on the spectra . as discussed in above , the captured matter will take about several weeks to reach the pulsar magnetosphere . before the accretion disk reaches to the light cylinder radius , the rotation activity of the pulsar is still turned on and produces the cold - relativist pulsar wind . then we can expect that the ics process of the cold - relativistic pulsar will boost up the uv photons from the disk to the higher energy photon . figure [ cpdisk ] summarizes the calculated spectra ( left panel ) and the integrated flux as a function of the accretion disk inclination angle ( right panel ) . the figure shows the results when the accretion disk extends from the light cylinder radius @xmath295 to @xmath296 . in addition , we assumed the mono - energetic distribution of the particles in the cold - relativistic pulsar wind . we took into account the energy loss of the pulsar wind due to the ics process . as figure [ cpdisk ] shows , if the initial lorentz factor is @xmath299 which produces the scattered photons with 1 - 10gev energy , the calculated flux level can reach an order of @xmath300 , which is comparable to that of the pulsed emissions . pulsar binary systems may provide a valuable laboratory to study the accretion process on rapidly spinning and strongly magnetized neutron stars . one may consider that once the accretion disk crosses the light cylinder , the rotation - powered activity would be quenched . this is because the copious plasma fills the acceleration region at the rotation - powered stage , where the charge density deviates from the so - called goldreich - julian charge density ( goldreich & julian 1969 ) . however , the recent observation suggests the rotation - powered activities and the accretion disk can co - exist in the magnetosphere . a transient - millisecond pulsar is a binary millisecond pulsar with low mass companion star , and it transits between the rotation - powered stage and the accretion stage ( archibald et al . 2009 ) . psr j1023 + 0038 transited from its rotation powered stage to the accretion stage at the latter in 2013 june , when the pulsed radio emission from this pulsar disappeared ( stappers et al . coherent pulsed x - ray emission discovered after 2013 june suggests an accretion of matter on the neutron star surface ( archibald et al . jaodand et al . ( 2016 ) , however , found that the spin - down rate in the accretion stage is higher than that in the rotation - powered stage , and they concluded that the rotation activity is still operating in the accretion stage . enhancement of observed gev emission in the accretion stage also suggests the survival of the rotation activity in the accretion stage ( takata et al . this model was also applied to another transient - millisecond pulsar psr j1227 - 4853 ( bednarek 2015 ) . gamma - ray binaries with a be companion may also offer opportunities to study this accretion process on young pulsars . the ics model of the pulsar wind could provide an explanation for the origin of the flare - like gev emissions from psr b1259 - 63/ls 2883 system after the second be disk passage of the pulsar . it has been observed that the gev peak in the orbital modulation occurs @xmath301 days after the x - ray peak ( tam et al . 2011 , 2015 ) . it has been considered that the phase of the x - ray peak corresponds to that of the pulsar / be disk interaction . so we may expect that the be disk matter starts to be captured by the pulsar around the orbital phase of the x - ray peak . since it will take several weeks to fully develop the accretion disc around the pulsar , the gev peak position will delay from the x - ray peak position . finally , the magnetic radius @xmath302 from the pulsar is defined as the distance where the dipole magnetic field of the pulsar begins to dominate the dynamics of the accreting matter , and it is estimated from @xmath303 with @xmath304 and @xmath305 , @xmath306 this radius is smaller than the light cylinder radius of psr j2032 + 4127 ( @xmath307 cm ) . the co - rotation radius ( @xmath308 ) is defined as the distance where the angular velocity of keplerian motion is equal to the spin angular velocity of the pulsar , namely @xmath309 which will be smaller than the magnetic radius , suggesting the accretion process is in propeller regime , and not all infalling matter will accrete on the pulsar surface . the outflow due to the propeller effect also could cause the high - energy emission ( e.g. papitto & torres , 2015 ) . in summary , we analyzed the swift and fermi - lat data of the psr j2032 + 4127/mt91 213 binary system , which is a candidate to be another tev gamma - ray binary . the x - ray flux has rapidly increased as the pulsar approaches periastron , which will occur in late 2017 or early 2018 , while the gev flux measured by fermi - lat shows no significant change . we investigated the x - ray emission by the synchrotron process of the pulsar wind particles accelerated at the shock due to the pulsar wind and stellar wind interaction . we argued that the increase in the observed x - rays is caused by ( 1 ) variation of the momentum ratio of the two winds along the orbital phase or ( 2 ) the evolution of the magnetization parameter of the pulsar wind with the radial distance as @xmath233 and @xmath310 . the current model predicts that the peak fluxes in the 0.3 - 10kev and 0.1 - 10tev energy bands are of the order of @xmath311 . the gamma - ray binary system could provide a unique laboratory to study the emissions from the cold - relativistic pulsar wind . the current model suggests that the ics process of the cold - relativistic pulsar wind could contribute to the fermi - lat observations at the periastron passage ( figure [ gevlight ] ) . for this binary system , the pulsar / be disk interaction will affect to the observed emissions , provided that the base density of the be disk is @xmath312 and the interaction occurs during @xmath313days and @xmath314days . for a be disk with higher base density @xmath89 , the pulsar / be disk interaction could form an accretion disk around the pulsar . the ics process of the cold - relativistic pulsar wind off the uv from the disk also provides the high - energy emission from the pulsar / be star binary system . future multi - wavelength observations will provide a unique opportunity to probe the pulsar wind properties and the various emission processes caused by the pulsar / be wind and pulsar / be disk interactions . we express our appreciation to an anonymous referee for useful comments and suggestions . we thank a. okazaki for the useful discussions on the be disk model . jt is supported by nsfc grants of chinese government under 11573010 and u1631103 . cwn and ksc are supported by grf grant under 17302315 . cyh is supported by the national research foundation of korea through grants 2014r1a1a2058590 and 2016r1a5a1013277 . all calculations were done under the high performance computing cluster ( hyperion ) of the institute of particle physics and astrophysics , hust . abdo et al . 2009 , apjl , 706 , 56 abdo et al . 2011 , apjl , 736 , 11 abdo , a. a. ; ajello , m. ; allafort , a. ; baldini , l. ; ballet , j. ; barbiellini , g. ; baring , m. g. ; bastieri , d. , et al . , 2013 , apjs , 208 , 17 acero , f. ; ackermann , m. ; ajello , m. ; albert , a. ; atwood , w. b. ; axelsson , m. ; baldini , l. ; ballet , j. et al . , 2015 , apjs , 218 , 23 ackermann , m. et al . 2012 , sci , 335 , 189 aharonian , f. et al . 2005 , a&a , 442 , 1 aharonian , f. et al . 2006 , a&a , 460 , 743 albert , j. et al . 2006 , sci , 312 , 1771 archibald , anne m. ; bogdanov , slavko ; patruno , alessandro ; hessels , jason w. t. ; deller , adam t. et al . , 2015 , apj , 807 , 62 archibald , anne m. ; stairs , ingrid h. ; ransom , scott m. ; kaspi , victoria m. ; kondratiev , vladislav i. , et al . , 2009 , sci , 324 , 1411 bednarek , w. , 2015 , mnras letter , 451 , 55 bednarek , w. , 2007 , a&a , 464 , 259 bednarek , w. , 2000 a&a 362 , 646 bednarek , w. , 1997 , a&a , 322 , 528 bogovalov , s. v. ; khangoulian , d. v. , 2002 , mnras letter , 336 , 53 bogovalov , s. v. ; khangulyan , d. v. ; koldoba , a. v. ; ustyugova , g. v. ; aharonian , f. a. , 2008 , mnras , 387 , 63 bosch - ramon , v. , 2013 , a & a , 560 , 32 bozzo , e. ; oskinova , l. ; feldmeier , a. ; falanga , m. , 2016 , a&a , 589 , 102 cerutti , b. ; dubus , g. ; henri , g. , 2008 a&a , 488 , 37 cerutti , b. , malzac , j. , dubus , g. , henri , g. , 2010 a&a , 519 , 81 camilo , f. ; ray , p. s. ; ransom , s. m. ; burgay , m. ; johnson , t. j. ; kerr , m. ; gotthelf , e. v. ; halpern , j. p. , et al . 2009 , apj , 705 , 1 canto , j. , raga , a.c . & wilkin , f.p . , 1996 , apj , 469 , 729 carciofi , a. c. ; bjorkman , j. e. , 2006 , apj , 639 , 1081 casandjian , j .- m . & grenier , i. a. , 2008 , a&a , 489 , 849 chernyakova , m. ; neronov , a. ; lutovinov , a. ; rodriguez , j. ; johnston , s. , 2006 , mnras , 367 , 1201 chernyakova , m. et al . mnras 454 , 1358 de la cita , v. m. ; bosch - ramon , v. ; paredes - fortuny , x. ; khangulyan , d. ; perucho , m. , 2016 , arxiv:1611.03788 dubus , g. , 2006 , a&a 456 , 801 dubus , g. ; cerutti , b. ; henri , g. , 2008 , a&a , 477 , 691 dubus , g. , cerutti , b. & henri , g. , 2010 , a&a , 516 , 18 dubus , g. , 2013 , the astronomy and astrophysics review , volume 21 , article i d . 64 dubus , g. ; cerutti , b. , 2013 , a&a , 557 , 127 dickey , john m. ; lockman , felix j. , 1990 , ara&a , 28 , 215 evans , p. a. ; beardmore , a. p. ; page , k. l. ; tyler , l. g. ; osborne , j. p. ; goad , m. r. ; obrien , p. t. ; vetere , l. , 2007 , a&a , 469 , 379 evans , p. a. ; beardmore , a. p. ; page , k. l. ; osborne , j. p. ; obrien , p. t. ; willingale , r. ; starling , r. l. c. ; burrows , d. n. et al . , 2009 , mnras , 397 , 1177 frank , j. , king , a. , & raine , d. 2002 , accretion power in astrophysics ( cambridge : cambridge univ . press ) georgy , c. ; meynet , g. ; maeder , a.,2011 , a&a , 527 , 52 goldreich , peter ; julian , william h. , 1969 , apj , 157 869 hibschman , j. a. , & arons , j. 2001 , apj , 554 , 624 hinton et al . 2009 , apjl , 690l , 101 ho , wynn c. g. ; ng , c .- y . ; lyne , andrew g. ; stappers , ben w. ; coe , malcolm j. ; halpern , jules p. ; johnson , tyrel j. ; steele , iain a. , 2016 , eprint arxiv:1609.06328 jaodand , amruta ; archibald , anne m. ; hessels , jason w. t. ; bogdanov , slavko ; dangelo , caroline r. , et al . 2016 , apj , 830 , 122 johnston , simon ; manchester , r. n. ; lyne , a. g. ; bailes , m. ; kaspi , v. m. ; qiao , guojun ; damico , n. , 1992 , apj ltter , 387 , 37 johnston , simon ; manchester , r. n. ; mcconnell , d. ; campbell - wilson , d. , 1999 , mnras , 302 , 277 johnston , simon ; ball , lewis ; wang , n. ; manchester , r. n. , 2005 , mnras , 358 , 1069 kalberla , p. m. w. ; burton , w. b. ; hartmann , dap ; arnal , e. m. ; bajaja , e. ; morras , r. ; p@xmath315ppel , w. g. l. , 2005 , a&a , 440 , 775 kapala , m. , bulik , t. , rudak , b. , dubus , g. & lyczek , m. , 2010 , proceedings of the 25th texas symposium on relativistic astrophysics . december 6 - 10 , 2010 . heidelberg , germany . editors : frank m. rieger ( chair ) , christopher van eldik and werner hofmann . kennel , c. f. ; coroniti , f. v. , 1984 , apj , 283 , 710 khangulyan , d. ; hnatic , s. ; aharonian , f.a . ; bogovalov , s.v . , 2007 , mnras , 380 , 320 khangulyan , d. ; aharonian , f. a. ; bogovalov , s.v . ; rib@xmath316 , m. , 211 , apj , 742 , 98 khangulyan , d. , aharonian , f.a . ; bogovalov , s.v . , & rib@xmath316 m. , 2012 , apjl , 752 , 17 khangulyan , dmitry ; bogovalov , sergey v. ; aharonian , felix a. , 2014 , international journal of modern physics conference series , 28 , 60169 kong , s.w . , yu , y.w . , huang , y.f . & cheng , k.s . , 2011 , mnras , 416 , 1067 kong , s. w. ; cheng , k. s. ; huang , y. f , 2012 , apj , 753 , 127 lee , umin ; osaki , yoji ; saio , hideyuki , 1991 , mnras , 250 , 432 lyne , a. g. ; stappers , b. w. ; keith , m. j. ; ray , p. s. ; kerr , m. ; camilo , f. ; johnson , t. j. , 2015 , mnras , 451 , 581 negueruela , ignacio ; rib@xmath317 , marc ; herrero , artemio ; lorenzo , javier ; khangulyan , dmitry ; aharonian , felix a. , 2011 , apjl , 732 , 11 okazaki , atsuo t. ; nagataki , shigehiro ; naito , tsuguya ; kawachi , akiko ; hayasaki , kimitake ; owocki , stanley p. ; takata , jumpei , 2011 , pasj , 63 , 893 owocki , s. p. ; romero , g. e. ; townsend , r. h. d. ; araudo , a. t. , 2009 , apj , 2009 , 696 , 690 owocki , stanley p. ; cohen , david h. , 2006 , apj , 648 , 565 papitto , a. ; torres , d. f. , 2015 , apj , 807 , 33 runacres , m. c. ; owocki , s. p. , 2002 , a&a , 381 , 1015 sefako , ramotholo r. ; de jager , ocker c. , 2003 , apj , 593 , 1013 sierpowska - bartosik , a. & torres , d.f . , 2008 , aph , 30 , 239 sierpowska - bartosik , a. , torres , diego f. , 2007 , apj letter , 671 , 145 sierpowska - bartosik , a. , bednarek , w. , 2008 , mnras , 385 , 2279 sierpowska , a. ; bednarek , w. , 2005 , mnras , 356 , 711 smith , m. a. , 2006 a&a , 459 , 215 snow , t. p. , jr . , 1981 , apj , 251 , 139 stappers , b. w. ; archibald , a. ; bassa , c. ; hessels , j. ; janssen , g. et al . atel , 5513 , 1 takahashi , t. et al . 2009 , apj , 697 , 592 takata , j. & taam , r.e . , 2009 , apj , 702 , 100 takata , j. ; okazaki , a. t. ; nagataki , s. ; naito , t. ; kawachi , a. ; lee , s .- h . ; mori , m. ; hayasaki , k. ; 2012 , apj , 750 , 70 takata , j. ; leung , gene c. k. ; tam , p. h. t. ; kong , a. k. h. ; hui , c. y. ; cheng , k. s. , 2014a , apj , 790 , 18 takata , j. ; li , k. l. ; leung , gene c. k. ; kong , a. k. h. ; tam , p. h. t. , et al . 2014b , apj , 785 , 131 tanaka , shuta j. ; takahara , fumio , 2010 , apj , 715 , 1248 tam , p.h.t . , huang , r.h.h . , takata , j. , hui , c.y . , kong , a.k.h . , cheng , k. s. , 2011 , apjl , 736 , 10 tam , p. h. t. ; li , k. l. ; takata , j. ; okazaki , a. t. ; hui , c. y. ; kong , a. k. h. , 2015 , apj letter , 798 , 26 tavani , m. , & arons , j. , 1997 , apj , 477 , 439 timokhin , a. n. ; harding , a. k. , 2015 , apj , 810 , 144 torres , d.f . , 2011 , proceedings of the 1st sant cugat forum on astrophysics , icrea workshop on the high - energy emission from pulsars and their systems , held in april , 2010 , ( preprint , arxiv:1008.0483 ) van soelen , b. ; meintjes , p. j. ; odendaal , a. ; townsend , l. j. , 2012 , mnras , 426 , 3135 waters , l. b. f. m. ; van den heuvel , e. p. j. ; taylor , a. r. ; habets , g. m. h. j. ; persi , p. , 1988 , a&a , 198 , 200 walder , rolf ; folini , doris ; meynet , georges , 2012 , ssrv , 166 , 145 xing , yi ; wang , zhongxiang ; takata , jumpei , apj , 828 , 61 yamaguchi , m. s. ; takahara , f. , 2010 , apj , 717 , 85 yamaguchi , m. s. ; takahara , f. , 2012,apj , 761 , 146	psr j2032 + 4127 is a radio - loud gamma - ray - emitting pulsar ; it is orbiting around a high - mass be type star with a very long orbital period of @xmath0years , and is approaching periastron , which will occur in late 2017/early 2018 . this system comprises with a young pulsar and a be type star , which is similar to the so - called gamma - ray binary psr b1259 - 63/ls2883 . it is expected therefore that psr j2032 + 4127 shows an enhancement of high - energy emission caused by the interaction between the pulsar wind and be wind / disk around periastron . ho et al . recently reported a rapid increase in the x - ray flux from this system . in this paper , we also confirm a rapid increase in the x - ray flux along the orbit , while the gev flux shows no significant change . we discuss the high - energy emissions from the shock caused by the pulsar wind and stellar wind interaction and examine the properties of the pulsar wind in this binary system . we argue that the rate of increase of the x - ray flux observed by swift indicates ( 1 ) a variation of the momentum ratio of the two - wind interaction region along the orbit , or ( 2 ) an evolution of the magnetization parameter of the pulsar wind with the radial distance from the pulsar . we also discuss the pulsar wind / be disk interaction at the periastron passage , and propose the possibility of formation of an accretion disk around the pulsar . we model high - energy emissions through the inverse - compton scattering process of the cold - relativistic pulsar wind off soft photons from the accretion disk .
let @xmath5 be a smooth 3-manifold . classical morse theory deals with a real - valued function @xmath6 . this function corresponds to a handle decomposition of @xmath5 namely @xmath7 , where @xmath8 is a collection of 0-handles , @xmath9 is a collection of 1-handles , @xmath10 is a collection of 2-handles and @xmath11 is a collection of 3-handles . in @xcite scharlemann and thompson introduce the concept of _ thin position _ for 3-manifolds ; the idea is to build the manifold as described before , with a sequence of 1-handles and 2-handles chosen to keep the boundaries of the intermediate steps as simple as possible . the morse theory of circle - valued maps @xmath12 , as in the real case , relates the topology of a manifold @xmath5 to the critical points of @xmath13 . morse - novikov theory was introduced by novikov @xcite to study these functions . see @xcite for a survey of these topics . recently there has been work on circle - valued morse theory on the complement of knots and links in @xmath14 . in @xcite , pajitnov , rudolph and weber introduced the concept of the _ morse - novikov _ number of a link @xmath15 . the _ morse - novikov number _ of a link , denoted by @xmath16 , is the least possible number of critical points of a regular circle - valued morse mapping @xmath17 . they proved that the morse - novikov number is subadditive with respect to the connected sum of knots ; i.e. @xmath18 . in @xcite , goda pointed out that there is a handle decomposition which corresponds to a circle - valued morse map , which he calls a heegaard splitting for sutured manifolds . we can consider more general circle - valued morse functions on knot complements which correspond to handle decompositions that do not necessarily arise from a heegaard splitting . analogous to scharlemann and thompson , we describe a process to re - order the handles of a handle decomposition of a knot exterior in such a way that the regular level surfaces are as simple as possible , giving rise to the definition of _ circular width of the knot exterior _ and _ circular thin position of the knot exterior_. similarly to regular thin position , circular thin position guarantees that all the level surfaces are either incompressible or weakly incompressible . hence when the knot exterior is in circular thin position we obtain a nice sequence of seifert surfaces which are alternately incompressible and weakly incompressible . in general we expect to see several such level surfaces . however there are some special cases . recall that a fibered knot @xmath19 is a knot with a seifert surface @xmath20 whose knot complement can be fibered over @xmath21 with fiber @xmath20 . in our context a fibered knot is a knot whose knot exterior has a circular thin position with one and only one incompressible level surface and none weakly incompressible level surface . this is the unique circular thin position for a fibered knot , see @xcite or @xcite , and as expected , circular thin position yields no additional seifert surfaces for the knot . the circular width of a fibered knot is defined to be zero . we define an almost fibered knot to be a knot whose complement possesses a circular thin position in which there is one and only one weakly incompressible seifert surface @xmath22 and one and only one incompressible seifert surface @xmath23 . goda @xcite showed that all non - fibered knots up to ten crossings are handle number one knots . in our context these knots have a circular thin position with one incompressible seifert surface @xmath23 of minimal genus and a weakly incompressible seifert surface @xmath22 with genus(s)= genus(f)+1 . thus , all non - fibered knots up to ten crossings are examples of almost fibered knots . goda s examples also illustrate that almost fibered knots do not have a unique circular thin position . he describes knots with two non - isotopic minimal genus seifert surfaces which can be used to find two different circular thin positions , both giving the structure of an almost fibered knot . given all these concepts and definitions we prove the main theorem : let @xmath19 . at least one of the following holds : 1 . @xmath4 is fibered ; 2 . @xmath4 is almost fibered ; 3 . @xmath4 contains a closed essential surface in its complement . moreover this closed essential surface is in the complement of an incompressible seifert surface for the knot ; 4 . @xmath4 has at least two non - isotopic incompressible seifert surfaces . we also study the behavior of circular width under two natural operations on knot exteriors . given two knots @xmath24 and @xmath25 in @xmath14 , we can take their connected sum , or we can glue their exteriors together along their common boundary ensuring that preferred longitudes match . in both cases , we find upper bounds for the circular width of the resulting manifold , which depend on the circular width of the original knot exteriors . it is natural to ask for an example of a knot which is neither fibered nor almost fibered . the candidate we propose is the connected sum of two almost fibered knots . as we will see in section [ section4 ] this connected sum inherits a circular structure from the knot summands . it seems to be hard to prove that this is indeed a circular thin position for the connected sum . in section [ section2 ] we review definitions concerning surfaces , circle - valued morse functions and heegaard splittings . we study circle - valued morse functions on knots in section [ section3 ] . we define and introduce the terminology of circular handle decomposition , circular width and almost fibered knots . we prove theorem [ my theo ] . section [ section4 ] is about the behavior of circular width under two knots operations : connected sum of knots and boundary sum of knot exteriors . using these two operations we construct new manifolds , in one case the exterior of the connected sum of two knots and in the other case a closed manifold . in both cases there is a natural circular decomposition inherited by these manifolds , so we can prove that the circular width of the manifolds is bounded above by an n - tuple which depends on the circular width of the original knot exteriors . i want to thank my advisor professor abigail thompson for her guidance , encouragement and for so many helpful conversations . this work is part of my doctoral dissertation at university of california , davis . this section is devoted to definitions related to knots and seifert surfaces , as well as to properties of seifert surfaces under two operations on knots . the definitions and operations are mostly classical ; see @xcite for more details . let @xmath4 be a knot in @xmath14 . the knot complement will be denoted by @xmath26 . an open tubular neighborhood of @xmath4 in will be denoted by @xmath27 and the exterior of the knot @xmath4 by @xmath28 . a seifert surface @xmath29 for a knot @xmath4 is an oriented compact 2-submanifold of @xmath14 with no closed components such that @xmath30 . the intersection of @xmath29 with @xmath3 , @xmath31 , is also called a seifert surface for @xmath4 . since @xmath20 is two sided we can specify a @xmath32 and a @xmath33 of @xmath20.we say that a disk @xmath34 , such that @xmath35 , lies on the @xmath32 ( resp . in the @xmath33 ) of @xmath20 if the collar of its boundary lies on the @xmath32 ( resp . in the @xmath33 ) of @xmath20 . we say that @xmath20 is _ compressible _ if there is a 2-disk @xmath36 such that @xmath37 does not bound a disk in @xmath20 . if @xmath20 is not compressible , it is said to be _ incompressible_. @xmath34 is a compressing disk for @xmath22 . we say that @xmath20 is _ strongly compressible _ if there are two compressing disks @xmath38 lying on the + side of @xmath20 and @xmath39 lying on the @xmath40side of @xmath20 with @xmath41 and @xmath42 disjoint essential closed curves in @xmath20 . otherwise we say that @xmath20 is _ weakly incompressible_. [ connected sum ] the _ connected sum of two knots _ @xmath24 and @xmath25 , denoted by @xmath43 , is constructed by removing a short segment from each @xmath44 and joining each free end of @xmath24 to a different end of @xmath25 to form a new knot . this operation is well - defined up to orientation . there is a 2-sphere @xmath45 that intersects @xmath46 in two points . @xmath45 is called a _ separating _ sphere . see figure [ knotsum ] . given seifert surfaces @xmath47 and @xmath48 for @xmath24 and @xmath25 , respectively , one may construct a seifert surface for the knot @xmath43 by taking a boundary connected sum of @xmath47 and @xmath48 , denoted by @xmath49 . figure [ boundarysum ] shows the simplest case for a boundary connected sum of seifert surfaces . now let us consider another operation on the exterior @xmath50 and @xmath51 of two knots @xmath24 and @xmath25 in @xmath14 . let @xmath47 and @xmath48 be seifert surfaces for @xmath24 and @xmath25 , respectively . [ boundary - sum ] let @xmath5 be the orientable closed 3-manifold obtained from @xmath50 and @xmath51 by identifying their boundaries via a homeomorphism @xmath52 such that @xmath53 , where @xmath54 for @xmath55 . denote this manifold by @xmath56 and call it the _ boundary sum _ of the knot exteriors @xmath50 and @xmath51 . under this identification @xmath47 and @xmath48 are glued together along their boundaries via @xmath57 , obtaining a closed embedded surface in @xmath5 . denote this surface by @xmath58 and call it a _ boundary sum _ of seifert surfaces @xmath47 and @xmath48 . the following lemma analyzes the behavior of incompressible seifert surfaces under the homeomorphism @xmath57 used to obtain the manifold @xmath59 . [ incomp_surf ] if @xmath47 and @xmath48 are incompressible seifert surfaces in @xmath50 and @xmath51 , respectively , then @xmath60 is incompressible in @xmath5 . let @xmath61 be the image of @xmath62 in @xmath5 and let @xmath63 be the image of @xmath64 in @xmath5 . suppose @xmath23 is compressible . then we can choose a nontrivial compressing disk @xmath65 with @xmath66 such that @xmath34 intersects @xmath63 in a minimal number of arcs . @xmath67 , since otherwise @xmath68 , contradicting the fact that @xmath69 is incompressible ( for @xmath70 . since @xmath23 is a two sided surface in @xmath5 , @xmath34 lies on one side of @xmath23 , say the + side . let @xmath71 be an outermost arc in @xmath34 . then there is an arc @xmath72 such that @xmath73 and @xmath74 share endpoints and bound a disk in @xmath34 . if @xmath75 is trivial in @xmath69 , then it can be pushed across @xmath63 reducing @xmath76 . hence @xmath77 must be essential in @xmath69 . suppose @xmath75 is in @xmath47 . we can cap off the edge @xmath73 in a neighborhood of @xmath63 , creating a compressing disk @xmath78 for @xmath47 . since @xmath47 is incompressible , @xmath79 bounds a disk in @xmath47 . using this disk we can push @xmath77 across @xmath63 , decreasing @xmath76 , which is a contradiction . therefore @xmath23 is incompressible . we will assume basic definitions and results from real - valued morse theory ; for details , see @xcite and @xcite . the following has been adapted from pajitnov s book on circle - valued morse theory , @xcite . let @xmath5 be a smooth compact 3-manifold . let @xmath13 be a smooth function from @xmath5 to the one dimensional sphere @xmath21 . the morse theory of circle - valued maps @xmath12 , as in the real case , relates the topology of a manifold @xmath5 to the critical points of @xmath13 . morse - novikov theory was introduced by novikov @xcite to study these functions . the motivation came from a problem in hydrodynamics . for a point @xmath80 choose a neighborhood @xmath81 of @xmath82 in @xmath21 diffeomorphic to an open interval of @xmath83 , and let @xmath84 . the map @xmath85 is then identified with a smooth map from @xmath86 to @xmath20 . thus all the local notions of critical points , non - degeneracy , index , etc . are defined in the same way as for the real - valued case . a smooth map @xmath87 is called a _ morse map _ , if all its critical points are non - degenerate . for a morse map @xmath87 we denote by @xmath88 the set of all critical points of @xmath13 , and by @xmath89 the set of all critical points of index @xmath90 . if @xmath5 is compact , the set @xmath88 is finite ; in this case we denote by @xmath91 the cardinality of @xmath88 and by @xmath92 the cardinality of @xmath89 . we turn our attention to circle - valued morse theory for the complement of a knot @xmath4 in @xmath14 . a _ circle - valued morse function _ on @xmath93 is a function @xmath94 which has only non - degenerate critical points . let @xmath4 be an oriented knot in @xmath14 . the manifold @xmath95 is not compact and to develop a reasonable morse theory it is natural to impose a restriction on the behavior of the morse map in a neighborhood of @xmath4 . this restriction will allow @xmath13 to have a finite set of critical points . we require the circle - valued morse map @xmath13 to behave nicely " in a neighborhood of @xmath4 . let @xmath4 be a knot in @xmath14 . a morse map @xmath96 is said to be _ regular _ if @xmath4 has a neighborhood framed as @xmath97 such that @xmath98 and the restriction @xmath99 is given by @xmath100 . the set of critical points of a regular morse map @xmath13 is finite . from now on we will be considering _ regular circle - valued morse functions _ on knot complements . for simplicity we will just refer to them as _ circle - valued morse functions_. recall that a knot @xmath19 is fibered if there is fibration @xmath101 behaving nicely " in a neighborhood of @xmath4 . this fibration is unique , see @xcite or @xcite . so if we consider a morse map @xmath96 with minimum number of critical points ( which is zero ) , then @xmath4 is a fibered knot , and @xmath13 is homotopic to @xmath102 . if @xmath4 is not fibered then any morse map @xmath96 will necessarily have critical points . it is natural to expect to find nice relationships between circle - valued morse functions on knot complements and the topology of the knot complement , just as in the real - valued case . we will discuss this relationship in section [ thin_circ_val ] . heegaard splittings were first introduced by poul heegaard in his dissertation in 1898 . he proved that a closed connected orientable compact 3-manifold contains a surface which decomposes the 3-manifold into two _ a _ handlebody _ is a connected compact orientable 3-manifold with boundary containing @xmath103 pairwise disjoint , properly embedded 2-disks such that the manifold resulting from cutting along the disks is a 3-ball . for manifolds with non - empty boundary , one needs the concept of compression body , introduced in @xcite . it is a generalization of a handlebody . definitions [ cbdef ] , [ triaddef ] and [ hsdef ] are from @xcite . [ cbdef ] a _ compression body _ @xmath104 is a cobordism rel @xmath105 between surfaces @xmath106 and @xmath107 such that @xmath108 2-handles @xmath109 3-handles and @xmath107 has no 2-sphere components . if @xmath110 and @xmath104 is connected , then @xmath104 is obtained from @xmath111 by attaching a number of 1-handles along disks on @xmath112 where @xmath107 corresponds to @xmath113 . denote by _ @xmath114 _ the number of 1-handles attached to @xmath115 . with @xmath116 a genus 2 surface with one boundary component and a genus 1 surface . @xmath117 is a genus 3 surface with one boundary component , width=188 ] [ triaddef ] a _ 3-manifold triad @xmath118 _ is a cobordism @xmath5 rel @xmath105 between surfaces @xmath119 and @xmath120 . thus @xmath119 and @xmath120 are disjoint surfaces in @xmath121 with @xmath122 such that @xmath123 . [ hsdef ] a _ heegaard splitting _ of @xmath118 is a pair of compression bodies @xmath124 such that @xmath125 , @xmath126 and @xmath127 , @xmath128 . @xmath22 is called a heegaard surface and @xmath129 . the _ genus _ of a heegaard splitting is defined by the genus of the heegaard surface . a heegaard splitting @xmath124 is said to be _ weakly reducible _ if there are disks @xmath130 and @xmath131 with @xmath132 an essential curve , for @xmath55 , and such that @xmath133 . if the heegaard splitting is not weakly reducible then it is said to be _ strongly irreducible_. the next lemma is proved in @xcite ; we will need it in section [ thin_circ_val ] . [ lemma1 ] if @xmath134 or @xmath135 are compressible in @xmath124 then @xmath124 is weakly reducible . [ thin_circ_val ] given a regular morse function @xmath96 , as in the case of real - valued morse functions , there is a correspondence between @xmath13 and a handle decomposition for @xmath3 , namely @xmath136 , where @xmath20 is a seifert surface for @xmath4 , @xmath137 is a regular level surface of @xmath13 , @xmath9 is a collection of 1-handles corresponding to index 1 critical points , and @xmath10 is a collection of 2-handles corresponding to index 2 critical points . we will call this decomposition a _ circular handle decomposition _ for @xmath3 . , width=151 ] let us denote by @xmath69 the surface @xmath138 and let @xmath139 be the surface @xmath140 , where @xmath141 means the closure . when @xmath142 , @xmath143 . every @xmath69 and @xmath144 contains a seifert surface for @xmath4 ; note that @xmath144 or @xmath69 may be disconnected . the surfaces @xmath69 and @xmath144 , for @xmath145 will be called _ level surfaces_. a level surface @xmath144 is called a _ thin surface _ and a level surface @xmath69 is called a _ thick surface_. let @xmath146collar of @xmath147 . @xmath148 is divided by a copy of @xmath69 into two compression bodies @xmath149collar of @xmath150 and @xmath151collar of @xmath152 . thus @xmath69 describes a heegaard splitting of @xmath148 into compression bodies @xmath153 and @xmath154 , where @xmath155 , @xmath156 , @xmath157 ( @xmath158 ) , @xmath159 . thus we can write @xmath160 . figure [ cirdec ] shows a schematic picture of a circular handle decomposition with level surfaces and compression bodies indicated . into compression bodies , width=188 ] we wish to find a decomposition in which the @xmath69 are as simple as possible . for a closed connected surface @xmath161 define the complexity of @xmath162 , @xmath163 , to be @xmath164 . for a connected surface @xmath22 with nonempty boundary we define the complexity , @xmath165 , to be @xmath166 , where @xmath167 denotes @xmath22 with its boundary components capped off with disks . if @xmath168 or @xmath169 , set @xmath170 . if @xmath22 is disconnected we define @xmath171 where @xmath69 are the components of @xmath22 . let @xmath4 be a knot in @xmath14 . let @xmath34 be a circular handle decomposition for @xmath3 . define the _ circular width of @xmath3 with respect to the decomposition d , @xmath172 _ , to be the set of integers @xmath173 . arrange each multi - set of integers in monotonically non - increasing order , and then compare the ordered multisets lexicographically . the _ circular width of @xmath3 , denoted @xmath174 _ , is the minimal circular width , @xmath172 over all possible circular decompositions @xmath34 for @xmath3 . @xmath3 is in _ circular thin position _ if the circular width of the decomposition is the circular width of @xmath3 . if a knot @xmath4 is fibered we define the circular width of @xmath4 , @xmath175 , to be equal to zero . analogous to @xcite , the following theorem holds ; [ theo1 ] if @xmath3 is in circular thin position then : 1 . each heegaard splitting @xmath176 is strongly irreducible . each @xmath144 is an incompressible surface in @xmath3 . 3 . each @xmath69 is a weakly incompressible surface . the lemma below is needed in the proof of the theorem . a proof of the lemma can be found in @xcite . [ lemma2 ] let @xmath23 be a surface . if @xmath177 is obtained from @xmath23 by a non - trivial compression then @xmath178 suppose @xmath176 is weakly reducible . then there are nontrivial compressing disks @xmath179 and @xmath180 with @xmath181 . compress @xmath69 towards @xmath153 ( resp . @xmath154 ) along @xmath182 ( resp . @xmath183 ) obtaining a surface @xmath184 ( resp . @xmath185 ) that divides @xmath153 ( @xmath154 ) into compression bodies @xmath186 ( resp . @xmath187 ) where @xmath188 , @xmath189 , @xmath190 and @xmath191 ( resp . @xmath192 , @xmath193 , @xmath194 and @xmath195 ) . + so we have obtained a new decomposition for @xmath3 whose width is the original width except for the integer @xmath196 that is replaced by @xmath197 and @xmath198 . by lemma [ lemma2 ] this new presentation has smaller circular width , which is a contradiction . hence @xmath176 is not weakly reducible . 2 . suppose @xmath144 is compressible . let @xmath34 be a compressing disk for a component of @xmath144 . let @xmath199 . by an innermost disk argument we can find a disk ( which we will also call @xmath34 ) so that @xmath200 . @xmath34 lies entirely inside either @xmath148 or @xmath201 , say the former . by lemma [ lemma1 ] we have that @xmath176 is weakly reducible , contradicting part ( 1 ) . 3 . by part ( 2 ) , @xmath23 is incompressible . hence we can assume that any compressing disk for @xmath69 lies in @xmath148 . any pair of disjoint compressing disks for @xmath69 in @xmath148 would contradict ( 1 ) . the converse of the theorem [ theo1 ] is not always true . a knot exterior @xmath3 could have a circular handle decomposition satisfying _ ( 1 ) , ( 2 ) _ and _ ( 3 ) _ of theorem [ theo1 ] , but such a decomposition need not be the thinnest . [ local_thin ] a circular handle decomposition @xmath34 for a knot exterior @xmath3 is called a _ circular locally thin _ decomposition if the thin level surfaces @xmath144 s are incompressible and the thick level surfaces @xmath69 s are weakly incompressible . if @xmath4 has a circular thin decomposition in which all 1-handles are added before all 2-handles , we call @xmath4 _ almost fibered_. @xmath4 is _ almost fibered _ if there is a seifert surface @xmath20 so that @xmath3 has a circular thin decomposition of the form @xmath202 . figure [ esquema_almost_fib ] shows a schematic picture of an almost fibered knot . examples of almost fibered knots are given by all non - fibered prime knots @xmath4 up to 10 crossings . in @xcite it is shown that these knots are handle number 1 and the decomposition arising from this handle number is realized by a minimal genus seifert surface . therefore all these knots are almost fibered . it is plausible to suspect that a minimal genus seifert surface always arises as part of a thin circular handle decomposition of a knot exterior . the only evidence to support this suspicion are the cases of fibered knots and almost fibered knots up to ten crossing . we can now state our main theorem : [ my theo ] let @xmath19 . at least one of the following holds : 1 . @xmath4 is fibered ; 2 . @xmath4 is almost fibered ; 3 . @xmath4 contains a closed essential surface in its complement . moreover this closed essential surface is in the complement of an incompressible seifert surface for the knot ; 4 . @xmath4 has at least two non - isotopic incompressible seifert surfaces . the proof of this theorem will follow from our definitions and a variation on a result of waldhausen ( proposition 5.4 in @xcite ) , applied to the double of @xmath3 . the waldhausen s result is the following : [ prop5.4 ] let @xmath5 be an irreducible 3-manifold . in @xmath5 let @xmath23 and @xmath203 be incompressible surfces , such that @xmath204 , and @xmath205 consists of mutually disjoint simple closed curves , with transversal intersection at any curve which is not in @xmath206 . suppose there is a surface @xmath207 and a map @xmath208 , such that @xmath209 is a covering map onto @xmath23 , and @xmath210 . then there is a surface @xmath211 and an embedding @xmath212 , such that @xmath213 , @xmath214 ( i.e. , a small piece of @xmath23 is parallel to a small piece of @xmath203 ) , and that moreover @xmath215 , and either @xmath216 , or @xmath217 and @xmath218 are disks . we use this to prove : [ product1 ] let @xmath5 be an irreducible 3-manifold . let @xmath23 and @xmath203 be isotopic , incompressible , closed , connected and disjoint surfaces in @xmath5 . then @xmath23 and @xmath203 are parallel , in other words they cobound a product region in @xmath5 . @xmath23 and @xmath203 have empty boundary and they are disjoint by hypothesis . since @xmath23 and @xmath203 are isotopic there exits @xmath219 such that @xmath220 is homeomorphism for every @xmath221 , @xmath222 and @xmath223 . define @xmath224 . the restriction of @xmath13 to @xmath225 is the identity on @xmath23 , so is a covering map onto @xmath23 . moreover ; @xmath226 so we can apply proposition [ prop5.4 ] . there is a subsurface @xmath227 which is parallel to a subsurface of @xmath228 . the intersection of @xmath217 with @xmath203 is precisely @xmath229 . since @xmath23and @xmath203 are disjoint it follows that @xmath217 is disjoint from @xmath203 as well , therefore @xmath230 . then @xmath217 and @xmath218 are not disks . so @xmath218 intersects @xmath23 in @xmath231 , but @xmath23 and @xmath203 are disjoint , then it follows that @xmath232 . the only possible subsurfaces of @xmath23 and @xmath203 with empty boundary are themselves . therefore @xmath23 and @xmath203 are parallel . we apply lemma [ product1 ] to the double of a knot exterior to obtain : [ lemma3 ] let @xmath4 be a knot in @xmath14 . if @xmath23 and @xmath203 are disjoint incompressible isotopic seifert surfaces in @xmath3 then @xmath23 and @xmath203 are parallel , that is they cobound a product region . let @xmath5 be the double of @xmath3 , i.e , @xmath5 is constructed by taking two disjoint copies of @xmath3 and glueing them together along their boundary . @xmath5 is an irreducible manifold . let @xmath177 ( and @xmath233 ) be the closed surface in @xmath5 obtained by gluing along the boundary two disjoint copies @xmath234 and @xmath235 of @xmath23 ( two disjoint copies @xmath236 and @xmath237 of @xmath203 ) . notice that @xmath177 and @xmath233 are disjoint , incompressible ( see lemma [ incomp_surf ] ) and isotopic . by lemma [ product1 ] , @xmath177 and @xmath233 are parallel . by construction the intersection of the images of @xmath238 in @xmath5 with @xmath177 and @xmath233 cobound a product annulus @xmath239 in the image of @xmath238 , which is contained in the product region bounded by @xmath177 and @xmath233 . hence we can split @xmath5 along @xmath238 to recover the manifold @xmath3 . the surfaces @xmath23 and @xmath203 inherit the parallelism of @xmath233 and @xmath177 . therefore the surfaces @xmath23 and @xmath203 cobound a product region in @xmath3 . we can now prove theorem [ my theo ] . [ proof of theorem [ my theo ] ] let @xmath34 be a circular thin decomposition of @xmath3 . @xmath240 . suppose @xmath4 is not fibered and not almost fibered . then @xmath241 ; so there is at least one thin level surface , @xmath235 , different from @xmath20 . consider @xmath242 . @xmath235 is an incompressible surface in @xmath3 , by part ( 2 ) of theorem [ theo1 ] . suppose @xmath235 is not connected . then @xmath235 contains a closed component . since each of its components is incompressible , ( 3 ) holds . otherwise @xmath235 is an incompressible seifert surface . if @xmath235 is isotopic to @xmath243 , by lemma [ lemma3 ] they are parallel , so they bound a product on one side . this implies that the decomposition is not thin , since it can be replaced on one side by a product , ( see figure [ isotopicf1_f2 ] ) . therefore @xmath235 is not isotopic to @xmath20 , and ( 4 ) holds . we describe two ways to construct new manifolds from @xmath50 and @xmath51 . we will analyze the effect of the constructions on the circular width . we note that the width , for compact orientable 3-manifolds , is additive " under connected sum of 3-manifolds : @xmath244 ( see @xcite ) . in @xcite , scharlemann and schultens analyze the behavior of generalized heegaard splittings by cutting along a family of essential annuli . let @xmath256 . there is a natural way to obtain a circular handle decomposition for @xmath3 as follows . starting with the seifert surface @xmath257 for @xmath4 , we attach the sequence of handles corresponding to @xmath50 , i.e. , we attach @xmath9 and @xmath10 , along the @xmath245 summand of @xmath20 . then we attach the sequence of handles corresponding to @xmath51 , i.e. , we attach @xmath258 and @xmath259 , along the @xmath246 component of @xmath20 . so we have : we have found an upper bound for the circular width of @xmath291 . the following results hold by our construction . the optimal bound occurs if @xmath245 and @xmath246 are minimal genus seifert surfaces . let @xmath24 be a fibered knot with fiber @xmath245 . let @xmath25 be an almost fibered knot whose thin circular handle decomposition consists of one 1-handle and one 2-handle . let @xmath246 and @xmath48 be the thin and thick level surfaces , respectively , of @xmath51 . suppose that @xmath246 is a minimal genus seifert surface . then the circular width of @xmath291 is given by : @xmath291 inherits a circular handle decomposition @xmath34 from @xmath50 and @xmath51 , which consists of one 1-handle and one 2-handle . the circular handle decomposition @xmath34 has the thin level surface @xmath303 , which is a minimal genus seifert surface for @xmath46 ; and the thick level surface @xmath304 . by corollary [ cw1 - 2 ] , @xmath305genus@xmath297 is an upper bound for the circular width of @xmath291 . if there were a thinner circular handle decomposition for @xmath274 , it would have a minimal genus seifert surface as a thin level surface . the number of 1-handles would have to be fewer than those in @xmath34 , hence zero . thus @xmath46 would be fibered . but the connected sum of two knots is fibered if and only if both knots are fibered . thus , we have that @xmath302genus@xmath297 . suppose that the given circular handle decompositions for @xmath50 and @xmath51 are thin . a natural question arises : is the circular handle decomposition induced on @xmath274 a thin circular decomposition ? in order to address this question we need to prove : 1 . the boundary connected sum of two incompressible seifert surfaces is incompressible 2 . the boundary connected sum of an incompressible seifert surface and a weakly incompressible seifert surface is a weakly incompressible seifert surface . let @xmath44 be a nontrivial knot in @xmath14 and @xmath144 its seifert surface , for @xmath55 . consider @xmath306 the connected sum of the knots and @xmath307 the boundary connected sum of the surfaces . let @xmath45 be the decomposition sphere . notice that @xmath308 is a properly embedded arc @xmath73 in @xmath3 . 1 . see @xcite . 2 . suppose @xmath234 is incompressible and @xmath235 is weakly incompressible . assume that @xmath23 is not weakly incompressible . then there exist compressing disks @xmath38 and @xmath39 lying on opposite sides of @xmath23 with @xmath309 . moreover we can choose them so that @xmath310 meets @xmath45 in a minimal number of arcs . notice that @xmath311 is non - empty since both disks can not be contained in @xmath234 or @xmath235 at the same time . consider @xmath312 an arc in @xmath313 which is outermost in @xmath38 , so @xmath312 cuts off a disk in @xmath38 with boundary @xmath312 and @xmath314 . if @xmath315 is trivial in @xmath234 or @xmath235 it can be pushed across @xmath45 ( taking any other arcs with it ) and reducing @xmath316 . thus @xmath315 is essential in @xmath234 or @xmath235 . @xmath315 is not essential in @xmath234 since @xmath234 is incompressible . hence @xmath315 is essential in @xmath235 . capping off @xmath312 in a neighborhood of @xmath45 gives rise to a compressing disk @xmath317 for @xmath235 . + if @xmath318 , then @xmath39 and @xmath317 are compressing disks for @xmath235 lying on opposite sites with @xmath319 , which contradicts the weak incompressibility of @xmath235 . + if @xmath320 , we can proceed as we did with @xmath38 to conclude that an outermost arc @xmath321 cuts off a disk in @xmath39 with boundary consisting of the arcs @xmath321 and @xmath322 , where @xmath323 is an essential arc in @xmath235 . cap off @xmath321 to obtain a compression disk @xmath324 for @xmath235 . figure [ arcosbeta ] illustrates how @xmath312 and @xmath321 must lie in @xmath45 . + and @xmath321.[arcosbeta],width=377 ] + in cases 1 - 2 we see that @xmath325 and @xmath326 can be made disjoint , contradicting weakly incompressibility of @xmath235 . + in case 3 , @xmath325 and @xmath326 intersect in one point . let @xmath327 be the union of the bicollar of @xmath317 and the bicollar of @xmath324 along the square where they intersect . let @xmath328 . we can slightly move @xmath329 so that @xmath235 cuts it into two hemispheres , each one on opposite sides , and @xmath330 is a single curve @xmath61 . the curve @xmath61 cuts off from @xmath235 a punctured torus . if @xmath61 is inessential then either @xmath61 bounds a disk in @xmath235 or @xmath61 is boundary parallel . @xmath61 can not bound a disk in @xmath235 , otherwise @xmath235 would be a torus . @xmath61 is not boundary parallel , otherwise @xmath25 bounds a disk implying that @xmath25 is the unknot . thus , @xmath61 is essential in @xmath235 . by cutting @xmath329 into the two hemispheres and pushing the two boundaries apart , we produce disjoint compressing disks on opposite sides of @xmath235 , contradicting weakly incompressibility . if @xmath50 and @xmath51 are provided with a circular thin decomposition , then @xmath274 inherits a circular handle decomposition in which @xmath331 and @xmath332 are incompressible and @xmath263 and @xmath333 are weakly incompressible . in other words the circular handle decomposition inherited by @xmath274 is circular locally thin . now let us consider another operation on the exterior @xmath50 and @xmath51 of two knots @xmath24 and @xmath25 in @xmath14 . let @xmath245 and @xmath246 be seifert surfaces for @xmath24 and @xmath25 , respectively . let @xmath5 be the orientable closed 3-manifold obtained from @xmath50 and @xmath51 by identifying their boundaries via a homeomorphism @xmath52 such that @xmath53 , where @xmath334 for @xmath55 . under this identification @xmath245 and @xmath246 are glued together along their boundary via @xmath57 obtaining a closed embedded surface in m , the _ boundary sum _ of @xmath50 and @xmath51 . let us denote this surface by @xmath335 . recall @xmath23 is called a _ boundary sum _ of the seifert surfaces @xmath245 and @xmath246 . ( see definition [ boundary - sum ] ) . there is a natural way to obtain a circular handle decomposition for @xmath5 , starting with @xmath341 . we attach the sequence of handles corresponding to @xmath50 , i.e. , we attach @xmath9 and @xmath10 , along the @xmath245 component of @xmath342 . then we attach the sequence of handles of @xmath51 , i.e. , @xmath258 and @xmath259 , along the @xmath246 component of @xmath342 . so we have : defining circular width , as well as circular width with respect to a circular handle decomposition , in the obvious way for the closed manifold @xmath5 we have that the circular width of @xmath5 with respect to the circular handle decomposition @xmath34 ( the one induced by attaching first handles of @xmath50 ) , @xmath361 , is the set ; [ cweq ] let @xmath24 be a fibered knot with fiber @xmath245 . let @xmath25 be an almost fibered knot whose thin circular handle decomposition consists of one 1-handle and one 2-handle . let @xmath246 and @xmath48 be the thin and thick level surfaces , respectively of @xmath51 . suppose that @xmath246 is a minimal genus seifert surface . then the circular width of @xmath367 is given by : to prove this corollary we need to guarantee that the boundary sum of two minimal genus seifert surfaces is minimal genus in its homology class in the manifold @xmath5 . to accomplish this we invoke the following result proved in @xcite . [ imm - emb ] if @xmath5 is obtained by performing zero frame surgery on a knot @xmath4 in @xmath14 , then @xmath5 is prime and the genus of @xmath4= min\{genus @xmath369 is an embedded , oriented nonseparating surface}. proposition [ imm - emb ] implies that a genus @xmath370 knot @xmath4 can not have an orientable surface in the homology class of the seifert surface with genus less than @xmath370 , even if the surface is allowed to have more than one boundary component . [ mingen ] let @xmath24 and @xmath25 be two knots in @xmath14 and let @xmath245 and @xmath246 be minimal genus seifert surfaces for @xmath24 and @xmath25 , respectively . then @xmath371 is minimal genus in its homology class @xmath372 $ ] in @xmath367 . if @xmath373 is not minimal genus in its homology class @xmath372 $ ] , let @xmath20 be a representative in @xmath372 $ ] with smaller genus . let @xmath63 be the separating incompressible torus given by the image of @xmath374 in @xmath5 . @xmath20 intersects @xmath63 in a finite collection of essential closed curves . these curves are parallel to the preferred longitude which is @xmath375 . since @xmath63 separates @xmath5 , after splitting @xmath5 along @xmath63 the surface @xmath20 is split into the pieces @xmath376 and @xmath377 . since @xmath20 is of smaller genus than @xmath373 , then either the genus of @xmath378 or @xmath379 is smaller than @xmath245 or @xmath246 . contradicting proposition [ imm - emb ] . therefore @xmath373 is minimal genus in its homology class @xmath372 $ ] in @xmath5 . [ proof of corollary [ cweq ] ] @xmath5 inherits a circular handle decomposition @xmath34 from @xmath50 and @xmath51 , which consists of one 1-handle and one 2-handle . the circular handle decomposition @xmath34 has the thin level surface @xmath373 , which is minimal genus in its homology class @xmath372 $ ] in @xmath5 ( by lemma [ mingen ] ) , and the thick level surface @xmath380 . by corollary [ cw2 - 1 ] , @xmath305genus@xmath297 is an upper bound for the circular width of @xmath5 . if there were a thinner circular handle decomposition for @xmath5 , it would have a minimal genus surface as a thin level surface . the number of 1-handles would have to be fewer than those in @xmath34 , hence zero . thus @xmath5 would be fibered . but this is not possible unless both knots @xmath24 and @xmath25 are fibered . thus we have that @xmath368genus@xmath297 . if we assume that @xmath50 and @xmath51 are in circular thin position , then we can ask if the inherited presentation for @xmath5 is circular thin as well or even locally circular thin . under some conditions we are able to prove local circular thinness . we need the analog to lemma [ 52 ] . first we introduce some definitions which are similar to those in @xcite . let @xmath22 be a seifert surface for a knot @xmath4 in @xmath14 . @xmath22 is _ boundary compressible _ if there is a disk @xmath381 such that @xmath382 consists of an essential @xmath73 arc in @xmath22 and an arc @xmath383 in @xmath238 . if @xmath22 is not boundary compressible , it is said to be _ boundary incompressible_. @xmath34 is a _ boundary compressing disk_. @xmath22 is _ strongly boundary compressible _ if there are boundary compressing disks @xmath38 and @xmath39 on opposite sides of @xmath22 with disjoint boundaries , or a boundary compressing disk and a compressing disk on opposite sides of @xmath22 with disjoint boundaries . an incompressible seifert surface is boundary incompressible . if not , using the boundary compressing disk and the irreducibility of @xmath3 , we can find a compressing disk of the seifert surface since @xmath238 is a torus . the following lemma is a variation of lemma 3 in @xcite . the main change is made on the separability of @xmath22 . it is replaced by the hypothesis of being 2-sided . the proof of lemma [ lemma - wbi ] is similar to the proof of lemma 3 . for sake of completeness we include the proof here . 1 . if @xmath384 is an incompressible seifert surface and @xmath385 is a weakly boundary incompressible seifert surface , then @xmath386 is a weakly incompressible surface in @xmath387 . if @xmath144 is an incompressible seifert surface in @xmath336 for @xmath55 then @xmath388 is incompressible in @xmath387 . 1 . suppose @xmath23 is strongly compressible . then there exist nontrivial compressing disks @xmath38 and @xmath39 lying on opposite sides of @xmath23 with @xmath389 and @xmath390 a minimal collection of arcs . notice that @xmath391 . otherwise @xmath38 and @xmath39 are both contained in @xmath51 because @xmath234 is incompressible . this contradicts the assumption that @xmath235 is weakly incompressible . + then @xmath392 and @xmath393 can not happen at the same time , hence * @xmath394 and @xmath395 or * @xmath396 and @xmath397 . + let @xmath398 be an outermost arc in @xmath38 , and let @xmath399 be the arc in @xmath41 so that @xmath400 cuts off a disk . if @xmath401 is trivial in @xmath234 or @xmath235 , we can push it across @xmath63 ( taking any other arc with it ) , reducing @xmath402 . so @xmath401 is essential in @xmath234 or @xmath235 . + if @xmath401 is in @xmath234 then the disk bounded by @xmath403 and @xmath401 is a boundary compressing disk for @xmath234 . because @xmath234 is boundary incompressible this is not possible , so @xmath401 must be in @xmath235 . hence @xmath38 contains a boundary compressing disk @xmath317 for @xmath235 . @xmath317 lies on the same side as @xmath38 . + if @xmath394 then it is a compressing disk for @xmath235 . @xmath39 is on the opposite side of @xmath317 and @xmath325 is disjoint from @xmath42 . this contradicts the weak incompressibility of @xmath235 . + if @xmath404 is non - empty then , as with @xmath38 , an outermost arc argument implies that there is a boundary compressing disk @xmath324 for @xmath235 . @xmath324 lies on the same side as @xmath39 . the disks @xmath317 and @xmath324 are disjoint and on opposite sides of @xmath235 . this contradicts that @xmath235 is weakly boundary incompressible . + therefore @xmath405 is weakly incompressible . the proof of this case is similar but easier . if @xmath50 and @xmath51 are provided with circular thin decompositions respectively , and if we further assume that the thin level surfaces for both decompositions are boundary incompressible and the thick levels for both decompositions are weakly boundary incompressible , then @xmath406 inherits a circular locally thin decomposition . 999 g. burde , h. zieschang , _ neuwirthsche knoten und flachenabbildungen _ , hamburg , vol . 31 , 1967 , 239 - 246 . a. casson , c. gordon , _ reducing heegaard splittings _ , topology and its applications , vol . 27 , 1987 , 275 - 283 . d. gabai , _ the murasugi sum is a natural geometric operation _ , 20 , ams , 1983 , 131 - 143 . d. gabai , _ foliations and the topology of 3-manifolds iii _ , j. differential geom . 26 , 479 - 536 , 1987 . h. goda , _ on handle number of seifert surfaces in @xmath14 _ , osaka j. math . , vol . 30 , 63 - 80 , 1993 . h. goda , _ circle valued morse theory for knots and links _ , clay math . proc . , vol 5 , 2006 , 71 - 99 . j. johnson , a. thompson , _ on tunnel number one knots that are not ( 1,n ) _ , arxiv : math/0606226v3 [ math.gt ] . y. matsumoto , _ an introduction to morse theory _ , translations of mathematical monographs , vol . 28 , ams , 2002 . j. milnor , _ morse theory _ , princeton university press , 1969 . novikov , _ multivalued functions and functional , an analogue of the morse theory _ , soviet math . dokl . , vol . 24 , 1981 , 222 - 226 . a. pajitnov , _ circle - valued morse theory_. de gruyter studies in mathematics 32 a. pajitnov , l. rudolph , c. weber , _ morse - novikov number for knots and links _ , st . petersburg math . 13 , no.3 , 2002 , 417 - 426 . a. ranicki , _ circle valued morse theory and novikov homology _ , lecture notes at the summer school on high - dimensional manifold topology , trieste , 2001 . d. rolfsen , _ knots and links _ , ams chelsea publishing,1976 . m. scharlemann , j. schultens , _ annuli in generalized heegaard splittings and degeneration of tunnel number _ , math . 317 ( 2000 ) , 4 , 783 - 820 . m. scharlemann , a. thompson , _ thin position for 3-manifolds _ , contemp . 164 , ams , 1994 , 231 - 238 . f. waldhausen , _ on irreducible 3-manifolds which are sufficiently large _ , ann . of math . ( 2 ) 87 , 1968 , 56 - 88 . w. whitten , _ isotopy types of knot spanning surfaces _ , topology , vol . 12 , 1973 , 373 - 380 .	a regular circle - valued morse function on the knot complement @xmath0 is a function @xmath1 which separates critical points and which behaves nicely in a neighborhood of the knot . such a function induces a handle decomposition on the knot exterior @xmath2 , with the property that every regular level surface contains a seifert surface for the knot . we rearrange the handles in such a way that the regular surfaces are as simple " as possible . to make this precise the concept of _ circular width for @xmath3 _ is introduced . when @xmath3 is endowed with a handle decomposition which realizes the circular width we will say that the knot @xmath4 is in _ circular thin position_. we use this to show that many knots have more than one non - isotopic incompressible seifert surface . we also analyze the behavior of the circular width under some knot operations .
