Datasets:

Hendrik-a

/

arxiv

like 0

we study the hypothesis that observed x - ray / extreme ultraviolet emission from coronae of magnetically active stars is entirely ( or to a large part ) due to the superposition of flares , using an analytic approach to determine the amplitude distribution of flares in light curves . the flare - heating hypothesis is motivated by time series that show continuous variability suggesting the presence of a large number of superimposed flares with similar rise and decay time scales . we rigorously relate the amplitude distribution of stellar flares to the observed histograms of binned counts and photon waiting times , under the assumption that the flares occur at random and have similar shapes . our main results are : ( 1 ) the characteristic function ( fourier transform of the probability density ) of the expected counts in time bins is = \}$ ] , where is the mean flaring interval , is the characteristic function of the flare amplitudes , and is the flare shape convolved with the observational time bin . ( 2 ) the probability of finding counts in time bins is = . ( 3 ) the probability density of photon waiting times is = with = the mean count rate . an additive independent background is readily included . applying these results to euve / ds observations of the flaring star ad leo , we find that the flare amplitude distribution can be represented by a truncated power law with a power law index of 2.3 0.1 . our analytical results agree with existing monte carlo results of and . the method is applicable to a wide range of further stochastically bursting astrophysical sources such as cataclysmic variables , gamma ray burst substructures , x - ray binaries , and spatially resolved observations of solar flares .

