article
stringlengths 0
745k
| abstract
stringlengths 0
92.7k
|
|---|---|
deploying more cellular base stations ( bs ) has been the main remedy to cope with relentless traffic growth . to reach the highest capacities ,
interest has lately been turning toward ultra - dense bs deployment @xcite , @xcite , where the density of bs may even exceed the typical number of users in a given area .
the impact of such extreme bs densification , however , has still not been explicitly analyzed . to be more specific , in an engineering perspective , the preceding work @xcite provides an analytic average spectral efficiency ( se ) calculation that reveals the se is independent of bs density .
the result is accurate only when bs density is low as it relies on an assumption that every bs has at least a single serving user . for densely deployed bs environment ,
the authors @xcite consider turned - off bss when having no serving users as in the third generation partnership project ( 3gpp ) release 12 specifications @xcite , and predict the se is a logarithmic function of bs density via its compatible performance metrics , service capacity , outage probability , and common ( worst user s ) rate respectively .
nevertheless , they can not represent the explicit relationship between the se and bs density due to the intractable forms of results .
motivated by these discussions on the bs densification , we derive an analytic se expression valid for general bs density by utilizing a stochastic geometric approach @xcite , and further provide its closed - form representation under ultra - dense bs environment .
the result verifies _ se logarithmically increases with bs density _ in ultra - dense networks where the bs densification makes interference restricted by turning off empty bss . in an economic perspective
, the authors @xcite consider a network operator s profit maximization based on user demand prediction in a single cell scenario neglecting the bs densification effect .
for a multi - cell network , the work @xcite deals with cost minimization when bs density is low . regarding the relationship between user demand and network supply , the authors @xcite provide an analytic approach although it resorts to an iterated simulation .
this study , leading from the preceding works , focuses on the question : how much amount of bs density and spectrum a network operator should invest in when user demand changes ? to capture the user demand variation , we consider two user demand characteristics : ( i ) user density and ( ii ) each user s sensitivity to his downloading rate .
we thereby answer the question via solving a profit maximization problem while considering the user demand meeting network s rate supply .
this paper examines the ramifications of bs density increase in downlink cellular networks from both engineering and economic points of view in terms of se and profit respectively .
the main contributions are listed below . 1 .
this paper derives the closed - form se in an ultra - dense downlink cellular network that is a logarithmic function of bs density .
2 . the paper provides network planning guidelines in terms of the profit optimal bs density and spectrum amount in closed forms .
we consider a downlink cellular network .
let @xmath0 denote bs coordinates in a two - dimensional euclidean plane , following homogeneous poisson point process with density @xmath1 .
similarly , user coordinates @xmath2 follow homogeneous poisson point process with density @xmath3 , independent of @xmath0 .
users are associated with their nearest bss , which correspondingly forms bs coverage regions whose boundaries comprise a two - dimensional voronoi tessellation @xcite .
each bs is tuned off when its coverage , a voronoi cell , is empty of serving users , otherwise transmitting with unity power . for a transmitted signal ,
we consider path loss attenuation with the exponent @xmath4 and rayleigh fading . , @xmath5 , and @xmath6 where user density @xmath7 .
compared to the exact se when @xmath8 , the analytic value achieves : 81.75% , 90.9% , and 97.96% for bs
density @xmath9 , @xmath10 , and @xmath11 respectively.,width=328 ] this section aims at providing a closed - form se , to be utilized for profit maximization in section [ sect : profitmax ] . to this end
, we consider interference - limited regime neglecting noise , and define se @xmath12 as ergodic rate per user for unity spectrum amount , @xmath13 $ ] in units of nats / sec / hz ( 1bit @xmath14 0.693 nats ) where @xmath15 denotes signal - to - interference ratio . for brevity , we neglect multiple user access , to be considered in section [ sect : cap_approx_ma ] . according to proposition 1 in @xcite ,
a bs s turned - on ( non - empty ) probability @xmath16 is given as @xmath17 .
applying this to the equation ( 16 ) in @xcite provides @xmath12 as : @xmath18^{-1}\hspace{-7pt } dt \label{eq : exact}\ ] ] where @xmath19 .
+ for a sparse ( @xmath20 ) cellular network when @xmath21 , the equation is in accord with the result in @xcite .
for explanatory convenience , let @xmath22 hereafter denote the se without multiple access in a sparse network , given as @xmath23^{-1 } dt .
\label{eq : sparse}\ ] ] this shows @xmath22 is independent of @xmath1 .
it implies , in other words , bs densification does not yield any gain in se under sparse environment since its desired signal power improvement is on the same order of amount as the aggregate interference increase , cancelled out each other at calculating @xmath15 .
for an ultra - dense ( @xmath24 ) network when @xmath25 by using taylor expansion , the equation hardly captures the relationship between @xmath12 and @xmath1 due to the complicated double integrations therein .
we take a detour this problem by deriving the closed - form approximation of @xmath12 in the following proposition .
interestingly , this simple expression shows that @xmath12 is a logarithmic function of @xmath1 . in other words ,
_ bs densification does increase the se whilst yielding diminishing returns _ under ultra - dense environment .
a stochastic geometric point of view interprets this phenomenon as follows . at a typical user ,
increasing @xmath1 shrinks each bs s coverage , simultaneously yielding : ( i ) the shortened distance to his associated bs and ( ii ) increased the number of empty cells ( or turned - off bss ) .
focusing firstly on the former , the shortened distance in the order @xmath26 ( see the average distance is @xmath27 in @xcite ) yields the received signal power increase from the associated bs , in the order @xmath28 . for the latter in an ultra - dense scenario , it makes almost all bss turned - off except for the ones serving users in their infinitesimal coverage regions .
consequently , the interfering bs locations ( or interferer density ) converge to the users ( or @xmath3 ) , delimiting the quantity of interference .
combining these results leads to the ever - increasing @xmath15 in the order @xmath29 , resulting in the logarithmic se increase .
[ fig : capapprox ] visually validates the tightness of the value in proposition 1 for different @xmath30 s . when @xmath8 , for instance , the difference between and is less than @xmath31 for @xmath32 .
thus , we hereafter regard as the approximation of @xmath33 .
additionally , it is worth mentioning that this result shows the bs ultra - densification gain in se . comparing to @xmath22 in a sparse scenario ,
ultra - densification provides the se gain : 162% , 250% , and 464% for @xmath9 , @xmath10 , and @xmath11 respectively when @xmath7 .
now we turn our attention to multiple access of users with a fixed amount of spectrum .
so far we have considered a bs serves all users in its coverage .
instead , we henceforth consider each bs serves at most a single user at a given time , who is selected according to a uniformly random scheduler @xcite .
according to proposition 2 in @xcite , a typical user s selection probability by the scheduler for a sparse network is @xmath34 , and for an ultra - dense network is @xmath11 . applying these results to the equations and yields the following corollary .
_ _ ( se with multiple access ) _ se with a uniformly random scheduler in a sparse or ultra - dense downlink cellular network is given as follows .
\label{eq : udnwma}\end{aligned}\ ] ] _ in a sparse scenario , increasing @xmath1 alleviates multiple access congestion , and thereby linearly increases se in spite of a constant @xmath22 independent of @xmath1 in . in an ultra - dense scenario , on the other hand , reducing the access congestion along with bs densification does not ameliorate se .
that is because multiple access of users barely occurs in the ultra dense network where almost all bss have at most a single user within their coverages .
this results in being equivalent to .
the following section utilizes these results so as to explore the economic impact of bs densification .
in this section , our objective is to maximize a network operator s profit when user demand varies , caused by the changes in the number of users @xmath3 and/or each user s rate sensitivity @xmath36 .
the operator is able to cope with these demand changes by adjusting his operating bs density @xmath1 and spectrum amount @xmath37 as well as price per nats / sec ( or per nat for unity time ) @xmath38 . since the profit is a joint function of user demand , price , average rate , and its operating cost , the operator s profit maximizing decision problem to clarify the adjustments in @xmath1 , @xmath37 , and @xmath38 is not trivial , being of our interest . to be more specific ,
we firstly predict average per - user demand @xmath39 , and then determine optimal @xmath1 and @xmath37 so as to maximize the average profit per unit area , formulated as : @xmath40 where @xmath41 and @xmath42 respectively denote bs and unit spectrum operating costs per unit area . in an operator s perspective , we divide this profit maximization problem into three sequential stages : stage 1 .
user demand prediction ; stage 2 .
price decision ; and stage 3 .
the decision on bs density and spectrum amount , resulting in average rate to be supplied .
[ fig:3stages ] elucidates these subdivided problems and their solving direction as well as the direction of network operation for given solutions of the problem . for the purpose of predicting user demand at stage 1 ,
consider a typical user s payoff @xmath43 having the following characteristics : logarithmically increasing with downloading rate @xcite , and linearly decreasing with cost under usage - based pricing @xcite .
let @xmath44 denote average rate per user , and define @xmath45 as a user s willingness - to - pay , assumed to be uniformly distributed from @xmath46 to @xmath36 , @xmath47 .
we hereafter interpret the maximum willingness - to - pay , @xmath36 , as the user s rate sensitivity .
correspondingly , we represent @xmath43 as : @xmath48^+.\ ] ] consider users try to maximize their payoffs . since @xmath43 is a concave function of @xmath44 , applying the first order necessary condition and taking average over @xmath45 yield the payoff maximizing average rate per user ( or average demand per user ) @xmath39 as follows .
@xmath49 [ fig : approx_lowden ] [ fig : approx_highden ] applying the result to the profit function in ` ( p1 ) ` reveals the profit decreases with @xmath38 .
this intuitively indicates attracting more users by reducing @xmath38 yields higher profit than increasing @xmath38 .
since @xmath39 is also a decreasing function of @xmath38 , the profit maximizing price @xmath50 at stage 2 is the price when the equality holds at the constraint in ` ( p1 ) ` , resulting in @xmath51 \nonumber \\ & \overset{(a)}{\approx } & \frac{b}{2(1 + w \gamma ) } \label{eq : optprice}\end{aligned}\ ] ] where ( a ) follows from taylor expansion for large @xmath52 .
this subsection aims at deriving optimal bs density @xmath53 and spectrum amount @xmath54 in closed forms so that their resultant average rate is provided to a typical user while maximizing profit .
exploiting @xmath50 , the equation in section [ sect : demandmodel ] , modifies ` ( p1 ) ` for stage 3 as follows .
@xmath55 for sparse cellular networks , applying to @xmath12 of ` ( p2 ) ` yields the following profit maximization problem .
@xmath56 for ultra - dense networks , in the same manner , applying to @xmath12 in ` ( p2 ) ` leads to the following profit maximization problem .
\r)^{-1 } \r\}^{-1}\\ & & \hspace{10pt } - \ ( c_b \lambda_b + c_w w \ ) \end{aligned}\ ] ] solving ` ( p3 ) ` and ` ( p4 ) ` provides profit maximizing @xmath53 and @xmath54 of sparse and ultra - dense networks respectively , as provided in the following proposition .
( optimal bs density and spectrum amount ) we interpret the above results in the following perspectives : 1 ) unit operating costs @xmath41 and @xmath42 and the resultant profit optimal operating costs @xmath58 and @xmath59 and 2 ) user demand comprising per - user rate sensitivity @xmath36 and user density @xmath3 . increasing unit bs operating cost @xmath41 leads to investing more in spectrum as the bs substitute , captured by increasing @xmath54 , and vice versa for unit spectrum operating cost @xmath42 increase . multiplying these unit operating costs by and yields the following profit optimal operating bs / spectrum cost ratio . _ _ ( optimal cost ratio ) _ profit maximizing ratio of bs and spectrum operating costs in a sparse or ultra - dense downlink cellular network is given as : @xmath60 _ in a sparse network , the operator should invest in bs operating cost as much as the spectrum cost since bs density and spectrum amount equally affect average rate ( see with the spectrum amount @xmath37 , shown in ` ( p3 ) ` ) . in an ultra - dense network , on the other hand , bs density less affects average rate than spectrum amount due to the densification s logarithmic impact on average rate .
this leads to the investment strategy that bs operating cost should be less than the spectrum cost , depending on @xmath30 ( note @xmath61 ) .
the straight lines on the bs density - and - spectrum planes ( bottom ) in fig .
3 illustrate such operating cost ratios . both profit optimal bs density and spectrum amount increase as user demand grows , but the optimal value increments incurred by user density are higher than the values by per - user rate sensitivity ( see the exponents of @xmath3 and b in and , visualized in fig .
while rate sensitivity growth solely increases user demand , user density growth not only increases the demand but also decreases average rate due to : ( i ) incurring more multiple access congestion in a sparse network or ( ii ) generating more interference in an ultra - dense network ( see the discussion after proposition 1 in section [ sect : cap_approx ] ) .
consequently , increasing user density requires more bs density and/or spectrum amount in order to keep up with the demand growth as well as to recover the average rate decline . such different impacts of user density and per - user rate sensitivity furthermore affect the network profitability .
for user density growth , fig .
4(a ) shows the profit of an ultra - dense network increases less than that of a sparse network .
an ultra - dense network requires much more bs density and spectrum amount to compensate the interference generation caused by user density growth , and it engenders too much cost increase worsening the network profitability . for this type user demand increase ,
bs ultra - densification is not preferable . for per - user rate sensitivity growth , in contrast , fig .
4(b ) depicts the profit of an ultra - dense network increases more than that of a sparse network thanks to the network s delimited interference , promoting ultra - dense bs deployment .
in this paper we have derived a closed - form relationship between bs density and se in an ultra - dense cellular network .
the se is shown to be a logarithmic function of bs density as the density grows .
this closed - form se expression was used to derive closed - form solutions for the optimal operating bs density and spectrum amount , for the traditional spectrum licensing ( auction ) case .
this expression could aid the operator in his decision if he should invest more in bs densification or bid for more spectrum for his network .
our results reveal some fundamentally unique characteristics of the ultra - dense network deployment , e.g. that the number of users has a larger impact on the optimal network configuration than each user s sensitivity to his downloading rate .
further , in our simplified model we see that the network operator should maintain a reasonable balance in investment , allocating about equal amounts to spectrum and bs deployment . to the best of our knowledge
, this paper is the first to simultaneously specify not only random spatial locations of both users and bss but also each user s demand model , bridging the gap between stochastic geometric and network economic analysis .
a weakness of the study is the very simple , homogeneous propagation model , i.e. using a constant path loss exponent .
the bss are typically in the same room as the users in line - of - sight ( los ) conditions , whereas the interfering bss are behind walls , i.e. in non - los conditions , creating a better sir .
further work should therefore involve a two - slope model , describing the los and non - los cases and giving a more direct relationship to the physical environment .
further extension to this work could also include milimeter - wave systems where non - los signals are very weak .
in addition , it does not seem likely that future short range systems will use traditionally licensed spectrum .
extending the economic analysis to spectrum sharing paradigms is another interesting avenue for future research .
this research was supported by the international research & development program of the national research foundation of korea ( nrf ) funded by the ministry of education , science and technology ( mest ) of korea ( grant number : 2012k1a3a1a26034281 ) .
let @xmath62 denote @xmath63 .
since @xmath64 , @xmath12 is lower bounded as : @xmath65^{-1 } dt \nonumber \\ & \overset{(a)}{\geq } & \int_{t>0 } \ [ 1-a ( e^t - 1)^{\frac{2}{\alpha } } \]^+ dt\\ & = & \log\ ( 1 + a^{-\frac{\alpha}{2}}\ ) + a \pi \csc\ ( \frac{2 \pi}{\alpha } \ ) \nonumber\\ & & \hspace{-5pt } -\frac{\alpha } { 2\ ( 1 + a^{\frac{\alpha}{2}}\ ) } \underbrace { \ , _
2f_1\left(1,1;1-\frac{2}{\alpha};1-\frac{1}{a^{\frac{\alpha}{2}}+1}\right)}_{(b ) } \label{eq : prop1pf}\end{aligned}\ ] ] where @xmath66 follows from taylor expansion , gaussian hypergeometric function @xmath67 , and @xmath68 rising factorial . the function @xmath69 monotonically increases with @xmath62 for all @xmath30 , having unity minimum value at @xmath70 . for @xmath71 ( or @xmath24 ) ,
therefore all terms in except @xmath72 become negligible , completing the proof . @xmath73 consider a sparse network .
for sufficiently large average rate ( or small @xmath74 ) , applying taylor expansion to the objective function in ` ( p3 ) ` leads to the following problem . @xmath75
the profit function is concave with respect to both @xmath1 and @xmath37 , so it has a unique maximum result . exploiting the first order necessary condition yields @xmath76^{\frac{1}{2 } } \lambda_u \quad \text{and } \label{eq : pfprop3densb}\\ w^*&=&\ [ \frac{b}{2 \gamma_\alpha c_w \lambda_b^ * } \]^{\frac{1}{2}}\lambda_u .
\label{eq : pfprop3w}\end{aligned}\ ] ] applying to proves the result . next ,
consider an ultra - dense network .
since the logarithmic function in ` ( p4 ) ` is not appropriate for deriving solutions in closed forms , we resort to considering its lower bound @xmath77 as the approximation that is tight for large @xmath78 ( or @xmath24 ) . applying taylor expansion
provides the formulation as shown below .
@xmath79 in the same way as the sparse network , exploiting the first order necessary condition leads to @xmath80^{\frac{1}{\frac{\alpha}{4}+1 } } \lambda_u \quad \text{and } \label{eq : pfprop3densbudn}\\ w^*&=&\ [ \frac{b { \lambda_u}^{\frac{\alpha}{4}+1 } } { 2 c_w}\ ( \frac{\rho_0}{{\lambda_b}^*}\)^{\frac{\alpha}{4 } } \]^{\frac{1}{2}}. \label{eq : pfprop3wudn}\end{aligned}\ ] ] applying to finalizes the proof . @xmath73 l. duan , j. huang , and b. shou , `` investment and pricing with spectrum uncertainty : a cognitive operator s perspective , '' _ ieee transactions on mobile computing _ , vol . 10 , no . 11 , pp . 1590 1604 , 2011
. s. sen , c. joe - wong , s. ha , and m. chiang , `` incentivizing time - shifting of data : a survey of time - dependent pricing for internet access , '' _ ieee communications magazine _ , vol .
50 , no .
11 , pp .
9199 , 2012 . | this paper investigates the relationship between base station ( bs ) density and average spectral efficiency ( se ) in the downlink of a cellular network .
this relationship has been well known for sparse deployment , i.e. when the number of bss is small compared to the number of users . in this case
the se is independent of bs density .
as bs density grows , on the other hand , it has previously been shown that increasing the bs density increases the se , but no tractable form for the se - bs density relationship has yet been derived . in this paper
we derive such a closed - form result that reveals the se is asymptotically a logarithmic function of bs density as the density grows .
further , we study the impact of this result on the network operator s profit when user demand varies , and derive the profit maximizing bs density and the optimal amount of spectrum to be utilized in closed forms . in addition
, we provide deployment planning guidelines that will aid the operator in his decision if he should invest in densifying his network or in acquiring more spectrum . ultra - dense cellular network , base station density , average spectral efficiency , spectrum amount , profit maximization , stochastic geometry , network economics . |
a description of the ipta data challenge and the data sets can be found at _ http://www.ipta4gw.org/_. the first round consisted of three parts - datasets 1,2 , and 3 , each with one open data set for testing purposes and one closed data set as a blind challenge .
the datasets were arranged in increasing order of difficulty , starting with evenly sampled data with bright gravitational wave signals and moving on to unevenly sampled data with weak gravitational wave signals .
the ipta data challenges are similar to the mock lisa data challenges @xcite , the main difference being that the people that put this challenge together have real detectors to work with as well .
my approach to the analysis was to use bayesian inference and markov chain monte carlo sampling to produce posterior distributions for the gravitational wave amplitude and the pulsar timing noise .
the bayesian approach requires a noise model - which fixes the form of the likelihood - a signal model , and priors on the model parameters . for stationary , gaussian noise
the likelihood is given by @xmath3 where @xmath4 is the covariance matrix , which depends on the noise in the detectors ( in this case the individual earth - pulsars arms ) and the amplitude of the stochastic gravitation wave background , and @xmath5 denotes the residuals after a signal model @xmath6 has been subtracted from the data @xmath7 . here
@xmath8 label which detector ( earth - pulsar arm ) , and latin indices @xmath9 label the data samples .
this expression for the likelihood applies to any collection of gravitational wave detectors with stationary , gaussian noise , and is widely used in theoretical studies of ground - based and space - based gravitational wave detectors @xcite .
it applies to any data representation , including time domain , frequency domain and wavelet domain . for evenly sampled time domain data with colored noise
it is best to transform to the frequency domain where @xmath10 , as this leads to substantial savings when computing the likelihood : @xmath11 versus @xmath12 floating point operations . in the current
setting we are looking for a stochastic gravitational wave signal , so we do nt have a deterministic signal model for the gw component of the data . instead , we are looking for correlations in the signals between earth - pulsar pairs @xmath13 , which depend on the angle @xmath14 between the line of sight to each pulsar and follow the hellings - downs @xcite correlation curve @xmath15 , which has the form @xmath16 we do however have a deterministic timing model that accounts for effects such as the pulsar spin - down , sky location and proper motion , and a host of other effects , such as relativistic time delays due to the earth s motion about the sun .
the leading terms in the timing model have the form @xmath17 the parameters @xmath18 are used to fit the pulsar spin - down , @xmath19 and @xmath20 account for the sky location and leading order time dilation effects , which introduce periodic variations with @xmath21 , @xmath22 and @xmath23 account for pulsar proper motion , while @xmath24 and @xmath25 account for second order time dilation effects and the eccentricity of the earth s orbit .
the publicly available tempo2 @xcite software package fits a more involved timing model to each individual pulsar using a chi - squared goodness of fit .
taking the tempo2 residuals and subtracting the raw data that was provided for the open data sets reveals a clear excess of power with a spectral slope of @xmath26 , in addition to a wide bump around @xmath27 from the @xmath19 and @xmath22 terms , along with a narrow spike at @xmath28 from the @xmath24 term
. a naive approach would be to take the tempo2 timing residuals and compare the the cross correlation between pulsar pairs to the hellings - downs correlation curve .
the problem with this approach is that the tempo2 fitting procedure introduces spurious features in the residuals , potentially removing or at least masking any gravitational wave signal that might be present .
a better approach is to combine the fitting of the timing model with the cross - correlation analysis .
i tried this approach , but the results were not entirely satisfactory .
for some reason the recovered gw amplitude always tended to get biased high .
i suspect that i might be missing some important contributions to the timing model .
i plan to re - visit this approach in an effort to understand why it did nt work as expected . simply ignoring the features introduced by tempo2 and using a model that only accounts for the gw signal and white noise in datasets 1 and 2 , or the red and white noise of dataset 3 , leads to gw amplitudes that appear to be biased high ( based on analysis of the open data sets ) .
the one caveat here is that i am not 100% sure about wether the quoted amplitudes for the gw signal refer to the astrophysical signal or the effective level registered in the detector .
the astrophysical level @xmath29 is related to the observed level in an earth - pulsar link @xmath30 by @xmath31 ( the two - thirds factor comes from sky and polarization averaging ) .
this corresponds to an amplitude ratio bias of @xmath32 .
if the factor of two - thirds was not included , then my results for the gw amplitude need to be multiplied by a factor of 1.225 . in any case , this factor is not large enough to account for the bias that i saw with the naive analysis , which showed a @xmath33 amplitude bias . to check my analysis codes in a more controlled fashion
, i wrote my own simulation software , using the same approach for generating stochastic backgrounds that we used in the mock lisa data challenges @xcite ( based on a adding together the response to independent stochastic signals from different patches of the sky using a fine healpix sky grid ) .
this simulation confirmed that the two - thirds factor should be included , and using my own simulated data , with no tempo2 processing , i was able to perfectly recover the injected gw amplitude and spectral slope , and the pulsar noise levels .
since the tempo2 fitting is done on a per - pulsars basis , the errors in the timing models should be uncorrelated between pulsars .
i therefore decided to model the timing model errors as an additional , uncorrelated red noise contribution to each pulsar , described by a spectral amplitude @xmath34 and slope @xmath35 .
for dataset 1 we were told that all the pulsars shared the same white noise level , so for that analysis the model i used had @xmath36 parameters ( since there are @xmath37 pulsars in the simulated array ) . for dataset 2
the white noise level in each pulsar was different , so the model had @xmath38 parameters .
for dataset 3 there is an additional red noise component , but i assumed that this was already taken care of by the red noise model that i introduced to account for the tempo2 fitting noise , so again the model had 110 parameters . moreover , if i understood the statement about the red noise level that was injected , it is essentially irrelevant to the analysis , as it lies below the fiducial 100 ns white timing noise for frequencies above 10 nhz ( here i am using the fact that 100 ns white timing noise corresponds to a spectral density of @xmath39 , and the red noise was quoted at @xmath40 , which cross at @xmath41 , which is at the very edge of the frequency band for a 5 year data set , and in a region where the fit to the pulsar spin - down removes any meaningful signal ) . in most runs i fixed the gravitational wave spectral slope at the expected value of @xmath42 to improve the amplitude recovery . with the model established , the next step is to decide on how to compute the likelihood .
the simplest and most robust approach is to implement ( [ like ] ) in the time domain , following the approach of van haarsteren _
et al . _
the downside of this approach is that the matrix inversion is very computationally expensive for an @xmath43 by @xmath43 matrix with @xmath44 rows and columns .
since the correlation matrices are positive definite , they can be inverted by way of cholesky factorization .
the matrix determinant is given by product of the squares of the diagonal elements of the cholesky decomposition .
i used the _ lapack _ routines dpotrf and dpotri to perform the factorization and inversion .
the codes were run on multi - core mac pro workstations , where an interesting side - effect of having the _ lapack _ routines as built in functions is that the matrix inversion is automatically parallelized , taking up an average of 4 cores . even with the parallel processing , each likelihood evaluation took @xmath45 seconds , making it difficult to produce long markov chains .
to adequately explore the posterior distribution we need many samples in the chains , with the required number of samples scaling roughly linearly with the model dimension .
the combination of the large model dimension and the high cost of generating the chains leads to long runs times ( or the need to secure larger computing resources ! ) . to shorten the `` burn - in '' phase of the analysis ( where the chains locate the region of high posterior mass ) ,
i first ran on each pulsar individually using a three parameter model consisting of the white noise level , the gravitational wave amplitude ( the slope was fixed at @xmath46 ) , and the low frequency cut - off in the spectrum .
the spectral estimates from the individual pulsar runs were used as a starting point for the full run on the pulsar network .
the low frequency cut - off parameter requires some explanation .
the time domain auto - correlation function for time lags @xmath47 , @xmath48 , forms a fourier pair with the frequency domain spectrum @xmath49 : @xmath50 setting @xmath51 for power - law power spectra leads to improper correlation functions . as van haarsteren _
et al . _
@xcite point out , the correlation function should not depend very strongly on @xmath52 so long as @xmath53 , as the fitting of the pulsar spin - down model removes the part of the spectrum that causes the divergence in the limit @xmath54 .
however , i found that the value of @xmath52 did affect the analysis quite significantly , so i made it a parameter to be determined from the data .
the chains generally favored values around @xmath55 .
the correlation functions used in the full analysis have the form @xmath56 for white noise with amplitude @xmath57 , and @xmath58 for a gw signal with amplitude @xmath59 at @xmath27 and spectral slope @xmath60 . here
@xmath61 , and the sum over @xmath62 needs to extend to @xmath63 if @xmath52 is allowed to go as high as @xmath64 .
the injected gw signals have spectral slope @xmath46 .
an expression similar to ( [ gw ] ) applies to the un - correlated red noise , but with @xmath15 replaced by @xmath65 . in the runs where i included a timing model of the form ( [ time ] ) for each pulsar , the mcmc alternated between updating the parameters in the correlation matrix ( which costs of order @xmath12 floating point operations ) , and updating the parameters in the timing model ( which costs of order @xmath66 floating point operations ) . because the timing model updates are so much cheaper , it made sense to do many timing updates for each correlation matrix update .
so while the timing model involves many more parameters than the spectral model , we are able to spend many more iterations exploring the much larger timing model parameter space for little additional computational cost .
however , as i remarked earlier , the runs that used the timing model did not do as good a job of recovering the amplitude of the gw signal as did the runs with an uncorrelated red noise model . in an attempt to speed up the calculation of the likelihood ,
i also considered frequency domain implementations of the likelihood calculation ( [ like ] ) . transforming the data to
the frequency domain turned out to be a highly non - trivial exercise due to the very red spectra and the uneven data sampling .
either one of these issues is easy to overcome on its own , but in tandem they are a bitch .
the approach i settled on combined several techniques that have been used previously in pulsar timing data analysis . to handle the unevenly sampled data i used an tempered mcmc algorithm to fit the timing residuals to a sum of sines and cosines using the log likelihood @xmath67\right)^2/\delta t^2 \ , , \ ] ] with @xmath68 ns ( the choice of @xmath69 is largely irrelevant since i used a tempered mcmc ) . to avoid the gibbs phenomenon from the finite - time window function and the very red spectra
, i set @xmath70 with @xmath71 .
i found that @xmath72 worked well , and that somewhat larger and smaller values also worked .
the sum over frequencies has to be extended beyond the usual nyquist frequency because of the un - even time - sampling . using the fourier coefficients @xmath73 and @xmath74 directly gave reasonable results , but i found that it was better to re - map the data onto the true time interval with uniform time sampling , then transform the data with a standard fft . to avoid the gibbs phenomena i applied a zero phase shift double difference filter in the time domain : @xmath75 = d[i-1]+d[i+1]-d[i]$ ] , and compenstated by dividing the fourier amplitudes derived from the @xmath76 sequence by the transfer function @xmath77 . applying this filtering procedure to the full data set leads to aliasing of high frequency power to low frequencies , so i used the fit derived from ( [ c2 ] ) to first split the data into high - passed and low - passed components .
the low frequency part was transformed with the filter , while the high frequency part was transformed without the filter .
the results were then added together to give the complete fourier transform .
the sum over frequencies in ( [ like ] ) was restricted to the range @xmath78 $ ] .
the lowest frequency bin was dropped as the fitting of the pulsar spin - down model takes out most of the very low frequency power , and including the @xmath64 frequency bin was found to bias the gw amplitude low .
i dropped the highest frequencies in part because they contain no information about the gw amplitude , and also because the interpolation of the data becomes less trustworthy one short time scales .
for unevenly sampled data , the fourier components are not strictly orthogonal , and @xmath79 . to keep the analysis simple i ignored this complication and assumed orthogonality . for the priors i used uniform distributions in the log of the spectral amplitudes , and uniform distributions in the spectral slope parameters .
i used a very wide prior range , and found that the distributions stayed well within these ranges , with the exception of the time domain analysis of the closed dataset3 data , where the low end of the gw prior was seen to cut - off the distribution , which suggests that the gw level was not confidently detected in that analysis .
i found that in most of the challenge data sets the gw amplitude was sufficiently high that there was no question about wether a detection had been made .
the one exception was the closed data from dataset3 . the gw amplitude flirted with zero during the time and frequency domain runs , indicating that a pure noise model may have greater bayesian evidence . to test this
i ran a trans - dimensional reversible jump markov chain monte carlo ( rjmcmc ) algorithm on the frequency domain data that allowed for transitions between a gw + noise model and a pure noise model .
the ratio of the time that the chains spent in each model provides the bayes factor , or evidence ratio for the two models .
if our prior belief is that the two models are equally likely , then this ratio is the betting odds for one model versus the other .
as we shall see , the odds turned out to favor the gw model by a fairly comfortable margin ( but perhaps by not enough to send a paper to nature if this was real data ) .
i begin by showing results from the frequency domain analysis .
the gw spectral slope was held fixed at @xmath80 for the analyses shown in figures 1 - 7 .
figure 1 shows the posterior distributions for the level of the white timing noise ( in seconds ) and the dimensionless gravitational wave amplitude @xmath59 using the open data from dataset 1 .
the white noise level came out a little above the injected value of @xmath81 s , while the gravitational wave amplitude peaked very close to the injected value of @xmath82 .
figure 2 shows the same quantities , but now for the closed dataset 1 data .
this time the white noise level was a fraction below 100 ns .
the gw amplitude distribution peaked at @xmath83 , which is significantly lower than the level seen in the open data set .
figure 3 shows the posterior distributions for the dimensionless gravitational wave amplitude for dataset2 for both the open and closed data .
the posterior distribution peaks very near the injected value of @xmath82 for the open data set , and the distribution for the closed data set peaks a little higher at @xmath84 . [ cols="^,^ " , ] returning to the frequency domain analysis , and now letting the spectral slope of the gw spectrum vary , i found the posterior distributions for the spectral slope @xmath60 and the gw amplitude @xmath59 shown in figure 8 . as expected , the spread in the amplitude distribution is wider than when @xmath60 was held fixed .
the distribution for @xmath60 is biased a little high relative to the injected value of @xmath46 , which may be due to the procedure used to perform the fourier transform , or more likely , to interplay with the red noise model used to account for the timing model errors .
i was able to get reasonable results from the time domain and frequency domain analyses .
there does appear to be a bias in the time domain estimates for the gravitational wave amplitude , which i attribute to a mis - match between the spectral model ( a simple power law ) and what is likely present in the data ( a power law that flatens out at low frequencies ) .
this should be easy to fix with a suitable change to the spectral model .
i remain puzzled as to why a simultaneous fit to the timing model and the gravitational wave correlation function did not produce better results
. this could be due to an implementation error , or it may be that the this approach is flawed in some more serious way .
the frequency domain analysis suggests that modeling the errors in the timing model as red `` modeling '' noise is quite effective .
i am still a little uneasy about the frequency domain analysis as the procedure i used to transform the data was fairly byzantine . as a general rule
i do nt like to specify the model used in the analysis in too rigid a fashion .
i prefer to let the data select the best model . in future
i would like to use a full rjmcmc analysis that is able to transition between models .
for example , when i look at the distributions for some of the red noise parameters it is clear that they are poorly constrained by the data .
a rjmcmc analysis would automatically discard these parameters from the model .
this approach may also cure some of the ills of the timing model analysis , where again it was clear that many of the timing model parameters were poorly constrained .
i have some ideas about how to speed up the interminably slow time domain analysis .
there the bottle neck comes from having to invert a large non - sparse matrix .
suppose however that we have found an approximate form for the matrix .
we can then use the cholesky factors to transform to a data representation where the a samples are approximately uncorrelated in time and space ( _ i.e. _ in both @xmath13 and @xmath85 ) . in this transformed representation `` nearby '' correlation matrices will be almost diagonal , and fast perturbative methods can be used to compute their inverse .
this work was supported by nasa grant nnx07aj61 g .
g. hobbs , f. jenet , k. j. lee , j. p. w. verbiest , d. yardley , r. manchester , a. lommen and w. coles _ et al .
_ , s. babak , j. g. baker , m. j. benacquista , n. j. cornish , j. crowder , s. l. larson , e. plagnol and e. k. porter _
et al . _ , class .
grav . * 25 * , 184026 ( 2008 ) | this is a very brief summary of the techniques i used to analyze the ipta challenge 1 data sets .
i tried many things , and more failed than succeeded , but in the end i found two approaches that appear to work based on tests done using the open data sets .
one approach works directly with the time domain data , and the other works with a specially constructed fourier transform of the data .
the raw data was run through tempo2 to produce reduced timing residuals for the analysis .
standard markov chain monte carlo techniques were used to produce samples from the posterior distribution function for the model parameters .
the model parameters include the gravitational wave amplitude and spectral slope , and the white noise amplitude for each pulsar in the array .
while red timing noise was only included in dataset 3 , i found that it was necessary to include effective red noise in all the analyses to account for some of the spurious effects introduced by the tempo2 timing fit .
this added an additional amplitude and slope parameter for each pulsar , so my overall model for the 36 pulsars residuals has 110 parameters . as an alternative to using an effective red noise model , i also tried to simultaneously re - fit the timing model model while looking for the gravitational wave signal , but for reasons that are not yet clear , this approach was not very successful .
i comment briefly on ways in which the algorithms could be improved .
my best estimates for the gravitational wave amplitudes in the three closed ( blind ) data sets are : ( 1 ) @xmath0 ; ( 2 ) @xmath1 ; and ( 3 ) @xmath2 . |
it is well known that the performance of a wireless network can be significantly improved by cooperative transmission among nodes in the network .
many cooperative transmission designs aim to exploit cooperative diversity that is inherently present in the network .
such designs have been suggested in @xcite for cellular networks .
recently there has been much interest in achieving cooperative diversity in a classical three - node relay channel @xcite , which represents the simplest wireless network that can derive advantages from cooperative transmission .
the relay channel has been thoroughly studied in @xcite .
bounds on the capacity have been given for the general relay channel , and the capacity has been calculated for the special case of degraded relay channels .
the coding techniques suggested in @xcite assume that the relay can operate in a full - duplex manner ; i.e. , it can transmit and receive at the same time .
it is commonly argued that full - duplex operation is not practical for most existing wireless transceivers .
thus the restriction of half - duplex operation at the relay is usually considered in cooperative transmission designs .
since the relay can not transmit and receive simultaneously , a time - division approach is employed in half - duplex relaying @xcite .
the source first transmits to the destination , and the relay listens and `` captures '' @xcite the transmission from the source at the same time .
then the relay aids the transmission by sending processed source information to the destination .
note that the source may still send data to the destination when the relay transmits .
several techniques to process and forward the received data by the relay have been suggested .
these techniques include the decode - and - forward ( df ) and amplify - and - forward ( af ) approaches @xcite . in the df approach , the relay decodes the received signal from the source and then forwards a re - encoded signal to the destination . in the af approach ,
the relay simply amplifies and forwards the signal received from the source to the destination .
the performance of the df approach is limited by the capability of the relay to correctly decode the signal received from the source .
this in turn depends on the quality of the link from the source to the relay . on the other hand
, the af approach performs poorly in low signal - to - noise ratio ( snr ) situations in which the relay forwards mainly noise to the destination .
in addition , the time - division approach leads to rate losses that are significant when the relay channel is to support high rates . some enhanced versions of the af and df approaches have been proposed to solve the rate loss problem . a distributed space - time - coding protocol is developed in @xcite .
an incremental af technique which requires feedback from the destination to the source is developed in @xcite .
the non - orthogonal af and dynamic df techniques suggested in @xcite allow the duration of the relay listening to the transmission from the source to adapt to the condition of the link from the source to the relay .
in particular , the dynamic df technique is shown to be superior to all the cooperative diversity techniques ( except perhaps the incremental relaying techniques ) mentioned above . a bursty af technique is also suggested in @xcite to solve the noise forwarding problem of the af approach when the snr is low .
it is shown that the bursty af technique achieves the best outage performance at the asymptotically low snr regime .
we note that all these cooperative diversity techniques mentioned so far are designed with the constraint that channel state information is not available at the source and the relay . some practical code designs for the df and space - time - coding approaches
have been suggested in @xcite and @xcite , respectively . when the links in the relay channel suffer from slow fading , it is conceivable that the channel state information ( or at least the channel quality information ) can be estimated and passed to the nodes .
the source and relay may then use this information to optimize the cooperative protocol to achieve better performance .
such a design has been considered in @xcite , in which optimal power control is performed at the source and relay in order to maximize the ergodic rates achieved by the df and compress - forward approaches . in this paper , we assume that the channel state information is available , and we develop time - division cooperative diversity designs that perform well in both high - rate and low - rate scenarios .
the main distinguishing feature of the proposed approach , compared with the cooperative designs mentioned above , is that we do not employ the approach of the relay `` capturing '' the transmission from the source to the destination .
instead , we divide the information to be sent to the destination into three flows .
the source employs cooperative broadcasting @xcite to intentionally send two _ distinct _ flows of data to the relay and destination , respectively , in the first time slot .
the relay helps to forward , in the df manner , the data that it receives to the destination in the second time slot , during which the source concurrently sends the remaining flow of data to the destination .
the transmit powers of the source and relay as well as the durations of the time slots are optimized according to the link conditions and the rate requirement .
this constitutes a form of optimal flow control .
due to the df nature of the proposed design , there is an implicit restriction on the decoding delay .
thus we will employ the capacity - versus - outage framework @xcite to evaluate the performance of the proposed design .
we will show that the proposed design can efficiently achieve cooperative diversity in both high - rate and low - rate scenarios . in particular , when the rate requirement is asymptotically small , the outage performance of the proposed design approaches that of full - duplex relaying with df , giving a second - order diversity performance . on the other hand ,
when the rate requirement is asymptotically large , the proposed approach still gives a close - to - second - order diversity performance .
moreover , the design also gives the best diversity - multiplexing tradeoff @xcite possible for the relay channel .
together with the application of optimal long - term power control @xcite , the design can give very good delay - limited rate performance again in both low - rate and high - rate scenarios .
we note that the two basic building blocks for the proposed approach are cooperative broadcasting ( cb ) in the first time slot and multiple access ( ma ) in the second time slot .
the combination of cb and ma allows distinct flows of data be sent through the relay and through the direct link from the source to the destination , and hence can be viewed as a generalized form of routing .
a practical advantage of the proposed design is that the basic building blocks are the well known cb and ma approaches .
practical ma coding designs have been well studied , e.g. see @xcite , while practical cb coding designs are currently available @xcite@xcite .
consider a classical three - node relay network , which consists of a source node 1 , a relay node 2 , and a destination node 3 as shown in fig .
[ f : relay ] .
we assume that each link in the figure is a bandpass gaussian channel with bandwidth @xmath2 and one - sided noise spectral density @xmath3 .
let @xmath4 denote the power gain of the link from node @xmath5 to node @xmath6 .
the link power gains are assumed to be independent and identically distributed ( i.i.d . ) exponential random variables with unit mean .
this corresponds to the case of independent rayleigh fading channels with unit average power gains .
the results in the sequel can be easily generalized to include the case of non - uniform average power gains . in this section ,
we consider the case in which the relay node is capable of supporting full - duplex operation .
our goal is to support an information rate of @xmath7 nats / s / hz from the source to the destination .
we assume that the link power gains change slowly so that they can be estimated , and hence the power gain information is available at all nodes .
the source and relay nodes can make use of this information to control their respective transmit power and time so that the total transmit energy is minimized . for convenience ,
we consider a slotted communication system with unit - duration time slots .
let @xmath8 be the total transmit energy of the source and relay needed to support the transmission of @xmath9 nats of information from the source to the destination in a time slot .
since the duration of a time slot is one , @xmath8 is also the total average transmit power required . although interpreting @xmath8 as the total average power may not carry any significant physical meaning , it is customary to speak of `` power '' rather than `` energy '' in for communication engineers .
unless otherwise stated , we will hereafter consider a normalized version of @xmath8 , namely the rate - normalized overall signal - to - noise ratio ( rnsnr ) of the network : @xmath10 the rnsnr can be interpreted as the additional snr , in db , needed to support the required rate of @xmath7 nats / s / hz , in excess of the snr required to support the same rate in a simple gaussian channel with unit gain .
this normalization is convenient as we will consider asymptotic cases when @xmath7 approaches zero and infinity .
we note that the use of the rnsnr to characterize our results has two important implications .
first , since the total transmit energy of the source and relay is used in defining the rnsnr , no individual power limits are put on the source and relay .
the results in this paper can be viewed as bounds if additional individual power limits are imposed .
our choice of focusing on the total energy comes from the viewpoint that the relay channel considered forms a small component of a larger wireless network . in this sense
, it is fairer to compare the total transmit energy incurred in sending information from the source to the destination by employing cooperative diversity to that incurred in direct transmission .
second , the normalization by the factor @xmath11 implies that the additional snr in db to combat fading can only be a _
constant _ over the snr required to achieve the target rate in a gaussian channel , regardless of the rate requirement .
that is , we restrict the snr to increase at the same rate as in a gaussian channel to cope with increases in the transmission rate through the relay channel . in a sense , this restriction enforces efficiency of energy usage . employing well known capacity bounds on the relay channel @xcite
, we can obtain the following bounds on the rnsnr to support required spectral efficiency of @xmath7nats / s / hz . [ thm : fd ] for any fixed positive link power gains @xmath12 , @xmath13 , and @xmath14 , define the bound @xmath15 and @xmath16 then @xmath17 is a sufficient condition in order to support the rate of @xmath7nats / s / hz from the source to destination .
also @xmath18 is a necessary condition in order to support the rate of @xmath7nats / s / hz from the source to destination .
see appendix [ app : fd ] .
the rnsnr @xmath19 is achieved by the df approach employing the block markov coding suggested in @xcite .
we also need to optimally allocate transmit energy between the source and relay nodes .
in addition , the availability of channel state information ( both magnitudes and phases of the fading coefficients of all three links ) as well as symbol timing and carrier phase synchronization at all the three nodes are implicitly assumed . the lower bound @xmath20 is based on the max - flow - min - cut bound in @xcite .
no known coding technique can achieve this bound .
in this section , we will consider the more practical scenario in which the relay node operates in the following half - duplex fashion .
we partition each unit time slot into two sub - slots with respective durations @xmath21 and @xmath22 , where @xmath23 . in the first time slot
, the source transmits while the relay and destination receive . in the second time slot ,
the source and relay transmit , and the destination receives . based on this half - duplex mode of operation , we will describe two cooperative communication protocols that make use of the two basic components of cooperative broadcasting ( cb ) from the source to the relay in the first time slot and multiple access ( ma ) from the source and relay in the second time slot
. the first protocol does not require phase synchronization among the three nodes , while the second protocol does so . in this protocol ,
the information from the source to the destination is divided into three flows of data @xmath24 , @xmath25 , and @xmath26 , where @xmath27 . in the first time slot
, the source sends , via cb , two flows of rates @xmath28 and @xmath29 to the destination and relay , respectively . in the second time slot ,
the relay and source send , via ma , two flows of rates @xmath30 and @xmath31 to the destination , respectively . the information flow of rate @xmath30 sent by the relay in the second time slot is from the flow of rate @xmath29 that it receives and decodes in the first time slot .
we choose @xmath21 , @xmath22 , @xmath24 , @xmath25 , and @xmath26 to minimize the total power transmitted by the source and relay to support the rate @xmath7nats / s / hz from the source to the destination . to determine the minimum rnsnr that can support the required rate when this protocol is employed , we start with the following lemma .
[ thm : cbma ] 1 . for @xmath32 ,
the infimum of the snr required so that the source can broadcast at rates @xmath28 and @xmath29 to the destination and relay , respectively , in the first time slot is @xmath33 for @xmath34 , @xmath35 .
2 . for @xmath36 ,
the infimum of the snr required so that the source and relay can simultaneously transmit at rates @xmath31 and @xmath30 , respectively , to the destination in the second time slot is @xmath37 for @xmath38 , @xmath39 .
see appendix [ app : cbma ] . with the help of lemma [ thm : cbma ] , we can now formulate the optimization of the parameters in protocol 1 as follows : @xmath40 where @xmath41 and @xmath42 are of the forms in lemma [ thm : cbma ] .
it is not hard to see that ( [ e : opt ] ) is a convex optimization problem and its solution provides the tightest lower bound for the snr required to support the rate of @xmath7nats / s / hz : [ thm : b1 ] let @xmath43 be the minimum value achieved in the optimization problem ( [ e : opt ] ) , normalized by the factor @xmath11 .
then @xmath44 is the infimum of the rnsnr required so that the rate of @xmath7nats / s / hz can be supported from the source to the destination by hdp1 . to describe the form of the rnsnr bound @xmath44
, we need to consider the following few cases .
this solution is established by applying the karush - kuhn - tucker ( kkt ) condition @xcite to the convex optimization problem ( [ e : opt ] ) as detailed in appendix [ app : soln ] . for notational convenience ,
we write @xmath45 as the harmonic mean the harmonic mean . ] of two real numbers @xmath46 and @xmath47 .
[ [ z_13-geq - m_hz_12z_23 ] ] @xmath48 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the solution is given by @xmath49 where @xmath21 and @xmath22 can be arbitrarily chosen as long as they satisfy the non - negativity and total - time requirements .
this corresponds to directly transmitting all data through the link from the source to destination , without utilizing the relay .
the resulting value of @xmath44 is @xmath50 [ [ z_13-m_hz_12z_23 ] ] @xmath51 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + define @xmath52 notice that @xmath53 and @xmath54 .
consider two sub - cases : 1 .
@xmath55 : + in this case , @xmath56 where the three snr terms @xmath57 , @xmath58 , and @xmath59 are respectively defined in ( [ e : ts1 ] ) , ( [ e : ts2 ] ) , and ( [ e : ts3 ] ) below .
+ the first snr term is given by @xmath60 define @xmath61 employing the well known inequalities @xmath62 for @xmath63 and @xmath64 for @xmath65 , it can be shown that @xmath66 . by simple calculus
, @xmath67 is the minimizing @xmath21 in ( [ e : ts1 ] ) above when @xmath68 .
when @xmath67 lies outside that range , the minimizing @xmath21 must be one of the boundary points .
when @xmath57 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath70 with @xmath71 and @xmath72 .
+ the second snr term is given by @xmath73 write the minimizing value of @xmath21 in the expression above as @xmath74 . when @xmath58 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath75 with @xmath76 and @xmath77 .
+ the third snr term is given by @xmath78 write the minimizing value of @xmath21 in the expression above as @xmath79 .
when @xmath59 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath80 with @xmath81 and @xmath82 .
@xmath83 : + in this case , @xmath84 where the three snr terms @xmath85 , @xmath86 , and @xmath87 are respectively defined in ( [ e : hs1 ] ) , ( [ e : hs2 ] ) , and ( [ e : hs3 ] ) below .
+ the first snr term is given by @xmath88 + \frac{1-t_{1}}{z_{23 } } \left[e^{k/(1-t_{1})}-1\right ] \right\}. \label{e : hs1}\ ] ] write the minimizing value of @xmath21 in the expression above as @xmath89 .
when @xmath85 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath90 with @xmath91 and @xmath92 .
+ the second snr term is given by @xmath93 write the minimizing value of @xmath21 in the expression above as @xmath94 .
when @xmath86 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath95 with @xmath96 and @xmath97 .
+ the third snr term is given by @xmath98 write the minimizing value of @xmath21 in the expression above as @xmath99 .
when @xmath100 is the minimum among the three terms inside the @xmath69 operator in ( [ e : b1a ] ) , the corresponding solution to the optimization problem ( [ e : opt ] ) is given by @xmath101 with @xmath102 and @xmath103 .
we are interested in characterizing the required rnsnr in the asymptotic scenarios as the required rate @xmath7 approaches zero and infinity , respectively .
the following corollary of theorem [ thm : b1 ] and the description of @xmath44 above provides such characterization : [ thm : b1asymp ] 1 .
@xmath44 is continuous and non - decreasing in @xmath7 for all @xmath104 .
2 . @xmath105 3 .
@xmath106 4 .
@xmath44 is continuous ( except at @xmath107 ) and non - increasing in each of @xmath13 , @xmath12 and @xmath14 for all @xmath108 .
see appendix [ app : b1asymp ] . from the solution of the optimization problem described in section [ se : b1k ] ( see the form of solution under ( [ e : hs1 ] ) ) , we observe that for a sufficiently low rate requirement , the most energy - efficient transmission strategy is to select between the direct link from the source to the destination and the relay path from the source to the relay and then to the destination .
the choice of which path to take is determined by comparing the power gains of the two paths .
we note that the power gain of the relay path is specified by the harmonic mean of the power gains of the links from the source to the relay and from the relay to the destination .
the form of @xmath109 in part 2 ) of corollary [ thm : b1asymp ] also suggests this strategy .
when the rate requirement is sufficiently high , the optimal strategy ( see the form of solution under ( [ e : ts1 ] ) ) is again to compare the path gains of the direct and relay paths .
if the direct path is stronger , all information is still sent through this path .
different from the low - rate case , if the relay path is stronger , most of the information is still sent through the direct path . only a fixed amount ( depends on the link power gains , but not on the rate regardless of
how high it is ) of information is sent through the relay path .
the reduction of this fixed amount of data through the direct path has the equivalent effect of improving the fading margin of the direct path and hence provides diversity advantage .
unlike the low - rate case , this strategy is not readily revealed by the form of @xmath110 in part 3 ) of corollary [ thm : b1asymp ] . in this protocol , the information from the source to the destination is again divided into three flows of data @xmath24 , @xmath25 , and @xmath26 , where @xmath27 . in the first time slot , the source sends , via cb , two flows of rates @xmath28 and @xmath29 to the destination and relay , respectively , as before . in the second time slot , the relay sends the information that it receives in the first time slot to the destination with a flow of rates @xmath30 .
the source , on the other hand , simultaneously sends two flows of information to the destination in the second time slot .
the first flow is the exact same flow of rate @xmath30 sent by the relay .
the other flow has rate @xmath31 containing new information .
like before , we choose @xmath21 , @xmath22 , @xmath24 , @xmath25 , and @xmath26 to minimize the total power transmitted by the source and relay to support the rate @xmath7nats / s / hz from the source to the destination . to send the same flow of data ( with rate @xmath30 ) in the second time slot , the source and relay use the same codebook .
the codeword symbols from the source and relay are sent in such a way that the corresponding received symbols arrive at the destination in phase and hence add up coherently . in order to do so , the source and relay need to be phase synchronized and to have perfect channel state information of the links .
we note that these two assumptions are also needed in the full - duplex approach described in section [ se : fd ] .
in addition , the codebooks used by the source to send the two different flows in the second time slot are independently selected so that the transmit power of the source is the sum of the power of the two codewords sent .
since the transmission procedure is the same as that of hdp1 in the first time slot , lemma [ thm : cbma ] part 1 ) gives the minimum snr that can support the required cb transmission in the first time slot .
the minimum snr required in the second time slot is given by the following lemma : [ thm : ipma ] for @xmath36 , suppose that the source transmits a flow of data at rate @xmath31 to the destination in the second time slot
. then the infimum of the snr required so that the source and relay can jointly send another in - phase flow of data at rate @xmath30 to the destination in the second time slot is @xmath111 for @xmath38 , @xmath112 .
see appendix [ app : ipma ] .
let us define @xmath113 .
then we note that the expression of @xmath114 above can be obtained by putting @xmath115 in place of @xmath14 in the expression of @xmath116 in lemma [ thm : cbma ] .
this means that as far as minimum snr is concerned , hdp2 is equivalent to hdp1 with the power gain of the link from the relay to the destination specified by @xmath115 instead . using this equivalence
, we obtain the following counterparts of theorem [ thm : b1 ] and corollary [ thm : b1asymp ] for hdp2 : [ thm : b2 ] let @xmath117 be obtained by replacing @xmath14 with @xmath118 in the description of @xmath44 given in section [ se : b1k ] . then @xmath119 is the infimum of the rnsnr required so that the rate of @xmath7nats / s / hz can be supported from the source to the destination by hdp2 .
we note that @xmath120 since hdp1 can be seen as an unoptimized version of hdp2 with zero power assigned to the transmission of the flow of rate @xmath30 from the source to the destination during the second time slot .
[ thm : b2asymp ] 1 .
@xmath119 is continuous and non - decreasing in @xmath7 for all @xmath104 .
2 . @xmath121 3 .
@xmath122 4 .
@xmath119 is continuous ( except at @xmath107 ) and non - increasing in each of @xmath13 , @xmath12 and @xmath14 for all @xmath108 . in parts 2 ) and
3 ) , @xmath123 , @xmath124 , and @xmath125 are the same as @xmath126 , @xmath127 , and @xmath67 , respectively , with @xmath14 replaced by @xmath115 .
in this section , we evaluate the performance of hdp1 and hdp2 , particularly in comparison to that of full - duplex relaying . as mentioned previously , we model the link power gains @xmath13 , @xmath12 , and @xmath14 as i.i.d . exponential random variables with unit mean .
the fading process is assumed to be ergodic and varies slowly from time slot to time slot .
hence the minimum rnsnr needed to support a given rate , or equivalently the maximum achievable rate for a given rnsnr , is a random variable .
thus we need to consider its distribution .
moreover , the two protocols , namely hdp1 and hdp2 , considered in section [ se : hd ] are based on the df approach .
the relay needs to decode in the first time sub - slot and then re - encode to forward to the destination in the second sub - slot .
hence the decoding delay is implicitly limited to one time slot . as a result
, the maximum ergodic rate achieved with optimal power control and infinite decoding delay @xcite does not apply here .
instead we will consider the capacity - versus - outage approach of @xcite ( see also @xcite ) that leads to performance measures like the outage probability @xcite , @xmath128-achievable rate @xcite , diversity - multiplexing tradeoff @xcite , and delay - limited achievable rate @xcite .
outage probability is defined as the probability of the event that the rate @xmath7 can not be supported at the rnsnr @xmath129 .
let us denote the outage probabilities of full - duplex relaying , full - duplex relaying with df , half - duplex relaying using hdp1 , and half - duplex relaying using hdp2 by @xmath130 , @xmath131 @xmath132 , and @xmath133 , respectively
. then by theorems [ thm : fd ] , [ thm : b1 ] , and [ thm : b2 ] , we have @xmath134 using these , we can obtain the following bounds on the outage probabilities .
let @xmath135 and @xmath136 be real - valued functions and @xmath137 be a constant .
we say that the function @xmath135 is of order @xmath138 asymptotically , denoted by @xmath139 , if @xmath140 .
moreover , we denote the @xmath141th - order modified bessel function of the second kind by @xmath142 .
[ thm : outprob ] 1 . for all @xmath104 , @xmath143 2 . for all @xmath104 , @xmath144 3 . for all @xmath104 , @xmath145 e^{-z }
dz ~\sim\mathcal{o}\left(\frac{4\log s}{s^2}\right).\end{aligned}\ ] ] 4 . for all @xmath104 , @xmath146 \cdot \left [ 1 - e^{-\frac{1}{s}}\right ] ~\sim \mathcal{o}\left(\frac{2}{s^2}\right).\ ] ] equality above
is achieved when @xmath7 approaches @xmath147 .
5 . for all
@xmath104 , @xmath148 dz ~\sim\mathcal{o}\left(\frac{4\log s}{s^2}\right).\end{aligned}\ ] ] 6 . for all @xmath104 ,
@xmath149 equality above is achieved when @xmath7 approaches @xmath147 .
see appendix [ app : outprob ] .
the various bounds in this theorem are illustrated in fig .
[ f : bound ] . for comparison purpose ,
it is easy to verify that the outage probability for direct transmission from the source to destination is @xmath150 . from parts 1 ) and 6 ) of the theorem , we see that @xmath151 .
hence full - duplex relaying provides a second - order diversity outage performance as expected .
in addition , when the rate requirement is small and the rnsnr is large , the loss in outage performance due to the restriction of half - duplex relaying is at most @xmath152db by using hdp2 .
if phase synchronization between the source and relay is impractical , then employing hdp1 results in an additional loss of about @xmath153db . comparing parts 2 ) and 6 )
, we see that hdp2 achieves the same outage performance as full - duplex relaying based on df at asymptotically small rates .
all these observations are readily illustrated in fig .
[ f : bound ] .
when the rate requirement increases , the loss of half - duplex relaying starts to increase . in figs .
[ f : p1 ] and [ f : p2 ] , we plot the outage probabilities achieved using hdp1 and hdp2 , respectively . in each of the figures
, we include the outage probabilities when the rate requirement approaches @xmath147 , @xmath154 , @xmath155 , @xmath156 , and @xmath157 bits / s / hz . for comparison , we also plot the lower bounds on outage probabilities for full - duplex relaying in parts 1 ) and 2 ) of theorem [ thm : fd ] and the outage probability for direction transmission in the figures .
all the results corresponding to hdp1 and hdp2 in the figures are obtained using monte carlo calculations . from fig .
[ f : p1 ] , for hdp1 , we see that the loss , with respect to full - duplex relaying at the outage probability of @xmath158 , is at most @xmath159db at @xmath154 bits / s / hz . the loss increases to about @xmath160db and @xmath161db when the rate increases to @xmath155 and @xmath156 bits / s / hz , respectively .
a similar trend is observed in fig .
[ f : p2 ] for hdp2 . at @xmath154 bits / s / hz
, the loss is about @xmath162db
. the loss increases to @xmath163db and @xmath164db when the rate increases to @xmath155 and @xmath156 bits / s / hz , respectively .
moreover , at all the values of @xmath7 considered for both hdp1 and hdp2 , the simulation results seem to indicate that the outage probability is of the order of @xmath165 for some constant @xmath137 , whose value is different for the different cases . when the rate requirement becomes asymptotically large , theorem [ thm : outprob ] parts 3 ) and 5 ) state that the outage probabilities for hdp1 and hdp2 are at most of order @xmath166 .
this implies that they both give close - to - second - order diversity performance at asymptotically high rates . from the simulation results shown in figs .
[ f : p1 ] and [ f : p2 ] , it appears that the outage probabilities for both hdp1 and hdp2 do in fact have the order of @xmath167 , where @xmath137 is about @xmath168 .
this corresponds to a performance loss of about @xmath169db at the outage probability of @xmath158 , and the bound in parts 3 ) and 5 ) is about @xmath162db loose ( cf .
[ f : bound ] ) .
we also note that hdp2 does not improve the outage performance , compared to hdp1 , at asymptotically large rates .
this is contrary to the finite rate cases in which hdp2 does provide performance advantage over hdp1 , although the amount of advantage decreases as the rate requirement increases . in summary , hdp1 seems to be of higher practical utility than hdp2 since the former does not require phase synchronization between the source and relay , while it only suffers from a performance loss of about @xmath153db . by using the standard sampling representation @xcite , the input - output relationship of the 3-node relay channel over a time slot can be written as @xmath170 where @xmath171 , @xmath172 , @xmath173 , @xmath174 , @xmath175 , and @xmath176 are the @xmath177-element transmit symbol vector at the source , transmit symbol vector at the relay , receive symbol vector at the destination , receive symbol vector at the relay , gaussian noise vector at the destination , and gaussian noise vector at the relay , respectively .
the dimension @xmath178 and is assumed to be large . conditioned on the link gain vector @xmath179 $ ] , the channel is memoryless and described by the gaussian conditional pdf @xmath180 .
an @xmath181-code @xcite over a time slot is one that consists of the encoding and decoding functions described in @xcite allowing one of @xmath182 messages to be sent from the source to destination in a time slot , and achieves the average ( averaged over all codewords sent at the source and relay , link and noise realizations ) error probability of @xmath183 , while the maximum ( over all codewords ) total transmit energy used in the time slot does not exceed @xmath184 .
since the link gain vector is available at all nodes , we allow the transmit powers of the source and relay to vary as functions of the link gains .
this corresponds to the application of power control @xcite . as a result
, the power control scheme is included implicitly in the code , and @xmath184 can be in general a function of the link gain vector @xmath185 . in most cases , we are interested in performing power control to minimize @xmath183 .
the rate @xmath7 is @xmath186-achievable if there exists a sequence of @xmath181-codes satisfying @xmath187 , @xmath188 , and @xmath189 almost surely ( a.s . ) . for half - duplex relaying , when the relay listens ( transmits ) , its transmit ( receive ) symbols
are restricted to zero . in the previous sections
, we have assumed that the relay first listens for @xmath21 seconds in a time slot and then transmits in the remaining @xmath22 seconds . in this case
, it is more convenient to describe the channel by the cb and ma conditional pdfs , @xmath190 and @xmath191 , for the first and second sub - slots , respectively .
we will also say that the rate @xmath7 is @xmath186-achievable with hdp1 ( hdp2 ) if there exists a sequence of @xmath181-codes with the cb and ma coding in the first and second sub - slots as described in hdp1 ( hdp2 ) , satisfying @xmath192 , @xmath193 , and @xmath194 a.s .
the theorem below states that the various @xmath128-achievable rates are characterized by the corresponding outage probabilities defined in the previous section .
[ thm : ec ] 1 . for all @xmath195 and @xmath196 , if the rate @xmath7 is @xmath186-achievable , then + @xmath197 .
2 . for all @xmath195 ,
the rate @xmath7 is @xmath186-achievable if @xmath198 .
3 . for all @xmath195 ,
the rate @xmath7 is @xmath186-achievable with hdp1 if @xmath199 .
4 . for all @xmath195 ,
the rate @xmath7 is @xmath186-achievable with hdp2 if @xmath200 .
see appendix [ app : ec ] .
it is also interesting to investigate the diversity - multiplexing tradeoff of @xcite for hdp1 and hdp2 . to this end
, we need to follow @xcite to change the parameterization of the outage probabilities from @xmath201 to @xmath202 , where @xmath203 is the snr and @xmath204 is the multiplexing gain ( @xmath205 ) defined by @xmath206 with the parameterization @xmath202 , the diversity orders @xcite achieved by hdp1 and hdp2 are defined as @xmath207 where @xmath208 is the average error probability of hdp@xmath5 at snr @xmath203 and multiplexing gain @xmath204 , for @xmath209 and @xmath210 , respectively .
then the diversity orders can be readily obtained in the following corollary of theorems [ thm : outprob ] and [ thm : ec ] .
[ thm : dm ] for @xmath209 and @xmath210 , @xmath211 .
hence hdp1 and hdp2 achieve the maximum diversity advantage possible for the relay channel when link gain information is available at all nodes . first , by theorem [ thm : ec ] , @xmath212 for @xmath209 and @xmath210 .
also notice that since @xmath213 , @xmath214 . hence @xmath215 . as a result , applying parts 3 )
6 ) of theorem [ thm : outprob ] with the parameterization @xmath202 , for sufficiently large @xmath203 and @xmath216 , @xmath217 where @xmath218 and @xmath219 . applying @xmath220 , dividing the result by @xmath221 , and finally taking limit as @xmath222 on each item in the inequality equation above give @xmath223 for @xmath209 and @xmath210 . on the other hand ,
part 1 ) of theorem [ thm : outprob ] and part 1 ) of theorem [ thm : ec ] force the error probability of any transmission scheme over the relay channel to be larger than @xmath224 for all @xmath196 . as a result , the maximum possible diversity order of any transmission scheme over the relay channel is @xmath225 .
thus we have the desired result .
when the average error probability decreases to zero , the @xmath128-achievable rates becomes the delay - limited rates @xcite .
we calculate the delay - limited rates achievable by hdp1 and hdp2 in this section .
we first employ the following definition @xcite as our definition of delay limited rate : the rate @xmath7 is @xmath8-achievable if there exists a sequence of @xmath181-codes over a time slot , satisfying @xmath226 , @xmath227 , and @xmath194 a.s . .
unfortunately , part 1 ) of theorem [ thm : ec ] forces the @xmath8-achievable rate of any transmission scheme over the relay channel to be zero as long as @xmath8 is finite .
this is due to the rayleigh fading nature of the links and the restriction that the total transmit energy in each time slot needs to be bounded by @xmath8 .
it turns out that more meaningful results can be obtained if we relax the latter restriction .
recall that the link power gains vary independently from time slot to time slot . with power control to maintain the error probability @xmath183 , the total transmit energy @xmath184 ( a function of @xmath185 ) of an @xmath181-code may vary from time slot to time slot .
this may require the total transmit energy to be very large in the worst faded time slots . as a relaxation of the transmit energy constraint ,
we require the average total transmit energy per time slot over many time slots to be bounded .
then the ergodicity of the fading process requires @xmath228 $ ] to be bounded .
this relaxation motivates the following definition : the rate @xmath7 is _ long - term _
@xmath8-achievable if there exists a sequence of @xmath181-codes over a time slot , satisfying @xmath226 , @xmath188 , and @xmath229 \leq p_t$ ] .
the following theorem then specifies the delay - limited rates achievable by hdp1 and hdp2 : [ thm : dlc ] for @xmath104 , define @xmath230 ( e^k - 1 ) n_0 w \\
p^{\mathrm{df}}_t(k ) & = & e[b_{\mathrm{df } } ] ( e^k - 1 ) n_0 w \\
p^{1}_t(k ) & = & e[b_{1}(k ) ] ( e^k - 1 ) n_0 w \\
p^{2}_t(k ) & = & e[b_{2}(k ) ] ( e^k - 1 ) n_0 w.\end{aligned}\ ] ] then @xmath231 . 1 . if the rate @xmath7 is long - term @xmath8-achievable , then @xmath232 .
the rate @xmath7 is long - term @xmath233-achievable .
the rate @xmath7 is long - term @xmath234-achievable with hdp1 .
the rate @xmath7 is long - term @xmath235-achievable with hdp2 .
see appendix [ app : dlc ] . in fig .
[ f : dlrate ] , we plot the rate @xmath7 against @xmath236 for @xmath237 . for @xmath238 ,
notice that @xmath239(e^k-1 ) \leq \frac{p_t^{i}(k)}{n_0w } \leq e[b_i(\infty)](e^k-1)\ ] ] by corollaries [ thm : b1asymp ] and [ thm : b2asymp ] , respectively . in each case
, we plot the lower and upper bounds instead .
the true @xmath236 curve lies between the bounding curves .
also the true curve approaches the lower bound when @xmath7 is small and the upper bound when @xmath7 is large . for comparison , we also plot the curve @xmath240 which corresponds to the snr required to achieve the rate @xmath7 in an additive white gaussian noise ( awgn ) channel with unit power gain .
thus , at each rate , the loss of performance , with respect to an awgn channel , in db for approach @xmath5 due to link fading and the restriction of one - slot decoding delay is @xmath241_{\mathrm{db } } = 10\log_{10}(e[b_i])$ ] .
the results obtained from numerical calculations are shown in table [ tb : loss ] . from the table
, we see that the loss when employing full - duplex relaying is between @xmath242db and @xmath243db , where the upper limit on the loss can be achieved by df with optimal power control .
the loss when using hdp1 with optimal power control ranges from @xmath244db to @xmath245db , while the loss when using hdp2 with optimal power control is between @xmath246db and @xmath247db .
the loss of performance of half - duplex relaying with respect to full - duplex relaying is at most @xmath248db .
this loss happens when the rate requirement is very large and hdp1 is employed .
when the rate is very small , the loss drops down to at most @xmath249db with the use of hdp2 . with the delay - limited rate as performance measure , hdp1 once again appears to be a good tradeoff between complexity and performance .
the maximum loss when using hdp1 instead of hdp2 is only @xmath250db .
with channel state information available at all nodes , we have shown that a half - duplex cooperative transmission design , based on optimizing distinct flows through the direct link from the source to the destination and the path via the relay , can effectively harness diversity advantage of the relay channel in both high - rate and low - rate scenarios .
specifically , the proposed design gives outage performance approaching that of full - duplex relaying using decode - and - forward at asymptotically low rates .
when the rate requirement becomes asymptotically large , the design still gives a close - to - second - order outage diversity performance .
the design also gives the best diversity - multiplexing tradeoff possible for the relay channel . with optimal long - term power control over the fading relay channel
, the proposed design can give delay - limited rate performance that is within a few dbs of the capacity performance of the additive white gaussian channel in both low - rate and high - rate scenarios .
in addition to the good performance , a perhaps more important advantage of the proposed relaying design is that only flow - level design is needed to optimize the use of the rather standard components of cooperative broadcasting and multiple access .
this advantage makes generalizations of the design to more - complicated relay networks manageable . in general
, the availability of channel information at the nodes appears to simplify cooperative transmission designs .
thus it is worthwhile to investigate how to effectively spread the channel state information in a wireless network .
let @xmath252 be the fraction of the total energy , @xmath8 , allocated to the source node .
then the transmit energy of the relay node is @xmath253 . by (
* theorem 1 ) ( also see @xcite ) , the following rate is achievable by the relay channel when the relay decodes and re - encodes its received signal : @xmath254\right ) , \right .
\nonumber \\ & & ~~~~~~~~~~~~~~~ \left .
c\left ( \frac{p_t}{n_0 w } \alpha \beta z_{12 } \right ) \right\ } , \label{e : rfdu}\end{aligned}\ ] ] where @xmath255 .
we can further maximize the rate by optimally allocating transmit energy between the source and relay , i.e. , @xmath256 where the second equality results from the fact that @xmath257 is an increasing function .
hence , the requirement that the rnsnr satisfies @xmath258 is sufficient for @xmath259 , where @xmath260 thus it reduces to solving the optimization problem in ( [ e : zfd ] ) .
under this case , notice that the first and second terms inside the @xmath69 operator are strictly decreasing and increasing in @xmath266 , respectively .
moreover the two terms equalize at some @xmath267 .
hence @xmath268 .
solving for the equalizing @xmath269 , we get @xmath270}.\ ] ] now we maximize @xmath271 over @xmath272 . for case a ) , @xmath273 . for case
b ) , a direct but tedious calculation shows that @xmath274 .
it is not hard to verify that the maximum value in case b ) is larger than the maximum value in case a ) .
hence @xmath275 , and the sufficient condition is @xmath276 .
first note that the capacity of the relay channel is upper bounded by the maximum sum rate of the cb channel from the source to the relay and destination .
this cb channel is a degraded gaussian broadcast channel , and the individual rates @xmath278 and @xmath279 from the source to the destination and relay , respectively , satisfy ( * ? ? ?
* ch . 14 ) @xmath280 for any @xmath252 . to have @xmath281
, we need @xmath282\\ & = & c\left(z_{13 } \frac{p_t}{n_0 w}\right),\end{aligned}\ ] ] where the last equality is obtained by choosing @xmath283 , due to the condition that @xmath284 .
hence @xmath285 .
this lower bound corresponds to sending all information directly form the source to the destination without using the relay . for the necessary condition
, we employ the max - flow - min - cut bound of ( * ? ? ?
* theorem 14.10.1 ) to obtain an upper bound , @xmath286 , on the rate of the relay channel .
it turns out @xcite that the expression for @xmath286 is obtained simply by replacing every occurrence of @xmath12 by @xmath287 in ( [ e : rfdu ] ) above .
in addition , the power optimization procedure in case 1 ) above carries through directly for this case with every occurrence of @xmath12 replaced by @xmath287 .
thus we obtain the necessary condition as @xmath288 . 1
. the case of @xmath34 trivially requires @xmath289 , and hence @xmath290 .
so we consider @xmath291 .
if @xmath292 , we have a degraded broadcast channel during this time slot .
thus rate constraints in ( [ e : cbregion ] ) must be satisfied with @xmath293 and @xmath294 .
combining the two inequalities to remove @xmath295 , it is easy to obtain the stated lower bound @xmath296 of the snr @xmath297 .
we note that this lower bound corresponds to the optimal choice @xmath298 . interchanging the roles of @xmath13 and @xmath12 , we get the stated snr lower bound for the case of @xmath299 .
the case of @xmath38 trivially requires @xmath300 , and hence @xmath301 .
so we consider @xmath302 .
the capacity region of this gaussian ma channel is specified by ( * ? ? ?
14 ) : @xmath303 p_t}{n_0 w } \right ) , \label{e : maregion } \end{aligned}\ ] ] for any @xmath252 , where @xmath295 and @xmath304 are the fractions of the transmit power assigned to the source and relay , respectively .
we want to optimally choose @xmath295 so that the snr @xmath297 required to satisfy ( [ e : maregion ] ) is minimized .
first , suppose that @xmath305 . then rearranging the second and third inequalities in ( [ e : maregion ] )
gives @xmath306 \frac{n_0 w}{p_t } - z_{23 } \right\ } , \end{aligned}\ ] ] respectively .
combining these two inequalities to remove @xmath295 , we obtain the stated lower bound @xmath307 .
we note that the corresponding optimal choice @xmath308 .
+ interchanging the roles of @xmath13 and @xmath14 , we get the stated snr lower bound for the case of @xmath309 .
when @xmath310 , the third inequality in ( [ e : maregion ] ) gives the stated lower bound @xmath307 .
the choice of optimal @xmath295 in this case is exactly the same as the one in the case of @xmath311 ( or @xmath312 ) .
suppose that @xmath21 and @xmath22 are fixed , satisfying both the non - negativity and total - time requirements .
then we can view the optimization problem ( [ e : opt ] ) as a convex optimization problem in @xmath24 , @xmath25 and @xmath26 . rewriting it in standard form @xcite : @xmath313 since this optimization problem is convex , the rate tuple @xmath314 is a solution if it satisfies the following kkt conditions : our approach to solve the original optimization problem ( [ e : opt ] ) is to first solve the sub - problem ( [ e : optsim ] ) for each pair of @xmath322 , and then minimize @xmath323 over all allowable pairs . to this end ,
we consider the following cases and obtain solution to the optimization problem ( [ e : optsim ] ) by directly checking the kkt conditions .
note that we assume in below that both @xmath21 and @xmath22 are positive .
for @xmath38 ( @xmath34 ) , lemma [ thm : cbma ] tells us that the transmission in the first ( second ) sub - slots reduces trivially to transmission over the direct link from the source to the destination .
hence @xmath324 . 1 . for @xmath342
, the following solution satisfies the kkt conditions : @xmath343 2 . for @xmath344
, the following solution satisfies the kkt conditions : @xmath345 3 . for @xmath346
, the following solution satisfies the kkt conditions : @xmath347 1 . for @xmath349
, the following solution satisfies the kkt conditions : @xmath350 2 . for @xmath351
, the following solution satisfies the kkt conditions : @xmath352 3 . for @xmath353
, the following solution satisfies the kkt conditions : @xmath354 for cases 1)4 ) , direction substitution of the solution yields @xmath355 .
since this solution is independent of the choice of @xmath322 , the solution to the optimization problem ( [ e : opt ] ) in these four cases is simply @xmath356 .
also note that these four cases can be collectively specified by the condition @xmath357 . for case 5a ) , the three functions @xmath57 , @xmath58 , and @xmath59 respectively described in ( [ e : ts1 ] ) , ( [ e : ts2 ] ) , and ( [ e : ts3 ] ) can be obtained by direct substitution of the solutions in the 3 sub - cases ( i. , ii . , and iii , respectively ) , and then minimizing the corresponding @xmath323 over the range of @xmath21 specified in each sub - cases .
hence the final solution of the optimization problem ( [ e : opt ] ) is obtained by finding the minimum among the these three functions . for case 5b ) , a similar procedure yields the fact that the solution to the optimization problem ( [ e : opt ] ) is the minimum among the three functions @xmath85 , @xmath86 , and @xmath100 respectively described in ( [ e : hs1 ] ) , ( [ e : hs2 ] ) , and ( [ e : hs3 ] ) . 1 .
fix @xmath13 , @xmath12 , and @xmath14 . from the description of @xmath44 in section [
se : b1k ] , since @xmath44 is trivially continuous and non - decreasing in @xmath7 when under the condition @xmath48 , it suffices to consider @xmath44 under the condition @xmath51 , which is assumed for the rest of the proof .
+ for convenience , let us denote the respective functions inside the @xmath69 operators of @xmath57 , @xmath58 , @xmath358 , @xmath85 , @xmath86 , and @xmath100 by the addition of the sign @xmath359 .
define , for @xmath360 , @xmath361 notice that for any fixed @xmath360 , @xmath362 is piecewise continuous , with six pieces over the respective ranges of @xmath7 that they are defined .
also it is easy to check that at each end point where two adjacent pieces meet , the values of the pieces coincide ( and hence the definition of @xmath362 above is valid ) .
thus @xmath362 is continuous in @xmath7 .
the same argument with fixed @xmath7 shows that @xmath362 is continuous in @xmath21 .
then @xmath363 and hence is continuous in @xmath7 .
+ now to show @xmath44 is non - decreasing in @xmath7 , it suffices to show that , for each fixed @xmath364 , the functions @xmath365 , @xmath366 , @xmath367 , @xmath368 , @xmath369 , and @xmath370 are all non - decreasing in the corresponding ranges of @xmath7 that the functions are used in the definition of @xmath362 in ( [ e : tb ] ) above . to this end
, we will repeatedly employ the following form of young s inequality : @xmath371 for nonnegative @xmath46,@xmath47 , and @xmath372 .
+ fix @xmath364 .
first let us consider @xmath57 .
for @xmath7 in the corresponding range in ( [ e : tb ] ) , @xmath373 but by young s inequality , @xmath374 thus @xmath375 is non - decreasing . + next
consider @xmath376 .
for @xmath7 in the corresponding range in ( [ e : tb ] ) , @xmath377 + \frac{1-t - a_1^t}{z_{23 } } - \frac{t}{z_{12 } } + \frac{t}{z_{13 } } \right\ } \cdot \frac{e^k}{\left(e^k-1\right)^2}.\ ] ] again by young s inequality , @xmath378 and @xmath379 hence @xmath366 is non - decreasing .
finally , we note that the non - decreasing nature of the functions @xmath367 , @xmath368 , @xmath369 , and @xmath370 can be proven in the same way .
2 . when @xmath7 is sufficiently small , @xmath380 . then a simple application of lhospital s rule gives the desired result .
3 . when @xmath7 is sufficiently large , @xmath381 .
then simply taking limit gives the desired result .
4 . fix @xmath7 and @xmath21 .
augment the definition of @xmath362 in ( [ e : tb ] ) by adding @xmath382 when @xmath48 .
for the rest of the proof , this augmented @xmath362 will be considered as a function of @xmath13 , @xmath12 , and @xmath14 , despite its notation .
then an argument similar to the one in part 1 ) can be employed to show that @xmath44 is continuous in each of @xmath13 , @xmath12 , and @xmath14 for every @xmath383 , except for @xmath107 at which @xmath44 becomes infinite .
+ again similar to part 1 ) , in order to show @xmath44 is non - increasing in each of @xmath13 , @xmath12 , and @xmath14 , we only need to show that for each fixed @xmath7 and @xmath21 , the functions @xmath384 , @xmath376 , @xmath385 , @xmath386 , @xmath387 , and @xmath388 are all non - increasing in each of @xmath13 , @xmath12 , and @xmath14 , over the respective ranges of these functions shown in ( [ e : tb ] ) . indeed , this fact can be shown by verifying that the derivatives involved are all non - positive .
the only interesting case is @xmath389 , which needs the use of young s inequality : @xmath390 -1 \right\ } \\ &
\leq & -\frac{1}{z_{13}^2 } \left\ { e^k \left(a_1 a_2\right)^{-t_1(1-t_1 ) } -1\right\ } \\ & = & -\frac{1}{z_{13}^2 } \left\ { e^{k - t_1(1-t_1)(\log a_1 + \log a_2 ) } -1\right\ } \\ & \leq & -\frac{1}{z_{13}^2 } \left\ { e^{k - t_1 k - ( 1-t_1 ) k } -1\right\ } = 0 , \end{aligned}\ ] ] where the second line is due to young s inequality and the last line is due to the fact that @xmath391 and @xmath392 in the range of interest of @xmath384 . the case of @xmath38 trivially requires @xmath300 , and hence @xmath393 .
so we consider @xmath302 .
suppose that the transmit power of the relay is @xmath394 and the transmit power of the source is @xmath395 , where @xmath396 is the power employed to transmit the flow of rate @xmath30 while @xmath397 is the power of the flow of rate @xmath31 .
then the transmission procedure in the second time slot of hdp2 ( cf .
section [ se : p2 ] ) describes the transmission over an equivalent two - user gaussian ma channel in which one user of rate @xmath30 has power @xmath398 and another user of rate @xmath31 has power @xmath399 . from the capacity region of this gaussian ma channel specified by ( * ? ? ?
14 ) , @xmath396 , @xmath394 and @xmath397 must satisfy : @xmath400 to minimize the total power ( energy ) given the rates of transmission , we consider the following optimization problem : @xmath401 where @xmath402 , @xmath403 , @xmath404 , and @xmath405 .
notice that @xmath406 . it can be shown that this is a convex optimization problem .
the power tuple @xmath407 is a solution if it satisfies the following kkt conditions : the condition k1 yields @xmath413 it is then easy to check that the following solution satisfies the kkt conditions : @xmath414 then normalizing the sum of this choice of @xmath396 , @xmath394 , and @xmath397 by @xmath415 gives the stated expression of @xmath416 in lemma [ thm : ipma ] . 1 . by theorem [ thm : fd ] , @xmath422 e^{-z } dz \\ & = & \int_{0}^{\infty } e^{-z } dz - \int_{\frac{1}{s}}^{\infty } e^{-z } dz - \int_{0}^{\frac{1}{s } } e^{-\frac{2}{s } + z } dz \\ & = & 1 - 2 e^{-\frac{1}{s } } + e^{-\frac{2}{s}}.\end{aligned}\ ] ] it is also easy to see that @xmath423 .
2 . by theorem [ thm :
fd ] , @xmath424 a simple calculation shows that @xmath425 .
conditioned on the event @xmath426 , @xmath427 .
hence @xmath428 e^{-z } dz \\ & = & \int_{0}^{\frac{1}{s } } \left ( e^{-z } - e^{-\frac{1}{s } } \cdot e^{-\frac{1}{s}+z}\right ) e^{-z } dz + \cdot \int_{\frac{1}{s}}^{\infty } \left ( e^{-z } - e^{-z } \cdot 1 \right ) e^{-z } dz \\ & = & \frac{1}{2}\left(1-e^{-\frac{2}{s}}\right ) - \frac{1}{s } e^{-\frac{2}{s}}.\end{aligned}\ ] ] it is also easy to see that @xmath429 .
3 . by corollary
[ thm : b1asymp ] , @xmath430 for all @xmath104 .
note that when @xmath7 is sufficiently large , @xmath431 .
now instead of choosing the optimal @xmath67 in ( [ e : ts1 ] ) , we choose @xmath432 and normalize the suboptimal solution by the factor @xmath433 . taking limit as @xmath434
, we get @xmath435 obviously , @xmath436 because of the suboptimality of the choice @xmath432 .
thus , @xmath437 . moreover , when @xmath438 , @xmath439 where the second line is due to the concavity of the square - root function .
hence @xmath440 + by claim [ c1 ] , @xmath441 e^{-z } dz \\ & = & 1- e^{\frac{1}{s } } - \int_{0}^{\frac{1}{s } } 2z k_1(2z ) e^{-3z } dz , \end{aligned}\ ] ] and @xmath442 e^{-z } dz . \end{aligned}\ ] ] + now by repeated uses of lhospital s rule , we have @xmath443 e^{-2u } \\ & = & 1 , \end{aligned}\ ] ] where the third equality is due to the fact that the derivative of @xmath444 is @xmath445e^{-u}$ ] @xcite , and the last equality is due to the facts that @xmath446 and @xmath447 @xcite . to find the asymptotic order of @xmath448 , let us write @xmath449 dz.\ ] ] first , we note that @xmath450 then again by repeated applications of lhospital s rule , we have @xmath451 dz } { -\frac{1}{2 } u \log u } \\ & = & 8 \lim_{u\rightarrow 0 } \frac { \int_{2\sqrt{2u}}^{\infty } [ k_1(x ) + k_0(x ) ] e^{-x } dx } { -\log u - 1 } \\ & = & 4 \lim_{u\rightarrow 0 } 2\sqrt{2u } \left [ k_1(2\sqrt{2u } ) + k_0(2\sqrt{2u } ) \right ] e^{-2\sqrt{2u } } \\ & = & 4 , \end{aligned}\ ] ] where the second equality is obtained by a change of integration variable after the use of lhospital s rule . as a consequence , @xmath452 .
4 . by corollary
[ thm : b1asymp ] , @xmath453 for all @xmath104 .
in addition , the bound is achieved as @xmath7 approaches zero .
thus @xmath454 \left[1 - \frac{2}{s } k_1\left(\frac{2}{s}\right ) e^{-\frac{2}{s } } \right ] , \end{aligned}\ ] ] where the last line is due to claim [ c1 ] and the bound is achieved as @xmath455 by monotone convergence .
moreover , @xmath456 \left [ 1 - e^{-\frac{1}{s } } \right]}{1/s^2 } & = & \lim_{u\rightarrow 0 } \frac{1 - 2uk_1(2u)e^{-2u}}{u } \cdot \lim_{u\rightarrow 0 } \frac{1-e^{-u}}{u } \\ & = & 2 \cdot 1 .
\end{aligned}\ ] ] 5 . by corollary [ thm : b2asymp ] and
similar to part 3 ) , we have @xmath457 by claim [ c1 ] , @xmath458 and @xmath459 dz .
\end{aligned}\ ] ] + now by repeated uses of lhospital s rule , we have @xmath460 as derived in part 3 ) , @xmath461 .
6 . by corollary [ thm : b2asymp ] and
similar to part 4 ) , we have @xmath462 dz\\ & = & 1 - e^{-\frac{1}{s } } - \frac{2}{s^2 } k_1\left(\frac{2}{s}\right ) e^{-\frac{2}{s } } , \end{aligned}\ ] ] where the equality in the third line is established by claim [ c1 ] and the bound is achieved as @xmath455 by monotone convergence .
moreover , @xmath463e^{-2u } \\ & = & \lim_{u\rightarrow 0 } \frac{-e^{-u } - 4u[k_1(2u)+k_0(2u)]e^{-2u}}{2 } + 1\\ & = & 1/2 + 1 = 3/2 .
\end{aligned}\ ] ] to prove part 1 ) of the theorem , we employ the fano inequality as in @xcite . for the remaining parts , the achievability proofs
are based on extending the feinstein lemma @xcite to the various cases of interest . 1 .
suppose that @xmath7 is @xmath186-achievable .
hence , for any @xmath464 , there is a sequence of @xmath181-codes satisfying @xmath465 , @xmath466 , and @xmath467 a.s .
for all sufficiently large @xmath177 .
let @xmath468 be the uniform random variable representing the message being set from the source to destination . since @xmath468 is independent of @xmath185 , conditioned on the link realization @xmath469 , by the fano inequality , @xmath470 \nonumber \\ & \geq & 1 - \frac{1}{n(k-\gamma ) } [ i(m;y^n| z=(z_{13},z_{12},z_{23 } ) ) + 1 ] , \label{e : fano } \end{aligned}\ ] ] for all sufficiently large @xmath177 . since the relay channel is memoryless conditioned on @xmath469 , by the same argument as in the proof of theorem 4 in @xcite @xmath471 where @xmath472\right ) , \right . \\ & & ~~~~~~~~~
c\left.\left ( \frac{p_t+\gamma}{n_0 w } \alpha \beta [ z_{12}+z_{13 } ] \right ) \right\}. \end{aligned}\ ] ] the two conditional mutual information terms on the right hand side of the first inequality in ( [ e : cib ] ) are based on the element - wise conditional pdf @xmath473 and conditional input pdf @xmath474 based on the code as in @xcite and the second inequality is due to the fact these mutual information terms are maximized by independent gaussian inputs @xmath475 and @xmath476 @xcite and @xmath467 for sufficiently large @xmath177 . + now putting ( [ e : cib ] ) into ( [ e : fano ] ) and noting that @xmath477 , we have @xmath478 \geq e\left[\max\left\{1 - \frac { r_{\mathrm{lb}}(z_{13},z_{12},z_{23};p_t+\gamma)}{k-\gamma } - \frac{1}{n(k-\gamma)},0\right\}\right]\ ] ] for sufficiently all large @xmath177 . since @xmath479 is arbitrary , @xmath480 \geq \delta \pr\left ( r_{\mathrm{lb}}(z_{13},z_{12},z_{23};p_t ) < ( 1-\delta)k \right),\ ] ] for all @xmath196 . from theorem [ thm : fd ] , we have then @xmath481 .
we employ the approach of block encoding and parallel gaussian channel decoding suggested in @xcite .
first we divide a time slot into @xmath482 sub - slots is an integer with no loss of generality . ] .
we are to send @xmath483 messages , each coming from one of @xmath484 possibilities , in the whole time slot .
thus in each sub - slot , we have @xmath485 symbols .
let @xmath486 be the gaussian distribution with mean @xmath487 and variance @xmath488 .
consider the following code construction .
+ [ [ codebook - generation ] ] codebook generation + + + + + + + + + + + + + + + + + + + + independently generate @xmath484 @xmath489-element vectors @xmath490 with all elements in the vectors distributed according to i.i.d .
@xmath491 similarly , independently generate @xmath484 @xmath489-element vectors @xmath492 with all elements in the vectors distributed according to i.i.d .
@xmath491 . + [
[ encoding ] ] encoding + + + + + + + + + let the @xmath493th message be @xmath494 , for @xmath495 , and @xmath496 , which is known to the relay and destination beforehand .
let @xmath263 .
in the @xmath493th sub - slot , the source sends @xmath497 and the relay sends @xmath498 , where @xmath499 is the estimate of @xmath500 that the relay obtains based on the signal that it receives in the @xmath501th sub - slot . in above ,
@xmath502 and @xmath503 , where @xmath504 .. + when the code defined above is employed , we can rewrite the relationship between the output and input symbols described by ( [ e : model ] ) as @xmath505 \tilde x_2^{\tilde n , k } + n^{\tilde n , k } \nonumber \\
y_1^{\tilde n , k } & = & \sqrt{\beta z_{12}p_1 } \tilde x_1^{\tilde n , k } + \sqrt{(1-\beta)z_{12 } p_1 } \tilde x_2^{\tilde n , k } + n_1^{\tilde n , k } , \label{e : model2 } \end{aligned}\ ] ] where the superscript @xmath493 denotes the @xmath493th sub - slot .
hence the relay channel can be alternatively specified by the conditional pdf @xmath506 with @xmath507 .
note that since the channel state information is available at the source and relay , @xmath295 and @xmath266 are functions of @xmath185 in general .
this corresponds to the application of flow control .
+ [ [ decoding ] ] decoding + + + + + + + + + consider decoding of the message @xmath508 ( @xmath509 ) under the assumption that the previous message @xmath500 has been correctly decoded , and hence is known , at both the relay and destination .
fix any @xmath510 , define the sets @xmath511 in the @xmath493th sub - slot , the relay outputs @xmath512 if and only if there is a unique @xmath5 ( from @xmath154 to @xmath484 ) such that @xmath513 .
this allows the encoding steps mentioned above to continue in the @xmath514th sub - slot . in the @xmath514th sub - slot , the destination outputs @xmath512 if and only if there a unique @xmath5 such that @xmath515 .
+ [ [ error - analysis ] ] error analysis + + + + + + + + + + + + + + + let @xmath516 be the average error probability of decoding the whole time slot and @xmath517 be the event of erroneous decoding in the @xmath493th sub - slot , for @xmath518
. then @xmath519\right\}\right).\ ] ] + consider the event @xmath520 .
the message @xmath521 is correctly decoded , and hence is known , at the relay and destination , while the message @xmath500 is corrected decoded , and hence is known , at the relay . by symmetry of the code generated , we can assume @xmath522 with no loss of generality .
for @xmath523 , write @xmath524 and @xmath525 . then @xmath526 + now for @xmath527 , @xmath528 where the independence between @xmath529 and @xmath530 in the second line is due to the fact that @xmath529 and @xmath530 are jointly gaussian and uncorrelated , and the inequality on the third line follows from the definition of @xmath531 .
similarly , we can employ the definition of @xmath532 to show that @xmath533 for @xmath534 .
hence @xmath535 \right ) + \sum_{k=1}^{\tilde n } \sum_{i=2}^{m_{\tilde
n } } \pr ( e_{i , k}^1 ) + \sum_{k=1}^{\tilde n } \sum_{i=2}^{m_{\tilde n } } \pr(e_{i , k}^2 ) \\ & \leq & \pr \left ( \cup_{k=1}^{\tilde n } \left[(e_{1,k}^1)^c \cup ( e_{1,k}^2)^c \right ] \right ) + 2\tilde n e^{-\tilde n\gamma}. \end{aligned}\ ] ] + putting the codes in the @xmath489 sub - slots together , we obtain a sequence of @xmath536-codes over the time slot with @xmath516 and @xmath184 respectively satisfying @xmath537 \right)$ ] and @xmath538 a.s . + further note that as @xmath177 ( and hence @xmath489 ) becomes large @xmath539\right )
\\ \lefteqn{\frac{1}{\tilde n } \log \frac{p_{y_1^{\tilde n , k}| \tilde x_1^{\tilde n , k } , \tilde x_2^{\tilde n , k},\alpha,\beta , z } ( y_1^{\tilde n , k}|\tilde x_1^{\tilde n , k},\tilde x_2^{\tilde n , k})}% { p_{y_1^{\tilde n , k}| \tilde x_2^{\tilde n , k},\alpha,\beta , z } ( y_1^{n , k}|x_2^{n , k } ) } \stackrel{\mathrm{a.s.}}{\longrightarrow } c\left(\frac{\alpha\beta z_{13 } p_t}{n_0w } \right ) } \end{aligned}\ ] ] for all @xmath518 , when the inputs symbols are gaussian distributed as described in the code generation step above .
+ now set @xmath540 and choose @xmath295 and @xmath266 so that the maximum rate @xmath541 in appendix [ app : fd ] is achieved . since the choice of @xmath542 is arbitrary , the code construction argument above shows the existence of a sequence of @xmath543-codes over the time slot satisfying @xmath544 , @xmath545 , and @xmath538 a.s . .
from theorem [ thm : fd ] , @xmath546 .
therefore if @xmath547 , then the rate @xmath7 is @xmath186-achievable .
we construct a code that conforms to hdp1 .
fix @xmath548 and @xmath549 .
write @xmath550 and @xmath551 .
+ [ [ codebook - generation-1 ] ] codebook generation + + + + + + + + + + + + + + + + + + + + independently generate @xmath552 @xmath553-element vectors @xmath554 with all elements in the vectors distributed according to i.i.d .
@xmath491 . independently generate @xmath555 @xmath553-element vectors @xmath556 with all elements in the vectors distributed according to i.i.d .
@xmath491 . similarly ,
independently generate @xmath557 @xmath558-element vectors @xmath559 with all elements in the vectors distributed according to i.i.d .
@xmath491 . independently generate @xmath555 @xmath558-element vectors @xmath560 with all elements in the vectors distributed according to i.i.d .
+ [ [ encoding-1 ] ] encoding + + + + + + + + + let @xmath561 . a message @xmath562 , with value from @xmath154 to @xmath182 , can be indexed by the triple @xmath563 , where @xmath5 ranges from @xmath154 to @xmath564 , @xmath6 ranges from @xmath154 to @xmath565 , and @xmath493 ranges from @xmath154 to @xmath557 .
also we divide a time slot with @xmath177 symbols into two sub - slots : the first with @xmath553 symbols and the second with @xmath558 symbols . in the first sub - slot
, the source sends @xmath566 , where @xmath567 .
the relay generates an estimate , @xmath568 , of @xmath6 . in the second sub - slot
, the source sends @xmath569 and the relay sends @xmath570 , where @xmath263 .
like before , @xmath295 and @xmath266 are the power control functions depending on the link gain vector @xmath185 . + as in part 2 ) above , when this code is used , the input - output relationship of the channel can be described by @xmath571 during the first sub - slot with @xmath572 corresponding to the codeword @xmath530 , @xmath573 corresponding to the codeword @xmath574 , and @xmath575 as the input to the cb channel . in the second sub - slot
, we have @xmath576 with @xmath577 and @xmath578 corresponding to @xmath579 and @xmath580 , respectively .
+ [ [ decoding-1 ] ] decoding + + + + + + + + + we combine the decoding approaches suggested for the cb and ma channels in @xcite and @xcite , respectively . fix any @xmath510 . define the sets @xmath581 in the first sub - slot , the relay sets @xmath582 if and only if there is a unique pair @xmath583 such that @xmath584 .
this allows the encoding step in the second sub - slot mentioned above .
the destination outputs @xmath5 if and only if there is a unique @xmath5 such that @xmath585 . in the second sub - slot , the destination outputs @xmath586 if there is a unique pair @xmath586 such that @xmath587 . finally , the estimate of the message is then @xmath588 .
+ [ [ error - analysis-1 ] ] error analysis + + + + + + + + + + + + + + + because of the symmetry of the code , we can assume @xmath589 .
let @xmath516 be the average error probability of decoding . for @xmath590 , write @xmath591 and for @xmath590 and @xmath592 , @xmath593 .
for @xmath592 and @xmath594 , @xmath595 .
then @xmath596 using the definitions of @xmath597 to @xmath598 and similar to part 2 ) ( see @xcite for the detailed arguments ) , one can show that the second , third , and fourth terms on the right hand side of ( [ e : hfd1err ] ) can be bounded by @xmath599 , @xmath600 , and @xmath601 , respectively .
+ as @xmath177 becomes large , @xmath602 when the inputs symbols are gaussian distributed as described in the code generation step above . for the cases of @xmath34 and @xmath603 , the channel reduces to the case of ma and cb , respectively .
hence the corresponding subset of code construction should be employed . + now let @xmath604 , @xmath605 , and @xmath606 such that @xmath607 and @xmath27 .
choose @xmath21 , @xmath295 , @xmath266 , @xmath24 , @xmath25 , and @xmath26 as functions of @xmath185 to minimize @xmath8 .
since @xmath608 is arbitrary , by theorem [ thm : b1 ] , the code construction above shows the existence of a sequence of @xmath181-codes over the time slot with satisfying @xmath609 , @xmath610 , and @xmath611 a.s .
+ indeed , to see that theorem [ thm : b1 ] applies here , it suffices to show that the optimization solution described in section [ se : b1k ] lies within the following region @xmath612 the last three inequality coincide with the ma region specified in part 2 ) of lemma [ thm : cbma ] ( see appendix [ app : cbma ] ) . for @xmath332 , it is easy to see the third inequality is redundant in place of the first two inequalities , which coincide with the cb region in part 1 ) of lemma [ thm : cbma ] . for @xmath613 , the optimal solution specified in appendix [ app : soln ]
can be achieved by the choice of @xmath34 , hence making only the last three inequalities matter .
all the arguments are essentially the same as in part 3 ) with the modification that the source sends @xmath614 and the relay sends @xmath615 , where @xmath616 is an additional power control component , in the second sub - slot .
we sketch the proof of the theorem , which employs results from @xcite directly .
below we use the index @xmath5 to denote one of the four cases of lower bound @xmath617 , decode forward @xmath618 , hdp1 @xmath619 , and hdp2 @xmath620 . for @xmath621 , replacing @xmath8 by @xmath622n_0w$ ] in the proofs parts 2 )
4 ) of theorem [ thm : ec ] given in appendix [ app : ec ] , we can show the existence of a sequence of @xmath623-codes with @xmath624n_0w+ \gamma$ ] a.s . and
@xmath625 , for any @xmath608 , whenever @xmath177 is sufficiently large .
note that we write the rnsnr @xmath626 to highlight the use of a general power control scheme which varies the total transmit energy ( rather than setting it to a fixed value as in the original proofs ) in each time slot according to the link gains .
consider the optimal power control function @xmath627 that solves the following optimization problem : @xmath628 \leq \bar s \stackrel{\triangle}{= } \frac{p_t}{n_0w ( e^k-1)}. \end{array } \label{e : optlt}\ ] ] write @xmath629 , @xmath630 , @xmath44 , and @xmath119 as @xmath631 for @xmath632 , and @xmath210 , respectively to highlight their dependence on @xmath185 .
define @xmath633 , where @xmath634 is the distribution function of the link gain vector @xmath185 .
let @xmath635 .
then proposition 4 of @xcite can be applied to solve the optimization problem in ( [ e : optlt ] ) , provided that @xmath631 is continuous for all @xmath636 and is non - increasing in each of the elements of @xmath185 ( c.f .
lemma 2 of @xcite ) .
this latter condition is established in corollaries [ thm : b1asymp ] and [ thm : b2asymp ] for @xmath209 and @xmath210 , respectively . for @xmath637 and @xmath638
, the condition can be checked by straightforward calculus .
the resulting solution is @xmath639 and @xmath640 .
now , if @xmath641 = \lim_{s\rightarrow\infty } \mathcal{s}_i(s)$ ] is finite , setting @xmath642 $ ] will make @xmath643 and hence @xmath644 as well as @xmath645 = e[b_i(z)]$ ] . for the cases of @xmath646 , and @xmath210 , this implies that the rate @xmath7 is long - term @xmath647(e^k-1)n_0w$]-achievable , long - term @xmath648(e^k-1)n_0w$]-achievable with hdp1 , and long - term @xmath649(e^k-1)n_0w$]-achievable with hdp2 , respectively . for @xmath637 ,
a slight modification to the proof of part 1 ) of theorem [ thm : ec ] shows that if the rate @xmath7 is long - term @xmath650-achievable , then the error probabilities of the sequence of codes ( and the corresponding power control schemes @xmath651 ) must satisfy @xmath652 \leq \bar
s + \gamma$ ] and @xmath653 for all @xmath654 and any @xmath510 , whenever @xmath177 is sufficiently large .
since @xmath655 is continuous in the second argument , this requires that @xmath656 for all sufficiently large @xmath177 .
the solution of ( [ e : optlt ] ) then implies that @xmath657 $ ] . since @xmath658 for all @xmath185
, it suffices to establish the finiteness of @xmath648 $ ] . from the proof of part 3 ) of theorem [ thm : outprob ] , @xmath659 thus it in turn suffices to establish the finiteness of @xmath660 $ ] .
indeed @xmath661 & = & \int_{\frac{1}{x } < \frac{1}{y}+\frac{1}{z } } \frac{1}{x } e^{-(x+y+z ) } dxdydz + \int_{\frac{1}{x } \geq \frac{1}{y}+\frac{1}{z } } \sqrt{\frac{2}{x}}\cdot\sqrt{\frac{1}{y}+\frac{1}{z } } e^{-(x+y+z ) } dxdydz \nonumber\\ & = & \underbrace{\int_{0}^{\infty } \frac{1}{x } \left [ 1 - 2xe^{-2x}k_1(2x)\right ] e^{-x } dx}_{a } \nonumber \\ & & ~~~ + \underbrace{\int_{0}^{\infty } \sqrt{\frac{2}{x } } e^{-x } \int_{x}^{\infty } 2 \sqrt{u } e^{-2u } \left[k_0(2u)+ k_1(2u)\right ] dudx}_{b } \label{e : ehb1}\end{aligned}\ ] ] where the second equality is obtained by using the integral representations of @xmath662 and @xmath663 in @xcite . again using the property of the modified bessel functions , it is easy to check that the integrand in the integral @xmath137 on the right hand side of ( [ e : ehb1 ] ) is bounded above over the range of @xmath664 and is bounded above by @xmath665 for @xmath666
. thus @xmath137 is finite . on the other hand
, we have @xmath667 du \right)\ ] ] where the first integral on the right hand side is @xmath668 ( see @xcite ) and the second integral is finite ( see @xcite ) .
thus @xmath448 is also finite .
a. sendonaris , e. erkip , and b. aazhang , `` user cooperation diversity
part ii : implementation aspects and performance analysis , '' _ ieee transactions on communications _ , vol .
11 , pp . 19391948 ,
2003 .
j. n. laneman .
d. n. c. tse , and g. w. wornell , `` cooperative diversity in wireless networks : efficient protocols and outage behavior , '' _ ieee transactions on information theory _ ,
30623080 , dec . 2004 .
j. n. laneman and g. w. wornell , `` distributed space - time - coded protocols for exploiting cooperative diversity in wireless networks , '' _ ieee transactions on information theory _ , vol .
10 , pp . 24152425 , oct .
k. azarian , h. el gamal , and p. schniter , `` on the achievable diversity - multiplexing tradeoff in half - duplex cooperative channels , '' _ ieee transactions on information theory _ , vol .
41524172 , dec . 2005 .
m. c. valenti and b. zhao , `` capacity approaching distributed turbo codes for the relay channel , '' in _ proc .
57th ieee semiannual vehicular technology conf .
( vtc03-spring ) _ , jeju island , korea , apr .
m. janani , a. hedayat , t. hunter , and a. nosratinia , `` coded cooperation in wireless communications : space - time transmission and iterative decoding , '' _ ieee transactions on signal processing _ , vol .
2 , pp . 362371 , feb . 2004 .
e. biglieri , j. proakis , and s. shamai ( shitz ) , `` fading channels : information - theoretic and communications aspects , '' _ ieee transactions on information theory _ ,
44 , no . 6 , pp . 26192692 , oct .
1998 .
w. yu , d. p. varodayan , and j. m. cioffi , `` trellis and convolutional precoding for transmitter - based interference presubtraction , '' _ ieee transactions on communications _ ,
53 , no . 7 , pp .
12201230 , july 2005 .
m. airy , a. forenza , r. w. heath , and s. shakkottai , `` practical costa precoding for the multiple antenna broadcast channel , '' in _ proc .
ieee 2004 global telecommunications conference ( globecom 04 ) _ , vol .
39423946 , nov .
2004 .
m. o. hasna and m .- s .
alouini , `` end - to - end performance of transmission systems with relays over rayleigh - fading channels , '' _ ieee transactions on wireless communications _ , vol .
2 , no . 6 , pp .
11261131 , dec . 2003 .
xie and p. r. kumar , `` a network information theory for wireless communication : scaling laws and optimal operation , '' _ ieee transactions on information theory _ , vol .
748767 , may 2004 . | this paper describes an approach for half - duplex cooperative transmission in a classical three - node relay channel . assuming availability of channel state information at nodes , the approach makes use of this information to optimize distinct flows through the direct link from the source to the destination and the path via the relay , respectively .
it is shown that such a design can effectively harness diversity advantage of the relay channel in both high - rate and low - rate scenarios .
when the rate requirement is low , the proposed design gives a second - order outage diversity performance approaching that of full - duplex relaying .
when the rate requirement becomes asymptotically large , the design still gives a close - to - second - order outage diversity performance .
the design also achieves the best diversity - multiplexing tradeoff possible for the relay channel . with optimal long - term power control over the fading relay channel ,
the proposed design achieves a delay - limited rate performance that is only @xmath0db ( @xmath1db ) worse than the capacity performance of the additive white gaussian channel in low- ( high- ) rate scenarios . |
the rich phenomenology of high-@xmath3 superconductors @xcite has stimulated strong experimental and theoretical interest in the field of strongly correlated electron systems .
apart from the anomalously high transition temperatures , these compounds are also of interest due to their unusual normal state properties .
most of these anomalous properties are found in spectra and transport quantities , i.e. are intimately linked to the dynamics of the electronic degrees of freedom .
thus , much of the experimental and theoretical effort has concentrated on the development of an understanding of the single - particle dynamics . among the fundamental and controversial questions are whether the cuprates can be described as a fermi liquid or not and what shape and volume a possible fermi surface will have . in this connection ,
one of the most informative experimental probes of the cuprates has become the angle - resolved photoemission spectroscopy(arpes ) .
the development in this field , again largely stimulated by the interest in the physics of high-@xmath3 , has led to a tremendous increase of the angular and energy resolution @xcite .
arpes is now able to access the low - energy behavior of single - particle spectra and especially the shape and topology of the cuprate fermi surface .
this has lead to the discovery of the shadow bands@xcite and the pseudo - gap formation in the underdoped cuprates@xcite .
recently , the single - particle self energy was extracted from the arpes data , showing extremely interesting behavior especially close to @xmath4 on the fermi surface@xcite .
a large number of the more recent arpes experiments have concentrated on fermi surface mapping .
some of these experiments seem to indicate that , at least for lsco , the fermi surface switches from being hole - like , centered at @xmath2 , for low doping to being electron - like , centered at the zone center @xmath5 for high doping with a volume consistent with the doping level@xcite . in other results , particularly for bilayer compounds like ybco or bi2212@xcite , the fermi surface seems to remain hole - like independent of the doping level . especially for those latter compounds ,
further complications in the experiments arise from superstructures due to umklapp scattering and the possibility of bilayer splittings in ybco and bi2212@xcite . furthermore , in the underdoped regime one can observe additional changes in the fermi surface topology that are frequently brought in connection with the possibility of stripe formation in lsco@xcite .
the presence of a large fermi surface has been taken as a validation of luttinger s theorem ; however , other results find that the fermi surface volume , at least in the underdoped regime , is too small @xcite , which would point towards a violation of luttinger s theorem .
furthermore , recent experiments seem to indicate that the low doping fermi surface near @xmath6 actually bifurcates into two parts , one electron - like and one hole - like@xcite .
this splitting has been interpreted in terms of strong interlayer coupling@xcite , but could equally well result from shadow fermi surface formation due to coupling of the electrons to strong antiferromagnetic fluctuations around the @xmath0 points@xcite .
thus , a consistent experimental picture concerning both shape and volume of the fermi surface is at present not available and a theoretical investigation of the generic features to be expected based on a model calculation is necessary .
early in the theoretical investigation of the high-@xmath3 cuprates it was realized that the 2d hubbard model @xmath7 in the intermediate coupling regime , or closely related models like the @xmath8-@xmath9 model probably capture the essential physics @xcite . in the wake of this conjecture ,
a huge effort has been directed to the study of these models@xcite .
there is now a general consensus that the appropriate parameter regime for the cuprates is the intermediate coupling regime where the coulomb parameter @xmath10 is roughly equal to the bandwidth .
however , this is the most complicated regime of the model since both weak and strong coupling perturbative approaches fail .
exact diagonalization of small clusters @xcite suffers from strong finite - size effects , often ruling out the reliable extraction of low - energy dynamics .
conventional quantum monte carlo for finite sized systems suffers from a severe minus sign problem in this parameter regime .
the resulting data is of insufficient quality to allow for reliable calculations of dynamic quantities at low enough temperatures .
high - temperature series has provided some of the most informative results for the fermi surface topology , but it does not yield spectra , and , so far , only results for the t - j model are available@xcite .
thus , also from a theoretical point of view , the question of whether a fermi surface does actually exist and what its topology is still is a matter of debate . in that connection it is of special interest that some experiments indicate a violation of luttinger s theorem ; if true , any theory , such as flex@xcite , based on a weak - coupling expansion around the non - interacting limit would be inadequate .
thus a treatment within a non - perturbative scheme clearly is desirable . in this paper
we therefore use the recently developed dynamical cluster approximation ( dca ) @xcite to study the low - energy behavior of the @xmath1 hubbard model in the intermediate coupling regime with nearest - neighbor hopping @xmath8 and on - site correlation @xmath10 equal to the band width @xmath11 .
the dca systematically incorporates non - local corrections to local approximations like the dynamical mean field , by mapping the lattice onto a self - consistently embedded cluster .
we solve the cluster problem using a combination of quantum monte carlo ( qmc ) and the maximum entropy method to obtain dynamics .
this technique produces results in the thermodynamic limit and has a mild minus - sign problem@xcite .
the paper is organized as follows .
the next section contains a brief introduction to the dca .
the numerical results will be presented in the third section followed by a discussion and summary .
a detailed discussion of the dca formalism was already given in previous publications @xcite .
the main assumption underlying the dca is that the single - particle self - energy @xmath12 is a slowly varying function of the momentum @xmath13 and can be approximated by a constant within each of a set of cells centered at a corresponding set of momenta @xmath14 in the first brillouin zone@xcite .
the single - particle lattice green functions are then coarse - grained or averaged within these cells , and used to calculate the lattice self energy and other irreducible quantities . within this approximation
, one can set up a self - consistency cycle similar to the one in the dynamical mean - field theory ( dmft)@xcite .
however , in contrast to the dmft , where only local correlations are taken into account , the dca includes non - local dynamical correlations .
the length scales of these non - local correlations can be varied systematically from short ranged to long ranged by increasing the number of coarse - graining cells .
the dca collapses to the dmft if one represents the brillouin zone by one cell only , thus setting the characteristic length scale to zero . by construction
, the dca preserves the translational and point group symmetry of the lattice .
comparisons to other extensions of the dmft developed during the past years@xcite either show that these are identical to the dca or converge more slowly as function of cluster size @xcite . for the impurity problem of the dmft a large set of reliable numerical techniques has been developed over the past ten years@xcite .
we have employed quantum monte carlo ( qmc ) and the non - crossing approximation ( nca ) @xcite as non - perturbative ( in @xmath10 ) methods to solve the cluster problem of the dca . for cluster sizes larger than @xmath15 however , only the qmc technique is presently available for the dca .
the cluster problem is solved using the hirsch - fye impurity algorithm @xcite modified to simulate an embedded cluster@xcite .
note that this problem is computationally much more difficult than that encountered in either the dmfa or in finite - sized simulations , since the block diagonal structure in space - time occurring in conventional finite - system simulations is not present here .
this increase in computation time is , however , partially compensated by a rather mild minus - sign problem , even for comparatively large values of @xmath10 and small temperatures@xcite . from the qmc data ,
the spectra are obtained by analytic continuation with the maximum entropy method@xcite .
finally , the self energy is interpolated on to the full brillouin zone using akima splines , which is a sensible step as long as the assumption of a slow variation in @xmath16-space is valid .
note that it is very important to interpolate irreducible quantities like the self energy and _ not _ for example the cluster green function itself .
for @xmath17ev , @xmath18ev , @xmath19 , @xmath20 and @xmath21 along certain high symmetry directions ( spectra for different @xmath22 are shifted along the y - axis ) .
the colored lines in figs .
( a ) and ( c ) indicate the spectra which cross the fermi energy with a peak closest to @xmath23 . in ( b ) , no such peak is found which crosses the fermi energy .
( d ) the imaginary part of the self energy versus frequency at the fermi surface crossing found in ( a ) and ( c ) . , width=288 ] for a proper description of the cuo@xmath24 planes of the high-@xmath3 cuprates within the hubbard model ( [ equ:1 ] )
it is generally accepted that the tight - binding dispersion has the form @xmath25{l}\displaystyle -2t\left(\cos(k_x)+\cos(k_y)\right)\\[5 mm ] \displaystyle -4t'\cos(k_x)\cos(k_y ) \end{array}\ ] ] with a nearest neighbor hopping amplitude @xmath26 and a next - nearest neighbor hopping amplitude @xmath27 , which in principle can have any sign . from bandstructure calculations and the general form of the measured fermi surface , especially in the overdoped regime ,
conventionally a negative @xmath27 is inferred@xcite .
such a negative @xmath27 would naturally lead to a fermi surface closed around the brillouin zone corner @xmath28 .
the interesting question , however , is whether the fermi surface closed around the @xmath28 point observed experimentally in the low - doping region is a mere bandstructure effect or induced by correlations .
weak coupling treatments of the @xmath1 hubbard model indicate that such a change of the shape , but not the volume , of the fermi surface due to the interactions is indeed possible@xcite .
thus , to obtain insight into the effects of correlations on the structure of the fermi surface and distinguish them from pure bandstructure effects , we concentrate on the case @xmath29 in this paper . in the following we set @xmath30ev in accordance with typical values extracted from the experiments and bandstructure and choose @xmath31ev .
this value of @xmath10 is sufficiently large that for @xmath32 a mott gap is present in the half - filled model @xcite .
we performed our simulations at a range of temperatures , but will present results for @xmath33ev only , which is roughly room temperature .
a pseudogap due to short - ranged spin correlations is also present in the weakly doped model for slightly lower temperatures @xcite .
@xmath34 along certain high symmetry directions .
other parameters as in fig .
[ akw_n0.95 ] .
the colored lines in figs .
( a ) and ( b ) indicate the spectra which cross the fermi energy with a peak closest to @xmath23 . in ( c ) , no such peak is found which crosses the fermi energy .
( d ) the imaginary part of the self energy versus frequency at the fermi surface crossing found in ( a ) and ( b ) . , width=288 ] the single - particle spectra for certain high symmetry directions are plotted in figs .
[ akw_n0.95 ] and [ akw_n0.80 ] for @xmath35 and @xmath36 , respectively .
we use the standard convention to identify the high symmetry points in the zone , @xmath5 , @xmath2 and @xmath6 . for @xmath35
the peak in the spectrum crosses the fermi energy along the @xmath37 and @xmath38 directions , while for @xmath36 the second crossing appears along @xmath39 .
the imaginary part of the self energy at these crossing points is plotted versus frequency in panel ( d ) of figs .
[ akw_n0.95 ] and [ akw_n0.80 ] .
one very interesting feature of the spectra at low doping , fig .
[ akw_n0.95 ] , is that the peak near ( @xmath40 ) broadens dramatically before crossing the fermi energy .
near @xmath0 , on the other hand , one does not observe any dramatic change in the spectrum when crossing the fermi energy .
this indicates that near ( @xmath41 hole - like quasi - particle ( qp ) excitations with @xmath42 appear to have longer lifetimes than electronic excitations with @xmath43 . this asymmetry between particles and holes near the fermi surface is a strong indication of non - fermi liquid ( nfl ) behavior , at least along the @xmath44-@xmath28 direction .
note that this is at least qualitatively in agreement with experimental arpes spectra@xcite . although these experiments on lsco had to be performed below @xmath3 in the underdoped sample@xcite ,
the authors infer from the general behavior of the spectra , that a peak crossing the fermi energy can be found along @xmath38 , while a rather broad structure is seen along @xmath37 around @xmath4
. it would clearly be interesting to have experimental arpes data above @xmath3 available for a more founded comparison .
it is also quite instructive to look at the imaginary part of the self - energy at the @xmath22-points where the peak in the spectrum crosses the fermi energy ( fig . [ akw_n0.95](d ) ) .
in particular at the crossing point @xmath45 close to @xmath46 , @xmath47 shows a striking asymmetry in the low - frequency regime as compared to @xmath48 , where @xmath49 is the crossing point close to @xmath46 .
in fact , @xmath47 starts to develop an additional feature which eventually leads to the formation of a pseudo gap in the spectra along the @xmath38 direction@xcite .
in addition , one observes a rather large residual scattering rate for both momenta @xmath50 and @xmath51 .
this can either be taken as further evidence for nfl behavior or as signal for the occurrence of a new very small low - energy scale@xcite .
obviously , we can not decide this question on the basis of the present data , but would have to look at much lower temperatures .
unfortunately , this is not possible at present .
note that in the low - energy regime the self energy displays significant @xmath52-dependence .
this clearly renders theories based on a local approximation like the dmfa inadequate at least for small doping . at higher doping ,
[ akw_n0.80 ] , the peaks in the spectrum close to the fermi energy are far sharper .
here it makes sense to speak of a conventional fermi liquid and quasi particles again .
as already mentioned , the fermi energy crossings can be found along @xmath37 and @xmath39 .
again , this is in qualitative accordance with arpes experiments for strongly overdoped lsco@xcite , although these experiments still find rather broad structures even in heavily overdoped samples .
in addition , there is no evidence for particle - hole asymmetry in our data . especially at @xmath4 the structure crossing the fermi level appears to be rather symmetric with respect to the crossing point .
units < 3in,3 in > x from 0 to 1 , y from 0 to 1 bottom invisible label @xmath53 ticks withvalues @xmath54 @xmath55 / at 0 1 / / left invisible label @xmath56 ticks withvalues @xmath55 / at 0 1 / / at 0.501 0.501 units < 3in,3 in > x from 0 to 1 , y from 0 to 1 bottom invisible label @xmath53 ticks withvalues @xmath54 @xmath55 / at 0 1 / / left invisible label @xmath56 ticks withvalues @xmath55 / at 0 1 / / at 0.501 0.501 the imaginary part of the self energy , shown in fig . [ akw_n0.80](d ) , has a broad maximum at @xmath23 with a very small residual scattering rate and changes little as @xmath16 moves along the fermi surface . this weak dependence on @xmath16
is an indication that approximations like the dmfa should be accurate here , i.e. that there is little effect of non - local correlations .
all indications are that for this doping regime standard fermi - liquid behavior has returned .
more evidence for nfl behavior can be seen in the shape of the fermi surface .
theoretically , the fermi surface can be defined in two different ways .
first , the gradient of the electronic distribution function , @xmath57 , has a maximum at the fermi surface . from a computational point of view
this quantity is very convenient , because it does not require the calculation of dynamical properties .
however , especially for a comparison with experimental fermi surface mappings based on arpes experiments , the approach via @xmath57 probably is not adequate , mainly due to the unknown influence of matrix elements in the experimental spectra@xcite .
an analysis of the fermi surface for the @xmath8-@xmath9 model based on a study of @xmath57 has been performed recently within a high - temperature expansion@xcite and been considered as clear evidence for a violation of luttinger s theorem and the possible formation of a non - fermi liquid at small doping . our results for that quantity
are collected in figs .