the high spectral resolution now available in the x - ray band led to a deeper understanding of the physical processes occurring in active galaxies . is a type 1 seyfert galaxy . for these objects , according to the standard model ( e.g. * ? ? ? * ) , our line of sight should intercept an outflowing , highly ionized gas . it is now recognized that many components , differing in ionization level , column density , and outflowing velocity , are actually constituting this outflow ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and that the uv and the x - ray band share at least some of these components ( e.g. * ? ? ? * ; * ? ? ? yet , the geometry , the distance , and therefore the density of the x - ray outflow is not yet clear ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the observation angle for seyfert 1 objects allows to investigate the innermost parts of the agn , from the broad line region ( blr , see e.g. * ? ? ? * ) , at a distance of few light days from the black hole , down to the accretion disk . the bulk of the x - rays are produced here ( e.g. * ? ? ? the blr is likely to produce x - ray radiation , but up to now this radiation was never convincingly observed in x - ray spectra @xcite . we processed the -letgs data of following the approach described in @xcite . the residuals to a basic continuum model show significant excess emission , especially in the 18 - 24 wavelength range , where the main oxygen line features are ( triplet and ) . we interpret these emission features as broad lines , possibly arising from the blr . the best fit shows that the excess can be modeled by gaussian profiles for , , , and . evidence of weaker broad excesses is found for and ( fig . [ f : best_fit ] ) and possibly neon ( see next paragraph for the modeling ) . the inclusion of broad emission lines is required to properly fit the absorbed spectrum . the best fit requires indeed absorption by two components of ionized gas . for the lower ionized component we find : @xmath0@xmath1 , and [email protected] , while for the high ionization component the parameters are : n@xmath3@xmath1 and [email protected] . the two absorbers seem to belong to systems with different outflow velocity , -220@xmath4 and -570@xmath5 kms@xmath6 for the low and high ionization gas , respectively . another component , whose main features are and has been modeled by a collisionally ionized gas at a temperature @xmath7 ev . shows also ionized absorption taking place at z=0 @xcite . in the letgs band the deepest features of this collisionally ionized component are and . thanks to the simultaneous fuse and hst observation @xcite , we could compare the x - ray line luminosity with the ones measured for the uv broad emission lines . we used cloudy @xcite for the modeling . the input ionizing spectral continuum was evaluated directly from the hst , fuse and data . we chose a grid of physical parameters for the blr clouds and then chose the best fit to our data through @xmath8 minimization . in fig . [ f : cloudy ] we display the intrinsic integrated luminosity measured for the x - ray and uv lines ( star points ) _ vs _ the rest frame wavelength of the lines , together with our best fit . two ionization components for the emitting medium are certainly needed in order to explain the uv and the lower ionization x - ray lines ( labeled as cold " and warm " in fig . [ f : cloudy ] ) . this result is based mainly on the more precise uv measurements . the higher ionization x - ray lines , such as , , , , and may require the contribution of a higher ionization component ( hot " in fig.[f : cloudy ] ) . the ionization parameters of these components are roughly log@xmath2 -1.5 , 1.6 and 3.6 for the cold " , warm " , and hot " components , respectively . the distance from the ionizing source of the emitting cloud is around @xmath9 cm , and the density ranges between @xmath10@xmath11 . the ionization parameter is given by : @xmath12 . here only the ionizing luminosity @xmath13 is directly measured . density effects on the emission may help in determining the parameters in the @xmath14 term . however , as a caveat , the @xmath15 and @xmath16 values are still coupled . r. antonucci , 1993 , ara&a , 31 , 473 n. arav , j. s. kaastra , g. a. kriss , k. t. korista , j. gabel , & d. proga , 2004 , astro - ph/0406200 j. a. baldwin , 1997 , asp conf . ser . 113 : iau colloq . 159 : emission lines in active galaxies : new methods and techniques , 80 e. behar et al . 2003 , apj , 598 , 232 g. j. ferland , et al . 1998 , pasp , 110 , 761 j. s. kaastra et al . 2000 , a&a , 354 , 84 j. s. kaastra et al . 2002 , a&a , 386 , 427 j. s. kaastra , n. arav , e. costantini et al . 2005 , in prep . s. kaspi et al . 2001 , apj , 554 , 216 y. krongold , f. nicastro , n. s. brickhouse , et al . 2003 , apj , 597 , 832 k. nandra , & k. a. pounds , 1994 , mnras , 268 , 405 p. m. ogle , k. o. mason , m. j. page , et al . 2004 , apj , 606 , 151 k. c. steenbrugge et al . , 2004 , a&a , accepted .	we present the -letgs analysis of , a bright seyfert 1 galaxy . the spectrum shows a variety of features arising from different physical environments in the vicinity of the black hole . we detect three absorption components , located at the redshift of the source . one of them is likely to arise from the host galaxy . an additional component due to absorption by a collisionally ionized gas at z=0 is also observed . such an absorption can be produced in the outskirts of our galaxy or in the local group . the emission spectrum is rich in narrow and broad emission features . we tested the hypothesis that the broad emission lines originate in the broad line region ( blr ) , as studied in the uv band . we find that at least 2 components with different ionization parameters are necessary to simultaneously interpret the broad emission line measured in the uv ( by and fuse ) and in the x - rays . address = sron , sorbonnelaan , 2 , 8534ca utrecht , the netherlands , altaddress = astronomical institute , utrecht university , p.o . box 80000 , 3508ta utrecht , the netherlands address = sron , sorbonnelaan , 2 , 8534ca utrecht , the netherlands address = sron , sorbonnelaan , 2 , 8534ca utrecht , the netherlands address = casa , university of colorado , 389 ucb , boulder , co , usa address = casa , university of colorado , 389 ucb , boulder , co , usa address = stsi , 3700 san martin drive , baltimore , md 21218 , usa
our starting point concerns nonnegative solutions with finite mass of the heat equation in @xmath0 @xmath1 it is straightforward to check that for any smooth enough solution of and any @xmath2 convex function @xmath3 , @xmath4 so that @xmath5 plays the role of a lyapunov functional . to extract some information out of such an identity , one needs to analyze the relation between @xmath5 and @xmath6 . this can be done using green s function or moment estimates , with the drawback that these quantities are explicitly @xmath7-dependent . it is simpler to rewrite the equation in self - similar variables and replace by the fokker - planck ( fp ) equation @xmath8 this can be done without changing the initial data by the time - dependent change of variables @xmath9 we shall restrict our approach to nonnegative initial data @xmath10 . by linearity , we can further assume that @xmath11 without loss of generality . we shall also assume that @xmath3 is defined on @xmath12 . up to the change of @xmath3 into @xmath13 such that @xmath14 , we can also assume that @xmath3 is nonnegative on @xmath12 and achieves its minimum value , zero , at @xmath15 . . has a unique nonnegative stationary solution @xmath16 normalized such that , namely @xmath17 if we introduce @xmath18 , then @xmath19 is a solution of the ornstein - uhlenbeck , or kolmogorov - fokker - planck ( kfp ) , equation @xmath20 with initial data @xmath21 . after identifying @xmath22 with the measure @xmath23 , the relevant lyapunov functional , or _ entropy _ , is @xmath24 and @xmath25 we shall restrict our study to a class of functions @xmath3 for which the entropy and the entropy production functional are related by the inequality @xmath26 for some @xmath27 ( it turns out that in the case of the gaussian measure we can choose @xmath28 ) . this allows us to prove that the entropy is exponentially decaying , namely @xmath29 if @xmath19 is a solution of and if @xmath30 is positive . a sufficient condition for such an inequality is that @xmath31 ( see for instance @xcite ) . at first sight , this may look like a technical condition but it has some deep implications . we are indeed interested in exhibiting a gradient flow structure for associated with the entropy or , to be more precise , to establish that , for some distance , the gradient flow of the entropy is actually . it turns out that is the natural condition as we shall see in section [ subsec : actiondensity ] . the entropy decays exponentially according to not only when one considers the @xmath32 norm ( the norm of the square integrable functions with respect to the gaussian measure @xmath22 ) , i.e. the case @xmath33 , or the classical entropy built on @xmath34 , for which is the gradient flow with respect to the usual wasserstein distance ( according to the seminal paper @xcite of jordan , kinderlehrer and otto ) . we also have an exponential decay result of any entropy generated by @xmath35 and more generally any @xmath3 satisfying . notice by the way that @xmath36 is compatible with if and only if @xmath37 and that @xmath34 appears as the limit case when @xmath38 . the exponential decay is a striking property which raises the issue of the hidden mathematical structure , a question asked long ago by f. poupaud . as already mentionned , the answer lies in the gradient flow interpretation and the construction of the appropriate distances . such distances , based on an action functional related to @xmath3 , have been studied in @xcite . our purpose is to exploit this action functional for the construction of gradient flows , not only in the case corresponding to but also for kfp equations based on general @xmath39-convex potentials @xmath40 . for the convenience of the reader , the main steps of the strategy have been collected in section [ sec : formal ] , without technical details ( for instance on the measure theoretic aspects of our approach ) . coming back to our basic example , namely the solution of , we may observe that a solution can easily be represented using the green kernel of the heat equation and our time - dependent change of variables . if @xmath36 , @xmath37 , we may observe that the exponential decay of the entropy can be obtained using the known properties of the heat flow and the homogeneity of @xmath41 , while the contraction properties of the heat flow measured in the framework of the weighted wasserstein distances introduced in @xcite can be translated into the exponential decay of the distance of the solution of to the gaussian measure @xmath22 , if we assume that @xmath42 is a probability measure . we shall however not pursue in this direction as it is very specific of the potential @xmath43 and of the heat flow ( for which an explicit green function is available ) . let us conclude this introductory section by a brief review of the literature on the functional inequalities based on entropies such that holds . such functionals are sometimes called _ @xmath44-entropies_. in this paper , we shall however avoid this denomination to prevent from possible confusions with the function @xmath45 and the functional @xmath46 used below to define the action and the weighted wasserstein distances @xmath47 . we shall refer to @xcite for a probabilistic point of view . a proof of under assumption and an hypothesis of convexity of @xmath40 can be found for instance in @xcite or in the more recent paper @xcite . this approach is based on the bakry - emery method @xcite and heavily relies on the flow of kfp or , equivalently , on the geometric properties of the ornstein - uhlenbeck operator ( using the _ carr du champ : _ see @xcite ) . strict convexity of the potential is usually required , but can be removed afterwards by various methods : see @xcite . for capacity - measure approaches of , we shall refer to @xcite . the inequality itself has been introduced in @xcite with a proof based on the hypercontractivity of the heat flow and spectral estimates , and later refined and adapted to general potentials in @xcite . concerning gradient flows and distances of wasserstein type , there has been a huge activity over the last years . we can refer to @xcite for fundamental ideas , and to two books , @xcite , for a large overview of the field . many other contributions in this area will be quoted whenever needed in the proofs . in section [ sec : intro ] , we have considered the case of the harmonic potential @xmath43 . we generalize the setting to any smooth , convex _ potential _ @xmath48 with @xmath49 and consider the _ reference measure _ @xmath22 given by @xmath50 where @xmath51 denotes lebesgues s measure on @xmath0 . we assume that @xmath52 next we define the _ action density _ @xmath53 as @xmath54 for some concave , positive , non decreasing function @xmath55 with _ sublinear _ growth . the function @xmath56 is therefore convex and also satisfies the condition @xmath57 our main example is @xmath58 for some @xmath59 . based on the action density , we can define the _ action functional _ by @xmath60 * the kolmogorov - fokker - planck ( kfp ) equation . * with the notations @xmath61 , the equation @xmath62 determines the _ kolmogorov - fokker - planck ( kfp ) flow _ @xmath63 . its first variation , @xmath64 , can be obtained as the solution of the _ modified kolmogorov - fokker - planck equation _ @xmath65 if @xmath66 , then @xmath67 , which can be summarized by @xmath68 by duality , using the notations @xmath69 and @xmath70 , if @xmath71 , we also find that @xmath72 , which amounts to @xmath73 ( see theorem [ thm : density_gradient_link ] for details ) . if @xmath74 , we define the semigroup @xmath75 acting on measures by @xmath76 . consider an _ entropy density _ function @xmath3 such that @xmath77 . if we define the _ entropy _ functional by @xmath78 and the _ entropy production _ , or generalized fisher information functional , as the action functional for the particular choice @xmath79 , i.e. @xmath80 then , along the kfp flow , we get @xmath81 for a solution @xmath19 of if @xmath82 notice that and are equivalent . see section [ subsec : actiondensity ] for more details . the main estimate for this paper goes as follows . [ thm : mainestimate ] under assumptions , if @xmath83 , @xmath84 and @xmath85 , then @xmath86 in particular the action functional decays exponentially if @xmath39 is positive : @xmath87 at formal level , this follows by an easy convexity argument . the rigorous proof requires many regularizations . see theorem [ thm : action_decay ] for a more detailed version of this result . now let us review some of the consequences of theorem [ thm : mainestimate ] . * entropy , entropy production and generalized poincar inequalities . * we can now apply theorem [ thm : mainestimate ] to the kfp flow . with @xmath79 , we find that the _ entropy production _ functional decays exponentially : @xmath88 if @xmath39 is positive . by integrating along the kfp flow when @xmath7 varies in @xmath12 , using and @xmath89 , we recover for @xmath90 the _ generalized poincar inequalities _ @xmath91 found by beckner in @xcite in the case of the harmonic potential and for @xmath58 , @xmath59 , and generalized for instance in @xcite . such inequalities interpolate between poincar and logarithmic sobolev inequalities . if we combine with , we find that the _ entropy _ decays according to @xmath92 by integrating from @xmath93 to @xmath7 the inequality @xmath94 which itself follows from and , we observe a _ first regularization effect _ along the kfp flow , namely @xmath95 if @xmath39 is positive , we can refine this estimate and actually prove by the same method that @xmath96 for any @xmath97 . * the @xmath55-wasserstein distance . * if @xmath98 is a measure with absolutely continuous part @xmath99 with respect to @xmath22 , and singular part @xmath100 , if @xmath101 is a vector valued measure which is absolutely continuous with respect to @xmath22 and has a modulus of continuity @xmath102 , i.e. if @xmath103 we can extend the action functional @xmath46 to the measures @xmath98 and @xmath101 by setting @xmath104 we shall say that there is an _ admissible path _ connecting @xmath105 to @xmath106 if there is a solution @xmath107}$ ] to the _ continuity equation _ @xmath108\;,\ ] ] and will denote by @xmath109 the set of all admissible paths . with these tools , we can define the _ @xmath55-wasserstein distance _ between @xmath105 and @xmath106 by @xmath110 notice that @xmath55 in `` @xmath55-wasserstein distance '' refers to the dependence of @xmath46 in @xmath55 through the action density @xmath45 , the usual wasserstein distance corresponding to @xmath111 . if @xmath112 is a curve of measures , its _ @xmath55-wasserstein velocity _ @xmath113 is determined by @xmath114 using the decomposition , we compute the _ derivative of the entropy along the curve _ @xmath112 as @xmath115 and find that @xmath116 by the cauchy - schwarz inequality . along the kfp flow , we know that @xmath117 which is the equality case in . this characterizes the kfp flow as the steepest descent flow of the entropy @xmath118 , i.e. this is a first charaterization of kfp as _ the gradient flow of @xmath118 _ with respect to the @xmath55-wasserstein distance . the kfp flow connects @xmath74 with @xmath119 and it has been established in @xcite that one can estimate the length of the path by @xmath120 ( see section [ subsec : weightedwasserstein ] for details ) . according to , we get @xmath121 this establishes the _ entropy production distance estimate _ @xmath122 along the kfp flow , we also find that @xmath123 using . by applying , this establishes the _ ( talagrand ) entropy distance estimate _ @xmath124 * contraction properties and gradient flow structure . * here as in @xcite , we use the technique introduced in @xcite and extended in @xcite : we consider a geodesic ( or an approximation of a geodesic ) , and evaluate the derivative of the action functional along a family of curves obtained by evolving the geodesic with the kfp flow . consider an @xmath125-geodesic @xmath126 connecting @xmath127 to @xmath128 , i.e. an admissible path in @xmath109 such that @xmath129 for any @xmath130 and observe that by , we know that @xmath131 is still an admissible curve connecting @xmath132 to @xmath133 . therefore yields @xmath134 which , by letting @xmath135 , proves that the kfp flow _ contracts the distance : _ @xmath136 see theorem [ thm : main_contraction ] for more details . next , we should again consider an @xmath125-geodesic , but for simplicity we assume that there is a geodesic @xmath126 connecting @xmath137 to @xmath138 , i.e. such that @xmath139 , and consider the path @xmath140 connecting @xmath141 to @xmath142 . notice that our notations mean that @xmath143 . since @xmath144 the path is admissible and , as a consequence , @xmath145 we can therefore differentiate the right hand side in the above inequality instead of the distance and furthermore notice that it is sufficient to do it at @xmath146 ; see theorem [ thm : metric_characterization ] and its proof for details . along the kfp flow we find that @xmath147 this is the strongest metric formulation of a _ @xmath39-contracting gradient flow_. here we have defined the relative entropy as @xmath148 if @xmath149 and @xmath74 , and @xmath150 otherwise . hence we recover a second characterization of the fact that kfp _ is the gradient flow of @xmath118 _ with respect to @xmath47 . as another consequence , the entropy _ @xmath118 is geodesically @xmath39-convex_. this follows from . fix a geodesic @xmath151 between @xmath152 and @xmath153 , follow the evolution of @xmath151 by kfp taking first @xmath152 and then @xmath153 fixed , and apply with @xmath154 and @xmath155 or @xmath156 . because of the minimality of the energy along the geodesic at time @xmath146 , by summing the two resulting inequalities we prove the convexity inequality of @xmath118 . see ( * ? ? ? * theorem 3.2 ) for more details . as a final observation , let us notice that , directly from the metric formulation , it follows that the kfp flow also has the following _ regularizing properties : _ @xmath157 the first estimate can indeed be obtained by integrating ( with @xmath158 and @xmath159 ) from @xmath93 to @xmath7 and recalling that @xmath160 is decreasing . as for the second one , we observe that also @xmath161 is decreasing by , so that and yield @xmath162 a further integration in time from @xmath93 to @xmath7 completes the proof . notice that it is crucial to start from a measure @xmath163 at _ finite _ distance from @xmath22 . in this section we first recall some definitions and results taken from @xcite . the measure @xmath22 and the functions @xmath45 and @xmath3 are as in section [ sec : formal ] , and we assume that conditions are satisfied . let @xmath48 be a @xmath39-convex and continuous potential . we assume that @xmath39 is nonnegative and @xmath39-convexity means that the map @xmath164 is convex . when @xmath40 is smooth in @xmath0 , this condition is equivalent to . we are assuming that @xmath165 is integrable in @xmath0 , so that we can introduce the finite , positive , log - concave measure @xmath22 defined by . for simplicity , we shall assume that @xmath22 is a probability measure , i.e. @xmath166 , which can always be enforced by replacing @xmath40 by @xmath167 . the potential @xmath40 being convex , the integrability of @xmath165 is equivalent to the property that @xmath168 at least linearly as @xmath169 ; see e.g. ( * ? ? ? * appendix ) . as a consequence , there exist two constants @xmath170 , @xmath171 such that @xmath172 we recall that non smooth , convex potentials @xmath40 can be approximated from below by an increasing sequence of convex potentials @xmath173 : @xmath174 moreover , the potentials @xmath173 are @xmath39-convex and , even in the case @xmath158 , they satisfy conditions with respect to constants @xmath175 and @xmath176 which are _ independent of @xmath177 . _ in particular , the log - concave measures @xmath178 weakly@xmath179 and monotonically converge in @xmath180 to @xmath22 . by this regularization techniques , many results could be extended to the case when @xmath40 is just lower semicontinuous and can take the value @xmath181 . as in section [ sec : formal ] , consider @xmath56 and @xmath55 on @xmath182 such that @xmath183 and @xmath184 . the following result has already been observed in @xcite but we reproduce it here for completeness . [ le : action1 ] with the notations of section [ sec : formal ] , the action density @xmath45 is _ convex _ if and only if @xmath55 is concave on @xmath182 or , equivalently , if @xmath56 satisfies condition . by standard approximations , it is not restrictive to assume that @xmath56 , @xmath185 . first of all observe that @xmath186 so that @xmath187 is nonpositive if and only if @xmath188 . next we evaluate the second derivative of @xmath45 along the direction of the vector @xmath189 as @xmath190 by minimizing with respect to @xmath191 , we get @xmath192\ ] ] if @xmath193 , with equality for the appropriate choice of @xmath194 . the convexity of @xmath45 is thus equivalent to @xmath195 if @xmath45 is convex , by choosing @xmath196 and using @xmath197 , we get @xmath198 ^ 2\quad\forall\;\rho>0\,,\quad\forall\;{{\mbox{\boldmath$w$}}}\in{{{\mathbb{r}}^d}}\,,\ ] ] which yields . conversely , the convexity of @xmath45 follows from @xmath199 . we can introduce a _ modulus of convexity _ as follows . assume that for some @xmath200 $ ] we have @xmath201 by , we obtain the refined estimate @xmath202 such a refinement has interesting consequences , which have been investigated in @xcite . the refined convexity assumption is equivalent to @xmath203 [ rem : main_example ] our main example is provided by the function @xmath204 which satisfies . when @xmath205 we simply get @xmath206 and for @xmath207 we have the @xmath208-homogeneous functional @xmath209 notice that the above considerations can be generalized to matrix - valued functions @xmath56 and @xmath55 : see ( * ? ? ? * example 3.4 ) . the _ action functional _ @xmath46 induced by @xmath45 has been defined by , with domain @xmath210 assuming as in section [ subsec : actiondensity ] that @xmath45 convex , it is well known that if @xmath211 and @xmath212 are such that @xmath213 for any @xmath214 and if @xmath215 in @xmath216 , and @xmath217 as @xmath218 , then by lower semi - continuity of @xmath46 , we have @xmath219 [ le : smooth_approximations ] consider two functions @xmath220 and @xmath221 such that @xmath222 and @xmath223 . then there exist two sequences @xmath211 and @xmath212 of bounded smooth functions ( with bounded derivatives of arbitrary orders ) such that @xmath224 and @xmath225 we first truncate @xmath99 and @xmath102 from above as follows . let @xmath226 and , for any @xmath214 , @xmath227 @xmath228 . we set @xmath229 and @xmath230 clearly @xmath231 , @xmath232 pointwise @xmath22 a.e . in @xmath0 , so that fatou s lemma yields @xmath233 since @xmath234 in @xmath216 as @xmath235 , we have @xmath236 and @xmath231 in @xmath216 . the dominated convergence theorem also yields @xmath232 in @xmath237 . finally , since @xmath238 and @xmath239 on @xmath240 , and since @xmath56 is non increasing , @xmath241 so that the `` @xmath242 '' in is in fact a limit . next we perform a lower truncation on @xmath99 . by a diagonal argument , it is sufficient to approximate the functions @xmath243 and @xmath244 we have just introduced , so we can assume that @xmath99 is essentially bounded by a constant @xmath245 and we omit the dependence on @xmath245 . for @xmath246 we now set @xmath247 . observe that @xmath248 so that @xmath249 on the set @xmath250 and , by convexity of @xmath56 , we get @xmath251 on the other hand , on the set @xmath252 , we have @xmath253 , and then @xmath254 . as a consequence , @xmath255 we can then pass to the limit as @xmath256 , since @xmath257 pointwise . the last step is to approximate the functions @xmath99 and @xmath102 , with @xmath258 , @xmath259 , by smooth functions . we consider a family of smooth approximations @xmath260 and @xmath261 obtained by convolution with a smooth kernel . we finally set @xmath262 and , in this framework , redefine @xmath263 . since @xmath264 converges to @xmath265 pointwise a.e . in @xmath0 and is uniformly bounded , we can pass to the limit as above when @xmath266 . since we assumed that @xmath55 is concave and strictly positive for @xmath267 , @xmath55 is an increasing map , so that @xmath56 is decreasing . we extend @xmath55 and @xmath56 to @xmath268 by continuity and we still denote by @xmath45 the lower semicontinuous envelope of @xmath45 in the closure @xmath269 . if @xmath270 then @xmath271 and @xmath272 . when @xmath273 we have @xmath274 and @xmath275 we also introduce the _ recession functional _ @xmath276 which is still a convex and lower semicontinuous function with values in @xmath277 $ ] , and @xmath208-homogeneous . it is determined by the behaviour of @xmath278 as @xmath279 . if we set @xmath280 we have @xmath281 and @xmath282 let @xmath283 be a nonnegative radon measure and let @xmath284 be a vector radon measure on @xmath0 . we write their lebesgue decomposition with respect to the reference measure @xmath22 as @xmath285 we can always introduce a nonnegative radon measure @xmath286 such that @xmath287 , @xmath288 , e.g. @xmath289 and define the _ action functional _ @xmath290 since @xmath291 is @xmath208-homogeneous , this definition is independent of @xmath141 . as we have done up to now , we shall simply write @xmath292 when there is no ambiguity on the reference measure @xmath22 . if @xmath55 has a _ sublinear growth _ , then @xmath293 and , as a consequence , if @xmath294 , then we have @xmath295 so @xmath296 is independent of the singular part @xmath100 . when @xmath55 has a _ linear growth _ , i.e. @xmath297 , if @xmath294 , then we have @xmath298 in both cases , one can choose @xmath299 , so that @xmath300 and if @xmath301 is finite , then we have @xmath302 while the last term simply drops if @xmath293 . [ le : action_lsc ] the action functional is lower semicontinuous with respect to the weak convergence of measures , i.e. if @xmath303 , @xmath304 and @xmath305 are sequences such that @xmath306 weakly in @xmath307 , @xmath308 weakly in @xmath307 and @xmath309 in @xmath310 as @xmath311 , then @xmath312 moreover , for every @xmath283 and @xmath284 such that @xmath313 , there exist sequences @xmath304 and @xmath305 for which @xmath314 such that @xmath315 the first statement is a well known fact about lower semicontinuity of convex integrals ( see e.g. @xcite ) . concerning the approximation property , general relaxation results provide a family of approximations in @xmath216 . we can then apply lemma [ le : smooth_approximations ] and a standard diagonal argument . denote by @xmath316 the collection of all borel subsets of @xmath0 , by @xmath307 the collection of all finite positive borel measures defined on @xmath0 and by @xmath317 the convex subset of all probability measures i.e. all @xmath283 such that @xmath318 . if @xmath310 is the set of the vector valued borel measures @xmath319 with finite variation , i.e. such that @xmath320 for any @xmath321 , then @xmath322 is in fact a finite positive measure in @xmath307 and @xmath101 admits the polar decomposition @xmath323 where the borel vector field @xmath102 belongs to @xmath324 . we can also consider @xmath101 as a vector @xmath325 of @xmath326 measures in @xmath327 . for any @xmath328 , let @xmath329 be the set of time dependent measures @xmath330}$ ] , @xmath331 such that 1 . @xmath332 is weakly @xmath179 continuous in @xmath333 , 2 . @xmath331 is a borel family with @xmath334 for any @xmath335 , 3 . @xmath336 is a distributional solution of @xmath337 as in @xcite , we define the weighted wasserstein distance as follows . [ def : main ] the @xmath338-wasserstein distance between @xmath105 and @xmath339 is defined by @xmath340^{1/2}\,:\\ ( \mu,{{\mbox{\boldmath$\nu$}}})\in\mathcal{ce}(0,1;{{{\mathbb{r}}^d}})\,,\;\mu_{t=0}=\mu_0\,,\;\mu_{t=1}=\mu_1\big\}\end{gathered}\ ] ] with @xmath341 if @xmath342 and @xmath343 , @xmath344 otherwise , and @xmath345 . we denote by @xmath346 $ ] the set of all measures @xmath347 which are at finite @xmath348-distance from @xmath141 . notice that in @xcite we were using the notation @xmath349 instead of @xmath348 . whenever there is no ambiguity on the choice of the measure @xmath22 , we shall simply write @xmath47 . the next result is taken from ( * ? ? ? * theorem 5.6 and proposition 5.14 ) [ thm:2 ] if @xmath45 satisfies and , the map @xmath350 is lower semicontinuous with respect to the weak @xmath179 convergence in @xmath333 . more generally , suppose that @xmath351 in @xmath333 , @xmath352 is monotonically decreasing w.r.t . @xmath177 and pointwise converging to @xmath55 , and @xmath353 , @xmath354 in @xmath333 as @xmath311 . then @xmath355 if moreover @xmath356 we have @xmath357 it is possible to reparametrize the path connecting @xmath105 to @xmath106 in the definition of @xmath348 and establish that , for any @xmath328 , @xmath358^{1/2}\,:(\mu,{{\mbox{\boldmath$\nu$}}})\in{\mathcal ce}(0,t;\sigma\to\eta)\big\}\ ] ] where @xmath359 denotes the set of the paths @xmath360 such that @xmath361 and @xmath362 . by ( * theorem 5.4 and corollary 5.18 ) , we have the [ existence : geodesics ] whenever the infimum in has a finite value , it is attained by a curve @xmath363 such that @xmath364 in this case we have the equivalent characterization @xmath365^{1/2}dt}\,:\,(\mu,{{\mbox{\boldmath$\nu$}}})\in{\mathcal ce}(0,t;\sigma\to\eta)\big\}\,.\ ] ] the curve @xmath366}$ ] associated to a minimum for is a constant speed mimimal geodesic : @xmath367\;.\ ] ] we may notice that the characterization of @xmath368 in terms of @xmath369 allows to consider the case @xmath370 . by ( * ? ? ? 1 ) ( also see @xcite ) , one knows that @xmath371 for any absolutely continuous curve @xmath372 such that @xmath373 and @xmath374 . now let us come back to the formal point of view of section [ sec : formal ] and establish in this framework . assume that @xmath19 is given by kfp and @xmath67 . the curve @xmath375 connects @xmath376 with @xmath119 and , using @xmath377 it follows that @xmath378 as already noted in section [ sec : formal ] ( equality case in ) . on the other hand , for any @xmath379 , @xmath380 , we have @xmath381 and so @xmath382 by taking first the infimum @xmath379 and then the limit @xmath383 , we also find @xmath384 thus proving the equality in the above inequality . this completes the proof of . let us consider now a function @xmath3 such that @xmath385 for any @xmath386 . among all possible choices of @xmath3 , we consider in particular the convex functions @xmath387 depending on @xmath388 and characterized by the conditions @xmath389 observe that @xmath390 has a strict minimum at @xmath388 and it satisfies the transformation rule @xmath391 independently of the choice of @xmath3 ( for a given function @xmath56 ) . when @xmath392 we obtain the logarithmic entropy density @xmath393 and the family @xmath394 which provides useful lower / upper bounds for @xmath3 . in fact , @xmath55 being concave , if @xmath273 , then @xmath395 if @xmath396 , so that @xmath397\;.\ ] ] on the other hand , when @xmath398 , we have @xmath399 , so that @xmath400 thus showing that @xmath401 has a superlinear growth as @xmath402 . we can therefore introduce the _ relative entropy functional _ @xmath403 in the particular case @xmath404 , we set @xmath405 since @xmath3 is convex and superlinearly increasing , if @xmath406 , then there exists a subsequence weakly converging to @xmath99 in @xmath216 and @xmath407 [ le : mccann ] if the function @xmath3 satisfies @xmath408 , @xmath409 and if holds , then @xmath3 also satisfies mccann s conditions , i.e. the map @xmath410 is convex and non increasing on @xmath182 or , equivalently , @xmath411 the convexity of @xmath3 indeed yields @xmath412 . consider the function @xmath413 and observe that @xmath414 , since @xmath415 and @xmath55 is concave so that , in particular , @xmath416 near @xmath417 , for some positive constant @xmath418 . on the other hand , we have @xmath419 and the function @xmath420 being positive , non increasing , we deduce that , so that @xmath421 . let us introduce the sobolev spaces @xmath422 for @xmath423 , @xmath222 , we define the _ entropy production functional _ as @xmath424 we also introduce the absolutely continuous functions @xmath425 and observe that @xmath426 is bounded if and only if @xmath427 has a linear growth as @xmath428 . in the case @xmath429 , @xmath404 , to the entropy functional @xmath430 corresponds the _ entropy production functional _ @xmath431 [ thm : production ] let @xmath99 be nonnegative function in @xmath216 . then @xmath423 and @xmath432 if and only if @xmath433 and in this case we have @xmath434 if @xmath435 and @xmath436 for some constant @xmath437 , then @xmath438 , @xmath439 moreover , the functional @xmath440 is lower semicontinuous with respect to the weak convergence in @xmath216 , i.e. if a sequence @xmath441 weakly converges to some @xmath99 in @xmath216 and @xmath442 , then @xmath423 and @xmath443 identity and @xmath444 are straightforward if @xmath99 takes its values in a compact interval of @xmath182 . the general case follows as in lemma [ le : smooth_approximations ] by a standard truncation argument , while the lower semicontinuity is a consequence of convexity . as in section [ sec : formal ] , let us introduce the differential operators @xmath445 which , with respect to the measure @xmath22 , satisfy the following `` integration by parts formulae '' against test functions @xmath446 : @xmath447 we consider the kolmogorov - fokker - planck equation @xmath448 for simplicity , we will consider equations in the whole @xmath449 ( corresponding to the finiteness assumption on the potential @xmath40 ) ; necessary adaptations when this is not the case are straightforward and left to the reader . we will also assume that @xmath450 based on the integration by parts formula , the variational formulation of in the hilbert space @xmath451 relies on the symmetric , closed dirichlet form @xmath452 where @xmath453 is endowed with its natural norm @xmath454 . using smooth approximations , it is not difficult to prove that @xmath453 is dense in @xmath451 . the abstract theory of variational evolution equation and the log - concavity of the measure @xmath22 yield the following result ( see e.g. ( * * thm . 6.7 ) ) . [ prop : fkpre ] assume that hold . for every @xmath455 , the solution of has the following properties : 1 . there exists a unique @xmath456 , @xmath457 , such that @xmath458 if @xmath459 , then @xmath19 satisfies the same uniform bounds . the semigroup @xmath460 is an analytic markov semigroup in @xmath451 which can be extended by continuity to a contraction semigroup in @xmath461 for every @xmath462 and to a weakly @xmath179 continuous semigroup in @xmath463 . 2 . for every @xmath99 , @xmath464 , we have @xmath465 3 . for every @xmath457 , @xmath466 maps @xmath463 into @xmath467 and @xmath468 into itself , with the uniform bound @xmath469_{{\rm lip}({{{\mathbb{r}}^d}})}\leq [ \rho]_{{\rm lip}({{{\mathbb{r}}^d}})}\quad\forall\;t\geq 0\;,\quad\forall\;\rho\in { \rm lip}_b({{{\mathbb{r}}^d}})\;.\ ] ] 4 . if @xmath470 , @xmath471 and @xmath472 , then the map @xmath473 is convex , @xmath474 for every time @xmath457 , and @xmath475}\int_{{{\mathbb{r}}^d}}\|x\|^2\rho_t\,d\gamma<\infty\;,\quad\frac d{dt}\,\mathcal h(\rho_t)=-\mathcal i(\rho_t)\;,\quad\frac d{dt}\big(e^{2\lambda t}\,\mathcal i(\rho_t)\big)\le 0\;.\ ] ] notice that the assumption @xmath455 is not needed in property 4 , according to ( * ? ? ? 6.7 ) . we first recall some basic results on measure - valued solutions of the fokker - planck ( fp ) equation @xmath476 solutions of are understood in the sense of distributions , i.e. for any @xmath328 and @xmath477\times{{{\mathbb{r}}^d}})$ ] , we have @xmath478 for any @xmath283 , we denote by @xmath479 , @xmath462 , the @xmath480-moment of @xmath98 , i.e. @xmath481 . by @xmath482 we denote the space of probability measures on @xmath0 with finite second moment @xmath483 . the _ relative entropy _ of @xmath98 with respect to @xmath22 is defined as @xmath484 given two probability measures @xmath98 and @xmath485 in @xmath317 , the classical _ wasserstein distance _ @xmath486 is defined as @xmath487^{1/2}\,:\,\sigma\in\gamma(\mu,\nu)\}$ ] . here @xmath488 is the set of all _ couplings _ between @xmath98 and @xmath485 : it consists of all probability measures @xmath489 on @xmath490 whose first and second marginals are respectively @xmath98 and @xmath485 , i.e. @xmath491 and @xmath492 for any @xmath493 . notice that the notation @xmath486 is not consistent with the one for weighted distances @xmath47 ; we shall however use it as it is classical . for a proof of the next results see e.g. ( * ? ? ? [ fpunique ] let @xmath494 . 1 . the fp equation has a unique solution @xmath495 in the class of weakly continuous maps @xmath496 with @xmath497 . the unique solution @xmath498 is continuous with respect to the wasserstein distance @xmath486 and lipschitz continuous in all compact intervals @xmath499\subset ( 0,+\infty)$ ] . it is characterized by the family of variational inequalities @xmath500 4 . in addition , it is stable : @xmath501 in @xmath482 implies that @xmath502 in @xmath482 for all @xmath503 . notice that the measure @xmath22 provides a _ stationary _ solution of . all solutions @xmath498 weakly converge to @xmath22 as @xmath504 . finally , @xmath498 is absolutely continuous with respect to @xmath22 for any @xmath457 , with density @xmath19 , and @xmath19 is a solution of the kfp flow . we consider the first variation of the kfp flow , i.e. the _ modified kolmogorov - fokker - planck equation _ @xmath505 for the vector field @xmath506 . in the hilbert space @xmath507 , we consider the continuous ( recall ) bilinear form @xmath508 we look for solutions @xmath509 solving the variational formulation @xmath510 observe that vector fields in @xmath511 belong to @xmath512 . actually the space of smooth compactly supported functions @xmath513 is dense in @xmath512 , and @xmath512 itself is dense in @xmath514 . notice moreover that if @xmath515 is a smooth convex function with bounded second order derivatives and @xmath516 , and @xmath517 ( with @xmath518 ) , an easy calculation shows that solutions of satisfy @xmath519 with these observations in hand , we can apply the variational theory of evolution equations and a simple regularization argument to prove the next result . [ prop : modified_kfp ] for every @xmath520 , there exists a unique solution @xmath521 of in @xmath522 with @xmath523 in @xmath514 . the semigroup @xmath524 is symmetric @xmath525 and satisfies @xmath526 and every convex function @xmath527 with @xmath516 . in particular @xmath524 can be extended by density to a contraction semigroup in @xmath528 , @xmath529 $ ] . the link between and is enlightened by the next result . [ thm : density_gradient_link ] if @xmath19 is a variational solution of the kfp equation with initial datum @xmath530 , then @xmath531 belongs to @xmath532 and it is the solution of the modified kfp equation with initial datum @xmath533 . in particular we have @xmath534 the same result holds if @xmath535 belongs to @xmath536 . since @xmath537 is dense in @xmath453 , we can assume that @xmath538 . then the regularity result of proposition [ prop : fkpre ] shows that @xmath539 for every @xmath97 . setting @xmath531 , we know ( see e.g. the argument in the proof of ( * ? ? ? * lemma 5.2 ) ) that @xmath540 . for a fixed @xmath541 , we can then evaluate @xmath542 with the notations @xmath543 and @xmath544 for @xmath545 , @xmath546 , @xmath547 @xmath326 , let us observe that @xmath548 and @xmath549 inserting this expression in and integrating by parts the first term we get @xmath550 combined with , this shows that @xmath531 satisfies the variational formulation of . the case of @xmath551 follows by a standard approximation procedure , the fact that @xmath552 , and the @xmath553-contraction property of @xmath524 . exactly like the ( k)fp equation , the modified system can be extended to vector - valued measures initial data . to @xmath554 , we associate the vector valued measures @xmath555 which satisfy the system @xmath556 in the weak sense , i.e. @xmath557 the semigroup can be extended to initial data which are vector valued measures with finite total variation using equi - integrability and moment estimates taken from @xcite . let @xmath558 be a nonnegative borel function such that @xmath559 and @xmath560 are finite . if @xmath313 , we have @xmath561 in particular , for every borel set @xmath562 we have @xmath563 which in particular yields ( @xmath564 ) @xmath565 if moreover @xmath566 , we can bound the first moment of @xmath322 by @xmath567 [ thm : vv - kfp ] for every @xmath568 with @xmath569 , there exists a unique solution @xmath570 in the class of weakly continuous maps @xmath571 with @xmath572}{{\sf m}}_1(\|{{\mbox{\boldmath$\nu$}}}_t\|)<+\infty$ ] , for every final time @xmath328 . when @xmath573 then @xmath574 . the map @xmath575 is stable in the following sense : if @xmath576 then @xmath577 in @xmath310 . we divide the proof in three steps . * step 1 . * let us first associate to @xmath578 the probability measure @xmath579 where the constant @xmath580 is a renormalization factor such that @xmath581 . observe that if @xmath582 is finite , then @xmath583 and @xmath584 we also choose the action density to be @xmath585 corresponding to @xmath111 , and observe that the corresponding functional writes @xmath586 [ prop : moment_estimate ] let us suppose that @xmath221 with @xmath587 and set @xmath588 , @xmath589 as in , @xmath590 , @xmath591 , @xmath592 . then latexmath:[\[\label{eq : cap4:12 } and for any @xmath457 , we have @xmath594 the first inequality of follows by the @xmath553-contraction property of @xmath524 . since the fp flow contracts the wasserstein distance by proposition [ fpunique ] and since @xmath22 is a stationary solution , the triangle inequality for the wasserstein distance and the fact that @xmath595 yield @xmath596 on the other hand , yields @xmath597 here we used the fact that @xmath598 . this will appear later as a consequence of theorem [ thm : action_decay ] , and is independent of the present result . combined with and , this proves the estimate on @xmath599 . applying and , we get . * step 2 : existence . * let us approximate a given @xmath568 with @xmath569 by a sequence @xmath600 as @xmath601 in @xmath310 with @xmath602 and @xmath603 . we set @xmath604 with @xmath605 so that @xmath606 solves . thanks to proposition [ prop : moment_estimate ] , we know that the first order moment of @xmath606 are uniformly bounded . this is sufficient to pass to the limit ( up to extraction of a suitable subsequence ) in and to find a solution @xmath607 which is weakly@xmath179 continuous in @xmath310 and satisfies the initial condition in the sense that @xmath608 in @xmath310 as @xmath609 . * step 3 : uniqueness and stability . * it follows by a standard duality argument , like in the case of equation . if @xmath610 and @xmath611 are two weakly continuous solutions of , their difference @xmath612 solves @xmath613 for every @xmath328 and @xmath614\times { { { \mathbb{r}}^d}};{{{\mathbb{r}}^d}})$ ] . by a mollification technique , it is not difficult to check that also holds for every function @xmath615\times{{{\mathbb{r}}^d}};{{{\mathbb{r}}^d}})$ ] with @xmath616 , @xmath617 and @xmath618 continuous and bounded in @xmath619\times{{{\mathbb{r}}^d}}$ ] . next , we introduce a family of smooth convex potentials @xmath173 with bounded derivatives of arbitrary orders , which satisfies a uniform lipschitz condition @xmath620 for some positive constant @xmath621 which is independent of @xmath177 and such that @xmath622 for a given @xmath623 , we consider the solution @xmath624 of the time reversed ( adjoint ) parabolic equation @xmath625 using a maximum principle that can be found in @xcite and the fact that the first and second order spatial derivatives of @xmath626 solve an analogous equation , standard parabolic regularity theory shows that @xmath626 is sufficiently regular to be used as a test function in and satisfies the uniform bound ( observe that the second and third derivatives of @xmath173 are still uniformly bounded ) @xmath627 this leads to @xmath628 since the first order moment of @xmath629 is uniformly bounded , we can pass to the limit as @xmath630 obtaining @xmath631 . as @xmath632 is arbitrary , we conclude that @xmath633 . the stability is then a simple consequence of uniqueness . we can prove now our main estimate , which is a refined version of theorem [ thm : mainestimate ] , under the assumption that @xmath634 \mbox{\emph{$(1-\beta)\,h\,h''+2\,\beta\,(h')^2\le0 $ holds in the sense of distributions.}}\end { array}\ ] ] notice that is equivalent to with @xmath635 . [ thm : action_decay ] assume that and hold , and let @xmath636 be such that @xmath223 . if is satisfied , then @xmath637 we first prove the result with the additional assumptions that @xmath638 and @xmath639 @xmath22 a.e . in @xmath0 . assume that @xmath55 is of class @xmath640 . it follows that @xmath641 and @xmath590 satisfy the same bounds and , for all @xmath457 , @xmath19 , @xmath642 and @xmath554 , @xmath643 . the function @xmath45 is of class @xmath2 in the strip @xmath644\times \{{{\mbox{\boldmath$w$}}}\in{{{\mathbb{r}}^d}}\,:\,\|{{\mbox{\boldmath$w$}}}\|\le w_{\rm max}\}\ ] ] and its differential @xmath645 can be decomposed as @xmath646 since @xmath647 and @xmath648 are bounded , the differential is also in @xmath649 . as a consequence , the time derivative of @xmath650 exists and @xmath651 in order to apply and we have to verify that all components of @xmath652 are in @xmath453 . we have already seen that they are in @xmath653 . let us compute their @xmath194-derivative : @xmath654 the above functions are in @xmath451 , since @xmath655 , @xmath656 , @xmath657 and @xmath554 are bounded , so we get @xmath658 recalling and the convexity assumption on @xmath40 , we find @xmath659 it follows from gronwall s lemma that for all @xmath660 , @xmath661 the result follows by passing to the limit as @xmath662 and recalling that @xmath45 is continuous and bounded on @xmath663 and @xmath664 , @xmath665 converge to @xmath99 , @xmath102 as @xmath666 in @xmath451 and @xmath514 respectively . the general result for an arbitrary concave function @xmath55 easily follows by approximating @xmath55 by a decreasing family of smooth concave functions in the interval @xmath667 $ ] . finally , the general case @xmath220 , @xmath221 , without upper and lower bounds , follows by approximation , using lemmas [ le : smooth_approximations ] and [ le : action_lsc ] . we can extend the results of theorem [ thm : action_decay ] to measure valued initial data . [ cor : action_decay2 ] assume that hold . let @xmath668 and @xmath669 with @xmath587 . then for every @xmath457 we have @xmath670 , @xmath671 with @xmath474 , @xmath672 if @xmath673 , and @xmath674 this follows directly from the measure formulation of the kfp flow ( proposition [ fpunique ] and theorem [ thm : vv - kfp ] ) . let us now consider the entropy functional @xmath675 , for a function @xmath3 as in section [ subsec : entropy ] . [ thm : ent - prod ] let @xmath676 with @xmath677 and let @xmath678 . then @xmath679 and @xmath680 for every @xmath457 , and we have @xmath681 as a consequence , we have @xmath682 for any @xmath97 and the following _ entropy entropy production inequality _ , or _ generalized poincar inequality _ , holds @xmath683 it is not restrictive to assume that @xmath684 . we first prove theorem [ thm : ent - prod ] for a function @xmath55 which grows at least linearly at @xmath685 , and therefore satisfies @xmath686 for some constant @xmath437 . the general result follows by writing @xmath55 as the limit of a decreasing sequence of such concave functions @xmath687 , observing that the corresponding actions @xmath688 and entropies @xmath689 converge increasingly to @xmath45 and @xmath3 respectively . by we know that @xmath690 is finite and therefore we have @xmath691 where the second estimate follows from . applying the chain rule for convex functionals in wasserstein spaces ( see for instance @xcite , we obtain that the map @xmath160 is absolutely continuous and @xmath692 by combining theorems [ thm : density_gradient_link ] and [ thm : action_decay ] applied with @xmath531 and differentiating with respect to @xmath7 , we get that @xmath693 . all other estimate are easy consequences that have already been established in section [ sec : intro ] . consider the space @xmath694 of probability measures at finite @xmath348 distance from @xmath22 . from , we know that @xmath22 has finite quadratic moments and , as a consequence of ( * ? ? ? * theorem 5.9 ) , any measure in @xmath694 also has finite quadratic moments . the same result holds for moments of higher order . let us first notice that since @xmath694 is stable under the action of the semigroup @xmath699 , it is sufficient to prove only at @xmath146 , under the assumption that @xmath98 writes as @xmath700 , for some @xmath701 . we make the additional assumption on the function @xmath55 that there exists some @xmath702 for which @xmath703 this assumption will be removed later in the proof . let @xmath704 fixed and @xmath705 , @xmath706 $ ] , be an admissible curve connecting @xmath141 to @xmath98 such that @xmath707 where @xmath708 . for any @xmath709 , we take @xmath710 . since @xmath55 is non decreasing , we still have @xmath711 notice that , thanks to ( * ? ? ? * theorem 5.17 ) ( also see theorem [ existence : geodesics ] ) , it is possible to assume that @xmath712 is constant with respect to @xmath706 $ ] . for @xmath457 , we set @xmath713 { { \mbox{\boldmath$w$}}}^{s , t}_\kappa = { { \mbox{\boldmath$r$}}}_{st}{{\mbox{\boldmath$w$}}}^s - t\,{{\mbox{\boldmath$d$}}}\rho^{s , t}_\kappa\;. \end{array } \right.\ ] ] it is clear that @xmath714 connects @xmath715 to @xmath716 . note that , thanks to the maximum principle , we have @xmath717 . we claim that it is admissible . indeed , @xmath718 & = & -s_{st}({\nabla\kern-3pt_\gamma\cdot}{{\mbox{\boldmath$w$}}}^s)+ { \nabla\kern-3pt_\gamma\cdot}(t\,{{\mbox{\boldmath$d$}}}\rho^{s , t}_\kappa)\;,\end{aligned}\ ] ] since @xmath719 is admissible . hence , @xmath720 it follows from the definition of @xmath721 that @xmath722 hence , with and , we obtain @xmath723\leq \frac 12\left[\mathscr e_\phi(\rho^{s , t}_\kappa,{{\mbox{\boldmath$w$}}}^{s , t}_\kappa)-\mathscr e_\phi(\rho^s_\kappa,{{\mbox{\boldmath$w$}}}^s)\right]+\frac{\varepsilon}{2}\;.\ ] ] by definition of @xmath724 , we have @xmath725 where @xmath726 and hence , @xmath727 recall that @xmath729 , with @xmath730 . then , acting as in the proof of proposition [ thm : production ] , we get that @xmath731 the assumption on @xmath55 then leads to @xmath732 the next step consists in proving that @xmath733 note that @xmath734 and the concavity of @xmath55 implies that @xmath735 hence @xmath736 since the kfp flow decreases the action . the bound immediately follows from the previous one and . as a consequence , we can apply the chain rule in wasserstein space , which implies that the function @xmath737 is absolutely continuous on @xmath738 $ ] and , for all @xmath739 $ ] , @xmath740 integrating on @xmath738 $ ] and letting @xmath741 go to @xmath93 finally leads to . let us go back to the proof of theorem [ thm : metric_characterization ] . we put and together and obtain @xmath742 + \,t\left[\psi(\sigma+\kappa\gamma\,\|\,\gamma)-\psi(\mathcal s_t\mu+\kappa\gamma\,\|\,\gamma)\right]+\frac{\varepsilon}{2}\;.\end{gathered}\ ] ] we then use the main estimate in theorem [ thm : action_decay ] with @xmath743 and to write @xmath744 & \leq & -\frac12\,i_\lambda(t)\,\mathscr e_\phi(\rho^s_\kappa,{{\mbox{\boldmath$w$}}}^s)\\ & \leq & -\frac12\,i_\lambda(t)\,w^2_{h,\gamma}(\mathcal s_t\mu+\kappa\gamma,\sigma+\kappa\gamma)\;,\end{aligned}\ ] ] where @xmath745 . it follows that @xmath746 + \frac12\,i_\lambda(t)\,w^2_{h,\gamma}(\mathcal s_t\mu+\kappa\gamma,\sigma+\kappa\gamma)\\ \label{eq : cap6:13 } \leq t\left[\psi(\sigma+\kappa\gamma\,\|\,\gamma)-\psi(\mathcal s_t \mu+\kappa\gamma\,\|\,\gamma)\right]+\frac{\varepsilon}{2}\;.\end{gathered}\ ] ] if we first let @xmath125 and then @xmath747 go to @xmath93 in the above estimate , we get that @xmath748+\frac12\,i_\lambda(t)\,w^2_{h,\gamma}(\mathcal s_t\mu,\sigma ) \leq t\left[\psi(\sigma\,\|\,\gamma)-\psi(\mathcal s_t\mu\,\|\,\gamma)\right]\ ] ] as soon as @xmath55 satisfies the assumption . now , any concave and non decreasing function @xmath55 can be decreasingly approched by a sequence @xmath749 satisfying , and the corresponding entropies converge increasingly . then , with theorem [ thm:2 ] , inequality turns out to be valid for any general @xmath55 . to complete the proof of theorem [ thm : metric_characterization ] , it just remains to divide by @xmath7 and let @xmath7 go to @xmath93 . * acknowledgements . * this work has been partially supported by the projects cbdif , evol and otarie of the french national research agency ( anr ) and by a prin08-grant from miur for the project _ optimal transport theory , geometric and functional inequalities , and applications . _	we use the distances introduced in a previous joint paper to exhibit the gradient flow structure of some drift - diffusion equations for a wide class of entropy functionals . functional inequalities obtained by the comparison of the entropy with the entropy production functional reflect the contraction properties of the flow . our approach provides a unified framework for the study of the kolmogorov - fokker - planck ( kfp ) equation . optimal transport , kantorovich - rubinstein - wasserstein distance , generalized poincar inequality , continuity equation , action functional , gradient flows , kolmogorov - fokker - planck equation 26a51 , 26d10 , 53d25
every device , may it be classical or quantum , operates surrounded by an environment which affects to a certain degree its properties . understanding the interplay between the different physical processes at work and the dissipative processes induced by the environment is thus of paramount importance . this is of particular interest for the development of technologies based on quantum mechanics as the environment can , for example , destroy the quantum coherence between different parts of the device and thus completely spoil its functionalities . various quantum systems affected by dissipative processes have been studied in solid - state physics @xcite , atomic and molecular physics @xcite and quantum optics @xcite . however , at the moment , a complete understanding of the physics of strongly correlated quantum many - body systems coupled to an environment is lacking . in recent years , due to rapid technological progress , cold atom experimentalists have been able to build quantum emulators , systems mimicking with a high degree of precision modeled hamiltonians . for this reason , these cold atoms have become an ideal test - bed to study quantum dynamical systems . in fact , the theoretical study of ultracold atoms paved the way to the first understanding of the interplay between interactions and dissipation . some notable works are ( i ) the discovery of zeno - like dynamics due to local losses of atoms in a double well geometry @xcite or in optical lattices @xcite , ( ii ) the existence of augmented stability against dissipation for weakly @xcite and strongly @xcite interacting systems and ( iii ) the emergence of a dynamical phase transition between the condensed and thermal steady states @xcite . also of great importance are works detailing methods to engineer dissipative processes generating dark states with precise quantum properties @xcite . in this work we will study a quantum well in a systematic way by investigating different boson densities . the system under study is a collection of ultracold atoms in a double well potential under the influence of a dissipative process . as a warm - up , we first consider a double well loaded with only two atoms . this exercise will help us gain a deeper analytical understanding of the combined effect of interaction and dissipation , and show that this interplay leads to a marked slowing down of the decoherence process . we then study a system with many atoms and demonstrate that the increased complexity of the system results in the emergence of a regime where coherence decays algebraically . hence , for large atom numbers and interaction strength the system is particularly resistant to dissipation . this paper is organized as follows . in section [ sec : model ] we introduce the model and the various relevant parameters . in section [ sec : two ] we first study in detail the case of two atoms in a double well before investigating in [ sec : many ] the dynamics for the same double well loaded this time with a large number of atoms . in section [ sec : conc ] we draw our conclusions . the properties of an ensemble of bosonic atoms in a double well can be described , in the single band approximation , by the hamiltonian @xmath0 here @xmath1 ( @xmath2 ) is a bosonic operator annihilating ( creating ) an atom at site @xmath3 and @xmath4 counts the number of atoms at site @xmath3 . dealing with a double well , @xmath3 can only take the values @xmath5 . the parameter @xmath6 is the hopping amplitude , and @xmath7 the interaction strength . as the ratio @xmath8 increases , the phase coherence between the two sites , @xmath9 , decreases . to represent a state vector , we use the fock basis @xmath10 where @xmath11 and @xmath12 label the number of atoms on sites @xmath13 and @xmath14 , respectively . as this ultracold gas of bosons is under the influence of a dissipative process , the system can be described by a density matrix @xmath15 which evolves following the master equation @xmath16 + \gamma\mathcal{d}({\hat{\rho } } ) \label{eq : master}\ ] ] where @xmath17 is the dissipator modeled by @xmath18\right ] \label{eq : dissipator}\ ] ] with coupling strength @xmath19 . it has been shown in @xcite that this dissipator , written in lindblad form @xcite , can be experimentally realized , for example , in the context of ultracold atoms exchanging energy with a red - detuned optical lattice . let us note that the considered dissipator leaves the fock states invariant . here we study the situation where the system is first prepared in the ground state of the hamiltonian . at time @xmath20 the dissipation is switched on and the evolution of the system under the influence of both the hamiltonian and the dissipator is investigated . having established the model and main notations in the previous section we now attempt to gain a deeper understanding of the physics of this system . the best starting point is to study a double well loaded with two atoms as this limit can be solved analytically . we show here that the system converges towards a unique steady state and that the relaxation process slows down as the interaction strength increases . as a reminder , we first provide here a brief summary of the physics of the hamiltonian system without dissipation . the dimension of the hilbert space is three and we choose as a basis the states @xmath21 , @xmath22 , @xmath23 in this order . the hamiltonian in this basis is thus given by @xmath24 the ground state of this hamiltonian , which we choose as the initial state for the time evolution , has an energy @xmath25 and is given by @xmath26 , where @xmath27 is the normalization constant @xmath28 . note that for large @xmath8 the ground state wavefunction tends towards @xmath29 with energy @xmath30 . with these results in mind , we can now address the following question : how does the system evolve under the effect of the dissipation considered ? using ( [ eq : master ] ) , we compute the time evolution of the nine elements of the density matrix @xmath15 . the evolution of the six complex independent elements @xcite is given by @xmath31 given that we prepare the system in the ground state of the hamiltonian , the initial state is symmetric under the exchange of the first and third element . for this reason we only focus on the symmetric subspace of the density matrix evolution . ( [ eq : eqrho ] ) is a linear equation and its symmetric eigenvalues @xmath32 are the roots of the polynomial @xmath33 \lambda + [ 9 ( \hbar\gamma)^2 + 16 j^2 + u^2 ] \lambda^2 + 6 \hbar\gamma \lambda^3 + \lambda^4 $ ] , and the steady state , with eigenvalue @xmath34 , corresponds to the density matrix @xmath35 which is the completely mixed state ( or the highest entropy state ) . the roots of @xmath36 are in general complex numbers with @xmath37 . studying the roots of @xmath36 , we identify dynamical transitions which arise when tuning @xmath8 and @xmath38 transforms a real into a complex eigenvalue ( or the converse ) . dynamically this implies that a simple exponential decay , the hallmark of real eigenvalues , can be transformed into an exponential decay accompanied by oscillations ( due to the imaginary part ) . the oscillations have their origin in the hamiltonian evolution . the presence of dynamical transitions becomes particularly interesting if one changes the value of the different parameters , for example @xmath7 or @xmath19 , in time , allowing for the modulation of the dynamics from an overdamped to an underdamped regime . ( divided by j ) versus the interaction @xmath8 for @xmath39 ( a , b ) , @xmath40 ( c , d ) and @xmath41 ( e , f ) . ] the various dynamical scenarios are depicted in fig . [ fig : eigs ] for different values of @xmath38 . for @xmath42 , as show in fig . [ fig : eigs](a , b ) , two eigenvalues are real and two are complex conjugates . for @xmath43 , fig . [ fig : eigs](c , d ) , there are three regimes as a function of @xmath8 : in the first regime , there are two complex and two real @xmath32 , in the second regime all the @xmath32 are real , and in the last regime again two eigenvalues are real and two are complex . for larger values of the dissipative coupling , @xmath44 , there are two regimes : for small @xmath8 , the decays are only exponential , while for large @xmath8 two eigenvalues are real and two are complex ( see fig . [ fig : eigs](e , f ) ) . for any value of @xmath38 , if @xmath8 is large enough the imaginary part of two of the eigenvalues will be proportional to @xmath45 indicating the strong influence of the hamiltonian part of the dynamics . the change in dynamics is exemplified in fig . [ fig : dynbeha ] where @xmath40 . in this figure , we plot the absolute value of the coherence versus time in a log - log form which highlight the change of sign of the coherence for @xmath46 ( blue line ) . we observe that , as expected , for both @xmath47 ( blue solid line ) and @xmath48 ( red dot - dashed line ) the evolution of the coherence is oscillatory while for @xmath49 ( green dashed line ) there are no oscillations but only a sum of exponential decays . the inset highlights the oscillatory behavior for @xmath48 . and for interaction @xmath46 ( blue solid line ) , @xmath49 ( green dashed line ) , @xmath48 ( red dot - dashed line ) . the inset shows the coherence versus time in a linear plot to highlight the oscillating behavior for @xmath48 . ] the long time dynamics of the system are governed by its slowest decaying state , those whose corresponding eigenvalues have the smallest ( in modulus ) real part . for this small system with only five symmetric states , the relevant ones are the steady state @xmath50 and the next slowest decaying state @xmath51 with corresponding @xmath52 . while the steady state is independent of @xmath8 , @xmath51 and its decay rate can change considerably with @xmath8 . to gain a deeper understanding of how the interaction affects the dynamics , we first present the non - interacting case @xmath53 and then cover the large interaction limit where @xmath54 and @xmath55 . for the non - interacting case , @xmath53 , the equation for the eigenvalues @xmath36 takes the simpler form @xmath56 \lambda + 5 \hbar\gamma \lambda^2 + \lambda^3\right\}$ ] . the slowest decaying eigenvalue at large @xmath38 is , in this regime , @xmath57 this is a clear manifestation of the quantum zeno effect : increasing the dissipative coupling @xmath19 suppresses the decay rate as the continuous density measurement prevents the system from evolving . an experimental observation of this effect was reported in @xcite . when the interaction is the dominant energy scale ( for large @xmath8 and large @xmath58 ) , we analytically find that @xmath59 similarly to the quantum zeno regime , here too the decay is strongly suppressed . however , as the physical mechanism behind this slowing down is different from the quantum zeno case , the dependence of @xmath60 on the system parameters is distinct from @xmath61 . remarkably , in this regime , the presence of interactions renders the system more robust against dissipation , as it increases the size of the energy gap that needs to be overcome to populate higher states . as reaching the steady @xmath62 requires a large amount of energy , of the order of @xmath7 , the dissipative process , heating up the system , will need a longer time to provide this energy at large @xmath7 . consequently , we refer to this phenomenon as `` interaction impeded decoherence '' . the cross - over between the two regimes is clearly illustrated in fig . [ fig : eigcrossinter](a ) where , for small interactions , @xmath32 decreases as @xmath19 increases , while , for large interactions , @xmath32 and @xmath19 increase together as indicated by the green arrows . versus interaction @xmath8 ( blue diamonds ) for different @xmath63 . the blue diamonds indicate the exact numerical values while the red - dashed line represents the approximate solution from ( [ eq : lqzi ] ) . ( b ) @xmath64 times @xmath38 versus interaction @xmath8 . ( c ) @xmath64 divided by @xmath38 versus interaction @xmath8 . the arrows indicate increasing @xmath19 values . ] for higher clarity we have also shown in fig . [ fig : eigcrossinter](b , c ) the value of , respectively , @xmath65 and @xmath66 . the collapse of the curve in fig . [ fig : eigcrossinter](b ) confirms the emergence of a quantum zeno behavior while the collapse of the curve in fig . [ fig : eigcrossinter](c ) signals the emergence of the `` interaction impeded coherence '' behavior . we gain deeper insight into the large interaction @xmath67 and/or large dissipation regime @xmath68 via an adiabatic elimination technique @xcite . the main idea behind this method is that at long times the dynamics is dominated by the diagonal elements of @xmath15 which dictate the full quantum dynamics . in fact , their evolution relates to the hopping parameter @xmath6 , the smallest parameter in the system , and connects the diagonal and the off - diagonal elements . however , the off - diagonal elements evolve very quickly , with a dephasing due to the interaction @xmath7 and/or a fast exponential decay due to @xmath19 . hence , the evolution of the off - diagonal terms can be integrated out and we are left with a coarse - grained evolution for the whole system . more precisely let us take , for example , the evolution of @xmath69 from ( [ eq : eqrho ] ) . this is given by @xmath70 here it should be noted that @xmath71 , being far from the diagonal , is much smaller and decays even faster . also as @xmath7 and @xmath19 are large , we can integrate ( [ eq : ae1 ] ) by parts to obtain @xcite @xmath72 likewise we have @xmath73 we now inject the information from ( [ eq : ae2 ] ) and ( [ eq : ae3 ] ) into the time evolution of the diagonal elements and , as anticipated earlier , the diagonal values of @xmath15 are determined by three independent equations . this results in @xmath74 eq . ( [ eq : ae4a ] ) shows consistently that @xmath50 is the steady state , ( [ eq : ae4b ] ) refers to the slow decaying anti - symmetric solution , which decays at a rate @xmath75 but is not relevant for our dynamics , and the last equation , ( [ eq : ae4c ] ) , represents the symmetric slowest decaying state with a rate @xmath76 from ( [ eq : lqzi ] ) , we derive both the quantum zeno behavior of ( [ eq : lqz ] ) and the `` interaction impeded decoherence '' behavior ( [ eq : li ] ) in their respective regimes ( small and large interactions limits ) . using ( [ eq : ae2 ] ) and ( [ eq : ae3 ] ) , we compute , at any given time , the coherence of the system knowing only the terms on the diagonal . in particular , the coherence of the slowest decaying state @xmath52 is given by @xmath77 which is proportional to @xmath7 when it is small compared to @xmath78 , and inversely proportional to @xmath7 when it is the dominant energy scale . thus the large interaction strengths affect the coherence in two ways : @xmath7 lowers the coherence of the slowest decaying state , but at the same time reduces the decay rate @xmath51 . we now want to understand how the dynamics is altered by the presence of many atoms . as for the two atom case the interaction was found to play a crucial role in the dynamics , we can expect that the presence of many interacting atoms will affect the dynamics even more dramatically . we first present a perturbative study at short times highlighting the subtle role played by interactions , we then cover in detail the strongly interacting regime where coherence decays algebraically with time . to illustrate how interaction influences decoherence , we study perturbatively the time evolution of the coherence . we first show analytically that for a non - interacting system the coherence decays following a simple exponential . to prove this statement it is sufficient to show that the evolution of the operator @xmath79 follows @xmath80 + \gamma\mathcal{d}(\hat{c } ) \nonumber \\ & = & \gamma\mathcal{d}(\hat{c } ) \nonumber \\ & = & -\gamma\hat{c}. \label{eq : masterfor c}\end{aligned}\ ] ] this expression is derived from the fact that @xmath81 and that , for @xmath53 , @xmath82=0 $ ] . in the non - interacting limit , the evolution of the coherence is then simply described by @xmath83 for non - zero interaction strength , and in particular for large values of @xmath7 , the decoherence follows a different law even for short times . to demonstrate this statement , we analyze the evolution of the density matrix @xmath15 up to third order in @xmath84 @xmath85 where each superimposed dot stands for a time derivative . as we use the ground state of the hamiltonian as the initial state , we have that @xmath86=0 $ ] and hence can write @xmath87\nonumber \\ \dddot{\hat{\rho}}_g&=&\mathcal{d}\left(\mathcal{d}\left(\mathcal{d}({\hat{\rho}}_g)\right ) \right ) -\frac i { \hbar } \mathcal{d}\left(\left[\hat{h},\mathcal{d}({\hat{\rho}}_g)\right ] \right ) \nonumber \\ & - & \frac i { \hbar } \left[\hat{h},\mathcal{d}\left(\mathcal{d}({\hat{\rho}}_g)\right)\right ] - \frac 1 { \hbar^2 } \left[\hat{h},\left[\hat{h},\mathcal{d}({\hat{\rho}}_g)\right]\right ] . \nonumber \end{aligned}\ ] ] the coherence is only dependent on the real part of the off - diagonal terms of the time - evolved density matrix , hence , for @xmath54 , we derive @xmath88 if @xmath53 , this expression would be the taylor expansion of an exponential decay . however , due to the finite interaction strength , the short time evolution is slowed down as @xmath7 opposes the effect of @xmath19 . this constitutes a first glimpse into the role played by interactions . in the following , we derive an expression for the decay of the coherence @xmath89 at large times and for a large number of atoms . this expression provides a much deeper understanding of the emerging physics . a very good first insight is provided by the study of the eigenvalues @xmath32 of the evolution of the density matrix @xmath15 . for the general case with @xmath90 atoms , the size of the density matrix @xmath15 is @xmath91 . the elements of the density matrix @xmath92 are such that @xmath93 where @xmath90 is the total number of atoms . rewriting the matrix @xmath15 in a vector form @xmath94 , we compute at each time @xmath95 the vector @xmath94 . this vector is given by the sum over all @xmath96 , each associated with a complex decay constant @xmath97 , @xmath98 here @xmath99 is the weight of each @xmath96 at time @xmath20 . the real parts of the different @xmath97 are plotted in fig . [ fig : eigenvalues ] . as a function of the number of atoms @xmath90 for @xmath100 and @xmath48 . the red - dashed line represent an algebraic fit that scales as @xmath101 . ] two aspects of this figure are worth noting : ( i ) the slowest decaying state is gapped from all others ( this gap depends on the number of particles as @xmath101 ) and ( ii ) for large @xmath90 an almost continuous band of slowly decaying states arises . the first point implies that past a certain critical time the decay will be fully exponential and solely due to the slowest decaying state @xmath51 ( note that the corresponding @xmath52 decreases as @xmath101 as shown by the linear fit in fig . [ fig : eigenvalues ] ) . this state , as expected , is mainly populated on the diagonal , see fig . [ fig : dmevensector ] , and these diagonal elements follow a particular distribution . for the slowest decaying symmetric state . this particular case is for @xmath102 , @xmath103 and @xmath100 . ] we observe that the matrix elements corresponding to large density imbalance between the left and right sites , for example @xmath104 or @xmath105 , are much more populated than those with balanced configurations , for example @xmath106 . for a large interaction strength , states with strong imbalance are highly energetic and thus populating them requires the dissipation to provide a tremendous amount of energy . moreover , in order to populate the state @xmath107 , with @xmath108 large , it is necessary to first populate state @xmath109 which in turns necessitates a long time . thus , the different configurations can be arranged into an energetical hierarchy . the second aspect , the presence of a band of eigenvalues , paves the way to the emergence of a new decay regime . this interesting behavior arises as similar exponential decays are summed over . the emergence of this regime is shown in fig . [ fig : decay_coherence ] where coherence decay for various values of @xmath8 and @xmath90 is plotted . on the left panels ( @xmath110,@xmath111,@xmath112 ) , we show how the slower decaying states decay in time , while on the right panels ( @xmath113,@xmath114,@xmath115 ) are plotted the weighted coherence of each eigenstate given by @xmath116 for small @xmath8 or @xmath90 the exponential decay of the slowest decaying state @xmath51 takes place earlier as the interaction is not strong enough to impede decoherence ( small interaction @xmath117 in fig . [ fig : decay_coherence](a , b ) ) or there are not enough states to form a band ( small atom number @xmath118 in fig . [ fig : decay_coherence](c , d ) ) . when both the interaction and atom number are sufficiently large , for example @xmath48 and @xmath119 as in fig . [ fig : decay_coherence](e , f ) , a power - law decay regime emerges over a large time . this regime is clearly shown on the log - log plot in the inset of fig . [ fig : decay_coherence](e ) . versus time for different values of @xmath90 and @xmath8 . right column : decay of the weighted coherence of each eigenstate @xmath120 versus time . the inset of fig . [ fig : decay_coherence](e ) presents a log - log of the coherence and highlight the power - law regime . the parameters are @xmath119 and @xmath117 for ( a ) and ( b ) , @xmath118 and @xmath48 for ( c ) and ( d ) , and @xmath119 and @xmath48 for ( e ) and ( f ) , while @xmath100 for all the subplots . ] to provide a better understanding of this emerging dynamical behavior and to pinpoint the algebraic nature of this regime , we first proceed with a coarse - graining of the evolution using adiabatic elimination and then the resulting equations are further approximated within a continuum limit . the end product of this procedure is a continuum classical diffusion equation with a non - uniform diffusion function . the solution of this diffusion equation is then used to demonstrate the existence of the algebraic regime @xcite . as for the two - atom problem described in section iii , we study the long time full quantum dynamics by only considering the diagonal elements of @xmath15 . the adiabatic elimination , in the regime of large dissipation @xmath38 and interaction @xmath8 , thus reduces the size of the system to @xmath121 . the system of equations describing the dynamics is derived by generalizing ( [ eq : ae2 ] ) to the limit @xmath122 to obtain @xmath123 this generalization is also applied to the master equation ( [ eq : master ] ) and gives @xmath124 [ eq : mastersimp ] for @xmath125 , @xmath90 even and @xmath126 . these last equations are used to accurately compute the dynamics and give access to the coherence of the system @xmath127 by performing the large @xmath90 limit , the discrete master equation ( [ eq : mastersimp ] ) is mapped onto a classical diffusion equation . the configuration space , indexed by @xmath108 is mapped to the coordinate @xmath128 $ ] which is a continuous variable in the limit @xmath129 . the boundaries of this system then become @xmath130 corresponding to the strongly imbalanced configurations of the double well , @xmath131 and @xmath132 , whereas the center , @xmath133 , corresponds to the balanced configuration @xmath134 . the diagonal elements of the density matrix @xmath15 are connected to a continuum probability density distributions @xmath135 via the relation @xmath136 . within this limit , the normalization of the probability density distribution is given by @xmath137 which is equivalent to @xmath138 . the initial state corresponds to @xmath139 , and the steady state is then given by the uniform distribution @xmath140 which represents the totally mixed state @xmath50 . to derive the differential equation for the probability density function @xmath135 , we rewrite ( [ eq : approxmodel2 ] ) as @xmath141 where @xmath142 . expanding the right - hand side of ( [ eq : continuum1 ] ) , to second order in @xmath143 , gives the diffusion equation @xmath144 \label{eq : continuum}\end{aligned}\ ] ] where @xmath145 . the diffusion function @xmath146 varies substantially as a function of @xmath147 : for the balanced configuration @xmath133 , @xmath148 diverges while for strongly imbalanced configurations @xmath149 it vanishes . thus , if the initial state is near the origin , which is the case when the evolution begins from the ground state , we observe a very fast initial diffusion ; when the probability density distribution approaches the boundaries the diffusion drastically slows down . for short times , such that @xmath150 is not affected by the boundaries , the diffusion equation ( [ eq : continuum ] ) is solved by using a scaling ansatz @xmath151 with @xmath152 . a scaling solution exists for @xmath153 and @xmath154 giving the differential equation for @xmath115 @xmath155 this expression is solved analytically giving @xmath156 the diffusion for the probability density function @xmath150 is thus , for short rescaled time @xmath154 , given by @xmath157 where the constant @xmath158 is the gamma function with argument @xmath159 . eq . ( [ eq : proba ] ) describes an anomalous diffusion process as @xmath160 which , since this formulation is valid for @xmath161 small , and as for normal diffusion @xmath162 , ( [ eq : proba ] ) is an example of super - diffusion . hence , the divergence in @xmath148 at @xmath163 leads to a highly accelerated initial diffusion . regarding the coherence @xmath89 , it is possible to obtain an analytical expression from the continuum limit of ( [ eq : cohediscr ] ) : @xmath164 this expression is in excellent agreement with the numerical simulations ( see fig . [ fig : coherencematch ] ) . hence , we demonstrated here how the power - law regime emerges for a large number of atoms and strong interactions . [ fig : coherencematch ] also shows clearly that if the interaction is not strong enough ( black continuous curve for @xmath46 ) the power - law regime does not emerge . it also shows that the algebraic region will increase when more atoms are in the system . in fact the power - law region is larger for large @xmath90 , for example @xmath165 ( blue squares ) , and it is smaller for fewer atoms , for example @xmath166 ( light blue stars ) . , @xmath48 and @xmath100 ( blue squares ) , for @xmath167 , @xmath103 and @xmath168 ( green diamonds ) , for @xmath166 , @xmath48 and @xmath169 ( light blue stars ) and for @xmath170 , @xmath46 and @xmath100 ( black continuous line ) . the red dashed line represents the analytical result of ( [ eq : coherence ] ) . ] from fig . [ fig : coherencematch ] , we also notice that for smaller @xmath19 the algebraic region begins at later times and is thus reduced ( see green diamonds curve for @xmath167 , @xmath103 and @xmath171 ) . the fact that the time - evolution of the coherence is independent of the interaction strength is particular to the coherence in this problem . to summarize we have shown here how the interaction between atoms can affect decoherence in a quantum system . we analyzed a double well setup by first considering only a system of two atoms and by subsequently extending our study to a system of many atoms . for two atoms , we discovered that various dynamical behaviors emerge and depend on the ratios between @xmath8 and @xmath58 . we also showed that strong interactions suppress strongly the detrimental effects due to dissipation . we refer to this process as `` interaction impeded decoherence '' . when the double well is loaded with many atoms , we found that the decoherence process is slowed down by interactions and that the nature of the decay is altered and follows an algebraic law . the experimental realization of this model is possible using cold atoms trapped in optical lattices . one of the main challenges would be to use a system that can be well described by a single band model despite the large interaction strength . this could be achieved using a very light species like lithium whose interaction can be tuned using a feshbach resonance @xcite . an alternate route would be to use two different kinds of bosonic atoms trapped in a single well potential and to control the effective tunneling and interaction respectively via two - photon combined microwave and radio - frequency pulses , and a feshbach resonance as in @xcite . we thank p. barmettler , h.p . breuer , j. dalibard , j .- eckmann , m. greiner , m. lukin and v. vuletic for fruitful discussions . we acknowledge anr ( famous ) , snsf ( division ii , manep ) , cifar , nserc of canada and the darpa - ole program for financial support .	we study the dynamics of bosonic atoms in a double well potential under the influence of dissipation . the main effect of dissipation is to destroy quantum coherence and to drive the system towards a unique steady state . we study how the atom - atom interaction affects the decoherence process . we use a systematic approach considering different atomic densities . we show that , for two atoms , the interaction already strongly suppresses decoherence : a phenomenon we refer to as `` interaction impeded decoherence '' . for many atoms , thanks to the increased complexity of the system , the nature of the decoherence process is dramatically altered giving rise to an algebraic instead of exponential decay .