astro-ph0312405

since the late 1930s the solar corona has been known to be much hotter than the photosphere , although the latter delineates the surface through which energy enters the corona . the corona is thus not in a state of simple heat flow equilibrium , and its high temperature must be supplied with some form of mechanical , electrical or radiative work which is dissipated at coronal heights . several heating mechanisms have been proposed , such as alfvn @xcite or ion cyclotron @xcite waves , or electric currents @xcite and magnetic reconnection @xcite converting magnetic shear into jets , and , ultimately , into heat . some of these processes operate in a continuous mode , whereas others invoke transients . comprehensive reviews may be found , e.g. , in @xcite , @xcite , @xcite , @xcite , and @xcite . with increasing sensitivity of the observations it was recognized that the ` quiet ' solar corona contains in fact ` microflares ' and ` nanoflares ' down to the actual resolution limit . clearly , these events are the signature of non - equilibrium processes , and therefore represent promising candidates for the primary heating mechanism of the solar corona @xcite . stars at high magnetic activity levels show properties that are reminiscent of continuous flaring processes , such as persistent coronal temperatures of 530 mk , high coronal electron densities , or continuous non - thermal radio emission . see @xcite for a summary of observational results . foremostly , x - ray and extreme ultraviolet ( euv ) time series of such stars reveal a high level of flux variability on time scales similar to flares @xcite . limited detector sensitivity could give the wrong impression that the emission is due to steadily emitting coronal sources , while new , highly sensitive observations in at least some cases show little indication of a base level emission that could be defined as being genuinely ` quiescent ' @xcite . a coronal heating mechanism that should successfully explain x - ray and euv emission from magnetically active stars inevitably has to address the question of continuous variability . in the context of microflare heating one is interested in the frequency distribution of flare amplitudes . in particular , it is essential to have reliable estimators of the small but frequent flares , since these presumably carry a major part of the total energy . however , these small flares substantially overlap , so that their amplitudes are hard to determine from the noisy observations . one way out is to use flare identification algorithms that also include flare superposition @xcite . another way out is to resort to a forward model , which constrains the flare amplitude distribution by model assumptions . within a monte carlo framework , this approach has been followed by @xcite and in @xcite . in the present article we explore _ analytical _ forward methods . we describe a statistical model of superimposed flares which provides an exact relation between the ( univariate ) probability density of the flare amplitudes and the ( univariate ) probability densities of the observed binned count rates and photon waiting times . as the method involves the histograms of binned count rates and photon waiting times only , it is insensitive to permutation of the time bins and to the presence of long data gaps . the price for this robustness is the loss of temporal information , which is replaced by the assumption that all flares have similar shape and occur at random . see @xcite and @xcite for a complementary approach focusing on the temporal autocorrelation of the light curves ( and thus on the first two moments of the univariate count rate distribution ) . in what follows we basically work with the observables ` counts ' and ` count rates ' , with emphasis on the effect of flare superposition and counting noise . the conversion into luminosity and flare energy is briefly discussed in section [ discussion_sect ] . while the energy release and relaxation processes are not known in detail , there is observational support for the similarity of flares on a global scale at least for ` simple ' flares . stellar flares of the type investigated in this paper can usually well be represented by a sharp rise and an exponential decay . examples supporting this view are found in @xcite , @xcite , @xcite , and @xcite . the observed durations of flares , as investigated in the solar case , are almost independent of the flare amplitude or radiated energy even if 3 - 5 observed orders of magnitude are included . ( @xcite , see also summary in @xcite ) . the decay times of flares studied below were found to scatter around a constant value in @xcite , independent of peak flux . since in stellar observations 3 orders of magnitude of flare energy are usually sufficient to explain the observed emission , we shall assume that a constant decay time is a good approximation for our case as well . for completeness , the case of a weak dependence of the decay time scale @xmath15 on radiated total energy @xmath16 , @xmath17 as suggested by one stellar investigation , was studied in both @xcite and @xcite , with the result that the main conclusions on the importance of small flares were reinforced . observations also suggest that individual flares occur at random and independently of each other ( for the solar case , see @xcite ) . this independence might suggest that flares trigger an exhaustive ` reset ' where mechanical stress ( and thus memory ) is completely destroyed . we shall thus adopt the following model for the instantaneous count rate in a fixed energy band : @xmath18 } \label{model}\ ] ] where @xmath19 represents the ( dimensionless ) flare shape , @xmath20 [ ct s@xmath21 are the flare amplitudes , and @xmath22 [ ct s@xmath21 is a background intensity . we shall agree that @xmath23 , and define @xmath24 to be the flare duration . all flare amplitudes @xmath20 are positive and drawn from the same probability density @xmath25 . ( from here on , @xmath9 denotes a generic real positive variable . ) the flare times @xmath26 occur at random with mean flaring interval @xmath3 [ s ] , and it is assumed that all @xmath20 , @xmath26 , and the background are statistically independent of each other . the observed counts are supposed to be a poisson process with time - varying intensity @xmath27 . such a non - uniform poisson process is not unambiguously defined , and we use here the construction shown in fig . [ poissonprocess_fig ] : the photon arrival times are distributed like the @xmath28-components of uniformly distributed points @xmath29 in the @xmath30-plane satisfying @xmath31 . note that the background is included in @xmath27 . when binned in time bins of duration @xmath0 , the counts are poisson distributed with parameter ( = expected number of counts in @xmath0 ) @xmath32 alternatively , we may work with the photon event list . in this case we consider the time intervals between consecutive photon arrivals ( ` photon waiting times ' ) . dimensional analysis shows that the flare - related part of the signal ( [ model ] ) is governed by the single dimensionless parameter @xmath33 . the conversion into discrete time bins involves a second dimensionless parameter , @xmath34 ( or , alternatively , @xmath35 ) . if the peaks are well - separated ( @xmath36 ) and counting noise is low , one can directly estimate the flare amplitudes from the observed peaks in the light curve . if flares overlap ( @xmath37 ) or counting noise is relevant , this is no longer possible . we resort then to the model ( [ model ] ) and ask for its predictions on the observed probability densities of binned counts @xmath7 and photon waiting times @xmath10 , which are available with high accuracy under stationary conditions . the goal is thus to establish the relation between @xmath25 on the one hand , and @xmath7 and @xmath10 on the other hand . the linearity of equation ( [ model ] ) , together with the statistical independence of its ingredients , suggests that a rigorous mapping from @xmath25 to @xmath7 and @xmath10 should exist . this is in fact the case , and the individual steps of this mapping are discussed in sections 3.1 - 3.4 . see table 1 for an overview on our notation . .notation@xmath38 [ cols="^,<,^ " , ] @xmath38the ( poisson- ) ` intensity ' is the instantaneous photon count rate ; the ( poisson- ) ` parameter ' is the expected number of photons in time bins @xmath0 . ` total ' means the sum of background and flares . we first ask for the probability that the flare signal @xmath39 assumes values between @xmath9 and @xmath40 . by statistical homogeneity , all times are equivalent . consider thus the time @xmath41 , and let it be centered in an auxiliary interval @xmath42 of length @xmath43 . neglecting flares outside @xmath42 as well as the background , the intensity at @xmath28 = 0 is @xmath44 . since @xmath45 is a sum of independent random variates , it is conveniently @xcite represented in a way which diagonalizes convolution , such as by its laplace- or fourier transform . we use here the latter , called the characteristic function @xcite and defined by @xmath46 where @xmath47 is the probability that @xmath45 is between @xmath9 and @xmath48 . the probability density is obtained from the characteristic function by fourier back transform , @xmath49 . we continue now with the calculation of the characteristic function @xmath50 . the essential point to notice is that @xmath50 can be partitioned according to the number of flares which occur in the interval @xmath42 . one can thus write @xmath51 , where @xmath52 is the probability of finding exactly @xmath53 flares in @xmath42 , and @xmath54 is the corresponding photon intensity from these @xmath53 superimposed flares in @xmath42 . it is understood that @xmath55 , because the case @xmath53 = 0 flare does not contribute to the count rate . since the times @xmath26 are independent and uniformly distributed in @xmath42 , the characteristic function of @xmath56 is @xmath57 the step from equation ( [ phi_q1 ] ) to equation ( [ phi_q2 ] ) also invokes the fact that all @xmath20 are independently @xmath58-distributed , and the definition of the characteristic function of the flare amplitudes , @xmath4 = @xmath59 . the step from equation ( [ phi_q2 ] ) to equation ( [ phi_q3 ] ) makes use of the symmetry of @xmath42 about @xmath28 = 0 . performing the sum over all possible numbers of flares and using equation ( [ phi_q3 ] ) , one finds that @xmath60 \big\ } \label{phi_ft } \ , .\ ] ] we now want to liberate ourselves from the artificial time interval @xmath42 , by extending it to infinity and thereby collecting all flare contributions . to this end we have to investigate whether the integral in equation ( [ phi_ft ] ) remains finite as @xmath61 . this turns out to be the case under rather weak conditions , one of which is as follows ( see appendix [ limit_appendix ] ) : if for some @xmath62 the flare shape @xmath63 decays faster than @xmath64 as @xmath65 , and if the probability density of the flare amplitudes decays no slower than @xmath66 as @xmath67 , then equation ( [ phi_ft ] ) remains well - behaved as @xmath61 . note that the above condition does not require the expectation value of the flare amplitudes to be finite ; this freedom will allow us to consider every power law index @xmath68 if the flare amplitudes have a power law distribution truncated at some lower cutoff . ( @xmath68 must be required to ensure that @xmath25 can be normalized . ) letting thus @xmath42 go to infinity , equation ( [ phi_ft ] ) becomes @xmath69 \big\ } \ , . \label{phi_f}\ ] ] the argumentation leading to equation ( [ phi_f ] ) can be repeated , with minor modifications , for the poisson parameter ( expected counts ) of time bins of duration @xmath0 . neglecting the background , the poisson parameter of time bin @xmath70 $ ] is @xmath71 , so that we merely have to replace @xmath19 by @xmath72 in order to obtain the characteristic function of @xmath73 , @xmath74 \big\ } \ , . \label{phi_f}\ ] ] equations ( [ phi_f],[phi_f ] ) are our central result ; they establish the connection between the probability density of flare amplitudes @xmath58 via @xmath75 and the distributions of photon intensities and expected counts , @xmath76 , @xmath77 via @xmath78 , @xmath79 . equation ( [ phi_f ] ) holds under the same conditions as equation ( [ phi_f ] ) . the function @xmath80 is a blurred version of the flare shape with resolution degraded to the observational time bin . it is easily seen that probabilistic normalization is preserved ( @xmath81=1 and @xmath82=1 if @xmath83=1 ) . it can also be shown that non - negative definiteness @xcite of @xmath4 is inherited under the transformations ( [ phi_f ] ) and ( [ phi_f ] ) , so that the @xcite theorem ensures that the fourier inverses of @xmath50 and @xmath1 represent , in fact , probability densities . the transformation @xmath84 redistributes amplitudes @xmath9 both towards higher ( flare overlap ) and lower ( flare tails ) values . it also broadens the probability density function by finite time binning , unless @xmath85 , in which case @xmath86 and @xmath87 . in the limit @xmath88 and if @xmath89 , normal behavior is approached , with @xmath90 concentrated about @xmath91 with relative spread of order @xmath92 . this expected result may be proven by formal expansion of equation ( [ phi_f ] ) about @xmath93 . so far the background has been neglected . when included , the total ( flare plus background ) photon poisson intensity at time @xmath28 = 0 is @xmath94 , and the total poisson parameter of the time bin @xmath95 $ ] is @xmath96 with @xmath97 . since the background is independent of the flares , the characteristic functions of the total photon poisson intensity and -parameter are @xmath98 with @xmath99 and @xmath100 the characteristic functions of @xmath101 and @xmath102 , respectively . if the background is constant then @xmath103 . the counts in the time bin [ @xmath104 are poisson distributed with parameter @xmath105 , where @xmath105 itself is @xmath106-distributed . the binned counts have thus the characteristic function @xmath107 which is @xmath108-periodic because the binned counts are integers . the probability density of the binned counts is obtained from fourier back transform of @xmath109 ; when expressed in terms of the original probability density @xmath25 , the probability of finding @xmath6 counts in bins of duration @xmath0 is @xmath110 if a gaussian approximation was used , with the binned counts modeled by @xmath111 with @xmath112 standard normal , then @xmath113 could assume any real number and the characteristic function is @xmath114 . this approximation becomes exact at @xmath115 , corresponding to large count rates . instead of time - binning the observed counts one may directly work with the photon event list , and consider the histogram of waiting times @xmath116 between consecutive photon arrivals . this has the advantage that no ad - hoc choice of time bins is needed , and that the waiting time bins can be adapted for optimum ( similar ) probability content . in addition , long data gaps ( @xmath117 ) and detector saturation ( @xmath118 ) are easily segregated , and the construction of the photon waiting time histogram amounts to a direct determination of the moment generating function of @xmath27 , which is ( almost ) the quantity provided by theory . the photon waiting times represent therefore a natural interface between theory and observation . to see why this is so , consider the ` waiting ' event that there is ( i ) one photon between @xmath119 and @xmath120 , ( ii ) no photon between @xmath120 and @xmath121 , and ( iii ) one photon between @xmath121 and @xmath122 ( the waiting time is thus @xmath123 ) . by the binomial theorem , the probabilities of ( i ) and ( iii ) are @xmath124 and @xmath125 , while from figure [ poissonprocess_fig ] we see that the probability of ( ii ) is @xmath126 . since ( i ) , ( ii ) and ( iii ) are independent of each other , the waiting event occurs with probability density @xmath127 . denoting the average over realizations of @xmath27 by angular brackets , and using the translation invariance @xmath128 , one finds that the probability density of photon waiting times @xmath129 is @xmath130 where the normalization constant @xmath131 is determined by the constraint @xmath132 = 1 , implying . ] that @xmath133 . while equation ( [ p_d ] ) is exact , it is generally difficult to calculate @xmath134 for arbitrary @xmath9 . this is , however , not needed because the photon waiting time is much shorter than the flare duration in any practically relevant case . one may thus resort to the approximation @xmath135 and @xmath136 the cumulative waiting time distribution is , accordingly , @xmath137 if the background is time - varying on time scale @xmath138 , the conditions @xmath139 and @xmath140 must be added to the right hand side of equations ( [ p_d0 ] , [ p_d1 ] ) . the range of validity of equation ( [ p_d0 ] ) was explored analytically ( appendix [ explicit_appendix ] ) and by monte carlo simulations ( section [ simulations ] ) . it was found to hold even if the assumptions on the right hand side are not strictly valid . the result ( [ p_d0 ] ) can also be derived in the following , perhaps more intuitive , way : let @xmath141 denote the probability that @xmath27 is between @xmath142 and @xmath143 in a random time interval of fixed duration . then , @xmath144 is the probability that a given photon has been emitted at poisson parameter @xmath142 . ( this probability is proportional to the total area of strips in the @xmath29 plane with @xmath145 see fig . [ poissonprocess_fig ] ) . assume now that @xmath27 varies only weakly during the local waiting time @xmath146 . thus @xmath144 is also the probability that a waiting time ` belongs ' to intensity @xmath142 . given the intensity @xmath142 , the photon waiting time has probability density @xmath147 , so that the ( marginal ) probability density of photon waiting times @xmath9 is @xmath148 , which is equation ( [ p_d0 ] ) . for special flare shapes and amplitude distributions , equations ( [ phi_f ] ) - ( [ p_d1 ] ) can be evaluated rigorously . this section summarizes results which are relevant for application to observational data ( section [ obs_sect ] below ) . exponential decay is a generic feature of many relaxation processes , and is a natural model of flares . observations of late - type main sequence stars often reveal a rather abrupt onset followed by an approximately exponential decay ( for stellar examples , see @xcite ) , with decay time almost independent of flare size ( e.g. , @xcite ; see also section ii ) . we shall thus use an exponential flare shape with constant decay time @xmath15 as a proxy for exact calculations . if the flare shape is @xmath149 with @xmath150 the heaviside step function , then @xmath151 ( see appendix [ explicit_appendix ] ) . the abrupt and artificial onset of the one - sided exponential flare shape may be softened by an exponential increase , thus creating a two - sided exponential shape : @xmath152 . since flare contributions from @xmath153 and @xmath154 to @xmath155 are statistically independent of each other , the characteristic function of their combined flare photon intensity is @xmath156 where @xmath157 is given by equation ( [ phi_f1 ] ) with @xmath15 replaced by @xmath158 . equations ( [ phi_f1 ] ) and ( [ phi_f_pm ] ) show that two - sided exponential flares have the same characteristic function @xmath78 as one - sided exponential flares with effective decay time @xmath159 . a similar result holds for @xmath79 if @xmath85 . from the observational point of view , this implies that one- and two - sided exponential flares can not be distinguished based on their short - term photon waiting time ( or binned count ) distribution . we therefore assume in the sequel that the flare shape is one - sided exponential , with the understanding that our results on @xmath25 are robust against the generalization of a one - sided to two - sided exponential shapes . for several forms of @xmath25 the characteristic function @xmath50 can be explicitly obtained . solvable cases include exponentials , power laws , and lvy distributions . details are given in appendix [ explicit_appendix ] ; here we cite the results for power laws , which are a traditional and observationally well confirmed model of the distributions of solar @xcite and stellar @xcite flares . if @xmath25 is a power law with exponent @xmath160 , truncated at a lower cutoff @xmath161 , @xmath162 and if the flare shape is exponential , then @xmath50 can be exactly expressed in terms of special functions ( appendix [ explicit_appendix ] ) . for practical purposes , these may be approximated by elementary functions . of particular interest is the cumulative photon waiting time distribution ( eq . [ p_d1 ] ) which predicts the probability content of arbitrary photon waiting time bins : @xmath163 with @xmath164^{-1}$ ] . the relative accuracy of the approximation ( [ pd_approx ] ) was found to be better than 0.001 if @xmath165 and @xmath160 @xmath166 2.01 . for @xmath160 @xmath166 2 the expectation value of the flare amplitudes exists and is given by @xmath167 . in the limit @xmath160 @xmath168 2 the right hand side of equation ( [ pd_approx ] ) converges to the unit step function @xmath169 ; this is also true if the exact equations ( [ phi_f_pl1 ] , [ phi_f_pl2 ] ) are used . the limit @xmath170 @xmath168 @xmath171 therefore predicts , theoretically , arbitrarily small photon waiting times . before application to real data , the results of section [ gensolution ] have been verified by several independent methods . the introduction of the characteristic function in section [ flare_contrib_sect ] may appear somewhat arbitrary , and the question arises whether a more direct approach exists . the answer is yes , but the calculation is tedious and difficult to make rigorous . we shall therefore only sketch it in appendix [ moment_appendix ] . here , we discuss validation based on an entirely independent analytical argument ( section [ peak_sect ] ) and on numerical simulations ( section [ simulations ] ) . for one - sided exponential flares and only for these the signal @xmath172 is a continuous - time markov process whose stationary measure can be obtained from a renewal argument . this provides an independent verification of equation ( [ phi_f1 ] ) , and has some interest on its own as it allows to relax the poisson assumption on the flare times ( see sect . [ discussion_sect ] below).the argument is as follows . discarding the background , the amplitudes at consecutive peaks are related by @xmath173 with @xmath174 and @xmath175 the flare waiting times . now , in the stationary case , the probability densities of @xmath176 and @xmath177 must agree . therefore their characteristic function @xmath178 must satisfy @xmath179 where @xmath180 is the flare waiting time distribution , and we have used the definition of the characteristic function and the fact that @xmath181 , @xmath176 and @xmath182 are statistically independent of each other . for uniformly distributed flares , the flare waiting times are exponentially distributed , @xmath183 . upon setting @xmath184 = @xmath185 , equation ( [ phi_eq ] ) simplifies thus to @xmath186 equation ( [ int_eq ] ) can be solved for @xmath187 by taking derivatives with respect to @xmath188 , yielding @xmath189 together with the initial ( normalization ) condition @xmath190 , equation ( [ diff_eq ] ) has the unique solution @xmath191 the exponential factor on the right hand side of equation ( [ phi_r_n ] ) acts as a low - pass filter , which broadens @xmath192 with respect to @xmath25 by an amount proportional to @xmath33 . as it must be , isolated flares reproduce the true flare amplitudes , @xmath193 . there is an apparent similarity between equations ( [ phi_f1 ] ) and ( [ phi_r_n ] ) , and the exact relation is as follows . by construction , the peaks @xmath176 are the sum of two independent random variates : the value of the continuous signal just before the jump , @xmath194 , plus the flare amplitude @xmath181 . from equation ( [ phi_r_n ] ) we see that @xmath194 has the characteristic function @xmath195 . since @xmath194 occurs with equal probability than any @xmath27 , the above expression must agree with the result of equation ( [ phi_f1 ] ) . this is in fact the case . equations ( [ phi_f ] , [ phi_f ] , [ p_c ] , [ p_d ] ) have also been verified by monte carlo simulations . in these simulations , a realization of @xmath27 is generated , and the binned poisson parameters @xmath196 = @xmath197 are numerically calculated . the binned counts are then simulated using a conventional poisson number generator , and their histogram is created . for the photon waiting times we proceed as indicated in figure [ poissonprocess_fig ] : a large number @xmath198 of points @xmath29 are uniformly distributed in the rectangle @xmath199 \times [ 0,y]$ ] , with @xmath200 and @xmath201 . the @xmath28-coordinates of points with @xmath31 constitute our photon list . note that the background is included in @xmath27 . the simulations are compared to numerical evaluations of eqns . ( [ p_c],[p_d],[p_d0 ] ) and to exact results ( appendix [ explicit_appendix ] ) . figure [ binned_counts_fig ] ( left column ) shows the theoretical ( gray ) and simulated ( black ) probability densities which occur in the transformation from the flare amplitudes into the binned counts . the flare amplitudes are uniformly distributed between 1 and 3 ct @xmath202 , the flare profile is a one - sided exponential with decay time @xmath15 = 8 s , the mean flaring interval is @xmath3 = 10 s , and the simulation involves @xmath203 time bins of duration @xmath0 = 3 s. these parameters represent marginally resolved , partially overlapping flares . accordingly , @xmath204 ( fig . [ binned_counts_fig ] line 2 left ) is smoothed with respect to @xmath25 ( line 1 left ) but has also a peak at @xmath9=0 which is due to low - count sequences between different flares . the binned background @xmath102 is modeled as chi square distributed with 3 degrees of freedom ( fig . [ binned_counts_fig ] line 3 left ) ; this ad - hoc choice yields a particularly simple characteristic function and is for illustration purposes only . the conversion of the binned poisson parameter ( fig . [ binned_counts_fig ] line 3 right ) into discrete counts renders the characteristic function @xmath108-periodic ( fig . [ binned_counts_fig ] line 4 right ) . for benchmarking purposes , the theoretical probability densities are obtained from numerical fourier inversion of the associated characteristic functions ( fig . [ binned_counts_fig ] right column , solid line - real part , dashed line - imaginary part ) . the transformation from the flare amplitudes into the photon waiting times is illustrated in figure [ waitingtime_fig ] . the flare amplitudes are exponentially distributed ( mean amplitude @xmath205 = 0.15 ct @xmath202 , mean flaring interval @xmath3 = 20 s ) and the flare shape is one - sided exponential ( @xmath15 = 40 s ) . the exponential flare amplitudes are chosen because they admit simple analytical results for @xmath10 ( see appendix [ explicit_appendix ] ) , which can be compared to the simulation . a constant background @xmath206 = 0.1 ct @xmath202 is present , and the simulated photon list includes @xmath207 photons . the theoretical @xmath208 ( fig . [ waitingtime_fig ] middle left gray ) is a chi square distribution with @xmath33 degrees of freedom ( eq . [ phi_f_exp ] ) , and @xmath10 ( fig . [ waitingtime_fig ] bottom left gray ) is given by the exact equation ( [ p_d_exp_bg ] ) . the characteristic functions of the flare amplitudes , flare photon intensity , and total ( flare plus background ) photon intensity are shown in the right column of figure [ waitingtime_fig ] . solid line denotes the real part , and dashed line denotes the imaginary part . a broad probability density corresponds to narrow characteristic functions , and vice versa . the constant background amounts to a modulation by @xmath209 . both binned counts and photon waiting times have also been simulated for power law flare amplitudes , which are used as a fitting template for the euve / ds observations below . figure [ sim_pl_fig ] shows the result of a simulation of one - sided exponential flares ( @xmath15 = 3000 s , @xmath3 = 333 s ) with lower cutoff @xmath210 = 0.007 ct @xmath202 and power law index @xmath160 = 2.36 . the simulation involves 3@xmath211 photons in 10@xmath212 time bins of duration @xmath0 = 100 s. again , the black line represents simulation , and the gray line represents theory , with @xmath10 being the short - term form ( eq . [ p_d0 ] ) computed by differentiation of the exact expression ( [ phi_f_pl1 ] ) for @xmath50 . a constant background @xmath206 = 0.09 ct @xmath202 is present , and an upper cutoff @xmath213 = 7 ct @xmath202 has been applied in the simulations and when evaluating @xmath7 from equation ( [ p_c ] ) . we have applied our theoretical results to a 3700 ks ( elapsed time ) observation of the active star ad leo acquired by the extreme ultraviolet explorer ( euve , e.g. @xcite ) deep survey ( ds ) instrument between april 2 and may 15 1999 . this time series has previously been analyzed by @xcite and in @xcite by monte - carlo simulations of the model ( eq . [ model ] ) , so that we can compare our findings with existing results . the deep survey instrument records photons between 52 and 246 with maximum sensitivity at 91 . figure [ euve_lc_fig ] shows an overview of the light curve , time - integrated over spacecraft orbits ( 5663 s each ) . during one spacecraft orbit , ad leo was in the euve field of view for typically 1500 s , followed by a @xmath214 4200 s interval without observations . furthermore , there are observation interrupts where the light curve of figure [ euve_lc_fig ] drops to zero . the period from 05/06 16:42 to 05/16 05:11 carries large contamination from energetic particles and is excluded from our analysis . figure [ euve_pd_fig ] shows the histograms of photon waiting times . the ` flares+background ' are collected across a spatial window centered at ad leo , while the ` background ' is determined from a larger disjoint area . the resolution of our photon arrival times is 0.1 s. this is not the full euve / ds time resolution , but has been considered sufficient with regard to the actual count rates . only photon waiting times from 0.25 s and 30 s are taken into account . the lower limit eliminates poor flare statistics and telemetry saturation at rare and extremely bright flares , and the upper limit eliminates low count rates and artifacts from ( much longer ) unobserved intervals . our selection comprises 118,637 photon waiting times , and the photon waiting time bin size is 0.1 s. the background waiting times ( fig . [ euve_pd_fig ] ` background ' black line ) are found to closely follow an exponential distribution ( fig . [ euve_pd_fig ] ` background ' gray line ) ; accordingly , we model the background as constant . when scaled to the ` flares+background ' collecting area , the best - fit background intensity is ( cf . @xcite ) @xmath215 once the background is known , we determine the flare amplitude distribution @xmath25 under the assumption that @xmath25 is a power law with exponent @xmath160 and lower cutoff @xmath210 ( eq . [ p_a_pl ] ) , and that the flare shape is one - sided exponential ( fig . [ euve_lc_fig ] inlet ) . these assumptions correspond up to a little relevant upper cutoff to those made by @xcite and in @xcite . we also assert that the short - term forms ( eqns . [ p_d0 ] , [ p_d1 ] ) of the photon waiting time distribution apply . our model parameters are thus ( @xmath216 ) . they are determined by minimizing @xmath217 where @xmath218 ( @xmath219 ) is the observed ( predicted ) number of waiting times in the @xmath6-th waiting time bin . the prediction is given by @xmath220 , and we use the exact expression ( [ phi_f_pl1 ] ) when computing the cumulative photon waiting time distribution ( eq . [ p_d1 ] ) . the sum in equation ( [ chi2 ] ) runs over @xmath221 = @xmath222 = 298 waiting time bins , and there are @xmath223 = @xmath221 @xmath224 3 = 295 degrees of freedom . the minimum reduced chi square is 0.989 , and the best - fit parameters are @xmath225 the resulting theoretical waiting time distribution is shown in figure [ euve_pd_fig ] ( ` flares+background ' gray line ) together with the observation ( ` flares+background ' black line ) . the errors of ( @xmath226 ) are not independent of each other , and the values given in equation ( [ results_d ] ) are projections of the 90% confidence domain ( see fig . [ euve_chi2_fig ] below ) . we also estimated the parameters ( @xmath226 ) from the observed histogram of binned counts ( fig . [ euve_pc_fig ] black line ) . the ( disjoint , adjacent ) time bins have duration @xmath0 = 100 s and contain , on average , 22 counts . time bins which would intersect the transition from or to an unobserved time interval are rejected . in order to abate the arbitrariness of choosing a binning offset , the starting time of the first time bin has been varied at random within @xmath0 , and the histogram of figure [ euve_pc_fig ] is an average over 50 random offsets . time bins with less than 2 counts and more than 80 counts are excluded , because they are affected by the binning offset and have poor statistics . the best - fit solution is again found by minimizing the chi square expression of equation ( [ chi2 ] ) , where @xmath218 and @xmath219 are now the observed and theoretical frequencies of @xmath6 counts in time bins of duration @xmath0 . the theoretical prediction is @xmath227 with @xmath131 = @xmath228 = 4959 and @xmath6 running over @xmath221 = @xmath229 = 78 photon number bins . since @xmath0 @xmath230 @xmath15 , we may use the short - term form of the probability density of binned counts , @xmath231 with @xmath232 and @xmath50 given by equation ( [ phi_f_pl1 ] ) . the minimum reduced chi square ( per @xmath223 = @xmath221 @xmath224 3 = 75 degrees of freedom ) is 1.003 , which is attained at @xmath233 the best - fit binned count distribution is superimposed in figure [ euve_pc_fig ] ( gray line ) . the errors given in equation ( [ results_c ] ) are again projections of the 90% confidence domain . the confidence domains of ( @xmath226 ) for both the photon waiting time and binned count methods are visualized in figure [ euve_chi2_fig ] . the contours represent chi square sections through the best fit solution ( crosses ) at levels @xmath234 , corresponding to ( 0.75 , 0.9 , 0.999 ) confidence levels ) was true then the probability that the observed @xmath235 ( eq . [ chi2 ] ) deviates from its expectation value @xmath223 by more than @xmath236 is about @xmath237 . the cited ` confidence levels ' represent the probabilities that this does not happen . ] . panels above the diagonal refer to the binned count method ; panels below the diagonal refer to the photon waiting time method . the formal 90% confidence level is marked boldface . we find that the relative error of @xmath160 is smaller than the errors of @xmath210 and @xmath33 for both the binned count and waiting time methods . the power law index @xmath160 can thus be estimated more reliably than the lower cutoff . the shape of the confidence domains indicates that the parameters are not independent of each other . the strongest interference is between the lower cutoff @xmath210 and the separation of individual flares , @xmath238 , which reflects the fact that the average flare photon flux , @xmath239 = @xmath240 , is sensitive to the product @xmath241 only . likewise , @xmath160 is positively coupled to both @xmath210 and @xmath33 because the function @xmath242 decreases with increasing @xmath160 ( @xmath160 @xmath166 2 ) . the results of the photon waiting time ( eq . [ results_d ] ) and binned count ( eq . [ results_c ] ) methods agree within the 90% errors , which is considered as satisfactory , although the binned count method systematically yields lower @xmath210 and larger @xmath33 than the photon waiting time method , whereas the product @xmath241 is similar in both cases ( 0.0541 for the waiting time method ; 0.0527 for the binned count method ) . the mean flaring interval @xmath3 and flare duration @xmath15 enter the short - term forms of binned count- and photon waiting time distributions only in the dimensionless combination @xmath33 . the decay constant @xmath15 of flares with amplitudes @xmath243 @xmath210 has been investigated by @xcite and in @xcite and found to be about 3000 s ; adopting this value , we find a mean flaring interval of @xmath244 ( average over both methods ) . the use of the short - term forms of the binned count- and photon waiting time distributions is justified , a posteriori , since @xmath0 @xmath230 @xmath15 , and since @xmath116 @xmath245 30 s @xmath230 @xmath15 and @xmath246 @xmath214 20 @xmath247 1 . in fact , direct numerical evaluations of the exact forms ( eqns . [ p_c ] , [ p_d ] ) at best - fit parameters ( eqns . [ results_d ] , [ results_c ] , [ flare_rate ] ) deviate from their short - term approximations by less than 2% . we have applied an analytical approach to the inversion of light curves to derive the amplitude distribution of constituent elementary flare light curves . the method uses either binned count light curves or the distribution of waiting times derived from photon lists . we have in particular considered flare shapes described by a one - sided exponential decay or a two - sided exponential profile , and we have studied flare amplitude distributions described by an exponential function and a truncated power law . the latter form has been favored in previous studies because solar flare statistics has revealed power - law distributions over a wide range in energy @xcite ; explicit stellar studies indicate power laws as well @xcite . we have thus used a power law to model euve / ds observations of ad leo , where we find a power law index of @xmath160 = 2.3 @xmath14 0.1 , a lower cutoff of @xmath210 = 0.004 @xmath14 0.002 ct @xmath202 , and that there are @xmath33 = 15 @xmath14 8 flares per flare duration above @xmath210 ; these values are compilations of the binned count- and photon waiting time results . our results agree with complementary numerical methods recently applied to similar data sets @xcite indicating power - law indices for the flare amplitude distribution of typically 2.0 2.4 , with most measurements concentrating around @xmath160 = 2.2 2.3 . while such values are necessary for power - law flare amplitude distributions to ` explain ' coronal heating on active stars , they are not sufficient . the emitted power observed from magnetically active coronae can only be due to the ensemble of flares _ if _ the power - law distribution extends to sufficiently low energies ; this conjecture is of course difficult to prove due to the strong superposition of near - simultaneous flares that can not be separated spatially , thus forming a pseudo - continuous emission level . when expressed in terms of flare energy , our numerical results ( eqns . [ results_d ] , [ results_c ] ) correspond to a flare peak luminosity and minimum flare energy of @xmath248 where it is assumed that one euve / ds count corresponds to @xmath249 erg emitted at ad leo ( see @xcite ) . alternatively , a fraction of the observed emission may be truly steady ( or variable on time scales much longer than flare time scales ) , but this level should then be below the lowest level seen in light curves between individual detected flares ; @xcite estimate that a mere 30% ( or less ) of the low - level emission of the dme ( binary ) star uv ceti could be ascribed to quasi - steady emission , the rest being recorded as variable on time scales of tens of minutes with shapes indicating continuous flaring . additional information contained in the differential emission measure distribution could help interpreting the physical processes participating in the heating of the coronal plasma @xcite . as mentioned in @xcite , a few words of caution are in order . a number of features can not be recognized in the data per se and would require additional information . specifically , the observations collect emission from across the visible hemisphere of the star , although heating mechanisms or , in our picture , flare amplitude distributions , may vary across various coronal structures ; further , some of the variability may be introduced by intrinsic changes in the large - scale magnetic field distributions ; emerging or disappearing active regions will certainly contribute to variability , as do changes in the visibility of individual regions on the star as they move into or out of sight due to the star s rotation . nevertheless , the light curve characteristics do not appear to be dominated by any of these longer - term effects , while variability on flare time scales is ubiquitous ; similar conclusions were drawn by @xcite , @xcite , @xcite , and @xcite . we therefore conclude that the present study further supports the view that stochastic flaring is of principal importance to coronae of magnetically active stars , and that coronae of such stars may significantly , if not entirely , be heated by flares . a practical benefit of the analytical method lies in its efficient assessment of the compatibility of model and observations . the computation of @xmath235 for a given set of parameters takes milliseconds on a medium - size work station , while a monte carlo simulation of comparable accuracy takes seconds or minutes due to slow ( @xmath250 ) convergence at rare large flares . the efficiency of the analytical approach allows to systematically explore the whole parameter space for the global chi square minimum . for instance , figure [ euve_chi2_fig ] has been obtained by a brute - force calculation of @xmath251 on a cubic lattice of dimension 90@xmath25295@xmath252100 , taking about one hour computation time , whereas a similar monte carlo simulation would take many weeks . the possibility for systematic exploration of the parameter space ( at low resolution ) is important because the chi square function has generally several local minima , so that traditional minimization routines such as conjugate gradients may fail . note that the simple contours of figure [ euve_chi2_fig ] depict only the local behavior around the global minimum . when comparing the binned counts- and photon waiting time methods it was found that the binned counts method has larger confidence domains than the photon waiting time method , but is also more robust in finding the global minimum . a good strategy is thus to start with the binned counts , using a low - resolution exploration of the parameter space . as soon as a the approximate global chi square minimum is found , this may be refined by the photon waiting time method . in any case the final solutions should be cross - checked for both methods , to ensure consistency and to obtain an error estimate for the parameters which goes beyond the errors of the separate methods . we would finally stress that our analytical method is in no way restricted to the particular time series studied in this paper . the method applies whenever the process can be described in the form of equation ( [ model ] ) , which may include models for so divers phenomena as cataclysmic variables , gamma ray burst substructures , and x ray binaries . there is also no need to restrict the process ( [ model ] ) a the real line ( see below ) . a drawback of the analytical method , compared to monte carlo simulations , is its limited capability to incorporate systematic instrumental artifacts . we close by mentioning a few straight forward extensions of the model ( eq . [ model ] ) . * flare clustering . * the assumption of independent flare times might be questionable in the light of more advanced avalanche models @xcite where flares may trigger each other . flare clustering can be accounted for by a flare waiting time distribution which does not create uniform flare times ( @xmath253 ) , but has a long - range tail permitting rare but large jumps . the ` clusters ' are between the jumps . if the flare amplitudes are still independently @xmath58-distributed , rise immediately and decay exponentially , then the functional equation ( [ phi_eq ] ) for the peak values still holds , but the separation into the form ( [ int_eq ] ) is no longer possible . however , equation ( [ phi_eq ] ) can be iterated : the sequence @xmath254 converges uniformly to the characteristic function of @xmath255 of clustered flares . all iterates are characteristic functions , and the characteristic function of the flare photon intensity is given by @xmath256 ( see section [ peak_sect ] ) . * ccd movies . * the arguments leading to equations ( [ phi_f ] ) and ( [ phi_f ] ) are not restricted to one - dimensional light curves . a useful generalization is to a sequence of spatially resolved ccd images od solar flares , where binning is over the space - time pixel @xmath257 , and the auxiliary interval @xmath42 of section 3.1 becomes @xmath258 . provided that the flares occur independently and uniformly in @xmath42 , the derivation of equation ( [ phi_f ] ) goes through without modifications if @xmath0 is replaced by @xmath259 and @xmath260 is replaced by @xmath261 with @xmath262 and @xmath263 the space - time flare shape . we thank marc audard for discussions on the euve data set . by definition , @xmath4 satisfies @xmath264 ( normalization ) . let us assume that @xmath265 as @xmath266 with @xmath62 and @xmath267 ; this is a valid form of @xmath4 which is realized , for @xmath268 , in lvy distributions @xcite . since @xmath269 as @xmath65 , the integral in equation ( [ phi_ft ] ) remains well - behaved for all @xmath188 when @xmath61 if @xmath270 , requiring that @xmath63 decays faster than @xmath64 as @xmath65 . so far about the decay of the flare shape . the decay of the probability density @xmath25 at large flare amplitudes @xmath9 is related by tauberian theorems @xcite to the small-@xmath188 limit of the characteristic function . for our assumption @xmath271 , one has that @xmath272 as @xmath67 . these are the conditions mentioned in section [ flare_contrib_sect ] . for special flare shapes and amplitude distributions , equations ( [ phi_f ] ) and ( [ phi_f ] ) can explicitly be evaluated . to this end , it is advantageous to cast them in the volterra form @xmath273 with kernels @xmath274 where @xmath275 is the @xmath53-th solution of @xmath276 ( kernel @xmath277 ) or @xmath278 ( kernel @xmath279 ) , respectively , and prime denotes derivative with respect to @xmath28 . the success of an explicit evaluation of equations ( [ volterra_f ] , [ volterra_f ] ) depends on the simplicity of @xmath4 and on the ability to invert the functions @xmath19 and @xmath80 for the kernels @xmath277 and @xmath279 . @xmath1 is generally harder to obtain than @xmath50 . if the flare shape is @xmath280 then the equation @xmath278 has the solutions @xmath119 = @xmath281 ( valid for the rise phase @xmath282 ) and @xmath121 = @xmath283 ( valid for the decay phase @xmath284 ) . the corresponding kernel contributions ( eq . [ kernel ] ) are @xmath285 and @xmath286 , and equation ( [ volterra_f ] ) becomes @xmath287 the mean of @xmath73 is @xmath288 , and its variance is @xmath289 . for short time bins ( @xmath85 ) one has that @xmath290 , and equation ( [ expdecay1 ] ) reduces to @xmath291 with @xmath50 given by equation ( [ phi_f1 ] ) . we present now several exactly solvable models for the flare amplitude distribution . for @xmath292 and one - sided exponential flare shape , all characteristic functions and the photon waiting time distribution can be rigorously calculated . the results are @xmath293 equation ( [ phi_f_exp ] ) indicates that @xmath208 is a chi square distribution with @xmath33 ( non - integer ) degrees of freedom . the combination of exponential flare amplitudes and exponential flare profiles has the peculiar property that @xmath294 ( and thus @xmath295 ) if @xmath296 . if a constant background @xmath206 [ ct @xmath202 ] is present , the exact ( eq . [ p_d ] ) and short - term ( eqns . [ p_d0 ] , [ p_d1 ] ) forms of the photon waiting time distributions are @xmath298 ^ 2 \hspace{-1 mm } + \frac{a^2 \tau e^{x/\tau}}{t } \big\ } \label{p_d_exp_bg } \\[2 mm ] p_\delta(x ) & \simeq & \frac{e^{-bx}(1+ax)^{-\tau / t}}{b+a\tau t^{-1 } } \big\ { \big ( b + \frac{a\tau t^{-1}}{1+ax } \big)^2 + \frac{a^2 \tau t^{-1 } } { ( 1+ax)^2 } \big\ } \ , .\label{p_d_exp_bg0}\end{aligned}\ ] ] equations ( [ p_d_exp_bg ] ) and ( [ p_d_exp_bg0 ] ) allow to benchmark ( and to be benchmarked by ) our monte carlo simulations ( section [ simulations ] ) . they also provide a measure of validity of the short - term approximation ( eqns . [ p_d0 ] , [ p_d1 ] ) : for example , for @xmath9 = @xmath299 , @xmath300 = 25 photons per flare , @xmath33 = 1 ( moderate flare overlap ) and @xmath206 = 0.2@xmath301 ( 20% background ) , the error of the short - term approximation is about 0.6% . if @xmath25 is a truncated power law ( eq . [ p_a_pl ] ) and if the flare shape is exponential , then @xmath50 may be expressed in terms of special functions . possible representations are @xmath302,[2,2,3-\nu],ia_0s ) \big\ } \label{phi_f_pl1 } \\ \phi_f(s ) & = & e^{\frac{\tau}{t}(\gamma - \frac{i\pi}{2 } + \frac{1}{\nu-1 } ) } \ , ( a_0s)^{-\tau / t } e^{-\frac{\tau}{t}\big\ { { \rm e_1}(-ia_0s ) + \nu ( -ia_0s)^{\nu-1 } \gamma(-\nu ,- ia_0s ) + \frac{e^{ia_0s}}{ia_0s } \big\ } } \label{phi_f_pl2}\end{aligned}\ ] ] where @xmath303 denotes the generalized hypergeometric function @xcite , @xmath304 is euler s constant , @xmath305 is the incomplete gamma function @xcite , @xmath306 is the exponential integral , and the principal branches of the @xmath307 and @xmath308 functions are to be used . equations ( [ phi_f_pl1 ] , [ phi_f_pl2 ] ) have been obtained by the computer algebra program maple . in numerical computations we use the form ( [ phi_f_pl1 ] ) with a taylor ( small - argument ) expansion of the hypergeometric function . for values of @xmath309 @xmath230 1 , this expansion converges rapidly , and machine precision is reached after @xmath310 6 orders if @xmath309 @xmath245 0.25 . in fact , already order two yields @xmath310 0.0005 rms relative precision for @xmath10 over the whole waiting time range 0 @xmath245 @xmath116 @xmath245 30s used in the euve data . this fact is exploited in the approximation ( [ pd_approx ] ) , which is based on an order - two taylor expansion of the hypergeometric function . if @xmath25 is a levy distribution @xcite with characteristic function @xmath311 , @xmath267 , and 0 @xmath310 @xmath312 @xmath310 1 , then @xmath25 vanishes at negative @xmath9 and therefore can model flare amplitudes . large flare amplitudes ( @xmath9 @xmath247 @xmath113 ) decay like @xmath25 @xmath214 @xmath313 , and superimposed flares have @xmath50 = @xmath314 with @xmath315 and @xmath316 is euler s constant ( see above ) . such lvy distributions might be a convenient model for small solar flares where power law indices @xmath310 2 have been reported @xcite . we ask for an elementary derivation of the results ( [ phi_f ] - [ p_d ] ) . the key observation is that equations ( [ phi_f ] ) and ( [ phi_f ] ) imply a simple relation between the moments of the flare amplitudes and the cumulants are defined by @xmath317 . cumulants of order @xmath318 can be expressed in terms of moments of order @xmath318 , and vice versa ; see , e.g. , @xcite . ] of the instantaneous photon intensity and the binned poisson parameter . this fact may be exploited to establish a relation between the moments of the flare amplitudes and the moments of the binned counts and photon waiting times , without recurrence to characteristic functions . we illustrate the procedure starting with the expectation value @xmath319 . interpreting the angular brackets as time average ( which is justified by the assumption of stationarity ) , the central limit theorem ensures that @xmath320 where @xmath321 is the number of flares in the interval @xmath42 ( see section [ flare_contrib_sect ] ) . when restricting the first sum in eq . ( [ time_mean ] ) to flares inside @xmath42 it has been supposed that contributions from outside @xmath42 decay sufficiently fast to be neglected , and we have used the fact that all @xmath20 and @xmath26 are statistically independent of each other . similarly , the second moment of @xmath45 is given by @xmath322 ^ 2 \ , dt + o\big({\cal n}_i^{1/2}\big ) \nonumber \\ & \to & \langle a \rangle^2 t^{-2 } \big ( \int \hspace{-1 mm } \xi(t ) \ , dt \big)^2 + \langle a^2 \rangle t^{-1 } \int [ \xi(t)]^2 \ , dt \label{time_var } \ , .\end{aligned}\ ] ] from equations ( [ time_mean ] ) and ( [ time_var ] ) we recognize that @xmath323 ^ 2 \ , dt$ ] : the variance of the instantaneous photon intensity equals the second moment of the flare amplitudes times a weight @xmath324 ^ 2 \ , dt$ ] . this finding can be generalized to higher cumulants if the division into @xmath325=@xmath53 and @xmath325@xmath326@xmath53 contributions in equation ( [ kl_split ] ) is replaced by a sum over partitions into equal and non - equal indices . we shall not discuss here the ( complicated ) combinatorial weights , but note that they correspond to those of the cumulants expressed in terms of moments ( e.g. , @xcite ) . as a consequence , @xmath327^n \ , dt \ \label{cumulants_f}\ ] ] equation ( [ cumulants_f ] ) is a weak form of equation ( [ phi_f ] ) , since it represents its taylor expansion around @xmath188=0 . it is easily seen that equation ( [ cumulants_f ] ) also applies to the binned poisson parameter @xmath73 if @xmath63 is replaced by @xmath328 . therefore the cumulants ( and thus the moments ) of both @xmath45 and @xmath73 can be expressed in terms of the moments of the flare amplitudes . the probability density of the binned counts @xmath6 is then given by @xmath329 , while the probability density of short ( section [ waitingtimesect ] ) photon waiting times @xmath9 is @xmath330 . although these series may be evaluated in special cases , it is generally much more convenient to work with the characteristic functions , which collect all the moments in a compact way . abramowitz , m. , & stegun , i. , 1970 , _ handbook of mathematical functions _ ( dover : dover publ . inc . ) ambruster , c. w. , sciortino , s. , & golub , l. 1987 , apj , 65 , 273 aschwanden , m. j. , tarbell , t. d. , nightingale , r. w. , schrijver , c. j. , title , a. , kankelborg , c. c. , martens , p. c. h. , & warren , h. p. , 2000 , apj , 535 , 1047 audard , m. , gdel , m. , & guinan , e. f. 1999 , , 513 , l53 audard , m. , gdel , m. , drake , j. j. , & kashyap , v. 2000 , apj , 541 , 396 audard , m. , gdel , m. , & skinner , s. l. 2003 , apj , 589 , 983 axford , w. i. , & mckenzie , j. f. , 1996 , _ astroph . space sci _ , 243 , 1 benz , a. o. , 1995 , in _ coronal magnetic energy release _ benz and a. krger ( berlin : springer ) benz , a. o. , & krucker , s. , 1999 , a&a , 341 , 286 bochner , s. , 1959 , _ lectures on fourier integrals _ ( princeton : univ . press ) butler , c. j. , rodon , m. , foing , b. h. , & haisch , b. m. 1986 , nature , 321 , 679 collura , a. , pasquini , l. , & schmitt , j. h. m. m. 1988 , , 205 , 197 crosby , n. b. , aschwanden , m. , & dennis , b. , 1993 , _ solar phys . _ , 143 , 275 datlowe , d. w. , elcan , m. j. , & hudson , h. s. 1974 , , 39 , 155 de jager , c. , heise , j. , van genderen , a. m. , foing , b. h. , ilyin , i. v. , kilkenny , d. s. , marvridis , l. , cutispoto , g. , rodon , m. , seeds , m. a. , et al . 1989 , a&a , 211 , 157 eadie , w. t. , drojard , d. , james , f. e. , roos , m. , & sadoulet , b. , _ statistical methods in experimental physics _ , 1971 ( amsterdam : north - holland ) feldman , u. , doschek , g.a . , & klimchuk 1997 , j.a . , _ apj _ , 474 , 511 gnedenko , b. v. , & kolmogorov , a. n. , 1954 , _ limit distributions for sums of independent random variables _ ( cambridge : addison - wesley ) gdel , m. , audard , m. , skinner , s. l. , & horvath , m. i. 2002 , apj , 580 , l73 gdel , m. , audard , m. , kashyap , v. , drake , j. , & guinan , e. , 2003 , apj , 582 , 423 gdel , m. , audard , m. , reale , f. , skinner , s. l. , & linsky , j. l. 2004 , a&a , in press gradshteyn , i. s. , & ryzhik , i. m. , 1980 , _ table of integrals , series and products _ ( new york : academic press ) heyvaerts , j. , & priest , e. , a&a , 1983 , 117 , 220 heyvaerts , j. , & priest , e. , a&a , 1984 , 137 , 63 hollweg , j. v. , coronal heating by waves , 1983 , in _ solar wind 5 _ , ed . m. neugebauer ( nasa cp-2280 , washington dc ) , 1 isliker , h. , a&a , 1996 , 310 , 672 kashyap , v. , & drake , j. j. 1999 , , 524 , 988 kashyap v. l. , drake , j. j. , gdel , m. , & audard , m. , 2002 , apj , 580 , 1118 kraev , u. , & benz , a. o. , 2001 , a&a , 373 , 318 krucker , s. , & benz , a. o. 1998 , , 501 , l213 lvy , p. , 1937 , _ thorie de laddition des variables alatoires _ ( paris : gauthiers - villars ) lin , r. p. , schwartz , r. a. , kane , s. r. , pelling , r. m. , & hurley , k. 1984 , apj , 283 , 421 lukacs , e. , 1970 , _ characteristic functions _ ( london : charles griffin ) malina , r. f. , & bowyer , s. , 1991 , in _ extreme - ultraviolet astronomy _ , ed . r. f. malina and s. bowyer ( new york : pergamon ) marsh , e. , & tu , c. y. , 1997 , a&a , 316 , l17 narain , u. & ulmschneider , p. 1990 , _ space science rev . _ , 54 , 377 narain , u. & ulmschneider , p. 1996 , _ space science rev . _ , 75 , 453 osten , r. a. , & brown , a. 1999 , apj , 515 , 746 parker , e. 1972 , apj , 174 , 499 pearce , g. , & harrison , r. a. 1988 , a&a , 206 , 121 priest , e. , & forbes , t. , 2000 , _ magnetic reconnection - mhd theory and applications _ ( cambridge : cambridge univ . press ) rota , g. , & shen , j. ( 2000 ) , _ j. of combinatorial theory _ , a 91 , 283 shimizu , t. 1995 , _ publ . soc . japan _ , 47 , 251 stein , m. l. 1999 , _ interpolation of spatial data _ , springer ulmschneider p. , rosner , r. , & priest , e. , 1991 , _ mechanisms of chromospheric and coronal heating _ ( berlin : springer ) vlahos , l. , georgoulis , m. , kluving , r. , & paschos , p. , 1995 , apj , 299 , 897 wheatland , m. s. 2000 , _ solar phys . _ , 191 , 381 wheatland , m. s. & craig , i. j. d. 2003 , apj , 595 , 458