[ 095_lowt_gradnk ] and [ 080_lowt_gradnk ] , where @xmath57 for the upper right quadrant of the first brillouin zone is shown in a density plot for @xmath35 and @xmath58 , respectively .
regions of large values of @xmath57 are colored in blue and violet , regions of small values in red . for comparison
the fermi surface for the non - interacting system is included ( black line ) .
for small doping , fig .
[ 095_lowt_gradnk ] , @xmath57 gives a rather broad structure around @xmath4 rather following the noninteracting fermi surface .
it is especially hard to define a fermi surface at all or extract a reliable estimate of the fermi surface volume from these results .
note also the strong bifurcation around the @xmath0 points .
we find that especially for small doping @xmath57 generally predicts that the fermi surface bifurcates near @xmath0 , with parts of the fermi surface along both @xmath38 and @xmath59 .
this is in striking contrast to the fact that the spectra in fig .
[ akw_n0.95 ] do not show a crossing along the direction @xmath60 , only a peak coming close to the fermi energy .
apparently , @xmath57 is unable ( at least at these temperatures ) to distinguish between a peak in the spectrum crossing or just simply approaching the fermi energy . in accordance with the spectra in fig .
[ akw_n0.80 ] , the plot of @xmath61 for large doping in fig .
[ 080_lowt_gradnk ] shows a fairly well defined fermi surface that coincides with the fermi surface of the non - interacting system .
again , in contrast to the sharp peaks found in the spectra , @xmath61 shows a substantial broadening , which in this case can however be explained by the conventional temperature broadening of fermi s function .
an alternative way of mapping the fermi surface , which also allows more direct contact with arpes experiments , is to make constant energy plots of the single - particle spectra at the fermi energy @xmath62 .
units < 3in,3 in > x from 0 to 1 , y from 0 to 1 bottom invisible label @xmath53 ticks withvalues @xmath54 @xmath55 / at 0 1 / / left invisible label @xmath56 ticks withvalues @xmath55 / at 0 1 / / at 0.501 0.501 units < 3in,3 in > x from 0 to 1 , y from 0 to 1 bottom invisible label @xmath53 ticks withvalues @xmath54 @xmath55 / at 0 1 / / left invisible label @xmath56 ticks withvalues @xmath55 / at 0 1 / / at 0.501 0.501 units < 3in,3 in > x from 0 to 1 , y from 0 to 1 bottom invisible label @xmath53 ticks withvalues @xmath54 @xmath55 / at 0 1 / / left invisible label @xmath56 ticks withvalues @xmath55 / at 0 1 / / at 0.501 0.510 the constant energy plots @xmath62 are shown in figs .
[ 095_lowt_nc ] , [ 090_lowt_nc ] and [ 080_lowt_nc ] for @xmath35 , @xmath63 and @xmath36 , respectively , for the upper right quadrant of the first brillouin zone .
the regions of high density are colored in violet and low density in red .
the solid black lines as before represent the non - interacting fermi surface . for @xmath35 ,
the fermi surface resulting from our calculations is hole - like , centered around @xmath64 , and encloses a volume larger than the non - interacting fermi surface , indicating a violation of luttinger s theorem . in addition , the shape of the fermi surface close to @xmath4 , especially the clear shift above @xmath4 can not be interpreted neither in terms of a simple tight - binding band structure nor a weak - coupling theory@xcite .
note also that in comparison to the @xmath57 result , no apparent electron like shadow band at @xmath46 can be seen .
with increasing doping a bifurcation of the fermi surface around @xmath46 starts to develop ( see fig .
[ 090_lowt_nc ] ) , signalling the incipient crossover from a hole like fermi surface to the expected electron like at large doping .
in addition , around @xmath4 the peak in the spectrum follows more or less the noninteracting fermi surface again .
this points towards a restoration of luttinger s theorem . for @xmath36 ,
the fermi surface is definitely electron like , centered at @xmath65 , and has essentially the same volume and shape as the non - interacting surface , indicating a return to fermi - liquid like behavior .
no apparent remnants of the hole - like fermi surface at lower doping are left . interpreting these bifurcations
seen most strongly around @xmath63 as shadow fermi surfaces , we find a strong reduction of the weight in these shadow features both for dopings less and larger than this `` optimal '' doping .
interestingly , a similar behavior was recently observed in the analysis of arpes data for bi2212@xcite .
however , in these experiments the fermi surface remained hole - like throughout the whole doping regime studied . whether this might be related to a finite @xmath27 neglected in our present calculations
will be discussed elsewhere .
the increasing precision and quality of experimental arpes spectra in recent years has led to a number of new results on the single - particle dynamics of the high-@xmath3 cuprates , both partially resolving long - standing issues and posing new questions and problems .
motivated by especially the interesting observations concerning the changes of fermi surface topology with doping , we have studied the two - dimensional ( 2d ) hubbard model with nearest neighbor hopping @xmath8 in the intermediate coupling regime ( on - site correlation @xmath10 equal to the bandwidth ) at the temperature @xmath33ev . to study the model in this most problematic parameter regime we used quantum monte carlo within the dynamical cluster approximation for the cluster size @xmath21 . since this method allows for controlled and reliable calculations of low - energy features within a non - perturbative scheme and in the thermodynamic limit , fundamental problems in this field
can be addressed .
these include the single - particle spectral properties at low energies , especially possible deviations from luttinger s theorem or the formation of non fermi liquid states , and the resulting topology of the fermi surface > from the two different ways to define the fermi surface , i.e. via @xmath61 and inspection of a constant energy scan @xmath66 the latter turned out to be the more precise .
the constant energy scans are able to distinguish the situation where a peak in the spectral crosses the fermi surface from that where it only approaches it .
thus they were free of spurious bifurcations observed in the @xmath61 plots . from the constant energy scans of the spectrum at the fermi energy we find that the fermi surface changes its topology compared to the non - interacting one as a function of decreasing doping .
while the latter is closed around the zone center for every finite doping , the interacting fermi surface at low doping , @xmath19 , is hole - like , closed around @xmath2 .
moreover , the form and shifts present in the fermi surface must be taken as clear evidence for a violation of luttinger s theorem .
non fermi liquid behavior is also evidenced by the spectrum .
a strong particle - hole asymmetry is found at the fermi surface crossing near @xmath4 with hole - like excitations having much longer life - times than electronic excitations .
furthermore the corresponding self - energy shows strong @xmath67-dependence rendering local approximations like the dmfa irrelevant .
additional structures in the self energy and a rather large residual scattering rate can also be interpreted as signs for non fermi liquid behavior .
however , to really distinguish a non fermi liquid from a fermi liquid with a possibly extremely small energy scale much lower temperatures must be studied .
with increasing doping the fermi surface bifurcates at @xmath46 around a doping @xmath68 and from the topology around @xmath4 one can infer a tendency towards restoration of luttinger s theorem and conventional fermi liquid behavior .
finally when @xmath34 the fermi surface is electron - like again , i.e. closed around the zone center , and coincides with the non - interacting fermi surface .
the corresponding spectra show well defined and sharp quasi - particle peaks around the fermi energy with particle - hole symmetry being recovered in the low - energy region .
although the hubbard model surely presents an oversimplification of the real cuprates , the common belief is that at least the essential qualitative features of the low energy dynamics should be reproduced . in particular
, the results presented here show some interesting qualitative agreement with experiments regarding the behavior of the different structures observed in the spectra .
moreover , these features could be related to at least an apparent violation of luttinger s theorem or possibly even non fermi liquid behavior at low doping . from our results
it becomes very clear that neither weak - coupling treatments nor local theories like the dmft are able to capture the essentials of the physics of the cuprate in the weakly doped regime .
there are , of course , a variety of further questions to be addressed .
for example , most of the interesting and controversial arpes results are for the system bi2212 , which actually is a bilayer system .
moreover , the simple nearest - neighbor tight binding band structure used in this paper is definitely not sufficient to describe the cuprates .
thus , a detailed study about the influence of a finite and negative @xmath27 and also a bilayer coupling on the topology of the fermi surface or more generally on the low - energy single particle dynamics is definitely necessary .
we believe that such an investigation can also address some of the still mysterious features in behavior of the fermi surface topology of the high-@xmath3 cuprates .
moreover it is very important to find new methods to solve the dca self consistency cycle at lower temperatures or preferably at @xmath69 .
this would enable us to clearly distinguish between a strong coupling fermi liquid and genuine non fermi liquid behavior at low doping .
borisenko , a.a .
kordyuk , s. legner , c. drr , m. knupfer , m.s . golden , j. fink , k. nenkov , d. eckert , g. yang , s. abell , h. berger , l. forr , b. liang , a. maljuk , c.t .
lin and b. keimer , phys .
b*64 * , 94513(2001 ) .
a. kampf , j.r .
schrieffer , phys .
b*41 * , 6399(1990 ) ; ibid * 42 * , 7967(1990 ) ; p.
aebi , j. osterwalder , p. schwaller , l. schlapbach , m. shimoda , t. mochiku and k. kadowaki , phys .
* 72 * , 2757(1994 ) .
a. fujimori , a. ino , t. yoshida , t. mizokawa , z .- x .
shen , c. kim , t. kakeshita , h. eisaki , s. uchida in _
open problems in strongly correlated electron systems _ , j. bona , p. prelovek , a. ramak and s. sakar ( eds . ) , kluwer academic publ . ( 2001 ) , p. 119 .
t. yoshida , x.j .
zhou , m. nakamura , s.a .
kellar , p.v .
bogdanov , e.d .
lu , a. lanzara , z. hussain , a. ino , t. mizokawa , a. fujimori , h. eisaki , c. kim , z .- x .
shen , t. kakeshita and s. uchida , phys .
b63 , 220501(2001 ) .
hettler , a.n .
tahvildar - zadeh , m. jarrell , t. pruschke and h. r. krishnamurthy , phys .
b * 58 * , 7475 ( 1998 ) ; m.h .
hettler , m. mukherjee , m. jarrell and h. r. krishnamurthy , phys .
b * 61 * , 12739 ( 2000 ) . | we discuss spectra calculated for the 2d hubbard model in the intermediate coupling regime with the dynamical cluster approximation , which is a non - perturbative approach .
we find a crossover from a normal fermi liquid with a fermi surface closed around the brillouin zone center at large doping to a non - fermi liquid for small doping .
the crossover is signalled by a splitting of the fermi surface around the @xmath0 point of the @xmath1 brillouin zone , which eventually leads to a hole - like fermi surface closed around the point @xmath2 .
the topology of the fermi surface at low doping indicates a violation of luttinger s theorem .
we discuss different ways of presenting the spectral data to extract information about the fermi surface .
a comparison to recent experiments will be presented . |
the capacity analysis of _ noncoherent _ fading channels has received considerable attention in recent years since it provides the ultimate limit on the rate of reliable communication on such channels .
proposed approaches to modeling noncoherent fading channels can be classified into two broad categories . the first is to model the fading process as a _ block - independent _ process . in the standard version of this model@xcite ,
the channel remains constant over blocks consisting of @xmath0 symbol periods , and changes independently from block to block .
the second is to model the fading process as a symbol - by - symbol _ stationary _ process . in this model
, the independence assumption is removed , but the block structure is not allowed . somewhat surprisingly , these two models lead to very different capacity results . for the standard block fading model ,
the capacity is shown @xcite to grow logarithmically with snr , while for the symbol - by - symbol stationary model , the capacity grows only double - logarithmically in snr at high snr if the fading process is _ regular _ @xcite . for symbol - by - symbol stationary gaussian fading channels ,
if the lebesgue measure of the set of harmonics where the spectral density of the fading process is zero is positive , the fading process is _ nonregular _ and the capacity grows logarithmically with snr @xcite .
this result is consistent with the capacity result for block - independent fading channels in the sense that the @xmath1 behavior in the high snr regime results from the rank deficiency of the correlation matrix of the fading process .
this point was elucidated in @xcite where a _ time - selective _ block fading model was considered in which the rank of the correlation within the block is allowed to be any number between one and the blocklength . however , the mechanisms that cause the rank deficiency in the block - independent fading and nonregular symbol - by - symbol stationary models are different . for the block - independent fading model
, the rank deficiency happens within each block .
but for the nonregular symbol - by - symbol stationary fading channel model , the correlation matrix of the fading process over any finite block can still be full - rank ; the rank deficiency in this case is in the asymptotic sense . in general
, the rank deficiency of the correlation matrix can be affected by both the short timescale correlation of the fading process as in the block - independent fading model and large timescale correlation as in the symbol - by - symbol stationary channel model . in order to capture both of these effects
, we model the fading process as a _ block - stationary _ gaussian process .
the block - stationary model was introduced and justified in @xcite .
we summarize the main points of the justification here . in the block - independent fading model ,
the channel is assumed to change in an i.i.d .
manner from block to block . the independence can be justified in certain time division or frequency hopping systems where the blocks are separated sufficiently in time or frequency to undergo independent fading .
the independence assumption is also convenient for information - theoretic analysis as it allows us to focus on one block in studying the capacity .
if the blocks are not separated far enough in time or frequency , the fading process can be correlated across blocks and the block - stationary model is more appropriate in this scenario . without time or frequency
hopping , the channel variations from one block to the next are dictated by the long term variations in the scattering environment .
if we assume that the variations in average channel power are compensated for by other means such as power control , it is reasonable to model the variation from block to block as stationary and ergodic .
[ rem : blockvsstat ] the block - stationary model does not imply that the fading process is stationary on a symbol - by - symbol basis as in the analysis of @xcite .
but as explained in @xcite , the symbol - by - symbol stationary model is not realistic for time intervals that are larger than that corresponding to a few wavelengths . for this reason it may be more accurate to model the fading process using a block fading model with possible correlation across blocks than it is to model it as a symbol - by - symbol stationary process . from the viewpoint of analysis ,
the block - stationary model generalizes all previously considered models discussed above and therefore so do the capacity results for this model .
more importantly , the block - stationary model provides us with a framework to study the interplay between many aspects of fading channels which are not captured in the aforementioned models , and allows us to identify the properties that are shared by different models and the properties that depend on channel modelling .
the channel capacity for the block - stationary model was only studied in @xcite under certain constraints on the correlation structure across blocks , which essentially disallow rank deficiency over the large timescale . in this paper
, we conduct a more complete study of the capacity for this channel model .
the following notation is used in paper . for deterministic objects ,
uppercase letters denote matrices , lowercase letters denote scalars , and underlined lowercase letters denote vectors .
random objects are identified by corresponding boldfaced letters .
for example , @xmath2 denotes a random matrix , @xmath3 denotes the realization of @xmath2 , @xmath4 denotes a random vector , and @xmath5 denotes a random scalar . for simplicity ,
sometimes we also use @xmath6 to denote the random vector @xmath7 .
although uppercase letters are typically used for matrices , there are some exceptions , and these exceptions are noted explicitly in the paper .
the operators @xmath8 , @xmath9 , @xmath10 , @xmath11 , and @xmath12 denote determinant , trace , conjugate , transpose and conjugate transpose , respectively .
we let @xmath13 denote the @xmath14 identity matrix for any positive integer @xmath15 , and let @xmath16 denote @xmath17 $ ] for random vectors @xmath18 and @xmath19 .
we consider a discrete - time channel whose time-@xmath20 complex - valued output @xmath21 is given by @xmath22 where @xmath23 is the input at time @xmath20 with peak power constraint @xmath24 ; @xmath25 models the fading process ; and @xmath26 models additive noise .
the processes @xmath25 and @xmath26 are assumed to be independent and have a joint distribution that does not depend on the input @xmath27 .
we assume that @xmath26 is a sequence of i.i.d .
circularly symmetric complex - gaussian random variables of zero mean and unit variance , i.e. , @xmath28 .
we assume that the fading process @xmath25 is a block - stationary process with @xmath29 and block length @xmath0 , i.e. , @xmath30 is a vector - valued stationary process . furthermore , we assume that @xmath31 is an ergodic process with a matrix spectral density function @xmath32 , @xmath33 .
specifically , @xmath34 where @xmath35 . since @xmath36 , @xmath37 , it is not hard to check that @xmath32 is hermitian , i.e. , @xmath38 .
moreover , we have @xmath39 ( @xmath33 ) ,
i.e. , @xmath32 is a positive semi - definite matrix .
there is an interesting relation between the matrix spectral density function and the asymptotic prediction error .
specifically , for the block stationary process @xmath40 , define the following prediction error covariance matrices : @xmath41 then @xmath42 and @xmath43 are related to the matrix spectral density function @xmath32 of @xmath25 as @xcite @xmath44&=&\exp\left\{\frac{1}{2\pi}\int_{-\pi}^{\pi}\log\det\left[s(e^{j\omega})+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}i_t\right]\mbox{d}\omega\right\},\label{identity2}\\ \det\left[\sigma(\infty)\right]&=&\exp\left\{\frac{1}{2\pi}\int_{-\pi}^{\pi}\log\det\left[s(e^{j\omega})\right]\mbox{d}\omega\right\}.\label{identity1}\end{aligned}\ ] ] the remainder of this paper is organized as follows . in section [ sec : asym_cap ] , we establish single - letter upper and lower bounds on channel capacity , and use these bounds to analyze the asymptotic capacity in the high snr regime . in section [ sec : discussion ] , we discuss the robustness of the asymptotic capacity results , and the interplay between the codeword length , communication rate and decoding error probability . in section [ sec : cap_energy ] , we adapt the formula of verd for capacity per unit cost @xcite to our channel model , and use it to derive an expression for the capacity per unit energy in the presence of a peak power constraint .
we summarize our results in section [ sec : concl ] .
we denote the capacity with peak power constraint @xmath45 by @xmath46 . for any @xmath47 and @xmath48 , let @xmath49 let @xmath50 be the set of probability distributions on @xmath51 .
since the channel is block - wise stationary and ergodic , a coding theorem exists and we have @xmath52 to derive a lower bound on @xmath46 for the channel model given in ( [ channelmodel ] ) , we adopt the interleaved decision - oriented training scheme proposed in @xcite with some modifications .
this scheme can also be viewed as a way of interpreting the computations in ( * ? ? ?
iv.e ) , ( * ? ? ?
v ) . let @xmath53 be a circularly symmetric distribution with @xmath54 $ ] .
construct the codebook @xmath55 with @xmath56 sub - codebooks @xmath57 , where codebook @xmath58 ( @xmath59 ) contains @xmath60 codewords of length @xmath61 generated independently symbol by symbol using distribution @xmath53 .
we assume that @xmath56 is a multiple of the block length @xmath0 , i.e. , @xmath62 for some positive integer @xmath63 .
now we multiplex ( or interleave ) these @xmath56 codebooks .
specifically , codebook @xmath58 ( @xmath59 ) is used at time instants @xmath64 . for codebook @xmath58
, its codeword can be successfully decoded if @xmath65 for sufficiently large @xmath61 . furthermore , using the facts that @xmath66 are i.i.d . and that the channel is stationary over time instants @xmath67 ,
we get @xmath68 this is to be expected since a channel with memory has a higher reliable communication rate than the memoryless channel with the same marginal transition probability .
thus , reliable communication at rate @xmath69 is possible for sub - codebook @xmath70 .
after @xmath66 is successfully decoded , the receiver can use these values as well as @xmath71 to estimate @xmath72 .
specifically , @xmath73 is used to estimate @xmath74 , @xmath75 . since @xmath54 $ ] , it is easy to verify the following markov chain condition @xmath76 to facilitate the calculation
, we assume that @xmath77 is used to estimate @xmath74 by forming the mmse estimate @xmath78 , @xmath75 .
the receiver decodes the codeword in codebook @xmath79 using @xmath80 as side information .
successful decoding is possible if @xmath81 similar to ( [ memoryhelp ] ) , we can use the lower bound @xmath82 to show that reliable communication at rate @xmath83 is possible for sub - codebook @xmath79 . by applying this procedure successively , we can conclude that for codebook @xmath58 , reliable communication is possible at rate @xmath84 thus , using this interleaved decision - oriented training scheme , we can have reliable communication at overall rate of @xmath85 we show in appendix [ app : monotonicity ] that @xmath86 is a monotone increasing sequence with @xmath87 now we let @xmath56 go to infinity ( i.e. , we let @xmath88 since @xmath0 is fixed ) , and we obtain @xmath89 \\ & = & \frac{1}{t}\sum\limits_{i=1}^ti\left({\text{\boldmath{$x$}}}_{i};{\text{\boldmath{$y$}}}_{i}\left|\mathbb{e}\left({\text{\boldmath{$h$}}}_{i}\left|\left\{{\text{\boldmath{$h$}}}_k+\frac{1}{x_{\min}}{\text{\boldmath{$z$}}}_k\right\}_{k=-\infty}^{i-1}\right.\right)\right.\right)\\ & = & \frac{1}{t}\sum\limits_{i=1}^ti\left({\text{\boldmath{$x$}}}_{i};{\text{\boldmath{$h$}}}_{i}{\text{\boldmath{$x$}}}_i+{\text{\boldmath{$z$}}}_i\left|\mathbb{e}\left({\text{\boldmath{$h$}}}_{i}\left|\left\{{\text{\boldmath{$h$}}}_k+\frac{1}{x_{\min}}{\text{\boldmath{$z$}}}_k\right\}_{k=-\infty}^{i-1}\right.\right)\right.\right).\end{aligned}\ ] ] this yields the single - letter lower bound @xmath90 where @xmath91 all have the same distribution @xmath53 , which is to be optimized later .
although channel estimation and communication are intertwined in this interleaved decision - oriented training scheme , the effect of channel memory is isolated from channel coding through interleaving .
this is because when @xmath56 is large enough , @xmath92 are roughly independent .
thus the codeword in codebook @xmath58 , which is transmitted over time instants @xmath93 , essentially experiences a memoryless channel .
this also suggests that as @xmath56 goes to infinity , the single - letter lower bound ( [ lowerbound ] ) provides a correct estimate of the rate supported by this interleaved decision - oriented training scheme .
we can see that the channel memory manifests itself in the lower bound ( [ lowerbound ] ) only through @xmath94 .
furthermore , in ( [ lowerbound ] ) , we can write @xmath95 as the sum of two independent random variables : the coherent fading component @xmath94 which is known the the receiver , and the non - coherent fading component @xmath96 which is unknown . isolating the effect of channel memory facilitates the channel code design : we only need to design channel codes for memoryless fading channels with different coherent and non - coherent components , instead of designing different codes for channels with different memory structures . to derive a single - letter upper bound on @xmath46
, we follow the approach in @xcite .
the capacity @xmath46 is given by @xmath52 by the chain rule , @xmath97 we can upper - bound @xmath98 as @xmath99 since @xmath100 is a sufficient statistic for estimating @xmath101 from @xmath102 , it follows that @xmath103 note that by the block stationarity of the fading process , @xmath104 depends on @xmath105 only through ( @xmath105 mod @xmath0 ) .
therefore , we obtain the single - letter upper bound @xmath106 now we proceed to show that the lower bound ( [ lowerbound ] ) and upper bound ( [ upperbound ] ) together characterize the asymptotic behavior of @xmath46 in the high snr regime .
[ lemma : rank ] for every @xmath107 $ ] , let @xmath108 be an @xmath14 symmetric positive semidefinite matrix .
we have @xmath109\mbox{d}\xi}{\log\epsilon}=\sum\limits_{i=0}^{m } ( m - i)\mu\left(\mbox{rank}\left(a(\xi)\right)=i\right).\end{aligned}\ ] ] where @xmath110 is the lebesgue measure of the set @xmath111 .
for the special case where @xmath112 for all @xmath113 $ ] and @xmath114 , we get @xmath115}{\log\epsilon}=m-\mbox{rank}(a).\end{aligned}\ ] ] see appendix [ app : rank ] .
[ lemma : computation ] if @xmath5 is uniformly distributed over the set @xmath116 , @xmath117 , @xmath118 , @xmath119 , and @xmath120 are all independent , then @xmath121 where @xmath122 is the euler constant .
[ th : asym_cap ] for the block - stationary gaussian fading channel model given in ( [ channelmodel ] ) , @xmath123}{t\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}=\frac{1}{2\pi t}\sum\limits_{i=0}^{t } ( t - i)\mu\left(\mbox{rank}(s(e^{j\omega}))=i\right ) .
\label{prelog}\end{aligned}\ ] ] the second equality in ( [ prelog ] ) follows from ( [ identity2 ] ) and lemma [ lemma : rank ] .
below we provide an intuitive explanation of this theorem based on the lower bound ( [ lowerbound ] ) .
the details of the proof are left to appendix [ app : asym_cap ] . in the lower bound ( [ lowerbound ] ) ,
let @xmath124 be uniformly distributed over the set @xmath116 , and write @xmath95 as @xmath125 where @xmath126 .
suppose @xmath127 , @xmath128 .
we can then write @xmath129 as @xmath130 where @xmath131 with @xmath132 . by viewing @xmath129 as the output of a coherent fading channel with the fading @xmath133 known at the receiver and noise @xmath134 , we get @xmath135 thus the lower bound ( [ lowerbound ] ) can be approximated by @xmath136 we then complete the proof by showing that @xmath137 is related to the matrix spectral density function @xmath32 by the equation @xmath138 theorem [ th : asym_cap ] generalizes many previous results on the noncoherent capacity for gaussian channels in the high snr regime as we illustrate in the following subsection .
_ constant fading within block _ for the special case where the fading remains constant within a block , i.e. , @xmath139 , for all @xmath140 , all the entries of @xmath141 for any fixed @xmath142 are identical .
this implies that , for any fixed @xmath143 , all the entries of @xmath32 are identical , which we shall denote by @xmath144 .
it is easy to see that @xmath144 is essentially the spectral density function of @xmath145 .
the rank of @xmath32 is 1 if @xmath146 , and is 0 if @xmath147 .
we therefore have @xmath148 when @xmath149 , we recover the result in @xcite that @xmath150 which illustrates the effect of large timescale correlation of the fading process on the pre - log term of the channel capacity in the high snr regime . when the fading is independent from block to block , we have @xmath151 , and thus recover the result in @xcite that @xmath152 which illustrates the effect of short timescale correlation of the fading process on the pre - log term of the capacity at high snr . _
time - selectivity within block _ in this example , we recover the main result in @xcite concerning the case where rank deficiency is caused purely by the correlation within a block . if @xmath153 , then is satisfied , for instance , when the fading process is independent from block to block . ]
@xmath154 to prove ( [ liangprelog ] ) , we first note that @xmath155 we therefore have the bound @xmath156}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\leq\lim\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{\log\det\sigma({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\leq\lim\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{\log\det\left[r(0)+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}i_t\right]}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}.\end{aligned}\ ] ] by lemma [ lemma : rank ] , @xmath156}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}&=&\lim\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{\log\det\left[r(0)+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}i_t\right]}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\\ & = & -t+\mbox{rank}(r(0)\end{aligned}\ ] ] which implies that @xmath157 therefore , by theorem [ th : asym_cap ] , @xmath158 it is worth noting that in this case the pre - log term of the capacity can be achieved by a scheme simpler than the aforementioned interleaved decision - oriented training scheme .
suppose the rank of @xmath159 is @xmath160 , so that @xmath159 has @xmath161 positive definite principal submatrix .
without loss of generality , suppose this submatrix is the covariance matrix of @xmath162 .
then @xmath163 can be represented as a linear combination of @xmath164 for any @xmath140 and @xmath165 .
the simpler scheme is described as follows : the transmitter sends deterministic training symbols with maximum power at time instants @xmath166 , i.e. , @xmath167 , where @xmath168 .
the receiver can form the mmse estimates @xmath169 , for @xmath170 and @xmath168 .
clearly , we have @xmath171 with the side information @xmath172 at the receiver , we can communicate reliably at time instants @xmath173 with rate at least @xmath174 .
let @xmath124 be uniformly distributed over the set @xmath175 . by lemma [ lemma : computation ] , @xmath176\\ & & + \log\left(1-\mbox{var}\left({\text{\boldmath{$h$}}}_{i}\left|\left\{{\text{\boldmath{$h$}}}_{j}+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_{j}\right\}_{j=1}^q\right.\right)\right)-\gamma-\log\frac{5e}{6}\\ & = & \log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+o(\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}).\end{aligned}\ ] ] threfore , the overall rate is lower - bounded by @xmath177 and the pre - log term is achieved .
this scheme has the following obvious advantages over the interleaved decision - oriented training scheme : ( i ) channel estimation and communication are completely decoupled ; and ( ii ) channel estimation is done locally since the estimate @xmath169 , @xmath178 only depends @xmath179 .
the following theorem generalizes ( * ? ? ?
* corollary 4.42 ) for regular gaussian fading processes to the block - stationary case .
[ th : asym_cap2 ] if @xmath180 , then @xmath181=-1-\gamma-\frac{1}{t}\log\det(\sigma(\infty))=-1-\gamma-\frac{1}{2\pi t}\int_{-\pi}^{\pi}\log\det\left[s(e^{j\omega})\right]\mbox{d}\omega . \label{fadingnum}\end{aligned}\ ] ] the second equality in ( [ fadingnum ] ) follows from ( [ identity1 ] ) .
see appendix [ app : asym_cap2 ] .
_ gauss - markov process _
suppose @xmath182 is a gauss - markov process with @xmath183 if @xmath184 , and @xmath185 otherwise . here
@xmath186 are complex numbers with @xmath187 . in this case
, we have @xmath188 therefore , by theorem [ th : asym_cap2 ] @xmath181=-1-\gamma-\log(1-|\rho_2|^2)-\frac{\log(1-|\rho_1|^2)-\log(1-|\rho_1|^2)}{t}.\end{aligned}\ ] ]
for simplicity , we assume in this section that the fading process is symbol - by - symbol stationary , i.e. , @xmath149 . in this case , theorem 1 is specialized to equation ( [ symbolprelog ] ) .
we can see that two fading processes with spectral density functions @xmath189 and @xmath190 can induce the same pre - log term in the high snr regime as long as @xmath191 .
but in the non - asymptotic regime , the capacities of these two channels may behave very differently .
so , for a fixed @xmath192 , it is natural to ask the question : which spectral density function @xmath144 gives the largest ( or smallest ) channel capacity at a given @xmath45 ?
this question is difficult to answer since we do not have a closed - form expression for noncoherent channel capacity .
we therefore turn to the lower bound ( [ lowerbound ] ) to formulate a closely - related problem .
when @xmath149 , the lower bound ( [ lowerbound ] ) can be reduced to @xmath193 we can see that the lower bound ( [ lowerboundsym ] ) depends on @xmath144 only through @xmath194 .
furthermore , if we fix the input distribution @xmath195 , then @xmath196 implies that @xmath197 we can therefore ask which @xmath144 gives the largest ( or smallest ) @xmath198 . more precisely , since @xmath199\mbox{d}\omega\right\}-\frac{1}{x^2_{\min}},\end{aligned}\ ] ] we can formulate the problem in the following form : @xmath200\mbox{d}\omega \label{optimization}\end{aligned}\ ] ] subject to @xmath201 where @xmath202 . due to the strict concavity of @xmath203 , it is easy to show that the maximizers of the optimization problem ( [ optimization ] ) are the set of spectral density functions @xmath144 with the property : @xmath204 this solution has the following interpretation . without constraints on the spectral density function ,
the worst fading process is the i.i.d .
gaussian process whose spectral density function is flat . with the constraint @xmath205 ,
the spectral density function @xmath144 can not be completely flat , but the worst fading process should have a spectral density function that is as flat as possible , i.e. , the correlation in the time domain should be the weakest possible .
note that the solution does not depend on @xmath206 .
we can use this fact to derive a universal lower bound on @xmath46 for the class of spectral density functions @xmath207 , which has further implications for the high snr asymptotic behavior of @xmath46 .
let @xmath208 be a maximizer of ( [ optimization ] ) .
we have @xmath209 for any spectral density function @xmath144 with @xmath205 , we have @xmath210 where @xmath5 is uniformly distributed over the set @xmath211 , @xmath212 , @xmath213 , @xmath119 , and @xmath120 are all independent
. we can further optimize over @xmath53 to tighten the lower bound ( [ universallower ] ) .
the minimizers of ( [ optimization ] ) do not exist .
consider the following set spectral density functions @xmath214 given by @xmath215\\ \frac{2\pi\theta^2 - 2\pi\theta+\alpha\theta+1}{\theta } & |\omega|\in(\pi-\frac{1}{2\theta},\pi ] \end{array}\right.&\\\end{aligned}\ ] ] where @xmath216 with @xmath217 we can compute @xmath218\mbox{d}\omega\\ & = & \lim\limits_{\theta\rightarrow\infty}\left[-2\alpha\log x_{\min}+\left(2\pi-\alpha-\frac{1}{\theta}\right)\log\left(\frac{1}{\theta}+\frac{1}{x^2_{\min}}\right)+\frac{1}{\theta}\log\left(\frac{2\pi\theta^2 - 2\pi\theta+\alpha\theta+1}{\theta}+x^2_{\min}\right)\right]\\ & = & -4\pi\log x_{\min}.\end{aligned}\ ] ] note that @xmath219\mbox{d}\omega<-4\pi\log x_{\min}$ ] .
therefore , as @xmath220 goes to infinity , @xmath219\mbox{d}\omega$ ] approaches the lower bound that is not attainable by any spectral density function .
intuitively , the fading process associated with @xmath221 becomes more and more deterministic as @xmath220 gets larger , and it can be verified that @xmath222 this result has interesting implications for the channel capacity .
[ pro : nonuniform ] for any @xmath223 , @xmath224 if @xmath225 , then @xmath226 see appendix [ app : nonuniform ] although by theorem [ th : asym_cap ] , for any fixed @xmath220 , the ratio between @xmath227 and @xmath1 converges to @xmath228 , proposition [ pro : nonuniform ] says that the convergence is not uniform with respect to @xmath220 .
this is intuitively clear because when @xmath220 is large , we have @xmath229 for @xmath230 $ ] .
therefore it can be expected that for a large range of snr , the channel capacity @xmath231 behaves like @xmath232 , which could be significantly larger than @xmath233 . for the extreme case where @xmath234 , by theorem [ th : asym_cap2 ] , for any fixed @xmath220 , the capacity @xmath235 grows like @xmath236 at high snr . but
proposition [ pro : nonuniform ] implies that even in this extreme case , the capacity @xmath227 of some @xmath220 can grow linearly with @xmath1 for a large range of snr .
this is consistent with the result in @xcite where is was shown that for the gauss - markov process with @xmath237 , the capacity @xmath46 grows like @xmath1 for a wide range of snr levels if @xmath238 is close 1 .
an intuitive explanation is that if @xmath238 is close 1 , the spectrum @xmath239 is approximately zero for all @xmath143 except those around zero , and we can expect from equation ( [ symbolprelog ] ) that @xmath46 should grow like @xmath1 for a wide range of snr .
but it should be noted that as opposed to a gauss - markov process , a general gaussian stationary process can not be characterized by a single parameter , and the behavior of @xmath46 can be much more complicated as shown in the following example .
[ ex : twolevel ] consider the spectral density function @xmath240\\ \frac{1-\alpha_1\epsilon_1-(\alpha_2-\alpha_1)\epsilon_2}{1-\alpha_2 } & |\omega|\in(\pi\alpha_2,\pi ] \end{array}\right.&\\\end{aligned}\ ] ] where @xmath241 , and @xmath242 ( note : for two positive numbers @xmath243 and @xmath244 , @xmath245 means @xmath246 is much greater than 1 ) .
we show in appendix [ app : twolevel ] that @xmath247 is approximately equal to @xmath248 for @xmath249 , and gradually decreases to @xmath250 as @xmath45 approaches @xmath251 .
this example shows that @xmath247 can be highly snr dependent . for a regular gaussian fading process
, it may require unreasonably high ( impractical ) values of snr in order for the noncoherent channel capacity @xmath46 to grow like @xmath236 , and the behavior of @xmath46 at moderate snr levels may depend highly on the spectral density function .
overall , the above analysis suggests that great caution should be exercised when using the asymptotic results in theorem [ th : asym_cap ] and theorem [ th : asym_cap2 ] to approximate the channel capacity @xmath46 at a finite snr level .
although the asymptotic capacity results might yield over - pessimistic approximations such as a @xmath252 growth with @xmath45 , they could also lead to over - optimistic conclusions . in the capacity analysis
, it is assumed that the codeword is of infinite length .
but when the length of codewords is finite , the situation can be dramatically different . by fano s inequality ,
the communication rate @xmath253 is upper - bounded by @xmath254 where @xmath61 is the codeword length and @xmath255 is the decoding error probability .
suppose we fix @xmath61 and @xmath255 , and let @xmath45 go to infinity .
for a symbol - by - symbol stationary gaussian fading process , even if @xmath256 , the correlation matrix of the fading process over any finite block length can still be full - rank .
note that @xmath257 is upper - bounded by the capacity of a block - independent gaussian fading channel with the correlation matrix of each block given by @xmath258 $ ] .
since @xmath258 $ ] is full rank , it follows from theorem [ th : asym_cap2 ] that @xmath257 , and hence @xmath253 , grows at most like @xmath236 as @xmath45 goes to infinity .
therefore , there is no nontrivial tradeoff between diversity and multiplexing in the sense of @xcite .
if we want @xmath253 to grow linearly with @xmath1 while having the decoding error probability @xmath255 bounded away from 1 , the codeword length @xmath61 must scale with @xmath45 .
it is of interest to determine how fast the codeword @xmath61 should scale with @xmath45 in order to guarantee that the rate @xmath253 can grow as @xmath1 with the decoding error probability not approaching 1 . more precisely ,
letting the rate @xmath259 , codeword length @xmath260 and decoding error probability @xmath261 all depend on @xmath45 , we wish to determine conditions on @xmath260 to guarantee the existence of a sequence of codebooks ( indexed by @xmath45 ) with rate @xmath259 and codeword length @xmath260 such that @xmath262 and @xmath263 where @xmath264 and @xmath265 .
now we proceed to derive a necessary condition on the growth rate of @xmath260 .
it follows by chain rule that @xmath266 by ( [ lateruse ] ) , we can upper - bound @xmath267 as @xmath268 since @xmath269 , and @xmath270\right\}\\ & \leq&\log\left[\frac{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\cdot\mbox{var}\left({\text{\boldmath{$h$}}}_k\left|\left\{{\text{\boldmath{$h$}}}_v+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_v\right\}_{v=1}^{k-1}\right.\right)}\right]\\&=&\log\left(\frac{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)-\log\mbox{var}\left({\text{\boldmath{$h$}}}_k+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_k\left|\left\{{\text{\boldmath{$h$}}}_v+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_v\right\}_{v=1}^{k-1}\right.\right)\\ & \leq&\log\left(\frac{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)-\log\mbox{var}\left({\text{\boldmath{$h$}}}_0+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_0\left|\left\{{\text{\boldmath{$h$}}}_v+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_v\right\}_{v =- n({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}^{-1}\right.\right),\end{aligned}\ ] ] it follows by fano s inequality and the condition @xmath271 that @xmath272 therefore , in order for @xmath273 we must have @xmath274 since @xmath275 is a monotone increasing function of @xmath61 , it is easy to see ( [ scaling ] ) implicitly provides us with a lower bound on the scaling rate of @xmath260 . in order to derive an explicit lower bound on the scaling rate of @xmath260 , we need to introduce a concept called _ transfinite diameter _ @xcite .
let @xmath276 be a compact set in the plane .
set @xmath277 and @xmath278^{\frac{2}{n(n-1)}}.\end{aligned}\ ] ] the transfinite diameter of @xmath276 is defined by @xmath279 we need the following facts regarding the transfinite diameter . 1 . for two compact sets @xmath280 and @xmath281 with @xmath282 , we have @xmath283 .
the diameter of the unit circle is 1 .
more generally , the diameter of an arc of central angle @xmath220 on the unit circle is @xmath284 ; 3 .
the transfinite diameter of any closed proper subset of the unit circle is less than 1 .
a full discussion of the transfinite diameter can be found in @xcite .
now return to the original problem .
since @xmath285 we have @xmath286 let @xmath287 .
it was shown in @xcite that if the set @xmath276 consists consist of a finite number of arcs of the unit circle , then @xmath288 under the conditions 1 .
the set @xmath276 consists consist of a finite number of arcs of the unit circle , 2 .
the set @xmath276 is a closed proper subset of the unit circle , it can be shown by using facts ( i ) , ( ii ) and ( iii ) that @xmath289 therefore , under conditions ( a ) and ( b ) , we have @xmath290 it is clear that in order to guarantee that ( [ scalelower ] ) is greater than or equal to @xmath291 , we must have @xmath292 which is a necessary condition on the scaling rate of @xmath260 .
in contrast , we show in appendix [ app : awgn ] and appendix [ app : rayleigh ] that for the awgn channel and memoryless coherent rayleigh fading channel , it is possible to have the rate @xmath259 grow linearly with @xmath1 with fixed codeword length @xmath61 and bounded decoding error probability at high snr .
for these two cases , to facilitate the calculation , we adopt the average power constraint .
but our main conclusion holds also under the peak power constraint .
in the preceding sections , we focused on the channel capacity in the high snr regime .
now we proceed to characterize the behavior of channel capacity in the low average power regime for the block - stationary gaussian fading channel model . to this end , we shall study the capacity per unit energy ( which is denoted by @xmath293 ) due to its intrinsic connection with the channel capacity in this regime .
the following theorem provides a general expression for the capacity per unit energy .
@xmath294 furthermore , the capacity per unit energy is related to the capacity by @xmath295 where @xmath296 is the channel capacity with average power constraint @xmath297 and peak power constraint @xmath45 .
the following theorem is an extension of ( * ? ? ?
* proposition 3.1 ) for the symbol - symbol stationary channel model to the block - stationary model .
for the block - stationary gaussian fading channel model given in ( [ channelmodel ] ) , @xmath298 where @xmath299\mbox{d}\omega,\end{aligned}\ ] ] and @xmath300 is an @xmath301 principal minor of @xmath32 with the indices of columns and rows specified by @xmath302 . the proof is omitted since it is almost identical to that for the symbol - by - symbol stationary fading channel @xcite .
the only difference is that although the capacity per unit energy of the block - stationary fading channel can be asymptotically achieved by temporal on - off signaling , we have to determine how to allocate on symbols in a block .
it can be shown that the optimal allocation scheme is given by @xmath303 , which is the minimizer of @xmath304 . here
@xmath303 might not be unique .
@xmath293 is a monotonically increasing function of @xmath45 .
it is easy to see that @xmath293 goes to 1 as @xmath305 , and goes to 0 as @xmath306 .
the following result provides a more precise characterization of the convergence behavior .
[ cor : snrasym ] at high @xmath45 , @xmath307 at low @xmath45 , if @xmath308\mbox{d}\omega<\infty,\end{aligned}\ ] ] then @xmath309\mbox{d}\omega+o({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}).\label{lowsnrasym}\end{aligned}\ ] ] by lemma [ lemma : rank ] , at high @xmath45 @xmath310\mbox{d}\omega=-\sum\limits_{i=0}^{|\mathcal{m}|}(|\mathcal{m}|-i)\mu(\mbox{rank}(s_{\mathcal{m}}(e^{j\omega}))=i)\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+o(\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}).\end{aligned}\ ] ] therefore , @xmath311\mbox{d}\omega+2\pi\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right\}\\ & = & 1+\max\limits_{\mathcal{m}\subseteq\{1,\cdots , t\}}\frac{\sum\limits_{i=0}^{|\mathcal{m}|}(|\mathcal{m}|-i)\mu\left(\mbox{rank}(s_{\mathcal{m}}(e^{j\omega}))=i\right)\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{2\pi|\mathcal{m}|{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}-\frac{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+o(\frac{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}})\\ & = & 1-\min\limits_{\mathcal{m}\subseteq\{1,\cdots , t\}}\frac{\sum\limits_{i=0}^{|\mathcal{m}|}i\mu\left(\mbox{rank}(s_{\mathcal{m}}(e^{j\omega}))=i\right)\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{2\pi|\mathcal{m}|{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+o(\frac{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}).\end{aligned}\ ] ] at low @xmath45 , using the second - order approximation @xcite , we obtain @xmath312=\mbox{tr}[s_{\mathcal{m}}(e^{j\omega})]{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\frac{1}{2}\mbox{tr}[s^2_{\mathcal{m}}(e^{j\omega})]{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^2+o({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^2).\end{aligned}\ ] ] therefore , @xmath313\mbox{d}\omega-\frac{1}{2}\int_{-\pi}^{\pi}\mbox{tr}[s^2_{\mathcal{m}}(e^{j\omega})]{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\mbox{d}\omega\right\}+o({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})\\ & = & \frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4\pi}\max\limits_{\mathcal{m}\subseteq\{1,\cdots , t\}}\frac{1}{|\mathcal{m}|}\int_{-\pi}^{\pi}\mbox{tr}[s^2_{\mathcal{m}}(e^{j\omega})]\mbox{d}\omega+o({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})\end{aligned}\ ] ] where the last equality follows from the fact that @xmath314\mbox{d}\omega=1.\end{aligned}\ ] ] using the inequality @xmath312\geq\mbox{tr}[s_{\mathcal{m}}(e^{j\omega})]{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\frac{1}{2}\mbox{tr}[s^2_{\mathcal{m}}(e^{j\omega})]{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^2,\end{aligned}\ ] ] we can upper bound @xmath293 by @xmath315\mbox{d}\omega.\end{aligned}\ ] ] it can be seen from corollary [ cor : snrasym ] that this upper bound is a good approximation of @xmath293 in the low @xmath45 regime .
now we proceed to compute @xmath293 in the following examples .
[ ex : blockindep ] when the channel changes independently from block to block , @xmath293 is equal to @xmath316\end{aligned}\ ] ] where @xmath317 is an @xmath301 principal minor of @xmath159 with the indices of columns and rows specified by @xmath302 . if we further let the fading remain constant within a block , then all the entries of @xmath159 are one .
it is not difficult to show that @xmath318=\frac{\log(1+|\mathcal{m}|{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}{|\mathcal{m}|{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\end{aligned}\ ] ] which is minimized when @xmath319 , i.e. , @xmath320 .
so we have @xmath321 as shown in @xcite .
consider the case in which the fading process satisfies the following conditions 1 .
all the off - diagonal entries of @xmath159 are equal to @xmath322 , where @xmath323 is a constant ; 2 .
all the entries of @xmath141 are equal to @xmath324 for any non - zero integer @xmath142 , where @xmath325 is a constant that depends only on @xmath142 .
we also know that the diagonal entries of @xmath159 are all one .
so for any fixed @xmath143 ( @xmath33 ) , all the diagonal entries of @xmath326 are identical , and all the off - diagonal entries of @xmath326 are identical .
it then follows from szasz s inequality @xcite that for any @xmath327 $ ] , @xmath328 is minimized when @xmath320 . in this case
we therefor have @xmath329 if the fading remains constant within a block , then for any fixed @xmath143 , all the entries of @xmath32 are identical , which we shall denote by @xmath144 . it can be shown that @xmath330=1+t{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}s(e^{j\omega}),\end{aligned}\ ] ] which yields @xmath331\mbox{d}\omega.\label{tsnr}\end{aligned}\ ] ] we can see from ( [ tsnr ] ) that @xmath293 is a monotonically increasing function of @xmath0 and @xmath45 .
intuitively , as @xmath0 gets larger and larger , the receiver can estimate the channel more and more accurately , and thus the capacity per unit energy of the non - coherent channel should converge to that of the coherent channel , which is equal to one ; as @xmath45 goes to infinity , @xmath293 should also converge to one since flash signaling can be used if there is no peak power constraint ( i.e. , @xmath332 ) @xcite .
moreover , ( [ tsnr ] ) provides a precise characterization of the interplay between the coherence time and signal peakiness , stating that the capacity per unit energy is unaffected as long as the product of @xmath0 and @xmath45 is fixed .
see @xcite for a related discussion . for the special case where the fading is a block gauss - markov process , i.e.
, all the entries of @xmath141 are equal to @xmath333 if @xmath334 and equal to @xmath335 if @xmath336 for some @xmath337 with @xmath338 , we have @xmath339 where @xmath340 and @xmath341 with @xmath342 .
the function @xmath343 is analytic and nonzero in a neighborhood of the unit disk .
thus , by jensen s formula @xmath344 from which we can recover ( * ? ? ?
* corollary 4.1 ) by setting @xmath149 . finding
the optimal @xmath303 is a difficult problem in general .
moreover , as shown in the following example , the optimal @xmath303 may depend on the snr level .
[ example ] let the fading process be independent from block to block with @xmath345 where @xmath346 $ ] .
it is shown in appendix [ app : snrdependent ] that 1 . when @xmath347 , the optimal @xmath303 is @xmath348 , and @xmath349 2 . when @xmath350 , @xmath351 and @xmath352 3 .
when @xmath353 , the optimal @xmath303 is @xmath354 , and @xmath355 it can be verified that this result is consistent with the asymptotic analysis in corollary [ cor : snrasym ] . since @xmath356 for any @xmath357 , it is easy to see that @xmath358 is minimized at @xmath359 if @xmath360 , and minimized at @xmath361 if @xmath353 . therefore , by ( [ highsnrasym ] ) , the optimal @xmath303 at high snr should be @xmath348 if @xmath360 , and should be @xmath354 if @xmath353 . since @xmath362 it follows that @xmath363 $ ] is maximized at @xmath359 if @xmath364 , and maximized at @xmath361 if @xmath365 .
therefore , by ( [ lowsnrasym ] ) , the optimal @xmath303 at low snr should be @xmath348 if @xmath364 , and should be @xmath354 if @xmath365 . intuitively , if @xmath238 is close to 1 , we can approximate @xmath159 by the all - one matrix , and then it follows from example [ ex : blockindep ] that the optimal @xmath303 is @xmath354 .
the approximation breaks down at high snr since corollary [ cor : snrasym ] implies that the optimal @xmath303 should be @xmath348 as @xmath305 .
we conducted a detailed study of the block - stationary gaussian fading channel model introduced in @xcite .
we derived single - letter upper and lower bounds on channel capacity , and used these bounds to characterize the asymptotic behavior of channel capacity .
specifically , we computed the asymptotic ratio between the non - coherent channel capacity and the logarithm of the snr in the high snr regime .
this result generalizes many previous results on noncoherent capacity .
we showed that the behavior of channel capacity depends critically on channel modelling .
we also derived an expression for the capacity per unit energy for the block - stationary fading model .
it is clearly of interest to generalize these results to the multi - antenna scenario , but such an extension seems technically nontrivial . another direction that we explored was the interplay between the codeword length , snr level , and decoding error probability .
we showed that for noncoherent symbol - by - symbol stationary fading channels , the codeword length must scale with snr in order to guarantee that the communication rate can grow linearly with @xmath1 with bounded decoding error probability , and we found a necessary condition for the growth rate of the codeword length .
we believe that a more complete characterization of the interplay between the codeword length , snr level , and decoding error probability is of both theoretical significance and practical value .
by the block stationarity of the fading process , we have @xmath366 since for any @xmath367 , @xmath368 form a markov chain , it follows by data processing inequality that @xmath369 equations ( [ mono1 ] ) and ( [ mono2 ] ) together imply that @xmath86 is a monotone increasing sequence . for every @xmath370 , we can construct a random variable @xmath371 independent of everything else such that @xmath372 in distribution .
clearly , @xmath373 as @xmath374 .
moreover , it is not difficult to show that @xmath375 combining ( [ mono1 ] ) and ( [ addnoiseequiv ] ) , we get @xmath376 since @xmath377 by ( * ? ? ?
* lemma 6.11 ) , we get @xmath378 since conditioned on @xmath379 , @xmath380 are jointly gaussian with uniformly bounded differential entropy for any realization of @xmath379 ( note : @xmath381 ) , it follows by dominated convergence theorem that @xmath382 therefore , @xmath383
by eigenvalue decomposition , we write @xmath384 and @xmath385 where @xmath386 is a unitary matrix , and @xmath387 is a diagonal matrix with nonnegative diagonal entries . since @xmath388 , define @xmath389 we have @xmath109\mbox{d}\xi}{\log\epsilon}&=&\lim\limits_{\epsilon\rightarrow 0}\frac{\int_{\xi_1}^{\xi_2}\log\det \left[d(\xi)+\epsilon i_m\right]\mbox{d } \xi}{\log\epsilon}\\ & = & \lim\limits_{\epsilon\rightarrow 0}\frac{-\sum\limits_{i=0}^m\int_{\omega_i}\log\det \left[d(\xi)+\epsilon i_m\right]\mbox{d } \xi}{\log\epsilon}.\end{aligned}\ ] ] for @xmath390 , ( possibly after permutating diagonal entries ) we can write @xmath391 , where @xmath392 , @xmath393
. therefore , @xmath109\mbox{d}\xi}{\log\epsilon } = \lim\limits_{\epsilon\rightarrow 0}\frac{-\sum\limits_{i=0}^m\int_{\omega_i}\sum\limits_{j=1}^i\log[d_j(\xi)+\epsilon]\mbox{d } \xi}{\log\epsilon}+\sum\limits_{i=0}^m(m - i)\mu(\mbox{rank}(a(\xi))=i)\end{aligned}\ ] ] where @xmath110 is the lebesgue measure of @xmath394 . by the argument in (
* section viii ) , it can be shown that @xmath395\mbox{d}\xi}{\log\epsilon}=0.\end{aligned}\ ] ] so we have @xmath396\mbox{d}\xi}{\log\epsilon } = \sum\limits_{i=0}^m(m - i)\mu(\mbox{rank}(a(\xi))=i).\end{aligned}\ ] ]
define @xmath397 in the lower bound ( [ lowerbound ] ) , let @xmath124 be uniformly distributed over the set @xmath398 . by lemma [ lemma : computation ] , @xmath399+\log\left(1-\sigma_i\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)+\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)-\gamma-\log\frac{5e}{6}\\ & = & -\log\left[\sigma_i\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)-\frac{12}{5{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right]+o(\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}).\end{aligned}\ ] ] since @xmath400 , it follows that @xmath401}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}.\end{aligned}\ ] ] let @xmath402 , where @xmath403 is a lower triangular matrix with unit diagonal entries , and @xmath404 .
we have @xmath405&=&\det\left[l\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)\right]\det\left[\lambda\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)\right]\det\left[l^\dagger\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)\right]\nonumber\\ & = & \det\left[\lambda\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)\right]\nonumber\\ & = & \prod\limits_{i=1}^t \sigma_i\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right).\label{ldl}\end{aligned}\ ] ] therefore , @xmath406}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\nonumber\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{-\log\det\left[\sigma\left(\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{4}\right)\right]}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\nonumber\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{-\log\det\left[\sigma\left({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right)\right]}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}.\label{preloginf}\end{aligned}\ ] ] we use ( [ upperbound ] ) derive an upper bound on @xmath407 .
first it is easy to see that @xmath408 it is shown in @xcite that @xmath409 now we proceed to upper - bound the first term in ( [ twoparts ] ) .
@xmath410\right\}\\ & \leq&\log\left[\frac{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\cdot\mbox{var}\left({\text{\boldmath{$h$}}}_k\left|\left\{{\text{\boldmath{$h$}}}_t+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_t\right\}_{t=-\infty}^{k-1}\right.\right)}\right]\\&=&\log\left[\frac{1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\cdot\sigma_k({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}\right].\end{aligned}\ ] ] therefore , @xmath411}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\nonumber\\ & = & \limsup\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{-\log\det\left[\sigma\left({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right)\right]}{\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}.\label{prelogsup}\end{aligned}\ ] ] the desired result follows by combining ( [ preloginf ] ) and ( [ prelogsup ] ) .
define @xmath412 similar to ( [ ldl ] ) , we have @xmath413=\prod\limits_{i=1}^t\sigma_i(\infty).\end{aligned}\ ] ] therefore , @xmath414>0 $ ] implies @xmath415 for all @xmath142 . note that if @xmath416 for some @xmath417 , then @xmath418 form a markov chain , and @xmath419 in the lower bound ( [ lowerbound ] ) , let @xmath420 be uniformly distributed over the interval @xmath421 $ ] . as @xmath422 grows sublinearly in @xmath1 to infinity ,
we get @xmath423\\ & \geq&\liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[i\left({\text{\boldmath{$x$}}}_{i};{\text{\boldmath{$h$}}}_i{\text{\boldmath{$x$}}}_i+{\text{\boldmath{$z$}}}_i\left|\mathbb{e}\left({\text{\boldmath{$h$}}}_{i}\left|\left\{{\text{\boldmath{$h$}}}_k+\delta{\text{\boldmath{$z$}}}_k\right\}_{k=-\infty}^{i-1}\right.\right)\right.\right)-\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right]\\ & = & -1-\gamma-\log\mbox{var}\left({\text{\boldmath{$h$}}}_i\left|\left\{{\text{\boldmath{$h$}}}_j+\delta{\text{\boldmath{$z$}}}_j\right\}_{j=-\infty}^{i-1}\right.\right)\end{aligned}\ ] ] where the last equality follows from ( * ? ? ?
* proposition 4.23 ) .
therefore , @xmath424\nonumber\\ & \geq&\liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[\frac{1}{t}\sum\limits_{i=1}^ti\left({\text{\boldmath{$x$}}}_{i};{\text{\boldmath{$h$}}}_i{\text{\boldmath{$x$}}}_i+{\text{\boldmath{$z$}}}_i\left|\mathbb{e}\left({\text{\boldmath{$h$}}}_{i}\left|\left\{{\text{\boldmath{$h$}}}_k+\frac{1}{x_{\min}}{\text{\boldmath{$z$}}}_k\right\}_{k=-\infty}^{i-1}\right.\right)\right.\right)-\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right]\nonumber\\ & = & -1-\gamma-\frac{1}{t}\sum\limits_{i=1}^t\log\mbox{var}\left({\text{\boldmath{$h$}}}_i\left|\left\{{\text{\boldmath{$h$}}}_j+\delta{\text{\boldmath{$z$}}}_j\right\}_{j=-\infty}^{i-1}\right.\right).\label{fadingnumber}\end{aligned}\ ] ] since ( [ fadingnumber ] ) holds for arbitrary positive @xmath425 , it follows that @xmath426&\geq&-1-\gamma-\frac{1}{t}\lim\limits_{\delta\rightarrow 0}\sum\limits_{i=1}^t\log\mbox{var}\left({\text{\boldmath{$h$}}}_i\left|\left\{{\text{\boldmath{$h$}}}_j+\delta{\text{\boldmath{$z$}}}_j\right\}_{j=-\infty}^{i-1}\right.\right)\nonumber\\ & = & -1-\gamma-\frac{1}{t}\log\det\left[\sigma(\infty)\right].\label{fadingnumberinf}\end{aligned}\ ] ] from the upper bound ( [ upperbound ] ) , we have @xmath427\nonumber\\ & \leq & \limsup\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[\frac{1}{t}\sum\limits_{k=1}^t\sup\limits_{p_{{\text{\boldmath{$x$}}}_k}\in\mathcal{p}_1({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}i\left({\text{\boldmath{$x$}}}_k,\mathbb{e}\left({\text{\boldmath{$h$}}}_k\left|\left\{{\text{\boldmath{$h$}}}_t+\frac{1}{\sqrt{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}}{\text{\boldmath{$z$}}}_{t}\right\}_{t=-\infty}^{k-1}\right.\right);{\text{\boldmath{$y$}}}_k\right)-\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right]\nonumber\\ & \leq&\limsup\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[\frac{1}{t}\sum\limits_{k=1}^t\sup\limits_{p_{{\text{\boldmath{$x$}}}_k}\in\mathcal{p}_1({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}i\left({\text{\boldmath{$x$}}}_k,\mathbb{e}\left({\text{\boldmath{$h$}}}_k\left|\{{\text{\boldmath{$h$}}}_j\}_{j=-\infty}^{k-1}\right);{\text{\boldmath{$y$}}}_k\right.\right)-\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right]\label{markov}\\ & \leq&\limsup\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[\sup\limits_{p_{{\text{\boldmath{$x$}}}_k}\in\mathcal{p}_1({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}i({\text{\boldmath{$x$}}}_1;{\text{\boldmath{$y$}}}_1)+\frac{1}{t}\sum\limits_{k=1}^t\sup\limits_{p_{{\text{\boldmath{$x$}}}_k}\in\mathcal{p}_1({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})}i\left(\left.\mathbb{e}\left({\text{\boldmath{$h$}}}_k\left|\{{\text{\boldmath{$h$}}}_j\}_{j=-\infty}^{k-1}\right.\right);{\text{\boldmath{$y$}}}_k\right|{\text{\boldmath{$x$}}}_k\right)-\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\right]\nonumber\end{aligned}\ ] ] where ( [ markov ] ) follows from the fact that @xmath428 form a markov chain
. it was shown in ( * ? ? ?
* corollary 4.19 ) that @xmath429=-1-\gamma.\end{aligned}\ ] ] furthermore , @xmath430 where ( [ markov2 ] ) follows from the fact that @xmath431 form a markov chain .
therefore , we get @xmath432&\leq&-1-\gamma-\frac{1}{t}\sum\limits_{k=1}^t\log\sigma_k(0)\nonumber\\ & = & -1-\gamma-\frac{1}{t}\log\det\left[\sigma(\infty)\right].\label{fadingnumbersup}\end{aligned}\ ] ] the desired result follows by combining ( [ fadingnumberinf ] ) and ( [ fadingnumbersup ] ) .
at high snr , we have @xmath433\mbox{d}\omega\right|_{\theta={\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{r } } & = & \alpha\log\left(\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)+\left(2\pi-\alpha-{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{-r}\right)\log\left({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{-r}+\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)\\ & & + { \mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{-r}\log\left(\frac{2\pi{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{2r}-2\pi{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{r}+\alpha{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{r}+1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^{r}}+\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)\\ & = & -2\pi\kappa\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c(r)+o(1)\end{aligned}\ ] ] and thus @xmath434 where @xmath435 , and @xmath436 in the lower bound ( [ lowerboundsym ] ) , let @xmath437 be uniformly distributed over the set @xmath438 . by lemma [ lemma : computation ] and equation ( [ predict ] )
, we get @xmath439+\log\left(1-\frac{e^{\frac{c(r)}{2\pi}}}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}-o\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\kappa}\right)\right)-\gamma-\log\frac{5e}{6}\\ & = & \kappa\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+o(\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}).\end{aligned}\ ] ] therefore , @xmath440 when @xmath225 , we have @xmath441 since the noncoherent channel capacity with peak power constraint @xmath442 is upper bounded by the coherent channel capacity with average power constraint @xmath443 , it follows that @xmath444 where @xmath445 .
therefore , @xmath446 the proof is complete .
in the lower bound ( [ lowerboundsym ] ) , let @xmath437 be uniformly distributed over the set @xmath438 . by lemma [ lemma : computation ] , we get @xmath447 where @xmath448^{1-\alpha_2}-\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}.\end{aligned}\ ] ] by specializing the upper - bound ( [ upperbound ] ) to the case where @xmath149 , we obtain @xmath449 where @xmath450\\ & = & -\log\left\{\left(\epsilon_1+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)^{\alpha_1}\left(\epsilon_2+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right)^{\alpha_2-\alpha_1}\left[\frac{1-\alpha_1\epsilon_1-(\alpha_2-\alpha_1)\epsilon_2}{1-\alpha_2}+\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right]^{1-\alpha_2}-\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}\right\}.\end{aligned}\ ] ] note that @xmath451 is the capacity of the memoryless noncoherent rayleigh fading channel ( see @xcite for a nonasymptotic upper bound ) , and we have @xmath452 for large @xmath45 .
we can compute that @xmath471 it can be verified that @xmath472 so the optimal @xmath303 is either @xmath348 or @xmath354 . setting @xmath473 yields @xmath474 which , after some algebraic manipulation , is equivalent to @xmath475 the above equation has two solutions @xmath476 @xmath477 can be discarded since it is always negative .
@xmath478 is positive for @xmath479 . when @xmath479 , it can be verified that @xmath480 if @xmath481 , and @xmath482 if @xmath483 .
when @xmath484 $ ] , @xmath478 is non - positive . in this case , we have @xmath480 for all @xmath48 . t. koch and a. lapidoth , the fading number and degrees of freedom in non - coherent mimo fading channels : a peace pipe , " _
proceedings 2005 ieee international symposium on information theory ( isit ) _ , adelaide , australia , sept . 4 - 9 , 2005
by the random coding bound @xcite , we have @xmath455 where @xmath456+\log\left\{e^r-\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}(e^r-1)}{2}\left[\sqrt{1+\frac{4e^r}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}(e^r-1)}}-1\right]\right\}\label{exponent1}\end{aligned}\ ] ] if @xmath457\leq r\leq\log(1+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})$ ] , and @xmath458 if @xmath459 $ ] . let @xmath460 where @xmath461 . by ( [ exponent1 ] ) , @xmath462+\log\left\{\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{\eta}-\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta)}{2\eta}\left[\sqrt{1+\frac{4}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta}}-1\right]\right\}\\ & = & \lim\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\frac{\eta}{2}\left[\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{\eta}+1-\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta}{\eta}\left(1+\frac{2}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta}\right)\right]+\log\left\{\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}{\eta}-\frac{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta)}{2\eta}\left[\frac{2}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta}-\frac{2}{({\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}-\eta)^2}\right]\right\}\\ & = & \eta-1-\log\eta\\ & > 0&.\end{aligned}\ ] ] for any @xmath463 , we can find an @xmath61 such that @xmath464 therefore , for any @xmath463 , there exist a sequence of codebooks with rate @xmath259 and fixed codeword length @xmath61 such that @xmath465 and @xmath271 .
choosing @xmath467 and @xmath468 , we get @xmath469\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[-\log\mathbb{e}_{{\text{\boldmath{$h$}}}}\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{1}{2}|{\text{\boldmath{$h$}}}|^2\right)^{-1}+\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c\right]\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left\{-\log\left[\int_{0}^\infty\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{t}{2}\right)^{-1}e^{-t}\mbox{d}t\right]+\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c\right\}\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left\{-\log\left[\int_{0}^1\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{t}{2}\right)^{-1}e^{-t}\mbox{d}t+\int_{1}^\infty\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{t}{2}\right)^{-1}e^{-t}\mbox{d}t\right]+\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c\right\}\\ & \geq&\liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left\{-\log\left[\int_{0}^1\left(\frac{1}{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}}+\frac{t}{2}\right)^{-1}\mbox{d}t+\int_{1}^\infty2e^{-t}\mbox{d}t\right]+\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c\right\}\\ & = & \liminf\limits_{{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}\rightarrow\infty}\left[-\log\left[2\log(2 + 2{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}^2)-2\log(2 + 2{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}})+2e^{-1}\right]+\log\log{\mathsf{s\hspace{-0.011in}n\hspace{-0.011in}r}}+c\right]\\ & = & -\log 2+c\end{aligned}\ ] ] which is positive if @xmath470 . | we consider a peak - power - limited single - antenna block - stationary gaussian fading channel where neither the transmitter nor the receiver knows the channel state information , but both know the channel statistics .
this model subsumes most previously studied gaussian fading models .
we first compute the asymptotic channel capacity in the high snr regime and show that the behavior of channel capacity depends critically on the channel model .
for the special case where the fading process is symbol - by - symbol stationary , we also reveal a fundamental interplay between the codeword length , communication rate , and decoding error probability .
specifically , we show that the codeword length must scale with snr in order to guarantee that the communication rate can grow logarithmically with snr with bounded decoding error probability , and we find a necessary condition for the growth rate of the codeword length .
we also derive an expression for the capacity per unit energy .
furthermore , we show that the capacity per unit energy is achievable using temporal on - off signaling with optimally allocated on symbols , where the optimal on - symbol allocation scheme may depend on the peak power constraint . wireless channels , noncoherent capacity , capacity per unit cost , block fading |
we recently studied effects of particle size and final - state interactions in the reaction @xmath0 within the framework of a constituent quark model @xcite .
the aim was to improve previous attempts @xcite to describe the lear data on @xmath0 differential cross sections @xmath2 and analyzing powers @xmath3 in the momentum range from 360 to 1550 mev/@xmath4 @xcite .
as of yet , theoretical approaches , whether using a baryonic or a quark picture @xcite , have not been successful in reproducing the characteristic double - dip structure of the @xmath3 observables nor the @xmath2 forward peaks at low momenta .
the large variations of the lear observables as a function of the angle portends the presence of a substantial number of partial wave amplitudes already at low energies .
in fact , the experimental data on differential cross sections as well as those on asymmetries point to a significant @xmath5 , @xmath6 and even higher @xmath7 contributions @xcite .
model calculations @xcite , however , lead invariably to scattering amplitudes which are strongly dominated by total angular momentum @xmath8 and @xmath9 .
this is due to a rather short range of the annihilation mechanism in these models . in order to study possible higher partial wave @xmath1 contributions ,
the role of final - state @xmath10 interactions has been investigated in ref . @xcite .
for example , in ref .
@xcite we made use of the non - planar @xmath11 quark - flow diagrams , in which a @xmath12 pair in either a @xmath13 or a @xmath14 state is annihilated and momentum is transferred to a remaining quark or antiquark as discussed in ref .
switching on the @xmath10 interaction moderately improved the fit of @xmath3 , in particular the double - dip structure at low momenta is slightly more pronounced .
this is caused by a readjustment of the strengths of the helicity amplitudes of different total angular momentum @xmath7 .
nonetheless , the predictions for @xmath2 showed only a modest improvement over the model without final - state interactions .
the main improvement obtained in ref .
@xcite , however , is due to a different aspect , namely a readjustment of the size parameters of the intrinsic proton and pion wave functions , so that the radii of the proton , antiproton , and pions increase . in a final fit
these radii are larger by about 7% than the respective measured charge distribution radii .
this runs contrary to the view that the valence quarks occupy a smaller volume than indicated by the charge radii .
nevertheless , introducing larger radii as well as final - state interactions , improves the quality of the fit in ref .
@xcite dramatically the forward and backward peaks of @xmath2 are very well reproduced and so are the characteristic double - dip structures of @xmath3 .
it should come as no surprise that the relative contribution of the higher @xmath1 amplitudes in ref .
@xcite turns out to be significantly larger than in ref .
@xcite in this paper , we report an alternative approach that also leads to enhancement of the higher partial waves .
it addresses the relativistic effects due to lorentz transformed intrinsic pion wave functions in the reaction @xmath0 .
the reason for this rather different approach stems from the observation that at the center - of - mass ( c.m . )
energies @xmath15 considered in the lear experiment @xmath0 , the produced pions are highly energetic .
the relativistic factor in this energy range is @xmath16 , which means the pions are ejected at speeds @xmath17 .
one therefore expects considerable relativistic effects due to their modified internal structure . in the c.m .
frame in which the transition amplitudes are calculated , the intrinsic pion wave functions employed must be lorentz transformed from the pion rest frame . in our model
@xcite the intrinsic pion as well as the proton and antiproton wave functions are of gaussian form in their respective rest frames .
we neglect the distortions of the proton and antiproton intrinsic wave function due to their much smaller kinetic energy and larger mass .
the relativistic effects have several consequences .
first , the intrinsic geometry of the pions is modified instead of spherical particles one deals now with highly flattened ellipsoids in the c.m .
obviously , the reaction geometry is altered also since the boosted pion wave functions enter the computation of the transition operators , which are obtained from an overlap integral of initial antiproton - proton , final pion - pion wave functions and the @xmath13 or @xmath14 annihilation mechanism .
details of this calculation were presented in ref .
@xcite . secondly , due to less overlap of the pion and antiproton - proton intrinsic wave functions , the annihilation range _ actually _ decreases but this effect is far from isotropic . as a result
the transition operators exhibit significant additional angle dependence . finally , the annihilation amplitudes , already non - local in the case of non - relativistic wave functions , become explicitly dependent on the c.m .
energy @xmath15 via the boost factor @xmath18 . in ref .
@xcite it was shown that the relativistic transformation of the spatial part of the intrinsic pion wave functions introduces new angle dependent terms in the transition operators .
these new terms also depend on the boost factor @xmath18 . in this paper
, we present a new fit to the lear data on @xmath0 using the @xmath11 transition operators of ref .
@xcite , which supports our claim that relativistic considerations lead to a strongly modified angular momentum content of the helicity amplitudes for @xmath19 .
especially the amplitudes for @xmath1 are amplified considerably .
even though we keep the original particle radii as in @xcite , i.e. smaller than the corresponding charge distribution radii , we achieve a very good reproduction of the lear observables @xmath2 and @xmath3 , comparable to the one in ref . @xcite .
we summarize the effects of lorentz transforming intrinsic pion wave functions on @xmath0 annihilation operators within the constituent quark model .
the actual details may be found in ref .
@xcite . in the pion rest frame ,
the radial part of its intrinsic wave function is described by a gaussian of the form @xmath20 which reads in the c.m .
frame @xmath21 \!\right \}\!\!.
\label{rwaveboost}\end{aligned}\ ] ] the vectors @xmath22 and @xmath23 are the quark and antiquark coordinates , @xmath24 is the pion coordinate , and @xmath25 is the size parameter .
as mentioned above , in previous work @xcite we varied the value of @xmath25 in order to obtain an increase in the annihilation range . in this paper
we use throughout the fixed value @xmath26@xmath27 , which corresponds to a pion radius of 0.48 fm @xcite . the new normalization factor
@xmath28 is due to the condition that @xmath29 be normalized to unity in the c.m .
wave functions of eq .
( [ rwaveboost ] ) are used to compute the transition operators for the @xmath14 and @xmath13 annihilation amplitudes .
it is shown in ref .
@xcite that these operators acquire new terms if compared with the non - relativistic expressions @xcite and which manifestly introduce additional angular dependence .
we here briefly recapitulate the results for the @xmath11 diagrams from @xcite .
the complete form of the @xmath11 transition operators for the _ vacuum _
@xmath14 amplitude is @xmath30 \times \nonumber \\
\times\ , \exp \{a\mathbf{r'}^2+b\mathbf{r}^2+d \mathbf{r}^2\cos^2\theta \ } . \label{tot1 } \end{aligned}\ ] ] the total @xmath13 amplitude for the _ longitudinal _ component is given by @xmath31 \times \nonumber \\ \times\ , \exp \{a\mathbf{r'}^2+b\mathbf{r}^2+d \mathbf{r}^2\cos^2\theta \ } \label{tot2}\end{aligned}\ ] ] and @xmath32 \times \nonumber \\ \times\ , \exp \{a\mathbf{r'}^2+b\mathbf{r}^2+d \mathbf{r}^2\cos^2\theta \ } \label{tot3}\end{aligned}\ ] ] for the _ transversal _ component .
the factor @xmath33 is an overall normalization and @xmath34 and @xmath35 are the relative @xmath36 and antiproton - proton coordinates , respectively .
the angle @xmath37 is between @xmath38 and @xmath39 in the c.m . frame . in the experiment it is the c.m .
angle between the antiproton beam direction and the outgoing @xmath40 direction .
the new terms mentioned previously are @xmath41 , @xmath42 , @xmath43 , and @xmath44 .
the selection rules for the @xmath11 diagrams discussed in ref .
@xcite are unchanged .
more precisely , @xmath45 and @xmath46 act in @xmath47 states with @xmath48 waves while @xmath49 contributes only to @xmath50 waves .
the explicit expressions for the coefficients of eqs .
( [ tot1])([tot3 ] ) are ( with @xmath51 being the proton size parameter defined similarly to @xmath25 in eq .
( [ pionnoboost ] ) ) : @xmath52 @xmath53 @xmath54 @xmath55 clearly , @xmath41 , @xmath42 , @xmath43 , and @xmath44 vanish in the non - relativistic limit @xmath56 .
a noticeable feature is that some of the new coefficients become relatively large for large @xmath18 values . in particular
, the @xmath57 term dominates the exponential part of eqs .
( [ tot1])([tot3 ] ) and , for example , @xmath58 and @xmath59 . on the other hand
, one finds that @xmath60 . at this stage
we will neglect additional relativistic corrections due to the ( anti)quark spinors in the final state pions since it was argued in ref .
@xcite that boosting of spin wave functions results in relatively smaller corrections .
the lear data for the @xmath2 and @xmath3 have been fitted using the transition operators in eqs .
( [ tot1 ] ) , ( [ tot2 ] ) , and ( [ tot3 ] ) . as before , taking only the @xmath11 topology into account , we use a distorted wave approximation ( dwa ) @xmath61 to obtain the scattering matrix elements and introduce final - state interactions as in ref .
the pion wave function @xmath62 is obtained from the coupled - channel model of ref .
@xcite and can be parameterized with phases @xmath63 and inelasticities @xmath64 .
the initial @xmath47 wave function @xmath65 is taken from an optical potential model @xcite .
both @xmath62 and @xmath65 still contain the angle dependence appropriate for the values of @xmath7 and @xmath66 .
we select from the specific energies where lear data are available @xcite the same set of five energies as in ref .
@xcite : @xmath67 , 123.5 , 219.9 , 499.2 , and 803.1 mev corresponding to antiproton momenta of respectively @xmath68 , 497 , 679 , 1089 , and 1467 mev/@xmath4 .
@xmath69 is the laboratory kinetic energy of the antiproton beam . in fitting the data ,
we proceed as in ref .
one starts with @xmath36 final - state plane waves , i.e. with @xmath70 and @xmath71 , and varies these parameters for @xmath19 .
the other parameters are the relative ( complex ) strength @xmath72 as defined by the @xmath11 transition amplitude @xmath73 , \ ] ] and the overall normalization @xmath74 .
the size parameters @xmath51 and @xmath25 of the quark model are kept fixed at the original values @xmath75@xmath27 and @xmath26@xmath27 @xcite . the aim is to find out how much improvement is obtained solely from introducing final - state interactions . at this initial stage ,
no relativistic effects have been included which implies @xmath76 .
the results of this fit are plotted in figs .
[ fig1][fig5 ] as dashed lines .
they will serve as reference point and were obtained previously in ref .
@xcite where the model with and without final - state interactions was discussed .
it was found that the improvement from just final - state interactions is very modest the double - dip structure in the analyzing power is somewhat more enhanced but the forward peaks in the differential cross sections at @xmath77 and 123.5 mev ( @xmath68 and 497 mev/@xmath4 ) as well as the backward peaks at @xmath78 and 803.1 mev ( @xmath79 and 1467 mev/@xmath4 ) are poorly reproduced .
next , we consider the effects of relativistic distortions of the intrinsic pion wave functions . hence ,
the transition operators of eqs .
( [ tot1])([tot3 ] ) must be used .
the exponential part is dominated by the @xmath44 term ( typically @xmath80@xmath27 , about two orders of magnitude larger than the @xmath81 term ) , which makes the partial wave amplitudes @xmath82 very sensitive to the value of the relative strength @xmath83 of the @xmath13 versus the @xmath14 mechanism and to the pion wave parameters @xmath64 and @xmath63 .
this is a consequence of the strong angle dependence of @xmath84 . in order for the partial wave expansion of the scattering amplitude @xmath85 to converge ,
all ingredients must be fine - tuned , in particular the phases @xmath63 and inelasticities @xmath64 for @xmath19 .
it should be emphasized that @xmath51 and @xmath25 remain fixed .
the improved new fit is illustrated in figs .
[ fig1][fig5 ] , where the predictions that include relativistic effects and final - state interactions are presented as solid curves .
the new fit is a significant improvement over the non - relativistic version of the @xmath11 annihilation model .
the differential cross sections @xmath2 are particularly well reproduced for the energies @xmath77 , 123.5 , and 219.9 mev ( @xmath68 , 497 , and 679 mev/@xmath4 ) . for @xmath78 and 803.1 mev ( @xmath86 and 1467 mev/@xmath4 ) the results for @xmath2 show improvement in forward and backward direction . at the two higher energies , the experimental double - well structure of the cross section
is reproduced but some problems remain near @xmath87 . as for the analyzing powers @xmath3 , we are able to reproduce their double - dip structure , already present in the lear data at the lowest energies . when the experimental errors are small , like for instance for @xmath88 and 499.2 mev , the fit is particularly successful .
at lowest energy , @xmath67 mev , the error bars in @xmath3 are rather large and we do not fit very well certain data points .
overall the improvement is clearly important and of similar quality as the results of ref .
@xcite which were obtained from increasing the particle radii .
.quark model parameters as function of @xmath69 .
[ cols="^ , > , > , > , > , > " , ] in tab .
[ tab1 ] we give the values of the quark model parameters as a function of @xmath69 while the @xmath10 phases @xmath63 and inelasticities @xmath64 are conveniently listed as a function of c.m .
energy @xmath15 in tab .
[ tab2 ] . in the quark model parameters of tab .
[ tab1 ] one notes that the relative strength @xmath89 of the @xmath13 versus the @xmath14 mechanism varies with energy and the average value @xmath90 .
there is no _ a priori _ reason why @xmath72 should be energy independent .
the energy variation of the inelasticities @xmath64 in the @xmath10 interaction observed in tab .
[ tab2 ] may be associated with the presence of @xmath10 resonances , in particular with the @xmath91 at @xmath92 mev in the @xmath5 amplitude .
we present a new fit to the lear @xmath0 data based on relativistic effects in the annihilation .
these effects are due to lorentz transformations of intrinsic pion wave functions from the pion rest frame to the c.m .
frame of the reaction .
the impact from the modified geometry of the pions is considerable and carries with it a much richer angle dependence of the transition operators .
the quality of this new fit is very similar to the one of a previous fit @xcite .
however , the approaches are very different . while in ref .
@xcite we achieved a good fit by means of particle radii increase , so the antiquarks and quarks do overlap more , in this work we keep the original particle sizes fixed but take into account the relativistic distortions of the pions in the c.m .
frame of the reaction .
we conclude that even though observed data at lear on @xmath2 and @xmath3 indicate strongly the presence of higher partial wave amplitudes and therefore seem to call for an increase of the annihilation range , one can obtain a similar improvement by additional angle dependence in the transition operators .
this additional angle dependence arises naturally when boosted intrinsic pion wave functions are employed in the quark model calculations , as becomes clear from the present fit .
relativistic effects in the reaction @xmath0 therefore should not be ignored .
b. el - bennich , w.m .
kloet , and b. loiseau , phys .
c * 68 * , 014003 ( 2003 ) .
g. bathas and w.m .
kloet , phys .
b**301 * * , 155 ( 1993 ) . g. bathas and w.m .
kloet , phys . rev .
c * 47 * , 2207 ( 1993 ) .
a. hasan _ et al .
b**378 * * , 3 ( 1992 ) .
b. moussallam , nucl . phys . *
a407 * , 413 ( 1983 ) ; _ ibid_. * a429 * , 429 ( 1984 )
. v. mull , j. haidenbauer , t. hippchen , and k. holinde , phys .
c * 44 * , 1337 ( 1991 ) .
v. mull , k. holinde , and j. speth , phys .
b * 275 * , 12 ( 1992 ) .
a.m. green and g.q .
liu , nucl .
phys * a486 * , 581 ( 1988 ) .
a. muhm , t. gutsche , r. thierauf , y. yan , and amand faessler , nucl . phys . *
a598 * , 285 ( 1996 ) .
y. yan and r. tegen , phys .
c * 54 * , 1441 ( 1996 ) ; nucl .
phys * a648 * , 89 ( 1999 )
oakden and m.r .
pennington , nucl .
* a574 * , 731 ( 1994 ) .
a. hasan and d.v .
bugg , phys .
b * 334 * , 215 ( 1994 ) .
kloet and f. myhrer , phys .
d * 53 * , 6120 ( 1996 ) .
martin and g.c oades , phys .
c * 56 * , 1114 ( 1997 ) ; _ ibid_. * 57 * , 3492 ( 1998 ) .
b. el - bennich and w.m .
kloet , [ arxiv : nucl - th/0403050 ] a.m. green and j.a .
niskanen , nucl .
a412 * , 448 ( 1984 ) ; _ ibid_. * a430 * , 605 ( 1984 ) ; mod .
lett a * 1 * , 441 ( 1986 ) .
kloet and b. loiseau , eur .
j. a * 1 * , 337 ( 1998 ) .
b. el - bennich , m. lacombe , b. loiseau , and r. vinh mau , phys .
c * 59 * , 2313 ( 1999 ) . | we present a new fit to the lear data on @xmath0 differential cross sections and analyzing powers motivated by relativistic considerations . within a quark model describing this annihilation
we argue , since the pions are highly energetic , that relativistic effects can not be neglected .
the intrinsic pion wave functions are lorentz transformed to the center of mass frame .
this change in quark geometry gives rise to additional angular dependence in the transition operators and results in a relative enhancement of higher @xmath1 partial wave amplitudes .
the fit to the data is improved significantly . |
when an intense electric field is applied to a suspension containing highly polarized particled embedded in a host fluid , the particles will form chains due to the induced dipole moments inside the particles , and hence complex anisotropic structures occur .
also , the apparent viscosity of the suspension will be enhanced greatly .
these structures can have anisotropic physical properties , such as the effective conductivity and permittivity @xcite . in this case ,
optical nonlinearity enhancement was found as well @xcite .
a convenient method of probing the nonlinear characteristics of the field - induced structures is to measure the harmonics of the nonlinear polarization under the application of a sinusoidal ( ac ) electric field @xcite .
therefore it is possible to perform a real - time monitoring of the field - induced aggregation process in such a suspension by measuring the optical nonlinear ac responses both parallel and perpendicular to the uniaxial anisotropic axis .
recently , an anisotropic maxwell - garnett theory @xcite was applied to study the nonlinear polarization of these anisotropic structures @xcite .
anisotropic optical nonlinear ac response was indeed found . when a nonlinear composite with nonlinear dielectric particles embedded in a host medium , or with a nonlinear host medium is subjected to a sinusoidal field
, the electrical response in the composite will in general be a superposition of many sinusoidal functions @xcite .
it is natural to investigate the effects of a nonlinear characteristics on the optical nonlinear ac response in a colloidal suspension which can be regarded as a nonlinear composite medium @xcite .
the strength of the nonlinear polarization should be reflected in the magnitude of the harmonics . for these field - induced anisotropic structures
, we found that as more and more chains are formed the harmonics of local field and induced dipole moment increase ( decrease ) for longitudinal ( transverse ) field case @xcite . on the other hand
, the effect of nonlinear characteristics on the inter - particle force has been analyzed in a colloidal suspension of nonlinear particles @xcite and further extended to a nonlinear host medium @xcite .
actually , the intrinsic dielectric dispersion often occurs due to the surface conductivity or the inhomogeneous structure ( e.g. , coating effect @xcite ) which often exist in colloidal suspensions .
hence it is of particular importance to discuss the effect of an intrinsic dielectric dispersion on the nonlinear polarization . in this regard , we will use a perturbation expansion method to calculate the nonlinear ac response of a nonlinear composite , i.e. , harmonics of local field as well as induced dipole moment .
let us consider the system in which the suspended particles of radius @xmath0 with a nonlinear dielectric constant are suspended in a host medium of linear dielectric constant @xmath1 .
the nonlinear characteristic gives rise to a field - dependent dielectric coefficient @xcite . in this case ,
the electric displacement - electric field relation inside the spheres is given by : @xmath2with @xmath3 , where @xmath4 and @xmath5 are the linear coefficient and the weak nonlinear coefficient of the suspended particles , respectively . here
@xmath6 is expressed as a debye expression , @xmath7 is the dielectric constant at high frequency , @xmath8 is the intrinsic dispersion of the particle , @xmath9 is the intrinsic characteristic frequency , with @xmath10 being the intrinsic relaxation time , and @xmath11 is the angular frequency of the external field . in eq.([nl - coefficient ] ) , we have adopted an approximation : the local field inside the particles is assumed to be uniform .
this assumption is called the decoupling approximation @xcite .
it has been shown that such an approximation yields a lower bound for the accurate result for the local field @xcite .
we further assumed that @xmath5 is independent of frequency , which is a valid assumption for low - frequency processes in colloidal suspensions .
as a result , the induced dipole moment under an applied field @xmath12 along @xmath13axis , with @xmath14 being time , is given by @xcite : @xmath15where @xmath16 is the effective dielectric constant of the suspension , and @xmath17 is the field - dependent dipole factor : @xmath18 in order to obtain the effective dielectric constant @xmath16 , we invoke the maxwell - garnett approximation ( mga ) for anisotropic composites @xcite . for the longitudinal field case when the ac field is applied along the uniaxial anisotropic axis , the mga has the form = f .
[ mga ] with @xmath19 being the volume fraction of particles , whereas for an transverse field case when the ac field is applied perpendicular to the uniaxial anisotropic axis , the mga expression is obtained by replacing @xmath20 with @xmath21 @xcite . here
@xmath22 and @xmath23 denote the local field factors parallel and perpendicular to the uniaxial anisotropic axis .
these local field factors are defined as the ratio of the local field in the particles to the lorentz cavity field @xcite . for isotropic composites , @xmath24 , while both @xmath20 and @xmath25 will deviate from unity for an anisotropic distribution of particles in composites .
these @xmath26 factors have been evaluated in a tetragonal lattice of dipole moments @xcite and in various field - structured composites @xcite .
we resort to the spectral representation approach @xcite to represent the local field inside the particle @xcite : |*e*_1|^2 = _ 0 ^ 1x , [ ee-1 ] with the material parameter @xmath27 as well as the spectral density function @xmath28 $ ] , which satisfies the sum rule @xmath29 . hence eq.([ee-1 ] ) has the form |*e*_1|^2 = ^2(t ) .
[ average1 ] actually , this is a self - consistent equation for @xmath30 , which can be solved at least numerically @xcite . in
what follows , we will use a perturbation method instead to obtain analytic expressions for harmonics of the local electric field and the induced dipole moment . on the other hand , due to the uniform local field in each particle , eq.([average1 ] ) can also be directly derived from @xcite : if we apply a sinusoidal electric field , i.e. @xmath31 , the induced dipole moment @xmath32 will depend on time sinusoidally , too . by virtue of the inversion symmetry
, @xmath32 is a superposition of odd - order harmonics such that = p_t+p_33t+ .
also , the local electric field contains similar harmonics = e_t+e_33t+ .
these harmonic coefficients can be extracted from the time dependence of the solution of @xmath32 and @xmath33 .
in this section , we will apply the perturbation expansion @xcite to extract the harmonics of the local electric field and the induced dipole moment .
it is known that the perturbation expansion method is applicable to weak nonlinearity only , limited by the convergence of the series expansion .
we expand @xmath32 and @xmath34 into a taylor expansion , taking @xmath35 as a perturbative quantity : = _ n=0^ _ _ 1=_1(_1|*e*_1|^2)^n , = _
m=0^ _ _ 1=_1(_1|*e*_1|^2)^m . in view of weak nonlinearity
, we can rewrite eqs.(10 ) and ( 11 ) , keeping the lowest orders of @xmath35 and @xmath36 : @xmath37 where @xmath38,\nonumber\\ j_1&=&\frac{3|\ep_2|}{\rho},\ \ j_3=\frac{27}{2}|\ep_2|^3\frac{f\beta-1}{\rho|\rho|}(\frac{1}{|\rho|^2}+\frac{1}{(\rho^{2})^*}),\nonumber\end{aligned}\ ] ] with @xmath39 . in the case of a sinusoidal field
, we can expand @xmath40 in terms of the first and the third harmonics .
the comparison with eq.(8 ) yields the harmonics of the induced dipole moment : p _
e+h_3_1 e^3 , p_3 = -h_3 _ 1 e^3 .
[ p3w ] similarly , we find the harmonics of the local electric field _ 1 ^ 1/2e _
= j_1_1 ^ 1/2e+j_3(_1 ^ 1/2e)^3 , _ 1 ^ 1/2e_3 = -j_3(_1 ^ 1/2e)^3 .
[ e3w ] in the above analysis , we have used the identity @xmath41 to obtain the first and the third harmonics .
similar analysis can be used to extract the higher - order harmonics , by retaining more terms in the series expansion .
for instance , to obtain fifth harmonics , we may take into account the identity @xmath42 , and then obtain the harmonics of induced dipole moment and local electric field , respectively : @xmath43 where @xmath44 + \frac{3\ep_2 ^ 2}{\rho(\ep_1 + 2\ep_2)^2}\nonumber\\ & & [ \frac{(1 + 3f - f\beta)\rho-(1-f\beta)(\rho+3f\ep_1 - 3f\ep_2)}{\rho}- \frac{2\rho+6f(\ep_1-\ep_2)}{\ep_1 + 2\ep_2 } ] \},\nonumber\end{aligned}\ ] ] and @xmath45 where @xmath46 because of the weak nonlinearity under consideration , the effect of high order harmonics is quite small .
thus , without loss of generality , we only consider the taylor expansion up to second term . in this regard , we will use eqs.([p3w ] ) and ( [ e3w ] ) to investigate the nonlinear ac response in the numerical calculations in next section .
we are now in a position to perform numerical calculations to investigate the dispersion effect on the harmonics of local electric field as well as induced dipole moment .
we let @xmath47 , @xmath48 , @xmath49 , where @xmath50 is the dielectric constant of free space . additionally , set @xmath51 to normalize the induced dipole moment , where @xmath52 is given by the anisotropic mga ( eq.([mga ] ) ) without nonlinear characteristic .
obviously , the expressions for the normalized local electric field and the normalized induced dipole moment are functions of a single variable @xmath53 @xcite . in view of the weak nonlinearity , we set @xmath54 for our calculation . in fig.1(a)@xmath55(c )
, we investigate the harmonics of local field and induced dipole moment versus frequency for different intrinsic dispersion strengths @xmath8 .
it is obvious that increasing the dispersion strength reduces the harmonics of both the local field and the induced dipole moment .
we find that , for increasing frequency , the harmonics of local field always increases .
however , the harmonics of induced dipole moment exhibits a more complicated behavior . for small @xmath8 ,
the ratio of the third to first harmonics of induced dipole moment decreases as the frequency increases , which is qualitatively in good agreement with the experimental data @xcite . however , for large @xmath8 , its harmonic ratio also increases as the frequency increases .
we expect this behavior can be demonstrated by experiment in the future . in fig.2 , we investigate the harmonics of local field and induced dipole moment versus frequency for different intrinsic relaxation times @xmath10 . a longitudinal field case ( @xmath56 ) is discussed in fig.2(a ) , while a transverse field case ( @xmath57 ) in fig.2(b ) .
it is shown that , in either case , increasing relaxation time @xmath10 may enhance the harmonics of both local field and induced dipole moment .
in addition , the transverse field case may predict stronger harmonics than the longitudinal field case within high - frequency region , which is actually valid for different dispersion strengths as well ( no figures shown herein ) . on the other hand , klingenberg claimed that increasing external electric field leads to increasing harmonics of electric current @xcite . in this sense , we have also calculated the effect of external electric field on the nonlinear ac response of the induced dipole moment ( no figures shown herein ) .
obviously , our calculations qualitatively demonstrate the experimental results @xcite as well .
here a few comments on our results are in order . in the present study
, we have examined the case of nonlinear particles suspending in a linear host .
our considerations may be extended to a nonlinear host medium @xcite .
we have studied the effect of an intrinsic dispersion on the nonlinear ac response of a colloidal suspension .
actually , our formalism can be readily generalized to discuss two or more intrinsic dielectric dispersions of particles ( or host fluid ) . in view of the fact that the intrinsic dielectric dispersion often occurs due to the surface conductivity or the inhomogeneous structure ( e.g. , coating effect @xcite )
, we believe such effect also plays an important role in the dielectric behavior of biological cells , such as the electrorotation , dielectrophoresis and electro - orientation .
the perturbation method is used in the present work , but we believe similar results may also be predicted by a self - consistent theory @xcite . in summary , a perturbation method has been employed to compute the local electric field and the induced dipole moment for suspensions in which the suspended particles have a nonlinear characteristic and an intrinsic dispersion , in an attempt to investigate the frequency effect on the nonlinear ac response .
the results showed , for weak intrinsic dispersion strength , the ratio of the third to first harmonics of the induced dipole moment decreases as the frequency increases , which is qualitatively in agreement with experimental result .
however , for a strong dispersion strength , the harmonics ratio increases as the frequency increases .
moreover , an increase in the intrinsic relaxation time may increase the strength of harmonics .
this work was supported by the research grants council of the hong kong sar government under project numbers cuhk 4284/00p and cuhk 4245/01p .
k.w.y . thanks prof .
a. r. day for his interest in our work and his suggestion to examine the effects of inhomogeneities on the dielectric response of biological cell suspensions .
j. e. martin , r. a. anderson and c. p. tigges , in _ nanostructured materials : clusters , composites , and thin films _ , edited by v. m. shalaev and m. moskovitz ( acs symposium series * 679 * , 1996 ) , chap . 6 . j. e. martin , r. a. anderson and c. p. tigges , j. chem . phys . * 108 * , 3765 ( 1998 ) ; * 108 * , 7887 ( 1998 ) . k. p. yuen , m. f. law , k. w. yu and ping sheng , phys .
e * 56 * , r1322 ( 1997 ) .
d. j. klingenberg , mrs bulletin * 23 * , 30 ( 1998 )
. o. levy , d. j. bergman and d. stroud , phys .
e * 52 * , 3184 ( 1995 ) .
g. q. gu , p. m. hui and k. w. yu , physica b * 279 * , 62 ( 2000 ) .
jones t. k. wan , g. q. gu and k. w. yu , phys .
e * 63 * , 052501 ( 2001 ) .
jones t. k. wan , g. q. gu and k. w. yu , int .
b * 14 * , 603 ( 2000 ) .
l. gao , jones t. k. wan , k. w. yu and z. y. li , phys .
e * 61 * , 6011 ( 2000 ) .
l. gao and z. y. li , j. appl .
* 87 * , 1620 ( 2000 ) ; see also l. gao , jones t. k. wan , k. w. yu and z. y. li , j. appl .
phys . * 88 * , 1893 ( 2000 ) .
k. w. yu , p. m. hui and d. stroud , phys .
b * 47 * , 14150 ( 1993 ) .
k. w. yu , p. m. hui and h. c. lee , phys .
a * 210 * , 115 ( 1996 ) ; see also w. m. v. wan , h. c. lee , p. m. hui and k. w. yu , phys . rev .
b * 54 * , 3946 ( 1996 ) . j. e. martin and r. a. anderson , j. chem
111 * , 4273 ( 1999 ) .
d. j. bergman , phys .
rep . * 43 * , 379 ( 1978 ) . | when a sinusoidal ( ac ) field is applied to a suspension containing nonlinear dielectric particles , the electrical response will generally consist of ac fields at frequencies of the higher - order harmonics .
the situation is further complicated by an intrinsic dielectric dispersion which often occurs due to the surface conductivity or inhomogeneous structure of the particles .
we perform a perturbation method to investigate the effect of intrinsic dielectric dispersion on the harmonics of local field as well as induced dipole moment .
the results showed , for weak intrinsic dispersion strength , the ratio of the third to first harmonics of the induced dipole moment decreases as the frequency increases , which is qualitatively in agreement with experimental result .
however , for a strong dispersion strength , the harmonics ratio increases as the frequency increases .
moreover , an increase in the intrinsic relaxation time may increase the strength of harmonics . |
an isolated classical system always has a fixed value of the total energy .
in contrast , an isolated quantum system can be in a superposition of states with different total energies @xcite .
this entails the following difficulty for the foundations of statistical physics : a typical isolated macroscopic quantum system is generally expected to thermalise under the action of its internal dynamics for the overwhelming majority of initial nonequilibrium states appearing in nature or created in a laboratory .
thermalisation implies that the density matrix of any small subsystem within the large system approaches a form consistent with the canonical gibbs density matrix ( canonical ensemble ) for the large system . a necessary condition for an isolated many - particle quantum system to thermalise is a sufficiently narrow initial probability distribution of its total energy , as is , for example , the case for the canonical and the microcanonical statistical ensembles ( see below ) . at the same time , for an isolated quantum system , the probability of occupying an energy eigenstate remains unchanged with time .
thus , a large isolated quantum system with a broad initial distribution of the total energy thermalises not to a conventional equilibrium state with a well - defined temperature but rather to a mixture or superposition of states with different temperatures .
the problem now is that the initial states characterised by the narrow distribution of total energy are _ not _ necessarily the most probable ones .
for example , non - gibbs equilibrium for small subsystems emerges when initial states are selected in the hilbert space of a typical many - particle system among quantum superpositions with a given energy expectation value and without any constraint on the width of the energy window for participating eigenstates .
the latter condition defines the `` quantum micro - canonical '' ( qmc ) ensemble @xcite .
given the above considerations , why do the initial nonequilibrium quantum states of macroscopic systems not normally exhibit the broad participation of energy eigenstates and hence non - gibbs statistics for small subsystems ?
in this paper , we address the above question by introducing the criterion of stability of quantum statistical ensembles with respect to local measurements and then apply this criterion to lattices of spins 1/2 .
similar ideas in other contexts have been considered in refs .
@xcite .
a quantum statistical ensemble for a macroscopic system is defined by the probability @xmath0 of occupying an eigenstate of total energy @xmath1 . given the density of energy states @xmath2 , the probability distribution of the total energy is @xmath3 we call @xmath4 broad , when @xmath5 , where @xmath6 is the variance of @xmath4 , @xmath7 is the average energy , and @xmath8 is the ground - state energy of the system .
as shown in appendix [ app_width ] , a canonical ensemble with a positive temperature @xmath9 is narrow from the above perspective , because , in this case , @xmath10 , where @xmath11 is the number of particles or microscopic subsystems in the system , cf . ref .
@xcite .
the article is organised as follows .
section [ sec_stab ] provides the definition of the stability criterion for quantum statistical ensembles . in sec .
[ sec_effect ] , the effects of local measurements are discussed qualitatively .
section [ sec_spinlat ] includes an analytical investigation of the stability of quantum ensembles for lattices of spins-1/2 , including noninteracting spins in a magnetic field and systems of interacting spins .
section [ sec_numerical ] presents the results of numerical investigations for systems of interacting spins .
final remarks and conclusions are given in sec .
[ sec_conclude ] .
longer derivations are included in the appendices .
let us now observe that accidental measurements of microscopic particles in a macroscopic system can not be excluded under any foreseeable natural or experimental conditions .
we , therefore , introduce the following stability criterion : _ a physically realisable quantum statistical ensemble describing a stationary state of a macroscopic system must be stable with respect to a small number of any arbitrarily chosen local measurements within the system .
the measurement is called `` local '' if the measured quantity is localised in the three - dimensional physical space .
the number of measurements @xmath12 is called small if @xmath13 .
the ensemble is called stable if _
@xmath14 _ where @xmath15 and @xmath16 are the probability distributions before and after the measurements , respectively . _ the measurements in question
are implied to occur naturally , for example , when a passing photon becomes entangled with the system and accidentally measured later . without the measurement ,
the above process would describe a decoherence event @xcite . in the following , we simplify the analysis by assuming random instantaneous projective measurements of individual particles in the system @xcite .
furthermore , we mainly focus on `` strong instability '' , which we define as the case when fewer than 10 measurements lead to @xmath17 independent of @xmath18 .
qualitatively , measurements can lead to both narrowing and broadening of @xmath4 .
the broadening effect of a single measurement is due to the off - diagonal elements of the projection operator describing the measurement in the basis of the total - energy eigenstates .
the narrowing effect originates from correlations between the total energy of the system and the measurement outcomes .
indeed , broad probability distributions @xmath4 can be considered as a mixture of microcanonical ( or canonical ) ensembles corresponding to different temperatures @xmath19 , which , in turn , imply different probability distributions of local variables .
when a given measurement outcome is more likely for one represented temperature than for another , the post - measurement distribution will be narrower than the initial one @xcite . in terms of the energy scales , the increase of the variance @xmath6 due to the off - diagonal elements of a local projection operator should normally be of the order of @xmath20 , where @xmath21 is an appropriately chosen single - particle energy .
the reason is that the quantum projections corresponding to the local measurements are one - particle or a few particle operators . at the same time
, the decrease of @xmath6 for broad @xmath4 can easily be of the order @xmath22 , i.e. , much larger .
since we focus primarily on broad @xmath4 , we neglect the broadening effect of the measurements unless explicitly specified otherwise .
we further limit our derivations to @xmath4 satisfying inequality @xmath23 where @xmath24 is the width of the energy distribution corresponding to the canonical ensemble with the same average energy as that of @xmath4 . for an ensemble with narrow energy distribution @xmath4 , such as the canonical ensemble
, the above - mentioned broadening effect also leads to heating , defined as the drift of the average energy @xmath7 in the direction of larger entropy @xmath25 ( larger value of @xmath26 $ ] ) .
this means that @xmath7 increases for positive temperatures and decreases for negative ones .
such a behaviour is consistent with the second law of thermodynamics , because local measurements of individual particles can be viewed as small nonadiabatic perturbations of the system . for positive temperatures @xmath9 ,
the increase of @xmath7 due to one measurement is of the order of the one - particle energy @xmath21 .
a broad ensemble can be considered as a mixture of canonical ensembles with different temperatures .
the heating for each of the contributing canonical ensembles occurs as explained above .
therefore , the overall heating effect for a broad ensemble corresponds to the combination of the heating effects for the individual canonical ensembles . in a system with a finite hilbert space ,
the asymptotic shape of @xmath4 corresponds to the canonical ensemble at infinite temperature , which is , in turn , proportional to the density of states @xmath2 of the system .
our definition of @xmath4 implies averaging over energy bins whose width @xmath27 satisfies the following inequalities : @xmath28 in the case of negative temperatures , @xmath29 should be used instead . the left inequality in eq .
together with the restriction @xmath30 and the inequality allows us to neglect heating described in the preceding part . the right inequality in eq .
assures that the energy eigenstates within each bin approximately correspond to the same density matrices of small subsystems within the entire system considered
. the probability distribution @xmath16 is defined in terms of the above energy bins as follows @xmath31 where the sum is taken over all energy eigenstates within the given bin and @xmath32 are the diagonal elements of the density matrix @xmath33 represented in the basis of the total - energy eigenstates .
now we consider a lattice of @xmath18 spins-1/2 and examine how multiple random local measurements affect @xmath4 .
we implement an individual local measurement by selecting a random spin at a random time and then measuring its projection on a random axis .
the measurements are assumed to occur very rarely with constant average rate per spin @xmath34 ( @xmath35 is much longer than the characteristic time of microscopic dynamics ) .
we label measurements by index @xmath12 .
each measurement is characterised by the parameters @xmath36 , where @xmath37 labels the lattice site of the measured spin and @xmath38 are the polar and azimuthal spherical angles indicating the orientation of the spin after the measurement .
the projection operator @xmath39 , which represents the @xmath12th measurement , is defined as @xmath40 where @xmath41 is the unit matrix acting on the hilbert space associated with the spin at lattice site @xmath42 and @xmath43 is the quantum state of a spin polarised into the direction given by the spherical angles @xmath44 .
the operator @xmath45 acts on the hilbert space associated with the spin at lattice site @xmath37 .
the operator @xmath39 satisfies the relations @xmath46 and @xmath47 .
the operator @xmath39 is related to the operator @xmath48 of the @xmath37th spin projection in the direction @xmath44 @xcite as @xmath49 where @xmath50 .
we denote the density matrix of the total system after @xmath12 measurements as @xmath33 and assume that the initial density matrix @xmath51 is diagonal in the basis of the energy eigenstates ( see appendix [ app_derivation_5 ] ) . the transformation from @xmath52 to @xmath53},\ ] ] where @xmath54 is the time of the @xmath12th measurement .
the corresponding probability distribution of the total energy after @xmath12 measurements @xmath16 can be expressed as @xmath55_{\text{diag}}\!(e)\ g_0(e),\nonumber\end{aligned}\ ] ] where @xmath56 is a normalisation factor , and @xmath57_{\text{diag}}\!(e)$ ] denotes the diagonal elements of the operator in the energy basis averaged over suitably chosen energy bins introduced in sec .
[ sec_coarse ] .
the derivation of eq . is given in appendix [ app_derivation_5 ] . since the measured spins are typically far away from each other , the effect of individual measurements in eq . normally factorises , which leads to @xmath58_{\text{diag}}\!(e)\ g_{n-1}(e),\ ] ] where @xmath59 is a normalisation factor .
the derivation of eq . is given in appendix [ app_derivation_6 ] .
below we consider the case when two measured spins are accidentally close to each other such that the corresponding effect does not factorise .
the stability measure averaged over all possible outcomes of @xmath12 measurements reads ( see appendix [ app_anexp ] ) @xmath60_{\text{diag}}\!(e ) - 1\bigg|g_0(e)de},\ ] ] where the bar denotes the result of averaging .
let us now turn to the example of noninteracting spins in magnetic field @xmath61 with hamiltonian @xmath62 where @xmath63 , @xmath64 , and @xmath65 are the spin operators on the lattice site . in this case , the outcome of a single spin measurement normally correlates with the total energy of the system and , therefore , leads to a significant narrowing of @xmath4 governed by eq . .
a calculation based on eq . gives ( see appendix [ app_derivation_9 ] ) @xmath66_{\text{diag}}\!(e ) = \frac{1}{2}-\cos(\vartheta_{n})\frac{e}{e_{\max}-e_{\min}},\ ] ] where @xmath67 and @xmath68 .
( solid red lines ) governed by eq . : @xmath69 $ ] ; @xmath70_{\text{diag}}\!(e)\ g_0(e)$ ] ; and @xmath71_{\text{diag}}\!(e)\ g_1(e)$ ] . here
@xmath72_{\text{diag}}\!(e)$ ] and @xmath73_{\text{diag}}\!(e)$ ] ( dashed blue lines ) correspond to two single - spin measurements with respective outcomes and @xmath74 substituted in eq .
.,scaledwidth=48.0% ] the action of transformation consists of `` cutting '' @xmath75 by function @xmath76_{\text{diag}}\!(e)$ ] and then renormalising the result .
this `` cutting '' normally makes @xmath16 narrower than @xmath75 .
the outcome of the next measurement can , in principle , lead to the opposite effect , but it is more probable that it will lead to further narrowing , because the probability of subsequent measurement outcomes is determined by the narrower @xmath16 . after many iterations , the drastic narrowing of @xmath4 becomes overwhelmingly probable .
figure [ fig_cut ] schematically illustrates the effect of a sequence of transformations applied to the initial two - peak distribution @xmath77,\ ] ] where @xmath78 is a dirac @xmath79 function . in this case ,
one of the two peaks dominates @xmath16 for @xmath80 .
figure [ fig_stab ] presents computed @xmath81 for the above @xmath15 . in each case
, we obtain that , after nine measurements , independent of the number of spins in the system , which implies `` strong instability '' . in appendix
[ app_anexp ] , we obtained the analytical approximation @xmath82 where @xmath83 with @xmath84_{\text{diag}}\!(e)/de| } \sim 1/(e_{\max}-e_{\min})$ ] . this approximation is illustrated in fig .
[ fig_stab ] . as a function of the number of measurements @xmath12 for a two - peak initial distribution @xmath16 .
points represent exact numerically computed results as explained in appendix [ app_fig ] .
lines correspond to the approximated expression with @xmath85 .
different colours represent different pairs of values for ( @xmath86,@xmath87 ) : blue ( circles ) ( @xmath88,@xmath89 ) , yellow ( squares ) ( @xmath88,@xmath90 ) , and green ( rhombi ) ( @xmath88,@xmath91 ) in units where @xmath92 and .,scaledwidth=48.0% ] let us now consider the initial gaussian distribution @xmath93}\ ] ] defined by parameters @xmath94 and @xmath95 , where .
after @xmath12 measurements , @xmath16 remains approximately gaussian with the width @xmath96 following the relation @xmath97 which is derived in appendix [ app_gauss ] .
this relation leads to when @xmath98 , which corresponds to the number of measurements @xmath99 .
according to our criterion , the border case for the ensemble stability corresponds to @xmath100 , which implies that the ensemble is unstable for @xmath101{n_s}$ ] .
an ensemble with @xmath102{n_s}$ ] may still become narrower due to measurements , but the criterion calls it `` stable '' , because the narrowing is relatively slow .
we finally note that , for @xmath103 , where is the characteristic single - spin energy , the decrease of the variance as a result of one measurement is @xmath104 .
this is of the same order of magnitude as the earlier mentioned increase of @xmath6 associated with the broadening effect of @xmath4 caused by the off - diagonal elements of the projection operator .
therefore , it is reasonable to expect that , for some @xmath103 , the narrowing effect of measurements would compensate the broadening effect , and hence such an ensemble is absolutely stable with respect to measurements .
remarkably , this @xmath95 is of the order of the width of the canonical ensemble for @xmath105 @xcite . given that the system is measured on average once per time @xmath106 , the above estimates for @xmath107 and for @xmath108 imply that the characteristic time to gain @xmath109 for a broad ensemble with the initial variance @xmath110 is @xmath111 for a macroscopic system , @xmath112 is , therefore , extremely short unless @xmath113 .
now we turn to the hamiltonian of nearest - neighbour interaction : @xmath114 where @xmath115 , @xmath116 , and @xmath117 are the coupling constants .
in contrast to the previous case , the outcome of a single - spin measurement here is not correlated with the total energy of the system , i.e. , @xmath76_{\text{diag}}\!(e)=const.$ ] , and , hence , does not induce narrowing of @xmath4 .
at least two accidental measurements sufficiently close in space and time are required for this .
let us consider two such measurements @xmath12 and @xmath118 at times @xmath119 .
the same treatment that led to eq . now gives ( see appendix [ app_derivation_13 ] ) @xmath120_{\text{diag}}\!(e)\ g_{n-2}(e),\ ] ] where @xmath121
. substituting eq . , we obtain in appendix [ app_derivation_13 ] @xmath122_{\text{diag}}\!(e)\\ & & = \frac{1}{4}+\frac{1}{2}\big[\left\{s_{n-1}(t_{n-1}),s_{n}(t_{n } ) \right\}\big]_{\text{diag}}\!(e)\nonumber\\ & & + \big[s_{n-1}(t_{n-1})s_{n}(t_{n})s_{n-1}(t_{n-1})\big]_{\text{diag}}\!(e),\nonumber\end{aligned}\ ] ] where @xmath123 is the anticommutator . once the spin orientations obtained in the and the @xmath12th measurements are specified and the energy @xmath1 is fixed , the last two terms on the right - hand side of eq .
can be considered equilibrium spin correlation functions for the microcanonical ensemble corresponding to energy @xmath1 , where @xmath124 represents the microcanonical average .
this definition is to be contrasted with the correlation functions for spin fluctuations @xmath125 .
the difference is essential in the presence of magnetic order . ] . for macroscopic systems ,
these correlation functions equal , in turn , the correlation functions for the canonical ensemble with temperature @xmath19 . to give an example ,
let us assume that the outcome of the first measurement is @xmath126 , @xmath127 ( spin 1 points into the positive @xmath128-direction ) and the outcome of the second measurement is @xmath129 , @xmath130 ( spin @xmath131 points into the positive @xmath132 direction ) .
for the two - spin term , we then obtain @xmath133_{\text{diag}}\!(e ) \!=\!\left\langle\ { s_{x}(\vec{r}_2,t_{2}),s_{z}(\vec{r}_{1},t_{1})\}\right\rangle_{t(e)},\ ] ] where @xmath134 is the position of the @xmath12th measured spin .
the above spin correlation functions depend on the time delay and the distance between the two measurements .
therefore , the cutting function in eq .
is strongly influenced by the presence of long - range magnetic order and integrals of motion .
we characterise the overall behaviour of the spin correlation functions by correlation time @xmath135 and correlation length @xmath136 .
if a system has constants of motion associated with spin polarisations , this corresponds to an infinite @xmath135 .
likewise , if a system exhibits long - range magnetic ordering in a range of energies @xmath1 , this means that for this range of energies @xmath136 is infinite .
[ it is worth noting that the correlation functions entering eq
. differ from the conventional correlation functions of spin fluctuations : for the latter the product of single - spin expectation values is subtracted . ] _ ordered phase : _ if a significant part of a broad @xmath15 corresponds to temperatures within the magnetically ordered phase , then the characteristic time @xmath112 to gain @xmath137 can be estimated by expression .
the reason is that , in the magnetically ordered phase , @xmath136 is infinite and , therefore , each measurement correlates with all previous ones in the sense that , for all subsequent measurements @xmath138 , @xmath139_{\text{diag}}\!(e)\neq0 $ ] .
the overall situation resembles the case of hamiltonian , with the external magnetic field replaced by the local field created by the ordered neighbours of each spin .
_ nonordered phase : _ in the nonordered ( paramagnetic ) phase , @xmath136 is of the order of the nearest - neighbour distance ( except for the energy range in the proximity to the magnetic phase transitions ) .
we make the assumption justified by the final result that @xmath140 for all energies @xmath1 , where @xmath135 is the correlation time introduced above .
therefore , we set the time delay entering the cutting function in eq . effectively to zero .
since @xmath141 , the outcome of the measurement addressing a nearest - neighbour of a previously measured spin is correlated with the total energy of the system .
the probability that the measurement does not address a nearest - neighbour of a previously measured spin is @xmath142 for @xmath143 , where @xmath144 is the number of nearest neighbours .
the probability that , among @xmath12 measurements , there is no nearest - neighbour pair measured , is @xmath145 for @xmath146 . using the relation @xmath147 , we finally obtain the probability that , after time @xmath148 , no pair of nearest neighbours is measured , @xmath149 accordingly , @xmath150 is the probability to measure at least one pair of nearest neighbours . therefore ,
for the paramagnetic phase and @xmath151 , we obtain @xmath152 is very large , @xmath112 is certainly much smaller than any reasonable correlation time @xmath153 .
this is a consistency check of our initial assumption @xmath154 . ] .
if only single - spin measurements were allowed , then the latter ensemble would be called stable according to our criterion .
however , the criterion admits any local measurements , including , for example , the measurements of the total spin of two neighbouring spins . for such measurements ,
the estimate remains valid , thus rendering the ensemble unstable .
we additionally note here that applying an external magnetic field to the paramagnetic phase would drastically shorten @xmath112 , because , in this case , single - spin measurements described by eq . would cause a much faster ensemble narrowing . beyond the hamiltonians and
, we expect that , for a broad class of systems with short - range interactions , local measurements whose possible outcomes correlate with the total energy of the system have a narrowing effect comparable to that of single - spin measurements for hamiltonian . in the general case ,
the analogue of the cutting function @xmath76_{\text{diag}}\!(e)$ ] is much more difficult to calculate directly but , otherwise , can be reasonably expected to be nonlinear with the energy - dependent slope of the order of @xmath155 .
the latter implies that the estimate for the ensemble - narrowing time and the argument for the proximity of the canonical ensemble to the absolute stability threshold remain valid .
is shown after each even - numbered measurement as indicated in the legend .
the system size is @xmath156.,scaledwidth=48.0% ]
in order to substantiate the conjectures made in the previous section , we numerically investigate a periodic chain of 24 governed by the hamiltonian with , , and @xmath157 .
the spins are measured in pairs : an odd - numbered measurement is done on a randomly selected spin , while one of its nearest neighbours is chosen for the next even - numbered measurement .
the time delays @xmath158 are chosen randomly from the interval @xmath159 $ ] .
the numerical techniques used for the calculations are described in appendix [ app_num ] . in fig .
[ fig_bins ] , we show a typical evolution of @xmath4 starting with a two - peak @xmath15 corresponding to the superposition of two quantum states representing canonical ensembles with temperatures @xmath160 and @xmath161 . in this example , one peak of @xmath4 becomes significantly suppressed , which represents the narrowing sketched in fig .
[ fig_cut ] .
adapting estimate to the present case of spin - pair measurements , we substitute @xmath162 , pair measurement time @xmath163 , pair energy @xmath164 , and @xmath165 ( see fig . [ fig_bins ] ) to obtain from eq .
@xmath166 corresponding to @xmath167 and within a factor of two consistent with the characteristic value @xmath168 extracted from fig . [ fig_bins ] .
we further tested the finite - size scaling of heating in small spin systems . as expected , the individual peaks in fig .
[ fig_bins ] also exhibit significant finite - size effects , namely peak broadening and the drift of the peak maxima associated with heating , which prevent us from computing @xmath169 representative of the macroscopic limit .
however , as we show below , the finite - size scaling of the heating is consistent with our expectation that they can be neglected in the macroscopic limit . in order to quantify the heating for the small system sizes available in numerical investigations , we calculate the deviation of the average energy @xmath7 from its initial value @xmath170 for a one - peak @xmath4 corresponding to the canonical ensemble with the temperature @xmath171 . in fig .
[ fig_en_drift ] , we show the evolution of @xmath172 as a function of @xmath12 . in this case , we randomly measure spins individually ( i.e. , not in pairs ) after time delays chosen randomly from the interval @xmath159 $ ] .
the results indicate that the heating becomes relatively weaker with increasing system sizes , which is in agreement with the general expectations from our analytical considerations .
is shown for the interacting spin system introduced in sec .