the panda spectrometer @xcite is one of the large apparatuses at the future facility for antiproton and ion research ( fair ) in europe . gluonic excitations and the physics of strange and charm quarks will be accessible with unprecedented accuracy , allowing high - precision tests of strong interaction . the fair facility ( shown in figure [ fair ] ) is under construction at the gsi laboratory in germany . science goals of the international fair project span a broad range of research fields , from atomic and plasma physics to the structure of the matter , from the quark - gluon structure of hadrons to physics of astronomical objects . the existing linear and sis18 accelerators at gsi will be used as injector for the new system . the hearth of fair is a double ring accelerator sis100/300 of 1 km diameter , which will provide primary beams of @xmath0u ions in a high charge state , at an energy up to 35 agev , and of protons up to 30 gev . the beams will be sent to several branches to be used by different experiments . secondary beams of radioactive ions of 2 agev energy and @xmath1 up to a momentum of 15 gev / c will be available . the @xmath1 will be pre - cooled in a system of storage and cooler rings ( resr / cr ) and then injected at 3.7 gev / c in the high energy storage ring ( hesr ) , where the panda detector will be installed . commissioning of @xmath1 beam for panda is forseen in 2015 . the hesr is a storage ring for @xmath1 in the momentum range from 1.5 to 15 gev / c . an important feature of hesr is the combination of phase - space cooled beams and internal target , which will allow experiments of unprecedented precision . two different operation modes will be available at hesr : high resolution mode using the electron cooling , allowing to achieve a very high momentum resolution @xmath2 at a moderate luminosity of @xmath3 , and high luminosity mode using the stochastic cooling , with greater momentum spread ( @xmath4 ) , but at @xmath5 @xmath6 . at hesr the combination of a cooled beam and an internal target will allow to study charmonium spectroscopy with unprecedented precision . indeed , the precision of the resonance mass and width measurements depends on the precision of the beam energy and beam energy width measurements , respectively . the determination of both these quantities is based on the measurement of the beam revolution frequency spectrum . in the panda experiment the detector is used to identify and reconstruct the final state and not to measure the parameters of the resonance . thus the experimental detector resolution determines the sensitivity to a given final state . since the @xmath1 can be cooled very effectively an excellent resonance mass resolution of 30 kev can be achieved , which has to be compared with previous best values obtained in @xmath7 annihilations ( 100 kev ) or in @xmath8 annihilations ( 10 mev ) . studies of hadron structure can be performed with different probes , each one with its specific advantages . the reason to use @xmath7 annihilations instead of @xmath8 ones is that all states can be directly formed and the mass resolution is very good . on the contrary , in @xmath8 interactions only states with the quantum numbers of the photon , j@xmath9=@xmath10 , can be formed , all other states being accessible by secondary decays with limited statistics , thus a moderate mass resolution can be achieved . moreover , @xmath7 gives access to boer - mulders function ( with no polarised beam ) and to transversity distribution . the panda physics programme @xcite concerns not only charmonium spectroscopy but spans many different topics , all of them addressed to answer to fundamental questions of hadron and nuclear physics . search for gluonic excitations , such as glueballs and hybrids , will be pursued in the charmonium region . properties of mesons with hidden and open charm in the nuclear medium will be studied to understand the origin of hadron masses . moreover , the open charm physics will be studied , measuring the rich spectrum of d mesons and their dominant and rare decays . since in panda the @xmath11 and @xmath12 hyperons will be copiously produced , it will be possible to generate double-@xmath13 hypernuclei , which allow to study the y - n and y - y interactions . last , but not least , the study of nucleon structure will be investigated , by measuring generalized parton distributions ( drell - yan and deeply virtual compton scattering , `` spin '' structure functions by polarised @xmath1 and proton time - like form factors up to 25 gev c.m . energy ( never reached before ) . such an ambitious programme requires a high performance spectrometer . the main requirements can be summarised as follows : * capability to handle an interaction rate up to 20 mhz , especially for detectors closed to the target and in the forward direction ; * full angular coverage for partial wave analysis ; * momentum resolution of 1@xmath14 ; * high vertex resolution @xmath15 m to reconstruct secondary vertex of d meson ; * good pid for charged particles in a large range of energies ; * efficient @xmath16 detection in a large range of energies ( 10 mev@xmath1710 gev ) ; * efficient event selection . the experimental set - up is shown in figure [ panda_det ] . the apparatus is a combination of a target and a forward spectrometer ( ts and fs in the following ) of modular design , optimized for the specific kinematics of the @xmath7 annihilation process with energy range 1.5@xmath1715 gev / c . the basic concept of the ts is a shell - like arrangement of various detector systems surrounding the interaction point of @xmath1 in a fixed target inside a magnetic field of 2 t provided by a large solenoid . to overcome the gap in acceptance and resolution in the forward region of such a system a fs with a dipole magnet of large gap ( 1 m ) will be used . it will cover angles below @xmath18 and @xmath19 in the vertical and horizontal directions , respectively . starting from the interaction point , we find a silicon microvertex detector , a main tracker system , a time - of - flight ( tof ) barrel and a barrel dirc . to cover the forward angles a system of drift chambers ( dc ) will be added inside the ts and in the fs , as well . the same thing for dirc , where an endcap will be used . the muon detector will be embedded in the return yoke of the magnet , either to cover the azimuthal angle or to cover the forward angles . in the fs different systems of dc , tof , rich , @xmath20 detectors and hadron calorimeter will be used . for the electromagnetic shower detection an em calorimeter will be available in the ts and a shashlyk calorimeter will be used in the fs . concerning the hydrogen internal target two different systems are under development , one based on pellet and one on cluster - jet . pellet target will guarantee to reach the maximum luminosity , since an averaged thickness of @xmath21 h/@xmath22 can be achieved . on the contrary cluster - jet target will assure to have a stable luminosity , since the density can be easily changed by means of the pressure . nuclear target , shaped in wires or foils will be also used for hypernuclear physics and for the study of charm in the nuclear medium . main feature of the silicon microvertex detector are the following : good vertex resolution , better than @xmath23 m , to reconstruct displaced secondary vertices from charmed and strange hadron decays ; high granularity ; low material budget and low power consumption . the structure of the detector consists of four barrel layers surrounding the interaction point and six disks in the forward region . the two innermost barrel layers are made out of custom hybrid active pixel sensors @xcite of @xmath24m@xmath25 size with an epitaxial sensor thickness less than @xmath26 m . the four first disks in the forward direction , as well as an inner part of the two outer disks , are also made out of hybrid pixel modules to cope with the high particles flux . for the readout system of the inner part a custom pixel chip ( topix @xcite ) , based on cmos 130 nm technology is under study at torino infn , the layout with analog and digital parts fits 100@xmath14 the pixel cell . in the second prototype currently developed the time - over - threshold has been implemented , in order to retain energy loss information . sufficient buffering to operate without trigger is forseen . the sensor and the readout chip are bump - bonded together including the support structure and the cooling system . the detector will have a radiation length x@xmath27 about 1@xmath14 . the reduced occupancy in the outer layers of the microvertex detector allows to use double silicon microstrips , with a corresponding reduction in the number of readout channels and material . both pitch size and thickness are 100 @xmath20 m ; such a system will use a standard readout chip developed in other experiments . the use of microstrip modules for the outer region allows to reduce total material budget of the microvertex detector to 4 - 6@xmath14 x@xmath27 . the total number of channel will be about 10@xmath28 . for the tracker system two different options are under investigation . one is based on straw tubes @xcite of mylar , 10 mm radius and 1.5 m long . the inner and outer layers in the axial direction and the intermediate layers are skewed of a small angle up to @xmath29 in order to measure the z coordinate . it is a very light detector with a momentum resolution of 1.5@xmath14 . the second option is a tpc with a multi - gem system for avalanche amplification ( which allows to suppress the ion feedback ) . due to the high rate pf @xmath1 annihilations ( @xmath30 ) the tpc has to operate continuously , therefore the gating technique can not be applied . the tpc is an ideal device for tracking due to low material budget , a momentum resolution of 1@xmath14 and a very good pid can be achieved , provided the problems of the space charge build - up and the continous sampling have been solved . particles emitted at polar angles below @xmath31 will be tracked in the dc placed downstream the target . a very good pid over a large range of energies is an essential requirement to fullfill the physics goals of panda . therefore , different processes will be used to assure redundancy : @xmath32erenkov effect above 1 gev / c , using different types of radiators ; energy loss below 1 gev / c ; tof below 1 gev / c ; energy loss at low energy by means of microvertex detector ; electromagnetic showers detection by means of em calorimeters . in the ts the em calorimeter @xcite will cover almost the full solid angle and will have a good efficiency and energy resolution up to 10 gev and down to 10 mev . it consists of one barrel and two endcaps , made of lead tungstate pwo - ii crystals of second generation , with enhanced light ouput . large area apd , which can be operated in presence of magnetic field , will be used for scintillating light read - out , allowing a fast timing at a level of about 1 ns . the pwo - ii crystals will be operated at a temperature down to t = -20@xmath33c in order to increase the light yield by a further factor 3 . an energy resolution of 2@xmath14 at 1 gev can be achieved . a shashlyk calorimeter will be installed in the fs beyond the dipole magnet , about 7 m from the interaction point . here a worse resolution of 4@xmath14 + const . at 1 gev can be achieved at a moderate cost . daq@xmath34trigger have to manage a high rate up to 20 mhz . no hardware trigger will be implemented and a continuous sampling will be done . the detector front ends will be capable to make autonomous data preprocessing . the data reduced in the preprocessing step will be marked by a precise time stamp and buffered for further processing . the data will be sent to powerful compute nodes , based on high density fpga preocessors . a configurable high speed network transports data through the levels . the last stage of network can be more traditional and it will be attached to the online reconstruction farm , where the event is fully reconstructed to perform the final selection before the mass storage . due to the high interaction rate a large bandwidth up to 200 gb / s is needed . d. calvo et al . , `` thinned epitaxial silicon hybrid pixel sensors for the panda experiment '' , nuc . @xmath34 meth . in phys . a594 ( 2008 ) 29 - 32 . d. calvo et al . , `` topix : the first prototype of pixel readout for panda experiment '' , proc . of rd07 , int . conf . on large scale applications and radiation hardness of semiconductor detectors , florence , italy , june 2007 .	the panda detector is under design to be installed at the hesr storage ring for antiproton of the future fair facility in darmstadt , germany . fundamental questions of hadron and nuclear physics interactions of antiprotons with nucleons and nuclei will be pursued using a multipurpose set - up which includes innovative detectors . here , the fair facility and the panda detector are described .
the investigation of semiconducting nanoclusters is one of the most promising directions in the search for new materials to construct optical and electronic devices , new laser materials , and biological markers @xcite . the properties of nanosize clusters are typically very different from their parent bulk compounds ; for example , surface passivated silicon nanoclusters show a number of interesting effects such as ultrabright luminescence @xcite and nonlinear optical effects @xcite , whereas crystalline bulk silicon is optically uninteresting because of its small and indirect gap . previous theoretical studies have employed various methods to model and understand the basic structural and electronic properties of nanosize clusters such as tight - binding @xcite , empirical pseudopotentials @xcite , density functional theory ( dft ) @xcite , gw - bethe salpeter ( gw - bse ) @xcite , and quantum monte carlo ( qmc ) @xcite . despite the significant progress that has been made , especially in interpreting the properties of larger clusters ( @xmath0 2 nm , @xmath0 500 atoms ) , many open questions still exist as to how new physical effects begin to dominate the electronic structure of these clusters as the surface to volume ratio increases . these effects include the possibility of surface states , reconstructions , and impurities . recent optical measurements of silicon nanocrystals synthesized in macroscopic quantities using a process of first etching and then sonically breaking up a silicon wafer @xcite appear to have predominantly spherical shapes , show ultrabright luminescence , and are produced in remarkably uniform batches . the smallest silicon nanoclusters in the range of 1 to 1.2 nm @xcite synthesized through this method exhibit significantly different properties compared with predictions from previous studies such as smaller absorption gaps and different red shifted emission peaks ( i.e. the stokes shift ) @xcite . understanding the differences in these two characteristics call for high quality electronic structure calculations which can precisely describe the effects of electron correlation in ground and excited states , charge transfers , and different surface bonding networks . in this article , we present a state - of - the - art computational study of @xmath1 1 nm hydrogenated si nanocluster prototype systems in order to understand the interplay between their electronic and optical properties , their surface states and structures , and to interpret their absorption and emission processes . we employ a combination of high accuracy _ ab initio _ methods including dft ( using the local density approximation ( lda ) and the perdew burke ernzerhof ( pbe ) @xcite and b3lyp flavors of the generalized gradient approximation ( gga ) ) and qmc to elucidate the electronic and structural properties of the most relevant structural prototypes . we focus on systems with 50 to 70 total atoms which are consistent with the observed sizes that emit in the blue or at the uv edge ( @xmath1 2.90 ev ) . we find that surface dimerization is considered as the best possible mechanism for the occurrence of smaller gaps and stokes shifts than would otherwise be expected in this size range @xcite . we use our derived models to study the structural and electronic effects related to the observed stokes shift , namely the energy difference between absorbed and emitted photons and the character of the corresponding exciton state . the high accuracy of our qmc calculations enables us to deduce that the mechanism most likely to be responsible for the stokes shift in these systems is due to _ a global relaxation of the cluster _ rather than the stretch mode of a single surface dimer @xcite . our results demonstrate that the accurate calculation of a combination of properties is critical for a full understanding of surface reconstructions , doping , and the stokes shift of nanoclusters in this size range . finally , we find that despite its agreement in determining optical gaps for spherical silicon nanocrystals , b3lyp compares less favorably with qmc results for systems with reconstructed surfaces or with oxygenated surfaces . to construct structural prototypes , we start from the bulk silicon lattice and choose a spherical region as this has the smallest surface area for a given number of atoms . the dangling bonds on the surface are saturated with hydrogen and the terminated si atoms are classified into -sih , -sih@xmath2 and -sih@xmath3 types . the structures with -sih@xmath3 are discarded as they are energetically less favorable and therefore prone to reactions with the environment . considering prototypes with -sih and -sih@xmath2 terminations in the 1 - 1.2 nm size regime leads to two likely ideal structures , si@xmath4h@xmath5 and si@xmath6h@xmath5 ( fig . [ fig : sih ] ) . in some cases , dft calculations using a planewave basis and norm - conserving pseudopotentials @xcite are used to establish the equilibrium geometries and the electronic structure of these systems with a kinetic energy cutoff of 35 ry , a gianozzi pseudopotential for hydrogen @xcite , and a hamann pseudopotential for all other atoms @xcite . in other cases , gaussian all - electron calculations using a 6 - 311g@xmath7 basis is used @xcite . as we have previously observed @xcite , the relaxed structures are very similar when calculated using either lda or gradient corrected ( pbe ) functionals and are fairly close to the original bulk derived structures with small adjustments of the bond lengths and angles , predominantly of the surface atoms . for each cluster , the highest occupied molecular orbital ( homo)/ lowest unoccupied molecular orbital ( lumo ) gap is calculated using lda , pbe , and the b3lyp functional . h@xmath5 , si@xmath4h@xmath5 , and si@xmath4h@xmath8 nanocrystals . the larger spheres represent si atoms while the smaller ones the hydrogens . note the surface reconstruction of the top dimer from si@xmath4h@xmath5 to si@xmath4h@xmath8 . ] although dft techniques produce accurate minimum energy structures of silicon nanoclusters and also reliably predict the trends in the optical gap of a given structural type as a function of size @xcite , a more accurate approach , for example , qmc which takes many - body effects into account is required to predict the difference in optical gaps between different classes of structural prototypes , such as clusters with different surface passivants , clusters with reconstructed surfaces , or clusters with amorphous - like geometries . in addition to the computed dft gaps , we therefore adopt a previously described qmc procedure @xcite to perform qmc calculations using the casino code @xcite for the optical gaps discussed in this paper . the smallest experimentally measured clusters are determined to be @xmath9 nm in diameter @xcite . 1 nm clusters are too small to observe crystallinity using transmission electron microscopy ( tem ) , so the structure must be discerned from other experimental methods , or deduced from physical properties such as the absorption gap and the stokes shift . indeed , even if tem or other methods were capable of resolving the crystallinity of 1 nm clusters , it could not predict the surface chemistry . in a previous work , we compared the absorption gap of 1 nm spherical silicon nanoclusters with two types of reconstructed surfaces @xcite . amongst the various structures analyzed , only clusters with reconstructed dimers yielded results consistent with the experimentally measured gaps . however , in addition to all possible oxygenated clusters which were not considered , a recent molecular dynamics calculation predicted that amorphous - like 1 nm clusters can have similar gaps to those with reconstructed dimers @xcite . here , we have recalculated the gaps of these ideal and reconstructed clusters , now using linear scaling qmc @xcite which has allowed us to obtain much smaller statistical errors , and compared them with the double cored clusters @xcite as well as various oxygenated clusters . for comparison , we also compute all relevant homo / lumo gaps within dft using various functionals , focusing on the b3lyp functional in order to make a more complete comparison of this functional with computationally demanding many - body qmc results . table 1 shows that the optical absorption gaps as calculated within qmc of two completely hydrogenated spherical crystalline nanoclusters about 1 nm in diameter , si@xmath4h@xmath5 and si@xmath6h@xmath5 ( 5.3 and 4.9 ev ) are larger than our experimentally measured value of @xmath1 3.5 ev@xcite , while the crystalline cluster with reconstructed dimers are in agreement with experiment as previously shown @xcite . a comparison with oxygenated clusters indicates that those with bridged oxygen are significantly higher than our experimental gaps , while those with double bonded oxygen are significantly lower , in both cases differing by over 1 ev . however , these gaps are consistent with the calculated gaps of double core amorphous - like clusters @xcite , suggesting both structures should be analyzed further . .calculated optical gaps ( ev ) of some prototype 1 nm clusters using three different methods . the statistical errors in the qmc values are in parenthesis . [ cols="<,^,^,^,^",options="header " , ] we also consider here the non - crystalline 1 nm clusters which have been predicted to form during chemical vapor deposition at high temperatures by quantum molecular dynamics ( qmd ) simulations @xcite . although not applicable to the sonification process demonstrated here , these non - crystalline clusters were shown to have a gap comparable to those with ideal reconstructed clusters . we calculate the stokes shift of these clusters to ascertain whether they have the 0.4 to 0.5 ev stokes shifts observed in our clusters and predicted for reconstructed clusters . surprisingly , we find very large stokes shifts , on the order of the gap size ! these non - crystalline 1 nm clusters behave more like small molecules @xcite than quantum dots . thus , the stokes shift has proved a very powerful characterization technique eliminating the `` double core '' non - crystalline clusters as being those observed here . in conclusion , we have carried out a thorough study of hydrogen terminated silicon nanoclusters in the 1 - 1.2 nm range . we investigated several prototype structures and compared their optical absorption gaps and stokes shifts with recent measurements and found that although a few other structures may yield a similar gap , most notably an amorphous - like double core cluster , only si@xmath4h@xmath8 yields both the correct gap and stokes shift . thus , we determine that _ both _ properties must be considered when evaluating candidate structures to interpret optical measurements . we determined that b3lyp generates inconsistent results for clusters with localized orbitals compared with qmc level calculations . the atomistic first - principles dft approaches coupled with qmc allowed us to study the optically induced excitons and to conclude that the most likely mechanism causing the stokes shift is the barrierless relaxation of the whole structure with the red shift of @xmath1 0.4 ev in agreement with experiment . comparison with further experimental data indicates that the si@xmath4h@xmath8 structural prototype is the most promising candidate of the possibilities we tested . we have also investigated the effect of doping with a number of atoms and molecular groups . like ideal unreconstructed structures , @xmath10 bonded passivants have a minimal effect on absorption gaps , as do single bridged passivants . therefore , other atoms or ligands may be used to functionalize these clusters with no discernible change to the gap . we thank erik draeger for providing qmd non - crystalline structures and for helpful discussions . in part this work was performed under the auspices of the u.s . department of energy by the university of california , lawrence livermore national laboratory under contract no . w-7405-eng-48 . we also gratefully acknowledge nsf support from the grant dmr-0102668 and m.n . acknowledges support by the nsf grant . part of the calculations has been done at ncsa , university of illinois . a. d. yoffe , adv . phys . * 50 * , 1 ( 2001 ) and references therein . c.c . chen , a.b . herhold , c.s . johnson , and a.p . alivisatos , science * 276 * , 398 ( 1997 ) . chan and s. nie , science * 281 * , 2016 ( 1998 ) . s. furukawa and t. miyasato , phys . b * 38 * , 5726 ( 1998 ) . w. l. wilson , p.f . szajowski , and l.e . brus , science * 262 * , 1242 ( 1993 ) . s. schuppler , s.l . friedman , m.a . marcus , d.l . adler , y.h . xie , f.m . ross , t.d . harris , w.l . brown , y.j . chabal , l.e . brus , and p.h . citron , phys . lett . * 72 * , 2648 ( 1994 ) . t. van buuren , l.n . dinh , l.l . chase , w.j . siekhaus , and l.j . terminello , phys . 80 * , 3803 ( 1998 ) . holmes , k.j . ziegler , r.c . doty , l.c . pell , k.p . johnston , and b.a . korgel , j. am . soc . * 123 * , 3743 ( 2001 ) . g. belomoin , j. therrien , a. smith , s. rao , r. twesten , s. chaieb , m.h . nayfeh , l. wagner , and l. mitas , appl . . lett . * 80 * , 841 ( 2002 ) . m. h. nayfeh , n. barry , j. therrien , o. akcakir , e. gratton , and g. belomoin , appl . lett . * 78 * , 1131 ( 2001 ) . j. p. proot , c. delerue and g. allan , appl . . lett . * 61 * , 1948 ( 1992 ) . wang and alex zunger , j. chem . 100 , 2394 ( 1994 ) . b. delley and e.f . steigmeier , phys . b * 47 * , 1397 ( 1993 ) . s. ogut , j. r. chelikowsky and s. g. louie , phys . lett . * 79 * , 1770 ( 1997 ) . m. rohlfing and s. g. louie , phys . 80 * , 3320 ( 1998 ) . bendict , a. puzder , a.j . williamson , j.c . grossman , g. galli , j.e . klepeis , j .- y . raty , and o. pankratov , phys . b * 68 * , 085310 ( 2003 ) . a. puzder , a. j. williamson , j.c . grossman , and giulia galli , phys . lett . * 88 * , 097401 ( 2002 ) . g. belomoin , j. therrien , and m.h . nayfeh , appl . . lett . * 77 * , 779 ( 2000 ) . i. vasiliev , s. ogut , j.r . chelikowsky , phys . lett . * 86 * , 1813 ( 2001 ) . williamson , j.c . grossman , r.q . hood , a. puzder , and g. galli , phys . * 89 * , 196803 ( 2002 ) . e. martin , c. delerue , g. allan , and m. lannoo phys . b * 50 * , 18258 ( 1994 ) . t. takagahara and k. takeda , phys . rev . b * 53 * , r4205 ( 1996 ) . g. allan , c. delerue and m. lannoo , phys . rev * 76 * , 2961 ( 1996 ) . perdew , k. burke , and m. ernzerhoff , phys . * 78 * , 1396 ( 1997 ) . l. mitas , j. therrien , r. twesten , g. belomoin , and m.h . nayfeh , appl . phys . lett . * 78 * , 1918 ( 2001 ) . the code gp 1.8.0 ( f.gygi , llnl ) was used . p. gianozzi ( private communication ) . d. hamann phys . b , * 40 * , 2980 ( 1989 ) . m.j . frisch _ et al . _ , a.4 ed . , gaussian , inc . , pittsburgh , pa ( 1998 ) . williamson , randolph q. hood and j. c. grossman , phys . lett . * 87 * , 246406 ( 2001 ) . a. puzder , a.j . williamson , j.c . grossman , and g. galli , j. chem . * 117 * 6721 ( 2002 ) . r. j. needs , g. rajagopal , m. d. towler , p. r. c. kent and a. j. williamson , casino version 1.0 user s manual , university of cambridge , cambridge 2000 . e. draeger , j.c . grossman , a. j. williamson , and g. galli , phys . . lett . * 90 * , 167402 ( 2002 ) . garoufalis , a.d . zdersis , and s. grimme , phys . lett . * 87 * , 276402 ( 2001 ) . z. zhou , l. brus , and r. friesner , nanoletters * 3 * , 163 ( 2003 ) . a. puzder , a.j . williamson , j.c . grossman , and g. galli , j. am . 125 * 2786 ( 2003 ) . wolkin , j. jorne , p.m. fauchet , g. allan , and c. delerue , phys . * 82 * , 197 ( 1999 ) . m. hirao and t. uda , int . j. quant . chem . * 52 * , 1113 ( 1994 ) . caldas physica status solidi b * 217 * , 641 ( 2000 ) . nayfeh , n. rigakis and z. yamani , phys . b * 56 * , 2079 - 2084 ( 1997 ) . a. franceschetti and s.t . pantelides , phys . b * 68 * , 033313 ( 2003 ) .	we evaluate the optical gap and stokes shift of several candidate 1 nm silicon nanocrystal structures using density functional and quantum monte carlo ( qmc ) methods . we find that the combination of absorption gap calculations and stokes shift calculations may be used to determine structures . we find that although absorption gaps calculated within b3lyp and qmc agree for spherical , completely hydrogenated silicon nanocrystals , they disagree in clusters with different surface bonding networks . the nature of the stokes shift of the ultrabright luminescence is examined by comparing possible relaxation mechanisms . we find that the exciton which reproduces the experimental value of the stokes shift is most likely a state formed by a collective structural relaxation distributed over the entire cluster .
femtoscopic bose - einstein correlations , also known as hbt / gglp effect , were investigated in ref . @xcite by cms for pp collisions at @xmath40.9 tev @xcite , 2.36 tev @xcite and 7 tev @xcite . such phenomenon was discovered by goldhaber , s. goldhaber , w. lee and a. pais ( gglp effect ) @xcite , being the analogous in high - energy collisions to a similar method proposed by r. hanbury - brown and r. q. twiss ( hbt effect ) @xcite for estimating angular dimensions of stars . a broad investigation was carried out in these studies in terms of the invariant relative momenum @xmath5 . in ref . @xcite , similarly to what was previously observed in @xmath6 collisions @xcite , an anticorrelation was reported . this result required a study of different fitting functions , as had been suggested in the case of small systems , such as the @xmath7 model @xcite , which considers strong correlations between the space - time coordinates and the momentum components of the emitted particles , and was found to describe better the overall behavior of the correlation functions . a natural extension of those analyses is to investigate the gglp / hbt correlations with respect to different components of the pair relative momentum , which allows for exploring the source sizes in different directions . such analyses have also been studies by other experiments at rhic and lhc @xcite . therefore , the hbt / gglp correlation is measured in two - dimensions ( 2-d ) as a function of the relative momenta along and transverse to the beam direction , @xmath8 and @xmath9 , respectively . in three - dimensions ( 3-d ) the bose - einstein correlations are studied in terms of @xmath8 , @xmath10 and @xmath11 , these last two obtained , respectively , by projecting the transverse component @xmath12 in orthogonal directions , i.e. , parallel to the average transverse momentum of the pair ( @xmath1 ) , and orthogonal to both @xmath8 and @xmath10 . using the same framework as in ref . @xcite , the hbt / gglp effect is further scrutinised here for charged hadrons produced in minimum bias events in pp collisions at @xmath13 and 7 tev with the data collected by cms at the cern lhc . new 1-d results at both these energies are discussed and then the analysis is extended to 2-d and 3-d cases . the data sets used in this analysis correspond to minimum bias samples obtained in pp collisions at 2.76 and 7 tev recorded with the cms detector at the lhc @xcite . a detailed description of the cms detector can be found in ref . @xcite . the minimum bias sample at @xmath14 2.76 tev was triggered on - line by requiring at least one track with @xmath15 gev to be found in the pixel tracker with @xmath16 for a pp bunch crossing . besides , in the offline analysis hadronic collisions were selected by requiring a coincidence of at least one forward hadronic ( hf ) calorimeter tower with more than 3 gev of total energy in each of the hf detectors . in the case of pp collisions at 7 tev events were selected by a trigger signal in each side of the beam scintillation counter ( bsc ) , coincident with a signal from either of the two detectors indicating the presence of at least one proton bunch crossing the interaction point ( ip ) . collision events were then selected offline by requiring a beam pickup for timing for the experiments ( bptx ) signal from both beams passing the ip . the set of pp collision events at @xmath13 tev used in this analysis comprises the data collected by cms in 2013 at the cern lhc ( 3.4 million events ) . at @xmath17 tev , a combined sample from pp collisions was considered , which uses data from three periods of the cms data taking , i.e. , commissioning run ( 23 million events ) , as well as from the runs 2010a ( 16 million events ) and run 2010b ( 4 million events ) , where the first is almost pileup free , while the later has a non - negligible fraction of events with multiple interactions . a filter was used for reducing the contamination in case of multiple vertices ( the reconstructed vertex with the largest number of associated tracks is selected at 7 tev , while in pp collisions at 2.76 tev , an additional primary vertex might be identified as originating from a second pp collision by looking at its properties ) . to assess the related systematic uncertainty , an alternative event selection for reducing pileup contamination was also investigated by considering only single reconstructed vertex events . in the case of pp collisions at 2.76 tev , three monte carlo samples were used . for obtaining the bec results , minimum bias events simulated with pythia 6 z2 tune @xcite were employed , whereas pythia d6 t and pythia z2star @xcite were used for estimating the systematic uncertainties related to the choice of monte carlo tune . each of them contained 2 million events . for the analysis at 7 tev about 33 million monte carlo events were simulated using pythia 6 z2 tune . the track selection employed in both cases above follows the same criteria as in ref . @xcite , and are discussed in details in these references , as well as in @xcite . the procedure adopted is the same as described in refs . although no particle identification is considered in this analysis , the contamination from non - pions is not expected to be sizeable , as discussed in @xcite , since pions are the dominant type of hadrons in the sample . for each event , the signal containing the bose - einstein correlations is identified by pairing same charge tracks from the same event and distributing them in bins of the relative momentum of the pair , for instance , @xmath18 , being @xmath19 the four - momenta of the individual particles in the pair . the background distribution or reference sample is formed similarly , by pairing charged particles from different events and within the same @xmath20 range ( where the full pseudorapidity interval is divided in three subranges @xmath21 , corresponding to @xmath22 , @xmath23 , and @xmath24 ) , as in ref . this mixed event technique is referred to as `` same track density in @xmath21 '' . a single ratio is then formed , having the signal pair distribution as numerator and the reference sample as denominator , with the appropriate normalization , i.e. , @xmath25 the invariant relative momentum of the pair is defined as @xmath26 ; @xmath27 is the integral of the signal pair distribution of all the events , whereas @xmath28 is the equivalent in the reference sample . a double ratio technique is then taken with the data and the monte carlo single ratios corresponding to the @xmath29 distributions , in terms of which the bose - einstein correlation ( bec ) effect is investigated @xcite , @xmath30 where @xmath31 is the single ratio computed with the simulated events generated without bec . in each case , the reference samples for data and simulation are obtained in the same way . this double - ratio procedure has the advantage of considerably reducing the sources of bias due to track inefficiency and other detector - related effects , as well as other bose - einstein correlations . the gglp / hbt method reflects not only the quantum statistics of the pair of identical particles , but is also sensitive to the underlying dynamics . in particular , in the case of charged hadrons , the correlation function may be distorted by strong , as well as by coulomb interactions . for pions , the strong interactions can usually be neglected in femtoscopic measurements . as in ref . @xcite , the depletion ( enhancement ) in the correlation function caused by the coulomb repulsion ( attraction ) of equal ( opposite ) charge pairs in the case of pions is corrected by weighting pair - wise with the inverse gamow factor @xcite . this factor , in case of same charge and opposite charge , is given by @xmath32 with @xmath33 , where @xmath34 the pion mass and @xmath29 the invariant relative momentum of the pair . for performing the multidimensional analysis , the double ratios are investigated in terms of the projections of the relative momentum @xmath35 in two or three directions . in the 2-d case , the decomposition is made in @xmath36 ( _ longitudinal _ component , along the beam direction ) , and @xmath37 ( _ transverse _ component ) . in the 3-d case , additional projections are considered in the transverse plane , resulting in @xmath10 ( _ outwards _ ) , and @xmath11 ( _ sidewards _ ) , respectively along the average transverse momentum of the pair , @xmath38 , and orthogonal to it ; @xmath10 , @xmath11 , and @xmath8 are mutually orthogonal . this decomposition of the relative momentum of the pair is also known as bertsch - pratt variables @xcite . the investigations are carried out in center - of - mass ( cm ) , i.e. , the lhc laboratory frame , as well as in the local co - moving system ( lcms ) , characterized by the frame in which the longitudinal component of the pair average momentum ( @xmath39 ) is zero . details about the 2-d and 3-d relative momentum projections , as well as the boost to the lcms can be found in ref . @xcite . the parameterizations used to fit the correlation functions in one- ( 1-d ) , two- ( 2-d ) and three - dimensions ( 3-d ) , respectively in terms of @xmath29 , @xmath40 and @xmath41 , are listed below . throughout this analysis @xmath42 is adopted . @xmath43 \ ; ( 1 + \delta \ ; q_{\textrm{inv } } ) , \label{eq:1d - levy}\ ] ] @xmath44 } \big\ } \times ( 1 + \alpha q_t + \beta q_l ) , \label{eq:2d - levy}\ ] ] @xmath45 } \big\ } \times \nonumber \\ & & ( 1 + \alpha q_s + \beta q_l + \gamma q_o ) . \label{eq:3d - levy}\end{aligned}\ ] ] in the above expressions , @xmath46 is the intercept parameter ( intensity of the correlation in the smallest bin of the pair relative momentum ) , @xmath47 , @xmath48 are constants . the exponent @xmath49 is the lvy index of stability satisfying the inequality @xmath50 . in all the above cases , if treated as a free parameter when fitting the double ratios , this exponent usually results into a number between the value characterizing the exponential ( @xmath51 ) and the gaussian ( @xmath52 ) functions . more details can be found in ref . for the sake of clarity , we denote the longitudinal component of the relative momentum in the cm frame as @xmath8 , and in the lcms , as @xmath53 . in particular , in the case @xmath51 the exponential term coincides with the fourier transform of the source function @xmath54 , characterised by a lorentzian distribution ; the radius parameters @xmath55 , @xmath56 , @xmath57 , correspond to the lengths of homogeneity fitted to the correlation function in 1-d , 2-d and 3-d , respectively . the additional polinomial terms are introduced for accommodating possible deviations of the baseline from unity at large values of these variables ( long - range correlations ) , as well as for allowing a better quality fit . in the 2-d and 3-d cases , @xmath51 leads to the so - called _ stretched exponential _ function . in eq . ( [ eq:2d - levy ] ) , @xmath58 and @xmath59 where @xmath60 and @xmath61 , originated in the mass - shell constraint ( @xmath62 ) ; @xmath63 is the angle between the directions of @xmath64 and @xmath65 and @xmath7 is the source life - time . in eq . ( [ eq:3d - levy ] ) @xmath66 , @xmath67 , @xmath68 and @xmath69 . both in eq . ( [ eq:2d - levy ] ) and ( [ eq:3d - levy ] ) a frame dependent cross - term , respectively proportional to @xmath70 and @xmath71 is present . however , when the analysis is performed in the lcms , this cross - term does not contribute for sources symmetric along the longitudinal direction . various sources of systematical uncertainties were considered in this analysis , as listed in table [ tab:1d - syst - summary ] , the first four being similar to what was discussed in ref . two additional studies were also performed : the effect of separating positive from negative charges in the single ratios ( since @xmath72 , as well as @xmath73 charges are added in the signal ; in the reference sample , besides these two combinations , the @xmath74 case is also added to the sample ) . the second study added is the effect of pileup events , investigated by comparing the results to the case where only single - vertex events are considered . the exponential function in eq.([eq:1d - levy ] ) ( lvy type with @xmath51 ) was adopted for this investigation . the total values of the systematic uncertainties are calculated by adding the individual contributions in quadrature . the systematic uncertainties estimated in 1-d and summarized in table [ tab:1d - syst - summary ] are extended to both the 2-d and the 3-d cases . l\|c\|c\|c\|c + @xmath75 & & + origin of systematics & @xmath46 & @xmath55 ( fm ) & @xmath46 & @xmath55 ( fm ) + monte carlo tune & 0.032 & 0.160 & 0.032 & 0.160 + reference sample & 0.009 & 0.047 & 0.051 & 0.188 + coulomb corrections & 0.016 & 0.009 & 0.018 & 0.020 + track cuts & 0.014 & 0.119 & 0.014 & 0.119 + charge dependence & 0.006 & 0.012 & 0.007 & 0.006 + pileup filter & 5.0 e-4 & 0.011 & 0.001 & 0.0025 + total & 0.040 & 0.206 & 0.065 & 0.275 + [ tab:1d - syst - summary ] the current analysis extends the previous one for pp at @xmath17 tev reported in ref . @xcite to full data sample , as well as at @xmath13 tev with full available statistics . the corresponding results for single and double ratios are shown in fig . [ fig:1-d - singleratios ] . the fits to the double ratios were produced with the exponential function in eq . ( [ eq:1d - levy ] ) , with @xmath51 . versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=38.0% ] versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=38.0% ] versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=44.0% ] versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=44.0% ] versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=41.0% ] versus @xmath76 ( acceptance and efficiency corrected ) , for pp collisions at 2.76 and 7 tev ( fit curves are proportional to @xmath77 ) . the inner error bars represent statistical uncertainties and the outer ones , statistical and systematic uncertainties added in quadrature . right : the anticorrelation s depth , @xmath78 , versus @xmath76 . , title="fig:",scaledwidth=41.0% ] figure [ fig : fits - to - dr - nch - kt-276 - 7tev ] shows results for the intercept parameter @xmath46 and the invariant radius @xmath55 from pp collisions at 2.76 and 7 tev ( full statistics ) , in terms of @xmath2 and @xmath1 . it can be seen that the results corresponding to the two energies are very similar for the different ( @xmath2 , @xmath1 ) combinations . the intercept @xmath46 decreases with increasing @xmath1 and @xmath2 , whereas @xmath55 steadily increases with multiplicity and seem to decrease with @xmath1 , at least for the two largest @xmath2 bins , showing that the lengths of homogeneity accessible to interferometric measurements decrease with the average pair momentum , as has been previously observed in several different collision systems and energy ranges @xcite.this behavior is expected in the case of emitting sources originated from expanding systems . in fig . [ fig : dr - radii - nch - dipdepth-276 - 7 ] ( left ) , @xmath55 is investigated as a function of the charged particle multiplicity , @xmath76 ( efficiency and acceptance corrected ) , where the curves are fits proportional to @xmath77 . the results for pp collisions at 2.76 tev and at 7 tev are consistent with former studies reported in ref . @xcite at 0.9 and 7 tev , showing a similar increase with @xmath77 . such results also suggest an approximate scaling property of the lengths of homogeneity with increasing collision center - of - mass energy . as discussed in @xcite , an anticorrelation ( value below unity ) was observed in the double ratios for values of @xmath29 away from the bose - einstein peak , whose depth was shown to decrease with increasing @xmath2 , for integrated values in @xmath1 . this dip structure is also observed in the present analysis and its depth is further investigated . more details and discussion on the results are presented in ref . an additional function , @xmath79 e^{-(qr_\alpha)^\alpha } ] \right\ } \cdot(1+\delta q)$ ] , was used to fit the data points , which better describes such anticorrelation , as discussed in @xcite . it is based on the so - called @xmath7 model @xcite , which parameterizes the time evolution of the source by means of a one - sided asymmetric lvy distribution . the _ dip _ s depth @xcite is estimated by the difference of the base - line function , @xmath80 , and the remaining fit function based on the @xmath7 model at its minimum , leading to the results shown in fig . [ fig : dr - radii - nch - dipdepth-276 - 7 ] ( right ) , where the results for pp collisions at both 0.9 and 7 tev from ref . @xcite and the new ones for 2.76 tev and the full sample at 7 tev are shown together ( see @xcite for details ) . the bec analysis is extended to the 2-d case in terms of the components @xmath81 of the pair relative momentum , with the data samples from pp collisions at @xmath82 tev and 7 tev . it is performed both in the cm frame and in the lcms , in which the cross - term depending on @xmath83 in 2-d , or @xmath84 , in 3-d , does not contribute , in case of longitudinally symmetric systems . ) , with data from pp collisions at 2.76 tev ( top ) and 7 tev ( bottom ) in the lcms corresponding to results integrated in all @xmath2 and @xmath1 bins . the right panel shows the corresponding 1-d projections of the single and double ratios in terms of @xmath53 ( for @xmath85 gev ) and @xmath9 ( for @xmath86 gev ) . gaussian , exponential and lvy ( with @xmath51 ) fit functions are shown superimposed to the data points . , title="fig:",scaledwidth=44.0% ] ) , with data from pp collisions at 2.76 tev ( top ) and 7 tev ( bottom ) in the lcms corresponding to results integrated in all @xmath2 and @xmath1 bins . the right panel shows the corresponding 1-d projections of the single and double ratios in terms of @xmath53 ( for @xmath85 gev ) and @xmath9 ( for @xmath86 gev ) . gaussian , exponential and lvy ( with @xmath51 ) fit functions are shown superimposed to the data points . , title="fig:",scaledwidth=51.0% ] ) , with data from pp collisions at 2.76 tev ( top ) and 7 tev ( bottom ) in the lcms corresponding to results integrated in all @xmath2 and @xmath1 bins . the right panel shows the corresponding 1-d projections of the single and double ratios in terms of @xmath53 ( for @xmath85 gev ) and @xmath9 ( for @xmath86 gev ) . gaussian , exponential and lvy ( with @xmath51 ) fit functions are shown superimposed to the data points . , title="fig:",scaledwidth=44.0% ] ) , with data from pp collisions at 2.76 tev ( top ) and 7 tev ( bottom ) in the lcms corresponding to results integrated in all @xmath2 and @xmath1 bins . the right panel shows the corresponding 1-d projections of the single and double ratios in terms of @xmath53 ( for @xmath85 gev ) and @xmath9 ( for @xmath86 gev ) . gaussian , exponential and lvy ( with @xmath51 ) fit functions are shown superimposed to the data points . , title="fig:",scaledwidth=51.0% ] as an illustration , the double ratios in the lcms are shown in fig . [ fig:2d_doubleratios - project-276 - 7tev - lcms ] , as a 2-d plot ( left panel ) in terms of @xmath87 , and the corresponding 1-d projections ( right panel ) for pp collisions at 2.76 tev ( top ) , and for 7 tev ( bottom ) . the 1-d projections , when plotted in terms of @xmath53 , considers only the first bin in @xmath9 ( i.e. , @xmath88 gev ) , and vice - versa , with the data superimposed by the gaussian , exponential and lvy ( with @xmath51 ) fit functions . ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] ) function are shown for pp collisions at 2.76 and 7 tev in the cm frame ( top ) and in the lcms ( bottom ) , as a function of @xmath89 and for different @xmath2 bins . the statistical uncertainties are indicated by error bars ( in some cases , smaller than the marker s size ) , whereas the systematic ones are indicated by empty ( at 2.76 tev ) or shaded boxes ( at 7 tev ) . , title="fig:",scaledwidth=32.7% ] analogously to the studies performed in 1-d the double ratios in 2-d were also investigated in terms of @xmath90 , in three intervals of the pair average momentum , @xmath1 , and in different @xmath2 bins . the results from the _ stretched exponential _ fit ( lvy - type with @xmath51 ) @xcite to the double ratios , performed both in the cm frame ( top ) and in the lcms ( bottom ) , are compiled in fig . [ fig:2d_radii_rl_rt - lambda-276_7tev - cm - lcms ] . the behaviour of the directional lengths of homogeneity is very similar in both frames , with @xmath91 ( @xmath92 ) and @xmath93 increasing with charged multiplicity , @xmath2 , and decreasing with the average transverse momentum , @xmath1 , at least in the larger multiplicity bins , a behaviour similar to that observed in the 1-d case and expected for expanding sources . another interesting feature of the data than can be observed in fig . [ fig:2d_radii_rl_rt - lambda-276_7tev - cm - lcms ] is that @xmath92 ( lcms ) @xmath94 ( cm ) for the same bins of @xmath2 and @xmath1 , suggesting an effect of lorentz boost contraction in the longitudinal length of homogeneity in the cm frame . regarding @xmath46 , as shown in fig . [ fig:2d_radii_rl_rt - lambda-276_7tev - cm - lcms ] , no significant sensitivity of the intercept is seen as a function of @xmath1 . however , within each @xmath1 range , @xmath46 slowly decreases with increasing track multiplicity in an similar way in both frames . . 2-d fit parameters for in the lcms [ cols="<,^,^ " , ] [ tab : radii-3d-7tev ] the fits to the 3-d double ratios were also investigated in three @xmath1 bins ( integrating over all @xmath2 ) . the fit parameters were obtained with gaussian , exponential and lvy ( with @xmath51 ) fit functions ( when treated as a fit parameter @xmath49 returned values close to unity also in the 3-d case , both in the cm and in the lcms ) . the results are compiled in fig . [ fig : fits_exp - gauss - levy - qsqlqo - kt_cm - lcms-7tev ] for the data from pp collisions at 7 tev , showing to depend noticeably on the type of fit used , the radius parameters being considerably larger in the case of the lvy - type function . the dependence on @xmath1 , however , seems to be similar for the three fit functions . the @xmath95 fit values are among the largest ( except for the gaussian fit ) and seem insensitive to @xmath1 , in both frames . also @xmath91 seems to be insensitive to @xmath1 , and is the smallest radius parameter in the cm frame . however , it shows opposite behavior in the lcms , where its decrease with increasing @xmath1 is more pronounced , also attaining the largest values of the three radius parameters , suggesting an effect related to the lorentz boost in the longitudinal direction , as in the 2-d case . the @xmath96 fit values are slightly smaller in the lcms as compared to the cm frame , and decrease moderately with increasing @xmath1 . its dependence on @xmath1 is similar in both frames , although it has a slightly steeper decrease with increasing @xmath97 in the lcms . ( efficiency and acceptance corrected ) , integrated in @xmath1 . the statistical uncertainties are indicated by error bars ( in some cases smaller than the marker s size ) , whereas the systematic ones are indicated by the shaded boxes . , title="fig:",scaledwidth=100.0% ] ( efficiency and acceptance corrected ) , integrated in @xmath1 . the statistical uncertainties are indicated by error bars ( in some cases smaller than the marker s size ) , whereas the systematic ones are indicated by the shaded boxes . , title="fig:",scaledwidth=100.0% ] ( efficiency and acceptance corrected ) , integrated in @xmath1 . the statistical uncertainties are indicated by error bars ( in some cases smaller than the marker s size ) , whereas the systematic ones are indicated by the shaded boxes . , title="fig:",scaledwidth=100.0% ] ( efficiency and acceptance corrected ) , integrated in @xmath1 . the statistical uncertainties are indicated by error bars ( in some cases smaller than the marker s size ) , whereas the systematic ones are indicated by the shaded boxes . , title="fig:",scaledwidth=100.0% ] the fits to the double ratios were also studied in four @xmath2 bins ( integrating over all @xmath1 ) . the corresponding results are shown as a function of @xmath98 ( efficiency and acceptance corrected ) in fig . [ fig : fits_exp - gauss - levy - qsqlqo - nch_cm - lcms-7tev ] , obtained both in the cm frame ( top ) and in the lcms ( bottom ) . a clear behavior can be seen , common to all fit functions and in all directions of the relative momentum components : the fit radius parameters @xmath95 , @xmath91 and @xmath96 increase with increasing average multiplicity , indicating an increase in the lengths of homogeneity with @xmath2 , similar to what was seen in the 1-d and 2-d cases . the intercept parameter @xmath46 was also studied as a function of @xmath1 and @xmath2 , both in the cm frame and in the lcms . the corresponding results are shown in fig . [ fig : lambda-3d - kt - nch - cm - lcms ] . the values of @xmath98 shown in the plots were corrected for efficiency and acceptance . a moderate decrease with increasing @xmath1 is observed . as a function of increasing @xmath2 , @xmath46 first decreases and then seems to saturate . in the 2-d and 3-d cases , the values of the longitudinal radius fit parameters coincide within the experimental uncertainties @xcite , as expected , since both correspond to the length of homogeneity in the beam direction . from figures [ fig : fits_exp - gauss - levy - qsqlqo - kt_cm - lcms-7tev ] and [ fig : fits_exp - gauss - levy - qsqlqo - nch_cm - lcms-7tev ] it can seen that there is an approximate scaling of @xmath91 ( @xmath99 ) and @xmath93 with @xmath2 , when comparing the results at 2.76 tev and at 7 tev . in fig . [ fig:3d - dip - structure - lcms ] the 3-d results for the double ratios in the lcms ( integrated in @xmath2 and @xmath1 ) are shown as 2-d projections in terms of pairs of @xmath100 ( the complementary one within @xmath101 gev ) . the plots are zoomed along the correlation function axis , cutting values above 1.2 . the corresponding 1-d projections along variable @xmath102 and @xmath10 ( other two variables within @xmath103 gev ) are also shown . ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=35.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=35.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=35.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=35.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=32.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=32.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=32.0% ] ) , ( @xmath104 ) , and ( @xmath105 ) , integrated in @xmath2 and @xmath1 , with @xmath106 gev , @xmath86 gev and @xmath107 gev , respectively . the bottom panel shows 1-d projections of the same data , with the complementary two variables constrained to be within the first bin ( i.e. , @xmath108 gev ) . , title="fig:",scaledwidth=100.0% ] the analysis discussed here extends the 1-d results reported in the two bec publications in @xcite by measuring the correlation functions in terms of different components of the pair relative momentum , as is usually studied by other experiments . this allows to investigate the extension of the source accessible to the correlation technique in different directions . two main projections are considered : in two - dimensions ( 2-d ) , the femtoscopic correlation is investigated as a function of the variables @xmath8 and @xmath9 , and in three - dimensions ( 3-d ) , as function of @xmath109 , being @xmath8 the same as in the 2-d case . for achieving this purpose , minimum bias events produced in proton - proton collisions at 2.76 and 7 tev ( full data sample ) are scrutinised in detail , as if by means of a magnifying lens . at 7 tev , the full data sample is used and the corresponding results are compared with the ones at the same energy recorded during the commissioning run at the lhc in 2010 . this analysis also extends the measurements of the bec correlations to the full minimum bias sample from pp collisions at 2.76 tev collected in 2013 , which is a very important baseline for the measurements of this second order interferometry in pbpb collisions at the same energy per nucleon . in 1-d are the results from both energies are compared with the ones in ref . @xcite at lower energies , as well as at @xmath110 tev , recorded during the commissioning run at the lhc in 2010 . in particular , comparisons showed that @xmath111 steadly increases with the charged multiplicity proportionally to @xmath77 . the measurements were performed both in the collision center - of - mass ( cm ) frame and in the local co - moving system ( lcms ) , where the average longitudinal momentum of the pair is zero . in the 2-d case , for integrated values of @xmath2 and @xmath1 , the lengths of homogeneity in the lcms suggest that the source is elongated along the beam direction , i.e. , @xmath112 . in the 3-d case , it was found that @xmath113 . in addition , it can be observed that the fit values for the longitudinal radius parameter , @xmath91 are consistent in 2-d and in 3-d cases , as should be expected , since they correspond to the length of homogeneity in the longitudinal direction in both cases . this conclusion is attained with respect to the @xmath91 fit parameter in the cm frame , as well as @xmath92 in the lcms @xcite . the anticorrelation observed in 1-d and reported in ref . @xcite , was also observed in minimum bias pp collisions at 2.76 tev and further investigated here with the full statistics at 7 tev . these new 1-d results are also compared with those in ref . @xcite , verifying a consistent behavior both for the invariant radius parameter and for the dip s depth measurements at 2.76 and 7 tev . such comparisons showed that the dip s depth decreases with increasing @xmath2 . cms collaboration , `` multidimensional analysis of bose - einstein correlations in pp collisions at 2.76 and 7 tev '' , cms pas fsq-13 - 002 ( 2014 ) ( https://twiki.cern.ch/twiki/bin/view/cmspublic/physicsresultsfsq13002 ) .	multidimensional two - particle bose - einstein correlation functions of charged hadrons are reported for @xmath0 collisions at 2.76 and 7 tev in terms of different components of the pair relative momentum , extending the previous one - dimensional ( 1-d ) analyses of cms . this allows for investigating the extension of the source accessible to the femtoscopic correlation technique in different directions , revealing a more detailed picture of the emitting source in these collisions at increasing energies . the measurements are performed for different intervals of the pair average transverse momentum , @xmath1 , and for increasing charged particle multiplicitiy , @xmath2 . results in 1-d , 2-d and 3-d show a decrease of the fit radius parameters with @xmath1 , whereas a clear rise with @xmath2 is observed in all cases . in addition , the fit radius parameters at both energies show close similarity in size and behavior within the same intervals of ( @xmath3 ) .