electron spin relaxation due to the dyakonov - perel mechanism is investigated in bilayer graphene with only the lowest conduction band being relevant . the spin - orbit coupling is constructed from the symmetry group analysis with the parameters obtained by fitting to the numerical calculation according to the latest report by konschuh _ et al . _ [ phys . rev . b * 85 * , 115423 ( 2012 ) ] from first principles . in contrast to single - layer graphene , the leading term of the out - of - plane component of the spin - orbit coupling in bilayer graphene shows a zeeman - like term with opposite effective magnetic fields in the two valleys . this zeeman - like term opens a spin relaxation channel in the presence of intervalley scattering . it is shown that the intervalley electron - phonon scattering , which has not been reported in the previous literature , strongly suppresses the in - plane spin relaxation time at high temperature whereas the intervalley short - range scattering plays an important role in the in - plane spin relaxation especially at low temperature . a marked nonmonotonic dependence of the in - plane spin relaxation time on temperature with a minimum of several hundred picoseconds is predicted in the absence of the short - range scatterers . this minimum is comparable to the experimental data . moreover , a peak in the electron density dependence of the in - plane spin relaxation time at low temperature , which is very different from the one in semiconductors , is predicted . we also find a rapid decrease in the in - plane spin relaxation time with increasing initial spin polarization at low temperature , which is opposite to the situation in both semiconductors and single - layer graphene . a strong anisotropy between the out - of- and in - plane spin relaxations at high temperature is also revealed with the out - of - plane spin relaxation time being about two orders of magnitude larger than the in - plane one . detailed comparisons of the temperature and electron density dependences of the spin relaxation with the existing experiments [ phys . rev . lett . * 107 * , 047206 ( 2011 ) , phys . rev . lett . * 107 * , 047207 ( 2011 ) and nano lett . * 11 * , 2363 ( 2011 ) ] are reported .