[ sec_numerical ] .
the energy distribution corresponds to the gibbs distribution with the temperature @xmath171 .
different system sizes were chosen as indicated in the legend .
the plotted values have been averaged over many iterations .
the points are connected by lines in order to guide the eye.,scaledwidth=48.0% ] for the numerical investigation of interacting spin systems , we use the approach established in refs .
this approach is based on the direct integration of the schrdinger equation and , hence , does not involve the diagonalisation of the hamiltonian matrix . for the numerical study , we employ the property of quantum typicality , which means that , for quantum systems with very large hilbert spaces , a single wave function representing a certain ensemble , for example , the ensemble given by @xmath4 , leads to the same average values of interest as the overall ensemble , cf .
eq . . therefore , we study the stability of energy distributions by considering randomly chosen wave functions .
we prepare the two - peak @xmath15 shown in fig . 3 by creating a quantum superposition of two pure states each corresponding to a canonical ensemble , one with temperature @xmath160 and the other with temperature @xmath161 .
in order to prepare the state at a temperature @xmath308 ( or @xmath309 ) , we use the method of imaginary - time evolution .
first , we randomly select a wave function corresponding to the infinite - temperature limit @xcite .
second , we use the relation @xmath310 , where @xmath311 .
for the imaginary time evolution , we use the same technique as for the one for the real - time evolution introduced in refs .
@xcite and also explained below .
for the real - time evolution , we use the fourth - order runge - kutta method @xcite @xmath312 where @xmath313 , @xmath314 , @xmath315 , and @xmath316 . for the schrdinger equation
, this method corresponds to the fourth - order taylor expansion of the time - evolution operator .
it was empirically shown in ref .
@xcite that this method is indeed accurate for the time intervals of interest in the present calculations . for the calculation of the energy distribution @xmath16
after the @xmath12th measurement at time @xmath237 , we perform the following fourier transform @xcite @xmath319 in order to do this , we first generate a discrete time series by time evolving the wave function immediately after each measurement . in our calculations ,
the length of the time steps is 0.025 and the number of these time steps is 2048 . after generating the time series ,
we multiply it by the kaiser - bessel window function in order to improve the energy resolution .
the kaiser - bessel window function is defined as @xmath320 where @xmath321 is the zeroth - order modified bessel function of the first kind , @xmath322 is a non - negative real number which determines the shape of the window , and @xmath323 . in our calculations ,
we use @xmath324 .
finally , we fourier transform the resulting time series and normalise the distribution in order to obtain @xmath16 . 54ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ]
+ 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop _ _ ,
( , ) http://stacks.iop.org/1367-2630/12/i=5/a=055006 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.180601 [ * * , ( ) ]
link:\doibase 10.1103/physrevlett.105.260402 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1103/physrevlett.96.050403 [ * * , ( ) ] http://dx.doi.org/10.1038/nphys444 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.190403 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.050405 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.010405 [ * * , ( ) ] link:\doibase 10.1038/nature06838 [ * * , ( ) ] https://arxiv.org/abs/1509.06411 [ ] link:\doibase 10.1103/physreve.80.051130 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/133/3/10.1063/1.3455998 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/133/3/10.1063/1.3456000 [ * * , ( ) ] http://stacks.iop.org/1402-4896/2012/i=t151/a=014078 [ * * , ( ) ] link:\doibase 10.1007/bf01883491 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.532661 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0550-3213(00)00447-8 [ * * , ( ) ] link:\doibase 10.1021/jp9039989 [ * * , ( ) ] link:\doibase 10.1007/s00220 - 011 - 1205 - 1 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.89.270403 [ * * , ( ) ] link:\doibase 10.1103/physrevb.93.184426 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.75.715 [ * * , ( ) ] http://dx.doi.org/10.1007/bf01725541 [ * * , ( ) ] @noop _ _ ( ) http://stacks.iop.org/0953-8984/14/i=15/a=201 [ * * , ( ) ] link:\doibase 10.1103/physrevb.77.184301 [ * * , ( ) ] @noop _ _ ( ) @noop _ _ , ed .
( ) http://arxiv.org/abs/1506.01213 [ ] http://dx.doi.org/10.1038/nature06057 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.044103 [ * * , ( ) ] @noop * * , ( ) @noop _ _ ( ) link:\doibase 10.1103/physreva.88.063609 [ * * , ( ) ] link:\doibase 10.1103/physrevx.4.031019 [ * * , ( ) ] http://stacks.iop.org/1367-2630/15/i=12/a=123012 [ * * , ( ) ] http://dx.doi.org/10.1038/nature02276 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.076401 [ * * , ( ) ] link:\doibase 10.1126/science.1240516 [ * * , ( ) ] http://dx.doi.org/10.1038/nphys2232 [ * * , ( ) ] http://dx.doi.org/10.1038/nature04693 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.050401 [ * * , ( ) ] http://stacks.iop.org/0953-8984/19/i=8/a=083202 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.070404 [ * * , ( ) ] link:\doibase 10.1103/physreve.62.4365 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.130403 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.120601 [ * * , ( ) ]
\(1 ) according to our stability criterion , ensembles with broad @xmath4 are not stable . in particular
, the qmc ensemble @xcite is not stable with respect to local measurements , because it implies @xmath4 having the form of two widely separated peaks corresponding to @xmath173 and @xmath174 , respectively @xcite .
\(2 ) our stability analysis can also be applied to unconventional quantum ensembles in experiments with nonmacroscopic isolated quantum clusters @xcite .
adapting the obtained results to such systems implies that the above experiments should avoid ( i ) external magnetic fields , ( ii ) long - range order , and ( iii ) local constants of motion .
in conclusion , we have shown that even relatively rare local measurements impose strict constraints on quantum statistical ensembles .
we introduced a stability criterion , according to which quantum statistical ensembles characterised by a total energy distribution significantly broader than that of a canonical ensemble are unstable .
this result justifies the use of statistical ensembles with narrow @xmath4 for equilibrium description of macroscopic systems .
the distribution @xmath4 can be approximated by a gaussian function @xmath180 , where @xmath7 is defined by the condition @xmath181 with @xmath182 .
expanding the exponent in eq . at @xmath7 up to the second order , we obtain @xmath183 where @xmath184 is the total specific heat of the system , @xmath185 , and @xmath186 is the inverse function of @xmath187 defined above . typically , @xmath188 .
therefore , @xmath189 , which means that @xmath4 is sharply peaked .
we denote the initial density matrix of the macroscopic system as @xmath51 .
we choose @xmath51 to be diagonal in the basis of the energy eigenstates @xmath190 , @xmath191 by doing this , we assume that the off - diagonal elements of @xmath51 , even if initially present , would have a negligible effect on the subsequent evolution of @xmath4 due to the rapid dephasing between states having macroscopically different values of energy . the transformation from @xmath52 to @xmath33 given in eq .
can be iterated to obtain the transformation from @xmath51 to @xmath33 , @xmath192 where @xmath193 , @xmath56 is a normalisation factor , and @xmath194 we are interested in the total - energy distribution and , therefore , focus on @xmath32 , i.e. , the diagonal elements of the density matrix in the total - energy basis . substituting eq .
and eq . into eq .
, we obtain after a transformation @xmath195 using the identity , we further rewrite eq . as
@xmath196 now we introduce the coarse graining of the energy axis , which means that we divide the energy axis into bins of width @xmath197 introduced above . substituting eq . into eq . , we obtain @xmath198 where @xmath199 [ without @xmath200 denotes a sum which is not restricted to the bin .
now we show that the first term in eq .
makes the main contribution to @xmath16 , while the last two terms in eq .
mainly compensate each other . to show this , we split each unrestricted sum into two sums , where one sum extends over the energy eigenstates within the bin , while the other one extends over the energy eigenstates outside the bin .
the latter sum is to be denoted as @xmath201 .
we also use the relation .
hence , we obtain for the last two terms in eq . @xmath202 now the two terms in eq . where both sums are restricted to bin@xmath203 cancel each other .
this can be readily seen after exchanging the summation indices @xmath42 and @xmath138 in one of these two terms .
the remaining terms in eq . can be further rewritten as @xmath204 the last two terms of eq .
contain off - diagonal elements @xmath205 corresponding to transitions between energy bins , because @xmath206 and @xmath207 lie in different bins .
let us denote the characteristic energy range @xmath208 of the off - diagonal elements as @xmath209 .
this range is limited by the condition @xmath210 therefore , only small energy intervals of length @xmath209 near the boundaries between the bins contribute to the sums . as shown in fig .
[ fig_els ] , we label these energy intervals as @xmath211 , @xmath212 , @xmath213 and @xmath214 . with such notations , eq .
can be rewritten as @xmath215 now we show that the last four terms in eq .
can be neglected in comparison with the first one , provided @xmath15 does not change too fast .
specifically , we impose the condition , which together with the inequalities yields @xmath216 where @xmath217 is the width of the energy distribution corresponding to the canonical ensemble with the same initial average energy as that of @xmath15 .
the condition in eq . must be satisfied within the energy interval , where @xmath15 is large enough to make a non - negligible contribution to the normalisation integral @xmath218 .
in the first term of eq . , we expect that , even if @xmath219 and @xmath220 fluctuate with respect to their bin - averaged values , they do it in an uncorrelated way . according to eq . , the bin - averaged value of @xmath219 is @xmath221 , where @xmath222 is the number of states within the bin .
we define the bin average of @xmath220 as @xmath223_{\text{diag}}\!(e)\equiv\frac{1}{n_\text{bin}(e)}\sum_k^{\text{bin}(e)}\langle e_k|{\cal o}^\dagger{\cal o}|e_k\rangle.\ ] ] given the right inequalities in eqs . and , both bin averages change very weakly over the bin size @xmath197 .
therefore , we can approximate the entire first term in eq . as @xmath224_{\text{diag}}\!(e)\ g_0(e).\ ]
] each of the remaining four terms in eq . has comparable values .
let us estimate the first of them .
we use the following inequality @xmath225 where , as before , the second sum extends over all energy eigenstates of the system . employing this inequality together with the assumptions used for deriving eq . , we obtain @xmath226_{\text{diag}}\!(e ) \nonumber\\ & & \approx \frac{1}{b } \left[{\cal o}^\dagger{\cal o}\right]_{\text{diag}}\!(e)\frac{\delta_{\cal o}}{\delta_\text{e}}g_0(e ) .
\nonumber\end{aligned}\ ] ] since @xmath227 , eqs . , and imply that @xmath228_{\text{diag}}\!(e)\ g_0(e),\ ] ] which is the same as eq . .
let us now consider the cutting function in eq . or ,
equivalently , in eq .
@xmath229_{\text{diag}}\!(e)\\ & & = \!\!\big[{\cal p}^{\dagger}_1e^{i{\cal h}(t_{2}-t_{1})}{\cal p}^{\dagger}_{2}\!\cdots\!{\cal p}^{\dagger}_{n}{\cal p}_{n}\!\cdots\!{\cal p}_{2}e^{-i{\cal h}(t_{2}-t_{1})}{\cal p}_1 \big]_{\text{diag}}\!(e)\nonumber\\ & & = \!\!\big[{\cal p}^{\dagger}_1(t_1){\cal p}^{\dagger}_{2}(t_2)\!\cdots\!{\cal p}^{\dagger}_{n}(t_n){\cal p}_{n}(t_n)\!\cdots\!{\cal p}_{2}(t_2){\cal p}_1(t_1 ) \big]_{\text{diag}}\!(e ) , \nonumber\end{aligned}\ ] ] where , in the last expression , we used , cf .
given that @xmath30 , it is rather unlikely that among the measured spins @xmath230 , @xmath231 , ... , @xmath37 there are two spins which are close to each other .
therefore , we assume the corresponding measurement operators @xmath232 to commute . rearranging the operators in the above expression and using @xmath233
, we obtain @xmath234_{\text{diag}}\!(e)=\big[{\cal p}_{n}(t_n)\cdots{\cal p}_{2}(t_2){\cal p}_1(t_1 ) \big]_{\text{diag}}\!(e).\ ] ] on the right - hand side of the above expression , there is the @xmath12-particle correlation function for the microcanonical ensemble . for systems with short - range interaction , we expect that correlation functions of mutually independent quantities factorise , which leads to @xmath235_{\text{diag}}\!(e)\\ & & = \big[{\cal p}_{n}(t_n)\big]_{\text{diag}}\!(e)\cdots\big[{\cal p}_{2}(t_2)\big]_{\text{diag}}\!(e)\big[{\cal p}_1(t_1 ) \big]_{\text{diag}}\!(e).\nonumber\end{aligned}\ ] ] this allow us to treat measurements iteratively .
since @xmath4 does not change with time between the measurements , the result does not depend on the particular values of @xmath236 , ... ,
@xmath237 and , according to eq . , . therefore
, we obtain @xmath238_{\text{diag}}\!(e)\ g_{n-1}(e),\ ] ] which is the same as eq . . for macroscopic systems , the above analysis does not depend on the particular value of @xmath35
however , for small clusters , eq . is valid as long as the delay between measurements @xmath35 is much longer than the characteristic time scales of the system s microscopic dynamics .
these characteristic time scales are , for example , of the order of @xmath239 for hamiltonian or of the order of for hamiltonian . there may be anomalous situations , when the expression for the measurement of a pair of neighbouring spins can not be approximated by the result of two successive applications of eq .
no matter how long the delay between the two measurements is .
an example is the ising model defined by @xmath240 in hamiltonian which has anomalously many local integrals of motion ( @xmath128 components of spins ) .
the hamiltonian @xmath241 with @xmath242 for spins in a magnetic field is diagonal in the product basis of individual spin states @xmath243 and @xmath244 .
therefore , we obtain for the expectation value of the @xmath128 polarisation of spin at lattice site @xmath37 , @xmath245 with @xmath246 , this yields @xmath247_{\text{diag}}\!({e})=-\cos(\vartheta_n)\frac{e}{e_{\max}-e_{\min}},\ ] ] which , taking into account the relation @xmath248 , results in eq . .
if two measurements @xmath118 and @xmath12 accidentally occur close in space and time , the corresponding projection operators @xmath299 and @xmath300 normally do not commute , and the effect of the two measurements does not factorise .
assuming that all other measured spins are far away from @xmath301 and @xmath37 , we obtain for the term describing the effect of the two measurements entering the expression in eq .
@xmath302_{\text{diag}}\!(e)$ ] , where @xmath121 . the expression for the modification of @xmath4 analogous to eq
. reads @xmath303_{\text{diag}}\!(e)\ g_{n-2}(e).\ ] ] for the function @xmath302_{\text{diag}}\!(e)$ ] , we obtain @xmath304_{\text{diag}}\!(e)\\ & & = \left[{\cal p}^\dagger_{n-1}(t_{n-1}){\cal p}^\dagger_{n}(t_{n}){\cal p}_{n}(t_{n}){\cal p}_{n-1}(t_{n-1})\right]_{\text{diag}}\!(e ) \nonumber\\ & & = \big[{\cal p}_{n-1}(t_{n-1}){\cal p}_{n}(t_{n}){\cal p}_{n-1}(t_{n-1})\big]_{\text{diag}}\!(e).\nonumber\end{aligned}\ ] ] substituting eq
. leads to eq . , which we rewrite here _
{ \text{diag}}\!(e)$ ] , we refer the reader to the discussion following eq . .
] , @xmath305_{\text{diag}}\!(e)\\ & & = \frac{1}{4}+\frac{1}{2}\big[\left\{s_{n-1}(t_{n-1}),s_{n}(t_{n } ) \right\}\big]_{\text{diag}}\!(e)\nonumber\\ & & + \big[s_{n-1}(t_{n-1})s_{n}(t_{n})s_{n-1}(t_{n-1})\big]_{\text{diag}}\!(e),\nonumber\end{aligned}\ ] ] where @xmath306 . when deriving eq .
, we assumed that the outcomes of single - spin measurements are not correlated with the total energy as explained in the main text .
this is formally expressed as @xmath307_{\text{diag}}\!(e)=0 $ ] .
we further note that the three - spin term in eq . should typically be significantly smaller than the two - spin term .
substituting @xmath249 $ ] into eq . , we obtain @xmath250_{\text{diag}}\!(e_1 ) - 1\bigg|}\\ & & + \frac{1}{2}\overline{\bigg|\prod_{i=1}^n\frac{1}{b_i}\left [ { \cal p}_i
\right]_{\text{diag}}\!(e_2 ) - 1\bigg|},\nonumber\end{aligned}\ ] ] where the bar denotes the result of averaging .
the probability for obtaining the given measurement outcomes equals the normalisation factor @xmath251 .
given eq .
, the above expression only depends on the polar angles @xmath252 , ... , @xmath253 describing the measurement outcome .
this leads to @xmath254_{\text{diag}}\!(e_1 ) - 1\bigg|\sin(\vartheta_1)\cdots\sin(\vartheta_n)\ d\vartheta_1\cdots d\vartheta_n\right.\nonumber\\ & & \left.+\int_0^\pi\prod_{i=1}^n b_i\bigg|\prod_{i=1}^n\frac{1}{b_i}\left [ { \cal p}_i \right]_{\text{diag}}\!(e_2 ) - 1\bigg|\sin(\vartheta_1)\cdots\sin(\vartheta_n)\ d\vartheta_1\cdots d\vartheta_n\right]\\ & = & \frac{1}{2}\left[\int_0^\pi\bigg|\prod_{i=1}^n\left [ { \cal p}_i \right]_{\text{diag}}\!(e_1 ) -
\prod_{i=1}^n b_i\bigg|\sin(\vartheta_1)\cdots\sin(\vartheta_n)\ d\vartheta_1\cdots d\vartheta_n\right.\nonumber\\ & & \left.+\int_0^\pi\bigg|\prod_{i=1}^n\left [ { \cal p}_i \right]_{\text{diag}}\!(e_2 ) -
\prod_{i=1}^n b_i\bigg|\sin(\vartheta_1)\cdots\sin(\vartheta_n)\ d\vartheta_1\cdots d\vartheta_n\right].\label{eqn_here}\end{aligned}\ ] ] using eq .
, we further obtain @xmath255_{\text{diag}}\!(e)g_0(e)\ de\\ & = & \frac{1}{2}\prod_{i=1}^n \left(\frac{1}{2}-\cos(\vartheta_i)\frac{e_1}{e_{\max}-e_{\min}}\right)\nonumber\\ & & + \frac{1}{2}\prod_{i=1}^n \left(\frac{1}{2}-\cos(\vartheta_i)\frac{e_2}{e_{\max}-e_{\min}}\right).\nonumber\end{aligned}\ ] ] substituting the last expression into eq . , the two summands in eq .
are identical and this leads to @xmath256\nonumber\\ \sin(\vartheta_1)\cdots\sin(\vartheta_n)\ d\vartheta_1\cdots d\vartheta_n.\nonumber\end{aligned}\ ] ] integrating the above expression numerically , we obtain the results shown in fig .
[ fig_stab ] .
in this section , we derive the analytical approximation for @xmath81 used in fig . 2 .
from relation , we obtain @xmath257_{\text{diag}}\!(e)\ g_0(e).\ ] ] the substitution of this expression into eq
. leads to @xmath258_{\text{diag}}\!(e)-1\big| g_0(e ) de,\ ] ] which , after averaging , gives eq . .
for the initial distribution @xmath249 $ ] , eq
. leads to @xmath259 with some probabilities @xmath260 and @xmath261 ( @xmath262 ) .
we further note that for @xmath263 , either @xmath264 or @xmath265 , such that @xmath266 . in order to find an approximate expression for @xmath81 ,
let us first observe that @xmath267 ^ 2}+\overline{\delta g^2(n-1)},\ ] ] because @xmath268 .
therefore , we make the ansatz @xmath269 , where @xmath270 . since the individual summands @xmath271 become smaller as @xmath272 approaches 1 , we make the rough approximation @xmath273 , which , in the continuum limit for @xmath12 , leads to the differential equation @xmath274 $ ] , where @xmath107 is some constant . assuming that @xmath275
, we obtain @xmath276 we further adopt an approximation @xmath277 where @xmath84_{\text{diag}}\!(e)/de|}$ ] , and @xmath278 is a fitting parameter . the expression in eq .
can be justified by the following considerations .
the parameter @xmath107 must be equal to 0 when @xmath279 .
also , @xmath107 must remain invariant under a sign change of @xmath280 .
therefore , the lowest - order term allowed in an analytical expansion of @xmath107 around 0 is proportional to @xmath281 .
assuming that the value of @xmath107 is only controlled by @xmath282 and @xmath280 , we conclude that @xmath282 must also enter quadratically in order for @xmath107 to be dimensionless .
the above approximation can be further supported by a more detailed calculation of @xmath81 in which case the expression for @xmath169 in eq .
is to be averaged over all possible measurement outcomes . for this average
, @xmath169 must be weighed by the probability for a given set of @xmath12 measurement outcomes , which equals the normalisation coefficient @xmath284_{\text{diag}}\!(e)\ g_0(e)de$ ] .
let us now make a simplifying assumption that the spin measurements are only done along the @xmath128 direction .
consequently , there are two possible measurement outcomes : positive ( @xmath285 ) and negative ( @xmath286 ) . according to eq .
, the projection operator is thus @xmath287_{\text{diag}}\!(e)=\frac{1}{2}\pm\frac{e}{e_{\max}-e_{\min}}$ ] . after averaging over these two possibilities for each measurement ,
we obtain @xmath288 where @xmath289 is the binomial distribution and @xmath290 .
the value of @xmath81 is governed by the overlap as function of @xmath138 between the two binomial distributions @xmath291 and @xmath292 . with an increasing @xmath12 ,
this overlap decreases approximately exponentially , which is consistent with the asymptotic behaviour of @xmath81 that follows from eq . for large @xmath12 .
for a gaussian distribution @xmath293\ ] ] with the variance @xmath294 , the `` cutting '' by the linear function @xmath76_{\text{diag}}\!(e)$ ] can be expressed as @xmath295_{\text{diag}}\!(e)\ e^{-\frac{(e - e_\text{av})^2}{2w^2_{g , n-1}}}\\ = e^{\ln(\left[{\cal p}_n\right]_{\text{diag}}\!(e))-\frac{(e - e_\text{av})^2}{2w^2_{g , n-1}}}.\nonumber\end{aligned}\ ] ] it changes the width according to the relation @xmath296'_{\text{diag}}(e_\text{av})}{\left[{\cal p}_n\right]_{\text{diag}}(e_\text{av})}\right)^2,\ ] ] where @xmath76'_{\text{diag}}\!(e_\text{av})=\frac{d\left[{\cal p}_n\right]_{\text{diag}}\!(e_\text{av})}{de_\text{av}}$ ] . from eq .
, we obtain @xmath297}\approx \frac{1}{w_{g,0}^2}+u^2\ n,\ ] ] where @xmath298'_{\text{diag}}(e_\text{av})|}$ ] , and we use @xmath76_{\text{diag}}(e_\text{av})\sim1 $ ] . | we introduce a stability criterion for quantum statistical ensembles describing macroscopic systems .
an ensemble is called `` stable '' when a small number of local measurements can not significantly modify the probability distribution of the total energy of the system .
we apply this criterion to lattices of spins-1/2 , thereby showing that the canonical ensemble is nearly stable , whereas statistical ensembles with much broader energy distributions are not stable . in the context of the foundations of quantum statistical physics ,
this result justifies the use of statistical ensembles with narrow energy distributions such as canonical or microcanonical ensembles . |
in 1994 , shor s ground - breaking paper presented an efficient quantum algorithm for integer factoring @xcite .
many other very important results have followed , but the true potential of quantum computers is still unknown .
one step towards understanding it is to consider what are the differences between quantum and classical computers in a setting where we are given the input via a black - box oracle . because we can not know the inner workings of the oracle , it is easier to prove lower bounds in this model and
therefore in many cases it is easier to show a separation between the best quantum and classical algorithms .
ever since deutsch presented a simple oracle problem that can be solved on a quantum computer using fewer oracle queries than on any classical computer @xcite , scientists have tried to come up with more and more problems for which quantum computers outperform their classical counterparts ; some very artificial , some quite natural .
bernstein and vazirani gave the first example of an oracle problem which can be solved in polynomial time on a quantum computer , but requires superpolynomial time on a classical computer @xcite .
shortly after that simon gave an example in which this separation is exponential @xcite .
yet it is unclear what are the best methods for the construction of efficient quantum algorithms .
while quantum fourier sampling is probably the most popular such method so far , many algorithms are also based on the concept of quantum walk . in particular , continuous - time quantum walks on graphs , introduced by farhi and gutmann @xcite , give rise to fast algorithms for nand tree evaluation @xcite and unstructured search @xcite .
continuous - time quantum walks are also known to be able to perform universal quantum computation @xcite , and they can solve some oracle problems exponentially faster than any classical algorithm @xcite . in this paper we introduce a new type of continuous - time quantum walk on graphs , the basis states of which are not vertices of the graph , but paths in it of a fixed length @xmath0 .
we define this new walk on an unweighted undirected graph @xmath1 as a regular continuous - time quantum walk ( the one introduced by farhi and gutmann ) on a more complex weighted undirected graph @xmath2 which is constructed from the original graph @xmath1 .
since , to the author of this paper , the discrete - time classical counterpart of this walk in some sense resembles the behavior of a ` snake ' of length @xmath0 which is placed on a graph along its edges and which makes random decisions in which direction to go next , we call this new walk the _ continuous - time quantum snake walk_. the main motivation for the continuous - time quantum snake walk is an extended version of the glued trees problem for which no efficient quantum algorithm yet is known .
childs et al . introduced the glued trees graph , consisting of two complete binary trees of the same height which are connected by a random cycle that alternates between the leaves of the trees @xcite . given
a glued trees graph via a black - box oracle and the label of the root of one tree , the glued trees problem is to determine the label of the root of the other tree .
while there is an efficient quantum algorithm based on the continuous - time quantum walk which solves this problem , no efficient quantum algorithm which finds a path connecting these two roots is known .
we hope that the continuous - time quantum snake walk might lead to a quantum algorithm efficiently finding such a path . in section [ sec : line ]
we analyze in detail the continuous - time quantum snake walk on the line .
even though most states stay localized throughout the quantum evolution , there are specific states which , under one reasonable assumption , move on the line as wave packets with momentum inversely proportional to the length of the snake .
this motion is similar to the regular continuous - time quantum walk on the line . in section [ sec
: gtg ] we discuss an algorithm based on the continuous - time quantum snake walk that solves the extended version of the glued trees problem mentioned above , that is , finds a path connecting the roots of the glued trees graph .
however , it is not clear whether this algorithm runs in polynomial time .
independently from this work , a recent paper by mcgettrick defines a discrete - time walk on the line similar to the continuous - time quantum snake walk on the line considered here @xcite .
in this section we define the main concept of this paper , the continuous - time quantum snake walk .
the main reason for considering this new type of quantum walk is its potential algorithmic applications . because of that we also define an extended version of the glued trees problem for which no efficient classical algorithm exists and no efficient quantum algorithm is known , but which might be solved efficiently using an algorithm based on the continuous - time quantum snake walk .
we will define the continuous - time quantum snake walk on one graph as the continuous - time quantum walk on another specific graph .
therefore , we define the continuous - time quantum walk first .
let @xmath3 be a weighted undirected graph , where @xmath4 is a set of vertices and @xmath5 is a weight function , which assigns a weight @xmath6 to every pair @xmath7 .
let @xmath8 be a hilbert space having an orthonormal basis @xmath9 called the _
standard basis _ , and let @xmath10 be a linear operator acting on @xmath8 such that @xmath11 for all @xmath12 . because @xmath13 is undirected with all weights being real , @xmath10 is hermitian . the _ continuous - time quantum walk _ on @xmath13
is defined as a quantum evolution in the space @xmath8 governed by the hamiltonian @xmath10 according to the schrdinger equation .
that is , if we fix an initial state of the walk to be @xmath14 , then we can look on the walk as a function which maps time @xmath15 to the state @xmath16 .
now suppose @xmath17 is an undirected unweighted graph , where @xmath18 is a set of vertices and @xmath19 is a set of edges .
let @xmath20 be the set of all paths in @xmath1 which have length @xmath0 .
here we assume that a path can visit a vertex multiple times .
let us call an element of @xmath20 a _ snake _ of length @xmath0 .
that is , a snake of length @xmath0 is a vector @xmath21 such that @xmath22 for all @xmath23 $ ] .
note that we assume all paths are directed ( @xmath24 and @xmath25 are not the same snake ) , even though the underlying graph @xmath1 is undirected . let @xmath26 be a snake of length @xmath0 .
we say that @xmath27 can move forward to a snake @xmath28 if there exists a vertex @xmath29 such that @xmath30 , and we write @xmath31 ( we think of @xmath32 as a predicate ) .
similarly , we say that @xmath27 can move backward to @xmath28 if there exists @xmath33 such that @xmath34 , and we write @xmath35 .
we also consider @xmath31 as a binary function on @xmath36 taking the value @xmath37 if and only if the predicate @xmath32 is true for a pair @xmath38 , and similarly for @xmath39 .
let @xmath40 , and let @xmath41 be a matrix whose rows and columns are labeled by the elements of @xmath20 and @xmath42 for all @xmath43 .
clearly @xmath31 if and only if @xmath44 , and therefore @xmath41 is a symmetric matrix .
we can look on @xmath41 as the adjacency matrix of a graph having @xmath20 as the set of vertices , and possibly some edges having weight @xmath45 instead of @xmath37 ( as a matter of fact , @xmath46 if and only if there are two adjacent vertices @xmath47 and @xmath48 such that @xmath49 and @xmath50 )
. let @xmath2 be a weighted graph with the set of vertices @xmath20 and the matrix of weights @xmath41 . the _ continuous - time quantum snake walk _ on the graph @xmath1
is defined as a continuous - time quantum walk on the weighted graph @xmath2 .
because we consider only continuous - time walks , we often refer to continuous - time quantum walks and continuous - time quantum snake walks , respectively , as regular quantum walks and quantum snake walks .
sometimes we use the word snake to refer , instead of an element of @xmath20 , to the content of a quantum register corresponding to the hilbert space @xmath51 . in some sense , we can think of the snake as being a path which can move form one position in the graph to another . in what sense the word snake
is used should be clear from the context . in order to provide an example of a quantum algorithm based on a quantum walk that can solve a certain black - box problem exponentially faster than any classical algorithm ,
childs et al .
introduced the glued trees graph @xcite .
the _ glued trees graph _ consists of two complete binary trees of height @xmath52 , which are connected by a random cycle that alternates between the leaves of the two trees .
an example of the graph is shown in figure [ fig : gluedtrees ] ( for @xmath53 ) .
let us also call @xmath52 and the roots of both trees , respectively , the height and the roots of the glued trees graph .
suppose we are given a glued trees graph via a black - box oracle ( given the label of a vertex as an input , the oracle outputs labels of all its neighbors ) and the label of one root @xmath54 .
the _ glued trees problem _ is to determine the label of the other root @xmath55 ( see figure [ fig : gluedtrees ] ) .
childs et al .
show that , if we start a continuous - time quantum walk at the root @xmath54 , the walk quickly ( in @xmath56 time ) traverses the graph and ends up in a superposition state which has a large overlap on the root @xmath55 .
we can implement continuous - time quantum walks on sparse graphs in the quantum circuit model efficiently ( see @xcite ) , therefore there is an efficient quantum query algorithm solving the glued trees problem .
it is also shown in @xcite that no classical query algorithm can solve this problem efficiently .
given a glued trees graph via a black - box oracle and the label of one of its two roots @xmath54 , the _ extended glued trees problem _ is to find a path in the graph connecting @xmath54 to the opposite root @xmath55 .
as far as i know , there is still no quantum algorithm known which solves the extended glued trees problem efficiently .
it is not even known if such an efficient algorithm exists ( * ? ? ?
* section 5.4 ) .
however , the hope is that the quantum snake walk might lead to an efficient algorithm .
as an example of a continuous - time quantum snake walk , let us consider the quantum snake walk on the line , that is , the graph @xmath17 , where @xmath57 is the set of vertices and @xmath58 is the set of edges .
this example is relatively simple compared to quantum snake walks on other graphs . nonetheless , understanding this walk
later helps us to analyze snake walks on more complex graphs .
it is convenient to think of the line also as the @xmath59-axis . for every snake @xmath60 on @xmath1 of length @xmath0 let @xmath61 be the start vertex of the snake and for @xmath23 $ ] let @xmath62 be such that @xmath63
this gives a one - to - one relation between the set of snakes @xmath20 and the set @xmath64 .
from now on let us consider any snake @xmath27 to be given as a pair @xmath65 , and let @xmath66 denote @xmath67 .
every snake on the line can move forward to two other snakes and move backward to two other snakes . if a snake @xmath68 moves forward , then the start vertex of the new snake is determined by @xmath69 .
its end vertex can either move in the positive or the negative direction of the @xmath59-axis . if a snake moves backward , the direction in which its start vertex moves is opposite to the direction of the first edge of the new snake .
in other words , for every @xmath70 we have @xmath71 this allows us to obtain the hamiltonian @xmath72 governing the quantum snake walk on the line .
after some derivations we get @xmath73 where @xmath74 and @xmath75 ( see ( * ? ? ?
* section 3.1 ) for details ) . for every @xmath76 we have @xmath77 , where @xmath78 is the dirac delta function ( see @xcite ) .
hence , for any value @xmath79 , if @xmath80 is an eigenvector of @xmath81 with the corresponding eigenvalue @xmath82 , then @xmath83 is an eigenvector of @xmath84 corresponding to the same eigenvalue @xmath82 .
let us focus on finding the eigenvalues and eigenvectors of @xmath81 for an arbitrary @xmath79 .
this task becomes much easier if we express @xmath81 in a different basis .
consider two pairs of orthonormal vectors : @xmath85 for @xmath86 $ ] let us define a unit vector @xmath87 as follows .
first , let @xmath88 . for @xmath89 $ ] , let @xmath90 written in binary using @xmath91 bits be @xmath92 .
then for @xmath89 $ ] we define @xmath93 as an example , for @xmath94 we have @xmath95 , @xmath96 , @xmath97 , @xmath98 , @xmath99 , @xmath100 , @xmath101 and @xmath102 . for any @xmath103 and @xmath104
the set @xmath105 is an orthonormal basis of @xmath106 , and ( * ? ? ?
* section 3.1.1 ) shows that @xmath81 expressed in this basis is @xmath107 from ( [ eq : hnkinhatbasis ] ) we see that @xmath108 whenever @xmath90 and @xmath109 have different greatest odd divisors , where we assume that the greatest odd divisor of @xmath110 is @xmath37
. therefore we can block diagonalize @xmath81 with respect to @xmath111 into @xmath112 blocks . to be more precise , let us define @xmath112 orthogonal projectors @xmath113 and [ piprojector ] @xmath114 for odd @xmath115 $ ] .
we have @xmath116 whenever @xmath117 .
therefore , in order to get the full eigenspectrum of @xmath81 , we consider the eigenvalues of each block separately . for odd @xmath115 $ ]
the block @xmath118 is basically twice the adjacency matrix of the line segment of length @xmath119 , whose eigenvalues and eigenvectors are well known ( for example , see @xcite )
. therefore the eigenvalues of this block are @xmath79-independent and the eigenvectors depend only on @xmath79 due to the fact that the basis @xmath111 is @xmath79-dependent . however , because this dependence is relatively simple , as shown in ( * ? ? ?
* section 3.1.3 ) , the following lemma holds . [ the : kindependent ] let @xmath120 and @xmath121 .
if @xmath122 , then @xmath123 , where @xmath124 .
the essence of lemma [ the : kindependent ] is that we can restrict our attention to the states in the space corresponding to the projector @xmath125 if we want to have an initial state which moves on the line further than distance @xmath126 , which for algorithmic interests seems to be a reasonable request .
the eigenvalues and eigenvectos of @xmath84 restricted to this subspace , namely @xmath127 , is determined by the eigenvalues and eigenvectors of the block @xmath128 .
let @xmath129 be a fixed set of @xmath130 orthonormal , @xmath79-independent vectors .
consider a linear @xmath79-dependent isometry @xmath131 we have @xmath132 , where @xmath133 which is the adjacency matrix of the graph given in figure [ fig : phigraph ] . .
] if @xmath134 , then @xmath135 and thus @xmath136 is the sum of two orthogonal operators @xmath137 and @xmath138 . the eigenspecturm of the former operator is @xmath45 times the eigenspectrum of the adjacency matrix of the line segment of length @xmath139 , and the eigenspecturm of the latter is trivial .
therefore , in the case when @xmath134 we can get all the eigenvalues and eigenvectors of @xmath136 : for every @xmath140 the unit vector @xmath141 is an eigenvector of @xmath136 with the corresponding eigenvalue @xmath142 , and @xmath143 is an eigenvector corresponding to the eigenvalue @xmath144 ( i.e. , @xmath145 ) .
unfortunately , in the general case we do not even know if there are closed form expressions for the eigenvalues of @xmath136
. however , the following lemma , which is proved in appendix [ app : pcondition ] , gives us some useful information about the eigenvalues and eigenvectors of @xmath136 for an arbitrary @xmath79 .
[ the : pcondition ] let us fix @xmath103 and @xmath146 .
the equation @xmath147 has @xmath130 distinct solutions in the interval @xmath148 , and , if @xmath149 is a solution of this equation , then @xmath150 is an ( unnormalized ) eigenvector of @xmath136 with the corresponding eigenvalue @xmath142 .
the eigenvalues of @xmath136 are all distinct for any value of @xmath79 , and they all depend on the value of @xmath79 .
we refer to ( [ eq : pcondition ] ) as the _
@xmath149-equation_. we do not know how to solve the @xmath149-equation , the solution of which would give us the full eigenspectrum of @xmath81 . despite that
, the @xmath149-equation allows us to obtain good approximations of eigenvalues and their derivatives . but let us first show that we can treat @xmath79-dependent eigenvalues of @xmath81 as infinitely differentiable functions of @xmath79 .
a complex - valued function is _ holomorphic _ if it is complex - differentiable in a neighborhood of every point in its domain .
holomorphic functions are known to be infinitely differentiable , and the class of holomorphic functions include polynomials , the exponential function , sine and cosine @xcite . the notion of holomorphic functions
can be generalized to vector - valued and operator - valued functions in an obvious way . from (
* chapter ii , 6.2 ) we have the following lemma [ lem : holo ] let @xmath151 be a finite complex euclidean space of dimension @xmath90 . consider a holomorphic operator - valued function @xmath152 which maps complex numbers to linear operators over @xmath151 such that @xmath153 is hermitian for all @xmath104 . then for @xmath104 there exist a family of orthonormal basis @xmath154\}$ ] of @xmath151 consisting of eigenvectors of @xmath153 and a family @xmath155\}$ ] consisting of eigenvalues of @xmath153 such that @xmath156 and @xmath157 are holomorphic functions of @xmath79 for all @xmath158 $ ] .
let @xmath156 be the @xmath109-th largest eigenvalue of @xmath136 .
because all the eigenvalues of @xmath136 are distinct and @xmath136 is holomorphic in @xmath79 ( which can be easily seen from ( [ eq : phik ] ) ) , according to lemma [ lem : holo ] , @xmath156 is a holomorphic function in @xmath79 for all @xmath159 $ ] . in order to give some intuition about the eigenspectrum of @xmath136 , in figure [ fig : eigenvalues ]
we show how the eigenvalues of @xmath136 and their derivatives depend on @xmath160 $ ] in the case when @xmath161 .
these plots are obtained via numerical computation .
( on the left ) and their derivatives ( on the right ) for @xmath161 . ] figure [ fig : eigenvalues ] suggests that the derivative of any eigenvalue is @xmath110 if and only if @xmath162 . the following theorem , which is proved in ( * ? ? ?
* section 3.2.1 ) using the @xmath149-equation , confirms this observation .
[ thm : lambdad ] for every @xmath159 $ ] the eigenvalue function @xmath156 is strictly decreasing in the interval @xmath148 and strictly increasing in the interval @xmath163 .
for the rest of the section we consider only the case when @xmath0 is even because it makes some calculations easier . also , out of all @xmath130 @xmath79-dependent eigenvalues of @xmath81 , let @xmath164 be the median one , that is , the @xmath165-th largest ; we consider only this particular eigenvalue .
the reasons for considering this particular eigenvalue are the fact that it can be well approximated and it seems to have the largest derivative , which most likely is good for a construction of states moving fast on the line .
note that the value of @xmath164 depends on @xmath0 .
let @xmath166 and let us use a prime to denote derivatives with respect to @xmath79 . in appendix [ app : hyplem ] we prove the following lemma .
[ lem : lambdaprime ] for every @xmath0 , @xmath167 is bounded between @xmath168 and @xmath169 and we have @xmath170 .
also , there exists @xmath171 such that for all @xmath172 we have @xmath173 if and only if @xmath174 .
the value of @xmath164 is bounded between @xmath175 and @xmath176 . as @xmath0 increases ,
the function @xmath177 and its derivatives @xmath178 and @xmath179 converge to the functions @xmath180 , @xmath181 and @xmath182 , respectively . in figure
[ fig : arctan ] we plot @xmath180 and its first two derivatives to get some intuition about the behaviour of this ` eigenvalue ' function .
( solid line ) and its first and second derivatives ( dashed and dotted lines , respectively ) . ] now that we have established some properties of the median eigenvalue @xmath164 , let us consider a quantum snake walk on the line having an initial state @xmath183 which is a uniform superposition of eigenvectors of @xmath84 corresponding to @xmath164 .
we will choose the phases of these eigenvectors in a particular manner so that @xmath183 seems to be highly localized .
then we will show that , under one reasonable assumption , for @xmath0 large enough the state @xmath183 evolves as a left - moving and a right - moving wave packet , each propagating with the same constant momentum .
but let us first give a precise definition of @xmath183 .
lemma [ the : pcondition ] implies that there exists a unique vector - valued function @xmath184 such that @xmath185 , @xmath186 and @xmath187 for all @xmath104 .
and lemma @xmath188 implies that @xmath189 is continuous in @xmath79 , and therefore so is @xmath190 .
hence , @xmath191 is an eigenvector of @xmath84 corresponding to the eigenvalue @xmath164 . for every @xmath192 define @xmath193 notice that @xmath194 .
the state @xmath183 obviously belongs to the subspace @xmath195\,\}\,)$ ] and remains in @xmath196 under the evolution of @xmath84 . because @xmath197 , where @xmath198 is the kronecker delta function , and @xmath199 for any @xmath200 , @xmath201 is an orthonormal basis of @xmath196 .
numerical results suggest that @xmath202 is highly localized ` around the value @xmath203 on the @xmath59-axis ' ( see appendix [ app : linenum ] ) .
for example , for @xmath204 , if we measure @xmath202 in the standard basis , with probability @xmath205 we obtain a snake starting at the position @xmath206 , with probability @xmath207 a snake starting at @xmath208 , with @xmath209 a snake starting at @xmath210 , and so on ( for even @xmath211 we never obtain a snake starting at position @xmath212 ) . also , the probability of obtaining a snake starting at the position at least @xmath0 units away from @xmath203 seems to decrease exponentially with @xmath0 , and the following hypothesis seems to hold .
[ hyp : local ] for every @xmath213 and @xmath70 let @xmath214 .
then for every polynomial @xmath215 we have @xmath216 basically , hypothesis [ hyp : local ] implies that for any @xmath217 , if @xmath218 is large , then for some @xmath219 $ ] the state @xmath220 has large overlap with the space of snakes starting at the position @xmath211 ; meanwhile , if @xmath221 is small for all @xmath219 $ ] , then the state @xmath220 has small overlap with the space of snakes starting at the position @xmath203 .
therefore , the inner products between @xmath220 and vectors of the basis @xmath201 allow us to specify where on the line the state @xmath220 is located .
let us analyze how the state @xmath183 evolves under the hamiltonian @xmath84 in the limit of an asymptotically large time of the evolution @xmath222 .
that is , for every @xmath223 let us consider @xmath224 for all @xmath225 such that @xmath226 , and let us think of @xmath227 as the momentum .
we have the following : @xmath228 let @xmath229 .
when @xmath222 is large , @xmath230 rapidly changes as @xmath231 changes .
this means that in an interval where @xmath231 changes the contribution from adjacent subintervals to the integral ( [ eq : xiz1xiz2int ] ) nearly cancels out @xcite .
therefore the most contribution to the integral comes from values of @xmath79 where the function @xmath231 is stationary , that is , where @xmath232 .
the method of stationary phase approximation makes this statement more rigorous , and we will use it to estimate the integral ( [ eq : xiz1xiz2int ] ) in the limit of asymptotically large @xmath222 . from the @xmath149-equation one can get that @xmath233 and @xmath234 , @xmath235 and @xmath236 . let us consider @xmath172 , where @xmath237 is the same as in lemma [ lem : lambdaprime ] .
lem : lambdaprime ] implies that the derivative @xmath167 is strictly increasing in the interval @xmath238 . for each @xmath239 let @xmath240 be the unique point in the interval @xmath238 such that @xmath241 .
notice that for @xmath227 such that @xmath242 the function @xmath231 has no stationary points .
the method of stationary phase approximation ( see @xcite ) applied to the integral ( [ eq : xiz1xiz2int ] ) gives us that , in the limit of asymptotically large @xmath222 , * if @xmath242 , then @xmath243 for every polynomial @xmath244 ; * if @xmath245 , then @xmath246{4\sqrt{3}(3n^2 + 4 ) } } \cdot \frac{1}{\sqrt[3]{t}};\ ] ] * if @xmath239 , then @xmath247 as we can see , if we measure @xmath248 in the basis @xmath249 , most likely we will obtain a state @xmath250 such that @xmath251 is close to @xmath110 . according to lemma [ lem : lambdaprime ] , @xmath251 is close to @xmath110 when @xmath240 is close to @xmath252 and @xmath253 ( see figure [ fig : arctan ] ) and therefore @xmath227 , the momentum , is close to @xmath254 . hence
, if indeed the state @xmath250 is localized around the position @xmath255 on the @xmath59-axis , as hypothesis [ hyp : local ] asserts , then we can see that for @xmath172 the state @xmath183 evolves as a left - moving and a right - moving wave packet each propagating with momentum @xmath256 . in appendix
[ app : linenum ] we present numerical data that suggest that this type of evolution can be already observed for small values of the time @xmath222 and length @xmath0 .
in this section we discuss how the continuous - time quantum snake walk might potentially lead to an efficient algorithm solving the extended glued trees problem , which we defined in section [ ssec : extdef ] .
our aim is to come up with an algorithm which makes @xmath257 oracle queries , where @xmath52 is the height of the glued trees graph and , without loss of generality , we assume each vertex is encoded using @xmath258 bit label .
let us consider the following expansion of the glued trees graph .
suppose @xmath1 is a glued trees graph of height @xmath52 , and @xmath54 and @xmath55 are its roots .
for @xmath259 we construct an _ expanded glued trees graph _ @xmath260 as follows : take @xmath261 instances of graph @xmath1 , create two complete binary trees @xmath262 and @xmath263 each of height @xmath264 , and then make bijective associations between leaves of @xmath262 and instances of @xmath54 and between leaves of @xmath263 and instances of @xmath55 to obtain @xmath260 .
an example of the graph @xmath265 for a given glued trees graph @xmath1 of height @xmath45 is given in figure [ fig : egtg ] .
basically , @xmath260 looks exactly like a glued trees graph of height @xmath266 , but instead of one cycle gluing the trees , now we have @xmath261 cycles . for the rest of the paper we will focus on the continuous - time quantum snake
walk on @xmath260 .
given an oracle for @xmath1 , since we know an efficient algorithm finding the label of @xmath55 @xcite , we can efficiently simulate the quantum snake walk on @xmath260 for @xmath267 and the length of the snake @xmath268 ( see ( * ? ? ? * section 4.1.3 ) ) .
the sketch of an algorithm potentially solving the extended glued trees problem would be as follows : 1 . choose @xmath269 such that @xmath270 ; 2 .
[ s : alg2 ] construct a particular initial superposition for a quantum snake walk on @xmath260 which contains only snakes on @xmath262 ( construction of any such state involves no oracle queries at all ) ; 3 . run a quantum snake walk on @xmath260 for a certain amount of time @xmath271 and measure , thus obtaining a snake of length @xmath0
if the obtained snake contains two vertices such that one belongs to @xmath262 and other to @xmath263 , then the snake has to go completely through one instance of @xmath1 , and from this snake we can extract a path connecting @xmath54 and @xmath55 ; otherwise return to step [ s : alg2 ] . for @xmath272 and @xmath273 .
one instance of the original glued trees graph @xmath1 is highlighted using thick edges . ]
first let us show that the analysis of the quantum snake walk on this relatively complex graph becomes much easier if we restrict the choice of initial state to certain superpositions .
for @xmath274 $ ] let @xmath275 be the set of all vertices in @xmath260 which are at distance @xmath276 from the root of @xmath262 ( see figure [ fig : egtg ] ) .
now , for @xmath274 $ ] and @xmath70 let @xmath277 be the set of all snakes @xmath278 such that @xmath279 and for all @xmath23 $ ] , if @xmath280 , then @xmath281 . for @xmath282 let @xmath283 be the uniform superposition over the elements of @xmath277 , and
let us refer to @xmath284 as the _ column subspace_. due to symmetry , the column subspace is invariant under @xmath285 , the hamiltonian governing the quantum snake walk on @xmath260 , and the action of @xmath285 on the column subspace does not depend on which cycle in particular glues the two binary trees in @xmath1 . also , let us call the subgraph of @xmath260 having the set of vertices @xmath286 and all the edges originally connecting @xmath287 and @xmath288 the _ glued part_. from now on let us consider @xmath266 such that @xmath289
. for @xmath290 and @xmath70 , if we run the quantum snake walk on @xmath260 starting from the state @xmath283 for time small enough , then we will never get close to the roots of @xmath260 . thus , in such a case the walk behaves almost exactly the same way on all @xmath291 having @xmath292 , and for the sake of simpler analysis we assume @xmath293 ( similarly as the regular quantum walk on a finite line segment , assuming the initial state is chosen to be far from the endpoints of the segment , can be approximated by the walk on the infinite line )
. therefore the hamiltonian governing the quantum snake walk on @xmath260 restricted to the column subspace effectively is @xmath294\,(\,|x+1,0j\rangle\langle x , j0|+|x , j0\rangle\langle x+1,0j|\ , ) \notag \\ & + \text{if}_{|j|_\pm>0}[1;2]\,(\,|x-1,1j\rangle\langle x , j1|+|x , j1\rangle\langle x-1,1j|\ , ) \notag \big ) \\ + \sum_{x= 1-\min(|j|_\pm,0)}^{+\infty } & \big ( ( \,|x+1,0j\rangle\langle x , j1|+|x , j1\rangle\langle x+1,0j|\ , ) \notag \\ & + 2\,(\,|x-1,1j\rangle\langle x , j0|+|x , j0\rangle\langle x-1,1j|\ , ) \\ & + \sqrt{2}\,(\,|x+1,0j\rangle\langle x , j0|+|x , j0\rangle\langle x+1,0j|\ , ) \notag \\ & + \sqrt{2}\,(\,|x-1,1j\rangle\langle x , j1|+|x , j1\rangle\langle x-1,1j|\ , ) \notag
\big ) \bigg ) , \end{aligned}\ ] ] where @xmath295 and @xmath296 $ ] is @xmath297 if @xmath298 , and @xmath299 otherwise ( see ( * ? ? ?
* section 4.2 ) for details ) . here
let us consider the continuous - time quantum snake walk on @xmath260 restricted to superpositions @xmath283 such that @xmath300 .
we care only about the first sum ( [ eqp : hhat ] ) in the expression for @xmath301 , and , because @xmath302 , similarly as before , we can approximate an action of @xmath301 by one of @xmath303 where @xmath304 the analysis of hamiltonian @xmath305 and the quantum evolution governed by it will highly resemble the analysis of @xmath84 above . let us redefine @xmath306 , @xmath307 , @xmath308 and @xmath309 , and let us define the orthonormal basis @xmath310 as before except that now we are using these new definitions of @xmath311 , @xmath312 , @xmath313 and @xmath314 to define @xmath87 for each @xmath86 $ ] . in this new basis @xmath315
again we can see that only @xmath130 eigenvalues of @xmath316 are @xmath79-dependent , and those are the eigenvalues of operator @xmath317 , where @xmath318 . along similar lines as for lemma [ the : pcondition ] we can prove the following lemma .
let us fix @xmath103 and @xmath104 .
the equation @xmath319 has @xmath130 distinct solutions in the interval @xmath148 , and , if @xmath149 is a solution of this equation , then @xmath320 is an eigenvalue of @xmath321 .
now suppose @xmath0 is even , and let @xmath322 be the median ( i.e. , the @xmath165-th largest ) eigenvalue of @xmath321 .
let @xmath323 .
similarly as before , we have for every @xmath0 , @xmath324 is bounded between @xmath325 and @xmath326 and we have @xmath327 .
also , there exists @xmath328 such that for all @xmath329 we have @xmath330 if and only if @xmath174 .
let @xmath331 be the unique eigenvector of @xmath321 corresponding to the eigenvalue @xmath322 such that @xmath332 and @xmath333 , let @xmath334 , and let @xmath335 similarly as for the quantum snake walk on the line , numerical results suggest that @xmath336 is localized on the tree @xmath262 at the distance @xmath337 away from the glued part .
if this property indeed holds , we can show that for @xmath329 and for asymptotically large time of the evolution @xmath222 the state @xmath336 evolves as two wave packets moving with momentum @xmath338 in the opposite directions , that is , one moving up the tree and the other down the tree .
however , now we need to be careful when we talk about asymptotically large @xmath222 , because we still want @xmath222 to be small enough so that the wave packets do not get close to the root of @xmath262 or the glued part . for the purpose of our algorithm , we are interested in only one of those two wave packets , the one which moves towards the glued part .
our hope is that in @xmath257 time this wave packet will propagate far enough to reach @xmath263 and with high probability will give us a snake containing vertices from both @xmath262 and @xmath263 .
the regular quantum walk on the line started from a single vertex evolves as two wave packets moving in the opposite directions with momentum @xmath45 ( * ? ? ?
* section 3.3.2 ) , but for any @xmath339 $ ] we can create a wave packet which moves with momentum @xmath227 in one single direction ( see @xcite ) .
we would like to do an analogous thing for the quantum snake walk .
for @xmath340 and @xmath341 consider the state @xmath342 where @xmath343 denotes the error function and @xmath344 denotes the real part function .
one can show that @xmath345 .
numeric results suggest that @xmath346 evolves as a gaussian - shaped wave packet moving towards the glued part of @xmath260 with momentum @xmath347 ( see appendix [ app : gmnum ] ) . for any snake in the superposition @xmath348 , other than the position of its initial vertex , we are also interested in how many different depths in the tree @xmath262 its vertices possesses . to be more precise , for a snake @xmath278 let @xmath349 be such that @xmath350 for all @xmath351 $ ] .
we define the _ span _ of @xmath27 to be @xmath352 and the _ span length _ of @xmath27 to be @xmath353 .
we are interested in snakes with reasonably large span length , because , after all , if we obtain a snake which contains a solution of the extended glued trees problem , the span length of this snake will be at least @xmath354 .
numerics presented in appendix [ app : gmnum ] suggest that , if we measure @xmath348 in the standard basis and output the span length of the obtained snake , then the expected value of this random variable is in @xmath355 , which is good .
however , the wave packet might drastically loose its average span length when it gets close to the glued part and the action of the operator @xmath301 can not be approximated anymore by one of @xmath305 . in appendix
[ app : unfort ] we describe numerical data which , unfortunately , suggest that it indeed may be the case .
let us consider yet another analogy to the regular quantum walk on lines .
consider the regular quantum walk on a graph @xmath356 constructed from some finite graph @xmath357 by attaching two semi - infinite lines to two vertices of @xmath357 . on either of those two lines we can construct a wave packet which moves towards the graph @xmath357 .
what are the probabilities of this wave packet being reflected form @xmath357 and it being transmitted through @xmath357 ( to the other line ) can be calculated by inspecting the eigenvalues and eigenvectors of the adjacency matrix of @xmath356 and then using the standard scattering theory ( see @xcite , for example ) .
similarly , for the quantum snake walk on @xmath260 we can construct wave packets which move on @xmath262 towards the glued part ( to be more precise , we seem to be able to construct such wave packets , since we have not rigorously proven that yet ) , we can also construct wave packets which move on @xmath262 away from the glued part , and we can do the same on @xmath263 . in order to understand what happens to the wave packets moving towards the glued part once they reach it , let us inspect particular eigenvectors and eigenvalues of @xmath301 .
let @xmath358 be the unique vector satisfying @xmath359 for all @xmath70 .
numerical data for @xmath0 up to @xmath360 agree with the following hypothesis .
[ hyp : evec ] for every @xmath104 @xmath322 is the eigenvalue of @xmath301 coresponding to the eigenvector @xmath361 where @xmath362 , @xmath363 let us denote the eigenvector given in hypothesis [ hyp : evec ] by @xmath364 . assuming hypothesis [ hyp : evec ]
, the standard scattering theory suggests that the wave packet ( started as ) @xmath346 , where @xmath340 and @xmath365 is assumed to be small , will be reflected from the glued part with probability @xmath366 and transmitted through it with the probability @xmath367 . also , @xmath368 , called the _ effective length _
, divided my the momentum of the wave packet describes for how much time the glued part delays the wave packet propagating through it .
it is really interesting that the transmission coefficient @xmath153 does not depend on @xmath0 , the length of the snake , and it still stays the same if we consider eigenvectors of @xmath301 corresponding to any other @xmath79-dependent eigenvalue of @xmath316 , not only the median eigenvalue .
even more , already in @xcite childs et al . proved that for the regular quantum walk on @xmath260 the probability of a wave packet corresponding to the vector @xmath369 being transmitted through the glued part is @xmath370 .
figure [ fig : transm ] shows that the wave packet @xmath346 has the highest probability of being transmitted through the glued part and has the minimal effective length of the glued part for @xmath371 .
it also seems that @xmath347 reaches its maximum at @xmath371 ( we have @xmath372 ) , therefore the faster a wave packet is , the most likely it will propagate through the glued part ( note : this is the case only for wave packets corresponding to the median eigenvalue ) .
propagating through the glued part on the left , and , on the right , its corresponding effective length of the glued part.,title="fig : " ] propagating through the glued part on the left , and , on the right , its corresponding effective length of the glued part.,title="fig : " ] even if we rigorously proved that we can construct localized wave packets having momenta as stated above , the span lengths of those wave packets are as stated , and so are the transmission probabilities , such a proof most likely would use the method of stationary phase approximation , which deals with the case when all ( i.e. , initial , reflected and transmitted ) wave packets are far away from the glued part .
but the problem is : what happens during the time when wave packets encounter the glued part is exactly what we are interested in for the analysis of our algorithm . unfortunately we can not obtain enough numerical data to have a good intuition what is the maximal probability obtaining a snake connecting @xmath262 and @xmath263 for any given @xmath52 and @xmath270 .
also , @xmath0 necessary for an efficient algorithm may be very large because we only restrict it to be in @xmath257 .
nonetheless , a numerical inspection of the vector @xmath364 done in appendix [ app : unfort ] suggests that the expected span length of snakes in a wave packet is significantly reduced when the wave packet propagates through the glued part .
we currently do not have any candidates for a pattern this reduction follows , so it is not clear if the reduction is exponential or not .
in this paper we introduced the continuous - time quantum snake walk and discussed one of its potential algorithmic applications , the extended glued trees problem . assuming the algorithm sketched in section [ ssec : algidea ] which uses wave packets as initial states indeed efficiently finds a path connecting both roots of the glued trees graph , it almost certainly will not find a shortest path .
it would be interesting to see if there exists an efficient algorithm finding a shortest path , or an efficient quantum algorithm for any other pathfinding problem , which is based on the quantum snake walk .
on the other hand , if the algorithm we propose is not efficient , despite that the wave packets ( most likely ) are being transmitted through the glued part with high probability and in polynomial time , it would suggest that possibly there are some fundamental limitations which restrict us from solving the extended glued trees problem efficiently .
finding and understanding such limitations would be an interesting and useful thing to do .
another potential direction for future research is to consider discrete - time analogues of the continuous - time quantum snake walk ( for example , by generalizing mcgettrick s results @xcite ) and their applications .
i wish to thank john watrous , richard cleve , and andrew childs for numerous helpful discussions regarding this work .
in particular , the idea to consider quantum walks whose states are paths is due to watrous .
this work was supported by a mike and ophelia lazaridis fellowship .
let @xmath373 be an eigenvector of @xmath136 and @xmath82 be its corresponding eigenvalue .
the maximum absolute column sum norm of ( [ eq : linehnk ] ) is at most @xmath374 ( it is exactly @xmath374 unless @xmath375 ) , which implies that all the eigenvalues of @xmath81 are at most @xmath374 by the absolute value , and therefore so are the eigenvalues of @xmath136 .
in fact , it is easy to see that @xmath376 would imply @xmath134 and @xmath377 , which is also the only case when @xmath378 .
hence , there exist unique @xmath379 and @xmath380 such that @xmath381 and @xmath382 .
we have @xmath383 from which we get @xmath384 .
now let us use induction .
let @xmath385 , and let us assume @xmath386 for all @xmath387 $ ] .
we have @xmath388 which gives us @xmath389 therefore , by induction , @xmath386 for @xmath390 $ ] .
similarly , @xmath391 implies @xmath392 .
finally , @xmath393 gives us @xmath394 one can see that ( [ eq : pcondition2 ] ) is both the necessary and sufficient condition for @xmath142 to be an eigenvalue of @xmath136 .
also , as shown above , an eigenvalue of @xmath136 uniquely ( up to a global factor ) determines the eigenvector corresponding to it .
this implies that all the eigenvalues of @xmath136 are distinct , thus ( [ eq : pcondition2 ] ) has @xmath130 distinct solutions in the interval @xmath148 .
for @xmath395 we have @xmath396 and the lemma holds , thus let us assume @xmath397 .
we know that @xmath398 and @xmath399 , therefore a corollary of theorem [ thm : lambdad ] is that @xmath400 .
let @xmath401 $ ] be such that @xmath402 , and therefore @xmath403 $ ] .
hence @xmath404 , where @xmath405 is the solution of @xmath406 which is just a rewritten @xmath149-equation . by simple trigonometric derivations this gives us @xmath407 notice that @xmath408 , and we already have an expression for @xmath409 .
thus we can obtain an expression for @xmath410 , from which @xmath411 because @xmath412 and @xmath413 for any @xmath414 , without the loss of generality let us assume @xmath415 ; hence @xmath416 .
therefore @xmath417 , @xmath418 and @xmath419 which implies @xmath420 since we already have expressions for @xmath167 and @xmath410 , we can get that @xmath421 the first term of this equality is in @xmath422 while the other two are in @xmath423 .
more precise analysis of the first term reveals that there is a constant @xmath424 such that @xmath425 .
regarding the second part of the lemma , along the similar lines as for @xmath426 , one can show that there also exists a constant @xmath427 such that @xmath428 .
because of that , there exist @xmath171 and @xmath429 such that @xmath430 for all @xmath431 $ ] and @xmath432 for all @xmath433 $ ] .
consider @xmath172 .
we have @xmath434 for all @xmath431 $ ] . the fact that @xmath435 and @xmath436 for all @xmath433 $ ] implies @xmath434 for @xmath437 .
thus , @xmath438 for all @xmath172 and @xmath439 .
fix even @xmath440 . for any given @xmath79 we can numerically calculate the median eigenvalue @xmath164 of @xmath136 and its corresponding eigenvector @xmath189 . then we can get a numerical approximation of @xmath183 by approximating the integral ( [ eq : defeta ] ) defining it by a riemann sum . for every unit vector @xmath441 , @xmath213 and @xmath70 let @xmath442 and let @xmath443 . then , in the standard basis , we have @xmath444 and @xmath445 . if we measure @xmath220 in the standard basis , @xmath446 is the probability of obtaining a snake which starts at the position @xmath203 ; therefore the probability distribution @xmath447 gives us good intuition about the ` location ' of the state @xmath220 on the line .
[ lem : eta0even ] for every even @xmath192 we have @xmath448 .
let @xmath203 be even , so that @xmath449 . from the definitions of @xmath183 and @xmath369
we can see that @xmath450 we are left to show that @xmath451 for all @xmath452 $ ] . for @xmath453 $ ] let @xmath454 be such that @xmath455 .
one can see that @xmath456 .
now , @xmath457 and @xmath458 imply @xmath459 for @xmath390 $ ] and @xmath460 . from the definition of @xmath461
we see that @xmath462 . ,
the probability @xmath463 of obtaining a snake starting at the position @xmath203 by measuring @xmath183 in the standard basis ( the linearly scaled plot on the left and the logarithmically scaled plot on the right ) .
note : numeric values below @xmath464 can not be trusted because they approach the precision of the floating point arithmetic used.,title="fig : " ] , the probability @xmath463 of obtaining a snake starting at the position @xmath203 by measuring @xmath183 in the standard basis ( the linearly scaled plot on the left and the logarithmically scaled plot on the right ) .
note : numeric values below @xmath464 can not be trusted because they approach the precision of the floating point arithmetic used.,title="fig : " ] as an evidence that @xmath183 is localized around the position @xmath110 on the line , in figure [ fig : eta0 ] we present numerically computed values of the probability @xmath463 in the case when @xmath204 for @xmath203 close to @xmath110 .
lemma [ lem : eta0even ] states that this probability is @xmath110 for even @xmath203 , therefore let us now focus on odd @xmath203 .
we can see that @xmath463 seem to decrease exponentially as @xmath337 increases , and especially large jump seem to be from @xmath465 to @xmath466 .
a similar pattern , a large gap between @xmath467 and @xmath468 , was observed for values of @xmath0 other than @xmath469 .
this is the main motivation why hypothesis [ hyp : local ] is stated in the way it is stated .
the reason why we choose @xmath37-norm in it instead of @xmath45-norm is a technicality : it allows us to make stronger statements about superpositions of vectors @xmath202 , where @xmath192 .
numerical data showed in figure [ fig : tail ] support hypothesis [ hyp : local ] .
-norm of the part of the state @xmath183 corresponding to snakes starting at least @xmath0 positions away from @xmath110 ( note : @xmath183 is @xmath0-dependent itself ) , that is , @xmath470 .
for @xmath395 this value can be shown to be @xmath110 . ] again , let @xmath204 , and let us compute how @xmath183 behaves under evolution of @xmath471 .
section [ ssec : linewave ] states that , under certain assumptions , for asymptotically large time @xmath222 the state @xmath183 evolves as two wave packets each moving with momentum @xmath472 .
we can see in figure [ fig:2waves ] that this kind of motion can be observed already for smaller values of @xmath222 .
notice that peeks of the probability distribution around positions @xmath473 get relatively higher for larger @xmath222 .
higher probabilities around the origin @xmath474 are likely there due to the fact that @xmath475 has a local maximum when @xmath476
. evolved for time @xmath477 and @xmath478.,title="fig : " ] evolved for time @xmath477 and @xmath478.,title="fig : " ]
the same way as in the case of the quantum snake walk on the line , we can express @xmath348 as an integral and then approximate it by a riemann sum . also , for every unit vector @xmath441 ( @xmath479 being the column subspace ) and @xmath192 let @xmath480 , which , if we measure @xmath220 in the standard basis , is the probability of obtaining a snake whose initial vertex is in @xmath275 .
figure [ fig : difwaves ] shows that the state @xmath346 evolves as a gaussian - shaped wave packet moving with momentum @xmath347 towards the glued part .
the diffusion of the wave packet can be explained by the heisenberg uncertainty principle : we can not know both the position and the momentum of the walk to an arbitrary precision . for @xmath481 ( the left plot ) and @xmath371 ( the right plot ) in the case
when @xmath161 and @xmath482 .
the initial position of the wave packet is given with the dashed lines , and the position after time @xmath483 with the solid lines .
numerically we get @xmath484 and @xmath485.,title="fig : " ] for @xmath481 ( the left plot ) and @xmath371 ( the right plot ) in the case when @xmath161 and @xmath482 .
the initial position of the wave packet is given with the dashed lines , and the position after time @xmath483 with the solid lines .
numerically we get @xmath484 and @xmath485.,title="fig : " ] for every @xmath192 and @xmath70 the span length of all snakes in @xmath277 is the same , and it depends only on @xmath486 . therefore we can easily construct the unique observable @xmath487 acting on @xmath106 such that for every @xmath441 , if we measure @xmath220 in the standard basis and return the span length of the resulting snake , the expected value of this random variable is @xmath488 .
for @xmath348 we have @xmath489 figure [ fig : span ] shows a plot of numerically obtained values of @xmath490 for all values of even @xmath0 up to @xmath491 . from this plot alone
it is not completely clear how fast @xmath492 grows with @xmath0 , but it seems to grow with the same rate as @xmath493 . for any @xmath70 we have @xmath494 , where @xmath495 is the hamming norm , therefore for @xmath486 uniformly chosen at random we have @xmath496 . because @xmath497 which is a quite balanced superposition over the vectors of @xmath498 , we conjecture that @xmath499 . for an arbitrary @xmath192 . ]
consider the walk on the expanded glued trees graph started as the wave packet @xmath346 , which moves towards the glued trees , and consider the maximal probability that , if at any point during the walk we make a measurement , we obtain a snake which contains vertices from both @xmath500 and @xmath501 .
we are interested to choose @xmath268 such that this probability is at least inverse - polynomially large in @xmath52 for @xmath502 .
it is not clear whether such choice is possible , and here we present numerical data suggesting that @xmath0 we need to choose may me very large , if not even exponentially large in @xmath52 .
we introduced the vector @xmath364 in hypothesis [ hyp : evec ] as an eigenvector of @xmath301 .
unless its coefficients @xmath503 behave very unexpectedly and @xmath504 , for @xmath505 and @xmath506 the state @xmath346 can be well approximated by a superposition of states @xmath364 , specifically , @xmath507 because @xmath364 is an eigenvector of @xmath301 for every @xmath104 , the state @xmath508 can be well approximated by a superposition of states @xmath364 too .
therefore , if for all @xmath104 the state @xmath364 has a small overlap with the space of snakes connecting vertices from both @xmath509 and @xmath510 , so does @xmath508 .
now , let us numerically calculate the coefficients @xmath503 and see what we obtain if we `` measure '' @xmath364 in the standard basis ( note that @xmath364 has an infinite euclidean norm ) . to be precise , we will be interested in the `` probabilities '' @xmath511 defined as follows
. for even @xmath512 and @xmath192 and for odd @xmath512 and @xmath513 let @xmath514 be the projector to the space of all snakes having span @xmath515 .
let @xmath516 .
we numerically calculated these values for @xmath517 and figure [ fig : four ] shows for every @xmath518 $ ] how @xmath519 depends on the value of @xmath203 ( in this particular case @xmath520 for all @xmath203 ) .
if we measure " @xmath521 .
note : the scaling of the plots differs.,title="fig : " ] if we
measure " @xmath521 .
note : the scaling of the plots differs.,title="fig : " ] + if we
measure " @xmath521 .
note : the scaling of the plots differs.,title="fig : " ] if we
measure " @xmath521 .
note : the scaling of the plots differs.,title="fig : " ] a similar pattern as in figure [ fig : four ] is observed if we consider values of @xmath511 for @xmath79 other than @xmath253 or `` measure '' eigenvectors of @xmath301 corresponding to other @xmath79-dependent eigenvalues of @xmath316 , not only the median eigenvalue .
this means that the overlap between these eigenvectors and the space of snakes having a small span length drastically increases close to the glued part , and decreases for snakes having a large span length .
therefore , the expected span length of wave packets will get smaller when they come close to the glued part , and will increase again once they have been reflected from or gone through it
. however , it is not clear what pattern this relative reduction of the span length follows when @xmath0 increases . | i introduce a new type of continuous - time quantum walk on graphs called the quantum snake walk , the basis states of which are fixed - length paths ( snakes ) in the underlying graph .
first i analyze the quantum snake walk on the line , and i show that , even though most states stay localized throughout the evolution , there are specific states which most likely move on the line as wave packets with momentum inversely proportional to the length of the snake .
next i discuss how an algorithm based on the quantum snake walk might potentially be able to solve an extended version of the glued trees problem which asks to find a path connecting both roots of the glued trees graph .
no efficient quantum algorithm solving this problem is known yet . |
the scaling behavior of directed percolation is recognized as the paradigm of the critical behavior of several non - equilibrium systems which exhibits a continuous phase transition from an active state to an absorbing non - active state ( see for instance @xcite ) .
the widespread occurrence of such systems in physics , biology , as well as catalytic chemical reactions is reflected by the well known universality hypothesis of janssen and grassberger that models which exhibit a continuous phase transition to a single absorbing state generally belong to the universality class of directed percolation @xcite . introducing additional symmetries
the critical behavior differs from directed percolation .
in particular particle conservation leads to the different universality class of absorbing phase transitions with a conserved field as pointed out in @xcite . in that work the authors introduced two models , the conserved lattice gas ( clg ) and the conserved threshold transfer process ( cttp ) .
the latter one is a conserved modification of the threshold transfer process introduced in @xcite .
both models display a continuous phase transition from an active to an inactive phase and are believed to belong to the same universality class @xcite .
the steady - state scaling behavior of the clg model was investigated recently .
the order parameter and its fluctuations were numerically examined in @xcite .
the scaling behavior in an external field conjugated to the order parameter was considered in @xcite .
furthermore a modified clg model was introduced which allows to determine analytically the steady - state mean - field scaling behavior of the universality class @xcite . on the other hand
the scaling behavior of the cttp was investigated in low dimensional ( @xmath0 ) systems only @xcite and no external field was applied .
therefore we consider in this work the cttp with and without an external field in various dimensions ( @xmath1 ) and determine a set of critical exponents .
all obtained results coincides with those of the clg model , strongly supporting the universality hypothesis of @xcite .
in this work we consider the cttp on simple cubic lattices of linear size @xmath2 in various dimensions with @xmath3 particles .
the lattice sites may be empty , occupied by one particle , or occupied by two particles .
empty and single occupied sites are considered as non - active whereas double occupied lattice sites are considered as active . in the latter case one tries to transfer both particles of each active site to randomly chosen empty or single occupied nearest neighbor sites .
if no active sites exist the system is trapped forever in a certain configuration , a so - called absorbing state . in the following
we denote the densities of active sites with @xmath4 and the density of particles on the lattice as @xmath5 , which is considered as the control parameter of the absorbing phase transition .
the density of active sites @xmath4 is the order parameter of the absorbing phase transition , i.e. , it vanishes at the critical density @xmath6 according to @xmath7 with the reduced control parameter @xmath8 and the order parameter exponent @xmath9 .
similar to equilibrium phase transitions it is possible in the case of absorbing phase transitions to apply an external field @xmath10 which is conjugated to the order parameter ( see for instance @xcite ) . as usually for continuous phase transitions the conjugated field has to destroy the disordered phase and
the associated linear response function @xmath11 has to diverge at the critical point ( @xmath12 , @xmath13 ) . in the case of an absorbing phase transition the external field acts as a spontaneous creation of active particles ,
i.e. , the external field destroys the absorbing state and thus the phase transition itself . but considering absorbing phase transitions with particle conversation one has to take care that the external field does not change the particle number .
a possible realization of the external field was developed in @xcite where the external field triggers movements of inactive particles which may be activated in this way .
the external field @xmath10 is another relevant scaling field and for sufficiently small values of @xmath10 the order parameter scales as @xmath14 with the critical field exponent @xmath15 and the scaling function @xmath16 .
choosing @xmath17 one recovers eq.([eq : order_par ] ) whereas @xmath18 leads at the critical density to @xmath19 in our simulations we start with randomly distributed particles .
all active sites are listed and this list is updated in a randomly chosen sequence . in the case
that an external field is applied the active particle creation is performed after each update step in order to mimic the external field . after a certain relaxation time the system reaches a steady state where the density of active sites at update step @xmath20 fluctuates around the average value @xmath21 which is interpreted as the order parameter @xmath22 ( see for instance figs.1 of @xcite ) . additionally to the order parameter we consider its fluctuations @xmath23 .
\label{eq : def_fluc}\ ] ] approaching the transition point the fluctuations diverge for zero - field according to @xmath24 the fluctuation exponent @xmath25 fulfills the scaling relation @xcite @xmath26 where the exponent @xmath27 describes how the spatial correlation length diverges at the transition point . in the critical regime
we assume that the fluctuations obey the scaling ansatz @xmath28 setting @xmath29 one recovers eq.([eq : fluc_crit ] ) for @xmath13 .
analogous to equilibrium phase transitions the susceptibility is defined as the derivative of the order parameter with respect to the conjugated field @xmath30 setting @xmath31 one gets that the susceptibility diverges for zero - field as required according to @xmath32 furthermore , one yields the scaling relation @xmath33 which corresponds to the well known widom equation of equilibrium phase transitions . using this scaling relation
one can calculate the value of the susceptibility exponent @xmath34 from the obtained values of @xmath9 and @xmath15 .
notice that in contrast to the scaling behavior of equilibrium phase transitions the non - equilibrium absorbing phase transition is characterized by @xmath35 .
as a function of the particle density for zero - field ( symbols ) and for @xmath36 and @xmath37 ( lines ) .
the inset displays the order parameter fluctuations @xmath38 for zero field ( symbols ) and for @xmath39 , @xmath36 and @xmath37 ( lines ) . , width=302 ] at the beginning of our analysis we consider the scaling behavior of the order parameter for @xmath40
. system sizes up to @xmath41 for @xmath42 , @xmath43 for @xmath44 , and @xmath45 for ( @xmath46 ) are considered . in each cases
we start the simulation with randomly distributed particles . after a certain transient regime
the system reaches a steady state where the density of active particles fluctuates around an average value which is interpreted as the order parameter . in the steady state
up to @xmath47 update steps for @xmath42 and @xmath48 for @xmath49 are performed to measure the average density of active sites . for zero - field
this procedure is repeated for at least 10 different initial configurations in order to get an accurate estimation of the order parameter close to the critical point ( @xmath50 , @xmath13 ) . in fig.[fig
: rho_a_1d ] we present the data of the one - dimensional order parameter at zero - field . approaching the transition point the corresponding correlation length increases and the system tends to the absorbing state if the correlation length is of the order of the system size . instead of a finite - size scaling analysis
( see for instance @xcite ) we take care of these finite - size effects in the way that we increase the system size before these finite - size effects occur and use only data from simulations that have not reached the absorbing state . as a function of the reduced particle density @xmath51 at zero - field for various dimensions @xmath52 .
the dashed line corresponds to a power - law behavior according to eq.([eq : order_par ] ) for @xmath53 .
for @xmath54 the data are shifted horizontally by a factor 1.5 in order to avoid an overlap . in the case of the four - dimensional model the dashed line corresponds to eq.([eq : order_par_dc_zero_field ] ) with @xmath55 .
, width=302 ] as a function of the reduced particle density @xmath51 at zero - field for various dimensions @xmath52 .
the dashed line corresponds to a power - law divergence [ eq.([eq : fluc_crit ] ) ] .
for @xmath56 the fluctuations are maximal at the transition point but finite . , width=302 ] decreasing the particle density the order parameter decreases and vanishes at the transition point . to determine the critical indices one varies the critical density @xmath6 until one obtains asymptotically a straight line in a log - log plot
the exponent is then obtained by a regression analysis .
the values of the order parameter as a function of the reduced particle density @xmath57 are plotted in fig.[fig : rho_a_all_d ] . in all cases
the asymptotic behavior ( @xmath58 ) of the order parameter obeys eq.([eq : order_par ] ) .
for @xmath42 we get @xmath59 and @xmath60 .
the latter value is smaller than the value @xmath61 estimated from significantly smaller system sizes ( @xmath62 ) @xcite .
furthermore our value differs from @xmath63 obtained from simulations of the one - dimensional fix - energy manna sandpile model @xcite that is expected to belong to the same universality class .
in the two dimensional case we obtain @xmath64 and @xmath65 . again
the order parameter exponent differs slightly from the previously reported result @xmath66 obtained from simulations of small lattice sizes ( @xmath67 ) @xcite .
but our value agrees with the estimate of the corresponding two - dimensional manna sandpile model @xmath68 @xcite .
the estimates of the three dimensional model are @xmath69 and @xmath70 . all
obtained critical exponents are listed in table [ table : critical_indicees ] . in fig.[fig :
rho_a_fluc_all_d ] we present the order parameter fluctuations as a function of the control parameter at zero field .
we observe for @xmath71 a power - law behavior according to eq.([eq : fluc_crit ] ) .
using a regression analysis we get the estimates @xmath72 for @xmath42 , @xmath73 for @xmath44 , and @xmath74 for @xmath46 . for various dimensions . for @xmath75 the curves
are shifted vertically in order to avoid overlaps . in the case of the four - dimensional model @xmath76
is plotted vs. @xmath77 with @xmath78 and @xmath79 ( see text ) .
, width=302 ] in the following we analyze the order parameter as a function of the control parameter @xmath51 for different fields from @xmath80 up to @xmath81 . the applied field results in a smoothing of the zero - field curve , i.e. , the order parameter increases smoothly with the control parameter for @xmath82 ( see fig.[fig : rho_a_1d ] ) .
according to the scaling ansatz of the order parameter [ eq.([eq : scal_ansatz ] ) ] we choose @xmath83 and get the scaling form @xmath84 thus one varies the exponent @xmath15 until the curves for different values of the driving field have to collapse onto the scaling function @xmath16 if one plots
@xmath85 as a function of @xmath86 .
convincing results are obtained for @xmath87 ( @xmath42 ) , @xmath88 ( @xmath44 ) , as well as @xmath89 ( @xmath46 ) and the corresponding scaling plots are shown in fig.[fig : rho_a_field_scal_all_d ] .
next we consider the scaling behavior of the order parameter fluctuations @xmath90 .
the fluctuation data for @xmath42 are shown for different values of the external field in the inset of fig.[fig : rho_a_1d ] . for finite fields the fluctuations display a peak .
approaching the transition point ( @xmath91 ) this peak becomes a divergence signalling the critical point . in order to analyze the scaling behavior of the fluctuations we use the scaling ansatz eq.([eq : fluc_scal_ansatz ] ) and set @xmath92 @xmath93 using the above determined values of @xmath6 , @xmath15 and @xmath25 we get good data collapses confirming the accuracy of our analysis .
( see fig.[fig : rho_a_fluc_scal_all_d ] ) furthermore we determine the susceptibility exponent @xmath34 .
using the scaling relation eq.([eq : widom ] ) one gets the estimates of the susceptibility exponents @xmath94 ( @xmath42 ) , @xmath95 ( @xmath44 ) , and @xmath96 ( @xmath46 ) . for various dimensions . for @xmath75 the curves
are shifted vertically in order to avoid overlaps .
the fluctuations diverges at the critical point for @xmath97 whereas a jump of the fluctuations is observed in higher dimensions at zero - field . in the in case of the four - dimensional model @xmath90
is plotted vs. @xmath98 with @xmath99 .
, width=302 ]
in the case of the four dimensional model we considered system sizes from @xmath100 up to @xmath101 .
at least @xmath102 update steps were used to reach the steady state close to the transition point and @xmath103 update steps were performed to determine the order parameter and its fluctuations . .
the data are rescaled according to eq.([eq : order_par_dc_zero_field ] ) . the assumed asymptotic scaling behavior ( dashed line )
is obtained for @xmath104 .
, width=302 ] at the upper critical dimension @xmath105 the scaling behavior of the cttp is affected by logarithmic corrections similar to the clg model @xcite .
as argued in @xcite the order parameter obeys in leading order the scaling ansatz @xmath106 where the exponents @xmath9 and @xmath15 are given by the corresponding mean - field values @xmath107 and @xmath108 @xcite , respectively .
thus , for zero field the asymptotic scaling behavior of the order parameter obeys @xmath109 with @xmath110 . in our analysis
we plot @xmath111 as a function of @xmath112 and vary the exponent @xmath113 as well as the critical density @xmath6 until one gets asymptotically a straight line .
the best result is obtained for @xmath104 , @xmath114 and the corresponding plot is shown in fig.[fig : rho_a_4d_log ] .
this value of @xmath113 differs from the corresponding value of the clg model @xmath115 @xcite .
similar to the lower dimensions we consider the scaling behavior of the order parameter as a function of the control parameter for different external fields .
choosing @xmath116 the scaling ansatz [ eq.([eq : scal_ansatz_dc ] ) ] yields in leading order @xmath117 where the scaling argument @xmath118 is given in leading order by @xmath119 with @xmath120 . varying the logarithmic correction exponents
one gets for @xmath78 and @xmath79 a convincing data - collapse , which is shown in fig.[fig : rho_a_field_scal_all_d ] . using the values @xmath121 and @xmath79
we get the estimation @xmath122 which agrees with @xmath104 obtained from numerical simulations in zero - field . on the other hand
this values differs from the corresponding estimations of the clg model @xmath115 , @xmath123 , and @xmath124 @xcite .
furthermore we consider how the logarithmic corrections affect the scaling behavior of the fluctuations at the upper critical dimension .
as pointed out in @xcite the order parameter fluctuations are expected to obey the scaling ansatz @xmath125 a good data collapse is observed for @xmath126 ( see fig.[fig : rho_a_fluc_scal_all_d ] ) which differs again from the corresponding value of the clg model @xmath127 @xcite .
a modified version of the cttp with random neighbor hopping was recently introduced in @xcite . there , unrestricted particle hopping breaks long range correlations and the scaling behavior is characterized by the mean - field values @xmath128 , @xmath107 , and @xmath108 which are calculated analytically . in our simulations of the five dimensional model we considered system sizes from @xmath100 up to @xmath129 whereas system sizes from @xmath130 up to @xmath131 are used for @xmath54 .
at least @xmath103 update steps were used to reach the steady state and @xmath103 update steps were performed to determine the order parameter and its fluctuations .
the values of the order parameter are plotted in fig.[fig : rho_a_all_d ] and the obtained critical densities are @xmath132 for @xmath133 and @xmath134 for @xmath54 , respectively . in both dimensions
the asymptotic scaling behavior of the order parameter is in agreement with the mean - field behavior @xmath107 .
the fluctuations of the order parameter @xmath90 are plotted in fig.[fig : rho_a_fluc_all_d ] .
analogous to the clg model the fluctuations are characterized by a jump at the transition point corresponding to @xmath135 @xcite . above the critical dimension , i.e. @xmath136
, the scaling behavior of the cttp is expected to obey again the scaling ansatzes eqs.([eq : scal_ansatz],[eq : fluc_scal_ansatz ] ) where the exponents are given by the mean - field values independently of the particular dimension .
the obtained data collapse of the order parameter curves are presented in fig.[fig : rho_a_field_scal_all_d ] and confirm the above scenario .
furthermore we consider the fluctuations above the upper critical dimension . according to the mean - field value @xmath135 @xcite we plot @xmath38 as a function of @xmath137 and
the obtained data collapses are shown in fig.[fig : rho_a_fluc_scal_all_d ] . as a function of the dimension @xmath52 .
the critical density of the mean - field solution is denoted by @xmath138 .
the dashed line corresponds to a power - law behavior [ eq.([eq : rho_c_d ] ) ] with an exponent @xmath139 . , width=302 ] , @xmath25 , and @xmath15 of the cttp and clg model ( obtained from @xcite ) for various dimensions .
the dashed lines are just to guide the eyes .
, width=302 ] finally we address the question how the critical densities depends on the dimension .
as can be seen from table[table : critical_indicees ] the critical density tends with increasing dimension to the mean - field value @xmath128 @xcite that corresponds to an infinite dimension .
our analysis reveals that the critical densities approaches that mean - field value according to @xmath140 with @xmath141 ( see fig.[fig : rho_c_d ] ) .
this behavior is different from that of clg models on simple cubic lattices which is characterized by an exponent @xmath142 @xcite .
we investigated the steady state scaling behavior of the cttp model in various dimensions .
the order parameter exponent , the fluctuation exponent and the external field exponent are determined and the corresponding values are listed in table[table : critical_indicees ] . for @xmath42 and @xmath44
our results of the order parameter exponents differ from previous simulations obtained from significantly smaller system sizes .
our values of the critical exponents @xmath9 , @xmath25 , and @xmath15 agree within the error - bars with the corresponding exponents of the clg model ( see fig.[fig : clg_cttp_exp ] ) , strongly supporting the conjecture @xcite that both models belong to the same universality class .
the picture is not so clear at the upper critical dimension @xmath105 .
although the exponents are identical the logarithmic correction exponents of the cttp and clg model are different .
this result is rather surprising since the logarithmic corrections exponents are a characteristic feature of the whole universality class ( see for instance @xcite ) .
we think that more than statistical uncertainties this result is caused by systematic uncertainties of our analysis . in all cases we focused our attention to the leading order of the scaling behavior . taking corrections to
the leading order into account may result in comparable values of the logarithmic correction exponents .
further investigations are needed to clarify this point .
llllll @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath34 + + @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 + @xmath160 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 + @xmath166 & @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath168 + @xmath171 & @xmath172 & @xmath148 & @xmath154 & @xmath173 & @xmath148 + @xmath174 & @xmath175 & @xmath148 & @xmath154 & @xmath173 & @xmath148 | we analyze numerically the critical behavior of an absorbing phase transition in the conserved transfer threshold process . we determined the steady state scaling behavior of the order parameter as a function of both , the control parameter and an external field , conjugated to the order parameter .
the external field is realized as a spontaneous creation of active particles which drives the system away from criticality .
the obtained results yields that the conserved transfers threshold process belongs to the universality class of absorbing phase transitions in a conserved field . |
in a hierarchical evolutionary scenario , galaxies experience accretion / merging events during their lifetime . while early - type galaxies in nearby clusters appear ( homogeneously ) old , the field early - type galaxy population seems to contain genuinely , recently _ rejuvenated _ objects ( see e.g. * ? ? ? * ) .
early - type galaxies showing fine structure , like shells , occupy a special position since they are believed to fill the gap between ongoing mergers and normal elliptical galaxies .
shells are faint , sharp - edged stellar features characterizing @xmath0 16.5% of the field early - type galaxy population and avoiding cluster environments ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
two major scenarios for their origin have been proposed : _ a ) _ a weak interaction between galaxies @xcite , _ b ) _ merging / accretion events @xcite . in the former scenario ,
weak interaction can form long lasting azimuthally distributed shells through the interference of density waves producing a thick disc population of dynamically cold stars . however
, this requires a cold thick disc , not found in ellipticals . in the merging / accretion events between galaxies of different masses ( mass ratios typically 1/10 - 1/100 , _
b ) _ scenario above ) shells are density waves formed from infalling stars from the companion during a minor merger .
major merger can also produce shells @xcite .
whatever the mechanism is , interaction / merging events seem to have played a significant role in the formation / evolution of the early - type class in the field and activated a new star formation event @xcite . whether there is a link between shell galaxies , the early phases of merging processes , and
the class of normal early - type galaxies is still an open question . to test the depth of the link ,
a multiwavelength approach is of paramount importance .
the uv data probe the ongoing / recent star formation processes and study their distribution across the galaxy while the x - ray emission is connected to the past star formation and metal enrichment history of the bulk of the galaxy and may disclose hidden agn activity .
we present _ galaxy evolution explorer _
( _ galex _ ) , xmm-_newton _ optical monitor ( om ) and xmm - epic observations of ngc 474 , ngc 7070a , and eso 2400100 , three shell galaxies in the @xcite catalogue .
[ cols= " < , < , < , < , < , < , < , < , < , < " , ] [ table1 ]
beside the presence of shells , clearly visible in all three , although more spectacular in ngc 474 , these galaxies share other properties : they are in low density environments and are interacting systems . however , the details of the interaction are quite different : companions are well separated in ngc 7070a @xcite and strongly interacting in ngc 474 as shown by h_i _ observations ( see e.g. * ? ? ? * ) .
two nuclei are embedded and interacting within the eso 2400100 envelope @xcite . furthermore
, dust is detected in ngc 474 and ngc 7070a , although only for the latter , there is kinematic evidence of an external acquisition .
table [ table1 ] gives the journal of the xmm - newton observations .
x - ray data were treated with the standard routines provided by sas version 7.0 as suggested by xmm - newton science analysis system : users guide available on line .
the x - ray observations of ngc 474 have been already discussed in @xcite . during x - ray observations
, shell galaxies were simultaneously imaged in the ultraviolet and optical bands with the optical monitor @xcite ) .
observations have been performed using uvw1 and uvm2 bands which cover the ranges 245 - 320 nm and 205 - 245 nm , respectively .
galaxies have been observed also in the u ( 300 - 390 nm ) and b ( 390 - 490 nm ) bands .
the point spread function fwhm is @xmath02.0 in uvw1 , and 1.8 in uvm2 sampled with 0476@xmath10476 pixels . in order to complete the uv information about our galaxies we searched the _
galex _ archive and retrieved the data for ngc 474 .
the _ galex _ mission and instruments are fully described in @xcite and @xcite .
the spatial resolution of the images is @xmath045 and 60 fwhm in fuv ( 135 - 175 nm ) and nuv ( 175 - 275 nm ) respectively , sampled with 15@xmath115 pixels .
we found _ galex _ data only for the ngc 474 , that was observed the october 5th , 2003 with an exposure of 1477 and 1647 seconds in nuv and fuv bands respectively .
we aim at obtaining crucial information about the time at which the accretion / merging phenomenon has occured that can be derived from the far uv and optical colours .
such information is important for a proper discussion of the x - ray data . according to the evidence provided by the sample of @xcite , the x - ray luminosity could evolve during the period that characterizes phases of the galaxy - galaxy encounter , the merging and the set - up of the merging towards a relaxed galaxy .
figure [ fig1 ] shows the isointensity contours from the adaptively smoothed xmm-_newton _ of ngc 7070a and eso 2400100 superposed on the dss plates .
it is evident that the emission from these two galaxies has a significantly different morphology . in ngc 7070a
the emission is rather compact both in the hard and soft bands . in eso 2400100
the soft band emission is significantly stronger than the hard one and extends further than in ngc 7070a .
spectral results are in agreement with the spatial picture : the nuclear source of ngc 7070a is the dominant component while in eso 2400100 the emission is indicative of a low temperature plasma .
@xcite showed that ngc 474 is at the bottom of the x - ray emission distribution of e galaxies : its x - ray emission is consistent with the low end of the expected emission from discrete sources .
_ galex _ data of ngc 474 ( figure [ fig2 ] ) show that the nuv emission extends to the galaxy bulge , while the fuv emission shows up only in the central regions of the galaxy . in the ngc 474 image obtained in the uvw1 and uvm2 filters outer shells
are visible as in the optical image @xcite .
both eso 2400100 ( figure [ fig3 ] ) and ngc 7070a have extension similar to that of the optical image in the om - uvw1 and in the _ galex _ nuv filter .
this implies that the uv emission comes from the same kind of stellar population .
the fuv emission , more concentrated toward the nucleus is most likely dominated by the emission of different types of hot stars ( see also * ? ? ?
* ) . in figure [ fig3 ] ( mid and left panels )
the uvw1 and uvm2 images show the two nuclei embedded in eso 2400100 .
the northern nucleus is significantly bluer than the southern one .
@xcite noticed that the two nuclei have a significantly different h@xmath2 line - strength indices ( 2.79 in the northern nucleus vs. 1.54 in the southern one ) suggesting a different stellar population composition .
the accretion of faint galaxies seems one of the drivers of the secular evolution of galaxies in loose , poor groups .
since shells are widely believed to be generated by an accretion event , shell galaxies potentially trace the typical secular evolution of early - type galaxies in such environments .
furthermore , there is a growing evidence of multiple accretion events in the same galaxy .
the complex shell system of ngc 474 is believed to be generated by two distinct accretion events @xcite . in this framework
the double nucleus in eso 2400100 is an interesting and puzzling case at the same time .
is the second nucleus the cause of the observed system of shells or another evidence of a new , on - going , accretion event ? | shell galaxies are considered the debris of recent accretion / merging episodes . their high frequency in low density environments suggest that such episodes could drive the secular evolution for at least some fraction of the early - type galaxy population .
we present here the preliminary results of ultraviolet and x - ray data for a sample of three shell galaxies , namely ngc 474 , ngc 7070a and eso 2400100 .
the far uv morphology and photometry are derived using the observations obtained with the _ galaxy evolution explorer _ and the xmm-_newton _ optical monitor .
we aim at investigating the _ rejuvenation _ processes in the stellar population using the uv information as well as at gaining information about the possible evolution with time of the x - ray emission due interaction / merging processes . |
first observed in the @xmath5 meson system at argus , the neutral b meson transition from the particle to anti - particle state , and vice versa , occurs through a second order weak transition or `` box diagram '' .
the frequency of the oscillation is proportional to the small difference in mass between the two eigenstates , @xmath6 , and for the @xmath5 system can be translated into a measurement of the ckm element @xmath7 .
@xmath7 can be used to constrain the unitarity triangle and thereby yield information on the _ cp _ violating phase @xcite .
@xmath2 has been precisely measured ( the world average is @xmath8 @xcite ) but large theoretical uncertainties dominate the extraction of @xmath7 from @xmath2 . this problem can be reduced if the @xmath9 mass difference , @xmath10 , is also measured .
@xmath7 can then be extracted with better precision from the ratio : @xmath11 where @xmath12 is estimated from lattice qcd calculations to be @xmath13 @xcite .
the above has motivated many experiments to search for @xmath9 oscillations and though a statistically significant signal hasnt been observed yet , a lower limit ( @xmath14 ps@xmath15 at 95% c.l . ) has been set .
since this current limit indicates that the @xmath9 oscillations are at least 30 times faster than the @xmath16 oscillations , a @xmath9 mixing measurement is experimentally very challenging .
the d experiment at the fermilab tevatron , a @xmath17 collider at 1.96 tev center of mass energy , is well equipped to search for @xmath9 oscillations .
the large muon acceptance and forward tracking coverage of the d detector ( pseudorapidity coverage of @xmath18 for the muon , @xmath19 for the tracking and @xmath20 for the silicon sub - detectors ) , along with a robust muon trigger are highly effective in exploiting the large @xmath21 cross - section resulting in some of the largest semileptonic @xmath9 yields .
[ dsphipi ] shows the yield for the decay @xmath22 in @xmath3250 pb@xmath15 of data .
the decay @xmath23 has also been reconstructed and is expected to contribute significantly to the total @xmath9 yield .
other decays including hadronic @xmath9 decays are being studied as well .
other elements essential to a mixing analysis are given by the expression for the average statistical significance : @xmath24 where @xmath25 is the number of signal events , @xmath26 is a measure of the effectiveness of a flavor tagger ( @xmath27 is the tagging efficiency while @xmath28 is the `` dilution '' ) , @xmath29 is the number of background events and @xmath30 is the proper time resolution . as indicated above ,
tagging the meson flavor ( @xmath9 or @xmath31 ) at decay time ( final - state tagging ) and at production time ( initial - state tagging ) are crucial . for the semileptonic modes used for mixing studies at d
the final state particles provide the decay - time tag . for initial - state tagging ,
different techniques are being studied and optimized by doing measurements of @xmath2 .
three tagging algorithms have been developed so far : opposite - side muon tagging , opposite - side jet charge tagging , and same - side ( soft - pion ) tagging .
the muon tagger relies on identifying the flavor of the other b meson in the event using the sign of the muon it decayed to - a negative muon corresponds to a @xmath32 quark , and vice - versa . for the decay used for @xmath16 mixing studies - @xmath33 where @xmath34 and @xmath35 - both muons having the same sign would indicate that one b hadron had oscillated while opposite signs would indicate that neither ( or both ) had oscillated .
[ mixing_slt ] shows the measured asymmetry between the non - oscillated and oscillated mesons as a function of the visible proper decay length ( vpdl ) .
the fit to the asymmetry gives @xmath36 ps@xmath15 with an efficiency of @xmath37 % and @xmath38 % .
the opposite - side jet charge tagging algorithm calculates a @xmath39-weighted average charge of tracks : @xmath40 .
a @xmath41 value corresponds to a @xmath32 quark for the b meson on the opposite side , and vice - versa .
the same - side pion tagging algorithm makes use of the fact that the charge of the fragmentation pion correlates with the flavor of the reconstructed b meson .
a positive pion corresponds to a @xmath42 quark ( i.e. @xmath16 ) if the reconstructed @xmath29 meson is neutral , but to a @xmath32 quark ( i.e. @xmath43 ) if the reconstructed @xmath29 meson is charged , and vice versa .
oscillations have been seen using both the opposite - side jet - charge and the same - side pion taggers , and work is ongoing to compute systematics errors and optimize performance .
lastly , since the significance scales as @xmath44 , for large values of @xmath10 , a precise measurement of the proper decay time is crucial . efforts are ongoing in this direction and we hope to have preliminary @xmath10 results soon .
a preliminary measurement of @xmath2 has been made with the upgraded d detector .
the opposite - side soft muon tagger was used to obtain @xmath36 ps@xmath15 .
this measurement , along with the development and optimization of other tagging algorithms , and the reconstruction of different @xmath45 decay modes , is an important step towards a @xmath0 mixing measurement .
d.abbaneo , `` review of experimental results on neutral b meson oscillations , '' arxiv : hep - ex/0012010 .
et al . _ , `` the review of particle physics by the particle data group , '' phys
. lett . * b592 * , 1 ( 2004 ) .
o.schneider , `` @xmath46 mixing , '' arxiv : hep - ex/0012010 . | measurement of the @xmath0 oscillation frequency via @xmath0 mixing analyses provides a powerful constraint on the ckm matrix elements .
the study of @xmath1 oscillations is an important step towards @xmath0 mixing and a preliminary measurement of @xmath2 has been made with @xmath3 250 pb@xmath4 of data collected with the upgraded run ii d detector .
different flavor tagging algorithms have been developed and are being optimized for use on a large set of @xmath0 mesons that have been reconstructed in different semileptonic decay modes .
= 14.5pt |
short gamma - ray bursts ( sgrbs ) are usually defined by the prompt duration , i.e. , gamma - ray bursts ( grbs ) with @xmath15 @xcite .
a significant fraction of the sgrbs is accompanied by longer duration ( up to @xmath1 ) extended emissions ( ees ) @xcite .
the isotropic energy of the prompt spike is @xmath16 @xcite , while that of ee is @xmath17 @xcite .
interestingly , in some cases , observed fluences of the ees are even larger than those of the prompt spikes , e.g. , @xmath18 for sgrb 050709 @xcite and @xmath19 for sgrb 080503 @xcite .
@xcite found that @xmath20 of _ swift _ bat sgrbs have ees in the x - ray band ( see their table .
3 ) . on the other hand
, @xcite searched ees in 296 batse sgrbs and found that the fraction of ees is @xmath21 , where they pointed out that this fraction should be regarded as the minimum value since dim and/or softer ees can not be detected by batse .
in fact , batse measures fluence above @xmath22 while bat does down to @xmath23 , and the 5 kev lower threshold energy yields an 18% increase in the ee population .
this suggests that a decrease in the threshold energy , say , as low as @xmath24 , might yield a dramatic increase in the ee population . here
, we consider the possible origin of such ees based on the compact binary ( neutron star ( ns)-ns or ns - black hole ( bh ) ) merger scenario @xcite .
if the maximum mass of a non - rotating ns is smaller than @xmath25 , the final outcome of such a merger will be a kerr bh with a mass of @xmath26 and a spin parameter of @xmath27 with an accretion disk with a mass of @xmath2 and possibly beyond , and a neutron - rich ejecta with a mass of @xmath6 with an expanding velocity of @xmath28 @xcite . here
, we adopt this situation .
if the maximum mass of a non - rotating ns is larger than @xmath25 , a rapidly rotating massive ns will be the final outcome and the magnetar activities may be responsible for the prompt spike and ees of the sgrbs as well as other electromagnetic counterparts @xcite . in this paper , we consider the huge rotational energy of a kerr bh to be the intrinsic energy budget of the ee ( see also * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
the mass formula of the kerr bh with a gravitational mass of @xmath29 and an angular momentum @xmath30 is written as @xcite @xmath31 where @xmath32 is the irreducible mass of the kerr bh .
writing @xmath33 , we have @xmath34 where @xmath35 is the well known kerr parameter and @xmath36 .
( [ eq : mass ] ) is rewritten as @xmath37 then , the energy available by the extraction of the angular momentum of the kerr bh is given by @xmath38 for @xmath39 , for example , @xmath40 is given by @xmath41 therefore , for an @xmath42 radiation efficiency , only @xmath43 of the rotational energy of the bh enables fueling of the ee even if the emission is isotropic .
the problem is how to extract the rotational energy of the kerr bh on a timescale of a few 10 s to @xmath1 .
one of the plausible mechanisms for extracting the rotation energy of the kerr bh is the blandford - znajek ( bz ) process @xcite . from the results of numerical simulations
, @xcite found that within factors of order unity , the bz power is expressed in units of @xmath44 by @xmath45 where @xmath46 and @xmath47 is the magnetic flux threading the horizon with @xmath48 being the strength of the magnetic field formed by the disk around the kerr bh .
recovering @xmath49 and @xmath50 , we have @xmath51 ^ 2 \left(\frac{gm_{bh}}{c^2 } \right)^2 c b^2.\ ] ] for @xmath39 , for example , @xmath52 is expressed as @xmath53 dividing eq .
( [ eq : del_e ] ) by eq .
( [ eq : l_bz ] ) , we have the characteristic time @xmath54 as @xmath55 this shows that if an accretion disk with @xmath56 and an accretion time @xmath1 exists , up to @xmath57 erg can be extracted from the kerr bh .
this is just the timescale of the ees and only @xmath43 efficiency is enough to explain the emissions even if they are isotropic .
this paper is organized as follows . in 2 ,
we analyze the time evolution of the neutrino - dominated accretion disk with finite mass and angular momentum around the bh , and estimate the resultant bz power and the duration . in 3 , we consider the interaction of the bz jets with the pre - ejected matter ( @xmath6 ) with an expanding velocity of @xmath28 , which produces mildly relativistic fireballs .
we calculate the dissipative photospheric emissions from such fireballs .
there , we also argue the detectability and the association with the observed ees .
4 is devoted to the discussion .
we use @xmath58 in cgs units unless otherwise noted .
@xcite calculated the steady - state solutions of a neutrino - dominated accretion disk around a kerr bh , which is also the case in our setup at the initial stage .
they assumed an accretion rate of @xmath59 constant , but took into account the full neutrino process and the kerr geometry .
their important conclusions are that ( i ) the pressure is dominated by baryons with @xmath60 , ( ii ) the disk is neutron dominated so that the electron fraction is as small as @xmath61 , ( iii ) the degeneracy of the electron is at most mild because , if the degeneracy is high , the neutrino cooling is lowered to increase the temperature , ( iv ) there is an ignition accretion rate for the neutrino cooling disk that is proportional to @xmath62 where @xmath63 is the parameter in the so - called @xmath63-disk model @xcite .
@xcite performed simpler newtonian calculations of such disks both numerically and analytically .
one of the important conclusions is that their analytical model fits well with the numerical ones by @xcite .
therefore , here we adopt a simple newtonian analytical model to mimic the neutrino - dominated accretion disk around the kerr bh .
one of the big differences from @xcite is that we take into account the time variation of the accretion rate for the finite disk mass and the finite disk angular momentum while they considered a constant accretion rate .
the structure of the accretion disk can be derived from @xcite @xmath64 where @xmath65 and @xmath66 are the accretion rate , @xmath67 in cylindrical coordinates , the density , the half thickness of the disk , the infalling velocity , the @xmath63 parameter , the pressure , and the angular frequency of the disk , respectively .
we can express @xmath68 by @xmath69 as @xmath70 @xmath71 @xmath72 where @xmath73 is the innermost stable circular orbit ( isco ) .. however we are using newtonian gravity so that the exact treatment of isco is not possible .
one can take into account the change of isco by putting different value of @xmath74 in all the equations . in this case , various quantities are modified by powers of the above factor , but our results do not change qualitatively . ]
the density @xmath69 can be determined by the energy equation and the equation of state . denoting the cooling rate @xmath75 , the energy balance is expressed as @xmath76 as for the energy loss rate , we consider two neutrino cooling processes relevant to the accretion disk we are interested in as @xcite @xmath77 where @xmath78 is the temperature in units of [ k ] .
the urca process is @xmath79 and the pair neutrino process is @xmath80 .
the urca process dominates over the pair process for @xmath81 . as for the pressure of the matter we should consider @xmath82 @xmath83 @xmath84 where @xmath85 is the radiation constant .
the relativistically degenerate pressure dominates over the gas and the radiation pressure for @xmath86 while the gas pressure dominates over the radiation pressure for @xmath87 let us consider the case for @xmath88 , where the pressure is determined by gas pressure and the cooling process is dominated by the urca process so that eqs .
( [ eq : q_eq ] ) and ( [ eq : q_urca ] ) give @xmath89 with @xmath90 . from eqs .
( [ eq : p_gene ] ) , ( [ eq : eos ] ) , and ( [ eq : cool_eq ] ) , @xmath69 is expressed as @xmath91 we then have @xmath92 @xmath93 @xmath94 @xmath95 @xmath96 @xmath97 then , the surface density of the disk ( @xmath98 ) is given by @xmath99 let us introduce the coordinate @xmath100 by @xmath101 .
we assume here that the disk has a minimum and a maximum @xmath100 as @xmath102 and @xmath103 , respectively . then for a given total mass @xmath104 and the total angular momentum @xmath105
, we have @xmath106 for a given value of @xmath104 , @xmath105 , @xmath63 , and @xmath29 , from eqs .
( [ eq : m_d ] ) and ( [ eq : j_t ] ) , we can determine @xmath107 and @xmath103 in general .
let us define a new variable @xmath108 by @xmath109 , that is , the mean value of the angular momentum is @xmath110 times the minimum value of the specific angular momentum at the isco .
then , @xmath103 is determined by @xmath111 it is easily shown that the left - hand side of eq .
( [ eq : x_max ] ) is a monotonically increasing function for @xmath112 and has a minimum value @xmath113 at @xmath114 so that for an arbitrary value of @xmath115 , there is an unique solution @xmath112 . in the accretion process
, the total angular momentum of the system should be conserved in our case , since the kerr metric has a rotational killing vector , i.e. , a stationary axisymmetric system .
some of the angular momentum is absorbed by the bh from the isco so that @xmath108 increases as a function of time .
note that the spin up of the bh due to accretion is negligible in our case .
if we denote the mass and the angular momentum of the accreted blob into the bh as @xmath116 and @xmath117 , we have @xmath118 the solution of eq .
( [ eq : beta ] ) is given by @xmath119 where @xmath120 and @xmath121 are the initial values of @xmath108 and @xmath104 , respectively .
therefore , @xmath122 in the later phase of the accretion so that @xmath123 and the @xmath124 ^ 2 $ ] will be a good approximation . inserting this expression to eq .
( [ eq : m_d ] ) , we have @xmath125^{-8/5}\alpha_{-1}^{6/5}\left(\frac{m_{bh}}{3 \ m_\odot}\right)^{-6/5}.\ ] ] integration of eq .
( [ eq : d_m_d ] ) yields @xmath126 the method we adopted here to solve the evolution of the accretion disk is similar to the quasi - static evolution of the star where the nuclear timescale is much longer than the free fall time so that at each time , the star can be regarded in gravitational equilibrium ( see , e.g. , * ? ? ?
* ) . in our case , from eq .
( [ eq : v_ratio ] ) , the accretion velocity is much smaller than the kepler velocity which determines the dynamical timescale so that we can regard the disk as stationary at each time .
the decrease in total mass and angular momentum can be regarded as a decrease in total nuclear energy and a change of the composition in the stellar evolution case , which are very slowly changing in a dynamical timescale .
let us assume that @xmath127 , @xmath128 , @xmath129 , and @xmath130 as suggested by numerical relativity calculations @xcite .
solid lines in fig .
[ mdot-1 ] show the accretion rates as a function of @xmath63 for the representative time such as 1 s , 3 s , 10 s , 30 s , 100 s , and 300 s. we are interested in the late time behavior ( @xmath131 s ) where the accretion rate decreases as a function of @xmath63 for a fixed time .
this is because @xmath132 in eq .
( [ eq : a ] ) is smaller for larger @xmath63 so that the accretion rate is smaller . physically , the accreting velocity is larger for larger viscosity as is clear from eq .
( [ eq : v_r ] ) , where the consumption of the disk mass is faster .
the dashed lines are the neutrino cooling ignition accretion rate obtained by @xcite for @xmath133 and @xmath134 ( eq .
42 of their paper ) .
neutrino cooling is effective only above these dashed lines . in fig .
[ l_bz ] , we show bz luminosity as a function of @xmath63 for typical time such as 1 s , 3 s , 10 s , 30 s , 100 s , and 300 s for @xmath127 , @xmath128 , @xmath129 , and @xmath130 . the magnetic field at the horizon is assumed to be determined by @xmath135 where @xmath136 is the pressure of the disk at the isco in eq .
( [ eq : p ] ) .
the bz luminosity is then given by eq .
( [ eq : l_bz ] ) .
we clearly see that the bz luminosity is higher for smaller @xmath63 for the same time .
this comes from both the strong @xmath63 dependence of the @xmath137 in eq .
( [ eq : p ] ) and the accretion rate in fig .
[ mdot-1 ] .
physically , if the viscosity is low , the accretion rate decreases slowly and the pressure is high due to the accumulation of matter , which yields a strong magnetic field in our scenario .
all these effects result in a larger bz luminosity for a smaller @xmath63 .
one might suspect that our approximation ( @xmath138 ) to the solution of eq .
( [ eq : x_max ] ) for given @xmath108 affects the result . to check this ,
we show in fig .
[ check_lc ] the time evolution of the luminosity for @xmath127 , @xmath128 , @xmath129 , @xmath130 , and @xmath139 .
the red and blue solid lines show the numerical solution , which is obtained by integrating eqs .
( [ eq : m_d ] ) , ( [ eq : j_t ] ) , and ( [ eq : x_max ] ) directly over time , the analytic approximation ( eq . [ eq : m_d_ana ] ) .
the dashed horizontal lines show the neutrino cooling ignition accretion rates for @xmath140 ( upper ) and @xmath141 ( lower ) , respectively .
we see that the analytic approximation only slightly overestimates the luminosity .
we can justify the use of the analytic approximate solution to eq .
( [ eq : x_max ] ) .
it is suggested that the typical @xmath142 from various numerical simulations @xcite . to have possible isotropic ee ,
the typical luminosity of @xmath143 is needed at @xmath144 a few 10 s. from fig .
[ mdot-1 ] , if @xmath145 , which is relatively smaller than that suggested by various numerical simulations , the accretion rate is above the ignition rate for neutrino cooling up to the observed time of @xmath146 s , and the luminosity can be above @xmath147 so that enough luminosity for ees seems to be obtained via the bz mechanism . however , from fig .
[ mdot-1 ] , the accretion rate is above the ignition rate for neutrino cooling only up to @xmath148 s. in this case the duration of ee is at most @xmath149 s. to increase the duration of ee , we need smaller @xmath63 than 0.01 .
when can the value of @xmath63 become smaller ? if the origin of viscosity is magnetic turbulence from the magneto rotational instability , there is a big difference from the usual situations , that is , the matter is neutron rich ( @xmath61 ) .
since neutron does not directly couple to the magnetic field , and @xmath61 , only 10% of the mass feels the viscosity first .
note that , for @xmath150 , @xmath151 , @xmath152 , which are typical values at the isco , the gyration radius of the proton @xmath153 cm is comparable to the mean free path of p - n collision with the cross section of @xmath154 .
then , the effective @xmath155 might be reduced , in principle , 10% compared to the usual case of proton (= hydrogen ) dominated gas , i.e. , @xmath156 in the conventional definition could give @xmath157 .
then , from fig
. [ mdot-1 ] , the duration can be even longer than @xmath158 s and the bz luminosity can supply an enough energy for the isotropic ee .
these arguments suggest that the rotational energy of the kerr bhs formed in compact binary mergers might supply the energy sources of the ees with the required duration even if the emissions are isotropic . in newtonian gravity , a numerical calculation of the accretion disk exists for the present problem , i.e. , the time evolution of an accretion disk of mass @xmath159 after the merger of an ns - ns binary @xcite .
our treatment also uses newtonian gravity so that we can compare our analytic results with their numerical calculations to confirm the quantitative agreement . in their calculations ,
the mass of the central bh is @xmath160 and they solved the time evolution of a z - direction integrated quantity such as the surface density @xmath98 with neutrino and advection cooling .
the initial surface density is given as @xmath161 with @xmath162 .
since @xmath163 peaks at @xmath164 , the initial specific angular momentum is @xmath165 . in our crude model , we assumed the disk boundary is at @xmath166 while in their calculation inner edge (= disk boundary ) is @xmath167 so that we define the coordinate @xmath100 by @xmath168 in this paragraph . in the newtonian calculation
, there is no isco so that the minimum specific angular momentum of their calculation is @xmath169 which is different from our model of @xmath170 . as for @xmath63
, they adopt 0.3 so that we need to rewrite eqs .
( [ eq : m_d ] ) and ( [ eq : j_t ] ) as @xmath171 defining @xmath108 by @xmath172 , we have the same equation as eq .
( [ eq : x_max ] ) .
the argument for deriving equations corresponding to eqs .
( [ eq : beta ] ) and ( [ eq : beta_int ] ) is also the same by changing @xmath173 and @xmath174 , which gives @xmath175 integration of eq .
( [ eq : m_d_metzger ] ) yields @xmath176 now let us compare eq .
( [ eq : m_d_metzger_int ] ) with those of the numerical calculations by @xcite .
their fig .
3 shows an accretion rate at @xmath177 for @xmath178 and @xmath179 are @xmath180 , @xmath181 and @xmath182 , respectively .
( [ eq : m_d_metzger_int ] ) yields @xmath183 , @xmath184 and @xmath185 .
we can say that our analytic model agrees rather well with the numerical calculations at isco especially for later times , which is indispensable for the use of our analytic model to study ees .
it has been shown that , in the late phase of the accretion of dense debris such as we consider here , the disk wind driven by energy injection via viscous heating and the recombination of nucleons into alpha - particles becomes relevant @xcite .
although our calculation does not include this effect , since such outflows are predominantly triggered after the viscous timescale of the disk , our results can be still viable up to this timescale , e.g. , @xmath186 for @xmath145 ( see fig .
in the previous section , we showed that the rotational energy of the kerr bh up to @xmath187 can be extracted as the poynting outflow via the bz process with a timescale of @xmath188 if the accretion of the debris @xmath189 occurs with @xmath145 .
hereafter , we argue the resultant emissions from such outflows and their detectability . in the course of compact binary mergers , a fraction of baryons of mass @xmath6 can be ejected with an expansion velocity of @xmath190 @xcite . a certain duration after the merger ,
say @xmath191 , the hypermassive ns collapses into a bh due to the loss of rotational support by emitting gravitational waves ( gws ) and/or poynting fluxes .
the poynting outflow by the bz process , which is relativistic , clashes with the pre - ejecta . the bz outflow or jet will be more or less beamed and drill through the pre - ejecta , forming a hot plasma cocoon surrounding the jet .
recently , such a situation has been investigated numerically @xcite , although the jet injection timescale is set to be @xmath192 s , considering jets responsible for prompt emissions of sgrbs .
these studies show that the jet dynamics are significantly affected by the pre - ejecta , especially for a jet luminosity of @xmath193 and a pre - ejecta mass ejection rate of @xmath194 , which we are interested in here .
in particular , if the jet launch is delayed more than @xmath195 s from the pre - ejecta launch , such a jet will be choked in the ejecta and a significant fraction of its energy will be dissipated inside the ejecta , forming a cocoon fireball . in our case ,
the assumed duration of the outflow injection is @xmath196 s , thus , the bz outflow would more easily penetrate the pre - ejecta if the onset time of outflow injection is not delayed significantly . nevertheless , even after penetration , a fraction of the bz - outflow energy can dissipate at the interaction surface with the pre - ejecta or the cocoon , which typically occurs at @xmath197 where @xmath54 corresponds to the time @xmath198 in the previous section .
note that the bz outflow is most likely magnetically dominated at the launching radius . in the case of magnetically dominated jets , the dynamics including the cocoon formation are different from those of hydrodynamical jets ( e.g. , * ? ? ?
* ) , and the energy dissipation process is rather uncertain .
hereafter , we simply parameterize such a dissipation process by the fraction of dissipated energy , @xmath199 and the beaming factor of the dissipation region , @xmath200 . after dissipation
, the heated pre - ejecta can be regarded as a fireball .
the temperature can be estimated as @xmath201 , or @xmath202 we note that the optical depth at around this radius is large , @xmath203 . here , @xmath204 is the opacity of the thomson scattering , @xmath205 is the mean density of the ejecta , and @xmath206 is the isotropic mass ejection .
though the mass ejection , in general , is unisotropic ( e.g. * ? ? ?
* ) , we can take into account this effect by changing @xmath200 and @xmath207 appropriately .
the fireball is accelerated due to the large internal energy , and the lorentz factor saturates at @xmath208 , or @xmath209 which occurs at @xmath210 hereafter , we simply assume @xmath211 in the acceleration phase . , the fireball is still magnetically dominated , and the evolution of the lorentz factor is generally different from the above scaling ; @xmath212 with @xmath213 . ] for @xmath214 , the fireball moves as a shell with a shell width @xmath215 so that the temperature decreases as @xmath216 , while in the lateral direction @xmath217 , it expands as @xmath218 ( e.g. , * ? ? ?
then , the fireball begins to expand almost spherically irrespective of the initial beaming angle beyond the radius given by @xmath219 the temperature decreases as @xmath220 for @xmath221 .
the photospheric emission from the fireball can be expected around the photospheric radius , @xmath222 the temperature of the fireball in the comoving frame evolves as @xmath223 for @xmath224 , @xmath216 for @xmath225 , and @xmath220 for @xmath226 . from eqs .