pks0349 - 27 is a prime example of a classical , double - lobed ( frii ) radio galaxy ( @xcite ) with a narrow emission line nuclear spectrum . it is also a spectacular example of a radio agn host with an extended gaseous component which has a structure reminiscent of a barred spiral ( @xcite ; see figure1b in section [ images ] ) . it has a redshift of @xmath319800kms@xmath0 ( @xcite ) , corresponding to a scale of @xmath30.96kpc@xmath4 this galaxy appears to fit into the `` rotator '' class of frii , double - lobed , radio galaxies defined by @xcite . earlier kinematic studies of this object have suggested that its centrally concentrated ionized gas is in rapid rotation as well as expansion ( with a rotation axis at @xmath3pa140 - 180@xmath5 , @xcite ) . its non - nuclear spectrum has been extensively observed ( op cit . and @xcite ) and incorporated in several analytical discussions of the type of ionization / excitation conditions found in the non - nuclear regions of active galaxies ( op cit . and @xcite ) . on the basis of hst , uv continuum images , @xcite have concluded that there is no strong uv emission from its nucleus ; however , they do report some evidence for extended uv emission in the @xmath33 region surrounding the nucleus . detailed radio continuum images of this object have been published by @xcite at 4.89 ghz and by @xcite at 1400 mhz . although it was initially thought to be a rare phenomenon , the presence of eelg is quite common in many strong radio galaxy hosts ( as reviewed and discussed by @xcite ) . it may well be related to the `` alignment '' effect seen in more distant , powerful radio galaxies ( cf . @xcite ) and frequently seems to reflect a different kinematic system from the stellar component of the radio galaxy host ( cf . the reviews in @xcite , and @xcite ) . ideas about the origins of this gas , its ionization / excitation parameters , and kinematic properties are reviewed in the references cited above as well as @xcite , and @xcite . these can be briefly summarized as follows : 1 . origins : the gas may be native , from stellar mass loss ; acquired in a `` cooling flow ; '' accreted from primordial cold gas ; or acquired from a gas - rich galaxy in a merger or close encounter . ionization / excitation : the gas may be photoionized by a hard continuum source originating in the nucleus ; shock heated by collisional impact due to radio jets or tidal disturbances from a merger or close encounter and post shock , radiative re - heating ; photoionized by starburst activity ; or a combination of these . kinematics : the gas may be in stable orbits about the host ; disrupted by collisions arising from mergers and tidal disruption ; or disrupted by the impact of a radio jet . nearby objects are particularly important for resolving the complex interactions noted above because high spatial resolution is needed to distinguish between the different physical zones which can arise within the clouds . the number of nearby , intrinsically powerful radio sources is , however , limited and thus it is important to subject those which do exist to intense analysis . this implies that these nearby objects should be restudied as instrumental capabilities improve our ability to measure such objects more accurately and with higher resolution . pks0349 - 27 is one of the few objects which is close enough for such detailed study and which also displays an extensive , highly structured ionized gas region in its host galaxy . in this paper we report new imaging and spectroscopic data for this object which identify regions of extreme kinematic disruption in the eelg of this galaxy s host . the analysis of these data is described in sections [ images ] and [ spectra ] . the implications of the observed kinematics and diagnostic emission line ratios for the origin and ionization / excitation of this galaxy s eelg are discussed in sections [ velkin ] , [ elr ] , and [ jets ] . as part of a program to examine the complex kinematics in frii radio galaxies we obtained narrow band h@xmath2 and continuum images and long slit spectra of pks0349 - 27 with the ctio 1.5 m and 4 m telescopes . the record of these observations is summarized in table [ tbl-1 ] . direct images of pks0349 - 27 were taken in january 1988 with the ctio 1.5 m , 888@xmath6800 ti ccd ( binned to 444@xmath6400 ) and @xmath7 cassegrain secondary ( lines 1 to 4 in table [ tbl-1 ] ) . the ccd scale was 0.31@xmath8 . the galaxy was observed through matched on - band off - band filters with the aim of minimizing psf differences between the image sets . the paired images were reduced using the routines in the iraf ccdred package and stsdas wfpc , crrej , which did an excellent job of removing `` cosmic rays . '' the resulting , composite continuum image is shown as a logarithmic grayscale plot in figure1a . the continuum image was fit with the stsdas isophote routine to produce a model continuum galaxy , which was then subtracted from the h@xmath2 image , yielding the model - subtracted image shown ( as grayscale ) in figure1b . this technique leaves intact the continuum sources , such as stars and other galaxies , which are not directly associated with the extended continuum of the host galaxy , but improves visibility of faint emission features by minimizing subtraction - induced noise . a clearer picture of the relationship between the optical h@xmath2 image and the radio structure is shown in figure2 . figure 2a is a 20 cm vla radio continuum image ( @xcite ) , and figure2b is the continuum subtracted ( not model subtracted ) h@xmath2 image . we note that all of the other `` sources '' in the field subtract out in figure2b , indicating that they are either pure continuum sources or emission line sources at a redshift which differs from that of pks0349 - 27 by at least 1700kms@xmath0 . figure2c is a superposition of the h@xmath2 image on the 20 cm image with the galaxy nucleus marked as a cross . the host galaxy is clearly displaced towards the brighter ( ne ) radio lobe , which has a hot spot that is seen in the 4.89 ghz map published by @xcite . the faint arcs ( marked as _ f1 , f2 , _ and _ f3 _ in figure2b ) are also present in the [ oiii ] image of @xcite . the feature designated _ f2 _ corresponds to the region designated `` c '' by @xcite . we note that the features _ _ f2 _ appear to be parallel to the polarization vectors seen in the 20 cm vla map of @xcite . although this alignment may not be significant , it does suggest the possibility that the extended radio plasma may be interacting with the ionized gas in this region . three long slit spectra were taken during two consecutive nights in february 1987 with the ctio 4 m spectrograph and the ti ccd . all exposures were bracketed with spectra of a hearne arc lamp taken with the spectrograph in the same pa as the galaxy spectrum and the telescope at the sky positions corresponding to the start and end positions for the intervening galaxy spectrum . the slit positions and observing details are shown in lines 5 through 7 in table [ tbl-1 ] . in all cases , the scale along the slit was 0.73@xmath9 and the slit width was 180 @xmath10 m ( 1.2@xmath11 ) . slit positions , widths and coverage are shown superimposed on the continuum image in figure1a . calibration stars ( ltt1788 and ltt3864 ) were taken at the beginning and end of each night . the slit in pa 83@xmath5 covers the extended arm - like features seen in the host ( shown in figure1b ) but does not lie along the radio axis , which is , instead , aligned along the inner bar - like structure near the nucleus . the direction of the radio axis is indicated by the dark bar in pa 56@xmath5 in figure1a . at radial distances which lie between 7 and 14 kpc@xmath12 from the nucleus , the spectra for pa 83@xmath5 exhibit strong bifurcation in the spatially extended emission lines . an image of the stronger spectral features for the extended emission line regions in this pa is shown in figure3 . the extreme distortions show up in both the permitted lines of h@xmath2 ( figure3b ) and the forbidden lines of [ oiii ] and [ nii ] ( figure3a and b ) . knots labeled a and b in figure2b correspond to the bifurcated spectral lines in figure3 , which are marked a1 and b2 . these are found at velocities which are close to the nuclear velocity and hereafter will be designated `` lo - vel '' clouds . the regions in the spectral lines marked a2 and b1 are found at velocities with much higher amplitudes relative to the nucleus , and hereafter will be designated `` hi - vel '' clouds . since these velocity anomalies appear in both the `` red '' spectra as well as the `` green '' spectra which were taken on different nights , they are certainly real . the spectra were reduced using a set of interactive routines ( written by sms ) . they were flat fielded , calibrated in wavelength , calibrated for s - distortion , which was minimal , and rebinned to remove s - distortion and to produce constant wavelength increments . a sinc function was used for this rebinning . this is the proper interpolation function for this type of evenly sampled spectral data , since it mimics the slit function and has a modulation transfer function tailored to finite data arrays . the wavelength ranges and dispersions are given in lines 5 - 7 of table [ tbl-1 ] . the internal consistency in the fits to the calibration spectra was roughly 0.25 pixels , corresponding to @xmath13 for the @xmath14 4895 - 5745 spectrum and @xmath15 for the @xmath14 6653 - 7338 spectra . these translate into velocity errors of @xmath321kms@xmath0 and @xmath314kms@xmath0 respectively . emission line positions in the calibrated spectra were measured using an interactive routine which was adapted from analysis routines developed by k. ford at carnegie institution of washington . this routine finds the line peak , calculates the line centroid and the center of a gaussian fit to its profile after background subtraction , and automatically steps onto the next radial distance along the direction perpendicular to the slit until the selection criteria are no longer met . the routine allows interactive setting of selection criteria and baseline intervals . the velocities reported here are based on measurements which have line strengths in excess of 10 times the rms in the baseline background and which agree to within 0.5 pixels between emission line positions determined by the two positioning routines . while analyzing the data , it became clear that the emission lines closer to the nucleus were composed of multiple peaks , representing at least 2 separate velocity systems . these peaks were measured using line plots and an interactive cursor and are discussed in the next paragraph . the error for these latter measurements can not be determined but is clearly greater than that for wavelength calibration fits and probably greater than that for line fitting further out from the nucleus . the systemic velocity from these measurements , obtained from the strongest emission lines [ sii ] , [ nii ] , h@xmath2 , and [ oiii ] is 19846@xmath1672kms.@xmath0 this is in surprisingly good agreement with the value of 19846@xmath16130kms@xmath0 found by @xcite with lower resolution spectra . the error of @xmath1672kms,@xmath0 however , far exceeds the value of @xmath3 14 to 21 km@xmath17 for the calibration spectra and that of @xmath18 km@xmath17 observed for other emission line galaxies measured with the same equipment over two different seasons ( cf . inspection of the nuclear spectra shows that there are , in fact , at least two separate velocity systems ; at 19840 @xmath19 km@xmath17 and 19908 @xmath20 km@xmath17 ) and possibly a third , at 19708 km@xmath17 seen only in the [ sii ] lines . thus , the velocity measurements reported here have been zeroed to a nuclear redshift velocity of 19800kms@xmath0 for purposes of determining the internal kinematics . the measured velocities are plotted in figure4 . the error bars have been omitted in this plot to allow cleaner delineation of the different line species . the scatter in the nuclear velocity measurements noted above is clearly seen in figure4a and appears to be associated with the velocity bifurcations seen at larger radii . the separate velocity systems are also seen in the pa 146@xmath5 spectrum but are easily lost in the steep velocity gradient across the nucleus in that position angle ( figure4b ) . this suggests that the dispersion in cloud velocities seen in the nuclear spectrum in both position angles is more likely the result of limited spatial resolution than true radial superposition of differentially moving clouds . the bifurcated velocities for the clouds labeled a1 , a2 , b1 , and b2 in figure3a are marked in figure4a . pa 83@xmath5 was the only position angle which was observed in _ both _ the red and the blue - green spectral region . the line ratios from these spectra for the emission lines [ oiii ] 5007/h@xmath1 , [ sii ] 6716 + 6731/h@xmath2 , and [ nii ] 6583/h@xmath2 are shown in table [ tbl-2 ] . in most cases only lines with signal to noise ( s / n ) @xmath213 were measured . in a few cases this criterion was relaxed to s / n@xmath222 . all errors listed in table [ tbl-2 ] reflect 1@xmath23 fluctuations from the measured line strengths . we note , however , that an occasional single emission line measurement can have an erroneous value which is well outside this error limit if it has been contaminated by a weak `` cosmic ray '' hit . the [ oi ] 6300 emission line is seen not only in the nuclear spectrum of this object , but is also visible as an extended feature , with the same characteristic shape as the stronger emission lines shown in figure3 . although detectable as a continuous feature in video images of the spectrum ( and recorded by @xcite ) , this line is far too weak in our observed ccd spectrum to meet the criteria imposed on the measurements reported in table [ tbl-2 ] . although we observed flux calibration standards for all of these spectra , it is clear from examination and comparison of these calibration spectra from night to night that the nights were non - photometric , with flux variations of a factor of 2 or 3 from one night to the next and up to 1.5 on the same night . thus it is not possible to obtain meaningful absolute values for the emission line strengths nor is it possible to obtain meaningful flux ratios for emission lines , such as [ oi]@xmath246300 and [ oiii]]@xmath245007 , which were taken with different grating settings on different nights . we have , however , determined the _ relative _ corrections to the ratios in table [ tbl-2 ] using the published absolute flux values for the stars observed . even in the most extreme case of [ oi]/h@xmath2 these are less than @xmath160.04 in the log . since this is much less than the errors from `` noise '' for most of the observed emission lines , any error in the _ relative _ calibration will be dominated by the emission line measurement errors themselves . we reiterate here the important observational point made by @xcite : absolute flux measurements are very difficult to accomplish , while diagnostic line ratios for adjacent emission lines observed simultaneously can be very accurate . the measured line ratios in table [ tbl-2 ] are plotted in a series of `` bpt '' diagrams ( cf . @xcite , and @xcite ) , shown in figures5 , [ fig7 ] , and [ fig8 ] . these plots will be described and discussed in detail in section [ elr ] . in addition to the logarithmic line ratios in table [ tbl-2 ] , we have calculated the line ratios for the [ sii ] @xmath14 6716/6731 doublet which can be used to estimate the gas cloud densities . although the errors are substantial for any one value of this ratio , the general _ trends _ in this density indicator are significant . if the eelg is optically thin and has a constant temperature , a trend toward higher values of the line ratio implies a trend towards lower values in the gas density ( cf . however , @xcite have found that for clouds with stratified densities and temperatures this line ratio may not be useful as a density indicator . the [ sii ] line ratios are plotted , with low and high density limits marked , in figure6a and they are discussed in section [ sii ] below . as noted in section [ eelg ] there are several different scenarios involving the formation and fueling of radio galaxies which may be relevant to an interpretation of the eelg in this object . in broad outline they are : 1 . possible galaxy collisions , mergers , and tidal forcing of the extended gas by a nearby companion . 2 . ionization and excitation of the eelg by continuum radiation from the nuclear `` engine '' . 3 . interaction between a radio plasma jet and the interstellar medium in the host galaxy . all of these possibilities have been discussed for pks0349 - 27 in earlier papers ( cf . @xcite and @xcite ) . however , the present data , with their better spectral resolution , provide additional information with new implications for these various scenarios . there are several studies in the literature which support the idea that agns are fueled by infalling material which has been channeled into a galaxy s nuclear regions by a merger or other tidal disturbance ( cf . @xcite and references therein ) . however , the short time scales involved in this process and the fact that some mergers and tidal disturbances _ do not _ fuel agns make it difficult to assess the importance of this mechanism for radio galaxies . there are several known examples of candidates for galaxy mergers in non - radio galaxies which show that the extended ionized gas in these systems is photoionized by starburst activity ( @xcite , @xcite , @xcite , @xcite , @xcite ) . thus , starbursts may be one way of identifying merger activity in agn hosts . the long , two - armed structure of the ionized gas , shown in figures1b and [ fig2]b , and the [ oiii ] images of @xcite and @xcite , is similar to the type of structure associated with tidally interacting galaxies ( @xcite and references therein ) . the suggestion of tidal interaction is reinforced by the presence of a companion galaxy with a radial velocity within @xmath3200kms@xmath0 of the pks0349 - 27 host which is seen at the end of the object s eastern arm in figure1b . this was first identified by @xcite , who measured its redshift . if the structure of the eelg is determined by a tidal encounter or a merger , then the most likely geometry is one in which we are viewing the orbit of the merging companion nearly edge - on ( cf . such systems often have gas velocities which are bifurcated or counter rotational ( op cit . ) . the data plotted in figure4a show that the most notable split in velocities occurs for gas at r@xmath257kpc@xmath12 . this type of velocity detail was not identified in earlier measurements of this object ( @xcite ) , either because it is peculiar to this particular position angle , unresolved in the earlier spectra , or because the line intensities in the low velocity system are weaker than in the high velocity one . since the extreme velocity separation of @xmath31000kms@xmath0 between the systems in a1 and b2 is far in excess of any velocity amplitudes reported by either @xcite or @xcite for the gas at pa 90/270,@xmath5 it appears likely that these earlier data refer to gas clouds associated with those labeled a2 and b1 in figures3 and [ fig4 ] . because these two systems ( a2 and b1 ) are also stronger ( figure3 ) , the presence of the weaker systems ( a1 and b2 ) in other position angles can not be ruled out by these earlier measurements . finally , although the split in velocity systems is most notable for the knots labeled a and b in figure2b , the velocity points in figure4a as well as the discussion in section [ velval ] show that this separation in cloud velocities extends into regions closer to the nucleus , but with much smaller amplitude . thus , there is a strong possibility that _ both _ velocity systems are present throughout the host galaxy but the weaker one has been undetected until now . @xcite present a composite rotation - expansion model for the gas in the host of pks349 - 27 . although the `` rotation '' velocities in figure4b ( pa 146@xmath5 ) are roughly consistent with that earlier model , the bifurcated velocities shown in figure4a ( pa 83@xmath5 ) show that the situation is much more complex than simple rotation - expansion . as a first step in resolving this complexity , we believe it is important to try to determine which of the different velocity clouds are physically associated . the two most likely patterns for the gas velocities shown in figure4a are : ( 1 ) the clouds comprising sets a2 and b2 may be physically associated with a high redshift ( `` hi - z '' ) system with respect to the host nucleus ; while those in a1 and b1 comprise a low redshift ( `` lo - z '' ) system . ( 2 ) the extreme velocity clouds for the set a1 and b2 may be physically associated with a system which has a high rotational velocity relative to the host galaxy nucleus ( `` hi - vel '' clouds ) , while the clouds ( a2 and b1 ) , comprise a separate , independent system with relatively low rotational velocities ( `` lo - vel '' clouds ) . we have used the bpt plots of measured emission line ratios in figures5 , [ fig7 ] , and [ fig8 ] to try to delineate the relationship between the different cloud systems . the solid curves in each diagram in figures5 through [ fig8 ] are taken from @xcite , and mark the empirical division between hii - region like objects , which lie to the left hand side of the line , and objects , such as agns , ionized by a harder radiation spectrum , which lie to the right hand side of the line . first , in an attempt to identify any gross differences in the physical conditions between the eelg , we have separated the emission line measurements into two different sets according to radial distance from the nucleus . those with projected distances @xmath257 kpc@xmath12 from the nucleus ( figures5a and b ) are plotted with the `` hi - vel '' clouds ( a1 and b2 ) as filled symbols and the `` lo - vel '' clouds as open symbols . those with projected distances @xmath267 kpc@xmath12 from the nucleus ( figures5c and d ) , are plotted with points to the west of the nucleus marked as filled symbols and points to the east of the nucleus marked as open symbols . figure5 allows us to immediately test the hypothesis that one of the velocity systems arises from gas which has been disturbed ( either by the radio plasma or as infall from a merging galaxy ) and the other originates in a more `` normal , '' gaseous component , ionized by recent starbursts . the results of this test are clear : in figure5 _ both _ systems exhibit the high excitation conditions which are the signature of ionization by a strong uv continuum or by shock heating . although there is more scatter in the points from regions @xmath257 kpc@xmath12 from the nucleus , all fill a similar region of the bpt diagrams , and all lie to the left ( the agn region ) of the dividing line between agn spectra and hii regions . thus our first conclusion from this data is that there is no detectable starburst activity in the eelg . the diagrams in figure5 show no significant change in the line ratios as a function of distance . this is in general agreement with the conclusions of @xcite and @xcite that there is no appreciable change in the ionization parameter , @xmath27 , of the eelg with radial distance from the central nuclear source . the conclusion they drew from this fact was that the gas density must decrease with radial distance . ( since @xmath28 , where @xmath29 is the nuclear ionizing photon luminosity , @xmath30 the radial distance from the nucleus , and @xmath31 , the hydrogen number density , then if @xmath27 is constant @xmath31 must decrease . ) however , a closer look at the bpt diagrams in figure5 and the [ sii ] plots in figure6a for points @xmath267kpc@xmath32 shows that the situation is more complex than this . the trend toward _ increasing _ density toward the east ( seen in figure6a ) does not fit into this picture . an increased density should lead to a _ decreasing _ ionization parameter ( decreasing [ oiii]/h@xmath1 ) in this direction , but the opposite is seen in figure5c and d. on the other hand , the clouds on the _ western _ side of the nucleus do seem to fit the central source picture . they tend to have lower ionization ( lower [ oiii]/h@xmath1 ) than those on the eastern side , and the increase in the [ sii ] ratios indicates that the clouds to the west also have decreasing density , balancing out the increase in @xmath30 . to help clarify this , the parameter which is most closely correlated with ionization ( [ oiii]/h@xmath1 , cf . @xcite ) , is plotted as a function of radial distance in figure6b . comparison of figures6a and b in the range -7@xmath337 shows that the trends predicted by a simple nuclear source photoionization model do not fit the observed density trends . these trends are significant within the measured errors and also consistent with the h@xmath2 image of the gas ( figures1b and [ fig2]b ) , which is fainter in the region to the west of the nucleus , showing that the gas there has a lower emission measure than the gas on the eastern side . we must conclude that either the gas on the western side of the eelg has @xmath34 with the number of ionizing photons falling off as @xmath35 while the eastern side is somehow receiving more ionizing photons at greater distances from the nucleus , or the eastern eelg clouds have an additional source of ionization / excitation which supplements the nuclear source . however , one interesting insight into the physical state of the eelg at @xmath367kpc@xmath37 can be obtained from the simple central source model . figure6 shows that the lo - z system has lower densities and a higher @xmath27 ( higher [ oiii]:h@xmath1 ) than the hi - z system . if _ the eelg is ionized by a central source , the relationship between @xmath27 and @xmath38 ( given above ) suggests that the two sets of clouds may well be at the same physical radius as well as the same projected radius from the nucleus . while the above discussion is consistent with conclusions of @xcite , @xcite , and @xcite , where it was noted that different measured line ratios in pks0349 - 27 were inconsistent with a global picture of ionization / excitation in the eelg based on photoionization by the ( hidden ) nuclear continuum , our findings are much more detailed and direct . the radial distributions of density and ionization reported here , coupled with the different characteristics of the two velocity systems seen in pa 83,@xmath5 strongly suggest that whatever the final interpretation of this object , it must take the velocity systems into account . in summary , these discrepancies in applying a central source model to the observed conditions in this galaxy s eelg when combined with the observation by @xcite that its nucleus has little or no observed uv flux , implies that the source of at least some of the ionization / excitation observed in the extended gas has a non - nuclear origin . in an attempt to identify common factors amongst gas clouds with @xmath367kpc@xmath32 we have sorted them into the different velocity sets defined in section [ velkin ] ( hi - lo - vel and hi - lo - z ) and plotted the bpt diagrams shown in figures7 and [ fig8 ] . the plots with the most overlap ( figure8 ) seem to be the ones in which the line ratios for the lo - z ( a1 and b1 ) clouds are plotted together ( figure8c and d ) . almost all of these have values of [ oiii]/h@xmath1@xmath3910 . the hi - z clouds ( a2 and b2 , figure8a and b ) also show good overlap and a slightly lower [ oiii]/h@xmath1 value ( @xmath38 ) but with more scatter . what this means physically is that the gas in the lo - z clouds has a slightly higher ionization / excitation state than that in the hi - z clouds . this is consistent with the [ sii ] line ratio plots in figure6a which indicate that the lo - z clouds have consistently lower densities , while the hi - z clouds show more scatter ( see section [ sii ] above ) . if these similarities in ionization and density imply physical association , then the velocities in figure4a can be reconciled in the following way : rotating the curves by 180@xmath5 and shifting them in the radial direction ( x - axis ) by @xmath32.9 kpc shows that the hook in the velocity curve at a1 and the flat region at b1 appear to be mirror images of the upper branch in the velocities at b2 and the flat part of the velocity plot at a2 . however , this transformation does not fit the velocity points traced by the [ sii ] measurements , which we noted in section [ velval ] appeared to have three velocity components near the nucleus . these [ sii ] velocities seem to trace a system with a slightly negative velocity slope near the nucleus . both of these features , asymmetrically shifted velocity curves which are mirror images , and small , inner systems which show counter rotation , are found in merging systems and tend to support the `` merger '' scenario for the origins of the gas ( cf . @xcite and references therein ) . although the symmetric velocity structure described in section [ sys ] above may be consistent with a tidal - interaction / merger , the magnitude of the velocity separations between the two cloud systems is also consistent with the pattern of disruption caused by the passage of a radio jet.pa 56@xmath5 in the original identification images by @xcite is most likely the inner part of the extended ionized gas seen in figure1b . ] the time scale for a disturbance in the gas to propagate from the location of the radio axis ( at pa56@xmath5 ) to the observed position in the ionized gas at pa83@xmath5 when the gas is moving at a relative velocity of @xmath3350kms,@xmath0 is @xmath310@xmath40 years . this is comparable to the age of this type of frii source ( 6 - 30@xmath41an , @xcite ) and since the observed radial velocity differences between regions b1 and b2 and a1 and a2 exceed @xmath3400kms,@xmath0 at points , we can not completely rule out collisional ionization of the eelg by the material ejected from the nucleus in a radio jet . however , there is some evidence that this type of jet - cloud collision generates ionization / excitation conditions which differ from both those for agn - like gas and hii regions ( @xcite ) . in the few clear examples of an immediate jet - cloud collision ( ngc7385 , @xcite , and pictor a , @xcite ) , the signature of such jet - ionized gas appears to be relatively low excitation ( with a value of [ oiii]/h@xmath424 ) along with agn - like values of [ sii]/h@xmath2 and [ oi]/h@xmath2 but relatively lower values of [ nii]/h@xmath2 ( similar to those found for hii regions ) . the observed line ratios in table [ tbl-2 ] lie outside these ranges . this does not completely rule out the possibility that a past disturbance by a radio jet has ionized the gas in pks0349 - 27 , but the recombination times for gas at the densities indicated by the [ sii ] line ratios in figure6a will be shorter than the propagation time scale noted above and thus an additional source for the present ionization is still necessary . although the material reported here provides a much more detailed picture of the ionized gas in pks0349 - 27 than previous observations , it raises several new questions . on the positive side , these measurements point up the importance of the greater accuracy possible when diagnostic line ratios are analyzed in terms of the bpt scheme proposed by @xcite . however , in spite of the observational advantages of this system , the discussions in sections [ sii ] and [ elr ] make it clear that it is difficult to compare the line ratio diagrams discussed here with either models or data reported in the more common systems which rely on [ oi]:][oiii ] or [ oii]:[oiii ] criteria for ionization / excitation determinations . hopefully , there will be soon more detailed theoretical models which include published line ratios calculated for the @xcite system . 1 . the ionization of the eelg in pks0349 - 27 does not come from starburst activity . 2 . the bifurcated velocity systems in pa83@xmath5 of this system appear to have similar emission line ratios ( and thus similar ionization parameters ) when grouped into hi - z and lo - z systems . 3 . there may be yet a third velocity system near the nucleus of this system which is in counter rotation . 4 . the gas clouds at projected distances closer than 7kpc@xmath43 to the nucleus of pks0349 - 27 have line ratios which are inconsistent with a simple central source ionization model and ( particularly on the eastern side ) require either an additional ionization source or a completely different mechanism which is unrelated to the nuclear source . the most likely conclusion from these facts is that the gas has been disturbed by a merger or tidal interaction with a nearby galaxy , but the possibility of disruption by a radio jet can not be completely ruled out . 1 . what is the detailed velocity field of the eelg in pks0349 - 27 ? is the weaker , high velocity system only seen in pa83,@xmath5 or does it extend into all areas around the nucleus ? 2 . what is the source of the `` additional '' ionization / excitation on the e side of the host how can we distinguish between shock heating due to the merger of two galaxies and that caused by the passage of a radio jet ? 3 . what is the detailed radio structure ( at arc second resolution ) op f pks0349 - 27 ? are the ionized gas filaments seen to the east of the nucleus interacting with the radio plasma ? if so , what are the implications for the ionization / excitation of the eelg ? these questions underscore the need for new high resolution studies , with improved equipment ( both radio and optical ) of not only this object but also other nearby , frii , radio galaxies for comparison . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . the ctio observations were done with some support from nsf ast-89 - 14567 and publication costs from a grant ( to sms ) from nasa , administered by the american astronomical society . the graphics in this paper were generated with the software package @xmath44 is copyright by the berkeley - illinois - maryland association ( bima ) project ( @xcite ) ] , using the @xmath45 graphics library . allen , m. , g. , dopita , m. , a. , tsvetanov , z. , i. , and southerland , r. , s. , 1998 , , 493 , 571 arnaboldi , m. ; oosterloo , t. ; combes , f. ; freeman , k. c. ; koribalski , b. , 1997 , , 113 , 585 baldwin , j.a . , phillips , m.m . , and terlevich , r . , 1981 , , 93 , 5 baum , s. a. , heckman , t. m. , bridle , a. , van breugel , w. , miley , g. k. , 1988 , , 68 , 643 baum , s. a. , and heckman , t. , 1989 , , 336 , 681 baum , s. a. , and heckman , t. , 1989 , , 336 , 702 baum , s. a. , heckman , t. m. , and van breugel , w. , 1992 , , 389 , 208 best , p. n. ; longair , m. s. ; rottgering , h. j. a. , 1996 , , 280 , l9 binette , l.,wilson , a. s. , storchi - bergmann , t. , 1996 , , 312 , 365 bolton , j. g. , clarke , m. e. , and ekers , r. d , 1965 , aust . j. ph . , 18 , 627 byrd , g. , g. and klari , m. , 1990 , , 99 1461 carilli , c. l. ; van gorkom , j. h. ; stocke , john t. , 1989 , natur . 338 , 134 carilli , c. l. , perley , r. a. , dreher , j. w. , and leahy , j. p. 1991 , , 383 , 554 carilli , c. l. ; van gorkom , j. h. ; stocke , john t. van gorkom , j. h. ; schweizer , francois 1994 , , 107 , 67 christiansen , w. n. , frater , r. f. , and watkinson , et al . , 1977 , , 181 , 183 danziger , i. j. , fosbury , r. a. , goss , w. m. , bland j. and boksenberg , a. , 1984 , , 208 , 589 ekers , r. d. , wall , j. v. , shaver , w. m. , goss , w. m. , fosbury , r. a. , danziger , i. j. , moorwood , a. f. m. , malin , d. f. , monk , a. s. , and ekers , j. a. , 1989 , , 236 , 737 ferguson , j. , w. , korista k. , t. , baldwin j. , a. , and ferland g. , j. 1997 , , 287 , 122 hansen , l. , jorgensen , h. e. , norgaard - nielsen , h. u. , 1987 , , 71 , 465 hibbard , j.e . , guhathakurta , p. , van gorkom , j.h . , and schweizer , f. , 1996 , , 107 , 67 hibbard , j.e . and van gorkom , j.h . , 1996 , , 111 , 655 koekemoer , a. m. , bicknell , g. v. , 1998 , , 497 , 662 mackenty , j. w. 1986 , , 308 , 571 mackenty , j.w . , simkin , s.m . , griffith , r.e . , ulvestad , j.s . , and wilson , a.s . 1994 , , 435 , 71 mccarthy , patrick j. ; spinrad , hyron , van breugel , wil , 1995 , , 99 , 27 mccarthy , patrick j. ; baum , stefi a. ; spinrad , hyron , 1996 , , 106 , 281 morgan , j. a. 1995 , in `` astronomical data analysis software and systems iv '' , pasp conf series 77 , editors r. a. shaw , h. e. payne , and j. j. e. hayes , p. 129 osterbrock d.,1989 , in `` astrophysics of gaseous nebulae and active galactic nuclei '' , ( university science books , mill valley , ca ) , p. 134 perley , r.a . , rser , h .- j . , and meisenheimer , k. 1997 , , 328 , 12 robinson , a. , binette , l. , fosbury , r. a.,e . , and tadhunter , c. n. , 1987 , , 227 , 97 searle , l. , and bolton , j. g. , 1968 , , 154 , 101 simkin , s. m. , 1976 , , 204 , 251 simkin , s. m. , bicknell , g. v. , and bosma , a. , 1984 , , 277 , 513 simkin , s. m. and callcut , j. , 1996 , iau symposium number 175 , editors r. ekers , c. fanti , l. padrirlli , ( kluwer academic pub . ) , p.183 simkin , s. m. and sadler , e. m. , 1989 , iau symposium 134 , editors d. osterbrock and j. miller , ( kluwer , dordrecht , the netherlands ) , p. 484 simkin , s. m. , sadler , e . m. , sault , r. , tingay , s. j. , and callcut , j. , 1999 , ( submitted , ms 39342 ) simkin , s. m. , grimberg , b.i . and sadler , e. m. , 1999 , , ( in preparation ) storchi - bergmann , t. , wilson , a. , s. , mulchaey , j. , s. , and binette , l. , 1996 , , 312 , 357 sulentic , j. w. ; marziani , p. ; zwitter , t. ; calvani , m. , 1995 , , 438 , l1 tadhunter , c. n. , fosbury , r. a. e. , and quinn , p. j. , 1989 , , 240 , 225 tadhunter , c. n. , morganti , r. , robinson , a. , dickson , r. , villar - martin , m. , fosbury , r. a. e. , 1998 , , 298 , 1035 toomre , a . , 1977 , in the evolution of galaxies and stellar populations , edited by b. m. tinsley and r. b. larson ( yale university observatory , new haven ) , p.401 van breugel , wh . , filippenko , a. , v.,heckman , t. , and miley , g. , 1985 , , 293 , 83 viegas , s.m . de gouveia dal pino , e.m . , 1992 , , 384 , 467 veilleux , s. , and osterbrock , d. , 1987 , , 63 , 295 weaver , k. a. ; wilson , a. s. ; baldwin , j. a. , 1991 , , 366 , 50 wilson , a. s. , binette , l. , and storchi - bergmann , t. , 1997 , , 482 , l131 zirbel , esther l. , and baum , stefi a. , 1998 , 114 , 177 ctio-1.5m&22/23 jan 1988 & 4000 s&1.02 & 7007&78-line & [email protected]&22/23 jan 1988 & 4000 s&1.12 & 6560&110-cont & [email protected]&23/24 jan 1988 & 4000 s&1.03 & 7007&78-line & [email protected]&23/24 jan 1988 & 4000 s&1.42 & 6560&110-cont & 1.78@xmath11ctio-4m - pa 146@xmath5&15/16 jan 1988 & 2000 s&1.30 & 6653 - 7338 &1.20 / px & 1.43@xmath11 ctio-4m - pa 83.1@xmath5&15/16 jan 1988 & 2000 s&1.56 & 6653 - 7338 &1.20 / px & 1.52@xmath11 ctio-4m - pa 83.1@xmath5 & 16/17 jan 1988 & 2000 s&1.19 & 4895 - 5745 &1.49 / px & 1.67@xmath11 & a1 and b2 & & & a2 and b1 & & -14.72 & 0@xmath46&-0.02@xmath47&0.45@xmath48 & -0.46@xmath49&-0.53@xmath50&0.84@xmath51 -14.02&&&&-0.32@xmath52&-0.44@xmath51&1.39@xmath53 -13.32&-0.17@xmath48&-0.12@xmath54&1.05@xmath55 & & & -12.61&-0.21@xmath56&-0.31@xmath57&1.26@xmath53 & -0.34@xmath58&-0.51@xmath59&1.45@xmath60 -11.91&-0.14@xmath48&-0.17@xmath61&0.91@xmath62 & -0.1@xmath63&-0.51@xmath64&0.79@xmath65 -11.21&-0.18@xmath51&-0.28@xmath66&0.97@xmath67 & -0.38@xmath68&-0.2@xmath69&1.01@xmath70 -10.51&-0.28@xmath51&-0.31@xmath51&1.12@xmath62 & -0.03@xmath56&-0.07@xmath63&0.75@xmath71 -9.81&-0.21@xmath72&-0.25@xmath52&1.12@xmath55 & -0.24@xmath73&-0.25@xmath73&0.79@xmath74 -9.11&&&&0.03@xmath75&0.04@xmath76&0.76@xmath77 -8.41&0.06@xmath78&-0.23@xmath79&-0.02@xmath80 1&-0.35@xmath61&-0.26@xmath48&1.57@xmath60 -7.71&-0.15@xmath63&-0.24@xmath81&1.18@xmath53 & & & -7.01&-0.26@xmath82&-0.35@xmath58&1.02@xmath83 & & & -6.31&-0.32@xmath73&0.04@xmath48&1.06@xmath77 & & & -4.91&-0.61@xmath84&-0.4@xmath85&0.91@xmath55 & & & -4.2&-0.39@xmath86&-0.12@xmath56&1.16@xmath87 & & & -3.5&-0.45@xmath88&-0.09@xmath89&0.97@xmath65 & & & -2.8&-0.09@xmath48&-0.03@xmath66&1.19@xmath70 & & & -2.1&-0.27@xmath61&-0.21@xmath48&0.85@xmath90 & & & -1.4&-0.19@xmath52&-0.16@xmath52&1.07@xmath70 & & & -0.7&-0.05@xmath91&-0.09@xmath91&0.85@xmath92 & & & 0 & -0.05@xmath93&-0.06@xmath94 & 0.87@xmath95 & & & 0.7&-0.03@xmath91&0.01@xmath91&0.91@xmath51 & & & 1.4&-0.02@xmath91&0.12@xmath91&0.74@xmath67 & & & 2.1&0.02@xmath52&0.21@xmath92&0.51@xmath66 & & & 6.31&0.08@xmath96&0.25@xmath97&0.45@xmath98 & & & 3.5&-0.27@xmath99&0.31@xmath56&1.03@xmath100 & & & 4.2&-0.22@xmath101&0.18@xmath89&0.55@xmath48 & & & 7.01&-0.07@xmath102&0.06@xmath103&0.34@xmath104 & & & 7.71&-0.39@xmath105&-0.12@xmath106&0.73@xmath74 & & & 8.41&-0.57@xmath107&-0.7@xmath108&0.64@xmath109 & & & 9.11&-0.1@xmath80&-0.52@xmath110&1.68@xmath111 & -0.55@xmath63&-0.59@xmath107&1.26@xmath112 9.81&-0.05@xmath113&-0.3@xmath106&0.98@xmath114 & -0.68@xmath54&-0.62@xmath66&1.27@xmath115 10.51&-0.3@xmath116&-0.21@xmath88&0.79@xmath117 & -0.43@xmath118&-0.5@xmath54&1.02@xmath92 11.21&-0.2@xmath89&-0.41@xmath107&0.71@xmath119 & -0.49@xmath52&-0.5@xmath52&1.08@xmath67 11.91&-0.21@xmath118&-0.76@xmath120&0.69@xmath112 & -0.35@xmath52&-0.31@xmath52&1.24@xmath62 12.61&&&&-0.29@xmath121 & 0@xmath122&0.99@xmath70	pks0349 - 27 is a classical frii radio galaxy with an agn host which has a spectacular , spiral - like structure in its extended emission line gas ( eelg ) . we have measured the velocity field in this gas and find that it splits into 2 cloud groups separated by radial velocities which at some points approach 400 kms.@xmath0 measurements of the diagnostic emission line ratios [ oiii]5007/h@xmath1 , [ sii ] 6716 + 6731/h@xmath2 , and [ nii]6583/h@xmath2 in these clouds show no evidence for the type of hii region emission associated with starburst activity in either velocity system . the measured emission line ratios are similar to those found in the nuclei of narrow - line radio galaxies , but the extended ionization / excitation can not be produced by continuum emission from the active nucleus alone . we present arguments which suggest that the velocity disturbances seen in the eelg are most likely the result of a galaxy - galaxy collision or merger but can not completely rule out the possibility that the gas has been disrupted by the passage of a radio jet .
the density functional theory ( dft ) has a important problem with the calculation of transition metal complexes . in spite of the recent development of the exchange - correlation functionals , it is difficult to evaluate the multiplet energies using the kohn - sham method , because these electronic structures are not always written in a single slater determinant . several methods have been proposed to calculate the multiplet energies using dft . one of these methods has been proposed by ziegler @xcite , wood @xcite and von barth @xcite . they have carried out the computation of the multiplet energies using the diagonal sum rule , in which the sum of the multiplet energies is equal to the corresponding sum of the single determinant energies . the method reproduces the multiplet energies of p electron system , but the method does not give correct multiplet energies for the d electron system . @xcite the time dependent density functional theory ( tddft ) has been recently developed to calculate excited states including the multiplet states . the method has been applied to p electron systems @xcite within the adiabatic approximation . the non - adiabatic tddft calculation , in which the exchange - correlation kernel becomes frequency dependent , is still difficult . another method was formally discussed in the grling s paper @xcite in which the energy functional depends on the multiplet states . they applied the optimized effective potential method to the calculation of the multiplet energies of p electron atoms . @xcite the optimized effective potential method ( opm ) was firstly proposed by sharp and horton@xcite . the method was applied to the neon and carbon atoms by talman et . grling pointed out that opm is formally regarded as dft because the orbital dependent energy functional implicitly depends on the density @xcite . the opm potential can be evaluated by the singular integral equation called as the opm equation . talman et . al.@xcite and engel et . al.@xcite solved the one - dimensional opm equation for an atom , in which the angular components of the opm equation for an atom are integrated out using the spherical symmetry . to efficiently calculate the opm potentials for atoms and molecules , grling et . al . expanded the opm potential using some base functions . @xcite the base expansion method gives the same total energy as calculated by talman s method . however , in the base expansion method , the opm potential shows spurious oscillations . to solve the problem , some methods have been proposed but still are not applied to d electron atoms . @xcite an early work of opm for the multiplet states is the opm calculation by aashamar @xcite who applied opm to the ground multiplet of the atom of which atomic number is less than 85 . the opm potentials for the p electron ground and excited state multiplets are calculated by nagy @xcite using the kli method.@xcite furthermore , aashamar also applied opm to the multiplet energies of p electron atoms using the multi - configuration hf form . @xcite the total energies of the multiplet states have been discussed in these studies , but the potentials of the multiplet states are not presented . sala and grling have been developed the lhf method to approximately calculate the exchange only opm potential . @xcite recently , the lhf method is extended to the open shell atoms and molecules.@xcite they applied the lhf method to the ground and excited multiplets of p electron atoms and molecules . however , in these papers , they did not treat the excited multiplet energies of the d electron atoms , which are difficult to calculate using the dft . furthermore , these studies give incorrect result caused by the approximation methods such as the kli or the lhf method . in the present paper , we apply opm to the ground and excited state multiplets of the 3d@xmath0 electron atoms using the total energy functional of the single - configuration hf @xcite form . following talman and engel@xcite , we solve the one - dimensional opm equation correctly . the solution is considered to give the most reliable result among all the available methods . we show that the differences between the opm potentials of the multiplet states can be approximated by the linear combination of the potentials derived from @xmath1,@xmath2 slater integrals . we numerically show that the opm exchange potential of multiplet state is approximated by the linear combination of the opm exchange potentials of the single determinants . in the present paper , we calculate the opm potential for eigenstates of an transition metal atom . each state is characterized by a definite value of the total orbital angular momentum @xmath3 and of the total spin angular momentum @xmath4 . these multiplet energies can be expressed as @xmath5 , where @xmath6 is the seniority number . the multiplet energy level is degenerate for possible directions of total momentums @xmath3 and @xmath4 . then , each multiplet wavefunction @xmath7 can be written as a linear combination of single slater determinant functions @xmath8 which are not always eigenfunctions of the atom . @xmath9 and @xmath10 are @xmath11 component of the total momentums @xmath3 and @xmath4 , respectively , and @xmath12 specifies different states with the same @xmath9 and @xmath10 component . each total energy of the single determinant is represented as @xmath13 . furthermore , we consider `` average energy of the configuration '' @xmath14 defined by slater . @xcite for the 3d@xmath0 multiplets , @xmath14 is expressed as @xmath15 to generally express these three kind of energies , @xmath5,@xmath13 and @xmath14 , we introduce @xmath16 where @xmath17 stands for @xmath18,@xmath19 and @xmath20 . the total energy @xmath16 is divided into the 1-body part @xmath21 and the 2-body part @xmath22 ; @xmath23 the 1-body part is constructed from the kinetic energy of electrons and the electron - nucleus coulomb energy . @xmath24 where @xmath25 represents the number of the electrons in the orbital @xmath26 . @xmath27 is the atomic number . @xmath28 and @xmath29 are the principal quantum number and the angular quantum number , respectively . in this study , we suppose that the opm potential @xmath30 is spherically symmetric . the single electron orbital is determined by the following equation @xmath31 where @xmath32 is the single electron energy . @xmath26 is the product of the distance from the nucleus @xmath33 and the radial component of the single electron wave function . the 2-body part @xmath22 is composed of the electron - electron interaction energy expressed as @xmath34 where the coefficients @xmath35 and @xmath36 depend on the multiplet states , the single determinants or the average energy of the configuration . the slater integrals @xmath37 and @xmath38 are defined as @xmath39 where @xmath40 is the smaller of @xmath33 , @xmath41 and @xmath42 is the larger . the opm potential @xmath43 is determined by requiring that @xmath16 be minimized for all @xmath44 obtained from equation ( [ p - equation ] ) . this results in @xmath45 where @xmath46 using equations ( [ total_energy]),([p - equation]),([oepeq - org2 ] ) , and the variation of the normalization condition @xmath47 with respect to @xmath48 , that is , @xmath49 , we can rearrange equation ( [ oepeq - org1 ] ) as @xmath50 the functional derivative @xmath51 appeared in equation ( [ oepeq - org1 ] ) is calculated using the perturbation theory : @xmath52 the functional derivative @xmath53 in equation ( [ oepeq - org3 ] ) is calculated as @xmath54 where @xmath55 is defined as @xmath56 . substituting equations ( [ oepeq - org4 ] ) and ( [ oepeq - org5 ] ) into equation ( [ oepeq - org3 ] ) , we obtain @xmath57 equation ( [ oep - equation ] ) is called as the opm equation . the opm potential can be obtained by self - consistently solving the opm equation and the single electron equation ( [ p - equation ] ) . to solve the opm equation ( [ oep - equation ] ) , we modified the code developed by fischer et . for the calculation of the green s function , we use the following expression @xmath58 where @xmath59 is the second solution of equation ( [ p - equation ] ) satisfying @xmath60 , and @xmath61 with @xmath62 . to evaluate @xmath26 and @xmath59 , we solve the single electron equation ( [ p - equation ] ) using the numerov s method with the transformation @xmath63 and @xmath64 . the lower bound of @xmath65 is set to -6.0 and the upper bound , which depends on the kind of atom , is set around 7 . the step size @xmath66 is @xmath67 . the integration in the opm equation is approximated as the sum of the product of integrand and step size . using this simple approximation with the step size @xmath68 , the opm equation is approximated to the set of linear equations for the opm potential . the opm equation ( [ oep - equation ] ) determines the solution @xmath43 only up to a constant . the constant is fixed by the physical requirement @xmath69 . in our program , we fix the constant using the asymptotic form of the opm potential @xmath70 at @xmath71 au , where @xmath72 is the index for homo . the self - consistent procedure converges satisfactorily in about 12 iterations if the average of the initial and final @xmath73 is taken at each iteration . the scf iteration has converged when @xmath74 , where @xmath75 is the number of the occupied orbitals . @xcite in opm , the single electron energy of homo @xmath76 is equal to the hf single electron expectation value for homo @xmath77@xcite . in our results of the calculations for the 3d transition metal atoms , the differences between the single particle energies ( @xmath78 ) are less than 0.005 au . the opm exchange potential @xmath79 is defined by subtracting spherical average of the coulomb potential from the opm potential @xmath80 : @xmath81 where the radial density @xmath82 is defined as @xmath83 firstly , we apply opm to the average energy @xmath14 defined by equation ( [ average_energy ] ) . the average opm potential @xmath84 and average single electron orbital @xmath85 are obtained by the self - consistent solution of the opm equation ( [ oep - equation ] ) and the single electron equation ( [ p - equation ] ) . the average radial density @xmath86 is calculated using equation ( [ density - def ] ) . in dft , the exchange energy is expressed as @xmath87 $ ] where @xmath88 is an electron density . to calculate averaged dft potential , we used spherical averaged density @xmath89 denoted as @xmath90 . then , the average energy for dft is defined as @xmath91.\end{aligned}\ ] ] the average density is calculated by the self - consistent solution of the kohn - sham equation with spherical exchange potential @xmath92}{\delta n^{\rm av}(\bm{r})}$ ] . in the bottom part of figure [ dftpot_mn ] , we show the x@xmath93 exchange potential ( @xmath94 ) , the becke s gga ( b88 ) exchange potential @xcite , and the opm exchange potential @xmath95 of mn@xmath96 . the opm exchange potential is close to the x@xmath93 and b88 exchange potential . the b88 exchange potential diverges at the nucleus , whereas the opm exchange potential does not diverge . the radial density @xmath86 is shown in the top part of figure [ dftpot_mn ] . there are some kinks in the opm exchange potential . the positions of the kinks correspond with the positions of troughs in the radial density . the kinks in the b88 exchange potential are smoother than that in the opm exchange potential . for the x@xmath93 potential , the kinks are not appeared . in table [ eng_aoc ] , we show the average total energies ( equation ( [ average_energy ] ) ) using hf method and opm . the average total energy of dft ( equation ( [ av - eng - dft ] ) ) using x@xmath93 and becke s gga ( b88 ) exchange functional is also tabulated in table [ eng_aoc ] . the total energy of the opm method is more closer to the hf total energy than that of the b88 and x@xmath93 . the difference between the hf and opm energies is order of 0.005 au throughout the transition metals . secondly , we discuss the opm exchange potentials for the 3d@xmath0 multiplet states . the 2-body part of the total energy of an atom ( equation ( [ eng_2body ] ) ) can be represented as the linear combination of @xmath37 and @xmath38 . for the 3d@xmath0 multiplet states , the coefficients @xmath97 and @xmath98 depend on the multiplet states , while the other coefficients do not . therefore , the 2-body part of the total energy is expressed as @xmath99 where @xmath100 is the 3d@xmath0 multiplet independent part of 2-body energy . since @xmath22 , @xmath37 and @xmath100 are functionals of density @xmath101 , we can define the potential for @xmath22 , @xmath37 and @xmath100 as @xmath102 , @xmath103 and @xmath104 , respectively . in the following , we derive the equation to determine these potentials . we define @xmath105 $ ] as a general expression of @xmath22 , @xmath37 and @xmath100 . the corresponding potential is defined as @xmath106 . from equation ( [ p - equation ] ) , the single electron orbital @xmath44 is considered to be a functional of @xmath107 . furthermore , we regard @xmath107 as a functional of @xmath108 . following chain rule of functional derivative , we get @xmath109}{\delta p^\xi_{n_i l_i}(r ' ) } \frac{\delta p^\xi_{n_i l_i}(r')}{\delta v^\xi_{\rm opm}(r '' ) } \frac{\delta v^\xi_{\rm opm}(r'')}{\delta n^\xi(r)}.\end{aligned}\ ] ] multiplying the both sides of equation ( [ goep - tmp1 ] ) by @xmath110 , and integrating the right - hand side over @xmath111 and @xmath112 , we obtain @xmath113}{\delta p^\xi_{n_i l_i}(r ' ) } \frac{\delta p^\xi_{n_i l_i}(r')}{\delta v^\xi_{\rm opm}(r''')},\end{aligned}\ ] ] where we used the following relation : @xmath114 . if we substitute @xmath22 for @xmath105 $ ] in equation ( [ goep - eq ] ) , we get the opm equation ( [ oepeq - org3 ] ) . furthermore , if we substitute @xmath115 and @xmath100 for @xmath105 $ ] , equation ( [ goep - eq ] ) defines the potential @xmath116 and @xmath117 , respectively . since equation ( [ goep - eq ] ) defines @xmath118 only up to a constant , we set the boundary condition @xmath119 . equation ( [ goep - eq ] ) is solved using the method described in section [ cal - method ] . functional derivative of equation ( [ eng_2body_sep ] ) with respect to @xmath120 leads the decomposition of the opm potential for the multiplet state @xmath121 ; @xmath122 in the top part of figure [ dens_v ] , we show the radial density @xmath123 of each multiplet states of v@xmath96 . in the bottom part of figure [ dens_v ] , we display the difference of the radial density @xmath124 . comparing the top and the bottom parts of figure [ dens_v ] , we find that the difference between the radial densities of the multiplet states are much smaller than the radial densities . so we neglect the difference of the radial components @xmath125 between the average configuration and the multiplet states , and we substitute @xmath126 and @xmath127 for @xmath128 and @xmath129 in equation ( [ sep - v ] ) , respectively . the difference between the opm exchange potential for the multiplet states @xmath130 and that for the average energy @xmath95 can be approximated as a linear combination of @xmath131 and @xmath132 : @xmath133 neglecting the multiplet dependence of the orbital , we can replace the orbital @xmath134 in equation ( [ eng_1body_sub ] ) and ( [ eng_2body_sep ] ) to @xmath135 . then , the 1-body and 2-body energies are approximated as @xmath136 in order to discuss the equality of equations ( [ ex - lc ] ) and ( [ eng_2body_sep_av ] ) , we compare the approximate opm exchange potentials of v@xmath96 multiplets with exact ones . the dotted line of figure [ f2f4lc_v ] exhibits the difference between the opm exchange potential of the multiplet states and that of the average energy calculated from equation ( [ ex - lc ] ) . in the solid line of figure [ f2f4lc_v ] , we show the same difference calculated from exact equation ( [ sep - v ] ) . the dotted line of figure [ f2f4lc_v ] close to the solid line . small differences come from the neglect of the multiplet dependence of the radial wave functions . in third column of table [ eng_mult ] , we show the approximate total energy of v@xmath96 multiplets calculated from equation ( [ eng_2body_sep_av ] ) . for comparison , the exact opm total energies are shown in second column of table [ eng_mult ] . if the total multiplet energies are deviate from the average of configuration , the approximate equation ( [ eng_2body_sep_av ] ) poorly predicts the total energies . however , the differences between the approximate and the exact total energies are no more than 0.002 au . for the other 3d@xmath0 transition metal atoms , the differences between the approximate and the exact energies are no more than 0.005 au . therefore , we can conclude that the equation ( [ ex - lc ] ) and ( [ eng_2body_sep_av ] ) work as the approximation formula of the exchange potential and the total energy for the multiplet states , respectively . in stead of showing the opm potentials for many multiplet states , we illustrated @xmath131 and @xmath132 multiplied by the occupation number @xmath137 in the top parts of figure [ f2pot ] and figure [ f4pot ] , respectively . as atomic number increases , the potentials @xmath138 and @xmath139 shrink and become high . this corresponds with the spread tendency of the electron density or the single electron orbital . in the bottom parts of figure [ f2pot ] and figure [ f4pot ] , we show functional derivative of @xmath140 and @xmath141 with respect to @xmath142 , respectively . maximum of @xmath143 is approximate to that of @xmath144 , but the shapes of the potentials are different . the two peaks are observed in @xmath138 , whereas , a single peak in @xmath143 . furthermore , at the nucleus , @xmath143 and @xmath138 are zero and nonzero value , respectively . finally , we discuss the relation between the opm exchange potential for the multiplet state and that for the single determinant . the multiplet states @xmath7 are represented as the linear combination of the single determinants @xmath8 . if the radial components of the wavefunctions do not depend on the multiplet states , the single determinant @xmath8 is represented as @xmath145 the 2-body energy of the single determinant @xmath146 is the expectation value of @xmath147 with respect to @xmath148 , which can be expressed as @xmath149 where @xmath150 , and @xmath151 is the spatial and spin coordinates . selecting some combination of single determinants , we solve simultaneous equations ( [ e2slmul ] ) with respect to the 2-body part of the multiplet state . the solution of the simultaneous equations is approximately valid because the orbital wave functions depend on multiplet states or single determinants . @xmath152 replaceing @xmath153 appearing in the derivation of ( [ energy_decomp ] ) with total hamiltonian , we obtain approximate equation for the total energy . @xmath154 from the variation of equation ( [ energy_decomp ] ) with respect to @xmath120 , the opm potential for the multiplet state @xmath155 is expressed as the linear combination of the opm potentials for the single determinants @xmath156 : @xmath157 following equation ( [ pot_decomp ] ) , we evaluate the approximate opm exchange potential for the multiplet state . figure [ oepdecomp_v ] presents the difference between the opm potential for v@xmath96 multiplet state and that for the average energy . the dotted line exhibits the approximate difference potential calculated from equation ( [ pot_decomp ] ) . we select the single determinant appearing in equation ( [ pot_decomp ] ) as @xmath158 , @xmath159 , @xmath160 , @xmath161 , @xmath162 , @xmath163 , @xmath164 , @xmath165 , @xmath166 , where the number represents the @xmath11 component of orbital angular momentum . the absence and presence of the line over the number stand for spin up and down state , respectively . in the solid line of figure [ oepdecomp_v ] , we show the difference potential calculated from exact equation ( [ sep - v ] ) . we find , from figure [ oepdecomp_v ] , that the approximate difference potential ( dotted line ) is very close to the exact one ( solid line ) . in the fourth column of table [ eng_mult ] , we show the approximate total energy calculated from equation ( [ total_energy_decomp ] ) . we compare the approximate total energy with exact opm total energy presented in second column of table [ eng_mult ] . the difference between the approximate total energy and the exact one is no more than 0.004 au . for the other 3d transition metal atoms , the difference between two energies is no more than 0.01 au . therefore , the exchange potential for the multiplet state can be evaluated using the theory such as the density functional theory , where the electronic structure is based on the single determinant , we applied opm to the multiplet energies of the 3d transition metal atoms using the total energy functional of the single - configuration hf @xcite form . the calculated opm exchange potential can be approximated by the @xmath167 and b88 exchange potentials in @xmath168 au . for near nucleus region ( @xmath169 au ) , the opm exchange potential strongly deviates from the @xmath167 and b88 exchange potentials . the difference between the opm exchange potential for the multiplet states and that for the average energy is quite smaller than the opm exchange potential . the opm exchange potential for the multiplet state of the 3d transition metal atom can be represented as the linear combination of @xmath170 and @xmath117 . we find that the opm exchange potential can be approximated by the linear combination of multiplet independent potentials @xmath171 and @xmath172 . this result indicates that the total energy functional can be represented as linear combination of the terms which are the product of the multiplet independent quantities derived from the slater integrals ( @xmath140 and @xmath141 ) and the multiplet dependent quantities ( @xmath173 , @xmath174 ) . as the representative of the potentials for the many multiplet states of the 3d transition metal atoms , we discussed the features of the potentials @xmath171 and @xmath172 . we find that the potentials @xmath171 and @xmath172 shrink , and become high as the atomic number increases , which is the same trend as observed in the wave function of the 3d transition metals . the multiplet state of the 3d transition metal atom is represented as the linear combination of the single determinants . from our calculation result , we find that the opm exchange potential of the multiplet state can be approximated as the linear combination of the opm exchange potentials of the single determinants . the result might be similar to the methods of ziegler @xcite , wood @xcite and von barth @xcite . however our result shows that the d electron multiplet state energy should be evaluated by including multiplet dependency to the total energy functional . one of the authors ( n.h . ) acknowledges a financial support from a grant from the ministry of education , science , sports and culture to promote advanced scientific research . 99 sharp r t and horton g k 1953 _ phys . rev . _ * 90 * 317 talman j d and shadwick w f 1976 _ phys . rev . _ a * 14 * 36 aashamar k , luke t m and talman j d 1979 3455 aashamar k , luke t m and talman j d 1981 803 weinert m , watson r e and fernando g w 2002 _ phys . rev . _ a * 66 * 032508 aashamar k , luke t m and talman j d 1979 _ phys . _ * 19 * 6 nagy 1999 2841 grling a 2000 _ phys . lett . _ * 85 * 4229 krieger j b , li y and iafrate g j 1992 _ phys . rev . _ a * 45 * 101 engel e , hck a and dreizler m 2000 _ phys . rev . _ a * 62 * 042502 sala f d , grling a 2003 _ j. chem . phys . _ * 118 * 10439 sala f d , grling a 2001 _ j. chem . * 115 * 5718 sala f d , grling a 2002 _ phys . lett . _ * 89 * 033003 sala f d , grling a 2002 _ j. chem . _ * 116 * 5374 grling a 1999 _ phys . * 83 * 5459 ivanov s , hirata s and bartlett r j 1999 _ phys . lett . _ * 83 * 5455 ivanov s , hirata s and bartlett r j 2002 _ j. chem . phys . _ * 116 * 1269 colle r and nesbet r k 2001 2475 yang w and wu q 2002 _ phys . * 89 * 143002 kummel s and perdew j p 2003 _ phys . lett . _ * 90 * 043004 kummel s and perdew j p 2003 _ phys . _ b * 68 * 035103 hirata s , ivanov s , grabowski i , bartlett r j , burke k and talman j d 2001 _ j. chem . phys . _ * 115 * 1635 engel e and vosko s h 1993 _ phys . rev . _ a * 47 * 2800 guan j , casida m e and salahub d r 2000 _ j. mol . ( theochem ) _ * 527 * 229 hirata s and head - gordon m 1999 _ chem . lett . _ * 302 * 375 wood j h 1980 1 von barth u 1979 _ phys . _ a * 20 * 1693 ziegler t , rauk a and baerends e 1977 _ j. theor . * 43 * 261 grling a and levy m 1993 _ phys . _ a * 50 * 196 grling a 1993 _ phys a * 47 * 2783 becke a d 1988 _ phys . rev . _ a * 38 * 3098 van leeuwen r and baerends e j 1994 _ phys . a * 49 * 2421 fischer c , brage t and jnsson p 2000 _ computational atomic structure _ ( institute of physics publishing ) slater j c 1974 _ the self - consistent field for molecules and solids : quamtum theory of molecules and solids _ vol 4 ( mcgraw - hill inc . ) .[eng_aoc ] the average energy of the configuration ( in au ) for 3d@xmath0 multiplets calculated by hartree - fock ( @xmath175 ) , opm ( @xmath176 ) , becke 88 ( @xmath177 ) , and x@xmath93 ( @xmath178,@xmath94 ) method . [ cols=">,>,>,>,>",options="header " , ]	we apply the optimized effective potential method ( opm ) to the multiplet energies of the 3d@xmath0 transition metal atoms , where the orbital dependence of the energy functional with respect to orbital wave function is the single - configuration hf form . we find that the calculated opm exchange potential can be represented by the following two forms . firstly , the difference between opm exchange potentials of the multiplet states can be approximated by the linear combination of the potentials derived from the slater integrals @xmath1 and @xmath2 for the average energy of the configuration . secondly , the opm exchange potential can be expressed as the linear combination of the opm exchange potentials of the single determinants .
there is a common and widely shared vision that next generation wireless networks will witness the proliferation of small - cells . as a matter of fact , researchers foresee network densification as one of key enablers of the 5 ^ -th^ generation ( 5 g ) wireless networks @xcite . although it refers to a concept rather than being a precise definition , the term _ ultra - dense networks _ is used to describe networks characterized by a massive and dense deployment of small - cells , in which the amount of base stations may grow up to a point where it will exceed the amount of user devices @xcite . as wireless networks evolve , performance requirements for the new technology are becoming more and more stringent . in fact , the 5 g requirements are set to a data rate increase up to a 1000-fold with respect to current 4 g systems@xcite , as well as for high energy efficiency @xcite in order to limit the energy expenditure of network operators . supported by recent results @xcite , cell densification as been put forward as the main enabler to achieve these target data rates . for example , the authors in @xcite have shown that the throughput gain is expected to grow linearly with the density of base stations per area ; this is a result of the simplified system model used during the investigation . namely , the assumption of a single slope path - loss model and that all base stations in the network are active . nonetheless , further work on cell densification has shown that , under less ideal assumptions , network performance may be different than what predicted in @xcite . in particular , when different path - loss models than single slope are used , the actual performance of cell densification is less optimistic than what estimated with single slope path - loss @xcite . in addition , if the base station density increases beyond the user density , like in a typical ultra - dense scenario , the network has been shown to experience a coverage improvement at the expense of a limited throughput gain @xcite . this implies that a larger number of bss will need to be deployed to meet a given data rate target , translating on higher network infrastructures costs . in recent years , stochastic geometry has been gradually accepted as a mathematical tool for performance assessment of wireless networks . in fact , one of the most important contribution to the study of cell densification can be found in @xcite , where the authors proposed a stochastic geometry - based framework to model single - tier cellular wireless networks ; by assuming a single slope path loss model , the authors have observed the independence of the signal - to - interference - plus - noise - ratio ( sinr ) and spectral efficiency ( se ) from the bs deployment density , with the main consequence being the linear dependence of the ase on the cell density . nevertheless , when the assumption of single - slope path - loss is dropped , it emerges that sinr , ase and coverage exhibit a different behaviour than what was found in @xcite . the authors in @xcite considered propagation models for millimeter waves . in @xcite , the authors extended the stochastic geometry framework proposed in @xcite to a multi - slope path loss model . the authors in @xcite developed a stochastic geometry framework for path - loss including line - of - sight ( los ) and non - line - of - sight ( nlos ) propagation for millimeter - waves . the effect of nlos propagation on the outage probability has been studied in @xcite , where the authors propose a function that gives the probability to have los at a given point depending on the distance from the source , on the average size of the buildings and on the density of the buildings per area . a stochastic geometry - based framework to study the performance of the network with a combined los / nlos propagation for micro - waves can be found in some previous work of ours @xcite and in @xcite . the common picture that emerges from the work @xcite is that of a non - linear behavior of the ase with the cell density ; overall , as a consequence of a different propagation model than the single slope path - loss , the spectral efficiency and coverage of the network do actually depend on the base station density . however , all these studies are based on the assumption of fully loaded networks , i.e. , all the base stations are active and have at least one user to serve ; thus , the applicability of the work above is limited to networks in which the density of base stations is lower than that of the users . in our paper , we broaden the study of network densification to the the case where there is no such a constraint in terms of bs density , i.e. , we also tackle partially loaded networks , in which some base stations might be inactive . work on stochastic geometry for partially loaded networks has been advanced in @xcite . the authors in @xcite studied the coverage in single tier networks , while multi - tier networks are addressed in @xcite . an analysis of the area spectral efficiency of partially loaded networks has been carried out in @xcite , while in @xcite the authors have extended the stochastic geometry - based model further to include multi - antenna transmission , and have also assessed the energy efficiency . overall , the authors in @xcite have shown that the network coverage improves as the base station density increases beyond the user density ; this is paid in terms of a lower throughput gain , which grows as a logarithmic function of the cell density . nonetheless , the authors in @xcite modeled the propagation according to a single slope path - loss model and did not investigate the effect of los / nlos propagation in partially loaded networks . to the best of our knowledge , currently there is no work that addresses ultra - dense scenarios with path - loss models different than the single slope for both fully and partially loaded networks . as a result of the combined effect of the path - loss model and of the partial load in ultra - dense networks , until now the behavior of the area spectral efficiency ( ase ) , coverage , and energy efficiency as the networks become denser , was unknown . this paper seeks to investigate the cell densification process in ultra - dense networks and evaluate the effect of los / nlos propagation on performance metrics such as coverage , spectral efficiency , area spectral efficiency , and energy efficiency . specifically , we use _ ultra - dense networks _ as a term to refer to those networks characterized by a very high density of base stations and that include both the cases of _ fully loaded networks _ ( i.e. , networks in which all the base stations are active ) and _ partially loaded networks _ ( i.e. , networks in which some base stations might be inactive and not transmit to any user ) . overall , the major contributions of our work can be summarized in the following points : * 1 ) stochastic geometry - based model for ultra - dense networks with los/ nlos propagation : * we propose a model based on stochastic geometry that allows us to study the signal - to - interference - plus - noise - ratio ( sinr ) distribution , the spectral efficiency and the area spectral efficiency of ultra - dense networks where the propagation has los and nlos components . we build on previous work @xcite and we adapt the model proposed by andrews et al . to the case of los / nlos propagation . in addition , our framework takes also into account the partial load of ultra - dense networks , in which a fraction of base stations may be inactive . * 2 ) study of cell densification , partial load and frequency reuse in networks where signal follows los / nlos propagation : * first , we investigate the effect of network densification on performance metrics such as sinr , spectral efficiency and area spectral efficiency in networks where the path - loss follows the los / nlos propagation . in particular , the ase gain becomes lower than linear at high cell densities , meaning that a larger number of bss would be necessary to achieve a given target with respect to the case of single slope path loss . moreover , the network coverage drops drastically as the bs density increases . then , we show that the performance drop due to los / nlos propagation gets mitigated by the usage of frequency reuse or if the base station density exceeds the user density , as it is likely to occur in ultra - dense networks . to the best of our knowledge , the combined effect of los / nlos propagation and partial load / frequency reuse has not been addressed before . * 3 ) investigation on the minimum transmit power per bs and energy efficiency for networks where signal propagates according to los / nlos path - loss : * as the cell density increases , the transmit ( tx ) power per base station can be lowered . we evaluate the minimum tx power per bs such that the network is guaranteed to be in the interference - limited regime , in which case the performance is not limited by the tx power . second , we make use of the tx power to determine the energy efficiency of the network when the propagation has los and nlos components . we show that the energy efficiency with los / nlos propagation drops considerably with respect to the case of single slope path - loss , making cell - densification costly for the network from an energetic stand - point . we further extend the study of the energy efficiency to frequency reuse and partial load . the remainder of this paper is organized as follows . in section [ sect : systemmodel ] we describe the system model . we show our formulation for computing the sinr , se and ase in section [ sect : sinr_se_ase ] and we address the energy efficiency in section [ sec : energy_efficiency ] . in section [ sect : results ] we present and discuss the results while the conclusions are drawn in section [ sect : conclusions ] . in this paper we consider a network of small - cell base stations deployed according to a homogeneous and isotropic spatial poisson point process ( sppp ) , denoted as @xmath0 , with intensity @xmath1 . further , we assume that each base station ( bs ) transmits with an isotropic antenna and with the same power , @xmath2 , of which the value is not specified , in order to keep our model general and valid for different base station classes ( e.g. , micro - bss , pico - bss , femto - bss ) ; we focus our analysis on the downlink . in our analysis , we considered the following path loss model : @xmath3 where @xmath4 and @xmath5 are the path - loss exponents for los and nlos propagation , respectively ; @xmath6 and @xmath7 are the signal attenuations at distance @xmath8 for los and nlos propagation , and @xmath7 can either refer to the signal attenuations at distance @xmath8 m or @xmath8 km ; this depends on the actual values given for the parameters of the channel model . ] respectively ; @xmath9 is the probability of having los as a function of the distance @xmath10 . the model given in is used by the 3gpp to model the los / nlos propagation , for example , in scenarios with heterogeneous networks ( * ? ? ? * table a.2.1.1.2 - 3 ) . the incorporation of the nlos component in the path loss model accounts for possible obstructions of the signal due to large scale objects ( e.g. buildings ) , which will result in a higher attenuation of the nlos propagation compared to the los path . we further assume that the propagation is affected by rayleigh fading , which is exponentially distributed @xmath11 . regarding the shadow fading , it has been shown that in networks with a deterministic , either regular or irregular , base station distribution affected by log - normal shadow fading , the statistic of the propagation coefficients converges to that of a network with sppp distribution as the shadowing variance increases @xcite . in other words , this sppp intrinsically models the effect of shadow fading . to ensure that our formulation and the outcomes of our study are general and not limited to a specific los probability pattern , we consider two different los probability functions . the first one , which is proposed by the 3gpp ( * ? ? ? * table a.2.1.1.2 - 3 ) to assess the network performance in pico - cell scenarios , is given below : @xmath12 with @xmath13 and @xmath14 being two parameters that allow to match the measurement data . unfortunately , this function is not practical for an analytical formulation . therefore , we chose to approximate it with a more tractable one , namely : @xmath15 where @xmath16 is a parameter that allows to be tuned to match . the second function is also suggested by the 3gpp ( * ? ? ? * table a.2.1.1.2 - 3 ) and is given below : @xmath17 from a physical stand point , the parameter @xmath16 can be interpreted as the los likelihood of a given propagation environment as a function of the distance . in our model , we always assume that : ( i ) the users are uniformly distributed ; ( ii ) that each user s position is independent of the other users position ; and ( iii ) each user connects only to one base station , the one that provides the strongest signal . we denote by @xmath18 the density of users per area ; whenever we consider a finite area @xmath19 , @xmath20 indicates the average number of users in the network . we also assume the users are served with full buffer , i.e. , the base station has always data to transmit to the users and make full use of the available resources . depending on the ratio between the density of users and the density of base stations , we distinguish between two cases , namely , fully loaded and partially ( or fractionally ) loaded networks . by fully loaded networks we refer to the case where each bs has at least one user to serve . with reference to a real scenario , fully loaded networks model the case where there are many more users than base stations , so that each base station serves a non - empty set of users . however , when the density of users is comparable with or even less that of base stations , some base stations may not have any users to serve and would then be inactive , meaning that they do not transmit and do not generate interference . when this occurs , we say that the network is partially loaded . the network can be modelled as partially loaded to study those scenarios characterized by high density of base stations and , in particular , scenarios where the density of base stations exceeds the density of users , such as in ultra - dense networks . to define formally the concepts of fully and partially loaded networks , we first need to introduce another concept , which is the _ probability a base station being active_. [ def : probability_of_activity ] the probability of a base station being active , denoted as @xmath21 , is the probability that a base station has at least one user to serve . this event implies that the base station is active and transmits to its users . [ def : fully_partially_loaded ] the network is said to be fully loaded if @xmath22 ; the network is said to be partially or fractionally loaded if @xmath23 . in this section we propose an analytical model to compute the sinr complementary cumulative distribution function ( ccdf ) , which allows us to asses key performance metrics such as coverage , spectral efficiency and ase . in order to compute the sinr tail distribution ( i.e. , the complementary cdf ) , we extend the analytical framework first proposed in @xcite so that to include the los and nlos components . from the slivnyak s theorem ( * ? ? ? * theorem8.10 ) , we consider the _ typical user _ as the focus of our analysis , which for convenience is assumed to be located at the origin . the procedure is composed of two steps : ( i ) we compute the sinr ccdf for the typical user conditioned on the distance from the user to the serving base station , denoted as @xmath24 ; ( ii ) using the pdf of the distance from the closest bs @xmath25 , which corresponds to the serving bs , we can average the sinr ccdf over all possible values of distance @xmath24 . let us denote the sinr by @xmath26 ; formally , the ccdf of @xmath26 is computed as : @xmath27=\mathrm{e_{r}}\big[\mathrm{p } \left[\gamma > y\|r\right]\big ] = \int _ { 0}^{+\infty } \mathrm{p}\left[\gamma > y\|r = r\right]f_{r}(r)\mathrm{d}r.