1302.5880

in recent years , single - layer graphene has received much attention due to its two dimensionality , dirac - like energy spectrum and outstanding spin - coherence properties.@xcite specifically , the hyperfine interaction and spin - orbit coupling ( soc ) are considerably weak,@xcite suggesting a long spin relaxation time ( srt ) in single - layer graphene . this long srt makes single - layer graphene a promising candidate for spintronic application . so far , a number of experiments on spin relaxation have been carried out , with the srts being orders of magnitude shorter than theoretical prediction.@xcite this pronounced discrepancy has provoked many investigations on the extrinsic effects of spin relaxation such as adatoms,@xcite curvature,@xcite substrate effects@xcite and contacts.@xcite consisting of two layer graphene , bilayer graphene ( blg ) also possesses good spin - coherence properties.@xcite additionally , in contrast to single - layer graphene , blg has a tunable gap caused by an out - of - plane electric field,@xcite which can be used to turn on or off the electronic transport . in addition , in the absence of space inversion symmetry , the intrinsic soc in the pseudospin space of blg provides an out - of - plane effective magnetic field near the dirac points whereas the intrinsic soc of single - layer graphene only induces a shift of the energy spectrum.@xcite moreover , the electron - phonon coupling becomes more complex in blg due to the existence of more phonon modes.@xcite so far , some efforts have been made to investigate the intravalley electron - phonon coupling.@xcite specifically , the intravalley electron - acoustic ( ac ) phonon coupling has been calculated by viljas and heikkil@xcite with the continuum theory and also by borysenko _ et al._@xcite from first principles . to explore the intravalley electron - optical ( op ) phonon coupling , the tight - binding model , density functional theory and first principles approach are employed by viljas and heikkil,@xcite cappelluti _ et al._@xcite and borysenko _ et al._,@xcite respectively . however , the intervalley electron - phonon coupling , which has been shown to play an important role in spin relaxation in rippled single - layer graphene,@xcite has not been reported in the previous literature . all these intriguing features make blg an attractive material for investigation . very recently , some attention has been devoted to the spin relaxation in blg.@xcite experimentally , the srts of blg on sio@xmath0 substrate@xcite or in freely suspended blg@xcite are reported to be of the order of @xmath1-@xmath2ns . both yang _ et al._@xcite and han and kawakami@xcite measured the temperature and carrier dependences of the srt . et al._@xcite showed that the srt is weakly dependent on temperature . however , han and kawakami@xcite reported a marked decrease of the srt with the increase of the temperature . as for the carrier density dependence , the decrease of the srt with increasing carrier density at low temperature is reported in both experiments . however , at room temperature , yang _ et al._@xcite observed an increase of the srt with increasing density whereas the srt measured by han and kawakami@xcite exhibits a marginal density dependence . in addition , yang _ et al._@xcite also showed that the srt scales inversely with the mobility at both room and low temperatures . in contrast to the above two experiments based on the micromechanically exfoliated blg samples , avsar _ et al._@xcite carried out the experiment in chemical vapor deposition grown blg . the carrier density and temperature dependences of the srt measured are comparable to the ones by yang _ et al._@xcite both quantitatively and qualitatively . motivated by these experiments , diez and burkard@xcite calculated the electron spin relaxation due to the dyakonov - perel ( dp ) mechanism@xcite with only the long - range electron - impurity scattering included in blg . they claimed to obtain a good agreement with the experimental data of yang _ et al._.@xcite it is noted that the effective soc ( referred to as the spin - orbit field in ref . ) used in their calculation is induced by an interlayer bias voltage together with the intrinsic soc in the pseudospin space.@xcite however , the intrinsic soc in the pseudospin space becomes incomplete in the presence of the interlayer bias since the interlayer bias breaks the space inversion symmetry and leads to additional extrinsic soc terms in the pseudospin space.@xcite also even with only the intrinsic soc in the pseudospin space in their calculation , they overlooked the out - of - plane component of the effective soc . as will be shown later , the leading term of this component leads to a zeeman - like term in the two valleys and makes a dominant contribution to the in - plane spin relaxation . moreover , they neglected the electron - phonon and electron - electron coulomb scatterings . in fact , the electron - electron coulomb scattering has been demonstrated to be very important for the spin relaxation in semiconductor systems@xcite and also single - layer graphene at high temperature.@xcite the electron - phonon scattering has also been shown to play an important role in spin relaxation in both semiconductors@xcite and single - layer graphene@xcite at high temperature . therefore , both have to be included in studying the spin relaxation . in the present work , with the electron - electron coulomb , ( both the intra- and inter - valley ) electron - phonon , possible short - range@xcite as well as long - range electron - impurity scatterings explicitly included , we investigate the electron spin relaxation due to the dp mechanism in blg with only the lowest conduction band being relevant by the kinetic spin bloch equation ( ksbe ) approach.@xcite the soc of the lowest conduction band near the dirac points @xmath3 is constructed from the symmetry group @xmath4 in the presence of an out - of - plane electric field.@xcite its three components @xmath5 @xmath6 can be written as ( see appendix [ appa ] ) @xmath7,\label{eq1}\\ \omega_y^{\mu}({\bf k})&=&-\alpha_1(k)\cos\theta_{\bf k}+ \mu[\alpha_2(k)\cos 2\theta_{\bf k}\nonumber\\ & & \mbox{}-\alpha_3(k)\cos 4\theta_{\bf k}],\label{eq2}\\ \omega_z^{\mu}({\bf k})&=&\mu\beta_1(k)+\beta_2(k)\cos 3\theta_{\bf k},\label{eq3}\end{aligned}\ ] ] with @xmath8 standing for k(k@xmath9 ) valleys and @xmath10 and @xmath11 representing the magnitude and the polar angle of the momentum @xmath12 relative to k(k@xmath9 ) points , respectively . the coefficients @xmath13 ( @xmath14-@xmath15 ) and @xmath16 are given in appendix [ appa ] by fitting to the numerical results calculated according to the latest report by konschuh _ et al._@xcite from first principles with both the intrinsic and extrinsic soc terms in the pseudospin space included . the out - of - plane component @xmath17 [ eq . ( [ eq3 ] ) ] , which has not been reported in the previous literature , is induced by the intrinsic soc in the pseudospin space in the absence of space inversion symmetry , whereas the in - plane component is contributed by both the intrinsic and extrinsic socs in the pseudospin space . this is very different from the case in single - layer graphene where the intrinsic soc in the pseudospin space only induces a shift of the energy spectrum in the conduction band.@xcite we find that the magnitudes of both the out - of- and in - plane components decrease with the increase of @xmath10 at large momentum , indicating a suppression of the inhomogeneous broadening@xcite by increasing momentum.@xcite this is very different from the case in both semiconductors and single - layer graphene.@xcite additionally , the leading term of the out - of - plane component , i.e. , @xmath18 , serves as a zeeman - like term with opposite effective magnetic fields perpendicular to blg plane in the two valleys , similar to the case in rippled single - layer graphene.@xcite this zeeman - like term , together with the intervalley scattering , opens a spin relaxation channel . this spin relaxation channel suppresses the in - plane srt _ significantly _ at high temperature by the intervalley electron - phonon scattering and also at low temperature by the short - range scattering.@xcite in addition , we find that although such zeeman - like terms also exist in the in - plane components of the soc , their contributions to the out - of - plane spin relaxation are marginal since these terms are of the high order of the momentum . this leads to a strong anisotropy between the in - plane and out - of - plane spin relaxations . we also find that the in - plane srt shows a marked nonmonotonic dependence on temperature with a minimum down to several hundred picoseconds in the absence of short - range scatterers . this minimum is comparable to the experimental data . the nonmonotonic behavior results from the crossover between the weak and strong intervalley electron - phonon scattering . moreover , we predict a peak in the density dependence at low temperature , which is very different from the one in semiconductors.@xcite as for high temperature , the srt shows a monotonic increase with increasing density . we also find that the in - plane srt decreases rapidly with increasing initial spin polarization at low temperature , which is very different from the previous studies in both semiconductors@xcite and single - layer graphene.@xcite finally , we report detailed comparisons of the temperature and electron density dependences with the existing experiments.@xcite this paper is organized as follows . in sec . ii , we present our model and construct the ksbes . the main results are given in sec . iii . in sec . iiia , we investigate the temperature , electron density and also initial spin polarization dependences of the electron spin relaxation in the absence of short - range scatterers . the anisotropy of the spin relaxation is also addressed in this part . in sec . iiib , we show detailed comparisons of the temperature and electron density dependences of the spin relaxation with the existing experiments with the inclusion of short - range scatterers . we summarize in sec . iv . we start our investigation from the ab - stacked blg on sio@xmath0 substrate . the spinless @xmath19-band structure can be described by the effective @xmath20 hamiltonian:@xcite @xmath21 in the on - site orbital bloch basis @xmath22 , @xmath23 , @xmath24 and @xmath25 , where a@xmath26 refers to the a sublattice in the lower layer , b@xmath0 to the b sublattice in the upper layer , etc . , and @xmath27 represents the two - dimensional wave vector counted from the @xmath28 point . here , @xmath29 and @xmath30 describe the nearest - neighbor intralayer and interlayer hoppings whereas @xmath31 and @xmath32 stand for the indirect hoppings between two layers . @xmath33 represents the asymmetry in the energy shift of the on - site energy due to the interlayer hopping . the potential @xmath34 with @xmath35 , @xmath36 and @xmath37 representing the electron charge @xmath38 , the out - of - plane electric field and the effective electrostatic bilayer distance , respectively . @xmath39 $ ] is the nearest - neighbor structural function of the graphene hexagonal lattice with @xmath40 denoting the lattice constant . near the dirac points , to the first order of the momentum @xmath41 originated from the k(k@xmath9 ) points , @xmath42 . then the effective @xmath20 hamiltonian near the dirac points can be written as @xmath43 by exactly diagonalizing this hamiltonian , one obtains the eigenvalue @xmath44 and the eigenfunction @xmath45 with @xmath46 being the band index including two conduction bands and two valence ones . in our calculation , we concentrate on the heavily @xmath47-doped case with only the lowest conduction band being relevant whereas other bands are separated by large energy intervals . for convenience , the eigenvalue and the corresponding eigenfunction of the lowest conduction band at @xmath12 are denoted by @xmath48 and @xmath49 , respectively . by incorporating the spin degree of freedom , the effective hamiltonian of the lowest conduction band is given by @xmath50 here , @xmath51 are the pauli matrices . @xmath52 stands for the effective magnetic field of the soc , which is constructed from the symmetry analysis with the parameters obtained approximately by fitting to numerical results based on the report by konschuh _ et al._.@xcite the analytical form of the soc with three components @xmath5 @xmath6 is shown in eqs . ( [ eq1]-[eq3 ] ) and the explicit numerical calculation is given in appendix [ appa ] . it is noted that at large momentum , the analytical result agrees with the numerical one fairly well ( see appendix [ appa ] ) . we then construct the microscopic ksbes@xcite to study the electron spin relaxation in blg . the ksbes read@xcite @xmath53 where @xmath54 represent the density matrices of electrons with the diagonal terms @xmath55 describing the distribution functions and the off - diagonal ones @xmath56 denoting the spin coherence . the coherent term is given by @xmath57 , \label{eq8}\end{aligned}\ ] ] where @xmath58\equiv ab - ba$ ] is the commutator ; @xmath59 represents the coulomb hartree - fock ( hf ) term.@xcite @xmath60 denotes the screened coulomb potential with its form given in appendix [ appb ] . the form factor @xmath61 . @xmath62 are the scattering terms including the electron - electron coulomb ( @xmath63 ) , long - range electron - impurity ( @xmath64 ) , intravalley electron - ac phonon ( @xmath65 ) , electron - op phonon ( @xmath66 ) , electron - remote - interfacial ( ri ) phonon ( @xmath67 ) and especially the intervalley electron - phonon scattering including the electron-@xmath68 phonon ( @xmath69 ) , electron-@xmath70 phonon ( @xmath71 ) , electron-@xmath72 phonon ( @xmath73 ) and electron-@xmath74 phonon ( @xmath75 ) scatterings . their detailed expressions can be found in ref . .@xcite here , @xmath68 , @xmath70 and @xmath72 are the in - plane phonon modes corresponding to the representations @xmath76 , @xmath77 and @xmath78 of group @xmath79 , respectively whereas @xmath74 stands for the out - of - plane phonon mode.@xcite it is noted that the intravalley electron - phonon scattering has been reported in the previous literature as mentioned in the introduction.@xcite however , to the best of our knowledge , there is no specific report to the intervalley electron - phonon scattering . in this work , we derive the intervalley electron - phonon scattering matrix elements using the tight - binding model according to the arxiv version of ref . . the scattering matrix elements are given by @xmath80,\label{eq9}\\ & & \hspace{-0.4cm}|m^{\rm ka_2^{\prime}}_{{\mu{\bf k}},{\mu^{\prime}{\bf k}^{\prime}}}|^2=\frac{3a^2{\gamma_4^{\prime}}^2}{2\rho l_4 ^ 2 } \frac{\hbar^2}{2\omega_{\rm ka_2^{\prime}}}\delta_{\mu^{\prime},-\mu } \big[|{\psi^{\mu}_{{\bf k}}}^{\dagger}\gamma^3_{\rm d}\gamma^5_{\rm d}\psi^{\mu^{\prime}}_{{\bf k}^{\prime}}|^2\nonumber\\ & & \mbox{}\hspace{0.1cm}+|{\psi^{\mu}_{{\bf k}}}^{\dagger}\gamma^0_{\rm d}\psi^{\mu^{\prime}}_{{\bf k}^{\prime}}|^2\big],\label{eq10}\\ & & \hspace{-0.4cm}|m^{\rm ke^{\prime}}_{{\mu{\bf k}},{\mu^{\prime}{\bf k}^{\prime}}}|^2=\frac{3a^2}{\rho l_4 ^ 2 } \frac{\hbar^2}{4\omega_{\rm ke^{\prime}}}\delta_{\mu^{\prime},-\mu } \big[|{\psi^{\mu}_{{\bf k}}}^{\dagger}(\gamma_3^{\prime}\gamma^1_{\rm d}\gamma^5_{\rm d}-i\gamma_3^{\prime}\gamma^2_{\rm d}\nonumber\\ & & \mbox{}\hspace{0.1cm}-\gamma_4^{\prime}\gamma^0_{\rm d})\psi^{\mu^{\prime}}_{{\bf k}^{\prime}}|^2+|{\psi^{\mu}_{{\bf k}}}^{\dagger } ( i\gamma_3^{\prime}\gamma^2_{\rm d}-\gamma_3^{\prime}\gamma^1_{\rm d}\gamma^5_{\rm d}-\gamma_4^{\prime}\gamma^0_{\rm d } ) \psi^{\mu^{\prime}}_{{\bf k}^{\prime}}|^2\nonumber\\ & & \mbox{}\hspace{0.1cm}+4|{\psi^{\mu}_{{\bf k}}}^{\dagger}\gamma_4^{\prime}\gamma^3_{\rm d}\gamma^5_{\rm d } \psi^{\mu^{\prime}}_{{\bf k}^{\prime}}|^2\big],\label{eq11}\\ & & \hspace{-0.4cm}|m^{\rm kz}_{{\mu{\bf k}},{\mu^{\prime}{\bf k}^{\prime}}}|^2=\frac{{\gamma_1^{\prime}}^2}{\rho } \frac{\hbar^2}{2\omega_{\rm kz}}\delta_{\mu^{\prime},-\mu } \label{eq12}\end{aligned}\ ] ] where @xmath81 is the mass density of the blg;@xcite @xmath82 , @xmath83 , @xmath84 and @xmath85 are the energy spectra of @xmath68 , @xmath70 , @xmath72 , and @xmath74 phonon modes , respectively;@xcite @xmath86 ( @xmath87 ) is the derivative of @xmath88 with respect to the corresponding hopping bond length;@xcite @xmath89 is the bond length of the interlayer hopping @xmath32 ; @xmath90 ( @xmath91-@xmath15 , @xmath92 ) , @xmath93 and @xmath94 are @xmath20 dirac matrices given in appendix [ appc].@xcite we will show that the above intervalley electron - phonon scattering plays a significant role in the in - plane spin relaxation in our investigation . the remaining scattering matrix elements such as @xmath63 , @xmath64 , and @xmath95 ( @xmath96 ) are given in appendix [ appb ] . it is noted that we also include the short - range scattering in our calculation with the scattering term laid out later in this paper . .parameters used in the calculation ( from ref . unless otherwise specified ) . [ cols= " < , < , < , < " , ] by numerically solving the ksbes , one obtains the time evolution of spin polarization along direction @xmath97 @xmath98/n_e$ ] with @xmath99 being the electron density . then the srt can be determined from the slope of the envelope of the spin polarization @xmath100.@xcite the initial spin polarization @xmath101 is set to be @xmath102 % and the spin - polarization direction @xmath97 is along the @xmath103-axis unless otherwise specified . it is noted that all the parameters used in our calculation are listed in table i. it is also noted that the soc is calculated both fully numerically following the approach by konschuh _ et al._@xcite and analytically by using eqs . ( [ eq1]-[eq3 ] ) . the srt calculated with the analytical form of the soc and with the explicit numerical one agrees fairly well with each other . in our calculation , the out - of - plane electric field is chosen to be a typical value @xmath104v / nm@xcite and the long - range impurity density is taken to be @xmath105@xmath106 . the corresponding mobility in our investigation is of the order of @xmath107@xmath108/(vs ) , which is about @xmath109-@xmath110 orders of magnitude larger than those reported in the existing experiments in blg on sio@xmath0 substrate@xcite and comparable to the ones in the experiments in freely suspended blg.@xcite it is noted that the short - range scattering is not included in this subsection . -axis @xmath111 ( @xmath112 ) and the calculation without the long - range electron - impurity ( @xmath113 ) , electron - ac - phonon ( @xmath114 ) , electron - ri - phonon ( @xmath115 ) , electron - electron coulomb ( @xmath116 ) or intervalley electron - phonon ( @xmath117 ) scattering as function of temperature @xmath118 . it is noted that when @xmath119 k , only the total srt and the calculation without the intervalley electron - phonon scattering are shown since the remaining ones almost coincide with the total srt . in the calculation , the electron density @xmath120@xmath106.,width=321 ] we first investigate the temperature dependence of the spin relaxation . in fig . [ fig1 ] , the srt along the @xmath103-axis @xmath111 is plotted as function of temperature @xmath118 . it is seen that @xmath111 changes little with the temperature at low temperature ( i.e. , @xmath121k ) . however , as @xmath118 becomes larger , the srt decreases dramatically with increasing temperature . in particular , the srt reaches down to several hundred picoseconds at @xmath122k , which is comparable to the experimental values.@xcite a detailed comparison with the existing experiments will be discussed in sec . when @xmath118 further increases , the srt begins to increase with @xmath118 . to understand the behavior of the spin relaxation , we show in fig . [ fig1 ] the temperature dependence of the srt calculated with the electron - electron coulomb , electron - ri - phonon , electron - ac - phonon , long - range electron - impurity , or intervalley electron - phonon scattering removed , separately . we find that at low temperature , the intervalley electron - phonon scattering is negligible . therefore , spins relax independently in the two valleys and the srt is solely determined by the intravalley spin relaxation channel . specifically , the long - range electron - impurity scattering is dominant , which leads to a weak temperature dependence of the srt.@xcite with the increase of the temperature , the contribution of the long - range electron - impurity scattering becomes marginal whereas the intervalley electron - phonon scattering becomes important . this intervalley scattering opens another spin relaxation channel together with the zeeman - like term with opposite effective magnetic fields in the two valleys , which is similar to the case in rippled single - layer graphene.@xcite this zeeman - like term is given by @xmath18 [ eq . ( [ eq3 ] ) ] approximately . then the spin relaxation due to the above intervalley spin relaxation channel can be approximated by the rate equations@xcite @xmath123 where @xmath124 , @xmath125\hat{{\bf z}}$ ] and @xmath126 represent the in - plane spin vectors in each valley , spin precession vector along the @xmath127-axis and the intervally electron - phonon scattering time , respectively . according to the report by zhang _ et al._,@xcite the in - plane srt is given by @xmath128 here , @xmath129 stands for the srt along the @xmath130-axis . since the system we investigate is always in the degenerate regime , we take @xmath131 approximately with @xmath132 representing the fermi wave vector . in our calculation , the srt along the @xmath103-axis is solely determined by the intervalley electron - phonon scattering at @xmath133-@xmath134k . at @xmath135k , @xmath136 , i.e. , the intervalley electron - phonon scattering is in the weak scattering limit . as a result , the srt @xmath137 according to eq . ( [ eq14 ] ) , which decreases with the enhancement of the intervalley scattering as the temperature increases . nevertheless , at @xmath138k , @xmath139 , i.e. , the intervalley scattering is in the strong scattering limit . then the srt @xmath140 $ ] from eq . ( [ eq14 ] ) , which increases with increasing of @xmath118 . therefore , the srt first decreases then increases with increasing temperature when the intervalley scattering changes from the weak to strong scattering limit . the crossover between the weak and strong intervalley scattering limit is determined by @xmath141.@xcite at this crossover point , @xmath142 here @xmath143ps , very close to the value @xmath144ps shown in the figure . it is noted that the electron - electron coulomb scattering is unimportant in the whole temperature regime . then we turn to study the electron - density dependence . in fig . [ fig2 ] , the srt along the @xmath103-axis is plotted against the electron density at @xmath145 @xmath146k . one observes a peak in the electron - density dependence of the srt at @xmath135k whereas the srt increases with the increase of the electron density when @xmath147k . we first focus on the case at @xmath135k and show the srt with the intervalley electron - phonon scattering excluded in fig . it is seen that when the electron density is low , the intervalley electron - phonon scattering is marginal . the srt is then determined by the intravalley spin relaxation channel and hence increases with the suppression of the inhomogeneous broadening@xcite as the electron density increases.@xcite the suppression of the inhomogeneous broadening can be understood from the decrease of the spin splitting with increasing momentum at high momentum [ see eqs . ( [ eq1]-[eq3 ] ) ] , as shown in fig . [ fig9 ] in appendix [ appa ] . this behavior is very different from the case in both semiconductors and single - layer graphene.@xcite when the electron density further increases , the intervalley electron - phonon scattering becomes important . this intervalley scattering is always in the weak scattering limit with @xmath148 and therefore @xmath137 according to eq . ( [ eq14 ] ) . as a result , the srt decreases with the increase of the intervalley scattering and hence the increase of the electron density.@xcite therefore , in the whole electron - density regime , a peak is observed . it is noted that this peak is very different from the one predicted by jiang and wu@xcite in semiconductors where the peak is attributed to the crossover from the nondegenerate - to - degenerate limit . as for the case at @xmath147k , the srt is always dominated by the intervalley electron - phonon scattering as shown in fig . [ fig2 ] by comparing the srt calculated with and without the intervalley electron - phonon scattering . since this intervalley scattering is in the strong scattering limit with @xmath149 , the srt @xmath140 $ ] from eq . ( [ eq14 ] ) . this srt increases with increasing electron density since both @xmath150 and the intervalley scattering time @xmath151 decrease with the increase of the electron density.@xcite -axis @xmath111 as function of the electron density @xmath99 with @xmath112 ( @xmath152 ) and without @xmath115 ( @xmath114 ) the intervalley electron - phonon scattering at @xmath153k.,width=321 ] -axis @xmath111 @xmath114 ( @xmath154 ) and the calculation without the coulomb hf term @xmath152 ( @xmath112 ) or the intervalley electron - phonon scattering @xmath113 ( @xmath115 ) as function of the initial spin polarization @xmath155 at @xmath153k . in the calculation , the electron density @xmath156@xmath106.,width=321 ] the initial spin polarization dependences of the srt along the @xmath103-axis at @xmath145 and @xmath157k are shown in fig . it is seen that the srt decreases rapidly with the increase of initial spin polarization @xmath155 at @xmath135k whereas the srt shows a slight increase with increasing @xmath155 at @xmath147k . the rapid decrease of the srt at @xmath135k is very different from the previous studies in both semiconductors@xcite and single - layer graphene@xcite where the srt increases significantly with increasing initial spin polarization . in the previous studies , the increase of the srt originates from the contribution of the coulomb hf term.@xcite this term serves as an effective magnetic field along the direction of the spin polarization , which blocks the spin precession induced by the dresselhaus@xcite and/or rashba@xcite socs and slows down the spin relaxation . however , the contribution of the coulomb hf term is marginal to the spin relaxation in our investigation , which is shown in fig . [ fig3 ] by comparing the the srt calculated with and without the coulomb hf term . here , the srt is determined by the intervalley electron - phonon scattering , as shown in fig . [ fig3 ] by comparing the calculation with and without the intervalley electron - phonon scattering . the intervalley electron - phonon scattering transfers electrons between the two valleys . the transferred electrons experience opposite effective magnetic fields induced by the zeeman - like term in the two valleys , which are not affected by the effective magnetic field induced by the coulomb hf term . moreover , the intervalley electron - phonon scattering is in the weak scattering limit at low temperature and hence the srt @xmath137 [ see eq . ( [ eq14 ] ) ] . the intervalley scattering time @xmath151 decreases significantly with the increase of the initial spin polarization as the density of states increases with the polarization . therefore , the srt shows a rapid decrease with the initial spin polarization at @xmath135k . as for the case of @xmath147k , the srt is also determined by the intervalley electron - phonon scattering in comparison with the case of @xmath135k . however , in contrast to the case of @xmath135k , the intervalley scattering is in the strong scattering limit . this makes the srt present an opposite trend to the case of @xmath135k . a mild increase of the srt results from the insensitivity of @xmath151 to the initial spin polarization at @xmath147k . @xmath158-axis @xmath111 @xmath159 as function of temperature @xmath118 . in the calculation , the electron density @xmath120@xmath106.,width=321 ] we also address the anisotropy of the spin relaxation with respect to the spin polarization direction . due to the existence of the out - of - plane effective magnetic field [ see eq . ( [ eq3 ] ) ] , the in - plane srts are identical.@xcite for comparison , we plot the temperature dependence of the srt along the @xmath103 ( @xmath127)-axis as the red curve with crosses ( blue curve with open squares ) in fig . it is seen that in comparison with the srt along the @xmath127-axis , the one along the @xmath103-axis is comparable at low temperature and orders of magnitude smaller at high temperature . as mentioned previously , the srt along the @xmath103-axis is strongly suppressed by the intervalley spin relaxation channel induced by the intervalley electron - phonon scattering and the zeeman - like term in the two valleys at high temperature . however , the contribution of this spin relaxation channel to the srt along the @xmath127-axis is marginal [ see eqs . ( [ eq1]-[eq2 ] ) ] . therefore , a strong anisotropy between the out - of- and in - plane spin relaxations is observed at high temperature . as function of temperature @xmath118 . @xmath116 ( @xmath114 ) corresponds to the case of the gate voltage v@xmath160v in the experiment of han and kawakami;@xcite @xmath112 ( @xmath115 ) stands for the case of electron density @xmath161@xmath106 in the experiment of avsar _ et al._;@xcite @xmath154 ( @xmath117 ) represents the case of @xmath162@xmath106 in the experiment of yang _ et al._;@xcite @xmath163 denotes the case of @xmath164@xmath106 in the experiment of neumann _ et al._.@xcite ( a ) red solid ( orange chain ) curve stands for our calculation corresponding to the case of v@xmath160v in the absence of the short - range scatterers . in the inset , we show the temperature dependence of the long - range impurity density @xmath165 corresponding to the case of v@xmath166v . ( b ) grey dashed and skyblue double - dotted chain curves are obtained from our calculations corresponding to the case of v@xmath166v by including the short - range scatterers , with a fixed short - range impurity density @xmath167@xmath106 and a tunable @xmath168 ( shown in the inset ) , respectively . we also show the dependence of @xmath169 on @xmath168 ( with the scale on the top of the frame ) at @xmath170k with other conditions the same as the case of v@xmath166v . the short - range potential strength @xmath171mevm@xmath172.,title="fig:",width=332 ] as function of temperature @xmath118 . @xmath116 ( @xmath114 ) corresponds to the case of the gate voltage v@xmath160v in the experiment of han and kawakami;@xcite @xmath112 ( @xmath115 ) stands for the case of electron density @xmath161@xmath106 in the experiment of avsar _ et al._;@xcite @xmath154 ( @xmath117 ) represents the case of @xmath162@xmath106 in the experiment of yang _ et al._;@xcite @xmath163 denotes the case of @xmath164@xmath106 in the experiment of neumann _ et al._.@xcite ( a ) red solid ( orange chain ) curve stands for our calculation corresponding to the case of v@xmath160v in the absence of the short - range scatterers . in the inset , we show the temperature dependence of the long - range impurity density @xmath165 corresponding to the case of v@xmath166v . ( b ) grey dashed and skyblue double - dotted chain curves are obtained from our calculations corresponding to the case of v@xmath166v by including the short - range scatterers , with a fixed short - range impurity density @xmath167@xmath106 and a tunable @xmath168 ( shown in the inset ) , respectively . we also show the dependence of @xmath169 on @xmath168 ( with the scale on the top of the frame ) at @xmath170k with other conditions the same as the case of v@xmath166v . the short - range potential strength @xmath171mevm@xmath172.,title="fig:",width=359 ] as mentioned in the introduction , experiments on spin relaxation have been carried out in blg on sio@xmath0 substrate@xcite or in freely suspended blg@xcite very recently by different groups . the in - plane srts in these experiments are of the order of @xmath1-@xmath2ns , which are comparable to our theoretical results at high temperature ( several hundred picoseconds ) . this motivates us to carry out a detailed comparison with the existing experiments . in these experiments , the gate voltage , temperature and electron density@xcite ( tuned by the gate voltage ) are given explicitly . however , some parameters such as the out - of - plane electric field and the impurity density are not available in the experiments . in our calculation , the out - of - plane electric field is taken to be the ratio of the gate voltage to the thickness of the substrate approximately.@xcite the impurity density is obtained by fitting to the mobility or the spin diffusion coefficient given in the experiments.@xcite in addition , we assume the initial spin polarization along the @xmath103-axis without loss of generality since the in - plane spin relaxations are identical as pointed out previously . we first compare the temperature dependence of the srts obtained from the experiments and calculation from our model . in these experiments , the mobilities are reported of the order of @xmath173-@xmath107@xmath108/(vs ) . the srts from different groups are comparable as shown in fig . however , the srt from the experiment of han and kawakami@xcite decreases with increasing temperature whereas the ones from the experiments of yang _ et al._@xcite and avsar _ et al._@xcite show a marginal temperature dependence . in addition , han and kawakami@xcite reported the longest srt among these experiments , indicating that their experimental data may contain less extrinsic effects . motivated by this , we show our comparison with the experiment of han and kawakami@xcite in fig . [ fig5](a ) . the experimental data with the gate voltage v@xmath160v corresponding to the electron density @xmath174@xmath106 , is labeled as @xmath175 . by fitting to the spin diffusion coefficient given in the experiment , we obtain the long - range impurity density @xmath165 shown in the inset as the curve with @xmath113 . in contrast to the long - range impurity scattering , the contribution of other scatterings to the spin diffusion coefficient ( mobility ) is marginal . then we calculate the srt corresponding to v@xmath160v shown as the red solid ( orange chain ) curve . a crossover is observed in the temperature dependence of the srt , similar to the one shown in sec . it is seen that the srt from our calculation is comparable to the experimental data at high temperature . this suggests that the intervalley spin relaxation channel induced by the intervalley electron - phonon scattering and the zeeman - like term in the two valleys addressed in sec . iiia , plays an important role at high temperature . nevertheless , the srts are still larger than the experimental data with an observable difference at high temperature . additionally , at low temperature , our results are orders of magnitude larger than the experimental ones since the intervalley electron - phonon scattering is marginal at low temperature and the srt is determined by the long - range electron - impurity scattering in our calculation . due to the existence of all these discrepancies , one has to take into account extrinsic effects such as adatoms,@xcite curvature,@xcite substrate effects@xcite and also contacts.@xcite in the above experiments on blg , the spin polarized electrons are all injected from ferromagnetic contacts.@xcite as reported , the ferromagnetic electrodes generate stray fields , which cause spin precession around the contact.@xcite as a result , the stray fields may have a strong influence on the observed experimental results . additionally , electron scattering by the tunneling contacts may also be responsible for the available experimental data.@xcite in addition to the above extrinsic effects , the short - range scatterers may also be an extrinsic source.@xcite here , we discuss the contribution of short - range scatterers@xcite in detail . the scattering term induced by the short - range scatterers is given by @xmath176 with @xmath168 and @xmath177 denoting the short - range impurity density and the constant short - range potential strength , respectively.@xcite this scattering term contributes to both the intra- and inter - valley scatterings . the intervalley scattering opens an intervalley spin relaxation channel together with the zeeman - like term in the two valleys [ see eq . ( [ eq3 ] ) ] , similar to the intervalley electron - phonon scattering . however , in contrast to the intervalley electron - phonon scattering , the intervalley scattering induced by the short - range scatterers is _ insensitive _ to the temperature and may play an important role in the in - plane spin relaxation especially at low temperature . here , we choose a fixed short - range impurity density @xmath167@xmath106 and potential strength @xmath171mevm@xmath172 . it is noted that although with this additional short - range scattering included , the mobility is still dominated by the long - range electron - impurity scattering since the contribution of the long - range electron - impurity scattering is about four orders of magnitude larger than that of the short - range one . we calculate the temperature dependence of the srt corresponding to the case of v@xmath166v in the experiment of han and kawakami@xcite by including the additional short - range scattering . the result is plotted as the grey dashed curve in fig . [ fig5](b ) . a crossover is also observed at high temperature due to the contribution of the intervalley electron - phonon scattering as pointed out previously . by comparing our calculation with the experimental data labeled as @xmath116 , we find that our result becomes comparable at low temperature . furthermore , by fitting to the experimental data with tunable short - range impurity density shown as the curve with @xmath152 in the inset , our result labeled as the skyblue double - dotted chain curve agrees fairly well with the experimental data in the temperature regime lower than the crossover . as for the temperature higher than the crossover , our result becomes even larger than the experimental one once the short - range scattering is included . this is because that the intervalley electron - phonon scattering is in the strong scattering limit . with the inclusion of the short - range scattering , the intervalley scattering becomes stronger and the srt becomes larger . this discrepancy may be attributed to other extrinsic effects . as mentioned above , the srt can be tuned by the short - range impurity density . we show the short - range impurity density dependence of the in - plane srt at @xmath170k as the purple dotted curve in fig . [ fig5](b ) , with all other conditions remaining the same as the case of v@xmath166v . a valley is also observed , similar to the case of the temperature dependence discussed above . at low temperature , the intervalley electron - phonon scattering is negligible . the srt is then determined by the intervalley short - range scattering . the valley originates from the crossover between the weak and strong intervalley short - range scattering limit by tuning the short - range impurity density . at the crossover point , the srt @xmath178ps according to eq . ( [ eq15 ] ) , close to the value @xmath179ps shown in the figure . this crossover point gives the minimum of the srt and can be taken as a criterion for determining the possible contribution of the short - range scattering . specifically , the crossover point @xmath180 , @xmath181 , @xmath182 , @xmath183 , and @xmath184ps corresponding to the case of the experimental data shown as @xmath114 , @xmath112 , @xmath115 , @xmath154 and @xmath117 , respectively . we find that both the experimental data of han and kawakami@xcite and avsar _ et al._@xcite are larger than the crossover point whereas the one of yang _ et al._@xcite is smaller . this suggests the experimental data of han and kawakami@xcite and avsar _ et al._@xcite can be explained by the short - range scatterers . as for the experiment of yang _ et al._,@xcite some other extrinsic effects have to be considered . as function of electron density . the dotted curves with symbols correspond to the experimental data of han and kawakami ( hk)@xcite at @xmath185k and avsar _ et al._@xcite and yang _ et al._@xcite at @xmath186k . the solid curves with symbols correspond to the experimental data at room temperature of han and kawakami,@xcite avsar _ et al._@xcite and yang _ et al._.@xcite red solid ( blue chain ) curve is from our calculation corresponding to the case of room ( low ) temperature in the experiment of han and kawakami.@xcite it is noted that the blue chain curve is calculated with the short - range scatterers with the short - range impurity density @xmath187@xmath106 and potential strength @xmath171mevm@xmath172.,width=321 ] then we turn to compare the electron density dependence of the srts from the experiments and our calculation . among the experiments , han and kawakami@xcite measured the longest srt at low temperature ( up to several nanoseconds ) shown as the dotted curve with @xmath114 in fig . this srt is about one order of magnitude larger than the ones from other experimental cases . in addition , the srts from the experiments show different dependences on electron density . specifically , at low temperature , both avsar _ et al._@xcite and yang _ et al._@xcite show a decrease of the srt with the increase of the electron density whereas a mild peak is observed by han and kawakami.@xcite at room temperature , the srts from avsar _ et al._@xcite and yang _ et al._@xcite increase with increasing electron density whereas han and kawakami@xcite reported an insensitive density dependence of the srt . we first show our comparison with the experimental data at low temperature . as shown in fig . [ fig5](a ) , our result is orders of magnitude larger than the experimental data of han and kawakami@xcite at low temperature without short - range impurities . however , with the inclusion of the short - range scatterers , our result shows good agreement with the experimental data . here , with the short - range scatterers explicitly included , we also show our comparison with the experimental data of han and kawakami.@xcite with a fixed short - range impurity density @xmath187@xmath106 and potential strength @xmath171mevm@xmath172 , the electron density dependence of the srt from our calculation is shown as blue chain curve in fig . we find that our result marginally agrees with the experimental data shown as the dotted curve with @xmath114 . better agreement can be obtained with tunable short - range impurity density . in addition , we also find that the experimental data of avsar _ et al._@xcite can be explained by the short - range scatterers . however , to account for the experimental data of yang _ et al._,@xcite one has to include other extrinsic effects . this is because the srt of yang _ et al._@xcite is smaller than the crossover point in the short - range impurity density dependence of the srt ( @xmath188-@xmath189ps ) , which is taken as a criterion for the possible contribution of the short - range scattering as mentioned previously . we then address our comparison with the experimental data at room temperature . as shown in fig . [ fig5](a ) , our result is larger than the experimental data of han and kawakami@xcite with an observable difference at room temperature without short - range impurities . with the short - range scatterers included , the srt from our calculation becomes even larger than the experimental data as mentioned previously , indicating that other extrinsic effects have to be taken into account . here , we show our comparison with the experiment of han and kawakami@xcite without extrinsic effects . the experimental data and our result are shown as the solid curve with @xmath113 and the red solid curve in fig . [ fig6 ] , respectively . we find that the srt in our calculation is about three times as large as the experimental one in low electron density regime and one order of magnitude larger than the experimental one in high density regime . in summary , we have investigated the electron spin relaxation due to the dp mechanism in blg with only the lowest conduction band being relevant . the soc of the lowest conduction band is constructed from the symmetry group analysis with the parameters obtained by fitting to the numerical calculation according to the latest report by konschuh _ et al._@xcite from first principles with both the intrinsic and extrinsic soc terms in the pseudospin space included . we find that the magnitudes of both the out - of- and in - plane components of the soc decrease with increasing momentum at large momentum , indicating a suppression of the inhomogeneous broadening with the increase of the momentum . this is different from the case in both semiconductors and single - layer graphene . additionally , the leading term of the out - of - plane component of the soc serves as a zeeman - like term with opposite effective magnetic fields in the two valleys , which is similar to the case in rippled single - layer graphene . this zeeman - like term , together with the intervalley electron - phonon and/or possible intervalley short - range scatterings , opens an intervalley spin relaxation channel , which has not been reported in the literature in blg . the intervalley electron - phonon scattering is derived by using the tight - binding model . in addition to the intervalley electron - phonon and short - range scatterings , we also include the long - range electron - impurity , electron - electron coulomb and intravalley electron - phonon scatterings to calculate the srt . we find that the in - plane srt is strongly suppressed by the intervalley electron - phonon scattering at high temperature . in contrast to the intervalley electron - phonon scattering , the intervalley short - range scattering is insensitive to the temperature and plays an important role in the in - plane spin relaxation especially at low temperature . in the absence of short - range scatterers , a marked nonmonotonic temperature dependence of the in - plane srt is predicted with a minimum srt down to several hundred picoseconds . this nonmonotonic behavior originates from the crossover between the weak and strong intervalley electron - phonon scattering . moreover , we predict a peak in the electron density dependence of the in - plane srt at low temperature , which is very different from the one in semiconductors . at high temperature , the in - plane srt increases monotonically with increasing density . we also find that the in - plane srt decreases rapidly with the increase of the initial spin polarization at low temperature . this is very different from the previous studies in both semiconductors and single - layer graphene where the srt increases significantly with increasing initial spin polarization . the physics is understood that the spin relaxation time is determined by the intervalley electron - phonon scattering , which transfers electrons between the two valleys . the transferred electrons experience opposite effective magnetic fields in the two valleys , which are not affected by the coulomb hf term . in addition , a strong anisotropy between the out - of- and in - plane spin relaxations is also addressed at high temperature where the out - of - plane srt is about two orders of magnitude larger than the in - plane one . as for low temperature , the out - of- and in - plane spin relaxation times are comparable . we also show our comparison with the existing experiments of han and kawakami,@xcite avsar _ et al._@xcite and yang _ et al._.@xcite we find that without intervalley scattering , the srt is orders of magnitude larger than the experimental data in the whole temperature regime . with the intervalley electron - phonon scattering explicitly included , the srt from our calculation becomes comparable to the experimental data at high temperature but still orders of magnitude larger than the experimental data at low temperature . in addition , a crossover in the temperature dependence of the in - plane srt is shown , which also results from the crossover between the weak and strong intervalley electron - phonon scattering . with the inclusion of the short - range scatterers , our result agrees fairly well with the experimental data in the temperature regime lower than the crossover . as for the temperature higher than the crossover , other extrinsic effects have to be included . moreover , a crossover point is also shown in the short - range impurity density dependence of the in - plane srt , which gives a minimum for determining the possible contribution of the short - range scattering . the experimental srt larger than the minimum can be explained by the possible short - range scatterers whereas other extrinsic effects have to be considered for experimental data smaller than the minimum . this work was supported by the national basic research program of china under grant no . 2012cb922002 and the strategic priority research program of the chinese academy of sciences under grant no . one of the authors ( l.w . ) would like to thank b. y. sun for double checking the derivation of the electron - phonon scattering matrix elements . , ( b ) @xmath190 and ( c ) @xmath191 as function of the out - of - plane electric field @xmath36 with @xmath192-@xmath92 . it is noted that @xmath193 , @xmath194 , @xmath195 , @xmath196 and @xmath197 ( @xmath14-@xmath15 ) are dimensionless.,title="fig:",width=257 ] , ( b ) @xmath190 and ( c ) @xmath191 as function of the out - of - plane electric field @xmath36 with @xmath192-@xmath92 . it is noted that @xmath193 , @xmath194 , @xmath195 , @xmath196 and @xmath197 ( @xmath14-@xmath15 ) are dimensionless.,title="fig:",width=257 ] , ( b ) @xmath190 and ( c ) @xmath191 as function of the out - of - plane electric field @xmath36 with @xmath192-@xmath92 . it is noted that @xmath193 , @xmath194 , @xmath195 , @xmath196 and @xmath197 ( @xmath14-@xmath15 ) are dimensionless.,title="fig:",width=257 ] the soc term of the lowest conduction band near the dirac points can be described by a @xmath198 matrix @xmath199 in the basis @xmath200 where @xmath201 and @xmath202 ( @xmath203 ) . the coefficients @xmath204 can be obtained by exactly diagonalizing the @xmath20 effective hamiltonian given in eq . ( [ eq5 ] ) . here , @xmath205 with @xmath206 ( @xmath207 ) being the spin - orbit matrix elements near the dirac points in the pseudospin space given by konschuh _ et al._.@xcite specifically , @xmath208+\frac{\mu-1}{2}[{c_3^{\mu}}^*c_1^{\mu}(i\lambda_4 ) \nonumber\\ & & \mbox{}\hspace{-0.4cm}+{c_1^{\mu}}^*c_2^{\mu}(-i\lambda_0)+{c_4^{\mu}}^*c_2^{\mu}(-i\lambda_4^{\prime})+{c_2^{\mu}}^*c_3^{\mu}(-i\lambda_3)\nonumber\\ & & \mbox{}\hspace{-0.4cm}+{c_3^{\mu}}^*c_4^{\mu}(i\lambda_0^{\prime})].\label{a7}\end{aligned}\ ] ] @xmath209 and @xmath210 are the strengths of single - layer - like intrinsic soc ; @xmath211 and @xmath212 represent single - layer - like extrinsic spin - orbit strengths ; @xmath213 , @xmath214 and @xmath215 stand for interlayer spin - orbit parameters.@xcite then the effective magnetic field of the soc @xmath216 in eq . ( [ eq6 ] ) is given by @xmath217 , @xmath218 and @xmath219 . it is noted that the contribution of other energy bands to the soc of the lowest conduction band is marginal in our calculation . and ( b ) @xmath220 as function of the out - of - plane electric field @xmath36 with @xmath192-@xmath221 . it is noted that @xmath222 , @xmath223 , @xmath224 and @xmath225 ( @xmath226 ) are dimensionless.,title="fig:",width=283 ] and ( b ) @xmath220 as function of the out - of - plane electric field @xmath36 with @xmath192-@xmath221 . it is noted that @xmath222 , @xmath223 , @xmath224 and @xmath225 ( @xmath226 ) are dimensionless.,title="fig:",width=283 ] as reported by konschuh _ et al._,@xcite the small group of the dirac points is @xmath4 in the presence of an out - of - plane electric field . following from the symmetry group @xmath4,@xcite we derive an analytical form of the soc near the dirac points [ see eqs . ( [ eq1]-[eq3 ] ) ] . in these equations , the coefficients @xmath13 ( @xmath14-@xmath15 ) and @xmath227 ( @xmath14-@xmath110 ) are obtained by fitting to the numerical calculation from eqs . ( [ a5 ] ) and ( [ a7 ] ) using the pad approximation.@xcite specifically , the coefficients @xmath13 ( @xmath14-@xmath15 ) in eqs . ( [ eq1]-[eq2 ] ) read @xmath228 the coefficients @xmath227 ( @xmath14-@xmath110 ) in eq . ( [ eq3 ] ) are given by @xmath229 here , the coefficient @xmath230ev , which is independent of the applied electric field . the electric field dependences of @xmath231 ( @xmath14-@xmath15 , @xmath192-@xmath92 ) and @xmath232 ( @xmath14-@xmath110 , @xmath192-@xmath221 ) are plotted in figs . [ fig7 ] and [ fig8 ] , respectively . it is noted that the pad approximation@xcite can give a precise description of @xmath13 ( @xmath14-@xmath15 ) and @xmath227 ( @xmath14-@xmath110 ) at large momenta near the dirac points , which can be populated and play an important role in spin relaxation according to the experimental conditions ( heavily doped).@xcite with only a simple linear @xmath10-order approximation , @xmath13 and @xmath227 agree with the numerical results only at very small momentum near the dirac points . we also show in fig . [ fig9 ] the spin splitting of the lowest conduction band calculated with analytical form of the soc ( blue dashed curve ) and with the explicit numerical one ( red solid curve ) . we find that the analytical result agrees fairly well with the numerical one at large momentum . in addition , the spin splitting decreases with increasing momentum at large momentum , which is very different from the case in both semiconductors@xcite and single - layer graphene.@xcite near the k point as function of the momentum @xmath233 under a typical electric field @xmath104v / nm . solid ( dashed ) curve represents the numerical ( analytical ) result . the electron - electron coulomb scattering matrix element is given by @xmath235 . the screened coulomb potential @xmath236 with @xmath237 being the screening under the random phase approximation.@xcite @xmath238 is the two - dimensional bare coulomb potential with @xmath239 and @xmath240 being the fermi velocity in single - layer graphene@xcite and the dimensionless wigner - seitz radius,@xcite respectively . @xmath241 is given by@xcite @xmath242 where @xmath243 . the long - range electron - impurity scattering matrix element @xmath244 , in which @xmath245 and @xmath246 stand for the impurity charge number and effective distance of the impurity layer to the blg sheet.@xcite it is noted that when we calculate the electron - electron coulomb and long - range electron - impurity scatterings , we take the distance between the graphene layers to be zero approximately.@xcite it is further noted that the coulomb potential in the hf term is in the static screening limit , i.e. , @xmath247 . then we turn to the electron - phonon scattering matrix elements . for the intravalley electron - ac - phonon scattering , the scattering matrix element @xmath248 where @xmath249 ; @xmath250 and @xmath251 represent the deformation potential@xcite and acoustic phonon velocity,@xcite respectively . for the intravalley electron - ri - phonon scattering , @xmath252 with @xmath253 being the thomas - fermi screening length.@xcite for sio@xmath0 substrate , the energy spectra of the two remote phonon modes are denoted as @xmath254 and @xmath255 , respectively ; @xmath256 and @xmath257 are the corresponding dimensionless coupling parameters.@xcite the matrix elements for the intravalley electron - op - phonon scattering ( @xmath66 ) are derived using the tight - binding model according to the arxiv version of ref . . the intravalley electron - op - phonon scattering includes both the electron in - plane ( @xmath258 ) and out - of - plane ( @xmath259 ) op phonon scatterings . the electron in - plane op phonon scattering matrix element is given by @xmath260\psi^{\mu}_{{\bf k}^{\prime}}|^2+|{\psi^{\mu}_{{\bf k}}}^{\dagger } [ i\gamma_0^{\prime}(\sin\theta_{\bf q}\sigma^{01}_{\rm d}-\cos\theta_{\bf q}\nonumber\\ & & \mbox{}\hspace{-1.9cm}\times\sigma^{02}_{\rm d})+a\gamma_4^{\prime}(i\cos\theta_{\bf q}\gamma^3_{\rm d}- \sin\theta_{\bf q}\gamma^3_{\rm d}\gamma^5_{\rm d})/(\sqrt{3}l_4)]\psi^{\mu}_{{\bf k}^{\prime}}|^2\nonumber\\ & & \mbox{}\hspace{-1.9cm}+|{\psi^{\mu}_{{\bf k}}}^{\dagger}[\gamma_0^{\prime}(\cos\theta_{\bf q}\sigma^{23}_{\rm d } -\sin\theta_{\bf q}\sigma^{13}_{\rm d})-a\gamma_3^{\prime}(\cos\theta_{\bf q}\gamma^1_{\rm d}\gamma^5_{\rm d}\nonumber\\ & & \mbox{}\hspace{-1.9cm}-i\cos\theta_{\bf q}\gamma^2_{\rm d}-i\sin\theta_{\bf q}\gamma^1_{\rm d}-\sin\theta_{\bf q}\gamma^2_{\rm d } \gamma^5_{\rm d})/(2\sqrt{3}l_3)]\psi^{\mu}_{{\bf k}^{\prime}}|^2\nonumber\\ & & \mbox{}\hspace{-1.9cm}+|{\psi^{\mu}_{{\bf k}}}^{\dagger}[i\gamma_0^{\prime}(\cos\theta_{\bf q}\sigma^{01}_{\rm d } + \sin\theta_{\bf q}\sigma^{02}_{\rm d})-a\gamma_4^{\prime}\gamma^3_{\rm d}(i\sin\theta_{\bf q}\nonumber\\ & & \mbox{}\hspace{-1.9cm}+\cos\theta_{\bf q}\gamma^5_{\rm d})/(\sqrt{3}l_4)]\psi^{\mu}_{{\bf k}^{\prime}}|^2\big\},\end{aligned}\ ] ] where @xmath261 is the energy of the in - plane op phonon modes;@xcite @xmath262 is the bond length corresponding to the interlayer hopping @xmath31 ; and @xmath263 and @xmath264 are @xmath20 dirac matrices given in appendix [ appc].@xcite here , @xmath265 is the polar angle of the momentum @xmath266 . the electron out - of - plane op phonon scattering matrix element reads @xmath267 being the energy of the out - of - plane op phonon mode.@xcite the relevant dirac matrices@xcite used in the electron - phonon scattering are @xmath268 @xmath269 @xmath270 with @xmath271 being @xmath198 unit matrix . yang , j. balakrishnan , f. volmer , a. avsar , m. jaiswal , j. samm , s. r. ali , a. pachoud , m. zeng , m. popinciuc , g. gntherodt , b. beschoten , and b. zyilmaz , phys . lett . * 107 * , 047206 ( 2011 ) . it is noted that due to the complicated energy spectrum of blg , we employ the single - layer graphene approximation to calculate the spin diffusion coefficient in blg according to the report by zhang and wu.@xcite