( [ eq : r_exp ] ) and ( [ eq : r_ph ] ) , @xmath227 is satisfied if @xmath207 is larger than the critical mass given by @xmath228 in the case of _ dirty _ fireballs with @xmath229 , @xmath227 is expected .
the temperature at the photospheric radius becomes @xmath230 , which is expressed as @xmath231 and the peak photon energy of the resultant photospheric emission can be estimated as @xmath232 , which is given by @xmath233 the peak intensity in the comoving frame can be approximated as @xmath234 , where @xmath235 with the planck constant @xmath236 .
is not the half thickness of the disk but the planck constant .
also we use @xmath108 as the spectral index in this section . ] in the observer frame , using the fact that @xmath237 is lorentz invariant ( e.g. , * ? ? ?
* ) , the corresponding energy flux is given by @xmath238 with @xmath239 and @xmath240 being the luminosity distance and the relativistic beaming effect , respectively , as @xmath241 @xmath242 in general , the observed duration of the photospheric emission is given by @xmath243.\ ] ] in the case of dirty fireballs , @xmath244 ( see eqs . [
eq : r_o ] and [ eq : r_exp ] ) , and thus the first term on the right - hand side of the above equation always larger than the second one . then , @xmath245 we remark that ee durations of @xmath246 do not necessarily require durations of the bz jet of @xmath247 if the fireball is dirty and the duration is determined by eq .
( [ eq : t_dur_dirty ] ) . for a fiducial parameter set ( @xmath248 ) , the emission is characterized by @xmath249 , @xmath250 , and @xmath251 from @xmath252 , and @xmath253 , @xmath254 , and @xmath255 from @xmath256 . in general
, the spectral shape is determined by additional dissipation processes , e.g. , internal shocks or magnetic reconnections occurring in @xmath257 , which may slightly boost the peak energy and most likely produce a quasi - thermal spectrum , @xmath258 for example , @xcite numerically calculated the photospheric emission from fireballs ( in their cases , hot - plasma cocoons ) including some dissipation processes at @xmath259 , and they showed that typically @xmath260 and @xmath261 .
then , for our fiducial parameters , @xmath262 from @xmath252 so that the spectrum may range from @xmath263 to @xmath264
. we plot the possible @xmath265 spectra in fig .
[ fig : ee_obs ] .
although @xmath266 has parameter dependancies like eq .
( [ eq : ee_peak_dirty ] ) , it is most likely that @xmath267 is below @xmath23 for @xmath229 and @xmath268 . in this case , the emission energy is outside of the coverage of batse and _ swift _ bat so that such ees have not been detected so far . on the other hand , the soft ees can be detected by soft x - ray survey facilities like wide - field maxi , which has a @xmath269 angular resolution .
let us estimate the possible detection rate of the soft ees discussed above , in particular , simultaneously with the gws from compact binary mergers .
@xmath270 where @xmath271 is the fraction of the soft ees in all sgrbs , and @xmath14 represents the ns - ns merger rate within the detection horizon of advanced ligo , advanced virgo , and kagra , @xmath272 , which corresponds to a redshift of @xmath273 and a comoving volume of @xmath274 .
we assume the standard @xmath275-cdm cosmology .
note here that @xmath276 can be expected given that the fraction of an ee burst is significantly larger in softer energy bands ; @xmath20 in the _ swift _ bat samples ( @xmath277 ) and @xmath21 in the batse samples ( @xmath278 ) .
we take into account the sky coverage of wide - field maxi @xmath279 and the anticipated duty cycle , @xmath280 for wide - field maxi and @xmath281 for the 2nd generation gw network . in the case of dirty fireballs
, the beaming factor can be relatively large , e.g. , @xmath282 , and the above estimate gives @xmath283 for @xmath284 and @xmath285 . with a planned detection threshold flux of wide - field maxi @xmath286 , such soft ees can be detectable from @xmath287 , which corresponds to a luminosity distance of @xmath288 and a comoving volume of @xmath289 .
the anticipated total detection rate can be estimated as @xmath290 next let us consider the observed ees in our scenario . as we argued above
, the dissipative photospheric emissions from dirty fireballs are too soft and too dim for _ swift _ bat as far as @xmath291 and @xmath268 .
importantly , the typical beaming angle of the observed ees can be estimated to be much smaller as follows .
the ees are being detected predominantly by _
swift _ bat with the image trigger . here
we set the trigger threshold as @xmath292 in 64 s @xcite . in this case , the trigger threshold by bat for a burst with @xmath293 can be calculated as @xmath294 . for an ee with a mean luminosity of @xmath295 ( like sgrb 061006 ) ,
the detection horizon also becomes @xmath296 , corresponding to @xmath297 and @xmath298 .
for an effective total observation time for bat of @xmath299 , and a sky coverage of @xmath300 , the total number of detectable ees becomes @xmath301 .
given that 14 ees have been identified in this interval @xcite , the fraction of ee bursts @xmath302 and the beaming factor @xmath303 can be constrained as @xmath304 , or @xmath305 which is small compared to the inferred beaming factor of the prompt spikes , @xmath306 @xcite .
we should mention that candidates of orphan " ees without prompt spikes , i.e. , long grbs with @xmath307 whose redshifts and host galaxies are not identified , have been detected by _
swift _ bat .
the detection rate of those candidates is roughly comparable to that of sgrbs with ees ] .
given this fact , the constraint on the beaming factor ( eq . [ eq : omega_ee ] ) is a lower limit and can be larger by a factor .
nevertheless , the possible simultaneous detection rate of such ees and gws from ns - ns mergers would be quite small , @xmath308 .
note that the estimated value is independent of the relatively uncertain binary ns merger rate . in the context of our picture
, a smaller @xmath200 corresponds to a relatively narrow fireball .
we note that the mass ejection associated with the bh - ns merger is orders of magnitude smaller in the polar direction than the angle averaged one , which has been confirmed by numerical simulations @xcite where the bh spin is set to be parallel to the orbital angular momentum . in general , it is expected that they are misaligned due to the kick velocity in the formation process of nss .
although more numerical simulations using general initial conditions are needed to know what happens in bh - ns mergers , the simulations by @xcite suggest that the matter along the bz jet axis is small compared with an ns - ns binary . as a result
, one can expect that a smaller fraction of energy is dissipated in a smaller region of the pre - ejecta compared with the previous ns - ns cases where the outflows via the bz process clash with denser pre - ejecta .
hereafter , we take @xmath309 , @xmath310 , @xmath311 as a fiducial value .
such fireballs are _ clean _ in that @xmath312 , i.e. , @xmath313 the lorentz factor of such clean fireballs and the comoving temperature at the photospheric radius are @xmath314 and @xmath315 , respectively , and the peak energy of the photospheric emission @xmath316 can be estimated as @xmath317 from eq .
( [ eq : f_peak_dirty ] ) , the corresponding peak flux is @xmath318 in the case of clean fireballs , the emission duration is @xmath319 . for a fiducial parameter set ( @xmath320 ) , the emission is characterized by @xmath321 , @xmath322 , and @xmath323 from @xmath324 ( see fig . [
fig : ee_obs ] where we also plot an observed flux spectrum of ee associated with sgrb 061006 at @xmath324 ) .
the observed ees can be consistently interpreted as the high energy tail of the photospheric emission from such a clean fireball . to test this scenario ,
simultaneous detections in the soft x - ray band ( @xmath325 ) are crucial .
as for the ees from fireballs , we need to model , e.g. , the subphotospheric dissipation processes in detail to predict the spectra more precisely .
nevertheless , a key message here is that the typical energy of ees is likely to be in soft x - ray bands . in our scenario
, this is essentially due to the relatively large launching radii of the fireball , @xmath326 ( eq . [ eq : r_o ] ) compared with that of the conventional grb fireball , @xmath327 , which results in a lower initial temperature ( eq . [ eq : t_o ] ) .
the importance of soft x - ray bands has been implied from the observed soft photon index of ees and the increase of the fraction of sgrb with ees in softer bands .
we strongly encourage soft x - ray survey facilities like wide - field maxi , which can provide a useful electromagnetic counterpart to gws from compact binary mergers with an angular resolution of @xmath12 .
such ee counterparts are also important in terms of time domain astronomy since they would be observed only @xmath328 after the mergers . if a larger detection rate as eq .
( [ eq : rate_soft_ee ] ) is realized , a statistical technique using a stacking approach might also be possible for the detection of gws , with the aid of soft ee counterparts . in our scenario , an ee duration of a few @xmath329
is attributed to a relatively small disk viscosity of @xmath145 and the effect of the disk spreading during the accretion . on the other hand
, such a relatively long duration also may be realized if the disk accretion is suspended , but still the rotation energy of the bh is extracted via the interaction between the bh and the disk magnetosphere@xcite .
so far , we have focused on the emission mechanism of ees , and not discussed that of the initial spike . in our picture , the initial spike can be provided by the initial inhomogeneity in the ejecta .
if there is a direction with low column density , either the bz jet , the neutrino - antineutrino pair annihilation jet , or the magnetic tower jet would cause the initial spike .
whatever origin of the initial spike is , our point here is that the major component of sgrbs can be the ees in terms of the energetics .
if a half opening angle of the outflow responsible for the initial spike is @xmath330 , the total energy of the initial spike can be two to three orders of magnitude smaller than that of the ees .
relatively soft ees without initial spikes might already have been detected by e.g. , swift , but misidentified as other types of events .
one might think that soft ees should have already been detected by maxi , though the rapid sky sweeping ( @xmath331 ) makes it difficult to identify the @xmath1 emissions .
our scenario can be more clearly tested by future soft x - ray observations .
we note that the pressure in eq.(29 ) is a factor of five larger than that estimated by the general relativistic calculation by @xcite , which is due to our newtonian treatment .
this causes a factor of five overestimate of the bz power through eq.(44 ) . to take into account this fact as well as the difference between our newtonian dynamics and the general relativistic one , we add a new phenomenological parameter @xmath332 in eq.(44 ) as @xmath333 fig .
5 shows the total energy of a bz jet as a function of @xmath63 for @xmath127 , @xmath128 , @xmath334 , and @xmath39 for three values of @xmath332 .
the total energy is obtained by integrating the bz luminosity over time with a the mass accretion rate that is larger than the ignition rate of neutrino cooling , which is determined by @xcite for @xmath141 ( eq .
42 of their paper ) .
the dashed lines show the rotational energy of a bh with @xmath39 and @xmath141 , respectively . for low @xmath63
, the figure shows that either the effect of back reaction is needed , @xmath332 is small or the disk accretion is suspended for a while as discussed in previous paragraphs in this paper .
therefore it is urgent to undertake more precise analysis for more reliable quantitative predictions .
nevertheless , qualitatively our newtonian model of ees presented in this paper is worthwhile . in the present paper , we treated the accretion disk in newtonian gravity for a constant accretion rate at first , then required a finite total disk mass and the total angular momentum .
then , we solved the evolution of the size of the disk and the accretion rate .
this treatment should be regarded as a crude approximation , and we should compare our results with the numerical simulations .
unfortunately , we could not find numerical simulations of the evolution of a finite sized accretion disk either in kerr or schwarzschild metric in the literature .
one of the basic problems here is the relativistic treatment of the @xmath63 viscosity .
a simple generalization of the navier stokes equation as the one in the text book , for example , _ fluid mechanics _ ( chapter 127 ) by landau & lifshitz violates the causality @xcite , i.e. , the information propagates with a speed faster than the light velocity .
the reason is simple and clear .
let us consider the non - relativistic one - dimensional diffusion equation for some quantity @xmath335 , @xmath336 where @xmath337 is the diffusion constant .
if we set the delta function source as @xmath338 the solution for @xmath339 is expressed as @xmath340 the above solution clearly shows that the initial disturbance at @xmath341 and @xmath342 propagates to any @xmath100 even for any very small value of @xmath343 , which means the causality is violated , i.e , the information propagates with infinite speed .
the simple rule of changing the non - relativistic equation into a general relativistic one such as changing the derivative to the covariant derivative and using the projection of the tensor with @xmath344 does not help to guarantee the causality .
we need to add several new terms with undetermined parameters to the basic equations such as done by @xcite .
if we can start from the general relativistic boltzmann equation , there is no problem as for the causality in principle .
however , in practice , we should treat the distribution function that depends on three coordinates and three momenta , but it is beyond the ability of the present computer power to simulate such a problem .
one may think that the general relativistic resistive mhd simulations in three dimension are enough to solve this problem since the causality is not violated in such a system .
however , the boltzmann equation should be solved anyway since the gyration radius of the proton is typically comparable to the mean free path of p - n collision at around the isco .
the authors thank kunihito ioka , tsvi piran , norita kawanaka , kenta hotokezaka , hiroki nagakura , koutaro kyutoku , peter meszaros , peter veres , shuichiro inutsuka , makoto takamoto , and sanemichi takahashi for useful comments .
this work is supported in part by the grant - in - aid from the ministry of education , culture , sports , science and technology ( mext ) of japan , no.23540305 ( t.n . ) , no.24103006 ( t.n . ) , no.23840023 ( y.s . ) and jsps fellowship ( k.k . ) . | we investigate the possible origin of extended emissions ( ees ) of short gamma - ray bursts with an isotropic energy of @xmath0 and a duration of a few 10 s to @xmath1 , based on a compact binary ( neutron star ( ns)-ns or ns - black hole ( bh ) ) merger scenario .
we analyze the evolution of magnetized neutrino - dominated accretion disks of mass @xmath2 around bhs formed after the mergers , and estimate the power of relativistic outflows via the blandford - znajek ( bz ) process .
we show that a rotation energy of the bh up to @xmath3 can be extracted with an observed time scale of @xmath4 with a relatively small disk viscosity parameter of @xmath5 .
such a bz power dissipates by clashing with non - relativistic pre - ejected matter of mass @xmath6 , and forms a mildly relativistic fireball .
we show that the dissipative photospheric emissions from such fireballs are likely in the soft x - ray band ( @xmath7 ) for @xmath8 possibly in ns - ns mergers , and in the bat band ( @xmath9 ) for @xmath10 possibly in ns - bh mergers . in the former case
, such soft ees can provide a good chance of @xmath11 for simultaneous detections of the gravitational waves with a @xmath12 angular resolution by soft x - ray survey facilities like wide - field maxi . here
, @xmath13 is the beaming factor of the soft ees and @xmath14 is the ns - ns merger rate detectable by advanced ligo , advanced virgo , and kagra . |
the holographic principle in field theory states that , in the presence of gravity , physics in a region of space - time can be encoded on a lower - dimensional surface .
the maldacena @xmath0 conjecture @xcite provides explicit examples for some of these holographic encodings .
for example , iib string theory on the background @xmath1 or @xmath2theory on the background @xmath3 are claimed to be mathematically equivalent ( dual ) to certain quantum field theories on the boundary at infinity of the relevant @xmath4 space .
the isometry groups of the @xmath4 spaces become conformal groups acting on the boundary quantum theories . in the case of @xmath5 ,
the quantum field theory is @xmath6 super yang - mills gauge theory with gauge group @xmath7 and coupling constant @xmath8 .
this particular yang - mills theory is also a conformal field theory @xmath9 . more specifically , if one introduces a generating function for correlation functions of operators @xmath10 in the yang - mills theory , then the @xmath0 correspondence states that @xcite , @xmath11.\ ] ] in this equation , on the left - hand side , we have the generating function with operators integrated against arbitrary density functions , @xmath12 , on the boundary of @xmath4 .
the boundary has dimension @xmath13 according to the version of the correspondence we consider .
the function @xmath14 $ ] is the string partition function with string coupling @xmath15 , evaluated in the @xmath4 background .
the fields , @xmath16 , satisfy the boundary conditions of approaching @xmath12 at infinity .
a limit can be taken to reduce the string theory to its supergravity limit .
this may be done by a combination of taking the string tension to infinity @xmath17 and sending @xmath18 .
a second limit is obtained by making the approximation @xmath19 of the supergravity partition function to @xmath20 , where the supergravity action @xmath21 $ ] is the action of the classical supergavity solution that satisfies the boundary conditions mentioned above .
although the strongest form of the correspondence is the string theory expression , in practice it is much easier to work with supergravity actions , and it is for this reason that most results have been obtained in this context . on the quantum field theory side of the correspondence , this approximation amounts to taking both @xmath22 and @xmath23 large .
the classical supergravity action should holographically capture the large-@xmath22 limit of strongly coupled quantum yang - mills theory .
a quite large amount of work has been produced in the last few years on the @xmath0 conjecture and an extensive introduction can be found in @xcite . in the @xmath0 correspondence
, the study of @xmath4 black holes has a direct interpretation in terms of the dual conformal field theory on its boundary .
the duality predicts that the retarded @xmath24 correlation functions are in one to one correspondence with green s functions on anti - de sitter space with appropriate boundary conditions @xcite .
furthermore , as mentioned in @xcite , it is assumed that a large static black hole in @xmath4 space corresponds to a thermal state in the @xmath24 on the boundary .
perturbing the black hole with a scalar field , for instance , is equivalent to perturbing this thermal state .
the perturbed system is expected , at late times , to approach equilibrium exponentially with a characteristic time - scale .
this time - scale is inversely proportional to the imaginary part of the poles of the correlators of the perturbation operator .
it seems that these relaxation time - scales are quite complicated to calculate in the @xmath24 , therefore their computation is conveniently replaced by the evaluation of the quasinormal frequencies in the @xmath4 bulk space . in @xcite and @xcite ,
the authors went beyond the scalar perturbation treated in @xcite and they considered electromagnetic and gravitational perturbations . in one of the most recent works on the subject , berti and kokkotas @xcite confirm and extend previous results on scalar , electromagnetic and gravitational perturbations of static @xmath4 black holes and analyse , for the first time , reissner - nordstrom anti - de sitter black holes and calculate their quasinormal frequencies .
naturally , perturbations associated with other kinds of fields than the scalars treated in @xcite will decay at different rates given by their quasinormal mode frequencies .
quasinormal modes have recently become relevant to quantum gravity for a different reason .
this arises from hod s proposal to describe quantum properties of black holes from their classical oscillation spectrum @xcite .
the quest for a consistent quantization of black holes was initiated many years ago by bekenstein @xcite who , based on semi - classical arguments , conjectured that the horizon area of a ( non extremal ) black hole should have a discrete eigenvalue spectrum .
the eigenvalues of this spectrum were expected to be uniformly spaced .
hod noted that the real parts of the asymptotic quasinormal frequencies of a static black hole of mass @xmath25 , as computed by nollert @xcite , can be written as @xmath26 , where @xmath27 is the black hole hawking s temperature .
he then exploited bohr s correspondence principle , asserting that transition frequencies at large quantum numbers should equal classical oscillation frequencies , to conjecture that variations in the black hole mass due to quantum processes should be given by @xmath28 .
finally , using the first law of black hole thermodynamics and the equivalence between entropy and the surface area , he deduced the equal spacing in the area spectrum of a static hole .
hod s conjecture has brought further motivations in the field of black hole physics and has provoked several attempts to evaluate highly damped quasinormal frequencies of kerr black holes .
it is , in fact , straightforward to ask whether this conjecture is valid for more general black holes than the static ones ; if asymptotic frequencies depend in these more general cases on black hole charge and/or angular momentum ; if the presence of a cosmological constant should imply a modification of the proposal . at the moment
these ideas of universality in hod s proposal have not been confirmed @xcite .
a brief outline of this paper is the following . in section
[ two ] we review some of the main features of @xmath4 spacetime and its static and rotating black hole solutions . in section
[ three ] we solve the radial teukolsky equation .
then , we concentrate on the angular equation and propose both a second order and a pade approximation of the associated eigenvalues . in section iv we introduce transformation theory and obtain the regge - wheeler equation .
these results are exploited for the explicit determination of the quasinormal frequencies and some considerations on their asymptotic spacing .
all the numerical results are reported in section [ numres ] .
we conclude by a final discussion of the results we have obtained and with some ideas for future work . in the appendix we give the explicit recurrence relation for the series solution of the radial teukolsky equation .
rotating black holes in four dimensions with a cosmological constant were first constructed by carter @xcite and demianski @xcite . in the asymptotically de sitter case ,
the black hole solutions have cosmological horizons .
thermal equilibrium is only possible when the black hole horizon has the same temperature as the de sitter horizon@xcite .
the asymptotically anti - de sitter case has no cosmological horizons and the black hole , even if it is rotating , can be in equilibrium with radiation out to the conformal boundary at infinity @xcite .
anti - de sitter space and the associated stationary black hole solutions can be described by a metric with radial component @xmath29 of the form @xmath30 the polynomial @xmath31 vanishes at the event horizon , which we shall denote by @xmath32 .
the event horizon is also a killing horizon , that is it is a surface on which a killing field @xmath33 become null . the surface gravity @xmath34 is defined in terms of @xmath33 by , @xmath35 for a stationary and axisymmetric black hole with timelike and axial killing fields @xmath36 and @xmath37 , the killing vector field @xmath33 is of the form @xmath38 the constant @xmath39 is the angular velocity of the horizon .
an explicit expression for the surface gravity of the event horizon is given by @xcite , @xmath40 the hawking temperature of the black hole is related to the surface gravity through the famous result , @xmath41 the version of the @xmath0 conjecture which concerns us here is the correspondence between @xmath2theory on @xmath42 and supersymmetric conformal yang - mills field theory on the conformal boundary at infinity .
crucial to the @xmath0 conjecture is the fact that the @xmath4 space and its related black hole solutions are conformal to a rotating version of the einstein static universe at infinity .
the angular velocity of the space at infinity is denoted by @xmath43 .
the combination @xmath44 form a set of thermodynamic variables @xcite . for kerr - ads spacetime with cosmological constant @xmath45 , @xmath46 in fig .
[ penroserads ] , we show the conformal compactification for the covering space of kerr-@xmath4 spacetime . the conformal boundary is the boundary of region i. the black hole event horizon is the largest real root of @xmath47 , @xmath48 where @xmath49 is the inverse of the anti - de sitter radius . from eq .
[ c4e2 ] and the definition of the surface gravity [ c4e0 ] , we derive an expression for the temperature @xmath27 , @xmath50 the rotation of the asymptotic space is given by @xcite @xmath51 fig .
[ c4f4 ] shows contours of constant @xmath34 in the @xmath52 and @xmath53 parameter space .
the @xmath54 solutions are extremal black holes and above this line in the diagram the solutions become naked singularities . fixing the value of the angular parameter @xmath55 and @xmath34 we have unique masses and therefore unique solutions for @xmath56 , but up to @xmath57 different solutions for @xmath58 .
black holes are plotted in the @xmath52 and @xmath53 parameter space . ]
the transformation from the metic parameters @xmath25 and @xmath55 to the thermodynamic variables becomes singular where the jacobian @xmath59 vanishes .
vanishing of the jacobian also corresponds to the divergence of the heat capacity at constant @xmath43 .
this critical line is plotted in fig .
[ c4f5 ] . another important feature of the parameter space is set by the condition @xmath60 .
if @xmath61 , then the killing field @xmath62 becomes non - timelike beyond a certain radius and the black hole radiation can not remain in thermal equilibrium .
this allows us to divide the paramter space into the four regions @xmath63 .
region @xmath64 is the region with no event horizons .
regions @xmath65 and @xmath66 both have negative specific heat .
region @xmath67 is the region of positive specific heat and @xmath60 .
and @xmath53 parameter space shows various stability limits , @xmath68 and the critical line where the thermal capacity @xmath69 diverges .
the parameter space is divided into regions : @xmath64 with only naked singularities ; @xmath65 where @xmath61 and @xmath70 ; @xmath66 where @xmath60 and @xmath70 and @xmath67 where @xmath60 and @xmath71 . ]
the full kerr @xmath4 metric can be found in @xcite , for instance , and it can be expressed as @xmath72 where @xmath73 and @xmath74 are spherical coordinates , with @xmath75 .
the forms @xmath76 and @xmath77 are @xmath78 and the metric functions are given by @xmath79 we shall make use later of the radial tortoise coordinate , defined by @xmath80 using the the definition of surface gravity given in eq .
[ c4e0 ] , applied to each of the roots of @xmath47 ( two of them complex ) , we can express @xmath81 in the form @xmath82 note that @xmath83 , with @xmath84 as @xmath85 as a consequence of the identity @xmath86 .
( in the ordinary kerr solution , @xmath87 as @xmath88 ) .
the generalisation of teukolsky s equations and their separation for asymptotically de - sitter and anti de - sitter spacetimes , was first derived by chambers and moss @xcite .
they looked for separable solutions for the weyl scalar of the form @xmath89 where @xmath90 can be found in @xcite , and @xmath91 is the angular mode number characteristic of the spherical harmonics .
( in @xcite , the time coordinate was scaled differently and the frequency was a factor of @xmath92 larger than the frequency which we are using here . ) in what follows we will drop the index @xmath91 when we talk about the angular function @xmath93 . following @xcite , we introduce new radial and angular derivatives @xmath94 with @xmath95 and @xmath96 defined as @xmath97 the angular and radial functions @xmath98 and @xmath99 , satisfy the two second order differential equations @xmath100 where @xmath101 is a separatin constant . in the non - rotating case ,
@xmath102 for @xmath103 .
let us start from the radial differential equation [ c4e5 ] , which can be re - written as @xmath104 this equation has four regular singular points at the four roots of @xmath105 ; one at the black hole outer horizon , @xmath32 , which is the largest root of @xmath105 , one at the black hole inner horizon , @xmath106 , and other two complex conjugate roots , @xmath107 and @xmath108 , which are unphysical roots .
this allows us to write @xmath109 as @xmath110 where @xmath111 is the inverse of the anti - de sitter radius .
next , we set @xmath112 so that we can write @xmath113 with @xmath114 note that @xmath115 are just the inverse of the roots of @xmath109 in terms of the new variable @xmath116 . using the same coordinate inversion we can rewrite @xmath95 as @xmath117 with @xmath118
it is also useful , for future reference , to re - write @xmath119 as @xmath120 in the new variable @xmath116 the radial equation becomes @xmath121 rescale @xmath122 as @xmath123 and insert it into equation [ c4e6 ] . after dividing everything by @xmath124
, we obtain @xmath125 rescale @xmath126 as @xmath127 where @xmath128 by writing the radial solution in this way we achieve the result of factoring out the behaviour of this function at the black hole event horizon @xmath129 .
substitute @xmath126 and its derivatives into [ c4e8 ] .
we finally get an equation in the function @xmath130 that reads @xmath131 the determination of the recurrence relation for the radial series solution @xmath130 can be found in appendix a of this paper .
the solution to the radial equation is strictly bound up with the determination of the eigenvalues @xmath132 of the angular equation , which appear explicitly in the coefficients of the recurrence relation associated with the series solution @xmath130 of the radial differential equation .
we could attempt to solve for both @xmath133 and @xmath101 simultaneously , but this would be a time - consuming procedure .
instead , we shall attempt to determine @xmath101 before we find @xmath133 . to this purpose
, we first propose a second order taylor approximation of @xmath101 for small @xmath134 .
we then improve this with the new numerical technique , already used in the context of kerr black holes @xcite , based on a pade approximating function for the separation constant .
we proceed in analogy with sasaki @xcite , who used a technique based on perturbation theory to evaluate the separation constant for kerr black holes in asymptotically flat spacetime .
our final result agrees with sasaki s result when we assume the cosmological constant to be equal zero .
we shall only consider @xmath135 .
let us start by rewriting the angular equation in the form @xmath136 since @xmath135 , this implies for @xmath96 the expression @xmath137 the equation can also be written as @xmath138 with @xmath139 first of all , it is straightforward to verify that when @xmath140 and @xmath141 we get from @xmath142 the expression we expect for static black holes in asymptotically flat spacetime , as can be verified from @xcite .
we then introduce the new variable @xmath143 , such that @xmath144 .
this gives @xmath145 next , we expand @xmath142 in powers of @xmath134 , @xmath146 where , @xmath147 we also expand @xmath148 and @xmath101 in powers of @xmath134 as @xmath149 @xmath150 with @xmath151 .
now , if we plug [ c4e9 ] , [ c4e10 ] and [ c4e11 ] into eq .
[ c4e12 ] , we have at the second order @xmath152 collecting the terms of the same order we obtain , @xmath153 the functions @xmath154 are the spheroidal harmonics of spin weight @xmath155 , for @xmath135 .
we will be interested in the explicit expressions of the functions @xmath156 , which can be represented using legendre polynomials @xmath157 , as @xmath158 we should mention that at the moment the spherical harmonics are not normalised .
it is convenient to normalise at least those we will be using .
the normalised spherical harmonics are @xmath159 since our purpose is to evaluate @xmath101 up to the second order , let us start with the calculation of @xmath160 in eq .
[ c4e11 ] . the first order correction to @xmath101
is obtained by multiplying equation [ c4e15 ] by @xmath156 from the left hand side and integrating it over @xmath161 .
this gives @xmath162 but , if @xmath135 , this integral is always zero for any value of @xmath91 as can , indeed , be verified in @xcite .
consequently , @xmath163 to obtain @xmath164 , we set @xmath165 we insert this expression into equation [ c4e15 ] , multiply it by @xmath166 and integrate over @xmath161 .
what we get is , @xmath167 the coefficients of the series on the left hand side , @xmath168 , can be derived from [ c4e17 ] and written as @xmath169 hence @xmath168 is generally non zero only for @xmath170 .
the second order correction , @xmath171 , is obtained by multiplying equation [ c4e16 ] by @xmath172 form the left hand side and integrating it over @xmath161 .
we find , @xmath173 but @xmath174\ ] ] and the sum is non zero only for @xmath175 .
it follows that , @xmath176\nonumber\\ & -&\left(c^{l-1}_{l}\right)^2\left[(l-2)(l+1)-(l-1)(l+2)\right ] + \int_{0}^{\pi}\hat s^{l}_{0}\;h_{2}\hat s^{l}_{0}\;d(\cos\theta).\end{aligned}\ ] ] if we , finally , evaluate the coefficients @xmath168 and the integral on the right hand side of [ c4e18 ] , when @xmath177 , what we obtain for @xmath171 is @xmath178 this gives us , for @xmath177 , up to the second order in @xmath134 , an expression for the separation constant @xmath101 that reads @xmath179 a similar calculation can be repeated for any other value of @xmath91 . for instance
, we have performed the same calculation for @xmath180 . in that case
, the taylor expansion of @xmath101 in the variable @xmath134 , up to the second order is @xmath181 in both cases , @xmath177 and @xmath180 , we recover the flat space results for @xmath101 as soon as we set @xmath140 , as can be checked in @xcite .
having showed how to obtain a second order approximation for the angular eigenvalues , we would like to partially overcome the incompleteness of this approach by a different scheme , which is based on the construction of a pade approximating function for @xmath101 .
the technique which we describe below has already been applied to find the quasi normal modes for kerr black holes in asymptotically flat space @xcite , and works well for all values of the parameter @xmath134 .
we will start from an analytical solution of the teukolsky angular equation in @xmath4 space - time .
this equation is @xmath182 if we replace the differential operators @xmath183 and @xmath184 with their explicit expressions [ c4e19 ] , we can rewrite [ c4e20 ] as @xmath185 this differential equation has a structure which is easily recognisable as being very similar to the structure of the radial equation , with the variable @xmath186 having assumed the role of @xmath187 times the radial variable @xmath188 .
it has four regular singular points and they correspond to the four roots of @xmath189 .
they are @xmath190 and @xmath191 .
as already done for the radial solution , it is possible to factorise out the leading behaviour at @xmath192 by rescaling the angular function @xmath148 as @xmath193 the variable @xmath194 , which is analogous to the radial tortoise coordinate @xmath195 , is defined by @xmath196 while @xmath197 this allows us , through the same logic used to deal with the radial function @xmath122 , to get to this simpler form for the angular equation @xmath198 next , we expand all the coefficients of [ c4e22 ] as taylor series around @xmath199 and insert them into the equation itself beside the expression for @xmath200 and its derivatives .
we obtain a four term recurrence relation of the form @xmath201 finally , we find the coefficients of [ c4e23 ] in the explicit form @xmath202 this allows us to determinate the coefficients @xmath203 of the series solution by recursion .
equation [ c4e22 ] has critical points at the north and south poles @xmath204 .
we require the solution which does not diverge at either pole .
this solution has @xmath205 , @xmath206 and @xmath207 consequently , we have @xmath208 at this point , we wrote a _
c++ _ computer program able to evaluate values for @xmath209 , which are identified with the roots of the left hand side of eq .
[ c4e31 ] , numerically . within the program , the values of @xmath101 which satisfy eq .
( [ c4e31 ] ) are found by the use of the newton - raphson method routine extended to the complex plane @xcite .
we have fit these numerical data for @xmath101 to pade approximating functions for _ real _ @xmath210 at a fixed value of @xmath52 , given by @xmath211 finally , the pade approximation and the numerical data are compared for _ imaginary _ @xmath210 . in the pade formula ,
the numerator is one degree higher than the denominator .
the reason for this choice is that the teukolsky angular equation belongs to the family of differential equations for angular prolate spheroidal wave functions , and the behaviour of the angular eigenvalues for large @xmath133 is known @xcite . in fig .
[ p1 ] and [ p2 ] we show how well the pade function fits the data for @xmath101 coming from the computer program .
we also show the improvement compared to the second order approximation and we plot results for the real and imaginary parts of @xmath101 .
is shown to fit the data for lambda with imaginary @xmath133 .
note that the pade approximation was constructed for real @xmath133 .
the angular parameter @xmath212 , angular mode number @xmath177 and @xmath135 .
we also show the improvement we achieve with our pade function technique comparing our results to a second order approximation of @xmath101 . ]
is shown , for @xmath213 , @xmath177 and @xmath135 . ]
the fit parameters of the rational function representing @xmath101 are shown in table [ t1 ] for values of the angular parameter @xmath52 ranging between @xmath214 and @xmath215 , for the @xmath177 angular mode number .
.this table shows the parameters we obtain by fitting the pade approximation for @xmath101 , to the numerical data coming out from the computer program for the calculation of the angular eigenvalues of the teukolsky angular equation .
these results have been obtained by fixing the angular mode number @xmath177 and @xmath135 . [
cols="^,^,^,^,^,^,^",options="header " , ]
we have to impose some boundary conditions on the radial equation before we can obtain an equation for the frequencies of the quasinormal modes
. we shall do this by extending the transformation theory of perturbations of black holes to ads space , allowing us to construct the regge - wheeler - zerilli equation for the underlying the metric perturbations .
the transformation theory for kerr black holes has been described in detail by chandrasekhar @xcite . following the kerr case , we reparameterise the weyl function @xmath216 by @xmath217 where @xmath218
we also adopt a new @xmath219 coordinate , @xmath220 the teukolsky equation for @xmath126 becomes @xmath221 where @xmath222 , and @xmath223 is a polynomial @xmath224 part of the aim of transformation theory is to construct a function @xmath225 which satisfies the regge - wheeler - zerilli equation @xmath226 with a potential function @xmath227 .
the function @xmath225 represents the metric perturbations of the black hole .
the two radial functions @xmath126 and @xmath225 are related by an expression of the form @xmath228 where @xmath122 and @xmath27 are functions of @xmath188 and @xmath229 is a constant .
( in this section , we follow chandrasekhar s notation and @xmath27 does not denote the temperature . )
the consistency of the two equations ( [ teq ] ) and ( [ zeq ] ) allows one to derive expressions for @xmath122 , @xmath27 and @xmath229 , @xmath230 where @xmath231 finding the constants @xmath34 , @xmath232 and @xmath233 requires a substantial amount of algebra .
some help can be obtained by using the starobinski constant @xmath234 for kerr - ads given in @xcite .
we obtain , @xmath235 there are four consistent solutions corresponding to the choice of signs in the above expressions .
the final ingredient in the transformation theory is the potential , @xmath236 there are four different potentials depending on the choice of signs for @xmath232 and @xmath233 .
some general features of the potential are that it vanishes as @xmath237 and approaches a non - zero constant as @xmath238 ( @xmath239 ) .
the basic situation can be interpreted as scattering from @xmath240 to a boundary at @xmath241 .
the potential has poles at @xmath242 for real @xmath243 , where @xmath244 ( the potential can also have poles where @xmath245 . ) note that the @xmath219 variable is not a single valued function of @xmath188 when @xmath243 .
this frequency range corresponds to the super - radiant regime of kerr . for kerr - ads ,
the existence of super - radiant frequencies can be prevented by the boundary conditions at @xmath241 . in the axisymmetric case ,
@xmath135 , the potential is always continuous and bounded . as with kerr
, two of the potentials can be complex when @xmath232 is imaginary .
suppose that we are interested in the evaluation of some correlation function of a polar quantity in the conformal field theory which resides on the boundary of our @xmath4 space .
the @xmath0 correspondence conjectures that this polar quantity can be associated to the polar metric perturbation of an @xmath4 black hole living in the bulk .
accordingly we must set the axial black hole perturbation to vanish on the boundary .
since the axial perturbation function is the regge - wheeler function @xmath225 with @xmath246 , this also means that we are led to assume that this @xmath225 goes to zero as @xmath188 goes to infinity . on the other hand , we have series solutions for @xmath247 .
let us consider the boundary conditions on this function .
first of all , we have from eq .
[ c4e32 ] @xmath248 we find , for @xmath249 , @xmath250 and @xmath251 as @xmath85 . the differential operator @xmath252 as @xmath85 yields @xmath253 ( recall @xmath254 ) .
if we rescale @xmath126 as @xmath255 with @xmath130 representing once again the series solution , then we find @xmath256 which represents the boundary condition for the axial problem we are considering .
this equation tells us that assuming dirichlet boundary condition on the regge - wheeler function @xmath225 , we are left with robin conditions on the solution of our radial teukolsky equation .
moreover , since @xmath257 as @xmath258 ( @xmath259 ) , the boundary conditions can be explicitly written in the form @xmath260 in practice , @xmath261 , and the numerical solutions described in the next section have used @xmath262 , in which case the boundary conditions reduce to those used in @xcite .
the transformation theory which we have obtained is sufficiently powerful to give us the frequencies of some special quasi normal modes .
the teukolski function @xmath126 gives rise to four independent radial functions , from which we select @xmath263 with oposite signs for @xmath232 . working back from eq .
( [ c4e32 ] ) , we can relate the radial functions by @xmath264 for some functions @xmath265 and @xmath266 .
now we take @xmath133 to be a root of @xmath267 , and solve for @xmath268 to obtain a quasi normal mode .
the frequencies are @xmath269 note that this is an implicit equation in the rotating case , since the constants depend on @xmath133 . in the non - rotating case ,
the frequencies reduce to @xmath270 , which was found in @xcite . in the rotating case
, the frequencies can be pure imaginary or complex .
these values provide a useful check on the accuracy of the numerical results presented in the next section .
in this section we will present some numerical results obtained by using the procedures outlined above for the solution of both the radial and the angular teukolsky equations .
the radial boundary conditions used and the motivations for choosing them are those specified in section [ rbcond ] . we replace the angular eigenvalue @xmath101 by the pade approximation and
then the quasi - normal frequencies @xmath133 are roots of eq .
( [ bcz ] ) .
small kerr anti de - sitter black holes , with the radial coordinate of the event horizon @xmath32 much smaller than the ads radius @xmath271 @xmath272 , represent an unstable thermodynamic phase .
hence , we concentrate on larger black hole radii . in order to avoid metric singularities ,
the black hole rotation parameter is constrained to be less than the ads radius , @xmath273 .
the black hole parameters which we have used are displayed in fig .
[ pars ] . ]
the quasinormal frequencies are decomposed into real and imaginary parts , @xmath274 .
having chosen the metric perturbation with time dependence @xmath275 , we have that the pure imaginary component @xmath276 is positive for all quasinormal frequencies . fig .
[ freqlbh ] , shows the evolution of the quasinormal frequencies of large black holes ( lbh ) , with respect to the angular parameter @xmath277 , for @xmath278 .
[ freq1 ] shows the evolution of the quasinormal frequencies of intermediate size black holes ( ibh ) , with respect to the angular parameter @xmath55 , for @xmath279 . .
results are plotted for the angular parameter @xmath280 .
the squares represent the results for schwarzschild @xmath4 black holes .
the frequencies are rescaled by @xmath32 , so that @xmath281 . ] .
results are plotted for the angular parameter @xmath282 .
the squares represent the results for schwarzschild @xmath4 black holes .
the frequencies are rescaled by @xmath32 , so that @xmath281 . ]
to obtain both pictures , we have used the pade approximation for @xmath101 and therefore we have been free to consider frequencies as large as we liked .
the squares represent the results for static @xmath4 black holes .
unfortunately , the numerical procedure described in section [ pap ] becomes unreliable for values of @xmath55 higher than @xmath283 ( for lbh ) and @xmath284 ( for ibh ) , precluding us from extending the present results to very rapidly rotating holes .
the quasinormal frequencies for rotating black holes show a behaviour which is very similar to the one found in the case of schwarzschild black holes @xcite .
they have a real part which displays an increase along with the imaginary part .
the spacing between consecutive modes tends to become regular as the overtone number @xmath285 gets larger , and it seems to depend on the angular parameter @xmath55 . since in @xmath4 space the first ten frequencies are usually enough to ascertain their asymptotic behaviour , we have computed the quasinormal frequencies for @xmath286 in the case of lbh , and @xmath287 in the case of ibh , up to the tenth mode .
what we have obtained for the spacing in the lbh regime is , @xmath288 while in the ibh regime we have , @xmath289 for the smaller hole , there appears to be evidence that the real part of the spacing decreases with @xmath55 whilst the imaginary part of the spacing increases with @xmath55 . from figs .
[ pars ] and [ c4f4 ] we see that the temperature is increasing with @xmath55 in this region of parameter space .
both the real and imaginary parts of the frequencies become proportional to the surface gravity @xmath34 for large @xmath34 , which agrees with the behaviour for static black hole quasinormal modes @xcite .
this is displayed in figures [ par1 ] and [ par2 ] in which we show the dependence of the real and imaginary parts of the first @xmath290 quasinormal frequency on the metric parameters for some values of @xmath291 .
note that these pictures have been obtained by exploiting the second order approximation for @xmath101 , and that the results for @xmath292 are those obtained by moss and norman in @xcite .
the surface gravity reaches a minimum close to the critical line in fig .
[ pars ] on which @xmath69 diverges .
there is some numerical evidence in figures [ par1 ] and [ par2 ] that the quasi normal frequencies are constant along the critical line , with @xmath293 . and
@xmath294 , where @xmath32 is the radius of the event horizon , @xmath34 represents the surface gravity of the black hole and @xmath271 is the anti - de sitter radius .
the upper branch corresponds to larger values of @xmath32 than the lower branch . ]
there have been investigations on @xmath4 black holes particularly focused on their thermodynamic properties since the beginning of the eighties @xcite .
recently , anti - de sitter space has appeared in the surprising new context of the maldacena conjecture relating the physics of @xmath4 in @xmath295 dimensions to a strongly coupled conformal ( gauge ) field theory @xmath9 in one dimension fewer .
this interest has led to the study of static , charged and rotating black holes in @xmath4 space in five and even seven dimensions , where the @xmath0 correspondence can be still formulated .
references on these investigations can be found , for instance , in @xcite .
large kerr @xmath4 black holes are stable against hawking evaporation .
this property is not shared by asymptotically flat kerr black holes .
furthermore , kerr-@xmath4 black holes can have rotating killing vectors which are time - like everywhere outside the horizon . following the orbits of these rotating killing vectors
, thermal radiation can rotate around the black hole with the same angular velocity of the hole , implying that the phenomenon known as super - radiance is not allowed , and the black hole is in thermal equilibrium with the thermal gas around it @xcite .
although research has been carried on for a few years in the field of quasinormal modes of black holes in @xmath4 space and results have been obtained for schwarzschild and reissner - nordstrom black holes , it is in this work that for the first time that frequencies of kerr @xmath4 black holes are computed .
the possibility is that , through the @xmath0 correspondence , such frequencies can be exploited in the calculation of the relaxation times of the perturbation of the corresponding thermal state in the conformal field theory on the boundary
. it would be desirable to analyse the kerr @xmath4 black holes in deeper detail and see how the frequencies relate to the thermodynamical properties of the black holes and whether there is a direct interpretation in terms of the @xmath24 .
in particular , it would be interesting to look for special behaviour along the phase transition line and to examine the non - axisymmetric case to find out what happens to the super radiant modes .
it is also worth mentioning that , although black hole solutions in higher dimensions than @xmath296 have been found @xcite , the calculation of their quasinormal frequencies has never been done .
therefore an extension of our work to @xmath297 and @xmath298 , where the @xmath0 correspondence can be still formulated , seems possible .
finally , it would be perhaps of some interest to further check if the ideas on black hole area quantisation and the relationship to highly damped quasinormal frequencies , can be translated into non - asymptotically flat spacetimes , as anti - de sitter space .
in this appendix we describe in details how to derive the recurrence relation of the series solution @xmath130 for the radial teukolsky equation associated to spin-@xmath299 perturbations of kerr anti - de sitter black holes . starting from equation [ c430 ] for @xmath130 , we can write its derivatives with respect to @xmath116 as @xmath300 we re - write eq . [ c437 ] as @xmath301 next , consider expressions [ c425 ] and [ c426 ] for @xmath302 and @xmath303 and call @xmath304 and @xmath305 the coefficients of @xmath306 and @xmath130 in eq .
[ a4 ] , @xmath307 and let us develop all these coefficients as taylor series around @xmath308 , which is the inverse of the radial coordinate of the event horizon of the black hole .
we have , @xmath309 @xmath310 @xmath311 note that @xmath312 .
in fact @xmath313 and @xmath314 is , by definition , the largest root of @xmath315 .
when , in eq .
[ a4 ] , we replace its coefficients with their taylor expansions and insert into the same equation the series expressions for @xmath130 and its first and second derivatives , we obtain the recurrence relation , @xmath316 where the coefficients are given by @xmath317\nonumber\\ \beta_{n}&=&(-x_{1})^{-n}\left[\frac{1}{2}p''(x_{1})n(n-1)+c'(x_{1 } ) ( n-1)+d(x_{1})\right]\nonumber\\ \gamma_{n}&=&(-x_{1})^{-(n-1)}\left[\frac{1}{6}p'''(x_{1})(n-1)(n-2)+\frac{1 } { 2}c''(x_{1})(n-1)+d'(x_{1})\right ] \nonumber\\ \delta_{n}&=&(-x_{1})^{-(n-2)}\left[a^2(n-2)(n-3)-4a^2 ( n-2)+6a^2\right]\end{aligned}\ ] ] and we impose the initial conditions , @xmath318 and @xmath319 .
note that @xmath320 where @xmath34 is the surface gravity .
we re - scale all the parameters with respect to @xmath321 , so that @xmath322 etc .
the re - scaled mass @xmath323 and the re - scaled surface gravity @xmath324 substituting everything into the equations for the coefficients of our recurrence relation , we finally obtain @xmath325 notice is that , since we are considering anti - de sitter space with a negative cosmological constant @xmath326 , @xmath327 this is the same limit mentioned in @xcite for schwarzschild anti - de sitter black holes .
99 g. t hooft , gr - qc/9310026 ( 1993 ) l. susskind , j. math . phys . *
36 * , 6377 ( 1995 ) j. maldacena , adv .
* 2 * , 231 ( 1998 ) s.s .
gubser , i. r. klebanov , a. m. polyakov , phys . lett .
b * 428 * , 105 ( 1998 ) e. witten , adv .
* 2 * , 253 ( 1998 ) o. aharony , s.s .
gubser , j. maldacena , h. ooguri , y. oz , phys . rept . * 323 * , 183 ( 2000 ) j. l. petersen , int . j. mod .
phys . a * 14 * , 3597 ( 1999 ) u.h .
danielsson , e. keski - vakkuri , m. kruczeski , nucl .
b * 563 * , 279 ( 1999 ) s. kalyana rama , b. sathiapalan , mod .
a * 14 * 2635 ( 1999 ) d. birmingham , i. sachs , s. n. solodukhin , phys .
lett . * 88 * , 151301 ( 2002 ) d. t. son , a. o. starinets , jhep * 0209 * , 042 ( 2002 ) g.t .
horowitz , and v. hubeny , phys . rev . * d62 * , 0240027 ( 2000 ) v. cardoso , j.p.s .
lemos , phys . rev .
d * 64 * , 084017 ( 2001 ) i.g .
moss , j.p .
norman , class .
* 19 * , 2323 ( 2002 ) e. berti , k.d .
kokkotas , phys . rev . *
d67 * , 064020 ( 2003 ) s. hod , phys .
lett . * 81 * , 4293 ( 2003 ) j. bekenstein , lett .
nuovo cimento * 11 * , 467 ( 1974 ) j. bekenstein , _ cosmology and gravitation _ edited by m. novello ( atlassciences , france 2000 ) , pp
. 1 - 85 , gr - qc/9808028 ( 1998 ) h. p. nollert , phys .
rev d * 47 * , 5253 ( 1992 ) v. cardoso , j. natario , r. sciappa , hep - th/0403132 ( 2004 ) r. m. wald , _ general relativity _
( university of chicago press , chicago , 1984 ) g. w. gibbons and s. w. hawking , phys .
d * 15 * 2738n ( 1977 ) f. mellor and i. g. moss , class .
quantum grav .
* 6 * 1379 ( 1989 ) s.w . hawking , c.j .
hunter , m.m .
taylor - robinson , phys .
rev d * 59 * , 064005 ( 1999 ) m. m. calarelli , g. cognola and d. klemm , class .
* 17 * , 399 ( 2000 ) g. w. gibbons , m. j. perry and c. n. pope , hep - th/0408217 s. w. hawking , g. f. r. ellis , _ the large scale structure of space - time _ , ( cambridge university press , 1973 ) b. carter , commun .
* 10 * , 280 ( 1968 ) m. demianski , acta astron . * 23 * , 211(1973 ) c.m .
chambers , i.g moss , class . quantum grav . * 11 * , 1035 ( 1994 ) s. chandrasekhar , _ the mathematical theory of black holes _
( claredon , oxford , 1983 ) m. giammatteo , gr - qc/0303011 ( 2003 ) y. mino , m. sasaki , m. shibata , h. tagoshi , t. tanaka , prog . theor .
suppl * 128 * , ( 1997 ) w.h . press , s.a .
teukoslky , b.f .
flannery , and w.t .
vetterling , _ numerical recipes , the art of scientific computing _ ,
( cambridge university press , 1986 ) m. abramowitz , _ handbook of mathematical functions _ , ( dover publications , inc .
, new york , 1965 ) g.b .
arfken , h.j .
weber , _ mathematical methods for physicists _ , ( academic press , london 1995 ) v. cardoso , o.j.c .
dias , phys rev .
d * 70 * , 084011 ( 2004 ) v. cardoso , r. konoplya , j.p.s .
lemos , phys .
d * 68 * , 044024 ( 2003 ) s.w .
hawking , don n. page , commun .
phys . * 87 * , 577 ( 1983 ) m. banados , a. gomberoff , c. martinez , class .
grav . * 15 * , 3575 ( 1998 ) m. banados , phys .
d * 57 * , 1068 ( 1998 ) s.w hawking , c.j .
hunter , m.m .
taylor - robinson , phys .
d * 59 * , 064005 ( 1999 ) s.w .
hawking , h.s .
reall , phys .
rev d * 61 * , 0240014 ( 2000 ) d. birmingham , class .
* 16 * , 1197 ( 1999 ) a.m. awad , c.v .
johnson , phys .
d * 63 * , 124023 ( 2001 ) a.m. awad , class .
* 20 * , 2827 ( 2003 ) | we investigate the quasinormal modes for gravitational perturbations of rotating black holes in four dimensional anti - de sitter ( ads ) spacetime .
the study of the quasinormal frequencies related to these modes is relevant to the ads / cft correspondence .
although results have been obtained for schwarzschild and reissner - nordstrom ads black holes , quasinormal frequencies of kerr - ads black holes are computed for the first time .
we solve the teukolsky equations in ads spacetime , providing a second order and a pade approximation for the angular eigenvalues associated to the teukolsky angular equation .
the transformation theory and the regge - wheeler - zerilli equations for kerr - ads are obtained . |
_ the economist _ @xcite is a well known weekly newspaper of economics and economics related subjects .
frequently , it publishes interesting quantitative information in paper and on the web , especially in the _ daily chart _ section . in occasion of the 2014 world cup it collected the amount of goals scored during all the world cup competitions , from 1930 to 2014 , minute by minute .
all the data is presented as interactive plot on the web @xcite .
we give in fig.[fig : economist - reduction ] a reduction of _ the economist _ original graphic . in our plot _ additional time _ ( called _ allowance for time lost _ in fifa documentation @xcite ) and _ extra time _ are not represented .
no distinction is made between _ goals _ , _ penalties _ and _ own goals_. in the original graphic it possible to see in which match each goal was scored and is possible to filter matches according to some categories .
the reader interested in these facets should definitely refer to the web .
all the data used for this paper and the analysis performed in _ r _
@xcite software are available at _ mingotti.uc3m.es / floor1/paper - goals_. just under the original plot the authors summarize the graphical information with the following sentence : _
`` ... one can expect a rush of goals in the last ten minutes of normal time , but the 18th and 75th minutes have proved fertile''_. we consider such conclusions hurried and possibly lacking in significance . in this paper
we will propose a simple and reasonable model that describes the minute by minute goal scoring frequency .
after that , we reconsider how much of the original analysis can be confidently accepted .
the goal scoring frequency has been already analyzed in literature for the _ national scottish league 1991 - 1992 _ in @xcite , the _
australian soccer league _ in @xcite and
european champions league _ in @xcite . the 1986 world cup in @xcite , 1990 world cup in @xcite and jointly 1998 , 2002 and 2006 _ world cups _ in @xcite .
as far as we could establish , a comprehensive study about goal scoring frequency in all _ world cups _ till present has never been approached .
article @xcite is the most similar to our work .
it focuses on world cup scoring frequency and it considers more than a single tournament . the authors concluded that dividing the match in two 45-min parts , most goals were scored in the second half . dividing the match in 15-min parts , most of the goals were scored in the last period ( 76 - 90 ) and there was a trend towards more goals scored as time progressed . in this paper
we will see all these conclusion can be confirmed and refined using the richer dataset of all world cups scores .
from the _ the economist _ original data set we consider only goals scored in _ regular time _ , that is in minutes between @xmath0 $ ] and @xmath1 $ ] .
we ignore _ additional time _ because its occurrence is decided match per match by the referee @xcite and _ extra time _ because it depends on the scoring situation at the end of regular time .
their goal scoring rate can not be directly compared with _
regular time_. we redraw the dataset in fig.[fig : economist - reduction ] with a _ loess _ smoother . in fig.[fig : loess ] the thick continuous line is the smooth _ loess _ non parametric fit for the relation between goals and minute .
the dashed lines represent the average number of goals in the first and second half respectively . just under the dashed lines , on their left side ,
do lay the numerical values of each mean .
the parameters used for the _ loess _ are the ones default with r @xcite from fig.[fig : loess ] we can see the average number of goals in the second half ( 28.16 ) is larger than in the first half ( 22.04 ) .
the difference between these averages @xmath2 is statistically greater than zero . in this work
_ significance level _ is always @xmath3 .
bootstrapping 10000 times we get @xmath4 is normally distributed as @xmath5 . observing the _ loess _ smooth fit in fig.[fig : loess ]
we set the following hypothesis up . in the first half of the game the scoring rate is constant , in the second it grows linearly with time .
we also notice that the first minutes could have an inferior scoring rate respect to the other minutes in the first half .
that is reasonable , the game starts always in the middle of the field , far from a comfortable scoring position . on the first half ,
our probabilistic model is that goals are distributed according to a @xmath6 where @xmath7 for all @xmath8 in @xmath9 . performing a @xmath10 _ goodness of fit _
test we get @xmath11 , homogeneity would be rejected .
but , if we remove just minute 18th , we get @xmath12 and homogeneity would be far from rejected .
we do nt want to be too much lighthearted in dropping observations to satisfy our model so we reshape the dataset and repeat the test . instead of considering a match
as composed of a sequence of single minutes , we consider it as a sequence of 2 , 3 and 5 minutes blocks .
for each minutes block we set the associate number of goals as the average number goals scored in each minute making the block .
performing again the @xmath10 _ goodness of fit _
test on the _ first half _
, now seen as sequence of 2,3,5 minutes blocks , we get _ pvalue _ respectively @xmath13 . with a minimal smoothing reshape
we see homogeneity test in far from rejected , without dropping any observation . in conclusion , we are confident that the _ first half _
can be considered homogeneous and minute 18th an outlier . in the _ second half _
our working hypothesis is a linear relation between minutes elapsed and goals scored .
we begin fitting a simple linear regression model with _ minute _ as independent variable . by the _
t - test _ , _ minute _ is a significant variable , @xmath14 .
residuals can be considered normally distributed according to _ kolmogorov - smirnov _ and _ shaipiro - wilk _ normality test , which give respectively _ pvalues _ @xmath15 .
@xmath16 is little and not much interesting for us .
the fitted model appears in eq.[eq : model-1 ] @xmath17 from eq.[eq : model-1 ] we can get the expected number of goals in minute @xmath18 .
we will call these values @xmath19 .
normalizing @xmath20 we get @xmath21 .
and now we can apply again the @xmath10 to see if the model for the _ second half _ is appropriate .
our null hypothesis is that goals realized in the second half ( @xmath22 ) are multinomially distributed with probability vector @xmath23 .
the @xmath10 test returns a pvalue of @xmath24 so the null hypothesis can not be rejected in the second half game .
let us join the half time models and check hypotheses against the full length match data . according to our models , the expected goal sequence is @xmath25 .
where @xmath26 for @xmath27 $ ] is the average number of goals scored in the first half and @xmath26 for @xmath28 $ ] is set by eq.[eq : model-1 ] .
normalizing @xmath29 we get a probabilities vector @xmath30 , a graphical representation of it appears in fig.[fig : model - scoring ] . using again a @xmath10 , we test if goals scored in all word cups matches minute per minute can be a realization of a _ multinomial distribution _ with probabilities vector @xmath30 . we get a @xmath31 . removing observation 18th
we get @xmath32 and we can not reject .
removing also the first three minutes improves the fit giving @xmath33 . to avoid removing observation
we reshape the dataset in blocks of 2,3,5 minutes as done previously .
the resulting _ pvalues _ are respectively @xmath34 so , the null hypothesis is never rejected .
fig.[fig : model - per - blocks ] illustrates goals smoothing on reshaping minute variable .
an analytic description of the probability distribution can be easily obtained once we move to a time continuous description . rescaling the match to a @xmath35 $ ] duration interval ,
if there will be a goal , the probability it will happen before time @xmath36 will be given by @xmath37 .
@xmath38 is defined in eq.[eq : cdf ] .
@xmath39 minutes 18th and 75th were called in the original document `` fertile '' minutes . using our model we have a different interpretation for each one of them . in fig.[fig : simulated - max ] we see the simulated distribution of the maximum in the first and second half according to the model . for the first half we see that an observation larger than the 18th ( 42 goals ) is improbable , its probability is approximately 0.0015 . on the contrary , one observation larger than the 75th ( 44 goals ) in the second half it is not improbable at all ( @xmath40 ) .
there is enough evidence to state that the goal scoring distribution in the world cup matches can be considered constant in the first half game and growing linearly in the second half . dividing the game in blocks of 2 , 3 or 5 minutes confirms the model is valid and makes is more robust to extreme values .
an analytic cumulative distribution function for goal scoring is presented in eq.[eq : cdf ] .
previous finding in article @xcite are confirmed , there are more goals in the second half respect to the first one . according to our model , if there is a goal in a match , the probability that it will be in the first half is @xmath41 .
the last part of the game is most probable for a goal , and there is a trend toward more goals as time passes , but only in the second half .
_ the economist _ conclusions about its dataset are only partially acceptable .
indeed , it is true that in on the last part of the game there are expected more goals .
it is not true that minute 75th is an especially `` fertile '' minute , a maximum of 44 goals can be easily realized by the variability of the second half game .
minute 18th is more interesting , according to our model the probability of observing a maximum of 42 goals in the first half is about @xmath42 .
we guess it has to be considered an outlier and removed because it is an isolated burst , it has no connections with its neighboring values . performing a mild smooth , as grouping each match as a sequence of 2 minutes removes all of its importance .
c michailidis , i michailidis , g papaiakovou , and i papaiakovou .
analysis and evaluation of way and place that goals were achieved during the european champions league of football 2002 - 2003 .
, 2(1):4854 , 2004 . | on the occasion of the last fifa world cup in brazil , _ the economist _ published a plot depicting how many goals have been scored in all world cup competitions until present , minute by minute .
the plot was followed by a naive and poorly grounded qualitative analysis . in the present article we use _ the economist _ dataset to check its conclusions , update previous results from literature and offer a new model .
in particular , it will be shown that first and second half game have different scoring rates . in the first half
the scoring rate can be considered constant . in the second it increases linearly with time .
* keywords .
* goal , distribution , football , soccer , fifa world cup , outlier , the economist |
70 years ago myron mathisson has presented the equations describing the motions of a spinning test particle in a gravitational field @xcite
@xmath0 @xmath1 where @xmath2 is the 4-velocity of a spinning particle , @xmath3 is the antisymmetric tensor of spin , @xmath4 and @xmath5 are , respectively , the mass and the covariant derivative with respect to the proper time @xmath6 ; @xmath7 is the riemann curvature tensor of the spacetime .
( throughout this paper we use units @xmath8 .
greek indices run 1 , 2 , 3 , 4 and latin indices 1 , 2 , 3 ; the signature of the metric ( ,,,+ ) is chosen . )
equations ( [ 1 ] ) , ( [ 2 ] ) were supplemented by the condition @xcite @xmath9 ( in special relativity condition ( [ 3 ] ) was introduced by j. frenkel @xcite . )
it is known @xcite that in the minkowski spacetime the mathisson equations ( [ 1])([3 ] ) have , in addition to usual solutions describing the straight worldlines , a family of solutions describing the oscillatory ( helical ) worldlines ( as a partial case , this family contains the circular solutions . ) in @xcite the oscillatory solutions of equations ( [ 1])([3 ] ) were connected with _
`` zitterbewegung''_. c. mller proposed another interpretation of these solutions : he pointed out that 1 ) in relativity the position of the center of mass of a rotating body depends on the frame of reference , and 2 ) condition ( [ 3 ] ) is common for the so - called proper and non - proper centers of mass @xcite .
the usual solutions describe the motion of the proper center of mass of a spinning body ( particle ) , and the helical solutions describe the motions of the family of the non - proper centers of mass @xcite
. later a. papapetrou derived equations ( [ 1 ] ) , ( [ 2 ] ) by the method which differs from mathisson s one @xcite , and , instead of ( [ 3 ] ) , the non - covariant condition @xmath10 was used @xcite in concrete calculations . to avoid the superfluous solutions of equations ( [ 1 ] ) , ( [ 2 ] ) , w. tulczyjew and w. dixon introduced the covariant condition @xcite @xmath11 where @xmath12 is the particle s 4-momentum .
in contrast to relation ( [ 3 ] ) the tulczyjew - dixon condition ( [ 5 ] ) picks out the unique worldline of a spinning test particle in the gravitational field .
that is , equations ( [ 1 ] ) , ( [ 2 ] ) under ( [ 5 ] ) do not admit the oscillatory solutions .
however , the question arises : is this worldline close , in the certain sense , to the usual ( non - helical ) worldline of equations ( [ 1 ] ) , ( [ 2 ] ) under condition ( [ 3 ] ) , for example , in a schwarzschild field ?
it is simple to answer this question when the relation @xmath13 takes place , because in this case condition ( [ 5 ] ) practically coincides with ( [ 3 ] ) .
for example , these conditions are equivalent for post - newtonian expansions @xcite
. however , _ a priori _ another situation is possible for the highly relativistic spinning particle .
naturally , this situation must be investigated carefully .
note that very condition ( [ 3 ] ) was derived in some papers by different methods @xcite , and we agree with the conclusion that this condition `` ... arises in a natural fashion in the course of the derivation '' , @xcite , p. 112 .
that is , condition ( [ 3 ] ) is necessary , though often , with high accuracy , it can be substituted by condition ( [ 5 ] ) .
we stress : the existence of the superfluous ( oscillatory ) solutions of equations ( [ 1 ] ) , ( [ 2 ] ) under condition ( [ 3 ] ) is not a reason to ignore this condition .
the point of importance is that just among all solutions of equations ( [ 1])([3 ] ) the single solution describing the motion of the particle s proper center of mass can be found .
obviously , it is necessary to know how this solution can be identified among others . in the focus of this paper is just the initial mathisson equations ( [ 1])([3 ] ) .
our purpose is to investigate non - oscillatory solutions describing highly relativistic motions of the spinning particle in a schwarzschild field .
note that the information on all possible types of motions of the spinning test particles in the gravitational fields is important for astrophysics , for more fine investigations of the gravitational collapse and other astrophysical phenomena .
this paper is organized as follows .
the known integrals of the strict mathisson equations ( [ 1])([3 ] ) in a schwarzschild field , the energy and angular momentum , are used in section 2 for reducing the order of differentiation in these equations .
the possible procedure of finding the values of the energy and momentum parameters , which correspond just to the motions of the proper center of mass , is discussed in section 3 .
this procedure is realized in section 4 for the motions close to highly relativistic equatorial circular orbits in a schwarzschild field . in section 5
the illustrations of computer calculation are presented .
we conclude in section 6 .
let us consider equations ( [ 1])([3 ] ) for schwarzschild s metric in the standard coordinates @xmath14 for the equatorial motions of a spinning particle with spin orthogonal to the motion plane @xmath15 .
then the non - zero components of the metric tensor @xmath16 are : @xmath17 where @xmath18 is the schwarzschild mass .
due to the symmetry of schwarzschild s metric equations ( [ 1 ] ) , ( [ 2 ] ) have the integrals of the energy @xmath19 and the angular momentum @xmath20 which for the equatorial motions can be written as @xcite @xmath21 @xmath22 for the equatorial motions with spin orthogonal to the motion plane @xmath15 equations ( [ 2 ] ) can be solved separately from ( [ 1 ] ) .
indeed , taking into account relations ( [ 3 ] ) and ( [ 8 ] ) it is not difficult to obtain from ( [ 2 ] ) all non - zero components @xmath23 @xmath24 @xmath25 where @xmath26 is the known constant of spin @xcite @xmath27 ( we stress that expressions ( [ 10 ] ) satisfy all equations of set ( [ 2 ] ) . )
now we shall consider equation ( [ 1 ] ) .
it is known that these equations under condition ( [ 3 ] ) contain the second order derivatives @xmath28 with respect to @xmath6 .
however , in our case of schwarzschild s metric , due to the integral @xmath19 and @xmath20 , it is possible to reduce the order of differentiation by the standard procedure of the differential equations theory . using ( [ 8])([10 ] ) and the relation @xmath29 after direct calculations ( which are quite simple but rather lengthly ) we get from equation ( [ 1 ] ) the two non - trivial equations for @xmath30 and @xmath31 : @xmath32 @xmath33^{1/2},\ ] ] @xmath34^{1/2},\ ] ] where a dot denote the usual derivatives with respect to @xmath6 .
( two other equations of set ( [ 1 ] ) are satisfied identically . )
so , equations ( [ 12 ] ) , ( [ 13 ] ) do not contain the third coordinate derivatives .
however , in these equations the quantities @xmath19 and @xmath20 are present as the parameters which are not determined by the initial values of @xmath35 , @xmath36 , @xmath37 , @xmath38 only .
( according to ( [ 9 ] ) , for the determination of @xmath19 and @xmath20 the second coordinate derivatives must be given as well . )
that is , equations ( [ 12 ] ) , ( [ 13 ] ) , as well as the initial mathisson equations ( [ 1 ] ) , describe the motions both of the proper center of mass and the non - proper centers .
in other words , equations ( [ 12 ] ) , ( [ 13 ] ) contain the non - oscillatory and oscillatory solutions .
the question of importance is : which values @xmath19 and @xmath20 correspond just to the proper center of mass at the arbitrary initial values of @xmath35 , @xmath36 , @xmath39 , @xmath40 for a spinning test particle ?
it is easy to answer this question when the motion of such a particle is close to the geodesic motion : then approximately @xmath41 and @xmath42 , i.e. , we write the relations for the geodesic motion .
however , we have not any proof that the worldline of a spinning particle is close to the corresponding geodesic worldline for all physically admitted initial values of @xmath35 , @xmath36 , @xmath39 , @xmath40 . on the contrary ,
our results of investigations of the gravitational spin - orbit interaction in a schwarzschild field @xcite show that just highly relativistic motions of a spinning particle must be studied carefully .
therefore , in the next section we shall consider the procedure of choosing the values @xmath19 and @xmath20 for the proper center of mass for highly relativistic motions . for further calculations it is convenient to write equations ( [ 12 ] ) , ( [ 13 ] ) in terms of the non - dimensional quantities @xmath43 ( in the following we shall put @xmath44 , without any loss in generality . )
then according to equations ( [ 12 ] ) , ( [ 13 ] ) we have the set of the first - order differential equations @xmath45 @xmath46^{1/2},\ ] ] @xmath47 @xmath48^{1/2},\ ] ] @xmath49 @xmath50 where @xmath51 for correct studying the physical conclusions of equations ( [ 15])([18 ] ) , it is necessary to take into account the wald condition for a test spinning particle @xcite , therefore we put @xmath52 because in the following we shall compare some solutions of equations ( [ 15])([18 ] ) with solutions of the geodesic equations in schwarzschild s metric , it is useful to write here the last equations for the equatorial motions in the coordinates @xmath53 @xmath54 @xmath55 using notation ( [ 14 ] ) we rewrite equations ( [ 21 ] ) , ( [ 22 ] ) as the set of the first - order differential equations @xmath56 @xmath57 plus two equations which coincide with ( [ 17 ] ) , ( [ 18 ] ) . equations ( [ 15 ] ) , ( [ 16 ] ) and ( [ 23 ] ) , ( [ 24 ] ) correspondingly are essentially different . as well as equations ( [ 12 ] ) ,
( [ 13 ] ) , the first two equations of set ( [ 15])([18 ] ) contain the constants of the energy and angular momentum of a spinning particle .
according to discussion above , different values of the parameters @xmath58 and @xmath59 in ( [ 15 ] ) , ( [ 16 ] ) at the fixed values of @xmath60 correspond to different centers of mass , namely to the single proper center of mass and to the set of non - proper centers .
the right - hand sides of equations ( [ 15 ] ) , ( [ 16 ] ) are too complicated for investigations in general case .
however , _ a priori _ we can not exclude the possibility to show the essential difference between the motions of the proper and non - proper centers of mass during the short time interval after the beginning of the corresponding motions , namely , when the displacement of the values @xmath61 from their initial values @xmath62 are considered in the linear approximations in the quantities @xmath63 ( in ( [ 27 ] ) we do not write the displacement @xmath64 because the right - hand sides of equations ( [ 15])([18 ] ) do not depend on @xmath36 and equation ( [ 18 ] ) is trivial , i.e. , according to ( [ 18 ] ) the value @xmath36 is determined by simple integration of @xmath65 . )
let us check this possibility , i.e. , consider equations ( [ 15])([17 ] ) in the linear in @xmath66 approximation .
then by direct calculations we obtain @xmath67 @xmath68 @xmath69 @xmath70 @xmath71 where the coefficients @xmath72 with the corresponding indexes are expressed through @xmath73 we shall use the expressions of these coefficients in the approximation @xmath74 according to notation ( [ 14 ] ) , relations ( [ 31 ] ) mean that the tangential component of the particle s initial velocity is highly relativistic , and , in addition , that the tangential component is much greater than the radial component .
first , it corresponds with our aim to investigate just highly relativistic motions .
second , because the deviation of a spinning particle is caused by the gravitational spin - orbit interaction , most clearly this deviation can be shown in the case when the tangential velocity is dominant ( in particular , for the circular or closer to the circular orbits @xcite ) . according to the theory of differential equations , the general solution of linear equations ( [ 28])([30 ] )
is determined by the combination of @xmath75 ( i=1,2,3 ) , where @xmath76 are the solutions of the third - order algebraic equation @xmath77 here the coefficients @xmath78 ( @xmath79 ) can be expressed through @xmath72 and depend both on @xmath80 and on the parameters @xmath81 . our task is to find such concrete values @xmath81 which at the fixed @xmath80 determine just the motion of the proper center of mass .
we begin by analyzing the expressions @xmath78 for the simple case when the worldline of a spinning particle is close to the corresponding geodesic worldline .
if the parameter @xmath82 in ( [ 20 ] ) is sufficiently small , for any fixed values @xmath83 the motion of the proper center of mass is close to the geodesic motion
. then we can write approximately @xmath84 ( see expressions ( [ 9 ] ) ) , and according to ( [ 14 ] ) , ( [ 19 ] ) we have @xmath85 let us consider expressions @xmath78 at @xmath81 from ( [ 41 ] ) .
it is easy to show that all these expressions have a common feature : all greatest terms with the large value @xmath86 are cancelled .
it means that the corresponding largest terms with @xmath86 are absent in the expressions @xmath87 as well .
so , just the values @xmath81 for the proper center of mass give the minimum values of @xmath87 .
we stress that similar situation takes place for motions of the proper center of mass in the minkowski spacetime : for the proper center @xmath88 ( it corresponds to the straightforward motions ) , and for the non - proper centers @xmath87 are proportional to @xmath89 ( the oscillatory motions ) .
in @xcite we have considered the case of the circular motion of the proper center of mass in a schwarzschild field with @xmath90 . by the notation ( [ 14 ] ) in this case we write @xmath91 for values
( [ 19 ] ) , from ( [ 9 ] ) we obtain @xmath92 let us estimate the values @xmath78
. taking into account ( [ 42 ] ) it is easy to check that , as well as in the previous case , in all expressions @xmath78 the largest terms with @xmath86 are cancelled .
it is naturally to suppose that the similar feature takes place not only in the two partial cases above .
therefore , below we shell check do the criterion of excluding the largest terms with @xmath86 in the expressions @xmath78 can be used for finding the values @xmath58 and @xmath59 which pick out just the motion of the proper center of mass .
as we pointed out in @xcite equations ( [ 1])([3 ] ) admit in a schwarzschild field the solutions describing the circular equatorial orbits in the region @xmath94 . by notation (
[ 14 ] ) for these orbits we have @xmath95 the expression @xmath65 from ( [ 43 ] ) is valid beyond the small neighborhood of the value @xmath96 .
( this neighborhood was considered in @xcite . ) at expressions ( [ 43 ] ) , it follows from ( [ 9 ] ) that @xmath97 it is interesting to consider the non - circular orbits which deviate from ( [ 43 ] ) due to @xmath98 under condition ( [ 31 ] ) .
let us consider the expressions @xmath78 in the case @xmath99 where according to ( [ 43 ] ) , ( [ 44 ] ) @xmath100 @xmath101 and @xmath102 , @xmath103 are some coefficients which do not depend on @xmath82 .
our task is to find such values @xmath102 , @xmath103 which ensure excluding the largest terms with @xmath86 in the expressions @xmath78 . using @xmath58 , @xmath59 from ( [ 46 ] )
we obtain the conditions under which the largest terms with @xmath86 in the expressions @xmath78 are cancelled : @xmath104 @xmath105 @xmath106 it is easy to check that at the value @xmath107 from ( [ 46 ] ) among three linear in @xmath102 , @xmath103 algebraic equations ( [ 48])([50 ] ) there are only two independent equations
. then the solution of ( [ 48])([50 ] ) is : @xmath108 @xmath109 relations ( [ 51 ] ) , ( [ 52 ] ) can be used in computer integration of equations ( [ 15])([18 ] ) under conditions ( [ 45 ] ) .
graphs of @xmath110 and @xmath111 according to equations ( [ 15])([18 ] ) under relations ( [ 51 ] ) , ( [ 52 ] ) are shown by the thick lines in figures 1 and 2 correspondingly .
for comparison , the thin lines in the same figures show graphs of @xmath110 and @xmath111 according to geodesic equations ( [ 23])([24 ] ) .
if relations ( [ 51 ] ) , ( [ 52 ] ) are violated , computer integration shows the oscillatory solutions .
similar situation takes place for orbits with @xmath112 .
an example of the oscillatory solution is presented in figure 3 .
1 . graph of @xmath110 according to equations ( [ 15])([18 ] ) under relations ( [ 51 ] ) , ( [ 52 ] ) at @xmath113 , @xmath114 , @xmath115 and @xmath116 determined by ( [ 45 ] ) , ( [ 46 ] ) ( thick line ) .
graph of @xmath110 according to geodesic equations ( [ 23])([24 ] ) at the same @xmath117 , @xmath118 , @xmath116 ( thin line ) .
the horizontal line @xmath119 corresponds to the horizon surface .
2 . graph of @xmath111 in the polar coordinates according to equations ( [ 15])([18 ] ) ( thick line ) and ( [ 23])([24 ] ) ( thin line ) .
the values @xmath117 , @xmath118 , @xmath116 are the same as in fig . 1 .
the circle @xmath119 corresponds to the horizon surface .
an example of the oscillatory solution for @xmath110 according to equations ( [ 15])([18 ] ) at @xmath120 , @xmath112 , @xmath121 and @xmath116 determined by ( [ 41 ] ) when relations ( [ 42 ] ) are slightly violated .
the computer integration of equations ( [ 15])([18 ] ) at relations ( [ 51 ] ) , ( [ 52 ] ) shows that these relations are suitable for choosing solutions , which describe the motions of the proper center of mass . as a result , according to figures 1 and 2 , we conclude that ( under those conditions indicated in the captions at these figures ) the force of the interaction of the particle s spin with the gravitational field acts as the repulsive one . due to this force
the spinning particle falls on the horizon surface during longer time as compared to the corresponding particle without spin ( figure 1 ) .
moreover , according to figure 2 , considerable space separation of the corresponding spinning and non - spinning particles takes place within a short time , i.e. , within the time of the particle s fall on the schwarzschild horizon .
20 m. mathisson , _ acta phys . pol .
_ * 6 * , 163 ( 1937 ) . j. frenkel , _ z. phys . _ * 37 * , 243 ( 1926 ) .
m. mathisson , _ acta phys .
pol . _ * 6 * , 218 ( 1937 ) .
j. weyssenhoff , a. raabe , _ acta .
phys . pol .
_ * 9 * , 7 ( 1947 ) . c. mller , _ commun .
dublin inst .
. studies _
a , * 5 * , 3 ( 1949 ) ; c. mller , _ ann .
henri poincare , _ * 11 * , 251 ( 1949 ) .
a. papapetrou , _ proc .
_ a , * 209 * , 248 ( 1951 ) .
e. corinaldesi , a. papapetrou , _ proc .
_ a * 209 * , 259 ( 1951 ) .
w. tulczyjew , 1959 _ acta phys .
pol . _ * 18 * , 393 ( 1959 ) ; w. dixon , _ proc .
_ a , * 314 * , 499 ( 1970 ) .
b.m . barker , r.f .
oconnell , _ gen .
grav . _ * 4 * , 193 ( 1973 ) ; m.d .
hartl , _ phys .
_ d * 67 * , 024005 ( 2003 ) .
b. mashhoon , _ j. math
* 12 * , 1075 ( 1971 ) .
h. taub , _ j. math
. phys . _ * 5 * , 112 ( 1964 ) s. ragusa , m. bailyn , _ gen .
grav . _ * 27 * , 163 ( 1995 ) .
r. micoulaut , _ z. phys . _
* 206 * , 394 ( 1967 ) .
r. plyatsko , _ phys .
_ d * 58 * , 084031 ( 1998 ) .
r. plyatsko , o. bilaniuk , _ class .
quantum grav . _ * 18 * , 5187 ( 2001 ) .
r. plyatsko , _ class .
quantum grav . _
* 22 * , 1545 ( 2005 ) .
r. wald , _ phys .
_ d * 6 * , 406 ( 1972 ) . | the mathisson equations under the frenkel - mathisson supplementary condition are studied in a schwarzschild field .
the choice of solutions , which describe the motions of the proper center of mass of a spinning test particle , is discussed , and the calculation procedure for highly relativistic motions is proposed .
the very motions are important for astrophysics while investigating possible effects of the gravitational spin - orbit interaction on the particle s world line and trajectory . |
it is now widely accepted that the horizontal structure of the sunspot penumbra is composed of two magnetic components ( solanki 2003 ; bellot rubio 2003 ) .
one of them possesses a somewhat inclined ( @xmath3 with respect to the vertical direction to the solar surface ) and strong ( @xmath4 g ) magnetic field , whereas the other is characterized by a weaker and more horizontal one ( lites et al .
1993 ; redi et al .
1998 ; bellot rubio et al . 2004
; borrero et al .
2004 , 2005 ) .
traditionally , these two magnetic components have been identified with a horizontal flux tube , that carries the evershed flow , and is embedded in a more vertical background magnetic field : _ uncombed _ model ( solanki & montavon 1993 ; schlichenmaier et al .
1998 ; borrero 2007 ) .
recently , this view has been challenged by spruit & scharmer ( 2006 ) and scharmer & spruit ( 2006 ) , who propose instead that the penumbra is formed by magnetic field - free plumes ( connected to the underlying convection zone ) that pierce the penumbral magnetic field from beneath .
this is the so - called _ gappy _ penumbral model .
so long as these two different magnetic structures ( weak / horizontal and strong / vertical ) have remained spatially ( horizontally ) unresolved , distinguishing between the uncombed and gappy penumbral scenarios has not been possible . however , with the new spectropolarimeter on board of the japanese spacecraft hinode ( kosugi et al .
2007 ; shimizu et al . 2007 ) it is now possible to obtain high spatial resolution ( @xmath5 " ) observations of the sunspot penumbra .
this could be sufficient to distinguish between the uncombed and gappy models , since they both postulate the existence of flux tubes or field - free gaps that are about 200 - 300 km in diameter ( martnez pillet 2000 ; spruit & scharmer 2006 ) .
this feature is particularly interesting , because the _ uncombed _ and _ gappy _ models predict very different vertical stratifications in the magnetic field strength across the weak / horizontal magnetic field component : which is identified with an embedded flux tube in the uncombed model , but with a field - free gap in the gappy model . in the latter ,
the magnetic field decreases monotonically with depth , whereas the former possesses a magnetic field that decreases with depth only initially , since once the boundary of the flux tube is reached , the magnetic field can either decrease or increase depending upon the strength of the magnetic field inside the tube . in this paper
we will focus on obtaining the vertical stratification of the magnetic field for penumbral filaments ( where the magnetic field is more horizontal and weaker ) using high spatial resolution spectropolarimetric observations from hinode , in order to establish which penumbral model is more realistic .
the observations are described in section 2 .
section 3 describes our data analysis and results from our inversion technique .
section 4 compares our findings with the predictions made by the uncombed and gappy penumbral models . in section 5
we make a thourough investigation of the effects of the scattered light .
finally , section 6 summarizes our findings .
on may 3rd 2007 , between 10:15 and 11:40 am ut , the active region ar 10953 was mapped using the spectropolarimeter of the solar optical telescope on - board of the hinode spacecraft ( lites et al . 2001 ) .
the active region was located at a heliocentric angle of @xmath6 .
it was scanned in a thousand steps , with a step width of 0.148 `` and a slit width of 0.158 '' .
the spectropolarimeter recorded the full stokes vector ( @xmath7 , @xmath8 , @xmath9 and @xmath10 ) of the pair of neutral iron lines at 630 nm with a spectral sampling of 21.53 m .
the integration time was 4.8 seconds , resulting in an approximate noise level of @xmath11 ( in units of the normalized continuum intensity ) . in the absence of the telluric oxygen lines we proceeded with two different wavelength calibration methods that were cross - checked for consistency .
the first method was obtained by matching the average quiet sun profile with the fts spectrum , whereas the second calibration assumes that the average umbral profile exhibits no velocities .
a map of the continuum intensity at 630 nm of the scanned region is shown in figure 1 .
the white arrow indicates the direction of the center of the solar disk . the penumbra on the center side is heavily distorted and therefore left out from our analysis . on the limb side the penumbra is more uniform , with radially aligned filaments .
the region enclosed by the white rectangle has been chosen for our study .
this sunspot has negative polarity ( magnetic field in the umbra points towards the solar interior ) , however the results presented hereafter are shown , in order to facilitate the interpretation , as if the sunspot had positive polarity .
we have applied the sir inversion code ( ruiz cobo & del toro iniesta 1992 ) to our spectropolarimetric observations to retrieve the physical properties of the solar atmosphere .
this code allows all relevant physical parameters to be a generic function of the optical depth : @xmath12 , @xmath13 , @xmath14 , @xmath15 , etc .
in addition , a depth - dependent temperature stratification @xmath16 models the atmospheric thermodynamics under local thermodynamic equilibrium ( lte ) conditions .
sir retrieves the values of the parameters at a number of optical depth points called _
nodes_. the final stratification is obtained by interpolating splines across those nodes .
note however , that sir employs equivalent response functions ( del toro iniesta 2003 ) , which ensures sensitivity to the atmospheric layers located between nodes .
each node represents a free parameter in the inversion . in our investigation
we will employ increasingly complex models ( i.e. : more free parameters ) according to the amount of information we hope to extract from the profiles .
given the high spatial resolution of hinode s observations , we will consider only one magnetic component .
a non - magnetic component is also considered to account for the scattered light . in this section ,
the scattered light profile is obtained by averaging the intensity profiles of those pixels with polarization signals below the noise level ( quiet sun granulation around the sunspot ) .
the same scattered light profile is used in the inversion of all pixels . in sect .
5 we make a thourough analysis of the effects that different treatments for the scattered light have on our results .
note that using one single magnetic component is equivalent to assuming that the penumbral structure is horizontally resolved .
this is clearly not the case if we look into continuum images at even higher spatial resolution ( scharmer et al .
however , our assumption would still be valid if the ( unresolved ) variations of the magnetic field inside the weak / horizontal magnetic component are much smaller than the differences between the weak / horizontal and strong / vertical components .
since the former remain unresolved , we can not assess the validity of this assumption .
this question should be addressed as better spectropolarimetric observations become available . in order to locate the intraspinal pixels we have carried out a first inversion where all physical parameters , with the exception of the temperature , are constant with optical depth .
we therefore have one single node for @xmath12 , @xmath13 , @xmath14 , @xmath15 . to account for unresolved velocity fields
, we also consider depth - independent micro and macroturbulent velocities : @xmath17 and @xmath18 .
another free parameter , @xmath19 , represents the fraction of the observed intensity , stokes @xmath7 , that corresponds to scattered light .
finally , three nodes are given to the temperature @xmath16 . in total
, this first inversion has 10 free parameters . since @xmath20 , @xmath21 , @xmath22 and @xmath23 are constant with optical depth , the retrieved values indicate some kind of average over the region where the spectral lines are formed .
westendorp plaza et al .
( 1998,2001 ) studied this issue in detail and found that the largest contribution for this pair of fe i lines ( sect .
2 ) comes from @xmath24 .
figure 2 displays the resulting values for the line - of - sight velocity and magnetic field vector in the selected box in fig .
1 . regions of weak , @xmath25 g , and highly inclined , @xmath26 , magnetic field can be clearly distinguished in this figure .
they are also characterized by the presence of large red - shifted velocities ( evershed flow ) .
these are the so - called penumbral intraspines , and therefore the most likely locations where field - free gaps or horizontal flux - tubes can be found . also visible are structures characterized by a stronger and more vertical magnetic field , as well as by a strongly reduced evershed flow .
these are usually referred to as spines .
spines and intraspines are also seen at moderate ( @xmath27 1 `` ) spatial resolution ( lites et al .
1993 ; stanchfield et al .
1997 ; mathew et al .
2003 ) but the associated changes in their properties ( field strength , inclination , etc ) are larger if observed at high spatial resolution ( bello gonzlez et al .
2005 ; langhans et al . 2005 ) .
bellot rubio et al . ( 2004 ) interprets this result as a consequence of these structures not being spatially resolved at 1 '' resolution .
as demonstrated by borrero et al .
( 2008 ) they are indeed horizontally resolved in hinode observations ( 0.32 " ) .
+ + in figure 2 we also indicate with black and white dots a large number ( total of 7520 ) of intraspinal pixels . they have been found as those satisfying the following conditions : @xmath28 g , and @xmath29 $ ] km s@xmath30 .
since the main difference between spines and intraspines is the presence of a strong evershed flow , we use @xmath23 to distinguish among them .
however , we do not consider the few pixels where @xmath31 km s@xmath30 , since they usually present extremely abnormal stokes v profiles , usually a sign of the existence of horizontally unresolved structure .
we do not constrain the values of the magnetic field inclination and strength ( here we use only a lower limit to avoid taking pixels outside the visible boundary of the sunspot ) because the magnetic properties of the spines in the outer penumbra are very similar to those of the intraspines in the inner penumbra .
the final selected pixels represent about 39 % of all penumbral pixels in figure 2 .
note that they are mostly located in the middle and outer penumbra : @xmath32 ( @xmath33 being the sunspot radius ; umbral - penumbral boundary is located at @xmath34 ) .
note also that , even though we have not constrained the values of the magnetic field strength an inclination , all intraspinal pixels are located in regions where the magnetic field is highly inclined and weak . in order to investigate the depth variation of the physical parameters in intraspines
, we performed a renewed inversion of the pixels selected in fig . 2 , where we now allow for two nodes in @xmath12 , @xmath13 , @xmath14 , @xmath15 ( linear variations with optical depth ) .
the total number of free parameters is now 14 .
results from this new inversion are presented in figure 3 : @xmath16 ( upper - left ) , @xmath15 ( upper - right ) , @xmath12 ( lower - left ) , and @xmath13 ( lower - right ) .
all inverted pixels display similar stratifications of @xmath15 and @xmath13 : both increase monotonically with optical depth : @xmath35 , @xmath36 . the magnetic field strength @xmath12 , however , can either increase ( in 66 % of inverted pixels : 4971 ) or decrease ( 34 % ; 2249 pixels ) with optical depth . in either case
, the retrieved gradient is relatively small : @xmath37 gauss km@xmath30 . an important feature to note
is that pixels displaying a decreasing magnetic field towards deeper layers , @xmath38 , are mostly located in the inner penumbra , whereas pixels showing @xmath39 are mostly found in the outer penumbra ( see figure 4 ) .
we now perform a more complex inversion of the same pixels as in sect . 3.2 . in this case
we allow for 4 nodes in @xmath12 , @xmath13 , @xmath14 , and @xmath15 .
these nodes are located at optical depth positions : @xmath40 $ ] . in total
, this new inversion has 22 free parameters .
figure 5 shows the results from the 4-node inversion of the 7250 intraspinal pixels selected in sect .
the stratifications are very similar to those already obtained through the 2-node inversion ( see fig . 3 ) . the larger scatter ( pixel - to - pixel variations ) in the 4-node inversion is due to the larger amount of free parameters , which are more weakly constrained by the observations . since now we allow for 4 nodes to the stratification of the magnetic field strength it is not easy to classify our results between those where the magnetic field increases or decreases with optical depth .
to showcase the differences between the possible stratifications we have taken separately those pixels where , in the 2-node inversion , showed @xmath38 ( family 1 ) or @xmath39 ( family 2 ) and obtained the averaged stratification for the 2 and 4-node inversion .
results for family 1 and 2 are presented in figures 6 and 7 respectively .
+ interestingly , the magnetic field strength , that in the 2-node inversion showed different gradients , now shows an initial decrease up to @xmath41 ( approximately 100 km above the continuum level ) , where it starts to increase again towards deep layers .
this happens for both families of magnetic structures , and thus could indicate that they are indeed closely related .
a closer look reveals that both families possess a similar magnetic field strength in deep layers : @xmath42 g , but slightly different higher up : @xmath43 g ( family 1 ; fig .
6 ) and @xmath44 g ( family 2 ; fig . 7 ) .
this effect explains why the 2-node inversion ( sect .
3.2 ) retrieves different overall gradients for the magnetic field strength : it is due to a large variation in the magnetic field at around @xmath45 since the magnetic field deeper down is basically the same in both cases . to further confirm these results
we have repeated our 4-node inversion with the nodes located at slightly different positions .
sir always places 2 nodes at the uppermost and deepest @xmath46-locations of the discretized atmosphere , while spreading the rest equidistantly in between .
therefore , to keep the same number of nodes and , at the same time , change their @xmath46-positions we must change the initial and last @xmath46-points of the atmosphere . in our first set of inversions
the atmosphere is discretized between @xmath47 $ ] . changing this to @xmath48 $ ] and
@xmath49 $ ] would position the 4 nodes at @xmath50 $ ] and @xmath51 $ ] , respectively .
we have inverted all pixels again in these two cases and confirmed that our results ( fig .
5,6 and 7 ) do not change .
this is due to the use that sir makes of equivalent response functions ( see section 3 ) .
if we consider a ray passing through the center of an intraspine , the _ gappy _ and _ uncombed _ penumbral models predict a very similar stratification of the magnetic field strength above the field - free gap or flux tube , but very different ones inside them .
figure 8 illustrates some possible stratifications predicted by these two models , where the upper boundary of the field - free gap and flux tube is located at @xmath52 . above
@xmath52 they both share the same stratification for the surrounding magnetic atmosphere . here
we present two examples , one where the surrounding magnetic field is weak : @xmath53 ( dashed line ; meant to represent the outer penumbra , @xmath54 ) , and another case where the surrounding field is stronger ( solid line ) : @xmath55 ( meant to represent to inner penumbra , @xmath56 ) .
note that the magnetic field strength in the surrounding atmosphere decreases towards deeper layers .
this is due to the fact that the vertical component of the surrounding field must vanish ( or nearly vanish in the case of a cusp - shaped boundary ) at the flux tube s or gap s boundary .
these two examples are actual solutions of analytical models ( fig . 5 in spruit & scharmer 2006 ; fig . 3 in scharmer & spruit 2006 ; see also eqs .
33 - 34 in borrero 2007 ) . below the boundary of the flux tube or field - free gap , @xmath52 , both models predict a very different situation . in the case of the _ gappy _ penumbra
this region is void of magnetic fields : @xmath57 ( hollow circles ) . in contrast
, the _ uncombed _ model assumes the existence of a flux tube where the magnetic field is strong @xmath58 g ( filled circles ) . if we compare fig . 8 with our 2-node inversion ( fig .
3 ) of intraspinal pixels we find that , on the one hand , the _ gappy _ penumbral model can only explain the slowly decreasing magnetic field , observed for 34 % of intraspinal pixels ( family 1 ) , if the @xmath2 level is formed above the gap s boundary , otherwise a much more sudden drop would be observed ( hollow circles in fig . 8) . on the other hand , this model does not offer any explanation for the 66 % of the intraspinal pixels that present an increasing magnetic field strength towards deeper layers ( family 2 ) .
however , the _ uncombed _ penumbral model can explain both observed situations .
it all depends on the strength of the flux tube s magnetic field as compared to the magnetic field high above it : @xmath59 versus @xmath60 .
a magnetic field that decreases smoothly towards the interior of the photosphere can be explained by a flux tube ( of any field strength ) whose upper boundary layer lies below @xmath2 . if the upper boundary is above @xmath2 , it can also be explained with a magnetic field inside the flux tube that is weaker than the magnetic field a few hundred kilometers above ( solid lines plus filled circles in fig .
in addition , a magnetic field that increases towards the interior of the photosphere is compatible with a flux tube with an upper boundary layer above @xmath2 , and with a stronger magnetic field than the one above ( dashed line and filled circles in fig .
8) .
a more complex ( 4-nodes ) inversion of intraspinal profiles indicates that , what appeared as two different families of structures using a 2-node inversion , are likely to correspond to one single kind of magnetic structure , where the magnetic field exhibits an initial decrease between @xmath61 $ ] , but increases between @xmath62 $ ] ( see figs . 6 - 7 ) .
while the _ gappy _ model offers no explanation for this effect , it can indeed be explained by the _ uncombed _
penumbral model , by means of a magnetic field whose strength decreases initially but increases once the line - of - sight crosses the flux tube s boundary ( see fig .
furthermore , although intraspinal families 1 and 2 appear to be the equivalent in the 4-node inversion , they still present a subtle yet important difference : family 1 ( more commonly found in the inner penumbra ; see fig .
4 ) displays a much stronger initial decrease as compared to family 2 , which is usually found in the outer penumbra ( compare lower - left panels in figs . 6 and 7 ) .
this can be explained , in terms of the _ uncombed _ model if , at small - intermediate radial distances , the horizontal flux tube possesses a weaker magnetic field than the field in the atmosphere in which it is embedded : @xmath63 at @xmath64 small ( compare solid line plus filled circles in fig . 8 with green solid in fig . 6 ) . as we move towards larger radial distances , and assuming that the magnetic field inside the flux tube remains constant , the surrounding magnetic field weakens and falls below the flux tube s field strength : @xmath65 at @xmath64 large ( compare dashed line plus filled circles in fig . 8 with solid green in fig .
note that the assumption that the magnetic field in the flux tube remains constant is in agreement with a surrounding magnetic field whose strength decays much more rapidly towards the outer penumbra than inside the flux tube ( see fig . 4 in borrero et al .
2004 ; fig . 6 in borrero
et al . 2005 and fig . 4 in borrero et al .
2006 ) . here
we find that this known feature of the penumbral intraspines helps to explain , within the frame of the uncombed model , differences in the stratification in the magnetic field strength across instrapines at different radial distances , as deduced from high resolution spectropolarimetric observations .
one of the most critical issues in the inversion of spectropolarimetric data is the treatment of the scattered light . in order to properly model its contribution , detailed measurements of the telescope s psf
are needed .
since these are not usually available , the scattered light is often treated as a non - polarized contribution to the total observed light ( see sect . 3 ) . in our study
this is particularly important because one of the models under study ( _ gappy _ model ) postulates the existence of field - free regions around the @xmath2 level in the penumbra .
these regions will naturally produce a non - polarized contribution to the total observed stokes vector .
therefore , there is a potential risk of not detecting the field - free gaps due to an incorrect treatment of the scattered light .
+ if our inversions are affected by this degeneracy between scattered light and field - free gaps , it is expected that those pixels where the intraspines are located show larger values for the amount of scattered light retrieved by the inversion ( @xmath19 ) . to study this possibility we have plotted in figure 9 the variations of the magnetic field strength ( top panel ) , line - of - sight velocity ( middle panel ) and inclination angle ( bottom ) along an azimuthal cut at @xmath66 .
other azimuthal cuts at different radial distances show very similar behaviors .
the values are taken at an optical depth of @xmath2 from the 2-node inversion in sect .
this plot includes , not only those pixels selected in fig . 2 as intraspines , but all of them . therefore regions where the magnitude of evershed flow is reduced and the magnetic field is more vertical and strong ( penumbral spines ) are also visible .
all three panels also show the amount of scattered light @xmath19 ( dashed lines ) .
there is no particular correlation between the location of penumbral intraspines ( high velocities , weak and very inclined fields ) and the regions where @xmath19 is largest .
similar variations are observed if we plot the values of the magnetic field strength and inclination , and line - of - sight velocity , at an optical depth of @xmath67 .
this rules out the possibility that our inversions do not show field - free regions , in the deep photospheric layers , where intraspines are located at the expense of an enhanced scattered light contribution .
recently , orozco surez et al . ( 2007a ; 2007b ) have presented inversions of stokes spectra measured using hinode s spectropolarimeter in the quiet sun .
these authors claim that for this instrument it is more appropriate to consider a local ( unpolarized ) scattered light profile .
this is obtained by averaging the observed stokes @xmath7 profiles over a small region ( about 1 arcsec ) around the pixel that is being studied . in this case ,
a different scattered light profile is used in the inversion of each pixel .
this approach can be justified by the fact that the focus of the narrow - band filter ( nbi ) on hinode is favored when simultaneous observations are carried out with both instruments . in our inversions , we have however used a global scattered light profile , where we average the stokes @xmath7 signal emerging from the quiet sun region far away from the sunspot . in order to test whether our results depend on the use of a different scattered light profile , we have repeated our 2-node inversion using the same approach as orozco surez et al .
the results are presented in fig .
fig . 3 ) .
the percentage of intraspinal pixels with @xmath68 is even larger than before ( 90 % ) . alternatively , orozco surez et al .
( 2007a ; 2007b ) point out that the most realistic way to account for the scattered light would be to consider a local * and * polarized scattered light profile , where not only stokes @xmath7 is averaged , but also stokes @xmath8 , @xmath9 and @xmath10 .
we have also tested this possibility .
unfortunately , this yields unrealistically high values for @xmath19 during the inversion : @xmath69 .
this indicates that the inversion code tries to dominantly reproduce the observed stokes profiles using the scattered light contribution .
this is not surprising since the neighboring stokes profiles often look very similar to those in the pixel under study .
therefore we conclude that this is not a reasonable approach when inverting sunspot data .
we can not rule out however , that this treatment will work in quiet sun regions . as a final test
, we have repeated our 2-node inversion but neglecting any scattered light : @xmath70 .
this test is very appropriate because , according to spruit & scharmer ( 2006 ) , inversions of spectropolarimetric data fail to detect field - free regions in the penumbra as a consequence of these being already included in the scattered light profile .
if their hypothesis is correct , not accounting for the scattered light contribution should uncover these regions with @xmath71 near @xmath2 .
results ignoring the effects of the scattered light ( figure 11 ) are essentially unchanged if compared to those where we used a global ( fig . 3 ) or local ( fig .
10 ) scattered light profile , with 68 % of the pixels showing @xmath68 .
the only difference is the increased temperatures obtained when we impose @xmath70 . in particular we do not see any pixel where the magnetic field reaches very small values in the deepest photospheric layers .
taking into account all tests carried out in this section , it seems unlikely that the scattered light can significantly bias our magnetic field stratifications , consequently making highly unlikely that we are missing the detection of field - free regions near @xmath2 .
the _ uncombed _ model postulates that penumbral intraspines are characterized by the presence of horizontal flux tubes embedded in a surrounding atmosphere that possesses an inclined magnetic field . according to this model ,
looking along these regions should reveal a magnetic field that smoothly decreases at first , but once the flux tube contribution starts , the field strength could either increase or decrease . alternatively ,
the _ gappy _ penumbral model postulates that instraspines correspond instead to regions where convective field - free gaps penetrate the penumbral field . in this case the magnetic field strength should also decrease with optical depth at first , but suffer a much larger drop once the line - of - sight crosses the field - free gap . in order to differentiate between these two models
, we have used polarimetric data at very high spatial resolution , recorded with the spectropolarimeter on - board of the japanese spacecraft hinode , to investigate the depth variation of the magnetic field strength in the penumbra .
we have selected a large number ( @xmath277500 ) of pixels that are representative of weak and horizontal magnetic fields ( i.e. : penumbral intraspines ) carrying strong evershed flows . from the inversion of the stokes profiles at these locations we find that the magnetic field strength can either increase or decrease with optical depth . a more detailed inversion of the average stokes vector over the selected pixels , shows that the magnetic field initially decreases , between @xmath72 $ ] , but increases thereafter until @xmath73 . the _ gappy _ penumbral model can explain a smoothly decreasing magnetic field strength only if the @xmath2 level is formed above the field free gap , otherwise a much more sudden decrease would be observed as the line - of - sight penetrates the field - free plasma .
a partial solution to this problem can be found if we assume that the gap is not fully evacuated of magnetic field .
however , it offers no explanation for about 66 % of the selected pixels , where an increasing magnetic field strength with optical depth is observed .
the absence of field - free gaps , as indicated by the inversion , does not in itself imply that there is no form of convection present in the penumbra , but rather suggests that the convective energy transport takes places in the presence of a magnetic field ( see zakharov et al .
2008 ; rempel & schssler 2008 ) .
an example is the roll convection proposed by danielson ( 1961 ) .
all inferred stratifications are compatible with the scenario proposed by the _
uncombed _ model . a magnetic field that decreases smoothly towards the interior of the photosphere can be explained by either a flux tube ( of any field strength ) whose upper boundary layer lies below @xmath2 or , if the upper boundary is above @xmath2 , with a magnetic field inside the flux tube that is weaker than the magnetic field a few hundred kilometers above .
in addition , a magnetic field that increases towards the interior of the photosphere is compatible with a flux tube with an upper boundary layer above @xmath2 , but with a stronger magnetic field than the one above .
this very same configuration can explain a magnetic field that first decreases and then increases with optical depth , as inferred from the inversion of averaged intraspinal profiles .
we have also studied the effects of the scattered light in our inversions .
we have seen that any inaccuracies in its treatment are unlikely to be a source of error in the stratification of the magnetic field strength .
it would be very desirable to make a robust confirmation of our findings , namely @xmath74 in the outer penumbra , for a larger number of sunspots at different heliocentric angles and including also the disk - ward side of the penumbra .
a natural extension of this work would be to use the fe i lines at 1.56 @xmath75 m ( which are formed deeper in the photosphere ) to confirm the absence of field - free regions around @xmath2 .
unfortunately , no such observations exist at the spatial resolution needed to resolve penumbral intraspines ( @xmath76 0.4 `` ) .
indeed , some studies at slightly lower resolution ( 0.6 - 0.7 '' ) have been presented by cabrera solana et al .
( 2008 ) , who used simultaneous observations of fe i 630 nm and 1.56 @xmath75 m recorded with the tip ( martnez pillet et al . 1999 ) and polis ( schmidt et al .
2003 ) instruments .
they found that in the outer penumbra , the horizontal magnetic field component ( carrying the evershed flow ) was no longer weaker than the more vertical one .
this can be used as an independent confirmation of our work , where we routinely find @xmath74 at large radial distances from the center of the sunspot .
in addition , flux tubes with stronger magnetic field than the one of the environment in which they are embedded are also necessary to explain certain aspects of the net circular polarization observed in the outer penumbra of sunspots ( tritschler et al . 2007 ; ichimoto et al .
2008 ) .
bello gonzlez , n. , okunev , o.v .
, domguez cerdea , i. , kneer , f. , puschmann , k.g .
2005 , a&a , 434 , 317 bellot rubio , l.r . , baltasar , h. & collados , m. 2004 , a&a , 427 , 319 bellot rubio , l.r .
2003 , in proceedings of the solar polarization conference .
eds : javier trujillo and jorge snchez almeida .
series , vol .
307 , p 301 .
borrero , j.m . ,
solanki , s.k . ,
bellot rubio , l.r .
, lagg , a. & mathew , s.k .
2004 , a&a , 422 , 1093 borrero , j.m . ,
lagg , a. , solanki , s.k . & collados , m. 2005 , a&a , 436 , 333 borrero , j.m . ,
solanki , s.k .
, lagg , a. , socas - navarro , h. & lites , b.w .
2006 , a&a , 450 , 383 borrero , j.m .
2007 , a&a , 471 , 967 borrero , j.m . ,
lites , b.w . & solanki , s.k .
2008 , a&a , 481 , l13 cabrera solana , d. , bellot rubio , l.r . ,
borrero , j.m . & del toro iniesta , j.c .
2008 , a&a , 477 , 273 danielson , r.e .
1961 apj , 134 , 289 del toro iniesta , j.c .
introduction to spectropolarimetry .
cambridge university press ( cambridge , uk ) ichimoto , k. , tsuneta , s. , suematsu , y. , katsukawa , y. , shimizu , t. , lites , b.w . ,
kubo , m. , tarbell , t.d . , shine , r.a . ,
title , a.m. , nagata , s. , a&a , 481 , l9 kosugi , t. , matsuzaki , k. , sakao , t. et al .
2007 , , 243 , 3 langhans , k. , scharmer , g. , kiselman , d. , lfdahl , m.g . &
berger , t.e .
2005 , a&a , 436 , 1087 lites , b.w . , elmore , d.f . , seagraves , p. & skumanich , a.p 1993 , apj , 418 , 928 lites , q.w . , elmore , d.f .
& streander , k.v .
2001 , in asp conf .
236 , advanced solar polarimetry , ed .
m. sigwarth ( san francisco : asp ) , 33 martnez pillet et al .
1999 , in asp conf .
183 , high resolution solar physics : theory , observations and techniques , ed .
rimmele , k.s .
balasubramanian & r.r .
radick ( san francisco : asp ) , 264 martnez pillet , v. 2000 , a&a , 361 , 734 mathew , s.k . , lagg , a. , solanki , s.k .
2003 , a&a , 410 , 695 orozco surez , d. , bellot rubio , l.r . &
del toro iniesta , j.c .
2007a , apj , 662 , l31 orozco surez , d. , bellot rubio , l.r . , del toro iniesta , j.c .
et al . 2007b , apj , 670 , l61 rempel , m. & schssler , m. 2008 , _ in preparation _ redi , i. , solanki , s.k . ,
keller , c.u .
& frutiger , c. 1998 , a&a , 338 , 1089 ruiz cobo , b. & del toro iniesta 1992 , apj , 398 , 375 scharmer , g.b . ,
gudiksen , b.v . ,
kiselman , d. , lfdahl , m. g. ; rouppe van der voort , l.h .
m. 2002 , nature , 420 , 151 scharmer , g. & spruit , h.c .
2006 , a&a , 460 , 605 schlichenmaier , r. , jahn , k. & schmidt , h.u .
1998 , a&a , 337 , 897 schmidt , w. , beck , c. , kentischer , t. , elmore , d. & lites , b.w .
2003 , astron .
nach . , 324 , 300 shimizu , t. , nagata , s. , tsuneta , s. et al .
2007 , , _ in press _ solanki , s.k .
, & montavon , c.a.p .
1993 , a&a , 275 , 283 solanki , s.k .
2003 , a&arv , 11 , 153 spruit , h.c . &
scharmer , g.b .
2006 , a&a , 447 , 343 stanchfield , d.c.h . ,
thomas , j.h . &
lites , b.w .
1997 , apj , 447 , 485 tritschler , a. , mller , d.a.n , schlichenmaier , r. & hagenaar , h.j .
2007 , apj , 671 , l85 westendorp plaza , c. , del toro iniesta , j.c . , ruiz cobo , b. , martnez pillet , v. , lites , b.w & skumanich , a. 1998 , apj , 494 , 453 westendorp plaza , c. , del toro iniesta , j.c . , ruiz cobo , b. , martnez pillet , v. , lites , b.w & skumanich , a. 2001 , apj , 547 , 1130 zakharov , v. , hirzberger , j. , riethmueller , t. , solanki , s.k . &
kobel , p. 2008 , a&a , _ submitted _ | the vertical stratification of the magnetic field strength in sunspot penumbrae is investigated by means of spectropolarimetric observations at high spatial resolution from the hinode spacecraft . assuming that the magnetic field changes linearly with optical depth we find that , in those regions where the magnetic field is more inclined and the evershed flow is strongest ( penumbral intraspines ) , the magnetic field can either increase or decrease with depth . allowing more degrees of freedom to
the magnetic field stratification reveals that the magnetic field initially decreases from @xmath0 until @xmath1 , but increases again below that .
the presence of strong magnetic fields near the continuum is at odds with the existence of regions void of magnetic fields at , or right below , the @xmath2 level in the penumbra .
however , they are compatible with the presence of a horizontal flux - tube - like field embedded in a magnetic atmosphere . *l * |
jets have been observed around a large variety of astrophysical objects , such as young stellar objects ( ysos ) , accreting white dwarfs , x - ray binaries ( xrbs ) , and active galactic nuclei ( agn ) , and are also thought to drive gamma - ray bursts ( grbs ) . in ysos jets facilitate angular momentum transport , allowing the central star to accrete more matter , and likely play a similar role in accreting black hole systems , where in agn they are thought to also affect the evolution of their host galaxy ( e.g. * ? ? ?
while there are enormous differences between the types and scales of objects around which jets can occur , the origins of jets seem to be remarkably similar , requiring the basic ingredients of infalling / collapsing , rotating matter , and magnetic fields . despite these seemingly simple initial conditions , there are currently many outstanding problems in our understanding of jets , from their creation to their matter content and internal physics .
one important facet of jets observationally is their hallmark synchrotron emission dominating the radio bands in particular , extending up to at least the near infrared ( nir ) in xrbs @xcite . at higher frequencies the typical power - law spectral energy distributions ( seds )
are generally interpreted as optically thin synchrotron emission from accelerated particles , both in agn @xcite and in xrbs @xcite .
an effective way to accelerate radiating particles into a power - law distribution is via diffusive shock acceleration ( e.g. * ? ? ?
* ; * ? ? ?
* ) off scattering centers in turbulent plasma flows .
once initiated , this process must to be distributed throughout the flow to account for the lack of spectral aging over vast distances along the jets .
but where does the acceleration itself begin , and what triggers it ?
there is increasing evidence that the start of this region is offset from the central compact object .
for example in the jet of the agn m87 , recent vlbi observations show the synchrotron emission starting at a region offset by @xmath0 from the core @xcite .
similarly , the start of power - law acceleration in compact jets would be indicated by a transition from optically thick emission , with a flat / inverted spectral index , to an optically thin power - law at a distinct location in the sed .
such a break has been observed directly so far only in one source , the galactic xrb gx 339 - 4 , in the nir @xcite , during the `` hard '' or `` nonthermally dominated '' accretion state associated with compact jet formation ( see state definitions in , e.g. , * ? ? ?
because of the stratification of emission regions in compact jets ( e.g. * ? ? ?
* ) , the lower the frequency where this turnover occurs , the further the location along the jet where particle acceleration starts . in xrbs , a break in the nir corresponds to an offset of @xmath1 @xmath2 , and models of the broadband data of most black hole xrbs in the hard state so far seem to require such a break .
the fact that the start of the acceleration region seems to occur at roughly the same location in several systems could be indicative of a critical point occurring in a magnetohydrodynamical ( mhd ) flow , particularly the magnetosonic modified fast point ( mfp ) . at the mfp the collimating magnetic field lines turn inwards towards the jet axis , potentially leading to recollimation shocks , while at the same time the flow becomes causally disconnected so shocks can occur without disrupting the flow upstream .
such a shock region thus would occur at a fixed location in the flow , closely connected to the mfp , and would be an ideal location for particle acceleration to begin .
we wish to investigate the feasiblity of this premiss in this paper .
because of the complexity involved in relativistic mhd including strong gravity , many groups are using the results of simulations to study the formation and development of jets .
these simulations often show the development of a steady outflow in which the magnetic field is remarkably self - similar and axisymmetric near the launch point ( see , e.g. , fig . 2 and fig
. 11 in * ? ? ?
however to study specifically the development of critical points in the flow and their dependence on external boundary conditions , current mhd simulations either do not extend far enough from the black hole , or if they do , they are too computationally expensive .
assuming that the resulting flows retain a self - similar structure , at least when gravity does not dominate as indicated in the simulations , we adopt the formalism developed by ( * ? ? ?
* hereafter vk03 ) . by assuming axisymmetry and a self - similar field line geometry , vk03 reduce the exact equations of special relativistic mhd to a one - dimensional problem .
although vk03 focused on jets in grbs , this treatment is also applicable to other mhd jets such as in agn and microquasars . in an earlier self - similar treatment that was non - relativistic
, @xcite presented a solution where the flow crosses the mfp at a finite height above the disk . in vk03 , however , they only found relativistic solutions with an mfp occurring at infinity ( meaning that the flow asymptotically approaches a perfect cylindrical geometry ) . in this paper
we extend the study of vk03 and derive new solutions in which the relativistic flow crosses the mfp at a finite location above the disk . in 2 we describe the vk03 model and our method for exploring the full parameter space of solutions . in
3 we present the first relativistic solutions that pass through an mfp . in 4
we discuss our results , and the dependence of the mfp location on the model parameters .
we also describe how this work sets the stage for further development to connect the flow to regions near the disk where gravity can no longer be ignored .
5 contains our conclusions .
during the acceleration and collimation of jets , magnetic fields are thought to efficiently extract rotational energy from either the compact object @xcite or the accretion disk @xcite .
the latter models are in the newtonian limit , with the matter considered cold , meaning there is negligible thermal pressure causing bulk acceleration to non - relativistic velocities .
these cold , non - relativistic solutions were generalized to the relativistic regime by @xcite , allowing the bulk velocity to attain relativistic speeds . vk03
further extended the solutions to include the `` hot '' regime , allowing the random motions of the particles to become relativistic and thus the jets to be hydrodynamically accelerated even at the base where temperatures are high .
it is this last scenario that we base this work upon .
starting from the equations of time - dependent special relativistic mhd , we make the following assumptions to render them more tractable : ideal mhd , no gravitational field or external force ( and thus self - similarity holds ) , axisymmetry , a zero azimuthal electric field and time independence . following the terminology of vk03 , after scaling the equations to make them non - dimensional , we are left with two coupled differential equations ( equations and in the appendix ) .
combining these two coupled differential equations , we obtain a single equation for @xmath3 ( equation , with @xmath4 the alfvnic mach number and @xmath5 the angle of the point on the field line with the axis of symmetry ) which acts like a `` wind equation '' , much akin to the wind equation of the parker solar wind model @xcite .
this wind equation , along with the other algebraic equations ( see appendix ) , can be solved for the velocity , magnetic and electric field strength , density and pressure along a field line . due to the self - similar assumption ,
once solutions are obtained for one field line , all other field lines can be obtained by simple scaling .
an example of this self - similarity and the meaning of some of the parameters used can be found in figure [ geometry ] . instead of the single critical ( or sonic ) point of the parker solar wind model , due to the inclusion of magnetic fields , the obtained wind equation has three locations where the denominator crosses zero .
starting from the accretion disk , these are the modified slow point ( msp ) , the alfvn point , and the mfp .
the msp and mfp are also called the slow and fast magnetosonic separatrix surfaces .
the alfvn point is the location where the relativistic collimation speed of the flow towards the axis ( @xmath6 ) is given by @xmath7 where @xmath8 is the lorentz factor , @xmath9 the strength of the magnetic field , @xmath10 the cylindrical radius in units of the light cylinder radius , @xmath11 the baryon rest - mass density , and @xmath12 the specific relativistic enthalpy ( the variables are described in more detail in [ variables ] ) .
the denominator of @xmath3 with the alfvn point divided out , can be expressed as @xmath13 \nonumber \\ & + \frac{u_\mathrm{s}^2}{c^2 } \frac{b_\theta^2 ( 1 - x^2)}{4 \pi \rho_0 \xi c^2},\end{aligned}\ ] ] with @xmath14 @xmath15 the velocity of light , @xmath16 the strength of the electric field , and @xmath17 the polytropic index .
the msp and mfp are , by definition , the locations where @xmath18 ( vk03 ) . at every critical point
, the numerator of @xmath3 should also pass through zero to ensure a smooth crossing .
this translates into a regularity condition at the critical points and fixes the value of a free parameter .
even though the msp should be crossed smoothly to obtain a solution that describes the entire jet from the accretion disk to the termination point , gravitational effects can not be ignored at the msp . since the equations do not include gravity ( as it is not compatible with the self - similarity assumption in relativistic flow ) , we do not try to fit for the msp .
therefore we fit two critical points , and correspondingly two parameters are fixed , in our approach described below @xmath19 for the alfvn point and @xmath20 for the mfp .
the physical importance of the mfp is that it is the location where not even the fastest signals can travel upstream anymore , meaning anything downstream from the mfp is causally disconnected from the region upstream .
if , for example , a shock were to exist beyond the mfp , it could not disrupt the flow leading to that shock , allowing it to be a permanent feature .
as mentioned above , at the mfp there is a component of the velocity heading towards the axis .
this can lead to a collimation shock shortly beyond the mfp , causing the magnetic energy to be converted to particle energy and the jet to become kinetically dominated .
another possibility is that the flow remains magnetically dominated , and , after reaching a minimum radius , bounces back , retaining an ordered magnetic field ( see , e.g. , * ? ? ?
this is not in conflict with the statement in vk03 that the only physically acceptable case in the super - alfvnic regime is for the flow to become asymptotically cylindrical , as this statement only applies to solutions with @xmath21 .
the prescription we are following from vk03 has 9 free parameters that determine the solution , whose effects are described below . following the same notation , a roman subscript a signifies the value of a variable at the alfvn point and an italic _ a _ denotes a value with respect to the poloidal magnetic flux function .
the exponent @xmath22 determines the current distribution .
a value @xmath21 corresponds to the current - carrying regime , with higher values of f ensuring faster collimation , but if @xmath23 we are in the return - current regime .
the restriction on @xmath22 is that it can not be negative : .
although we consider @xmath22 to be a free parameters , for this paper we chose to keep it fixed at 0.75 .
the adiabatic index @xmath17 can have values of @xmath24 for relativistic and @xmath25 for non - relativistic solutions .
@xmath26 gives the angle where the alfvn point is located with respect to the axis of symmetry .
it is limited by the value of @xmath27 : .
@xmath27 gives the poloidal slope of the field line at the alfvn point with respect to the accretion disk .
this in turn is limited by @xmath26 : .
@xmath28 is the radius squared of the alfvn point in terms of the light cylinder radius . for @xmath29
the solution becomes more force - free .
the allowed values are .
@xmath20 is the magnetization parameter in the monopole solution of @xcite and is related to the mass - to - magnetic flux ratio .
the constraint is .
@xmath30 is the dimensionless adiabatic coefficient and is constant along a field line . for a large value of @xmath30 ,
the specific relativistic enthalpy of the matter is high , for @xmath31 equation shows @xmath32 and the flow is cold . therefore .
@xmath19 is the derivative of @xmath33 with respect to the polar angle @xmath5 at the alfvn point . accelerating flow implies .
@xmath34 , with reference magnetic field @xmath35 and reference length @xmath36 , is the scaling of the solution , relating the dimensionless values to physical dimensions .
we do not yet apply our solutions to specific black hole systems , so this parameter is not used here but it will be important for future applications of our solutions .
the smooth crossing of the alfvn point is ensured by calculating @xmath19 from the corresponding regularity condition , given by equation . in the same way we determine @xmath20 by crossing the mfp , although this parameter sometimes can have two values ( see figure [ 3d ] ) .
this leaves @xmath22 , @xmath17 , @xmath26 , @xmath27 , @xmath28 , and @xmath30 ( and @xmath34 ) to satisfy the boundary conditions at the source and , indirectly , at the end of the jet .
there are also parameters derived from the above values : @xmath37 determines the total energy - to - mass - flux ratio ( @xmath38 ) and is conserved along a field line .
this parameter is determined from equation and .
@xmath39 is the value of the magnetization function @xmath40 , defined as the poynting - to - matter energy flux , at the alfvn point .
this parameter is determined from equation . as @xmath37 can not be negative , from equation
follows .
and finally we describe the other variables used in the equations : @xmath41 determines the specific ( per baryon mass ) relativistic enthalpy ( @xmath12 ) .
if @xmath42 we have cold , pressureless matter .