\ ] ] the key elements of this procedure are the pdf of the distance to the nearest base station @xmath28 and the tail probability of the sinr conditioned on @xmath24 , @xmath29 $ ] . the methodology to compute each of these elements while modelling the los and nlos path loss components will be exposed next . the set of the base stations locations originates an sppp , which we denote by @xmath30 . and use @xmath31 and instead of @xmath32 for convenience of notation . ] as a result of the propagation model we have adopted in our analysis ( see section [ subsect : channelmodel ] ) , the user can either be in los or nlos with any base station @xmath32 of @xmath33 . now , we perform the following mapping : we first define the set of los points , namely @xmath34 , and the set of nlos points , @xmath35 . then , each point @xmath32 of @xmath33 is mapped into @xmath34 if the base station at location @xmath32 is in los with the user , while it is mapped to @xmath35 if the base station at location @xmath32 is in nlos with the user . since the probability that @xmath32 is in los with the user is @xmath36 , it follows that each point @xmath32 of @xmath33 is mapped with probability @xmath36 into @xmath34 and probability @xmath37 into @xmath35 . given that this mapping is performed independently for each point in @xmath33 , then from the `` thinning theorem '' ( * ? ? ? * theorem 2.36 ) it follows that the processes @xmath34 and @xmath35 are sppps with density @xmath38 and @xmath39 , respectively . note that , because of the dependence of @xmath40 and @xmath41 on @xmath31 , @xmath34 and @xmath35 are inhomogeneous sppps . to make the formulation more tractable , we consider @xmath34 and @xmath35 to be independent processes ; because the union of two sppps processes is an sppp of which the density is the sum of the densities of the individual sppps ( * ? ? ? * preposition 1.3.3 ) , the union of @xmath34 and @xmath35 is an sppp with density @xmath42 , i.e. , it is an sppp with the same density as that of the original process @xmath33 . hence , the assumption of independence between @xmath34 and @xmath35 does not alter the nature of the process @xmath33 . given that we have two inhomogeneous sppp processes , it is not trivial to obtain the distribution of the minimum distance of the user to the serving base station , which will be necessary later on to compute the sinr cdf . in fact , assuming the user be in los with the serving base station at a distance @xmath43 , there might be an interfering bs at a distance @xmath44 which is in nlos with the user . this is possible because the nlos propagation is affected by a higher attenuation than the los propagation . hence , to make our problem more tractable , we map the set of points of the process @xmath35 , which corresponds to the nlos base stations , into an equivalent los process @xmath45 ; each point @xmath46 located at distance @xmath47 from the user is mapped to a point @xmath48 located at distance @xmath49 from the user , so that the bs located at @xmath48 provides the same signal power to the user with path - loss @xmath50 as if the base station were located at @xmath31 with path - loss @xmath51 . [ def : mapping_function ] we define the mapping function @xmath52 as : @xmath53 @xmath54 [ def : inv_mapping_function ] the inverse function @xmath55 is defined as : @xmath56 @xmath57 where @xmath58 while @xmath59 . it is important to notice that , from the `` mapping theorem '' ( * ? ? ? * theorem 2.34 ) , @xmath45 is still an sppp . using the mapping we introduced in section [ subsect : mapping ] , we can compute the pdf @xmath25 of the minimum distance @xmath24 between the user and the serving bs . we first compute the probability @xmath60 $ ] ; the pdf can be ultimately obtained from the derivative of @xmath60 $ ] as @xmath61 ) $ ] . let @xmath62 be the ball of radius @xmath63 centred at the origin @xmath64 . moreover , we use the notation @xmath65 to refer to number of points @xmath66 contained in @xmath67 @xcite . using the mapping we introduced in section [ subsect : mapping ] the probability @xmath60 $ ] can be found as : @xmath68=\mathrm{p}\left [ \phi_{\mathrm{l}}\left(b(0,r)\right)=0\cap\phi_{\mathrm{eq}}\left(b\left(0,r\right)\right)=0\right]\ ] ] @xmath69\ ] ] @xmath70\cdot\mathrm{p}\left[\phi_{\mathrm{nl}}\left(b\left(0,d_{\mathrm{eq}}^{-1}(r)\right)\right)=0\right],\ ] ] where equality @xmath71 comes from the mapping defined in and in , while equality @xmath72 comes from the independence of the processes @xmath34 and @xmath35 . by applying the probability function of inhomogeneous sppp ( * ? ? ? * definition 2.10 ) , we obtain the following , @xmath73=\exp\bigg(-\int_{b(0,r)}\lambda_{\mathrm{l}}(x)\mathrm{d}x\bigg ) \exp\bigg(-\int_{b\left(0,d_{\mathrm{eq}}^{-1}(r)\right)}\lambda_{\mathrm{nl}}(x)\mathrm{d}x\bigg).\ ] ] from , we can obtain @xmath25 , first , by integrating and , second , by computing its first derivative in @xmath74 . the formulation in is general and thus can be applied to several los probability functions @xmath9 . below , we provide the expression of the pdf of the distance from the ue to the serving bs for the los functions and , respectively . [ res : pdf_p_l_exp_square ] if the los probability function is as in and if we denote @xmath75 by @xmath76 , the pdf of the distance to the serving bs is : @xmath77 @xmath78 [ res : pdf_p_l_exp ] if the los probability function is as in and if we denote @xmath75 by @xmath76 , the pdf of the distance to the serving bs is : @xmath79 @xmath80 @xmath81 we refer to the appendix for the details of the @xmath25 we have given in and in . in this framework , we include the cases of partially loaded networks and of frequency reuse and we treat them separately . first , we identify the set of bss that are active , i.e. , those having one or more users to serve . as we focus the analysis on the typical user , we can also identify the set of bss that act as interferers for that user ; an active bs ( excluding the one serving the user ) is seen as an interferer if that bs transmits over the same band used to serve that user . in the following , we denote by @xmath82 the set of active bss , while we denote by @xmath83 the set of the interfering bss . let us consider first the case of frequency reuse , in which all the base stations are active , but each of these only uses a portion of the spectrum , in order to reduce the interference . since all the base stations are supposed to be active , the process @xmath82 is the same as @xmath33 . however , we assume each base station selects a channel in a random manner using , for instance , frequency aloha spectrum access @xcite . with a frequency reuse factor of @xmath84 , each base station uses @xmath85 out of @xmath84 channels , which is chosen independently from the other base stations . hence , each bs interferes with a given user with probability @xmath86 ; this is equivalent to carrying out a thinning of the original process @xmath33 with probability @xmath86 ; from the thinning theorem , we obtain that @xmath83 is a homogeneous process with density @xmath87 . in regards to the partially loaded networks , we recall from section [ subsect : user_distribution ] that a fraction of the base stations might be inactive and , as such , would not generate interference . assuming all the bss transmit over the same band , in partially loaded networks the active base stations are the only bss that generate interference to the users with the exception of the serving bs . thus , we can write @xmath88 , where @xmath89 is the serving base station ; moreover , from the palm theorem @xcite , @xmath83 and @xmath82 are characterized by the same density . to obtain the process of active bss @xmath82 from the original the process @xmath33 , we first assume that each ue deployed in the network connects to the closest bs ; , with los / nlos propagation the serving bs might not be the closest one to the user . nonetheless , this does not alter the validity of the explanation we are giving in this section . ] finally , only the bss which are assigned one or more users are active . yet , this is equivalent to performing a thinning of the original process to obtain @xmath82 . however , the fact that a base station is picked to be part of @xmath82 ( i.e. there exists a user for which this bs is the closest one ) depends on the positions of the neighbouring base stations , which implies that the base stations are not chosen independently of one another @xcite . as the independence is one of the necessary conditions in order to have an sppp , it follows that @xmath82 is not an sppp . although @xmath82 can not be formally regarded as an sppp , it has been shown that the actual sppp obtained through the thinning of @xmath33 well approximates @xmath82 @xcite . specifically , the authors in @xcite have shown that , ( i ) the probability @xmath21 of a base station to be active ( i.e. , to have users to serve ) can be well approximated once the density of users @xmath18 and density of base stations @xmath1 are known , and , ( ii ) the process @xmath82 of active base stations can be well approximated by thinning the original process @xmath82 with probability @xmath21 , which is given below @xcite : @xmath90 from the thinning theorem , it follows that the resulting process obtained through thinning as described above is an sppp with density @xmath91 ; moreover , @xmath92 . in light of these findings , we approximate @xmath82 with an sppp of density @xmath93 . fig . [ fig : prob_a_vs_density ] shows how the probability @xmath21 and the intensity of the interfering bss @xmath94 vary as functions of the ratio @xmath95 . drops as the ratio @xmath95 is close to or greater than 1 , i.e. , as @xmath1 approaches @xmath18 . as a result of this , the density of active bss as well as the density of interfering bss converge to @xmath18 as @xmath1 approaches or overcomes @xmath18 . ] the probability @xmath29 $ ] can be computed as in ( * ? ? ? * theorem 1 ) ; we skip the details and provide the general formulation : @xmath96=\mathrm{p}\left[\frac{g k_{\mathrm{l}}r^{-\beta_{\mathrm{l}}}}{\sigma^{2}+i_{r}}>y\right ] = e^{-\mu y k_{\mathrm{l}}^{-1}r^{\beta_{\mathrm{l}}}\sigma^{2}}\mathcal{l}_{i_{r}}(\mu y k_{\mathrm{l}}^{-1}r^{\beta_{\mathrm{l}}}),\ ] ] where @xmath97 is the rayleigh fading , which we assume to be an exponential random variable @xmath98 ; @xmath99 is the variance of the additive white gaussian noise normalized with the respect to the transmit power , @xmath100 is the interference conditioned on @xmath74 , i.e. , @xmath101 @xmath102 where @xmath103 and @xmath104 are independent and identically distributed @xmath105 fading coefficients . please , note that the overall interference accounts only for the active base stations . compared to the formulation of @xmath106 proposed in @xcite , in our case we have to deal with two non - homogeneous sppp , namely @xmath34 and @xmath35 instead of a single homogeneous sppp . the laplace transform @xmath106 can be written as follows : @xmath107.\end{aligned}\ ] ] given that @xmath34 and @xmath35 are two independent sppp , we can separate the expectation to obtain : @xmath108 \\ \nonumber \mathrm{e}_{\phi_{\mathrm{nl}}\cap\phi_{\mathrm{a}},g_j}\bigg[\exp\bigg(-s\sum\limits_{\{j:\ : x_j\in \phi_{\mathrm{nl}}\cap\phi_{\mathrm{a}},\ : \\|x_j\\| > d_{\mathrm{eq}}^{-1}(r)\}}g_j k_{\mathrm{nl}}\\|x_j\\|^{-\beta_{\mathrm{nl}}}\bigg)\bigg].\end{aligned}\ ] ] by applying the probability generating functional ( pgfl ) for sppp ( which holds also in case of inhomogeneous sppp @xcite ) to , we obtain the following result : the laplace transform @xmath106 for los / nlos propagation with model given in is : @xmath109p_{\mathrm{l}}(v)v\mathrm{d}v\bigg)\ ] ] @xmath110p_{\mathrm{nl}}(v)v\mathrm{d}v\bigg).\ ] ] the laplace transform in along with and can be plugged in to obtain the sinr ccdf through numerical integration . similarly to ( * ? ? ? * section iv ) we compute the average spectral efficiency and the ase of the network . first , we define the ase as : @xmath111 } { a\cdot\mathrm{bw_{a}}}=\frac{\lambda_{\mathrm{a}}\cdot \mathrm{e}[\mathrm{c } ] } { n},\ ] ] where @xmath112 is the available bandwidth , @xmath113 is the used bandwidth , @xmath114 $ ] is the average spectral efficiency , @xmath19 is the area , @xmath115 is the number of active base stations within the area @xmath19 , and @xmath84 is frequency reuse factor . the average rate @xmath114 $ ] can be computed as follows @xcite : @xmath116 = \mathrm{e}\left[\log_{2}(1+\gamma)\right]=\int _ { 0}^{+\infty}\mathrm{p}\left[\log_{2}(1+\gamma)>u\right]\mathrm{d}u \\ \nonumber = \int_{0}^{+\infty}\int _ { 0}^{+\infty}\mathrm{p}\left[\log_{2}(1+\gamma)>u\|r = r\right]f_{r}(r)\mathrm{d}r\mathrm{d}u.\end{aligned}\ ] ] @xmath117 where @xmath106 is given in . similarly to the sinr ccdf , can be evaluated numerically . we evaluate the tx power in order to compute the overall power consumption of the wireless network . ideally , to ensure the network performance not be limited by the transmit power , @xmath2 should be set in order to guarantee the interference - limited regime , i.e. , the transmit power should be high enough so that the thermal noise power at the user receiver can be neglected with respect to the interference power at the receiver . in fact , when the network is in the interference - limited regime , the transmit power is high enough that any further increase of it would be pointless in terms of enhancing the sinr , since the receive power increment would be balanced by the exact same interference increment . practically , we refer to the outage probability @xmath118 $ ] as a constraint to set the power necessary to reach the interference limited regime . when the tx power is low , small increments of @xmath2 yields large improvements of the outage @xmath119 ; however , as @xmath2 increases , the corresponding outage gain reduces , until @xmath119 eventually converges to its optimal value @xmath120 , which would be reached in absence of thermal noise . it is reasonable to assume the network be in the interference - limited regime when the following condition is met : @xmath121 where @xmath122 is a tolerance measure setting the constraint in terms of the maximum deviation of @xmath119 from the optimal value @xmath123 . . provides us with a metric to compute the transmit power , but does not give us any indication on how to find @xmath2 as a function of the density @xmath1 . unfortunately , we can not derive a closed - form expression for the transmit power that satisfies , as we do not have any closed - form solution for the outage probability @xmath124 $ ] . we then take a different approach to calculate the minimum transmit power . in alg . [ algo : power ] we proposed a simple iterative algorithm that finds the minimum transmit power satisfying by using numerical integration of . this algorithm computes the outage probability corresponding to a given @xmath2 ; starting from a low value of power , it gradually increases @xmath2 by a given power step @xmath125 , until is satisfied . to speed up this procedure , the step granularity is adjusted from a coarse step @xmath126 up to the finest step @xmath127 , which represents the precision of the power value returned by alg . [ algo : power ] . inputs : 1 . vector of the power steps in dbm @xmath128 $ ] , @xmath129 is the length of vector @xmath130 ; 2 . outage sinr threshold @xmath131 ; 3 . outage tolerance @xmath132 ; initialize variables : 1 . @xmath133 , where @xmath134 is the awgn power in dbm over the bandwidth @xmath135 2 . @xmath136 find optimal outage @xmath137 $ ] by integrating with parameter @xmath138 find @xmath139 $ ] by integrating with parameter @xmath140 set granularity of the power step @xmath141 increase the current power with step @xmath125 , i.e , @xmath142 find @xmath139 $ ] by integrating with parameter @xmath140 update the final value of power , i.e. , @xmath136 remove the last power increment before increasing the granularity , i.e. , @xmath143 output : @xmath144 is the power in dbm s.t . is satisfied . in this subsection , we characterize the energy efficiency of the network as a function of the base station density to identify the trade - off between the area spectral efficiency and the power consumed by network . we define the _ energy efficiency _ as the ratio between the overall throughput delivered by the network and the total power consumed by the wireless network , i.e. , we define the energy efficiency as follows : @xmath145 where @xmath146 is the network throughput , which can be written as @xmath147 , with * @xmath148 * denoting the bandwidth and @xmath149 denoting the area spectral efficiency ; @xmath150 is the overall power consumption of the network . when we compute the power consumption of each bs , we need to take into account that a fraction of the base stations may be inactive and model the power consumption accordingly . for active base station , we model the power consumption @xmath151 of the base station assuming that @xmath152 is the sum of two components , i.e. , @xmath153 ; the first , denoted by @xmath154 , takes into account the energy necessary for signal processing and to power up the base station circuitry . this power @xmath154 is modelled as a component being independent of the transmit power and of the base station load @xcite . the second component , denoted by @xmath155 , takes into account the power fed into the power amplifier which is then radiated for signal transmission . the power @xmath155 is considered to be proportional to the power transmitted by the base station ; we can thus write @xmath156 , where @xmath157 takes into account the losses of the power amplifier ( i.e. , we assume @xmath157 to be the inverse of the power amplifier efficiency ) . in the case of inactive base stations , we assume that the bs switches to a stand - by state for energy saving purposes @xcite , in which it does not transmit ( i.e. , @xmath158 ) and reduces the circuitry power consumption . therefore , the power required to maintain the stand - by state can be modelled as @xmath159 , where @xmath160 is power saving factor that describes the relative power consumption of the circuitry with respect to the active case ; note that @xmath161 . finally , the overall power consumption of the network due to both active and inactive base station can be expressed as follows : @xmath162 @xmath163 the energy efficiency for the cases of fully loaded networks and partially loaded networks is addressed in the following sub - sections . in this section we study the energy efficiency @xmath164 trend as a function of @xmath1 ; we focus on fully loaded networks , i.e. , @xmath165 and @xmath166 . unfortunately , the analysis of the derivative of @xmath167 is not straightforward , as we have a closed - form solution neither for the throughput @xmath146 nor for the transmit power @xmath168 . one feasible way to get around this burden is to approximate @xmath146 and @xmath168 with functions in the form : @xmath169 the model in has two advantages : ( i ) it can be easily derivated and , thus , is apt to investigate the existence of optima ; ( ii ) it is well suited to fit the non - linear behaviour of ase and tx power . in fact , we have shown in our previous work @xcite that both @xmath146 and @xmath168 can be approximated with a piece - wise function in the form , where the parameters @xmath170 and @xmath171 can be obtained , for instance , by linear regression in the logarithmic domain for a given range of values of @xmath1 . backed by the conclusions from previous work @xcite , we approximate the throughput as @xmath172 and the transmit power as @xmath173 , within a given interval of @xmath1 . under these assumptions , the energy efficiency becomes : @xmath174 the derivative of @xmath164 is given below : @xmath175 let us note that @xmath176 , @xmath154 , @xmath157 and @xmath177 are positive ; moreover it is reasonable to assume that @xmath178 ( i.e. , the area spectral efficiency is an increasing function of the density ) and that @xmath179 , i.e. , the transmit power per bs is a decreasing function of the density . in the following paragraphs , we study the behaviour of the energy efficiency as function of the density @xmath1 by analyzing the derivative @xmath180 . we distinguish the following three cases . this occurs if the ase growth is linear or superlinear , i.e. , if @xmath181 . from this , if follows that also @xmath182 holds true ; in this case , @xmath180 is strictly positive , meaning that the energy efficiency increases as the density increases . this occurs if the ase growth is sublinear , i.e. , if @xmath183 , and , in addition , @xmath184 . then , @xmath180 is strictly negative and so the energy efficiency is a monotonically decreasing function of the density @xmath1 . if ase gain is sublinear ( i.e. @xmath183 ) but grows with a slope @xmath185 sufficiently high , ( i.e. , @xmath186 ) , then we obtain that the derivative @xmath187 nulls for @xmath188 is positive for @xmath189 and is negative for @xmath190 . therefore , @xmath191 is a global maximum of the energy efficiency . as a whole , the behavior of the spectral efficiency is due to how the growths of the ase of the tx power relate among each other as the base station density increases . if the ase grows rapidly enough to counterbalance the total power increase of the network given by the addition of base stations , then the energy efficiency increases with the bs density ; this means that adding base station is profitable in terms of energy efficiency . else , adding bss turns not to be profitable from energy efficiency point of view . for partially loaded networks , we only analyze the case where @xmath192 , as the opposite case of @xmath193 leads back to fully loaded networks . by using lhpital s rule , one can show that can be approximated by @xmath194 , for @xmath1 is sufficiently greater than @xmath18 . by applying this approximation to , we obtain : @xmath195 it is known from @xcite that , as the bs density increases , the main contribution to the total power consumption is due to the circuitry power @xmath196 , while the transmit power becomes negligible for the overall power balance . therefore , to make the problem more tractable , we can further approximate the total power in as @xmath197 . from , by using the approximation @xmath172 for the throughput and @xmath198 for the power , we obtain the following expression for the energy efficiency : @xmath199 to analyze the behaviour of the energy efficiency as a function of @xmath1 , we follow the same approach as in section [ subsec : energy_eff_fully_loaded ] and we compute the derivative of @xmath164 , which is given below : @xmath200 as the ase is known to be sub - linear for partially loaded networks @xcite , we assume @xmath201 ; moreover , the power saving factor @xmath160 satisfies @xmath202 . therefore , the derivative @xmath203 nulls for : @xmath204 it is positive for @xmath205 while it is negative for @xmath206 . hence , @xmath207 is a local maximum of the energy efficiency for partially loaded networks and the energy efficiency decreases for densities @xmath208 . note that , this result holds for @xmath1 sufficiently greater than @xmath18 . in this section we present and discuss the results we obtained by integrating numerically the expressions of outage probability , of the spectral efficiency ( se ) , and of the ase . in section [ subsect : rate_ase_density ] , [ sub : frequency_reuse ] and [ sub : partially_loaded_results ] we assume the network to be interference - limited ( i.e. , we set the thermal noise power to 0 ) , while the noise is taken into account in section [ sub : txpowerperbs_results ] and [ sub : energy_efficiency_results ] . the parameters we used to obtain our results are specified in table [ table : parameters ] . .parameters for results section [ cols="<,<",options="header " , ] in this subsection we assume the network to be fully loaded and with frequency reuse 1 . we compared the results for two los probability functions , namely and ; we also compared the results for los / nlos propagation with those obtained with a the single slope path - loss model . we first analyze the outage probability ( defined as @xmath209 $ ] ) results , which have been obtained by numerical integration of . we show the outage probability results in fig . [ fig : outage_full_load ] , where we can see the different effect of the los / nlos propagation with respect to the single slope path - loss ( pl ) . with single - slope pl , the outage is constant with the bs density . in contrast , with los / nlos propagation , there is a minimum in the outage curves , which is achieved for density @xmath1 = 50 - 100bss / km@xmath210 , depending on the los probability function . within this range of densities , the user is likely to be in los with the serving bs and in nlos with most of the interfering bs , meaning that the interference power is lower than the received power . at densities @xmath1 greater than 300bss / km@xmath210 , the outage starts growing drastically and , depending on the los likelihood , can reach 32 - 43% . this is due to the fact that more and more interfering bss are likely to enter the los region , causing an overall interference growth and thus a reduction of the sir . at densities @xmath1 smaller than 100bss / km@xmath210 , the serving bs as well as the interfering bss are likely to be in nlos with the user . because of this , both the receive power and the overall interference increase at the same pace and , as a consequence , the sir remains constant , and so does the outage . let us note that , the los probability function affects the outage curves at intermediate values of the bs density ( e.g. 10 - 300 bss / km@xmath210 ) . at low densities , all the bss are likely to be in nlos with the user , while at high densities the serving bs and the strongest interferers are likely to be in los bss are likely to be in los with the user . the results of the ase are shown in fig . [ fig : ase_full_load ] . compared to the single - slope pl , which shows a linear growth of the ase with the density @xmath1 , with the los / nlos propagation we observe a different behaviour of the ase . in particular , we observe a lower steepness of the ase curve at high bs densities , which is due to the effect of the interfering bss entering the los region and , thus , increasing the total interference power . to assess steepness of the ase , we can use linear regression to interpolate the ase curve with the model given in . in particular , we can approximate the ase @xmath211 with a piece - wise function of the kind @xmath212 , where @xmath213 and @xmath185 are given for given intervals of @xmath1 . we specifically focus on @xmath185 , which gives the steepness of the ase curve . with reference to the ase curve ( solid - blue curve in fig . [ fig : ase_full_load ] ) obtained with as a los probability function , the the value of the parameter @xmath185 turns to be 1.15 within the range of @xmath1 1 - 50 bss / km@xmath210 , 0.48 within the range 50 - 500 bss / km@xmath210 and 0.81 within the range 500 - 10000 bss / km@xmath210 . to have a comprehensive view of the frequency reuse as an interference mitigation scheme , we need to assess the trade - off between the ase and the network coverage probability , defined as @xmath214 $ ] . the results of this trade - off are shown in fig . [ fig : ase_coverage_tradeoff ] , where we plotted the network coverage against the ase for different frequency reuse factors and base station densities . , and compare the combined los / nlos model with the single slope one . ] firstly , we focus on the los / nlos propagation ; we can notice from this plot that , if we fix the bs density , higher frequency reuse factors enhance the network coverage but , on the other hand , determine a drop of the ase . this is in line with what one would expect from frequency reuse . nonetheless , if we have no constraint in the choice of the bs density , the ase vs coverage trade - off improves as the frequency reuse factor increases . in fact , the trade - off curve we obtain for a given reuse factor @xmath215 lies on the top - right hand side with respect to the curve for reuse factor @xmath216 . this means that , by increasing the reuse factor and the base station density at the same time , it is possible to achieve better performance than with a lower frequency reuse factors ; note , though , that this is true when there is no constraint in terms of bs density . by looking at the single slope pl curve in fig . [ fig : ase_coverage_tradeoff ] , it appears that higher frequency reuse factors should still be preferred in order to improve the ase vs coverage trade - off . however , unlike with the los / nlos path loss , increasing the bs density enhances the ase with no loss in terms of network coverage . yet , modelling the signal propagation with the combined los / nlos path loss yields different results than with the single - slope pl . in this subsection we show the results of the cell densification for partially loaded networks with los / nlos propagation . differently from the case of fully loaded networks , we recall that a fraction of the bss may be inactive and , thus , the density of interfering bss @xmath94 does not necessary follow the trend of bs density @xmath1 ( see section [ subsec : partially_loaded_networks ] and fig . [ fig : prob_a_vs_density ] ) . in fig . [ fig : outage_partial_load ] and [ fig : ase_partial_load ] we show the outage probability and the ase curves , respectively , as function of the base station density for difference user densities . to better understand the effect of the partial load on the network performance , we compare these curves with those for fully loaded networks . we reported the values of the probability @xmath21 of a bs being active over the outage and ase curves . we observe that , as long as @xmath217 , the deviation from the fully loaded network case is minimal . however , as soon as @xmath1 approaches the value of user density @xmath18 , the probability @xmath21 drops and , as a consequence , the density of interfering @xmath94 bss grows slowly with @xmath1 , up to the point where it saturates and converges to @xmath18 ( see fig . [ fig : prob_a_vs_density ] ) . at the same time , as @xmath1 increases , the distance from ue to the serving bs tends to decrease , leading to an increment of the received power . overall , the fact that @xmath94 saturates whereas the received power keeps growing as @xmath1 increases has a positive impact on the sir ; as a result , the outage probability ( see fig . [ fig : outage_partial_load ] ) and the spectral efficiency improve once the density @xmath1 approaches or overcomes @xmath18 . in regards to the ase trend , we show the results in fig . [ fig : ase_partial_load ] . according to , the ase trend is the combined outcome of the increase of the spectral efficiency and of the density of the active base stations . as the density of base stations increases and approaches the user density @xmath18 , the density of active base stations will converge to @xmath18 ( see fig . [ fig : outage_partial_load ] ) ; given that the density of active bss remains constant , the only contribution to the ase increase will be given by the spectral efficiency improvement . as a matter of fact , we can see that , with respect to the case of fully loaded networks , the ase curves show a lower gain when the density @xmath1 approaches @xmath18 . to assess steepness of the ase , we applied linear regression to the ase curves in order to obtain the value of the parameter @xmath185 corresponding to different intervals of @xmath1 ; we specifically consider the approximation for the curve corresponding to @xmath218ues / km@xmath210 ( red curve in fig . [ fig : ase_partial_load ] ) . these values are @xmath219 within the density range 1 - 50 bss / km@xmath210 , @xmath220 within the density range 50 - 500 bss / km@xmath210 and @xmath221 within the density range 500 - 10000 bss / km@xmath210 . in fig . [ fig : transmit_power ] we show the simulation results of the transmit power per base station @xmath168 , which has been computed by using algorithm [ algo : power ] as explained in section [ sub : computing - the - transmit ] . in this figure we compare the results we obtained using the _ single slope _ and the _ combined los / nlos _ path loss models . db and for tolerance @xmath222 . the plot compares the tx power per bs for single slope - slope and los / nlos path - loss for fully loaded networks . it also provides the curve for frequency reuse factor 2 and 3 and for partially loaded network with @xmath218ues / km@xmath210 . ] as we can see from this plot , the behaviour of the transmit power as a function of the bs density @xmath1 is different in the two cases of single slope and combined los / nlos propagation . with reference to fig . [ fig : transmit_power ] , with single slope path loss , the power decreases linearly ( in logarithmic scale ) with the density ; in the case of combined los / nlos propagation , the transmit power exhibits different slopes as the as the base station density increases . we used linear regression to assess the slopes of the tx power curves ( indicated by @xmath223 , as explained in section [ subsec : energy_eff_fully_loaded ] ) within different density intervals . with reference to the curve corresponding to fully loaded networks with los / nlos propagation ( solid - blue curve in fig . [ fig : transmit_power ] ) , the values of ( @xmath224 , @xmath223 ) are @xmath225 within the @xmath1 range 1 - 60 bss / km@xmath210 , @xmath226 within the @xmath1 range 60 - 300 bss / km@xmath210 and @xmath227 within the range 300 - 10000bss / km@xmath210 . the fact that the transmit power per base station decays more or less steeply with the density @xmath1 depends on how quickly the interference power increases or decreases with @xmath1 . as we explained in section [ sub : computing - the - transmit ] , the transmit power per base station @xmath168 has to be set so that the network is interference limited . thus , if the channel attenuation between the interferer and the user decreases quickly as the density increases , a lower transmit power will be enough to guarantee that the interference power is greater than the noise power . in other words , if the interferer - to - user channel attenuation tends to decrease quickly as the density increases , so does the transmit power and vice - versa . for instance , for @xmath228$]bss / km@xmath210 , the probability of having interferers in los with the user rises and , as a consequence , we have a lower attenuation of the channel between the interfering base station and the user . hence , the @xmath168 which guarantees the interference - limited regime will also decrease steeply with @xmath229 as @xmath1 increases . on the contrary , for @xmath230 bss / km@xmath231 , most of the interferers will have already entered the los zone , meaning that the interferer - to - user channel attenuation drops less rapidly than for @xmath232 bss / km@xmath231 ; for this reason , also @xmath168 will decrease less rapidly with @xmath233 . let us note that , with increasing reuse factors @xmath84 , the tx power decreases , as indeed a smaller bandwidth is used and , thus , the noise power is lower . one of the most surprising outcomes of our study on los / nlos propagation for ultra - dense networks is the effect of cell - densification on the energy efficiency of the fully loaded network , of which we show the results in fig . [ fig : energy_efficiency_v1 ] . the difference between the energy efficiency with single - slope and with los / nlos path - loss is noticeable . in the case of single - slope pl , due to the linear growth of the ase , @xmath164 is a monotonically increasing function of the density @xmath1 ( see section [ subsect : energy_eff_increasing ] ) . in the case of los / nlos propagation , from fig . [ fig : energy_efficiency_v1 ] we observe that the energy efficiency exhibits a maximum , which is achieved for a given density @xmath191 . to explain this , we consider the case frequency reuse @xmath234 ( solid - blue curve in fig . [ fig : energy_efficiency_v1 ] ) ; from and with the values of the parameters @xmath196 ( given in table [ table : parameters ] ) , @xmath224 and @xmath223 ( given in section [ sub : txpowerperbs_results ] ) , and @xmath185 ( given in section [ subsect : rate_ase_density ] ) , the optimal point @xmath191 is approximately 100bss / km@xmath210 . beyond this point , the ase gain is too low to compensate power consumption increase in the network , leading to a drop in terms of energy efficiency . from fig . [ fig : energy_efficiency_v1 ] , we can note that frequency reuse reduces the energy efficiency compared to @xmath234 . as a result of the lower ase achieved at higher frequency reuse factors @xmath84 , the energy efficiency drops as @xmath84 increases . in fig . [ fig : energy_efficiency_part_load ] we show the energy efficiency for partially loaded networks , for a user density @xmath18 of 1000 ues / km@xmath210 . as we are dealing with partially loaded networks , we are interested in the bs densities @xmath235 , where energy efficiency strongly depends on the power saving factor @xmath160 of the bss in stand - by state . this is because the parameter @xmath160 determines the energy saving of the inactive bss , which become more numerous as the density @xmath1 increases . depending on the value of @xmath160 , according to a local maximum may even occur at @xmath236 . with @xmath237 and with the values of @xmath185 given in section [ sub : partially_loaded_results ] , the local maximum turns to be @xmath238bss / km@xmath210 . for higher values of @xmath160 , @xmath207 is smaller than or too close to @xmath18 to be considered as a reliable estimate of a maximum ; we recall from section [ subsec : energy_eff_part_loaded ] that this estimate can be reckoned as reliable only if @xmath207 is sufficiently greater than @xmath18 . in fact , we observe from fig . [ fig : energy_efficiency_part_load ] that there is no local maximum beyond @xmath18 for @xmath239 or @xmath240 . in this paper , we have proposed a stochastic geometry - based framework to model the outage probability , the area spectral efficiency ( ase ) of fully loaded and partially loaded ultra - dense networks ( udns ) , where the signal propagation accounts for los and nlos components . we also studied the energy efficiency of udns resulting from this propagation model . as the main findings of our work , we have shown that , with los / nlos propagation , massive cell densification determines a deterioration of the network coverage at high cell densities , if the network is fully loaded . moreover , the ase grows less steeply than a linear function at high cell densities , which implies that a larger number of base stations would be required to achieve a given throughput target with respect to the case of single slope path - loss . in regards to the energy efficiency , cell densification turns out to be inefficient for the network from an energetic point of view . in partially loaded networks , when the base station density exceeds that of the users , cell densification results in a coverage improvement . overall , based on our findings , we can conclude that udns are likely face coverage issues in highly crowded environments with many users , which represent the worst case scenario for ultra - dense networks . n. bhushan , l. junyi , d. malladi , r. gilmore , d. brenner , a. damnjanovic , r. sukhavasi , and s. geirhofer , `` network densification : the dominant theme for wireless evolution into 5 g , '' _ ieee communications magazine _ , 2014 . j. park , s .- l . kim , and j. zander , `` asymptotic behavior of ultra - dense cellular networks and its economic impact , '' in _ ieee global communications conference ( globecom 2014 ) _ , dec . 2014 , pp . 49414946 . c. galiotto , n. k. pratas , n. marchetti , and l. doyle , `` a stochastic geometry framework for los / nlos propagation in dense small cell networks , '' _ ieee icc 2015 _ , 2015 . [ online ] . available : http://arxiv.org/abs/1412.5065v2 c. li , j. zhang , and k. b. letaief , `` throughput and energy efficiency analysis of small cell networks with multi - antenna base stations , '' _ ieee transactions on wireless communications _ , vol . 13 , no . 5 , pp . 2505 2517 , mar . 2014 . h. s. dhillon , r. k. ganti , and j. g. andrews , `` load - aware modeling and analysis of heterogeneous cellular networks , '' _ ieee transactions on wireless communications _ , 12 , no . 4 , pp . 16661677 , apr . 2013 . c. galiotto , i. m. gomez , n. marchetti , and l. doyle , `` effect of los / nlos propagation on area spectral efficiency and energy efficiency of small - cells , '' _ ieee globecom 2014 _ , 2014 . [ online ] . available : http://arxiv.org/abs/1409.7575	this paper aims at investigating the achievable performance and the issues that arise in ultra - dense networks ( udns ) , when the signal propagation includes both the line - of - sight ( los ) and non - line - of - sight ( nlos ) components . backed by an analytical stochastic geometry - based model , we study the coverage , the area spectral efficiency ( ase ) and the energy efficiency of udns with los / nlos propagation . we show that when los / nlos propagation components are accounted for , the network suffers from low coverage and the ase gain is lower than linear at high base station densities . however , this performance drop can partially be attenuated by means of frequency reuse , which is shown to improve the ase vs coverage trade - off of cell densification , provided that we have a degree of freedom on the density of cells . in addition , from an energy efficiency standpoint , cell densification is shown to be inefficient when both los and nlos components are taken into account . overall , based on the findings of our work that assumes a more advanced system model compared to the current state - of - the - art , we claim that highly crowded environments of users represent the worst case scenario for ultra - dense networks . namely , these are likely to face serious issues in terms of limited coverage . ultra - dense , los / nlos , area spectral efficiency , partially loaded , energy efficiency , coverage .
the lepton flavor violating ( lfv ) z decays are clean from the theoretical point of view since they are free from the long distance effects . on the other hand , they are rich in the sense that they exist at least at one loop level and carry a considerable information about the free parameters of the model used . therefore , it is worthwhile to analyze these decays and there is an extensive work related to them in the literature @xcite-@xcite . since the lepton flavor is conserved in the sm , for the flavor violation in the lepton sector , one needs to extend the sm . the so called @xmath3sm model , which is constructed by taking neutrinos massive and permitting the lepton mixing mechanism @xcite , is one of the candidate . however , the theoretical predictions for the branching ratios ( brs ) of the lfv z decays in this model are extremely small in the case of internal light neutrinos @xcite @xmath4 and they are far from the experimental limits obtained at lep 1 @xcite : @xmath5 and from the improved ones at giga - z @xcite : @xmath6 with @xmath7 . notice that these numbers are obtained for the decays @xmath8 , namely @xmath9 other possible scenarios to enhance the brs of the corresponding lfv z decays are the extension of @xmath3sm with one heavy ordinary dirac neutrino @xcite , the extension of @xmath3sm with two heavy right - handed singlet majorana neutrinos @xcite , the zee model @xcite , the model iii version of the two higgs doublet model ( 2hdm ) , which is the minimal extension of the sm @xcite , the supersymmetric models @xcite , top - color assisted technicolor model @xcite . in this work , we study the lfv processes @xmath0 , @xmath1 and @xmath2 in the framework of the 2hdm in the split fermion scenario @xcite . this scenario is based on the idea that the hierarchy of fermion masses are coming from the overlap of the fermion gaussian profiles in the extra dimensions and reached great interest in the literature @xcite-@xcite . in @xcite the explicit positions of left and right handed components of fermions in the extra dimensions have been predicted . @xcite is devoted to the restrictions on the split fermions in the extra dimensions using the leptonic w decays and the lepton violating processes . the cp violation in the quark sector has been analyzed in @xcite and to find stringent bounds on the size of the compactification scale 1/r , the physics of kaon , neutron and b / d mesons has been studied in @xcite . in @xcite the rare processes in the split fermion scenario and in @xcite the shapes and overlaps of the fermion wave functions in the split fermion model have been considered . in @xcite the electric dipole moments of charged leptons have been predicted , and in @xcite the radiative lfv decays have been examined , in the framework of this scenario . recently , the higgs localization in the split fermion models has been analyzed in @xcite . in our calculations we estimate the sensitivities of brs of the lfv z decays to the compactification scale and the gaussian widths of the charged leptons in the extra dimensions . we make the analysis including a single extra dimension and observe that the enhancement in the brs of these decays are small . however , in the case of two extra dimensions , especially the one that charged leptons are restricted to the fifth extra dimension , with non - zero gaussian profiles , there is a considerable enhancement in the brs of the decays under consideration , even more than one order for the compactification scale interval considered . the paper is organized as follows : in section 2 , we present the effective vertex and the brs of lfv z decays in the split fermion scenario . section 3 is devoted to discussion and our conclusions . in appendix section , we give the explicit expressions of the form factors appearing in the effective vertex . the extremely small theoretical values of the brs of the lfv z boson decays forces one to go beyond and search a new mechanism to enhance these numerical values near to the experimental limits . from the theoretical point of view , the existence of the flavor changing neutral currents ( fcncs ) is essential to create the lfv processes and , the multi higgs doublet models , which are constructed by extending the higgs sector of the sm , are among the possible models which permits the fcnc currents at tree level . the 2hdm is one of the candidate for the multi higgs doublet model and , in general , it permits the fcnc at tree level . the lfv z decay @xmath10 can be induced at least in the one loop level in the framework of the 2hdm and new higgs scalars play the main role for the large brs of these decays . furthermore , the inclusion of the spatial extra dimensions brings additional contributions to the physical quantities of the decays under consideration . here , we respect the idea that the hierarchy of lepton masses are coming from the lepton gaussian profiles in the extra dimensions , so called split fermion scenario . the lfv z decay @xmath10 exist with the help of the yukawa interactions and , in a single extra dimension , respecting the split fermion scenario , it reads @xmath11 where @xmath12 and @xmath13 denote chiral projections @xmath14 , @xmath15 is the new scalar doublet and @xmath16 are the flavor violating yukawa couplings in five dimensions and they are complex in general . the lepton fields @xmath17 ( @xmath18 ) are the zero mode lepton doublets ( singlets ) with gaussian profiles in the extra dimension which is represented by the coordinate @xmath19 and they can be defined as @xmath20 where @xmath21 ( @xmath22 ) are the lepton doublets ( singlets ) in four dimensions with family indices @xmath23 and @xmath24 , @xmath25 , satisfying the property @xmath26 , is the parameter representing the gaussian width of the leptons in five dimensions and n is the normalization factor , @xmath27 . here the coordinates @xmath28 and @xmath29 represent the positions of the peaks of left and right handed parts of @xmath30 lepton in the fifth dimension and they are obtained by assuming that the mass hierarchy of leptons are coming from the relative positions of the gaussian peaks of the wave functions located in the extra dimension @xcite . the observed lepton masses are the sources to calculate these coordinates and in @xcite one possible set of locations of the lepton fields in the fifth dimension has been estimated as @xmath31 now , we would like the present the higgs sector of the model under consideration . the higgs doublets @xmath32 and @xmath15 are chosen as @xmath33\ , ; \phi_{2}=\frac{1}{\sqrt{2}}\left(\begin{array}{c c } \sqrt{2 } h^{+}\\ h_1+i h_2 \end{array}\right ) \,\ , , \label{choice}\end{aligned}\ ] ] and the vacuum expectation values are , @xmath34 this choice brings the possibility that the sm ( new ) particles are collected in the first ( second ) doublet and the higgs fields @xmath35 and @xmath36 become the mass eigenstates @xmath37 and @xmath38 respectively since no mixing occurs between two cp - even neutral bosons @xmath39 and @xmath37 at tree level in this case . as it is seen in eq . ( [ lagrangian ] ) , the new higgs field @xmath15 is responsible for the lfv interaction at tree level . with the addition of extra dimensions , after the compactification on the orbifold @xmath40 , the new higgs field @xmath15 can be expanded as @xmath41 where @xmath42 ( @xmath43 ) is the higgs doublet in the four dimensions ( the kk modes ) including the charged higgs boson @xmath44 ( @xmath45 ) , the neutral cp even - odd higgs bosons @xmath37- @xmath38 ( @xmath46- @xmath47 ) . the non - zero @xmath48 kk mode of the charged higgs mass is @xmath49 , and the neutral cp even ( odd ) higgs mass is @xmath50 , ( @xmath51 ) , with the @xmath52th level kk particle mass @xmath53 . the @xmath10 decay exist at least at one loop level in the 2hdm with the help of the internal neutral higgs particles @xmath37 and @xmath38 . in fig . [ fig1ver ] the necessary 1-loop diagrams , the self energy and vertex diagrams , are given . with the inclusion of extra dimensions , there exists the additional contributions due to the kk modes of neutral higgs particles . at this stage one needs the lepton - lepton-@xmath54 ( @xmath55 ) interaction , which is modified in the case of the split fermion scenario and these vertex factors @xmath56 in the vertices @xmath57 , with the right ( left ) handed @xmath30 flavor lepton fields @xmath58 in five dimensions ( see eq . ( [ gaussianprof ] ) ) , are obtained by the integration over the fifth dimension . finally , the vertex factor for @xmath48 kk mode higgs fields s read @xmath59 \,\ , . \label{vij1}\end{aligned}\ ] ] for @xmath60 , this factor becomes @xmath61 and we define the yukawa couplings in four dimensions as @xmath62 now , we would like to present the general effective vertex for the interaction of on - shell z - boson with a fermionic current : @xmath63 where @xmath64 is the momentum transfer , @xmath65 , @xmath66 ( @xmath67 ) is vector ( axial - vector ) coupling , @xmath68 ( @xmath69 ) magnetic ( electric ) transitions of unlike fermions . here @xmath70 ( @xmath71 ) is the four momentum vector of lepton ( anti - lepton ) . the vector ( axial - vector ) @xmath66 ( @xmath67 ) couplings and the magnetic ( electric ) transitions @xmath68 ( @xmath69 ) including the contributions coming from a single extra dimension can be obtained as @xmath72 where @xmath73 are the couplings without scalar boson @xmath74 kk mode contributions and they can be obtained by taking @xmath60 in eq . ( [ fvame ] ) . on the other hand the couplings @xmath75 are the ones due to the kk modes of the scalar bosons @xmath74 ( see eq . ( [ fvame ] ) ) . here the summation over the index @xmath23 represents the sum due to the internal lepton flavors , namely , @xmath76 . we present @xmath75 in the appendix , by taking into account all the masses of internal leptons and external lepton ( anti - lepton ) . if we consider two extra dimensions where all the particles are accessible , the couplings @xmath75 appearing in eq . ( [ fvameex ] ) should be replaced by @xmath77 and they read @xmath78 where the summation would be done over @xmath79 except @xmath80 . ( see the appendix for their explicit forms ) . finally , the br for @xmath81 can be written in terms of the couplings @xmath66 , @xmath67 , @xmath68 and @xmath69 as @xmath82 where @xmath83 and @xmath84 is the total decay width of z boson . in our numerical analysis we consider the br due to the production of sum of charged states , namely @xmath85 since the lfv z decays @xmath86 , @xmath87 , exist at least in the one loop level in the 2hdm , the internal leptons and new scalar bosons drive the interaction and the physical quantities of these decays are sensitive to the yukawa couplings @xmath88 , which are among the free parameters of the model with the definition @xmath89 where n denotes the word `` neutral '' . ] . furthermore , respecting the split fermion scenario , which is based on the idea that the hierarchy of fermion masses are due to the fermion gaussian profiles in the extra dimensions , there arise new parameters , namely , the compactification radius , the fermion widths and their locations in the new dimesion(s ) . in this scenario , the yukawa couplings in four dimensions appear with a multiplicative exponential suppression factors after the integration of the extra dimension ( see eq . ( [ coupl4 ] ) ) . this factor is coming from the different locations of various flavors and their left and right handed parts of lepton fields , in the yukawa part of the lagrangian . here , we consider that the couplings @xmath90 are smaller compared to @xmath91 , respecting the sher scenario @xcite , since latter ones contain heavy flavors . furthermore , we assume that , in four dimensions , the couplings @xmath92 is symmetric with respect to the indices @xmath23 and @xmath24 and choose the appropriate numerical values for the yukawa couplings , by respecting the current experimental measurements . the upper limit of @xmath93 is predicted as @xmath94 ( see @xcite and references therein ) by using the experimental uncertainty , @xmath95 , in the measurement of the muon anomalous magnetic moment and assuming that the new physics effects can not exceed this uncertainty . using this upper limit and the experimental upper bound of br of @xmath96 decay , br @xmath97 @xcite , the coupling @xmath98 can be restricted in the range , @xmath99 ( see @xcite ) . for the yukawa coupling @xmath100 , we have no explicit restriction region and we use the numerical values which are greater than @xmath101 . our study is devoted to the prediction of the effects of the extra dimensions on the br of the lfv processes @xmath102 , in the split fermion scenario , in the framework of the 2hdm . the compactification scale @xmath103 , which is one of the free parameter of the model , should be restricted . in the literature , there exist numerous constraints for this scale , in the case of the single extra dimension , in the split fermion scenario : * @xmath104 due to the direct limits from searching for kk gauge bosons . * @xmath105 from @xmath106 , @xmath107 from @xmath108 and @xmath109 from the upper limit of the @xmath110 , @xmath111 @xcite . * a far more stringent limit @xmath112 @xcite coming from the precision electro weak bounds on higher dimensional operators generated by kk exchange in our numerical analysis , we choose an appropriate range for the compactification scale @xmath103 , by respecting these limits in the case of a single extra dimension . for two extra dimensions we used the same broad range for @xmath103 . throughout our calculations we use the input values given in table ( [ input ] ) . .the values of the input parameters used in the numerical calculations . [ cols= " < , < " , ] in the present work , we estimate the brs of the lfv z decays in a single and two extra dimensions , by considering split leptons with a possible set of locations . for a single extra dimension ( two extra dimensions ) we use the estimated location of the leptons given in eq . ( [ location ] ) ( eq . ( [ location2 ] ) ) to calculate the lepton - lepton - higgs scalar kk mode vertices . in the case of two extra dimensions we study two possibilities : the leptons are restricted to the fifth extra dimension , with non - zero gaussian profiles and the leptons have non - zero gaussian profiles also in the sixth dimension . in the former one , the enhancements in the brs of the present decays are relatively large due to the well known kk mode abundance of higgs fields . however , in the latter the additional exponential factor appearing in the second summation further suppresses the brs . [ zmuerr ] is devoted to the compactification scale @xmath103 dependence of the br @xmath113 for @xmath114 , @xmath115 and @xmath116 . here the solid ( dashed , small dashed , dotted ) line represents the br without extra dimension ( with a single extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in the fifth extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in both extra dimensions ) . it is observed that br is at the order of the magnitude of @xmath117 without extra dimensions and it is the weakly sensitive to the parameter @xmath103 , for the @xmath118 , for a single extra dimension . in the case of two extra dimensions , the enhancement of the br is relatively larger , near one order of magnitude , for @xmath119 , due to the higgs scalar kk mode abundances . however , these contributions do not increase extremely due to the suppression exponential factor appearing in the summations . the enhancement in the br becomes weak for @xmath120 . furthermore , the numerical values of brs are slightly greater in the case that the leptons have non - zero gaussian profiles only in the fifth extra dimension . in fig . [ ztauerr ] , we present the compactification scale @xmath103 dependence of the br @xmath121 for @xmath114 , @xmath122 and @xmath123 . here the solid ( dashed , small dashed , dotted ) line represents the br without extra dimension ( with a single extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in the fifth extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in both extra dimensions ) . this figure shows that the br is at the order of the magnitude of @xmath124 without extra dimensions . similar to the previous decay , the br is weakly sensitive to the parameter @xmath103 , for the @xmath118 , for a single extra dimension . in the case of two extra dimensions , the enhancement of the br is almost one order larger than the one with a single extra dimension , for @xmath119 and this enhancement becomes weak for @xmath120 . [ ztaumurr ] represents the compactification scale @xmath103 dependence of the br @xmath125 for @xmath126 , @xmath122 and @xmath123 . here the solid ( dashed , small dashed , dotted ) line represents the br without extra dimension ( with a single extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in the fifth extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in both extra dimensions ) . for this decay , the br is observed at the order of the magnitude of @xmath127 without extra dimensions . the br is weakly sensitive to the parameter @xmath103 , for the @xmath118 , for a single extra dimension . in the case of two extra dimensions , the enhancement of the br is more than one order larger than the one with a single extra dimension , for @xmath119 and this enhancement becomes also weak for @xmath120 . now we would like to estimate the sensitivity of the brs of the z decays under consideration to the gaussian widths , @xmath128 of leptons , where @xmath129 is the free parameter which regulates the amount of width in the extra dimension . [ zmuero ] ( [ ztauero ] * ; * [ ztaumuro ] ) shows the parameter @xmath129 dependence of the br of the decay @xmath130 ( @xmath131 * ; * @xmath125 ) for @xmath132 , and the real couplings @xmath133 , @xmath134 ( @xmath135 , @xmath136 * ; * @xmath135 , @xmath137 ) . here the solid ( dashed , small dashed , dotted ) line represents the br without extra dimension ( with a single extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in the fifth extra dimension , with two extra dimensions where the leptons have non - zero gaussian profiles in both extra dimensions ) . it is observed that the br of the decay @xmath130 ( @xmath131 * ; * @xmath125 ) increases almost @xmath138 ( @xmath138 * ; * @xmath138 ) for @xmath139 and reaches @xmath140 ( @xmath141 * ; * @xmath141 ) for @xmath142 , in the case of a single extra dimension . this shows that the sensitivities of the brs of the present decays to the parameter @xmath129 are not weak , especially for its small values . for two extra dimensions , these sensitivities increases considerably . for the decay @xmath143 ( @xmath121 * ; * @xmath125 ) , the br enhances more than one order ( more than one order * ; * @xmath144 ) for the intermediate values of the parameter @xmath129 , in the case of two extra dimensions where the leptons have non - zero gaussian profiles in the fifth extra dimension . this enhancement reaches @xmath145 ( @xmath145 * ; * @xmath145 ) for @xmath142 . in the case that the leptons have non - zero gaussian profiles in both extra dimensions , the br enhances less than one order ( less than one order * ; * more than one order ) , for the intermediate values of the parameter @xmath129 , for the decay @xmath143 ( @xmath146 * ; * @xmath147 ) . it is shown that the sensitivities of the brs of studied lfv decays are considerably large for two extra dimensions and the brs enhances more than one order , especially for the lfv z decays with heavy lepton flavors . as a summary , the br is weakly sensitive to the parameter @xmath103 for @xmath118 for a single extra dimension , however , there is an enhancement in the br , even more than one order for the scale @xmath148 . this enhancement decreases with the increasing values of the scale @xmath103 . furthermore , the br is sensitive to the parameter @xmath129 especially for two extra dimensions case . therefore , the lfv z decays are worthwhile to study and with the help of the forthcoming more accurate experimental measurements of the these decays , especially the @xmath149 one , the valuable information can be obtained to detect the effects due to the extra dimensions in the case of split fermion scenario . this work has been supported by the turkish academy of sciences in the framework of the young scientist award program . ( eoi - tuba - gebip/2001 - 1 - 8 ) here we present the explicit expressions for @xmath150 , @xmath151 , @xmath152 and @xmath153 @xcite ( see eq . ( [ fvameex ] ) ) : @xmath154 where @xmath155 and @xmath156 the parameters @xmath157 and @xmath158 are @xmath159 and @xmath160 and the masses @xmath161 read @xmath162 , where @xmath13 is the compactification radius . in eq . ( [ etava ] ) the flavor changing couplings @xmath163 represent the effective interaction between the internal lepton @xmath23 , ( @xmath164 ) and outgoing ( incoming ) @xmath165 one . the parameters @xmath166 , @xmath167 read @xmath168\ , \ , , \nonumber \\ c'_n \,(f , i)&=&\cos[\frac{n\,(y_{f l}+y_{i r})}{2\ , r}]\ , . \label{coeff}\end{aligned}\ ] ] in the case of two extra dimensions which all the particles feel , the parameters @xmath166 and @xmath167 are replaced by @xmath169\ , \ , , \nonumber \\ c'_{(n , s)}(f , i)&=&\cos[\frac{n\,(y_{f l}+y_{i r } ) + s\,(z_{f l}+z_{i r})}{2\ , r}]\ , , \label{coeff22}\end{aligned}\ ] ] and the exponential factor @xmath170 becomes @xmath171 . furthermore , the masses @xmath161 are replaced by @xmath172 , @xmath173 , with @xmath174 . here we use a possible positions of left handed and right handed leptons in the two extra dimensions , by using the observed masses . with the assumption that the lepton mass matrix is diagonal , one of the possible set of locations for the gaussian peaks of the lepton fields in the two extra dimensions can be obtained as @xcite @xmath175 where the numbers in the parenthesis denote the y and z coordinates of the location of the gaussian peaks of lepton flavors in the extra dimensions . here we choose the same numbers for the y and z locations of the gaussian peaks . finally , the couplings @xmath176 may be complex in general and they can be parametrized as @xmath177 where @xmath178 denote the lepton flavors and @xmath179 are cp violating parameters which are the possible sources of the lepton edm . however , in the present work we take these couplings real . 1 t. riemann and g. mann , `` nondiagonal @xmath180 decay : @xmath181 '' , in _ proc . of the int . neutrino82 , 14 - 19 june 1982 , balatonfred , hungary _ ( a. frenkel and e. jenik , eds . ) , vol . ii , pp . 58- , budapest , 1982 , scanned copy at http://www.ifh.de/@xmath182riemann , g. mann and t. riemann , _ annalen phys . _ * 40 * ( 1984 ) 334 . j. i. illana , m. jack and t. riemann , _ hep - ph/0001273 _ ( 2000 ) , j. i. illana , and t. riemann , _ phys . * d63 * 053004 ( 2001 ) . collaboration , c. caso _ et al . _ , _ eur . phys . j. _ * c3 * ( 1998 ) 1 . collaboration , r. akers _ et al . _ , _ z. phys . _ * c67 * ( 1995 ) 555 . collaboration , o. adriani _ _ , _ phys . lett . _ * b316 * ( 1993 ) 427 . collaboration , p. abreu _ et al . _ , _ z. phys . _ * c73 * ( 1997 ) 243 . g. wilson , `` neutrino oscillations : are lepton - flavor violating z decays observable with the cdr detector ? '' and `` update on experimental aspects of lepton - flavour violation '' , talks at desy - ecfa lc workshops held at frascati , nov 1998 and at oxford , march 1999 , transparencies obtainable at + http://wwwsis.lnf.infn.it/talkshow/ and at + http://hepnts1.rl.ac.uk/ecfa_desy_oxford/scans/0025_wilson.pdf . a. ghosal , y. koide and h. fusaoka , _ phys.rev . _ * d64 * ( 2001 ) 053012 . e. o. iltan , i. turan , _ phys.rev . _ * d65 * ( 2002 ) 013001 ; e. o. iltan , _ eur . phys . j _ * c41 * ( 2005 ) 233 . j. illiana , m. masip , _ phys . rev . _ * d67 * ( 2003 ) 035004 . j. cao , z. xiong , j. mi . yang , _ eur . phys . j. _ * c32(2004 ) 245 . c. yue , w. wang , f. zhang , _ j. phys . _ g30 * ( 2004 ) 1065 . m. a .perez , g. t. velasco , j. j. toscano , _ int.j.mod.phys . _ * a19 * ( 2004 ) 159 ; a. flores - tlalpa , j.m . hernandez , g. tavares - velasco , j.j . toscano , _ phys . * d65 * ( 2002 ) 073010 , r. hawkings and k. mnig , _ eur . direct _ * c8 * ( 1999 ) 1 ; v. ganapathi , t. weiler , e. laermann , i. schmitt , and p. zerwas , _ phys . rev . _ * d27 * ( 1983 ) 579 , m. clements , c. footman , a. kronfeld , s. narasimhan , and d. photiadis , _ phys . _ * d27 * ( 1983 ) 570 . b. pontecorvo , _ zh . eksp . _ * 33 * ( 1957 ) 549 , z. maki , m. nakagawa , and s. sakata , _ prog . _ * 28 * ( 1962 ) 870 , b. pontecorvo , _ sov . * 26 * ( 1968 ) 984 . n. arkani - hamed and m. schmaltz , _ phys . rev . _ * d61 * , 033005 ( 2000 ) ; n. arkani - hamed , y. grossman and m. schmaltz , _ phys . * d61 * ( 2000 ) 115004 . e. a. mirabelli , schmaltz , _ phys . _ * d61 * ( 2000 ) 113011 . w. f. chang , i. l. ho and j. n. ng , _ phys . rev . _ * d66 * 076004 ( 2002 ) . g. c. branco , a . gouvea , m. n. rebelo , _ phys . * b506 * ( 2001 ) 115 . w. f. cang , j. n. ng , _ jhep _ * 0212 * ( 2002 ) 077 . b. lillie , j. l. hewett , _ phys . _ * 68 * ( 2003 ) 116002 . y. grossman and g. perez , _ phys . _ * d67 * ( 2003 ) 015011 ; _ pramana _ * 62 * ( 2004 ) 733 . e. o. iltan , hep - ph/0503001 ( 2005 ) . e. o. iltan , hep - ph/0504013 ( 2005 ) . a. delgado , a. pomarol , m. quiros , _ jhep _ * 0001 * ( 2000 ) 030 ; y. grossman , _ int . phys . _ * a15 * ( 2000 ) 2419 ; d. e. kaplan and t. m. tait , _ jhep _ * 0111 * ( 2001 ) 051 ; g. barenboim , _ et . _ , _ phys . rev . _ * d64 * ( 2001 ) 073005 ; w. f. chang , i. l. ho and j. n. ng , _ phys . _ * d66 * ( 2002 ) 076004 ; w. skiba and d. smith , _ phys . * d65 * ( 2002 ) 095002 ; y. grossman , r. harnik , g. perez , m. d. schwartz and z. surujon , _ phys . rev . _ * d71 * ( 2005 ) 056007 ; p. dey , g. bhattacharya , _ phys . rev . _ * d70 * ( 2004 ) 116012 ; y. nagatani , g. perez , _ jhep _ * 052 * ( 2005 ) 068 ; r. harnik , g. perez , m. d. schwartz , y. shirman , _ jhep _ * 0503 * ( 2005 ) 068 ) ; c. yue , d. yu , l. liu , _ phys . rev . _ * d69 * ( 2004 ) 095003 ; c. yue , h. li , y. zhang . _ phys * b536 * ( 2002 ) 67 . z. surujon , hep - ph/0507036 . t. p. cheng and m. sher , _ phy . _ * d35 * ( 1987 ) 3383 . e. o. iltan , h. sundu , _ acta phys.slov._ * 53 * ( 2003 ) 17 . m. l. brooks et . mega collaboration , _ phys . lett . _ * 83 * , 1521 ( 1999 ) . e. o. iltan , _ phys . rev . _ * d64 * , 115005 , ( 2001 ) ; _ phys . _ * d64 * 013013 , ( 2001 ) . t. g. rizzo , j. d. wells , _ phy . rev . _ * d61 * ( 2000 ) 016007 .	we predict the branching ratios of @xmath0 , @xmath1 and @xmath2 decays in the framework of the 2hdm , in the split fermion scenario . we observe that the branching ratios are not sensitive to a single extra dimension , however , this sensitivity is considerably large for two extra dimensions .
over the last couple of decades , simulation science has become as important as theoretical and experimental science . the success of simulation science hinges on the ability to perform the calculations efficiently . the inner most kernel in these calculations is often the solution of algebraic systems , including , but not limited to , systems of linear and nonlinear equations , eigenvalue problems , optimization problems , and sensitivity analysis . tops , which stands for _ towards optimal petascale simulations _ , is a multi - institutional scidac applied math center that focuses on the development of solvers for tackling these algebraic systems , as well as the deployment of such technologies in large - scale scientific applications , particularly those of interest to the u.s . department of energy . in this paper , we highlight two specific areas of tops : eigenvalue calculations and optimization . in particular , we highlight some accomplishments we have achieved in collaboration with computational physicists in unedf . the goal of the unedf scidac application project @xcite is to obtain a comprehensive understanding of nuclei and their reactions based on the most accurate knowledge of the strong nuclear interaction . eigenvalue calculations come up in the solution of the nuclear schrdinger equation @xcite . the eigenvalues and the eigenvectors correspond to the energy states and wave functions . numerical optimization techniques are needed in building the next generation of nuclear energy functionals , which will provide nuclear physicists better tools for predicting the properties and behavior of atomic nuclei over the entire nuclear table . in nuclear configuration interaction calculation , it is sometimes necessary to investigate , among others , nuclear level densities as a function of the total angular momentum @xmath0 and excitation energy , and to evaluate scattering amplitudes as a function of @xmath0 @xcite . we will refer to this as a _ total - j calculation _ in this paper . in this type of calculation , we are interested in computing a relatively large number of states with a prescribed @xmath0 value . one brute - force approach to a total - j calculation is to simply compute a large number of eigenvalues and wave functions of a nuclear many - body hamiltonian , for example in an m - scheme basis ( good angular momentum projection along the z - axis ) , and select , among these wave functions , the ones that have a prescribed @xmath0 value . this approach is appropriate when the number of desired energy states and wave functions is small ( e.g. , ten to twenty states ) . when that is not true , or when certain properties of a nucleus pertaining to a fixed @xmath0 are to be calculated , the brute - force approach may require computing a very large number of wave functions , and the computational cost for performing this type of calculation may be prohibitively high . furthermore , even if we can afford to perform this type of calculation , this may not be an efficient use of resources because we compute a large number of wave functions only to throw away most of them because they do not have the desired @xmath0 value . we have developed an alternative approach where we construct an invariant subspace @xmath1 that contains all wave functions associated with a fixed @xmath0 value in advance and project the nuclear many - body hamiltonian into this subspace to produce a projected hamiltonian with the minimum dimension consistent with that chosen @xmath0 . a sparse matrix diagonalization procedure @xcite is then applied to this projected hamiltonian to obtain the desired energy states and their corresponding wave functions . to construct @xmath1 , we need to work with the total angular momentum square operator @xmath2 and compute the null space of @xmath3 , where @xmath4 is a known eigenvalue of @xmath2 . when the many - body basis states associated with the configuration space are properly ordered and grouped , @xmath2 becomes block diagonal : @xmath5 therefore , the task of computing the desired null space of @xmath6 reduces to that of computing the desired null spaces of @xmath7 , for @xmath8 . however , because the dimensions of the @xmath9 s vary over a wide range ( e.g. , from 1 to more than 36,000 for @xmath10c , @xmath11 ) , it is difficult to maintain a good load balance in the null space calculation . here , @xmath12 is a parameter limiting the total number of oscillator quanta allowed in the many - body states . we developed a multi - level task and data distribution scheme to achieve optimal parallel performance in the null space calculation by 1 . limiting the granularity of the parallelism ; that is , we try to divide the overall task into many small tasks of limited sizes so that good load balance arises from distributing these small tasks evenly among different processors . 2 . limiting the communication overhead incurred in the null space calculation so that the overall time of the computation can be minimized . to achieve these inherently conflicting goals , we classified @xmath9 blocks into small , medium and large groups based on the estimated computational loads associated with computing the desired null space of @xmath13 , and the estimated ratio of communication volume to floating point operations count . the small @xmath9 blocks are distributed among all processors based on their computational load by a greedy algorithm . the null spaces of these matrices are computed by a sequential rank - revealing qr subroutine . no communication is involved in these calculations . each one of the medium - sized @xmath9 blocks is assigned to a subgroup of processors by the same greedy algorithm . the null space calculation for such a block is parallelized among processors within the same subgroup , which will incur some communication overhead . finally , the desired null space calculation for a large @xmath9 block is carried out in parallel on all processors . we implemented three different algorithms for computing the null space of @xmath7 for medium and large blocks . 1 . randomized rank - revealing qr ( rqr ) . the algorithm performs two standard qr factorizations of dense matrices without pivoting . although we do not take advantage of the sparsity of @xmath9 in this approach , it is more efficient than other approaches when the dimension of the desired null space is relatively large ( e.g. larger than 10% of the dimension of @xmath9 ) . shift - invert lanczos ( sil ) , which requires solutions of sparse linear systems . 3 . polynomial accelerated subspace iteration ( pasi ) . we apply a standard subspace iteration @xcite to the matrix @xmath14 , where @xmath15 is a polynomial that assumes the value of 1 at @xmath16 , and has a much smaller magnitude ( than 1 ) in other parts of the spectrum of @xmath9 . table [ tab : dist_comparison ] shows that our load balance scheme is much better than a brute - force approach of distributing @xmath9 in a cyclic fashion to different processors . table [ tab : compare_methods ] shows that pasi is more efficient when j=0 . the randomized qr algorithm appears to be more efficient for larger j values . however , when j becomes very large , which typically results in smaller dimension of the null space , pasi becomes more efficient again . [ tab : dist_comparison ] .rqr decomposition vs pasi for different @xmath0 values . both methods use the greedy load balancing technique . times are in seconds . [ cols="^,^,^,^,^,^ " , ] optimization plays a central role in the building of next - generation nuclear energy functionals , including functionals based on density functional theory ( dft ) and/or ab initio calculations . in the case of dft - based functionals , for example , a primary computational bottleneck is determining parameter values so that the functional agrees with data on a set of observables such as binding energies , radii , and odd - even staggering @xcite . mathematically , we need to solve the optimization problem @xmath17 where @xmath18 parameter values must be determined from a set of data of @xmath19 observables . challenges in solving this problem include the computational expense of , and the noise resulting from , the iterative calculations performed when simulating the theoretical observable @xmath20 , and the fact that derivatives of some simulated observables with respect to certain parameters @xmath21 may not be available ( or even exist ) for use by an optimization algorithm . quantifying the computational noise for a deterministic simulator across 2,049 nuclei . ] quantifying the computational noise for a deterministic simulator across 2,049 nuclei . ] as part of scidac efforts , we have developed pounders ( practical optimization using no derivatives for sums of squares ) , an algorithm for derivative - free optimization of nonlinear least - squares problems such as ( [ eq : chi2 ] ) . a key benefit of pounders is that it works with the individual residuals @xmath22 rather than the aggregated fit function @xmath23 . as a result , pounders can take advantage of the availability of the derivatives of some observables ( e.g. , binding energies ) and can approximate nonlinearities in @xmath24 using simulations at fewer @xmath25 values . as part of the tops collaboration , pounders is now available through the open - source toolkit for advanced optimization ( tao ) @xcite . figure [ fig : pounder ] quantifies the computational savings in this ability to exploit the sums of squares structure in ( [ eq : chi2 ] ) for a fit to 2,049 binding energies . by working with the residuals , the pounders variants obtain far better fits in far fewer evaluations than the analogous variants of pounder , a similar algorithm that does not have access to the residuals . warm _ variants illustrate the benefit of using external simulations , done as part of an initial experimental design , to warm start the optimization . the savings in figure [ fig : pounder ] can be substantial . for the more complex functional optimized in @xcite , each evaluation of @xmath24 requires 14.4 cpu hours . the resulting parameterization is then used to perform a simulation of nuclei across the nuclear table in a calculation requiring 9,000 processors for more than half a day @xcite . mathematical work has also contributed to the sensitivity analysis of nuclear energy functionals . though the simulations are typically deterministic , the aforementioned computational noise can obfuscate the number of reliable digits in computed functional properties . the ecnoise algorithm described in @xcite estimates a standard deviation - like quantity using only a few simulations . figure [ fig : noise ] illustrates the relative noise in the computed binding energies for 2,049 nuclei with the parameterization obtained from the pounders optimization in figure [ fig : pounder ] . estimates of the noise can , for example , reveal limitations on the predictability of computed functional observables and can enable stable approximations of the noisy derivatives needed for sensitivity analysis . our work on eigenvalue calculations has made several impacts on nuclear structure calculations . earlier collaborations with nuclear physicists led to significant improvements to an eigensolver for configuration interaction calculation , which was subsequently used in predicting the properties of @xmath26f before the isotope was observed experimentally . the work described in this paper takes configuration interaction calculation one step further . it enables our physics collaborators to efficiently compute energy states of a nuclei with a prescribed total angular momentum instead of computing many energy states and identifying those corresponding to the prescribed total angular momentum . the optimal parameters we have delivered to our physics collaborators have resulted in realistic functionals that are now being explored by a variety of groups outside of the unedf collaboration ( as evidenced in the most recent justipen conference ( http://massexplorer.org/justipen/index.php ) ) . for example , our current results show remarkable power for predicting fission barrier heights , which is a first step toward a microscopic understanding of fission . these results are a consequence of including a richer set of experimental data and more free parameters , resulting in problems that can be solved only by an efficient , state - of - the - art optimization algorithm . 5 http://www.unedf.org . p. sternberg , e.g. ng , c. yang , p. maris , j.p . vary , m. sosonkina , and h.v . le . accelerating configuration interaction calculations for nuclear structure . in _ proceedings of the 2008 acm / ieee conference on supercomputing _ ( austin , texas , november 15 - 21 , 2008 ) . conference on high performance networking and computing . ieee press , piscataway , nj , 1 - 12 . lehoucq , d.c . sorensen , p. vu , and c. yang . : _ an implementation of the implicitly re - started arnoldi iteration that computes some of the eigenvalues and eigenvectors of a large sparse matrix _ , 1995 . available from ftp.caam.rice.edu under the directory pub / software / arpack .	at the heart of many scientific applications is the solution of algebraic systems , such as linear systems of equations , eigenvalue problems , and optimization problems , to name a few . tops , which stands for towards optimal petascale simulations , is a scidac applied math center focused on the development of solvers for tackling these algebraic systems , as well as the deployment of such technologies in large - scale scientific applications of interest to the u.s . department of energy . in this paper , we highlight some of the solver technologies we have developed in optimization and matrix computations . we also describe some accomplishments achieved using these technologies in unedf , a scidac application project on nuclear physics .