we study the x ray spectral variability of the narrow line seyfert 1 galaxy ngc 4051 as observed during two _ xmm newton _ observations . to gain insight on the general behaviour , we first apply model independent techniques such as rms spectra and flux - flux plots . we then perform time resolved spectral analysis by splitting the observations into 68 spectra ( 2 ks each ) . the data show evidence for a neutral and constant reflection component and for constant emission from photoionized gas , which are included in all spectral models . the nuclear emission can be modelled both in terms of a `` standard model '' ( pivoting power law plus a black body component for the soft excess ) and of a two component one ( power law plus ionized reflection from the accretion disc ) . both models reproduce the source spectral variability and can not be distinguished on a statistical ground . the distinction has thus to be made on a physical basis . the standard model results indicate that the soft excess does not follow the standard black body law ( l ) despite a variation in luminosity by about one order of magnitude . the resulting temperature is consistent with being constant and has the same value as observed in pg quasars . moreover , although the spectral slope is correlated with flux , which is consistent with spectral pivoting , the hardest photon indexes are so flat (.31.4 ) as to require rather unusual scenarios . furthermore , the very low flux states exhibit an invertedflux behaviour which disagrees with a simple pivoting interpretation . these problems can be solved in terms of the two component model in which the soft excess is not thermal , but due to the ionized reflection component . in this context , the power law has constant slope ( about 2.2 ) and the slope flux correlation is explained in terms of the relative contribution of the power law and reflection components which also explains the shape of the flux flux plot relationship . the variability of the reflection component from the inner disc closely follows the predictions of the light bending model , suggesting that most of the primary nuclear emission is produced in the very innermost regions , only a few gravitational radii from the central black hole . [ firstpage ] galaxies : individual : ngc 4051 galaxies : active galaxies : seyfert x - rays : galaxies