@xmath41 will drop from the high temperatures at the base of the jet as matter is mainly accelerated hydrodynamically and at some point above the disk drop down to 1 . from this point on
all acceleration is magnetic .
this variable is determined from equation .
@xmath4 is the alfvnic mach number , the velocity of the flow in terms of the alfvn velocity .
@xmath43 is the radius in terms of @xmath44 and is therefore equal to 1 at the alfvn point . to find solutions with an mfp we obtained expressions for @xmath45 , given by equation , and for @xmath46 by combining the derivative of the bernoulli equation with the transfield equation using the determinant method . because @xmath46 is very unstable near the alfvn point , as the numerator has a first order zero point there and the denominator a second order one , we reverted to the bernoulli equation , given by , to determine @xmath47 . to start off the integration from the alfvn point
we specify @xmath22 , @xmath17 , @xmath26 , @xmath28 , @xmath27 , @xmath30 and an initial guess for @xmath20 .
we determine @xmath19 from the alfvn regularity condition , equation . integrating outward from the alfvn point , we determine whether the numerator or denominator crosses zero first and adjust @xmath20 accordingly until both cross at the same time .
we then proceed to explore the range of solutions which cross both the alfvn point and mfp ( see figure [ solutionc ] ) . for plotting purposes ,
we have divided out the factor @xmath48 in both the numerator and denominator . as discussed above ,
because there is no gravity in the model , we do not try explicitly to cross the modified slow point ( msp ) .
gravitational effects should play a large role close to the black hole , and the self - similar equations can not predict accurately where the msp is located . to begin our exploration of parameter space , we chose a solution from @xcite known to have an mfp ( specified below their figure 4 ) with parameters @xmath49 , @xmath50 , @xmath51 , @xmath52 , @xmath53 , @xmath54 , @xmath55 , @xmath56 and @xmath57 . by comparing terms in @xcite and vk03 , it is possible to translate these parameters to the parameters used in vk03 . because we are using the relativistic equations from vk03 , the parameters of our first solution ( see table [ parametertable ] ) differ slightly from the corresponding parameters above and we vary @xmath20 to obtain a critical solution again . from this solution
we were able to traverse parameter space , while allowing only critical solutions with an mfp . by increasing @xmath28
we were able to obtain higher velocities of the jet . after achieving relativistic velocities , we focused on finding solutions with higher values of @xmath30 , as the solutions so far were cold .
but for a fixed value of @xmath28 there is a maximum value of @xmath30 that produces critical solutions crossing the mfp .
this is due to the fact that the collection of all solutions form a surface in the multidimensional parameter space , and we had reached a maximum for @xmath30 for the fixed values of the other parameters ( see figure [ 3d ] ) .
the surfaces of valid solutions have roughly the same appearance for the explored range of @xmath26 and @xmath28 . to describe the effect of @xmath26 and @xmath28 on the solutions , we approximate the graph as a cone with the base in the @xmath58-plane and with the maximally allowed value of @xmath30 as its height
. if we increase @xmath44 the base of the cone shrinks while moving to higher @xmath20 , and the height decreases . the latter limits the maximum value that @xmath28
can be increased to .
if we decrease @xmath26 , the height increases , indirectly allowing higher values for @xmath28 . the area of the base becomes bigger and shifts to higher @xmath27 and @xmath20 , with the upper surface becoming steeper .
due to the shape of the surface , it is possible to increase any two parameters of @xmath28 , @xmath30 and @xmath20 at the expense of the third . by extending our search to three parameters ( @xmath27 , @xmath30 and @xmath20 ) , we were able to move around this point and continue increasing @xmath30 .
this revealed a multidimensional surface that is double - valued in @xmath20 .
an example of this surface is shown in figure [ 3d ] , which also includes our most relativistic solution presented below , solution @xmath15 .
the numerator , denominator , and acceleration of @xmath33 in this latter solution are also plotted in figure [ solutionc ] .
in this section we present the various solutions crossing the mfp , from the first one that we found to one with relativistic temperature and bulk flow we sought , while describing the features particular to a certain solution .
the parameters of our solutions are given in table [ parametertable ] and the main properties in figure [ overview ] .
as we are not yet applying our solutions to specific black hole systems , we do not use the scaling parameters @xmath34 and @xmath36 . [ cols="<,^,^,^,^,^,^,^,^,^,^ " , ] + note.the values for the first six parameters ( @xmath22 through @xmath27 ) are exact , for the last four ( @xmath20 through @xmath37 ) they are rounded off . [ parametertable ] this solution is the closest to the non - relativistic parameter values given in @xcite that conforms to the alfvn regularity condition ( arc ) , which is the transfield equation at the alfvn point .
the solution crosses the mfp at @xmath60 or @xmath61 ( @xmath62 ) .
the alfvn point is located at @xmath63 or @xmath64 .
the top left panel of figure [ overview ] gives the meridional projection of the magnetic field lines .
the jet overcollimates after a maximum radius of almost 16 times the alfvn radius at @xmath65 or @xmath66 , shortly before the mfp .
the second left panel of figure [ overview ] shows the flow is cold throughout ( @xmath67 ) due to the very small value for @xmath30 and low @xmath28 .
as @xmath68 at the alfvn point , equation gives a value for @xmath69 very close to 1 .
the energy of the matter @xmath41 ( including the dominant rest mass energy ) is much higher than the energy in the magnetic field throughout .
therefore , even though the magnetic acceleration is efficient , the jet is not accelerated to relativistic velocities ( @xmath70 ) .
the third center panel shows the `` causal connection '' opening angle @xmath71 and the opening half - angle of the outflow , which goes from @xmath64 to a few degrees overcollimation .
although the causal connection opening angle has little importance for non - relativistic flows , as it remains very close to @xmath72 since @xmath73 , it is shown for completeness .
after our first solution , we increased the velocity of our jet by increasing @xmath28 .
as the top center panel in figure [ overview ] shows , after a long period where the field line remains almost parabolic , the jet in this solution overcollimates as well .
this is caused by magnetic hoop stresses and may allow a shock region to develop beyond the mfp . the alfvn point is again located at @xmath63 and the mfp at @xmath75 or @xmath76 ( @xmath77 ) . the second center panel shows the lorentz factor and the enthalpy of the flow .
the flow here also is seen to be cold ( @xmath78 ) , meaning the jet is mainly magnetically accelerated , which is again due to the small value of @xmath30
. the poynting - to - mass flux ratio ( @xmath79 ) decreases , showing magnetic energy being transferred into kinetic energy , with the flow reaching a lorentz factor of 2.8 .
the bottom center panel shows the squares of the light cylinder radius , @xmath80 , and the alfvnic mach number , @xmath4 .
when @xmath81 the light surface is reached , which is the radius where the field circular velocity reaches the speed of light . after having achieved a relativistic solution for the cold plasma case , we would like to find solutions with an increased flow temperature . to do
so requires increasing the value of @xmath41 .
as @xmath33 is given by @xmath82 at the alfvn point , equation shows that increasing @xmath28 and/or @xmath30 has the desired effect .
unfortunately , for larger @xmath28 the maximum value of @xmath30 decreases . by choosing a lower value for @xmath26 the attainable values for @xmath28 and @xmath30
are increased , leading to a warmer flow .
this solution is shown in the third column of figure [ overview ] .
the alfvn point is located at @xmath83 or @xmath84 and the mfp at @xmath85 or @xmath86 ( @xmath87 ) .
near the beginning of the flow @xmath88 . the lorentz factor of the flow at the mfp is 8.3 , which is mainly due to the high initial poynting flux .
it can be seen in the second right panel of figure [ overview ] that @xmath41 always drops to 1 and from the fourth right panel that @xmath33 always dominates @xmath89 near the mfp . as the lorentz factor is given by @xmath90
the final lorentz factor is approximately @xmath37 .
this means that while the jets may start as poynting flux - dominated , eventually they convert most of their poynting flux and become kinetic energy - dominated .
since we are interested in the location of the mfp , we would like to know how it depends on the model parameters . as it is challenging to sample the full parameter space , we will focus on the region around the solution closest to observed jets , solution @xmath15 . by allowing all parameters to vary and looking at the effect this has on the location where the mfp occurs , we can draw the following conclusions : the mfp moves outward ( smaller @xmath5 ) when the alfvn point occurs at a smaller angle , ( lower @xmath26 ) , when the temperature at the base of the flow is increased ( higher @xmath30 ) , when the flow at the alfvn point is already close to collimation ( higher @xmath27 ) , or when the alfvn point moves closer to the light cylinder radius , making the flow more force - free ( higher @xmath44 ) .
we have succeeded in obtaining new , solutions for a relativistic , magnetized flow that smoothly crosses the mfp at a finite height ( @xmath91 ) above the system equator .
these solutions suggest that it should be possible to construct better , more mhd - consistent jet models where the location of the acceleration region is determined _ a priori _ from the physical boundary conditions .
so far none of the solutions derived remains poynting flux - dominated up to the mfp , which is probably due to the relatively small value of @xmath20 found so far .
increasing @xmath28 and especially decreasing @xmath26 will allow higher values of @xmath20 to be used .
the same can also be done at the expense of @xmath30 .
we also have described some of the relations between the different parameters ( see [ variables ] ) and the effect they have on each other .
having a more force - free solution ( higher @xmath28 ) decreases the allowed range of temperatures ( @xmath30 ) and collimation at the alfven point ( @xmath27 ) , but at the same time allows a higher value of @xmath20 that provides a critical solution . keeping all other parameters fixed , there is a maximum value of @xmath28 for which a solution is possible at all .
this maximum may be increased by moving the alfvn point closer to the disk ( smaller @xmath26 ) .
this change has the additional effects of allowing a broader range for the collimation angle at the alfvn point ( @xmath27 ) while at the same time shifting this range towards higher collimation .
it also allows for a higher temperature of the flow ( @xmath30 ) or a higher magnetic field strengths ( @xmath20 ) .
any pair of parameters @xmath28 , @xmath30 and @xmath20 may be increased at the expense of the third .
two parameters not varied so far are @xmath22 and @xmath17 .
higher values for @xmath22 should ensure faster collimation and might therefore be very important for the exact location of the mfp .
similarly , we may want to explore the @xmath92 case for jets with extremely relativistic temperature .
however , for the weaker jets in agn and xrbs that we plan to target , the radiating particle distributions are generally thought to peak at mildly relativistic energies .
if the start of the particle acceleration region in steady jets is indeed associated with the magnetosonic fast critical point in the bulk flow , then our results support the conclusion that such a region could occur at a fairly stable location about the launch point .
all of the solutions found overcollimate shortly before the mfp , as would be expected for the initiation of shock development . by starting with a solution that crosses the mfp in the non - relativistic case of @xcite
, we were able to extend the solution through the multidimensional parameter space towards relativistic velocities and temperatures , while retaining the critical point .
this feature sets our results apart from the work of vk03 , whose formulism we adopted , in that they are immediately applicable to observed compact jet sources with an optically thick - to - thin break in the synchrotron spectrum , such as hard state xrbs and weakly accreting agn .
our most promising solution is a jet outflow with mildly relativistic temperature , and lorentz factor of @xmath93 , also appropriate for the steady jets in both xrbs as well as agn .
it is clear that a wide range of parameter space is left still unexplored , that can be exploited for matching physical boundary conditions appropriate to known astrophysical sources .
however , before a radiative model can be constructed around the dynamical `` backbone '' provided by the solutions presented here , a prescription for including gravity must be included , to extend these solutions through the msp and allow connection with a physical model of the accretion flow / corona .
we are currently working on matching these necessarily non - self similar solutions to those presented here , which will be presented in a separate work .
once this solution is in place , we will have a much more physically consistent model ( compared to , e.g. , * ? ? ?
* ) to use in the fitting of data from accreting black holes across the mass scale , which show compact , steady jets . as there seems to be no shortage of possible solutions , we are confident that we can match physical boundary conditions with critical solutions .
p. polko and s. markoff gratefully acknowledge support from a netherlands organization for scientific research ( nwo ) vidi fellowship .
in addition , s. markoff is grateful for support from the european community s seventh framework program ( fp7/2007 - 2013 ) under grant agreement number itn 215212 `` black hole universe '' .
part of the research described in this paper was carried out at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration .
we would like to thank the anonymous referee for helpful comments that improved this manuscript .
here we list the equations we have used for reference .
see [ variables ] for a description of the parameters and variables .
the transfield equation is given by @xmath98 & = ( f-1 ) \frac{x_{\mathrm{a}}^4
\mu^2 x^2}{f^2 \sigma_{\mathrm{m}}^2 } \left ( \frac{1 - g^2}{1 - m^2 - x^2 } \right)^2 \nonumber \\ & - \sin^2(\theta ) \frac{m^2 + f
x^2 - f + 1}{\cos^2(\psi + \theta ) } \nonumber \\ & - \frac{x_{\mathrm{a}}^4 \mu^2 x^2}{f^2 \sigma_{\mathrm{m}}^2 m^2 } \left ( \frac{g^2 - m^2 - x^2}{1 - m^2 - x^2 } \right)^2 \nonumber \\ & + 2 \frac{\gamma - 1}{\gamma } \frac{f - 2}{f^2 \sigma_{\mathrm{m}}^2 } \frac{\xi ( \xi - 1 ) x^4}{m^2 } \label{transfield}\end{aligned}\ ] ] the magnetization function @xmath40 and the fractions at the alfvn point are given by @xmath99 we obtain @xmath100 by inserting these relations into the bernoulli equation .
@xmath101 ^ 2 } \left[x_{\mathrm{a}}^2 \xi_{\mathrm{a}}^2 + \frac{f^2 \sigma_{\mathrm{m}}^2 \left(1 - x_{\mathrm{a}}^2 \right)^2 \sin^2(\theta_{\mathrm{a}})}{x_{\mathrm{a}}^2 \cos^2(\theta_{\mathrm{a}}+ \psi_{\mathrm{a } } ) } \right ] \label{mu2}\ ] ] the alfvn regularity condition is obtained by substituting these relations into the transfield equation .
@xmath102 \bigg\ } \nonumber \\ = \left [ x_{\mathrm{a}}^2 - \sigma_{\mathrm{a}}(1 - x_{\mathrm{a}}^2 ) \right]^2 - ( f - 1 ) \sigma_{\mathrm{a}}^2 ( 1 - x_{\mathrm{a}}^2 ) - 2 \frac{\gamma - 1}{\gamma } ( f - 2 ) \frac{\xi_{\mathrm{a}}- 1}{\xi_{\mathrm{a } } } \left\ { x_{\mathrm{a}}^2 - \left [ x_{\mathrm{a}}^2 - \sigma_{\mathrm{a}}(1 - x_{\mathrm{a}}^2 ) \right]^2 \right\ } \label{arc}\end{aligned}\ ] ] [ windequation ] @xmath103 where @xmath104 } \right ) \bigg\ } \\
b_1 & = - 2 f^2 \sigma_{\mathrm{m}}^2 m^4 \sin^2(\theta ) \tan(\psi + \theta ) g^2 ( 1 - m^2 - x^2)^2 \\ c_1 & = x^4 \cos^2(\psi + \theta ) \mu^2 \frac{{\mathrm{d}}g^2}{{\mathrm{d}}\theta } \left [ - \left ( 1 - m^2 - x_{\mathrm{a}}^2 \right)^2 + 2 x_{\mathrm{a}}^2 \left ( g^2 - m^2 -x^2 \right ) ( 1 - x_{\mathrm{a}}^2 ) \right ] \nonumber \\ & \qquad + \frac{{\mathrm{d}}g^2}{{\mathrm{d}}\theta } ( 1 - m^2 -x^2 ) \left ( 1- m^2 - 3 x^2 \right ) \left [ \xi^2 x^4 \cos^2(\psi + \theta ) + f^2 \sigma_{\mathrm{m}}^2 m^4
\sin^2(\theta ) \right ] \nonumber \\ & \qquad + 2 f^2 \sigma_{\mathrm{m}}^2 m^4 \sin^2(\theta ) g^2 ( 1 - m^2 - x^2)^2 \left [ \frac{\cos(\theta)}{\sin(\theta ) } - \frac{1}{g^2 } \frac{{\mathrm{d}}g^2}{{\mathrm{d}}\theta } + \tan(\psi + \theta ) \right ] \\
a_2 & = -\sin^2(\theta ) \tan(\psi + \theta ) \\
b_2 & = \sin^2(\theta)\frac{1 - m^2 - x^2}{\cos^2(\psi + \theta ) } \\ c_2 & = -\sin^2(\theta)\frac{1- m^2 - x^2}{\cos^2(\psi + \theta ) } + \sin^2(\theta ) \tan(\psi + \theta ) \left [ x_{\mathrm{a}}^2 \frac{{\mathrm{d}}g^2}{{\mathrm{d}}\theta } + ( 1 - m^2 - x^2 ) \frac{1}{g } \frac{{\mathrm{d}}g}{{\mathrm{d}}\theta } \right ] \nonumber \\ & \qquad + ( f-1 ) \frac{x_{\mathrm{a}}^4
\mu^2 x^2}{f^2 \sigma_{\mathrm{m}}^2 } \left ( \frac{1 - g^2}{1 - m^2 - x^2 } \right)^2 - \sin^2(\theta ) \frac{m^2 + f x^2 - f + 1}{\cos^2(\psi + \theta ) } \nonumber \\ & \qquad - \frac{x_{\mathrm{a}}^4 \mu^2 x^2}{f^2 \sigma_{\mathrm{m}}^2 m^2 } \left ( \frac{g^2 - m^2 - x^2}{1 - m^2 - x^2 } \right)^2 + 2 \frac{\gamma - 1}{\gamma } \frac{f - 2}{f^2 \sigma_{\mathrm{m}}^2 } \frac{\xi ( \xi - 1 ) x^4}{m^2}\end{aligned}\ ] ] | observations of relativistic jets from black holes systems suggest that particle acceleration often occurs at fixed locations within the flow .
these sites could be associated with critical points that allow the formation of standing shock regions , such as the magnetosonic modified fast point . using the self - similar formulation of special relativistic magnetohydrodynamics by @xcite
, we derive a new class of flow solutions that are both relativistic and cross the modified fast point at a finite height .
our solutions span a range of lorentz factors up to at least 10 , appropriate for most jets in x - ray binaries and active galactic nuclei , and a range in injected particle internal energy . a broad range of solutions exists , which will allow the eventual matching of these scale - free models to physical boundary conditions in the analysis of observed sources . |
the focus of this paper is on the following structured minimization @xmath1 where @xmath2 is a continuously differentiable ( may be nonconvex ) function that is bounded below ; @xmath3 denotes the @xmath0-norm of a vector ; parameter @xmath4 is used to trade off both terms for minimization . due to its structure , problem ( [ probtype ] ) covers a wide range of apparently related formulations in different scientific fields including linear inverse problem , signal / image processing , compressive sensing , and machine learning .
a popular special case of model ( [ probtype ] ) is the @xmath0-norm regularized least square problem @xmath5 where @xmath6 @xmath7 is a linear operator , and @xmath8 is an observation . model ( [ onenorm ] ) mainly appears in compressive sensing an emerging methodology in digital signal processing , and has attracted intensive research activities over the past years @xcite .
compressive sensing is based on the fact that if the original signal is sparse or approximately sparse in some orthogonal basis , an exact restoration can be produced via solving problem ( [ onenorm ] ) .
another prevalent case of ( [ probtype ] ) that has been achieved much interest in machine learning is the linear and logistic regression . given the training date @xmath9^\top\in\mathbb{r}^{m\times n}$ ] and class labels @xmath10 .
a linear classifier is a hyperplane @xmath11 , where @xmath12 is a set of weights and @xmath13 is the intercept .
a frequently used model is the @xmath14-loss support vector machine @xmath15 because of the `` max '' operation , the @xmath14-loos function is continuous , but not differentiable . based on the conditional probability , another popular model is the logistic regression @xmath16 obviously , the logistic loss function is twice differentiable .
although the models of these problems have similar structures , they may be very different from real - data point of view .
for example , in compressive sensing , the length of measurement @xmath17 is much smaller than the length of original signal @xmath7 and the encoding matrix @xmath18 is dense .
however , in machine learning , the numbers of instance @xmath17 and features @xmath19 are both large and the data @xmath18 is very sparse . since the @xmath0-regularized term is non - differentiable when @xmath20 contains values of zero , the use of the standard unconstrained smooth optimization tools are generally precluded . in the past decades ,
a wide variety of approaches has been proposed , analyzed , and implemented in compressive sensing and machine learning literatures .
this includes a variety of algorithms for special cases where @xmath21 has a specific functional form such as the least square ( [ onenorm ] ) , the square loss ( [ svm ] ) and the logistic loss ( [ logmodel ] ) . in the following ,
we briefly review some of them in each literature .
the first popular approach falls into the coordinate descent method . at the current iterate @xmath22 , the simple coordinate descent method updates one component at a time to generate @xmath23 , @xmath24 , such that @xmath25 , @xmath26 , and solves a one - dimensional subproblem @xmath27 where @xmath28 is defined as the @xmath29-th column of an identity matrix
clearly , the objective function has one variable , and one non - differentiable point at @xmath30 . to solve the logistic regression model ( [ logmodel ] ) ,
bbr @xcite solves the sub - problem approximately by the use of trust region method with newton step ; cdn @xcite improves bbr s performance by applying a one - dimensional newton method and a line search technique . instead of cyclically updating one component at each time , the stochastic coordinate descent method @xcite randomly selects the working components to attain better performance ; the block coordinate gradient descent algorithm cgd @xcite is based on the approximated convex quadratic model for @xmath31 , and selects the working variables with some rules .
the second type of approach is to transform model ( [ probtype ] ) into an equivalent box - constrained optimization problem by variable splitting .
let @xmath32 with @xmath33 and @xmath34 .
then , model ( [ probtype ] ) can be reformulated equivalently as @xmath35 the objective function and constraints are smooth , and therefore , it can be solved by any standard box - constrained optimization technique . however , an obvious drawback of this approach is that it doubles the number of variables .
gpsr @xcite solves ( [ subbound ] ) , and subsequently solves ( [ onenorm ] ) , by using barzilai - borwein gradient method @xcite with an efficient nonmonotone line search @xcite .
it is actually an application of the well - known spectral projection gradient @xcite in compressive sensing .
trust region newton algorithm tron @xcite minimizes ( [ subbound ] ) , then solves the logistic regression model ( [ logmodel ] ) , and exhibits powerful vitality by a series of comparisons . to solve ( [ svm ] ) and ( [ logmodel ] ) , the interior - point algorithm @xcite
forms a sequence of unconstrained approximations by appending a ` barrier ' function to the objective function ( [ subbound ] ) which ensures that @xmath36 and @xmath37 remain sufficiently positive . moreover , truncated newton steps and preconditioned conjugate gradient iterations are used to produce the search direction .
the third type of method is to approximate the @xmath0-regularized term with a differentiable function .
the simple approach replaces the @xmath0-norm with a sum of multi - quadric functions @xmath38 where @xmath39 is a small positive scalar .
this function is twice - differentiable and @xmath40 .
subsequently , several smooth unconstrained optimization approaches can be applied , based on this approximation .
however , the performance of these algorithms is much influenced by the parameter values , and the condition number of the corresponding hessian matrix becomes larger as @xmath39 decreases .
the nesterov s smoothing technique @xcite is to construct smooth functions to approximate any general convex nonsmooth function .
based on this technique , nesta @xcite solves problem ( [ onenorm ] ) by using first - order gradient information .
the fourth type of approach falls into the subgradient - based newton - type algorithm .
the important attempt in this class is from andrew and gao @xcite , who extend the well - known limited memory bfgs method @xcite to solve @xmath0-regularized logistic regression model ( [ logmodel ] ) , and propose an orthant - wise limited memory quasi - newton method owl - qn . at each iteration
, this method computes a search direction over an orthant containing the previous point . the subspace bfgs method subbfgs @xcite involves an inner iteration approach to find the descent quasi - newton direction and a subgradient wolfe - condtions to determine the stepsize which ensures that the objective functions are decreasing .
this method enjoys global convergence and is capable of solving general nonsmooth convex minimization problems . finally , to solve model ( [ onenorm ] ) , besides gpsr and nesta , there are other numerous specially designed solvers . by an operator splitting technique , hale , yin and zhang derive the iterative shrinkage / thresholding fixed - point continuation algorithm ( fpc ) @xcite . by combining the interior - point algorithm in @xcite , fpc
is also extended to solve large - scale @xmath0-regularized logistic regression in @xcite .
twist @xcite and fista @xcite speed up the performance of ist and have virtually the same complexity but with better convergence properties .
another closely related method is the sparse reconstruction algorithm sparsa @xcite , which is to minimize non - smooth convex problem with separable structures .
spgl1 @xcite solves the lasso model ( [ onenorm ] ) by the spectral gradient projection method with an efficient euclidean projection on @xmath0-norm ball .
the alternating directions method yall1 @xcite , investigates @xmath0-norm problems from either the primal or the dual forms and solves @xmath0-regularized problems with different types .
all the reviewed algorithms differ in various aspects such as the convergence speed , ease of implementation , and practical applicability .
moreover , there is no enough evidence to verify that which algorithm outperforms the others under all scenarios .
although much progress has been achieved in solving the problem ( [ probtype ] ) , these algorithms mainly deal with the case where @xmath31 is a convex function even a least square . in this paper , unlike all the reviewed algorithms , we propose a barzilai - borwein gradient algorithm for solving @xmath0-regularized nonsmooth minimization problems . at each iteration , we approximate @xmath31 locally by a convex quadratic model , where the hessian is replaced by the multiplies of a spectral coefficient with an identity matrix . the search direction is determined by minimizing the quadratic model and taking full use of the @xmath0-norm structure .
we show that the generated direction is descent which guarantees that there exists a positive stepsize along the direction . in our algorithm
, we adopt the nonmonotone line search of grippo , lampariello , and lucidi @xcite , which allows the function values to increase occasionally in some iteration but decrease in the whole iterative process .
the attractive property of the nonmonotone line search is that it saves much number of function evaluations which should be the main computational burden in large dataset .
the method is easily performed , where only the value of objective function and the gradient of the smooth term are needed at each iteration .
we show that each cluster of the iterates generated by this algorithm is a stationary point of @xmath41 . in this paper , although we mainly consider the @xmath0-regularizer , the @xmath14-norm regularization problem and the matrix trace norm problems can also be readily included in our framework .
thus , this broaden the capability of the algorithm .
we implement the algorithm to solve problem ( [ probtype ] ) where @xmath31 is a nonconvex smooth function from cuter library to show its efficiency .
moreover , we also run the algorithm to solve @xmath0-regularized least square , and do performance comparisons with the state - of - the - art algorithms nesta , cgd , twist , fpc and gpsr .
the comparisons results show that the proposed algorithm is effective , comparable , and promising .
we organize the rest of this paper as follows . in section [ algorithmsec ] ,
we briefly recall some preliminary results in optimization literature to motivate our work , construct the search direction , and present the steps of our algorithm along with some remarks . in section
[ theory ] , we establish the global convergence theorem under some mild conditions . in section [ secexten ] , we show that how to extend the algorithm to solve @xmath14-norm and matrix trace norm minimization problems . in section [ expnoncov ] ,
we present experiments to show the efficiency of the algorithm in solving the @xmath0-regularized nonconvex problem and least square problem .
finally , we conclude our paper in section [ concludsec ] .
first , consider the minimization of the smooth function without the @xmath0-norm regularization @xmath42 the basic idea of newton s method for this problem is to iteratively use the quadratic approximation @xmath43 to the objective function @xmath21 at the current iterate @xmath22 and to minimize the approximation @xmath43 .
let @xmath44 be twice continuously differentiable , and its hessian @xmath45 be positive definite .
function @xmath31 at the current @xmath22 is modeled by the quadratic approximation @xmath43 , @xmath46 where @xmath47 .
minimizing @xmath48 yields @xmath49 which is newton s formula and @xmath50 is the so - called newton s direction .
for the positive definite quadratic function , newton s method can reach the minimizer with one iteration . however , when the starting point is far away from the solution , it is not sure that @xmath51 is positive definite and newton s direction @xmath52 is a descent direction .
let the quadratic model of @xmath31 at @xmath53 be @xmath54 finding the derivative yields @xmath55 setting @xmath56 , @xmath57 , and @xmath58 we get @xmath59 for various practical problems , the computing efforts of the hessian matrices are very expensive , or the evaluation of the hessian is difficult ; even the hessian is not available analytically .
these lead to the quasi - newton method which generates a series of hessian approximations by the use of the gradient , and at the same time maintains a fast rate of convergence . instead of computing the hessian @xmath60
, quasi - newton method constructs the hessian approximation @xmath61 , where the sequence @xmath62 possesses positive definiteness and satisfies @xmath63 in general , such @xmath64 will be produced by updating @xmath65 with some typical and popular formulae such as bfgs , dfp , and sr1 .
unfortunately , the standard quasi - newton algorithm , or even its limited memory versions , does nt scale well enough to train very large - scale models involving millions of variables and training instances , which are commonly encountered , for example , in natural language processing .
the main computational burden of newton - type algorithm is the storage of a large matrix at per - iteration , which may be out of the memory capability for a pc .
it should be develop a matrix - free algorithm to deal with large - scale problems but also belongs to the quasi - newton framework .
for this purpose , it would like to furthermore simplify the approximation hessian @xmath65 as a diagonal matrix with positive components , i.e. , @xmath66 with an identity matrix @xmath67 and @xmath68 .
then , the quasi - newton condition changes to the form @xmath69 multiplying both sides by @xmath70 , gives @xmath71 similarly , multiplying both sides by @xmath72 , yields @xmath73 observing both formulae , it indicates that if @xmath74 , the matrix @xmath75 is positive definite , which ensures that the search direction @xmath76 is descent at current point .
the formulae ( [ alpha1 ] ) and ( [ alpha2 ] ) were firstly developed by barzilai and borwein @xcite for the quadratic case of @xmath31 .
this method essentially consists the steepest descent method , and adopts the choice of ( [ alpha1 ] ) or ( [ alpha1 ] ) as the stepsize along a negative gradient direction .
barzilai and borwein @xcite showed that the corresponding iterative algorithm is r - superlinearly convergent for the quadratic case .
raydan @xcite presented a globalization strategy based on nonmonotone line search @xcite for the general non - quadratic case .
other developments in barzilai and borwein gradient algorithm can be found in @xcite . due to its simplicity and numerical efficiency
, the barzilai - borwein gradient method is very effective to deal with large - scale smooth unconstrained minimization problems .
however , the application of the barilai - borwein gradient algorithm to @xmath0-regularized nonsmooth optimization is problematic since the regularization is non - differentiable . in this subsection , we construct an iterative algorithm to solve the @xmath0-regularized structured nonconvex optimization problem .
the algorithm can be described as the iterative form @xmath77 where @xmath78 is a stepsize , and @xmath52 is a search direction defined by minimizing a quadratic approximated model of @xmath41 .
now , we turn to our attention to consider the original problem with @xmath0-regularizer . since @xmath0-term is not differentiable , hence , at current @xmath22 , objective function @xmath41 is approximated by the quadratic approximation @xmath79 , @xmath80\triangleq q_k(d),\end{aligned}\ ] ] where @xmath81 is a small positive number .
the term in @xmath82 $ ] can be considered as an approximate taylor expansion of @xmath83 with a small @xmath81 , and the case @xmath84 reduces the equivalent form @xmath83 .
minimizing ( [ direcmodel ] ) yields @xmath85 where @xmath86 , @xmath87 , and @xmath88 denote the @xmath89-th component of @xmath22 , @xmath90 , and @xmath91 respectively .
the favorable structure of ( [ subd ] ) admits the explicit solution @xmath92 hence , the search direction at current point is @xmath93.\ ] ] where @xmath94 and @xmath95 are interpreted as componentwise and the convention @xmath96 is followed .
when @xmath97 , ( [ direction ] ) reduces to @xmath98 , i.e. , the traditional barizilai - borwein gradient algorithm in smooth optimization .
the key motivation for this formulation is that the optimization problem in eq .
( [ subd ] ) can be easily solved by exploiting the structure of the @xmath0-norm .
[ ineq ] for any real vectors @xmath99 and @xmath100 , the following function @xmath101 is non - decreasing @xmath102 note that @xmath103 hence , it reduces to prove that @xmath104 is non - decreasing for each @xmath89 .
+ ( a ) . when @xmath105 and @xmath106 .
it is clear that @xmath107 .
+ ( b ) . when @xmath105 and @xmath108 , we have @xmath109 ( c ) .
when @xmath110 and @xmath106 , we have @xmath111 ( d ) .
when @xmath110 and @xmath106 , we have @xmath112 .
+ it is not difficult to see that @xmath104 is non - decreasing at each case . hence , @xmath101 is non - decreasing .
the following lemma shows that the direction defined by ( [ direction ] ) is descent if @xmath113 .
[ descent ] suppose that @xmath68 and @xmath52 is determined by ( [ direction ] )
. then @xmath114+o(\theta ) \quad \theta\in(0,h],\ ] ] and @xmath115 by the differentiability of @xmath31 and the convexity of @xmath116 , we have that for any @xmath117 $ ] ( @xmath118 $ ] ) , @xmath119,\end{aligned}\ ] ] which is exactly ( [ des1 ] ) . noting that @xmath52 is the minimizer of ( [ direcmodel ] ) and @xmath117 $ ] , from ( [ direcmodel ] ) and the convexity of @xmath116 , we have @xmath120 hence
, @xmath121 the last three terms of the left side in ( [ sanbuineq ] ) can be re - organized as @xmath122\big\}\nonumber\\ = & \frac{\mu}{h}\big\{\|x_k+hd_k\|_1-\|x_k\|_1-\theta\big[h\cdot \frac{\|x_k+h^2d_k\|_1 -\|x_k\|_1}{h^2}\big]\big\}\nonumber\\ \geq & \frac{\mu}{h}\big\{\|x_k+hd_k\|_1-\|x_k\|_1-\theta\big[h\cdot \frac{\|x_k+hd_k\|_1 -\|x_k\|_1}{h}\big]\big\}\nonumber\\ = & \frac{\mu}{h}(1-\theta)\{\|x_k+hd_k\|_1-\|x_k\|_1 \}\label{imporineq},\end{aligned}\ ] ] where the inequality is from lemma [ ineq ] . combining ( [ sanbuineq ] ) with ( [ imporineq ] ) , it produces @xmath123 dividing both sides of ( [ y416 ] ) by @xmath124 and noting @xmath117 $ ] , we get the desirable result ( [ des2 ] ) .
when the search direction is determined , a suitable stepsize along this direction should be found to determine the next iterative point . in this paper , unlike the traditional armijo line search or the wolfe - powell line search , we pay particular attention to a nonmonotone line search strategy .
the traditional armijo line search requires the function value to decrease monotonically at each iteration . as a result
, it may cause the sequence of iterations following the bottom of a curved narrow valley , which commonly occurs in difficult nonlinear problems . to overcome this difficultly , a credible alternative is to allow an occasional increase in the objective function at each iteration . to easy comprehension of the proposed algorithm , we briefly recall the earliest nonmonotone line search technique by grippo , lampariello , and lucidi @xcite .
let @xmath125 , @xmath126 and @xmath127 be a positive integer .
the nonmonotone line search is to choose the smallest nonnegative integer @xmath128 such as the stepsize @xmath129 satisfing @xmath130 where @xmath131 if @xmath132 , the above nonmonotone line search reduces to the standard armijo line search . for the @xmath0-regularized nonsmooth problem ( [ probtype ] ) , based on lemma [ descent ] ,
the inequality ( [ gll1 ] ) should be modified as @xmath133 where @xmath134 from ( [ des2 ] ) , it clear that @xmath135 whenever @xmath113 .
hence , this shows that @xmath78 given by ( [ gll2 ] ) is well - defined . in light of all derivations above
, we now describe the nonmonotone barzilai - borwein gradient algorithm ( abbreviated as nbbl1 ) as follows . ' '' '' 1.5 mm [ alg1 ] * ( nbbl1 ) * 1.5 mm ' '' '' choose @xmath136 and constants @xmath4 .
constants @xmath137 , @xmath126 , @xmath138 , @xmath139 $ ] and positive integer @xmath127 .
set @xmath140 .
+ * step 1 . *
stop if @xmath141 .
otherwise , continue .
+ * step 2 .
* compute @xmath52 via ( [ direction ] ) . + * step 3 . *
compute @xmath78 via ( [ gll2 ] ) . + * step 4 .
* let @xmath142 .
+ * step 5 . *
let @xmath143 .
go to step 1 .
-3 mm 1 mm ' '' '' 1 mm [ remark1 ] we have shown that if @xmath68 , then the generated direction is descent .
however , in this case , the condition @xmath68 may fail to be fulfilled and the hereditary descent property is not guaranteed any more . to cope with this defect
, we should keep the sequence @xmath144 uniformly bounded ; that is , for sufficiently small @xmath145 and sufficiently large @xmath146 , the @xmath147 is forced as @xmath148 this approach ensures that @xmath147 is bounded from zero and subsequently ensures that @xmath52 is descent at per - iteration .
[ remark2 ] from lemma [ descent ] , it is clear that there exists a constant @xmath117 $ ] such that @xmath149 is a descent point in sense of ( [ des1 ] ) .
hence , in practical computation , it is suggested to choose the initial stepsize as @xmath150 .
this section is devoted to presenting some favorable properties of the generated direction and establishing the global convergence of algorithm [ alg1 ] subsequently .
our convergence result utilizes the following assumptions .
[ assu1 ] the level set @xmath151 is bounded .
[ station ] suppose that @xmath68 and @xmath52 is defined by ( [ direction ] ) with @xmath139 $ ] .
then @xmath22 is a stationary point of problem ( [ probtype ] ) if and only if @xmath152 .
if @xmath113 , then lemma [ descent ] shows that @xmath52 is descent direction at @xmath22 , which implies that @xmath22 is not a stationary point of @xmath41 . on the other hand , if @xmath152 is the solution of ( [ subd ] ) , for any @xmath153 with @xmath154 we have @xmath155 since @xmath156 , this together with ( [ yun17 ] ) yields @xmath157-\big[\frac{\mu}{h}\|x_k+\alpha hd\|_1-\frac{\mu}{h}\|x_k\|_1\big]}{\alpha}\big).\\ \geq & \lim_{\alpha \downarrow 0 } \frac{-\frac{\lambda_k\alpha^2}{2 } \|d\|_2 ^ 2+o(\alpha)}{\alpha}\\ = & 0,\end{aligned}\ ] ] where the second inequality is from lemma [ ineq ] . hence , @xmath22 is a stationary point of @xmath41 .
the proof of the following lemma is similar with the theorem in @xcite .
[ funclem ] let @xmath158 be an integer such that @xmath159 then the sequence @xmath160 is nonincreasing and the search direction @xmath161 satisfies @xmath162 from the definition of @xmath163 , we have @xmath164 .
hence @xmath165 moreover , by ( [ gll2 ] ) , we have for all @xmath166 , @xmath167 by assumption [ assu1 ] , the sequence @xmath160 admits a limit for @xmath168 .
hence , it follows that @xmath169 on the other hand , from the definition of @xmath170 in ( [ deltak ] ) and the inequality ( [ des2 ] ) , it is not difficult to deduce that @xmath171 combining ( [ alphadelk ] ) , yields @xmath172 which shows the desirable result ( [ limdkzero ] ) . [ theorem ] let the sequence @xmath173 and @xmath174 generated by algorithm [ alg1 ] . then , there exists a subsequence @xmath175 such that @xmath176 from @xcite , it is clear that ( [ limdkzero ] ) also implies @xmath177 now , let @xmath178 be a limit point of @xmath173 , and @xmath179 be a subsequence of @xmath173 converging to @xmath178 . then by ( [ dklim ] ) either @xmath180 , which implies @xmath181 , or there exists a subsequence @xmath182 ( @xmath183 ) such that @xmath184 in this case , we assume that there exists a constant @xmath185 such that @xmath186 since @xmath78 is the first value for satisfying ( [ gll2 ] ) , it follows from step 3 in algorithm [ alg1 ] that there exists an index @xmath187 such that , for all @xmath188 and @xmath189 , @xmath190 since @xmath31 is continuous differentiable , by the mean - value theorem on @xmath31 , we can find there exists a constant @xmath191 , such that @xmath192 by combining with ( [ great ] ) , we have @xmath193 since @xmath150 and @xmath194 in ( [ dkalph ] ) , we have @xmath195 as @xmath196 .
it follows from lemma [ ineq ] that @xmath197 subtracting both sides of ( [ datlakk2 ] ) by @xmath170 and noting the definition of @xmath170 , it is clear that @xmath198\nonumber\\ > & -(1-\delta)\delta_k\nonumber\\ \geq & ( 1-\delta)\frac{\lambda_{(\min)}}{2}\|d_k\|_2 ^ 2\label{datlakk3}.\end{aligned}\ ] ] taking the limit as @xmath189 , @xmath196 in the both sides of ( [ datlakk3 ] ) and using the smoothness of @xmath31 , we obtain @xmath199 which implies @xmath200 as @xmath189 , @xmath196 .
this yields a contradiction because ( [ neq ] ) indicates that @xmath201 is bounded .
in this section , we show that our algorithm can be readily extended to solve @xmath14-norm and matrix trace norm minimization problems in machine learning ; thus , broaden the applicable range of our approach significantly .
firstly , we consider the @xmath14-regularization problem @xmath202 it is not difficult to deduce that , the search direction @xmath52 is determined by minimizing @xmath203 from @xcite , the explicit solution is @xmath204 i.e. , @xmath205.\ ] ] now , we consider the matrix trace norm minimization problem @xmath206 where the functional @xmath207 is the trace norm of matrix @xmath208 , which is defined as the sum of its singular values .
that is , assume that @xmath208 has @xmath209 positive singular values of @xmath210 , then @xmath211 .
the matrix trace norm is alternatively known as the schatten @xmath0-norm , ky fan norm , and nuclear norm @xcite .
such problem has been received much attention because it is closely related to the affine rank minimization problem , which has appeared in many control applications including controller design , realization theory and model reduction . as it has been done in the previous sections
, we can readily reformulate ( [ direcmodel ] ) as the following quadratic model to determine the search direction , @xmath212 to get the exact solution of ( [ dkmatrix ] ) , we now consider the singular value decomposition ( svd ) of a matrix @xmath213 with rank @xmath209 , @xmath214 where @xmath215 and @xmath216 are @xmath217 and @xmath218 matrices respectively with orthonormal columns , and the singular value @xmath219 is positive .
for each @xmath220 , we let @xmath221_+),\ ] ] where @xmath82_+=\max\{0,\cdot\}$ ] .
it is shown that @xmath222 obeys the following nuclear norm minimization problem @xcite , i.e. , @xmath223 comparing ( [ dkmatrix ] ) to ( [ nclear ] ) , we deduce that @xmath224_+\big),\ ] ] or , equivalently , @xmath225.\ ] ] subsequently , it is easily to derive the nonmonotone barzilai and borwein gradient algorithmic framework for solving @xmath14-norm and matrix trace norm regularization problems .
in this section , we present numerical results to illustrate the feasibility and efficiency of nbbl1 .
we partition our experiments into three classes based on different types of @xmath31 . in the first class
, we perform our algorithm to solve @xmath0-regularized nonconvex problem . in the second class ,
we test our algorithm to solve @xmath0-regularized least squares which mainly appear in compressive sensing . in the third class ,
we compare some state - of - the - art algorithms in compressive sensing to show the efficiency of our algorithm .
all experiments are performed under windows xp and matlab 7.8 ( 2009a ) running on a lenovo laptop with an intel atom cpu at 1.6 ghz and 1 gb of memory .
our first test is performed on a set of the nonconvex unconstrained problems from the cuter @xcite library .
the second - order derivatives of all the selected problems are available .
since we are interested in large problems , we only consider the problems with size at least @xmath226 . for these problems
, we use the dimensions that is admissible of the
double large " installation of cuter .
the algorithm stops if the norm of the search direction is small enough ; that is , @xmath227 the iterative process is also stopped if the number of iterations exceeds @xmath228 without achieving convergence . in this experiment ,
we take @xmath229 , @xmath84 , @xmath230 , @xmath231 . in the line search , we choose @xmath232 , @xmath233 , @xmath234 and @xmath235 .
we test nbbl1 with different parameter values @xmath236 .
the numerical results are presented in table [ results ] , which contains the name of the problem ( problem ) , the dimensions of the problem ( dim ) , the number of iterations ( iter ) , the number of function evaluations ( nf ) , the cpu time required in seconds ( time ) , the final objective function values ( fun ) , the norm of the final gradient of @xmath31 ( normg ) , and the norm of final direction ( normd ) . from table [ results ] , we see that nbbl1 works successfully for all the test problems in each case .
particularly , nbbl1 always produces great accuracy solutions within little consuming time .
the proposed algorithm requires large number of iterations for some special problems , such as problems fletcher , noncvxu2 , broydn7d with parameter @xmath97 , problems fletcher and broydn7d with @xmath237 , problem fletcher with @xmath238 , and vardim with @xmath239 .
however , if lower precision is permitted , the number can be decreased dramatically .
the first part of table [ results ] presents the numerical results of nbbl1 for solving a smooth nonconvex minimization problem without any regularization . from the last second column in this part , we observe that the norm of the final gradient is sufficiently small
. the important observation verifies that the proposed algorithm is very efficient to solve unconstrained smooth minimization problems .
it is not a pleasant supervise , because our algorithm reduces to the well - known nonmonotone barzailai - borwein gradient of raydan @xcite in this case .
let @xmath178 be a sparse or a nearly sparse original signal , @xmath240 ( @xmath241 ) be a linear operator , @xmath242 be a zero - mean gaussian white noise , and @xmath8 be an observation which satisfies the relationship @xmath243 recent compressive sensing results show that , under some technical conditions , the desirable signal can be reconstructed almost exactly by solving the @xmath0-regularized least square ( [ onenorm ] ) .
in this subsection , we perform two classes of numerical experiments for solving ( [ onenorm ] ) by using the gaussian matrices as the encoder . in the first class
, we show that our algorithm performs well to decode a sparse signal , while in the second class we do a series of experiments with different @xmath81 to choose the best one .
we measure the quality of restoration @xmath244 by means of the relative error to the original signal @xmath178 ; that is @xmath245 in the first test , we use a random matrix @xmath18 with independent identically distributions gaussian entries .
the @xmath246 is the additive gaussian noise of zero mean and standard deviation @xmath247 . due to the storage limitations of pc
, we test a small size signal with @xmath248 , @xmath249 .
the original contains randomly @xmath250 non - zero elements . besides
, we also choose the noise level @xmath251 .
the proposed algorithm starts at a zero point and terminates when the relative change of two successive points are sufficient small , i.e. , @xmath252 in this experiment , we take @xmath253 , @xmath254 , @xmath255 , @xmath256 . in the line search , we choose @xmath257 , @xmath233 , @xmath234 and @xmath235 .
the original signal , the limited measurement , and the reconstructed signal are given in figure [ fig1 ] .
and @xmath258 positive non - zero elements ; middle : the noisy measurement with length @xmath259 ; right : recovered signal by nbbl1 ( red circle ) versus original signal ( blue peaks).,title="fig : " ] and @xmath258 positive non - zero elements ; middle : the noisy measurement with length @xmath259 ; right : recovered signal by nbbl1 ( red circle ) versus original signal ( blue peaks).,title="fig : " ] and @xmath258 positive non - zero elements ; middle : the noisy measurement with length @xmath259 ; right : recovered signal by nbbl1 ( red circle ) versus original signal ( blue peaks).,title="fig : " ] comparing the left plot to the right one in figure [ fig1 ] , we clearly see that the original sparse signal is restored almost exactly .
we see that all the blue peaks are circled by the red circles , which illustrates that the original signal has been found almost exactly .
all together , this simple experiment shows that our algorithm performs quite well , and provides an efficient approach to recover large sparse non - negative signal .
we have clearly known that the last term in the approximate quadratic model ( [ direcmodel ] ) is equivalent to @xmath83 exactly when @xmath84 .
next , we provide evidence to show that other values can be potentially and dramatically better than @xmath84 .
we conduct a series of experiments and compare the performance at each case . in our experiments
, we set all the parameters values as the previous test except for @xmath260 .
we present , in figure [ figh ] , the impact of the parameter @xmath81 values on the total number of iterations , the computing time , and the quality of the recovered signal . in each plot
, the level axis denotes the values of @xmath81 from @xmath261 to @xmath262 in a log scale . in log scale.,title="fig : " ] in log scale.,title="fig : " ] in log scale.,title="fig : " ] in figure [ figh ] , the number of iterations , the computing time and the quality of restorations are greatly influenced by the @xmath81 values .
generally , as @xmath81 increases , nbbl1 always has good performance .
the right plot clearly demonstrates that the relative error decreases dramatically at the very beginning and then becomes slightly after @xmath263 .
however , the quality of restoration can not be improved any more after @xmath264 . on the other hand ,
the left and the middle plots show that the number of iterations and the computing time slightly increase after @xmath265 .
taking three plots together , these plots verify that the performance of nbbl1 is sensitive to the @xmath81 values , and the value @xmath266 $ ] may be the better choice .
the third class of the experiment is to test against several state - of - the - art algorithms which are specifically designed in recent years to solve @xmath0-regularized problems in compressive sensing or linear inverse problems .
it is difficult to compare each algorithm in a very fair way , because each algorithm is compiled with different parameter settings , such as the termination criterions , the staring points , or the continuation techniques .
hence , as usual , in our performance comparisons , we run each code from the same initial point , use all the default parameter values , and only observe the convergence behavior of each algorithm to attain a similar accuracy solution .
nesta uses nesterov s smoothing technique @xcite and gradient method @xcite to solve basis pursuit denoising problem .
the current version is capable of solving @xmath0-norm regularization problems with different types including ( [ onenorm ] ) . in this experiment , we test nesta with continuation ( named nesta - ct ) for comparison , where this algorithm solves a sequence of problems ( [ onenorm ] ) by using a decreasing sequence of values of @xmath267
. additionally , nesta - ct uses the intermediated solution as a warm start for the next problem . in running nesta ,
all the parameters are taken as default except tolvar is set to be @xmath268 to obtain similar quality solutions with others .
gpsr - bb ( gradient projections for sparse reconstruction ) @xcite reformulates the original problem ( [ onenorm ] ) as a box - constrained quadric programming problem ( [ subbound ] ) by splitting @xmath32 .
figueiredo , nowak and wright use a gradient projection method with barziali - borwein steplength @xcite for its solution .
moreover , the nonmonotone line search @xcite is also used to improve its performance . for the comparison with gpsr - bb
, we use its continuation variant and set all parameters as default .
the well - known cgd uses gradient algorithm to solve ( [ onesub ] ) in order to obtain the search direction @xmath269 in @xmath270 , where @xmath271 is a nonempty subset of @xmath272 , and choose the index subset @xmath271 a gauss - southwell rule .
the iterative process @xmath142 continues until some termination critera are met , where @xmath273 with @xmath274 and the stepsize @xmath78 by using a armijo rule . in running cgd
, we use the code cgd in its matlab package , and set all the parameter as default except for init=2 to start the iterative process at @xmath275 .
twist is a two - step ist algorithm for solving a class of linear inverse problems .
specifically , twist is designed to solve @xmath276 where @xmath18 is a linear operator , and @xmath277 is a general regularizer , which can be either the @xmath0-norm or the tv .
the iteration framework of twist is @xmath278 where @xmath279 are parameters , @xmath280 and @xmath281 we use the default parameters in twist and terminate the iteration process when the relative variation of function value falls below @xmath282 .
fpc is the fixed - point continuation algorithm to solve the general @xmath0-regularized minimization problem ( [ probtype ] ) , where @xmath31 is a continuous differentiable convex function . at current @xmath22 and any scalar @xmath220 ,
the next iteration is produced by the so - called fixed point iteration @xmath283 where sgn " is a componentwise sign function . in order to obtain a good practical performance , a continuation approach is also augmented in fpc .
moreover , the fpc is further modified by using barzilai - borwein stepsize ( code fpc - bb in matlab package fpc_v2 ) .
the continuation and barzilai - borwein stepsize techniques make fpc - bb faster than fpc . in running of fpc - bb ,
we use all the default parameter values except we set xtol = 1e-5 to stop the algorithm when the relative change between successive points is below xtol . in this test
, @xmath18 is a partial discrete cosine coefficients matrix ( dct ) , whose @xmath17 rows are chosen randomly from the @xmath284 dct matrix .
such encoding matrix @xmath18 does not require storage and enables fast matrix - vector multiplications involving @xmath18 and @xmath285 .
therefore , it is able to be used to test much larger size problems than using gaussian matrices . in nbbl1 , we take @xmath286 , @xmath265 , @xmath255 , @xmath256 . in the line search , we choose @xmath287 , @xmath233 , @xmath288 and @xmath235 . in this comparison , we let @xmath289 , @xmath290 .
the original signal @xmath178 contains @xmath291 number of nonzero components , where floor is a matlab command used to round an element to the nearest integers towards minus infinity .
moreover , the observation @xmath292 is contaminated by gaussian noise with level @xmath293 .
the goal is to use each algorithm to reconstruct @xmath178 from the observation @xmath292 by solving ( [ onenorm ] ) with @xmath294 .
all the tested algorithms start at @xmath275 and terminate with different stopping criterions to produce similar quality resolutions . to specifically illustrate the performance of each algorithm , we draw four figures to show their convergence behavior from the point of objective function values and relative error as the iteration numbers and computing time increase , which given in figure [ figcom ] . + from the top plots in figure [ figcom ] , nbbl1 usually decreases relative errors faster than nesta - ct , cgd and gpsr - bb throughout the entire iteration process , and meanwhile requires less number of iterations .
the top right plot shows that twist needs less steps than nbbl1 to obtain similar level of relative error .
however , twist is much slower because it has to solve a de - noising subproblem ( [ denois ] ) at each iteration .
unfortunately , nbbl1 needs further improvement to challenge the well - known code fpc - bb .
we now turn our attention to observe the function values behavior of each algorithm .
similarly , nbbl1 is superior to nest - ct , cgd , gpsr - bb and twist from the computing time points of view .
fpc - bb reaches the lowest function values at the very beginning , and then starts to increase it to meet nearly equal final values at the end . in this test
, cgd appears to be much slower than the others , because it is sensitive to the choice of starting points .
if cgd starts at @xmath295 with all the other settings unchanged , its performance should be significantly improved @xcite . taking everything together , from the limited numerical experiment
, we conclude that nbbl1 provides an efficient approach for solving @xmath0-regularized nonsmooth problem and is competitive with or performs better than nesta - ct , gpsr - bb , cgd , twist and fpc - bb .
in this paper , we proposed , analyzed , and tested a new practical algorithm to solve the separable nonsmooth minimization problem consisting of a @xmath0-norm regularized term and a continuously differentiable term .
the type of the problem mainly appears in signal / image processing , compressive sensing , machine learning , and linear inverse problems .
however , the problem is challenging due to the non - smoothness of the regularization term .
our approach minimizes an approximal local quadratic model to determine a search direction at each iteration .
the search direction reduces to the classic barzilai - borwein gradient method in the case of @xmath97 .
we show that the objective function is descent along this direction providing that the initial stepsize is less than @xmath81 .
we also establish the algorithm s global convergence theorem by incorporating a nonmonotone line search technique and assuming that @xmath31 is bounded below .
extensive experimental results show that the proposed algorithm is an effective toll to solve @xmath0-regularized nonconvex problems from cuter library .
moreover , we also run our algorithm to recover a large sparse signal from its noisy measurement , and numerical comparisons illustrate that our algorithm outperforms or is competitive with several state - of - the - art solvers which specifically designed to solve @xmath0-regularized compressive sensing problems . unlike all the existing algorithms in this literature , our approach uses an linear model to approximate @xmath83 for computing the search direction with a small scalar @xmath81 ; that is @xmath296 although the equations may hold exactly in the case of @xmath84 , a series of numerical experiments show that @xmath266 $ ] may produce better performance with suitable experiment settings .
this approach is distinctive and novel ; therefore , it is one of the important contributions of this paper . as we all know , the nonmonotone barzilai - borwein gradient algorithm of raydan @xcite is very effective for smooth unconstrained minimization , and its remarkable effectiveness in signal reconstruction problems involving @xmath0-regularized problems has not been clearly explored . hence , our approach can be considered as a modification or extension , to broaden the university of @xcite .
moreover , the numerical experiments illustrate that our approach performs comparable to or even better than several state - of - the - art algorithms . surely , this is the numerical contribution of our paper .
although the proposed algorithm needs further improvement to challenge the well - known code fpc - bb , the enhancement of it to deal with non - convex problems is noticeable .
our algorithm is readily to solve the @xmath0-regularized logistic regression , the @xmath14-norm and matrix trace norm minimization problems in machine learning .
however , we do not test them in this paper .
this should be interesting for further investigations .
e. cands , j. romberg , and t. tao , _ robust uncertainty principles : exact signal reconstruction from highly incomplete frequence information _ ,
ieee transactions on information theory , 52 ( 2006 ) , 489 - 509 .
m. figueiredo , r.d .
nowak , and s.j .
wright , _ gradient projection for sparse reconstruction : application to compressed sensing and other inverse problems _ , ieee journal of selected topics in signal processing , ieee press , piscataway , nj , 2007 , 586 - 597 .
s. kim , k. koh , m. lustig , s. boyd , and d. gorinevsky , _ an interior - point method for large - scale @xmath0-regularized least squares _
, ieee journal of selected topics in signal processing , 1 ( 2007 ) , 606 - 617 .
wright , r.d .
nowak and m.a.t .
figueiredo , _ sparse reconstruction by separable approximation _ , in roceedings of the international conference on acoustics , speech , and signal processing , 2008 , 3373 - 3376 .
j. yu , s.v.n .
vishwanathan , s. gnter , and n.n .
schraudolph , _ a quasi - newton approach to nonsmooth convex optimization problems in machine learning _
, journal of machine learning research , 11 ( 2010 ) , 1145 - 1200 .
yuan , k.w .
chang , c.j .
hsieh , and c.j .
lin , _ a comparison of optimization methods and software for large - scale @xmath0-regularized linear classification _ , journal of machine learning research , 11 ( 2010 ) , 3183 - 3234 . | this paper is devoted to minimizing the sum of a smooth function and a nonsmooth @xmath0-regularized term .
this problem as a special cases includes the @xmath0-regularized convex minimization problem in signal processing , compressive sensing , machine learning , data mining , etc .
however , the non - differentiability of the @xmath0-norm causes more challenging especially in large problems encountered in many practical applications .
this paper proposes , analyzes , and tests a barzilai - borwein gradient algorithm . at each iteration ,
the generated search direction enjoys descent property and can be easily derived by minimizing a local approximal quadratic model and simultaneously taking the favorable structure of the @xmath0-norm . moreover , a nonmonotone line search technique is incorporated to find a suitable stepsize along this direction .
the algorithm is easily performed , where the values of the objective function and the gradient of the smooth term are required at per - iteration . under some conditions ,
the proposed algorithm is shown to be globally convergent .
the limited experiments by using some nonconvex unconstrained problems from cuter library with additive @xmath0-regularization illustrate that the proposed algorithm performs quite well .
extensive experiments for @xmath0-regularized least squares problems in compressive sensing verify that our algorithm compares favorably with several state - of - the - art algorithms which are specifically designed in recent years .
nonsmooth optimization , nonconvex optimization , barzilai - borwein gradient algorithm , nonmonotone line search , @xmath0 regularization , compressive sensing 65l09 , 65k05 , 90c30 , 90c25 |
massive lepton pair production and inclusive direct photon production in hadronic collisions have historically provided an important tool to gain access to parton distributions in hadrons . moreover , direct photons , i.e. photons not from hadronic decay , can be also a powerful probe of the initial state of matter created in heavy ion collisions , since they interact with the medium only electromagnetically and therefore provide a baseline for the interpretation of jet - quenching models . in the parton model ,
the feynman diagrams for partonic subprocesses that are present in drell - yan ( dy ) lepton pair production and in inclusive direct photon production are different , and the connection between both production mechanisms within a unique approximation scheme is not obvious .
since in the target rest frame the dy process looks like bremsstrahlung of a virtual photon decaying into a lepton pair , we will show that the color dipole formalism defined in this frame is well suited to describe both production processes in a unified framework free of parameters . as an illustrative example
, we study dilepton spectra in @xmath0-gev @xmath1 collisions from the e866 experiment @xcite , inclusive direct - photon spectra in @xmath1 at @xmath2 gev from the phenix experiment @xcite , and @xmath3 collisions at @xmath4 tev from the cdf experiment @xcite .
there have been already some attempts to describe the dy transverse momentum distribution in the color dipole approach @xcite , but unfortunately the experimental data that was used for comparison is not fully kinematically in the range of validity of the model .
here we confront the dipole approach with experimental data that is in a region where the model is supposed to be at work .
furthermore , we will also study the inclusion of both non - perturbative primordial transverse momenta and dglap evolution . despite many years of intense studies ,
a satisfactory description of all existing inclusive direct photon production data in hadronic collision , based on perturbative qcd ( pqcd ) calculations , seems to be evasive @xcite .
this letter is an alternative attempt .
we shows that the color dipole approach can successfully describe inclusive photon production in hadron - hadron collisions .
the color dipole formalism , developed in @xcite for the case of the total and diffractive cross sections , can be also applied to radiation @xcite .
although in the process of electromagnetic bremsstrahlung by a quark no dipole participates , the cross section can be expressed via the more elementary cross section @xmath5 of interaction of a @xmath6 dipole . nevertheless , this is a fake , or effective dipole .
similar to a real dipole , where color screening is provided by interactions with either the quark or the antiquark , in the case of radiation the two amplitudes for radiation prior or after the interaction screen each other , leading to cancellation of the infra - red divergences @xcite .
the transverse momentum @xmath7 distribution of photon bremsstrahlung in quark - nucleon interactions , integrated over the final quark transverse momentum , was derived in @xcite in terms of the dipole formalism , @xmath8 where @xmath9 and @xmath10 and @xmath11 are the quark - photon transverse separations in the two radiation amplitudes contributing to the cross section , eq .
( [ m1 ] ) , which correspondingly contains double - fourier transformations .
the parameter @xmath12 is the relative fraction of the quark momentum carried by the photon , and is the same in both amplitudes , since the interaction does not change the sharing of longitudinal momentum .
the transverse displacement between the initial and final quarks is @xmath13 and @xmath14 respectively .
since the amplitude of quark interaction has a phase factor @xmath15 , where @xmath16 is the impact parameter of collision , the transverse displacement between the initial and final quarks leads to the color screening factor @xmath17 . in eq .
( [ m1 ] ) @xmath18 stands for transverse and @xmath19 for longitudinal photons .
the energy dependence of the dipole cross section , which comes via the variable @xmath20 , where @xmath21 is the projectile four - momentum and @xmath22 is the four - momentum of the dilepton , is generated by additional radiation of gluons which can be resummed in the leading @xmath23 approximation . in order to obtain the hadron cross section from the elementary partonic cross section
( [ m1 ] ) , one should sum up the contributions from quarks and antiquarks weighted with the corresponding parton distribution functions ( pdfs ) @xcite , @xmath41 @xmath42 the pdfs of the projectile enter in a combination which can be written in terms of proton structure function @xmath43 .
notice that with our definitions the fractional quark charge @xmath44 is not included in the lc wave function of eq .
( [ wave ] ) , and that the factor @xmath45 in eq .
( [ con0 ] ) accounts for the decay of the photon into the lepton pair .
we use the standard notation for the kinematical variables , @xmath46 denotes the momentum fraction that the photon carries away from the projectile hadron in the target frame , we define @xmath47 , @xmath48 is the feynman variable and @xmath49 , where @xmath50 and @xmath51 denote the longitudinal and transverse momentum components of the photon in the hadron - hadron center of mass frame , @xmath37 is the center of mass energy squared of the colliding protons and @xmath33 is the dilepton mass .
we also need to identify the scale @xmath52 entering in the proton structure function in eq .
( [ con ] ) , and relate the energy scale @xmath53 of the dipole cross section entered in eq .
( [ di ] ) to measurable variables . from our previous definition , and following previous works @xcite we have that @xmath54 . at zero transverse momentum , the dominant term in the lc wavefunction eq .
( [ wave ] ) is the one that contains the modified bessel function @xmath55 .
this function decay exponentially at large values of the argument , so that the mean distances which numerically contribute are of order @xmath56 . on the other hand ,
the minimal value of @xmath12 is @xmath57 , and therefore the virtuality @xmath58 which enters into the problem at zero transverse momentum is @xmath59 .
thus the hard scale at which the projectile parton distribution is probed turns out to be @xmath60 .
notice that in the previous studies , @xmath61 @xcite and @xmath62 @xcite were used for the scale @xmath63 .
nevertheless , these different choices for @xmath63 bring less than about a @xmath64 effect at small @xmath65 values .
the dipole cross section is theoretically unknown , although several parametrizations have been proposed in the literature . for our purposes , here we consider two parametrizations , the saturation model of golec - biernat and wsthoff ( gbw ) @xcite and the modified gbw coupled to dglap evolution ( gbw - dglap ) @xcite . in the gbw model
@xcite the dipole cross section is parametrized as , @xmath66 where the parameters , fitted to dis hera data at small @xmath53 , are given by @xmath67 mb , @xmath68 , where @xmath69 .
this parametrization gives a quite good description of dis data at @xmath70 .
a salient feature of the model is that for decreasing @xmath53 , the dipole cross section saturates for smaller dipole sizes , and that at small @xmath71 , as perturbative qcd implies , @xmath72 vanishes .
this is the so - called color transparency phenomenon @xcite .
one of the obvious shortcoming of the gbw model is that it does not match with qcd evolution ( dglap ) at large values of @xmath63 .
this failure can be clearly seen in the energy dependence of @xmath73 for @xmath74 , where the the model predictions are below the data @xcite . a modification of the gwb dipole parametrization model , eq .
( [ gbw ] ) , was proposed in ref .
@xcite @xmath75 where the scale @xmath76 is related to the dipole size by @xmath77 here the gluon density @xmath78 is evolved to the scale @xmath76 with the leading order ( lo ) dglap equation @xcite .
moreover , the quark contribution to the gluon density is neglected in the small @xmath53 limit , and therefore @xmath79 where @xmath80 and @xmath81 denote the qcd splitting function and coupling , respectively .
the initial gluon density is taken at the scale @xmath82 in the form @xmath83 where the parameters @xmath84 , @xmath85 , @xmath86 and @xmath87 are fixed from a fit to the dis data for @xmath70 and in a range of @xmath58 between @xmath88 and @xmath89 @xmath90 @xcite .
we use the lo formula for the running of the strong coupling @xmath91 , with three flavors and for @xmath92 .
the dipole size determines the evolution scale @xmath76 through eq .
( [ scale ] ) .
the evolution of the gluon density is performed numerically for every dipole size @xmath71 during the integration of eq .
( [ m1 ] ) .
therefore , the dglap equation is now coupled to our master equations ( [ con0],[con ] ) .
it is important to stress that the gbw - dglap model preserves the successes of the gbw model at low @xmath63 and its saturation property for large dipole sizes , while incorporating the evolution of the gluon density by modifying the small-@xmath71 behaviour of the dipole size . the proton structure function in eqs .
( [ con0],[con ] ) is parametrized as @xmath93^{b(x ) } \left(1+\frac{c(x)}{q^{2}}\right),\ ] ] with @xmath94 , @xmath95 gev , and the functions @xmath96 and @xmath97 are parametrized in terms of @xmath98 parameters fitted to different experiments , and whose functional forms can be found in the appendix of ref .
this parametrization is only valid in the kinematic range of the data sets which cover correlated regions in the ranges @xmath99 and @xmath100 .
before we proceed to present the results in the color dipole approach , some words regarding the validity of this formulation are in order . although both valence and sea quarks in the projectile are taken into account through the proton structure function eqs .
( [ con0 ] , [ con ] ) , the color dipole picture accounts only for pomeron exchange from the target , while ignoring its valence content . in terms of regge phenomenology
, this means that reggeons are not taken into account , and as a consequence , the dipole approach predicts the same cross sections for both particle and antiparticle induced dy reactions .
therefore , in principle this approach is well suited for high - energy processes , i.e. small @xmath65 .
the exact range of validity of the dipole approach is of course not known a priori , but there is evidence @xcite in its favor for values of @xmath101 . in our case , however , we use a parametrization of the dipole cross section fitted to dis data for bjorken-@xmath70 and for energy scales @xmath102 . given these restrictions , at present there are not many data for dy cross section at low @xmath65 .
notice also that some data are integrated over @xmath36 and @xmath33 , and are therefore contaminated by contributions not included into the color dipole approach .
-gev @xmath1 collisions at @xmath103 and @xmath104gev .
we show the result of the gbw dipole model ( dashed line ) and the gbw - dglap model ( dotted line ) .
we also show the result when a constant primordial momentum @xmath105 is incorporated within the gbw - dglap dipole model ( solid line ) .
experimental data are from ref .
the e866 error bars are the linear sum of the statistical and systematic uncertainties .
an additional @xmath106 uncertainty in the experimental data points due to the normalization is also common to all points .
[ fig-2],width=8 ] , except for @xmath107.[fig-3],width=8 ] we compare the present approach to data for @xmath0-gev @xmath1 collisions from e886 @xcite , which are not integrated over @xmath36 and @xmath33 , and correspond to the lowest @xmath65 values , i.e. lightest @xmath33 and highest @xmath36 .
we selected a @xmath36 bin where @xmath108 , with an average value @xmath109 . within this bin we selected two bins with the lightest average values for @xmath33 , one for @xmath110 , with an average value of @xmath111 gev , and the other for @xmath112 , with an average value @xmath113 gev .
the experimental data are plotted , with errors , in figs .
[ fig-2 ] and [ fig-3 ] . in our calculations
we have taken the experimental average values for @xmath36 and @xmath33 . in figs .
[ fig-2 ] , [ fig-3 ] we show the result obtained by the dipole approach , for both the gbw and the gbw - dglap dipole models . at low transverse
momentum @xmath114 gev both model predictions are almost identical , but at higher @xmath7 the dipole parametrization improved by dglap evolution bends down towards the experimental points improving the result .
this is more obvious for higher values of @xmath33 .
notice that for the case of a smaller value of @xmath65 with a lighter @xmath33 , fig .
[ fig-3 ] , where the dipole approach is better suited , the gbw model without inclusion of the dglap evolution already provides a good description of the data .
we stress that the theoretical curves in figs .
[ fig-2 ] , [ fig-3 ] are the results of a parameter free calculation .
as we already pointed out , varying the quark mass @xmath34 leaves the numerical results almost unaffected .
notice also that in contrast to the lo parton model , no @xmath115-factor was introduced , since the dipole parametrization fitted to dis data already includes contributions from higher order perturbative corrections as well as non - perturbative effects contained in dis data .
one of the data point which surprisingly is left out from our theoretical computation curves for both values of @xmath33 , is the one at the lowest @xmath7 . in the dipole approach
the dy cross section is finite at @xmath116 due to the saturation of the dipole cross section , which is in striking contrast to the lo pqcd correction to the parton model , where one needs to resume the large logarithms @xmath117 from soft gluon radiation in order to obtain a physically sensible results at @xmath116 @xcite .
one of the possible reasons behind the lack of agreement between our result and the experimental data at @xmath118 may be due to a soft non - perturbative primordial transverse momentum distribution of the partons in the colliding protons .
such a primordial transverse momentum may have various non - perturbative origins , e. g. finite size effects of the hadron , instanton effects , pion - cloud contributions .
moreover , in the parton model it has been shown that even within the next - to - leading order ( nlo ) pqcd correction , experimental data of heavy quark pair production @xcite , direct photon production @xcite and dy lepton pair production @xcite can be only described if an average primordial momentum as large as @xmath119 gev is included ( see also ref .
such a large value for the initial transverse momentum strongly indicates its perturbative origin in the parton model , and in principle must have been already included in the pqcd correction .
therefore , in the pqcd approach , it is still an open question how to separate what is truly intrinsic and what is pqcd generated transverse momentum . however , in the dipole approach all perturbative and non - perturbative contributions , apart from the finite - size effect of hadrons , are already encoded into the cross section via fitting the dipole parameters to dis data .
therefore , we expect that in the dipole approach the primordial momentum should have a purely non - perturbative origin , and to be considerably less than in the parton model .
one may introduce an intrinsic momentum contribution in the following factorized form @xmath120 where the function @xmath121 denotes the cross section defined in eqs .
( [ con0],[con ] ) .
we assume that the initial @xmath7 distribution @xmath122 has a gaussian form , @xmath123 where @xmath124 is the square of the two - dimensional width of the @xmath7-distribution for an incoming quark , and also that @xmath124 is a constant independent of the hard scale @xmath52 , since the pqcd radiation - generating transverse momenta are already taken into account in our approach .
the differential cross section convoluted with the primordial momentum distribution in the gbw - dglap dipole model are shown in figs .
[ fig-2 ] and [ fig-3 ] with curves denoted with gbw - dglap - primordial .
a value around @xmath125 can describe the experimental points at low @xmath7 for both sets of data plotted in figs .
[ fig-2 ] and [ fig-3 ] .
this value , as we expected , is lower than the primordial momentum which has been used in the parton model .
the experimental data points for @xmath126 should be taken with some precaution , since there exists some disagreement between different experiments for dy lepton pair production at low @xmath7 . indeed ,
although the e772 and e866 measurements @xcite have good agreement among them over a wide range of values , they disagree at @xmath126 . therefore , the discrepancy between our theoretical results and experimental data at @xmath126 might be just in fact an artifact of the experiments .
next we calculate the inclusive direct photon spectra within the same framework . for direct photon
we have @xmath35 , and we assume again a quark mass @xmath40 gev . as illustrative examples
we compare our results with the phenix and cdf experiments .
notice that direct photon problem ( with @xmath35 ) , compared to the massive virtual photon case ( with @xmath33 as big as @xmath127 gev ) , is numerically more involved since the integrand in eq .
( [ m1 ] ) is divergent when @xmath128 . at rhic energy @xmath2 gev .
experimental data are from ref . @xcite . in the down panel
we use the gbw - dglap dipole model result for the theory .
the error bars are the linear sum of the statistical and systematic uncertainties .
[ fig-4],width=8 ] tev .
experimental data are for inclusive isolated photon from cdf experiment for @xmath3 collision at cdf energy and @xmath129 @xcite .
the nlo qcd curve is from the authors of reference @xcite ( given in table 3 of ref .
@xcite ) and use the cteq5 m parton distribution functions with the all scales set to the @xmath7 . in the down panel
we use the gbw - dglap dipole model result for the theory .
the error bars are the linear sum of the statistical and systematic uncertainties .
[ fig-5],width=8 ] in fig .
[ fig-4 ] we show the differential cross section obtained from the gbw and the gbw - dglap dipole models at midrapidities , for @xmath1 collisions at rich energies @xmath130 gev .
the experimental data are from the phenix measurements for inclusive direct photon production at @xmath131 @xcite .
we have also checked out that the effect of the incorporation of the same transverse primordial momentum @xmath132 which can describe the dilepton spectra at low @xmath7 , will be in this case too small to improve the results at the range of @xmath7 of the experimental data . without a physically sound guiding principle ,
however , the introduction of a higher value of intrinsic momentum is somehow unsatisfactory and will not be further discussed here .
notice also that , in contrast to the parton model , we have not included any photon fragmentation function @xcite for computing the cross section , since the dipole formulation already incorporates all perturbative ( via pomeron exchange ) and non - perturbative radiation contributions .
it has been shown that the nlo pqcd prediction @xcite are also consistent with the rhic data within the uncertainties @xcite . in fig .
[ fig-5 ] we show the dipole models predictions for inclusive prompt - photon production at midrapidities , and for cdf energies @xmath4 tev .
the experimental points are taken from cdf data for inclusive isolated - photon , averaged over @xmath129 @xcite . at lower transverse
momentum @xmath133 gev the gbw dipole model can reproduce rather fairly the experimental data , and at higher @xmath7 values dglap evolution significantly improves the results .
in the collider experiments at the tevatron , in order to reject the overwhelming background of secondary photons which come from the decays of pions , isolation cuts are imposed @xcite .
these cuts affect the direct - photon cross section , in particular by reducing the fragmentation effects .
isolation conditions are not imposed in our calculation , although the experimental data is for isolated photon .
however , it has been shown that the cross section does not vary by more than @xmath134 under cdf isolation conditions and kinematics
. therefore , the main source of uncertainty in our approach is due to the fact that the experimental points are averaged over rapidity and contaminated by reggeon contributions which are ignored in the dipole approach .
one should also notice that the parametrizations of the dipole cross section and proton structure function employed in our computation have been fitted to data at considerably lower @xmath63 values ( see previous section ) .
the nlo pqcd calculation for direct photon production at the tevatron energy was performed in ref .
new independent measurement of direct photon at the tevatron energy which is in agreement with previous cdf measurement @xcite , provided further evidence that the shape of the cross section as function of @xmath7 can not be fully described by the available nlo pqcd computation @xcite . in this letter
, we showed that both direct photon production and dy dilepton pair production processes can be described within the same color dipole approach without any free parameters .
in contrast to the parton model , in the dipole approach there is no ambiguity in defining the intrinsic transverse momentum .
such a purely non - perturbative primordial momentum improves the results in the case of dilepton pair production , but does not play a significant role for direct photon production at the given experimental range of @xmath7 .
we also showed that the color dipole formulation coupled to the dglap evolution provides a better description of data at large transverse momentum compared to the gbw dipole model .
the authors would like to thank t. isobe for providing the experimental data in ref .
this work was supported in part by fondecyt ( chile ) grants 1070517 and 1050519 and by dfg ( germany ) grant pi182/3 - 1 .
20 j. c. webb , fermilab - thesis-2002 - 56 , hep - ex/0301031 .
phenix collaboration , phys .
. lett . * 98 * , 012002 ( 2007 ) .
cdf collaboration , phys .
lett . * 73 * , 2662 ( 1994 ) ; * 74*,1891 ( 1995 ) .
m. a. betemps , m. b. gay ducati , m. v. t. machado , j. raufeisen , phys . rev . *
d67 * , 114008 ( 2003 ) .
p. aurenche , m. fontannaz , j. ph .
guillet , b. kniehl , e. pilon and m. werlen , eur .
j. * c9 * , 107 ( 1999 ) .
a. b. zamolodchikov , b. z. kopeliovich and l. i. lapidus , jetp lett .
* 33 * , 595 ( 1981 ) .
kopeliovich , proc . of the workshop hirschegg 95 : dynamical properties of hadrons in nuclear matter , hirschegg january 16 - 21 , 1995 , ed . by h. feldmeyer and w. nrenberg , darmstadt , 1995 , p. 102
( hep - ph/9609385 ) .
b. z. kopeliovich , a. schaefer and a. v. tarasov , phys . rev . *
c59 * , 1609 ( 1999 ) .
b. z. kopeliovich , j. raufeisen and a. v. tarasov , phys .
* b503 * , 91 ( 2001 ) . b. z. kopeliovich , j. raufeisen and a. v. tarasov , phys . rev . * c62 * , 035204 ( 2000 ) .
j. raufeisen , j .- c .
peng and g. c. nayak , phys . rev .
* d66 * , 034024 ( 2002 ) .
k. golec - biernat and m. wusthoff , phys . rev . * d59 * , 014017 ( 1999 ) ; * d60 * , 114023 ( 1999 ) .
j. bartels , k. golec - biernat and h. kowalski , phys .
rev . * d66 * , 014001 ( 2002 )
. m. a. j. botje , qcdnum16 : a fast qcd evolution , zeus note 97 - 006 , 1997 .
smc collaboration , phys . rev . *
d58 * , 112001 ( 1998 ) .
p. chiappetta and h. j. pirner , nucl
b291 * , 765 ( 1987 ) .
m. n. mangano , p. nason , and g. ridolfi , nucl .
phys . * b373 * , 295 ( 1992 ) .
fermilab e706 collaboration , phys .
lett . * 81 * , 2642 ( 1998 ) ; l. apanasevich _
_ , phys . rev . *
d59 * , 074007 ( 1999 ) .
d. c. hom _ et al .
* 37 * , 1374 ( 1976 ) ; d. m. kaplan _
et al . _ ,
lett.*40 * , 435 ( 1978 ) .
wang , phys . rev . *
c61 * , 064910 ( 2000 ) . l. e. gordon and w. vogelsang , phys . rev . *
d48 * , 3136 ( 1993 ) ; * d50 * , 1901 ( 1994 ) . e. l. berger and j .- w .
qiu , phys . rev .
* d44 * , 2002 ( 1991 ) .
p. aurenche _
et al . _ ,
b140 * , 87 ( 1984 ) ; nucl . phys .
* b297 * , 661 ( 1988 ) ; h. baer _ et al .
_ , phys . rev . *
d42 * 61 ( 1990 ) ; phys .
lett . * b234 * 127 ( 1990 ) .
s. catani , m. fontannaz , j. ph .
guillet and e. pilon , jhep * 0205 * , 028 ( 2002 )
. m. gluck , l. e. gordon , e. reya and w. vogelsang , phys .
73 * , 388 ( 1994 ) .
cdf collaboration , phys .
. * d70 * , 074008 ( 2004 ) . | drell - yan dilepton pair production and inclusive direct photon production can be described within a unified framework in the color dipole approach .
the inclusion of non - perturbative primordial transverse momenta and dglap evolution is studied .
we successfully describe data for dilepton spectra from @xmath0-gev @xmath1 collisions , inclusive direct photon spectra for @xmath1 collisions at rhic energies @xmath2 gev , and for @xmath3 collisions at tevatron energies @xmath4 tev , in a formalism that is free from any extra parameters . |
[ sec : intro ] dynamic networks , instantiated , e.g. , by wireless sensor networks , mobile ad - hoc networks and vehicle area networks , are becoming ubiquitous nowadays .
the primary properties of such networks are sets of participants ( called processes in the sequel ) that are a priori unknown and potentially changing , time - varying connectivity between processes , and the absence of a central control .
dynamic networks is an important and very active area of research @xcite . accurately modeling dynamic networks is challenging , for several reasons : first , process mobility , process crashes / recoveries , deliberate joins / leaves , and peculiarities in the low - level system design like duty - cycling ( used to save energy in wireless sensor networks ) make static communication topologies , as typically used in classic network models , inadequate for dynamic networks .
certain instances of dynamic networks , in particular , peer - to - peer networks @xcite and inter - vehicle area networks @xcite , even suffer from significant churn , i.e. , a large number of processes that can appear / disappear over time , possibly in the presence of faulty processes @xcite , and hence consist of a potentially unbounded total number of participants over time .
more classic applications like _ mobile ad - hoc networks _ ( manets )
@xcite , wireless sensor networks @xcite and disaster relief applications @xcite typically consist of a _ bounded _ ( but typically unknown ) total number of processes .
second , communication in many dynamic networks , in particular , in wireless networks like manets , is inherently broadcast : when a process transmits , then every other process within its transmission range will observe this transmission either by legitimately receiving the message or as some form of interference .
this creates quite irregular communication behavior , such as capture effects and near - far problems @xcite , where certain ( nearby ) transmitters may `` lock '' a receiver and thus prohibit the reception of messages from other senders .
consequently , the `` health '' of a wireless link between two processes may vary heavily over time @xcite . for low - bandwidth wireless transceivers , an acceptable link quality
usually even requires communication scheduling @xcite ( e.g. , time - slotted communication ) for reducing the mutual interference .
overall , this results in a frequently changing spatial distribution of pairs of nodes that can communicate at a given point in time . as a consequence , many dynamic networks ,
in particular , wireless ones @xcite , are not adequately modeled by means of bidirectional links : fading and interference phenomenons @xcite , including capture effects and near - far problems , are _ local _ effects that affect only the receiver of a wireless link .
given that the sender , which is also the receiver of the reverse link , resides at a different location , the two receivers are likely to experience very different levels of fading and interference @xcite .
this effect is even more pronounced in the case of time - slotted communication , where forward and backward links are used at different times .
consequently , the existence of asymmetric communication links can not be ruled out in practice : according to @xcite , 80% of the links in a typical wireless network are asymmetric . despite these facts ,
most of the dynamic network research we are aware of assumes bidirectional links @xcite .
the obvious advantage of this abstraction is simplicity of the algorithm design , as strong communication guarantees obviously make this task easier .
moreover , it allows the re - use of existing techniques for wireline networks , which naturally support bidirectional communication
. however , there are also major disadvantages of this convenient abstraction : first , for dynamic networks that operate in environments with unfavourable communication conditions , e.g. in disaster relief applications or , more generally , in settings with various interferers and obstacles that severely inhibit communication , bidirectional links may simply not be achievable . for implementing distributed services in such settings , algorithms that do not need bidirectional links are mandatory .
second , the entire system needs to be engineered in such a way that bidirectional single - hop communication can be provided within bounded time .
this typically requires relatively dense networks and/or processes that are equipped with powerful communication interfaces , which incur significant cost when compared to sparser networks or / and cheaper or more energy - saving communication devices . and last but not least , if directed single - hop communication was already sufficient to reach some desired goal ( say , reaching some destination process ) via multi - hop messages , waiting for guaranteed single - hop bidirectional communication would incur a potentially significant , unnecessary delay . obviously ,
in such settings , algorithmic solutions that do not need bidirectional single - hop communication could be significantly faster . in this paper
, we thus restrict our attention to dynamic networks consisting of an _ unknown but bounded _
total number of processes , which are interconnected by _ directed _ communication links .
the system is assumed to be synchronous , hence time is measured in discrete _ rounds _ that allow the processes to exchange at most one message .
time - varying communication is modeled as a sequence of _ communication graphs _ , which contain a directed edge between two processes if the message sent in the corresponding round is successfully received .
a bidirectional link is modeled by a pair of directed links that are considered independent of each other here . a natural approach to build robust services despite the dynamic nature of such systems
is to use some sort of distributed agreement on certain system parameters like schedules , frequencies , and operating modes , as well as on application - level issues : such a solution allows to use arbitrary algorithms for generating local proposals , which are supplied as inputs to a consensus algorithm that finally selects one of them consistently at all processes .
as opposed to master - slave - based solutions , this approach avoids the single point of failure formed by the process acting as the master .
the ability to reach _ system - wide _ consensus is hence the most convenient abstraction one could provide here .
the first major contribution of our paper is hence a suite of impossibility results and a consensus algorithm for directed dynamic networks that , to the best of our knowledge , works under the weakest communication guarantees sufficient for consensus known so far .
obviously , however , one can not reasonably assume that every dynamic network always provides sufficiently strong communication guarantees for solving consensus .
fortunately , weaker forms of distributed agreement are sufficient for certain applications . in case of determining communication schedules @xcite , for example , which are used for staggering message transmission of nearby nodes in time to decrease mutual interference , it usually suffices if those processes that have to communicate regularly with each other ( e.g. , for implementing a distributed service within a partition ) agree on their schedule .
a more high - level example would be agreement on rescue team membership @xcite in disaster relief applications . for such applications , suitably designed @xmath2-set agreement algorithms @xcite , where processes must agree on at most @xmath2 different values system - wide , are a viable alternative to consensus ( @xmath10 ) .
this is particularly true if such a @xmath2-set agreement ( i ) respects partitions , in the sense that processes in the same ( single ) partition decide on the same value , and ( ii ) is _ gracefully degrading _
, in the sense that the actual number @xmath2 of different decision values depends on the _ actual _ network topology in the execution : if the network is well - behaved , the resulting @xmath2 is small ( ideally , @xmath10 ) , whereas @xmath2 may increase under unfavorable conditions . whereas any gracefully degrading algorithm must be @xmath2-uniform , i.e. , unaware of any a priori information on @xmath2
, it should ideally also be _
, i.e. , produce the smallest number @xmath2 of different decisions possible .
the second major contribution of our paper are several impossibility results for @xmath2-set agreement in directed dynamic networks , as well as the , to the best of our knowledge , first instance of a worst - case @xmath2-optimal @xmath2-set agreement , i.e. , a consensus algorithm that indeed degrades _ gracefully _ to general @xmath2-set agreement . * detailed contributions and paper organization . * in [ sec : model ] , we introduce our detailed system model , which adopts the _ message adversary _ notation used in @xcite .
it consists of an ( unknown ) number @xmath11 of processes , where communcation is modeled by a sequence of directed communication graphs , one for each round : if some edge @xmath12 is present in the communication graph @xmath13 of round @xmath14 , then process @xmath15 has received the message sent to it by @xmath16 in round @xmath14 .
the message adversary determines the set of links actually present in every @xmath13 , according to certain constraints that may be viewed as network assumptions . with respect to consensus
, we provide the following contributions : 1 .
in [ sec : consimposs ] , we show that communication graphs that are weakly connected in every round are not sufficient for solving consensus , and introduce a fairly weak additional assumption that allows to overcome this impossibility .
our message adversary @xmath17 requires that the communication graph in every round is weakly connected and has one ( possibly changing ) strongly connected component ( called a _ root component _ ) that has no in - coming links from processes outside .
note carefully that every directed graph has at least one root component .
since this assumption is still too weak for solving consensus , @xmath17 also requires that , eventually , there will be @xmath1 consecutive rounds where the processes in the root component remain the same , although the connection topology may still change .
we use the term _ vertex - stable root component _ ( vsrc ) for this requirement . in [ sec : consensus ] , we provide a consensus algorithm that works in this model , and prove its correctness . our algorithm requires a window of stability of @xmath18 rounds , where @xmath19 is the _ dynamic network causal diameter _ of the network (= the number of rounds required to reach all processes in the network from any process in the vertex - stable root component via multi - hop communication ) .
2 . in [ sec : consimposs ] , we show that any consensus and leader election algorithm has to know an a priori bound on @xmath19 . since @xmath20 is a trivial bound on @xmath19 , this implies that no uniform algorithm , i.e. , no algorithm unaware of @xmath11 or @xmath19 , can solve consensus in our model .
in addition , we prove that consensus is impossible both under @xmath6 and under @xmath21 , which shows that @xmath19 is a lower bound for the window of stability of vsrcs .
we also demonstrate that neither reliable broadcast , atomic broadcast , nor causal - order broadcast can be implemented under @xmath17 .
the same is shown to be true for counting , @xmath2-verification , @xmath2-token dissemination , all - to - all token dissemination , and @xmath2-committee election . with respect to @xmath2-set agreement and gracefully degrading consensus
, we provide the following contributions : 1 . in [ sec :
impossibility - proofs ] , we provide a fairly weak natural message adversary @xmath0 that is still too strong for solving @xmath2-set agreement : it reveals that the restriction to at most @xmath2 simultaneous vsrcs in every round is _ not _ sufficient for solving @xmath2-set agreement if just a single vsrc is vertex - stable for less than @xmath22 rounds : a generic reduction of @xmath2-set agreement to consensus introduced in @xcite , in conjunction with certain bivalence arguments , is used to construct a non - terminating run in this case . moreover , _
eventual _ stability of _ all _ vsrcs is also not enough for solving @xmath2-set agreement , not even when it is guaranteed that ( substantially ) less than @xmath2 vsrcs exist simultaneously .
the latter is a consequence of some adversarial partitioning over time , which could happen in our dynamic networks . 2 . in [ sec : sufficiency
] , we show that the message adversary @xmath7 , which combines @xmath8 ( ensuring termination ) with some information flow guarantee @xmath9 between certain vsrcs ( ensuring @xmath2-agreement ) , is sufficient for solving @xmath2-set agreement .
basically , @xmath9 guarantees that at most @xmath2 vsrcs exist in a run that are not affecting each other significantly . despite being fairly strong
, the resulting message adversary @xmath7 allows to implement a @xmath2-uniform @xmath2-set agreement algorithm , which naturally respects partitions and is _ worst - case @xmath2-optimal _ , in the sense that no algorithm can solve @xmath23-set agreement under @xmath7 . to the best of our knowledge
, it is the first gracefully degrading consensus algorithm proposed so far . as a final remark
, we note that the ultimate goal of the latter part of our research are network assumptions for every @xmath24 , which are both necessary and sufficient for solving @xmath2-set agreement . knowing or at least approaching this border is interesting for several reasons : first , it is interesting from a theoretical point of view : @xmath2-set agreement has been a major target for the study of solvability in asynchronous systems with failure detectors since decades .- set agreement is still unknown @xcite .
interestingly , @xcite revealed that there are relations between this classic model and dynamic networks . ]
second , striving for weak network assumptions is always advantageous w.r.t . the assumption coverage in real systems , as they are typically more likely to hold in a given dynamic network .
finally , a set of network assumptions close to the necessary and sufficient ones is needed for @xmath2-optimal @xmath2-set agreement algorithms : whereas our worst - case @xmath2-optimal algorithm only needs a single worst - case run under @xmath7 where it can not solve @xmath23-set agreement , a @xmath2-optimal algorithm must solve @xmath2-set agreement for the _ smallest _ @xmath2 possible in _ every _ run .
we believe that our work constitutes a significant step towards identifying the exact solvability border of @xmath2-set agreement : since necessary and sufficient network conditions in our model must lie somewhere in between ( 3 ) and ( 4 ) , we managed to tightly `` enclose '' them . further tightening the gap and eventually closing it ,
is a topic of future research .
dynamic networks have been studied intensively in research ( see the overview by kuhn and oshman @xcite and the references therein ) . besides work on peer - to - peer networks like @xcite , where the dynamicity of nodes ( churn ) is the primary concern , different approaches for modeling dynamic connectivity have been proposed , both in the networking context and in the context of classic distributed computing .
casteigts et al .
@xcite introduced a comprehensive classification of time - varying graph models .
* there is a rich body of literature on dynamic graph models going back to @xcite , which also mentions for the first time modeling a dynamic graph as a sequence of static graphs .
a more recent paper using this approach is @xcite , where distributed computations are organized in lock - step synchronous rounds .
communication is described by a sequence of per - round communication graphs , which must adhere to certain network assumptions ( like @xmath25-interval connectivity , which says that there is a common subgraph in any interval of @xmath25 rounds ) .
afek and gafni @xcite introduced message adversaries for specifying network assumptions in this context , and used them for relating problems solvable in wait - free read - write shared memory systems to those solvable in message - passing systems .
raynal and stainer @xcite also used message adversaries for exploring the relationship between round - based models and failure detectors . besides time - varying graphs , several alternative approaches that consider missing messages as failures have also been proposed in the past : moving omission failures @xcite , round - by - round fault detectors @xcite , the heard - of model @xcite and the perception - based failure model @xcite . * agreement problems .
* agreement problems in dynamic networks with undirected communication graphs have been studied in @xcite ; agreement in directed graphs has been considered in @xcite . in particular , the work by kuhn et al .
@xcite focuses on the @xmath26-coordinated consensus problem , which extends consensus by requiring all processes to decide within @xmath26 rounds of the first decision .
since they consider only undirected graphs that are connected in every round , without node failures , solving consensus is always possible . in terms of the classes of @xcite ,
the model of @xcite is in one of the strongest classes ( class 10 ) in which every process is always reachable by every other process . on the other hand ,
@xcite do consider directed graphs , but restrict the dynamicity by not allowing stabilizing behavior .
consequently , they also belong to quite strong classes of network assumptions in @xcite . in sharp contrast
, the message adversary tolerated by our algorithms does not guarantee bidirectional ( multi - hop ) communication between all processes , hence falls between the weakest and second weakest class of models defined in @xcite .
the leader election problem in dynamic networks has been studied in @xcite , where the adversary controls the mobility of nodes in a wireless ad - hoc network .
this induces dynamic changes of the ( undirected ) network graph in every round and requires any leader election algorithm to take @xmath27 rounds in the worst case , where @xmath28 is a bound on information propagation .
regarding @xmath2-set agreement in dynamic networks , we are not aware of any previous work except @xcite , where bidirectional links are assumed , and our previous paper @xcite , where we assumed the existence of an underlying _ static _ skeleton graph ( a non - empty common intersection of the communication graphs of all rounds ) with at most @xmath2 _ static _ root components .
note that this essentially implies a directed dynamic network with a static core .
by contrast , in this paper , we allow the directed communication graphs to be fully dynamic . in @xcite , we provided @xmath2-set agreement algorithms for partially synchronous systems with weak synchrony requirements . * degrading consensus problems .
* we are also not aware of related work exploring gracefully degrading consensus or @xmath2-uniform @xmath2-set agreement .
however , there have been several attempts to weaken the semantics of consensus , in order to cope with partitionable systems and excessive faults .
vaidya and pradhan introduced the notion of _ degradable _ agreement @xcite , where processes are allowed to also decide on a ( fixed ) default value in case of excessive faults .
the _ almost everywhere agreement _ problem introduced by @xcite allows a small linear fraction of processes to remain undecided .
aguilera et .
@xcite considered quiescent consensus in partitionable systems , which requires processes outside the majority partition not to terminate .
none of these approaches is comparable to gracefully degrading @xmath2-set agreement , however : on the one hand , we allow more different decisions , on the other hand , all correct processes are required to decide and every decision must be the initial value of some process . ingram et .
@xcite presented an asynchronous leader election algorithm for dynamic systems , where every component is guaranteed to elect a leader of its own .
whereas this behavior clearly matches our definition of graceful degradation , contrary to decisions , leader assignments are revocable and the algorithm of @xcite is guaranteed to successfully elect a leader only once the topology eventually stabilizes .
we consider a synchronous distributed system made up of a fixed set of distributed processes @xmath29 with @xmath30 , which have fixed unique ids and communicate via unreliable message passing . for convenience , we assume that the unique i d of @xmath31 is @xmath32 , and use both @xmath33 and @xmath32 for denoting this process ; `` generic '' processes will also be denoted by @xmath16 , @xmath15 etc .
similar to the @xmath34 model @xcite , we assume that processes organize their computation as an infinite sequence of communication - closed @xcite lock - step rounds .
for every @xmath35 and each round @xmath36 , let @xmath37 be the state of @xmath16 at the beginning of round @xmath14 , taken from the set @xmath38 of all states @xmath16 can possibly enter ; @xmath39 is taken from the set of @xmath16 s initial states @xmath40 .
the round @xmath14 computation of process @xmath16 is determined by two functions that make up @xmath16 s algorithm : the message sending function @xmath41 determines the message @xmath42 , taken from a suitable message alphabet @xmath43 , sent to all other processes in the system by @xmath16 in round @xmath14 , based on @xmath16 s state @xmath44 at the beginning of round @xmath14 . for simplicity
, we assume that some ( possibly @xmath45 ) message is sent to all in a round where there is no proper algorithm message to be broadcast . a receiver may omit to receive a message sent to it in a round , and senders do not know ( without receiving explicit feedback later on ) who successfully received their message .
the transition function @xmath46 takes @xmath16 s state @xmath44 at the beginning of round @xmath14 and a set @xmath47 of pairs of process ids and messages , which contains the round @xmath14 messages received by @xmath16 from other processes in the system , and computes the successor state @xmath48 .
we assume that , for each process
@xmath15 , there is at most one @xmath49 such that @xmath50 is the message @xmath15 sent in round @xmath14 .
note that neither @xmath51 nor @xmath52 need to involve @xmath11 , i.e. , the algorithms executed by the processes may be uniform with respect to the network size @xmath11 .
= [ circle , draw = gray , fill = lightgray!30,thick , inner sep=0pt , minimum size=4 mm ] = [ p , double ] = [ ->,black , thick , auto ] = [ ->,black , thick , densely dotted , auto ] = [ draw = none , opacity=0 ] the evolving nature of the network topology is modeled as an infinite sequence of simple directed graphs @xmath53 , which is determined by an omniscient _ message adversary _ @xcite that has access to the processes states .
[ def : commgraphs ] for each round @xmath14 , the _ round @xmath14 communication graph _
@xmath54 is a simple directed graph with node set @xmath55 and edge set @xmath56 , where @xmath57 _ iff _ @xmath15 successfully receives @xmath16 s round @xmath14 message ( in round @xmath14 ) .
the set @xmath58 denotes @xmath15 s _ in - neighbors _ in @xmath13 ( excluding @xmath15 ) .
note that we will sloppily write @xmath59 to denote @xmath60 , as well as @xmath61 to denote @xmath62 .
[ fig : graphs ] shows a sequence of communication graphs for a network of @xmath63 processes , for rounds @xmath64 to @xmath65 . for deterministic algorithms ,
a run is completely determined by the initial states of the processes and the sequence of communication graphs .
we emphasize that @xmath16 does not have any _ a priori _ knowledge of its neighbors , i.e. , @xmath16 does not know who receives its round @xmath14 message , and does not know who it will receive from in round @xmath14 before its round @xmath14 computation .
since every @xmath66 can range arbitrarily from @xmath11 isolated nodes to a fully connected graph , there is no hope to solve any non - trivial agreement problem without restricting the power of the adversary to drop messages has no outgoing / incoming edges to / from some other processes in @xmath13 . ]
to some extent .
inspired by @xcite , we encapsulate a particular restriction , e.g. , that every communication graph must be strongly connected , by means of a particular _ message adversary_. note that [ def : ma ] generalizes the notation introduced in @xcite , which just specified the _ set _ of communciation graphs the adversary may choose from in every round , to sets of _ sequences _ of communication graphs .
[ def : ma ] a message adversary _ adv _ ( for our system @xmath67 of @xmath11 processors ) is a set of sequences of communication graphs @xmath68 . a particular sequence of communication graphs @xmath69 is _ feasible _ for _ adv _ , if @xmath70 . informally
, we say that some message adversary _ adv _ guarantees some property , called a _ network assumption _ , if every @xmath71 satisfies this property . for our system @xmath67 of @xmath11 processes , this introduces a natural partial order of message adversaries , where @xmath72 is weaker than @xmath73 ( denoted @xmath74 ) iff @xmath75 , i.e. , if it can generate at most the communication graph sequences of @xmath73 . as a consequence , an algorithm that works correctly under message adversary
@xmath73 will also work under @xmath72 . to formally introduce the consensus and
@xmath2-set agreement problem studied in this paper , we assume some finite set @xmath76 and consider the set of possible initial states @xmath40 ( of process @xmath16 ) to be partitioned into @xmath77 subsets @xmath78 $ ] , with @xmath79 .
when @xmath16 starts in a state in @xmath78 $ ] , we say that @xmath80 is @xmath16 s input value , denoted @xmath81 .
moreover , we assume that , for each @xmath82 , there is a set @xmath83\subset\s_p$ ] of decided states such that @xmath83\cap\d_p[w]=\emptyset$ ] if @xmath84 and @xmath85 $ ] is closed under @xmath16 s transition function , i.e. , @xmath52 maps every state in this subset to this subset ( for all possible sets @xmath86 of received messages ) .
we say that @xmath16 has _ decided _ on the output value ( also called decision value ) @xmath80 , denoted @xmath87 , when it is in some state in @xmath83 $ ] .
when @xmath16 performs a transition from a state outside of the set of decided states to the set of decided states , we say that @xmath16 _
decides_. [ def : consensus ] algorithm @xmath88 solves _ consensus _ , if the following properties hold in every run of @xmath88 : * if process @xmath16 decides on @xmath89 and @xmath15 decides on @xmath90 , then @xmath91 . *
if @xmath92 , then @xmath80 is some @xmath93 s initial value @xmath94 .
* every process must eventually decide .
for the _ @xmath2-set agreement problem _
@xcite , we assume that both @xmath95 and @xmath96 to rule out trivial solutions .
[ def : ksa ] algorithm @xmath88 solves _ @xmath2-set agreement _
, if the following properties hold in every run of @xmath88 : * at most @xmath2 different decision values are obtained system - wide in any run . * if @xmath92 , then @xmath80 is some @xmath93 s initial value @xmath94 .
* every process must eventually decide . clearly , consensus is the special case of @xmath64-set agreement ; set agreement is a short - hand for @xmath20-set agreement .
a consensus or @xmath2-set agreement algorithm is called _ uniform _ , if it does not have any a priori knowledge of the network ( and hence of @xmath11 ) .
a @xmath2-set agreement algorithm is called _ @xmath2-uniform _ , if it does not require a priori knowledge of @xmath2 .
we will now define the cornerstones of the message adversaries used in our paper , which culminate in [ def : d - bounded - vsrc ] and [ def : e - network - vertex - stable - roots ] .
message adversaries such as @xmath17 ( [ ass : window ] ) and @xmath0 ( [ ass : inter ] ) will be defined implicitly , by defining the properties of the sequences of feasible communication graphs . informally , most of those will rest on the pivotal concept of _ root components _ , which are strongly connected components in @xmath13 without _ incoming _ edges from processes outside the component .
our message adversaries will be required to eventually guarantee root components that are vertex - stable , i.e. , to consist of the same _ set _ of nodes ( with possibly varying interconnect ) during a sufficiently large number of consecutive rounds .
vertex - stability will eventually guarantee that all members can receive information from each other .
[ def : rc ] a _ root component _ @xmath97 , with non - empty set of vertices @xmath98 , is a _ strongly connected component _
( scc ) in @xmath13 that has no incoming edges from other components , formally @xmath99 . by contracting sccs , it is easy to see that every weakly connected directed simple graph @xmath100 has at least one root component , see [ lem : root ] . hence , if @xmath100 has @xmath2 root components , it has at most @xmath2 weakly connected components ( with disjoint root components , but possibly overlapping in the remaining processes ) .
[ def : vsrc ] a sequence of consecutive rounds with communication graphs @xmath101 for @xmath102 $ ] , @xmath103 , contains an @xmath104-_vertex - stable root component _ @xmath105 , if , for @xmath106 , every @xmath101 contains a root component @xmath107 with the same set of nodes @xmath108 ( but possibly varying interconnection topology ) .
we will abbreviate @xmath105 as an @xmath104-vsrc or @xmath109-vsrc if only the length of @xmath104 matters , and sometimes denote an @xmath104-vsrc @xmath105 by its vertex set @xmath108 if @xmath104 is clear from the context .
note carefully that we assume @xmath110 here , since @xmath111 $ ] ranges from the _ beginning _ of round @xmath112 to the _ end _ of round @xmath113 ; hence , @xmath114 $ ] is not empty but rather represents round @xmath14 .
the most important property of a vsrc @xmath105 is that information is guaranteed to spread to all its vertices @xmath108 if the interval @xmath104 is large enough , as proved in [ lem : bncd ] below . to express this formally ,
we need a few basic definitions and lemmas . similarly to the classic
`` happened - before '' relation @xcite , we say that a process _
@xmath16 causally influences @xmath15 in round @xmath14 _ , denoted by @xmath115 , iff either ( i ) @xmath15 has an incoming edge @xmath116 from @xmath16 in @xmath66 , or ( ii ) if @xmath117 , i.e. , we assume that @xmath16 always influences itself in a round . given a sequence of communication graphs @xmath118 , we say that there is an _ causal influence chain _ of length @xmath119 starting from @xmath16 in round @xmath14 to @xmath15 , denoted by @xmath120}{\leadsto } q)$ ] , if there exists a sequence of not necessarily distinct processes @xmath121 such that @xmath122 for @xmath123 .
if @xmath2 is irrelevant , we just write @xmath115 or just @xmath124 and say that @xmath16 ( in round @xmath14 ) causally influences @xmath15 .
this allows us to define the notion of a dynamic causal distance between processes as given in [ def : dd ] .
[ def : dd ] given a sequence of communication graphs @xmath118 , the _ dynamic causal distance _
@xmath125 from process @xmath16 ( in round @xmath14 ) to process @xmath15 is the length of the shortest causal influence chain starting in @xmath16 in round @xmath14 and ending in @xmath15 , formally @xmath126}{\leadsto}q)\}$ ] .
we define @xmath127 and @xmath128 if @xmath16 never influences @xmath15 after round @xmath14 .
note that , in contrast to the similar notion of dynamic distance defined in @xcite , the dynamic causal distance in our _ directed _ graphs is not necessarily symmetric : if the adversary chooses the graphs @xmath66 such that not all processes are strongly connected , the causal distance between two processes can even be finite in one and infinite in the other direction .
in fact , even if @xmath66 is strongly connected for round @xmath14 ( but not for rounds
@xmath129 ) , @xmath130 can be infinite .
however , the following [ lem : cd ] shows that the causal distance in successive rounds can not arbitrarily decrease .
[ lem : cd ] given a sequence of communication graphs @xmath118 , for every two processes @xmath131 it holds that @xmath132 . as a consequence , if @xmath133 , then also @xmath134 .
since @xmath135 in every round @xmath14 , the definition of dynamic causal distance trivially implies @xmath136 .
analogous to the dynamic diameter defined for undirected communication graphs in @xcite , we now define the _ dynamic causal diameter _
@xmath137 for round @xmath138 in a @xmath104-vsrc @xmath105 as the largest round @xmath138 dynamic causal distance @xmath139 between any pair of processes @xmath140 : [ def : cdiam ] given a sequence of communication graphs @xmath118 , let @xmath111 $ ] , @xmath141 , be a nonempty interval of indices in this sequence.$ ] with a vsrc @xmath105 '' . ]
assume that the subsequence of communication graphs @xmath101 for @xmath142 contains an @xmath104-vsrc @xmath105 with node set @xmath108 .
then , the _ dynamic causal diameter _ of @xmath105 for round @xmath138 is defined as @xmath143 . obviously , it may be the case that @xmath144 in general .
however , if @xmath109 is sufficiently large , the following [ lem:2 ] reveals that @xmath145 .
[ lem:2 ] given some @xmath111 $ ] and a vsrc @xmath105 with @xmath146 , if @xmath147 , then @xmath148 \colon \cdiam^x(r^i ) { \leqslant}|r|-1 $ ] .
fix some process @xmath149 and some @xmath138 where @xmath150 .
let @xmath151 , and define for each @xmath152 the set @xmath153 .
@xmath154 is hence the set of processes @xmath155 such that @xmath156}{\lt}}q)$ ] holds . using induction
, we will show that @xmath157 for @xmath158 .
induction base @xmath159 : @xmath160 follows immediately from @xmath151 .
induction step @xmath161 , @xmath158 : clearly the result holds if @xmath162 , thus we consider round @xmath163 and @xmath164 : it follows from strong connectivity of @xmath165 that there is a set of edges from processes in @xmath166 to some non - empty set @xmath167 .
hence , we have @xmath168 , which implies @xmath169 by the induction hypothesis . thus , in order to guarantee @xmath170 and thus @xmath171 , choosing @xmath2 such that @xmath172 and @xmath173 is sufficient .
since @xmath174 , both conditions can be fulfilled by choosing @xmath175 . moreover , due to the definition of @xmath166 , it follows that @xmath176 for all @xmath155 .
since this holds for any @xmath16 and any @xmath177 , the statement of [ lem:2 ] follows .
[ lem:2 ] thus implies that information available at any node @xmath149 at the beginning of round @xmath178 $ ] has spread to all other nodes in @xmath108 by the end of round @xmath113 , i.e. , during @xmath104 . on the other hand
, it may be the case for some particular vsrc @xmath105 with @xmath179 that the information available at the beginning of some round @xmath106 has already spread to all other nodes in @xmath108 by the end of round @xmath113 .
[ lem : infprop ] reveals that this implies that the information available at any round @xmath180 $ ] has also been spread to all nodes in @xmath108 by the end of round @xmath113 .
[ lem : infprop ] suppose that @xmath105 for @xmath111 $ ] is an @xmath104-vsrc of size @xmath181 , such that there is some @xmath182 $ ] with @xmath183 .
then , for every @xmath184 $ ] , it holds that @xmath185 .
[ lem : cd ] reveals that for all @xmath186 , we have @xmath187 , which implies @xmath188 for every @xmath189 where @xmath190 and proves our lemma . conversely , assume that some particular vsrc @xmath105 is such that information available at the beginning of round @xmath112 reaches all members of @xmath108 by the end of some round @xmath191 , i.e. , @xmath192 for some @xmath193 .
can we infer something about @xmath137 for later rounds @xmath194 in this case ? in particular , will information available at the beginning of round @xmath195 be spread to all nodes by the end of round @xmath113 ? unfortunately , in general , this is not the case , as the following simple example for @xmath196 $ ] and @xmath197 shows : if @xmath198 is the complete graph whereas @xmath199 is a ring , @xmath200 , but information propagation starting at round 2 does not reach all other nodes by the end of of round @xmath201 .
this stimulated the following [ def : d - bounded - vsrc ] , which parameterizes the worst - case information propagation in a vsrc via a parameter @xmath28 that represents its dynamic causal diameter .
informally , it guarantees that messages sent by any process in @xmath108 , in any but the last @xmath202 rounds of @xmath104 , reach all members of @xmath108 within @xmath104 .
[ def : d - bounded - vsrc ] an @xmath104-vertex - stable root component @xmath105 with @xmath111 $ ] is _
@xmath28-bounded _ , with _ dynamic causal diameter _
@xmath203 , if either @xmath204-bounded that is too short to be interesting .
obviously , such a vsrc need not guarantee information propagation within @xmath28 rounds .
note that it would actually be possible to write @xmath205 here , as our algorithms do not even consider @xmath28-vsrcs as interesting ; we chose the present definition for consistency with [ def : e - network - vertex - stable - roots ] for @xmath19-network - bounded vsrcs , however . ] or else @xmath206 \colon \cdiam^x(r^i ) { \leqslant}d$ ] .
[ lem:2 ] showed that every sufficiently long vsrc @xmath105 is @xmath28-bounded for @xmath207 ; all sufficiently long vsrcs are hence necessarily @xmath208-bounded . on the other hand ,
choosing some @xmath209 can be used to force the message adversary to speed - up information propagation accordingly .
for example , we show in [ sec : expander ] that certain expander graph topologies ensure @xmath210 .
to formalize information propagation from root components to the rest of the network , one has to account for the fact that a process @xmath15 outside any root component may be reachable from _ multiple _ root components in general . intuitively speaking , this models dynamic networks that do not `` cleanly '' partition .
given a sequence of communication graphs @xmath118 containing a set @xmath211 of @xmath212 @xmath104-vsrcs , all vertex - stable in the same interval @xmath111 $ ] , let the round @xmath138 _ dynamic network causal diameter _
@xmath213 be the maximum , taken over all processes @xmath214 , of the minimal dynamic causal distance @xmath139 from _ some _ process @xmath215 in round @xmath138 , formally @xmath216 .
[ def : e - network - vertex - stable - roots ] will be used in the sequel to guarantee that every process in the network receives a message from some member of at least one vsrc in @xmath211 within @xmath19 rounds if @xmath217 .
[ def : e - network - vertex - stable - roots ] a set @xmath211 of @xmath212 @xmath104-vsrcs with @xmath111 $ ] is @xmath19-network - bounded , with _
dynamic network causal diameter _
@xmath218 , if either @xmath219 or else @xmath220:\ , \cheight^x(s^i ) { \leqslant}h$ ] .
note that [ def : e - network - vertex - stable - roots ] guarantees @xmath221}{\leadsto } q)$ ] for _ at least one _ but not for all @xmath222 .
moreover , @xmath16 ( and hence @xmath223 ) may be different for different starting rounds @xmath138 in @xmath104 .
a comparison of [ def : e - network - vertex - stable - roots ] and [ def : d - bounded - vsrc ] reveals that it always holds that @xmath224 . moreover , in the case @xmath225 ( where @xmath226 contains a single root component @xmath105 only ) , [ def : e - network - vertex - stable - roots ] is exactly [ def : d - bounded - vsrc ] with the dynamic causal diameter @xmath137 replaced by the dynamic network causal diameter @xmath227 . finally , analogous to [ lem:2 ] , the following [ lem : bncd ] shows that the dynamic network causal diameter @xmath19 is bounded by @xmath20 , provided @xmath228 .
[ lem : bncd ] suppose there is some interval @xmath111 $ ] where there is a set @xmath211 of exactly @xmath212 @xmath104-vertex - stable root components . if @xmath229 and @xmath230 , then @xmath226 is @xmath20-network bounded .
let @xmath231 and fix any @xmath138 where @xmath232 .
define , for each @xmath152 , the set @xmath233 .
@xmath234 is hence the set of processes @xmath15 such that @xmath235}{\leadsto } q)$ ] holds for at least one @xmath236 . using induction
, we will show that @xmath237 for @xmath238 .
induction start @xmath239 follows immediately from @xmath240 with @xmath212 .
induction step @xmath241 : first assume that already @xmath242 ; since @xmath243 , we are done . otherwise , consider round @xmath163 and @xmath244 : since every node @xmath245 is in a weakly connected component containing at least one root in every round , hence also in @xmath246 , there is a set of edges from processes in @xmath247 to some non - empty set @xmath248 .
hence , we have @xmath249 , which implies @xmath250 by the induction hypothesis .
thus , in order to guarantee @xmath251 and thus @xmath252 , choosing @xmath2 such that @xmath253 and @xmath254 is sufficient . since @xmath255 , both conditions can be fulfilled by choosing @xmath256 . moreover , due to the definition of @xmath247 , it follows that for all @xmath214 there is some @xmath257 with @xmath258 , implying @xmath259 . since this holds for any @xmath260 following [ def : e - network - vertex - stable - roots ] , this implies [ lem : bncd ] .
we conclude this section with an example of a network topology that guarantees a dynamic causal network diameter @xmath19 that is much smaller than @xmath20 , which justifies why we introduced this parameter ( as well as @xmath28 ) explicitly in our model .
an undirected graph @xmath100 is an _ @xmath262-vertex expander _ if , for all sets @xmath263 of size @xmath264 , it holds that @xmath265 , where @xmath266 is the set of neighbors of @xmath267 in @xmath100 , i.e. , those nodes in @xmath268 that have a neighbor in @xmath267 .
( explicit expander constructions can be found in @xcite . ) as we need an expander property for _ directed _ communication graphs , we consider , for a vertex / process set @xmath267 and a round @xmath14 , both the set @xmath269 of nodes outside of @xmath267 that are reachable from @xmath267 and the set of nodes @xmath270 that can reach @xmath267 in @xmath14 . [
ass : fast ] ensures an expansion property both for subsets @xmath267 chosen from root components ( property ( a ) ) and other processes ( properties ( b ) , ( c ) ) .
[ ass : fast ] there is a fixed constant @xmath262 and a fixed set @xmath108 such that the following conditions hold for all sets @xmath271 : if @xmath272 and @xmath273 , then @xmath274 and @xmath275 .
if @xmath276 and @xmath277 , then @xmath278 .
if @xmath276 and @xmath279 , then @xmath280 . the following [ lem : expgraphs ] shows that ( 1 ) [ ass : fast ] does not contradict the existence of a single root component and that ( 2 ) these expander topologies guarantee both a dynamic causal diameter @xmath210 for @xmath104-vsrcs with @xmath281 and a dynamic causal network diameter @xmath282 .
[ lem : expgraphs ] there are sequences of graphs @xmath283 with a single root component in every @xmath13 where [ ass : fast ] holds and where , for any such run , there is an interval @xmath104 during which there exists a @xmath28-bounded and @xmath19-network - bounded @xmath104-vertex stable root component with @xmath210 and @xmath282 .
we will first argue that _ directed _ graphs with a single root exist that satisfy [ ass : fast ] .
consider the simple _ undirected _ graph @xmath284 that is the union of an @xmath262-vertex expander on @xmath105 with member set @xmath108 , and an @xmath262-vertex expander on @xmath285 .
we turn @xmath286 into a directed graph by replacing every edge @xmath287 with oriented directed edges @xmath288 and @xmath289 .
this guarantees properties ( a)-(c ) . in order to guarantee the existence of exactly one root component ,
we drop all directed edges pointing to @xmath105 from the remaining graph , i.e. , we remove all edges @xmath288 where @xmath290 and @xmath155 , which leaves properties ( a)-(c ) intact and makes the @xmath108 from [ ass : fast ] the single root component of the graph .
we stress that the actual topologies chosen by the adversary might be quite different from this construction , which merely serves us to show the existence of such graphs .
we also recall that our message adversaries like the one given in [ ass : window ] will rely on vertex - stable root components @xmath105 , which only require that the set of its vertices @xmath108 remain unchanged , whereas the interconnect topology can change arbitrarily . adding [ ass : fast ]
does of course not change this fact .
we will first show that the `` per round '' expander topology stipulated by [ ass : fast ] is strong enough to guarantee that every sufficiently long vsrc is @xmath28-bounded with @xmath210 . for @xmath291 ,
let @xmath292 be the set of processes @xmath15 in @xmath105 with @xmath111 $ ] and @xmath281 such that @xmath293}{\lt}}q)$ ] , and @xmath294 .
the result @xmath210 follows immediately from [ lem:2 ] if @xmath295 , so assume that @xmath296 and consider some process @xmath297 . for round @xmath112 , property ( a )
yields @xmath298 .
in fact , for all @xmath32 where @xmath299 , we can apply property ( a ) to get @xmath300 , hence @xmath301 .
let @xmath302 be the smallest value such that @xmath303 , which guarantees that @xmath304 .
that is , @xmath305 .
now consider any @xmath155 and define @xmath306 as the set of nodes that causally influence the set @xmath307 in round @xmath308 , for @xmath309 .
again , by property ( a ) , we get @xmath310 , so @xmath311 . from the definition of @xmath302 above
, we thus have @xmath312 .
since @xmath313 , it follows that every @xmath314 influences every @xmath155 within @xmath315 rounds .
while the above proof has been applied to the starting round @xmath316 only , it is evident that it carries over literally also for any @xmath317 , which shows that @xmath105 is indeed @xmath28-bounded .
what remains to be shown is that @xmath19-network - boundedness with @xmath318 also holds .
we use properties ( b ) and ( c ) similarly as in the above proof : for any round @xmath319 $ ] , we know by ( b ) that any process @xmath149 has influenced at least @xmath320 nodes by round @xmath321 where @xmath322 by arguing as for the @xmath154 sets above . now ( c ) allows us to reason along the same lines as for the sets @xmath323 above . that is , any @xmath15 in round @xmath324 will be influenced by at least @xmath320 nodes .
therefore , any @xmath16 will influence every @xmath245 by round @xmath324 , which completes the proof .
this confirms that sequences of communication graphs with @xmath325 and @xmath261 indeed exists and are compatible with message adversaries such as @xmath17 stated in [ ass : window ] below .
in this section , we will prove that some a priori knowledge of the dynamic network causal diameter and the existence of a stable interval of a certain minimal size are inevitable for soving consensus in our model .
moreover , we will introduce the message adversary @xmath17 , which will be shown in [ sec : consensus ] to be weak enough for solving consensus if @xmath326 , albeit it is too strong for solving other standard problems in dynamic networks like reliable broadcasting . since consensus is trivially impossible for an unrestricted message adversary , which may just inhibit any communication in the system , we start from a message adversary that guarantees weakly connected communication graphs @xmath13 in every round @xmath14 .
however , it is not difficult to see that this not sufficient for solving consensus , even when all @xmath327 are the same , i.e. , in a static topology : consider the case where @xmath100 contains two root components @xmath328 and @xmath329 ; such a graph obviously exists , cp .