let @xmath5 be a graph with vertices @xmath6 and @xmath7 edges . @xmath5 is a _ minimally rigid graph _ ( also called _ laman graph _ ) if @xmath8 and every subset of @xmath9 vertices spans at most @xmath10 edges . an embedding of the graph @xmath11 on a set of points @xmath12 is a mapping of the vertices to points in the euclidean plane @xmath13 . the edges @xmath14 of @xmath5 are mapped to straight line segments @xmath15 . an embedding is _ planar _ or _ non - crossing _ if no pair of segments @xmath16 and @xmath17 corresponding to non - adjacent edges @xmath18 have a point in common . an embedded laman graph on the _ generic _ point sets is called _ laman framework _ and has the special property of being _ minimally rigid _ @xcite , when viewed as bar - and - joint _ frameworks _ with fixed edge - lengths , which motivates the tremendous interest in their properties . let @xmath19 be a set of non - crossing edge ( bar ) on @xmath20 . a laman framework containing @xmath19 is called _ f - constrained_. in this paper we give an algorithm for enumerating all the _ f - constrained non - crossing laman frameworks_. in this paper we only consider planar embedded laman graphs , namely laman frameworks . then we simply denote a vertex @xmath21 by @xmath22 and an edge @xmath15 by @xmath14 . * novelty . * to the best of our knowledge , this is the first algorithm proposed for enumerating ( without repetitions , in polynomial time and without using additional space ) all the @xmath19-constrained non - crossing laman frameworks . we achieve @xmath3 time per graph in @xmath4 space by using reverse search . for the unconstrained case , using a different method , the same time and space complexity results we obtained in @xcite . the method presented here uses matroid property and is considerably simpler than @xcite . * historical perspective . * the reverse search enumeration technique of avis and fukuda @xcite has been successfully applied to a variety of combinatorial and geometric enumeration problems . the necessary ingredients to use the method are an implicitly described connected graph on the objects to be generated , and an implicitly defined spanning tree in this graph . in this paper we supply these ingredients for the problem of generating laman frameworks . relevant to the historical context of our work are the results of bereg @xcite using reverse search combined with data - specific lexicographic orderings to enumerate triangulations and pointed pseudo - triangulations of a given point set . we notice in passing that there exist several other algorithms for enumerating ( pseudo-)triangulations @xcite , but they are based on different techniques . also relevant is the pebble game algorithm of jacobs and hendrickson @xcite for 2-dimensional rigidity , see also @xcite . our complexity analysis relies on the recent results , due to lee , streinu and theran @xcite , regarding the detailed data - structure complexity of finding and maintaining rigid components during the pebble game algorithm . indeed , the time - space trade - off of our algorithm is inherited from @xcite . * related previous works . * to better put our problem ( and solution ) in context and relate it to previous work , we briefly discuss now the difference between generating arbitrary laman frameworks , as opposed to just pointed pseudo - triangulations . a _ pointed pseudo - triangulation _ is a special case of a non - crossing laman framework @xcite , where every vertex in the embedding is incident to an angle larger than @xmath23 . pseudo - triangulations are connected via simple _ flips _ , in which the removal of any non - convex - hull edge leads to the choice of a _ unique _ other edge that can replace it , in order to restore the pointed pseudo - triangulation property . the flip graph of all pointed pseudo - triangulations is a connected subgraph of the graph of all the laman frameworks . in fact , it is the one - skeleton of a polytope @xcite , and the reverse search technique can be directly applied to it . bereg s efficient algorithm makes use of specific properties of pointed pseudo - triangulations which do not extend to arbitrary non - crossing laman frameworks . in particular , remove - add flips are _ not unique _ , relative to the removed edge , in the case of a non - crossing laman framework . in our previous paper @xcite we showed that a set of all non - crossing laman frameworks on a point set is connected via these flips . the proof relies on some properties of one - degree - of - freedom mechanisms which is obtained when removing one edge from the laman frameworks , which can not extend to edge constrained case . in this paper we use triangulation as an important tool . we use the fact that ( constrained ) non - crossing laman frameworks form the bases of a matroid defined on any triangulation of @xmath20 to define a parent function . the edge constrained case appears in an application to structural engineering as will be seen below , and the proposed algorithm will much reduce the output size compared with the unconstrained case . therefore , we can enumerate non - crossing laman frameworks for previously unsolved real problems . * motivating application . * we describe now briefly how this problem came to our attention via the work of the third author . graph theoretical approaches are widely used in _ structural mechanics _ @xcite , where the edges and vertices in the graph represent the bars and rotation - free joints of a structure called a _ truss_. it is well - known that the stiffest truss under static loads is statically determinate that is equivalent to a laman graph @xcite . another bar - joint system , which is widely used in industrial application , is a link mechanism that is unstable and generates large deformation or changes the direction of the nodal displacement . the applications of link mechanisms can be found in , e.g. , suspension of automobiles , robot hands , umbrellas , crank shafts , etc . kawamoto et al.@xcite presented a method by enumeration of the planar graphs to find an optimal mechanism . however , their method is developed for their specific problem , and no general approach was given . recently , a new type of mechanism called compliant mechanism has been developed and applied mainly in the field of micro - mechanics . a compliant mechanism has flexible parts to stabilize the structure , which is contrary to the conventional unstable mechanism . although a compliant mechanism is usually modelled as a continuum with elastic joints , it is possible to generate the similar mechanism by a bar - joint system . ohsaki and nishiwaki @xcite presented a method for generating flexible bar - joint mechanisms using nonlinear programming approach , and found that the optimal structure is statically determinate , i.e. minimally rigid . they utilized snapthrough behaviour to generate multi - stable mechanism that has multiple self - equilibrium states . such mechanism can be used for switching device , robot hand , gripper , deployable structure , etc . in their method , the optimal locations of bars and joints are found from the highly connected initial structure that has bars between all the pairs of the nodes whose distances are small enough . combining an implementation of our our earlier method for generating unconstrained laman frameworks @xcite with this nonlinear programming approach , we could obtain many new compliant mechanisms with up to 10 joints @xcite . however the number of laman frameworks grows too rapidly to allow a complete enumeration for significantly larger examples . in view of practical requirements , the optimal structure should not have intersecting members , and some pre - selected members should always exist . therefore , the computational cost may be much reduced if the candidate set of statically determinate non - crossing trusses ( non - crossing laman frameworks ) are first enumerated . thus , it is desirable to enumerate laman frameworks for which the existing edges are specified , and this is the problem considered in the current paper . let @xmath24 be a non - crossing laman framework on a given point set @xmath20 . a _ mechanism _ is a flexible framework obtained by removing one or more edges from a laman framework . its _ number of degrees of freedom _ or _ dof _ s , is the number of removed edges . we will encounter mostly _ one - degree - of - freedom ( 1dof ) mechanisms _ , which arise from a laman framework by the removal of one edge . in particular , a mechanism with @xmath25 dofs has exactly @xmath26 edges , and each subset of @xmath27 vertices spans at most @xmath10 edges . a subset of some @xmath27 vertices spanning _ @xmath10 edges is called a _ rigid block_. a _ maximal block _ is called a _ rigid component _ or a _ body_. the laman frameworks on the generic point set with @xmath0 points form the set of _ bases _ of the _ generic rigidity matroid _ ( or simply called _ laman matroid _ ) on @xmath28 , see @xcite . the bases have all the same size @xmath29 . bases may be related via the _ base exchange _ operation , which we will call a _ flip _ between two laman frameworks . two laman frameworks @xmath30 and @xmath31 are connected by a flip if their edge sets agree on @xmath32 elements . the flip is given by the pair of edges @xmath33 not common to the two bases , @xmath34 , @xmath35 . using flips , we can define a graph whose nodes are _ all _ the laman frameworks on @xmath0 vertices , and whose edges correspond to flips . it is well - known that the graph whose nodes are the bases of a matroid connected via flips , is connected . but a priori , the subset of _ ( constrained ) non - crossing laman frameworks _ may not necessarily be . we will prove this later in section [ sec : main ] . reverse search is a memory efficient method for visiting all the nodes of a connected graph that can be defined implicitly by an adjacency oracle . it can be used whenever a spanning tree of the graph can be defined implicitly by a _ parent _ function . this function is defined for each vertex of the graph except a pre - specified _ root_. iterating the parent function leads to a path to the root from any other vertex in the graph . the set of such paths defines a spanning tree , known as the _ search tree_. let @xmath36 be a triangulation on a given set of @xmath0 points @xmath20 in the plane containing @xmath25 triangles . the angle vector of @xmath36 is the vector of @xmath37 interior angles sorted into non - decreasing order . let @xmath19 be a non - crossing edge set on @xmath20 . an _ f - constrained triangulation _ on @xmath20 is one that contains @xmath19 as a subset . many facts about these triangulations are contained in the survey by bern and eppstein @xcite . if @xmath19 is an independent set in the laman matroid on @xmath28 , then a laman framework containing @xmath19 is called _ @xmath19-constrained_. the following lemma clearly holds from the known fact about matroids ( see , e.g.@xcite ) . [ lemma : minor ] let @xmath19 be a non - crossing edge set on @xmath20 that is an independent set in the laman matroid on @xmath28 . let us fix @xmath19-constrained triangulation @xmath36 on @xmath20 . then a set of bases of the laman matroid on @xmath36 whose edge set contains @xmath19 forms matroid . [ lemma : trilam ] let @xmath19 be a non - crossing edge set on @xmath20 that is an independent set in the laman matroid on @xmath28 . every @xmath19-constrained triangulation @xmath36 on @xmath20 contains an @xmath19-constrained laman framework . since @xmath36 is statically rigid , it is generically rigid , and hence contains a base @xmath38 of the laman matroid ( see , e.g. whiteley @xcite ) . since @xmath19 is independent in the laman matroid , it can be extended to a base by adding edges from @xmath39 . two points @xmath40 and @xmath41 are _ visible _ if no edge of @xmath19 properly intersects the segment @xmath42 . @xmath42 is _ visible _ to point @xmath43 if the triangle @xmath44 is not properly intersected by an edge of @xmath19 . ( cdt ) an f - constrained delaunay triangulation ( cdt ) contains the edge @xmath42 between points @xmath40 and @xmath41 in @xmath20 if and only if @xmath40 is visible to @xmath41 , and some circle through @xmath40 and @xmath41 contains no point of @xmath20 visible to segment @xmath42 . we call @xmath42 a d(elaunay)-edge . ( definition 1 , @xcite ) if @xmath20 has four or more co - circular points , using a linear transformation as described in @xcite , we may transform @xmath20 into a point set @xmath45 with a unique cdt . @xmath20 and @xmath45 have the same non - crossing laman frameworks . then we will assume in what follows that @xmath20 has a unique cdt . ( d - flip ) let @xmath46 be an edge of @xmath47 which is the diagonal of a convex quadrilateral @xmath48 contained in @xmath36 . the edge flip which replaces @xmath46 by edge @xmath49 is a d - flip if the circumcircle of the triangle @xmath44 contains the point @xmath50 in its interior . ( equivalently the circumcircle of the triangle @xmath51 contains the point @xmath41 in its interior . ) we call @xmath46 a _ @xmath19-illegal edge _ , otherwise it is called _ @xmath19-legal edge_. note that a d - flip increases the angle vector lexicographically . this can be used to prove the following . [ fact : angle ] the cdt has the lexicographically maximum angle vector of all @xmath19-constrained triangulations on @xmath20 . ( theorem 1 , @xcite ) . [ fact : d - flip ] an @xmath19-constrained triangulation @xmath36 can be converted to the cdt by at most @xmath4 d - flips , taken in any order . ( lemma 4 , @xcite ) . now let us consider non - crossing laman frameworks . a _ constrained delaunay laman framework ( cdlf ) _ is an @xmath19-constrained laman framework that is a subset of the @xmath52 . ( l - flip ) an l - flip is an edge insertion and deletion that takes a laman framework @xmath24 to a new laman framework @xmath53 . [ theo1 ] every @xmath19-constrained non - crossing laman framework @xmath24 can be transformed to a cdlf by at most @xmath4 l - flips . before giving the proof of theorem [ theo1 ] , we define the underlying triangulation @xmath54 for a non - crossing laman framework @xmath24 , which is @xmath24-constrained delaunay triangulation . [ teigi : l - constrained_delaunay ] let @xmath24 be a non - crossing laman framework . the @xmath24-constrained delaunay triangulation @xmath54 is constructed by adding edges to @xmath24 in the following way . first , we add convex hull edges if missing in @xmath24 , and for each planar face of @xmath24 we compute its ( internal ) constrained delaunay triangulation , and add these new edges . by this construction , we have the following fact : [ fact : under_tri ] let @xmath24 be a @xmath19-constrained non - crossing laman framework . then all @xmath19-illegal edges in @xmath54 are the edges of @xmath55 . suppose that @xmath24 has an @xmath19-illegal edge . then , from fact [ fact : under_tri ] , such a @xmath19-illegal edge , say @xmath46 , is among the edges of @xmath55 . consider removing @xmath46 . the updated underlying triangulation @xmath56 contains @xmath57 , and by lemma [ lemma : trilam ] and the matroid property shown by lemma [ lemma : minor ] , there is some edge @xmath58 in @xmath56 such that @xmath59 is a non - crossing laman framework . the fact that @xmath46 is @xmath19-illegal edge implies that there exists at least one d - flip when updating @xmath54 to @xmath56 , which lexicographically increases the angle vector . from fact [ fact : angle ] and [ fact : d - flip ] , repeating this procedure @xmath4 times , we eventually reach the @xmath19-constrained delaunay triangulation @xmath60 , and @xmath61 is the required cdlf . for edges @xmath62 with @xmath63 and @xmath64 with @xmath65 , we use the notation @xmath66 or @xmath67 when @xmath68 is lexicographically smaller than @xmath69 i.e. , either @xmath70 or @xmath71 and @xmath72 , and @xmath73 when they coincide . for an edge set @xmath74 we use the notations @xmath75 and @xmath76 to denote the lexicographically largest and smallest labelled edges in @xmath74 , respectively . ( lexicographic edge list ) let @xmath77 and @xmath78 be the lexicographically ordered edge lists . then @xmath79 is lexicographically smaller than @xmath80 if @xmath81 for the smallest @xmath22 such that @xmath82 . [ theo2 ] the set of @xmath19-constrained non - crossing laman frameworks on a point set @xmath20 is connected by @xmath4 edge flip operations . by the previous theorem , from any non - crossing laman framework we can perform l - flips @xmath4 time to reach a cdlf , say @xmath24 . let @xmath83 be the cdlf with lexicographically smallest edge list . we show that we can do edge flips from @xmath24 to @xmath83 , at most @xmath84 times , maintaining the non - crossing laman property . indeed , delete from @xmath24 the largest indexed edge @xmath46 in @xmath85 . by the matroid properties , there will always be an edge in @xmath86 to insert that maintains the laman framework property . in fact , the 1dof mechanism @xmath57 is not maximum component and can be extended to a base by adding an edge from @xmath86 . planarity is maintained since both @xmath24 and @xmath83 are subsets of the cdt which is non - crossing . let @xmath87 be the number of edges of the triangulation on @xmath20 . by euler s formula , @xmath88 holds , where @xmath89 is the number of points on the convex hull of @xmath20 . then @xmath24 has at most @xmath84 edges which are not in @xmath83 . since we can replace one of such edges by an edge of @xmath83 by one edge flip , we eventually reach @xmath83 by @xmath84 edge flips . let @xmath90 be a set of @xmath19-constrained non - crossing laman frameworks on @xmath20 , and @xmath91 be a set of the cdlfs . clearly @xmath92 holds . let @xmath83 be the cdlf with lexicographically smallest edge list . we define the following _ parent function _ @xmath93 based on theorem [ theo1 ] and [ theo2 ] . [ def : parent ] ( parent function ) let @xmath94 with @xmath95 . @xmath96 is the parent of @xmath24 , where + case 1 : @xmath97 , + @xmath98 and @xmath99 , + case 2 : @xmath100 , + @xmath101 and @xmath102 . to simplify the notations , we denote the parent function depending on case 1 and case 2 by @xmath103 and @xmath104 , respectively . in fig.[fig : parent ] , we show the example of @xmath105 in which @xmath24 is not a cdlf : removing the largest indexed @xmath19-illegal edge @xmath106 , and inserting the smallest indexed edge @xmath107 in @xmath108 , we obtain another non - crossing laman framework shown in the rightmost and upper corner of fig . [ fig : parent ] . , and the dotted lines represent additional edges for the underlying triangulations . ] the reverse search algorithm can be considered on the underlying graph in which each vertex corresponds to a non - crossing laman framework and two frameworks are _ adjacent _ if and only if one can be obtained from the other by a l - flip . then , for @xmath109 the local search is given by an _ adjacency function _ , @xmath110 , defined as follows : @xmath111 where @xmath112 and @xmath113 . the number of candidate edge pairs @xmath33 is @xmath3 . let @xmath114 and @xmath115 be the list of edges of @xmath53 and @xmath28 ordered lexicographically , let @xmath116 and @xmath117 be the number of elements of @xmath114 and @xmath115 and let @xmath118 and @xmath119 be the @xmath22-th elements of @xmath114 and @xmath115 , respectively . we also denote the above defined adjacency function by @xmath120 for which @xmath121 with @xmath122 and @xmath123 with @xmath124 . then , based on the algorithm in @xcite , we describe our algorithm in fig . [ algo : reverse ] . ' '' '' height 0.8pt enumerating @xmath19-constrained non - crossing laman frameworks . ' '' '' @xmath125 the cdlf with lexicographically smallest edge list ; @xmath126 ; @xmath127 ; @xmath128 ; @xmath129 ; @xmath130 ; @xmath131 ; @xmath132 ; @xmath127 ; @xmath128 ; line 4 ; @xmath133 ; @xmath134 * then * @xmath135 ; @xmath136 ; determine integers pair @xmath137 such that @xmath138 ; @xmath139 ; ; ' '' '' as we will show later , both the parent function and the adjacency function need @xmath4 time for each process . then , the while - loop from line 4 to 17 has @xmath140 iterations which require @xmath141 time if simply checking the line 8 and 10 . in order to improve @xmath141 time to @xmath3 time we claim the following two lemmas : [ lemma : parent1 ] let @xmath24 and @xmath53 be two distinct cdlfs which are subgraphs of @xmath142 for which @xmath143 for @xmath112 and @xmath113 . then , @xmath144 holds if and only if @xmath145 and @xmath146 satisfy the following conditions : * @xmath147 , * @xmath148 , * @xmath149 , * @xmath150 . [ lemma : parent2 ] let @xmath24 and @xmath53 be two distinct non - crossing laman frameworks for which @xmath143 for edges @xmath112 and @xmath113 , and let @xmath151 . then , @xmath152 holds if and only if @xmath145 and @xmath146 satisfy the following conditions : * @xmath145 is @xmath19-legal edge in @xmath153 , * @xmath154 , * @xmath155 . * @xmath156 . we will explain later ( in the proof of theorem [ theo : time ] ) how lemmas [ lemma : parent1 ] and [ lemma : parent2 ] are used to obtain @xmath3 time for generating each output of our algorithm . notice that for @xmath53 and @xmath94 such that @xmath157 , at most one of @xmath144 and @xmath152 holds from the conditions ( b ) of lemmas [ lemma : parent1 ] and [ lemma : parent2 ] . ( `` only if''-part . ) since @xmath144 holds , @xmath145 and @xmath146 must be chosen as @xmath58 and @xmath46 in case 1 of definition [ def : parent ] . from definition [ def : parent ] , @xmath158 holds . since @xmath159 , @xmath160 and @xmath161 holds , thus we have ( b ) . similarly since @xmath162 , we have ( a ) . from @xmath163 , we have @xmath164 let @xmath165 . suppose ( c ) does not hold . then @xmath166 holds . ( note that the equality does not hold since @xmath112 . ) then from eq.([eq : lem1_eq ] ) and @xmath167 ( which comes from definition [ def : parent ] ) , @xmath168 thus , @xmath69 would have been selected instead of @xmath145 when the parent function @xmath169 is applied to @xmath24 , which contradicts @xmath163 . hence , ( c ) holds . finally , let @xmath170 , and suppose that ( d ) does not hold and @xmath171 holds . ( note that the equality does not hold since @xmath124 . ) since @xmath172 , @xmath173 then @xmath174 would have been selected instead of @xmath146 when the parent function @xmath169 is applied to @xmath24 , which contradicts @xmath175 . thus , ( d ) holds . ( `` if''-part . ) from ( a ) and ( b ) , @xmath176 . since @xmath177 from ( a ) , ( d ) implies @xmath178 thus , @xmath179 holds , and hence @xmath169 chooses @xmath146 for an edge @xmath46 to be deleted from @xmath24 . from this we have @xmath180 . since @xmath181 from ( b ) , ( c ) implies @xmath182 since @xmath183 , @xmath184 . thus , @xmath169 chooses @xmath145 for an edge to be added , and @xmath185 returns @xmath53 . ( `` only if''-part . ) since @xmath152 holds , @xmath145 and @xmath146 must be chosen as @xmath58 and @xmath46 of definition [ def : parent ] . from @xmath175 , we have @xmath186 since @xmath187 from definition [ def : parent ] and fact [ fact : under_tri ] , @xmath58 is @xmath19-legal in @xmath56 and hence @xmath19-legal in @xmath188 . thus , from @xmath163 , ( a ) holds . since @xmath145 is @xmath19-legal in @xmath153 , we have @xmath189 let @xmath190 . suppose that ( c ) does not hold . then @xmath166 holds . ( note that the equality does not hold since @xmath112 . ) from eq.([eq:1dof2 ] ) and ( [ eq : under_tri ] ) , @xmath191 then , @xmath69 would have been selected when the parent function is applied to @xmath24 , which contradicts @xmath163 . hence ( c ) holds . next let us consider @xmath146 . since @xmath192 is @xmath19-illegal in @xmath54 from definition [ def : parent ] , we have @xmath193 . thus ( b ) holds . finally , ( d ) must hold since the parent function remove an lexicographically largest illegal edge in @xmath54 . ( `` if''-part . ) from ( a ) , @xmath145 is @xmath19-legal in @xmath153 . then , we have @xmath194 . suppose that @xmath146 is @xmath19-legal in @xmath195 , @xmath196 holds and @xmath146 is in @xmath197 , which contradicts the condition ( b ) . therefore , @xmath146 is @xmath19-illegal edge in @xmath54 , and ( d ) says that @xmath146 is the lexicographically largest @xmath19-illegal edge in @xmath54 . thus , @xmath198 chooses @xmath146 for an edge @xmath46 to be deleted from @xmath24 , and @xmath199 holds . from @xmath199 and the condition ( c ) , @xmath200 holds . thus , @xmath201 . ( note that @xmath202 , because @xmath203 and @xmath204 hold from ( a ) . ) thus , @xmath198 chooses @xmath145 for an edge to be added , and @xmath105 returns @xmath53 . by lemmas [ lemma : parent1 ] and [ lemma : parent2 ] , we will describe an @xmath3 algorithm in the proof of the following theorem [ theo : time ] . before it , we give you the simple observation for checking the condition ( d ) in the lemma [ lemma : parent2 ] efficiently : [ ob : condition_d ] let @xmath153 be a @xmath53-constrained delaunay triangulation constrained by edges of a non - crossing laman framework @xmath53 , and let @xmath205 be a @xmath19-legal edge in @xmath153 and @xmath113 be an edge intersecting no edge of @xmath53 . then @xmath206 holds . this comes from the following two facts that : * @xmath194 holds , and * @xmath145 is @xmath19-legal in @xmath207 . the fact ( 1 ) clearly hold since @xmath145 is @xmath19-legal edge in @xmath153 . using lemma 4.2 in @xcite the fact ( 2 ) immediately follows . let us show how to update @xmath153 to @xmath207 . the edge @xmath146 intersects some edges in @xmath208 , and we denote a set of such edges by @xmath209 . first , we delete the edges in @xmath209 . the resulting graph @xmath210 has a hole bounded by the polygon @xmath211 . when inserting @xmath146 into @xmath210 , @xmath146 splits @xmath211 into two polygon @xmath212 and @xmath213 . then , @xmath214 holds from the fact in @xcite , i.e. @xmath207 is obtained from @xmath153 by ( delaunay ) triangulating only @xmath212 and @xmath213 independently . this implies that @xmath145 never flips even if @xmath145 is not a constrained edge . ( note that @xmath146 does not intersect @xmath145 from the lemma assumption . ) hence , @xmath145 is @xmath19-legal in @xmath207 . [ theo : time ] the set of all @xmath19-constrained non - crossing laman frameworks on a given point set can be reported in @xmath3 time per output using @xmath4 space , or @xmath1 time using @xmath2 space . as described in section [ sec : main ] , we use a linear transformation if necessary to get a unique cdt . the complexity of testing the uniqueness of a cdt is @xmath4 by simply testing the circumcircle of each triangle in the cdt to see there is another point other than vertices of the triangle on the circumcircle . given a non - crossing laman framework @xmath109 and @xmath53-constrained delaunay triangulation @xmath153 . the algorithm will check @xmath215 or @xmath216 at line 10 depending on the edge pair @xmath33 . here we will show that each condition in lemmas [ lemma : parent1 ] and [ lemma : parent2 ] can be checked in @xmath217 time for each of the @xmath3 edge pairs @xmath33 by the following way . first , for all edges @xmath218 , we calculate the number of edges @xmath205 intersecting @xmath146 , which we denote by cross_n@xmath219 . if cross_n@xmath220 , we delete @xmath146 from @xmath115 since @xmath221 is never non - crossing for any @xmath222 . if cross_n@xmath223 , we associate a pointer of the edge @xmath145 intersecting @xmath146 with @xmath146 , and we denote such @xmath145 by cross_e@xmath219 . next , for each @xmath222 , we attach two flags to @xmath145 which represent that @xmath145 satisfies the condition ( a ) of lemmas [ lemma : parent1 ] and [ lemma : parent2 ] , respectively . this preprocessing can be done in @xmath2 time by simply scanning all edges of @xmath114 . by this we can check the condition ( a ) in lemmas [ lemma : parent1 ] and [ lemma : parent2 ] in @xmath217 time . similarly , we attach two flags to @xmath218 which represent that @xmath146 satisfies the condition ( b ) of lemmas [ lemma : parent1 ] and [ lemma : parent2 ] . this process can be done in @xmath4 time . by this we can check the condition ( b ) in lemmas [ lemma : parent1 ] and [ lemma : parent2 ] in @xmath217 time . we can calculate the lexicographically largest edge in @xmath224 in @xmath2 time . by this we can check the condition ( d ) in lemmas [ lemma : parent1 ] in @xmath217 time . now let us consider how to identify a set of edges @xmath225 satisfying the condition ( d ) in lemma [ lemma : parent2 ] in @xmath3 time with @xmath4 space . ( in the case of @xmath1 time algorithm this process must be skipped , and the condition ( d ) in lemma [ lemma : parent2 ] will be checked simply by updating @xmath153 to @xmath226 for each pair @xmath33 using @xmath2 time and @xmath2 space by applying the algorithm by chin and wang@xcite . ) it can be done regardless of the removing edge @xmath145 if the condition ( a ) in lemma [ lemma : parent2 ] holds . from observation [ ob : condition_d ] , we can say that the condition ( d ) holds if and only if @xmath146 is lexicographically largest @xmath19-illegal edge in @xmath227 when cross_n@xmath228 . and it is sufficient to check the condition ( d ) only in @xmath229cross_e@xmath230 when cross_n@xmath223 . updating the delaunay triangulation takes @xmath2 time ( see @xcite for a linear time update of the constrained delaunay triangulation ) . thus we can attach a flag to @xmath218 in @xmath2 time which represents whether @xmath146 satisfies the condition ( d ) of lemma [ lemma : parent2 ] or not , thus this preprocessing for all edges in @xmath115 takes @xmath3 time . by using the above mentioned data , we will show that for a fixed @xmath222 , the inner while - loop from line 6 to 16 can be executed in @xmath4 . in order to efficiently test the condition ( c ) of lemmas [ lemma : parent1 ] and [ lemma : parent2 ] , we prepare the data structure proposed by lee , streinu and theran @xcite in @xmath4 time for maintaining _ rigid components _ obtained by deleting @xmath145 . this data structure supports a _ pair - find _ query which determines whether two vertices are spanned by a common component in @xmath217 time using @xmath4 space , or @xmath2 time with @xmath2 space . from this , we can calculate @xmath231 ( i.e. , determine whether @xmath232 ) in @xmath217 time with @xmath4 space , or @xmath2 time with @xmath2 space , for each edge @xmath218 . also , we can compute @xmath233 and @xmath234 in @xmath2 time with @xmath4 space , or @xmath4 time with @xmath2 space . from @xmath69 and @xmath174 we can check condition ( c ) in lemmas [ lemma : parent1 ] and [ lemma : parent2 ] in @xmath217 time . we have to determine which of @xmath215 and @xmath216 should be checked for every edge @xmath218 . this is simply done by checking whether @xmath235 or @xmath236 . when @xmath161 , we check whether @xmath215 holds or not . when @xmath236 , we check whether @xmath216 holds or not . both can be done in @xmath217 time with @xmath237 space , or @xmath2 time with @xmath2 space . by using the above mentioned data structure for maintaining the rigid components , we can perform both parent function and adjacency function in @xmath4 time with @xmath4 space , or @xmath3 time with @xmath2 space . thus , we have an @xmath3 algorithm using @xmath4 space , or @xmath1 algorithm using @xmath2 space . figure [ fig : search_tree ] illustrates an example of the search tree of a set of @xmath19-constrained non - crossing laman frameworks on six points , where @xmath238 . we have presented an algorithm for enumerating all the constrained non - crossing laman frameworks . we note in passing that the techniques in this paper can also be used to generate all @xmath19-constrained non - crossing spanning trees of a point set since they also form bases of the graphic matroid on any triangulation of @xmath20 . the unconstrained case was considered in @xcite . o. aichholzer , g. rote , b. speckmann , and i. streinu . the zig - zag path of a pseudo - triangulation . in _ proc . 8th international workshop on algorithms and data structures ( wads ) _ , lecture notes in computer science 2748 , pages 377388 , ottawa , canada , 2003 . springer verlag . d. avis , n. katoh , m. ohsaki , i. streinu , and s. tanigawa . enumerating planar minimally rigid graphs . to appear in _ proc . 12th annual international computing and combinatorics conference ( cocoon 2006 ) _ , taipei , 2006 i. beichl , and f. sullivan . coping with degeneracies in delaunay triangulation . in _ modelling , mesh generation and adaptive numerical methods for partial differential equations , j.e . flaherty et al . _ , pages 2330 , springer , new york , 1995 . a. berg and t. jordn . algorithms for graph rigidity and scene analysis . in g. d. battista and u. zwick , editors , _ proc . 11th annual european symposium on algorithms ( esa ) _ , volume 2832 of _ lecture notes in computer science _ , pages 7889 . springer , 2003 . a. dumitrescu , b. grtner , s. pedroni , and e. welzl . enumerating triangulation paths . , 20(1 - 2):312 , 2001 . a preliminary version in proceedings of the twelfth canadian conference on computational geometry , 2000 ( cccg00 ) , 233 - 238 . n. katoh , m. ohsaki , t. kinoshita , s. tanigawa , d. avis and i. streinu . enumeration of optimal pin - jointed bistable mechanisms . in _ proc . 4th china - japan - korea symp . of structural and mechanical systems _ , kunming , nov 2006 . g. rote , f. santos , and i. streinu . expansive motions and the polytope of pointed pseudo - triangulations . in j. p. boris aronov , saugata basu and m. sharir , editors , _ discrete and computational geometry - the goodman - pollack festschrift _ , algorithms and combinatorics , pages 699736 . springer verlag , berlin , 2003 .	in this paper we present an algorithm for enumerating without repetitions all the non - crossing generically minimally rigid bar - and - joint frameworks under edge constraints ( also called constrained non - crossing laman frameworks ) on a given generic set of @xmath0 points . our algorithm is based on the reverse search paradigm of avis and fukuda . it generates each output graph in @xmath1 time and @xmath2 space , or , slightly different implementation , in @xmath3 time and @xmath4 space . in particular , we obtain that the set of all the constrained non - crossing laman frameworks on a given point set is connected by flips which restore the laman property . _ key words : _ geometric enumeration ; rigidity ; constrained non - crossing minimally rigid frameworks ; constrained delaunay triangulation .
rf information and energy transfer , which treats wireless receivers as either conventional information decoding ( i d ) receivers or energy harvesting ( eh ) receivers , has emerged as an attractive paradigm for green communication in the next - generation wireless networks @xcite . in particular , transmit beamforming to improve the quality - of - service of rf information and energy transfer has drawn significant research interest @xcite . there are two important beamforming problems in optimizing the harvested energy : ( i ) maximization of the total harvested energy ( i.e. , ` sum eh maximization problem ' ) , and ( ii ) maximization of the most disadvantaged eh receiver in the network ( i.e. , ` max - min eh problem ' ) . both problems are subject to the minimum sinr constraint at the i d receivers and the total transmit power constraint at the base station ( bs ) . they are indefinite quadratic programs in the beamforming vectors , which are typically recast as nonconvex rank - one matrix constrained semidefinite programs ( sdps ) in the beamforming outer products . the rank - one matrix constraints are then omitted , allowing for suboptimal sdp relaxation @xcite . meanwhile , it is realistic for the users located in the vicinity of the bs to conduct both i d and eh functions by a signal splitting ( ss ) based receiver @xcite . a problem of particular interest is how to jointly design the transmit beamforming vectors and the receive ss ratios in order to maximize the harvested energy . finding an efficient computational solution for this problem is a major challenge in itself . due to the strong coupling of the beamforming vectors and the ss ratios in the optimization objective , the sum eh maximization problem can not be recast as a matrix rank - one constrained sdp to accommodate the conventional sdp relaxation approach @xcite . for the max - min eh problem , the beamforming vectors and ss ratios can be decoupled via a bisection search for the worst eh receiver . nevertheless , the resultant sdp relaxation only provides an upper bound performance for this problem as this approach is less likely to generate a rank - one matrix solution . randomization in conjunction with linear programming must be further employed to generate the beamforming vectors , which however could be far away from the actual optimal solutions @xcite . this letter aims to develop an efficient computational method for the two aforementioned joint design eh optimization problems . our research contributions are summarized as follows . * we propose a new path - following method for their solutions . each iteration of the proposed algorithms requires solving a second - order cone program ( socp ) in the beamforming vectors and ss ratios . we mathematically show that the algorithms progress at every iteration and converge to locally optimal solutions . in addition , our simulation results with practical parameter values show that the obtained locally optimal solutions are the same as the upper bound given by the exhaustive search over the domain of ss ratios ( for the sum eh maximization problem ) or that by the sdp relaxation - based bisection search ( for the max - min eh problem ) . this demonstrates the usefulness of our solutions . * our algorithms are computationally efficient and simple to implement . in contrast , the exhaustive search and the sdp relaxation - based bisection search can only achieve the upper bounds after solving many sdps in the beamforming outer products of a substantially increased dimension . such techniques are computationally prohibitive in practical networks . we consider an energy - constrained small - cell wireless network where a bs with @xmath0 antennas transmits to @xmath1 single - antenna users ( ues ) . let @xmath2 be the flat fading channel vector between the bs and ue @xmath3 , which includes the effects of large - scale pathloss and small - scale fading . let @xmath4 ( with @xmath5 ) denote the intended message for ue @xmath6 , which is beamformed by the vector @xmath7 at the bs . the baseband signal received by ue @xmath6 is expressed as : @xmath8 where @xmath9 is the zero - mean circularly symmetric additive white gaussian noise ( awgn ) with variance @xmath10 introduced by the receive antenna . we assume perfect channel state information ( csi ) is available at the bs , where details of the csi acquisition process can be found in @xcite . the first term in is the intended signal for ue @xmath6 while the second term represents the interference . given that rf eh is only practical when the bs - ue distance is sufficiently small @xcite , our model divides the bs coverage area into ( i ) the eh zone ( near the bs ) , and ( ii ) the i d zone ( outside the eh zone ) . if a ue resides inside the eh zone , it can conduct both eh and i d ; otherwise , it only performs i d . assume there are @xmath11 and @xmath12 ues residing in the eh and i d zones , respectively , where @xmath13 . let us index the eh - id ues by @xmath14 and the id - only ues by @xmath15 . at an eh - id ue @xmath16 , the signal splitter divides the received signal @xmath17 into two parts in the proportion of @xmath18 , where @xmath19 is termed as the ss ratio for ue @xmath20 . the first part @xmath21 forms an input to the i d receiver as : @xmath22 where @xmath23 is the additional noise with variance @xmath24 introduced by the i d receiver circuitry . from , we write the general expression of the sinr at the input of the i d receiver of ue @xmath25 ( i.e. , either an eh - id ue or an id - only ue ) as : @xmath26 where @xmath27_{n\in { \cal n}}$ ] and @xmath28 with @xmath29 for @xmath30 and @xmath31 for @xmath32 . also at the eh - id ue @xmath16 , the second part @xmath33 of the received signal @xmath17 is processed by an eh receiver . the energy harvested by ue @xmath20 is thus given by : @xmath34 where @xmath35 , and the constant @xmath36 denotes the efficiency of energy conversion at the eh receiver . upon defining @xmath37_{n_1\in\mathcal{n}_1}$ ] , the sum eh maximization problem is formulated as : [ eq : p1 ] @xmath38 constraint caps the total transmit power at a predefined value @xmath39 , and constraint ensures that the received sinr by ue @xmath6 be greater than a predefined threshold @xmath40 . note that is a nonconvex optimization problem because is not concave . in the relaxation approach , by omitting the difficult constraints @xmath41 for the beamforming outer products @xmath42 , one uses @xmath43 , @xmath44 and @xmath45 to form [ sdp1 ] @xmath46 with linear function @xmath47 and convex function @xmath48 , where @xmath49_{\eta\in{{\cal n}}}$ ] . the multiplicative objective function in ( [ sdp1a ] ) remains nonlinear nonconcave in @xmath50 and @xmath51 , making ( [ sdp1 ] ) still computationally difficult . in the following section , we propose an efficient path - following method to solve directly in the beamforming vectors @xmath52 and the scalar ss ratios @xmath53 since the power constraint ( [ eq:01a ] ) is convex quadratic , let us first deal with the sinr constraint . for @xmath54 , one has @xmath55 and @xmath56 for @xmath57 , where @xmath58 and @xmath59 is the real part of a complex number @xmath60 . therefore , can be recast as @xmath61 @xcite . the latter is equivalent to the following soc : @xmath62 where @xmath63 is an @xmath64 column vector , @xmath65 is an auxiliary variable , and @xmath66 for @xmath67 whereas @xmath68 for @xmath69 . this means ( [ eq : p1 ] ) is a nonconcave maximization problem subject to convex constraints . to develop a path - following procedure for the computational solution of ( [ eq : p1 ] ) , the following lemma provides an effective concave lower approximation for the multiplicative objective in @xcite . [ prop ] _ for any @xmath70 , @xmath71 and @xmath72 , @xmath73 the following relations hold true : _ [ eq : p_max_min11 ] @xmath74 for @xmath75 @xmath76 \\ -{\displaystyle}\frac{p_{n_1}(\mathbf{w}^{(\kappa)})(1-(\alpha_{n_1}^{(\kappa)})^2)^2}{1-\alpha_{n_1}^2 } , \end{array}\ ] ] which is concave in @xmath77 . see the appendix . with lemma [ prop ] , we propose algorithm [ alg:1 ] to solve problem ( [ eq : p1 ] ) . initialize @xmath78 . find a feasible point @xmath79 of by solving the socp : @xmath80 s.t . ( [ soc1 ] ) , ( [ soc2 ] ) . solve the socp : @xmath81 to generate a feasible point @xmath82 for . set @xmath83 . note that @xmath84 by ( [ in1 ] ) and @xmath85 by ( [ in2 ] ) , while @xmath86 as long as @xmath87 because @xmath88 is also feasible to ( [ itea ] ) . as a result , @xmath89 i.e. , @xmath82 is a better feasible point to ( [ eq : p1 ] ) than @xmath88 . the sequence @xmath90 of improved feasible points to ( [ eq : p1 ] ) converges at least to its local optimum , which also satisfies the karush - kuhn - tucker condition @xcite . simulation results in sec . [ sec : simulation ] further show that our algorithm achieves the computable upper bound , implying that a global optimum is attained in the practical settings considered in the simulations . our proposed method can be extended to solve the following max - min eh problem : @xmath91 in this case , instead of ( [ itea ] ) , we solve the following socp to generate @xmath92 from @xmath93 : @xmath94 * remark . * unlike @xcite , here we consider both eh and i d functionalities for the near ues . the nature of the optimization problems is thus nontrivially changed , requiring a different solution approach as has been proposed in this letter . the difference - of - two - convex - functions based optimization approach in @xcite is not suitable for the sum eh maximization problem . not only would the problem dimension be increased , it is also not easy to find a good feasible initial point to use in this approach due to the nonconvex constraints . recently , @xcite has employed the socp ( [ soc1])-([soc2 ] ) to express the sinr constraint ( [ eq : c1 ] ) in the problem of minimizing the total transmit power @xmath95 . the eh threshold is then set to make the eh constraints automatically satisfied by the sinr constraints ( [ soc1])-([soc2 ] ) . the approach in @xcite is thus inapplicable to our eh optimization problems ( [ eq : p1 ] ) and ( [ eq : p2 ] ) . we consider a small - cell network with @xmath96 ues and @xmath97 antennas at the bs . unless specified otherwise , the bs - to - ue distance is set as @xmath98 m and @xmath99 m for the @xmath100 eh - id ues and the @xmath101 id - only ues , respectively . later in fig . [ fig : comp ] , we will consider both @xmath98 m and @xmath102 m distances for the @xmath11 eh - id ues since commercial radiative wireless charging systems ( e.g. , the cota systems @xcite ) can deliver power up to @xmath103 ft ( i.e. , @xmath102 m ) . we assume a simplified path loss model with carrier center frequency of @xmath104 mhz , transmit antenna gain of @xmath105 dbi , reference distance of @xmath106 m and path loss exponent of @xmath107 @xcite , ( * ? ? ? we generate a rician fading channel with a rician factor of @xmath108 db . for simplicity and without loss of generality , we assume that @xmath109 and @xmath110 . thus , we set @xmath111 , @xmath112 dbm , and @xmath113 dbm . the numerical results are averaged over @xmath114 random channel realizations . [ fig : pow ] plots the optimized sum harvested energy for @xmath115 dbm and @xmath116 db . on average , algorithm 1 requires only @xmath117 iterations ( i.e. , solving @xmath117 socps in @xmath53 and @xmath118 ) . [ fig : pow ] also confirms that the sum harvested energy increases when more transmit power is available . for the typical value of @xmath119 dbm used in a small - cell bs , the sum harvested energy for the three eh - id ues is found as @xmath120 dbm for @xmath97 bs antennas . [ fig : gam ] plots the sum harvested energy for @xmath121 db and @xmath119 dbm . as can be seen from the figure , increasing the sinr threshold reduces the harvested amount of energy . this is because more received power is diverted to the i d receiver to meet the minimum sinr requirement , leaving less power for the eh receiver . [ fig : gam ] shows that the sum harvested energy is in the range of @xmath122 dbm to @xmath123 dbm for the practical network parameters considered . _ convergence and complexity : _ figs . [ fig : pow ] and [ fig : gam ] demonstrate that algorithm 1 practically yields the globally optimal solution of ( [ eq : p1 ] ) . it achieves the upper bound obtained by a branch - and - bound ( bb ) procedure @xcite over the domain @xmath124 that finds the optimal ss ratios @xmath125_{n_1\in{{\cal n}}_1}$ ] of the nonconvex program ( [ sdp1 ] ) . at each iteration of this procedure , an upper bound of the nonconvex program ( [ sdp1 ] ) over @xmath126\subset ( 0,1)^{n_1}$ ] is provided by the sdp @xmath127,\forall n_1}\ { \displaystyle}\sum_{n_1\in\mathcal{n}_1}\zeta_{n_1}(1-\alpha_{n_1}^2)\tilde{p}_{n_1}(\mathbf{w})\ \text{s.t.}\ \eqref{sdp1b}-\eqref{sdp1c}$ ] . the incumbent ss ratios are updated by solving ( [ sdp1 ] ) at a fixed @xmath128 , @xmath129 , which is an sdp in the beamforming outer products @xmath130 . while much more efficient than the intensive grinding , the computational complexity of this exhaustive search over the domain @xmath124 is still prohibitively high . furthermore , the total dimension of the beamforming outer products @xmath130 is @xmath131 compared with @xmath132 of the beamforming vectors @xmath118 . in our simulations , we have also observed that this search gives solution @xmath130 with rank greater than one in @xmath133 and almost @xmath134 of all cases for @xmath135 db and @xmath136 db , respectively . fig . [ fig : comp ] plots the optimized energy harvested by the worst user over a range of transmit power budget @xmath115 dbm . here , the distances from the bs to all eh - id ues are set as @xmath98 m and @xmath102 m. it is observed that our algorithm achieves the upper bound given by the sdp - based bisection search with considerably less computational complexity . the bisection search locates the largest @xmath137 such that the sdp @xmath138 is feasible in the beamforming outer products @xmath130 and the scalar ss ratios @xmath53 . on average , our algorithm converges after @xmath139 iterations ( i.e. , solving @xmath139 socps ) whereas the bisection search solves @xmath140 sdps of a much larger dimension . furthermore , the latter often yields a rank - greater - than - one matrix @xmath130 . particularly , for @xmath141 bs antennas , we have observed @xmath130 with rank greater than one in @xmath142 of the simulation cases . to generate a rank - one matrix for extracting beamforming vectors , a randomization step in conjugation with linear programming is required in such approach . this incurs extra computational overhead while the performance may suffer ( * ? ? ? 3 ) as the extracted point may be far away from the actual optimum . finally , [ fig : comp ] confirms that the received power threshold of @xmath143 dbm to @xmath144 dbm required to activate practical eh receivers @xcite is met by our algorithm , even with the most conservative choice of simulation parameters ( i.e. , the smallest number of antennas @xmath145 , a maximum distance of @xmath102 m from the bs to eh - id ues , and the smallest bs power @xmath146 dbm ) . we have proposed successive second - order cone programming algorithms for the joint design of transmit beamforming vectors and receive signal splitting factors . the objective is to maximize either the sum eh or the energy harvested by the receiver with the least favorable channel conditions under the ue sinr and the bs power constraints . simulation results with practical parameters have confirmed the merits of the proposed algorithms . as function @xmath147 is convex in @xmath148 , the so - called perspective of @xmath149 defined as @xmath150 is also convex in @xmath151 and @xmath152 @xcite . this gives @xmath153 for @xmath154 and applying ( [ ap1 ] ) for @xmath155 and @xmath156 , @xmath157 , we have @xmath158 similarly , @xmath159 recall that the lhs of ( [ in1 ] ) is @xmath160 . then , @xmath161\nonumber\\ & \qquad + 2\sigma_a^2(1-(\alpha_{n_1}^{(\kappa)})^2)-\sigma_a^2(1-(\alpha_{n_1}^{(\kappa)})^2)^2/(1-\alpha_{n_1}^2)\nonumber\\ & = 2(1-(\alpha_{n_1}^{(\kappa)})^2)[{\displaystyle}\sum_{{\color{black } \eta}\in\mathcal{n } } \re\{(\mathbf{w}_{{\color{black } \eta}}^{(\kappa)})^h\mathbf{h}_{n_1}\mathbf{h}^h_{n_1 } \mathbf{w}_{{\color{black } \eta } } \}+\sigma_a^2]\nonumber\\ & \qquad-{\displaystyle}p_{n_1}(\mathbf{w}^{(\kappa)})(1-(\alpha_{n_1}^{(\kappa)})^2)^2/(1-\alpha_{n_1}^2),\end{aligned}\ ] ] showing the inequality in ( [ in1 ] ) . the equality in ( [ in2 ] ) is obvious . x. chen , z. zhang , h. h. chen , and h. zhang , `` enhancing wireless information and power transfer by exploiting multi - antenna techniques , '' _ ieee commun . magazine _ , vol . 53 , no . 4 , pp . 133141 , apr . 2015 . s. timotheou , i. krikidis , s. karachontzitis , and k. berberidis , `` spatial domain simultaneous information and power transfer for mimo channels , '' _ ieee trans . wireless commun . _ , vol . 14 , no . 8 , pp . 41154128 , aug . 2015 . a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , `` relaying protocols for wireless energy harvesting and information processing , '' _ ieee trans . wireless commun . _ , vol . 12 , no . 7 , pp . 36223636 , jul . 2013 . i. krikidis , s. sasaki , s. timotheou , and z. ding , `` a low complexity antenna switching for joint wireless information and energy transfer in mimo relay channels , '' _ ieee trans . _ , vol . 62 , no . 5 , pp . 15771587 , may 2014 . d. ng , e. lo , and r. schober , `` robust beamforming for secure communication in systems with wireless information and power transfer , '' _ ieee trans . wireless commun . _ , vol . 13 , no . 8 , pp . 45994615 , aug 2014 . q. shi , w. xu , t .- h . chang , y. wang , and e. song , `` joint beamforming and power splitting for miso interference channel with swipt : an socp relaxation and decentralized algorithm , '' _ ieee trans . signal process . _ , vol . 62 , no . 61946208 , dec . 2014 .	we consider the joint design of transmit beamforming and receive signal - splitting ratios in the downlink of a wireless network with simultaneous radio - frequency ( rf ) information and energy transfer . under constraints on the signal - to - interference - plus - noise ratio ( sinr ) at each user and the total transmit power at the base station , the design objective is to maximize either the sum harvested energy or the minimum harvested energy . we develop a computationally efficient path - following method to solve these challenging nonconvex optimization problems . we mathematically show that the proposed algorithms iteratively progress and converge to locally optimal solutions . simulation results further show that these locally optimal solutions are the same as the globally optimal solutions for the considered practical network settings . energy harvesting , nonconvex optimization , path - following algorithm , signal splitting , transmit beamforming
the cosmic microwave background ( cmb ) is now a well known probe of the early universe . the temperature fluctuations in the cmb , especially the so - called acoustic peaks in the angular power spectrum of cmb anisotropies , capture the physics of primordial photon - baryon fluid undergoing oscillations in the potential wells of dark matter ( hu et al . 1997 ) . the associated physics involving the evolution of a single photon - baryon fluid under compton scattering and gravity are both simple and linear , and many aspects of it have been discussed in the literature since the early 1970s ( peebles & yu 1970 ; sunyaev & zeldovich 1970 ) . the gravitational redshift contribution at large angular scales ( e.g. , sachs & wolfe 1968 ) and the photon - diffusion damping at small angular scales ( e.g. , silk 1968 ) complete this description . by now , the structure of the first few acoustic peaks is well studied with nasa s wilkinson microwave anisotropy probe ( wmap ) mission ( e.g. , spergel et al . 2003 ) , while in the long term , esa s planck surveyor , will extend this to a multipole of @xmath1 2000 . given the improved resolution and better frequency coverage , a variety of studies related to secondary anisotropies , e.g. , sunyaev - zeldovich effect ( sunyaev & zeldovich 1980 ; cooray , hu & tegmark 2000 ) is expected . beyond temperature fluctuations , detections of the polarization anisotropy , related to density or scalar perturbations , have now been made by dasi ( kovac et al . 2003 ; leitch et al . 2004 ) and cbi ( readhead et al . 2004 ; cartwright et al . 2005 ) , while the temperature - polarization power spectrum is measured by wmap ( kogut et al . 2003 ) . an interesting result from wmap data is related to the very large ( @xmath2 ) angular scale polarization signal , which probes the local universe and associated astrophysics instead of physics at the last scattering surface probed with polarization at tens of arcminute scales . this local universe contribution arises when the universe reionizes again and the temperature quadrupole begins to rescatter to produce a new contribution to the polarization anisotropy at angular scales corresponding to the horizon at the new scattering surface ( zaldarriaga 1997 ) . the large - scale excess signal measured in the wmap temperature - polarization cross - correlation power spectrum is interpreted as rescattering with an optical depth to electron scattering of @xmath3 ( kogut et al . 2003 ) . such an optical depth suggests early reionization ; for example , if the universe reionized completely and instantaneously at some redshift , the measured optical depth suggests a reionization redshift of @xmath1 17 ( @xmath4 ) . such a high optical depth , in the presence of a @xmath5 1% neutral hydrogen fraction from z @xmath1 6 sloan quasars ( fan et al . 2002 ) , suggests a complex reionization history . a complex , and patchy , reionization , however , is expected if the reionization is dominated by the uv light from first luminous objects , though to explain the high level of the wmap s optical depth requires high star formation efficiency at redshifts of order 15 with a possibility for a population of metal - free massive pop iii stars ( e.g. , cen 2003 ) . in general , one can only extract limited information related to the reionization history from polarization peak at tens of degrees or more angular scales ( e.g. , hu & holder 2003 ) . one reason for this is the large cosmic variance associated with measurements at multipoles of @xmath1 10 . to extract detailed information of the reionization process , one can move to temperature fluctuations generated during the reionization era at angular scales corresponding to few arcminutes and below . at these small angular scales , potentially interesting contributions from high redshifts include a scattering contribution related to moving electrons in ionized patches and fluctuations in the moving electron population ( e.g. , santos et al . 2003 ) , and a thermal sunyaev - zeldovich effect related to the first supernovae bubbles ( oh et al . the same redshift ranges are expected to be illuminated in the near - infrared band between 1 to 3 @xmath6 m in the form of spatial fluctuations to the cosmic ir background ( e.g. , santos et al . 2002 ; salvaterra et al . 2003 ; cooray et al . potentially interesting studies to extract more information related to the reionization process include brightness temperature fluctuations of the redshifted 21 cm line emission by neutral hi prior to and during hydrogen ( e.g. , zaldarriaga et al . 2004 ) , and cross - correlation studies between small angular scale effects in cmb and maps at other wavelengths ( cooray 2004a ; cooray 2004b ; salvaterra et al . 2005 ) . upcoming experiments such as the south pole telescope and the atacama cosmology telescope , combined with near - ir and 21 cm maps , will provide some of the first possibilities in this direction . we first summarize basic aspects related to inflation under the assumption of a single scalar field model . our derivations , however , are applicable to multi - field theories with few alterations ( lidsey et al . 1997 ) . in general , constraints on inflation are obtained by comparing theoretically calculated scalar and tensor metric gravitational waves perturbations to direct ( in the case of space - based detectors ) and indirect ( such as cmb ) measurements of these perturbations . the tensor and scalar perturbations are determined by the evolution of the scalar field , @xmath7 , ( the inflaton field ) and the scale factor , @xmath8 , with time . the equation of motion for a scalar field with a self - interaction potential , @xmath9 , is given by , @xmath10 where @xmath11 is the hubble parameter and the prime refers to differentiation with respect to the field , @xmath7 . the evolution of the hubble parameter is determined by the energy density , @xmath12 , which we suppose is dominated by the energy density associated with the inflaton field , @xmath13 where @xmath14 . the initial curvature redshifts away by the time the scales corresponding to the current hubble horizon left the inflationary horizon . in general , an exact analytic solution for the spectrum of scalar and tensor perturbations does not exist . in order to facilitate analytic solutions , one generally resorts to the slow - roll expansion , where the equations of motion are expanded in terms of the inflaton potential and its derivatives ( liddle , parsons & barrow 1994 ) . motivation for this expansion can be understood by noting that the equation of motion for the inflaton field is formally equivalent to the equation of motion of a mass moving under a potential @xmath15 with damping due to the cosmological expansion . the slow roll expansion supposes that the motion of the field reaches a terminal velocity in which the potential energy dominates over the kinetic energy and the acceleration of the field is negligible . in order to quantify these assumptions , the slow - roll parameters are used @xmath16 under the assumption that @xmath17 , the approximate slow - roll evolution equations are @xmath18 on a given length scale , scalar perturbations start as quantum fluctuations in otherwise classical fields well within the horizon . here , flat space - time quantum field theory is appropriate . as the scale factor quasi - exponentially increases , the length scale is stretched beyond the inflationary horizon . the quantum fluctuations are then amplified and become classical perturbations to the metric ( guth & pi 1982 ) . in order to study scalar and tensor perturbations , we consider their power spectra . it can be shown that within the slow roll approximation the amplitude of power spectra can be written as ( stewart & lyth 1993 , lidsey et al . 1997 ) , @xmath19 where @xmath20 and @xmath21 refer to scalar and tensor perturbations respectively . the right hand sides of the above equations are to be evaluated when a given comoving length scale crosses the inflationary horizon with @xmath22 . in order to quantify how these power spectra change with scale , one considers their power - law spectral indices ( liddle and lyth 1992 ) , @xmath23 if the potential is flat enough , higher order spectral parameters , which are higher order in the slow roll parametersm , can be ignored ( stewart & lyth 1993 ) . as long as the slow roll parameters can be considered constant , we are able to write the power spectra as power - laws with spectral indices given by the above expressions . it can then be shown that , @xmath24 currently , cmb data probe up to @xmath25 corresponding to @xmath26 . current observations constrain @xmath27 at scales corresponding to the current hubble horizon . therefore , although cmb observations do cover a few orders of magnitude in wavenumber , to within second order in the slow roll parameters , they can be considered constant over the range probed by the cmb . this implies that the perturbation power spectra will appear as approximate power laws , expanded around some pivot wavenumber , @xmath28 ( leach et al . 2002 ) , @xmath29 in the case of wmap these parameters are evaluated at @xmath30 ( spergel et al . 2003 ; peiris et al . 2003 ) . using the expressions for the amplitudes and spectral indices in terms of the slow roll parameters , we are able to solve for the potential and its first two derivatives and thereby use observations to constrain the form that this potential can take . we have , @xmath31 and all of the spectral parameters are to be evaluated at the pivot wavenumber , @xmath28 . the last expression is known as the consistency relation and is true only in single - field inflationary model . the expression relaxes to an inequality in the case of multi - field models . it is a consequence of the fact that both scalar and tensor spectra are determined from the dynamics of a single scalar field , so they are not independent ( lidsey et al . 1997 , liddle & lyth 2000 ) . though acoustic oscillations in the temperature anisotropies of cmb suggest an inflationary origin for primordial perturbations , such as in the wmap data ( peiris et al . 2003 ) , it has been argued for a while that the smoking - gun signature for inflation would be the detection of a stochastic background of gravitational waves ( e.g. , kamionkowski & kosowsky 1999 ) . since cmb probes horizon to super - horizon scales , inflation is the only known mechanism to casually produce a stochastic background of gravitational waves at such large wavelengths ( e.g. , starobinsky 1979 ) . note that the amplitude of the gravitational wave background is highly unknown . since the amplitude of these inflationary gravitational waves is proportional to @xmath32 , the value of the inflaton potential @xmath9 during inflation , the amplitude of gravitational - wave induced b - mode polarization anisotropies directly constrains the energy scale of inflation @xmath33 . relative to cobe normalization , the relation between the energy scale and the ratio of tensor - to - scalar fluctuations is @xmath34 ( turner & white 1996 ) , the current limit from cmb anisotropy data , based on large angular - scale cmb temperature fluctuations , suggests an upper limit of order @xmath35 gev ( melchiorri & odman 2003 ) . instead of temperature fluctuations ( where dominant scalar perturbations confuse the detection of tensor contribution ) , the gravitational - waves can be studied through their distinct signature in the cmb polarization in the form of a contribution to the curl , or magnetic - like , component ( kamionkowski et al . 1997 ; seljak & zaldarriaga 1997 ) . there is no contribution from the dominant scalar , or density - perturbation , contribution to these curl modes in linear theory . secondary effects related to density perturbations , however , do produce signals in the curl - mode and confuse the detection of inflationary gravitational - wave signal . among these , the few percent conversion of e to b from cosmic shear modification to the polarization pattern by the intervening large scale structure is important ( zaldarriaga & seljak 1998 ) . these lensing - induced curl modes introduce a noise from which gravitational waves must be distinguished in the cmb polarization . gravitational lensing of the cmb photons by the mass fluctuations in the large - scale structure is now well understood ( e.g. , hu 2000 ; hu & cooray 2001 ) . the lensing effect can be described through a remapping process involving angular deflections resulting along the photon path : @xmath36 where the deflection angle on the sky is given by the angular gradient of the lensing potential , @xmath37 , which itself is a projection of the gravitational potential , @xmath38 ( see e.g. kaiser 1992 ) , @xmath39 the quantities here are the conformal distance or lookback time , from the observer ( @xmath40 ) and the analogous angular diameter distance @xmath41 . the lensing potential in equation ( 16 ) can be related to the well known convergence generally encountered in conventional lensing studies involving galaxy shear ( kaiser 1992 ) @xmath42 where note that the 2d laplacian operating on @xmath38 is a spatial and not an angular laplacian . expanding the lensing potential to fourier moments , @xmath43 we can write the usually familiar quantities of convergence and shear components of weak lensing as @xmath44 taking the fourier transform , as appropriate for a flat - sky , we write @xmath45 where @xmath46 the lensed power spectrum , according to the present formulation , is discussed in hu ( 2000 ) and we can write @xmath47 \ , c_l^{\theta}+ { \int { d^2 { \bf l}_1 \over ( 2\pi)^2 } } c_{\| { { \bf l}}- { { \bf l}}_1\|}^{\theta}c^{\phi\phi}_{l_1 } [ ( { { \bf l}}- { { \bf l}}_1)\cdot { { \bf l}}_1]^2 \ , . \nonumber \\ \label{eqn : ttflat}\end{aligned}\ ] ] since the lensing effect involves the angular gradient of cmb photons and leaves the surface brightness unaffected , its signatures are at the second order in temperature . effectively , lensing smoothes the acoustic peak structure at large angular scales and moves photons to small scales . note that the approach above treats the deflection angle as a perturbative parameter . such an approximation breaks down at angular scales corresponding to roughly the rms deflection angle of a cmb photon ( @xmath1 a few arcminutes ) . while the approach can be extended to higher order ( e.g. , cooray 2004c ) , the numerical calculation becomes increasingly time consuming . the development of numerically fast , but still accurate , approaches ( e.g. , challinor & lewis 2005 ) will aid in analysis data from future high resolution cmb experiments . to put rough estimates on the expected changes due to lensing , the mean gradient is of order @xmath1 15 @xmath6k arcmin@xmath48 and with deflection angles or order 0.5 arcmin or so from massive clusters , the lensing effect results in temperature fluctuations of order @xmath1 5 to 10 @xmath6k . in fig . 1 , we show the effect of lensing on temperature fluctuations where power is transferred from acoustic peaks to the damping tail of anisotropy spectrum . to calculate the lensing contribution to the polarization , we follow the notation in hu ( 2000 ) . introducing , @xmath49 where @xmath50 represent combinations of the stokes parameters . the lensing effect is again to move the photon propagation directions on the sky . in fourier space , assuming flat - sky , we can consider the e- and b - mode decomposition introduced in kamionkowski et al . ( 1997 ; seljak & zaldarriaga 1997 ) such that @xmath51 and @xmath52 since primordial @xmath53 , due to the gravitational wave contribution , is small , we can make the useful approximation that @xmath54 \phi({{\bf l}}-{{\bf l}}_1 ) ( { { \bf l}}-{{\bf l}}_1 ) \cdot { { \bf l}}_1 \nonumber \\ \tilde b({{\bf l } } ) & = & - { \int { d^2 { \bf l}_1 \over ( 2\pi)^2 } } \left[e({{\bf l}}_1 ) \sin 2 ( \varphi_{{{\bf l}}_1 } -\varphi_{{\bf l}})\right ] \phi({{\bf l}}-{{\bf l}}_1 ) ( { { \bf l}}-{{\bf l}}_1 ) \cdot { { \bf l}}_1 \ , , \end{aligned}\ ] ] under this approximation , the lensed polarization power spectra can be expressed in terms of @xmath55 and , in the case of b - modes , we can write @xmath56 ^ 2 c^{ee}_{l_1 } [ 1 - \cos(4\varphi_{{{\bf l}}_1 } ) ] \ , , \nonumber \\\end{aligned}\ ] ] where @xmath57 is the contribution due to igws and the contribution in the second line is the confusion that must be removed . to achieve this separation , one must obtain information on the integrated potential out to last scattering surface . one can effectively extract information related to the lensing potential , and the integrated dark matter density field responsible for the lensing effect , via quadratic statistics applied to cmb temperature and polarization maps ( hu 2001a , hu 2001b ; hu & okamoto 2003 ; cooray & kesden 2003 ; kesden et al 2003 ) or likelihood - based techniques ( hirata & seljak 2003 ) . in figure 2 , we show the contribution from dominant scalar modes to the polarization in the gradient component and from gravitational - waves to the curl polarization . unlike the b modes generated by tensor perturbations ( the igws ) , @xmath58 is dominated by larger amplitude scalar perturbations . the expected few - percent conversion of e modes creates a large signal in the b - mode power spectrum ( zaldarriaga & seljak 1998 ) . for tensor - to - scalar ratios @xmath59 below @xmath60 or @xmath33 below @xmath61 gev the igw signal is completely confused by the lensing contaminant ( lewis et al . 2002 ) . to bypass this limit one must separate the lensing induced b - modes from those due to igws ( kesden et al . 2002 ; knox & song 2002 ; seljak & hirata 2004 ) . clearly , the lensing confusion could be exactly removed if one knew the three - dimensional distribution of mass out to the cmb last - scattering surface . however , knowledge of the projected quantity @xmath62 is sufficient . one way to estimate @xmath62 is by using quadratic estimators or maximum likelihood methods to statistically infer the deflection - angle field given temperature and polarization anisotropies ( hu 2001b ; hu & okamoto 2002 ) . these method allow one to detection down @xmath63 gev ( kesden et al . 2002 ; knox & song 2002 ) . in the case of likelihood methods , one does not have an obvious ultimate limit , however , the residual lensing noise is still above the detector noise even for highest sensitive polarization maps . while this is encouraging , it is useful to consider alternative techniques to reduce the confusion . a useful way to estimate @xmath62 , but out to a high - redshift , is by observing the weak - lensing distortions of the shapes of objects of a known average shape at redshift @xmath64 . as the cmb source redshift is at @xmath65 , observations of the weak lensing of galaxies , out to a redshift of @xmath66@xmath67 , can not be used to effectively delens cmb maps because a large fraction of the lensing contamination ( 55% at @xmath68 ) comes from structure at @xmath69 . a possibility exists in the form anisotropies in the 21 cm brightness temperature fluctuations during the era and prior to reionization ( sigurdson & cooray 2005 ) . for example , using a quadratic reconstruction of the deflection field of 21 cm anisotropies out to a source redshift of 30 , in conjunction with a low - resolution but high signal - to - noise cmb polarization experiment , could detect @xmath70 or @xmath71 gev if 21 cm fluctuations measurements are limited to @xmath72 . if the resolution in 21 cm observations is improved so that one can measure 21 cm brightness temperature anisotropy spectrum out to @xmath73 , one can improve down to @xmath74 or @xmath75 gev . this limit is comparable to using cmb lensing information alone from high sensitive maps at 2 arcminute resolution ( seljak & hirata 2004 ) . to achieve such a resolution one requires roughly a space mission with an aperture of diameter @xmath76 meters . in the case of cmb lens - cleaning with information from 21 cm observations , as the igw signal is at multipoles of 100 ( in the case of recombination bump ) and/or 20 ( in the case of reionization with optical depths around 0.1 ) , the cmb data need not be high resolution . since lensing information is not extracted from cmb observations , the optimal observing strategy is to integrate over a few square degree patch of the sky as proposed in kamionkowski & kosowsky ( 1998 ) . thus , it is conceivable that the inflationary probe of nasa s beyond einstein program can be designed in combination with a cosmic 21-cm radiation experiment . such a cosmic 21-cm radiation experiment could be extremely exciting for reasons involving 21 cm fluctuations alone where one probes the small scale structure of the universe at high redshifts . for a bias - limited reconstruction out to a @xmath77 , the limit on the tensor - to - scalar ratio is @xmath78 or @xmath79 gev . for @xmath80 , the maximum redshift where 21-cm fluctuations are expected to be nonzero , one could probe down to @xmath81 gev . beyond cmb polarization data , there is long - term interest , as part of nasa s beyond einstein program , to launch a space - based laser interferometer ( big bang observer ) to directly detect the presence of relic stochastic gravitational background today from inflation . the cmb detection and the detection of the laser interferometers provide unique opportunities to study properties related to inflationary physics . to understand requirements for the direction detection , we summarize some of the key aspects related to the stochastic background . the evolution of tensor perturbations , for scales inside the horizon subsequent to inflation , is governed by the massless klein - gordon equation @xmath82 subject to the boundary conditions , @xmath83 and @xmath84 . @xmath85 is conformal time , which is related to cosmic time by the differential relation , @xmath86 . we define @xmath87 . with this definition , we are able to rewrite the above equation , @xmath88 there are two limiting behaviors for @xmath89 : before horizon entry @xmath90 ; after horizon entry @xmath89 will oscillate which implies @xmath91 will oscillate with an amplitude decreasing as @xmath92 . therefore , the current spectrum of gravitational waves is determined by the primordial power spectrum and by the rate at which scales enter the horizon ( i.e. the evolution of the scale factor ) : @xmath93 . at horizon crossing , note that @xmath94 . during radiation domination , @xmath95 , so that @xmath96 and during matter domination @xmath97 , so that @xmath98 . from these relations , we find that , @xmath99 in order to quantify these scalings , a transfer function is used ( turner , white & lidsey 1993 ) . we note that during the matter dominated epoch , @xmath100 and the above equation becomes a bessel equation ( see , discussions in pritchard & kamionkowski 2004 ) . requiring that the solution approach unity at @xmath101 , we find that @xmath102 using this , we define the transfer function , @xmath103 ( where @xmath104 is the wavenumber of matter - radiation equality @xmath105 mpc@xmath48 ) as , @xmath106 where the average is taken over several periods around the time that we want to evaluate the transfer function . from our scaling arguments , the transfer function must take the form of @xmath107 the spectrum as seen by a gravitational wave detector is ( turner , white & lidsey 1993 ) , @xmath108 where @xmath109 mpc for @xmath110 km / s mpc . @xmath111 is the ratio of the energy density in gravitational energy to that of the closure density , @xmath112 . in order to determine the values for @xmath113 and @xmath114 we need to numerically integrate the evolution equation and determine @xmath115 . since space - based gravitational wave detectors that are under consideration for detection of inflationary gravitational waves have sensitivities around a frequency of @xmath116hz ( @xmath117 mpc@xmath48 ) , our analysis considers only wave numbers in excess of @xmath104 . therefore , the values determined by turner , white and lidsey ( 1993 ; @xmath118 ) are applicable , even though they assumed the universe was filled with only matter and radiation . here , we have ignored one important detail ; the right hand side of time evolution equation is not strictly zero . instead , tensor perturbations are sourced by anisotropic stress in the cosmic fluid . the largest source of such stresses comes from the free - streaming of relic neutrinos ( weinberg 2004 ) . it can be shown that these neutrinos cause the amplitude of gravitational waves that enter the horizon well before matter radiation equality to be damped by a factor of @xmath119 . this will influence the amplitude of gravitational waves detectable with space - based detectors . similarly , the amplitude seen by cmb polarization is damped between a few to a few ten percent with longest waves damped less . in fig . 4 , we show the amplitude of tensor fluctuations related to a background of inflationary gravitational waves as a function of the gravitational wave frequency . we plot roughly 25 orders of magnitude in the frequency from super - horizon to horizon scales probed by cmb to a million - km wavelength modes detectable with space - based interferometers . for reference , we also show various existing limits on the stochastic backgrounds based on calculations related to the big bang nucleosynthesis ( allen 1996 ) to timing - limit for millisecond pulsars ( e.g. , thorsett & dewey 1996 ) . an extended discussion related to such limits are in seto & cooray ( 2005 ) . the left - panel of fig . 3 shows @xmath120 which is the fractional density contribution from a background of stochastic gravitational waves with the density @xmath121 : @xmath122 for reference , in the right - panel , we also plot the dimensional strain @xmath123 given by the @xmath124 where @xmath125 is the power spectral density of gravitational waves , @xmath126 . the two quantities , @xmath120 and @xmath127 are related through ( maggiore 2000 ) , @xmath128 in addition to single detector noise levels , in fig . 4 , we also show the improvement when the signal from two independent detectors can be combined through a correlation analysis between the two detectors ( e.g. , maggiore 2000 ) . such a correlation increases sensitivity to a stochastic background such that , @xmath129 where @xmath130 is the bandwidth over which the signals can be correlated , @xmath21 is the integration time , and @xmath131 is a filling factor , which accounts for the fact that a primordial stochastic background will be isotropic on the sky , but the detector will only be sensitive to a fraction of the sky . for omni - directional interferometers , such as the laser interferometer space antenna ( lisa ) , @xmath132 , while @xmath133 is the char acteristic strain due to detector noise . the single - detector sensitivity is @xmath134 . here , @xmath135 is the _ overlap - reduction factor _ that takes in to account the relative orientation and configuration of two interferometer systems . this correction results in an overall reduction of noise from the naive estimates that do not take in to account detail configuration and motion of the detectors with respect to each other in the orbit ( e.g. , ungarelli & vecchio 2001 ) . for a correlation analysis , the increase in sensitivity is under the assumption that the detector noises are independent between the two detectors , while the only correlation expected is due to stochastic signals such as inflation . we note here that for the single detector case , the minimum observable strain is independent of the integration time while for a correlation analysis , long - term observations are essential . while lisa will not allow an opportunity for such a correlation analysis , some concept studies for the big bang observer consider two systems such that the improvement related to the correlation analysis can be exploited . in fig . 5 , we concentrate on the frequency regime relevant for space - based detectors . in addition to detector noise and the expected level of amplitude related to inflationary gravitational waves , based on the energy scale of inflation , we also show the confusion related to foreground stochastic backgrounds related to merging compact binaries . these signals include cosmological neutron star - neutron star binaries ( e.g. , schneider et al . 2001 ) , white dwarf - white dwarf binaries ( e.g. , farmer & phinney 2003 ) , neutron star - white dwarf binaries ( e.g. , cooray 2004d ) , and the background related to single rotational neutron stars ( ferrari et al . 1999 ) . in addition to these compact binaries , a cosmological background of hypothetical pop iii supernovae could potentially contribute a substantial background at frequencies corresponding to the big bang observer ( e.g. , buonanno et al . 2004 ) . to jointly constrain the inflaton potential using observations at horizon - scale wavelengths with cmb and million - km scale with space - based interferometers , one can extend the cmb - based technique discussed in turner ( 1993 ) . the detectable amplitude , and the spectral index , with space - based detectors can be used to establish the inflaton potential amplitude and its first derivative . combining this information with cmb , at roughly 30 orders of magnitude difference in scales , should provide additional information to reconstruct the shape of the potential . while a priori assumed potential can easily be constrained based on the shape of the potential al , for a general reconstruction of the potential shape , the simple approximations considered in the literature , based on power - law expansions of the potential , however , may not be appropriate given the large difference in scales . a detailed analysis on the extent to which inflation models can be studies with joint constraints between cmb and direct gravitational - wave detections will be discussed elsewhere ( smith et al . 2005 ) . _ acknowledgments : _ author thanks marc kamionkowski , mike kesden , naoki seto , kris sigurdson , and tristan smith for collaborative work . the author especially thanks tristan smith for his help with writing review sections related to inflation and the direct detection of inflationary gravitational waves described in figures 4 and 5 . this work was supported by the sherman fairchild foundation , nasa , and the department of energy . challinor , a. , lewis , a. , astro - ph/0502425 cooray , a. 2004a , phys . d. , 70 , 063509 cooray , a. 2004b , phys . d. , 70 , 023508 cooray , a. 2004c , new astron . , 9 , 173 cooray , a. 2004d , mon . not . soc . , 354 , 25 cooray , a. , hu , w. , tegmark , m. 2000 , apj , 544 , 1 cooray , a. , baumann , d. , sigurdson , k. 2004 , in `` background microwave radiation and intracluster cosmology '' , eds . f. melchiorri & y. rephaeli , italian physical society ( astro - ph/0410006 ) .	the curl - modes of cosmic microwave background ( cmb ) polarization probe horizon - scale primordial gravitational waves related to inflation . a significant source of confusion is expected from a lensing conversion of polarization related to density perturbations to the curl mode , during the propagation of photons through the large scale structure . either high resolution cmb anisotropy observations or 21 cm fluctuations at redshifts 30 and higher can be used to _ delens _ polarization data and to separate gravitational - wave polarization signature from that of cosmic - shear related signal . separations based on proposed lensing reconstruction techniques for reasonable future experiments allow the possibility to probe inflationary energy scales down to 10@xmath0 gev . beyond cmb polarization , at frequencies between 0.01 hz to 1 hz , space - based laser interferometers can also be used to probe the inflationary gravitational wave background . the confusion here is related to the removal of merging neutron star binaries at cosmological distances . given the low merger rate and the rapid evolution of the gravitational wave frequency across this band , reliable removal techniques can be constructed . we discuss issues related to joint constraints that can be placed on the inflationary models based on cmb polarization information and space - based interferometers such as the big bang observer .