astro-ph0602191

ngc 4051 is a nearby ( z=0.0023 ) low luminosity narrow line seyfert 1 ( nls1 ) galaxy which exhibits extreme x ray variability in flux and spectral shape on both long and short timescales . the source sometimes enters relatively long and unusual low flux states in which the hard spectrum becomes very flat ( @xmath3 ) while the soft band is dominated by a much steeper component ( @xmath4 , or blackbody with @xmath5 kev ) . most remarkable is the 1998 _ bepposax _ observation reported by guainazzi et al ( 1998 ) in which the source reached its minimum historical flux state , the spectral slope was @xmath6 , and the overall spectrum was best explained by assuming that the nuclear emission had switched off leaving only a reflection component from distant material ( clearly visible in the hard spectrum ) which still responded to previous higher flux levels . a different interpretation assumes that the nuclear emission in low flux states originates so close to the central black hole that only a tiny fraction can escape the gravitational pull of the hole so that the nuclear continuum is virtually undetectable at infinity ( e.g. miniutti & fabian 2004 ) . in general , the 210 kev spectral slope appears to be well correlated with flux . however , the correlation is not linear with the slope hardening rapidly at low fluxes and reaching an asymptotic value at high fluxes ( see e.g. lamer et al 2003 , based on _ rxte _ monitoring data over about three years ) . this behaviour is relatively common in other sources as well , such as mcg6 - 30 - 15 , 1h 0419577 , ngc 3516 ( among many others ) and might be due to flux correlated variations of the power law slope produced in a corona above an accretion disc as originally proposed by haardt & maraschi ( 1991 ; 1993 ) and haardt et al . ( 1994 ) and related to changes in the input soft seed photons ( e.g. maraschi & haardt 1997 , haardt , maraschi & ghisellini 1997 , poutanen & fabian 1999 , zdziarski et al 2003 ) . on the other hand , such slope flux behaviour can be explained in terms of a two component model ( mchardy et al 1998 ; shih , iwasawa & fabian 2002 ) in which a constant slope power law varies in normalisation only , while a harder component remains approximately constant hardening the spectral slope at low flux levels only , when it becomes prominent in the hard band . in the well studied case of mcg6 - 30 - 15 , detailed spectral and variability analysis ( e.g. fabian & vaughan 2003 ; vaughan & fabian 2004 ) allowed the presence of two main spectral components to be disentangled , namely a variable power law component ( plc ) with constant slope , and a reflection dominated component ( rdc ) from the inner accretion disc which does not follow the plc variation , confirming for this particular case the validity of the two component model and ruling out spectral pivoting as the main source of spectral variability . the contribution from a distant reflector is limited to 10 per cent or less ( see e.g. the long _ chandra _ observation reported by young et al 2005 ) i.e. the rdc is completely dominated by lowly ionized reflection from the inner disc , including the broad relativistic fe line . from a theoretical point of view , the lack of response of the disc rdc to the plc variations has been interpreted in terms of strong gravitational light bending by miniutti et al ( 2003 ) and miniutti & fabian ( 2004 ) , which predicts this behaviour during normal flux periods and gives a correlation at low flux levels only . it should be stressed that , in the framework of the two component model , the 210 kev slope flux correlation in ngc 4051 can not entirely be explained by the contribution of a distant reflector in the hard band ( the presence of which is required by the detection of a narrow component to the iron line ) . indeed , as shown by lamer et al ( 2003 ) the correlation is still present even if the distant reflector contribution is taken into account . even at low flux levels , soft and hard components are significantly variable and well correlated , which excludes any dominant extended emission ( torus and/or extended scattering region ) . extended emission is undetected by _ chandra _ on @xmath7 100 pc scales ( collinge et al 2001 ; uttley et al 2003 ) . therefore , the residual slope flux correlation can be due to intrinsic pivoting or to an additional , almost constant , spectral component ( i.e. rdc from the disc , as for mcg6 - 30 - 15 ) . throughout the paper for `` two component model '' we mean a parametrization of the nuclear emission in terms of two components . other spectral components such as absorption , emission from photoionized gas , and reflection from distant matter can be present but do not represent nuclear emission . other important clues on the nature of the spectral variability in ngc 4051 ( and other sources as well ) comes from the so called flux flux plot analysis , first presented by churazov , gilfanov & revnivtsev ( 2001 ) to demonstrate the stability of the disc emission in the high / soft state of cyg taylor , uttley & mchardy ( 2003 ) have applied the same technique to a few seyfert galaxies by using _ rxte _ data . in the case of ngc 4051 , uttley et al ( 2004 ) have performed such an analysis by using the same _ xmm newton _ data we are presenting here and showed that the distribution of high , intermediate , and low flux state data points are smoothly joined together , indicating that the same process causing the spectral variability in high and intermediate flux states continues to operate even at the very low flux levels probed by the _ xmm newton _ data ( uttley et al 2004 ) . this evidence seems to challenge the interpretation of the spectral variability as due to variable absorption by a substantial column of photoionized gas ( pounds et al . 2004 ) . here we report results based on the two _ xmm newton _ observations of ngc 4051 . the data have been previously analysed by uttley et al ( 2004 ) and pounds et al ( 2004 ) who reach different conclusions on the nature of the spectral variability in ngc 4051 and on the main spectral components that come into play . our analysis is complementary to previous studies and offers a novel point of view that may also be relevant for other similar sources . ngc 4051 has been observed twice by _ xmm newton_. the first observation took place during revolution 263 ( 2001 may ) for @xmath7117 ks , the second during revolution 541 ( 2002 november ) for @xmath752 ks . these data are public and have been obtained from the data archive . the analysis has been made with the latest version of the sas software , starting from the odf files . during the first observation the epic pn and the mos2 were operated in small window mode with the medium filter . no significant pile - up is present in the pn data , so single and double pixel events were selected . the mos2 data show @xmath71% of pile - up when the count rate is higher than 5 counts / s ( during the first observation it sometimes reaches 15 counts / s ) . this could invalidate the results of time resolved spectral variability , because , even if pile - up is not strong , its effect is increasing with flux . thus , for the time resolved spectral analysis , we use the mos2 data selecting only single events and extracting the source photons within a circular region of 30@xmath8 radius when the count rate is lower than 5 counts / s , and within an annular region with 7@xmath8 and 30@xmath8 radii , during the higher count rate intervals . the mos1 data were in fast uncompressed mode and , due to cross calibration uncertainty , we did not used those . the second observation was timed to coincide with an extended period of low x ray emission from . during this observation the pile - up is negligible in all the epic cameras and single plus double pixel events are used for mos1 , mos2 and pn cameras . the background spectrum has been taken from source free regions in the same chip as the source and ancillary and response files have been created . after filtering out periods of high background , we obtain net exposures of about 97 ks ( @xmath769 ks for the pn small window mode ) and 39 ks for the first and second observation respectively . at the moment the pn camera is not well calibrated below 0.5 kev . for the data we are presenting here , for example , the ratio between the mos and pn camera below 0.5 kev is of about 20 per cent . thus , the data from the pn camera are used in the 0.510 kev band only , while mos data are used down to 0.2 kev . all spectral fits include absorption due to the line - of - sight galactic column of @xmath9=1.32@xmath1010@xmath11 @xmath12 ( elvis et al . in the subsequent spectral analysis , errors are quoted at the 90 per cent confidence level ( = 2.7 for one interesting parameter ) . the _ xmm newton _ broadband light curve of ngc 4051 exhibits large amplitude count rate variations in both observations , typical for this source and for nls1 galaxies in general . fast large amplitude variability is superimposed on longer trends which are mainly characterized by persistent low flux periods in which variability is suppressed . the _ xmm newton _ light curves shown in fig . [ lcs ] are a good representation of this behaviour which can be described in more general terms by the rms flux relation in which the absolute rms variability is proportional to the source flux ( uttley & m@xmath13hardy 2001 ) . light curves such as those shown in fig . [ lcs ] are just a realization of the rms flux relation which naturally produces large amplitude spikes at high fluxes and periods of relative quiescence at low fluxes ( e.g. uttley , m@xmath13hardy & vaughan 2005 ) . the general shape of the broadband spectrum of ngc 4051 in both observations can be roughly described by the presence of a power law continuum , reflection from distant matter including a narrow 6.4 kev fe line , and a prominent soft excess below about 1 kev . the fe line is unresolved and fluxes are consistent with each other in the two observations . a warm absorber imprints its presence in the spectrum particularly through two edge like features at 0.74 and 0.87 kev ( o and o ) . even modelling the absorption , some structures persist between 0.8 kev and 1 kev , particularly evident in the low flux state . this further structure can be interpreted as a broad absorption structure ( as suggested by pounds et al 2004 ) , emission around 0.9 kev ( see collinge et al 2001 , uttley et al 2004 ) , or a combination of the two . spectral variability is clearly present between the two observations with the source being much harder in the second , lower flux observation . moreover , when the source is faint and hard , the spectrum above @xmath7 2 kev appears to be more curved than in the high state . to illustrate the differences in spectral slope and hard curvature between the two observations we show the two 310 kev time averaged spectra in the top panel of fig . [ test ] . the best fitting power law slope is indicated , the fit including a narrow 6.4 kev fe line as well . the spectral curvature can be interpreted as the effect of absorption by a relatively high column of ionized gas or could be the signature of a relativistically blurred reflection component from the accretion disc . two main mechanisms have been proposed so far to explain the spectral variability of ngc 4051 . pounds et al ( 2004 ) performed a comparative analysis of the two _ xmm newton _ observations based on the time averaged spectra and on the simultaneous high resolution rgs data in the soft band . their main conclusions is that the spectral variability between the high flux ( rev . 263 ) and low flux ( rev . 541 ) observations is mainly due to variable absorption by a substantial column of photoionized gas in the line of sight . the authors suggest that in the low state ( observed about 20 days after the source entered one of its characteristic low state periods ) part of the gas had recombined in response to the extended period of lower x ray flux so providing a partial coverer ( with @xmath14 @xmath12 ) . the variable opacity of the absorber could then explain in a natural way both the flat spectrum and the hard curvature that is observed in the second low flux observation . this interpretation is challenged by the analysis by uttley et al ( 2004 ) who proposed spectral pivoting as the main driver of the spectral variability in ngc 4051 and pointed out that a variation in the absorber opacity in the line of sight during the low state is not consistent with the nature of the spectral variability , as revealed by flux flux plot analysis , in which no sharp transition is seen between the two observations . the flux flux relationship rather indicates that the spectral variability is flux dependent and the low flux states of the first observation are virtually identical in spectral shape to the high flux state of the second . this would rule out that the low flux spectral shape during the second observation depends on the fact that data were obtained following an extended low flux state allowing the absorbing gas to recombine . it is clear that the pounds et al interpretation implies that the spectral shape is different in the two observations . for instance , the low flux states of the first ( high flux ) observation should not be as flat and curved as the spectrum in the second observation . this idea can be easily checked by comparing two spectra : the first one is representative of the very low flux state during the first observation ( around 60 ks , see fig . [ lcs ] ) and the second one is extracted from the second observation . we choose the two spectra to have the same exposure and approximately the same count rate . the two spectra are shown in the bottom panel of fig . [ test ] in the hard band . the best fitting power law slope is also indicated . the fit also includes fe emission at 6.4 kev . firstly , the spectral slope is exactly the same in the two spectra . therefore the slope is count rate dependent and the time averaged flat spectrum during the low flux observation has thus little to do with the extended period of low flux preceding the second observation . secondly , the two spectra also seem to have approximately the same hard curvature . when fitted with a partial covering model , both spectra require a large column density of @xmath15@xmath16 @xmath12 and a covering fraction of 6070 per cent . this demonstrates that not only the apparent spectral slope is flux dependent , but also that the hard curvature depends on flux only . from the above analysis , it seems likely that the spectral variability in ngc 4051 is simply flux dependent and that the second low flux observation is not peculiar but indeed identical to low flux states during the first observation . as already mentioned by uttley et al ( 2004 ) one possible explanation for the flux dependent spectral variability is spectral pivoting . in this paper we shall also explore an alternative explanation invoking a reflection component from the accretion disc . before considering time resolved spectral analysis , we perform some model independent tests with the aim of identifying the main spectral components and their relative contribution to the spectrum and variability , in order to have a guide for our subsequent analysis . we start our analysis of the spectral variability of by calculating the rms spectrum , which measures the total amount of variability as a function of energy ( edelson et al . 2002 ; vaughan et al . 2004 ; ponti et al . [ rms ] shows the rms spectra during rev . 263 and 541 . the rms spectra have been calculated with time bins of @xmath72 ks ( the same as the subsequent time resolved spectral analysis ) and with energy bins chosen in order to have at least 300 counts per energy time bin . in this way the poissonian noise is negligible compared to the rms value and thus the use of the uncertainty given in ponti et al . ( 2004 ) is valid . during the high flux observation of rev . 263 ( fig . [ rms ] ) the variability rapidly increases toward softer energies , i.e. the broadband emission tends to steepen as it brightens . some reduction in the variability is seen below about 800 ev followed by a plateau below 500 ev . a similar rms shape has been observed in other sources ( ponti et al . 2005 ; vaughan et al . 2003 ; vaughan et al . 2004 ) and the two simplest explanations for the broadband trend invoke either spectral pivoting of the variable component , or the two component model ( see e.g markowitz , edelson & vaughan 2003 ) . another important feature of the rms spectrum is a drop of variability at 6.4 kev , the energy of the narrow fe line which is seen in the time averaged spectrum . such a drop shows that the narrow fe line ( and therefore the associated reflection continuum ) is much less variable than the continuum ( if variable at all ) indicating an origin in distant material . some structure is also present around 0.9 kev , with the possible presence of either a drop at that energy or two peaks . in the bottom panel of the same figure , we show the rms spectrum obtained for the low flux observation during rev 541 . the variability below 3 kev is strongly suppressed with respect to the high flux observation . moreover , the shape of the rms spectrum is totally different and rather unusual . the trend of increasing variability toward softer energies breaks down completely and the most striking feature is the marked drop of variability around 0.9 kev . this feature , as well as the other drop of variability at @xmath70.55 kev , could have the same nature as the structure seen in the high flux rms spectrum , but is much more significant and prominent here . as a sanity check , we show in fig . [ lcsoft ] the source light curve around 0.9 kev and in two bands below and above for comparison . the source appears to be constant around 0.9 kev ( @xmath17=144.2 for 140 degrees of freedom ) . the constant hypothesis is unacceptable fits in the other two bands . as in the high flux observation , the variability is also suppressed around 6.4 kev . the drop at 67 kev is here even more dramatic due to the reduced continuum level and therefore increased visibility of the constant narrow fe line . the rms spectrum analysis given above indicate the presence of two constant or weakly variable components . the first is related to x ray reflection and fe emission and is seen in both observations . the second is associated with the drop in variability around 0.9 kev which is clearer in the low flux data where the contribution of the continuum is lower . [ rms ] shows a clear minimum at 6.4 kev indicating that the narrow fe emission line is less variable than the continuum in both observations . the fe line is clearly detected in the time averaged spectra of both observations . in the low flux one , the line is unresolved , has an energy of @xmath18 kev , a flux of @xmath1910@xmath20 ph s@xmath21 @xmath12 , and an equivalent width of about 260 ev . the line energy and flux are consistent with measurements obtained in the high flux observation and also with previous data from _ chandra _ ( collinge et al 2001 ) and _ bepposax _ ( guainazzi et al 1998 ) . as mentioned , a constant and narrow fe emission line is also required by the rms spectra . since the fe line must be associated with a reflection continuum , in all subsequent fits we always include a constant and neutral reflection model from magdziarz & zdziarski ( 1995 ) and a narrow fe line . the reflection continuum normalization is chosen such that the fe line equivalent width ( with respect to such continuum only ) is consistent with the _ bepposax _ observation ( ew@xmath22700 ev ) by guainazzi et al ( 1998 ) . this is because that observation caught ngc 4051 in a particularly low flux state in which the primary continuum had switched off leaving only reflection from distant matter in the hard band , thereby providing a most useful way to constrain that component ( under the reasonable assumption that it remains constant ) . the second constant component is clearer in the low flux state rms spectrum ( bottom panel of fig . [ rms ] ) and appears as a deep minimum , that reach almost null variability , at around 0.9 kev and with a lower variability at @xmath70.55 kev ( see also fig . [ lcsoft ] ) . these prominent drops indicate the presence of constant emission around @xmath70.9 and @xmath70.5 kev , as the drop at 6.4 kev is easily interpreted as constant fe emission . moreover , an emission like feature is seen in the pn spectrum of the low flux observation and in previous _ low flux spectra as well ( collinge et al 2001 ; uttley et al 2003 ) . the energy of the feature in the pn spectrum is @xmath7 0.9 kev , consistent with ne ( and possibly fe emission plus o recombination continuum , or rrc ) . a second feature is also detected around 0.50.6 kev , where emission from o and o is expected . as an illustration , in the top panel of fig.[r12 ] , we show the ratio of the low flux pn and mos data to a simple continuum model comprising galactic absorption power law and blackbody in the relevant energy band . the energies of the emission features and their intensities over the continuum correspond with the two drops seen in the rms spectrum . the larger strength of these features during rev . 541 is due to the lower continuum . in fact , the same drop of variability at about 0.9 kev observed during the rev . 263 can be reproduced by the same constant emission lines with a higher continuum . to understand the nature of these constant emission components , we then inspected the simultaneous high resolution rgs data to better search for possible narrow emission lines that could explain the features seen at ccd resolution . as already shown by pounds et al ( 2004 ) the high resolution rgs data during the low flux _ xmm newton _ observation indeed show an emission line spectrum which is very similar to that of typical seyfert 2 galaxies ( e.g.kinkhabwala et al 2002 ; bianchi et al 2005 ) . we take the same continuum as in the ccd data and add a o edge at 0.74 kev to account for the absorption seen there . the most prominent emission lines are due to the o triplet ( dominated by the forbidden and intercombination lines and with no clear resonance ) , the o k@xmath23 line and the n k@xmath23 line ( see fig . [ rgs ] ) . the ne ( forbidden ) line is also clearly detected at 0.905 kev with a flux of @xmath24 ph s@xmath21 @xmath12 , not enough to account for the pn feature around 0.9 kev . however , the ne line sits on top of a broad feature most likely due to o rrc and possibly unresolved fe emission lines . when fitted with a crude gaussian model in the rgs data , such a feature has an energy of @xmath25 kev and a flux of @xmath26 ph s@xmath21 @xmath12 . by combining the ne line with this broad feature ( @xmath27 ev ) we obtain a flux of ( 0.7 - 1.1)@xmath28 ph s@xmath21 @xmath12 around 0.9 kev . we then included ( with all parameters fixed to the rgs results ) the n and o lines in the mos / pn model , together with a o edge at 0.74 kev ( fixed energy ) and a gaussian emission line with free energy , width , and normalization to account for the blend of ne line plus broad feature around 0.9 kev . the resulting fit is acceptable and is shown as a ratio plot in the bottom panel of fig . as for the emission feature at 0.9 kev , we measure an energy of @xmath29 kev and a flux of @xmath30 ph s@xmath21 @xmath12 , consistent with the rgs upper limit . as mentioned in pounds et al ( 2004 ) excess absorption might be present around 0.76 kev ( possibly related to a m shell unresolved transition array from fe ) . if included in both ccd and rgs data , this has the effect of reducing the broad 0.9 kev feature intensity , making so much more difficult to reproduce the variability drop in the rms spectra . the above analysis demonstrates that emission from photoionized gas is present in the low flux data and detected both in the rgs and in the lower resolution ccd detectors . such components , as shown by the rms spectra , are constant over time , thus then are likely to come from relatively extended gas ( see pounds et al 2004 ) . a constant emission component , indeed , does explain the unusual drop of variability around 0.9 kev in the low flux rms spectrum . the line intensities are also consistent ( though not detected ) with the higher flux observation data . they are not detected because of the much higher continuum level . we then included , in addition to the constant neutral reflection from distant material , such lines with fixed energies and intensities ( as obtained from the low flux rgs data ) in all subsequent fits . another model independent approach that provides information on the dominant driver of the spectral variations is the flux flux technique ( see taylor , uttley & m@xmath13hardy 2003 ; uttley et al 2004 ) . we recall here that a linear flux flux relationship reveals that the data can be described by a simple two component model where one component varies in flux but not in spectral shape , while the other remains constant both in flux and spectral shape . an offset on the hard or soft axis allows one to infer in which of the two bands the constant component contributes the most . on the other hand , if the flux flux relationship has a power law functional form , then the data are consistent with spectral pivoting of the variable component . uttley et al ( 2004 ) plotted the 210 kev band count rate versus the 0.10.5 kev one and found a very accurate power law flux flux relationship which was interpreted as evidence for spectral pivoting of the variable component . the choice of the soft and hard energy ranges by uttley et al ( 2004 ) is driven by the requirement of having well separated bands to best reveal the spectral variability . however the 0.10.5 kev band does not seem to have the same variability properties of the broadband continuum . this is clearly seen in the top panel of fig . [ rms ] in which the soft plateau indicates the presence of a component with different variability properties than the continuum above 1 kev . such a component is likely to be that producing the soft excess . therefore we consider as the reference soft band the 11.4 kev range to avoid possible contamination from undesired components . in all previously reported analysis of the x ray spectrum of ngc 4051 , the intermediate 11.4 kev band is dominated by a power law continuum which is not strongly affected by absorption , soft excess , or reflection . as for the hard band , we consider the 4 - 10 kev range , simply to allow enough leverage for spectral variability to show up clearly . in the top panel of fig . [ ff1 ] , we show the flux flux relationship in the chosen hard and soft bands . the two _ xmm newton _ observations are shown with different symbols . as already pointed out by uttley et al ( 2004 ) the low flux state data join gently on to the normal / high flux state ones , showing that the spectral variability is likely to be due to the same mechanism at all flux levels . there is no dramatic discontinuity between the two observations suggesting , once again , that the spectral changes are not due to a dramatic change in one of the physical parameters ( such a substantial change in the absorbing column , as suggested by pounds et al 2004 ) . the flux flux relation is clearly non linear with some curvature showing up especially at low fluxes . in the bottom panel , we show a binned version of the same plot ( minimum of 10 points per bin ) . in the same figure , the solid line is the best fit for a power law model representing a situation in which the main driver of the spectral variability is spectral pivoting ( see e.g. taylor , uttley & m@xmath13hardy 2003 for details ) . the fit is performed as in uttley et al ( 2004 ) with a power law relationship including constant components which are required in both the hard and the soft band . the power law relationship does not provide a good fit to the data ( @xmath31 for 22 degrees of freedom ) , mainly because the model has too much curvature at high fluxes and not enough at low fluxes . this does not exclude spectral pivoting as the main driver for the spectral variability , but suggests that we need to look for an alternative explanation . in fact , if the lowest flux data points are excluded , the flux flux relationship is linear with high accuracy . this is shown as a dotted line in the bottom panel of fig . [ ff1 ] which is the best fit linear relationship obtained when the lowest flux data points are ignored . we obtain a very good fit ( @xmath32 for 15 degrees of freedom ) with large hard offset of @xmath33 . as mentioned , a linear relationship suggests that the spectral variability is due to a two component model , i.e. to the relative contribution of a variable component whose spectral shape is flux independent ( e.g. a constant @xmath2 power law varying in normalization only ) and a constant or weakly variable harder component . if the flux flux relationship at normal / high fluxes is indeed linear , one can compute the contribution of the constant component in the 410 kev from the y axis offset @xmath34 . since the mean 4 - 10 kev rate for the normal / high flux data points is @xmath35 , this means that the putative constant component contribution in the 410 kev band is as large as 50 per cent in the normal / high flux states . however , if indeed the two component model is a fair representation of the spectral variability at normal / high fluxes , the low flux data imply that the spectral variability must smoothly change regime at low fluxes producing the gentle low flux bending in the relationship . such a transition could be reproduced if , below a given flux level , the constant component ( or part of it ) is not constant anymore but rather follows the changes of the variable one . in fact , any two component model in which the less variable component varies together with the variable one at low fluxes and saturates to an approximately constant flux at higher fluxes would produce the same kind of relationship as observed here . in order to check if the source spectral variability could be due to the two component model we try to reproduce the rms spectrum during rev . 263 with a simple phenomenological parametrization ( the rms spectrum during rev . 541 is not considered because its shape is mainly dictated by the constant components coming from distant material ) . the model is composed of a direct power law component ( plc ) and an ionized reflection dominated component ( rdc , from ross & fabian 2005 ) the spectrum of which is convolved with a kernel with fixed inner and outer disc radius ( to @xmath36 and @xmath37 respectively ) , and fixed disc inclination ( @xmath38 ) . in the rms simulation the contribution from neutral distant reflection is also considered . we assume that all the source variability is due to the variation of the normalisation of the direct plc that has a spectral index fixed at @xmath2=2.4 ( lamer et al . 2003 ) and that the reflection component is completely constant not only in normalisation , but also in spectral shape ( i.e. ionisation and relativistic blurring parameters ) . to simulate the rms we vary the plc normalization from the lowest to the highest value observed . for the rdc we have chosen the best fit values of the spectrum during the low flux state of rev . 263 ( @xmath3950 and disc emissivity index of @xmath406 ) . figure 8 shows this simulated rms ( dotted line , green in the colour version ) and , for comparison , the simulated rms with the reflection parameters of @xmath41=5 and @xmath23=4 ( solid line , blue in the colour version ) , that also roughly reproduce the low flux spectrum . the interplay between the plc and the rdc can roughly reproduce the shape of the rms ( within @xmath420.02 ) . this means that the two component model can explain the bulk of the spectral variability requiring no further variable component at low energy . the variability in the soft band can be explained by a variable power law and a constant soft excess ( which is in this case due to constant rdc ) clearly , a completely constant reflection component is an over approximation . in fact , although , it is possible to reproduce the bulk of the variability , some variations of the normalization of the rdc are required in order to match with more details the rms spectra and to reproduce the constant spectral shape below 0.5 kev ( see section 6.3.2 ) . in sources such as ngc 4051 where large amplitude fast variability is associated with spectral variability , time resolved or flux resolved spectroscopy provides much more detailed information on the nature of the source than time averaged spectroscopy . in the particular case of ngc 4051 , flux resolved spectroscopy could be a risky procedure to adopt because of the small central black hole mass ( and therefore the short dynamical timescale associated with it ) . as an example , if we were to consider the very high flux state of the first _ xmm newton _ observation ( see left panel of fig . [ lcs ] ) and make a selection say above 40 counts / s , we would accumulate spectra from intervals sometimes spaced by 70 ks . this timescale is about 140 times the dynamical timescale at 10 @xmath43 ( where @xmath44 ) for a black hole mass of @xmath45 ( e.g. shemmer et al 2003 ) which is , in our opinion , too large a factor to extract any truly significant information on the nature of the spectral variability of the source . we therefore explore time resolved spectroscopy on the shortest possible timescale ( set by requiring good quality time resolved individual spectra ) . in the following , we present results obtained by performing such an analysis on a 2 ks timescale ( about four dynamical timescales at 10 @xmath43 ) . the 2 ks slices that have been used in the analysis are shown in fig . [ lcs ] and have been numbered for reference . it is well known that the 210 kev spectral slope in ngc 4051 ( and other sources ) is correlated with the source flux . in the top panel of fig . [ gf1 ] , we show such a correlation when all the 2 ks spectra are fitted with a simple power law model in the 210 kev band . data for the two observations are shown with different symbols . the photon index increases with flux and seems to saturate around @xmath46 with a @xmath47 between the low and high flux states . however , the power law model is clearly inadequate to fit the 210 kev spectra in ngc 4051 . this is because of the presence of a narrow 6.4 kev fe k@xmath23 line and the associated reflection continuum in the hard band . as shown above , such a component is constant and likely originates in some distant material ( such as the torus ) . we then fit again the 2 ks data by including the constant reflection component and show in the bottom panel of fig . [ gf1 ] the resulting @xmath2flux relationship . the @xmath48 between the low and high flux states is now reduced to about 0.70.8 . again the photon index seems to saturate at high fluxes to @xmath46 . this value is slightly different from the one ( @xmath49 ) observed during a long rxte observation campaign ( lamer et al . the difference could be due to long term spectral variability . moreover , the photon index seems to saturate not only at high fluxes but also at low fluxes ( to @xmath50@xmath51 ) . since the two asymptotes are not extremely well defined , such behaviour could still be consistent with spectral pivoting , even though it should be stressed that the low flux photon index is still somewhat hard to be explained in terms of standard comptonization models and would suggest some peculiar physical scenario ( such as a photon starved corona ) . on the other hand , the @xmath2flux relationship seen in the bottom panel of fig . [ gf1 ] can be explained if an additional and weakly variable component is present in the 210 kev band . if so , such a component dominates the low flux states and the hard photon index measured there is just a measure of its intrinsic spectral shape in the 210 kev band , while it is overwhelmed by the power law at high fluxes where @xmath46 is then the intrinsic power law photon index . in other words , as an alternative to spectral pivoting , the @xmath2flux relationship can be explained by assuming the presence of i ) a constant reflection component from distant matter ; ii ) a variable power law with constant or weakly variable slope @xmath46 ; iii ) an additional component with approximate spectral shape of @xmath52@xmath51 in the 210 kev band which must be almost constant at normal / high fluxes and is allowed to vary significantly only at low fluxes . we point out here that our flux flux plot analysis showed that the two component model might indeed be appropriate in the normal / high flux states . if so , the constant component contributes by about 50 per cent in the 410 kev band ( from the y axis offset in the linear fit ) and this is much too large a fraction to be accounted for by constant reflection from distant material ( and associated narrow fe line ) which contributes in that band by only 10 per cent . the two component model interpretation of the flux flux relationship at normal / high fluxes implies that an additional constant component is present in the normal / high flux states and contributes by about 40 per cent to the 410 kev band . such a component naturally produces the observed @xmath2flux relationship because as the flux drops the spectrum becomes more and more dominated by it . as a first step , we consider a simple continuum model comprising a power law plus black body ( bb ) emission to model the prominent soft excess . this is , in many respects , the `` standard model '' to fit agn x ray spectra and , though often crude and phenomenological , provides useful indications that can guide further analysis . the power law slope is free to vary to account for possible spectral pivoting . we add to the model neutral photoelectric absorption with column density fixed to the galactic value . to account for the presence of absorbing ionized gas in the line of sight we include two edges ( o and o ) . we searched for possible variations of their energies without finding any evidence , so we fixed the absorption energies to the rest frame ones . this is clearly a crude approximation for the effects of the warm absorber in ngc 4051 , but the quality of the 2 ks spectra is not such to suggest the use of more sophisticated models . in addition , the constant components discussed above ( reflection from distant matter and emission from photoionized gas ) are also included in the spectral model . the overall model has six free parameters only . the resulting fits are very good with a reduced @xmath17 ranging between 0.9 and 1.2 with a mean close to unity . therefore , from a statistical point of view , a single model is able to describe the spectral variability of the source in a very satisfactory way . below , we discuss the main results of the fitting procedure and their main physical consequences . in fig . [ bbpo ] ( top panel ) we show the relation between the measured bb temperature and its intensity , which is proportional to the luminosity ( in the present case a@xmath53=10@xmath54 correspond to l@xmath53=9.7@xmath1010@xmath55 ergs s@xmath21 , assuming h@xmath56=71 mpc km@xmath21 s@xmath21 ) . the bb luminosity spans about one order of magnitude , while the bb temperature ranges from about 90 ev to 140 ev with an average of 110120 ev . such a temperature is somewhat high , but can still be consistent with that expected from a black hole with mass @xmath57 ( shemmer et al 2003 ; m@xmath58hardy et al 2004 ) accreting at high rate ( for the given black hole mass , the expected disc temperature is of the order of 70 ev for an object accreting at the eddington rate ) . we point out that even in the lowest flux states the temperature ( about 100 ev ) is larger than predicted and , in the framework of standard disc models , would already require the system to be accreting at the eddington rate . if the soft excess in ngc 4051 is really due to bb emission from a constant area , we should be able to detect the bb law ( l@xmath59 ) in data spanning one order of magnitude in bb luminosity . however , such a relation is ruled out . a fit of the l@xmath53 versus temperature data ( top panel of fig . [ bbpo ] ) with the bb law ( leaving only the normalization as a free parameter ) produces a very bad fit with @xmath60 for 67 degrees of freedom . the relation appears to be much steeper ( with an index of 6.36.4 rather than 4 ) , but even so , a power law fit is unacceptable . the measured temperature appears to be much more constant ( the fit with a constant gives a @xmath61 for 67 degrees of freedom ) than predicted by standard disc models . the properties of the soft excess in ngc 4051 are remarkably similar to those of 26 bright radio quiet quasars ( considered as a population ) studied e.g. by gierlinski & done ( 2004 ) and selected from the _ xmm newton _ archive by their strong uv continuum ( such that they have strong accretion disc emission ) . the quasar sample spans a wide range of black hole masses and luminosities and should therefore exhibit a wide range of disc temperatures ( between a few ev to about 70 ev ) . however , when fitted with a cool comptonized component , the measured temperature of the soft excess is remarkably constant throughout the sample with a mean of 120 ev ( and very small variance ) . it is a rather surprising coincidence that this is exactly the average temperature we measure for the soft excess in ngc 4051 , considering also that the black hole mass in ngc 4051 is about 3 orders of magnitude smaller than the typical black hole mass in the quasar sample . moreover , no correlation between the measured soft excess temperature of quasars and the expected disc temperature was found , in line with the results presented above . such discrepancies naturally raises questions on the real nature of the x ray soft excess . as pointed out by gierlinski & done ( 2004 ) , the remarkable constancy of the `` soft excess temperature '' might indicate an origin in atomic rather than thermal physical conditions . possible candidates seems to be absorption ( e.g. from a disc wind , as suggested by gierlinski & done 2004 ) and/or ionized reflection ( e.g. from the accretion disc itself , as suggested e.g. in ross & fabian 2005 ) . the `` standard model '' of the source emission seems to be inadequate to describe the corona parameters as well . in the bottom panel of fig . [ bbpo ] we show the power law slope as a function of the 0.510 kev flux . in the normal / high flux observation the photon index steepens with flux , as already pointed out in the 210 kev analysis . this behaviour is consistent with spectral pivoting . once again , the lowest @xmath2 are of the order of 1.4 which is a hard spectral shape to be accounted for by standard comptonization models and might require a photon starved corona scenario . the main problems arise at very low fluxes , where the @xmath2flux relation is inverted in a manner that is not consistent with simple spectral pivoting . such behaviour could indicate the presence of an additional soft ( steep ) component which becomes prominent at low fluxes and is not properly accounted for by the black body component . in fact , the power law is trying to fit the soft data by steepening the index at low fluxes and leaves significant residuals in the hard band where a slope of about 1.31.4 would be more appropriate even at very low fluxes ( see fig . [ gf1 ] ) . as a final comment , the mean optical depth of the ionized absorber ( here modelled crudely by the o and o edges ) is of the order of 0.10.3 , consistent with the value obtained from the rgs analysis ( pounds et al . these values are left free to vary and they seem to suggest some variations during the two observations . nevertheless , the uncertainty associated with these is so big that a fit to the optical depth with a constant during the first observation results in a @xmath17 of 26 ( o ) and 47 ( o ) for 47 degrees of freedom , while during the second low flux observation the @xmath17 are 29 and 7 for 19 degrees of freedom . the results of the fits with the `` standard model '' presented above suggest that the soft excess is not ( or not completely ) truly disc black body emission . moreover , if the spectral variability is due to spectral pivoting , the power law slope at low fluxes is somewhat too hard to be accounted for by standard comptonization ( haardt et al . in addition , a steeper component appears in the soft band at the lowest flux levels . we then explore here an alternative model to describe the spectral variability of ngc 4051 . in order to reproduce the flux flux plot and the @xmath2flux relationship without imposing spectral pivoting , we need two main continuum components . the first one is variable in normalization only retaining an approximately constant spectral shape at all flux levels . we make here the assumption that this can be represented by a power law component ( plc ) . its slope should be similar to the high flux asymptote of the @xmath2flux relationship , i.e. @xmath62 2.2 . the other component must be almost constant in normal / high flux states where its contribution to the 410 kev band is of the order of 40 per cent . in the lower flux states it has instead to correlate with the plc in order to reproduce the flux flux plot . from the 210 kev @xmath2flux relationship we have also some indication that its spectral shape in the hard band can be approximated by a slope of @xmath631.31.4 . the obvious candidate is a relativistically blurred reflection dominated component ( rdc ) from a ionized accretion disc . such a component is characterised by an x ray spectrum which is hard in the hard band ( @xmath631.31.6 depending on the parameters ) and steep in the soft band therefore naturally producing a soft excess . as pointed out e.g. by ross & fabian ( 2005 ) when fitted with a thermal model in the _ xmm newton _ band , such a soft excess has a temperature of the order of 150 ev , very similar to the constant temperature of radio quiet bright quasars and to the one we have measured in ngc 4051 . since the rdc is produced in the inner disc by irradiation by the plc , the two components should always be well correlated while , in the present case , the flux flux plot would require that the rdc is well correlated with the plc at low fluxes but is then only weakly variable at higher fluxes despite large changes in the plc flux . this is however precisely the main feature of the gravitational light bending model proposed by miniutti et al ( 2003 ) and miniutti & fabian ( 2004 ) . in this model the variability is dominated by a power law component ( plc ) which changes in flux but not in spectral shape . the plc is ( naively ) assumed to have constant intrinsic luminosity and all the plc variability is due to light bending effects as the primary plc sources have different positions at different times with respect to the central black hole . the plc illuminates the accretion disc giving rise to a reflection component which is affected by the relativistic effects arising in the inner disc . such a reflection dominated component ( rdc ) does not respond trivially to the plc variability because of strong gravity effects ( see miniutti & fabian 2004 for more details ) . at low fluxes the rdc is indeed well correlated with the plc , but it gradually varies less and less as the flux increases reaching a regime in which it becomes almost constant in a range in which the plc varies up to a factor 5 . therefore , the model successfully reproduces the two component model ( variable plc plus constant rdc ) at mean / high fluxes and since it predicts a correlation between the two components at low fluxes , it has the potential of explaining the curved flux flux plot relationship without invoking spectral pivoting . in this framework , low flux states correspond to situations in which the plc originates few @xmath43 from the black hole strongly illuminating the innermost disc . light bending reduces at the same time the observed plc flux at infinity . as the plc source is further away from the black hole , more photons can escape to infinity ( so that the observed plc flux increases ) and the illuminating flux on the disc is reduced . the model therefore predicts also a relation between the emissivity profile of the rdc on the disc and flux . the lower the flux , the more centrally concentrated the disc irradiation and the steeper the emissivity . the two component model we have tested comprises then i ) a plc with fixed slope at @xmath64 ( consistent with the high flux asymptote of the @xmath2flux relationship ) and variable normalization and ii ) an ionized rdc ( from ross & fabian 2005 ) the spectrum of which is convolved with a kernel with fixed inner and outer disc radius ( to @xmath36 and @xmath37 respectively ) , and fixed disc inclination ( @xmath38 ) . the inner disc radius corresponds to the innermost stable circular orbit around a kerr black hole . the only free parameters of the relativistic kernel is then the index @xmath65 of the disc emissivity profile ( @xmath66 ) . the ionized reflection model is appropriate for solar abundances and has ionization parameter and normalization as free parameters , while the photon index of the illuminating power law in the rdc model is tied to that of the plc and therefore fixed to @xmath64 . as for the `` standard model '' discussed in the previous section , the overall spectral model also includes galactic absorption , constant emission from photoionized gas and from a distant reflector and the o and o edges with fixed energies . the number of free parameters in the model is just the same as in the `` standard model '' case ( 6 free parameters ) . the model reproduces very well the data at all flux levels with a reduced @xmath17 between 0.8 and 1.2 . given that the standard model and the two component one have the same number of free parameters , we can directly compare the resulting averaged reduced @xmath17 which are @xmath67 for the standard model and @xmath68 for the two component one . therefore , the two component model is statistically indistinguishable from the standard one and has to be considered as a possible alternative to be accepted or rejected on a physical rather than statistical basis . as mentioned , the plc used to model the data has a fixed slope of @xmath64 at all flux levels . in fact , we also tested a variable @xmath2 fit to the data and found that all 68 spectra are consistent with @xmath64 with only three exceptions in which the statistics is significantly better with a different power law slope ( with small @xmath69 ) . therefore , we demonstrate here by direct spectral fitting that the curvature in both the flux flux plot and the @xmath2flux relationship does not imply spectral pivoting but is only consistent with it . the second characteristic of the model is that the soft excess is not due to a thermal component anymore ( see fig.[refl_emod ] ) . it is in fact the result of including ionized reflection from the disc , which naturally produces a soft excess . the `` constant temperature '' problem can then be naturally solved because of the very non thermal nature of that component . the same kind of model successfully reproduces the soft excesses in other sources as well ( e.g. fabian et al 2004 ; 2005 ) . the ionization parameter of the reflection spectrum is not very well constrained by the data and most spectra are consistent with @xmath41 between 50 and 300 erg cm s@xmath21 with only few exceptions below and above ( and no clear trend with flux ) . the reason for the non accurate measurement of @xmath41 is probably twofold . firstly , the spectra lack sharp emission lines which would help in constraining the ionization state ( if present , the emission lines are broadened and skewed because emitted from the inner disc ) . secondly , we are using a reflection model with a single uniform ionization parameter over all the disc surface from the innermost to the outermost regions . this is clearly a rough approximation for a real case in which the ionization parameter has a structured radial profile . such approximation is likely to play a role in the non accurate measurement of @xmath41 . on the other hand , the disc reflection emissivity index of the relativistic blurring model is better constrained . in fig . [ blur ] we show the emissivity index @xmath65 as a function of the plc flux . some trend can be seen in the figure with low flux states generally corresponding to steeper emissivity profiles ( with @xmath65 of the order of 5 ) than high flux ones . as mentioned above , this result is in line with the predictions of the light bending model ( miniutti & fabian 2004 ) . the other main prediction of the light bending model is the rdc versus plc relation . as mentioned , the model predicts a correlation at low flux levels and an almost constant rdc at higher flux levels , despite large variation in the plc . in fig . [ gflux ] we show the 0.510 kev flux of the rdc versus the plc flux in the same band . the rdc is well correlated with the plc at low flux levels ( the solid line represent perfect correlation between the two components ) . however , as the plc flux increases , the correlation clearly breaks down and the rdc is much less variable ( about a factor 2.5 ) than the plc ( about a factor 7 ) in very good agreement with the qualitative predictions of the model . the main difference between the predicted and observed behaviour is that some residual correlation is seen at normal / high flux ( while the model predicts an almost constant rdc ) . such residual correlation can be easily interpreted by intrinsic luminosity variations of the plc source which are not accounted for in the model . we finally remark that , if the light bending interpretation is correct , the rdc versus plc behaviour requires the primary source of plc to be centrally concentrated above the accretion disc within 15 gravitational radii at most from the central black hole ( see miniutti & fabian 2004 ) . in particular , the correlation between the two component is predicted to occur only if the primary source of the plc is within @xmath7 5 gravitational radii from the hole . therefore , one consequence of this interpretation is that low flux states are characterized by an extremely compact region of primary emission , well inside the relativistic region around the black hole . we tested both spectral models further by comparing the rms spectra with synthetic ones , obtained from the best fitting models discussed above . obviously , the better the fits to the 2 ks spectra are , the more closely the rms spectra will be matched , but the comparison is not trivial because the @xmath17 obtained from the time resolved spectroscopy is a global quantity in the energy space ( in fact it depend over all the energies considered ) while the rms is a local quantity ( it does not depend on what is the model at other energies ) . it is possible , for example , that the spectral model slightly fails always in the same energy band . if this is the case , a comparison between observed and simulated rms spectra is useful . in fig . [ rms2 ] we show the observed and synthetic rms spectra for both observations . in back ( blue in the colour version ) we plot the rms spectra obtained from the two component model best fitting parameters and in grey ( green ) that from the standard bb plus power law model . both models reproduce well the broadband shape of the rms spectra in the two observations with residuals of the order of few per cent only . this is possible , allowing the three parameters of the rdc component ( ionization , emissivity index and normalization ) to vary ( see sect . we do not exclude the presence of further soft component with a possibly different physical origin , but we show that that is not required when variations of the parameters of the rdc are considered . in particular , the sharp features in the rms spectra have been reproduced thanks to the inclusion of the two constant components due to distant neutral reflection and photoionized gas emission . this reinforces our previous qualitatively interpretation indicating that the emission line intensities and their constancy ( not only during each observation but also between the two one year apart ) are needed in order to reproduce the rms spectra . the variability spectrum during the rev . 263 is less affected by the two constant components and carries more informations about the nuclear variability . its shape can be explained equivalently well by the two models investigated . in the `` standard scenario '' the gradual drop of variability with energy is mainly due to the steepening of the power law slope with flux , while the lower variability at low energy is due to a less variable bb component . in the two component model scenario it is expected that the source has lower variability both a high and low energy , because these are the regions where the almost constant relativistic reflection component is more important . in order to estimate the contribution , to the overall spectral variability , due to the variations of the warm absorber , we have simulated the rms spectra both with constant and varying optical depth edges . the inserts of figure [ rms2 ] show the enlargement of the rms in the soft energy band , where the effect of the warm absorber are stronger . the grey ( green in the colour version ) line shows the synthetic rms obtained starting from the best fitting parameters with the `` standard model '' and with varying optical depth edges , while the black ( orange in the colour version ) line shows the rms expected from refitting the data with the imposition of a constant warm absorber . we show here only the results obtained considering the standard model , but similar results are obtained with the two component model . the inserts of figure [ rms2 ] show that , when the constancy of the optical depth is imposed , some deviations in the reproduction of the rms are present . the intensity of these deviations are of the order of few percent suggesting that a variation of the warm absorber is possible but the bulk of the spectral variability is due to the interplay of the back body and power law components ( power law and reflection , in the two component model ) if the continuum is left completely free to vary . the analysis we presented above shows that rms spectra , flux - flux plots and individual 2 ks spectra can be explained either in terms of a standard back body plus power law model or of a two component model . in both cases , the presence of constant emission from photoionized gas and from a distant reflector are required . the standard model comprises a blackbody component which provides the soft excess and a power law with variable slope ( and normalization ) . the blackbody temperature is surprisingly constant given that its luminosity spans one order of magnitude and does not seem to follow the blackbody law ( @xmath70 ) . moreover the average temperature we measure ( about 110 - 120 ev ) is remarkably similar to the average temperature measured in a sample of 26 radio quiet quasars with typical black hole masses three orders of magnitude larger than in ngc 4051 . these surprising results suggest that a pure thermal component is not appropriate to account for all the soft excess in ngc 4051 . in fact , as shown in crummy et al . ( in prep . ) the soft excess in type 1 active galactic nuclei is consistent , like here , with being produced by ionized reflection . the power law slope in correlates with flux except at very low flux levels where the relation expected from simple spectral pivoting breaks down . moreover the hardest photon index we measure ( about 1.31.4 ) seem too hard to be explained in terms of standard comptonization models . it should be stressed also that while spectral pivoting describes well the flux flux plot relationship between the 0.10.5 kev and 210 kev bands ( see uttley et al 2004 ) , it fails ( or at least reduces its accuracy ) when the reference soft band is chosen to be between 1 kev and 1.4 kev . this is somewhat surprising because the latter band is dominated ( whatever spectral model is preferred ) by the power law component , while at softer energies a contribution from the prominent soft excess is unavoidable . in the context of the two component model , the soft excess is instead accounted for by ionized reflection from the accretion disc which simultaneously hardens the hard spectral shape as the flux drops . the apparent constant temperature of the soft excess is then explained , because the soft excess spectral shape is dictated by atomic rather than thermal mechanisms . the two component model is consistent with the variability being dominated by a constant slope power law which varies in normalization only . the @xmath2flux relationship is then interpreted as spurious and due to the relative flux of the power law and reflection components . the rdc is in fact correlated with the plc at low fluxes and varies with much smaller amplitude at normal / high fluxes . such behaviour explains not only the spurious @xmath2flux relationship , but is also consistent with the flux flux plot analysis . we investigated the x ray spectral variability of the narrow line seyfert 1 galaxy with model independent techniques ( rms spectra and flux flux plots ) and with time resolved spectral variability . the main features of the x ray emission of are here summarized . 1 . the data of show evidence for a distant neutral and constant reflection component contributing by about 10 per cent in the 410 kev band at mean fluxes . 2 . during the low flux observation a photoionized gas imprints its presence in the soft x ray band . this emission is consistent with being constant not only during , but also between the two observations one year apart . it is not clearly detected during the high flux observation because of a much stronger continuum . + the constancy of this component , as well as the previous one , is primarily shown thanks to the powerful rms spectra . 3 . the nuclear emission has been interpreted both in a `` standard scenario '' ( consisting of bb plus power law emission ) and by a two component ( plc plus ionized rdc from the disc ) . both models reproduce the rms spectra , and describe the time resolved spectra in a comparable way from a statistical point of view . our results are consistent with the warm absorber playing a minor role in the spectral variability . the standard model results indicate that the bb emission does not follow the l@xmath0 relation , even if more than an order of magnitude in luminosity is spanned . the best fit bb temperature is almost constant , clustering its value between 0.1 and 0.12 kev . the power law slope is consistent with spectral pivoting and steepens from 1.31.4 to about 2.2 with flux . the hardest photon indexes are so flat to require rather unusual scenarios such as a photon starved corona . moreover , the very low flux states exhibit an inverted @xmath2flux behaviour which disagree with a simple pivoting interpretation . it is then possible that further unmodelled soft / steep component manifest itself at very low fluxes . 5 . in the framework of the two component model the soft excess is interpreted as the soft part of ionized reflection from the accretion disc . the constant temperature problem is then solved in terms of atomic rather than truly thermal processes . moreover , the rdc explains the @xmath2flux relationship at all flux levels in terms of the relative contribution of a plc with constant slope @xmath64 and the rdc . the variability properties of the rdc with respect to the plc explain also the reason why the flux flux plot is curved without the need for invoking spectral pivoting . the rdc is well correlated with the plc at low fluxes and varies with much smaller amplitude than it at normal / high fluxes , in line with the predictions of the light bending model in which the variability is mainly due to the fact that primary sources close to the black hole appear fainter than sources far from it because of light bending , which reduces the net observed flux at infinity as the source ( of the plc ) is closer and closer to the hole . we then propose the light bending model as a possible alternative explanation for the spectral variability of ngc 4051 in which most of the nuclear emission is emitted from the very inner regions of the flow , only few gravitational radii from the central black hole . from a physical point of view , the model does not require extreme parameters . the only true requirement is that the plc originates in some compact region above the accretion disc and that the active region changes location in time ( or disappears and a new one appears at a different location ) but always within about 15 gravitational radii from the center . we stress here that if , as it is generally believed for high accretion rate sources ( such as ngc 4051 ) , the bulk of the x ray emission comes from the inner regions of the accretion flow , the effects we are talking about are simply unavoidable . no radiation can escape from the inner regions of the flow without being affected by gravitational light bending and other relativistic effects . this is true not only for disc emission ( and reflection ) but also for the primary radiation . the authors would like to thank mauro dadina , giorgio palumbo , giuseppe malaguti , pierre olivier petrucci , roderick johnstone michael mayer and the anonymous referee for many useful comments , suggestions and technical help . based on observations obtained with xmm - newton , an esa science mission with instruments and contributions directly funded by esa member states and nasa . gp thanks the european commission under the marie curie early stage research training programme for support . gm and ki thank the pparc for support . acf thanks the royal society for support .