[ lem : root ] below .
if all processes in @xmath328 start with initial value @xmath330 and all processes in @xmath329 start with initial value @xmath64 , they must decide on their own initial value and hence violate agreement .
after all , no process in , say , @xmath328 ever has an incoming link from any process not in @xmath328 .
we hence restrict our attention to message adversaries that guarantee a _ single _
root component in @xmath66 for any round @xmath14 .
[ fig : graphs ] showed a sequence of graphs where this is the case .
some simple properties of such graphs are asserted by [ lem : root ] .
[ lem : root ] any @xmath13 contains at least one and at most @xmath11 root components ( isolated processes ) , which are all disjoint .
if @xmath13 contains a single root component @xmath97 , then @xmath13 is weakly connected , and there is a directed ( out - going ) path from every @xmath331 to every @xmath332 . we first show that every weakly connected directed simple graph @xmath100 has at least one root component . to see this , contract every scc to a single vertex and remove all resulting self - loops .
the resulting graph @xmath333 is a directed acyclic graph ( dag ) ( and of course still weakly connected ) , and hence @xmath333 has at least one vertex @xmath108 ( corresponding to some scc in @xmath100 ) that has no incoming edges . by construction , any such vertex @xmath108 corresponds to a root component in the original graph @xmath100 .
since @xmath13 has at least @xmath64 and at most @xmath11 weakly connected components , the first statement of our lemma follows . to prove the second statement
, we use the observation that there is a directed path from @xmath334 to @xmath80 in @xmath100 if and only if there is a directed path from the vertex @xmath335 ( containing @xmath334 ) to the vertex @xmath336 ( containing @xmath80 ) in the contracted graph @xmath333 . if there is only one root component in @xmath100 ,
the above observations imply that there is exactly one vertex @xmath108 in the contracted graph @xmath333 that has no incoming edges .
since @xmath333 is connected , @xmath108 has a directed path to every other vertex in @xmath333 , which implies that every process @xmath297 has a directed path to every vertex @xmath15 , as required .
it follows from @xcite that assuming a single root component makes consensus solvable if the root component is static . in this paper
, we allow the root component to change throughout the run , i.e. , the ( single ) root component @xmath97 of @xmath66 might consist of a different set of processes in every round round @xmath14 .
however , it will turn out that a sufficiently long interval of vertex - stability is indispensable for solving consensus in this setting . in the sequel
, we will consider the message adversary @xmath17 stated in [ ass : window ] , which implicitly enforces the dynamic network causal diameter @xmath19 according to [ def : e - network - vertex - stable - roots ] and is parameterized by some stability window duration @xmath337 . [ ass : window ] the message adversary @xmath17 is the set of all sequences of communication graphs @xmath283 , where 1 . for every round @xmath14 , @xmath13 contains exactly one root component @xmath97 , 2 .
all vertex - stable root components occurring in any @xmath283 are @xmath19-network - bounded , 3 . for each @xmath283
, there exists some @xmath338 and an interval of rounds @xmath339 $ ] with a @xmath19-network - bounded @xmath340-vertex - stable root component .
note that item ( ii ) has been added to the above definition solely for the sake of our consensus algorithm in [ sec : consensus ] .
all the impossibility results and lower bounds in this section hold also when ( ii ) is dropped or replaced by something ( like @xmath28-bounded vsrcs , as in [ ass : inter ] ) that does not affect item ( iii ) .
first , we relate the message adversary in [ ass : window ] to the classification of @xcite : [ lem : classification ] reveals that it is stronger than the weakest class that requests one node that eventually reaches all others , but weaker than the second class that requests one node that is reached by all .
by contrast , models like @xcite that assume bidirectionally connected graphs @xmath13 in every round belong to the strongest classes ( class 10 ) in @xcite . [
lem : classification ] in every sequence @xmath283 of communication graphs feasible for @xmath341 , 1 .
there is at least one process @xmath16 such that @xmath342 is finite for all @xmath245 , and this causal distance is in fact at most @xmath343 .
conversely , for @xmath344 , the adversary can choose some sequence @xmath283 where no process @xmath16 is causally influenced by all other processes @xmath15 , i.e. , @xmath345 .
[ ass : window ] guarantees that there is ( at most ) one root component @xmath97 in every @xmath13 , @xmath36 .
since we have infinitely many graphs in @xmath283 but only finitely many processes , there is at least one process @xmath16 in @xmath97 for infinitely many @xmath14 .
let @xmath346 be this sequence of rounds .
moreover , let @xmath151 , and define for each @xmath152 the set @xmath347 . using induction , we will show that @xmath348 for @xmath158 .
consequently , by the end of round @xmath349 at latest , @xmath16 will have causally influenced all processes in @xmath67 .
induction base @xmath159 : @xmath350 follows immediately from @xmath151 .
induction step @xmath161 , @xmath158 : first assume that already @xmath351 ; since @xmath352 , we are done .
otherwise , consider round @xmath353 and @xmath354 : since @xmath16 is in @xmath355 , there is a path from @xmath16 to any process @xmath15 , in particular , to any process @xmath15 in @xmath356 .
let @xmath357 be an edge on such a path , such that @xmath358 and @xmath359 .
clearly , the existence of this edge implies that @xmath360 and thus @xmath361 . since this implies @xmath362 by the induction hypothesis , we are done . finally , at most @xmath343 rounds
are needed until all processes @xmath15 have been influenced by @xmath16 , i.e. , @xmath363 : a pigeonhole argument reveals that at least one process @xmath16 must have been in the root component for @xmath20 times after so many rounds . after all , if every @xmath16 appeared at most @xmath364 times , we could fill up at most @xmath365 rounds . by the above result
, this is enough to secure that some @xmath16 influenced every @xmath15 .
the converse statement ( ii ) follows directly from considering a static star , for example , i.e. , a communication graph where there is one central process @xmath366 , and for all @xmath14 , @xmath367 . clearly , @xmath366 can not be causally influenced by any other process , and @xmath368 for any @xmath369 .
on the other hand , this topology satisfy [ ass : window ] , which includes the requirement of at most one root component per round .
next , we examine the solvability of several broadcast problems @xcite under the message adversary of [ ass : window ] , summarized in [ thm : impossibleproblems ] .
although there is a strong bond between some of these problems and consensus in traditional settings , they are _ not _ implementable under our assumptions basically , because there is no guarantee of ( eventual ) bidirectional communication .
[ thm : impossibleproblems ] under the message adversary @xmath17 given in [ ass : window ] , for any @xmath1 , neither _
reliable broadcast _ , _ atomic broadcast _ , nor _ causal - order broadcast _ can be implemented .
moreover , there is no algorithm that solves _ counting _ , _ @xmath2-verification _ , _
@xmath2-token dissemination _
, _ all - to - all token dissemination _ , and _ @xmath2-committee
election_. we first consider reliable broadcast , which requires that when a correct process broadcasts @xmath370 , every correct process eventually delivers @xmath370 .
suppose that the adversary chooses the communication graphs @xmath371 , which matches [ ass : window ] .
clearly , @xmath15 is a correct process in our model . since @xmath16 never receives a message from @xmath15 , @xmath16
can trivially never deliver a message that @xmath15 broadcasts .
for the token dissemination problems stated in @xcite , consider the same communication graphs and assume that there is a token that only @xmath372 has .
since no other process ever receives a message from @xmath372 , token dissemination is impossible . for counting , @xmath2-verification , and @xmath2-committee election
, we return to the static star round graph @xmath367 with central node @xmath366 considered in the proof of [ lem : classification ] .
as the local history of any process is obviously independent of @xmath11 here , it is impossible to solve any of these problems .
we will now show that every correct solution for consensus , as well as for the related leader - election problem , requires some a priori knowledge of the dynamic network causal diameter of the communication graphs generated by the adversary . recall that a uniform algorithm does not have any priori knowledge of the network ,
i.e. , does not even know upper bounds for the dynamic network causal diameter @xmath19 ( and hence for @xmath11 ) .
[ thm : impossdiameter ] there is no uniform algorithm that can solve consensus under the message adversary @xmath17 given in [ ass : window ] , for any @xmath1 .
assume for the sake of a contradiction that there is such a uniform algorithm @xmath373 , w.l.o.g . for a set of input values
@xmath76 that contains 0 and 1 .
consider a run @xmath374 of @xmath373 on a communication graph @xmath100 that forms a ( very large ) static directed line rooted at process @xmath16 and ending in process @xmath15 .
process @xmath16 has initial value @xmath375 $ ] , while all other processes have initial value @xmath330 .
clearly , the uniform algorithm @xmath373 must allow @xmath16 to decide on @xmath80 by the end of round @xmath376 , where @xmath376 is a constant ( independent of @xmath19 and @xmath11 ; we assume that @xmath11 is large enough to guarantee @xmath377 ) .
next , consider a run @xmath378 of @xmath373 that has the same initial states as @xmath374 , and communication graphs @xmath379 that , during rounds @xmath380 $ ] , are also the same as in @xmath374 ( defining what happens after round @xmath376 will be defered ) . in any case , since @xmath374 and @xmath378 are indistinguishable for @xmath16 until its decision round @xmath376 , it must also decide @xmath80 in @xmath378 at the end of round @xmath376 . however , since @xmath381 , @xmath15 has not been causally influenced by @xmath16 by the end of round @xmath376 .
hence , it has the same state @xmath382 both in @xmath378 and in @xmath383 . as a consequence
, it can not have decided by round @xmath376 : if @xmath15 decided @xmath80 , it would violate agreement with @xmath16 in @xmath383 .
now assume that runs @xmath384 , @xmath383 are actually such that the stable window occurs later than round @xmath376 , i.e. , @xmath385 , and that the adversary just reverses the direction of the line then : for all @xmath386 , @xmath387 , @xmath15 is the root and @xmath16 is the last process of the resulting topology .
observe that the resulting @xmath378 still satisfies [ ass : window ] , since @xmath15 itself forms the only root component .
now , @xmath15 must eventually decide on some value @xmath388 in some later round @xmath389 , but since @xmath15 has been in the same state at the end of round @xmath376 in both @xmath378 and @xmath383 , it is also in the same state in round @xmath389 in both runs .
hence , its decision contradicts the decision of @xmath16 in @xmath390 .
we now use a more involved indistinguishability argument to show that a slightly weaker problem than consensus , namely , leader election is also impossible to solve uniformly under the message adversary @xmath341 . the classic leader election problem ( cf .
@xcite ) assumes that , eventually , exactly one process irrevocably elects itself as leader ( by entering a special elected state ) and every other process elects itself as non - leader ( by entering the non - elected state ) .
non - leaders are not required to know the process i d of the leader . whereas it is easy to achieve leader election in our model when consensus is solveable , by just reaching consensus on the process ids in the system , the opposite is not true : since the leader elected by some algorithm need not be in the root component that exists when consensus terminates , one can not use the leader to disseminate a common value to all processes in order to solve consensus atop of leader election .
[ thm : impossdiameterle ] there is no uniform algorithm that can solve leader election under the message adversary @xmath17 given in [ ass : window ] , for any @xmath1 .
we assume that there is a uniform algorithm @xmath373 that solves the problem .
consider the execution @xmath391 of @xmath373 in a static unidirectional chain of @xmath370 processes , headed by process @xmath16 with i d @xmath392 : since @xmath16 has only a single out - going edge and does not know @xmath11 , it can not know whether it has neighbors at all . since it might even be alone in the single - vertex graph consisting of @xmath16 only
, it must elect itself as leader in any @xmath391 , @xmath393 , after some @xmath394 rounds ( @xmath394 may depend on @xmath392 , however , as we do not restrict @xmath373 to be time - bounded ) .
let @xmath392 and @xmath395 be two arbitrary different process ids , and let @xmath394 resp.@xmath396 be the termination times in the executions @xmath391 resp .
@xmath397 , for any @xmath370 , @xmath398 ; let @xmath399 .
we now build a system consisting of @xmath400 processes .
to do so we assume a chain @xmath401 of @xmath402 processes headed by @xmath16 ( with i d @xmath392 ) and ending in process @xmath403 , a second chain @xmath404 of @xmath402 processes headed by @xmath15 ( with i d @xmath395 ) and ending in process @xmath372 , and the process @xmath14 .
now consider an execution @xmath405 , which proceeds as follows : for the first @xmath25 rounds , the communication graph is the unidirectional ring created by connecting the above chains with edges @xmath406 , @xmath407 and @xmath408 ; its root component clearly is the entire ring . starting from round @xmath402 on , process @xmath14 forms the single vertex root component , which feeds , through edges @xmath408 and @xmath409 the two chains @xmath404 and @xmath410 , with @xmath410 being @xmath401 with all edges reversed .
note that , from round @xmath402 on , there is no edge connecting processes in @xmath401 with those in @xmath404 or vice versa .
let @xmath302 be the process that is elected leader in @xmath405 .
we distinguish 2 cases : 1 .
if @xmath411 , then consider the execution @xmath412 that is exactly like @xmath405 , except that there is no edge @xmath406 during the first @xmath25 rounds : @xmath16 with i d @xmath392 is the single root component here .
clearly , for @xmath16 , the execution @xmath412 is indistinguishable from @xmath413 during the first @xmath414 rounds , so it must elect itself leader .
however , since no process in @xmath415 ( including @xmath416 ) is causally influenced by @xmath16 during the first @xmath25 rounds , all processes in @xmath417 have the same state after round @xmath25 ( and all later rounds ) in @xmath412 as in @xmath405 .
consequently , @xmath302 also elects itself leader in @xmath412 as it does in @xmath405 , which is a contradiction .
2 . on the other hand ,
if @xmath418 , we consider the execution @xmath419 , which is exactly like @xmath405 , except that there is no edge @xmath408 during the first @xmath25 rounds : @xmath15 with i d @xmath395 is the single root component here .
clearly , for @xmath15 , the execution @xmath419 is indistinguishable from @xmath420 ( made up of the chain @xmath404 ) during the first @xmath421 rounds , so it must elect itself leader .
however , since no process @xmath403 in @xmath422 ( including @xmath416 ) is causally influenced by @xmath15 during the first @xmath25 rounds , @xmath403 has the same state after round @xmath25 ( and all later rounds ) in @xmath419 as in @xmath405 .
consequently , @xmath302 also elects itself leader @xmath419 as it does in @xmath405 , which is again a contradiction .
this completes the proof of [ thm : impossdiameterle ] .
the goal of this section is to show that root components @xmath105 must be vertex - stable sufficiently long for solving consensus in our model .
in essence , what is needed for this purpose is that every member of the set @xmath108 of processes in @xmath105 is able to reach the entire network . recalling [ def : e - network - vertex - stable - roots ] , this requires @xmath109 to be at least @xmath19 and hence @xmath423 in [ ass : window ] . in order to show that @xmath424 is indeed necessary in our setting
, we will now consider a stronger message adversary @xmath425 given in [ ass : too - short ] below : it is stronger than @xmath424 as its stability interval is shorter , but still slightly weaker than @xmath426 , in that it also guarantees one process to be reached from the processes in @xmath108 within @xmath19 rounds , despite the too short stability interval @xmath104 .
note carefully that , since there is only one such process , it would be reached if @xmath109 was actually @xmath19 .
this property is formally captured by _ almost @xmath427-network - bounded _
vsrcs introduced in [ def : almoste - network - vertex - stable - roots ] , which is slightly weaker than [ def : e - network - vertex - stable - roots ] in that @xmath104-vsrc s with @xmath428 are no longer arbitrary .
[ def : almoste - network - vertex - stable - roots ] an @xmath104-vertex - stable root component @xmath105 with @xmath111 $ ] is _ almost @xmath427-network - bounded _ , with _
dynamic network causal diameter _
@xmath218 , if either @xmath429 or else @xmath430 $ ] there exists a unique @xmath245 with @xmath431 , while for all @xmath432 we have @xmath433 . [ ass : too - short ] the message adversary @xmath425 is the set of all sequences of communication graphs @xmath283 , where 1 . for every round @xmath14 , @xmath13 contains exactly one root component @xmath97 , 2 . all vertex - stable root components
@xmath105 occurring in any @xmath283 are @xmath19-network - bounded , 3 . for each @xmath283 , there exists some @xmath338 and an interval of rounds @xmath434 $ ] with an almost @xmath427-network - bounded @xmath340-vertex - stable root component .
note carefully that [ ass : too - short ] allows the message adversary to choose _ any _ communication graph sequence that is consistent with the conditions stated therein .
in particular , @xmath425 can choose a sequence of communication graphs that ensures a dynamic causal distance @xmath19 between _ any _ specific @xmath435 and @xmath15 in a vsrc with @xmath428 .
moreover , we have the following [ lem : advrelation ] that relates our message adversaries : [ lem : advrelation ] it holds that @xmath436 , so that every sequence of communication graphs generated by the message adversary @xmath424 is also feasible for @xmath425 .
a comparison of [ ass : too - short ] and [ ass : window ] reveals that they differ only in item ( iii ) .
since almost @xmath427-network - bounded is slightly weaker than @xmath19-network - bounded , as the adversary needs to guarantee a network causal distance @xmath437 of at most @xmath438 from every @xmath149 to every @xmath439 in the former , @xmath440 follows : after all , @xmath426 assumes a @xmath19-network - bounded vsrc . on the other hand , [ ass : too - short ]
does not forbid the message adversary to generate a sequence of communication graphs that adheres to [ ass : window ] with @xmath441 , which also confirms @xmath442 and completes our proof .
we will now prove that the message adversary @xmath425 , and hence by [ lem : advrelation ] also @xmath21 , is too strong for solving consensus : processes can _ withold _ information from each other , which causes consensus to be impossible @xcite . in order to simplify our proof , we assume that the adversary has to fix the start of @xmath434 $ ] and the set of root members @xmath108 in the eventually generated root component @xmath443 before the beginning of the execution ( but given the initial values ) . note that this does not strengthen the adversary , and hence does not weaken our impossibility result : for deterministic algorithms , the whole execution depends only on the initial values and the sequence of the @xmath66 s , so the adversary could simulate the execution and determine every @xmath444 based on this . [
lem : neighbours - are - bivalent ] consider two runs of a consensus algorithm @xmath373 under message adversary @xmath425 , for some a priori fixed @xmath434 $ ] and set of processes @xmath108 in @xmath443 , which start from two univalent configurations @xmath445 and @xmath446 that differ only in the state of one process @xmath16 at the beginning of round @xmath14 .
then , @xmath445 and @xmath446 can not differ in valency .
the proof proceeds by assuming the contrary , i.e. , that @xmath445 and @xmath446 have different valency .
we will then apply the same sequence of round graphs to extend the execution prefixes that led to @xmath445 and @xmath446 to get two different runs @xmath447 and @xmath448 .
it suffices to show that there is at least one process @xmath15 that can not distinguish @xmath447 from @xmath448 : this implies that @xmath15 will eventually decide on the same value in both executions , which contradicts the assumed different valency of @xmath445 and @xmath446 .
our choice of the round graphs depends on the following exhaustive cases : 1 . for @xmath290
, we let the adversary choose any root component @xmath449 consisting of the processes in @xmath108 , for all @xmath450 . obviously , every process ( i.e. , we can choose any ) @xmath155 has the same state throughout @xmath447 and @xmath448 .
2 . for @xmath149 and @xmath451 ,
we choose any root component @xmath449 consisting of the processes in @xmath108 for @xmath452 , and @xmath453 for @xmath454 , where @xmath15 is the process that does not hear from any process in @xmath108 ( and hence from @xmath16 ) within @xmath340 according to [ def : almoste - network - vertex - stable - roots ] .
hence , @xmath15 has the same state in @xmath447 and @xmath448 , both during @xmath340 and afterwards , where it is the single root .
3 . for @xmath149 and @xmath455
, we choose graphs @xmath456 where @xmath453 and @xmath16 has only in - edges for @xmath457 ; @xmath15 ( satisfying @xmath458 and hence @xmath459 ) is again the `` distant '' process allowed by [ def : almoste - network - vertex - stable - roots ] . from @xmath460 on , we choose the same graphs @xmath456 as in case ( ii ) .
it is again obvious that @xmath15 has the same state throughout @xmath447 and @xmath448 , since @xmath16 can not communicate to any process before @xmath340 and does not reach @xmath15 within @xmath340 . in any case , for process @xmath15 , the sequence of states in the extensions starting from @xmath445 and @xmath446 is hence the same .
therefore , the two runs are indistinguishable for @xmath15 , which can not hence decide differently .
this provides the required contradiction to the different valencies of @xmath445 and @xmath446 .
the next [ lem : graph - seq ] establishes connectedness of the successor graph of a configuration @xcite .
[ lem : graph - seq ] for any two round @xmath14 graphs @xmath333 and @xmath461 , we can find a finite sequence of graphs @xmath462 , each with a single root component , where any two consecutive graphs differ only by at most one edge .
we say that the configurations @xmath445 resp .
@xmath446 reached by applying @xmath333 resp .
@xmath461 to the same configuration @xmath463 are _ connected _ in this case .
moreover , our construction guarantees that if the root components of @xmath333 and @xmath461 consist of the same set of processes @xmath464 , the same is true for all @xmath465 .
first , we consider two cases with respect to the members @xmath466 and @xmath467 of the respective root components : ( a ) @xmath468 , ( b ) @xmath469 . moreover , for the second part of the proof , we also consider a special case of ( b ) : ( b ) @xmath470 . for case ( b ) ( and thus also for ( b ) ) , we consider @xmath471 . for case
( a ) , we construct @xmath472 from @xmath333 as follows : let @xmath473 and @xmath474 , then @xmath472 has the same edges as @xmath333 plus @xmath475 , thus @xmath476 ( recall that @xmath477 must be reachable from @xmath466 already in @xmath333 ) .
so , now we have that in both cases @xmath333 and @xmath472 differ in at most one edge .
moreover , there is a nonempty intersection between @xmath328 and @xmath467 . in the first phase of our construction ( which continues as long as @xmath478 ) , we construct @xmath479 from @xmath480 , @xmath291 , by choosing one edge @xmath481 from @xmath482 and let @xmath479 have the same edges as @xmath465 plus @xmath483 .
clearly , @xmath465 and @xmath479 differ in at most one edge . moreover
, when adding an edge , we can not add an additional root component , so as long as we add edges we will have that @xmath479 has a single root component @xmath484 .
when we reach a point in our construction where @xmath485 , the first phase ends .
as @xmath465 now contains all the edges in @xmath461 , i.e. , @xmath486 , we have @xmath487 . in the second phase of the construction
, we remove edges . to this end , we choose one edge @xmath481 from @xmath488 , and construct @xmath479 from @xmath465 by removing @xmath483 .
again we have to show that there is only one root component .
since we never remove an edge in @xmath489 , @xmath465 always contains a directed path from some @xmath490 to both @xmath80 and @xmath392 that only uses edges in @xmath489 . as @xmath491 ,
this also holds for @xmath479 .
since there is only one root component in @xmath461 , this implies that there is only one in @xmath479 .
let @xmath492 be the last graph constructed in the first phase , and @xmath493 the last graph constructed in the second phase .
it is easy to see that @xmath494 , which implies that @xmath495 and hence @xmath496 .
this completes the proof of the first part of our lemma . to see that the second part also holds , we consider case ( b ) in more detail and show by induction that @xmath497 . for the base case , we recall that @xmath498 and thus @xmath499 . for the induction step ,
we consider first that the step involves adding an edge @xmath481 ( phase 1 ) : adding an edge can only modify the root component when @xmath500 and @xmath501 .
since such an edge @xmath483 is not in @xmath489 ( as it has the same root component as @xmath502 ) , we can not select it for addition , so the root component does not change .
if , on the other hand , the step from @xmath465 to @xmath479 involves removing the edge @xmath481 ( phase 2 ) , we only need to consider the case where @xmath503 .
( if @xmath500 , then also @xmath504 so the root component can not change by removing @xmath483 . )
but since we never remove edges from @xmath489 , this implies that even after removing @xmath483 there is still a path from @xmath80 to @xmath392 , so the root component can not have changed .
the proof of the following impossibility result follows roughly along the lines of the proof of ( * ? ? ?
* lemma 3 ) .
it shows , by means of induction on the round number , that a consensus algorithm @xmath373 can not reach a univalent configuration after any finite number of rounds .
[ thm : consensusimp ] there is no algorithm that solves consensus under the message adversary @xmath425 , and hence none under @xmath21 .
we follow roughly along the lines of the proof of ( * ? ? ?
* lemma 3 ) and show per induction on the round number , that no algorithm @xmath373 can reach a univalent configuration by round @xmath14 , for any @xmath36 . since no process can have decided in a bivalent configuration , this violates the termination property of consensus . for the base case ,
we consider binary consensus only and argue similar to @xcite but make use of our stronger validity property : let @xmath505 be the initial configuration , where the processes with the @xmath138 smallest ids start with @xmath64 and all others with @xmath330 .
clearly , in @xmath506 all processes start with @xmath330 and in @xmath507 all start with @xmath64 , so the two configurations are @xmath330- and @xmath64-valent , respectively . to see that for some @xmath138 @xmath505 must be bivalent , consider that this is not the case , then there must be a @xmath505 that is @xmath330-valent while @xmath508 is @xmath64-valent .
but , these configurations differ only in @xmath509 , and so by [ lem : neighbours - are - bivalent ] they can not be univalent with different valency . for the induction step
we assume that there is a bivalent configuration @xmath463 at the beginning of round @xmath510 , and show that there is at least one such configuration at the beginning of round @xmath14 .
we proceed by contradiction and assume all configurations at the beginning of round @xmath14 are univalent .
since @xmath463 is bivalent and all configurations at the beginning of @xmath14 are univalent , there must be two configurations @xmath445 and @xmath446 at the beginning of round @xmath14 which have different valency . clearly , @xmath445 and @xmath446 are reached from @xmath463 by two different round @xmath510 graphs @xmath511 and @xmath512 .
[ lem : graph - seq ] shows that there is a sequence of graphs such that @xmath445 and @xmath446 are connected .
each pair of subsequent graphs in this sequence differs only in one link @xmath513 , such that the resulting configurations differ only in the state of @xmath392 .
moreover , if the root component in @xmath333 and @xmath461 is the same , all graphs in the sequence also have the same root component .
since the valency of @xmath445 and @xmath446 was assumed to be different , there must be two configurations @xmath514 and @xmath515 in the corresponding sequence of configurations that have different valency and differ only in the state of one process , say @xmath16 . applying [ lem :
neighbours - are - bivalent ] to @xmath514 and @xmath515 again produces a contradiction , and so not all successors of @xmath463 can be univalent .
we have hence established that @xmath425 is too strong for consensus , which implies the same for @xmath21 according to [ lem : advrelation ] .
in this section , we show that it is possible to solve consensus under the message adversary @xmath517 given in [ ass : window ] .
the underlying idea of our consensus algorithm is to use flooding to propagate the largest input value to everyone .
however , as [ ass : window ] does not guarantee bidirectional communication between every pair of processes according to [ lem : classification ] , flooding is not sufficient : the largest input value could be hidden at a single process @xmath16 that never has outgoing edges .
if such a leaf process @xmath16 would never accept smaller values , it is impossible to reach agreement ( without potentially violating validity ) .
thus , we have to find a way to force @xmath16 to accept also a smaller value . a well - known technique to do so is _ locking _ a candidate value .
obviously , we do not want a leaf process to lock its value , but rather some process(es ) that will be able to impose their locked value , i.e. , can successfully flood the system .
in addition , we may allow processes that have successfully locked a value to decide only when they are sure that every other process has accepted their value as well . according to [ def : e - network - vertex - stable - roots ] , both can be guaranteed when these processes have been in a vertex stable root component long enough which is ( amply ) guaranteed by @xmath517 .
the first major ingredient of our consensus algorithm is a network approximation algorithm ( described in [ sec : approxalgo ] ) , which allows processes to detect their root membership in ( past ) rounds .
the core of our consensus algorithm ( presented in [ sec : consensuscore ] ) then exploits this knowledge for reaching agreement on locked values and imposes the resulting value on all processes in the network . as we will see
, the main complication comes from the fact that a process can detect whether it has been part of the root component of round @xmath14 only with some latency . according to our system model
, no process @xmath16 has any initial knowledge of the network . in order to learn about vsrcs , for example
, it hence needs to _ locally _ acquire such knowledge . process @xmath16 achieves this by means of [ alg : approx ] , which maintains a _ network estimate _ @xmath518 in a local variable . of process @xmath16 in round @xmath14 _ before _ the round @xmath14 computation finishes as @xmath519 ; we usually suppress the superscript when it refers to the current round . ]
@xmath518 is a graph that holds the local estimates of every communication graph @xmath66 that occurred so far , simply by labeling an edge @xmath520 with the set of round numbers of every @xmath66 once @xmath16 received evidence that @xmath520 was present in round @xmath14 .
initially , @xmath518 consists of process @xmath16 only . in every round ,
every process @xmath16 broadcasts its current @xmath518 and fuses it with the network estimates received from its neighbors . in more detail
, @xmath16 updates @xmath518 whenever @xmath521 , by adding @xmath522 if @xmath15 is @xmath16 s neighbour for the first time , or by updating the label of the edge @xmath523 to @xmath524 ( [ line : addedge1 ] and [ line : addedge2 ] ) .
moreover , @xmath16 also receives @xmath525 from @xmath15 and uses this information to update its own knowledge : the loop in [ line : forloop ] ensures that @xmath16 has an edge @xmath526 for each @xmath527 in @xmath525 , where @xmath25 is the set of rounds previously known to @xmath16 . given @xmath518 ,
we use @xmath528 with @xmath330 for intervals that span at most the last @xmath529 rounds , i.e. , any older information could safely be removed from the approximation graph , resulting in a message complexity that is polynomial in @xmath11.]@xmath530 to denote the current estimate of @xmath531 contained in @xmath518 .
formally , @xmath528 is the graph induced by the set of edges @xmath532 as the information about @xmath15 s neighbors in @xmath531 might take many rounds to reach some process @xmath16 ( if it ever arrives at @xmath16 ) , @xmath528 may never be fully up - to - date , and as only reported edges are added to the estimate ( but not all reports need to reach @xmath16 ) , @xmath528 will be an under - approximation of @xmath531 .
for example , a process @xmath16 that does not have any incoming links from other processes , throughout the entire run of the algorithm , can not learn anything about the remaining network , i.e. , @xmath518 will permanently be the singleton graph .
[ alg : approx ] finally provides an externally callable function @xmath533 , which will be used by the core consensus consensus algorithm to find out whether the calling process @xmath16 was member in an @xmath104-vsrc @xmath105 and to query the set of all members @xmath108
. we will prove in [ lem : cpr2root ] below that the latter is the case if @xmath528 is strongly connected and consists of the same non - empty set @xmath108 of processes for all @xmath534 .
informally , this is due to the fact that the members of an @xmath104-vsrc will not be able to acquire knowledge of the topology outside @xmath105 within @xmath104 , as they do not have incoming links from outside .
provides externally callable function @xmath535 .
* variables and initialization : * @xmath536 initially @xmath537 * emit round @xmath14 messages : * send @xmath538 to all current neighbors * round @xmath14 : computation : * replace @xmath483 with @xmath539 in @xmath540 where @xmath541 [ line : addedge1 ] add @xmath542 to @xmath540 [ line : addedge2 ] @xmath543 [ line : forloop ] replace @xmath544 in @xmath540 with @xmath545 ; add @xmath527 if no such edge exists [ line : forloopend ] let @xmath546 be induced graph of @xmath547 let @xmath548 be @xmath546 if it is strongly connected , or the empty graph otherwise.[line : cps ] return @xmath549 return @xmath550 we start our analysis of [ alg : approx ] with [ lem : asubsetg ] , which shows that @xmath528 underapproximates @xmath531 in a way that consistently includes neighborhoods .
its proof uses the trivial invariant asserting @xmath551 at the end of every round @xmath552 .
[ lem : asubsetg ] if @xmath528 contains @xmath513 at the end of some round @xmath14 , then ( i ) @xmath553 , i.e. , @xmath554 , and ( ii ) @xmath528 also contains @xmath555 for every @xmath556 .
we first consider the case where @xmath552 , then at the end of round @xmath14 @xmath528 is empty , i.e. , there are no edges in @xmath528 . as the precondition of the lemma s statement is false
, the statement is true . for the case
where @xmath557 , we proceed by induction on @xmath14 : induction base @xmath558 : if @xmath528 contains @xmath513 at the end of round @xmath558 , it follows from @xmath559 at the end of every round @xmath552 , for every @xmath245 , that @xmath560 , since @xmath16 is the only processor that can have added this edge to its graph approximation . clearly , it did so only when @xmath561 , i.e. , @xmath562 , and included also @xmath555 for every @xmath563 on that occasion
. this confirms ( i ) and ( ii ) .
induction step @xmath564 , @xmath557 : assume , as our induction hypothesis , that ( i ) and ( ii ) hold for any @xmath565 at the end of round @xmath14 , in particular , for every @xmath566 .
if indeed @xmath513 in @xmath528 at the end of round @xmath567 , it must be contained in the union of round @xmath14 approximations @xmath568 and hence in some @xmath569 ( @xmath570 or @xmath571 ) at the end of round @xmath14 .
note that the edges ( labeled @xmath567 ) added in round @xmath567 to @xmath518 are irrelevant for @xmath528 here , since @xmath572 .
consequently , by the induction hypothesis , @xmath562 , thereby confirming ( i ) . as for (
ii ) , the induction hypothesis also implies that @xmath555 is also in this @xmath569 .
hence , every such edge must be in @xmath573 and hence in @xmath528 at the end of round @xmath567 as asserted .
the following [ lem : cpr2root ] shows that locally detecting @xmath528 to be strongly connected ( in [ line : cps ] of [ alg : approx ] ) implies that @xmath16 is in the root component of round @xmath403 .
this result rests on the fact that @xmath528 underapproximates @xmath531 ( [ lem : asubsetg].(i ) ) , but does so in a way that never omits an in - edge at any process @xmath574 ( [ lem : asubsetg].(ii ) ) .
[ lem : cpr2root ] if the graph @xmath575 ( [ line : cps ] ) with @xmath576 is non - empty in round @xmath14 , then @xmath16 is member of @xmath577 , i.e. , @xmath149 .
for a contradiction , assume that @xmath575 is non - empty ( hence @xmath528 is an scc by [ line : cps ] ) , but @xmath290 . since @xmath16 is always included in any @xmath518 by construction and @xmath528 underapproximates @xmath531 by [ lem : asubsetg].(i ) , this implies that @xmath528 can not be the root component of @xmath531 .
rather , @xmath528 must contain some process @xmath392 that has an in - edge @xmath513 in @xmath531 that is not present in @xmath528 . as @xmath392 and hence some edge @xmath578
is contained in @xmath528 , because it is an scc , [ lem : asubsetg].(ii ) reveals that this is impossible . from the definition of the function @xmath533 in [ alg : approx ] and [ lem : cpr2root ] , we get the following [ cor : stable2root ] .
[ cor : stable2root ] if the function @xmath533 evaluates to @xmath579 at process @xmath16 in round @xmath14 , then @xmath580 where @xmath581 , it holds that @xmath16 is a member of @xmath582 , i.e. , @xmath149 .
the following [ lem : root2cpr ] proves that , in a sufficiently long @xmath111 $ ] with a @xmath104-vertex - stable root component @xmath105 , every member @xmath16 of @xmath105 detects an scc for round @xmath112 ( i.e. , @xmath583 ) with a latency of at most @xmath28 rounds ( i.e. , at the end of round @xmath584 ) . informally speaking , together with [ lem : cpr2root ] , it asserts that if there is an @xmath104-vertex - stable root component @xmath105 for a sufficiently long interval @xmath104 , then a process @xmath16 observes @xmath585 from the end of round @xmath584 on iff @xmath149 .
[ lem : root2cpr ] consider an interval of rounds @xmath111 $ ] , such that there is a @xmath28-bounded @xmath104-vertex - stable root component @xmath105 and assume @xmath586 .
then , from the end of round @xmath584 onwards , we have @xmath587 , for every process in @xmath435 .
consider any @xmath588 . at the beginning of round @xmath589
, @xmath15 has an edge @xmath590 in its approximation graph @xmath525 with @xmath591 iff @xmath592 .
since processes always merge all graph information from other processes into their own graph approximation , it follows from the definition of a @xmath28-bounded @xmath104-vertex - stable root component ( [ def : d - bounded - vsrc ] ) in conjunction with the fact that @xmath593 that every @xmath435 has these in - edges of @xmath15 in its graph approximation by the end of round @xmath594 .
since @xmath105 is a vertex - stable root - component , it is strongly connected without in - edges from processes outside @xmath105 .
hence @xmath587 from the end of round @xmath584 on , as asserted .
this immediately gives us the following [ cor : allrootdec ] , which ensures that in a sufficiently long @xmath104-vsrc @xmath105 , with @xmath111 $ ] and member set @xmath108 , every @xmath149 detects its membership in the @xmath340-vsrc @xmath443 , @xmath595 \subseteq i$ ] , with a latency of at most @xmath28 rounds .
[ cor : allrootdec ] consider an interval of rounds @xmath111 $ ] , with @xmath596 , such that there is a @xmath28-bounded vertex - stable root component @xmath105 .
then , from the end of round @xmath113 on , a call to @xmath597)$ ] returns @xmath108 at every process in @xmath108 . together , [ cor : stable2root , cor : allrootdec ] reveal that @xmath598 _ precisely _ characterizes the caller s actual membership in the @xmath599$]-vsrc @xmath105 in the communication graphs from the end of round @xmath113 on .
* variables and initialization : * @xmath600 , initially own input value @xmath601 initially @xmath602 @xmath603 initially @xmath330 * emit round @xmath14 messages : * send @xmath604 to all neighbors [ line : senddec ] send @xmath605 to all neighbors * round @xmath14 computation : * [ line : secondif ] [ line : firstif ] @xmath606 decide on @xmath607 and set @xmath608[line : decother ] @xmath609 [ line : updatexp ] @xmath610 @xmath611 [ line : lock ] decide on @xmath607 and set @xmath608 [ line : decideown ] [ line : decown ] @xmath612 as explained in [ sec : consensus ] , the core consensus algorithm stated in [ alg : consensus ] builds upon the network approximation algorithm given as [ alg : approx ] : relying on [ cor : stable2root ] , every process uses @xmath613 provided by [ alg : approx ] to detect whether it has been in the vertex - stable root component of some past round(s ) .
since [ cor : allrootdec ] reveals that @xmath613 has a latency of up to @xmath614 rounds for reliably detecting that a process is in the vertex - stable root component of some ( interval of ) rounds , our algorithm ( conservatively ) looks back @xmath28 rounds in the past when locking a value . in more detail , [ alg : consensus ] proceeds as follows : initially , no process has locked a value , that is , @xmath615 and @xmath616 .
processes try to detect whether they are privileged by evaluating the condition in [ line : if - g ] .
when this condition is true in some round @xmath302 , they lock the current value ( by setting @xmath617 and @xmath618 to the current round ) , unless @xmath619 is already @xmath620 . note that our locking mechanism does not actually protect the value against being overwritten by a larger value being also locked in @xmath302
; it locks out only those values that have older locks @xmath621 .
when the process @xmath370 that had the largest value in the root component of round @xmath302 detects that it has been in a vertex - stable root component in all rounds @xmath302 to @xmath622 ( [ line : if - lock - heard ] ) , it can decide on its current value . as all other processes in that root component must have had @xmath370 s value imposed on them , they can decide as well . after deciding ,
a process stops participating in the flooding of locked values , but rather ( [ line : senddec ] ) floods the network with @xmath623 .
since the stability window guaranteed by [ ass : window ] with @xmath624 is large enough to allow every process to receive this message , all processes will eventually decide .
before we turn our attention to the correctness proof of [ alg : consensus ] , we need to define how the network approximation algorithm and the core consensus algorithm are combined to form a joint algorithm in our computation model .
informally , we assume that ( i ) the complete round @xmath14 computing step of the network approximation algorithm is executed just before the round @xmath14 computing step of the consensus algorithm , and that ( ii ) the round @xmath14 message of the former is piggybacked on the round @xmath14 message of the latter .
consequently , the round @xmath14 computing step of the consensus core algorithm , which terminates round @xmath14 , can already access the _ result _ of the round @xmath14 computation of the network approximation algorithm , i.e. , its state at the _ end _ of round @xmath14 .
consequently , [ cor : stable2root , cor : allrootdec ] reveal that a call to @xmath533 with @xmath625 $ ] by @xmath16 in the transition function of round @xmath113 ( or later ) returns @xmath626 _ precisely _ when a vsrc @xmath105 containing @xmath16 existed .
formally , let @xmath627 , @xmath628 , @xmath629 , @xmath630 be the set of states , message alphabet , transition function , and message sending function of the network approximation algorithm , with @xmath631 , @xmath632 and @xmath633 denoting its state at the beginning of round @xmath14 , the message sent in round @xmath14 , and the set of messages received in round @xmath14 .
analogously , let @xmath634 , @xmath635 , @xmath636 , @xmath637 , @xmath638 , @xmath639 and @xmath640 be the corresponding entities for the core consensus algorithm ; note that @xmath641 , albeit the core consensus algorithm only reads ( but never writes ) the graph approximation @xmath518 ( when calling @xmath613 ) . for the joint algorithm , we define the joint state space as @xmath642 and the joint message alphabet as @xmath643 .
we assume that there are projection functions @xmath644 resp .
@xmath645 which , given @xmath646 , can be used to obtain the corresponding @xmath647 resp .
the joint message sending function @xmath649 just computes the pair of messages @xmath650 via @xmath651 and @xmath652 .
the joint transition function @xmath653 first applies @xmath629 to @xmath654 to compute ( i ) @xmath655 and ( ii ) an intermediate state @xmath656 that is identical to @xmath638 except that @xmath657 is replaced by the newly computed @xmath658 .
@xmath636 is then applied to @xmath656 to compute the state @xmath659 , which finally results in @xmath660 .
all this happens atomically and instantaneously at the round switching time .
our correctness proof starts with the validity property of consensus according to [ def : consensus ] .
[ lem : validity ] every decision value is the input value of some process .
processes decide either in [ line : decother ] or in [ line : decown ] .
when a process decides via the former case , it has received a @xmath661 message , which is sent by @xmath15 iff @xmath15 has decided on @xmath662 in an earlier round . in order to prove validity , it is thus sufficient to show that processes can only decide on some process input value when they decide in [ line : decown ] , where they decide on their current estimate @xmath663 .
let the round of this decision be @xmath14 .
the estimate @xmath663 is either @xmath16 s initial value , or was updated in some round @xmath664 in [ line : updatexp ] from a value received by way of one of its neighbors @xmath665 message . in order to send such a message , @xmath15 must have had @xmath666 at the beginning of round @xmath667 , which in turn means that @xmath662 was either @xmath15 s initial value , or @xmath15 has updated @xmath662 after receiving a message in some round @xmath668 . by repeating this argument , we will eventually reach a process that sent its initial value , since no process can have updated its decision estimate prior to the first round .
the following [ lem : assorted - properties ] states a number of properties maintained by our algorithm when the first process @xmath16 has decided .
essentially , they say that there has been a vertex - stable root component in the interval @xmath669 $ ] centered around the lock round @xmath302 ( but not earlier ) , and asserts that all processes in that root component chose the same lock round @xmath302 .
[ lem : assorted - properties ] suppose that process @xmath16 decides in round @xmath14 , no decisions occurred before @xmath14 , and @xmath670 , then * @xmath16 is in the vertex - stable root component @xmath105 with @xmath669 $ ] and member set @xmath108 , * @xmath671 , * @xmath672 , where @xmath466 is the members set of the vsrc @xmath673 , and * all processes in @xmath108 executed [ line : lock ] in round @xmath302 , and no process in @xmath674 can have executed [ line : lock ] in a round @xmath675 . item ( i ) follows since [ line : if - g ] has been continuously @xmath620 since round @xmath302 and from [ lem : cpr2root ] .
as for item ( ii ) , @xmath676 follows from the requirement of [ line : if - lock - heard ] , while @xmath677 follows from ( i ) and the fact that by [ lem : root2cpr ] the requirement of [ line : if - lock - heard ] can not be , for the first time , fullfilled strictly after round @xmath678 . from [ lem : root2cpr ] , it also follows that if @xmath679 , then the condition in [ line : if - g ] would return true already in round @xmath680 , thus locking would occur already in round @xmath680 .
since @xmath16 did not lock in round @xmath680 , ( iii ) must hold .
finally , from ( i ) , ( iii ) , and [ lem : root2cpr ] , it follows that every other process in @xmath108 also has @xmath681)=\true$ ] in round @xmath302 . moreover , due to ( iii ) ,
@xmath682)=\false$ ] in round @xmath680 , which causes all the processes in @xmath108 ( as well as those in @xmath674 ) to set @xmath618 to 0 . since @xmath683)$
] can not become true for any @xmath684 at a process @xmath685 , as @xmath686 for any @xmath687 by [ cor : stable2root ] , ( iv ) also holds .
the following [ lem : highlander ] asserts that if a process decides , then it has successfully imposed its proposal value on all other processes .
[ lem : highlander ] suppose that process @xmath16 decides in [ line : decideown ] in round @xmath14 and that no other process has executed [ line : decideown ] before @xmath14 .
then , for all @xmath15 , it holds that @xmath688 . using items ( i ) and ( iv ) in [ lem : assorted - properties ]
, we can conclude that @xmath16 was in the vertex - stable root component of rounds @xmath670 to @xmath622 and that all processes in it member set @xmath108 have locked in round @xmath302 .
therefore , in the interval @xmath689 $ ] , @xmath302 is the maximal value of @xmath618 .
more specifically , all processes @xmath15 in @xmath108 have @xmath690 , whereas all processes @xmath372 in @xmath691 have @xmath692 during these rounds by [ lem : assorted - properties].(iv ) .
let @xmath693 have the largest proposal value @xmath694 among all processes in @xmath108 .
since @xmath370 is in @xmath108 , there is a causal chain of length at most @xmath19 from @xmath370 to any @xmath245 .
note carefully that guaranteeing this property requires item
( ii ) of [ ass : window ] , as the first decision ( in round @xmath14 ) need not occur in the eventually guaranteed @xmath695-vsrc but already in some earlier `` spurious '' vsrc . since no process executed [
line : decideown ] before round @xmath14 , no process will send @xmath696 messages in @xmath689 $ ] .
thus , all processes continue to execute the update rule of [ line : updatexp ] , which implies that @xmath697 will propagate along the aforementioned causal path to @xmath15 .
[ thm : consensus ] let @xmath698 be the beginning of the stability window guaranteed by the message adversary @xmath517 given in [ ass : window ] .
then , [ alg : consensus ] in conjunction with [ alg : approx ] solves consensus by the end of round @xmath699 .
validity holds by [ lem : validity ] . considering [ lem : highlander ]
, we immediately get agreement : since the first process @xmath16 that decides must do so via [ line : decown ] , there are no other proposal values left in the system .
observe that , so far , we have not used the liveness part of [ ass : window ] .
in fact , [ alg : consensus ] is always safe in the sense that agreement and validity are not violated , even if there is no vertex - stable root component .
we now show the termination property . by [ cor : allrootdec ] , we know that every process in @xmath149 evaluates the predicate @xmath700)=\true$ ] in round @xmath701 , thus locking in that round .
furthermore , [ ass : window ] and [ cor : allrootdec ] imply that at the latest in round @xmath702 every process @xmath149 will evaluate the condition of [ line : if - lock - heard ] to @xmath620 and thus decide using [ line : decown ] .
thus , every such process @xmath16 will send out a message @xmath703 . by [ def : e - network - vertex - stable - roots ] and [ ass : window ] , we know that every @xmath245 will receive a @xmath696 message at the latest in round @xmath704 and decide by the end of this round .
in this section , we will turn our attention from consensus to general @xmath2-set agreement and prove related impossibility results and lower bounds .
we will accomplish this by showing that certain `` natural '' message adversaries do not allow to solve @xmath2-set agreement .
for example , as excessive partitioning of the system into more than @xmath2 root components makes @xmath2-set agreement trivially impossible , one natural assumption is to restrict the maximum number of root components per round in our system to @xmath2 . [
ass : inter ] below defines the generic message adversary @xmath0 , which allows at most @xmath2 vsrcs per round and guarantees a common window of vertex stability of duration at least @xmath1 .
note that it implicitly involves both the dynamic causal diameter @xmath28 and the dynamic network causal diameter @xmath705 according to [ def : d - bounded - vsrc ] and [ def : e - network - vertex - stable - roots ] ( that have be enforced by the message adversary ) .
[ ass : inter ] the message adversary @xmath0 is the set of all sequences of communication graphs @xmath283 , where 1 . for every round @xmath14 , @xmath13 contains at most @xmath2 root components , 2 .
all vertex - stable root components occurring in any @xmath283 are @xmath28-bounded , 3 . for each @xmath283
, there exists some @xmath338 and an interval of rounds @xmath339 $ ] where @xmath706 @xmath19-network - bounded vertex - stable root components @xmath707 exist simultaneously .
like for [ ass : window ] , item ( ii ) has only been added for the sake of the @xmath2-set agreement algorithm ( [ alg : ksa ] ) ; the impossibility results and lower bounds also hold when ( ii ) is dropped or replaced by something that does not affect item ( iii ) .
observe that @xmath708 is the same as @xmath17 except that item ( ii ) requires all vsrcs to be @xmath28-bounded instead of @xmath19-network - bounded .
note also that the message adversary @xmath709 guarantees at most @xmath2 vsrcs in every @xmath13 , @xmath36 .
we will now prove that it is impossible to solve @xmath2-set agreement for @xmath710 under the message adversary @xmath711 , even under the slightly weaker version of this message adversary stated in [ thm : d - static - roots ] below .
we will use the generic impossibility theorem provided in ( * ? ? ?
1 ) for this purpose . in a nutshell ,
the latter exploits the fact that @xmath2-set agreement is impossible if @xmath2 sufficiently disconnected components may occur and consensus can not be solved in some component .
we first introduce the required definitions : two executions of an algorithm @xmath712 are _ indistinguishable _ ( until decision ) for a set of processes @xmath713 , denoted @xmath714 , if for any @xmath715 it holds that @xmath16 executes the same state transitions in @xmath262 and in @xmath405 ( until it decides ) .
now consider a model of a distributed system @xmath716 that consists of the set of processes @xmath67 and a _ restricted model _ @xmath717 that is computationally compatible to @xmath43 ( i.e. , an algorithm designed for a process in @xmath43 can be executed on a process in @xmath718 ) and consists of the set of processes @xmath719 .
let @xmath373 be an algorithm that works in system @xmath716 , where @xmath720 denotes the set of runs of algorithm @xmath373 on @xmath43 , and let @xmath721 be a nonempty set of processes .
given any restricted system @xmath722 , the _ restricted algorithm _
@xmath723 for system @xmath718 is constructed by dropping all messages sent to processes outside @xmath713 in the message sending function of @xmath373 .
we also need the following similarity relation between runs in computationally compatible systems ( cf .
* definition 3 ) ) : let @xmath724 and @xmath725 be sets of runs , and @xmath713 be a non - empty set of processes .
we say that _ runs @xmath725 are compatible with runs @xmath724 for processes in @xmath713 _ , denoted by @xmath726 , if @xmath727 . [ thm : impossibility ] let @xmath728 be a system model and consider the runs @xmath729 that are generated by some fixed algorithm @xmath72 in @xmath730 , where every process starts with a distinct input value .
fix some nonempty and pairwise disjoint sets of processes @xmath731 , and a set of distinct decision values @xmath732 .
moreover , let @xmath733 and @xmath734 .
consider the following two properties : * for every set @xmath735 , value @xmath736 was proposed by some @xmath737 , and there is some @xmath738 that decides @xmath736 . * if @xmath739 then @xmath93 receives no messages from any process in @xmath713 until every process in @xmath740 has decided .
let @xmath741 and @xmath742 be the sets of runs of @xmath72 where ( dec-@xmath740 ) respectively both , ( dec-@xmath713 ) and ( dec-@xmath740 ) , hold . is by definition compatible with the runs of the restricted algorithm @xmath743 . ]
suppose that the following conditions are satisfied : @xmath744 is nonempty .
@xmath745 .
in addition , consider a restricted model @xmath746 such that the following properties hold : there is no algorithm that solves consensus in @xmath747 .
. then , @xmath72 does not solve @xmath2-set agreement in @xmath730 .
the proof of [ thm : d - static - roots ] below utilizes [ thm : impossibility ] in conjunction with the impossibility of consensus under @xmath21 established in [ thm : consensusimp ] .
[ thm : d - static - roots ] there is no algorithm that solves @xmath2-set agreement with @xmath749 processes under the message adversary @xmath711 stated in [ ass : inter ] , for any @xmath750 , even if there are @xmath23 root components @xmath751 that are vertex - stable all the time , i.e. , in @xmath752 $ ] ( and only root component @xmath753 is vertex - stable for at most @xmath754 rounds ) .
suppose that there is a @xmath2-set algorithm @xmath373 that works correctly under the assumptions of our theorem . for @xmath10 , [ thm : d - static - roots ]
is implied by [ thm : consensusimp ] , since @xmath5 is the same as @xmath755 if item ( ii ) is dropped in both definitions . to prove the theorem for @xmath756 , we will show that the conditions of the generic [ thm : impossibility ] are satisfied , thereby providing a contradiction to the assumption that @xmath373 exists .
let @xmath757 for @xmath758 and let @xmath759 .
consequently , @xmath740 = @xmath760 and @xmath761 . *
( a ) * the set of runs @xmath762 of @xmath373 where no process in @xmath740 receives any message from @xmath713 before it dedices is nonempty : we choose the communication graph in every round to be such that @xmath740 has no incoming links from @xmath713 until every process in @xmath740 has decided . since any such sequence of communication graphs satisfies the assumptions of our theorem , @xmath763 . *
( b ) * the set of runs @xmath764 of @xmath373 where both ( i ) some process in every @xmath765 decides @xmath736 and ( ii ) no process in @xmath740 receives any message from @xmath713
before it decides satisfies @xmath766 : let @xmath767 be the set of runs where processes @xmath33 have unique input values @xmath768 , @xmath769 , the communication graph in every round is such that @xmath770 are isolated , and @xmath771 are weakly connected ( with a single root ) until every process has decided . by the assumptions of our theorem ,
@xmath767 is non - empty .
since ( i ) the processes in @xmath740 never receive a message from a process in @xmath713 in both @xmath744 and @xmath767 , and ( ii ) the initial values of the processes in @xmath740 are not restricted in @xmath767 in any way , it is easy to find , for any run @xmath772 , a run @xmath773 such that @xmath774 . because obviously @xmath775 , we have established @xmath776 . *
( c ) * consensus is impossible in @xmath777 : let @xmath740 be the partition containing the @xmath2^th^ root component @xmath753 , which is perpetually changing in every round , except for some interval of rounds @xmath778 $ ] , where @xmath779 , for some fixed @xmath698 . during this interval , let the topology of @xmath740 be such that there exists some @xmath780 and some @xmath781 with @xmath782 . since @xmath783 , such a topology ( e.g. a chain with head @xmath16 and tail @xmath15 ) can be created by the message adversary @xmath21 underlying [ thm : consensusimp ] exists . hence , consensus is impossible in @xmath740 . *
( d ) * @xmath784 : fix any run @xmath785 and consider a run @xmath786 , where every process in @xmath740 has the same sequence of state transitions in @xmath787 as in @xmath788 .
such a run @xmath787 exists , since the processes in @xmath740 can be disconnected from @xmath713 in every round in @xmath720 , so @xmath774 . since [ thm : d - static - roots ] tells us that no @xmath2-set agreement algorithm ( for @xmath750 ) can _ terminate _ with insufficient concurrent stability of the at most @xmath2 root components in the system , it is tempting to assume that @xmath2-set agreement becomes solvable if a round exists after which all communication graphs remain the same .
however , we will prove in [ thm : rgst - impossibility ] below that this is not the case for any @xmath789 .
we will again use the generic [ thm : impossibility ] , this time in conjunction with the variant of the well - known impossibility of consensus with lossy links @xcite provided in [ lem : lossylink - simulation ] , to prove that ensuring at most @xmath2 different decision values is impossible here , as too many decision values may originate from the unstable period .
[ lem : lossylink - simulation ] let @xmath790 be a two - processor subsystem of our system @xmath791 .
if the sequence of communication graphs @xmath13 , @xmath36 , of @xmath43 are restricted by the existence of a round @xmath792 such that ( i ) for @xmath793 , @xmath794 and/or @xmath795 , and no other edges incident with @xmath16 or @xmath15 are in @xmath13 , and ( ii ) for @xmath796 , there are no edges incident with @xmath16 and @xmath15 at all in @xmath13 , then consensus is impossible in @xmath718 .
up to @xmath667 , this is ensured by the impossibility of 2-processor consensus with a lossy but at least unidirectional link established in ( * ? ? ? * lemma 3 ) .
after @xmath667 , this result continues to hold ( and is even ensured by the classic lossy link impossibility @xcite ) .
hence , consensus is indeed impossible in @xmath718 .
[ thm : rgst - impossibility ] there is no algorithm that solves @xmath2-set agreement for @xmath797 processes under the message adversary @xmath798 , for every @xmath799 .
suppose again that there is a @xmath2-set algorithm @xmath373 that works correctly under the assumptions of our theorem .
we restrict our attention to runs of @xmath720 where , until @xmath800 , ( i ) the same set of @xmath23 root components @xmath801 with @xmath759 exists in every round , and ( ii ) two remaining processes @xmath802 exist , which are ( possibly only uni - directionally , i.e. , via a lossy link ) connected in every round , without additional edges to or from @xmath713 .
after @xmath800 , the communication graph remains the same , except that the processes in @xmath740 are disconnected from each other and there is an edge from , say , @xmath803 to some process in @xmath713 in every round .
note that these runs satisfy [ ass : inter ] for @xmath804 , as the number of root components never exceeds @xmath2 .
moreover , we let the adversary choose @xmath800 sufficiently large such that the processes in @xmath713 have decided .
since the processes in @xmath735 ( @xmath805 ) never receive a message from the remaining system before @xmath800 , in which case they must eventually unilaterally decide , we can safely assume this .
we can now again employ the generic impossibility in this modified setting .
the proofs of properties ( a ) , ( b ) and ( d ) remain essentially the same as in [ thm : d - static - roots ] .
it hence only remains to prove : consensus is impossible in @xmath777 : this follows immediately from [ lem : lossylink - simulation ] with @xmath806 .
the following [ thm : strong - rgst - impossibility ] reveals that even ( considerably ) less than @xmath2 root components per round before stabilization and a single perpetually stable root component after stabilization are not sufficient for solving @xmath2-set agreement .
[ thm : strong - rgst - impossibility ] there is no algorithm that solves @xmath2-set agreement for @xmath797 processes under the message adversary @xmath807 , for every @xmath799 , even if @xmath808 , @xmath809 , where @xmath100 contains only a single root component .
we show that , under the assumption that @xmath373 exists , there is a sequence of communication graphs that is feasible for our message adversary that leads to a contradiction .
we choose @xmath768 for all @xmath31 and let @xmath810 for @xmath811 .
if @xmath2 is even , let @xmath812 ; if @xmath2 is odd , let @xmath813 . in any case , let @xmath814 .
finally , let @xmath815 .
note that @xmath740 may be empty , while all @xmath735 are guaranteed to contain at least one process since @xmath816 .
for all rounds , the processes in @xmath740 have an incoming edge from a process in one of the @xmath735 .
we split the description of the adversarial strategy into @xmath817 phases in each of which we will force some @xmath735 to take @xmath818 decisions . to keep processes @xmath819 with @xmath820 from deciding on the same value before their respective phase @xmath32
, the adversary restricts @xmath13 such that ( i ) there are no links to @xmath735 from any other @xmath821 and ( ii ) either the edge @xmath116 or @xmath822 or both are in @xmath13 , in a way that causes [ lem : lossylink - simulation ] to apply .
note carefully that any such @xmath13 indeed has no more than @xmath823 root components . in the initial phase
, @xmath824 is forced to decide : since @xmath825 has no incoming edges from another node in @xmath13 , this situation is indistinguishable from a run where @xmath825 became the single root after @xmath800 .
thus , by the correctness of @xmath373 , @xmath825 must eventually decide on @xmath826 . at this point ,
the initial phase ends , and we can safely allow the adversary to modify @xmath13 in such a way that @xmath825 has an incoming edge from some other process .
we now proceed with @xmath827 phases : in the @xmath32^th^ phase , @xmath811 , the adversary drops any link between the processes @xmath828 ( and does not provide an incoming link from any other process , as before ) in any @xmath13 .
since , for both @xmath16 and @xmath15 , this is again indistinguishable from the situation where they become the single root after @xmath800 , both will eventually decide in some future round ( if they have not already decided ) .
since the adversary may have chosen a link failure pattern in earlier phases that causes the impossibility (= forever bivalent run ) of [ lem : lossylink - simulation ] to apply , as @xmath829 , it follows that @xmath373 and hence @xmath830 can not have solved consensus in @xmath735 .
since @xmath373 solves @xmath2-set agreement , @xmath16 and @xmath15 must hence decide on two _ different _ values .
moreover , since neither @xmath16 nor @xmath15 ever received a message from a process not in @xmath735 , their decision values must be different from the ones in all former phases .
finally , after @xmath16 and @xmath15 have made their decisions , the adversary may again modify @xmath13 such that they have an incoming edge from some other process , thereby reducing the number of root components by two and preserving the maximum number @xmath823 of root components , and continue with the next phase .
if @xmath2 is even , then the final phase @xmath827 forces two more decisions just as described above ; otherwise , @xmath831 provides one additional decision value ( which happens concurrently with the initial phase here ) . in either case
, we have shown that all @xmath33 with @xmath832 have decided on different values , which contradicts the assumption that a correct algorithm @xmath373 exists .
note that [ thm : strong - rgst - impossibility ] reveals an interesting gap between @xmath201-set agreement and 1-set agreement , i.e. , consensus : it shows that @xmath201-set agreement is impossible with @xmath833 root components per round before and a single fixed root component after stabilization .
by contrast , if we reduce the number of root components per round to a single one before stabilization ( and still consider a single fixed root thereafter ) , even @xmath64-set agreement becomes solvable @xcite .
in this section , we will provide a message adversary @xmath9 ( [ ass : stablestream ] ) that is sufficiently weak for solving @xmath2-set agreement if combined with @xmath834 ( [ ass : inter ] ) .
although we can of course not claim that it is a strongest one in terms of problem solvability ( we did not even define what this means ) , we have some indications that it is close to the solvability / impossibility border . to illustrate some of the ideas that will be used in our message adversary for general @xmath2-set agreement , we start with the simple case of @xmath20-set agreement ( also called _ set agreement _ ) first .
note that [ thm : d - static - roots ] does not apply here . to circumvent the impossibility result of [ thm : strong - rgst - impossibility ]
, it suffices to strengthen the assumption of at most @xmath20 root components in every round such that the generation of too many decision values during the unstable period is ruled out . a straightforward way to achieve
this is to just forbid @xmath11 different decisions obtained in root components consisting of a single process .
achieving this is easy under the @xmath835-influence message adversary given in [ asm : influenceset ] , the name of which has been inspired by the @xmath835 failure detector @xcite .
[ asm : influenceset ] the message adversary @xmath836 is the set of all sequences of communication graphs @xmath283 , where in any set @xmath837 of @xmath11 root components consisting of single processes @xmath838 , @xmath839 , occuring in any run the following holds : there are two indices @xmath32 , @xmath840 such that @xmath841 _ influences _
@xmath842 , denoted @xmath843 , in the sense that there exists a causal chain starting after @xmath844 that ends before or at the beginning of @xmath845 .
it is easy to devise a set agreement algorithm that works correctly in a dynamic network under [ asm : influenceset ] , provided ( a bound on ) @xmath11 is known : in [ alg : setagreement ] , process @xmath33 maintains a proposal value @xmath736 , initially @xmath846 , and a decision value @xmath847 , initially @xmath848 , which are broadcast in every round .
if @xmath33 receives no message from any other process in a round , it decides by setting @xmath849 . if @xmath33 receives a message from some @xmath93 that has already decided ( @xmath850 ) , it sets @xmath851 .
otherwise , it updates @xmath736 to the maximum of @xmath736 and all received values @xmath852 . at the end of round @xmath11 , a process that has not yet decided sets @xmath853 , and all processes terminate .
* set agreement algorithm , code for process @xmath33 : * @xmath854 // initial value @xmath855 * emit round @xmath14 messages : * send @xmath856 to all * receive round @xmath14 messages : * receive @xmath857 from all current neighbors * round @xmath14 : computation : * @xmath858 @xmath859 @xmath853 @xmath853 ; terminate [ thm : setagreement ] [ alg : setagreement ] solves @xmath20-set agreement in a dynamic network under message adversary @xmath836 given in [ asm : influenceset ] .
termination ( after @xmath11 rounds ) and also validity are obvious , so it only remains to show @xmath20-agreement .
assume , w.l.o.g .
, that the processes @xmath860 are ordered according to their initial values @xmath861 , and let @xmath862 be the set of different values ( in @xmath847 or , if still @xmath863 , in @xmath736 ) present in the system at the beginning of round @xmath864 ; @xmath865 is the set of initial values .
obviously , @xmath866 , and since @xmath20-agreement is fulfilled if @xmath867 , we only need to consider the case where all @xmath846 are different
. consider process @xmath803 : if @xmath803 gets a message from some other process @xmath93 in round 1 , @xmath868 as ( i ) @xmath803 does not decide on its own value and sets @xmath869 and ( ii ) no process that receives a message containing @xmath870 from @xmath803 takes on this value .
hence , @xmath20-set agreement will be achieved in this case .
otherwise , @xmath803 does not get any message in round 1 and hence decides on @xmath870 . proceeding inductively ,
assume that @xmath871 has decided on @xmath872 by round @xmath873 , and received only messages from processes with smaller index in rounds @xmath874 and no message in round @xmath2 .
now consider process @xmath33 : if @xmath33 gets a message from some process @xmath93 with @xmath875 in some round @xmath876 , with minimal @xmath2 , before it decides , then @xmath877 as ( i ) @xmath33 does not decide on its own value and sets @xmath878 , ( ii ) @xmath33 did not send its value to any process in @xmath879 before their decisions , and ( iii ) no process with index larger than @xmath32 that receives a message containing @xmath846 from @xmath33 takes on this value . hence , @xmath20-set agreement will be achieved in this case . otherwise ,
if @xmath33 gets a message from some process @xmath880 in round @xmath32 , it will decide on @xmath881 s decision value @xmath872 and hence also cause @xmath882 . in the only remaining case , @xmath33 does not get any message in round @xmath32 and hence decides on @xmath846 , which completes the inductive construction of @xmath883 for @xmath884 .
now consider @xmath885 in round @xmath11 in the above construction of @xmath886 : [ asm : influenceset ] prohibits the only case where @xmath20-agreement could possibly be violated , namely , when @xmath885 also decides on @xmath887 : during the first @xmath11 rounds , we would have obtained @xmath11 single - node root components no two of which influence each other in this case .
thus , we can not extend the inductive construction of @xmath888 to @xmath889 , as the resulting execution would be infeasible . whereas the set agreement solution introduced in the previous subsection
is simple , it is apparent that [ asm : influenceset ] is quite demanding . in particular
, it requires explicit knowledge of ( a bound on ) @xmath11 .
we will now provide a message adversary @xmath9 ( [ ass : stablestream ] ) , which is sufficient for general @xmath2-set agreement if combined with @xmath834 ( [ ass : inter ] ) .
we obtained this combination by adding some additional properties to the necessary network conditions implied by our impossibility [ thm : d - static - roots , thm : strong - rgst - impossibility].-set agreement could be to generalize [ asm : influenceset ] : one could e.g. assume that at least two out of every set of @xmath20 different root components consisting of 1 or 2 processes are influenced by a common predecessor root component .
whereas this assumption does not require vertex stability of root components , it effectively ensures that information propagates not slower as in vsrcs .
owing to this fact , it also prohibits the existence of the node @xmath15 in [ ass : too - short ] with causal distance @xmath28 from @xmath16 in the root component , thereby causing the proof of [ thm : d - static - roots ] to fail . working out the details
may turn out difficult , though : after all , unlike single - process roots , larger root components suffer from the problem that its members can not always determine whether the root was a vsrc or not .
influence must hence be conservative , in the sense that it involves even _ potential _ 2-process roots . ] to avoid non - terminating (
i.e. , forever undecided ) executions as predicted by [ thm : d - static - roots ] , we require the _ stable interval _ constraint guaranteed by the message adversary @xmath834 to hold . the parameter @xmath28 , which can always be safely set to @xmath890 according to [ lem : infprop ] , allows to adapt the message adversary to the actual dynamic causal diameter guaranteed in the vsrcs of a given dynamic network .
note that , since @xmath203 , rounds where no message is received are not forbidden here ( in contrast to [ asm : influenceset ] ) . in order to also circumvent executions violating the @xmath2-agreement property established by [ thm : strong - rgst - impossibility ]
, we introduce the _ majority influence _ constraint guaranteed by the message adversary @xmath9 given in [ ass : stablestream ] below .
like [ asm : influenceset ] for set agreement , it guarantees some ( minimal ) information flow between sufficiently long - lasting vertex - stable root components that exist at different times . as visualized in [ fig : model_majorinf ] , it implies that the information available in any such vsrc originates in at most @xmath2 `` initial '' vsrcs .
thereby , it enhances the very limited information propagation that could occur in our model solely under @xmath891 , which is too strong for solving @xmath2-agreement .
= [ circle , draw , text = black , inner sep=3mm , scale=0.6 ] = [ circle , fill = black!20,draw = black , text = black , inner sep=3mm , scale=0.6 ] = [ line width=0.8mm , draw = black ] \(a ) @xmath892 ; ( b ) [ right of = a ] @xmath893 ; ( c ) [ right of = b ] @xmath894 ; ( d ) [ right of = c ] @xmath895 ; \(i ) [ below left of = b ] @xmath896 ; ( k ) [ below right of = b ] @xmath897 ; \(f ) [ below left of = k ] @xmath898 ; ( g ) [ right of = f ] @xmath899 ; ( h ) [ right of = g ] @xmath900 ; ( e ) [ left of = f , ] @xmath901 ; \(a ) edge node ( b ) ( b ) edge[myarrows ] node ( c ) ( c ) edge[myarrows ] node ( d ) \(i ) edge node ( f ) ( i ) edge node ( b ) ( k ) edge node ( d ) \(e ) edge node ( f ) ( f ) edge[myarrows ] node ( g ) ( f ) edge node ( k ) ( g ) edge[myarrows ] node ( h ) ; ( @xmath902 )
( @xmath903)node[below right ] @xmath403 ; formally , given some run @xmath787 , we denote by @xmath904 the set of all root components that are vertex - stable for at least @xmath1 consecutive rounds in @xmath787 .
let @xmath905 be vertex - stable in @xmath906 $ ] and @xmath907 be vertex - stable in @xmath908 $ ] with @xmath909 ; note that @xmath910 for every @xmath911 .
[ def : influence ] given any two @xmath912 , we say that some process @xmath913 _ influences _ some process @xmath914 and write @xmath915 with @xmath916 iff there exists a causal chain from @xmath16 to @xmath15 starting after @xmath917 that ends before or at the beginning of @xmath918 , i.e. , @xmath919 . in this case
, we also say that @xmath920 ( weakly ) influences @xmath921 and write @xmath922 , using the relation @xmath923 here .
we will also need stronger notions of influence , which are based on the following [ def : influence - set ] : [ def : influence - set ] given any two @xmath924 , their influence set is @xmath925 .
the _ majority influence _ between the nodes in @xmath926 and @xmath927 guarantees that @xmath926 influences a set of nodes in @xmath927 , which is greater than any set influenced by vsrcs not already known by the processes in @xmath926 ( and greater than or equal to any set influenced by vsrcs already known by the processes in @xmath926 ) .
majority influence is hence a very natural way to discriminate between strong and weak influence between vsrcs , see [ def : stronginfluence ] below .
[ def : majorityinfluence ] we say that a vsrc @xmath928 exercises a _ majority influence _ on a vsrc @xmath929 , denoted @xmath930 with @xmath931 , iff @xmath932 with @xmath933 it holds that @xmath934 and @xmath932 with @xmath935 it holds that @xmath936 . with these preparations , we are now ready to specify a message adversary @xmath9 given in [ ass : stablestream ] . [
ass : stablestream ] the message adversary @xmath9 is the set of all sequences of communication graphs @xmath283 , where in every run @xmath937 with @xmath938 s.t.@xmath939 with @xmath940 . informally speaking , [ ass : stablestream ] ensures that all but at most @xmath2 `` initial '' vsrcs in @xmath941 are majority - influenced by some earlier vsrc in @xmath941 ( see [ fig : model_majorinf ] ) .
note carefully , though , that [ ass : stablestream ] neither prohibits partitioning of the system in more than @xmath2 simultaneous vsrcs nor directly exhibits a @xmath2-quorum property , cf .
the well - known quorum failure detector @xmath942 @xcite that is known to be necessary ( but not sufficient ! ) for solving @xmath2-set agreement : after all , one could e.g. choose @xmath943 vsrc s @xmath893 , @xmath895 and @xmath944 in [ fig : model_majorinf ] without finding any pair among those which are majority - influenced by a common predecessor vsrc .
therefore , @xmath9 alone is too strong for solving @xmath2-set agreement .
the same is true for an alternative to [ ass : stablestream ] that just ensures a @xmath2-quorum ( unless acyclicity could be guaranteed as well ) .
conversely , if majority influence was replaced by strong influence according to [ def : stronginfluence ] , a quorum property could be easily established : starting out from an arbitrary set of @xmath945 @xmath946-vsrcs , we could go back along the ( acyclic ) majority influence relation until we end up in the set @xmath947 guaranteed by [ ass : stablestream ] .
if a @xmath2-set agreement algorithm relied on @xmath946-vsrcs for decisions , this would guarantee that no more than @xmath2 decision values ( possibly fabricated in the `` initial '' @xmath946-vsrcs ) can be produced . a message adversary equivalent to [ ass : stablestream ] with strong majority
would be fairly weak , however .
these observations indicate that @xmath948 is indeed reasonably close to the @xmath2-set agreement solvability border .
we conclude this section with some straightforward stronger assumptions , which also imply [ ass : stablestream ] and can hence be handled by the algorithm introduced in [ sec : algorithms ] : replacing majority influence in [ def : majorityinfluence ] by majority intersection @xmath949 , which is obviously the strongest form of influence . requiring @xmath950 , i.e. , a majority intersection with respect to the number of processes in @xmath927 .
this could be interpreted as a changing vsrc , in the sense of `` @xmath927 is the result of changing a minority of processes in @xmath926 '' .
although this restricts the rate of growth of vsrcs in a run , it would apply , for example , in case of random graphs where the giant component has formed @xcite . in this section
, we provide a @xmath2-set agreement algorithm and prove that it works correctly under the message adversary @xmath948 , i.e. , the conjunction of [ ass : inter , ass : stablestream ] .
note that the algorithm needs to know @xmath28 , but neither @xmath11 nor @xmath19 .
it consists of a `` generic '' @xmath2-set agreement algorithm , which relies on the network approximation algorithm of [ sec : approxalgo ] for locally detecting vertex - stable root components and a function @xmath951 that extracts candidate decision values from history information .
our implementation of @xmath951 uses a vector - clock - like mechanism for maintaining `` causally consistent '' history information , which can be guaranteed to lead to proper candidate values thanks to @xmath948 . in sharp contrast to classic @xmath2-set agreement algorithms ,
the algorithm is _
@xmath2-uniform _ , i.e. , the parameter @xmath2 does not appear in its code . rather , the number of system - wide decision values is determined by the number of ( certain ) @xmath946-vsrcs occurring in the particular run . as a consequence ,
if the network partitions into @xmath2 weakly connected components , for example , decision values need _ not _ involve @xmath2 isolated partitions : obviously , @xmath2 isolated partitions in the communication graph also imply @xmath2 root components , but @xmath2 root components do not imply a partitioning of the communication graph into @xmath2 weakly connected components one process may still be connected to several components . ]
all processes in a component obtain the same decision value . on the other hand ,
if the network remains well - connected , the algorithm guarantees a unique decision value system - wide .
our algorithm is in fact not only @xmath2-uniform but even worst - case @xmath2-optimal , in the sense that ( i ) it provides at most @xmath2 decisions system - wide in all runs that are feasible for @xmath948 , and ( ii ) that there is at least one feasible run under @xmath948 where no correct @xmath2-set agreement can guarantee less than @xmath2 decisions .
( i ) will be proved in [ sec : corproof ] , and ( ii ) follows immediately from the fact that a run consisting of @xmath2 isolated partitions is also feasible for @xmath948 .
our algorithm can hence indeed be viewed as a consensus algorithm that degrades gracefully to @xmath2-set agreement , for some @xmath2 determined by the actual network properties .
01 -10 like the consensus algorithm in [ sec : consensus ] , our @xmath2-set agreement algorithm consists of two reasonably independent parts , the network approximation algorithm [ alg : approx ] and the @xmath2-set agreement core algorithm given in [ alg : ksa ] . as in [ sec : consensuscore ] , we assume that the complete round @xmath14 computing step of the network approximation algorithm is executed just before the round @xmath14 computing step of the @xmath2-set algorithm , and that the round @xmath14 message of the former is piggybacked on the round @xmath14 message of the latter . recall that this implies that the round @xmath14 computing step of the @xmath2-set core algorithm , which terminates round @xmath14 , can already access the result of the round @xmath14 computation of the network approximation algorithm , i.e. , its state at the end of round @xmath14 .
* variables and initialization : * @xmath952 [ * ] : = \emptyset$ ] / * @xmath953[r]$ ] holds @xmath33 s estimate of the locks learned by @xmath93 in round @xmath14 * / @xmath954[0 ] : = \set{(\set{p_i } , x_i , \r-1)}$ ] / * virtual first lock @xmath955 at @xmath33 * / [ line : inithist ] @xmath956 // most recent lock round , @xmath848 if none @xmath957 // @xmath33 s decision , @xmath848 if undecided * emit round r messages : * send @xmath958 to all neighbors [ line : sendhistory ] * receive round r messages : * for all @xmath93 in @xmath33 s neighborhood @xmath959 , receive @xmath960 * round r computation : * decide @xmath961 and set @xmath962 [ line : adoptdec ] // update @xmath963 with @xmath964 received from neighbors @xmath965 // remember current history @xmath966[r ' ] : = \hist_i[x][r ' ] \cup \hist_j[x][r']$ ] [ line : remotehistoryupdate ] // locally add all newly learned locks : @xmath954 : = \hist_i \setminus \hist'_i$ ] [ line : learnlock ] //
perform state transitions ( undecided , locked , decided ) : @xmath967 [ line : lock ? ] @xmath968 @xmath969 [ line : setlock ] @xmath954[r ] : = \hist_i[i][r ] \cup \lock$ ] //
create new lock [ line : addnewlock ] @xmath956 //
release unsuccessful lock [ line : releaselock ] [ line : decide ? ] decide @xmath970 and set @xmath971 [ line : setdecision ] let @xmath267 be the multiset @xmath972[r'']$ ] + let @xmath973 be the set of the most frequent elements in @xmath267 [ line : bounded - history ] let @xmath974 [ line : hottestmajority ?
] let @xmath80 be @xmath975 of the single element @xmath976 @xmath977 [ line : createlockmajority ] [ line : ambiguousmajority ? ] @xmath978 [ line : createlockmaximum ] //
deterministic choice @xmath979 [ line : returnnewlock ] the general idea of our core @xmath2-set agreement algorithm in [ alg : ksa ] is to generate new decision values only at members of @xmath946-vsrcs , and to disseminate those values throughout the remaining network .
using the network approximation @xmath980 , our algorithm causes process @xmath33 to make a transition from the initially _ undecided _ state to a _ locked _ state when it detects some minimal `` stability of its surroundings '' , namely , its membership in some @xmath981-vsrc @xmath28 rounds in the past ( [ line : lock ? ] ) . note that the latency of @xmath28 rounds is inevitable here ,
since information propagation within a @xmath981-vsrc may take up to @xmath28 rounds due to @xmath28-boundedness , as guaranteed by item ( ii ) in [ ass : inter ] .
if process @xmath33 , while in the locked state , observes some period of stability that is sufficient for locally inferring a consistent view among _ all _ vsrc members ( which occurs when the @xmath981-vsrc has actually extended to a @xmath946-vsrc ) , @xmath33 can safely make a transition to the _ decided _ state ( [ line : setdecision ] ) .
the decision value is then broadcast in all subsequent rounds , and adopted by any not - yet decided process in the system that receives it later on ( [ line : adoptdec ] ) .
note that @xmath834 ( [ ass : inter ] ) guarantees that this will eventually happen .
since locking is done optimistically , however , it may also happen that the @xmath981-vsrc does not extend to a @xmath946-vsrc ( or , even worse , is not recognized to have done so by some members ) later on . in this case , @xmath33 makes a transition from the locked state back to the undecided state ( [ line : releaselock ] ) .
unfortunately , this possibility has severe consequences : meachanisms are required that , despite possibly inconsistently perceived unsuccessful locks , ensure both ( a ) an _ identical _ decision value among all members of a @xmath946-vsrc who successfully detect this @xmath946-vsrc and thus reach the decided state , and ( b ) no more than @xmath2 different decision values originating from different @xmath946-vsrcs .
both goals are accomplished by a particular selection of the decision values ( using function @xmath982 ) , which ultimately relies on an intricate utilization the network properties guaranteed by our message adversary @xmath948([ass : inter , ass : stablestream ] ) : our algorithm uses a suitable _ lock history _ data structure for this purpose , which is continuously exchanged and updated among all reachable processes .
it is used to store sets of _ locks _ @xmath983 , which are created by every process that enters the locked state : @xmath108 is the vertex - set of the detected @xmath981-vsrc , @xmath80 is a certain proposal value ( determined as explained below ) , and @xmath984 is the round when the lock is created . in more detail , the lock history at process @xmath33 consists of an array @xmath953[r]$ ] that holds @xmath33 s ( under)approximation of the locks process @xmath93 got to know in round @xmath14 .
it is maintained using the following simple update rules : _ local lock creation : _ apart from the single _ virtual _ lock @xmath985 created initially by @xmath33 in [ line : inithist ] ( which guarantees a non - empty lock history right from the beginning ) , all regular locks created upon @xmath33 s transition from the undecided to the locked state are computed by the function @xmath982 in [ line : setlock ] .
any lock locally created at @xmath33 in round @xmath14 ( that is , in the round @xmath14 computing step of the core @xmath2-set agreement algorithm that terminates round @xmath14 ) is of course put into @xmath954[r]$ ] . _
remote lock learning : _ since all processes exchange their lock histories , @xmath33 may learn about some lock @xmath986 created by process @xmath987 in round @xmath667 from the lock history @xmath988[r']$ ] received from some @xmath93 later on . in this case , @xmath986 is just added to @xmath966[r']$ ] ( [ line : remotehistoryupdate ] ) .
_ local lock learning : _ in order to ensure that the lock histories of all members of a @xmath946-vsrc are eventually consistent , which will finally ensure identical decision values , _ every _ newly learned remote lock @xmath989[r']$ ] obtained in ( ii ) is also added to @xmath954[r]$ ] .
note that the update rules ( i)+(ii ) resemble the ones of vector clocks @xcite .
clearly , @xmath954[r']$ ] will always be accurate for current and past rounds @xmath664 , while @xmath953[r']$ ] may not always be up - to date , i.e. , may lack some locks that are present in @xmath990[r']$ ] .
nevertheless , if @xmath33 and @xmath93 are members of the same @xmath946-vsrc @xmath105 with @xmath991 $ ] , [ def : d - bounded - vsrc ] ensures that @xmath33 and @xmath93 have consistent histories @xmath953[r']$ ] and @xmath992[r']$ ] at latest by ( the end of ) round @xmath993 , for any @xmath994 $ ] .
hence , if @xmath33 creates a new lock @xmath986 when it detects , in its round @xmath14 computing step , that it was part of a @xmath981-vsrc that was stable from @xmath995 to @xmath996 , it is ascertained that any other member @xmath93 will have locally learned the same lock @xmath986 in the same round @xmath14 , provided that the @xmath981-vsrc in fact extended to a @xmath946-vsrc .
the resulting consistency of the histories is finally exploited by the function @xmath997 , which computes ( the value of ) a new local lock ( [ line : setlock ] ) created in round @xmath14 . as its input parameters , it is provided with the members @xmath108 of the detected @xmath981-vsrc and its starting round @xmath998 . @xmath982 first determines a multiset @xmath267 , which contains all locks locally known to the members @xmath999 by round @xmath995 ( [ line : bounded - history ] ) . note that the multiplicity of some lock @xmath1000 in @xmath267 is just the number of members of @xmath108 who got to know @xmath986 by round @xmath995 , which is just @xmath1001 according to [ def : influence - set ] . in order to determine a proper value for the new lock to be computed by @xmath982 , we exploit the fact that @xmath9 ( given in [ ass : stablestream ] ) ensures majority influence according to [ def : majorityinfluence ] : if the set @xmath1002 , containing the most frequent locks in @xmath267 with the same maximal lock creation round , contains a single lock @xmath986 only , its value @xmath1003 is used .
note that the restriction to the maximal lock creation date automatically filters unwanted , outdated locks that have merely been disseminated in preceding @xmath946-vsrcs , see ( 1 ) below .
otherwise , i.e. , if @xmath1002 contains multiple candidate locks , a consistent deterministic choice , namely , the maximum among all lock values in @xmath267 , is used ( [ line : createlockmaximum ] ) . as a consequence , at most @xmath2 different decision values will be generated system - wide . given the various mechanisms employed in our algorithm and their complex interplay , the question about a more light - weight alternative solution that omits
some of these mechanisms might arise .
we will proceed with some informal arguments that support the necessity some of the pillars of our solution , namely , ( 1 ) the preference of most recently created locks in @xmath982 , ( 2 ) the creation of a new lock at every transition to the locked state , and finally ( 3 ) the usage of an a priori unbounded data structure @xmath963 . although these arguments are also `` embedded '' in the correctness proof in the following section
, they do not immediately leap to the eye and are hence provided explicitly here .
the preference of most recently created lock in @xmath982 , which is done by selecting the set @xmath1002 in [ line : hottestmajority ? ] , defeats the inevitable `` amplification '' of the number of processes that got to know some `` old '' lock : all members of a @xmath946-vsrc have finally learned _ all _ `` old '' locks that were only known to _ some _ of its members at the starting round of the vsrc initially . in terms of multiplicity in @xmath267
, this would falsely make any such old lock a preferable alternative to the most recently created lock . instead of creating new locks at every newly detected @xmath981-vsrc
, it might seem sufficient to simply update the creation time of an old lock that ( dominantly ) influences a newly detected vsrc .
this is not the case , however : consider a hypothesized algorithm where new locks are only generated if no suitable old locks can be found in the current history , and assume a run where two vsrcs with vertex sets @xmath1004 and @xmath1005 that are both stable for @xmath981 rounds and two root components @xmath1006 and @xmath1007 that are stable for @xmath946 rounds are formed .
let these vsrcs be such that @xmath223 is formed before @xmath1008 if @xmath1009 and let there be no influence among the processes of @xmath1010 , apart from their influence on each other when they are members of the same vsrc .
first , let the processes of @xmath328 lock on some old lock @xmath1011 .
then , assume that the processes of @xmath329 lock on some lock is known to @xmath1012 and has a more recent creation time than @xmath1011 ] @xmath1014 , a lock not known in @xmath328 .
since @xmath1005 , if @xmath1015 is sufficiently well connected , @xmath803 might lock on @xmath1011 in @xmath1015 , because @xmath1011 is known to both @xmath803 and @xmath1016 while @xmath986 is known merely to @xmath803 at the start of @xmath1015 .
subsequently , this results in the situation in @xmath1017 where there is neither a clear majority ( @xmath1011 and @xmath986 are known to both members of @xmath1017 ) nor a clear most recently adopted lock ( for @xmath803 , it seems that @xmath1011 is the most recent lock , while for @xmath1012 , it seems that @xmath986 is more recent ) .
consequently , in @xmath1017 , it is not clear whether to lock on @xmath1003 or on @xmath1018 .
nevertheless , the processes of @xmath1017 should be able to determine that they must lock on @xmath986 and not on @xmath1011 , since @xmath1019 holds in our example : @xmath1020 and @xmath1021 .
we can therefore conclude that merely adopting old locks is insufficient . since the stabilization round @xmath800 , as implied by [ ass : inter ] ,
may be delayed arbitrarily , an unbounded number of @xmath946-vsrcs can occur before @xmath800 .
since any of those might produce a critical lock , in the sense of exercising a majority influence upon some later @xmath946-vsrc , no such lock can safely be deleted from @xmath963 of any @xmath33 after bounded time . in this final subsection
, we will prove the following [ thm : kset ] : [ thm : kset ] [ alg : ksa ] solves @xmath2-uniform @xmath2-set agreement in a dynamic network under the message adversary @xmath948 , which is the conjunction of [ ass : inter ] and [ ass : stablestream ] .
the proof consists of a sequence of technical lemmas , which will finally allow us to establish all the properties of @xmath2-set agreement given in [ sec : model ] .
first , validity according to [ def : ksa ] is straightforward to see , as only the values of locks are ever considered as decisions ( [ line : setdecision ] ) .
values of locks , on the other hand , are initialized to the initial value of a process ( [ line : inithist ] ) and later on always have values of previous locks assigned to them ( [ line : createlockmajority , line : createlockmaximum ] ) .
note that the claimed @xmath2-uniformity is obvious , as the code of the algorithm does not involve @xmath2 . to establish termination , we start with some simple properties related to setting locks at all members of vertex stable root components . [
lem : minlock ] apart from processes adopting a decision sent by another process , only processes part of a vertex stable root with interval length greater than @xmath28 ( resp .
@xmath1022 ) lock ( resp .
decide ) .
the if - statement in [ line : lock ? ] ( resp .
[ line : decide ? ] ) is evaluated to true only if @xmath613 detects a stable member set @xmath108 in some interval @xmath104 of length @xmath981 ( resp . of length @xmath946 ) or larger , which implies by [ cor : stable2root ] that @xmath105 is indeed a @xmath981-vsrc ( resp .
@xmath946-vsrc ) .
[ lem : maxlock ] all processes part of a vertex stable root @xmath1023}$ ] with interval length greater than @xmath1022 , which did not start already before @xmath112 , lock , i.e. set @xmath1024 , in round @xmath1025 . because @xmath1023}$ ] is @xmath28-bounded by [ ass : inter ] , [ cor : allrootdec ] guarantees that @xmath1026 returns @xmath108 from round @xmath1025 ( of the @xmath2-set - algorithm ) on , and that it can not have done so already in round @xmath1027 .
hence , @xmath1028 in round @xmath1025 , the if - statement in [ line : lock ? ] is entered and @xmath1024 is set in [ line : setlock ] .
[ lem : maxdecide ] all processes part of a vertex stable root @xmath1023}$ ] with interval length greater than @xmath1029 , which did not start already before @xmath112 , have decided by round @xmath1030 .
it follows from [ lem : maxlock ] that all members of the vsrc @xmath1023}$ ] set @xmath1024 in round @xmath1025 . as the vsrc remains stable also in rounds @xmath1031 , [ line : releaselock ] will not be executed in these rounds , thus @xmath1032 remains unchanged .
consequently , due to [ cor : allrootdec ] , the if - statement in [ line : decide ? ] will evaluate to true at the latest in round @xmath1033 , causing all the processes to decide via [ line : setdecision ] by round @xmath1030 as asserted .
[ lem : termination ] the algorithm eventually terminates at all processes . for a contradiction , assume that there is @xmath1034 which has not terminated after the stable interval guaranteed by [ ass : inter ]
this implies that @xmath93 is not part of a root component during this stable interval , because [ lem : maxdecide ] ensures termination by @xmath1035 at the latest for the latter .
hence , @xmath93 did not get a decide message either . from [ def : e - network - vertex - stable - roots ] , it follows that there exists a causal chain of length at most @xmath19 to @xmath93 from some member @xmath33 of a vsrc after its termination .
therefore , it must receive the decide message by @xmath1036 at latest .
although we now know that all members of a vsrc that is vertex stable for at least @xmath1029 rounds will decide , we did not prove anything about their decision values yet . in the sequel , we will prove that they decide on the _ same _ value .
[ lem : d - consistent - history ] given some vsrc @xmath105 with @xmath111 $ ] and @xmath1037 , in all rounds @xmath1038 $ ] it holds that @xmath1039[r ' ] = \bigcup_{r ' { \leqslant}a } \hist_j[j][r']$ ] by the @xmath28-boundedness of @xmath105 , a message from round @xmath112 has reached every member of @xmath108 by round @xmath584 .
moreover , no message sent by a process not in @xmath108 during @xmath104 can reach a member of @xmath108 during @xmath104 because @xmath105 is a root component .
therefore , since @xmath963 is sent by each process @xmath33 in every round ( [ line : sendhistory ] ) and @xmath33 adds only newly learned entries to @xmath963 ( [ line : addnewlock , line : learnlock ] ) , all these updates of @xmath963 during @xmath104 , regarding any round @xmath1040 , occur at the latest in round @xmath584 .
[ lem : consistent - decisions ] all processes of a vsrcs @xmath105 of @xmath941 with @xmath111 $ ] adopt the same lock ( and hence decide the same ) .
such a lock is created by @xmath1041 in round @xmath1025 , when it recognizes @xmath105 as having been vertex - stable for @xmath981 rounds according to [ lem : maxlock ] .
as the lock ( value ) is computed based on @xmath963 present in round @xmath1025 , which is consistent among all vsrc members by [ lem : d - consistent - history ] , the lemma follows .
finally , we show that , given that the system satisfies [ ass : stablestream ] , there will be at most @xmath2 decision values in any run of [ alg : ksa ] , which proves @xmath2-agreement : since there are at most @xmath2 vsrcs of @xmath941 that are not majority - influenced by other vsrcs , it remains to show that any majority - influenced vsrc decides the same as the vsrc it is majority - influenced by . in order to do so
, we will first establish a key property of our central data structure @xmath963 .
[ lem : historyapproximation ] given @xmath1042}$ ] , @xmath1043}$ ] with @xmath1044 and any @xmath1045 .
let @xmath986 be a lock known to all members of @xmath926 by @xmath1046 , i.e. , for all @xmath1047 it holds that , by the end of round @xmath1046 , @xmath1048[r']$ ] . for any process @xmath1049
, it holds that if there exists some @xmath1047 , s.t .
@xmath1050 , then @xmath1051[r']$ ] .
assume there exists a @xmath1047 s.t .
@xmath1050 but @xmath1052[r']$ ] .
the definition of @xmath1050 implies that there exists a causal chain from @xmath33 to @xmath93 that ends before @xmath93 becomes a part of @xmath927 .
since processes send their own history in every round according to [ line : sendhistory ] , every message in this causal chain consisted of a @xmath1053 containing @xmath986 and thus @xmath93 put @xmath986 into its @xmath990[r]$ ] via [ line : remotehistoryupdate ] if @xmath1054[r']$ ] did not already contain @xmath986 . given @xmath1042}\in \mathbb{v}_{2d+1}$ ] and
@xmath1043 } \in \mathbb{v}_{2d+1}$ ] , assume that the processes of @xmath926 created the ( same ) lock @xmath986 when locking .
if @xmath1055 , then the processes of @xmath927 will choose a lock @xmath1011 where @xmath1056 ( and hence decide the same as the processes of @xmath926 ) . from the definition of @xmath1057 ( [ def : majorityinfluence ] ) , it follows that no vsrc @xmath105 of @xmath1058 has a larger influence set on @xmath927 than @xmath926 . by [ lem : minlock ]
, this implies that no lock that was generated by some @xmath105 in @xmath1058 can be known to more members of @xmath927 than the lock @xmath986 generated by @xmath926 .
since process @xmath33 puts only newly learned locks into @xmath963 ( [ line : addnewlock , line : learnlock ] ) , by [ lem : historyapproximation ] , this means that in round @xmath1059 no `` bad '' lock @xmath1060 is present in more elements of @xmath1061[r']$ ] than @xmath986 .
we now show that @xmath1062 for all @xmath1060 occuring in as many elements of @xmath267 as @xmath986 with @xmath1063 .
obviously , the only locks @xmath1060 that could occur in as many elements of @xmath267 as @xmath986 are locks that have been in @xmath963 of some @xmath1047 at the beginning of round @xmath1064 already . since for any such @xmath1060 , @xmath986 was created after @xmath1060 , by [ line : createlockmajority , line : createlockmaximum ] , we have that @xmath1062 , as claimed . because in round @xmath1065 , at all processes @xmath1066 of @xmath927 , [ lem : d - consistent - history ] implies that @xmath1067[r ' ] = \bigcup_{r ' { \leqslant}r_{suc } } \hist_j[i][r']$ ] , when locking in round @xmath1065 according to [ lem : maxlock ] , every @xmath33 of @xmath927 will find @xmath986 as the unique most common lock in the elements of @xmath267 with maximal @xmath984 .
this leads to the evaluation of the if - statement in [ line : hottestmajority ? ] to true and to the creation of a new lock @xmath1011 , where @xmath1068 in [ line : createlockmajority ] , as asserted .
this finally completes the proof of [ thm : kset ] .
we introduced a framework for modeling dynamic networks with directed communication links under generalized message adversaries that focus on vertex - stable root components .
we presented related impossibility results and lower bounds for consensus , as well as a message adversary that is much stronger than the ones known so far for solving consensus , along with a suitable algorithm and its correctness proof .
moreover , we made a significant step towards determining the solvability / impossibility border of general @xmath2-set agreement in our model .
we provided several impossibility results and lower bounds , which also led us to the , to the best of our knowledge , first gracefully degrading consensus/@xmath2-uniform @xmath2-set agreement under fairly strong message adversaries proposed so far . | we study distributed agreement in synchronous directed dynamic networks , where an omniscient message adversary controls the presence / absence of communication links .
we prove that consensus is impossible under a message adversary that guarantees weak connectivity only , and introduce vertex - stable root components ( vsrcs ) as a means for circumventing this impossibility : a @xmath0 message adversary guarantees that , eventually , there is an interval of @xmath1 consecutive rounds where every communication graph contains at most @xmath2 strongly connected components consisting of the same processes ( with possibly varying interconnect topology ) , which have at most out - going links to the remaining processes .
we present a consensus algorithm that works correctly under a @xmath3 message adversary , where @xmath4 is the dynamic causal network diameter .
our algorithm maintains local estimates of the communication graphs , and applies techniques for detecting network stability and univalent system configurations .
several related impossibility results and lower bounds , in particular , that neither a @xmath5 message adversary nor a @xmath6 one allow to solve consensus , reveal that there is not much hope to deal with ( much ) stronger message adversaries here . however , we show that gracefully degrading consensus , which degrades to general @xmath2-set agreement in case of unfavorable network conditions , allows to cope with stronger message adversaries : we provide a @xmath2-uniform @xmath2-set agreement algorithm , where the number of system - wide decision values @xmath2 is not encoded in the algorithm , but rather determined by the actual power of the message adversary in a run : our algorithm guarantees at most @xmath2 decision values under a @xmath7 message adversary , which combines @xmath8 ( with some small value of @xmath1 , ensuring termination ) with some information flow guarantee @xmath9 between certain vsrcs ( ensuring @xmath2-agreement ) . since related impossibility results reveal that a @xmath0 message adversary is too strong for solving @xmath2-set agreement and that some information flow between vsrcs is mandatory for this purpose as well , our results provide a significant step towards the exact solvability / impossibility border of general @xmath2-set agreement in directed dynamic networks |
the quantum monte carlo ( qmc ) technique is becoming an increasingly important tool in the study of condensed matter @xcite .
competitive in accuracy with high - level quantum chemistry methods , it has the enormous advantage of being practicable for large systems containing hundreds of atoms .
the power of qmc in overcoming the deficiencies of density functional theory ( dft ) has been amply demonstrated by recent applications , including the energetics of point defects in silicon @xcite and carbon @xcite , the reconstruction of the si ( 001 ) surface @xcite and its interaction with h@xmath0 @xcite , and the calculation of optical excitation energies @xcite .
nevertheless , the classes of materials to which qmc has been applied have so far been rather limited .
oxide materials are likely to be a very fruitful field for the application of qmc , but in exploring this field it is clearly important to study the capabilities of the techniques for the simplest possible oxides .
we present here qmc calculations on mgo , focusing on its elementary bulk properties , including the equilibrium lattice parameter of the rock - salt structure , the stable structure under ambient conditions , and the pressure of the b1-b2 transition between the rock - salt and cscl structures .
two types of qmc are relevant here . in the first , known as variational monte carlo ( vmc ) , a trial many - electron wave function is constructed as the product of a slater determinant of single - electron orbitals and the so - called jastrow correlation factor .
the latter is parameterised , and the parameter values are obtained using a stochastic optimisation procedure . since vmc by itself is not usually accurate enough , the optimised many - electron wave function produced by vmc is then used in diffusion monte carlo ( dmc ) @xcite , which improves the ground - state estimate by performing an evolution in imaginary time . in principle
, the ground - state energy would be exact , but to overcome the fermion sign problem we use the standard `` fixed - node approximation '' @xcite . in practice ,
only the valence electrons are treated explicitly , the interactions between the valence and core electrons being represented by pseudopotentials .
this introduces additional approximations , including the `` pseudopotential locality approximation '' .
the calculations are performed on periodically repeated cells , and system size errors need to be carefully treated . however , in many cases , the overall errors within qmc can be made much smaller than those within dft , and qmc has already been important in revealing , quantifying and overcoming dft errors in such quantities as defect formation energies , surface reaction energies and energy barriers @xcite .
the three main purposes of this work are : first , to establish the technical feasibility of performing qmc on mgo ; second , to study differences between dft and qmc predictions for the properties of bulk mgo ; and third , to prepare the way for qmc work on more challenging oxides .
as one of the simplest oxides , mgo has often been used as a paradigm for testing theoretical techniques . for qmc , the issue of technical feasibility is a non - trivial one , since the computing effort required to obtain accurate results with dmc depends heavily on the ability of vmc to deliver wave functions which are sufficiently close to the exact ground - state wave function .
if sufficiently accurate trial wave functions can not be obtained , the dmc calculations may even become unstable .
the electronic simplicity of mgo is expected to ease the task of finding good trial wave functions .
the concerns of this work are not purely technical .
the pressure - induced transition from the rock - salt to cscl structure in mgo has been much studied because of its geological interest ( see , e.g. , @xcite and references therein ) .
we also expect that the present work will provide the basis for applying qmc to several important and controversial questions related to ionic materials such as mgo , including the adsorption and dissociation of molecules on their surfaces ( see , e.g. , ref .
@xcite ) , the self - trapping of hole centres and their trapping at defects ( see , e.g. , ref .
@xcite ) , and the conflict between theory and experiment for the slope , @xmath1 , of their melting curves @xcite . beyond this , we hope that the experience gained here will help to prepare the way for the application of qmc to transition - metal oxides such feo , where electron correlation is highly non - trivial .
dmc studies of nio @xcite and mno @xcite have already shown the feasibility of calculations on these materials , but those studies did not include energy - volume curves , and the unit cells used were not large enough to give the accuracy required here . in the following section we summarise the qmc techniques used here . in sec . 3 , we present tests on the magnitude of various errors , including system - size errors , and we report our results for the total energy as a function of volume for the rock - salt and cscl structures , and the transition pressure between the two .
discussion , prospects for future work , and conclusions are presented in sec . 4 .
detailed descriptions of vmc and dmc have already been reported @xcite , so here we only outline rather briefly the main features of the present work . all the qmc calculations presented here were performed using the casino code , the technical details of which are given in ref .
@xcite .
our trial wave functions are of the slater - jastrow type : @xmath2 where @xmath3 and @xmath4 are slater determinants of up- and down - spin single - electron orbitals , and @xmath5 is the so - called jastrow factor , describing the correlation between the electrons .
we use single - electron orbitals obtained from dft calculations .
these single - particle orbitals fix the nodal surface ( the surface in configuration space on which the wave function vanishes and across which it changes sign ) . within this `` fixed - node approximation '' dmc gives a variational upper bound to the ground state energy , rather than the exact ground state energy .
however , because of its large band - gap , we expect that a single slater determinant will give a good description of the nodal surface of mgo .
the function @xmath6 appearing in the jastrow factor is a sum of parametrized one - body and two - body terms @xcite , the latter being designed to satisfy the cusp conditions .
the free parameters in @xmath6 are determined by requiring that the variance of the local energy in vmc be as small as possible @xcite . in the present work ,
the many - body wave function represents explicitly only valence electrons , whose interactions with the ionic cores are represented by pseudopotentials .
we used pseudopotentials generated within hartree - fock ( hf ) theory , but including scalar relativistic effects @xcite .
there is evidence to show that hf theory provides better pseudopotentials for use within qmc than dft @xcite .
the core radii of our pseudopotentials are @xmath7 , @xmath8 , @xmath9 , @xmath10 .
the single - particle orbitals entering the slater determinants of eq .
( [ eq : sj ] ) are the most important component of the wave function .
filippi and fahy @xcite have developed a method for optimising orbitals within vmc , which achieved an energy reduction in diamond from optimising lda orbitals of 0.040(16 ) ev per atom .
et al . _
@xcite found that in bulk silicon using lda orbitals in a vmc calculation gave an energy 0.024(4 ) ev per atom lower than hf orbitals .
these energy changes would be significantly reduced in dmc and it appears that hf and lda orbitals are sufficient for these systems . in this work
we have used lda orbitals obtained using the plane - wave pseudopotential dft code pwscf @xcite .
the basis set used to represent the single - particle orbitals in the qmc calculations themselves is not plane - waves , which become very inefficient for large systems , because the computation cost of evaluating an orbital is proportional to the system size .
instead , we use a b - spline basis , also known as blip functions , consisting of piecewise continuous localised cubic spline functions centred on a regular grid of points . a detailed explanation of blip functions , and
their great advantages for qmc calculations has been reported elsewhere @xcite .
the blip basis is closely related to a plane - wave basis , and for a plane - wave cut - off energy @xmath11 ( @xmath12 is the electron mass ) , there is a natural choice of blip - grid spacing @xmath13 given by @xmath14 . in the same way that plane - wave convergence
is achieved by increasing @xmath15 , blip convergence is achieved by reducing @xmath13 . because of the relationship between plane - waves and blips , it is straightforward to transform the plane - wave coefficients from the pwscf calculations into the blip coefficients needed for the qmc calculations , as explained in more detail in our earlier paper @xcite . for qmc calculations on perfect crystals , there is a useful device which allows a considerable saving of memory .
instead of constructing single - particle orbitals at a given * k*-point ( e.g. , the @xmath16 point ) for the large repeating cell , we construct them for the primitive cell at the corresponding set of * k*-points .
the plane wave coefficients from this calculation are then converted to blip coefficients on points of the blip - grid within the primitive cell . at run - time
, a simple conversion allows these stored coefficients to be used to calculate the required values of the single - particle orbitals at any point in the large repeating cell .
the key point here is that it is unnecessary to store blip coefficients at grid points covering the entire large repeating cell .
an important source of error in qmc calculations using periodic boundary conditions is the limited size of the repeating cell , and the convergence of the qmc energy with respect to the size of the simulation cell must be carefully investigated . to improve this convergence , we follow the common practice @xcite of correcting for this error by using separate dft calculations : we add to the dmc energies the difference @xmath17 between the dft - lda energy calculated with a very large set of * k*-points and the dft - lda energy calculated using the same sampling as in the dmc calculation .
the question of correcting for finite size errors in the coulomb energy has been addressed in recent papers @xcite , and a method known as the model periodic coulomb ( mpc ) interaction has been developed .
the finite size error in the ewald interaction energy arises from the exchange - correlation energy , which can be written as the interaction of the electrons with their exchange - correlation holes .
the interaction with the hole should have the standard @xmath18 form , but within periodic boundary conditions this must be replaced by a periodic interaction .
the mpc interaction maintains the correct ewald interaction for evaluating the hartree energy while for the exchange - correlation energy a periodically repeated potential based on the @xmath18 form is used .
this significantly reduces the finite size errors in the interaction energy , although effects due to the squeezing of the exchange - correlation hole into a finite cell still remain .
we also mention two other technical points relating to size effects .
dmc calculations require the use of real trial wave functions .
however , these can be constructed using single - particle orbitals obtained either from calculations at the @xmath16-point or , in general , * k*-points which correspond to one half of a reciprocal lattice vector of the simulation cell @xcite .
the difference between qmc energies obtained in these two ways can be used as an indication of the system size errors .
the other point is that a given physical crystal structure can be treated using large repeating cells associated with different bravais lattices .
since mgo in both the rock - salt and cscl structures has cubic symmetry , it is most natural to use bravais lattices for the repeating cell having simple - cubic ( sc ) , body - centred - cubic ( bcc ) or face - centred - cubic ( fcc ) symmetries .
we expect that the fcc repeating geometry will give the best convergence with respect to system size , and we shall present results which illustrate this effect .
the number of walkers in dmc simulations is governed by a population control algorithm , which has the purpose of maintaining this number roughly constant . in order to minimise statistical bias in the total energy ,
the calculations need to be run with a large population of walkers . for our dmc calculations
we have used a target population of 640 walkers , which also makes it efficient to run on massively parallel machines , with parallelism achieved by distributing walkers across processors .
for the imaginary time evolution of the walkers we found that a time step of 0.005 a.u .
gave time step errors in the dmc energy of less than 10 mev / atom .
we have found that the quality of our slater - jastrow trial wave function is improved if a large plane - wave cut - off is used in generating the single - particle orbitals , and a correspondingly small blip - grid spacing is used in representing them . in order to investigate this question
, we performed a series of vmc calculations , and calculated the standard error in the energy as a function of plane - wave cut - off .
the blip - grid spacing was taken to be related to the plane - wave cut - off by the ` natural ' formula mentioned in sec .
the calculations were performed without a jastrow factor , because this makes it possible to check some components of the total energy against dft calculations .
these calculations were performed on a 16-atom cell for mgo in the rock - salt structure with a lattice parameter of @xmath19 , which is close to the zero pressure equilibrium lattice parameter .
results are presented in fig .
[ fig : cutoff ] .
we notice that by increasing the pw cutoff from 680 ev to 6800 ev the standard error in the energy is reduced by a factor of @xmath20 .
this means that qmc runs performed using the trial wave function obtained with the largest cutoff can be 4 times shorter , in order to achieve the same statistical accuracy .
more importantly , we found that using a very large pw cutoff was essential for having stable dmc runs .
we were unable to perform any useful dmc simulation with cutoff energies less than 2712 ev .
we have made extensive tests on system size effects .
we divide our discussion of these tests into two parts : first , tests on the rock - salt structure at low pressures , which are relevant to the equilibrium properties of this structure ; second , tests on both the rock - salt and cscl structures at high pressures , which are relevant to the determination of the transition pressure .
as we shall see , these two sets of tests involve somewhat different questions . in fig .
[ fig : size1 ] , we show the vmc energy per atom of mgo as a function of the number of atoms in the repeating cell , using both the standard ewald interaction and the mpc interaction . for these calculations we used a plane - wave cut - off of 6800 ev and the associated natural blip - grid spacing for the description of the single - particle orbitals .
we note that the mpc results appear to converge considerably faster than the ewald ones , and that for a system of 54 atoms the mpc energy is already converged to better than @xmath21 50 mev / atom .
we therefore decided to use this cell size to evaluate the energy - volume curve presented in the following section .
the results we report here for the cscl structure were performed with the standard ewald method rather than the mpc method . in order to check system - size errors thoroughly , we found it essential to perform tests on large systems of up to over 600 atoms .
we made extensive tests on the cscl structure to compare sc and fcc repeating geometries and to examine the effect of using different sampling wavevectors .
the sampling wavevectors we used are the @xmath16-point @xmath22 , and the wavevector @xmath23 , where @xmath24 specifies the dimension of the repeated cell .
( in more detail , @xmath24 is the length such that with sc repeating geometry the primitive translation vectors are @xmath25 , @xmath26 and @xmath27 , while with fcc geometry they are @xmath28 , @xmath29 and @xmath30 . )
since the wavevector @xmath23 lies on the zone boundary of the brillouin zone associated with the periodically repeated supercell , we refer to sampling using this wavevector as `` zone - boundary '' sampling .
these tests were performed at the volume 8.77 @xmath31/atom , which is close to the zero pressure equilibrium volume .
the tests were all performed using vmc , and we used the jastrow factor optimised using a 16 atom cell for all system sizes , because re - optimising the jastrow factor introduces small `` jumps '' in the energy .
since we needed to go to large system sizes , we decided to reduce the plane - wave cut - off from 6800 ev to 2712 ev , because this gave a considerable reduction in the memory required ; with this lower cut - off , the standard error in the energy fluctuations is only slightly larger ( see fig .
[ fig : cutoff ] ) , and dmc calculations are still stable .
results of these tests are shown in fig .
[ fig : size2 ] .
we see that convergence to within less than 50 mev / atom is obtained for systems larger than 108/128 atom .
we also note that convergence is better with the fcc than with the sc repeating geometry , as expected , and that there is little to choose between @xmath16-point and `` zone - boundary '' sampling .
we have therefore performed all further calculations using fcc geometry and @xmath16-point sampling . since calculation of the transition pressure requires qmc calculations for the two structures at high pressures , we have performed further vmc calculations at volumes 4.23 and 4.41 @xmath31/atom for the cscl and the rock - salt structures respectively , close to the transition , using the ewald interaction .
results of these tests are reported in table [ tab : size ] .
we see that for calculations on the cscl structure , using a cell containing 108 atoms with fcc repeating geometry , the error is about 50 mev / atom .
the error is approximately the same for the rock - salt structure with a 128-atom cell , so that the error in the energy difference between the two structures is less than our target accuracy of 30 ev / atom . in fig .
[ fig : nacl ] we display dmc energies as a function of volume for mgo in the nacl structure .
the length of these simulations was typically 6000 steps , resulting in a statistical error bar of less than 10 mev / atom .
these energy points were then used to fit the parameters of the birch - murnaghan equation of state @xcite : @xmath32 \ ; , \ ] ] where @xmath33 , @xmath34 is the equilibrium volume , @xmath35 is the zero - pressure bulk modulus , @xmath36 is its derivative with respect to pressure at zero pressure , and @xmath37 is the energy at the minimum .
the fitted curve is also reported on the same figure .
the values of the fitted parameters are reported in table [ tab : murna_parameters ] together with other theoretical results and experimental data .
a comparison of the qmc results with experimental values shows that our calculated lattice parameter of @xmath38 is smaller than the measured value of @xmath39 @xcite , and our bulk modulus @xmath40 gpa is greater than the measured value @xmath41 gpa @xcite .
however , two kinds of corrections need to be made .
it is known from earlier dft calculations @xcite that room temperature thermal pressure due to lattice vibrations increases @xmath42 by @xmath43 and decreases @xmath35 by 10 gpa .
we should also correct for pseudopotential errors . to estimate these ,
we have compared the predictions of pseudopotential and all - electron hf calculations using the crystal @xcite code ( see table 1 ) .
this shows that the pseudopotentials we have used underestimate @xmath42 by 0.10 and overestimate @xmath35 by 15 gpa . combining these two corrections , our revised qmc values are @xmath44 and @xmath45 gpa , which are very close to the experimental values . in fig .
[ fig : transition ] we report the dmc energy for mgo in the nacl and the cscl structures evaluated for volumes corresponding to roughly one half of the zero pressure equilibrium volume , which is the region of volumes in which the transition occurs .
these energy points have also been fitted to the birch - murnaghan equation of state , which we have then used to compute the enthalpies of the two structures , which are displayed in the upper part of fig .
[ fig : enth ] , from which we infer a transition pressure of about 597 gpa .
we note that the slopes of the two curves are very similar , and that an error of about 1 mev / atom in the relative enthalpy results in an error in the transition pressure of about 1 gpa .
we do not expect our dmc results to be more accurate than about 20 mev / atom , so our computed transition pressure should be considered to have an error bar of about 20 gpa . for comparison , we also report in the lower part of fig .
[ fig : enth ] the enthalpy evaluated with dft - lda and the same pseudopotentials , from which we deduce a transition pressure of 569 gpa .
an important conclusion from this work is that it is technically feasible to carry out diffusion monte carlo calculations on mgo at the level of accuracy required to compute quantities such as the equilibrium lattice constant , the bulk modulus , and the b1-b2 transition pressure . as we mentioned in the introduction , this is a non - trivial conclusion , because dmc calculations succeed in practice only if the many - body trial wave function is sufficiently close to the true ground state wave function . even though mgo should be a favourable oxide in this respect , it was still essential to pay careful attention to the accurate representation of the single - electron orbitals in order to bring statistical fluctuations under control .
we have shown that , provided corrections are made for thermal effects and errors due to imperfections of our pseudopotentials , the qmc predictions of lattice parameter @xmath42 and bulk modulus @xmath35 agree with experimental values to within @xmath46% for @xmath42 and to within experimental error ( @xmath20% ) for @xmath35 . in future work
, there should be scope for further improvement in the pseudopotentials .
it is interesting to note that our qmc prediction for @xmath42 is almost identical to the hf prediction with the same pseudopotentials .
this might seem to suggest that correlation effects are negligible in mgo . however , this is certainly not the case .
the correlation energy is at least the difference between the hf and the qmc total energies .
we find that for the rock - salt structure near the equilibrium volume , this difference is @xmath47 ev / atom .
the close agreement between the hf and qmc lattice parameters therefore indicates that this rather large correlation energy depends only weakly on volume .
our qmc value of @xmath48 gpa for the b1-b2 transition pressure supports the most recent dft predictions of a pressure in the region of @xmath49 gpa , which is beyond the region of geophysical interest ( the pressure at the core - mantle boundary of the earth is 135 gpa ) .
the detailed value we have found may suffer from a pseudopotential error , but at present we are unable to quantify this . with these encouraging results for mgo
, we believe that there are now good prospects for extending qmc methods to studying the more challenging problems involving mgo mentioned in the introduction , and to studies of transition metal oxides . lda ( and generalised gradient approximation ) calculations are unsuitable for transition metal oxides because they lead to an incorrect filling of the energy levels .
unrestricted hf and b3lyp orbitals have already been used with some success in dmc studies of transition metal oxides @xcite , and one might also consider using lda+u or sic ( self - interaction corrected ) dft orbitals .
towler and needs @xcite found that unrestricted hf orbitals gave a lower energy than b3lyp orbitals for nio .
transition metal oxides are clearly a case where it would be useful to optimise the orbitals in the presence of the jastrow factor .
.dft and hf values for the lattice constant and bulk modulus of the nacl phase of mgo , and the equilibrium pressure for the b1-b2 transition .
see the original references for details . [
cols="<,^,^,^",options="header " , ] | we report diffusion monte carlo ( dmc ) calculations on mgo in the rock - salt and cscl structures .
the calculations are based on hartree - fock pseudopotentials , with the single - particle orbitals entering the correlated wave function being represented by a systematically convergeable cubic - spline basis .
systematic tests are presented on system - size errors using periodically repeating cells of up to over 600 atoms .
the equilibrium lattice parameter of the rock - salt structure obtained within dmc is almost identical to the hartree - fock result , which is close to the experimental value . the dmc result for
the bulk modulus is also in good agreement with the experimental value .
the b1-b2 transition pressure ( between the rock - salt and cscl structures ) is predicted to be just below 600 gpa , which is beyond the experimentally accessible range , in accord with other predictions based on hartree - fock and density functional theories . |
there is no statistically significant large scale anisotropy in the observed arrival distribution of ultra - high energy cosmic rays ( uhecrs ) above @xmath2 ev @xcite .
this may imply an extragalactic origin of cosmic rays above @xmath2 ev , combined with the change of spectral slope of the observed energy spectrum at @xmath3 ev @xcite .
another important feature of the uhecr arrival distribution is the small scale clusterings of the arrival directions @xcite .
the current agasa data set of 57 events above 4 @xmath4 ev contains four doublets and one triplet within a separation angle of 2.5@xmath5 .
chance probability to observe such clusters under an isotropic distribution is only about 1 @xmath6 @xcite . on the other hand
, the cosmic - ray energy spectrum does not show the gzk cutoff @xcite because of photopion production with the photons of the cosmic microwave background ( cmb ) and extends above @xmath7 ev @xcite .
the discrepancy between the agasa and the high resolution fly s eye ( hires ; * ? ? ?
* ) , which reports the cosmic ray flux with the gzk cut - off around @xmath7 ev @xcite , remains to be one of the major open question in astroparticle physics .
this issue is left for future investigation by new large - aperture detectors under development , such as south and north auger project @xcite , the euso @xcite , and the owl @xcite experiments . in our recent work (
* hereafter paper i ) , we perform numerical simulations for propagation of uhe protons in intergalactic space , and examine whether the present agasa observation above @xmath8 ev can be explained by a bottom - up scenario in which the source distribution of uhecrs is proportional to that of galaxies .
we use the optical redshift survey ( ors ; * ? ? ?