solid molecular hydrogen is the simplest and most fundamental quantum molecular crystal . at high pressure it is predicted to become not only a metallic atomic solid @xcite but also a high-@xmath2 superconductor @xcite and perhaps a superfluid @xcite . although there are experimental results for the structure and phase diagram of high - pressure hydrogen at room temperature , @xcite the crystal structures appearing in the complex low - temperature high - pressure phase diagram are not yet completely determined . spectroscopic investigations using infra - red ( ir ) and raman methods suggest the existence of three phases @xcite . phase i , which is stable up to 110 gpa , is a molecular solid composed of quantum rotors arranged in a hexagonal - close - packed ( hcp ) structure . changes in the low - frequency regions of the raman and ir spectra imply the existence of a second phase above 110 gpa ; this is known as phase ii , or the broken symmetry phase . the continuous behavior of the raman vibron frequency across the i to ii transition suggests that phase ii may consist of an arrangement of molecules on an hcp lattice , as in the p@xmath3/m structure . calculations based on density functional theory ( dft ) predict orthorhombic cmc@xmath4 , pca@xmath4 , and p@xmath4/c structures @xcite . the molecular centres in these structures lie close to those of an hcp lattice , but are incompatible with neutron data for solid deuterium @xcite . the appearance of phase iii at 150 gpa is accompanied by a large discontinuity in the raman spectrum and a strong rise in the spectral weight of the ir molecular vibrons . although the structures of phase ii and phase iii are not known for certain , recent x - ray results @xcite at pressures up to 183 gpa suggest that the hydrogen molecules in both phases lie close to the points of an hcp lattice . recent infrared and optical absorption results @xcite indicate that the molecular structure of phase iii persists up to 360 gpa , with no evidence of the onset of a metallic state . computational methods used to investigate the high - pressure low - temperature phase diagram of solid molecular hydrogen include density functional theory @xcite and quantum monte carlo ( qmc ) @xcite . approaches used to explore the phase space of possible structures include the evolutionary method of oganov and glass @xcite and the static ab - initio structure searching method of pickard and needs , @xcite which has also been employed to determine the ground - state structure of _ atomic _ metallic hydrogen @xcite . recently , a series of four papers @xcite presented a fresh look at the high - pressure hydrogen problem . by studying the equalization of the intra- and inter - molecular h - h bond lengths in dense hydrogen , these papers investigated the four most stable structures found by pickard and needs @xcite using dft in the static approximation ( i.e. , ignoring the zero - point vibrational energy of the h atoms ) . however , all of the recent calculations of the metallization pressure and phase diagram are based on the single - particle mean - field - like dft method . this has various drawbacks , such as the band - gap problem and the sensitivity of the results to the choice of exchange - correlation ( xc ) functional , which affects both the relative stabilities of different phases and the predicted metallization pressure @xcite . furthermore , attempts to explore the phase space of possible structures have been based primarily on dft in the static approximation , which ignores the proton zero - point energy ( zpe ) . since the zpe is normally larger than the energy differences between various candidate structures , structure - searching methods including an approximate zpe should be considered . most of the existing dft calculations of solid hydrogen have used the local density approximation ( lda ) or the generalized gradient approximation ( gga ) to the unknown exchange - correlation functional . neither of these functionals adequately describes van der waals ( vdw ) interactions . in this paper , we use qmc simulations to determine the ground - state phase diagram of solid molecular hydrogen in the pressure range @xmath5 gpa . we study , systematically and accurately , the two main factors affecting the quality of the calculated phase diagram : finite - size errors and proton zero - point energy . to investigate the mechanism of metallization and determine the pressure at which the band gap closes , we also carry out many - body perturbation theory calculations within the @xmath0 approximation . we focus on four specific structures , with space groups p@xmath3/m , c2/c , cmca-12 , and cmca . according to ground - state static dft calculations using the perdew - burke - ernzerhof ( pbe ) version of the generalized gradient approximation @xcite ( gga ) , these are stable in the pressure ranges @xmath6105 , 105270 , 270385 , and 385490 gpa , respectively @xcite . the fixed - node diffusion quantum monte carlo ( dmc ) algorithm @xcite yields the energy expectation value of the variationally optimal wave function with the same nodal surface ( surface of zeros ) as a chosen many - electron trial function . the method has a long history and has been thoroughly benchmarked @xcite . for small molecules , where almost exact results can be obtained using quantum chemical methods , ground - state total energies calculated using fixed - node dmc are consistently more accurate than energies calculated dft , typically by a factor of 10 or more . unlike dft , fixed - node dmc also provides a good description of the dependence of the total energy on coordination number ( very important when comparing the energies of different crystal structures ) and a fairly good description of the potential surface during bond - breaking processes . for large systems such as the ones studied in this paper , fixed - node dmc is the most accurate known ground - state total energy method and the only possible benchmarks are against experiment . again , however , dmc consistently outperforms dft . for example , the measured cohesive energy of crystalline si is 4.62 ev / atom ; fixed - node dmc simulations yields 4.63 ev / atom @xcite ; and dft in the local spin - density approximation yields 5.28 ev / atom . in this unusually favorable case , the dmc result is 65 times more accurate than the dft result . the ability of dmc to describe changes in coordination is illustrated by the example of the si self - interstitial defect @xcite . the dmc value of the defect formation energy , which is in good agreement with experiment , is about 1 ev larger than the pw91 gga value and 1.5 ev larger than the lda values . the main drawback of dmc , which explains why it is not more widely used , is the computational cost : a dmc simulation typically costs about @xmath7 times more than a dft simulation of the same system . dmc can also be used to obtain limited information about excited states . however , other approaches such as the @xmath0 method used in this work are generally better for excitation spectra . the results of @xmath0 calculations of the band gaps of @xmath8-bonded materials generally match experiment to within a few tenths of an ev . our fixed - node dmc simulations used the casino qmc code @xcite and a trial function of slater - jastrow ( sj ) form , @xmath9\det[\psi_{n}({\bf r}_i^{\uparrow})]\det[\psi_{n}({\bf r}_j^{\downarrow } ) ] , \ ] ] where @xmath10 is a @xmath11-dimensional vector that defines the positions of all @xmath12 electrons , @xmath13 is the position of the ith spin - up electron , @xmath14 is the position of the jth spin - down electron , @xmath15 $ ] is the jastrow factor , and @xmath16 $ ] and @xmath17 $ ] are slater determinants of spin - up and spin - down one - electron orbitals . these orbitals were obtained from dft calculations using the plane - wave - based quantum espresso code @xcite . a norm - conserving pseudopotential constructed within dft using the pbe exchange - correlation functional was employed . @xcite previous work has shown that the pseudopotential approximation has a very small impact on the results of calculations of high - pressure solid hydrogen @xcite . we chose a very large basis - set cut - off of 300 ry to guarantee converge to the complete basis set limit @xcite . the plane - wave orbitals were transformed into a blip polynomial basis @xcite . our jastrow factor consists of polynomial one - body electron - nucleus ( en ) and two - body electron - electron ( ee ) terms , the parameters of which were optimized by variance minimization at the variational monte carlo ( vmc ) level @xcite . geometry and cell optimizations were carried out using dft with a dense @xmath18 * k-point mesh . the bfgs quasi - newton algorithm , as implemented in the quantum espresso package , was used for both cell and geometry optimization , with convergence thresholds on the total energy and forces of 0.01 mry and 0.1 mry / bohr , respectively , to guarantee convergence of the total energy to less than 1 mev / proton and the pressure to better than 0.1 gpa . an investigation of the proton - proton distance of accurately relaxed structures in the pressure range of 100 - 450 gpa reveals two main features . first , as the pressure increases , the intra - molecular h@xmath19 bond length ( nearest - neighbor distance ) increases slightly , rising from 0.735 to 0.778 . this could be a consequence of a weakening of the h@xmath19 bonds as the coordination of the h atoms increases . second , unlike the intra - molecular h@xmath19 bond length , the inter - molecular distance decreases smoothly as the pressure rises . details of the behavior of the inter - atomic distance with pressure and of the experimental evidence are presented in ref @xcite . our recent dft work @xcite showed that hybrid exchange - correlation functionals including a small fraction of exact exchange favor insulating phases relative to metallic phases . we therefore investigate how the choice of density functional used to calculate the one - electron orbitals appearing in the slater determinants ( and thus to define the fixed nodal surface ) affects our qmc results . for this purpose , we compared the vmc and dmc energies of the insulating p@xmath3/m phase ( which has the largest band gap of the four structures considered ) using orbitals calculated using the semi - local pbe gga functional @xcite and the pbe0 hybrid functional including 25% exact exchange @xcite . the vmc and dmc calculations were performed at the @xmath20 point of the brillouin zone of a simulation cell containing 128 hydrogen atoms and @xmath21 replicas of the p@xmath3/m unit cell . the vmc and dmc energies at pressures 100 gpa , 200 gpa and 300 gpa , are given in . @lllll pressure & pbe - vmc & pbe0-vmc & pbe - dmc & pbe0-dmc + 100 & -15.1335(4 ) & -15.1397(4 ) & -15.2327(3 ) & -15.2338(3 ) + 200 & -14.5898(4 ) & -14.5979(4 ) & -14.6856(3 ) & -14.6867(3 ) + 300 & -14.0473(4 ) & -14.0524(4 ) & -14.2184(3 ) & -14.2193(3 ) + the difference between the pbe and pbe0 energies is about 6 mev / atom at the vmc level and 1 mev / atom at the dmc level . similar calculations for the cmca-12 structure at two different pressures gave similar results . this demonstrates that , in the case of solid molecular hydrogen , dmc energies calculated using pbe and pbe0 orbitals differ negligibly , even though the change of functional has a large effect on the dft phase diagram @xcite . the very weak dependence of the dmc energy on the choice of single - particle orbitals suggests , furthermore , that the fixed - node errors are small . this contrasts with the case of 3d transition metal compounds @xcite , where changing the dft exchange - correlation functional does affect the dmc results . when simulating infinite periodic systems or finite systems subject to periodic boundary conditions , it is not possible to use the familiar @xmath22 form of the coulomb interaction because the sums over images of the simulation cell do not converge absolutely . the standard solution to this problem is to replace the coulomb interaction by the ewald interaction . the three - dimensional ewald interaction is the periodic solution of poisson s equation for a periodic array of point charges embedded in a uniform neutralizing background and is therefore appropriate for an electrically unpolarized , neutral system @xcite . all dmc results reported in this paper were calculated using the ewald @xcite electron - electron interaction . in dft , the limit of infinite system size can be approached by improving the @xmath23-point sampling , but qmc calculations have to be performed using increasingly large simulation cells . the resulting finite - size ( fs ) effects are a significant source of error . the following qmc results used c2/c , cmca-12 , p@xmath3/m and cmca simulation cells containing 96 , 96 , 128 , and 144 atoms , respectively , equivalent to @xmath21 , @xmath21 , @xmath21 and @xmath24 repetitions of the corresponding unit cells . qmc fs errors are conventionally separated into one - body and two - body contributions . one - body ( independent - particle ) errors arise from the non - interacting kinetic , potential and hartree energies ; two - body errors arise from the effects of exchange and correlation on the coulomb and kinetic energies . this work accounts for fs errors by correctly and accurately combining the use of twist - averaged boundary conditions ( tabc ) @xcite with dft - based corrections obtained using the size - dependent kwee - zhang - krakauer ( kzk ) functional @xcite . by construction , kzk corrections remove the finite - size errors in dmc calculations of cubic homogeneous systems exactly ; our systems are neither cubic nor homogeneous , but we show below that the kzk approach still works remarkably well . our dmc - tabc energies were averages over 12 random twists , corresponding to 12 randomly translated monkhorst - pack @xmath25-point meshes . the kzk correction is given by @xmath26 , where @xmath27 is the dft energy computed within the local density approximation ( lda ) using a fully converged k*-point mesh ( which in this work means @xmath28 ) , and @xmath29 is obtained from dft - kzk calculations employing the same simulation cell and twists as our dmc - tabc results . in the case of the metallic cmca structure , we also carried out dmc - tabc calculations using 24 twists ; the change in the total energy was about 1.4 mev / proton at @xmath30 gpa , implying that dmc energies obtained by combining tabc and kzk fs - corrections converge very rapidly as the number of twists is increased . @ll n & dmc - tabc + 64 & -14.1829(6 ) + 96 & -14.1931(3 ) + 128 & -14.1936(3 ) + 144 & -14.1927(4 ) + to check the accuracy and convergence of the dmc energy as a function of the number of protons @xmath12 in the simulation cell , we performed calculations for different @xmath12 . shows the results for the metallic cmca phase at @xmath31 gpa . the convergence of the tabc - dmc energy with increasing @xmath12 is excellent and there is no difficulty in reaching the accuracy of 1mev per proton necessary for high - pressure hydrogen phase diagram calculations . we emphasize the importance of our approach to the treatment of finite - size errors as an application of tabc to real systems @xcite . to find the metallization pressure , quasi - particle band gaps were calculated within the @xmath0 approximation using the yambo code @xcite . non - self - consistent @xmath0 calculations were performed starting from dft wave functions , using a plasmon pole approximation for the dielectric function @xcite . following convergence checks , the numbers of unoccupied bands and @xmath23-points were set to 200 and 512 , respectively . we also checked convergence with respect to the number of plane waves , which was set to 1800 for the p@xmath3/m structure and 2700 for the cmca-12 and c2/c structures . it is important to point out that our @xmath0 and dmc results are independent of one another . dmc is used to calculate the enthalpy as a function of pressure , whist @xmath0 is used to calculate the quasi - particle band gap as a function of pressure . both methods use the same relaxed dft crystal structures , but they are otherwise unrelated . in the following , we present the metallization pressure of the cmca-12 phase as obtained using the pbe , pbe0 , and @xmath0 approximations . we use the simplest estimate of the metallization pressure , which is the pressure at which the energy band gap closes . to estimate the metallization pressure from the calculated band gaps at a few different pressures , we linearly extrapolate the band gap as a function of pressure . figure [ cmca12 ] illustrates band - gap closure in the cmca-12 phase . the pbe , pbe0 , and @xmath0 results show that the band gap of cmca-12 vanishes at pressures of 245 , 309 , and 373 gpa , respectively . the pbe metallization pressure is in agreement with the value estimated by pickard and needs @xcite . from band - structure calculations we see that the cmca-12 phase has an indirect band gap ; the linear decrease of the band gap with pressure is not unexpected in indirect gap structures and is also reported experimentally @xcite . as usual , band gaps calculated using the semi - local pbe functional are much too small . a significant contributor to this band - gap underestimation is the inability of _ any _ analytic functional of the density to represent the discontinuous change in exchange - correlation potential that occurs as a single electron is added to a system with a filled band @xcite . by including a fraction of the exact non - local orbital - dependent exchange energy , the pbe0 hybrid functional circumvents this problem to some extent . however , even the pbe0 band gap is considerably smaller than the @xmath0 band gap , which is expected to be more accurate . ( color online ) band gap of the cmca-12 crystal structure as a function of pressure as calculated using the pbe , pbe0 , and @xmath0 approximations . ] our @xmath0 results show that the band gaps of the p@xmath3/m , c2/c , and cmca-12 structures vanish at the static pressures of 446 , 398 , and 373 gpa , respectively . semilocal dft , hybrid - dft @xcite , and @xmath0 calculations all show that p@xmath3/m has the largest band gap and , consequently , the largest metallization pressure of the three insulating structures considered . the @xmath0 indirect band gap and its pressure dependence agree very well with recently reported experimental results @xcite . in the following , we present enthalpy - pressure diagrams obtained from dmc simulations at the static level of theory , which ignores proton zpe . figure [ h - stat - p - stat ] illustrates @xmath32 as a function of pressure . within the static approximation , the p@xmath3/m structure is the most stable up to pressures of approximately 250 gpa , but is less stable than both c2/c and cmca-12 by 350 gpa . the error bars make it difficult to establish exactly where the transition from p@xmath3/m to c2/c or cmca-12 takes place or to say which of c2/c or cmca-12 is favored . the transition from c2/c or cmca-12 to the metallic cmca phase happens around 375 gpa . thus , within the static approximation , dmc predicts that the hexagonal close packed p@xmath3/m structure is more stable than the other structures investigated over the entire pressure range at which phases i and ii exist experimentally , and that it remains more stable well above the measured transition to phase iii . as already discussed above , recent x - ray results @xcite at pressures up to 183 gpa suggest that hcp symmetry is a strong candidate not only for phase ii but also for phase iii . we claim that our static dmc phase diagram is much more accurate than the static dft phase diagram . one reason is the importance of vdw interactions between hydrogen molecules , which significantly influence the phase diagram , especially at low densities . qmc calculations have recently shown that the vdw interaction has a substantial effect on the transition pressures between the crystalline phases of ice at high pressure @xcite and one would expect them to matter even more in solid molecular hydrogen . an important consequence is that transition pressures obtained from standard dft functionals , which neglect vdw forces , are likely to be unreliable . because of this recognition , some very recent work @xcite has used van der waals density functionals to study solid h. there is no need to emphasize that dmc provides a highly accurate description of exchange and correlation in relatively weakly correlated materials and that it includes an excellent quantitative treatment of vdw forces . combining our static dmc phase diagram and @xmath0 results suggests : ( a ) that the well known rule of thumb `` the lower the energy the wider the gap '' @xcite is valid here ; and ( b ) that metallization occurs via a transition from the zero band - gap cmca-12 structure to the metallic cmca structure at around 375 gpa . the agreement between the pressure at which the @xmath0 band gap of the cmca-12 structure closes and the pressure at which dmc predicts a phase transition from cmca-12 to the metallic cmca structure is remarkable but fortuitous , since including the effects of phonon zero - point energy changes the picture . the above scenario is based on the static approximation . in reality , even in the ground state , h@xmath19 molecules are quantum oscillators with a natural frequency of @xmath33@xmath34 hz ( for the h h bond stretching ) and a significant zpe . following standard practice in the field , our treatment of zpe uses the harmonic approximation . a full analysis of anharmonicity is beyond the scope of this paper , but estimating the zpe within the harmonic approximation is feasible @xcite and widely believed to produce accurate results in the case of the molecular solid hydrogen . the harmonic approximation has also been used for atomic metallic hydrogen @xcite . to include the effects of zpe , phonon frequencies were calculated using density functional perturbation theory as implemented in quantum espresso @xcite . the zpe per proton was estimated as @xmath35 , where @xmath36 . here @xmath37 and @xmath38 are the number of vibrational modes in the unit cell and the number of phonon wave vectors @xmath39 , respectively , @xmath40 is the angular frequency of the phonon in band @xmath41 with wave vector @xmath39 , and the summation over @xmath39 includes all @xmath23-points on a @xmath42 grid in the brillouin zone . mcmahon @xcite demonstrated that a @xmath39-point grid of this density is sufficient to converge zpe differences between structures to within a few percent . in agreement with previous dft results @xcite , we find that the zpe is large , structure dependent , and density dependent , ranging from 257 mev per proton at low density to 351 mev per proton at high density . the p@xmath3/m structure has the largest and the cmca structure the smallest zpe over the whole range of pressures considered . the c2/c and cmca-12 structures have almost the same zpe , which lies between the p@xmath3/m and cmca values . by defining @xmath43 , where @xmath44 , we obtain the dynamic phase diagram of solid molecular hydrogen illustrated in fig . [ h - dyn - p - stat ] . bearing in mind the error bars shown in fig . [ h - dyn - p - stat ] , we see that p@xmath3/m is the most stable phase up to a pressure of approximately 220 gpa . this is in good agreement with experimental results @xcite . as discussed above , because of the tiny enthalpy difference between the c2/c and cmca-12 phases , it is difficult to determine their pressure stability ranges accurately . however , we are able to conclude that we expect to observe the following structural transitions with increasing pressure : p@xmath3/m @xmath45 c2/c @xmath45 cmca-12 @xmath45 cmca . comparing the static and dynamic results , we see that the proton zpe has two main effects . first , although the order in which the phases appear with increasing pressure is unaffected , their enthalpies and regions of stability change . if we were to take an optimistic view and ignore the error bars , we would say that p@xmath3/m , c2/c , cmca-12 , and cmca phases are now stable in the pressure ranges @xmath6220 gpa , 220360 gpa , 360430 gpa , and @xmath46430 gpa , respectively . as in the static case , however , the error bars make it difficult to discriminate between the c2/c and cmca-12 structures over a wide range of pressures , this time from 300 to approximately 450 gpa . second , the inclusion of zpe increases the pressure of the transition to the metallic cmca phase by @xmath47 , which is close to the @xmath48 increase previously reported @xcite . if the band gap is of order 0.5 ev ( the vibron energy ) or less , which is comparable to the cmca-12 gap in this pressure range , the inclusion of zpe effects is also expected to increase the gap . @xcite the increase is small , however , and the gap still closes before the transition to the cmca structure takes place . this leaves a region of 2030 gpa where the cmca-12 phase persists but is no longer insulating . we , also , found that the zp pressures are quite small and has no significant effect on the phase diagram . a comparison of the densities of different structures at the same pressure shows that , in phases ii and iii : density(cmca ) @xmath46 density(cmca-12 ) @xmath46 density(c2/c ) @xmath46 density(p@xmath3/m ) . as the pressure is increased , the @xmath49 term in the enthalpy rises faster for the low - density insulating phases than for the metallic phase , reducing the relative stability of the insulating phases and leading eventually to the metal - insulator transition . the difference between the densities of the insulating phases and the metallic cmca phase reduces at high pressure , especially in the case of the cmca-12 structure . the differences between the dmc and dft phase diagrams are primarily due to the internal energy term in the enthalpy . for pressures up to 220 gpa , the dmc energy of the low density p@xmath3/m phase lies further below the energies of the other phases , especially the high - density metallic cmca phase , than does the dft energy . our results , at both the static and dynamic levels , predict that the p@xmath3/m phase is the most stable of the structures considered for a large range of pressures up to 220 gpa in the dynamic case or 280 gpa in the static case . this prediction is consistent with the latest low temperature x - ray powder - diffraction results @xcite , which indicate that the hydrogen molecules in phases ii and iii are still in a hexagonal structure and that phase iii remains hexagonal up to the maximum experimental pressure of 183 gpa . the nature of the difference between phases ii and iii remains to be established , but may be related to a classical orientational ordering of the h@xmath19 molecules @xcite . in conclusion , we have used dmc simulations to study the phase diagram of the four most stable structures of high pressure hydrogen found from dft calculations . we accurately computed the fs corrections by correctly combining the use of tabc with the dft - kzk method . we also carried out @xmath0 calculations to find the metallization pressure . our @xmath0 results show that the band gap of the insulating cmca-12 phase closes at 373 gpa , whilst our dmc simulations in the dynamic approximation show that the cmca-12 structure may remain stable up to 430 gpa , at which point it transforms to the metallic cmca phase . we propose that the semimetallic state observed experimentally at around 360 gpa @xcite may correspond to the cmca-12 structure near the pressure at which the band gap closes . at high pressures our dmc results are in qualitative agreement with previous dft calculations @xcite , and suggest that the metallization of solid molecular hydrogen happens through the transformation of the cmca-12 phase to the cmca structure . our dmc results indicate that the hcp structure p@xmath3/m , which has the lowest density and the largest band gap of the phases considered , is the most stable molecular structure up to 220 gpa . we acknowledge interesting and helpful discussions with richard needs and chris pickard .	we use the diffusion quantum monte carlo ( dmc ) method to calculate the ground state phase diagram of solid molecular hydrogen and examine the stability of the most important insulating phases relative to metallic crystalline molecular hydrogen . we develop a new method to account for finite - size errors by combining the use of twist - averaged boundary conditions with corrections obtained using the kwee - zhang - krakauer ( kzk ) functional in density functional theory . to study band - gap closure and find the metallization pressure , we perform accurate quasi - particle many - body calculations using the @xmath0 method . in the static approximation , our dmc simulations indicate a transition from the insulating cmca-12 structure to the metallic cmca structure at around 375 gpa . the @xmath0 band gap of cmca-12 closes at roughly the same pressure . in the dynamic dmc phase diagram , which includes the effects of zero - point energy , the cmca-12 structure remains stable up to 430 gpa , well above the pressure at which the @xmath0 band gap closes . our results predict that the semimetallic state observed experimentally at around 360 gpa [ phys . rev . lett . * 108 * , 146402 ( 2012 ) ] may correspond to the cmca-12 structure near the pressure at which the band gap closes . the dynamic dmc phase diagram indicates that the hexagonal close packed @xmath1 structure , which has the largest band gap of the insulating structures considered , is stable up to 220 gpa . this is consistent with recent x - ray data taken at pressures up to 183 gpa [ phys . rev . b * 82 * , 060101(r ) ( 2010 ) ] , which also reported a hexagonal close packed arrangement of hydrogen molecules .
many of the over 380 exoplanets discovered to date orbit solar mass ( 0.71.5 @xmath1 ) stars . this observational bias occurs because the main targets of previous doppler spectroscopy - based exoplanet searches have been solar - type stars . these studies have revealed a variety of their planetary systems ( e.g. @xcite ; @xcite ) whose statistical properties are now used to constrain planet - formation models ( e.g. @xcite ) . however , only about 35 and 25 planets were detected around intermediate - mass ( 1.55 @xmath1 ) and low - mass ( @xmath3 0.7 @xmath1 ) stars in surveys of evolved g - k type ( sub ) giants and k - m type dwarfs , respectively , so far . the planetary systems around such stars are not well understood . clarifying the relationship between stellar mass and planetary system is important for understanding general planet formation because planet formation should depend on the properties of the protoplanetary disks , which be affected by the condition of host stars . evolved intermediate - mass ( sub ) giant stars are suitable targets for doppler spectroscopy - based planet searches because these stars have low surface activity and their spectra exhibit many sharp absorption lines . although the number of substellar companions found orbiting such stars is still small , some planetary system properties have begun to emerge . for example , the masses of planets and their host stars are correlated : more massive substellar companions tend to exist around more massive stars ( e.g. @xcite ) . this correlation suggests that the mass ranges of the brown dwarf desert depend on the host star mass @xcite . the planet occurrence rate also depends on the host star mass : the planet frequency around 1.52 @xmath1 giant stars is higher than that around lower - mass stars @xcite . moreover , the semi - major axes of planetary systems seem to be correlated with host star : the range for planets orbiting intermediate - mass giant stars exceeds 0.6 au , and the range for planets around solar - type stars is large than 0.02 au ( e.g. @xcite ) . these properties of substellar systems around intermediate - mass giant stars seem to be not similar to those around solar - type stars . a unified understanding of planetary systems over a wide range of host star masses will provide valuable insights into the dependence of substellar systems on the central stars and into general plant formation mechanisms . in 2005 , we started a joint planet search program between korean and japanese researchers to search for planets around gk - type giant stars using a precise doppler technique with using the 1.8-m telescope at boao and the 1.88-m telescope at oao @xcite . this survey program is an extended version of the ongoing oao planet search program @xcite and part of an international collaboration among researchers from korea , china and japan ( an east - asian planet search network , eaps - net ; @xcite ) . the collaboration aims at clarifying the properties of planetary systems around intermediate - mass stars by surveying more than 800 gk giants for planets at oao , boao , the xinglong station ( china ) , and the subaru telescope . for the korean - japanese planet search program , we selected about 190 target stars from the @xmath4 catalog based on the following criteria : color - index 0.6 @xmath3 @xmath5 @xmath3 1.0 , absolute magnitude @xmath63 @xmath3 @xmath7 @xmath3 2 , declination @xmath8 @xmath2 @xmath625 @xmath9 , and visual magnitude 6.2 @xmath3 @xmath10 @xmath3 6.5 @xcite . these targets are fainter than those of the oao and xinglong program ( @xcite ; @xcite ) . we divided the targets into two parts : one for boao and the other for oao . each is observed independently at the assigned observatory , although a star that exhibits a large radial velocity variation is observed intensively at both observatories . radial velocity observations at boao are carried out with the 1.8-m telescope and the boao echelle spectrograph ( boes ; @xcite ) , a fiber - fed high resolution echelle spectrograph . we place an iodine ( i@xmath11 ) cell in the optical path in front of the fiber entrance of the spectrograph @xcite for precise wavelength calibration and use a 200-@xmath12 m fiber , obtaining a wavelength resolution @xmath13 = @xmath14/@xmath15 @xmath16 51000 . the spectra cover the 350010500 wavelength range . echelle data reduction is performed using the iraf software package in the standard manner . we use the 5000 to 5900 wavelength range , a region covered by many i@xmath11 absorption lines , for radial velocity measurements . we also make use of ca ii h lines at around 3970 as chromospheric activity diagnostics . radial velocity analysis is performed using the spectral modeling technique described in @xcite , which is based on the method of @xcite , and improve and optimize for boes data analysis @xcite . we employ the extraction method described in @xcite to prepare a stellar template spectrum from stellar spectra taken through the i@xmath11 cell ( star+i@xmath11 spectra ) . the technique allows us to achieve a long - term doppler precision of 14 m s@xmath17 over 5.5 yr . radial velocity observations at oao are carried out with the 1.88-m telescope and high dispersion echelle spectrograph ( hides ; @xcite ) attached to the coud focus of the telescope . we use an i@xmath11 cell placed in the optical path in front of the slit of the spectrograph @xcite as a precise wavelength calibrator . we set the slit width to 200 @xmath12 m ( 0.76 " ) , providing a spectral resolution of 63000 . until november 2007 , we had taken star+i@xmath11 spectra with the 5000 to 6200 wavelength range , and had taken stellar spectra at the same wavelength range without the i@xmath11 cell for abundance analysis . since the hides ccd system was upgraded to a three ccd mosaic in december 2007 , we obtain simultaneous spectra in the full 3750 to 7550 wavelength range . we use the 5000 to 5900 wavelength ranges of star+i@xmath11 spectra for radial velocity measurements . echelle data reduction is performed using the iraf software package in the standard manner . stellar radial velocities are derived from the i@xmath11-superposed stellar spectrum using the spectral modeling techniques detailed in @xcite , giving a doppler precision of less than 8 m s@xmath17 over 5.5 yr . up to now , we have been monitoring the radial velocities of sample stars using 1.8-m boao telescope and 1.88-m oao telescope for 5.5 yr and identified many candidate stars with large radial velocity variations . and then , two massive intermediate - mass stars with a periodic variation caused by a substellar companion are found among them : a brown dwarf - mass companion orbiting hd119445 @xcite and a planetary companion orbiting a giant star . we monitored the radial velocity of hd 119445 for 2.3 years from the beginning of the survey at both observatories . the observed radial velocities of hd 119445 are shown in figure [ fig1 ] . the best - fit keplerian orbit derived from both the boao and oao data has a period @xmath18 @xmath19 410.2 @xmath20 0.6 days , a velocity semi - amplitude @xmath21 @xmath19 413.5 @xmath20 2.6 m s@xmath17 , and an eccentricity @xmath22 @xmath19 0.082 @xmath20 0.007 . the best - fit curve is shown in figure [ fig1 ] as a solid line overlaid on the observed velocities . the rms scatter of the residuals to the best - fit is 13.7 m s@xmath17 . this values is comparable to the typical radial velocity scatter ( 1020 m s@xmath17 ) of the g - type giants @xcite . adopting a stellar mass @xmath23 = 3.9 @xmath20 0.4 @xmath1 for hd 119445 , we obtained a semi - major axis @xmath24 = 1.71 @xmath20 0.06 au and a minimum mass @xmath25 = 37.6 @xmath20 2.6 @xmath0 for a brown dwarf - mass companion @xcite . a large radial velocity variation in the giant star with a planetary companion was found in the early boao radial velocity survey and we made intensive follow - up observations of the star at boao and oao . the observed radial velocities of the star are shown in figure [ fig2 ] . a best - fit keplerian orbit derived from both the boao and oao velocity data by a least square fit has a period @xmath18 @xmath19 157.57 d , a velocity semi - amplitude @xmath21 @xmath19 35.2 m s@xmath17 , and an eccentricity @xmath22 @xmath19 0.085 . the best - fit curve is shown in figure [ fig2 ] as a solid line overlaid on the observed velocities . the rms of the residuals to the best - fit are 11.2 m s@xmath17 for boao and oao data . this values is comparable to the typical radial velocity scatter of the g - type giants @xcite . adopting a host star s mass @xmath23 = 2.4 ( 2.02.6 ) @xmath1 , which was estimated from evolutionary track and fundamental stellar parameters of @xmath26 = 43 @xmath27 , @xmath28 = 4861 k , and [ fe / h ] = 0.15 , we obtained a semi - major axis @xmath24 = 0.77 au and a minimum mass @xmath25 = 1.7 @xmath0 for a planetary companion . their host stars , hd119445 and the giant star , have masses of 3.9 @xmath1 and 2.4 @xmath1 , respectively . the brown dwarf - mass companion orbiting hd 119445 is most massive substellar companions , and the planetary companion orbiting the giant star is lowest - mass companion , among those discovered around massive intermediate - mass ( 1.95 @xmath1 ) stars . in figure [ fig1 ] , we plot masses of the companions detected within semi - major axis of 3 au by precise doppler surveys against their host star s masses ; solar - mass stars ( 0.7 @xmath1 @xmath29 @xmath23 @xmath3 1.5 @xmath1 , _ open triangles _ ) , intermediate - mass subgiants and giants ( 1.5 @xmath1 @xmath29 @xmath23 @xmath29 5 @xmath1 , _ filled circles _ ) , intermediate - mass dwarfs ( _ open circles _ ) , and hd 119445 and the giant star ( _ stars _ ) ( e.g. _ the extrasolar planets encyclopadia _ ; updated version of figure 5 in @xcite ; this work ) . the detectable companion mass for a given host star mass depends on the orbital separation of its companion and the radial velocity jitter of the host star . assuming that typical radial velocity jitters @xmath30 for solar - mass stars , intermediate - mass subgiants ( 1.51.9 @xmath1 ) and giants ( 1.95 @xmath1 ) are @xmath16 5 m s@xmath17 , @xmath16 7 m s@xmath17 and @xmath16 20 m s@xmath17 , respectively , we estimated the lower limits of companion masses detectable by precise doppler surveys around a solar - mass star and intermediate - mass subgiant and giant at 3 au ( sold lines in figure [ fig1 ] ) , corresponding to companion masses that provide the amplitude of three times of typical radial velocity jitters . we also indicate detectable masses for these stars at 0.02 au ( _ dotted lines _ ) and 0.6 au ( _ dot - dashed lines _ ) , corresponding to the semi - major axes of the known innermost planets orbiting solar - type and intermediate - mass evolved stars . two unpopulated regions of substellar companions orbiting intermediate - mass subgiants and giants seem to exist in region ( a ) and ( b ) . a possible host star - companion mass correlation considered from unpopulated region ( a ) and ( b ) supports the current view that more massive substellar companions tend to exist around more massive stars , that are derived from the results of planet searches around various mass stars ( @xcite ; @xcite ) . all of the brown dwarf - mass companions to intermediate - mass evolved stars were found around those with @xmath31 2.7 @xmath1 and there seems to be a paucity of such companions around those with 1.52.7 @xmath1 ( region ( a ) in figure [ fig1 ] ) . considering the smaller number of survey targets of @xmath31 2.7 @xmath1 ( e.g. , 35 % of the 300 oao targets ; @xcite ) compared with that of 1.52.7 @xmath1 , frequency of brown dwarf companions may become higher as stellar mass increases . this might favor gravitational instability in protostellar disks @xcite rather than fragmentation of proto - stellar clouds @xcite as the formation mechanism of brown dwarf - mass companions because stellar systems with larger difference in mass between primary and secondary stars are more difficult to form by the latter mechanism @xcite . also , there seems to be a possible paucity of lower - mass companions around [email protected] @xmath1 stars , in particular , a lack of planetary companions around 34 @xmath1 stars ( region ( b ) in figure [ fig1 ] ) . although it is basically difficult to detect planets around such `` noisy '' stars with large intrinsic radial velocity variability ( @xmath30 @xmath16 20 m s@xmath17 ) , planets with mass @xmath31 2.63.3 @xmath0 ( @xmath31 5.77.4 @xmath0 ) and @xmath24 @xmath19 0.6 au ( @xmath24 @xmath19 3.0 au ) should be above the current detection limit ( 3 @xmath30 @xmath16 60 m s@xmath17 ) . @xcite predicted that the frequency of giant planets has a peak near 3 @xmath1 stars based on a core accretion scenario taking account of the movement of snow line along the evolution of accretion and the central stars . moreover , if a formation mechanism works that invokes capturing of solid bodies migrating inward at the inner edge of the inactive magnetorotational instability - dead zone inside of the protoplanetary disk , gas giant planets could be formed efficiently at around 1 au around intermediate - mass stars before the planetary disks deplete @xcite . increasing the number of known massive planetary companions around massive intermediate - mass stars by further radial velocity surveys would be of great interest to understand the formation mechanisms of giant planets around intermediate - mass stars . of 5 m s@xmath17 for solar - mass stars ( 0.7 @xmath1 @xmath29 @xmath23 @xmath3 1.5 @xmath1 ) , 7 m s@xmath17 for intermediate - mass subgiants ( 1.5 @xmath1 @xmath29 @xmath23 @xmath29 1.9 @xmath1 ) and 20 m s@xmath17 for intermediate - mass giants ( 1.9 @xmath1 @xmath3 @xmath23 @xmath29 5 @xmath1 ) . _ dotted _ and _ dot - dashed lines _ indicate the detection limits for companions at 0.02 au and 0.6 au in solar mass stars and intermediate - mass evolved stars , respectively . two regions devoid of substellar companions are denoted by ( a ) and ( b ) . ] this research was supported as a korea - japan joint research project under the japan - korea basic scientific cooperation program between korea science and engineering foundation ( kosef ) and japan society for the promotion of science ( jsps ) . bate , m. r. 2000 , _ mnras _ , 314 , 33 butler , r. p. , _ et al . _ 2006 , _ apj _ , 646 , 505 butler , r. p. , marcy , g. w. , williams , e. , mccarthy , c. , dosanjh , p. , & vogt , s. s. 1996 , _ pasp _ , 108 , 500 hatzes , a. p. , guenther , e. w. , endl , m. , cochran , w. d. , dllinger , m. p. , & bedalov , a. 2005 , _ a&a _ , 437 , 743 hekker , s. , snellen , i. a. g. , aerts , c. , quirrenbach , a. , reffert , s. , & mitchell , d. s. 2008 , _ a&a _ , 480 , 215 ida , s. , & lin , d. n. c. 2004 , _ apj _ , 604 , 388 izumiura , h. 1999 , in proc . 4th east asian meeting on astronomy , ed . chen ( kunming : yunnan observatory ) , 77 izumiura , h. 2005 , journal of korean astronomical society , 38 , 81 johnson , j. a. , aller , k. m. , howard , a. w. , & crepp , j. r. 2010a , _ pasp _ , 122 , 905 johnson , j. a. , _ et al . _ 2010b , _ apjl _ , 721 , l153 kambe , e. , _ et al . _ 2002 , _ pasj _ , 54 , 865 kennedy , g. m. , & kenyon , s. j. 2008 , _ apj _ , 673 , 502 kim , k .- _ 2002 , journal of korean astronomical society , 35 , 221 kim , k .- _ 2007 , _ pasp _ , 119 , 1052 kretke , k. a. , lin , d. n. c. , garaud , p. , & turner , n. j. 2009 , _ apj _ , 690 , 407 liu , y .- j . , _ et al . _ 2008 , _ apj _ , 672 , 553 lovis , c. , & mayor , m. 2007 , _ a&a _ , 472 , 657 omiya , m. , _ et al . _ 2009 , _ pasj _ , 61 , 825 rice , w. k. m. , armitage , p. j. , bonnell , i. a. , bate , m. r. , jeffers , s. v. , & vine , s. g. 2003 , _ mnras _ , 346 , l36 sato , b. , kambe , e. , takeda , y. , izumiura , h. , & ando , h. 2002 , _ pasj _ , 54 , 873 sato , b. , kambe , e. , takeda , y. , izumiura , h. , masuda , s. , & ando , h. 2005 , _ pasj _ , 57 , 97 sato , b. , _ et al . _ 2008 , _ pasj _ , 60 , 539 takeda , y. , sato , b. , & murata , d. 2008 , _ pasj _ , 60 , 781 udry , s. & santos , n.c . 2007 , ara&a , 45 , 397	a korean - japanese planet search program has been carried out using the 1.8 m telescope at bohyunsan optical astronomy observatory ( boao ) in korea , and the 1.88 m telescope at okayama astrophysical observatory ( oao ) in japan to search for planets around intermediate - mass giant stars . the program aims to show the properties of planetary systems around such stars by precise doppler survey of about 190 g or k type giants together with collaborative surveys of the east - asian planet search network . so far , we detected two substellar companions around massive intermediate - mass giants in the korean - japanese planet search program . one is a brown dwarf - mass companion with 37.6 @xmath0 orbiting a giant hd 119445 with 3.9 @xmath1 , which is the most massive brown dwarf companion among those found around intermediate - mass giants . the other is a planetary companion with 1.8 @xmath0 orbiting a giant star with 2.4 @xmath1 , which is the lowest - mass planetary companion among those detected around giant stars with @xmath2 1.9 @xmath1 . plotting these systems on companion mass vs. stellar mass diagram , there seem to exist two unpopulated regions of substellar companions around giants with 1.53 @xmath1 and planetary companions orbiting giants with 2.44 @xmath1 . the existence of these possible unpopulated regions supports a current characteristic view that more massive substellar companions tend to exist around more massive stars . address = korea astronomy and space science institute , 61 - 1 hwaam - dong , yuseong - gu , daejeon 305 - 348 , south korea , email : [email protected] address = korea astronomy and space science institute , 61 - 1 hwaam - dong , yuseong - gu , daejeon 305 - 348 , south korea , email : [email protected] address = okayama astrophysical observatory , national astronomical observatory of japan , asakuchi , okayama 719 - 0232 , japan , altaddress = department of astronomical science , the graduate university for advanced studies , shonan village , hayama , kanagawa 240 - 0193 , japan address = korea astronomy and space science institute , 61 - 1 hwaam - dong , yuseong - gu , daejeon 305 - 348 , south korea , email : [email protected] address = tokyo institute of technology , 2 - 12 - 1-s6 - 6 ookayama , meguro - ku , tokyo 152 - 8550 , japan address = korea astronomy and space science institute , 61 - 1 hwaam - dong , yuseong - gu , daejeon 305 - 348 , south korea , email : [email protected] address = department of astronomy and atmospheric sciences , kyungpook national university , daegu 702 - 701 , south korea address = okayama astrophysical observatory , national astronomical observatory of japan , asakuchi , okayama 719 - 0232 , japan address = hiroshima astrophysical science center , hiroshima university , higashi - hiroshima , hiroshima 739 - 8526 , japan address = tokushima science museum , asutamu land tokushima , itano - gun , tokushima 779 - 0111 , japan address = kobe science museum , 7 - 7 - 6 minatojimanakamachi , chuo - ku , kobe , hyogo 650 - 0046 , japan address = bisei spaceguard center , japan spaceguard association , 1716 - 3 okura , bisei - cho , ibara , okayama 714 - 1411 , japan address = liberal arts education center , tokai university , 1117 kitakaname , hiratsuka , kanagawa 259 - 1292 , japan
the sphere packing problem asks how to arrange congruent balls as densely as possible without overlap between their interiors . the _ density _ is the fraction of space covered by the balls , and the problem is to find the maximal possible density . this problem plays an important role in geometry , number theory , and information theory . see @xcite for background and references on sphere packing and its applications . although many interesting constructions are known , provable optimality is very rare . aside from the trivial case of one dimension , the optimal density was previously known only in two @xcite , three @xcite , and eight @xcite dimensions , with the latter result being a recent breakthrough due to viazovska ( see @xcite for expositions ) . building on her work , we solve the sphere packing problem in twenty - four dimensions : [ thm : main ] the leech lattice achieves the optimal sphere packing density in @xmath0 , and it is the only periodic packing in @xmath0 with that density , up to scaling and isometries . in particular , the optimal sphere packing density in @xmath0 is that of the leech lattice , namely @xmath1 for an appealing construction of the leech lattice , see section 2.8 of @xcite . it is unknown in general whether optimal packings have any special structure , but our theorem shows that they do in @xmath0 . the optimality and uniqueness of the leech lattice were previously known only among lattice packings @xcite , which is a far more restrictive setting . recall that a lattice is a discrete subgroup of @xmath2 of rank @xmath3 , and a lattice packing uses spheres centered at the points of a lattice , while a periodic packing is the union of finitely many translates of a lattice . lattices are far more algebraically constrained , and it is widely believed that they do not achieve the optimal density in most dimensions ( for example , see @xcite for an example in @xmath4 of a periodic packing that is denser than any known lattice ) . by contrast , periodic packings at least come arbitrarily close to the optimal sphere packing density . the proof of theorem [ thm : main ] will be based on the linear programming bounds for sphere packing , as given by the following theorem . [ thm : lp ] let @xmath5 be a schwartz function and @xmath6 a positive real number such that @xmath7 , @xmath8 for @xmath9 , and @xmath10 for all @xmath11 . then the sphere packing density in @xmath2 is at most @xmath12 here @xmath13 means @xmath14 when @xmath3 is odd , and the fourier transform is normalized by @xmath15 where @xmath16 denotes the usual inner product on @xmath2 . without loss of generality , we can radially symmetrize @xmath17 , in which case @xmath18 is radial as well . we will often tacitly identify radial functions on @xmath0 with functions on @xmath19 and vice versa , by using @xmath20 with @xmath21 to denote the common value @xmath22 with @xmath23 . all fourier transforms will be in @xmath0 unless otherwise specified . in other words , if @xmath17 is a function of one variable defined on @xmath19 , then @xmath24 means @xmath25 where @xmath26 satisfies @xmath27 . optimizing the bound from theorem [ thm : main ] requires choosing the right auxiliary function @xmath17 . it was not previously known how to do so except in one dimension @xcite or eight @xcite , but cohn and elkies conjectured the existence of an auxiliary function proving the optimality of the leech lattice @xcite . we prove this conjecture by developing an analogue for the leech lattice of viazovska s construction for the @xmath28 root lattice . in the case of the leech lattice , proving optimality amounts to achieving @xmath29 , which requires that @xmath17 and @xmath18 have roots on the spheres of radius @xmath30 about the origin for @xmath31 . see @xcite for further explanation and discussion of this condition . furthermore , the argument in section 8 of @xcite shows that if @xmath17 has no other roots at distance @xmath32 or more , then the leech lattice is the unique optimal periodic packing in @xmath0 . thus , the proof of theorem [ thm : main ] reduces to constructing such a function . the existence of an optimal auxiliary function in @xmath0 has long been anticipated , and cohn and miller made further conjectures in @xcite about special values of the function , which we also prove . our approach is based on a new connection with quasimodular forms discovered by viazovska @xcite , and our proof techniques are analogous to hers . in sections [ sec : a ] and [ sec : b ] we will build two radial fourier eigenfunctions in @xmath0 , one with eigenvalue @xmath33 constructed using a weakly holomorphic quasimodular form of weight @xmath34 and depth @xmath32 for @xmath35 , and one with eigenvalue @xmath36 constructed using a weakly holomorphic modular form of weight @xmath37 for the congruence subgroup @xmath38 . we will then take a linear combination of these eigenfunctions in section [ sec : proof ] to construct the optimal auxiliary function . throughout the paper , we will make free use of the standard definitions and notation for modular forms from @xcite . we begin by constructing a radial eigenfunction of the fourier transform in @xmath0 with eigenvalue @xmath33 in terms of the quasimodular form @xmath40 where @xmath41 and the variable @xmath42 lies in the upper half plane . as mentioned in the introduction , we follow the notation of @xcite . in particular , @xmath43 denotes the eisenstein series @xmath44 which is a modular form of weight @xmath45 for @xmath35 when @xmath45 is even and greater than @xmath32 ( and a quasimodular form when @xmath46 ) . furthermore , we normalize @xmath47 by @xmath48 recall that @xmath47 vanishes nowhere in the upper half plane . this function @xmath49 is a weakly holomorphic quasimodular form of weight @xmath34 and depth @xmath32 for the full modular group . specifically , because @xmath50 we have the quasimodularity relation @xmath51 where @xmath52 and @xmath53 it follows from setting @xmath54 in that @xmath55 as @xmath56 , while the @xmath57-series for @xmath49 shows that @xmath58 as @xmath59 . we define @xmath60 for @xmath61 , which converges absolutely by these bounds . [ lemma : plusone ] the function @xmath62 analytically continues to a holomorphic function on a neighborhood of @xmath63 . its restriction to @xmath63 is a schwartz function and a radial eigenfunction of the fourier transform in @xmath0 with eigenvalue @xmath33 . we follow the approach of @xcite , adapted to use modular forms of different weight . substituting @xmath64 yields @xmath65 where we have shifted contours as in the proof of proposition 2 in @xcite . now the quasimodularity relation and periodicity modulo @xmath33 show that @xmath66 thus , @xmath67 which gives the analytic continuation of @xmath68 to a neighborhood of @xmath63 by and . essentially the same estimates as in proposition 1 of @xcite show that it is a schwartz function . specifically , the exponential decay of @xmath69 as the imaginary part of @xmath42 tends to infinity suffices to bound all the terms in , which shows that @xmath68 and all its derivatives decay exponentially . taking the @xmath70-dimensional radial fourier transform commutes with the integrals in and amounts to replacing @xmath71 with @xmath72 . therefore @xmath73 now setting @xmath74 shows that @xmath75 thus , and the fact that @xmath49 is periodic modulo @xmath33 show that @xmath76 , as desired . for @xmath61 , we have @xmath77 by . by the quasimodularity relation , @xmath78 thanks to the @xmath57-expansions with @xmath79 , we have @xmath80 as @xmath56 , where @xmath81 let @xmath82 then @xmath83 for @xmath61 . the integral @xmath84 is analytic on a neighborhood of @xmath19 , and hence holds for all @xmath6 . note in particular that @xmath68 maps @xmath63 to @xmath85 by ( or by via analytic continuation ) . equation implies that @xmath86 vanishes to second order whenever @xmath87 with @xmath88 , because @xmath89 has no poles at these points . furthermore , this formula implies that @xmath90 @xmath91 @xmath92 @xmath93 and @xmath94 the taylor series expansion is @xmath95 around @xmath96 . if we rescale @xmath68 so that its value at @xmath97 is @xmath33 , then the value at @xmath98 becomes @xmath99 and the derivative there becomes @xmath100 , and the derivative at @xmath32 becomes @xmath101 . the taylor series becomes @xmath102 however , the higher order terms in this taylor series do not appear to be rational , because they involve contributions from the integral in . we arrived at the definition of @xmath49 via the ansatz that @xmath103 should be a holomorphic quasimodular form of weight @xmath104 and depth @xmath32 for @xmath35 . the space of such forms is five - dimensional , spanned by @xmath105 , @xmath106 , @xmath107 , @xmath108 , and @xmath109 . within this space , one can solve for @xmath49 in several ways . we initially found it by matching the numerical conjectures from @xcite , but in retrospect one can instead impose constraints on its behavior at @xmath97 and @xmath110 , namely and . this information is enough to determine @xmath49 and hence the eigenfunction @xmath68 , up to a constant factor . next we construct a radial eigenfunction of the fourier transform in @xmath0 with eigenvalue @xmath36 . we will use the notation @xmath111 @xmath112 and @xmath113 for theta functions from @xcite . these functions satisfy the transformation laws @xmath114 where @xmath115 , @xmath116 , and @xmath117 for a function @xmath118 on the upper half plane and a matrix @xmath119 . let @xmath120 which is a weakly holomorphic modular form of weight @xmath37 for @xmath38 , and let @xmath121 and @xmath122 note that @xmath123 , which follows from the jacobi identity @xmath124 . using these @xmath57-expansions , we find that @xmath125 as @xmath56 , and @xmath126 as @xmath59 . let @xmath127 for @xmath61 , where the integral converges by the above bounds . [ lemma : minusone ] the function @xmath128 analytically continues to a holomorphic function on a neighborhood of @xmath63 . its restriction to @xmath63 is a schwartz function and a radial eigenfunction of the fourier transform in @xmath0 with eigenvalue @xmath36 . as in the proof proposition 6 from @xcite , we substitute @xmath129 and shift contours to show that for @xmath61 , @xmath130 here , we have used @xmath131 , and we have shifted the endpoints from @xmath132 to @xmath110 ( which is justified because the inequality @xmath61 ensures that the integrand decays exponentially ) . finally , applying @xmath133 yields @xmath134 which yields the analytic continuation to @xmath135 , and essentially the same estimates prove that it is a schwartz function . to show that the @xmath70-dimensional radial fourier transform @xmath136 satisfies @xmath137 , we follow the approach of proposition 5 from @xcite . as in the proof of lemma [ lemma : plusone ] , @xmath138 and the change of variables @xmath139 yields @xmath140 finally , @xmath137 follows from the equations @xmath141 where the first two equations amount to the definition of @xmath142 and the third follows from @xmath123 . for @xmath61 , we have @xmath143 from the @xmath57-expansion , we have @xmath144 as @xmath56 , and @xmath145 thus , for all @xmath146 , @xmath147 by analytic continuation . this formula implies that @xmath148 vanishes to second order whenever @xmath149 with @xmath88 . furthermore , it implies that @xmath150 @xmath151 and @xmath152 the taylor series expansion is @xmath153 around @xmath96 , and @xmath154 maps @xmath63 to @xmath85 . to obtain the definition of @xmath155 , we began with the ansatz that @xmath156 should be a holomorphic modular form of weight @xmath157 for @xmath38 . the space of such forms is eight - dimensional , spanned by @xmath158 with @xmath159 , and the subspace of forms satisfying the linear constraint @xmath160 is three - dimensional . as in the case of @xmath49 in section [ sec : a ] , one can solve for @xmath155 in several ways . in particular , within the subspace satisfying @xmath123 , the asymptotic behavior specified by and determines @xmath155 up to a constant factor . we can now construct the optimal auxiliary function for use in theorem [ thm : lp ] . let @xmath161 then @xmath162 , and the quadratic taylor coefficients of @xmath17 and @xmath18 are @xmath163 and @xmath164 , respectively , as conjectured in @xcite . the functions @xmath17 and @xmath18 have roots at all of the vector lengths in the leech lattice , i.e. , @xmath30 for @xmath31 . these roots are double roots except for the root of @xmath17 at @xmath32 , where @xmath165 ( in accordance with lemma 5.1 in @xcite ) . furthermore , @xmath17 has the value @xmath99 and derivative @xmath166 at @xmath98 , while @xmath18 has the value @xmath99 and derivative @xmath167 there . we must still check that @xmath17 satisfies the hypotheses of theorem [ thm : lp ] . we will do so using the approach of @xcite , with one extra complication at the end . for @xmath61 , equations and imply that @xmath168 where @xmath169 to show that @xmath170 for @xmath171 with equality only at @xmath6 of the form @xmath30 with @xmath172 , it suffices to show that @xmath173 . specifically , @xmath174 can not be identically zero since then @xmath17 would vanish as well ; given that @xmath174 is continuous , nonpositive everywhere , and negative somewhere , it follows that @xmath175 for all @xmath6 for which it converges ( i.e. , @xmath61 ) . because @xmath176 showing that @xmath173 amounts to showing that @xmath177 the formula @xmath178 immediately implies that @xmath179 , and so to prove it suffices to prove that @xmath180 . we prove this inequality in lemma [ lemma : phinonpos ] by bounding the truncation error in the @xmath57-series and examining the leading terms ( splitting into the cases @xmath181 and @xmath182 ) . it follows that @xmath170 for @xmath183 , as desired . for @xmath61 , the analogous formula for @xmath18 is @xmath184 where @xmath185 to show that @xmath186 for @xmath61 , it suffices to show that @xmath187 for all @xmath188 , i.e. , @xmath189 for the same reason as we saw above with @xmath190 . this inequality is lemma [ lemma : ineq2 ] . the formula in fact holds for @xmath191 , not just @xmath61 . to see why , we must examine the asymptotics of @xmath192 . there is no problem with the integral in as @xmath59 , because @xmath192 vanishes in this limit by and . however , the exponential growth of @xmath192 as @xmath56 causes divergence when @xmath6 is too small for @xmath193 to counteract this growth . to estimate the growth rate , note that by and , the @xmath194 terms cancel in the asymptotic expansion of @xmath192 as @xmath56 , which means that @xmath195 . thus , the formula for @xmath24 converges when @xmath191 , and it must equal @xmath24 by analytic continuation . note that it can not hold for the whole interval @xmath196 , because that would force @xmath18 to vanish at @xmath98 , which does not happen . thus , and the inequality @xmath197 in fact prove that @xmath198 for all @xmath199 . when @xmath200 , this inequality no longer implies that @xmath186 , which is a complication that does not occur in @xcite . instead , we must analyze @xmath192 more carefully . as @xmath56 , equations and show that @xmath201 we will ameliorate this behavior by subtracting these terms over the interval @xmath202 . they contribute @xmath203 which is nonnegative for @xmath200 , and the remaining terms @xmath204 converge for all @xmath205 . the integrand @xmath192 in the first integral is nonnegative , and thus to prove that @xmath186 for @xmath206 it suffices to prove that @xmath207 for @xmath208 , which is lemma [ lemma : bleadingterms ] . combining the results of this section shows that @xmath17 satisfies the hypotheses of theorem [ thm : lp ] , and thus that the leech lattice is an optimal sphere packing in @xmath0 . furthermore , @xmath17 has no roots @xmath61 other than @xmath209 with @xmath31 , and as in section 8 of @xcite this condition implies that the leech lattice is the unique densest periodic packing in @xmath0 . this completes the proof of theorem [ thm : main ] . the proof in section [ sec : proof ] requires checking certain inequalities for quasimodular forms on the imaginary axis . fortunately , these inequalities are not too delicate , because equality is never attained . the behavior at infinity is easily analyzed , which reduces the proof to verifying the inequalities on a compact interval , and that can be done by a finite calculation . thus , these inequalities are clearly provable if true . the proof of the analogous inequalities in @xcite used interval arithmetic , but in this appendix we take a different approach , based on applying sturm s theorem to truncated @xmath57-series . we have documented the calculations carefully , to facilitate checking the proof . computer code for verifying our calculations is contained in the file ` appendix.txt ` , which is available as an ancillary file at the arxiv.org e - print archive , where this paper is available as . our code is for the free computer algebra system pari / gp ( see @xcite ) , but the calculations are simple enough that they are not difficult to check in any computer algebra system . to prove each inequality , we approximate the modular form using @xmath57-series and prove error bounds for truncating the series , which we then incorporate by adding them to an appropriate term of the truncated series . the result is nearly a polynomial in @xmath57 , with the possible exceptions being factors of @xmath210 ( where @xmath54 ) , and we bound those factors so as to reduce to the case of a polynomial in @xmath57 . furthermore , we bound any factors of @xmath211 so that the coefficients become rational . finally , we use sturm s theorem with exact rational arithmetic to verify that the truncated series never changes sign . to prove the error bounds , we need to control the growth of the coefficients . we first multiply by @xmath212 to clear the denominators that appear in , , and . the advantage of doing so is that the coefficients of the numerator grow only polynomially . to estimate the growth rate , we bound the coefficient of @xmath213 in @xmath214 by @xmath215 in absolute value , in @xmath216 by @xmath217 , and in @xmath218 by @xmath219 . it is also not difficult to show that the coefficient of @xmath220 in @xmath221 , @xmath222 or @xmath223 is at most @xmath215 in absolute value . and @xmath223 have nonnegative coefficients , and their sum is the theta series of the @xmath224 root lattice in the variable @xmath225 , from which one can bound their coefficients . furthermore , @xmath226 . ] multiplying series is straightforward : if @xmath227 and @xmath228 , then the coefficients of @xmath229 are bounded by @xmath230 . when we add two @xmath57-series with coefficients bounded by different powers of @xmath231 , we typically produce an upper bound by rounding up the lower power for simplicity . using these techniques leads to explicit polynomial bounds for the coefficients of @xmath232 , @xmath233 , and @xmath234 by using their definitions in terms of eisenstein series and theta functions . these bounds are inefficient , but they suffice for our purposes . when @xmath208 , @xmath235 is small enough that these coefficient bounds yield a reasonable error term . when @xmath182 , we replace it with @xmath236 ( via @xmath237 ) and compute the corresponding @xmath57-expansion . first , we prove this inequality for @xmath208 , in which case @xmath240 . the bounds described above show that the coefficient of @xmath213 in @xmath232 is at most @xmath241 in absolute value , and exact computation shows that @xmath242 thus , the sum of the absolute values of the terms in @xmath232 for @xmath243 amounts to at most @xmath244 . let @xmath245 be the sum of the terms with @xmath246 . we use sturm s theorem to check that @xmath247 never changes sign on @xmath248 as a polynomial in @xmath57 , and we observe that it is negative in the limit as @xmath249 . this proves that @xmath250 for @xmath208 . using , the bound for @xmath182 is equivalent to showing that @xmath251 for @xmath208 . again we multiply by @xmath212 to control the coefficients . this case is more complicated , because there are factors of @xmath210 and @xmath211 . we replace factors of @xmath211 with rational bounds , namely @xmath252 or @xmath253 based on the sign of the term and whether it is a positive power of @xmath211 ( so that we obtain a lower bound ) , and we similarly use the bounds @xmath254 ; the latter bound follows from @xmath235 and @xmath255 . to estimate the error bound from truncation , we use @xmath256 ; the result is that the error from omitting the @xmath213 terms with @xmath243 is at most @xmath244 . these observations reduce the problem to showing that a polynomial in @xmath225 with rational coefficients is positive over the interval @xmath257 . using sturm s theorem , we check that it holds over the larger interval @xmath258 . note that we could have avoided fractional powers of @xmath57 in this proof if we had used a different upper bound for @xmath210 , but fractional powers will be needed to handle @xmath142 and @xmath155 in any case . we will use the bounds such as @xmath259 from the preceding proof systematically in the remaining proofs . we use exactly the same technique as in the proof of lemma [ lemma : phinonpos ] . for @xmath208 , removing the @xmath261 and higher terms in the @xmath57-series for @xmath262 introduces an error of at most @xmath244 , and sturm s theorem shows that the resulting polynomial has no sign changes . note that @xmath142 involves powers of @xmath225 , and so we must view the truncated series as a polynomial in @xmath225 rather than @xmath57 . for @xmath182 , we apply relations and to reduce the problem to showing that @xmath263 for @xmath208 . when we multiply by @xmath212 and remove the @xmath261 and higher terms , the error bound is at most @xmath244 , and sturm s theorem completes the proof . as in the previous proof , this case involves handling factors of @xmath210 and @xmath211 , but they present no difficulties . as usual , we multiply @xmath265 by @xmath212 and compute its @xmath57-series . our usual truncation bounds show that removing the @xmath261 and higher terms introduces an error bound of at most @xmath244 , and sturm s theorem again completes the proof . t. hales , m. adams , g. bauer , d. t. dang , j. harrison , t. l. hoang , c. kaliszyk , v. magron , s. mclaughlin , t. t. nguyen , t. q. nguyen , t. nipkow , s. obua , j. pleso , j. rute , a. solovyev , a. h. t. ta , t. n. tran , d. t. trieu , j. urban , k. k. vu , and r. zumkeller , _ a formal proof of the kepler conjecture _ , to appear in forum of mathematics , pi .	building on viazovska s recent solution of the sphere packing problem in eight dimensions , we prove that the leech lattice is the densest packing of congruent spheres in twenty - four dimensions , and that it is the unique optimal periodic packing . in particular , we find an optimal auxiliary function for the linear programming bounds , which is an analogue of viazovska s function for the eight - dimensional case .