"we describe a new numerical 3d relativistic hydrodynamical code and present the results of three va(...TRUNCATED)

astro-ph0202402

"collimated extragalactic flows , exhibiting a complex pattern of internal structures , often with h(...TRUNCATED)

"we consider the exponential matrix representing the dynamics of the fermi - bose model in an undepl(...TRUNCATED)

1211.1806

"the fermi - bose model under study here forms the underlying basis for a range of phenomena in cond(...TRUNCATED)

"a one - parameter generalisation of the symmetric macdonald polynomials and interpolations macdonal(...TRUNCATED)

0903.3996

"let @xmath3 be a partition , i.e. , @xmath4 is a weakly decreasing sequence of nonnegative integers(...TRUNCATED)

"it is widely recognised that the effect of doping into a mott insulator is complicated and unpredic(...TRUNCATED)

1606.09311

"electron - hole doping asymmetry in mott insulators has been one of the key questions related to th(...TRUNCATED)

"current theories of structure formation predict specific density profiles of galaxy dark matter hal(...TRUNCATED)

1011.3041

"weak gravitational lensing is a powerful technique for studying the distribution of matter in the u(...TRUNCATED)

"we construct one - zone steady - state models of cosmic ray ( cr ) injection , cooling , and escape(...TRUNCATED)

0910.0478

"the radio and fir luminosities of star - forming galaxies are linearly correlated over three decade(...TRUNCATED)

"we investigate lattices that can be represented as sublattices of the lattice of all convex subsets(...TRUNCATED)

math0609114

"a _ lattice of sets _ is a family @xmath1 of sets which is a lattice with respect the inclusion rel(...TRUNCATED)