* ) to construct realistic source models of uhecrs . in paper
i , we calculate both the energy spectrum and arrival directions of uhe protons , and compare the results with the agasa observation above @xmath8 ev .
we find that the large - scale isotropy and the small - scale anisotropy of the uhecr arrival distribution observed by the agasa can be reproduced when @xmath9 of the ors sample more luminous than @xmath10 mag are selected as uhecr sources , in the case of weak extragalactic magnetic field ( egmf @xmath11 ng ) . in terms of the source number density , this constraint corresponds to @xmath12 mpc@xmath13 .
the small scale anisotropy can not be well reproduced in the case of strong egmf ( @xmath14 ng ) , because the correlation at small scale between events which originate from a single source is eliminated , or the correlation continues to larger angle scale , due to large deflection when uhecrs propagate in the egmf from sources to the earth . although @xcite and @xcite conclude that the expected small - scale anisotropy and large - scale isotropy for local enhancement of uhecr sources in the lsc in the presence of the strong egmf ( @xmath15 g ) are in marginal agreement with the agasa , the consistency is somewhat worse than that predicted by our scenario for @xmath16 ng .
of course , we can not draw any firm conclusion about the strength of the egmf , considering the current limited amount of data .
however , we assume extremely weak egmf throughout the paper . if local enhancement of uhecr sources in the lsc @xcite is disfavored from the observations , there is no way that explains the observed extension of the cosmic - ray spectrum beyond the gzk cutoff .
our conclusion in paper i is that a large fraction of cosmic rays above @xmath7 ev observed by the agasa experiment might originate in the top - down scenarios , or that the energy spectrum measured by the hires experiment might be better .
as mentioned above , we obtain the constraint on the source number density as @xmath12 mpc@xmath13 by comparing our model prediction with the agasa data only above @xmath8 ev .
it is very important to examine whether our source model can explain the agasa data including lower energy @xmath17 one . on the other hand ,
the arrival directions of uhecrs above @xmath2 ev are modified by the galactic magnetic field ( gmf ) by a few @xmath18 degrees . in order to accurately calculate the expected uhecr arrival distribution and compare with the observations , the effect of the gmf should be taken into account .
the first step of the studies of uhecr propagation in the gmf is found in @xcite .
@xcite calculate the expected arrival distribution of uhecrs above @xmath19 ev for several source location scenarios .
they perform numerical simulations of uhecr propagation in the galaxy injected from sources toward the earth
. the radius of the earth ( detector ) must be so small that the unavoidable smearing in arrival angle is kept smaller than the accuracy of arrival direction determination @xmath20 a few degree @xcite . in this case
, the number fraction of injected uhecrs arriving at the earth is very small .
this requires a large number of particles to be propagated , which takes enormous cpu time . in this paper
, we present a new method for calculating uhecr arrival distribution which can be applied to several source location scenarios including modifications by the gmf .
we numerically calculate the propagation of anti - protons from the earth toward the outside of the galaxy ( in this study , we set a sphere centered around the galactic center with radius @xmath2140 kpc as the boundary condition ) , including the effects of lorentz force due to the gmf . the anti - protons
are ejected isotropically from the earth . by this calculation , we can obtain the trajectories and the sky map of position of anti - protons that have reached to the boundary at radius @xmath21 40 kpc .
next , we regard the obtained trajectories as the ones of protons from the outside of the galaxy toward the earth . also , we regard the obtained sky map of position of anti - proton at the boundary as relative probability distribution ( per steradian ) for protons to be able to reach to the earth for the case in which the flux of the uhe protons from the extra - galactic region is isotropic ( in this study , this flux corresponds to the one at the boundary @xmath22 kpc ) . this treatment is supported by the liouville s theorem . when the flux of the uhe protons at the boundary is anisotropic ( e.g. , the source distribution is not isotropic ) , this effect can be included by multiplying this effect ( that is , by multiplying the probability density of arrival direction of uhe protons from the extra - galactic region at the boundary ) to the obtained relative probability density distribution mentioned above . by adopting this new method , we can consider only the trajectories of protons which arrive to the earth , which , of course , helps us to save the cpu time efficiently and makes calculation of propagation of crs even with low energies ( @xmath3 ev ) possible within a reasonable time .
we also consider the energy loss processes of uhe protons in the intergalactic space , which is not taken into account by @xcite . with this method
, we calculate the uhecr arrival distribution above @xmath2 ev for our source scenario which can explain the current agasa observation above @xmath23 ev . using the harmonic amplitude and the two point correlation function as statistical quantities , we compare our model prediction with the agasa observation .
we also demonstrate the arrival distribution of uhecrs with the event number expected by future experiments such as south and north auger project @xcite , the euso @xcite , and the owl @xcite experiments . in section [ gmf ] ,
we introduce the gmf model .
we explain the method for calculating uhecr arrival distribution in section [ method ] .
results are shown in section [ result ] . in section [ summary ] , we summarize the main results .
in this study , we adopt the gmf model used by @xcite , which is composed of the spiral and the dipole field . in the following
, we briefly introduce this gmf model .
faraday rotation measurements indicate that the gmf in the disk of the galaxy has a spiral structure with field reversals at the optical galactic arms @xcite .
we use a bisymmetric spiral field ( bss ) model , which is favored from recent work @xcite .
the solar system is located at a distance @xmath24 kpc from the center of the galaxy in the galactic plane . the local regular magnetic field in the vicinity of the solar system
is assumed to be @xmath25 in the direction @xmath26 where the pitch angle is @xmath27 @xcite . in the polar
coordinates @xmath28 , the strength of the spiral field in the galactic plane is given by @xmath29 where @xmath30 g , @xmath31 kpc and @xmath32 .
the field decreases with galactocentric distance as @xmath33 and it is zero for @xmath34 kpc . in the region around the galactic center ( @xmath35 kpc ) the field is highly uncertain , and thus assumed to be constant and equal to its value at @xmath36 kpc . the spiral field strengths above and below the galactic plane are taken to decrease exponentially with two scale heights @xcite @xmath37 where the factor @xmath38 makes the field continuous in @xmath39 .
the bss spiral field we use is of even parity , that is , the field direction is preserved at disk crossing .
observations show that the field in the galactic halo is much weaker than that in the disk . in this work
we assume that the regular field corresponds to a a0 dipole field as suggested in @xcite .
in spherical coordinates @xmath40 the @xmath41 components of the halo field are given by : @xmath42 where @xmath43 is the magnetic moment of the galactic dipole .
the dipole field is very strong in the central region of the galaxy , but is only 0.3 @xmath44 g in the vicinity of the solar system , directed toward the north galactic pole .
there is a significant turbulent component , @xmath45 , of the galactic magnetic field .
its field strength is difficult to measure and results found in literature are in the range of @xmath46 @xcite .
however , we neglect the random field throughout the paper , in order to make easy to see the effects of the regular field , such as the arrangement of the clustered event in the order of their energies ( section [ arrival_future ] )
. possible dependence of the results on the random field is discussed in the section [ statistics_arrival ] .
the energy spectrum of uhecrs injected at extragalactic sources is modified by the energy loss processes when they propagate in the intergalactic space .
this subsection provides the method of monte carlo simulations for propagation of uhe protons in intergalactic space .
we assume that uhecrs are protons injected with a power law spectrum within the range of ( @xmath2 - @xmath47)ev .
10000 protons are injected in each of 31 energy bins , that is , 10 bins per decade of energy .
then , uhe protons are propagated including the energy loss processes ( explained below ) over @xmath48 gpc for @xmath49 gyr .
we take a power law index as 2.6 in order to fit the calculated energy spectrum to the one observed by the agasa @xcite .
uhe protons below @xmath50 ev lose their energies mainly by pair creations and adiabatic energy losses , and above it by photopion production @xcite in collisions with photons of the cmb .
we treat the adiabatic loss as a continuous loss process .
we calculate the redshift @xmath39 of source at a given distance using the cosmological parameters @xmath51 km s@xmath52 mpc@xmath52 , @xmath53 , and @xmath54 .
similarly , the pair production can be treated as a continuous loss process considering its small inelasticity ( @xmath55 ) .
we adopt the analytical fit functions given by @xcite to calculate the energy loss rate for the pair production on isotropic photons .
the same approach has been adopted in our previous studies ( paper i , * ? ? ?
* ; * ? ? ?
on the other hand , protons lose a large fraction of their energy in the photopion production .
for this reason , its treatment is very important .
we use the interaction length and the energy distribution of final protons as a function of initial proton energy which is calculated by simulating the photopion production with the event generator sophia @xcite . in this study
, we neglect the effect of the egmf because of the following two reasons .
first , numerical simulations of uhecr propagation in the egmf including lower energy ( @xmath3 ev ) ones take a long cpu time .
second , we show in our previous study that small scale clustering can be well reproduced in the case of weak egmf @xmath56 ( paper i ) . @xcite and @xcite show that the expected small scale anisotropy for local enhancement scenario of uhecr sources in the presence of strong egmf ( @xmath57 g ) in the local super cluster is marginally consistent with the agasa observation .
however , the consistency of small - scale anisotropy is somewhat worse than that predicted by our scenario in the case of weak egmf ( paper i ) .
although we can not draw any firm conclusion because of the limited amount of data , we assume extremely weak egmf throughout the paper . in this study
, we apply the method for calculating the uhecr arrival distribution with modifications by the gmf ( section [ calc_arrival ] ) to our source location scenario , which is constructed by using the ors @xcite galaxy catalog . as mentioned in section [ intro ]
, we show in paper i that the arrival distribution of uhecrs observed by the agasa can be reproduced when @xmath9 of the ors galaxies more luminous than @xmath58 is selected as uhecr sources .
we consider only this source model throughout the paper .
it is unknown how much an ultimate uhecr source contribute to the observed cosmic ray flux . in paper
i , we thus consider the two cases in which ( 1 ) all galaxies inject the same amount of cosmic rays , or ( 2 ) they inject cosmic rays proportionally to their absolute luminosity .
however , we find that the results in the two cases do not differ from each other , as far as we focus on the luminous galaxies as uhecr sources .
accordingly , we restrict ourselves to the case that all galaxies inject the same amount of cosmic rays . in order to calculate the energy spectrum and the distribution of arrival directions of uhecrs realistically
, there are two key elements of the galaxy sample to be corrected .
first , galaxies in a given magnitude - limited sample are biased tracers of matter distribution because of the flux limit @xcite .
although the sample of galaxies more luminous than @xmath10 mag is complete within 80 @xmath59 mpc ( where @xmath60 is the hubble constant divided by 100 km s@xmath52 and we use @xmath61 ) , it does not contain galaxies outside it for the reason of the selection effect .
we distribute sources of uhecrs outside 80 @xmath59 mpc homogeneously .
their number density is set to be equal to that inside 80 @xmath59 mpc .
we do not take into account luminosity evolution for simplicity .
second , our ors sample does not include galaxies in the zone of avoidance ( @xmath62 ) . in the same way
, we distribute uhecr sources in this region homogeneously , and calculate its number density from the number of galaxies in the observed region . in this subsection ,
we present the method of calculation of uhecr arrival distribution with modifications by the gmf .
we start by injecting anti - protons from the earth isotropically , and follow each trajectory until \1 .
anti - proton reaches a sphere of radius @xmath63 kpc centered at the galactic center , or \2 . the total path length traveled by anti - proton is larger than @xmath64 kpc , by integrating the equations of motion in the magnetic field
it is noted that we regard these anti - protons as protons injected from the outside of the galaxy toward the earth .
the number of propagated anti - proton is 2,000,000 .
we have checked that the number of trajectories which are stopped by the limit ( 2 ) is smaller than @xmath65% of the total number .
the energy loss of protons can be neglected for these distances .
accordingly , we inject anti - protons with injection spectrum similar to the observed one @xmath66 .
( note that this is not the energy spectrum injected at extragalactic sources . )
the result of the velocity directions of anti - protons at the sphere of radius @xmath63 kpc is shown in the right panel of figure [ event ] in the galactic coordinate . from liouville
s theorem , if the cosmic - ray flux outside the galaxy is isotropic , one expects an isotropic flux at the earth even in the presence of the gmf .
this theorem is confirmed by numerical calculations shown in figure 6 of @xcite , which is the same figure as our figure [ event ] except for threshold energy .
thus , the mapping of the velocity directions in the right panel of figure [ event ] corresponds to the sources which actually give rise to the flux at the earth in the case that the sources ( including ones which do not actually give rise to the flux at the earth ) are distributed uniformly .
we calculate the uhecr arrival distribution for our source scenario using the numerical data of the propagation of uhe anti - protons in the galaxy .
detailed method is as follows . at first
, we divide the sky into a number of bins with the same solid angle .
the number of bins is taken to be @xmath67 .
we then distribute all the trajectories into each bin according to their directions of velocities ( source directions ) at the sphere of radius @xmath63 kpc .
finally , we randomly select trajectories from each bin with probability @xmath68 defined as @xmath69 here subscripts j and k distinguish each cell of the sky , @xmath70 is distance of each galaxy within the cell of ( j , k ) , and the summation runs over all of the galaxies within it .
@xmath71 is the energy of proton , and @xmath72 is the energy spectrum of protons at our galaxy injected at a source of distance @xmath70 .
the normalization of @xmath68 is determined so as to set the total number of events equal to a given number , for example , the event number of the current agasa data . when @xmath73 , we newly generate events with number of @xmath74 , where @xmath75 is the number of trajectories within the sky cell of @xmath76 .
the arrival angle of newly generated proton ( equivalently , injection angle of anti - proton ) at the earth is calculated by adding a normally distributed deviate with zero mean and variance equal to the experimental resolution @xmath77 @xmath78 for @xmath79 ev @xmath80 to the original arrival angle .
we perform this event generation 20 times in order to calculate the averages and variances , due to the finite number of the simulated events , of the statistical quantities ( section [ statistics ] ) . in this subsection
, we explain the two statistical quantities , the harmonics analysis for large scale anisotropy @xcite , the two point correlation function for small scale anisotropy .
the harmonic analysis to the right ascension distribution of events is the conventional method to search for global anisotropy of cosmic ray arrival distribution . for a ground - based detector like the agasa
, the almost uniform observation in right ascension is expected . the @xmath81-th harmonic amplitude @xmath82 is determined by fitting the distribution to a sine wave with period @xmath83 . for a sample of @xmath84 measurements of phase , @xmath85 , @xmath86 , @xmath87 , @xmath88 ( 0 @xmath89 )
, it is expressed as @xmath90 where , @xmath91 , @xmath92 .
we calculate the harmonic amplitude for @xmath93 from a set of events generated by the method explained in the section [ calc_arrival ] .
if events with total number @xmath84 are uniformly distributed in right ascension , the chance probability of observing the amplitude @xmath94 is given by , @xmath95 where @xmath96 the current agasa 775 events above @xmath2 ev is consistent with isotropic source distribution within 90 @xmath6 confidence level @xcite .
we therefore compare the harmonic amplitude for @xmath97 with the model prediction .
the two point correlation function @xmath98 contains information on the small scale anisotropy .
we start from a set of generated events . for each event , we divide the sphere into concentric bins of angular size @xmath99 , and count the number of events falling into each bin .
we then divide it by the solid angle of the corresponding bin , that is , @xmath100 , \label{eqn100}\end{aligned}\ ] ] where @xmath101 denotes the separation angle of the two events .
@xmath99 is taken to be @xmath102 in this analysis .
the agasa data shows correlation at small angle @xmath103 with 2.3 ( 4.6 ) @xmath104 significance of deviation from an isotropic distribution for @xmath79 ev @xmath105 @xcite .
in this subsection , we present the results of the arrival distribution of uhecrs above @xmath2 ev , using the method explained in the section [ calc_arrival ]
. at first , figure [ source ] shows the distribution of the sources for a specific source selection when @xmath9 of the ors galaxies more luminous than @xmath106 is randomly selected as uhecr sources in the galactic coordinate .
we show only the sources within @xmath107 mpc from us for clarity , as circles of radius inversely proportional to their distances . it is noted that the sources outside 113@xmath108 mpc are randomly distributed because the ors sample does not contain any galaxy outside it .
we show in figure [ esp ] the expected energy spectrum for the source model of figure [ source ] .
the injection spectrum is set to be @xmath109 .
the contributions from sources at different distances are also shown .
we also show the observed cosmic - ray spectrum by the agasa experiments @xcite .
the resultant spectrum is in good agreement with the one observed by the agasa , except for @xmath110 ev .
as mentioned above , we conclude in paper i that a large fraction of cosmic rays above @xmath7 ev observed by the agasa experiment might originate in the top - down scenarios .
accordingly , we consider only cosmic rays with @xmath111 ev throughout the paper . as mentioned above
, @xcite does not take the energy loss processes in the intergalactic space into account .
thus , they can not include the effects of difference between resultant energy spectra injected at different distances into numerical calculations . in our calculations , however , sources at larger distance mainly contribute to the cosmic ray flux at lower energies as is evident from figure [ esp ] .
this enables us to calculate the arrival distribution of uhecrs under more realistic situations .
given the source distribution and the resultant energy spectrum as a function of the source distance , we can calculate the right hand side of eq .
[ def_pselec ] .
then we perform the selection of trajectories according to the probability @xmath112 , as explained in the section [ calc_arrival ] .
one realization of the event generations is shown in figure [ selected_event ] .
the events are shown by color according to their energies .
this figure corresponds to figure [ event ] .
that is , the injection directions of anti - protons at the earth ( figure [ event ] , left ) corresponds to the arrival directions of protons ( figure [ selected_event ] , left ) .
similarly , the arrival directions of anti - protons at the sphere of galactocentric radius of @xmath63 kpc ( figure [ event ] , right ) does to the directions of the sources which actually give rise to the cosmic - ray flux ( figure [ selected_event ] , right ) . for the source model of figure
[ source ] , the nearest source is located at @xmath113 and 64 mpc from us . a number of the simulated events are clustered at this direction as seen in figure [ selected_event ] .
furthermore , these events are aligned in the sky according to the order of their energies , reflecting the direction of the gmf at this direction . as we will show in the section [ arrival_future ] , this interesting feature of the uhecr arrival distribution becomes evident with increasing amount of the event number . in this subsection , we show the results of the statistical quantities on the uhecr arrival distribution above @xmath2 ev . in the last section ,
we showed the results for a specific source scenario when @xmath9 of the ors galaxies more luminous than @xmath106 is randomly selected as uhecr sources .
however , the statistical quantities presented in this section are calculated with not only the statistical error but also the variation between different selections of source from our ors sample .
the upper panels of figure [ amp ] shows the first and the second harmonics predicted by our source model as a function of the cosmic - ray energies for @xmath114 g , where @xmath115 is the strength of the gmf in the vicinity of the solar system .
it is noted that we calculate the harmonic amplitudes for the simulated events within only @xmath116 in order to compare with agasa data .
the errorbars represent the statistical fluctuations due to the finite number of the simulated events , which is set to be equal to that observed by the agasa ( 775 events for @xmath79 ev ) .
the event selections are performed @xmath117 times .
the shaded regions represent 1 @xmath104 total error due to not only the statistical error but also the source selections from our ors sample .
the random source selections are performed @xmath118 times .
the region below the histogram is expected values for the statistical fluctuation of isotropic source distribution with the chance probability larger than 10@xmath6 .
it is clear that our source model predicts the large - scale isotropy fully consistent with that expected by uniform source distribution within 1 @xmath104 total error ( statistical one plus source selection ) .
we have checked that 27 source distributions out of 100 predict the sufficient large - scale isotropy within 1 @xmath104 statistical error . in order to investigate the effects of the gmf on the large - scale anisotropy
, we also calculate the harmonic amplitude for the case of @xmath119 g. for @xmath119 g , the predicted arrival distribution is relatively more isotropic than that for @xmath114 g. we also note that this tendency can be seen at lower energies ( @xmath120 ev ) .
because the deflection angle of cosmic rays with such energies by the gmf is about @xmath121 , the harmonic amplitude of arrival distribution of uhecrs can be affected by anisotropy of the event distributions which is caused by the events aligned according to the order of their energies , reflecting the direction of the gmf . in figure
[ 2p ] , we show two point correlation function predicted by our source model for @xmath122 ( left ) and @xmath79 ev ( right ) . it is noted that we calculate two point correlation function for the simulated events within only @xmath116 in order to compare with agasa data .
the errorbars represent the statistical fluctuations due to the finite number of the simulated events , which is set to be equal to that observed by the agasa ( 775 events for @xmath79 ev ) .
the shaded regions represent 1 @xmath104 total error due to not only the statistical error but also the source selections from our ors sample .
the event selections and the random source selections are performed @xmath117 times and @xmath118 times , respectively .
the event numbers shown in this figures are averaged over all trials of the event selections and the source selections .
the histograms represent the agasa data . however , the agasa data for @xmath79 ev are fitted to the result of our calculation at larger angle ( @xmath123 ) , since we can not know the normalization of the agasa data with this energy .
clearly visible is that large peak at small angle scale is too strong compared with the agasa observation @xcite .
we have checked that when extremely nearby sources are selected by accident , predicted small - scale anisotropy becomes to be very strong .
this is the reason for too large peak at small angle scale in figure [ 2p ] , where the averages and the variances are calculated including such source distributions .
provided that extremely nearby sources are selected by accident , not only the small - scale anisotropy but also the large - scale isotropy is inconsistent with the agasa observation .
accordingly , we calculate two point correlation function only for the source distributions which predict the large - scale isotropy consistent with uniform source distribution within 1 @xmath104 statistical error .
the number of such source distributions is 27 out of all the 100 source selections .
the result is shown in figure [ 2p_1sig ] .
two point correlation function in figure [ 2p_1sig ] exhibits a structure that is similar to that seen in the agasa data , that is , large peak at small angle scale followed by a tail at large angles . for @xmath79
ev , a peak at small angles is somewhat weaker than that for @xmath122 ev because of the large deflection by the gmf .
this feature is also seen in agasa data @xcite . from this result , it can be understood that the source distributions which predict sufficient large - scale isotropy also predict the small - scale anisotropy that is similar to that seen in the agasa data .
however , we should note that the peak at small angle scale is still relatively strong compared with the agasa .
there may be two possible explanations for this fact .
first , we neglect the effects of the extragalactic magnetic field in this study , in order to save the cpu time . if we can include this effect by future studies , strong correlation at small scale will be reduced because of the deflection of uhecrs in the intergalactic space .
second , we also neglect the random component of the gmf in order to make easy to see the effect of the regular field .
this may also relax the large peak at small angle scale , provided that there is the random component with same level of strength with the regular component .
these issues are left for future investigations . in this subsection
, we demonstrate the results of the uhecr arrival distribution above @xmath79 ev with the event number expected by future experiments , such as auger , euso and owl .
the results for the source model of figure [ source ] are shown in figure [ selected_many ] .
the events are shown by color according to their energies .
it is noted that the expected event rate by the auger experiment is @xmath124 per year above @xmath2 ev .
these results are extended versions of our previous study @xcite , where we predict the uhecr arrival distribution above @xmath8 ev without modifications by the gmf .
remarkable feature is the arrangement of clustered events at the directions of nearby sources ( see figure [ source ] ) .
the events are aligned according to the order of their energies reflecting the direction of the gmf .
we will be able to obtain some kind of information about the gmf and the chemical composition of uhecrs .
in forthcoming work , we plan studies about new statistical quantities which allow us to obtain such invaluable information .
in this paper , we presented a new method for calculating the arrival distribution of uhecrs , which can be applied to several source location scenarios , including modifications by the gmf .
we performed numerical simulations of uhe anti - protons , which are injected isotropically at the earth , in the galaxy and recorded the directions of velocities at the earth and outside the galaxy for all of the trajectories .
it is noted that we regard these anti - protons as protons injected from the outside of the galaxy toward the earth .
we then selected some trajectories so that the resultant mapping of the velocity directions outside the galaxy of the selected trajectories corresponds to our source location scenario , applying liouville s theorem .
there are two points of our improvement over the work of @xcite .
first , we calculated only the trajectories actually reaching the detectors by propagating anti - protons backwards from earth , instead of propagating protons from the source and selecting those reaching the earth .
this helps us to save the cpu time efficiently and makes calculations of propagation of cosmic rays even with lower energies ( @xmath3 ev ) possible within a reasonable time .
second , we considered energy loss processes of uhe protons in the intergalactic space , which is not taken into account in @xcite .
this enables us to include the effects of difference between resultant energy spectra injected at different distances into numerical calculations .
we can calculate the arrival distribution of uhecrs under more realistic situation . as an application of this method
, we calculate the uhecr arrival distribution above @xmath2 ev for the source model which is adopted in our recent study ( paper i ) and can explain the current agasa observation above @xmath8 ev .
we found that the predicted large - scale anisotropy is fully consistent with uniform source distribution , in the same manner as the current agasa data . in order to investigate the effects of the gmf on the large - scale anisotropy , we calculated the harmonic amplitude for the case of @xmath119 g. for @xmath119 g
, the predicted arrival distribution is relatively more isotropic than that for @xmath114 g. this would be due to anisotropy of the event distributions which is caused by the events aligned according to the order of their energies , reflecting the direction of the gmf .
it is also found that the calculated two point correlation function is similar to that of agasa data , when we restrict our attention to the source distributions which predict sufficient isotropic arrival distribution of uhecrs
. there may be effects of the extragalactic magnetic field and the random component of the gmf on the large peak of two point correlation function .
these issues are left for future studies .
finally , we demonstrated the uhecr arrival distribution above @xmath2 ev with the event number expected by future experiments in the next few years .
the interesting feature of the resultant arrival distribution is the events aligned according to the order of their energies , reflecting the directions of the galactic magnetic field .
this is also pointed out by @xcite .
this feature will become clear with increasing amount of data , and allow us to obtain some kind of information about the composition of uhecrs and the gmf .
in forthcoming work , we plan studies about new statistical quantities which allow us to obtain such invaluable information . in the present work ,
we calculate the arrival distribution of uhecrs for our source location scenario which is adopted in our previous study ( paper i ) .
however , it should be mentioned that the same results would be obtained if the sources were truly drawn at random , as far as the source number density is @xmath12 mpc@xmath13 .
the results such as the events aligned according to the order of their energies are independent on our assumption about the source distribution . in this study
, we calculate the harmonic amplitude and two point correlation function , which are only published quantities on uhecrs observed by the agasa including lower energy ones ( @xmath3 ev ) .
in particular , the agasa observation has published neither existence nor non - existence of the events aligned according to the order of their energies .
if more detailed event data with @xmath125 ev are published , we may obtain more strong constraint on our source model , other than the source number density , using another statistical quantities .
this is also one of future study plans .
the authors acknowledge fruitful discussions with s. tsubaki . this research was supported in part by giants - in - aid for scientific research provided by the ministry of education , science and culture of japan through research grant no.s14102004 and no.14079202 . | we present a new method for calculating arrival distribution of ultra - high energy cosmic rays ( uhecrs ) including modifications by the galactic magnetic field .
we perform numerical simulations of uhe anti - protons , which are injected isotropically at the earth , in the galaxy and record the directions of velocities at the earth and outside the galaxy for all of the trajectories .
we then select some of them so that the resultant mapping of the velocity directions outside the galaxy of the selected trajectories corresponds to a given source location scenario , applying liouville s theorem .
we also consider energy loss processes of uhe protons in the intergalactic space . applying this method to our source location scenario which is adopted in our recent study and
can explain the agasa observation above @xmath0 ev , we calculate the arrival distribution of uhecrs including lower energy ( @xmath1 ) ones .
we find that our source model can reproduce the large - scale isotropy and the small - scale anisotropy on uhecr arrival distribution above @xmath2 ev observed by the agasa .
we also demonstrate the uhecr arrival distribution above @xmath2 ev with the event number expected by future experiments in the next few years .
the interesting feature of the resultant arrival distribution is the arrangement of the clustered events in the order of their energies , reflecting the directions of the galactic magnetic field .
this is also pointed out by @xcite .
this feature will allow us to obtain some kind of information about the composition of uhecrs and the magnetic field with increasing amount of data . |
moving target defense ( mtd ) is believed to be a game changer " for cyber defense .
although there have been many studies on specific mtd techniques , the power of mtd is often demonstrated via simulation . rigorously characterizing the power of mtd is an important problem and
is closely related to the well known hard problem of security metrics . in this paper
, we initiate the study of a novel approach for characterizing the power of mtd .
we propose to use the cyber epidemic dynamics approach to characterize the power of _ classes _ of mtd techniques .
we define and investigate two novel and complementary security measures that are applicable when using mtd to achieve a certain defense goal . the first measure is centered on the _ maximum _ portion of time ( in the equilibrium ) during which the system can afford to stay in an undesired / insecure configuration ( or posture ) , without considering the cost of deploying mtd .
the second measure is centered on the _ minimum _ cost when the system has to stay in an undesired / insecure configuration ( or posture ) for a predetermined portion of time .
our analytic studies lead to algorithms for deploying mtd in such optimal fashions . to our knowledge , this is the first systematic study on characterizing the power of classes of mtd techniques . rigorously characterizing the power of mtd ( or any defense ) would be a core problem in the science of security . indeed , the present study initiates a paradigm for measuring the power of mtd ( or other kinds of defense techniques whose deployment can make a global difference ) .
the paradigm is centered on measuring the degree of undersired / insecure configurations that can be tolerated by deploying advanced defense such as mtd .
the specific criterion used in the present paper , namely that the attacks are eventually wiped out in a certain sense , can be substituted by other criteria .
one possible candidate is the containment of malware infections to a certain tolerable level @xcite ( e.g. , by appropriately choosing threshold cryptosystems ) .
the present paper does not introduce any new mtd technique .
rather , it studies how to systematically characterize the power of mtd and optimally launch mtd .
existing studies for a similar purpose are often based on simulation @xcite .
there is effort at analytically evaluating the power of some _ specific _ mtd techniques from a _
view @xcite ; in contrast , we characterize the power of _ classes _ of mtd techniques from a _ global _ view .
cyber epidemic dynamics was rooted in biological epidemic dynamics @xcite .
the first cyber epidemic models @xcite were limited by their _ homogeneity _
assumption that each computer / node has the same effect on the others .
recently , models that are more appropriate for studying cyber security problems have been proposed @xcite . as we will elaborate later
, the basic idea underlying these models is to use a graph - theoretic abstraction to represent the _ attack - defense _ structure , and use _ parameters _ to represent attack and defense capabilities .
cyber epidemic dynamics is a special kind of _ cybersecurity dynamics _
@xcite , we will use the cyber epidemic dynamics model in @xcite as the starting point of our study .
this model @xcite describes _ reactive adaptive defense _
( i.e. , the defender aims to adjust its defense to control / contain the global security state ) .
we extend this model to accommodate mtd , a kind of _ proactive defense _ , and the resulting model is analyzed using different skills @xcite .
we mention that the effect of dynamic structures in cyber epidemic models is studied in @xcite , where the structure dynamics however follows a _ deterministic and periodic _ process , rather than _ adaptively _ scheduled by using ( for example ) mtd .
we also mention that the effect of dynamic semi - heterogeneous structures ( i.e. , clustered networks ) , rather than arbitrary heterogeneous structures , is studied in @xcite .
these studies @xcite consider static parameters only and do not have any of the measures we propose to use .
the rest of the paper is organized as follows . in section [ sec :
models - accommodating - mtd ] , we present a classification of mtd techniques and describe a cyber epidemic dynamics model that can accommodate mtd . in section [ sdvsss ] ,
we characterize the power of mtd that induces dynamic parameters . in section [ dsvsss ] , we characterize the power of mtd that induces dynamic attack - defense structures . we discuss the limitations of the present study in section [ sec : extension - and - discussion ] .
we conclude the paper in section [ sec : conclusion ] .
as mentioned in section [ sec : related - work ] and elaborated later , cyber epidemic dynamics models use a graph - theoretic abstraction to represent the _ attack - defense _ structure , and use _ parameters _ to represent attack and defense ecapabilities .
this suggests us to classify mtd techniques based on what they will induce changes to the attack - defense structure and/or parameters .
example techniques that fall into this class are _ ip address ( and tcp port ) randomization _ and _ dynamic access control_. the basic idea underlying ip address and tcp port randomization is to frequently shuffle the ip addresses of computers by using various methods .
one method is to use virtual machine techniques , such as migrating ensembles of virtual machines @xcite and others @xcite .
another method is to use networking techniques , such as network address space randomization ( nasr ) whereby ip addresses can be dynamically assigned ( in lieu of dhcp ) to render the attacker s hitlist useless @xcite , ip hopping @xcite and others @xcite .
a recent variant also considers constraints and how to minimize the operation cost @xcite .
the basic idea underlying dynamic access control is to dynamically regulate which computers or which network address space can directly have access to the services in which other network address space .
for example , certain servers on a campus network only accommodate service requests from certain classrooms . by dynamically randomizing ip addresses of authorized computers ( e.g. , using aforementioned techniques ) ,
some compromised computers can not launch direct attacks against some target computers .
four kinds of techniques fall into this class : instruction - level , code - level , memory - level , and application - level .
one _ instruction - level _ technique is called instruction set randomization ( isr ) , which aims to randomize the instructions of each process so that the attacker can not inject executable malicious code @xcite .
isr uses a program - specific key to encrypt the instructions of a program and the processor uses the same key to decrypt and run the instructions , where encryption is often based on binary transformation tools , and decryption is often based on dynamic binary instrumentation tools @xcite , emulators @xcite , or architectural support @xcite .
one _ code - level _ technique is _ code randomization _
code randomization offers fine - grained protection against code reuse attacks by substituting / reordering instructions , inserting nops , and re - allocating registers .
code randomization operations can be conducted at the compiler @xcite or virtual machine level @xcite , or via static binary rewriting @xcite and runtime binary rewriting @xcite .
dynamically generated code can be randomized as well @xcite .
one _ memory - level _ technique is called address space layout randomization ( aslr ) , which defeats code - injection attacks by randomizing the memory layout of a program ( e.g. , stack ) either at the compile time or at the runtime @xcite .
aslr can protect an executable ( including the associated static data , stack , heap and dynamic libraries ) @xcite and the operating system kernel @xcite , but can not defeat code reuse attacks .
one _ application - level _ technique is called @xmath0-version programming @xcite , by which the defender can dynamically use different implementations of the same program function .
another technique is called _
proactive cryptography_. cryptographic properties proven in abstract models are undermined by attacks ( e.g. , malware ) that can compromise cryptographic keys .
threshold cryptography can avoid this single - point - of - failure because it split " a key into @xmath1 pieces such that compromising fewer than @xmath2 pieces will not cause exposure of the key , while the cryptographic function can be computed when any @xmath2 of the @xmath1 pieces participate @xcite .
proactive cryptography can render the compromised pieces of a key useless once the pieces are re - shuffled @xcite .
the defender can utilize honeypot - like techniques to capture new attacks . however , the attacker can tomograph " honeypots and bypass the ip addresses monitored by them . therefore
, the defender can dynamically change the ip addresses monitored by honeypots @xcite .
the basic idea underlying cyber epidemic dynamics models ( see , for example , @xcite ) can be explained as follows .
cyber attacks are often launched from compromised computers against vulnerable computers .
this means that there exists an @xmath3 relation , which captures that an attacker ( e.g. , compromised computer ) can _ directly _ attack a victim ( e.g. , vulnerable ) computer . in the extreme case where any computer can attack any other computer
, this relation induces a complete graph structure . in general ,
any graph structure can be relevant .
the resulting graph structures are called _ attack - defense structures _
, where compromised computers / nodes may be detected and cured , but may later get attacked again .
such models can naturally abstract attack and defense capabilities as _ parameters _ that are associated to the nodes and edges of attack - defense structures . a core concept in cyber epidemic dynamics models
is the so - called _ epidemic threshold _ , namely a
_ sufficient condition _ under which the epidemic dynamics converges to the clean state .
we adapt the cyber epidemic dynamics model introduced in @xcite , which considers _ reactive adaptive defense _ , to accommodate mtd ( i.e. , _ proactive defense _ ) .
specifically , the afore - discussed * class i * mtd techniques can be accommodated with _
dynamic attack - defense structures _ , because they can cause that an infected computer may be able to attack a vulnerable computer at time @xmath4 but not at time @xmath5 ( e.g. , the vulnerable computer s ip address has been changed ) . * class ii * mtd techniques can be accommodated with dynamic parameters because they can affect capabilities of attacker and defender over time . *
class iii * mtd techniques can be accommodated with dynamic attack - defense structures ( because an ip address assigned to honeypot at time @xmath4 may be re - assigned to a production computer at time @xmath5 ) _ and _ dynamic parameters ( because the defender could learn zero - day attacks from honeypot - captured data , and identify and disseminate countermeasures to prevent / detect such attacks ) . as such , our characterization study can accommodate the three classes of mtd techniques .
specifically , we consider a cyber epidemic dynamics model with dynamic attack - defense structure @xmath6 , where @xmath7 is the set of nodes ( e.g. , computers ) and @xmath8 is the set of edges at time @xmath2 such that @xmath9 means that node @xmath10 can attack node @xmath11 at time @xmath2 .
. we may think @xmath13 as well .
let @xmath14 $ ] denote the adjacency matrix of @xmath15 , where @xmath16 if @xmath17 and @xmath18 otherwise .
naturally , we have @xmath19 for all @xmath20 ( i.e. , a computer does not attack itself ) .
suppose any node @xmath21 has two possible states : _ secure _ or _
infected_. a node @xmath21 is _ secure _ if it is vulnerable but not successfully attacked yet , and _ infected _ if it is successfully attacked .
let @xmath22 and @xmath23 respectively be the probabilities that @xmath21 is _ infected _ and _ secure _ at time @xmath2 , where @xmath24 .
let @xmath25 be the probability that at time @xmath2 , an _ infected _ node @xmath20 successfully attacks a _ secure _ node @xmath21 over @xmath17 .
let @xmath26 be the probability that an _ infected _ node @xmath21 becomes _ secure _ at time @xmath2 .
suppose the attacks are independently launched .
the probability that a _ secure _
node @xmath21 becomes _ infected _ at time @xmath2 is @xcite : @xmath27 the master dynamics equation is @xcite : @xmath28 this is the starting point of our study .
table [ table : notations ] lists the main notations used in the paper .
.main notations used throughout the paper[table : notations ] [ cols= " > , < " , ] let @xmath29 denote the clean state or equilibrium @xmath30 for all @xmath21 , namely that there are no _ infected _ computers in the equilibrium ( i.e. , the spreading dies out ) .
cyber epidemic threshold is a sufficient condition under which the dynamics converges to @xmath29 . in the special case that @xmath31 and @xmath32 are independent of @xmath2 , it is known @xcite that the dynamics converges to @xmath29 if @xmath33 where @xmath34 is the largest ( in modulus ) eigenvalue of the adjacent matrix @xmath35 of @xmath36 . if @xmath37 , the dynamics does not converge to @xmath29 at least for some initial values .
suppose the defender is confronted with configuration or posture @xmath38 , under which condition ( [ eqn : static - epidemic - threshold ] ) does not hold .
suppose the defender can launch combinations of mtd techniques to induce configurations @xmath39 for @xmath40 , each of which satisfies condition ( [ eqn : static - epidemic - threshold ] ) .
if the defender can always assure @xmath41 for * any * @xmath42 and some @xmath40 , the problem is solved because the defender can make the dynamics converge to @xmath29 by launching mtd to induce @xmath43 .
however , it would be more realistic that the defender can maintain such configurations as @xmath43 ( @xmath40 ) for a small period of time , because the attacker can introduce ( for example ) zero - day attacks to force the system to depart from configuration @xmath43 and enter configuration @xmath44 .
moreover , the system may have to stay in configuration @xmath44 at least for some period of time because @xmath45 is necessary for facilitating some applications .
figure [ fig : switching - illustration ] illustrates the idea of using mtd to make the overall dynamics converge to @xmath29 , while allowing the system to stay for some portion of time in the undersired configuration @xmath44 , which violates condition . during time interval @xmath46 , in @xmath47 during time interval @xmath48 because the defender launches mtd , etc .
although @xmath44 violates condition , the _ overall dynamics _ can converge to @xmath29 because of mtd .
note that @xmath47 and @xmath49 may reside in between two @xmath44 s ( i.e. , launching two combinations of mtd to induce @xmath47 and @xmath49 before returning to @xmath44).,scaledwidth=50.0% ] the preceding discussion leads us to define two measures of power of mtd .
the first definition captures the _ maximum _ time the system can afford to stay in the undersired configuration @xmath44 while the overall dynamics converges to @xmath29 because of mtd , without considering cost .
[ definition - measure - without - cost ] _ * ( power of mtd without considering cost ) * _ consider undesired configuration @xmath50 that violates condition ( [ eqn : static - epidemic - threshold ] ) . suppose the defender can launch mtd to induce configurations @xmath51 for @xmath52 $ ] , where each @xmath43 satisfies condition ( [ eqn : static - epidemic - threshold ] ) . denote by @xmath53 for @xmath54 , where @xmath55 is the adjacency matrix of @xmath56 .
we say mtd is @xmath57-powerful if it can make the overall dynamics converge to @xmath29 , while allowing the system to stay in configuration @xmath44 for the maximum @xmath58-portion of time in the equilibrium .
the second definition captures the _ minimum _ cost with respect to a given portion of time , @xmath59 , during which the system must stay in configuration @xmath44 .
[ definition - measure - with - cost ] _ * ( power of mtd while considering cost ) * _ consider undesired configuration @xmath50 that violates condition ( [ eqn : static - epidemic - threshold ] ) , and the potion of time @xmath59 that the system must stay in configuration @xmath44 .
suppose the defender can launch mtd to induce configurations @xmath51 for @xmath60 , where each @xmath43 satisfies condition ( [ eqn : static - epidemic - threshold ] ) . denote by @xmath61 for @xmath60 , where @xmath62 is the adjacency matrix of @xmath63 .
consider cost function @xmath64 such that @xmath65 is the cost of launching mtd to induce configuration @xmath43 for @xmath60 , where @xmath66 for @xmath67 .
for give cost function @xmath68 , we say mtd is @xmath69-powerful if the overall dynamics converges to @xmath29 at the minimum cost @xmath70 , where @xmath71 ( @xmath72 ) is the portion of time the system stays in configuration @xmath43 in the equilibrium .
* remark*. definitions [ definition - measure - without - cost]-[definition - measure - with - cost ] characterize the power of mtd from two complementary perspectives . definition [ definition - measure - without - cost ] does not explicitly mention @xmath73 because the problem of computing @xmath74 is orthogonal to the existence of @xmath75 .
nevertheless , all of our results allow to explicitly compute @xmath76 .
definition [ definition - measure - with - cost ] explicitly mentions @xmath73 because they are essential to the definition of minimum cost , where @xmath77 is not a parameter of the cost @xmath78 because the system must stay in @xmath44 for a predetermined portion of time @xmath59 .
in this section we characterize the power of mtd that induces dynamic parameters but keeps the attack - defense structure intact ( i.e. , @xmath36 is independent of time @xmath2 throughout this section ) .
let @xmath35 be the adjacency matrix of @xmath36 .
we first recall the following theorem from @xcite and present a corollary of it .
[ sd_theorem ] _ ( @xcite ) _ consider configurations @xmath79 , + where the dynamic parameters @xmath26 and @xmath25 are driven by some ergodic stochastic process .
let @xmath80 and @xmath81 be the respective expectations of the stationary distributions of the process .
suppose convergences @xmath82 and + @xmath83 are uniform with respect to @xmath84 almost surely .
if @xmath85 , the dynamics converges to @xmath29 almost surely ; if @xmath86 , there might exist infected nodes in the equilibrium .
[ corollary : single - configuration ] consider configurations @xmath79 , where @xmath87 are driven by a homogeneous markov process
@xmath88 with steady - state distribution @xmath89 $ ] and support + @xmath90 , meaning @xmath91 and @xmath92 .
if @xmath93 the dynamics will converge to @xmath29 ; if @xmath94 the dynamics will not converge to @xmath29 at least for some initial value scenarios . in this case , despite that @xmath95 violates condition ( [ eqn : static - epidemic - threshold ] ) , the system needs to stay as much as possible in configuration @xmath44 .
fortunately , the defender can exploit mtd to make the overall dynamics converge to @xmath29 .
this is possible because mtd can induce configurations @xmath96 for @xmath97 , where each @xmath43 satisfies condition ( [ eqn : static - epidemic - threshold ] ) .
denote by @xmath98 for @xmath99 . without loss of generality , suppose @xmath100 . according to corollary [ corollary : single - configuration ] , if @xmath101 then the dynamics will converge to @xmath29 .
since inequality ( [ without - cost1 ] ) is strict and is a linear function of @xmath59 , in order to reach the maximum @xmath75 we need to introduce a sufficiently small constant @xmath102 and replace condition ( [ without - cost1 ] ) with @xmath103 theorem [ theorem : sd - ss ] constructively identifies the maximum @xmath75 , the maximal portion of time the system can afford to stay in @xmath44 . [
theorem : sd - ss ] suppose configuration @xmath95 violates condition ( [ eqn : static - epidemic - threshold ] ) .
suppose mtd - induced configurations @xmath96 for @xmath104 satisfy condition ( [ eqn : static - epidemic - threshold ] ) as @xmath105 .
the maximal potion of time the system can afford to stay in configuration @xmath44 is @xmath106 while the system will stay in configurations @xmath107 respectively with portions of time given by @xmath108 in other words , mtd is @xmath109-powerful .
( [ without - cost2 ] ) implies @xmath110 which means @xmath111 moreover , this maximum @xmath75 can be reached if all the equalities in eqs .
( [ without - cost - proof1 ] ) and ( [ without - cost - proof2 ] ) hold , namely @xmath112 this means that the defender only needs to launch the mtd that induces configuration @xmath113 .
theorem [ theorem : sd - ss ] says that although the defender can launch mtd to induce a set of @xmath114 configurations , @xmath107 with @xmath105 , only @xmath115 matters .
this means that @xmath116 is indicative of the capability of a configuration .
figure [ fig : pi-1-dependence ] plots the dependence of @xmath75 on @xmath117 and @xmath118 with @xmath119 .
we observe that for fixed @xmath120 , the maximum portion of time @xmath75 monotonically non - decreases in @xmath118 .
for example , by fixing @xmath121 , @xmath75 is a non - decreasing curve in @xmath118 . on @xmath117 and @xmath118 .
, scaledwidth=46.0% ] in this case , configuration @xmath95 is given and the time the system must stay in @xmath44 is predetermined as @xmath59 .
the defender wants to deploy mtd to make the overall dynamics converge to @xmath29 , while minimizing the cost of launching mtd .
denote the mtd - induced configurations by @xmath96 for @xmath122 .
note that cost may only be considered for @xmath123 because when @xmath124 , it is more meaningful to maximize @xmath75 ( i.e. , the preceding case ) . since we have proved that @xmath125 is necessary to make the dynamics converge to @xmath29 , @xmath126 is the natural upper bound on @xmath59 ( i.e. , if @xmath59 is above the upper bound , we can not assure the dynamics will converge to @xmath29 , regardless of the cost ) . consider cost function @xmath127 as in definition [ definition - measure - with - cost ] .
such that @xmath128 is the cost of launching mtd to induce configuration @xmath43 for @xmath122 , where @xmath129 for @xmath67 .
the objective is to minimize , for given cost function @xmath130 and any constant @xmath131 , the following cost : @xmath132 subject to @xmath133 since @xmath134 is equivalent to @xmath135 , eq .
( [ restrict_con1 ] ) is equivalent to : @xmath136 since the objective is linear and the optimal solution would get on bound of the non - closed constraint ( [ restrict_con2 ] ) , we need to introduce a sufficiently small constant @xmath137 and replace constraint ( [ restrict_con2 ] ) with @xmath138 theorem [ theorem : sd - ss - minimum - cost ] shows how to find the minimum cost @xmath139 according to constraints ( [ objective - function ] ) and ( [ restrict_con3 ] ) , and therefore gives an algorithm for the optimization problem . proof of theorem [ theorem : sd - ss - minimum - cost ] is deferred to the appendix .
[ theorem : sd - ss - minimum - cost ] suppose configuration @xmath95 violates condition ( [ eqn : static - epidemic - threshold ] ) .
suppose mtd - induced configurations @xmath96 for @xmath104 satisfy condition ( [ eqn : static - epidemic - threshold ] ) .
suppose @xmath140 , where @xmath141 , is the potion of time the system must stay in @xmath44 .
suppose @xmath130 is the cost function as discussed above .
define @xmath142 and for @xmath143 , @xmath144 if @xmath145 , the minimal cost under constraint ( [ restrict_con3 ] ) is @xmath146 which is reached by launching mtd to induce configuration @xmath147 only . if @xmath148 , the minimal cost under constraint ( [ restrict_con3 ] ) is @xmath149 denote by @xmath150 .
the minimal cost is reached by launching mtd to induce configurations @xmath151 respectively with portions of time @xmath152 $ ] : @xmath153=\frac{1}{\mu_{m^*}-\mu_{l^*}}\left[\begin{array}{cc } ( \mu_{m^*}-\delta)+\pi_1 ( \mu_1-\mu_{m^*})\\ -(\mu_{l^*}-\delta)+\pi_1 ( \mu_{l^*}-\mu_1 ) \end{array}\right],\end{aligned}\ ] ] where @xmath137 is some constant .
that is , mtd is + @xmath154-powerful .
when not considering cost , theorem [ theorem : sd - ss ] constructively gives a method for optimally launching mtd .
when considering _ arbitrary _ cost function @xmath130 , theorem [ theorem : sd - ss - minimum - cost ] constructively shows how to find the minimum cost @xmath139 according to constraints ( [ objective - function ] ) and ( [ restrict_con3 ] ) , and therefore gives a method for computing the minimum cost and the corresponding strategy for optimally launching mtd .
theorems [ theorem : sd - ss]-[theorem : sd - ss - minimum - cost ] suggest many possible ways / algorithms to achieve the goal , with algorithm [ algorithm : sd - ss - strategy ] being a concrete example .
input : initial configuration @xmath95 , mtd - induced configurations @xmath155 for @xmath104 and @xmath156 , constant @xmath157 determining time resolution , optional cost function @xmath130 , @xmath158 ( @xmath159 ) , optional @xmath59 + output : optimal mtd strategy compute @xmath75 according to eq .
( [ eq : pi-1- * ] ) wait for time @xmath160 launch mtd to make system stay in @xmath115 for time @xmath161 stop launching mtd compute @xmath162 according to eq .
( [ sdvsss - k * ] ) compute @xmath163 according to eq .
( [ sdss - cost ] ) compute @xmath164 according to eq .
( [ eqn : sdvsss - deploy ] ) set @xmath165 and @xmath166 wait for time @xmath167 @xmath168 launch mtd to make system stay in @xmath169 for time @xmath170 set @xmath171 @xmath172 stop launching mtd and wait for time @xmath167 @xmath168 specifically , lines 2 - 7 describe the algorithm corresponding to theorem [ theorem : sd - ss ] ( i.e. , not considering cost ) , where line 4 instructs the defender not to launch mtd so that the system stays in configuration @xmath44 for a period of time @xmath173 , and line 5 instructs the defender to launch mtd to make the system stay in configuration @xmath115 for the period of time @xmath174 . on the other hand , lines
9 - 26 describe the algorithm corresponding to theorem [ theorem : sd - ss - minimum - cost ] ( i.e. , considering cost ) .
if @xmath145 , the defender needs to make the cyber system stay alternatively in configurations @xmath44 and @xmath47 .
if @xmath148 , the defender needs to make the cyber system stay alternatively in configurations @xmath44 for a period of time @xmath173 , in configuration @xmath175 for a period of time @xmath176 and/or in configuration @xmath177 for a period of time @xmath178 .
depending on the random coins flipped on lines 21 and 24 , possible configuration sequences include : @xmath179 and @xmath180 .
another algorithm for achieving the same goal it to make the system in @xmath181 periodically for periods of time @xmath182 , respectively .
the computational complexity of algorithm [ algorithm : sd - ss - strategy ] is straightforward .
when not considering cost , the algorithm incurs @xmath183 computational complexity .
when considering cost , line 9 incurs @xmath184 complexity for searching @xmath162 according to , line 11 incurs @xmath185 complexity for searching the optimal @xmath186 and @xmath187 according to , and all other steps incur @xmath183 complexity .
algorithm [ algorithm : sd - ss - strategy ] applies to _ arbitrary _ cost function @xmath130 .
we make a further observation on theorem [ theorem : sd - ss - minimum - cost ] , which says that for any given @xmath188 and cost function @xmath130 , if @xmath145 , the minimum cost is reached by inducing configuration @xmath47 ; if @xmath148 , eqs .
( [ sdss - cost ] ) and ( [ sdss - minimum - cost ] ) indicate that the minimum cost is dependent upon the property of @xmath130 .
now we show that when @xmath130 is convex or concave , which may hold for most scenarios , we can obtain closed - form results on @xmath189 , and therefore algorithm [ algorithm : sd - ss - strategy ] is naturally simplified .
recall that @xmath190 .
define @xmath191 .
it can be verified that @xmath192\\ = & \pi_1^*f(\mu_1)+(1-\pi_1^*)f(\mu_m)\\ & + r(\mu_l,\mu_m)[(\delta-\pi_1^*\mu_1)-\mu_m(1-\pi_1^*)].\end{aligned}\ ] ] * if @xmath130 is convex , namely @xmath193 , then for fixed @xmath194 ( or @xmath195 ) , @xmath196 is monotonically non - decreasing in @xmath195 ( or @xmath194 ) .
note that @xmath197 , where @xmath198 . for fixed @xmath194 ( or @xmath195 ) , @xmath199 is monotonically non - decreasing ( or non - increasing ) in @xmath195 ( or @xmath194 ) .
the minimum cost is @xmath200 having identified @xmath201 , one can compute @xmath202 according to .
thus , lines 11 and 12 are simplified by this analytical result , with the complexity of searching for the optimal solution ( i.e. , @xmath162 in this case ) reduced to @xmath184 . * if @xmath130 is concave , namely @xmath203 , then for fixed @xmath194 ( or @xmath195 ) , @xmath196 is monotonically non - increasing ( or non - decreasing ) in @xmath195 ( or @xmath195 ) .
the minimum cost is @xmath204 similarly , having identified @xmath205 , one can compute @xmath206 according to .
thus , lines 11 and 12 are simplified by this analytical result , with the complexity of searching for the optimal solution reduced to @xmath183 .
the above discussion suggests the following : if @xmath130 is convex , the defender only needs to launch mtd to induce configurations @xmath207 ; if @xmath130 is concave , the defender only needs to launch mtd to induce configurations @xmath208 . to illustrate the influence of @xmath130 on the power of mtd
, we set @xmath209 , @xmath210 , @xmath211 , @xmath212 , @xmath213 , @xmath119 , @xmath214 , @xmath215 , @xmath216 , @xmath217 , and @xmath218 . from eq .
( [ eq : pi-1- * ] ) , we get @xmath219 .
we set @xmath220 , which means @xmath221 .
figure [ fig : optimal - f ] plots the total cost @xmath222 with different cost functions @xmath130 , where @xmath223 .
the shadow area in the @xmath224-plane is the constrain slope of @xmath225 with respect to condition ( [ restrict_con3 ] ) .
note that @xmath226 , which is linear non - increasing in @xmath227 for fixed @xmath228 ( also in @xmath228 for fixed @xmath227 ) .
for convex function @xmath229 , the above analysis revealed that the minimum cost is reached at @xmath230=[\pi_{l^*},\pi_{m^*}]$ ] as given by eq .
( [ eqn : sdvsss - deploy ] ) , namely by launching mtd to induce configurations @xmath231 .
figure [ fig : optimal - f1 ] shows that the minimum cost is reached at @xmath232=[0,0.3,0.1]$ ] and the minimum cost is @xmath233 , which matches the analytic result given by eq . . for concave
function @xmath234 , figure [ fig : optimal - f2 ] shows that the minimum cost is reached at @xmath235=[0.2,0,0.2]$ ] and the minimum cost is @xmath236 , which matches the analytic result given by eq . .
in this section , we characterize the power of mtd that induces dynamic attack - defense structures @xmath15 , while the parameters @xmath32 are kept intact . more specifically ,
suppose configuration @xmath237 violates condition .
suppose mtd - induced configurations @xmath238 for @xmath239 and @xmath240 satisfy condition .
we want to identify a markov process strategy @xmath241 , defined over @xmath242 , to make the dynamics converge to equilibrium @xmath29 , while staying in configuration @xmath44 as much as possible or minimizing the cost of launching mtd . throughout this section ,
let @xmath243 be the adjacency matrix of @xmath244 for @xmath245 .
we start with a general result where one or more configurations violating condition .
[ dsvsss - theorem ] consider @xmath238 for @xmath246 , where @xmath247 for some @xmath248 violate condition but @xmath249 for some @xmath250 satisfy condition .
then , the overall dynamics converges to @xmath29 almost surely under markov process strategy @xmath241 with infinitesimal generator @xmath251 defined as : 1 . for @xmath252 , @xmath253}{\frac{jc+n'-1-j}{n'-1}-a}$ ] ; 2 . for @xmath254 , @xmath255}{b-\frac{c(j-1)}{n'-1}-\frac{n'-j}{n'-1}}$ ] ;
@xmath256 for all @xmath257 and @xmath258 .
suppose ( if needed , with reordering ) @xmath259 for any @xmath252 , @xmath260 $ ] is a hurwitz matrix @xcite ( i.e. , real parts of all eigenvalues are negative ) , meaning that there exist positive definite matrices @xmath261 and a constant @xmath137 such that @xmath262)^s=[\gamma \lambda_1(a_k)-\beta + \frac{\delta}{2 } ] p_k<0 $ ]
. we can find positive definite matrices @xmath263 with @xmath264 and positive constants @xmath265 such that @xmath266 and @xmath267\}^s\leq [ \gamma \lambda_1(a_\ell)-\beta+\frac{\delta}{2}]p_\ell,~\forall~\ell\leq j.\end{aligned}\ ] ] by combining ( i)-(iii ) in the condition of the theorem , we obtain @xmath268\}^s+\frac{1}{2}\sum_{r=1}^{n ' } q_{rm } p_r\leq-\frac{\delta}{2 } i_n.\end{aligned}\ ] ] since the parameters are static and the structures are driven by markov process @xmath241 , the dynamics of @xmath22 for @xmath21 is : @xmath269\bigg](1-i_v(t))-\beta i_v(t).\end{aligned}\ ] ] since @xmath270\bigg](1-i_v(t))\leq\sum\limits_{u\in v}\gamma ( a_{\sigma_t})_{vu}i_u(t)\ ] ] always holds , we define a new variable @xmath271 with dynamics : @xmath272 note that any sample point @xmath273 corresponds to a deterministic @xmath274 .
let @xmath275^{\top } , y(t)=[y_1(t),\cdots , y_n(t)]^{\top}\end{aligned}\ ] ] be the solutions of systems ( [ sys1 ] ) and ( [ compare ] ) under the markov switching process @xmath274 respectively . from the comparison theory of differential equations , we know @xmath276 holds if @xmath277 , which implies that @xmath278\leq \mathbb e[\|y(t)\|^2]$ ] .
let @xmath279 and @xmath280 .
the joint process @xmath281 is a strong markov process and the infinitesimal generator of the process is : @xmath282 then , we have @xmath283 from the dynkin formula @xcite and eq .
( [ dsvsss - slow - con ] ) , we have @xmath284\}^sy(\tau)d\tau\\\nonumber & & + \frac{1}{2}\mathbb e\int_0^t e^{\zeta\tau}y(\tau)^{\top}\sum_{r=1}^{n'}q_{r,\sigma_\tau}p(r ) y(\tau)d\tau\\\nonumber & \leq & v(y_0,0,\sigma_0).\end{aligned}\ ] ] hence , we have @xmath285 & \leq & \mathbb{e}\bigg[\|y(t)\|^2\bigg]\leq \frac{2\mathbb{e } v(y(t),t,\sigma_t)}{a}\\ & \leq & \frac{2 v(y_0,0,\sigma_0)}{a}e^{-\zeta t},\end{aligned}\ ] ] which implies that @xmath286 converges to zero almost surely for all @xmath11 .
this completes the proof .
[ theorem : dynamic - structure - only ] suppose configuration @xmath237 violates condition and mtd - induced configurations @xmath238 for @xmath245 satisfy condition .
denote by @xmath287 for @xmath288 . without loss of generality ,
suppose @xmath289 . under the definition of @xmath290 in theorem
[ dsvsss - theorem ] , the maximum portion of time the system can afford to stay in configuration @xmath44 is @xmath291}}{\frac{b-1}{2b[-\mu_1+\delta]}+\frac{c - a}{2a[\mu_{n'}-\delta]}},\end{aligned}\ ] ] which is reached by launching mtd to induce configuration @xmath292 .
that is , mtd is @xmath293-powerful .
the infinitesimal generator @xmath290 defined in the proof of theorem [ dsvsss - theorem ] specifies the desired law @xmath241 , which can guide the deployment of mtd to force the overall dynamics converge to @xmath29 .
note that @xmath294 in eq .
( [ dsvsss - theorem - order ] ) represents the number of configurations that violate condition .
hence , @xmath295 in the present scenario . for each @xmath296 , let @xmath297 ; then @xmath298 is the portion of time in configuration @xmath299 .
let @xmath300 be as defined in the proof of theorem [ dsvsss - theorem ] .
consider configurations @xmath301 , where @xmath302 , and @xmath303 for @xmath304 and @xmath305 ( which will be determined below ) are mtd - induced configurations . under the definition of @xmath290 in theorem [ dsvsss - theorem ]
, we have @xmath306 } , ~~x_{k_l}\geq\frac{\frac{c+m-1}{m}-a}{2a[\mu_{k_l}-\delta]},~l=1,\cdots , m.\end{aligned}\ ] ] this means that the dynamics converges to @xmath29 , while staying in configuration @xmath307 for a portion of time @xmath59 , where @xmath308}}\\ & \leq \frac{x_1}{x_1+\frac{c+m-1-am}{2a[\max_{l}\mu_{k_l}-\delta]}}\leq \frac{x_1}{x_1+\frac{c - a}{2a[\mu_{k_{n'}}-\delta]}}\\ & \leq\frac{\frac{b-1}{2b[-\mu_1+\delta]}}{\frac{b-1}{2b[-\mu_1+\delta]}+\frac{c - a}{2a[\mu_{n'}-\delta]}}.\end{aligned}\ ] ] we see that the maximum @xmath59 , namely @xmath75 , is reached when + @xmath309 .
this completes the proof .
theorem [ theorem : dynamic - structure - only ] further confirms that @xmath310 is indicative of the capability of a configuration in terms of `` forcing '' the overall dynamics to converge to @xmath29 .
( [ dsvsss - pi-1 ] ) says that @xmath75 is monotonically increasing in @xmath311 for fixed @xmath120 and decreasing in @xmath120 for fixed @xmath311 .
figure [ fig : pi-1-dependence-2 ] confirms this property with @xmath312 , @xmath313 , @xmath314 , @xmath119 , while eq .
( [ dsvsss - pi-1 ] ) leads to @xmath315 .
on @xmath117 and @xmath311 . , scaledwidth=46.0% ] * remark*. in theorem [ theorem : dynamic - structure - only ] we consider a single configuration @xmath44 that violates condition
. we can extend theorem [ theorem : dynamic - structure - only ] to accommodate multiple configurations that violate condition , because theorem [ dsvsss - theorem ] can accommodate this scenario .
however , if the goal is to maximize the time that the system can stay in the configurations that violate condition , the optimal solution is with respect to configuration @xmath43 , where @xmath294 is given by eq .
, such that the system will stay in @xmath43 for the maximum portion of time @xmath75 given by theorem [ theorem : dynamic - structure - only ] .
the goal is to make the dynamics converge to @xmath29 , at minimum cost of launching mtd , while the system stays a predetermined @xmath59 portion of time in @xmath44 .
the preceding case proved that @xmath316}}{\frac{b-1}{2b[-\mu_1+\delta]}+\frac{c - a}{2a[\mu_{n'}-\delta]}}$ ] is necessary to make the dynamics converge to @xmath29 .
let @xmath317 for @xmath288 .
let the cost function @xmath318 : @xmath319 be the same as discussed in definition [ definition - measure - with - cost ] , namely that @xmath320 is the cost of launching mtd to induce @xmath321 for @xmath245 , where @xmath322 for @xmath67 .
let @xmath241 be the desired law " for deploying mtd , and denote by @xmath323 $ ] its infinitesimal generator .
let @xmath324 , where @xmath325 is the expectation of sojourn time in configuration @xmath321 .
then , the portion of time in @xmath321 is @xmath326 .
our goal is to find the law " under which the cost is minimum .
theorem [ dsvsss - theorem ] specifies the desired markov
law " @xmath241 via its infinitesimal generator .
now we consider cost function @xmath327 with respect to @xmath328 $ ] .
[ dsvsss - theorem - b ] given configuration @xmath44 that violates condition and mtd - induced configurations @xmath321 for @xmath245 that satisfy condition .
suppose @xmath140 , where @xmath350}}{\frac{b-1}{2b[-\mu_1+\delta ] } + \frac{c - a}{2a[\mu_{n'}-\delta ] } } , $ ] is the portion of time that the system must stay in @xmath44 .
denote by @xmath351 where @xmath352},\\ \delta(k_1,\cdots , k_{m'})&=&\frac{1-\pi_1 } { \pi_1 } \bar{x}_1-\sum_{l=1}^{m'}\bar{x}_{k_l}(m').\end{aligned}\ ] ] we want to find @xmath353 such that @xmath354 for given cost function @xmath318 with arbitrary constant @xmath355 , the minimum cost is @xmath356 which is reached by launching mtd to induce configuration + @xmath357 via the following deployment strategy : @xmath358 hence , mtd is @xmath359-powerful . by comparing the costs of these stationary points , we find the minimum cost of launching these @xmath376 configurations is @xmath377 this complete the proof .
* remark*. similar to theorem [ theorem : dynamic - structure - only ] , in theorem [ dsvsss - theorem - b ] we consider a single configuration @xmath44 that violates condition
. we also can extend theorem [ dsvsss - theorem - b ] to accommodate multiple configurations that violate condition , because theorem [ dsvsss - theorem ] can accommodate this scenario .
the extension is straightforward because the portions of time that are allocated to the violating configurations are fixed and not involved in the definition of cost .
theorems [ theorem : dynamic - structure - only ] and [ dsvsss - theorem - b ] are constructive and lead to algorithm [ alg : ds - ss - strategy ] that can guide the deployment of optimal mtd .
input : configuration @xmath44 , optional cost function @xmath318 , mtd - induced @xmath321 for @xmath245 and @xmath240 , constant @xmath157 ( determining time resolution ) , @xmath158 ( @xmath378 ) , optional @xmath140 + output : optimal mtd strategy compute @xmath75 according to eq .
( [ dsvsss - pi-1 ] ) wait for time @xmath160 launch mtd to force the system to stay in configuration @xmath379 for time @xmath380 stop launching mtd find indices @xmath381 according to eq .
( [ dsvsss - withcost - opt ] ) set @xmath382 as defined in eq .
( [ dsvsss - withcost - deployment ] ) and @xmath332 for @xmath383 wait for time @xmath167 @xmath384 launch mtd to make the system stay in @xmath385 for time @xmath386 set @xmath387 @xmath388 stop launching mtd to make the system stay in @xmath44 for time @xmath389 @xmath384 in algorithm [ alg : ds - ss - strategy ]
, lines 2 - 7 correspond to the case of not considering cost , where each step incurs @xmath183 computational complexity .
lines 9 - 21 correspond to the case of considering cost .
specifically , line 9 incurs complexity @xmath390 , which is not infeasible because in practice @xmath391 ( i.e. , the number of mtd - induced configurations ) is often small .
possible instances of configurations the system will stay include @xmath392 and + @xmath393 .
first , the present study assumes that the attack - defense structures and parameters are given .
it is sufficient for characterizing the power of mtd . nevertheless , it is important to study how to obtain such structures and parameters .
second , the present study does not allow the attacker to choose when to impose configuration @xmath44 .
it is important to give the attack the freedom in choosing when to impose @xmath44 .
this incurs technical difficulties . for example , the portion of time in the violating configuration may not be fixed at @xmath77 , which breaks the setting of the optimisation problem .
third , it is interesting to extend the model to accommodate heterogeneous @xmath394 and @xmath395 .
however , this will make the model difficult to analyze mainly because of accommodating @xmath395 .
we have introduced an approach of using cyber epidemic dynamics to characterize the power of mtd .
the approach offers algorithms for optimally deploying mtd , where
optimization " means maximizing the portion of time the system can afford to stay in an undesired configuration , or minimizing the cost of launching mtd when the system has to stay in an undesired configuration for a predetermined portion of time .
we have discussed the limitations of the present study , which should inspire fruitful future research .
* acknowledgement*. wenlian lu was supported by a marie curie international incoming fellowship from european commission ( no .
fp7-people-2011-iif-302421 ) , national natural sciences foundation of china ( no . 61273309 ) , and program for new century excellent talents in university ( no .
ncet-13 - 0139 ) .
shouhuai xu was supported in part by aro grant # w911nf-12 - 1 - 0286 and afosr grant fa9550 - 09 - 1 - 0165 .
any opinions , findings , and conclusions or recommendations expressed in this material are those of the author(s ) and do not necessarily reflect the views of any of the funding agencies .
m. atighetchi , p. pal , f. webber , and c. jones . adaptive use of network - centric mechanisms in cyber - defense . in _
ieee symposium on object - oriented real - time distributed computing , 2003 _ , pages 183192 , 2003 .
t. jackson , b. salamat , a. homescu , k. manivannan , g. wagner , a. gal , s. brunthaler , c. wimmer , and m. franz .
diversifying the software stack using randomized nop insertion in s. jajodia , a. ghosh , v. swarup , c. wang , and x. wang , editors , _ moving target defense _ ,
pages 7798 . for any @xmath396
, we have @xmath397 , meaning that the dynamics will converge to @xmath398 by launching mtd to induce configuration @xmath399 with portion of time @xmath400 . for any @xmath401
, we have @xmath402 , meaning that condition ( [ corollary : ave1 ] ) for the dynamics to converge to @xmath398 is not satisfied even if the defender launches mtd to induce configuration @xmath321 with portion of time @xmath400 .
if @xmath145 , the dynamics will converge to @xmath398 by launching mtd to induce configuration @xmath403 . since @xmath130 is non - decreasing ,
we have @xmath404 it can be seen that the equality above can be guaranteed by taking @xmath405 and @xmath406 for @xmath407 .
if @xmath148 , we use lagrange multipliers to calculate the minimum cost @xmath408 , subject to constraint ( [ restrict_con3 ] ) .
since lagrange multipliers require equality constraints , we introduce vector of variables @xmath409 $ ] with @xmath410 satisfying @xmath411 and variable @xmath412 with @xmath413 , such that solving the minimization problem is equivalent to finding the minimum of the following function @xmath414 subject to @xmath415 where @xmath416,@xmath417 , and @xmath418 have continuous first partial derivatives . to solve this variant problem
, we introduce lagrange multipliers @xmath419 and @xmath420 and the lagrange function as follows @xmath421\\ & + \alpha_2[h_2(x)+\pi_1 -1].\end{aligned}\ ] ] denote by @xmath422 the minimum of @xmath423
. there exist @xmath424 and @xmath425 such that @xmath426 is a stationary point for the lagrange function @xmath427 , i.e. , with gradient @xmath428 .
then , we are to solve @xmath428 : @xmath429x_2=0\\ ~~~~~~~~~\vdots\\ \nabla_{x_n}\lambda=2[f(\mu_n)+\alpha_1\mu_n+\alpha_2]x_n=0\\ \nabla_{\zeta}\lambda=-2\alpha_1\zeta=0\\ \nabla_{\alpha_1}\lambda=\sum_{j=2}^n\mu_j x_j^2+\pi_1 \mu_1-\delta-\zeta^2=0\\ \nabla_{\alpha_2}\lambda=\sum_{j=2}^n x_j^2+\pi_1
-1=0 , \end{array}\right.\end{aligned}\ ] ] _ case 1 : the optimal strategy is that the defender launches mtd to induce only one configuration @xmath430
. _ then , @xmath396 must hold and the cost is @xmath431 . hence , the minimum cost of launching mtd to induce a single configuration is @xmath432 which is reached when inducing configuration @xmath433 .
_ case 2 : the optimal strategy is that the defender launches mtd to induce two configurations @xmath434 . _ the minimum cost will be reached at @xmath435 because the cost function @xmath436 is non - decreasing in @xmath67 .
firstly , we look for the minimal cost when launching mtd to induce configurations @xmath437
. this requires to find stationary points of @xmath427 such that @xmath438 and @xmath439 for @xmath440 .
it can be verified that the following points are stationary points for @xmath427 : @xmath441= \frac{1}{\mu_k-\mu_1 } & \left[\begin{array}{c } ( \mu_{m}-\delta)+\pi_1 ( \mu_1-\mu_{m})\\ -(\mu_{l}-\delta)+\pi_1 ( \mu_{l}-\mu_1 ) \end{array}\right],\\ & ~\zeta=0,~ { x}_k^2=0,~k\geq 2,~ k\neq l , m.\end{aligned}\ ] ] and the corresponding cost is @xmath199 .
hence , the minimum cost of inducing two configurations is @xmath442 which is reached by launching mtd to induce configurations @xmath151 according to @xmath164 in eq .
( [ eqn : sdvsss - deploy ] ) .
_ case 3 : the optimal strategy is that the defender launches mtd to induce @xmath443 configurations @xmath444 . _ to find the minimum cost , we need to find stationary points of @xmath427 such that @xmath445 and @xmath439 for @xmath446 .
that is , @xmath447 thus , the stationary point @xmath372 should satisfy @xmath448 the cost at this stationary points @xmath372 becomes @xmath449 if ( [ multiple_points ] ) does not hold , then there is no stationary point @xmath372 in the form @xmath445 and @xmath439 for @xmath446 , meaning that there is no minimum cost when inducing these configurations . | moving target defense ( mtd ) can enhance the resilience of cyber systems against attacks .
although there have been many mtd techniques , there is no systematic understanding and _ quantitative _ characterization of the power of mtd . in this paper
, we propose to use a cyber epidemic dynamics approach to characterize the power of mtd .
we define and investigate two complementary measures that are applicable when the defender aims to deploy mtd to achieve a certain security goal .
one measure emphasizes the _ maximum _ portion of time during which the system can afford to stay in an undesired configuration ( or posture ) , without considering the cost of deploying mtd .
the other measure emphasizes the _ minimum _ cost of deploying mtd , while accommodating that the system has to stay in an undesired configuration ( or posture ) for a given portion of time .
our analytic studies lead to algorithms for optimally deploying mtd . |
as a part of the third phase of the sloan digital sky survey ( sdss - iii ; * ? ? ?
* ) , the marvels ( _ _ m__ulti - object _ _ a__po _ _ r__adial _ _ v__elocity _ _ e__xoplanet _ _ l__arge - area _ _ s__urvey ) project is searching for substellar companions by monitoring the radial velocities ( rvs ) of 3330 fgk stars @xcite .
this sample size is large enough for the project to find relatively rare objects , such as brown dwarf ( bd ) companions to solar - type stars .
the paucity of observed bd companions to solar - type stars with separations of @xmath85 au is typically referred to as the bd desert @xcite . since the size of the marvels sample allows us to begin to quantify how arid the bd desert may be , any marvels discovery of a bd in the desert ( or lack thereof ) is a step toward increasing our understanding of bd formation .
in addition to its large homogeneous target sample , marvels differs from other surveys for substellar companions in two key ways .
first , the project employs a dispersed fixed - delay interferometer ( dfdi ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
second , it uses a multi - object spectrograph to observe 60 stars simultaneously @xcite .
the dfdi prototype instrument was used to discover the first extrasolar planet around hd 102195 in 2006 with this new rv method @xcite .
the marvels dfdi technique combines an interferometer with a medium resolution spectrograph ( @xmath912,000 ) in order to obtain a precision of @xmath5100 m s@xmath7 .
given its rv precision and survey design to monitor each target with at least 24 rv measurements over at least 1 yr , marvels is sensitive to bd and low - mass stellar companions with periods ranging from a few days to hundreds of days .
nonetheless , certain specific types of astrophysical false positives can mimic substellar companions unless additional vetting is performed .
this paper describes just such a case , tyc-3010 - 1494 - 1 ( hereafter tyc 3010 ) , a stellar binary that initially appeared as a single star with a substellar companion and that , through a confluence of orbital parameters , continued to masquerade as such despite a disconcertingly extensive amount of observation and analysis . when we began analysis of tyc 3010 , marvels and its pilot project had already detected two bd candidates orbiting late f stars in the bd desert ( ( * ? ? ?
* ; * ? ? ?
* at present , we have three more candidates in the desert : ) @xcite ) . the marvels discovery data indicated that tyc 3010 possessed a substellar companion with a minimum mass of @xmath10 and that it was on a @xmath5238-day moderately eccentric orbit with an rv amplitude of @xmath51.5 km s@xmath7 ( see the top panel of figure [ fig : mc10rv_curve_bd ] ) .
however , given the cadence of marvels and the period of the orbit , there were significant gaps in the phase coverage and additional observations with a different spectrograph were required to constrain the rv solution .
initially , the follow - up data remained fully consistent with the bd companion scenario .
however , during the course of the program , we found two rv points that were shifted by @xmath520 km s@xmath7with respect to most of our data ; while investigating the source of these anomalous points , we realized that a few similar points had been rejected from our marvels discovery data by the team s outlier rejection procedures ( see bottom panel of figure [ fig : mc10rv_curve_bd ] ) .
examining the cross - correlation function ( ccf ) of the anomalous rv points ( in both the discovery and subsequent data ) revealed evidence that there were two components in the ccf , which suggested that the companion to the primary was most likely a stellar - mass secondary .
finally , including the initially flagged outlier measurements and disentangling the rv measurements of the two components , the system was found to be a nearly equal - mass stellar binary ( @xmath11 ) on a highly eccentric orbit ( @xmath12 ) .
evidently , for a system like tyc 3010 , it is possible to clip just a few measurements and obtain an apparently reasonable solution that is convincing but completely incorrect . as large scale rv and transit surveys for exoplanets
become more common , it is increasingly inevitable that any and all forms of astrophysical false positives , despite their rarity , will be found .
indeed , the first bd candidate discovered by the marvels project , , appeared to exhibit evidence for an additional planet - mass companion , but turned out instead to likely be a quadruple system , comprising four stars with no detected bd or planetary - mass companion @xcite .
akin to tyc 3010 , is a double - lined spectroscopic binary ; the stars have relative rvs which are sufficiently low that they are always blended , even at the resolution of the hobby - eberly telescope ( het ; @xmath13 mode ) .
thus , with both and tyc 3010 , we actually measure a flux - weighted mean of two sets of stellar spectral lines .
this flux - weighted mean exhibits a suppressed velocity shift that mimics a single - lined binary with a bd secondary .
both systems possess geometries that allow them to masquerade as less massive systems : is nearly face - on , which leads to low projected velocities , while tyc 3010 is on a highly elliptical orbit with a semi - major axis oriented nearly perpendicular to our line of sight .
similarly , @xcite describe what at first appeared to be a transiting bd companion to an f star from the tres transit survey , but turned out instead to be an f star blended with a g+m stellar eclipsing binary .
the system that we describe here follows these unfortunate examples , and is similarly pernicious . in the following sections , we present our analysis as a kind of cautionary tale for other rv surveys to avoid similar false positives . in section
[ sec : data ] , we describe the spectroscopic and photometric data obtained for tyc 3010 . in section [ sec : results ] , we discuss in detail the nature of the evidence that led us to conclude that tyc 3010 was an eccentric stellar binary instead of a bd companion to a solar - type star
. we also present the properties we derived for both components of the spectroscopic binary . in section [ sec : discussion ] , we discuss the circumstances that allowed this false positive to masquerade for so long and through several vetting steps as a compelling detection of a substellar companion , and we describe methods that the marvels team and other rv surveys can use to recognize this kind of astrophysical false positive in the future . finally , in section [ sec : summary
] , we conclude with a summary of the main results .
we obtained a total of 65 rv measurements from the sloan 2.5 m , the apo 3.5 m , and the het 9.2 m telescopes .
we will briefly summarize the characteristics of the data from all three telescopes .
for more details of the analysis , please see @xcite , @xcite , and @xcite .
a total of 28 spectra ( see table [ tbl : sing_rvs ] ) of tyc 3010 were obtained with the sloan 2.5 m telescope @xcite at apache point observatory ( apo ) .
the multi - fiber marvels spectrograph @xcite can simultaneously measure the rvs of 60 stars during each telescope pointing .
both beams of the interferometer are imaged onto the detector , so each 50-minute observation results in two fringed spectra in the wavelength range of @xmath5500570 nm with a resolving power of @xmath14 .
the marvels interferometer delay calibrations are described in @xcite . for more details on how the data were reduced and analyzed to yield rvs , see @xcite .
cccc 2454927.82470 & m & 62.681 & 0.148 + 2454928.85061 & m & 62.564 & 0.139 + 2454964.76792 & m & 61.479 & 0.108 + 2454965.77714 & m & 61.374 & 0.113 + 2454994.69536 & m & 59.933 & 0.115 + 2455193.91250 & m & 62.102 & 0.165 + 2455197.96727 & m & 61.753 & 0.134 + 2455198.94828 & m & 61.714 & 0.095 + 2455199.96552 & m & 61.664 & 0.139 + 2455200.98947 & m & 61.585 & 0.097 + 2455201.97760 & m & 61.587 & 0.116 + 2455202.99063 & m & 61.528 & 0.149 + 2455258.88272 & m & 39.192 & 0.091 + 2455259.83118 & m & 41.327 & 0.092 + 2455260.82412 & m & 45.097 & 0.145 + 2455261.82050 & m & 48.416 & 0.096 + 2455280.77587 & m & 61.103 & 0.105 + 2455280.76844 & m & 61.174 & 0.117 + 2455283.81484 & m & 61.411 & 0.154 + 2455284.75054 & m & 61.461 & 0.112 + 2455311.68421 & m & 62.493 & 0.209 + 2455313.62591 & m & 62.402 & 0.174 + 2455369.64423 & m & 62.531 & 0.333 + 2455551.99403 & m & 62.788 & 0.120 + 2455552.98222 & m & 62.856 & 0.104 + 2455553.98561 & m & 62.795 & 0.121 + 2455556.97163 & m & 62.821 & 0.123 + 2455557.97465 & m & 62.801 & 0.104 + 2455471.98302 & a & 60.138 & 0.116 + 2455519.95995 & a & 61.359 & 0.052 + 2455519.98157 & a & 61.371 & 0.051 + 2455637.88366 & a & 62.452 & 0.055 + 2455637.92209 & a & 62.278 & 0.048 + 2455654.83350 & a & 62.390 & 0.059 + 2455665.65219 & a & 62.323 & 0.065 + 2455665.69165 & a & 61.664 & 0.075 + 2455669.60113 & a & 61.827 & 0.052 + 2455686.82409 & a & 60.946 & 0.076 + 2455695.66512 & a & 60.949 & 0.039 + 2455695.70529 & a & 60.931 & 0.053 + 2455703.61994 & a & 60.942 & 0.116 + 2455709.77767 & a & 59.749 & 0.098 + 2455903.90846 & h & 62.448 & 0.051 + 2455917.87269 & h & 62.237 & 0.060 + 2455928.84083 & h & 61.759 & 0.046 + 2455940.80855 & h & 61.122 & 0.058 + 2455946.80490 & h & 60.285 & 0.055 + 2455950.80134 & h & 59.539 & 0.045 + 2455953.82447 & a & 58.385 & 0.049 + 2455954.00566 & h & 58.467 & 0.050 + cccccc 2455725.68377 & a & 46.012 & 0.167 & 75.222 & 0.257 + 2455735.62781 & a & 43.197 & 0.251 & 81.056 & 0.175 + 2455956.76037 & h & 53.788 & 0.030 & 69.104 & 0.063 + 2455959.78075 & h & 51.409 & 0.025 & 71.807 & 0.055 + 2455964.75592 & h & 44.163 & 0.026 & 80.066 & 0.055 + 2455964.83117 & a & 44.884 & 0.071 & 79.700 & 0.372 + 2455967.75334 & h & 36.755 & 0.030 & 88.389 & 0.062 + 2455967.82824 & a & 37.684 & 0.076 & 88.651 & 0.471 + 2455968.74640 & h & 34.456 & 0.025 & 90.966 & 0.054 + 2455971.73989 & h & 36.359 & 0.029 & 88.685 & 0.060 + 2455972.97350 & h & 40.368 & 0.025 & 84.354 & 0.054 + 2455976.73787 & h & 50.072 & 0.024 & 73.253 & 0.051 + 2455977.71541 & h & 51.788 & 0.028 & 71.580 &
0.058 + 2455978.71767 & h & 52.990 & 0.026 & 69.840 & 0.056 + 2455979.71599 & h & 54.160 & 0.029 & 68.450 & 0.062 + as described below , it proved essential to examine the ccfs of the individual spectra .
however , performing a cross - correlation on a dfdi spectrum requires a few steps beyond what one performs for a typical slit or cross - dispersed echelle spectrograph . in both cases the images
are reduced using standard techniques ( bias subtraction , trace correction , flat fielding etc . )
once a fully processed two - dimensional spectrum has been extracted , there is a divergence in the techniques . in the case of a normal spectrum ,
one merely sums the flux in the slit ( channel ) direction to produce a one - dimensional spectrum . this approach is not possible in the dfdi technique because the fringing pattern will introduce false fluctuations in total flux if one just sums in the slit direction .
these fluctuations will be a function of the phase of the fringe pattern in each pixel channel .
to correct for this effect , a sinusoidal function of the form @xmath15 is fit to each pixel column . for the purposes of cross - correlation the only term of interest is @xmath16 , or the mean flux in each channel .
a one dimensional spectrum is then constructed using the @xmath16 term in each channel . from this point
forward the ccf is determined using standard techniques .
a total of 19 rv observations were taken with the apo 3.5 m telescope using the arc echelle spectrograph ( arces ; * ? ? ?
this spectrograph operates in the optical regime from @xmath53,60010,000 with a resolving power of @xmath17 .
the first set of observations were taken from 2010 october to 2011 june .
the second set of observations , which were undertaken with the goal of increasing phase coverage of periastron , were obtained during 2012 january february .
as shown in tables [ tbl : sing_rvs ] and [ tbl : doub_rvs ] , there were 15 arces points observed outside of periastron , and 4 points during periastron ( the first two of these periastron points are where we initially resolved both the primary and secondary spectral lines
see bottom panel of figures [ fig : mc10rv_curve_bd ] , [ fig : spectra ] , and [ fig : ccfs]and began to suspect that the system might be a double - lined spectroscopic binary ) . to achieve high - accuracy rv measurements with the echelle spectrograph
, we obtained a thorium - argon ( thar ) exposure after every science exposure . in order to place tyc 3010 on an absolute rv scale ,
we also frequently bracketed our observations of tyc 3010 with observations of the rv standard hd 102158 , which has an absolute rv of 28.122 km s@xmath7 @xcite . from the standard deviation of the 13 rv measurements we obtained for hd 102158 ( see table [ tbl : rvstd ] ) , we were able to determine that the arces spectrograph possesses an rv stability of @xmath50.5 km s@xmath7 .
two of the arces spectra were taken with longer exposure times in order to achieve a high signal - to - noise ratio ( s / n ) for deriving the fundamental stellar parameters ( see section [ sec : starpars ] ) .
these two spectra were taken with an exposure time of 200 s and with the default slit setting described in @xcite .
the data were reduced with iraf , and after barycentric corrections and continuum normalization , the two spectra were combined to produce a final spectrum with an s / n of @xmath5170 per resolution element at @xmath56500 . however , once we realized that tyc 3010 was a double - lined spectroscopic binary , we re - derived the spectroscopic parameters with a double - lined spectrum obtained near periastron , as described in section [ sec : starpars ] . upon realizing the eccentric binary - star nature of the object from the apo 3.5 m data , observations where initiated with the 9.2 m het @xcite and the higharcsec resolution spectrograph ( hrs ;
* ) at a resolving power of using a 2 arcsec optical fiber .
a total of 18 observations were obtained to completely cover periastron , and thereby fully constrain the orbit .
the queue - scheduled observing mode of the het @xcite is extremely well suited for investigating objects that require monitoring over a long timespan , as well as targeted observations near periastron passage . for wavelength calibration ,
thar images were obtained immediately before and after the science exposure to aid in calibrating any possible instrument drift .
the data were reduced and wavelength calibrated using custom optimal extraction scripts written in idl .
rvs were measured using two different techniques , which we describe below .
the het observations clearly resolve the orbit for tyc 3010 , and constrain the eccentricity to a value of @xmath2 ( see section [ sec : sb2_soluxn ] ) .
ccc 2455654.81577 & 28.734 & 0.052 + 2455665.67180 & 28.050 & 0.034 + 2455665.71479 & 28.169 & 0.043 + 2455669.58461 & 27.736 & 0.045 + 2455686.80825 & 27.470 & 0.054 + 2455695.68591 & 28.030 & 0.038 + 2455695.72570 & 28.090 & 0.034 + 2455703.60341 & 28.244 & 0.039 + 2455709.76151 & 27.476 & 0.030 + 2455725.66705 & 26.692 & 0.123 + 2455735.61237 & 28.126 & 0.109 + 2455964.81467 & 28.093 & 0.058 + 2455967.81369 & 27.728 & 0.050 + rvs were measured using a cross - correlation mask derived from national solar observatory fourier transform spectroscopic solar data @xcite , and a technique similar to that described by @xcite . the resultant ccf encodes information from the @xmath5400600 nm region , and we elected not to use redder wavelengths due to issues with telluric contamination .
figure [ fig : ccfs ] shows the resulting ccf for an epoch during periastron and one outside of periastron ; as is the case for the arces data , during periastron the primary and secondary peaks are clearly visible in the het ccfs , but outside of periastron only a single peak is resolved .
the centroid of the ccf peak is determined by fitting a gaussian .
this technique has been used successfully for isolated stars to derive precise rvs by the teams using fiber - fed high resolution spectrographs ( e.g. , harps , sophie , elodie , coralie ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) , since psf stability is an important component of deriving precise rvs with this technique .
any mismatch between the ccf and the simple gaussian model is absorbed as a zero - point offset in the derived rvs as long as the psf is stable ( resulting in a stable ccf shape ) .
the het / hrs spectrograph is also fiber - fed , enabling this technique to also be applied to binary stars .
this method is computationally efficient , and also does not require that the spectra be normalized , resulting in a quick turn around in determining rvs once the data are in hand .
the rvs derived enabled us to plan and obtain observations as soon as the peaks began to separate on the approach to peri - passage .
table [ tbl : sing_rvs ] shows the het rvs obtained with this technique for those epochs where the ccf appears as a single peak .
while the ccf mask technique described above works quite well , it does not yield the best rvs possible for spectra with two ccf peaks since only one mask ( g2 spectral type ) was used in determining peak positions .
once all the data were in hand , we were able to apply the two - dimensional cross - correlation algorithm , todcor @xcite .
todcor can simultaneously cross - correlate two stellar templates against a blended target stellar spectrum to disentangle the stellar rvs of the components as well as derive a flux ratio .
we used todcor along with hrs observations of hd161237 ( g5v ) and hd 198596 ( k0v ) as templates to measure the rvs of tyc 3010 .
the hrs spectrum was divided into different bandpasses , and each bandpass was solved independently following @xcite and the resulting cross - correlation surface combined with a maximum likelihood analysis .
further details on our implementation of the todcor algorithm , as well as details of our custom hrs spectral extraction pipeline , can be found in @xcite .
table [ tbl : doub_rvs ] shows the rvs of the primary and secondary determined using this algorithm at those epochs where the ccf is double peaked .
we add 0.05 km s@xmath7 in quadrature to the todcor formal errors to account for additional noise effects like wavelength calibration , small tracking induced psf changes , etc .
while the het observed the target on 18 epochs , the secondary rvs are only reliably measured for 11 epochs .
these are the epochs where the primary and secondary peaks are sufficiently separated to determine an independent rv for each .
while rvs can be determined for the other 7 epochs , they are rvs of blended spectra , and the associated systematic error is not only larger , but also more difficult to quantify .
since both peaks are unambiguously detected in todcor at these epochs , we are also able to measure the secondary to primary flux ratio , @xmath18 , which we determine to be @xmath19 by averaging the flux ratio of the templates ( g5v and k0v ) over four bandpasses spanning 4663 - 5863 . finally , the mass ratio derived from these 11 epochs is @xmath11 .
the marvels team obtained lucky imaging for tyc 3010 in order to detect any spatially resolvable companions . in 2011 april , using the fastcam @xcite instrument on the 1.5 m tcs telescope at observatorio del teide in spain , we obtained 47,000 frames in the @xmath20-band with a 70 ms exposure time for each frame .
data processing was accomplished with a custom - made idl pipeline . as described in @xcite ,
the best frames are selected via the brightest pixel ( bp ) method .
the frames with the brightest @xmath21% of bps are combined to generate a final image , where @xmath22 \{@xmath23 } for tyc 3010 .
figure [ fig : lucky_exp ] shows the resulting final images for each particular percentage of the best frames .
no companions are detected , but we can place constraints on the upper limit of the masses of resolvable companions . using the spectroscopic @xmath24 for tyc 3010 ( see section [ sec : starpars ] ) , and the relations from @xcite , we determine the bolometric magnitude . combining the bolometric magnitude with mass luminosity relations @xcite , we convert the detection limit for the @xmath20-band magnitude into a lower limit for the masses of detectable companions at different separations . at the @xmath25 level , where @xmath26 is defined in @xcite as the rms of the counts within concentric annuli centered on tyc 3010 , and using 8 pixel boxes , we can rule out the presence of detectable companions above a mass of @xmath27 outside of 50 au ( see figure [ fig : ao_mass_limit ] ) .
in addition to the lucky imaging , we were also able to obtain adaptive optics ( ao ) images of tyc 3010 on 2012 october 21 ut using the nirc2 imager at keck ( instrument pi : keith matthews ; * ? ? ? * ) .
observations consist of a sequence of nine dithered frames in the @xmath28 filter ( central @xmath29 m ) using the narrow camera ( plate scale = 10 mas pix@xmath7 ) setting .
each frame consisted of 20 coadds with 0.1814 s of integration time per coadd , totaling 32.65 s of on - source exposure time .
images were processed using standard techniques to remove hot pixels , subtract the sky - background , and align and coadd the cleaned frames .
no candidate companions were identified in either raw or processed images .
figure [ fig : ao_mass_limit ] shows our sensitivity to off - axis sources as a function of angular separation .
our diffraction - limited observations rule out the presence of companions 6.5 magnitudes fainter than the primary star for separations beyond 0.5@xmath30(@xmath25 ) . using theoretical isochrones from @xcite
, we convert this magnitude limit to a mass upper limit , as shown in figure [ fig : ao_mass_limit ] ; we can exclude companions with a mass above outside of 100 au .
in this section we present the orbit solution of the tyc 3010 system .
first we show how the data initially suggested a spurious solution in which tyc 3010 is a single star with a bd companion .
next we present the correct solution , in which tyc 3010 is shown to be a double - lined spectroscopic stellar binary ( sb2 ) with two solar - type stars , and we provide a full characterization of the system properties . of the 28 rv measurements collected with the marvels instrument , 24 passed the data quality checks and were therefore included in the automated orbit solution fitting procedures .
for the arces data , the first 14 consecutive rv points obtained during the initial set of observations were fully consistent with our working solution , that tyc 3010 was a candidate bd ( see figure [ fig : mc10rv_curve_bd ] and table [ tbl : orb_params ] ) .
these rv points are well fit by a solution consistent with a substellar object ( @xmath31 ) orbiting in the bd desert around a solar - type star .
a robust fit to the low amplitude ( @xmath512 km s@xmath7 ) variations was found with the exofast program @xcite , which uses a set of markov chain monte carlo trials to find the best fit .
this solution , shown in figure [ fig : mc10rv_curve_bd ] ( top panel ) , is a very convincing fit to the 38 originally included marvels ( red points ) and apo ( blue points ) measurements .
this fit yielded a @xmath32 of 34.63 after scaling the error bars to force @xmath32/dof@xmath51 .
these scalings were not unreasonable compared to other marvels candidates .
as noted previously , four of the original marvels rv measurements were initially rejected as outliers .
the outlier rejection procedure included a 40@xmath26 statistical clipping to avoid phase wrapping , and rejection of consecutive points deviating by a large systematic offset from the bulk of the measurements .
the latter rejection step was specifically implemented in an attempt to account for cases of fiber mis - pluggings , which are known to happen on occasion , in which the wrong star is observed for a few observations in a row and those few measurements appear at a very different systemic velocity relative to the majority of the measurements .
the four rejected marvels measurements are also shown in figure [ fig : mc10rv_curve_bd ] ( bottom panel , red points ) near hjd 2455250 .
the final ( correct ) orbit solution is also shown ( see details below ) , but it must be noted that this final orbit solution is only a good fit after properly disentangling the rvs from epochs where just a single set of spectral lines is resolved ; it is not a good fit to the _ directly observed _ single - lined rv measurements , since these are in fact a flux - weighted average of the true primary and secondary rvs .
the six outlier " measurements from this first set of observations ( four marvels points and two arces points ) appear systematically displaced by 1520 km s@xmath7 relative to the other 38 measurements , which are well fit by the spurious orbit ( solid curve ) but not by the correct orbit ( dashed curve ) . ) on the mass of companions as a function of angular separation . with this upper limit , we can rule out the presence of companions above a mass of @xmath27 outside of @xmath550 au , and above a mass of @xmath33 outside of @xmath5100 au .
, scaledwidth=38.0% ] lcrcr @xmath34 & & @xmath35 & & @xmath36 + @xmath37 & & @xmath38 & & @xmath39 + @xmath40 & & @xmath41 & & @xmath42 + @xmath43 & & @xmath44 & & @xmath45 + @xmath46 ( km s@xmath7 ) & & @xmath47 & & @xmath48 + @xmath49 ( km s@xmath7 ) & & ... & & @xmath50 + @xmath51 ( km s@xmath7 ) & & @xmath52 & & @xmath53 + @xmath54 & & ... & & @xmath55 + cccc @xmath18 ( 2000 ) & 11 00 11.45 & & ( 1 ) + @xmath56 ( 2000 ) & + 39 43 24.74 & & ( 1 ) + pmra [ mas yr@xmath7 ] & @xmath58 & 1.7 & ( 1 ) + pmde [ mas yr@xmath7 ] & @xmath59 & 1.6 & ( 1 ) + b@xmath60 & @xmath61 & @xmath62 & ( 1 ) + v@xmath60 & @xmath63 & @xmath64 & ( 1 ) + @xmath65 & @xmath66 & @xmath67 & ( 2 ) + @xmath68 & @xmath69 & @xmath70 & ( 2 ) + @xmath71 & @xmath72 & @xmath73 & ( 2 ) + @xmath74 & @xmath75 & @xmath76 & ( 2 ) + @xmath77 & @xmath78 & @xmath79 & ( 2 ) + @xmath80 & @xmath81 & @xmath82 & ( 2 ) + @xmath83 & 9.977 & 0.021 & ( 3 ) + @xmath84 & 9.554 & 0.016 & ( 3 ) + @xmath85 & 9.488 & 0.019 & ( 3 ) + wise1 ( 3.4 @xmath86 m ) & 9.407 & 0.006 & ( 4 ) + wise2 ( 4.6 @xmath86 m ) & 9.482 & 0.006 & ( 4 ) + wise3 ( 12 @xmath86 m ) & 9.470 & 0.038 & ( 4 ) + k , @xmath87 , [ fe / h]@xmath88 , and @xmath89 , yielding a distance of @xmath90 pc . _
bottom : _ a second nextgen fit that uses two stellar components ( corresponding to the primary and secondary stars of tyc 3010 ) with one of the components constrained to the spectroscopically determined stellar parameters for the primary ( @xmath91 k , @xmath92 , [ fe / h]@xmath93 ) .
this fit estimates the secondary stellar parameters to be @xmath94 k , @xmath95 , @xmath96 , and the distance to tyc 3010 to be @xmath97 pc , with an @xmath98 ( @xmath99dof@xmath100 ) .
+ , title="fig:",scaledwidth=35.0% ] k , @xmath87 , [ fe / h]@xmath88 , and @xmath89 , yielding a distance of @xmath90 pc .
_ bottom : _ a second nextgen fit that uses two stellar components ( corresponding to the primary and secondary stars of tyc 3010 ) with one of the components constrained to the spectroscopically determined stellar parameters for the primary ( @xmath91 k , @xmath92 , [ fe / h]@xmath93 ) .
this fit estimates the secondary stellar parameters to be @xmath94 k , @xmath95 , @xmath96 , and the distance to tyc 3010 to be @xmath97 pc , with an @xmath98 ( @xmath99dof@xmath100 ) .
+ , title="fig:",scaledwidth=35.0% ] in addition , as we have done with all marvels candidates , we performed a fit to the spectral energy distribution ( sed ) of the system to verify that it is consistent with a single stellar source and to provide a consistency check on the spectroscopically determined stellar properties ( see below ) .
we constructed the sed using fluxes ( see table [ tbl : mc10ab_cat_props ] ) from the tycho catalogue @xcite , apass ( _ _ a__avso _ _ p__hotometric _ _ a__ll-__s__ky _ _ s__urvey ; data release 6 , see * ? ? ?
* ) , two micron all sky survey @xcite , and _ wise _ @xcite .
nextgen models ( hauschildt et al . 1999 ) are used to generate theoretical seds by holding @xmath101 , @xmath102 , and [ fe / h ] at the spectroscopically determined values ( see below ) , and the maximum extinction @xmath103 was limited to 0.05 mag based on the dust maps of @xcite .
the best fit model can be seen in the top panel of figure [ fig : sed ] ; it corresponds to an @xmath103 of @xmath104 , and a distance of @xmath105 pc .
this single - star sed fit to the available photometry spanning 0.212@xmath86 m is quite good , with the only hint of a discrepancy being a mild excess that appears in the _ galaxy evolution explore ( galex ) _ near - uv ( nuv ) passband , despite the lack of any strong emission in the observed ca hk lines . however , this by itself was not deemed to be a compelling reason to suspect the high quality orbit solution .
thus , at this point in our analysis , fully 38 rv measurements from two separate instruments were well fit by the same orbit solution of a single , solar - type star with a @xmath106 companion on a modestly eccentric orbit .
the sed of tyc 3010 was furthermore consistent with being a single solar - type star , and the lack of any companions in the high - resolution imaging ruled out a blend scenario in which the rv variations might be caused by a binary beyond 0.5@xmath30 of the line of sight .
only four of the discovery rv measurements appeared to be discrepant , and these were rejected for what appeared to be good reasons , behaving not unlike fiber mis - pluggings that the marvels team had observed in other stars before .
however , the last two rv measurements from the first set of arces observations appeared as strong outliers ( see figure [ fig : mc10rv_curve_bd ] , blue points near hjd 2455730 ) . as they were observed with a standard echelle spectrograph , these could not be attributed to fiber mis - pluggings , and inspection of the ccfs revealed double lines ( see bottom panel of figure [ fig : ccfs ] ) , immediately nullifying the bd companion hypothesis .
238 days , an eccentricity of @xmath50.79 , with @xmath10715.38 km s@xmath7 and @xmath10817.50 km s@xmath7 . finally , for the rv points outside of periastron , it was necessary to de - blend the observed rvs with the method described in section [ sec : rv_fitting ] .
+ , scaledwidth=45.0% ] to further confirm that tyc 3010 was indeed a stellar binary , we closely observed the next peripassage with the hrs spectrograph on het . with het , we obtained complete coverage of periastron , permitting a complete double - lined orbit solution . in this section we present the correct orbit solution for tyc 3010 ,
including all the points from the discovery and subsequent data , which shows that tyc 3010 is an sb2 with a period of @xmath3 days , an eccentricity of @xmath109 , and a mass ratio of @xmath11 . with this eccentricity and orbital period , tyc 3010
lies near the upper bound of ( but within ) the distribution of orbital eccentricities of solar - type binaries with orbital periods of 100300 days ( see , e.g. , * ? ? ?
* ; * ? ? ?
the orbital parameters for the binary are summarized in table [ tbl : orb_params ] , the rv solution is shown in figure [ fig : mc10rv_curve_sb ] , and a schematic of the orbit is shown in figure [ fig : mc10_schematic ] . in this section
we also describe our determination of the stellar parameters for the primary in tyc 3010 , and we estimate its mass and radius using the relations described in @xcite . since the secondary is comparable in mass to the primary , we had to take special care in accounting for the flux contamination from the secondary , both in our determination of the stellar parameters and with the rv values that we measured for the system outside of periastron .
for the orbital solution of the binary , we used the rv fitting software described in @xcite .
since we do not resolve two sets of spectral lines for the phases outside of periastron , most of the rv points correspond to a flux - weighted average of the primary and secondary rvs . in
order to de - blend the flux - weighted rvs that we measured , and derive the corresponding primary rvs , we used the following prescription . we treat the blended velocities as a flux - weighted average of the primary and secondary velocities : @xmath110 where @xmath111 and @xmath112 are the primary and secondary velocities respectively , and @xmath113 and @xmath114 are the primary and secondary fluxes .
we normalize the flux weights by setting the sum of the fluxes , @xmath115 , to unity . using the flux ratio , @xmath116 , from the todcor analysis ( which was only performed for the het / hrs epochs where it was possible to resolve two sets of spectral lines )
, we can solve for @xmath113 and @xmath114 in terms of @xmath18 : @xmath117 in addition , we can use the mass ratio , @xmath118 , from the rv solution to write @xmath112 in terms of @xmath111 , since @xmath119 .
@xmath120 returning to ( [ eqn : vblend ] ) , we can now write @xmath121 with equation ( [ eqn : deblend ] ) , we can iteratively solve for a final set of de - blended rvs for the primary .
for the first iteration , we provide an initial guess for @xmath122 by performing a joint fit to the primary rvs ( blended@xmath123unblended ) combined with the secondary rvs ( unblended ; only measured during periastron ) . inserting this initial guess for @xmath122 into equation ( [ eqn : deblend ] )
, we derive an initial set of de - blended primary rvs .
then we perform another joint fit to the primary ( de - blended+unblended ) and secondary ( unblended ) rvs to refine our value for @xmath122 .
we repeat the process until @xmath122 converges .
the value we find for @xmath122 ( @xmath124 ) from this de - blending analysis is in excellent agreement with the value for @xmath122 ( @xmath50.88 ) that we found from the ratio of the primary and secondary rvs that were measured for the 11 het / hrs epochs where two peaks were resolved in the ccfs .
thus , @xmath122 has been determined very precisely by the orbital solution ( better than 3% ) , and is more precise than the individual quoted errors on the masses . as a further consistency check on @xmath18 and @xmath122
, we also note that according to the relationship between mass and bolometric luminosity from @xcite , there should be a relationship between @xmath18 and @xmath122 . since @xmath18 is derived from a set of finite wavelength bands , it is not bolometric .
however , since the stars have temperatures that are not too dissimilar , @xmath18 is likely to be approximately equal to the ratio of the bolometric luminosities . for stars with @xmath125 , a fit to the @xcite data yields @xmath126 .
thus , @xmath127 , so @xmath128 , which is within 3@xmath26 of the value obtained from the rv analysis .
the stellar parameters for the primary were determined with a double - lined spectrum obtained near periastron ( see section [ sec : arces_data ] ) .
the spectroscopic analysis used to determine the atmospheric parameters is similar to the one described in @xcite , where we use two independent methods that require the conditions of excitation and ionization equilibria for and lines .
these methods are referred to as the `` bpg '' ( brazilian participation group ) method and the `` iac '' ( instituto de astrofsica de canarias ) method .
crr parameter & value & uncertainty + @xmath129 & 0.335 & 0.035 + @xmath54 & 0.878 & 0.016 + a@xmath130 & 0.03 & 0.02 + @xmath131 ( pc ) & 225 & 40 + & tyc 3010 a & tyc 3010 b + @xmath24 ( k ) & @xmath132 & @xmath133 + @xmath102 ( cgs ) & @xmath134 & @xmath135 + @xmath136fe / h@xmath137 $ ] & @xmath138 & ... + @xmath139 ( @xmath140 ) & @xmath141 & @xmath142 + @xmath143 ( @xmath144 ) & @xmath145 & @xmath146 the `` bpg '' analysis was done in local thermodynamic equilibrium ( lte ) using the 2002 version of moogchris / moog.html ] @xcite and one - dimensional plane - parallel model atmospheres interpolated from the odfnew grid of models @xcite . in previous marvels papers ( e.g. , * ? ? ?
* and references therein ) , the equivalent widths ( ews ) of the fe lines were determined in an automated fashion . however , in this case , the ews were manually measured to carefully account for visible blends on the fe lines from the secondary s spectrum .
we note that contaminations from very weak lines could have affected the ew measurements . in order to correct the ews measured for the primary for the veiling from the continuum flux of the secondary star
, we followed a procedure similar to the one described in section 5.2.1 of @xcite . according to their prescription ,
we can relate the value of the true equivalent width ( ew@xmath147 ) of a given line to the observed equivalent width ( ew@xmath148 ) through the following relationship , @xmath149 where @xmath150 is the so - called veiling factor for the primary .
the veiling factors for the two components are related by @xmath151 where @xmath152 and @xmath153 are the fluxes for the primary and secondary .
furthermore , the veiling factors satisfy the equation @xmath154 to simplify our analysis , we treated the veiling factors and flux ratio as if they were wavelength independent . using the average flux ratio derived by todcor ( @xmath155 ;
see section [ sec : todcor ] ) , and the added constraint from equation [ eqn : veiling_norm ] , we find the veiling factor for the primary to be @xmath156 .
thus , after correcting the ews , we find the stellar parameters to be @xmath91 k , @xmath92 , and [ fe / h]@xmath93 ( see table [ tbl : mc10ab_sys_props ] )
. the uncertainties for these parameters are larger than the typical errors that we achieve with our spectroscopic analysis because of the flux contamination from the secondary star .
the `` iac '' analysis extracted the stellar parameters of the primary and secondary stars by considering veiling factors that were wavelength - dependent .
these veiling factors are estimated using low - resolution kurucz fluxes ( * ? ? ?
* and references therein ) and the following equation : @xmath157 where @xmath158 and @xmath159 correspond to the surface brightness of the primary and the secondary respectively . to determine the ratio of the radii , we derived an empirical mass radius relationship from a sample of 55 stars from @xcite , with the masses restricted to 0.7 m@xmath160 1.4 m@xmath161 .
we fit a function to the data of the form @xmath162\ ] ] where @xmath163 and @xmath164 .
thus , the ratio of the radii for the components of tyc 3010 can be written as @xmath165 the mass ratio was determined from the todcor analysis to be @xmath166 , so we find that @xmath167 . as a first guess , we adopt the above values to estimate the stellar mass and radius of the primary @xcite , from solar - scaled theoretical isochrones @xcite .
the mass ratio allows us to derive a first guess of the @xmath168 value for the secondary to be roughly 5100 k , assuming @xmath169 and the same metallicity as the primary .
the stellar radii we get from the comparison with isochrones are 0.89 @xmath170 and 0.77 @xmath170 , and thus the ratio is @xmath171 , which is very similar to the value previously estimated ( @xmath172 ) .
thus , the derived veiling factors lie in the range @xmath173 and @xmath174 in the spectral region 45007000 .
we then measure automatically , using the code ares @xcite , the ews of the and lines @xcite for both stellar components and correct them using the wavelength - dependent veiling factors .
we then use the code stepar @xcite to automatically derive the stellar parameters of each component and we get @xmath175 k , @xmath176 , [ fe / h]@xmath177 and @xmath178 from 162 and 18 lines .
the uncertainties are unexpectedly large and may be due to the contamination of neighboring lines of other elements of the companion star .
thus the results for the secondary are fairly tentative and the errors are even larger .
we were only able to measure 64 and 3 lines to get @xmath179 k , @xmath180 , [ fe / h]@xmath181 and @xmath182 .
compared to the `` bpg '' analysis , the lower @xmath183 of the primary may be related to the different methods used to derive the veiling factors .
nevertheless , the `` iac '' stellar parameters for the primary star are very similar to those previously derived and are actually consistent within the large uncertainties so we decide to adopt the `` bpg '' values .
, @xmath184 ) , and the error bars correspond to the 68.27% confidence intervals .
the contours are lines of equal probability density which enclose 68% , 90% , and 95% of the cumulative probability relative to the maximum of the probability density . in the top and
right panels , the probability distribution ( solid line ) and cumulative probability ( dashed line ) are shown for the mass and radius respectively.,scaledwidth=38.0% ] with the `` bpg '' stellar parameters for the tyc 3010 primary , we again performed a fit to the observed sed of the system as in section [ sec : spurious ] , but now also including the contribution of the secondary star .
once again , nextgen models @xcite are used to generate theoretical seds by holding @xmath101 , @xmath102 , and [ fe / h ] at the spectroscopically determined values for the primary , while the @xmath101 for the secondary is found by the value that minimizes @xmath185 ( @xmath99dof@xmath100 ) .
the best fit model can be seen in the bottom panel of figure [ fig : sed ] ; it corresponds to an @xmath103 of @xmath186 , and a distance of @xmath187 pc .
compared to the sed fit performed in section [ sec : spurious ] , which assumed a single stellar contribution , this two - component sed fit no longer exhibits an excess in the _ galex _ nuv passband , and more generally is an excellent fit to all of the available photometry .
finally , from this two - component fit to the sed , we also obtain a set of values for the stellar parameters of the secondary of tyc 3010 .
we find that @xmath188 k , @xmath189 , and [ fe / h]@xmath190 . and [ fe / h]@xmath191 .
ages ( in gyr ) of 1.0 , 5.0 , 8.0 , and 11.0 are represented by blue dots , and the 1@xmath26 deviations from the evolutionary track are shown in the shaded region.,scaledwidth=34.0% ] given the spectroscopic stellar parameters , we can derive the mass and radius of the tyc 3010 primary star using the empirical relationships described in @xcite .
figure [ fig : mc10_mass_radius ] shows the result of a set of mcmc trials for the best estimate of the mass and radius . for the precise parameters of the primary ( @xmath192 k , @xmath193 , [ fe / h]=0.09 ) , the torres relations give 0.98 @xmath140 and 0.75 @xmath144 .
once one includes the fairly large uncertainties in the stellar parameters , the median values for the mass and radius become @xmath194 and @xmath195 , respectively .
the means are @xmath196 and @xmath197 , so the distributions are quite skewed as shown in figure [ fig : mc10_mass_radius ] . compared to a yonsei - yale evolutionary track ( see figure [ fig : y2_models ] ) , we do not have a strong constraint on the age , but tyc 3010 is unlikely to have evolved off the main sequence . we can also derive the mass and radius for the secondary given the stellar parameters determined from the two - component sed fit and the @xcite relations .
we find that @xmath198 and @xmath199 .
this value for the mass of the secondary agrees within 1@xmath26 of the value that can be derived using the primary mass we determined above and the mass ratio from the rv solution , i.e. , @xmath200 .
the rv signal from tyc 3010 initially seemed to indicate that it was a bd orbiting a solar - type star in the bd desert . over 80% of the marvels discovery data agreed with this interpretation , and there seemed to be plausible reasons for excluding the outliers .
however , once similar outliers were found in the subsequent observations , we began to suspect the validity of the bd interpretation . in this section , we discuss in detail why we initially favored the bd interpretation , as well as how this conclusion was abruptly overturned by a few surprising data points . in the discovery data
, there were four outliers in total , each offset by @xmath520 km s@xmath7 from the rest of the data .
the most anomalous of the outliers was extracted from a spectrum with a low s / n , so its rv value did not seem trustworthy . the remaining outliers
( considering that they corresponded to a @xmath520 km s@xmath7 offset in rv that was only captured once during the three orbits contained in the discovery data ) , also seemed likely to be spurious .
the marvels spectrograph is a fiber - fed spectrograph that can observe 60 objects simultaneously .
each fiber is plugged by hand to observe the correct target , and occasionally a mistake may occur . indeed , the marvels data vetting procedures were evolved to specifically include an outlier rejection step that sought to mitigate such errors , by searching for consecutive strings of measurements that were offset from the bulk of the data in a similar fashion to how these four measurements behave . remarkably , excluding these few apparent `` outliers''and in fact _ only _ by excluding them permits a convincing orbit solution .
it is not intuitive that this should be the case , in particular because only @xmath515% of the measurements are excluded ( including both the discovery data and the initial follow - up data which appeared to corroborate the spurious solution ) and because the resulting solution is so dramatically different from the true solution .
evidently , a system such as tyc 3010 ( with its extreme eccentricity , leading to punctuated large rv excursions , and its orbital orientation being nearly perpendicular to the line of sight , leading to very small rv variations for @xmath595% of the orbit ) is able to mimic a more circular orbit of a low - mass companion about a single star .
moreover , the similarity of the two stars in tyc 3010 leads to a combined light sed that is only slightly different from that of a single star at a nearer distance .
thus many lines of evidence supported the initial solution , considering that the bd interpretation appeared to be supported by two years of discovery rv data , six months of additional rv observations , lucky imaging , and a well - constrained sed . indeed , when the two follow - up rv measurements observed near periastron appeared , indicating a possible problem with the original orbit solution , we began to search for reasons to suspect the validity of these two anomalous points .
at first , we thought the situation might be similar to the fiber mis - pluggings believed to have occurred with the discovery data , and we considered that the arces outliers were the result of pointing at the wrong star . but after investigating the data from those two nights , we confirmed that we had observed the correct target .
next we learned of a recent change that had been made to the arces instrument : the thar lamp had recently been replaced .
the thar lamp is used to perform the wavelength calibration , and it was plausible that the new lamp might have caused problems with the wavelength solution .
therefore , the arces outliers may have merely been the result of an artificial doppler shift generated by an incorrect wavelength solution . in the end
, we were only able to accept that the bd interpretation was incorrect after we inspected the ccf for each of the outliers .
the ccfs for the outliers both showed two peaks instead of one , indicating the presence of a second stellar component .
furthermore , the secondary peak was comparable in height to the primary peak ( see bottom panel of figure [ fig : ccfs ] ) , which led us to suspect that tyc 3010 was in fact a spectroscopic stellar binary .
but how did most of the data that we had for tyc 3010 conspire to imply that it was a much less massive system ?
the period , shape , and orientation of the orbit with respect to the line of sight ( see figure [ fig : mc10_schematic ] ) made it such that for most of the orbit the two stars possess relatively low rvs with respect to each other .
in particular , the difference between the magnitude of their rvs is smaller than the typical ccf width for our instruments , resulting in their ccf peaks being blended into one .
since the flux ratio is not too different from unity , and the mass ratio is also close to unity , for epochs where the spectral lines are blended , there is a near - cancellation ( or strong suppression ) of the true orbital velocities for the primary and secondary , which are nearly equal in magnitude but oppositely signed ( see equation [ eqn : deblend ] , and recall that @xmath201 is what we actually measure ) .
thus , for @xmath595% of the orbit , the amplitude of the variations ( @xmath512 km s@xmath7 ) suggest a bd companion to a solar - type star ; furthermore , the eccentricity and the orbital period ensure that the stars spend a long time ( @xmath57 months ) away from periastron , which is precisely the moment when the rvs of the components are disparate enough for it to be fairly easy to resolve the two sets of spectral lines , and the large rv amplitude ( @xmath51520 km s@xmath7 ) is indicative of a stellar binary with two solar - type stars .
moreover , the orientation makes it so that only a relatively small component of the orbital velocities is directed along our line of sight . finally , the cadence of the marvels survey made it unlikely to observe multiple epochs of periastron .
for any given rv survey , the lower the resolution of the spectrograph , the more vigilant one must be for these kinds of false positives .
for tyc 3010 in particular , a spectrograph with a resolution of @xmath202 is required to resolve the spectral lines throughout most of the orbit .
but in general , as the resolution ( and cadence of observations ) decreases , the wider the range of eccentricities , arguments of periastron , and orbital periods by which stellar binaries could masquerade as substellar companions for significant fractions of their orbits . furthermore , longer period orbits ( @xmath203yr ) should be handled with special care , for in these cases the phase coverage is more likely to be incomplete . in order to survey @xmath53,000 stars over four years , marvels required a cadence that made it less likely to observe multiple epochs of periastron for a binary with the period of tyc 3010 . for marvels and similar rv surveys for substellar companions
, it can be costly to use precious resources to examine false positives .
therefore , in this section , we describe a method that the marvels team currently employs to identify binaries like tyc 3010 during the candidate - vetting process . for typical rv surveys today , a standard line bisector analysis can usually be performed to assess the presence of blended double - lined binaries . however , this was not possible for the marvels discovery data due to its limited spectral resolution .
thus , following our experience with tyc 3010 , marvels has developed an internal pipeline for inspecting the widths of the ccf peaks for all of our candidates .
this way , we can readily monitor the ccfs for signs that indicate that there may be more than one stellar component present ( e.g. , the large excursions in the width of the ccf peak that occur near periastron for tyc 3010 ; see figure [ fig : ccf_widths ] ) .
there are two properties of the ccfs that we now monitor : ( 1 ) the average width of the ccf peak compared to other stars in the survey , and ( 2 ) any other significant changes in the shape of the ccf over time . for a typical solar - type star that is not rotating too rapidly ( i.e. , the kinds of stars that marvels targets )
, one would expect the width of the ccf peak to be @xmath510 km s@xmath7 , which is largely the result of thermal broadening and micro - turbulence .
however , when binary systems like tyc 3010 are unresolved , the widths of the ccf peak are broader ( @xmath520 km s@xmath7 ) , indicating that there may be multiple stellar components contributing to the flux from the system ( see figure [ fig : ccf_widths ] ) .
in fact , an atypically broad ccf peak could also be the result of a single star rotating atypically fast , so a broad peak is not in itself sufficient to identify the system as a binary .
nevertheless , a broad peak should be taken as a sign to proceed with caution .
furthermore , changes in the skewness of the ccf peak might provide an even more sensitive diagnostic for these kinds of systems .
thus , by monitoring changes in the ccf peak , even if one misses the small fraction of the orbit where , depending on the resolution , the ccf peak either broadens dramatically or separates into distinct peaks ( or if one is suspicious of the relatively few epochs where the system happened to be caught near periastron ) , it is possible to flag systems like tyc 3010 , which may contain much more mass than most of the rv data suggests .
the case of tyc 3010 is also a pertinent lesson on how important it is to handle outliers carefully , especially in this era of large surveys where thousands of objects must be screened for the most favorable candidates .
we possessed plausible reasons for suspecting that the outliers in the discovery data might be spurious ( known issues with fiber mis - pluggings ; low s / n ; and the outliers were only detected during one of the three orbits observed ) .
moreover , and perhaps ironically , the spurious orbit solution is actually a better fit to the discovery data ( excluding the outliers ) than the true orbit solution , because of the need to disentangle the primary and secondary rv components from the ( apparently ) single - lined rv measurements .
however , even when faced with such a compelling initial solution and sensible reasons for considering the outliers to be invalid , it is imperative to investigate further and provide evidence that the reasons for rejecting the outliers are not only plausible but justified .
furthermore , when the analysis is distributed among multiple team members like it is within marvels , it is necessary to make sure each step of the analysis is documented as clearly as possible . for marvels ,
the members who perform the candidate - vetting are usually different from those who perform the subsequent analysis for each candidate , so it is important for each team member to be able to readily discover if any outliers were rejected and why .
marvels has now modified its internal analysis tracking system in order to make the entire analysis process more transparent . finally , if we had been monitoring the widths of the ccf peaks , we could have considered the evidence of the broad peak , as well as the changing peak width around periastron , though in truth neither the changing width nor the broad peak by themselves would have likely been sufficiently compelling to reject the initial orbit solution . in the end ,
the most important part of our analysis was to strategically focus our het / hrs observations on periastron , the phase where the outliers occurred and where it was easiest to resolve the spectral lines .
this strategy would have been more difficult with a conventionally scheduled telescope , but was readily achieved with the queue - scheduled nature of the het .
we have demonstrated , using high resolution spectroscopy , that tyc 3010 is an sb2 .
we have shown how , with a spectrograph below a given resolution ( @xmath204 ) , the eccentricity and the orientation of the system with respect to our line - of - sight allowed a large fraction of the rv curve to appear remarkably similar to the kind of signal one would expect from a bd secondary as opposed to a stellar - mass secondary . furthermore , as a result of the cadence of the marvels survey and the orbital period of the system , we were more likely to miss periastron during a given orbit .
thus , we were more susceptible to rejecting the periastron points we did obtain as outliers , even though these points are where the spectral lines are most widely separated , and thereby where it is easiest to determine that the system is an sb2 . finally , we concluded with a word of warning to rv surveys , since for a given resolution and cadence , there are a range of orbital parameters that can make a stellar - mass binary companion appear to be substellar . the lower the resolution or cadence , the greater the number of stellar binaries that can masquerade in a fashion similar to tyc 3010 . therefore ,
if other surveys can carefully monitor the widths of the ccf peaks for their targets ( or monitor their line bisectors if they have high enough resolution ) , and when possible , focus their resources on observations of peripassage , then we hope that they will be able to avoid similar astrophysical false positives .
this research was partially supported by the vanderbilt initiative in data - intensive astrophysics ( vida ) and nsf career grant ast 0349075 ( cem , kgs , lh , jp ) , nsf aapf ast 08 - 02230 ( jpw ) , nsf career grant ast 0645416 ( ea ) , cnpq grant 476909/2006 - 6 ( gfpm ) , faperj grant apq1/26/170.687/2004 ( gfpm ) , nsf career grant ast-1056524 ( bsg , jde ) , and a papdrj capes / faperj fellowship ( lg ) .
based on observations with the sdss 2.5-meter telescope .
funding for the marvels multi - object doppler instrument was provided by the w.m .
keck foundation and nsf grant ast-0705139 .
the marvels survey was partially funded by the sdss - iii consortium , nsf grant ast-0705139 , nasa with grant nnx07ap14 g and the university of florida .
the center for exoplanets and habitable worlds is supported by the pennsylvania state university , the eberly college of science , and the pennsylvania space grant consortium .
data presented herein were obtained at the hobby - eberly telescope ( het ) , a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen .
the het is named in honor of its principal benefactors , william p. hobby and robert e. eberly .
this research has made use of the simbad database , operated at cds , strasbourg , france , and the aavso photometric all - sky survey ( apass ) , funded by the robert martin ayers sciences fund .
it also made use of the iraf software distributed by the national optical astronomy observatory , which is operated by the association of universities for research in astronomy ( aura ) under cooperative agreement with the national science foundation .
this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation .
we also make use of data products from the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration .
funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s .
department of energy office of science .
the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , university of florida , the french participation group , the german participation group , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . ,
j. , lee , b. , de lee , n. , et al .