it is generally challenging to understand the fundamental group of a locally complicated space @xmath4 a common tactic is to consider the image @xmath5 as a subgroup of the first shape homotopy group @xmath6 via the natural homomorphism @xmath7 arising from the ech expansion . in particular , if @xmath8 is injective , @xmath2 is said to be @xmath9-shape injective and one gains a characterization of the elements of @xmath10 as sequences in an inverse limit of fundamental groups of polyhedra . thus @xmath11 can be thought of as the data of the fundamental group forgotten when passing to the first shape group . in this paper , we develop a new lens in which to study @xmath8 and identify new characterizations of @xmath11 in terms of familiar subgroups of the fundamental group . the spanier group @xmath12 of a space @xmath2 , introduced in frvz11 , is a subgroup of @xmath10 useful for identifying loops deformable into arbitrarily small neighborhoods @xcite . moreover the existence of generalized covering maps @xcite is intimately related to both @xmath13 and @xmath11 . it is apparently an open question to understand exactly when @xmath14 . on the one hand , for a general space @xmath15 the inclusion @xmath16 holds @xcite . on the other hand , examples show inclusion can be strict if @xmath2 fails to be non - locally path connected @xcite . our main result , theorem [ leftexactsequence ] , captures a decent class of spaces such that @xmath14 . _ if _ @xmath2 _ is a locally path connected paracompact hausdorff space , then _ @xmath11 _ is precisely the spanier group _ @xmath17 . in particular if @xmath2 is a peano continuum then @xmath18 a variety of 2-dimensional examples help motivate this paper s content as follows . the occurrence of @xmath9-shape injectivity ( i.e. the vanishing of @xmath19 ) has been studied in a number of cases : for example if @xmath2 has lebesgue covering dimension @xmath20 @xcite , if @xmath2 is a subspace of a closed surface ( or a planar set ) @xcite , or if @xmath2 is a fractal - like tree of manifolds @xcite , then @xmath21 . the equality @xmath22 now indicates @xmath2 is @xmath9-shape injective iff @xmath23 . at the other extreme , the equality @xmath24 can happen if @xmath2 is a 2-dimensional peano continuum ( e.g. if @xmath2 is the join of two cones over the hawaiian earring ) . in particular , this equality will occur if every loop in @xmath2 is a `` small loop '' @xcite or , more generally , if @xmath2 is a spanier space @xcite . for such spaces , understanding @xmath10 is equivalent to understanding @xmath25 finally , there is a rich middle groundwhere @xmath11 lies strictly between @xmath26 and @xmath27 two important such peano continua appear in @xcite , one of which is homotopically path hausdorffand one of which is strongly homotopically path hausdorff . in both cases , the verification of the aforementioned properties and the attempts to manufacture generalized notions of covering spaces , rely heavily on an understanding of @xmath17 . to investigate when @xmath28 , we first modify the familiar definition . the spanier group with respect to an open cover @xmath29 ( denoted @xmath30 ) was introduced in spanier s celebrated textbook @xcite . recall @xmath12 is defined as the intersection of the groups @xmath30 over all open covers @xmath29 . while the usual groups @xmath30 are useful for studying covering space theory and its generalizations , our modified version , the so - called _ thick spanier group of _ @xmath2 _ with respect to _ @xmath29 ( denoted @xmath31 ) is useful for studying the shape homomorphism @xmath32 . in particular , the main technical achievement of this paper , theorem [ shortexactsequence ] , states : _ if _ @xmath29 _ is an open cover consisting of path connected sets and _ @xmath33 _ is a canonical map to the nerve of _ @xmath29 _ _ , then there is a short exact sequence _ _ @xmath34 & { \pi^{sp}(\mathscr{u},x_0)}\ar[r ] & \pi_1(x , x_0 ) \ar[r]^-{p_{{\mathscr{u}}\ast } } & \pi_{1}(\|{n(\mathscr{u})}\|,u_0 ) \ar[r ] & 1}.\ ] ] the utility of this short exact sequence is the identification of a convenient set of generators of @xmath35 . the same sequence with the ordinary spanier group @xmath36 in place of @xmath37 fails to be exact even in simple cases : see example [ circle ] . this paper is structured as follows : in section 2 , we include necessary preliminaries on simplicial complexes and review the constructions of the ech expansion of a space @xmath2 in terms of canonical maps @xmath33 ( where @xmath38 is the nerve of an open cover @xmath29 of @xmath2 ) , the first shape group , and the homomorphism @xmath8 . in section 3 , we define and study thick spanier groups and their relationship to ordinary spanier groups . thick spanier groups are constructed by specifying generators represented by loops lying in pairs of intersecting elements of @xmath29 ; section 4 is then devoted to characterizing generic elements of @xmath39 in terms of a homotopy - like equivalence relation on loops which depends on the given open cover @xmath29 . section 5 is devoted to a proof of theorem [ shortexactsequence ] guaranteeing the exactness of the above sequence . sections 6 and 7 include applications of the above level short exact sequences : in section 6 , we obtain an exact sequence : @xmath34 & { { \pi^{sp}}(x , x_0)}\ar[r ] & \pi_1(x , x_0 ) \ar[r]^-{\psi_{x } } & { \check{\pi}_{1}(x , x_0)}}\ ] ] and thus identify @xmath3 and @xmath40 for a large class of spaces ( see theorem [ leftexactsequence ] ) . finally , in section 7 , we find application in the theory of a topologically enriched version of the fundamental group . by construction , the first shape group @xmath41 is the inverse limit of discrete groups ; the fundamental group @xmath42 inherits the structure of a topological group when it is given the pullback topology with respect to @xmath43 ( the so - called _ shape topology _ ) . we apply the above results to identify a convenient basis for the topology of @xmath42 and show this topology consists precisely of the data of the covering space theory of @xmath2 . throughout this paper , @xmath2 is assumed to be a path connected topological space with basepoint @xmath44 . we call upon the theory of simplicial complexes in standard texts such as @xcite and @xcite . much of the notation used is in line with these sources . if @xmath45 is an abstract or geometric simplicial complex and @xmath46 is an integer , @xmath47 denotes the n - skeleton of @xmath45 and if @xmath48 is a vertex of @xmath45 , @xmath49 and @xmath50 denote the open and closed star of the vertex @xmath48 respectively . when @xmath45 is abstract , @xmath51 denotes the geometric realization . if vertices @xmath52 span an n - simplex of @xmath51 , then @xmath53 $ ] denotes the n - simplex with the indicated orientation . finally , for each integer @xmath46 , @xmath54 denotes the n - th barycentric subdivision of @xmath45 . we frequently make use of the standard abstract 2-simplex @xmath55 which consists of a single 2-simplex and its faces and whose geometric realization is @xmath56 . the boundary @xmath57 is the 1-skeleton @xmath58 and its realization @xmath59 is homeomorphic to the unit circle . if @xmath45 is a geometric subcomplex of a subdivision of @xmath60 containing the origin , then the origin is assumed to be the basepoint . we use the following conventions for paths and loops in simplexes and general spaces . a _ path _ in a space @xmath2 is a map @xmath61\rightarrow x $ ] from the unit interval . the _ reverse path _ of @xmath62 is the path given by @xmath63 and the constant path at a point @xmath64 will be denoted @xmath65 . if @xmath66\rightarrow x$ ] are paths in @xmath67 such that @xmath68 , the concatenation of this sequence is the unique path @xmath69 , sometimes denoted @xmath70 , whose restriction to @xmath71 $ ] is @xmath72 . a path @xmath61\to x$ ] is a _ loop _ if @xmath73 . quite often it will be convenient to view a loop as a map @xmath74 where @xmath59 is identified with @xmath75/\{0,1\}\cong s^1\subseteq \mathbb{r}^{2}$ ] by an orientation preserving homeomorphism in @xmath76 . a loop @xmath77 is _ inessential _ if it extends to a map @xmath78 and is _ essential _ if no such extension exists . two loops @xmath79 are _ freely homotopic _ if there is a homotopy @xmath80\to x$ ] such that @xmath81 and @xmath82 . if @xmath83 are loops based at @xmath84 , then they are _ homotopic rel . basepoint _ if there is a homotopy @xmath85 as above such that @xmath86)=x_0 $ ] . we use special notation for edge paths in a geometric simplicial complex @xmath45 : if @xmath87 are vertices in @xmath45 , we identify the oriented 1-simplex @xmath88 $ ] with the linear path from vertex @xmath89 to @xmath90 on @xmath88 $ ] . similarly , @xmath91 $ ] denotes the constant path at a vertex @xmath48 . an _ edge path _ is a concatenation @xmath92\ast [ v_1,v_2]\ast \dots \ast [ v_{n-1},v_n]$ ] of linear or constant paths in the 1-skeleton of @xmath45 . if @xmath93 , then @xmath94 is an _ edge loop_. it is a well - known fact that all homotopy classes ( rel . endpoints ) of paths from @xmath95 to @xmath89 are represented by edge paths . we now recall the construction of the first shape homotopy group @xmath41 via the ech expansion . for more details , see @xcite . let @xmath96 be the set of open covers of @xmath2 direct by refinement . similarly , let @xmath97 be the set of open covers with a distinguished element containing the basepoint , i.e. the set of pairs @xmath98 where @xmath99 , @xmath100 and @xmath101 . we say @xmath102 refines @xmath98 if @xmath103 refines @xmath29 as a cover and @xmath104 . the nerve of a cover @xmath105 is the abstract simplicial complex @xmath106 whose vertex set is @xmath107 and vertices @xmath108 span an n - simplex if @xmath109 . the vertex @xmath110 is taken to be the basepoint of the geometric realization @xmath38 . whenever @xmath102 refines @xmath98 , construct a simplicial map @xmath111 , called a _ projection _ , given by sending a vertex @xmath112 to a vertex @xmath113 such that @xmath114 . in particular , @xmath115 must be sent to @xmath110 . any such assignment of vertices extends linearly to a simplicial map . moreover , the induced map @xmath116 is unique up to based homotopy . thus the homomorphism @xmath117 induced on fundamental groups is independent of the choice of simplicial map . recall that an open cover @xmath118 of @xmath2 is normal if it admits a partition of unity subordinated to @xmath118 . let @xmath119 be the subset of @xmath97 ( also directed by refinement ) consisting of pairs @xmath98 where @xmath118 is a normal open cover of @xmath2 and such that there is a partition of unity @xmath120 subordinated to @xmath118 with @xmath121 . it is well - known that every open cover of a paracompact hausdorff space @xmath2 is normal . moreover , if @xmath122 , it is easy to refine @xmath123 to a cover @xmath124 such that @xmath125 is the only element of @xmath126 containing @xmath44 and therefore @xmath127 . thus , for paracompact hausdorff @xmath2 , @xmath119 is cofinal in @xmath97 . the _ first shape homotopy group _ is the inverse limit @xmath128 given an open cover @xmath129 , a map @xmath130 is a _ ( based ) canonical map _ if @xmath131 for each @xmath113 and @xmath132 . such a canonical map is guaranteed to exist if @xmath133 : find a locally finite partition of unity @xmath120 subordinated to @xmath118 such that @xmath121 . when @xmath113 and @xmath134 , determine @xmath135 by requiring its barycentric coordinate belonging to the vertex @xmath136 of @xmath38 to be @xmath137 . according to this construction , the requirement @xmath121 gives @xmath132 . a canonical map @xmath138 is unique up to based homotopy and whenever @xmath102 refines @xmath98 ; the compositions @xmath139 and @xmath138 are homotopic as based maps . therefore the homomorphisms @xmath140 satisfy @xmath141 . these homomorphisms induce a canonical homomorphism @xmath142)=\left([p_{\mathscr{u}}\circ \alpha]\right)\ ] ] to the limit . it is of interest to characterize @xmath143 since , when @xmath144 , @xmath41 retains all the data in the fundamental group of @xmath2 . a space for which @xmath144 is said to be @xmath145-_shape injective_. we begin by recalling the construction of spanier groups frvz11,spanier66 . let @xmath146 denote the set of homotopy classes ( rel . endpoints ) of paths starting at @xmath44 , i.e. the star of the fundamental groupoid of @xmath2 at @xmath44 . as in @xcite , multiplication of homotopy classes of paths is taken in the fundamental groupoid of @xmath2 so that @xmath147[\beta]=[\alpha\ast\beta]$ ] when @xmath148 . [ spaniergroupdef ] _ let @xmath29 be an open cover of @xmath2 . the _ spanier group of _ @xmath2 _ with respect to _ @xmath29 is the subgroup of @xmath42 , denoted @xmath36 , generated by elements of the form @xmath147[\gamma][\overline{\alpha}]$ ] where @xmath147\in \widetilde{x}$ ] and @xmath149\to u$ ] is a loop based at @xmath150 for some @xmath151 . _ note , in the above definition , @xmath147 $ ] is left to vary among all homotopy ( rel . endpoint ) classes of paths starting at @xmath44 and ending at @xmath152 . thus @xmath36 is a normal subgroup of @xmath42 . also observe that if @xmath103 is an open cover of @xmath2 which refines @xmath29 , then @xmath153 . _ the _ spanier group _ of @xmath2 is the intersection @xmath154 of all spanier groups with respect to open covers of @xmath2 . equivalently , @xmath13 is the inverse limit @xmath155 of the inverse system of inclusions @xmath156 induced by refinement in @xmath96 . _ note the spanier group of @xmath2 is a normal subgroup of @xmath42 . _ in the previous two definitions , we are actually using the _ unbased _ spanier group as defined by the authors of @xcite . these authors also define the _ based _ spanier group by replacing covers with covers by pointed sets ; the two definitions agree when @xmath2 is locally path connected . the unbased version is sufficient for the purposes of this paper since our main results apply to locally path connected spaces . _ to study @xmath157 , we require similar but potentially larger versions of the spanier groups constructed above . _ let @xmath29 be an open cover of @xmath2 . the _ thick spanier group of _ @xmath2 _ with respect to _ @xmath29 is the subgroup of @xmath42 , denoted @xmath37 , generated by elements of the form @xmath147[\gamma_1][\gamma_2][\overline{\alpha}]$ ] where @xmath147\in \widetilde{x}$ ] and @xmath158\to u_1 $ ] and @xmath159\to u_2 $ ] are paths for some @xmath160 . _ with respect to @xmath29.,height=144 ] [ tspanier1 ] for every open cover @xmath29 of @xmath2 , @xmath37 is a normal subgroup of @xmath42 such that @xmath161 . if @xmath103 is an open cover of @xmath2 which refines @xmath29 , then @xmath162 . the subgroup @xmath163 is normal in @xmath164 for the same reason @xmath165 is normal . we have @xmath166 since the generators @xmath147[\gamma_1][\gamma_2][{\overline{\alpha}}]$ ] of @xmath163 where @xmath167 have image in @xmath168 are precisely the generators of @xmath165 . the second statement follows as it does for ordinary spanier groups . [ circle ] _ it is not necessarily true that @xmath169 even in the simplest cases . suppose @xmath170 is an open cover of the unit circle @xmath171 consisting of two connected intervals @xmath172 such that @xmath173 is the disjoint union of two connected intervals . since both @xmath172 are simply connected , we have @xmath174 . on the other hand , @xmath37 contains a generator of @xmath175 and thus @xmath176 . _ [ naturalityofspanier ] _ spanier groups and thick spanier groups with respect to covers are natural in the following sense : if @xmath177 is a map such that @xmath178 and @xmath179 is open cover of @xmath180 , then @xmath181 is an open cover of @xmath2 . observe if @xmath182 is the homomorphism induced on fundamental groups , then @xmath183 and @xmath184 . _ _ the _ thick spanier group _ of @xmath2 is the intersection @xmath185 of all thick spanier groups with respect to open covers of @xmath2 . equivalently @xmath186 is the inverse limit @xmath187 of the inverse system of inclusions @xmath188 induced by refinement in @xmath96 . _ the following proposition follows directly from previous observations . [ spanierinclusion ] [ tspanier2 ] for every space @xmath2 , @xmath189 . if every open cover @xmath29 of @xmath2 admits a refinement @xmath103 such that @xmath190 , then @xmath191 . we apply proposition [ tspanier2 ] to find two conditions ( those in proposition [ pcintersections ] and theorem [ paracompact1 ] ) guaranteeing the equality of the spanier group and thick spanier group . [ pcintersections ] if @xmath103 is an open cover of @xmath2 such that @xmath192 is path connected ( or empty ) for every pair @xmath193 , then @xmath194 . consequently , if every open cover @xmath29 of @xmath2 has an open refinement @xmath103 with this property , then @xmath191 . suppose @xmath126 is as in the statement of the proposition . it suffices to show @xmath195 . let @xmath196[\gamma_1][\gamma_2][{\overline{\alpha}}]$ ] be a generator of @xmath197 where @xmath147\in { \widetilde{x}}$ ] , and @xmath198\to v_i$ ] for @xmath199 . since , by assumption , @xmath200 is path connected and the points @xmath201 and @xmath202 both lie in @xmath200 , there is a path @xmath203\to v_{1}\cap v_{2}$ ] from @xmath204 to @xmath205 . note @xmath206[\gamma_1\ast{\overline{\beta}}][{\overline{\alpha}}]$ ] and @xmath207[\beta\ast\gamma_2][{\overline{\alpha}}]$ ] are generators of @xmath208 . since @xmath209 , it follows that @xmath210 . the second statement in the current proposition now follows from proposition [ tspanier2 ] . [ starcover ] _ let @xmath45 be a geometric simplicial complex and @xmath211 be a vertex . it is well - known that the open star @xmath49 is contractible ( and thus path connected ) . therefore @xmath212 is an open cover of @xmath45 such that @xmath213 . + moreover , if @xmath214 , then the open 1-simplex @xmath215 and hence there is a canonical strong deformation from @xmath216 onto @xmath217 . for a given point @xmath218 fix the coefficients of @xmath219 and @xmath220 while linearly contracting to @xmath221 the remaining barycentric coordinates of @xmath62 . hence , since @xmath217 is contractible , @xmath216 is contractible and , in particular , path connected . thus it follows from proposition [ pcintersections ] that @xmath222 . _ it is known that @xmath223 for any space @xmath2 : see proposition 4.8 of @xcite . thus the spanier group of @xmath2 vanishes whenever @xmath2 is @xmath145-shape injective . we use proposition [ pcintersections ] to prove the analogous inclusion for the thick spanier group ; note the proof is similar to that in @xcite . [ inclusion1 ] for any space @xmath2 , @xmath224 . let @xmath225 and suppose @xmath226 . it suffices to show @xmath227 . recall that for each @xmath228 , we have @xmath229 . therefore @xmath230 is an open cover of @xmath2 which refines @xmath118 . since @xmath231 by assumption , @xmath232 is a product of generators of the form @xmath147[\gamma_1][\gamma_2][{\overline{\alpha}}]$ ] where @xmath147\in { \widetilde{x}}$ ] and @xmath198\to p_{{\mathscr{u}}}^{-1}(st(u_i,{n(\mathscr{u})}))$ ] for @xmath233 , @xmath234 . note @xmath235 is a path in @xmath236 based at @xmath110 with @xmath237 and @xmath238\to st(u_i,{n(\mathscr{u})})$ ] for @xmath234 . thus if @xmath239 is the cover of @xmath236 by open stars , then @xmath240[\gamma_1][\gamma_2][{\overline{\alpha}}])=[p_{{\mathscr{u}}}\circ\alpha ] [ p_{{\mathscr{u}}}\circ\gamma_1][p_{{\mathscr{u}}}\circ\gamma_2][\overline{p_{{\mathscr{u}}}\circ\alpha}]\]]is a generator of @xmath241 . but it is observed in example [ starcover ] that @xmath242 . since generators multiplying to @xmath232 lie in the subgroup @xmath243 of @xmath164 , so does @xmath232 . in the non - locally path connected case , the inclusion @xmath224 need not be equality ; a counterexample is given later in example [ example1 ] . the following theorem calls upon the theory of paracompact spaces . recall if @xmath29 is an open cover of @xmath2 and @xmath64 , the _ star of _ @xmath244 _ with respect to _ @xmath29 is the union @xmath245 . barycentric refinement _ of @xmath29 is an open refinement @xmath103 of @xmath29 such that for each @xmath64 , there is a @xmath113 with @xmath246 . it is known that a @xmath247 space is paracompact if and only if every open cover has an open barycentric refinement @xcite . [ paracompact1 ] if @xmath2 is @xmath247 and paracompact , then @xmath191 . we apply the second statement of proposition [ tspanier2 ] . given an open cover @xmath118 of @xmath2 take @xmath126 to be a barycentric refinement of @xmath118 . suppose @xmath196[\gamma_1][\gamma_2][{\overline{\alpha}}]$ ] is a generator of @xmath197 where @xmath147\in { \widetilde{x}}$ ] and @xmath198\to v_i$ ] for @xmath199 , @xmath234 . since @xmath248 and @xmath126 is chosen to be a barycentric refinement , we have @xmath249 for some @xmath228 . note @xmath250 is a loop in @xmath136 based at @xmath150 and @xmath196[\gamma_{1}\ast\gamma_{2}][{\overline{\alpha}}]$ ] . thus @xmath251 and the inclusion @xmath252 follows . if @xmath2 is metrizable , then @xmath191 . while the definition of the thick spanier group @xmath37 in terms of generators is often convenient , it is desirable to characterize generic elements . we do this by replacing homotopies as continuous deformations of paths with finite step homotopies through open covers . this approach is closely related to that of dugundji in @xcite and differs from the notions of @xmath29-homotopy in kennison89 and @xcite . fix an open cover @xmath29 of @xmath2 and define a relation @xmath253 on paths in @xmath2 . if @xmath254\to x$ ] are paths in @xmath2 , let @xmath255 when @xmath256 , @xmath257 , and there are partitions @xmath258 and @xmath259 and a sequence @xmath260 of open neighborhoods such that @xmath261)\subseteq u_i$ ] and @xmath262)\subseteq u_i$ ] for @xmath263 . note if @xmath103 is an open refinement of @xmath29 , then @xmath264 @xmath265 @xmath255 . _ two paths @xmath266 and @xmath267 in @xmath2 are said to be @xmath29__-homotopic _ _ if @xmath256 , @xmath257 , and there is a finite sequence @xmath268 such that @xmath269 for each @xmath270 . the sequence @xmath271 is called a @xmath29__-homotopy from _ _ @xmath266 _ to _ @xmath267 . if a loop @xmath266 is @xmath29-homotopic to the constant path , we say it is _ _ null-__@xmath29_-homotopic_. _ note @xmath29-homotopy defines an equivalence relation @xmath272 on the set of paths in @xmath2 . [ homotopicimpliesuhomotopic ] if two paths are homotopic ( rel . endpoints ) , then they are @xmath29-homotopic . suppose @xmath273 are paths in @xmath2 with @xmath256 and @xmath257 and @xmath274\times [ 0,1]\to x$ ] is a homotopy such that @xmath275 , @xmath276 , @xmath277 , and @xmath278 for all @xmath279 $ ] . find an integer @xmath280 such that for each square @xmath281\times \left[\frac{j-1}{n},\frac{j}{n}\right]$ ] , @xmath282 , @xmath283 for some @xmath284 . let @xmath285 be the path which is the restriction of @xmath85 to @xmath75\times \{j / n\}$ ] . we have @xmath286 , @xmath287 , and @xmath288 for each @xmath270 and thus @xmath289 . observe that @xmath29-homotopy respects inversion and concatenation of paths in the sense that if @xmath290 and @xmath291 then @xmath292 and @xmath293 when the concatenations are defined . since inessential loops are null-@xmath29-homotopic by lemma homotopicimpliesuhomotopic , the set @xmath294 of null-@xmath29-homotopic loops is a subgroup of @xmath42 . in fact , @xmath294 is normal in @xmath42 since , if @xmath266 is null-@xmath29-homotopic and @xmath295\in \pi_{1}(x , x_0)$ ] , then @xmath296 we now compare @xmath37 and @xmath294 as subgroups of @xmath42 . [ uhomotopy ] suppose @xmath29 is an open cover of @xmath2 consisting of path connected sets . if @xmath266 and @xmath267 are @xmath29-homotopic paths with @xmath297 , then @xmath298\in \pi^{sp}(\mathscr{u},x_0)$ ] . first , consider the case when @xmath299 . take open neighborhoods @xmath300 and partitions @xmath258 and @xmath259 such that @xmath261)\subseteq u_i$ ] and @xmath262)\subseteq u_i$ ] . since @xmath301 is path connected , there are paths @xmath302\to u_i$ ] from @xmath303 to @xmath304 for @xmath305 . additionally , there are paths @xmath306\to u_{i+1}$ ] from @xmath303 to @xmath304 for @xmath305 . let @xmath307 be the constant path at @xmath44 and @xmath308 be the constant path at @xmath257 . observe @xmath309 is a loop in @xmath310 based at @xmath303 for @xmath305 . let @xmath311 and @xmath312 be the paths given by restricting @xmath266 and @xmath267 to @xmath313 $ ] and @xmath314 $ ] respectively . note the following paths @xmath315 are generators of @xmath165 : 1 . @xmath316[\alpha_1\ast\epsilon_1\ast\overline{\beta_{1}}][\overline{c_{x_0}}]$ ] 2 . @xmath317 [ \overline{\delta_{i-1}}\ast\alpha_{i}\ast \epsilon_{i}\ast \overline{\beta_{i } } ] [ \overline{\beta_1\ast \beta_2\ast\dots\ast\beta_{i-1}}]$ ] for @xmath318 3 . @xmath319[\overline{\delta_{n-1}}\ast \alpha_n \ast \overline{\beta_{n}}][\overline{\beta_1\ast \dots\ast\beta_{n-1}}]$ ] since @xmath166 , we have @xmath320 for @xmath263 . additionally , for @xmath305 , @xmath321[\overline{\epsilon_{i}}][\delta_i][\overline{\beta_1\ast \dots \ast \beta_i}]\]]is a generator of @xmath163 . note @xmath322=[\alpha_1][\alpha_2]\dots [ \alpha_n][\overline{\beta_{n}}]\dots [ \overline{\beta_{2}}][\overline{\beta_{1}}]= g_1h_1g_2h_2\dots g_{n-1}h_{n-1}g_n\]]and thus @xmath298\in { \pi^{sp}(\mathscr{u},x_0)}$ ] . for the general case , suppose there is a @xmath118-homotopy @xmath268 where @xmath288 for each @xmath270 . the first case gives @xmath323\in { \pi^{sp}(\mathscr{u},x_0)}$ ] for each @xmath263 and thus @xmath322=[\gamma_{0}\ast\overline{\gamma_{1 } } ] [ \gamma_{1}\ast\overline{\gamma_{2}}]\dots [ \gamma_{n-1}\ast\overline{\gamma_{n}}]\in { \pi^{sp}(\mathscr{u},x_0)}.\ ] ] [ characterizethickspangrp ] for every open cover @xmath29 , @xmath324 . moreover , if @xmath29 consists of path connected sets , then @xmath325 . since @xmath326 is a subgroup of @xmath164 , the inclusion @xmath327 follows from showing @xmath326 contains the generators of @xmath163 . suppose @xmath295=[\alpha][\gamma_1][\gamma_2][{\overline{\alpha}}]$ ] is such a generator where @xmath328 has image in @xmath233 . we claim if @xmath329 , then @xmath330 . find a sequence of neighborhoods @xmath331 and a partition @xmath332 such that @xmath333)\subseteq v_j$ ] . since @xmath334 we may assume @xmath335)\subseteq u_1\cap u_2 $ ] . for @xmath336 , let @xmath337\to [ 0,c]$ ] be the order - preserving , linear homeomorphism . define a partition @xmath338 by @xmath339 if @xmath340 , @xmath341 , and @xmath342 for @xmath343 . rewrite the sequence @xmath344 in @xmath118 as @xmath345 . note @xmath346)\subseteq w_k$ ] for @xmath347 . define a partition @xmath348 by @xmath349 if @xmath340 , @xmath350 , and @xmath351 for @xmath343 . note @xmath352)\subseteq w_k$ ] for @xmath347 and thus @xmath330 . since @xmath353 is an inessential loop , @xmath267 is null-@xmath118-homotopic by lemma [ homotopicimpliesuhomotopic ] . in the case that @xmath118 consists of path connected sets , note that if @xmath266 is a null-@xmath118-homotopic loop based at @xmath44 , then @xmath147=[\alpha\ast \overline{c_{x_0}}]\in { \pi^{sp}(\mathscr{u},x_0)}$ ] by lemma [ uhomotopy ] . this gives the inclusion @xmath354 . the following theorem , which identifies the kernel of the homomorphism @xmath140 for certain covers @xmath29 , is the main technical achievement of this paper . [ shortexactsequence ] suppose @xmath355 is such that @xmath29 consists of path connected sets and @xmath130 is a canonical map . the sequence @xmath34 & { \pi^{sp}(\mathscr{u},x_0)}\ar[r]^-{i_{{\mathscr{u } } } } & { \pi_{1}(x , x_0)}\ar[r]^-{p_{{\mathscr{u}}\ast } } & \pi_{1}(\|{n(\mathscr{u})}\|,u_0 ) \ar[r ] & 1}\ ] ] where @xmath356 is inclusion is exact . the above theorem is proved in the following three subsections in which the hypotheses of the statement are assumed ; exactness follows from theorems [ exactness1 ] , [ exactness2 ] and [ exactness3 ] . note if @xmath357 is a normal refinement of @xmath98 , where @xmath103 also consists of path connected sets and @xmath358 is a canonical map , then there is a morphism @xmath34 & { \pi^{sp}(\mathscr{v},x_0)}\ar[d]^{i_{{\mathscr{u}}{\mathscr{v } } } } \ar[r]^{i_{{\mathscr{v } } } } \ar[r ] & { \pi_{1}(x , x_0)}\ar[d]^{id } \ar[r]^-{p_{{\mathscr{v}}\ast } } & \pi_{1}(\|{n(\mathscr{v})}\|,v_0 ) \ar[d]^{p_{{\mathscr{u}}{\mathscr{v}}\ast } } \ar[r ] & 1\\ 1 \ar[r ] & { \pi^{sp}(\mathscr{u},x_0)}\ar[r]_{i_{{\mathscr{u } } } } & { \pi_{1}(x , x_0)}\ar[r]_-{p_{{\mathscr{u}}\ast } } & \pi_{1}(\|{n(\mathscr{u})}\|,u_0 ) \ar[r ] & 1}\ ] ] of short exact sequences of groups where the left square consists of three inclusion maps and ( at least ) one identity . [ sesofinversesystem ] suppose @xmath2 is paracompact hausdorff and @xmath359 is a directed subset of @xmath119 such that if @xmath360 , then @xmath29 contains only path connected sets . there are level morphisms of inverse systems @xmath361 and @xmath362 such that @xmath34 & { \pi^{sp}(\mathscr{u},x_0)}\ar[r]^-{i_{{\mathscr{u } } } } & { \pi_{1}(x , x_0)}\ar[r]^-{p_{{\mathscr{u}}\ast } } & \pi_{1}(\|{n(\mathscr{u})}\|,u_0 ) \ar[r ] & 1}\ ] ] is exact for each @xmath363 . moreover , @xmath34 & ( { \pi^{sp}(\mathscr{u},x_0)},i_{{\mathscr{u}}{\mathscr{v}}},\lambda ' ) \ar[r]^-{i } & ( { \pi_{1}(x , x_0)},id,\lambda ' ) \\ & \ar[r]^-{p } & ( \pi_{1}(\|{n(\mathscr{u})}\|,u_0),p_{{\mathscr{u}}{\mathscr{v}}\ast},\lambda ' ) \ar[r ] & 1}\ ] ] is an exact sequence in the category of pro - groups . the level sequences , i.e. the sequences for each element of @xmath364 , are exact by theorem [ shortexactsequence ] . this gives rise to the exact sequence in the category of pro - groups ( * ? ? ? ii , 2.3 , theorem 10 ) . theorems [ characterizethickspangrp ] and [ shortexactsequence ] combine to give a characterization of @xmath35 in terms of the @xmath29-homotopy of the previous section . if @xmath29 consists of path connected sets and @xmath365\to x$ ] is a loop based at @xmath44 , then @xmath147\in \ker p_{\mathscr{u}\ast}$ ] if and only if @xmath266 is null-@xmath29-homotopic . [ homotopicpaths ] let @xmath45 be a simplicial complex and @xmath367\to \|k\|$ ] be loops where @xmath368 and @xmath369 . if there are vertices @xmath370 and partitions @xmath371 such that @xmath372)\cup l^{\prime}([t_{k-1},t_k])\subset st(v_k , k)$ ] for each @xmath373 , then @xmath374 and @xmath375 are freely homotopic in @xmath51 . if @xmath376 , then @xmath374 and @xmath375 are homotopic rel . basepoint . since a non - empty intersection of two stars in a simplicial complex is path connected , there is path @xmath377\to st(v_{k},k)\cap st(v_{k+1},k)$ ] from @xmath378 to @xmath379 for @xmath380 . let @xmath381 be a path in @xmath382 from @xmath383 to @xmath384 . each loop @xmath385}\ast \gamma_{k}\ast \overline{l'\|_{[t_{k-1},t_k]}}\ast \overline{\gamma_{k-1}}$ ] , @xmath386 has image in the star of a vertex and is therefore inessential . thus @xmath374 and @xmath387 are freely homotopic . if @xmath376 , the same argument produces a basepoint - preserving homotopy when we take @xmath388 to be the constant path at @xmath376 . [ surjectivity ] if @xmath29 consists of path connected sets and @xmath389\ast [ u_2,u_3]\ast \dots \ast[u_{n-1},u_n]$ ] is an edge loop in @xmath38 such that @xmath390 , then there is a loop @xmath365\to x$ ] based at @xmath44 such that @xmath391 and @xmath94 are homotopic rel . basepoint . since each @xmath392 is path connected and @xmath393 , there is a loop @xmath365\to x$ ] based at @xmath44 such that @xmath394\right)\subseteq u_j$ ] for each @xmath395 . we claim @xmath94 is homotopic to @xmath396 . recall @xmath397 for each @xmath398 so that @xmath399 is an open refinement of @xmath118 . consequently , for each @xmath395 , there is a subdivision @xmath400 and a sequence @xmath401 such that @xmath402)\subseteq v_{w_{j}^{k}}\subseteq w_{j}^{k}$ ] for @xmath403 . we may assume @xmath404 for @xmath405 and , since @xmath406 , we may assume @xmath407 . note that @xmath408)\subseteq st(w_{j}^{k},{n(\mathscr{u})})$ ] . + define an edge loop @xmath387 based at the vertex @xmath110 as@xmath409\right)\ast \dots \ast\left(\ast_{k=1}^{m_j-1}[w_{j}^{k},w_{j}^{k+1}]\right)\ast \dots\ast \left(\ast_{k=1}^{m_n-1}[w_{n}^{k},w_{n}^{k+1}]\right)\ ] ] to see @xmath387 is homotopic to @xmath374 , observe the set of intervals @xmath410 $ ] , and consequently the set of neighborhoods @xmath411 , inherits an ordering from the natural ordering of @xmath75 $ ] . suppose the intervals @xmath410 $ ] are ordered as @xmath412 $ ] and the neighborhoods @xmath411 are ordered as @xmath413 so that @xmath414 . for @xmath415 , let @xmath416 be the barycenter of @xmath417 $ ] . find a partition @xmath418 such that @xmath419)\subseteq st(w_\ell,{n(\mathscr{u})})$ ] for @xmath420 by choosing @xmath421 so that @xmath422 for @xmath420 . lemma [ homotopicpaths ] applies and gives a basepoint preserving homotopy between @xmath374 and @xmath387 . it now suffices to show the edge loops @xmath387 and @xmath389\ast [ u_2,u_3]\ast \dots \ast[u_{n-1},u_n]$ ] are homotopic . we do so by showing @xmath423 is homotopic to a concatenation of inessential loops based at @xmath110 . first , for each @xmath395 , let @xmath424\ast \dots \ast[u_{j-1},u_j]$ ] . observe @xmath425 $ ] is constant and @xmath426 . + 1 . for @xmath395 and @xmath427 , let @xmath428\ast [ w_{j}^{k},w_{j}^{k+1}]\ast[w_{j}^{k+1},u_j]\ast \overline{e_{j}}.\ ] ] these loops are well - defined and inessential since @xmath429 determines a 2-simplex in @xmath430 . even in the case two or more of @xmath392 , @xmath411 , @xmath431 are equal , @xmath432 is still inessential . 2 . for @xmath405 , let @xmath433\ast [ w_{j}^{m_j},u_{j+1}]\ast[u_{j+1},u_j]\ast \overline{e_{j}}.\ ] ] these loops are well - defined and inessential since @xmath434 . therefore the product @xmath435is inessential . we observe @xmath62 is homotopic to @xmath423 by reducing words . indeed , for each @xmath405 , @xmath436 reduces to @xmath437\ast \left(\ast_{k=1}^{m_j-1}[w_{j}^{k},w_{j}^{k+1}]\right)\ast [ w_{j}^{m_j},u_{j+1}]\ast \overline{e_{j}}\ ] ] and the last factor @xmath438 reduces to @xmath439\ast \left(\ast_{k=1}^{m_n-1}[w_{n}^{k},w_{n}^{k+1}]\right)\ast \overline{e_{n}}$ ] ( recall that @xmath440 ) . using the fact that @xmath441 and @xmath404 to identify constant paths , it follows that @xmath62 reduces to @xmath442 . [ exactness1 ] if @xmath355 where @xmath29 consists of path connected sets and @xmath130 is a canonical map , then @xmath366 is surjective . since edge loops in @xmath236 represent all homotopy classes in @xmath443 , the theorem follows from lemma [ surjectivity ] . the flavor of the proof of the following lemma is quite similar to that of lemma [ surjectivity ] but holds when the elements of @xmath29 are not necessarily path connected . [ inclusion2 ] let @xmath29 be an open cover of @xmath2 . suppose @xmath167 are paths in @xmath2 such that @xmath445 and @xmath446 . if @xmath328 has image in @xmath447 , then @xmath448 is an inessential loop in @xmath38 . we again use that @xmath449 is open refinement of @xmath118 . find partitions @xmath450 and @xmath451 and a sequence of neighborhoods @xmath452 such that @xmath453)\subseteq v_{w_{j}^{k}}\subseteq w_{j}^{k}$ ] for @xmath454 and @xmath403 . since @xmath446 , we may assume @xmath455 and since @xmath445 , we may assume @xmath456 . observe that @xmath457)\subseteq st(w_{j}^{k},{n(\mathscr{u})})$ ] . + let @xmath387 be the loop defined as the concatenation @xmath409\right)\ast \left(\ast_{k=1}^{m_2 - 1}[w_{2}^{k},w_{2}^{k+1}]\right).\ ] ] find a partition @xmath458 such that @xmath459)\subseteq st(w_{j}^{k},{n(\mathscr{u})})$ ] . by lemma [ homotopicpaths ] , @xmath374 and @xmath387 are freely homotopic . + it now suffices to show @xmath387 is inessential . we do so by showing @xmath387 is homotopic to a concatenation of inessential loops . first , let @xmath460 $ ] and @xmath461\ast[u_1,u_2]$ ] . 1 . for @xmath462 , let @xmath463\ast [ w_{1}^{k},w_{1}^{k+1}]\ast [ w_{1}^{k+1},u_1]\ast \overline{e_{1}}.\ ] ] 2 . let @xmath464\ast [ w_{2}^{1},u_2]\ast [ u_2,u_1]\ast \overline{e_{1}}$ ] . 3 . for @xmath465 , let @xmath466\ast [ w_{2}^{k},w_{2}^{k+1}]\ast [ w_{2}^{k+1},u_2]\ast \overline{e_{2}}.\ ] ] 4 . let @xmath467=[w_{1}^{1},u_1]\ast[u_1,u_2]\ast[u_2,w_{1}^{1}]$ ] . note the loop @xmath468 is well - defined and inessential since @xmath469 determines a 2-simplex in @xmath430 . similarly , the loops defined in 2.-4 . are well - defined and inessential . consequently , the concatenation @xmath470 is an inessential loop based at @xmath471 . we observe @xmath62 is homotopic to @xmath387 by reducing words . indeed , @xmath472 reduces to @xmath473\right)\ast [ w_{2}^{1},u_{2}]\ast \overline{e_{2}}$ ] and @xmath474 reduces to @xmath475\ast \left(\ast_{k=1}^{m_2 - 1}[w_{2}^{k},w_{2}^{k+1}]\right)$ ] . [ exactness2 ] if @xmath355 and @xmath130 is a canonical map , then @xmath476 . suppose @xmath196[\gamma_1\ast\gamma_2][{\overline{\alpha}}]$ ] is a generator of @xmath163 where @xmath328 is a path with image in @xmath233 . since @xmath477[p_{{\mathscr{u}}}\circ(\gamma_1\ast\gamma_2)][\overline{p_{{\mathscr{u}}}\circ \alpha}]$ ] and @xmath478 is inessential by lemma [ inclusion2 ] , @xmath479 is trivial in @xmath443 . to show the inclusion @xmath480 holds under the hypotheses of theorem [ shortexactsequence ] , we work to extend loops @xmath74 to maps on the 1-skeleton of subdivisions of @xmath60 . _ an edge loop @xmath374 is _ short _ if it is the concatenation @xmath481\ast [ v_1,v_2]\ast [ v_2,v_3]$ ] of three linear paths . spanier edge loop _ is an edge loop of the form @xmath482 where @xmath94 is an edge path and @xmath374 is a short edge loop . _ [ fishnet ] let @xmath85 be a subgroup of @xmath42 and @xmath280 be an integer . let @xmath483 be a based map on the 1-skeleton of @xmath484 , and @xmath485 be the restriction . if @xmath486\in h$ ] for every spanier edge loop @xmath487 in @xmath484 , then @xmath295\in h$ ] . it follows from elementary planar graph theory that @xmath488 is generated by the homotopy classes of spanier edge loops . thus if @xmath486\in h$ ] for every spanier edge loop @xmath487 in @xmath489 , then @xmath490 . in particular @xmath295\in h$ ] . let @xmath491 be the first barycentric subdivision of @xmath59 . the 0-skeleton consists of six vertices : let @xmath492 be the vertices @xmath493 of @xmath59 respectively and @xmath494 be the vertex which is the barycenter of the edge opposite @xmath495 for @xmath496 . _ let @xmath118 be an open cover of @xmath2 . a map @xmath497 is @xmath118__-admissible _ _ if there are open neighborhoods @xmath498 such that @xmath499 and @xmath500 for @xmath496 . _ _ the condition @xmath500 in the previous definition means precisely that @xmath501\cup [ v_1,m_3])\subseteq u_1 $ ] , @xmath502\cup [ v_2,m_3])\subseteq u_2 $ ] , and @xmath503\cup [ v_3,m_2])\subseteq u_3 $ ] . _ [ uadmissible ] let @xmath29 be an open cover of @xmath2 consisting of path connected sets . if @xmath497 is @xmath29-admissible and @xmath365\to x$ ] is a path from @xmath44 to @xmath504 , then @xmath147[\delta][{\overline{\alpha}}]\in \pi^{sp}(\mathscr{u},x_0)$ ] . since @xmath505 is @xmath118-admissible there are sets @xmath506 such that @xmath507 contains a point @xmath508 and @xmath509 for @xmath496 . whenever @xmath510 and @xmath511 , let @xmath512 be the path which is the restriction of @xmath513 to @xmath514 $ ] . with this notation @xmath515= [ \gamma_{1,3}][\overline{\gamma_{2,3}}][\gamma_{2,1}][\overline{\gamma_{3,1}}][\gamma_{3,2 } ] [ \overline{\gamma_{1,2}}].\ ] ] note @xmath516 for all choice of @xmath517 . thus if @xmath518 are distinct , the endpoint of @xmath512 lies in @xmath519 . define six paths in @xmath2 : whenever @xmath510 and @xmath511 , let @xmath520 be a path in @xmath392 from @xmath521 to @xmath508 . such paths are guaranteed to exist since each element of @xmath118 is path connected . now given any path @xmath365\to x$ ] from @xmath44 to @xmath504 let @xmath522[\gamma_{1,3}\ast\eta_{1,3}\ast\overline{\eta_{1,2}}\ast\overline{\gamma_{1,2}}][{\overline{\alpha}}]\\ \zeta_{2 } & = & [ \alpha\ast\gamma_{1,2}\ast\eta_{1,2}][\overline{\eta_{1,3}}][\eta_{2,3}][\overline{\alpha\ast\gamma_{1,2}\ast \eta_{1,2}}]\\ \zeta_{3 } & = & [ \alpha\ast\gamma_{1,2}\ast\eta_{1,2}][\overline{\eta_{2,3}}\ast\overline{\gamma_{2,3}}\ast\gamma_{2,1}\ast\eta_{2,1 } ] [ \overline{\alpha\ast\gamma_{1,2}\ast\eta_{1,2 } } ] \\ \zeta_{4 } & = & [ \alpha\ast\gamma_{1,2}\ast\eta_{1,2}][\overline{\eta_{2,1}}][\eta_{3,1 } ] [ \overline{\alpha\ast\gamma_{1,2}\ast\eta_{1,2}}]\\ \zeta_{5 } & = & [ \alpha\ast\gamma_{1,2}\ast\eta_{1,2}][\overline{\eta_{3,1}}\ast\overline{\gamma_{3,1}}\ast\gamma_{3,2}\ast\eta_{3,2 } ] [ \overline{\alpha\ast\gamma_{1,2}\ast\eta_{1,2}}]\\ \zeta_{6 } & = & [ \alpha\ast\gamma_{1,2}][\eta_{1,2}][\overline{\eta_{3,2 } } ] [ \overline{\alpha\ast\gamma_{1,2 } } ] \end{array}\ ] ] note each @xmath523 is written as a product to illustrate that @xmath524 , @xmath525 , @xmath526 and @xmath527 , @xmath528 , @xmath529 . since @xmath166 , we have @xmath530 for each @xmath531 . a straightforward check gives @xmath147[\delta][{\overline{\alpha}}]=\zeta_1\zeta_2\zeta_3\zeta_4\zeta_5\zeta_6 $ ] and thus @xmath147[\delta][{\overline{\alpha}}]\in { \pi^{sp}(\mathscr{u},x_0)}$ ] . [ admissibletheorem ] if @xmath483 is a map such that the restriction of @xmath532 to the boundary @xmath533 of every 2-simplex @xmath534 in @xmath484 is @xmath29-admissible and @xmath535 , then @xmath295\in \pi^{sp}(\mathscr{u},x_0)$ ] . by lemma [ fishnet ] it suffices to show @xmath536\in { \pi^{sp}(\mathscr{u},x_0)}$ ] for any spanier edge loop @xmath487 in @xmath489 . suppose @xmath537 is such a spanier edge loop where @xmath94 is an edge path from @xmath538 to a vertex of @xmath533 and @xmath374 is a short edge loop traversing @xmath533 . let @xmath539 and @xmath540 . since @xmath513 is @xmath118-admissible by assumption , @xmath486=[\alpha][\delta][{\overline{\alpha}}]\in{\pi^{sp}(\mathscr{u},x_0)}$ ] by lemma [ uadmissible ] . [ technical ] suppose @xmath355 where @xmath29 consists of path connected sets and @xmath130 is a canonical map . if @xmath541 is a loop such that @xmath295\in \ker p_{\mathscr{u}\ast}$ ] , then there exists an integer @xmath280 and an extension @xmath542 of @xmath267 such that the restriction of @xmath532 to the boundary @xmath533 of every 2-simplex @xmath534 in @xmath484 is @xmath29-admissible . since @xmath543 is a null - homotopic loop based at the vertex @xmath110 , it extends to a map @xmath544 . since , by the definition of @xmath545 , we have @xmath397 for each @xmath228 , the cover @xmath399 is an open refinement of @xmath118 . additionally , if @xmath546 , the collection @xmath547 is an open cover of @xmath548 . following theorem 16.1 ( finite simplicial approximation ) in @xcite , find a simplicial approximation for @xmath549 using the cover @xmath179 : let @xmath550 be a lebesgue number for @xmath179 so that any subset of @xmath548 of diameter less than @xmath550 lies in some element of @xmath179 . choose an integer @xmath551 such that each simplex in @xmath489 has diameter less than @xmath552 . thus the star @xmath553 of each vertex @xmath554 in @xmath489 lies in a set @xmath555 for some @xmath556 . the assignment @xmath557 on vertices extends to a simplicial approximation @xmath558 of @xmath549 , i.e. a simplicial map @xmath559 such that @xmath560 for each vertex @xmath554 ( * ? ? ? * lemma 14.1 ) . we construct an extension @xmath561 of @xmath267 such that the restriction of @xmath532 to the boundary of each 2-simplex of @xmath489 is @xmath118-admissible . first , define @xmath532 on vertices : for each vertex @xmath562 , pick a point @xmath563 . in particular , if @xmath554 is a vertex of the subcomplex @xmath564 of @xmath565 , take @xmath566 . this choice is well defined since whenever @xmath567 , we have @xmath568 and thus @xmath569 . observe if @xmath570 $ ] is any 2-simplex in @xmath571 , then @xmath572)$ ] is a simplex of @xmath430 spanned by the set of vertices @xmath573 ( possibly containing repetitions ) . by the construction of @xmath430 , the intersection @xmath574 is non - empty . extend @xmath532 to the interiors of 1-simplices of @xmath575 ; to define @xmath532 on a given 1-simplex @xmath576 $ ] , consider two cases : case i : suppose 1-simplex @xmath576 $ ] is the intersection of two 2-simplices in @xmath484 . extend @xmath532 to the interior of @xmath576 $ ] by taking @xmath577}$ ] to be a path in @xmath578 from @xmath579 to @xmath580 such that if @xmath581 is the barycenter of @xmath576 $ ] , then @xmath577}([a , m])\subseteq u_a$ ] and @xmath577}([m , b])\subseteq u_b$ ] . such a path is guaranteed to exist since @xmath582 , @xmath583 are path connected and have non - trivial intersection . case ii : suppose 1-simplex @xmath576 $ ] is the face of a single 2-simplex or equivalently that @xmath576 $ ] is a 1-simplex of @xmath564 . in this case , let @xmath577}=\beta_{[a , b]}$ ] . again , note @xmath584 . since @xmath585 and @xmath586 , we have @xmath587))=h(int([a , b]))\subseteq st(u_a,{n(\mathscr{u})})\cap st(u_{b},{n(\mathscr{u})}).\]]thus @xmath588))\subseteq v_{u_a}\cap v_{u_b}\subseteq u_a\cap u_{b}$ ] . note if @xmath581 is the barycenter of @xmath576 $ ] , then @xmath589)\subseteq u_a$ ] and @xmath590)\subseteq u_b$ ] . now @xmath561 is a map extending @xmath267 such that the restriction of @xmath532 to the boundary @xmath533 of every 2-simplex @xmath534 in @xmath489 is @xmath118-admissible . [ exactness3 ] if @xmath355 where @xmath29 consists of path connected sets and @xmath130 is a canonical map , then @xmath591 . the theorem follows directly from lemma [ admissibletheorem ] and lemma [ technical ] . theorem [ shortexactsequence ] allows us to characterize the kernel of the natural map @xmath593 as the spanier group @xmath13 . [ leftexactsequence ] if @xmath2 is a locally path connected , paracompact hausdorff space , then there is a natural exact sequence @xmath34 & { { \pi^{sp}}(x , x_0)}\ar[r ] & { \pi_{1}(x , x_0)}\ar[r]^-{\psi_{x } } & { \check{\pi}_{1}(x , x_0)}}\ ] ] where the second map is inclusion . it suffices to show @xmath594 . the inclusion @xmath595 holds for arbitrary @xmath2 ; this fact follows from propositions [ spanierinclusion ] and [ inclusion1 ] and is proved directly in @xcite . since @xmath2 is paracompact hausdorff , @xmath596 by theorem [ paracompact1 ] . thus , it suffices to show @xmath597 . suppose @xmath295\in \ker\psi_{x}$ ] ( equivalently @xmath295\in \ker p_{{\mathscr{w}}\ast}$ ] for every @xmath598 ) and @xmath118 is a given open cover of @xmath2 . it suffices to show @xmath295\in { \pi^{sp}(\mathscr{u},x_0)}$ ] . let @xmath125 be a path connected neighborhood of @xmath44 contained in @xmath599 . every point @xmath600 is an element of some @xmath601 . let @xmath602 be a path connected neighborhood of @xmath244 contained in @xmath603 which is disjoint from @xmath604 . now @xmath605 is an open cover of @xmath2 which contains path connected sets , is a refinement of @xmath118 , and such that there is a single set @xmath125 having the basepoint @xmath44 as an element . since @xmath2 is paracompact hausdorff , @xmath126 is normal and therefore @xmath127 . since @xmath295\in \ker p_{{\mathscr{v}}\ast}$ ] and @xmath606 by theorem [ shortexactsequence ] , we have @xmath295\in { \pi^{sp}}({\mathscr{v}},x_0)$ ] . note that @xmath607 since @xmath126 refines @xmath118 . thus @xmath295\in { \pi^{sp}(\mathscr{u},x_0)}$ ] . regarding naturality , it is well - known that @xmath157 is natural in @xmath2 . thus it suffices to check that if @xmath177 , @xmath178 is a map , then @xmath608 . this follows directly from remark [ naturalityofspanier ] . [ shapevanishing ] if @xmath2 is a locally path connected , paracompact hausdorff space , then @xmath2 is @xmath145-shape injective if and only if @xmath609 . [ example1 ] _ theorem [ leftexactsequence ] fails to hold in the non - locally path connected case . a counterexample is the compact space @xmath610 of @xcite obtained by rotating the closed topologist s sine curve so the linear path component forms a cylinder and connecting the two resulting surface components by attaching a single arc . we have @xmath611 and @xmath612 , yet @xmath613 . _ the identification @xmath614 in theorem leftexactsequence allows us to give two alternative characterizations of @xmath592 : one in terms of @xmath29-homotopy ( from section [ sectionuhomotopy ] ) and one in terms of covering spaces of @xmath2 . here covering spaces and maps are meant in the classical sense @xcite . [ uhomotopycharacterization ] suppose @xmath2 is a locally path connected , paracompact hausdorff space and @xmath365\to x$ ] is a loop based at @xmath44 . the following are equivalent : 1 . @xmath147\in\ker\psi_{x}$ ] 2 . for every open cover @xmath29 of @xmath2 , @xmath266 is null-@xmath29-homotopic . 3 . for every covering map @xmath615 , @xmath616 , the unique lift @xmath617\to y$ ] of @xmath266 ( i.e. such that @xmath618 ) starting at @xmath619 is a loop in @xmath180 . @xmath620 2 . ) by theorems [ paracompact1 ] and [ leftexactsequence ] , we have @xmath621 . since @xmath2 is locally path connected , every open cover @xmath118 admits an open refinement @xmath126 whose elements are path connected and thus @xmath622 by theorem [ characterizethickspangrp ] . since @xmath622 for all @xmath126 in a cofinal subset of @xmath623 the equality @xmath624 follows . ( 1 . @xmath620 3 . ) since @xmath625 , we show @xmath626 is the intersection of all images @xmath627 where @xmath615 , @xmath616 is a covering map . this equality follows directly from the following well - known result from the covering space theory of locally path connected spaces ( see lemma 2.5.11 and theorem 2.5.13 in @xcite ) : given a subgroup @xmath85 of @xmath164 , there is an open cover @xmath118 of @xmath2 such that @xmath628 if and only if there is a covering map @xmath615 , @xmath616 with @xmath629 . [ example2 ] _ consider the peano continua @xmath630 and @xmath631 of @xcite constructed as subspaces of @xmath632 ( these spaces also appear as the spaces @xmath633 and @xmath634 in @xcite respectively ) . the space @xmath630 is constructed by rotating the topologists sine curve @xmath635 $ ] about the central axis " ( i.e. the portion @xmath636 $ ] of the z - axis ) and attaching horizontal arcs so @xmath630 is locally path connected and so the arcs become dense only on the central axis . the space @xmath631 is an inverted version of @xmath630 in the sense that @xmath637 is now rotated around the vertical line passing through @xmath638 and arcs are attached becoming dense around the outside cylinder . take the basepoint @xmath639 of @xmath631 to lie on the outer cylinder.the authors of @xcite observe both spaces have non - vanishing spanier group and thus fail @xmath145-shape injectivity . the spanier group of @xmath630 is known to be non - trivial ( * ? ? ? * prop 3.2(1 ) ) , however slightly more effort is required to show a simple closed curve @xmath374 , based at @xmath639 , traversing the outer cylinder of @xmath631 is homotopically non - trivial ( * ? ? ? * lemma 3.1 ) and satisfies @xmath640\in \pi^{sp}(z^{\prime},z_0)$ ] . to do so , one can , given an open cover @xmath29 of @xmath2 , look for small arcs in the intersections of elements of @xmath29 and factor @xmath640 $ ] as a product of generators of @xmath641.using the above results , we can observe @xmath640\in \pi^{sp}(z^{\prime},z_0)$ ] without having to construct explicit factorizations and without having to rely on convenient covers with enough " arcs lying in intersections . for instance , to see @xmath640\in \pi^{sp}(z^{\prime},z_0)$ ] , let @xmath29 be an open cover of @xmath631 and find a subdivision of @xmath642 so @xmath643)\subseteq u_i$ ] for @xmath644 and @xmath645 . there is an arc @xmath266 connecting @xmath646 to the surface portion with image in @xmath647 and a loop @xmath375 on the surface portion based at @xmath150 such that @xmath648)\subseteq u_i$ ] for each @xmath531 . thus @xmath374 is @xmath29-homotopic to the homotopically trivial loop @xmath649 . since @xmath640 $ ] is null-@xmath29-homotopic for any given @xmath29 , we have @xmath640\in \ker\psi_{x}=\pi^{sp}(z^{\prime},z_0)$ ] by corollary uhomotopycharacterization . _ _ if , in addition to the hypotheses on @xmath2 in theorem leftexactsequence , @xmath119 admits a cofinal , directed subsequence @xmath650 where @xmath651 refines @xmath652 ( for instance , when @xmath2 is compact metric ) , the exact sequence in theorem [ leftexactsequence ] extends to the right via the first derived limit @xmath653 for non - abelian groups ( * ? ? ? ii , 6.2 ) . in doing so , we obtain a connecting function @xmath513 and an exact sequence @xmath34 & { { \pi^{sp}}(x , x_0)}\ar[r ] & { \pi_{1}(x , x_0)}\ar[r]^-{\psi_{x } } & { \check{\pi}_{1}(x , x_0)}\ar[r]^-{\delta } & \varprojlim^{1}{\pi^{sp}}({\mathscr{u}}_n , x_0 ) \ar[r ] & \ast } \ ] ] in the category of pointed sets . thus elements in the image of @xmath157 are precisely those mapped to basepoint of the first derived limit of the inverse sequence @xmath654 of thick spanier groups . _ for a general space @xmath2 , the fundamental group @xmath655 admits a variety of distinct natural topologies @xcite , one of which is the pullback from the product of discrete groups @xcite described as follows . by definition , the geometric realization of a simplicial complex is locally contractible . hence the inverse limit space @xmath6 admits a natural topology as a subspace of the product of discrete groups @xmath656 to impart a topology on @xmath10 , recall the homomorphism @xmath657 and declare the open sets of @xmath10 to be precisely sets of the form @xmath658 where @xmath659 is open in @xmath660 we refer to this group topology on @xmath655 as the _ shape topology_. the space @xmath6 is hausdorff ( since @xmath6 is a subspace of the arbitrary product of hausdorff spaces ) . hence , if @xmath661 is one - to - one , then @xmath10 is hausdorff ( since in general the preimage under a continuous injective map of a hausdorff space is hausdorff ) . conversely , if @xmath661 is not one - to - one , then @xmath10 is not hausdorff ( since , with the pullback topology , if @xmath662 and @xmath663 then @xmath244 and @xmath664 can not be separated by open sets ) . the foregoing is summarized as follows . [ opensubgroup ] _ note if @xmath665 is a topological group , @xmath85 is a subgroup of @xmath665 , and @xmath45 is a subgroup of @xmath85 which is open in @xmath665 , then @xmath85 is also open in @xmath665 since @xmath85 decomposes as a union of open cosets of @xmath45 . _ suppose @xmath118 is an open cover of @xmath2 . pick a set @xmath666 such that @xmath101 . since @xmath2 is locally path connected , paracompact hausdorff , there exists @xmath127 refining @xmath123 such that @xmath126 consists of path connected sets ( e.g. see the the second paragraph in the proof of theorem [ leftexactsequence ] ) . let @xmath667 be a based canonical map . recall @xmath668 is discrete and observe @xmath669 is continuous . additionally , @xmath670 by theorem [ shortexactsequence ] . thus @xmath671 is an open subgroup of @xmath164 . since @xmath126 refines @xmath118 , we have @xmath672 and thus @xmath163 is open by remark [ opensubgroup ] . since @xmath2 is paracompact hausdorff , there is an open refinement @xmath673 of @xmath118 such that @xmath674 ( see the proof of theorem [ paracompact1 ] ) . by the previous paragraph , @xmath675 is open . thus @xmath165 is open by remark [ opensubgroup ] . the hypotheses on @xmath2 allow us to replace @xmath119 with the cofinal directed subset @xmath364 consisting of pairs @xmath123 such that @xmath118 contains only path connected sets . we alter notation for brevity . for @xmath676 , let @xmath677 be the identity of @xmath678 . additionally , let @xmath679 , and @xmath680 . if @xmath681 in @xmath364 , let @xmath682 be homomorphism induced by the projection . thus @xmath683 and @xmath684 . + in @xmath685 , a basic open neighborhood of the identity is of the form @xmath686where @xmath687 is a finite subset of @xmath364 . therefore the sets @xmath688 form a neighborhood base at the identity of @xmath164 . by theorem [ shortexactsequence ] , @xmath689 for each @xmath690 . thus @xmath691 for each finite set @xmath692 . for given @xmath687 , take @xmath693 such that @xmath694 refines @xmath550 for each @xmath695 . this gives @xmath696 . consequently , the open subgroups @xmath697 for a neighborhood base at the identity . by theorem [ thickspanierbasis ] , the thick spanier groups @xmath37 form a neighborhood base at the identity of @xmath42 . for every open cover @xmath118 of @xmath2 , the spanier group @xmath36 is an open subgroup of @xmath698 by lemma [ tspanieropen2 ] and @xmath161 by proposition [ tspanier1 ] . thus the spanier groups @xmath36 form a neighborhood base at the identity . @xmath620 2 . ) follows directly from corollary [ spanierbasis ] and remark [ opensubgroup ] . @xmath620 3 . ) is a well - known result from covering space theory ( used above in the proof of corollary [ uhomotopycharacterization ] ) . thus the well - known classification of covering spaces can be extended to non - semilocally simply connected spaces in the following way : if @xmath2 is locally path connected , paracompact hausdorff , then there is a canonical bijection between the equivalence classes of connected coverings of @xmath2 ( in the classical sense ) and conjugacy classes of open subgroups of @xmath42 with the shape topology .	we develop a new route through which to explore @xmath0 the kernel of the @xmath1shape group homomorphism determined by a general space @xmath2 , and establish , for each locally path connected , paracompact hausdorff space @xmath2 , @xmath3 is precisely the spanier group of @xmath2 .
we would like to thank z. nussinov for pointing us to us the recent quant - ph reference by golestanian and kardar , and to gilda reyes for help with the preparation of the manuscript . [ 1 ] h. b. g. casimir , proc . k. ned akad wet . * 51 * , 793 ( 1948 ) . [ 2 ] see , e.g. , c. itzykson and j. b. zuber , _ quantum field theory _ , mcgraw hill ( 1985 ) , p. 137 . j. sparany , _ nature _ * 180 * , 334 ( 1957 ) and _ physica _ * 24 * , 751 ( 1958 ) . [ 4 ] s. k. lamoreaux , phys . lett . * 78 * , 5 ( 1997 ) . [ 5 ] p. w. miloni , r. j. cook , and m. e. goggin , phys . a38 , 1621 ( 1988 ) . [ 6 ] s. nussinov , polarized / magnetic versions of the casimir forces " , hep - ph/9509228 ( september 1995 ) . [ 7 ] this method , utilizing direct path integration , has been independently discovered by one of us ( o. keneth ( 1996 ) , unpublished ) , while attempting an exact evaluation of the casimir effect for different directions of conductivity suggested in ref . ( 6 ) above . we have found that m. kardar and collaborators ( h. li , r. golestanian , m. goulian ) suggested this method sometime ago and applied it to a broad range of dynamical and other casimir effects though not to the specific electromagnetic effects of interest here ; see , e.g. h. li and m. kardar , phys . 67 * , 3275 ( 1991 ) , phys . a46 , 6490 ( 1992 ) , r. golestanian and m. kardar , quant - ph/9802017 ( february 1998 ) . [ 8 ] since the current @xmath111 is conserved , we can adopt this particular choice . [ 9 ] notice that the original one - half factor from eq . ( 10 ) survives ; we have to integrate over both the real and imaginary parts of @xmath112 but thanks to eq . ( 9 ) , only over half the @xmath57 values say those with @xmath113 larger than 0 . [ 10 ] it should be emphasized though , that this finite result is not obtained by discarding what formally looks like self - interactions " pertaining to each plate separately , namely , the @xmath114 and @xmath115 terms . indeed , these diagonal terms are crucial for making the gaussian integral convergent and meaningful . [ 11 ] h. g. b. casimir and d. polder , phys * 73 * , 369 ( 1948 ) . for a more recent , detailed , derivation , see g. feinberg and j. sucher , phys . rev . a2 , 2395 ( 1970 ) . [ 12 ] there are also figure eight - type configurations in which the b field lines thread the two conducting rings in opposite directions and these would induce a repulsive interaction . however , the b line can not self - intersect . the need for spatial avoidance causes the flux lines to be longer and the action suppression to be stronger for such configurations . [ 1 ] the casimir energy versus the angle between the conducting directions .	a new version of the casimir effect where the two plates conduct in specific , different , directions is considered . by direct functional integration the evaluation of the casimir energy as a function of the angle between the conduction directions is reduced to quadratures . other applications of the method and a magnetic casimir variant are mentioned . hep - th/9802149 + february 14 , 1998 + 2.cm * a new variant of the casimir effect and its exact evaluation . * 2.cm oded kenneth and shmuel nussinov 0.5 cm the casimir force per unit area [ 1 ] @xmath0 attracts two parallel conducting plates at a distance @xmath1 apart in vacuum . its independence of atomic and qed parameters reflects the perfect conductor idealization where all details are subsumed into boundary conditions @xmath2 . these , in turn , quantize the @xmath3 component of the wave number vector for modes in the region between the plates , @xmath4 . the problem then reduces to evaluating the change in vacuum energy of all the transverse modes inside this region : @xmath5 a careful regularization of this formally divergent expression yields [ 2 ] @xmath6 and @xmath7 . the tiny casimir forces are elusive . past efforts [ 3 ] verified eq . ( 1 ) rather roughly and only recently a 5% precision experiment was done [ 4 ] . in this paper we focus mainly on a new variant of the casimir effect : each of the two plates conduct in a specific direction : @xmath8 for plate one and @xmath9 on plate two so that only the components @xmath10 and @xmath11 need to vanish on plates i and ii respectively , and @xmath12 is arbitrary . the two polarizations contribute equally to @xmath13 . by the above twist " these polarizations can generate a controlled @xmath14 and a casimir torque , @xmath15 . the casimir force can be derived also [ 5 ] by evaluating the pressure imbalance due to reflection of vacuum modes " off the outside surface of the plate and of the ( fewer ) internal modes off the inside surfaces . if the two plates are replaced by arrays of only vertical ( or horizontal ) conducting wires ( mimicking the anisotropic conductivities ) then only the @xmath16 polarized modes will be reflected suggesting that we have half the casimir force . we also expect further reduction [ 6 ] of the casimir force as @xmath17 , the angle between the directions of the two sets of wires ( or the directions of conductivity in the two plates ) increases from @xmath18 to @xmath19 . we will next present an exact evaluation of @xmath20 for general @xmath17 s using an altogether different method [ 7 ] . we add to the free electromagnetic action @xmath21 a lagrange multiplier term @xmath22 ( where the last integral is only over the area - time of the plates ) . the functional integration over @xmath23 ensures the vanishing of the transverse electric field @xmath24 over the plates . since we want the @xmath25 field to vanish only up to gauge transformations we integrate only over conserved currents @xmath26 . also since only transverse components of @xmath27 have to vanish we allow @xmath26 to have components only in the three - dimensional area - time of the plates i.e. @xmath28 . thus we will write for the partition function @xmath29 the only difference in the case of interest , where each plate conducts in a specific direction , is that only one component of @xmath24 has to vanish on each plate . this can be achieved by further restricting the allowed currents @xmath26 . hence eq . ( 6 ) and its consequences remain correct as long as we remember to interpret @xmath30 differently . changing the order of integration and doing first the standard @xmath31 integration yields : @xmath32 where @xmath33 is the massless feynman propagator [ 8 ] ( constant s coefficients such as the @xmath34 above contribute only an overall multiplicative factor or an additive term to the energy and will be discarded henceforth ) . denoting the currents on the first and second plates by @xmath35 and @xmath36 , we have the more explicit expression for @xmath37 : @xmath38 or after a wick rotation @xmath39 where we think of @xmath40 as of ordinary 3-vectors in ordinary 3-dimensional euclidean space ( although it is actually spanned by @xmath41 ) . fourier transforming in @xmath42 this becomes : @xmath43 where @xmath44 and we used translation invariance . in the usual case of two conducting plates both @xmath45 and @xmath46 have two transverse degrees of freedom fixed by the current conservation condition : @xmath47 . in the case of specific conduction directions , @xmath45 , and likewise @xmath46 , have only one allowed non - zero component determined by current conservation and by the demand that its spatial part @xmath48 is along the direction of conduction . let us denote the cosine of the angle between the directions of @xmath49 and @xmath50 by @xmath51 ( with @xmath52 an ordinary euclidean vector ) . then we can write for @xmath37 : @xmath53 where the @xmath54 are scalars and the reality of @xmath55 implies : @xmath56^ * = j_i(-\vec k)\ ] ] since the action is quadratic , @xmath37 is given by the corresponding determinant which is just the product of the two - dimensional determinants corresponding to the various value of @xmath57 . hence @xmath58 @xmath59 where the area - time @xmath60 came from density of states factor [ 9 ] . it corresponds to having @xmath61 the usual quantization of continuous modes in a box of volume @xmath62 . note that the last integral in eq . ( 11 ) is well defined and convergent . to obtain it , we discarded the infinite @xmath63 term which does not depend on @xmath1 or the angle @xmath17 between the directions of conductivity in the two plates and hence does not contribute to casimir forces / torque [ 9 ] . identifying @xmath64 we get finally : @xmath65 or using integration by parts ( with respect to @xmath66 after separating @xmath67 ) : @xmath68 where in the last step we multiplied numerator and denominator by @xmath69 and changed variables @xmath70 . we next find an explicit expression for @xmath71 . to this end , let us denote by @xmath72 the two unit vectors in the planes of plates and perpendicular to the direction of conduction in the first and second plate respectively . the direction @xmath73 is then determined by the conditions @xmath74 . hence @xmath75 choosing @xmath76 and using polar coordinates decomposition of @xmath66 : @xmath77 we find @xmath78 ^ 2 } { ( 1-\sin^2\theta\cos^2\varphi)(1-\sin^2\theta\cos^2(\varphi-\beta))}\ ] ] for @xmath79 the integral can be calculated analytically @xmath80 in this case @xmath81 and eq . ( 14 ) yields @xmath82 this is exactly half the usual casimir energy , as expected since only one polarization contributes providing a nice check of the calculation . for general @xmath17 numerical integration of eq . ( 15 ) yields @xmath83 as plotted in fig . 1 . 1.0 cm in addition to the @xmath17 dependent @xmath84 , @xmath85 is a casimir torque tending to align the plates so that @xmath79 , i.e. parallel conductivity directions and minimal energy are achieved . since torques are easier to measure , @xmath86 could perhaps be tested to a better accuracy than @xmath87 or @xmath88 though extreme , global , flatness will be required to avoid friction due to microscopic roughness , as the circular plates rotate relative to each other . we note that for non - circular plates , or when we have broad conducting stripes , another aligning torque results from the preference to have maximal adjacent plate area . clearly , this trivial geometrical effect is not the subject of interest here . the following comments are in order : \(i ) the mode sum / integral in eq . ( 2 ) involved in the usual derivation is very different from that in the present derivation . in particular , beyond our discarding of the @xmath89 independent infinite part in the last step of eq . ( 11 ) , no regularization is required here . also extending the standard method to derive our result for @xmath90 , seems rather difficult . \(ii ) the only relevant distance in this problem is the space - like plate separation involved here . this problem as well as any other static casimir calculation can therefore be done directly by the euclidean path integral with @xmath91 and no wick rotation is needed . \(iii ) the present method was applied also to other geometries ( spherical , cylindrical , etc . ) the known casimir energies were retrieved though not with extra ease or rigor . \(iv ) for the case of two magnetic / electric polarizable objects much smaller than their separation the present formalism readily yields the usual casimir polder result for the potential between polarizable atoms [ 10 ] . \(v ) finite temperature can be easily incorporated into the present approach , by replacing the @xmath92 integration in eq . ( 14 ) by a discrete sum . we can also generalize to the case of @xmath93 dimensions , by replacing @xmath94 there by @xmath95 . \(vi ) the present approach can be readily generalized to time dependent boundary condition . as noted also by golestanian and kardar ( see ref . 7 above ) , this can result in casimir radiation " . in particular we find that the lowest order ( two - photon amplitude is simply given by @xmath96\ ] ] with @xmath97 the polarizability of a neutral and small ( relative to @xmath98 ) object which moves along a trajectory @xmath99 with @xmath100 the proper time . all of the points ( iii)-(iv ) will be elaborated in a longer paper by o. kenneth . returning to the main theme of this letter , we note that the naive argument of ref . ( 6 ) that @xmath101 should vanish [ due to the fact that for two arrays of orthogonal wires , the @xmath102 polarized modes are free to escape from the left , say , and the @xmath103 polarized modes from the right ] fails . as indicated in fig . 1 , this is definitely not the case . indeed one simple approach views the ordinary attractive casimir effect as the attraction between patches of charges formed on one plate by charge separation due to quantum fluctuations and patches of opposite sign , image " , charges induced on the other plate . clearly this mechanism can operate , albeit with reduced strength , even if the two plates are made of conducting stripes pointing in orthogonal directions . this brings us to the final subject that we would like to mention , namely , an analog interpretation of magnetic casimir attraction between two conducting rings . consider then two parallel conducting rings of size @xmath1 and at a distance @xmath1 apart . the magnetic vacuum fluctuations include closed b field lines which link both rings . these will induce , by faraday s law , parallel currents in the two rings . thus , regardless of the sign of the b fluctuation and of the ensuing circulating current , the resulting current current forces will be attractive [ 12 ] . the fluctuations of interest are of scale @xmath104 when the above current - current forces on the various segments of the rings add coherently corresponding to the net current flow in the rings @xmath105 . if there is no net global flux change in the rings due to the vacuum fluctuation there will be - in this approximation - no net current and no net force . how will this force be modified if the rings become superconducting ? ideally , the superconducting rings impose a new integral constraint , namely that the total fluxes threading the various superconducting rings must be integer multiples of the flux quantum : @xmath106 . this implies however a strong exponential suppression . thus if we have a fluctuation with roughly constant @xmath107 on scale @xmath1 : @xmath108 the action of such a configuration will therefore be : @xmath109 the exponential suppression @xmath110 \approx \exp\left[-\frac{n^{2}}{\alpha_{em}}\right]$ ] renders such fluctuation and the attendant magnetic casimir forces completely negligible . the above considerations suggest that if the casimir force between conducting rings is constantly monitored as the temperature of the system is lowered below the superconducting critical temperature , then the quenching of part of the magnetic casimir force reduces the observed effect . hopefully , this amusing effect can eventually be observed , but we will not elaborate here on the conditions necessary for this .

Subsets and Splits

Data Mining Abstracts

This query searches for and retrieves 2000 documents containing the phrase 'data mining' in the abstract, providing limited but specific document samples.