2009 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7440 , society of photo - optical instrumentation engineers ( spie ) conference series , t. j. 2004 , in astronomical society of the pacific conference series , vol .
318 , spectroscopically and spatially resolving the components of the close binary stars , ed .
r. w. hilditch , h. hensberge , & k. pavlovski , 159165 , a. , rebolo , r. , lpez , r. , et al .
2008 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 7014 , society of photo - optical instrumentation engineers ( spie ) conference series , f. , mayor , m. , delabre , b. , et al .
2000 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4008 , society of photo - optical instrumentation engineers ( spie ) conference series , ed .
m. iye & a. f. moorwood , 582592 , l. w. , adams , m. t. , barnes , t. g. , et al .
1998 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol .
3352 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . l. m. stepp , 3442 , r. g. 1998 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 3355 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . s. dodorico , 387398 , s .-
hildebrand , r. h. , hobbs , l. m. , et al .
2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4841 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . m. iye & a. f. m. moorwood , 11451156 | we report the discovery of a highly eccentric , double - lined spectroscopic binary star system ( tyc 3010 - 1494 - 1 ) , comprising two solar - type stars that we had initially identified as a single star with a brown dwarf companion . at the moderate resolving power of the marvels
spectrograph and the spectrographs used for subsequent radial - velocity ( rv ) measurements ( @xmath0 ) , this particular stellar binary mimics a single - lined binary with an rv signal that would be induced by a brown dwarf companion ( @xmath1 ) to a solar - type primary .
at least three properties of this system allow it to masquerade as a single star with a very low - mass companion : its large eccentricity ( @xmath2 ) , its relatively long period ( @xmath3 days ) , and the approximately perpendicular orientation of the semi - major axis with respect to the line of sight ( @xmath4 ) . as a result of these properties , for @xmath595% of the orbit the two sets of stellar spectral lines are completely blended , and the rv measurements based on centroiding on the apparently single - lined spectrum is very well fit by an orbit solution indicative of a brown dwarf companion on a more circular orbit ( @xmath6 ) . only during the @xmath55% of the orbit near periastron passage does the true , double - lined nature and large rv amplitude of @xmath515 km s@xmath7 reveal itself .
the discovery of this binary system is an important lesson for rv surveys searching for substellar companions ; at a given resolution and observing cadence , a survey will be susceptible to these kinds of astrophysical false positives for a range of orbital parameters . finally , for surveys like marvels that lack the resolution for a useful line bisector analysis , it is imperative to monitor the peak of the cross - correlation function for suspicious changes in width or shape , so that such false positives can be flagged during the candidate vetting process . |
harvesting with optimum packet scheduling policy is significantly important for increasing the lifetime and sustainability of wireless sensor networks ( wsns ) and to achieve the demands of green communications @xcite .
the scarcity and sporadic availability of the energy make it necessary to store it and utilize optimally .
therefore , optimum power management and data transfer schemes are significantly important for wsns .
the correlation of data observed in wsns is one of the most important factors to be taken into account for designing optimum power scheduling algorithms for the energy harvesting transmitters in order to consume the available resources more efficiently @xcite . in this article ,
previous optimal solutions defined for gaussian mac are extended to include common data observed at the energy harvesting transmitters .
optimum online and offline packet scheduling in energy harvesting communication systems are recently investigated for single hop , multiple - access channel ( mac ) and broadcast systems .
optimal packet scheduling for single - user energy harvesting communication systems are presented in @xcite . in @xcite ,
a directional water - filling algorithm optimizes the throughput for a single - user fading channel with additive gaussian noise with finite capacity rechargeable batteries under offline and online knowledge . in @xcite ,
a two - hop relaying communication network with energy harvesting rechargeable nodes is formulated for the offline end - to - end throughput maximization as a convex optimization problem . besides that
, similar analyses are achieved for the mac schemes . in @xcite ,
optimal packet scheduling problem is solved in a 2-user mac system with energy harvesting transmitters where the energy harvesting times and harvested energy amounts are known before the transmission .
the karush - kuhn - tucker ( kkt ) solution and the generalized iterative backward water - filling algorithm are presented .
however , these studies do not consider common data and correlation among the collected sensor data which is the main contribution achieved by the article .
in addition , the optimum policy is not compared with no power scheduling case and uniform power scheduling polices .
furthermore , in @xcite , the proposed optimum scheduling policies are extended by including one - way energy transfer capability between two transmitter nodes .
moreover , in @xcite , the capacity region of gaussian mac with amplitude constraints and batteryless energy harvesting transmitters are analysed . on the other hand , in @xcite , optimal continuous - time online power policies for energy harvesting macs
are presented . in @xcite , energy harvesting transmitter and receiver pair
is considered in a utility maximization framework achieving power policy using a water - filling approach . in @xcite , optimal transmit power policy for energy harvesting transmitters in a gaussian mac
is presented by also considering storage losses .
however , these studies do not consider the common data in an optimum packet scheduling framework .
on the other hand , data correlation in wsns are significantly important to save the power - bandwidth resources @xcite . in @xcite , the explicit characterization of the capacity region in a gaussian mac channel with common data and fading
is considered .
the optimum power allocation achieving the arbitrary rate tuples on the boundary of the capacity region are presented and numerically computed .
however , the study does not consider optimum power scheduling and operates only in a single time interval .
the current study extends the optimum solution to include multiple time intervals in an optimum power scheduling policy as the main contribution .
moreover , in @xcite , the capacity region of the discrete p - transmitter / q - receiver mac defined as general mac ( gmac ) with a common message is derived as a generalization of @xcite . in @xcite
, information - theoretic results and power allocation policies in combination with joint source - channel codes on the transmission of memoryless dependent sources through a memoryless fading mac are analysed . in @xcite , 2-user mac with common message ( maccm ) and mac with conferencing encoders ( macce ) with channel state information ( csi )
are analyzed .
the capacity results for the gaussian mac with cooperative encoders and with additive interference known non - causally to both encoders are presented .
however , none of these studies combine energy harvesting and optimum packet scheduling in a gaussian mac with common data . to the best of our knowledge , in this work
, optimum offline packet scheduling solution , for the first time , is given for gaussian mac with common data and energy harvesting transmitters .
the kkt solution is given , for the first time , for optimum packet scheduling problem for gaussian mac with common data and energy harvesting transmitters .
an efficient iterative water - filling algorithm is introduced for the optimum solution although the optimum solution and the determination of the water levels are more complicated compared with the mac without common data .
the departure region or the capacity boundary surface for the data rates of the individual and the common data messages is numerically simulated by using the defined optimum water - filling algorithm .
the optimum packet scheduling solution is compared with the solutions defined for gaussian mac without common data , the case for no power scheduling and uniform power scheduling framework by showing the advantages of the proposed solution .
the remainder of the paper is organized as the following . in section [ sysmodel ] , power scheduling policy and the system model for the gaussian mac with common data
are defined .
then , in section [ ops ] , data throughput maximization problem is defined . in section [ optw ] , kkt solution for the defined problem
is proposed and efficient iterative water - filling algorithm is presented . in section [ simulr1 ] ,
a simulation study is performed illustrating optimum scheduling policy , the departure region boundary surfaces and the comparison of the proposed solution with no scheduling and uniform power scheduling cases , and the comparison with the previous solutions defined for mac without common data .
then , in section [ fw ] , future work and open issues are discussed .
finally , in section [ conc ] , the conclusions are given .
+ in this paper , energy harvesting gaussian mac with two transmitters and one receiver is considered as shown in fig .
[ fig1 ] @xcite while incorporating the common data model given in @xcite .
each user has their individual data packets and also a common message known by both transmitters .
it is assumed that the amounts of harvested energy and the harvesting times are known before the data transmission .
similar to @xcite , energy harvesting times are put in ascending order and the length of the time interval between two energy harvesting instants @xmath0 and @xmath1 is denoted by @xmath2 while energy harvesting starts at @xmath3 and ends at @xmath4 .
it is assumed that the final deadline time instant to transmit data bits is @xmath5 with the @xmath6th time interval length being equal to @xmath7 .
for example , 1st user harvests @xmath8 at the time instant @xmath9 and 2nd user harvests @xmath10 at the time instant @xmath1 possibly both users harvesting energy at any single time instant .
an illustrative energy harvesting scenario is shown in fig .
[ fig0 ] .
+ in this article , stochastic nature of energy harvesting and the time - varying channel states are not formulated due to complexity of the issue .
the harvested energy is stored in sensor nodes and the problem is simplified by assuming that the data packets are available before the transmission and information about the energy harvesting times is available .
therefore , in this article we consider an offline solution with a deterministic system setting instead of an online solution with stochastic energy , data and channel states leaving the consideration as a future work .
the target is to find the maximum data throughput regions , i.e. , maximum departure region @xcite , for three independent messages , i.e. , @xmath11 , @xmath12 and @xmath13 , where @xmath11 is known by both the users , for any given deadline time @xmath5 and propose a water - filling algorithm finding the optimal solution . the system model for the gaussian mac
is shown in fig .
the data packets @xmath14 are available at the beginning before the transmission .
the packets are modulated into message sequences @xmath15 and 1st node knows both @xmath11 and @xmath12 while 2nd node knows both @xmath11 and @xmath13 . the transmitted symbols , i.e. , @xmath16 , is a function of @xmath17 for @xmath18 $ ] .
each node @xmath19 for @xmath18 $ ] has the available power @xmath20 in units of ( w ) for transmission .
each symbol consists of the addition of the symbols for the independent message , i.e. , @xmath21 , and the common message , i.e. , @xmath22 .
each node @xmath19 assigns the power levels @xmath23 to its own symbol for the independent message and @xmath24 to the symbol for the common message . in the receiver ,
a beam - forming gain occurs for the common message such that @xmath25 is larger than @xmath26 .
the inputs and the output at some specific time are related as follows , @xmath27 \\
y & = & h_1 \,x_1 + h_2 x_2 + z \nonumber \\
\label{eq3 } & = & h \ , \left ( \sqrt { p_1 } x_1^ { ' } + \sqrt { p_2 } x_2^ { ' } + \sqrt { p_0 } x_0^ { ' } \right ) + z \end{aligned}\ ] ] where @xmath28 and @xmath29 is a zero mean gaussian noise sample . the aim is to formulate the effect of common data on optimum power scheduling in energy harvesting mac in a deterministic setting and the fading channel coefficient is assumed constant during transmission .
the capacity of the channel in a time interval @xmath30 with the total power @xmath31 is denoted by @xmath32 and given by the following , @xmath33 where _ log _ denotes base @xmath34 logarithm , @xmath35 ( hz ) refers the total bandwidth , @xmath36 ( w / hz ) is the noise spectral density and @xmath37 is the fixed path loss . throughout the article , @xmath38 is denoted by the power constant @xmath39 ( w ) such that @xmath40 .
the total transmitted bits in the time interval @xmath30 can be represented by @xmath41 .
the capacity region of the gaussian mac with the common data is given by the following @xcite , @xmath42 where @xmath43 , @xmath44 and @xmath45 , @xmath46 , @xmath47 , @xmath48 is the variable adjusting the contribution to @xmath49 by each node and @xmath50 .
a rate triplet @xmath51 is achievable if a sequence of @xmath52 codes exist where the average probability of error for decoding messages correctly approaches zero as @xmath53 goes to infinity @xcite .
the capacity region @xmath54 is defined by the closure of the set of achievable @xmath51 rate triplets .
the form of @xmath54 is illustrated in fig .
[ fig2 ] for some specific @xmath55 and @xmath56 .
+ the data rates of main boundary points @xmath57 on the three dimensional ( 3d ) capacity curve are listed as follows @xcite , @xmath58 where @xmath59 = @xmath60 , @xmath61 = @xmath62 .
it is proved in @xcite that all the points on the capacity region of gaussian mac are achieved by some point on the line segment @xmath63 of @xmath54 for some @xmath43 , @xmath44 .
furthermore , the union @xmath64 is denoted by the capacity region @xmath65 where the set of achievable @xmath51 in ( [ eq6 ] ) is given by @xmath66 and @xmath67 is the following where @xmath68 , @xmath69 it is stressed out that the boundary surface of @xmath65 can be found with the following optimization problem by varying rewarding values @xmath70 , @xmath71 it is observed in @xcite that various regions of @xmath70 values achieve @xmath65 at the defined boundary points as in the following , @xmath72 therefore , an optimization framework can cover the whole capacity region by optimizing the solution at the defined boundary points .
these points are extended to include multiple time intervals in the following sections .
now , after defining the system model and capacity region for gaussian mac with common data , the capacity maximization convex problem and its solution are analysed .
in this section , firstly , the data throughput or the maximum departure region under a deadline time constraint @xmath5 is defined .
then , the convexity of the region is proven and the capacity maximization problem is defined by using lagrange multipliers for the defined convex optimization formulation .
the problem formulation is achieved by using kkt conditions @xcite .
the departure region for the overall harvesting duration is characterized with _
lemma [ lemma1 ] _ which can be proved by using the similar approaches in @xcite , @xcite and@xcite regarding the ergodic capacity region for the overall transmitted bits within a finite amount of time .
[ lemma1 ] the departure region in @xmath6 time intervals for a time - nonvarying fading gaussian mac with common data and harvested energies denoted by the vectors @xmath73 is given by @xmath74 where @xmath75 is the set of departure triplets ( @xmath76 ) s.t .
@xmath77 where @xmath78 \big \rbrace $ ] , @xmath79 $ ] with @xmath80 denoting the total harvested energy until the @xmath53th time interval and @xmath81 is the energy available at the beginning , @xmath82 is the total harvested energy to be consumed until the final deadline time @xmath5 , @xmath83 and @xmath84 are the assigned power levels for the 1st and 2nd nodes , respectively , satisfying @xmath85 , @xmath86 while the following causality conditions hold for the total consumed energy regarding its relation to the harvested energy for @xmath87 $ ] , @xmath88 the approaches similar to the study in @xcite developed for the mac without common data can be used , especially _ lemma 1 _ and _ 2_. it can be easily proved that the optimal power allocation policy does not change the transmission rate or power during energy harvesting times .
furthermore , the rate triplets @xmath14 given in ( [ lemma1_2 ] ) define the maximum departure region for any feasible power policy @xmath89 by recursively combining the individual three dimensional ( 3d ) capacity regions for each time interval instead of 2d capacity region combinations in @xcite .
in other words , for the first time interval , the 3d capacity region @xmath90 gives the maximum departure region . then , for the second time interval , any point on the capacity region of the first time interval can be taken as the origin and the overall capacity region for the total of first two time intervals are found by combining two of the regions as shown in ( [ lemma1_2 ] ) .
it can be applied recursively to the next time intervals . by using the similar approach for the capacity region of gaussian mac with common data @xcite ,
the capacity region is unchanged if the inequalities in @xmath85 , @xmath86 are changed with equality .
furthermore , the departure region points on the boundary surface of @xmath91 is the set of all departure triplets @xmath92 such that @xmath93 is a solution to the following problem for some value of @xmath94 @xcite , @xmath95 where @xmath96 , @xmath97 , the conditions in ( [ eq22 ] ) are satisfied by replacing @xmath98 with @xmath99 and @xmath100 with @xmath101 , @xmath102 and @xmath103 for @xmath104 $ ] , @xmath105 $ ] and @xmath106 $ ] .
therefore , by maximizing @xmath107 for specific @xmath108 the points on the boundary region can be achieved .
then , we can prove that the departure region is convex .
[ lemma2 ] @xmath109 is a convex region and @xmath110 is a convex set . the proof of lemma [ lemma2 ] is given in appendix [ app0 ] . by using the the convexity ,
kkt conditions can be utilized to transform the problem into more convenient forms to be solved as the following @xcite , @xmath111 where @xmath112 corresponds to the extension of the boundary points defined in ( [ eq12 ] ) to the @xmath6 time intervals , e.g. , point @xmath113 for @xmath114 s.t .
@xmath115 , @xmath116 , @xmath117 , and the following constraints are satisfied for @xmath118 $ ] , @xmath119 $ ] , @xmath120 $ ] and @xmath106 $ ] , @xmath121 taking the derivative with respect to @xmath122 for @xmath123 $ ] and equalizing to zero gives the following , @xmath124 = p_0(n ) \left [ 2 \lambda_{2,n}^{p } \right ] + \lambda_{3,n } - \lambda_{4,n } \end{aligned}\ ] ] where @xmath125 for @xmath126 $ ] .
if @xmath127 and @xmath128 , the quadratic expression can be removed from the lagrangian function by calculating @xmath122 as the following , @xmath129 for the other cases where @xmath130 or @xmath131 and @xmath128 , the quadratic terms similarly disappear leading to a convex set of equations .
now , the problem in ( [ eqoptprob2 ] ) can be solved as a convex optimization problem leading to the unique global optimum solution .
the solution is found by solving kkt optimality conditions with unique kkt multipliers for the global solution .
@xmath132 departure triplets are found by varying the rewarding values of @xmath133 such that @xmath134 , @xmath135 and @xmath136 are utilized for the half part of the capacity region .
the other half part can be found by changing the roles of the nodes and @xmath137 and @xmath138 . for the case of @xmath139 , point @xmath113 in the boundary region
maximizes the capacity and @xmath140 becomes @xmath141 .
for the case of , @xmath142 , @xmath143 is equal to zero until @xmath5 where the boundary point @xmath144 maximizes the capacity and @xmath140 becomes @xmath145 . and finally , for @xmath146 , in the optimum solution , @xmath147 and @xmath143 will be zero and all the power will be consumed for the common data with @xmath148 in the objective function .
now , the solution for three different cases of rewarding values @xmath149 are analyzed .
this case corresponds to the boundary point @xmath113 , and taking the derivative of ( [ eqoptprob2 ] ) with respect to @xmath150 and equalizing to zero give the following kkt equalities for @xmath151 $ ] , @xmath152 for @xmath122 in ( [ eq54 ] ) where @xmath153 for @xmath154 $ ] and @xmath155 is replaced with @xmath156 in order to simplify the notation without changing the final solution for @xmath157 . for @xmath158 , three different regions of @xmath159
can be observed , i.e. , @xmath127 , @xmath130 and @xmath160 .
for the first case , i.e. , @xmath127 , @xmath122 is obtained in ( [ eq54 ] ) .
if the resulting expression is inserted into ( [ eq52 ] ) , the following can be obtained , @xmath161 where @xmath162 is defined as @xmath163 furthermore , the overall power consumed in the time interval @xmath53 should satisfy the total power equalities as @xmath164 and @xmath165 .
if the expressions of @xmath122 and @xmath166 in ( [ eq54 ] ) and ( [ eq55 ] ) , respectively , in terms of @xmath167 and @xmath168 are inserted to these power equalities , the resulting equations are obtained for @xmath167 and @xmath168 in terms of @xmath169 as the following , @xmath170 where @xmath18 $ ] and @xmath171 $ ] .
for the second and third cases of @xmath122 , i.e. , @xmath130 and @xmath160 , the following is obtained from ( [ eq52 ] ) , @xmath172 therefore , @xmath167 and @xmath168 are expressed in terms of @xmath169 .
for the other cases , the equalities in ( [ eq50]-[eq52 ] ) can be used to extract the values of @xmath167 and @xmath168 whenever the respective lagrange multipliers , i.e. , @xmath173 or @xmath174 , are zero corresponding to @xmath175 or @xmath176 , respectively .
now , the water - filling algorithm for the defined solution is provided finding the global optimum solution in an efficient way .
[ lemma3 ] the optimization regarding the equalities in ( [ eq50]-[eq52 ] ) is achieved by defining and equalizing the following water levels @xmath177 for each time interval @xmath171 $ ] such that @xmath178 is satisfied for @xmath179 $ ] and the water levels are defined as follows @xmath180 and @xmath181 , @xmath182 obtained by ( [ eq50 ] - [ eq52 ] ) , ( [ eqlambdaopt ] ) are defined in table [ taboptlambda ] where @xmath183 denotes @xmath184 and the water levels satisfy the following , @xmath185 & \mbox{if } \lambda_{1,n}^{p } = \lambda_{1,n+1}^{p } \ , \ & \ , \lambda_{2,n}^{p } = \lambda_{2,n+1}^{p } \end{cases}\ ] ] [ 1]>m#1 @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath49 & @xmath147 & @xmath143 + - & @xmath190 & @xmath191 & @xmath183 & @xmath191 & @xmath191 & @xmath183 + @xmath192 & - & @xmath183 & @xmath193 & @xmath191 & @xmath183 & @xmath193 + @xmath194 & @xmath195 & @xmath196 & @xmath183 & @xmath191 & @xmath183 & @xmath183 + @xmath197 & @xmath198 & @xmath183 & @xmath183 & @xmath183 & @xmath193 & @xmath199 + @xmath200 & @xmath201 & @xmath196 & @xmath183 & @xmath196 & @xmath191 & @xmath183 + @xmath202 & @xmath198 & @xmath196 & @xmath183 & @xmath196 & @xmath196 & @xmath193 + @xmath203 & @xmath198 & @xmath183 & @xmath196 & @xmath183 & @xmath183 & @xmath183 + [ taboptlambda ] the proof of theorem [ lemma3 ] is given in appendix [ app0 ] .
the constraints and the components of the objective functions for the other 2 cases of @xmath204 , i.e. , @xmath135 and @xmath205 , are found with a similar approach to the case 1 .
therefore , in the following , only the kkt conditions are presented for thee cases without the detailed proofs .
this case corresponds to the boundary point @xmath144 , and similar to the approach performed for @xmath134 , taking the derivative of ( [ eqoptprob2 ] ) with respect to @xmath206 gives the following for @xmath151 $ ] , @xmath207 = p_0(n ) \left [ 2 \lambda_{2,n}^{p } \right ] + \lambda_{3,n } - \lambda_{4,n } \end{aligned}\ ] ] if @xmath127 , ( [ eq54 ] ) is satisfied and putting into ( [ eq57 ] ) , the equation becomes as the following , @xmath208 it can be observed that @xmath209 becomes always zero and the water level @xmath210 is absent compared with the first case .
then , it can be proved that @xmath211 and @xmath212 .
therefore , the optimization algorithm will check for the inequalities @xmath213 for @xmath214 and @xmath215 , and equalities @xmath216 for @xmath217 or @xmath218 .
furthermore , the table regarding @xmath219 is modified as shown in table [ taboptlambda2 ] .
@xmath220 & @xmath221 & @xmath188 & @xmath189 & @xmath49 & @xmath147 & @xmath143 + - & @xmath190 & @xmath191 & @xmath183 & @xmath191 & @xmath191 & @xmath183 + @xmath192 & - & @xmath183 & @xmath193 & @xmath191 & @xmath183 & @xmath191 + @xmath222 & @xmath223 & @xmath183 & @xmath183 & @xmath183 & @xmath193 & @xmath191 + @xmath202 & @xmath223 & @xmath196 & @xmath183 & @xmath196 & @xmath196 & @xmath191 + [ taboptlambda2 ] this case corresponds to the boundary point @xmath225 and taking the derivative of ( [ eqoptprob2 ] ) with respect to @xmath226 and equalizing to zero give the following kkt conditions for @xmath151 $ ] , @xmath227 furthermore , ( [ eq54 ] ) is satisfied and putting into ( [ eq57 ] ) , the equation becomes as the following , @xmath228 it is observed that @xmath229 for @xmath171 $ ] , the comparison for @xmath230 is removed and only @xmath231 is compared between neighbouring time intervals .
the table for @xmath219 is modified as shown in table [ taboptlambda3 ] .
@xmath232 & @xmath233 & @xmath188 & @xmath189 & @xmath49 & @xmath147 & @xmath143 + - & @xmath234 & @xmath191 & @xmath183 & @xmath196 & @xmath191 & @xmath193 + @xmath234 & - & @xmath183 & @xmath193 & @xmath183 & @xmath191 & @xmath191 + @xmath235 & @xmath236 & @xmath183 & @xmath183 & @xmath183 & @xmath193 & @xmath193 + [ taboptlambda3 ] it is difficult to implement the solution defined in theorem [ lemma3 ] in a water - filling algorithm since at each time interval either of the nodes or both of the nodes can transfer energy to the neighbouring next time interval by looking only at the neighbouring time intervals with possibly leading to sub - optimum solution .
therefore , an iterative water - filling algorithm is defined similar to @xcite in a way realizing power scheduling node by node iteratively while fixing the energy levels in the other node .
then , the following lemma is proved for @xmath114 and it can be easily extended to the other 2 cases of @xmath237 and @xmath238 .
[ lemma4 ] the optimization regarding the equalities in ( [ eq50]-[eq52 ] ) can be achieved with an iterative backward water - filling algorithm given in algorithm [ alg1 ] by only satisfying @xmath239 for the power scheduling of the 1st node where the power levels in 2nd is fixed , and @xmath240 for the power scheduling of the 2nd node where the power levels in 1st is fixed for @xmath241 $ ] whenever power transfer from the time interval @xmath53 to @xmath242 occurs . the proof is given in appendix [ app0 ] . the iterative water - filling algorithm given in algorithm [ alg1 ] looks at the difference between water - levels and using an iterative weighted search algorithm , water - levels are optimized .
the algorithm firstly optimizes the power levels in all time intervals based on the available amount of power in each node and using the iterative algorithm based on the regions and solutions in table [ tab_app1 ] in appendix [ app1 ] .
then , the power levels in the nodes are iteratively optimized by equalizing @xmath239 and @xmath240 for the power scheduling of the 1st and 2nd node , respectively , while fixing the power levels in the other node fixed . for the neighbouring time intervals @xmath53 and @xmath242 if the equalities can not be achieved by the power transfer then no action is taken for that specific time interval @xmath53 .
initialize @xmath243 for @xmath244 $ ] and @xmath171 $ ] by finding the optimum @xmath245 , @xmath246 , @xmath209 with the solutions in table [ tab_app1 ] .
fix 2nd node fixed and satisfy @xmath247 for the 1st node .
update @xmath248 , @xmath249 , @xmath250 , @xmath251 for @xmath252 $ ] .
fix the 1st node fixed and satisfy @xmath253 for the 2nd node .
update @xmath248 , @xmath249 , @xmath254 , @xmath255 for @xmath252 $ ] .
the proposed algorithm is simulated for the case that the 1st user and 2nd user harvest energies @xmath256 $ ] and @xmath257 $ ] mj at the time instants @xmath258 $ ] and @xmath259 $ ] seconds , respectively , @xmath260 time intervals are considered until @xmath261 seconds , bandwidth @xmath262 mhz , noise spectral density @xmath263 w / hz and path loss @xmath264 are considered for generic analyses used in the literature @xcite . in the following figures , @xmath265 and @xmath266
denote consumed and harvested energy , respectively , for the nodes @xmath267 $ ] . . , title="fig:",width=288 ]
+ the departure region boundary surface formed from the scattered points of the sample points is shown in fig .
[ figure_ns_1 ] which resembles the capacity boundary surface for the single time interval shown in fig .
it can be observed that there is a linear region on the capacity boundary surface between the points obtained with @xmath268 .
this region corresponds to sampling boundary points between @xmath113 and @xmath269 as shown in fig .
furthermore , 3 sample points labelled with the labels @xmath270 , @xmath271 and @xmath272 on the @xmath268 curve are taken which corresponds to sampling the point @xmath269 at different @xmath273 .
these points are used to compare the effect of the amount of @xmath274 on the optimum power scheduling scenario .
these points correspond to @xmath275 for the case when no common data is transmitted similarly to the study in @xcite , @xmath276 mbit where common data and the distinct data of each user are transmitted and the case for the maximum amount of common data of @xmath277 mbit corresponding to also the maximum amount of total data rate . .
, title="fig:",width=288 ] + firstly , the comparison between the defined optimization framework and the optimum backward iterative water - filling algorithm denoted as owa in @xcite is shown in fig .
[ figure_ns_2 ] .
it is observed that the proposed solution gives the same result with @xcite when no common data is transmitted , e.g. , @xmath275 .
furthermore , the linear region between the points @xmath113 and @xmath269 can be observed .
the constant @xmath274 contours are given in fig .
[ figure_ns_3 ] .
while @xmath274 increases , the amount of power used for @xmath278 and @xmath279 decreases and the boundary shows a curved behaviour with a decreasing amount of linear region as shown in fig .
[ figure_ns_3 ] .
contours for the optimum scheduling algorithm.,title="fig:",width=288 ] + contours.,title="fig:",width=288 ] + the optimum scheduling algorithm is denoted as opt - s and compared with two basic algorithms denoted as uni - s and no - s which represent no scheduling and the scheduling algorithm distributing the power uniformly until @xmath5 , respectively . in no - s algorithm , at each time interval , the power levels are optimized to maximize the data throughput only in that time interval by using the solutions in table [ tab_app1 ] in [ app1 ] . in uni - s algorithm ,
the harvested power is distributed uniformly for the next - coming time intervals until @xmath5 . as shown in fig .
[ figure_ns_4 ] , no - s algorithm gives a maximum amount of common data rate of approximately @xmath280 mbit with non - zero @xmath278 and @xmath279 since in some time intervals the harvested power is not utilized for @xmath274 due to the reason that only one of the nodes harvest energy and the optimal solution does not allow allocating power to common data .
the uni - s and opt - s algorithms perform better compared with the no - s algorithm with larger @xmath274 and a larger capacity boundary volume .
opt - s algorithm is better compared with uni - s algorithm as shown in fig .
[ figure_ns_4 ] such that a larger @xmath278 and @xmath279 curve is obtained for constant @xmath274 contours and the maximum amount of @xmath274 is bigger .
.,title="fig:",width=288 ] + the water - filling and the power scheduling profiles of the nodes are shown and compared in figs .
[ figure_ns_5 ] - [ figure_ns_8 ] for the sample points @xmath270 , @xmath271 and @xmath272 described in fig .
[ figure_ns_1 ] . as shown in fig .
[ figure_ns_5 ] , when there is no power scheduling , the water levels are not in equilibrium resulting a lower data rate for the common data @xmath274 . on the other hand , optimization algorithm leads to the equilibrium of the water levels such that @xmath281 for @xmath282 and @xmath283 for @xmath284 where previously harvested power is transferred to 3rd and 5th time intervals for the first node and to 2nd , 4th and 5th intervals for the second node .
, @xmath271 and @xmath272 for varying @xmath274.,title="fig:",width=288 ] + it can be observed that as @xmath274 increases , the power levels consumed by each node in different time intervals are more homogenized in order to maximize @xmath274 as shown in fig .
[ figure_ns_6 ] since for a given @xmath285 more amount of @xmath49 is obtained whenever @xmath188 and @xmath189 are closer to each other . the same situation is observed in fig .
[ figure_ns_7 ] where more homogeneous and increasing power consumption is realized for each node as @xmath274 increases . at the sample points a , b an c. , title="fig:",width=288 ] + furthermore , as shown in fig .
[ figure_ns_8 ] , for the case when no common data is transmitted and @xmath268 , the optimal power scheduling optimizes as if there is a single node with the total power @xmath286 @xcite where the total consumed power monotonically increases and uses all the harvested energy until the transmission rate changes .
however , as @xmath274 increases , some of the harvested power is saved for future use in a way to maximize @xmath274 for a given @xmath278 , @xmath279 data rate .
for example , at @xmath287 ( sec ) not all the harvested energy is used although there is an increase in the data rate for the common data at the sample boundary points b an c as shown in figs .
[ figure_ns_6 ] and [ figure_ns_8 ] . at the sample points a , b an c. , title="fig:",width=288 ] + as a result , the optimum solution defined in this article extends the previously defined optimum solution for gaussian mac without common data and performs better than no power scheduling and uniform power scheduling cases .
furthermore , although water - filling levels are more complicated due to the complexity of the optimum departure region for gaussian mac with common data , very efficient and simple water - filling algorithm is defined for the optimum solution .
there is a multitude of assumptions defined in the article which leads to a set of future work topics .
first of all , the optimum power allocation solution for @xmath288-user gaussian mac with common data in a single time interval , e.g. the study in @xcite , can be utilized for an optimum packet scheduling policy with energy harvesting transmitters and various levels common data sharing mechanisms between @xmath288 nodes .
furthermore , another open issue is to combine the time - varying fading effects and stochastic nature of energy harvesting and packet reception in a way to obtain the optimum online solution for gaussian mac with common data .
furthermore , the online extension can be improved more by adapting the scheduling algorithm to the level of the correlation among the transmitters , e.g. , in a temporally or spatially correlated wsn . moreover , the utilization of common data beam - forming to transfer not only data but also energy , e.g. , in a sensor networking architecture with limited amount of powers , is an alternative open issue to be analyzed .
time varying fading effects can be considered to adapt the transmission policy based on the current channel states . therefore , an optimization framework can be defined deciding on the trade off between the amount of common data to transfer more energy to the next node and the amount of data for the independent messages .
in this paper , optimum offline packet scheduling policy is developed for 2-user gaussian mac with common data and energy harvesting transmitters .
the optimum solution is derived by using kkt conditions .
the proposed solution is implemented with an efficient iterative water - filling algorithm .
the proposed optimum solution algorithm is numerically simulated to find the optimum departure region .
the optimum solution is compared with the solutions defined for gaussian mac without common data , no scheduling and uniform power scheduling solutions with the result of being the best among them and extending the previous solutions for gaussian mac without common data .
finally , a set of open issues and future work studies are defined which includes considering stochastic nature of the system leading to online solution , time - varying fading effects , both energy and data transfer possibility and the optimum offline solution for m - user gaussian mac with common data .
we can use a similar approach to @xcite and extend the proof to include multiple time intervals . in the following
, @xmath289 denotes @xmath14 , @xmath290 $ ] and @xmath291 $ ] .
we need to show that for any @xmath292 , @xmath293 @xmath294 @xmath295 is in @xmath110 . both @xmath296 and @xmath297 satisfy that @xmath298 for some @xmath299 such that the following conditions hold @xmath300 for @xmath301 $ ] and @xmath302 or @xmath303 .
let us define @xmath304 for @xmath154 $ ] and the following for @xmath104 $ ] @xmath305 by using the concavity of log function , the following can be proved , @xmath306 similarly , the following can be proved , @xmath307 then , by using ( [ l2_eq1]-[l2_eq4 ] ) , the following can be proved , @xmath308 similarly , it can be proved that @xmath309 it can be easily proved that @xmath310 by using ( [ l2_eq3]-[l2_eq4 ] ) .
then , inserting into ( [ l2_eq9]-[l2_eq10 ] ) and extending the result for the time interval @xmath311 , the following equations can be obtained , @xmath312 as a result , @xmath313 .
therefore , @xmath314 @xmath315 @xmath316 and @xmath293 @xmath294 @xmath295 is in @xmath110
. the optimum solution of @xmath317 regarding the single time step @xmath53 with the available power levels @xmath83 and @xmath84 for that iteration is found by using the iterative descent algorithm defined in @xcite with 8 different lagrange multipliers regions and the corresponding solutions as shown in table [ tab_app1 ] in appendix [ app1 ] .
there is a difference between the optimum solution for the single time step between @xcite and the current study .
the lagrange multipliers obtained in the solution for the single time step are the solutions for the power inequalities corresponding to @xmath83 and @xmath84 rather than ( [ c1_eq]-[c5_eq2 ] ) in the proposed solution .
therefore , the multipliers @xmath167 and @xmath168 include the effects of the all time intervals .
compared with @xcite , whenever the water levels between time intervals @xmath53 and @xmath242 are equalized , there can be potentially at most @xmath318 different combinations of optimality regions defined in table [ tab_app1 ] .
furthermore , in the optimum scheduling solution , there can be 3 different cases corresponding to the transfer of energy between two time intervals , i.e. , power transfer to the next time interval by the 1st node , 2nd node and both of the nodes , complicating the analysis more .
a large set of optimality relations exist between the water levels @xmath319 and @xmath320 based on ( [ eq50]-[eq52 ] ) . for example , some of the multipliers will be zero whenever the corresponding power levels @xmath321 are greater than zero .
on the other hand , @xmath167 and @xmath168 are already decreasing functions based on their definition , i.e. , @xmath322 , and the same for the @xmath323 .
@xmath167 is equal to @xmath324 if the 1st node transfers the stored energy from the time interval @xmath53 to @xmath242 , and similarly @xmath168 is equal to @xmath325 for the 2nd node and both the equalities hold if both the nodes transfer energy to the next time interval .
moreover , it is observed that @xmath326 . as a result , by using these set of observations and the equalities in ( [ eq50]-[eq52 ] ) , it can be observed that @xmath327 for @xmath328 $ ] is always satisfied possibly leading to @xmath329 for some specific combinations of optimality regions between neighbouring time intervals @xmath53 and @xmath242 . due to space limitations , the result is not given for all different optimality region combinations , however , they can be shown easily .
for the comparison regarding @xmath330 and @xmath331 , the fact that @xmath332 if the @xmath311th node transfers stored energy from the time interval @xmath53 to @xmath242 is utilized for @xmath267 $ ] .
in fact , @xmath333 is represented in terms of @xmath334 , @xmath246 , @xmath209 , @xmath83 , @xmath335 in the corresponding time interval @xmath53 .
therefore , instead of equalizing lagrange multipliers @xmath333 and @xmath336 , new water levels @xmath337 and @xmath338 represented in terms of the power levels are defined and equalized . when the 2nd node has fixed energy in one of iterations , the only water levels to be compared are the pairs ( @xmath339 , @xmath340 ) and ( @xmath341 , @xmath342 ) for @xmath241 $ ] . in the same manner , when the 1st node has fixed energy in one of iterations , the only water levels to be compared are the pairs ( @xmath343 , @xmath344 ) and ( @xmath345,@xmath346 ) for @xmath241 $ ]
. these can be proved by removing the inequalities from ( [ eq50]-[eq52 ] ) including the corresponding multipliers regarding the fixed node .
it can be easily proved by comparing the multipliers for the two neighbouring time intervals such that the inequalities and equalities in table [ tab0_app0 ] hold for the optimum solution .
then , it can be easily proved by using the solutions in table [ tab_app1 ] such that equalizing ( @xmath341 , @xmath342 ) for @xmath241 $ ] satisfies the inequalities or the equalities regarding ( @xmath339 , @xmath340 ) and ( @xmath343 , @xmath344 ) in table [ tab0_app0 ] .
therefore , there is no need to compare the 1st and 2nd water levels and it is enough to equalize ( @xmath341 , @xmath342 ) and ( @xmath345 , @xmath346 ) between neighbouring time intervals .
region indices @xmath347 & region indices @xmath348 & + @xmath34 , @xmath349 , @xmath350 & @xmath34 , @xmath349 , @xmath350 & [ -0.3em]@xmath351 & + @xmath34 , @xmath349 , @xmath350 & @xmath352 , @xmath353 , @xmath354 , @xmath355 & & @xmath356 + @xmath352 , @xmath353 , @xmath354 , @xmath355 & @xmath34 , @xmath349 , @xmath350 & & @xmath357 + @xmath352 , @xmath353 , @xmath354 , @xmath355 & @xmath352 , @xmath353 , @xmath354 , @xmath355 & & @xmath358 + + region indices @xmath347 & region indices @xmath348 & + @xmath34 , @xmath352 , @xmath353 & @xmath34 , @xmath352 , @xmath353 & [ -0.3em]@xmath359 & + @xmath34 , @xmath352 , @xmath353 & @xmath349 , @xmath350 , @xmath354 , @xmath355 & & @xmath360 + @xmath349 , @xmath350 , @xmath354 , @xmath355 & @xmath34 , @xmath352 , @xmath353 & & @xmath361 + @xmath349 , @xmath350 , @xmath354 , @xmath355 & @xmath349 , @xmath350 , @xmath354 , @xmath355 & & @xmath362 + the following definitions are used next , @xmath363 , @xmath364 , @xmath365 , @xmath366 , @xmath367 , @xmath368 , @xmath369 .
the regions are defined for varying @xmath370 with the corresponding solutions .
@xmath371 refers the region , and @xmath372 refers the solution in that region where @xmath373 refers to tzero power region .
kkt or lagrange multiplier regions and the solutions for the optimum power levels are given in table [ tab_app1 ] resembling the structure in @xcite .
index & region & solution + @xmath374 & @xmath375 , g > \mu_0 $ ] & @xmath376 + @xmath34 & @xmath377 , \frac{\mu_0}{g } > 1 $ ] & @xmath378 + @xmath379 & @xmath380 & @xmath381 + @xmath382 & @xmath383 & @xmath384 + @xmath385 & @xmath386 & @xmath387 + @xmath388 & @xmath389 & @xmath390 + @xmath391 & @xmath392 & @xmath393 + @xmath394 & @xmath395 & @xmath396 + + @xmath374 & @xmath397 & @xmath376 + @xmath34 & @xmath398 & @xmath399 + @xmath379 & @xmath400 & @xmath401 + @xmath382 & @xmath402 & @xmath403 + + @xmath374 & @xmath404 & @xmath405 + @xmath34 & @xmath406 & @xmath407 + b. gulbahar and o. b. akan , `` information theoretical optimization gains in energy adaptive data gathering and relaying in cognitive radio sensor networks , '' _ ieee transactions on wireless communications _ , vol . 11 , no . 5 , pp .
17881796 , 2012 .
o. kaya and s. ulukus , `` achieving the capacity region boundary of fading cdma channels via generalized iterative waterfilling , '' _ ieee transactions on wireless communications _ , vol . 5 , no . 11 , pp .
32153223 , 2006 .
b. gurakan et al .
, `` two - way and multiple - access energy harvesting systems with energy cooperation , '' _ ieee forty sixth asilomar conference on signals , systems and computers ( asilomar ) _ , pp .
5862 , 2012 .
n. liu and s. ulukus , `` capacity region and optimum power control strategies for fading gaussian multiple access channels with common data , '' _ ieee transactions on communications _ , vol .
54 , no .
18151826 , 2006 .
r. rajesh and v. sharma , `` joint source - channel coding over a fading multiple access channel with partial channel state information , '' _ ieee global telecommunications conference ( globecom ) _ , pp .
17 , 2009 .
d. n. c. tse and s. v. hanly , `` multiaccess fading channels .
i. polymatroid structure , optimal resource allocation and throughput capacities , '' _ ieee transactions on information theory _
44 , no . 7 , pp .
27962815 , 1998 . | the lifetime and the sustainability of the wireless sensor networks ( wsns ) can be increased with energy harvesting transmitters utilizing optimum packet scheduling . on the other hand , wsns
are observed to collect spatially or temporally correlated data which should be taken into account for the optimum packet scheduling in an energy harvesting system .
however , the solutions available for 2-user multiple - access channel ( mac ) systems with energy harvesting transmitters do not consider the common data or the correlation among the data . in this paper , optimal packet scheduling for energy harvesting 2-user gaussian mac with common data
is achieved by assuming deterministic knowledge of the data and energy packets , i.e. , offline solution .
the optimum departure region is found by using karush - kuhn - tucker ( kkt ) conditions generalizing the solutions obtained for the mac without common data .
an efficient iterative backward water - filling algorithm is defined .
the optimum solution is numerically compared with the case of no scheduling , uniform power scheduling and the previous solutions defined for the mac without common data by showing the improvement obtained with the optimization .
submitted paper energy harvesting , mac , common data , correlation , packet scheduling . |
newtonian gravity theory has served physics and technology faithfully for well over three centuries . nevertheless , it has long been known that it is only an approximation to a relativistic gravitation theory , usually identified with einstein s 1915 general relativity ( gr ) .
gr has correctly predicted subtle effects in the dynamics of the solar system , in the celebrated hulse - taylor double pulsar , and has anticipated the existence of those exotic denizens of the universe , black holes , now confirmed by myriad observations of galactic nuclei , compact galactic x - ray sources , etc .
the nonlinear aspects of gravitation in gr are crucial to these new settings : gravitation is more difficult than newton supposed . but up until a few decades ago astronomers took comfort in the belief that if one sidesteps situations with extremely strong gravitational fields , or with extremely rapid motions , then newtonian theory is a good approximation to the truth .
but is that so ?
consider the situation in the realm of galaxies .
speeds are low compared to light s , and gravitational fields are weak there .
so newtonian physics
_ should _ describe motions extremely well . yet
the accelerations of stars and gas clouds in the outskirts of spiral galaxies , and of galaxies swimming in the large clusters of galaxies , well exceed newtonian predictions made on the basis of the matter actually visible in these systems .
those accelerations are inferred by combining doppler measured velocities with geometric assumptions , e.g. that in a particular spiral galaxy stars move in circles on a plane .
the geometric assumptions can be checked , at least statistically , and the acceleration discrepancy is found to be real and generic . sometimes the problem is characterized thus : rotation or random velocities are much bigger than expected .
i stress acceleration because it is the quantity which directly measures the strength of the gravitational field . in any case , what is wrong ?
it was realized by ostriker and peebles @xcite that massive halos would stabilize disk galaxies against the bar forming instabilities which , in newtonian theory , are endemic to systems with little velocity dispersion . since about third of the spiral galaxies have no significant bar , this justified the hypothesis that disk galaxies ( including spirals ) are often embedded in massive but invisible halos .
thus started the trend to resolve the acceleration discrepancy by postulating the existence of much dark matter ( dm ) in systems ranging from the very tenuous dwarf spheroidal galaxies with visible masses @xmath1 to the great clusters of galaxies with observed masses in the @xmath2 ballpark @xcite , in brief in any system where an acceleration discrepancy exists .
the dm s role is to provide the missing gravitational pull to account for the excessive accelerations ( an easy introduction to the dm paradigm is provided by khalil and mu~ noz @xcite ) .
but thirty years of astronomical exploration and laboratory experiments @xcite have yet to provide independent evidence of dm s existence , e.g. @xmath3 rays from its decay .
is there an alternative to dm ?
this review focuses on the modified newtonian dynamics scheme put forward by milgrom @xcite as an alternative to the systematic appeal to dm .
to pass judgment on dm and alternatives to it , it is well to take stock of the two overarching empirical facts in the phenomenology of disk galaxies .
first , ever since the work of bosma as well as rubin and coworkers @xcite , it has been clear that gas clouds in the disks of spiral galaxies , which serve as tracers of the gravitational potential , circle around each galaxy s center with a ( linear ) velocity which first rises as one moves out of the center , but hardly drops as the radius grows to well beyond the visible disk s edge .
yet a falloff of the velocity with radius was naively expected because , to judge from the light distribution in spiral galaxies , most of their visible mass is rather centrally concentrated .
newtonian gravitation would thus predict that `` rotation curves '' ( rc ) should drop as @xmath4 outside the bright parts of these galaxies , but this is not seen in over a hundred extended rcs measured with sufficient precision @xcite .
in fact in many spiral galaxies , particularly those possessing high surface brightness , the extended rc becomes flat away from the central parts ( fig .
[ figure : fig1 ] ) . is plotted vs galactocentric radius @xmath5 in kiloparsecs ( kpc ) ; 1 kpc @xmath6 3000 light years .
it is seen that the rcs are flat to well beyond the edges of the optical disks ( @xmath7 kpc ) .
graph from ref . , reprinted with permission from the annual review of astronomy and astrophysics , volume 39 ( c)2001 by annual reviews www.annualreviews.org,width=264 ] second , as originally pointed out by tully and fisher ( tf ) @xcite , the rotation velocity @xmath8 and the blue band luminosity @xmath9 of a galaxy
are simply correlated ( fig .
[ figure : fig2 ] ) . in a more informative form
@xcite , the tf law states that for disk galaxies the luminosity in the near infrared band , @xmath10 ( a good tracer of stellar mass ) , is proportional to the fourth power of the rotation velocity in the flat part of the rc , with a universal proportionality constant . in units of @xmath11 vs. the asymptotic rotational velocity @xmath8 in km / s .
the straight line ( mine ) has slope exactly 4 and corresponds to mond s prediction , eq .
( [ tflaw ] ) , for a reasonable mass ( in units of @xmath12 ) to @xmath13 ratio @xmath14 .
graph reproduced from ref . by permission of the american astronomical society.,width=226 ] the flat extended rc s have provided the logic underpinning the dm paradigm as follows . in the flat part of a rc ,
the rotation velocity is @xmath15 independent , so the centripetal acceleration goes as @xmath16 .
thus in the plane of the galaxy the gravitational field must be decreasing as @xmath16 . according to poisson s equation of newtonian theory , such a gravitational field , if assumed spherically symmetric ,
must be sourced by a mass distribution with @xmath17 profile ( isothermal sphere model ) . since the visible mass density ( in stars and gas ) in the inner disk drops much faster than this , it is consistent to assume that the total mass distribution in the outer parts is quasispherical .
the conclusion is that each spiral galaxy must be immersed in a roundish dm halo with mass density profile tending , at large @xmath15 , to @xmath17 .
but questions plague the halo hypothesis .
the halo dm , though much searched for , has never been detected directly @xcite .
cosmogonic simulations of dm halo formation within newtonian physics predict that such a halo would have a density profile behaving like @xmath16 for small @xmath15 and gradually steepening to @xmath18 asymptotically . only by fine tuning the halo s parameters
is it feasible to match this kind of profile to the observed flat rcs as well as would the isothermal sphere ( the need to fine tune parameters of a halo model in order to fit the shape of the galaxy s rc was already clear long ago @xcite ) . and the predicted @xmath16 `` cusp '' in the density profile is also observationally problematic @xcite .
the dm halo hypothesis is evidently an attempt to resolve the acceleration discrepancy within orthodox gravitation theory .
but in the coming to terms with this discrepancy , suspicion fell on newtonian gravity already early in the game .
zwicky , who had exposed the acceleration discrepancy in clusters of galaxies @xcite , opined much later that the discrepancy may reflect a failure of conventional physics @xcite .
concrete proposals for departures from the newtonian inverse square law in various settings were put forward by a number of workers @xcite .
all of these still regarded gravitation as a linear interaction with the strength of the field proportional to its source s mass .
as milgrom realized @xcite , such a modification of the gravity law , if relevant on the scales of galaxies , is incompatible with the tf law . for one
it would imply that a mass @xmath19 at @xmath20 generates at @xmath21 the acceleration field @xmath22 , with @xmath23 depending only on fundamental constants and on the relative positions of source and field points in accordance with the principle of homogeneity of space .
then , obviously , for matter orbiting with speed @xmath8 at radius @xmath5 in the outskirts of a galaxy ( with its source mass , @xmath24 , concentrated at a fairly definite distance @xmath5 ) , the acceleration , @xmath25 should be equal to @xmath26 .
this would require not only @xmath27 , but also @xmath28 , the last incompatible with the tf law .
then again @xmath29 with the correct dimensions , @xmath30 , can not be built by using only @xmath31 and the scale of length @xmath32 at which gravity departs from newtonian form . and inclusion of mass in the construction ( presumably the source s mass ) would be against the spirit of linearity .
the bottom line is that linear gravity , even if non - newtonian , is incompatible with the tf law .
milgrom @xcite proposed a novel paradigm which can be interpreted as reflecting non - newtonian as well as nonlinear character of gravity already at the nonrelativistic level .
he called the scheme modified newtonian dynamics ( mond ) .
the essential part of it is the relation @xmath33 between the acceleration @xmath34 of a particle and the ambient conventional newtonian gravitational field @xmath35 .
if the function @xmath36 were unity , this would be usual newtonian dynamics .
milgrom assumes that the positive smooth monotonic function @xmath36 approximately equals its argument when this is small compared to unity ( deep mond limit ) , but tends to unity when that argument is large compared to unity ( fig .
[ figure : fig3 ] ) .
the @xmath37 is a natural constant , approximately equal to @xmath38 .
it is a fact that the centripetal accelerations of stars and gas clouds in the outskirts of spiral galaxies tend to be below @xmath37 .
function : the `` simple '' function @xmath39 ( solid ) and the `` standard '' function @xmath40 ( dotted).,width=226 ] how does the mond formula ( [ mond ] ) help ?
consider again stars or gas clouds orbiting in the disk of a spiral galaxy ( mass @xmath24 ) with speed @xmath41 at radius @xmath15 from its center .
we must identify @xmath42 with the centripetal acceleration @xmath43 . sufficiently outside the main mass distribution we may estimate @xmath44 . and at sufficiently large @xmath15 , @xmath42 will drop below @xmath37 and we shall be able to approximate @xmath45 . putting all this together
gives @xmath46 from which we may conclude two things .
first , @xmath41 well outside the main mass distribution becomes independent of @xmath15 , that is , the rc flattens at some value @xmath47 .
second , from the coefficients follows @xmath48 introducing the ratio of mass to luminosity @xmath49 , @xmath50 , we have the added prediction @xmath51 . although @xmath50 in the blue band varies somewhat with galaxy color , this last results explains the tf law , fig .
[ figure : fig2 ] .
theoretically @xmath50 is much less variable in the near infrared ( k ) band @xcite , which accounts for the extra sharpness of the tf relation in that band ( see fig . 2 in ref . ) . in units of the solar mass @xmath12 vs. the observed asymptotic rotational velocity @xmath47 in
km / s . for each of the 60 galaxies from ref .
( deep blue circles ) the mass _ in stars _ comes from a fit of the _ shape _ of the rc with mond , whereas for the eight dwarf spirals ( light blue circles ) the mass in stars ( relatively small ) is inferred in ref . directly from the luminosity . the green line , with slope 4 , is mond s prediction , eq .
( [ tflaw ] ) .
graph reproduced from ref . by permission of the american astronomical society.,width=226 ] the impressive agreement of eq .
( [ tflaw ] ) with observations is most clear from the baryonic tf law @xcite , fig .
[ figure : fig4 ] . in full agreement with mond ,
the observed baryonic mass in spiral galaxies is accurately proportional to the fourth power of the asymptotic rotation velocity .
mond s single formula thus unifies the two overarching facts of spiral galaxy phenomenology .
mond is similarly successful in explaining the detailed _ shapes _ of rcs .
[ figure : fig5 ] shows the measured rcs of fifteen galaxies together with the mond fits based on surface photometry in the optical band and radio measurements .
the only parameter adjusted in these impressive fits is the disk s stellar mass , or equivalently , the stellar mass - to - luminosity ratio @xmath50 .
the trend of the so determined values of @xmath50 with galaxy color jibes with that predicted by stellar evolutionary models @xcite ; in this mond does better than dm halo models @xcite . to make similarly successful fits without just `` putting dm where it is needed '' ,
dm theorists must adjust two extra parameters apart from @xmath50 in standardized halo models . as stressed in a recent critical reappraisal @xcite
, those galaxies where one - parameter mond fits do worse than halo fits often display complicating factors that could mitigate its less than striking performance . in disk galaxies
mond is unquestionably more economical , and thus more falsifiable , than the dm paradigm .
when applied to many galaxies , the mond fits mentioned serve to determine @xmath37 with some accuracy ; the value @xmath52 is often used @xcite .
it is significant that @xmath37 as deduced from rc fits agrees well with the value for @xmath37 obtained by comparing @xmath53 with the empirical coefficient in the tf law : the roles of @xmath37 are different in the two subjects , and they are tied together only in mond .
milgrom @xcite notes a few additional conceptually distinct roles played by @xmath37 in extragalactic astrophysics .
an early example is that @xmath37 sets a special scale of mass surface density @xmath54 , and wherever in a system the actual surface mass density drops below @xmath55 , newtonian gravitational behavior gives way to mond dynamics @xcite .
( in galaxies like ours this occurs some way out in the disk , and that is why the rc may exhibit a brief drop before becoming flat asymptotically ) . on this basis milgrom
predicted that were disk galaxies with surface mass density everywhere below @xmath55 to exist , they should show especially large acceleration discrepancies .
a population of such galaxies became known in the late 1980 s @xcite , and all facets of milgrom s prediction were subsequently confirmed @xcite .
mond obviously predicts that in these so called low surface brightness galaxies , the shapes of the rcs , which tend to be _ rising _ throughout the visible disk , should be independent of the precise way @xmath56 switches from @xmath57 to unity for @xmath58
. constraints of space forbid me from delving into other mond successes , e.g. for elliptical and dwarf spheroidal galaxies .
but excellent reviews of these and other phenomenological aspects of mond exist @xcite .
it is well , however , to stress one prominent empirical difficulty faced by the mond formula ( [ mond ] ) .
clusters of galaxies exist which comprise hundreds of galaxies of various kinds moving in their joint gravitational field with velocities of up to @xmath59 . the newtonian virial theorem ( gravitational potential energy of a quiescent system equals minus twice its kinetic energy )
can be used to estimate a cluster s mass since the kinetic term is proportional to the total mass while the potential one goes like the square of mass . in this way
it is determined that the total gravitating mass in a typical cluster is 5 - 10 times the mass actually seen in galaxies as well as in hot x - ray emitting gas , an oft constituent of clusters .
a similar story is told by analyses of the hydrostatics of the hot gas in light of its measured temperature .
finally , the gravitational lensing by several clusters has led to estimates of their masses commensurate with the mentioned diagnosis .
when the mond formula , or even better , a mond analog to the virial theorem @xcite , is used to estimate the masses of clusters , one finds an improvement but not a full resolution to the acceleration discrepancy .
clusters still seem to contain a factor of two more matter than actually observed in all known forms @xcite .
the optimist will stress that mond has alleviated the discrepancy without even once overcorrecting for it ; the pessimist @xcite will view this finding as damaging mond s credibility .
but one should keep in mind that clusters may contain much invisible matter of rather prosaic nature , either baryons in a form which is hard to detect optically , or massive neutrinos @xcite .
however , the option that clusters contain non - baryonic cold dm between the galaxies , while logically possible , seems hardly justifiable in view of mond s overall philosophy .
what is the physical basis of mond s success in unifying a lot of extragalactic data ?
there are certain things the mond formula can not be .
first , it is unlikely to be just a recipe for the way dm is distributed spatially in astronomical objects , as sometimes proposed @xcite .
such proposals strive to explain how the scale @xmath37 enters into spiral galaxy properties through some regularity in the cosmogony of dm halos .
but strong arguments exists against an intrinsic scale of the halos @xcite . and
it is hard to see how such an explanation could give an account of the multiple roles of @xmath37 in different systems @xcite .
the mond formula , conceived as _ exact _ , can not be a generic modification of the inertia aspect of newton s second law @xmath60 ( here @xmath61 is the sum of _ all _ forces acting on the particle including the newtonian gravity force ) , an alternative proposed originally by milgrom @xcite ( see a latter elaboration in ref . ) . to see why consider a binary system with unequal masses @xmath19 and @xmath62 evolving under mutual gravity alone .
the time derivative of @xmath63 , as calculated from eq .
( [ mond ] ) , does not vanish in general if at least one of the accelerations is comparable to or smaller than @xmath37 , since the @xmath36 s will generally not be equal .
so in the proposed interpretation , the mond formula does not conserve momentum . by the same token
it does not conserve angular momentum , nor energy .
milgrom sidesteps the problem by stipulating that the mond formula is only valid for test particles moving on a given background , e.g. , stars moving in the collective gravitational field of a galaxy .
we may inquire more generally , does the mond formula , or some closely related one , represent a modification of inertia in test particle motion ? to comply with the conservation laws we likely want the formula to arise from a lagrangian .
the kinetic part of the lagrangian should give us the @xmath64 part with whatever `` corrections '' are required .
however , it is a theorem that no mond - like dynamics exists that simultaneously has a newtonian limit for @xmath65 ( all accelerations are large ) , is galilei invariant , and is derivable from a _ local _
action @xcite .
( `` local '' means that the relevant lagrangian can be written as a single integral over volume . )
this prohibition is even more stringent from a relativistic standpoint @xcite .
accordingly milgrom introduced a _ nonlocal _
action , i.e. one which is a functional of complete orbits , but can not be reduced to an integral over a lagrangian @xcite .
this approach does not quite reproduce formula ( [ mond ] ) generically , but that formula is recovered for the special case of circular orbits . of course circular stellar orbits _ a la _ mond is really all that one needs to analyze data on spiral galaxies .
however , it is known that elliptical galaxies comprise highly radial stellar orbits too , yet elliptical galaxies also seem to be well described by mond @xcite , and so the nonlocal modified inertia approach might be found wanting here . on
the plus side one may mention the pioneer anomaly @xcite .
the pioneer 10 and pioneer 11 spacecraft , as they travelled almost radially in the outer parts of the solar system , were found to be subject to an anomalous sunward acceleration of order @xmath66 ( not far from @xmath37 ) which does not fall off measurably with distance from the sun .
were such an acceleration to reflect a generic gravitational field , it would have affected the outer planet ephemerides to an intolerable extent @xcite .
( an ephemerides is the calculated position of a celestial body as a function of time , past or future . ) by contrast , milgrom s nonlocal theory does predict different modifications of newtonian dynamics for radial ( pioneer ) and circular ( planetary ) orbits , though details have yet to be worked out @xcite .
milgrom has also speculated @xcite that modified inertia may have its origin in an effective interaction of bodies with the vacuum .
this is motivated by the well known fact that an object swimming through a fluid has its inertial mass increased by interacting with fluid degrees of freedom .
of course , by local lorentz invariance inertial motion through the vacuum is possible as usual .
however , accelerated motion is distinguished from inertial motion ( even before we come to inertia ) by the perception that the accelerated system is immersed in a thermal bath whose temperature is proportional to the acceleration ( unruh radiation @xcite ) .
accordingly , milgrom suggests that accelerated motion with respect to the vacuum may shape the inertial characteristics of objects in a way compatible with mond . an alternative speculation
@xcite is that mond style modified inertia and its particular scale @xmath37 may be generated by a symmetry in a higher dimensional spacetime , just as we think of ordinary ( newtonian or einstenian ) inertia as reflecting lorentz symmetry in minkowski spacetime . in one implementation of this program , with the extremely symmetric desitter spacetime playing the role of the embedding spacetime , a relation between the cosmological constant and @xmath37 surfaces , which is actually approximately satisfied if the observed acceleration of the hubble expansion is interpreted as reflecting a nonzero cosmological constant .
what if , instead of the modified inertia interpretation , we follow milgrom s alternative proposal @xcite to regard mond as a modification of newtonian gravity ? this entails a change in the r.h.s . of @xmath60 , with the new gravitational force possibly depending nonlinearly on its sources
can such implementation be protected from violation of the conservation laws ?
an affirmative answer to this query is readily available by starting from a modification of the lagrangian of newtonian gravity with milgrom s @xmath37 playing the role of characteristic scale , to wit @xmath67d^3x .
\label{lagrangian}\ ] ] here @xmath68 is some positive function , @xmath69 is the matter s mass density and @xmath70 , the field in the theory , is to be identified with the gravitational potential that drives motion , i.e. @xmath71 . the lagrangian s kinetic part is the most general one that depends only on first derivatives of @xmath70 , and is consistent with the isotropy of space ( only @xmath72 appears ) .
this aquadratic lagrangian theory ( aqual ) @xcite reduces to newton s in the limit @xmath73 , when it readily leads to poisson s equation for @xmath70 .
more generically it yields the equation @xmath74&=&4\pi g\rho , \label{aqual } \\ \mu(\surd x ) & \equiv & f'(x ) .
\label{f}\end{aligned}\ ] ] one identifies the @xmath36 function here with mond s @xmath36 .
comparison of the aqual equation ( [ aqual ] ) with poisson s for the same @xmath69 shows that @xmath75 where @xmath76 is some calculable vector field .
now in highly symmetric situations ( spherical , cylindrical or plane symmetry ) , use of gauss theorem with the aqual equation shows that @xmath77 has to vanish . in light of the relation @xmath78 ,
aqual then reproduces the mond formula ( [ mond ] ) exactly .
however , the curl term does not usually vanish in situations with lower symmetry ( milgrom and brada exhibit a curious exception @xcite ) .
we conclude that the mond formula is generically an approximation to the solution of the aqual equation .
how good an approximation ?
work by milgrom and brada @xcite has shown that often the curl is just a 5 - 10% correction .
this explains how the mond formula manages to be so successful while not constituting by itself a physically consistent theory .
now almost all mond fits to observed rcs are calculated using that formula exclusively .
it should thus be clear that modest departures of empirical rcs from mond s predictions do not constitute evidence against the paradigm .
the era of rc modeling with the full aqual eq .
( [ aqual ] ) is still in its early stages @xcite .
i mentioned in sec . 4 that the mond formula is inconsistent with the conservation laws .
the aqual theory is free of this problem .
this needs no special proof ( although explicit proofs exist @xcite ) : a theory based on a lagrangian which does _ not _ depend _ explicitly _ on coordinates , directions and time ( all true for aqual s lagrangian ) automatically conserves momentum , angular momentum and energy . at the level of the first integral ( [ integral ] ) of the aqual equation , it is the curl term which protects aqual gravity from violating the conservation laws .
aqual consistently embodies the weak equivalence principle ( all bodies , regardless of inner structure , and starting from the same initial conditions , follow the same path in a gravitational field ) .
as remarked by milgrom @xcite , the mond formula is ambiguous in this respect . in a star accelerations of individual ions
greatly exceed @xmath37 in magnitude .
thus for each ion @xmath36 should be very close to unity , and we would guess that @xmath79 for the star as a whole . accordingly ,
if regarded as a collection of ions , the star would follow a newtonian orbit in the galaxy s field .
this conclusion is entirely at variance with the evidence from the flat rcs , or from the dictate of the mond formula when applied to the star regarded as structureless .
aqual predicts that the center of mass ( cm ) of a collection of particles , no matter what the internal gravitational field , will follow the orbit determined by @xmath80 with @xmath70 the external potential coming from the aqual equation @xcite .
as mentioned , this acceleration will approximate that predicted by the mond equation .
aqual does indeed supply a physical basis for mond .
a system with sub-@xmath37 internal gravitational field immersed in an external field @xmath81 is predicted by aqual to exhibit quasi - newtonian internal dynamics @xcite .
originally milgrom conjectured this `` external field effect '' to explain why no acceleration discrepancy is evident in the loose open star clusters residing in the milky way @xcite : the galaxy s field , still stronger than @xmath37 in its inner parts , suppresses the mond behavior expected in light of the weak internal fields . by contrast , if the said system is immersed in an external field @xmath82 , which is , however , _
stronger _ than its internal fields , the internal dynamics is predicted to be quasi - newtonian , but with the effective gravitational constant @xmath83 @xcite .
a useful product of aqual is a formula for the force between two masses in motion about each other , e.g. a binary galaxy .
the mond formula can not be used directly here because , as mentioned in sec .
4 , it would suggest that the binary s center of mass should accelerate ( nonconservation of momentum ) . instead , @xmath70 has to be computed from the aqual equation , and
then @xmath84 has to be integrated over one galaxy s volume to obtain the force on it , as effectively done by milgrom @xcite . for point masses @xmath19 and @xmath62 separated by distance @xmath15 , the force is @xmath85 with @xmath86^{1/2}$ ] ; milgrom gives the graph of @xmath87 . in the newtonian limit ( @xmath88 ) , @xmath89 . in the deep mond limit ( @xmath90 )
, milgrom @xcite shows analytically that the force is @xmath91 $ ] .
this is most elegantly computed by exploiting the discovery that in that limit the aqual equation is conformally invariant , an approach that also yields the exact virial relation in the deep mond regime @xcite .
aqual permits a study of galaxy disc stability in the mond paradigm .
i mentioned in sec .
i that newtonian disk instabilities motivated the idea that dm halos surround disk galaxies . in the mond paradigm
there is not dm , so how do a fraction of the spiral galaxies in the sky avoid forming bars ? using a mixture of mond formula and aqual equation arguments , milgrom @xcite showed analytically that mond enhances local stability of disks against perturbations as compared to the newtonian situation .
brada and milgrom @xcite combined an @xmath92-body code with a numerical solver for the aqual equation to verify this and to demonstrate that mond enhances global stability .
both works conclude that the degree of stabilization saturates in the deep mond regime ; this means no disk in mond is absolutely stable .
christodoulou @xcite and griv and zhytnikov @xcite reach similar conclusions with @xmath92-body simulations based on just the mond formula .
recently tiret and combes @xcite studied _ evolution _ of bars in spiral disks using @xmath92-body simulations built on the aqual equation .
they find that bars actually form more rapidly in mond than in dm halo newtonian models , but that mond bars then weaken as compared to those in newtonian models .
the long term distribution of bar strengths predicted by mond is in better agreement with the observed distribution than that obtained from newtonian halo models .
aqual is also important for studies of two - body relaxation in stellar systems .
two - body relaxation is the process whereby distant gravitational encounters of pairs of stars in a galaxy or a cluster redistribute the system s energy and help it to approach some sort of equilibrium .
in newtonian theory two - body relaxation is excruciatingly slow . but
as binney and ciotti show @xcite , two - body relaxation in the framework of aqual proceeds faster . in fact , if the system is deep in the mond regime , the speedup factor is the square of the acceleration discrepancy .
a related effect is dynamical friction , whereby a massive object moving through a collection of lighter bodies , e.g. a globular star cluster coursing among the stars of its mother galaxy , loses energy by virtue of its kicking the background objects gravitationally as it sweeps by them . as a result the heavy object settles in the gravitational well confining it . in aqual for systems deep in the mond regime , dynamical friction
is speeded up over the newtonian value by one power of the acceleration discrepancy @xcite . both of the above phenomena open a new window onto the mond - newtonian gravity dichotomy .
_ suppose _ aqual is the correct nonrelativistic description of gravity .
in regard to rcs of spiral galaxies , newtonian theory with dm could be `` saved '' by imagining the dm in each galaxy to be distributed in just such a way that the resulting total @xmath93 is just the @xmath70 generated in aqual theory by the baryonic ( visible ) matter .
admittedly , the required dm distribution might be peculiar ( negative density somewhere @xcite , or incompatible with cosmogonic simulations ) . but leaving these options aside , observational uncertainties might make it difficult to tell the two theories apart , as recent fits of rcs of nearby galaxies based on high quality data testify @xcite
. yet consideration of the mentioned dissipative effects might remove the effective degeneracy between the two models of a galaxy offered by the two theories . in view of the generic prediction of newtonian cosmogonic simulations that the dm distribution in a galaxy rises to a cusp at the center ( see sec .
i ) , newtonian dynamical friction should have caused the globular clusters of many a dwarf spheroidal galaxy to already have merged at its center and there lost their integrity @xcite . while globular clusters are indeed absent from several such dwarf companions of the milky way , e.g. draco , sextans and carina @xcite , the fornax dwarf spheroidal has five globular clusters near , but _ not at _ , its center .
goerdt , moore , et al .
@xcite regard this as showing that the dm near the core of fornax has a flat density profile , in contradiction to the dm simulations .
in fact , there is other growing evidence negating dm density cusps @xcite .
now , some of the dwarf spheroidal satellites of the galaxy including fornax are in the deep mond regime . following binney and ciotti @xcite , snchez - salcedo , reyes - iturbide and hernandez @xcite point out that the mentioned speedup of settling by dynamical friction in aqual should long ago have caused the globular clusters in fornax to amalgamate at its center .
aqual thus has a problem explaining the observations of fornax ; however , as we see the dm alternative has its problems too .
can mond give a picture of the late stages of galaxy formation ?
a step in this direction was taken by nipoti , londrillo and ciotti @xcite on the basis their earlier work with @xmath92-body codes incorporating the full aqual equation @xcite .
they simulate the dissipationless collapse of a cloud of particles which ends up in the mond or deep mond regimes .
they note that phase mixing ( the disappearance of structure in the initial conditions due to commingling of particle orbits ) proceeds slower than in the counterpart newtonian simulations , and that the final velocity distribution is anisotropic and favors radial orbits .
current thought in cosmogony is that elliptical and spheroidal galaxies are the outcome of anisotropic collapse _ after _
star formation is complete .
nipoti et al .
note that the end - products of some of their simulations resemble , in their surface density and velocity dispersion profiles , the dwarf - elliptical and dwarf - spheroidal galaxies in the sky .
but other final states have trouble fitting in all the accepted empirical correlations between luminosity , radius and velocity dispersion of elliptical galaxies .
why is a relativistic version of mond important ?
although most astronomical systems with significant acceleration discrepancy are nonrelativistic in the extreme , there are two important exceptions .
cosmology , which is important in itself and as a framework for the process of galaxy formation , is widely regarded as requiring the presence of much dm on the largest scales , and cosmology is universally believed to require relativistic treatment .
gravitational lensing by galaxies and clusters of galaxies has become a salient astronomical tool in the last two decades , and , since it discloses acceleration discrepancies , it is commonly regarded as supporting the need for dm . since light propagates with speed @xmath94 , nonrelativistic mond can not even begin to formulate the problem of gravitational lensing .
the above cases drove the relativistic formulation of aqual @xcite , which i shall refer to as raqual .
raqual retains einstein s equations of gr as a tool for deriving the spacetime metric @xmath95 from the energy - momentum tensor of the matter ( one of whose components represents matter density ) .
raqual further stipulates that matter and radiation play out their dynamics , not in the arena of @xmath95 , but in that of @xmath96 , where @xmath97 denotes a scalar field with dimensions of squared velocity .
since all matter is treated alike in this respect , the exquisitely tested weak equivalence principle is safeguarded , but since the metric governing gravitational field dynamics is different from the one experienced by matter , the strong equivalence principle fails .
this is essential , for otherwise we should end up back with gr as the relativistic gravity theory , and gr in the nonrelativistic limit leads to newtonian theory , not to mond .
how is @xmath97 determined in raqual ?
one takes the lagrangian for @xmath97 to be the covariant version of that for @xmath70 in aqual , eq .
( [ lagrangian ] ) .
this means replacing @xmath98 as well as @xmath99 , where @xmath100 is the determinant of the matrix of components of @xmath95 .
but in raqual one does not include a term in lieu of @xmath101 ; the coupling between @xmath97 and matter , which is to provide the source of the equation for @xmath97 in accordance with its lagrangian , is automatically generated by the @xmath102 factor in the metric @xmath103 with which we built the _ matter _
s lagrangian . for time
independent situations with nonrelativistic matter sources , the @xmath97 equation reduces to eq .
( [ aqual ] ) with @xmath104 .
however , it would be a mistake to conclude that in raqual @xmath97 is the gravitational potential . in gr for weak gravity and nonrelativistic motion ,
the potential in which matter moves , @xmath93 , is related to the metric by @xmath105 ( @xmath106 is the time coordinate ) . because , in this linearized theory , the most relevant einstein equation reduces to poisson s , @xmath93 coincides with the usual newtonian potential . in raqual
@xmath95 still satisfies einstein - like equations , so the above relation applies as well .
but the stipulation that matter s dynamics go forward in the metric @xmath103 determines the nonrelativistic potential in which matter moves , now called @xmath70 , to be @xmath107 .
writing in linearized theory @xmath108 , we have @xmath109 .
now to first approximation @xmath110 ; thus to leading order @xmath111 consequently , in raqual the gravitational potential traced by the matter s motions is the sum of the newtonian potential and the aqual scalar field , both generated by the same baryonic matter .
the @xmath97 does the job of the dm s potential in conventional approaches , but its source is ordinary matter . let us compare raqual with the successful aqual .
we take the @xmath36 in eq .
( [ aqual ] ) with @xmath104 to be approximately equal to its argument for @xmath112 , and to level off at a value @xmath113 when @xmath114 . then for strong @xmath115 poisson s equations gives @xmath116 so that the true potential is @xmath117 .
this means gravity is newtonian with effective gravitational constant @xmath118 . for @xmath112 , the raqual equation
tells us that @xmath119 is much stronger than @xmath120 .
consequently , by eq .
( [ sum ] ) , @xmath121 so that in the weak field regime raqual s gravitational field is close to aqual s .
thus in raqual we have a relativistic theory which in the nonrelativistic limit inherits the good traits of mond by way of aqual . however , on the relativistic front raqual hit two roadblocks . already the original paper @xcite remarks that perturbations of @xmath97 from a static background for which @xmath112 , e.g. @xmath97 waves travelling in the reaches of a galaxy , propagate superluminally , i.e. , outside the light cone of @xmath103 .
nowadays one hears the view that superluminality by itself is not sufficient reason for causal problems @xcite .
this is by no means the standard view , and so the specter of acausality motivated disillusionment with raqual @xcite , and formulation of a two - scalar field theory , phase - coupled gravity ( pcg ) @xcite intended to prevent superluminality . in the pcg
lagrangian the new scalar @xmath122 contributes a quadratic kinetic part , just as does @xmath97 ; there is also a simple coupling between @xmath122 and the kinetic term for @xmath97 which gives the theory its mond character .
although pcg generates an intolerable drift of the kepler constant @xmath123 in the solar system , and marginally contradicts the observed precession of mercury s perihelion @xcite , the theory has merits for both galaxy dynamics and cosmology @xcite .
however it may be , raqual and pcg were both finally disqualified as consistent representations of gravity by their inability to cope with observations of gravitational lenses . discovered in 1979 , gravitational lenses have been intensively studied ever since . in strong gravitational lensing a cluster of galaxies ( but sometimes a single galaxy ) forms a few images of a single quasar in the background by bending the incoming light rays with its gravitational field .
this provided a novel tool for determination of masses of large astronomical systems by an elementary use of gr . by the late 1980 s it was apparent that masses of clusters so determined were larger than the observed baryonic mass in them , and commensurate with the masses determined from motions of the galaxies comprising them .
the novelty was that the acceleration discrepancy was now put in evidence by a relativistic phenomenon .
many pounced upon the finding as added support for dm s presence in clusters .
raqual can not deal with the new evidence ; the reason is instructive .
maxwell s equations are known to be conformally invariant : replacement of the metric @xmath95 with which they are formulated by @xmath124 does not change the form of the equations .
it follows that study of light propagation can not distinguish between the metrics @xmath95 and @xmath103 of raqual .
accordingly , although maxwell s equations are supposed to be written with @xmath103 , the metric for studying gravitational lensing might just as well be @xmath95 .
but this last obeys einstein s equations whose sources are the baryonic matter s energy - momentum tensor and that of @xmath97 .
however , for systems as distended as clusters or galaxies , this last is negligible .
the light bending , then , is predicted by raqual to be nearly the same as predicted by gr _ without _ dm . yet there is observational evidence that the inner parts of clusters gravitationally lense light more strongly than would be expected from gr _ with no _ dm . the same problem recurs in pcg , in which the two metrics are , again , conformally related in the same manner as in raqual .
it is thus quite clear that if a scalar field is to be the vehicle for mond effects , then regardless of the form of its dynamics , the relation between the metrics @xmath103 and @xmath95 must be non - conformal .
i attempted @xcite to construct such a coupling out of scalar field alone , to wit @xmath125 with @xmath126 and @xmath127 functions of the invariant @xmath128 .
however , sanders and i found that avoidance of propagation of gravitational waves which are superluminal with respect to the matter metric @xmath103 requires that @xmath129 , while @xmath130 is required so that both metrics have lorentzian signatures @xcite .
we then calculated that for equal sources of einstein s equations , the light bending in the proposed modification of raqual is _ weaker _ than that in gr .
as mentioned , in systems of the scale of galaxies and clusters of galaxies , the scalar field provides very little energy or momentum density as compared to the matter .
thus in regard to light ray bending , the new theory performs worse than raqual .
actually the above conclusion could be avoided if @xmath131 is a _
timelike _ 4-vector .
however , we would expect that deep inside or near a galaxy , or a massive cluster of galaxies which is certainly virialized , the scalar field , whose principal source is the matter therein , should be quasistatic .
then @xmath131 is expected to be spacelike , i.e. to give lensing weaker than needed .
the problem is thus still present . to overcome it sanders @xcite proposed to replace @xmath131 in its
above role by a nondynamical vector @xmath132 which in an isotropic cosmological model points precisely in the time direction , and approximately so in the presence of mass concentrations .
he further related the metrics by @xmath133 sanders further supposed the action for the scalar field @xmath134 , which supersedes @xmath97 , to be of raqual form , or more generally with the function @xmath68 depending also on the scalar @xmath135 .
the timelike nature of the vector is conducive to enhancement of the lensing , and such a theory , named stratified theory , can cope with the observations of gravitational lensing .
however , it was clear to sanders that this is just a toy theory since a globally timelike vector field , such as @xmath136 , determines a universal preferred frame of reference , an aether .
additionally , since the vector field is supposed constant , the theory is not covariant .
the mentioned problems are avoided by making the vector field a dynamical one , as first done in t@xmath0v@xmath137s , my tensor - vector - scalar theory of gravity @xcite .
t@xmath0v@xmath137s retains the relation ( [ gtilde ] ) between a gravitational metric @xmath95 and a physical ( or observable ) metric @xmath103 .
the matter lagrangian is built exclusively with @xmath103 ; this guarantees that the weak equivalence principle will be obeyed precisely .
the conventional gr einstein - hilbert lagrangian is used to give dynamics to @xmath95 , which then trivially induce dynamics for @xmath103 .
the scalar field @xmath134 is provided with a raqual type lagrangian . for convenience
this is couched in terms of a quadratic form in the 4-gradient of @xmath134 in interaction with a second scalar field , @xmath122 , which also occurs in a potential - like term .
( however , no derivatives of @xmath122 enter , so @xmath122 is not dynamical , and if eliminated , the explicit aquadratic lagrangian for @xmath134 shows up . )
the mentioned quadratic form is not the usual one , but rather @xmath138 ; it is introduced to forestall the superluminal propagation that afflicts raqual .
the kinetic part of the @xmath136 lagrangian is quadratic in its first derivatives , and exactly analogous to that for a gauge field .
this is not the only quadratic form possible , but it is the one in which the einstein equations for @xmath95 will not have second derivatives of @xmath136 in their sources . as mentioned , it is imperative that @xmath136 be a timelike vector . in t@xmath0v@xmath137s
this is accomplished by introducing , alongside the kinetic term , a lagrange multiplier term of the form @xmath139 which forces the norm of @xmath136 to be negative ( and @xmath136 to be of unit length as a bonus ) . whereas raqual is a one parameter ( @xmath37 ) theory with one free function ( @xmath36 ) , t@xmath0v@xmath137s has one free function , @xmath140 [ distinct from the @xmath68 in eq .
( [ lagrangian ] ) ] that determines the _ shape _ of the potential - like term , and three parameters in addition .
one of these is a scale of length @xmath32 in the @xmath134 lagrangian that sets the strength of the potential , while a dimensionless one , @xmath141 , determines the scale of its argument .
the third parameter , @xmath142 , also dimensionless , sets the strength of the @xmath136 lagrangian .
one can form the acceleration scale @xmath143 which parallels the role of milgrom s constant in mond and aqual .
it is found that the limit of t@xmath0v@xmath137s with @xmath144 , @xmath145 and @xmath146 amounts to gr .
in fact @xmath65 in this limit , so that all accelerations are to be regarded as strong , and gravitation as conventional . how about nonrelativistic motion ?
let @xmath147 be the coeval value of @xmath134 in the cosmological model in which the nonrelativistic system is embedded
. this will be the asymptotic boundary value of @xmath134 for any local solution of the @xmath134 equation . from a linearized version of einstein s equations , and use of the relation ( [ gtilde ] ) , it follows @xcite that the effective nonrelativistic gravitational potential is @xmath148 ( this corrects a sign error in refs . and ) . here
@xmath93 comes from the usual poisson equation while @xmath134 is found from the scalar equation ; in both the source is the baryonic mass density @xmath69 .
thus far all studies have assumed that @xmath149 and @xmath150 , in which case eq .
( [ sum2 ] ) essentially coincides with eq .
( [ sum ] ) in raqual . for this same reason
any time variation of @xmath151 @xcite should be negligible .
this is of particular significance for the above derived value of @xmath37 which , strictly speaking , includes a factor @xmath151 [ for example eq . ( 6.7 ) of ref . ] . for a static situation
the equation for @xmath134 takes a form like eq .
( [ aqual ] ) , with @xmath36 ( essentially @xmath152 ) expressed in terms of @xmath153 through t@xmath0v@xmath137s s free function @xmath140 . by a suitable choice of
@xmath68 one can reproduce any @xmath36 function that would be relevant in aqual , with the scale thereon agreeing with @xmath37 in eq .
( [ a0 ] ) . by analogy with our discussion of raqual
we see that t@xmath0v@xmath137s can have mond phenomenology and also a newtonian limit .
the transition between newtonian and mond regimes takes place as @xmath153 sweeps through the value @xmath37 , near the point where @xmath154 . actually for systems strongly departing from sphericity ,
the situation is not as clear , but it seems that at the nonrelativistic level t@xmath0v@xmath137s implements the mond paradigm overall . when it comes to details , my original choice for @xmath68 @xcite can be criticized : while it leads to flat outer rcs , it does not give a satisfactory account of the transition part of the rcs of several well measured galaxies , including our own @xcite . by contrast , the mond formula with the `` simple '' variant of migrom s @xmath36 ( fig .
[ figure : fig3 ] ) leads to very good fits for many rc s @xcite .
forms of the t@xmath0v@xmath137s function @xmath68 which would do a correspondingly good job have been proposed by zhao and famaey @xcite , sanders @xcite and famaey , et al .
@xcite .
sanders has also proposed a variation on t@xmath0v@xmath137s involving a second scalar @xmath122 with pcg - like dynamics instead of raqual ones @xcite .
this bi - scalar tensor vector theory ( bstv ) has three free functions and a free parameter .
the theory is not especially needed to obviate superluminal propagation since t@xmath0v@xmath137s seems to do well on this @xcite .
however , bstv , like pcg , is a more appropriate frame for generating cosmological evolution of @xmath37 . since numerically @xmath155
, it is often argued that @xmath37 must be determined by cosmology , and should thus vary on a hubble timescale @xcite .
as mentioned , @xmath37 in t@xmath0v@xmath137s is essentially set by the parameters @xmath32 and @xmath141 ; it should be nearly constant in the expanding universe .
it is otherwise in bstv .
discrimination between constant and evolving @xmath37 may be possible with good rcs of disk galaxies at redshifts @xmath1565 .
such curves are just now coming into range .
in fact , the data in ref . put the @xmath157 galaxy bzk-15504 right on the mond derived tully - fisher law with the standard value for @xmath37 @xcite . thus the meagre data available today are consistent with no @xmath37 evolutions on the hubble timescale .
recently zlosnik , ferreira and starkman @xcite have clarified the relation between t@xmath0v@xmath137s and the so called einstein - aether gravity theories in which a timelike unit vector field ( but no scalar ) plays a role alongside the metric @xcite .
they do this by re - expressing t@xmath0v@xmath137s solely in terms of the observable metric @xmath103 and the vector field .
the @xmath134 is eliminated with the help of the constraint imposed through the lagrange multiplier term .
there emerges an einstein - aether like theory in which the metric @xmath103 also satisfies einstein - like equations .
however , in contrast to orthodox einstein - aether theories , the vector kinetic action in the theory in question is a generic quadratic form in the vector s derivatives contracted into polynomials of the vector s components .
this form of t@xmath0v@xmath137s is more complex than the original one , but there are some circumstances in which it may be more convenient in ferreting out consequences of the theory . in t@xmath0v@xmath137s in either form the vector is normalized with respect to @xmath95 , not @xmath103 .
zlosnik , et al .
have also proposed a variant tensor - vector theory with a timelike vector normalized with respect to @xmath103 , and which also has mond like behavior @xcite .
this is constructed by taking the vector s action as a function @xmath158 of @xmath159 , the quadratic form in the derivatives of the vector field of orthodox einstein - aether theories .
the zlosnik , et al .
theory has four parameters : a length scale and three dimensionless parameters .
the form of the free function @xmath160 can be deduced approximately from the requirement that mond arise in the nonrelativistic quasistatic limit , and from the stipulation that a cosmology built on this theory shall have an early inflationary period and an accelerated expansion at late times . in summary
, the relativistic implementations of mond involve either one or three free functions ; those with one free function have either three or four free parameters .
as stressed earlier , the particular structure of t@xmath0v@xmath137s reflects the desire to encompass in the mond paradigm the observation that what passes , from a dynamical point of view , for dm in galaxies and clusters of galaxies also lenses light to a commensurate degree .
how does t@xmath0v@xmath137s measure up to the task ?
now the measured gravitational lensing by galaxies and clusters of galaxies takes place over cosmological distances .
the light rays in t@xmath0v@xmath137s are null geodesics of the observable metric @xmath103 ; to compute @xmath103 one needs @xmath161 and @xmath95 .
isotropic cosmological models in t@xmath0v@xmath137s closely resemble the corresponding cosmological models of gr ; they sport a @xmath162 pointed precisely in the time direction and feature very slow change of @xmath134 @xcite .
a consequence is that @xmath103 in a t@xmath0v@xmath137s isotropic cosmological model differs little from the metric in the corresponding gr cosmology .
thus the cosmological facet of gravitational lensing is very much like in gr .
the second facet is the local bending of light rays in a mass vicinity . as in isotropic cosmological models , so in static situations , the solution of the vector s equation has @xmath162 pointed precisely in the time direction . to compute the bending in linearized theory , the approximation commonly used in the business
, one also needs the scalar field @xmath134 and the metric @xmath95 , both to first order in @xmath93 .
the final result is that the line element takes the form @xmath163 with @xmath70 given by eq .
( [ sum2 ] ) .
note that the same potential @xmath70 appears in both terms of this isotropic form of the line element .
hence light ray bending , which leans on both to equal degree , measures the same gravitational potential as do dynamics which are sensitive only to the temporal part of the line element .
this mirrors the situation in gr whose line element is obtained by sending @xmath164 in eq .
( [ le ] ) .
thus half of the problem that plagues raqual and similar theories is overcome : the acceleration discrepancy makes itself felt equally through the gravitational lensing as through the dynamics .
the second half of the problem revolves about the mass distribution that generates @xmath70 . in gr @xmath70
is all there is , and its laplacian , as determined from the lensing observations or from the dynamics , will give the _ total _ ( baryonic plus dark ) mass distribution directly .
because dm is not seen directly , this prediction can only be judged by the plausibility of the derived distribution of dm .
by contrast in t@xmath0v@xmath137s the measured @xmath70 is to be decomposed into two parts in the manner of eq .
( [ sum2 ] ) , with each generated by the same _ baryonic _ mass density @xmath69 , one part through poisson s equation , and the second through a highly nonlinear aqual type equation . evidently ,
in a t@xmath0v@xmath137s model of a gravitational lense , the baryonic matter will be distributed differently from the total matter in a gr model of the same lense . and when the lensing system is not spherically symmetric , the centers of the two distributions may be offset .
chiu , ko and tian @xcite have explored light ray bending by a pointlike mass @xmath24 in t@xmath0v@xmath137s .
they note that the deflection angle in the deep mond regime [ impact parameter @xmath165 approaches a constant , as might have been expected from naive arguments , but is less predictable in the intermediate regime @xmath166 .
thus calculations of lensing based on a mixture of mond and gr motives @xcite can easily mislead .
chiu et al .
work out the lens equation in t@xmath0v@xmath137s , which controls the amplifications of the various images in the strong lensing of a distant source , and remark that for two images the difference in amplifications is no longer unity as in gr , and may depend on the masses . with a photometric survey like the sloan digital survey it may be possible to check for this effect . finally , these authors work out the gravitational time delay in t@xmath0v@xmath137s , which could be of use in interpreting differential time delays in doubly imaged variable quasars . more phenomenologically oriented , zhao , bacon , taylor and horne @xcite compare t@xmath0v@xmath137s predictions with a large sample of galaxy strong lenses which each produce two images of a quasar .
they model the galaxy baryonic mass distribution either by point masses or by the popular hernquist profile .
galaxy masses are estimated by comparing observations both with predicted image positions and with predicted amplifications ratios ; the two methods are found to give consistent results , themselves well correlated with the luminosities of the galaxies .
the corresponding mass - to - light ratios are found to be in the normal range for stellar populations , with some exceptions .
how frequently should strong lensing occur ?
this question is taken up in the context of t@xmath0v@xmath137s by chen and zhao @xcite . again modeling the mostly elliptical galaxies with hernquist profiles , they compute the probability of two images occurring as a function of their separation .
their prediction falls somewhat beneath the frequency observed in the lensing surveys , though they consider this still acceptable .
the predictions are sensitive to the assumed mass profile , as well as to the assumed shape of the @xmath36 function , which chen and zhao assume to switch brusquely from linear in the argument to unity
. gravitational lensing by the colliding galaxy clusters 1e0657 - 56 has been claimed to give theory independent proof of dm dominance at large scales @xcite . in this system ( fig .
[ figure : fig6 ] ) a smaller cluster , the `` bullet '' , has crashed through a larger one and the intracluster gas of both has been stripped by the collision , the bullet s gas trailing behind its galaxy component
. weak lensing ( distorted but unsplit images ) mapping shows the lensing mass to be concentrated in the two regions containing the galaxies , rather than in the two clouds of stripped gas which contains the lion s share of baryonic mass @xcite .
collisionless dm would indeed move together with the galaxies .
hence the inference that much dm continues to accompany the bullet .
angus , famaey and zhao @xcite ( see also ref . ) point out that in the very asymmetric system 1e0657 - 56 , mond , aqual and t@xmath0v@xmath137s all predict substantially different gravitational field distributions ( compare eqs .
( [ mond ] ) and ( [ integral ] ) ) , a situation which confuses clowe et al.s `` theory independent '' inference . whereas angus , famaey and zhao consider it possible to explain the lensing with a reasonable purely baryonic matter distribution , a later paper by angus , shan , et al .
@xcite concludes that dark matter is needed after all .
this is hardly surprising ; as we saw in sec . 3 , pure mond
does not fully account for the acceleration discrepancy in the dynamics of quiescent galaxy clusters @xcite .
but dm models of the bullet clusters within gr are not without their problems .
farrar and rosen @xcite note that the relative velocity of the clusters is too high as compared to those seen in dm simulations of structure formation . to remove the contradiction they propose that a non - gravitational attraction of a new sort acts only between clumps of dm .
but is assuming existence of dm together with a new interaction specific to it more parsimonious than a modification of standard gravity such as mond ?
with my original choice of @xmath68 , t@xmath0v@xmath137s cosmological models with baryonic matter content alone can be very similar to the corresponding gr models by virtue of the scalar @xmath134 s energy density remaining relatively small @xcite .
in particular chiu et al .
@xcite note that these t@xmath0v@xmath137s models give a reasonable relation between redshift and angular distances , and are thus as effective as gr models in providing the scaffolding for the analysis of cosmologically distant gravitational lenses .
how would changing @xmath68 affect the evolution of a cosmological model ?
this is studied exhaustively by bourliot et al .
@xcite following an earlier exploration by skordis et al .
bourliot et al .
display a large set of variants of my @xmath68 for which , while the scalar energy density may not be negligible , it tracks or mimicks the behavior of another energy component of the model , e.g. the radiation s during radiation dominance .
they also characterize shapes of @xmath68 which can lead to future singularities in cosmology , and which presumably should be avoided .
mond critics have always held up the complicated power spectrum of cosmological perturbations ( background radiation or baryons ) as a proof that dm is needed on the cosmic level ; after all the spectrum is said to be well fit by the `` concordance '' dm model of the universe .
this argument took for granted that mond could never measure up to the test . with t@xmath0v@xmath137s on the scene one can face the question technically . in a massive work skordis has provided the full covariant formalism for evolution of cosmological perturbations in t@xmath0v@xmath137s @xcite . and using this skordis et al .
@xcite have shown that , without invoking dark matter , t@xmath0v@xmath137s can be made consistent with the observed spectrum of the spatial distribution of galaxies and of the cosmic microwave radiation if one allows for contributions by massive neutrinos ( in the still allowed mass range ) and the cosmological constant . in this approach the role of dark matter in standard cosmology is taken over by a feedback mechanism involving the scalar field perturbations .
dodelson and liguori @xcite have independently calculated perturbation growth , and stressed that it is rather the vector field in t@xmath0v@xmath137s which is responsible for growth of large scale structure without needing dm for this @xcite .
it thus seems there is potential in t@xmath0v@xmath137s to do away with cosmic dm , as well as with dm in galaxies .
much attention is commanded today by the mystery of the `` dark energy '' , the agent responsible for the observed acceleration of the hubble expansion .
such agent is obligatory in the context of gr cosmological models .
quite naturally many have wondered whether t@xmath0v@xmath137s could obviate this need .
diaz - rivera , samushia , and ratra @xcite have found exact desitter solutions of t@xmath0v@xmath137s cosmology which can represent either early time inflation epochs or the late time acceleration era . in another t@xmath0v@xmath137s study ,
hao and akhoury @xcite have concluded that with a suitable choice of the t@xmath0v@xmath137s function @xmath68 , the scalar field can play the role of dark energy . and
zhao @xcite maintains that with zhao and famaey s choice of @xmath68 @xcite , cosmological models can be had that evolve at early times like those of standard cold dark matter cosmology , and display late time acceleration with the correct present hubble scale , all this without assuming dm or dark energy . for the related einstein - aether theory , zlosnik et al .
@xcite remark that with suitable choice of their theory s @xmath158 , the vector field can both drive early inflation as well as double for dark energy at late times .
it may be unnecessary to go far from earth to tell t@xmath0v@xmath137s , bstv and similar theories apart from gr ; the solar system can be turned into a sieve for the correct modified gravity . between any two bodies in the solar system
there is an extremum point , strictly a saddle point , of some relevant field . in aqual
this would be the @xmath70 potential and in t@xmath0v@xmath137s the @xmath134 field .
near such a point the gradient of the said field is small , and departures from newtonian gravity are significant ( @xmath36 in aqual falls short of unity , and the derivative of @xmath68 in t@xmath0v@xmath137s is driven away from the pole which signals newtonian behavior ) . a detailed study by magueijo and me @xcite with t@xmath0v@xmath137s shows that the anomalous regions , e.g. between sun and jupiter and earth and moon , are small , but gives some hope that with the fine guidance of space probes like esa s lisa - pathfinder ( scheduled for liftoff in october 2009 ) , it may be possible to diagnose mond - like effects . well away from the saddle points , but still in the region occupied by the planets ,
t@xmath0v@xmath137s predicts small departures of @xmath134 from @xmath16 form , i.e. small departures from newtonian behavior as encoded in @xmath70 @xcite .
similar predictions issue from bstv @xcite .
these are constrained by the observed constancy of kepler s constant @xmath123 out to the orbits of the major planets , and by the absence of significant departures from gr s predictions for the precessions of the perihelia of mercury and the asteroid icarus . according to sanders
, t@xmath0v@xmath137s can satisfy these constraints with choices of the function the @xmath68 proposed by refs . and ,
which , as mentioned in sec . 7 , also lead to correct galaxy rcs .
bstv can do just as well when its extra parameter , @xmath167 , is chosen properly @xcite .
both t@xmath0v@xmath137s and bstv with the above choices predict that beyond 100 astronomical units ( au ) from the sun , there is a fairly @xmath15-independent attractive component of the sun s gravitational field of strength @xmath168 .
the pioneer anomaly , as measured at distances beyond 20 au , is also @xmath15-independent but stronger .
is this evidence for mond ?
unfortunately , the interpretation of the pioneer anomaly as an almost constant sunward acceleration seems to clash with limits on the variation of kepler s constant set with spacecraft at the distance of uranus and neptune @xcite and with the latest ephemerides of the outer planets @xcite .
it thus seems prudent to suspend judgment until a prosaic origin for pioneer anomaly , e.g. drag by unknown matter , can be excluded with high probability .
the above focuses on the nonrelativistic limit . how do relativistic mond theories fare with regard to the lowest order relativistic effects in the solar system ? as for any metric theory of gravity , these effects can be calculated from the post - newtonian ( ppn ) parameters of the theory of choice . for any metric theory it is possible to parametrize the first and second order departures of the metric from minkowski form in terms of a set of ten intuitive looking potentials @xcite .
the newtonian potential @xmath93 figures in the list , and occurs in the corrections both linearly as well as squared ; two others are @xmath169 where @xmath170 denotes the fluid velocity in the system .
the post - newtonian ( ppn ) parameters are the dimensionless coefficients multiplying the diverse potentials in the correction .
some examples will clarify this .
the correction to the space - space part of the metric , @xmath171 , is @xmath172 ; it defines the ppn parameter @xmath3 .
the correction to the temporal metric component , @xmath173 , starts with the terms @xmath174 , which define the ppn parameter @xmath175 , and also includes a @xmath176 term whose coefficient brings in a further ppn parameter , @xmath177 .
each time - space metric component @xmath178 gets corrections proportional to the @xmath179 and @xmath180 ( @xmath181 denotes the velocity of the chosen coordinate system with respect to the cosmological matter comoving frame ) . with these last terms come two additional ppn parameters , @xmath182 and @xmath183 , which are called preferred frame parameters . a third one , @xmath184 ,
is associated with a correction of the form @xmath185 to @xmath173 .
there are four more ppn parameters for a total of ten . the parameters @xmath175 and @xmath3 are both unity for gr and for some of it competitors .
they have also been computed to be unity in t@xmath0v@xmath137s @xcite , which thus fares as well as gr in reference to gravitational light bending , perihelia precessions of the planets , and the radar time delay .
sanders @xcite has provided a simple argument that @xmath175 and @xmath3 are always unity in a class of theories ( including bstv ) which start from the disformal relation ( [ gtilde ] ) .
next in order of relevance are the preferred frame parameters @xmath186 and @xmath184 .
these vanish in gr ; after all , gr has no preferred frames . in t@xmath0v@xmath137s or bstv at least one of the @xmath187s
should be nonvanishing because the vector @xmath132 establishes a locally preferred frame , with the vector pointing out the time direction in that frame .
sanders argues heuristically that the @xmath187s should be strongly suppressed in both theories @xcite . but
explicitly computing the @xmath187 s for relativistic mond theories should be a high priority because they are subject to tight experimental bounds .
the odyssey in search of a relativistic embodiment of the mond paradigm has led , not to one relativistic mond , but to many .
t@xmath0v@xmath137s by itself exemplifies a family of theories .
first there is the freedom in the choice of the function @xmath68 , which is still only modestly constrained @xcite .
next , the coefficient of the second term in the factor @xmath188 in the scalar s lagrangian can be changed to any other negative number . in t@xmath0v@xmath137s
the lagrangians for @xmath134 and @xmath136 are formulated upon the metric @xmath189 ; a different but related theory emerges if one uses instead the background of @xmath190 .
additionally one can replace the aquadratic lagrangian for @xmath134 by a pcg - style lagrangian for two scalars , as indeed done in sanders bstv .
finally , one can altogether dispense with the scalar fields and go the way of the zlosnik - ferreira - starkman theory @xcite . while the lack of uniqueness of relativistic mond is a nuisance
, it does add needed flexibility to the search for a `` final '' fundamental theory with which to underpin the mond paradigm .
it is already clear that some of the above mentioned theories may not be in full accord with the facts .
for example , the published t@xmath0v@xmath137s has an exact mond limit at low accelerations , yet as mentioned in sec . 3
, mond can not handle the dynamics of clusters of galaxies without invoking additional unseen matter .
an interesting escape is suggested by sanders @xcite . in bstv
cosmology , just as in the pcg one @xcite , the @xmath122 field can undergo oscillations which generate bosonic particles early on .
if the bstv parameters are right , these bosons can be trapped by clusters ( but not by galaxies ) , and can thus comprise the additional unseen matter .
the charm of this resolution is that this new cold dark matter emerges from the modified gravity theory itself , and is not a separate invention .
relativistic mond as here described has developed from the ground up , rather than coming down from the sky : phenomenology , rather than pure theoretical ideas , has been the main driver .
actually a large industry flourishes on the sidelines with imaginative ideas from first principles regarding the essence of mond .
i have not touched here on these motley approaches because they have given so little that is observationally viable .
neither have i dwelt here on modified gravity theories that are not mond motivated or oriented .
but the time may be ripe for turning to a more deductive approach to mond .
now that we have some idea of what constitutes a viable relativistic mond theory , it should be easier to single out theoretical frameworks which might yield a promising candidate for the fundamental mond theory either as a limiting case , as an effective theory , after dimensional reduction , etc .
to give an example , let us recall that brans - dicke modified gravity plus maxwellian electromagnetism in the real world can both be recovered by dimensional reduction of 5-d einstein gravity theory .
t@xmath0v@xmath137s has the same number of degrees of freedom as that pair of theories ; specifically , its vector field has three degrees of freedom on account of the normalization condition , and the electromagnetic vector potential also has three on account of gauge freedom .
might some variant of t@xmath0v@xmath137s arise from dimensional reduction of a pure gravity theory in 5-d ?
if so , this would both ameliorate the common feeling that t@xmath0v@xmath137s is unduly complicated , and point the way to a lode for theories which might do better justice to the observations in the spirit of the mond paradigm .
i thank mordehai milgrom , bob sanders and stacy mcgaugh for many useful remarks on the original manuscript , and the last two as well as douglas clowe for providing figures .
research on this subject has been supported by grant 694/04 from the israel science foundation established by the israel academy of sciences and humanities .
finzi , a. , 1963 , mon . not .
soc . , * 127 * , 21 ; tohline , j. e. , 1982 , in _ internal kinematics and dynamics of galaxies _ , edited by a. athanassoula ( dordrecht : reidel ) , p.205 ; sanders , r. h. , 1984 , astron .
astroph . , * 136 * , l21 ; and 1986 , astron .
astroph . , * 154 * , 135 ; goldman , i. , 1986 , astron . astroph . , * 170 * , l1 ; kuhn , j. r. and kruglyak , l. , 1987 , astroph .
, * 313 * , 1 ; and others . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ jacob bekenstein _ received his ph .
d. ( 1972 ) from princeton university .
he worked at the ben gurion university in israel , where he became full professor in 1978 and the arnow professor of astrophysics in 1983 . in 1990
he moved to the hebrew university of jerusalem where he is the polak professor of theoretical physics .
a member of the israel academy of sciences and humanities , of the world jewish academy of sciences , and of the international astronomical union , bekenstein is a laureate of the rothschild prize and of the israel national prize .
his scientific interests include gravitational theory , black hole physics , relativistic magnetohydrodynamics , galactic dynamics , and the physical aspects of information theory .
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ | no more salient issue exists in contemporary astrophysics and cosmology than that of the elusive `` dark matter '' . for many years already milgrom s paradigm of modified newtonian dynamics ( mond ) has provided an alternative way to interpret observations without appeal to invisible dark matter .
mond had been successful in elucidating economically the dynamics of disk galaxies of all scales , while doing less well for clusters of galaxies ; in its original form it could not address gravitational lensing or cosmology . after reviewing some of the evidence in favor of mond , i recollect the development of relativistic formulations for it to cope with the last deficiency .
i comment on recent work by various groups in confronting t@xmath0v@xmath0s , a relativistic embodiment of mond , with observational data on gravitational lensing and cosmology . throughout
i ask what sort of physics can be responsible for the efficacy of mond , and conclude with an appraisal of what theoretical developments are still needed to reach a full description of the world involving no unobserved matter . |
graphene layers display a number of interesting properties and potential applications owing to the linearly - dispersing bands found near the @xmath0 point in the brillouin zone @xcite .
however , in order to get a complete characterization of graphene other regions in the brillouin zone need to be considered .
in particular , unoccupied states in the vicinity of the @xmath1 point can play a significant role in the transport of currents@xcite .
indeed , it is well known the importance of conduction band minima valleys located around @xmath1 in ballistic electron emission processes , where currents are injected on a substrate under voltage - dependent matching restrictions ( e.g. @xmath2-conservation@xcite ) that are relevant to design field emission transistors@xcite . on the other hand ,
the transport of heat also involves other regions of the brillouin zone apart from the @xmath0 point .
recently it has been shown how the thermal conductivity of few layers of graphene supported on silicon dioxide depend crucially on flexural modes@xcite that are sensitive to the electronic structure around @xmath1@xcite .
therefore , the dielectric response of very few layers of graphene is a key physical element to effectively design devices based on graphene . the dielectric response
is directly related , and therefore can be investigated , by trapping electrons in the region of unoccupied states between the fermi level and the vacuum level .
these states , bound by their self - induced long - range image potential , are called image - potential states @xcite .
the experimental @xcite , and theoretical @xcite study of image - potential states constitutes an ideal probe to better understand the properties of graphene layers .
the image force is a non - local effect asymptotically dominated by correlation effects @xcite . in order to study the infinite rydberg series arising from the image potential
one needs to compute an effective one - dimensional potential , @xmath3 , representing the real part of the quasi - static self - energy for an external probe charge .
this self - induced potential is a continuous function spanning from inside the material , where it represents the exchange and correlation energy , to the vacuum region , where it should have the correct coulomb - like asymptotic behavior @xmath4 .
such a potential can only be obtained from a non - local spatial formalism , since a local approach results in a correlation potential decaying exponentially in the vacuum region , following the density behavior outside the solid @xcite . for a self - consistent first - principles theory such a non - local functional dependence requires costly numerical calculations @xcite .
therefore , it is useful and natural to search for simpler ways to obtain such an effective potential , which is the basic ingredient needed to understand the physics of image - potential states bound by an ultra - thin polarizable layer like graphene .
the simplest of these alternatives is to introduce a set of fitting parameters to continuously join solutions valid either inside or outside the solid .
this point of view has been taken , e. g. by silkin _
et al . _ to study image - potential states in free - standing graphene , joining a function with the correct classical asymptotic behavior outside to a first - principles calculation inside a graphene layer @xcite .
such _ ab - initio _ calculations depends on choosing a model for the exchange and correlation potential ; local density approximation ( lda ) has been employed by silkin _
et al . _
furthermore , it needs setting a very large unit cell to minimize effects between charged periodic images ( i. e. @xmath5 vacuum separator was used to reach convergence , implying a large number of plane - waves and a serious computational effort ) .
finally , such a calculation needs to be supplemented by a few adjustable parameters , e. g. the choosing of a matching point to join the inside potential to the asymptotic classical potential . in this paper
we analyze an alternative that makes use of a minimal free parameter set and makes a simple , flexible and accurate basis for interpreting experimental results .
we use a well - known model for the reflection of electromagnetic fields at the surface ( infinite barrier specular model @xcite ) , coupled with a non - local static dielectric response so the desired self - energy can be obtained @xcite . for the electrodynamics model ,
two free parameters are introduced , i. e. the electronic polarizability of the thin slab , which is determined by the fermi - thomas screening wavelength inside the slab ( electron density of the material ) , and a geometrical dimension given by the layer thickness .
this approach leads in a natural way to a potential with proper physical features : it is continuous and finite over the full spatial domain and it has the right asymptotic behavior towards the vacuum region .
recently , ghaznavi et al.@xcite have studied the non - linear screening of a external charge near a doped graphene layer by solving a fermi - thomas model via a non - linear integral equation .
the non - linear image potential shows changes up to @xmath6 ev with respect to the classical image potential , and supports the use of rpa dielectric response for graphene . in the case of a graphene layer laying on a metallic support another free parameter
may be introduced in this model : the wave - function penetration in the material .
image - potential states are supported in many metallic surfaces for energies between the vacuum level and the fermi level due to the existence of surface band gaps that prevent the penetration of the wave function towards the bulk and allows the existence of bound states @xcite .
therefore , penetration of wave functions is determined by the band structure of particular materials and surface orientations in a way that can not be incorporated in our model , except by including a free parameter that globally determines this penetration . here
we shall first discuss two limiting cases : 0 or 100% penetration of the graphene layer , and then the continuous evolution from one limit to the other . since for image - potential states we are interested in regions in reciprocal space with @xmath7 near @xmath1 and energies between the vacuum level and the fermi energy , it is well justified to model graphene as a polarizable electron gas with a quadratic energy dispersion , as seen from the relevant bands for graphite and graphene@xcite .
a connection between the interlayer states and the image potential states on graphite , graphene and other carbon - based materials has been established by different groups , e.g. silkin et al.@xcite , feng et al.@xcite , and hu et al.@xcite .
these authors have pointed out that such states exist universally on two - dimensional materials .
the static dielectric response , @xmath8 , has been modeled by a random phase approximation ( rpa ) , and by its small @xmath9 expansion , the fermi - thomas approximation ( ft ) @xcite .
while the rpa yields a more accurate description of excitation in the material , introducing the ft allows us to write the potentials as analytic expressions or quasi - analytical ones which merely depend on a final numerical step involving the simple integration of a function decaying quickly for large values of the argument .
a single parameter , the screening constant related to the density of states at the fermi level , @xmath10 , fixes both the scales for energies and lengths facilitating the rescaling of results for different materials and sizes ( atomic units are used throughout the paper , except where explicitly stated otherwise ) .
taking graphite as a model ( 2 g/@xmath11 , 2s@xmath12 2p@xmath12 ) , a typical value for graphene is @xmath13 ( @xmath14 ) , although its precise value should depend on factors like doping , external potentials , etc ; this is accommodated in our results through the simple aforementioned scaling with @xmath15 .
the other parameter used to characterize a thin slab is its width , @xmath16 . for a single atom thick layer of graphene a reasonable value for @xmath17 ,
should be related to the spatial extension of @xmath18 carbon orbitals , @xmath19 a.u .
the value of this parameter turns out not to be critical for this work because wave functions spread over regions much larger than @xmath17 .
the theoretical results are compared to experimental results for graphene on various substrates .
the presence of the substrate leads to a charge transfer from or to the graphene which can be described also as doping .
the resulting work function change due to graphene has been modeled by giovanetti _
et al._@xcite .
we find a good agreement with the experimental data .
the doping changes the available screening charge and leads in turn to a change in the energy of the image - potential states .
the experimental data from two - photon photoemission agree with the calculated dependence .
, of a unit test charge at different positions outside and inside a thin slab ( @xmath20 and @xmath21 ) .
the two thick vertical lines ( black ) mark the slab region .
black continuous line corresponds to ft ( eqs .
( [ eq : ftvi ] ) and ( [ eq : ftvii ] ) ) , while black dots have been obtained using the rpa .
black dashed line : ft approximant to rpa ( @xmath22 ) .
blue dotted line : asymptotic law with image plane at @xmath23 ( @xmath24 ) . dashed line ( green ) : @xmath25 ( repulsive term ) for @xmath26 .
horizontal dashed and dotted lines show the first four eigenvalues ( red and blue for even and odd , respectively ) ; the horizontal thick line ( brown ) gives an approximate value for the work function in graphene ( @xmath27 ev ) . ] for an external probe charge near a slab ( @xmath28 ) we seek the potential acting on @xmath29 by the polarization charges induced in the medium by @xmath29 itself .
this is obtained by computing the total potential , and subtracting the charge s own naked potential . to ensure the proper boundary conditions , and according to the specular reflection model at the surface ,
auxiliary pseudo - media are introduced for the polarizable slab and the vacuum that reduce the calculation to matching solutions obtained in different regions of space for homogeneous media everywhere @xcite .
details of the calculation for the thin slab are given in [ slab ] .
the resulting potential depends on a number of integrals that include the dielectric response of the system : for the rpa these are computed numerically . on the other hand , within the ft approximation an expression
that only depends on a single numerical integration can be obtained , @xmath30}\right ) \left(\chi+\kappa \tanh\left[\chi d \right]\right ) } \label{eq : ftvi}\ ] ] @xmath31 @xmath32 @xmath33\right ) { \mathrm{csch } \,}\left[2 \chi d \right]\right ) } { 2 \kappa \chi \cosh\left[2 \chi d \right]+ \left(2 \kappa^2+k_\mathrm{ft}^2\right ) \sinh \left[2 \chi d \right ] } \bigg \rbrack \bigg \rbrace\ ] ] where @xmath34 .
this is an useful expression that can be computed very efficiently .
we shall see that for the purpose of computing the energy levels of image - potential states it makes an excellent approximation to the more costly rpa calculation . in figure [ fgr : lam28 ]
we show the potential for a slab occupying the region @xmath35 ; both in the ft approximation ( black continuous line ) , and in the rpa one ( black dots ) . in the region
determining the rydberg series ( @xmath36 ) , both approaches yield similar values and agree with the correct asymptotic power - law . near the center of the slab , ft overestimates the interaction over rpa by about 30 - 40% , @xmath37[eq : ch ] a difference that is reflected mainly in the lowest state ( node - less ) that has a significant weight in the central part of the slab where the difference between rpa and ft is larger .
this state is located outside the window between the vacuum level and the fermi energy and is not part of the image - potential states series .
for @xmath38 the @xmath39 integral in equations ( [ eq : ftvi ] ) and ( [ eq : ftvii ] ) can be evaluated analytically to obtain the following useful particular values ( see [ asymptotics ] ) : @xmath40 , and @xmath41 .
we remark that the latter value corresponds to the coulomb hole @xcite .
furthermore , for @xmath42 ( vacuum region ) the potential is well approximated by a classical law , @xmath43 , corrected by an image plane , @xmath44 .
the value of @xmath44 can be obtained by expanding the integrals for @xmath45 , which fixes the position of the image plane in this model : @xmath23 ( dashed blue line in figure [ fgr : lam28 ] , see [ asymptotics ] ) .
finally , equation ( 3 ) suggests a simple procedure to find a good approximant to the rpa potential .
given a physically representative value for @xmath46 , one can find another value , @xmath47 , such that @xmath48 .
for the purpose of obtaining energy eigenvalues of @xmath49 , @xmath50 is a good approximation to the full rpa potential that can be computed quick and easy .
we solve numerically the schrdinger equation @xcite to compute the eigenvalues and eigenfunctions corresponding to the rpa and ft model potential described above .
quite generally , bound states can be unequivocally labeled by the number of nodes @xmath51 , with energies increasing as the number of nodes increases .
furthermore , as long as the potential is symmetric , eigenfunctions have either even or odd parity , for even or odd number of nodes , @xmath51 .
.[tbl : be ] binding energies , @xmath52 ( ev ) , measured with respect to the vacuum level and labeled according to the number of nodes in wave functions .
top three rows : results for rpa and fermi - thomas ( eq . [ eq : ftvii ] ) are given for a free - standing graphene ultra - thin layer and compared with similar results obtained by from an _ ab - initio _ lda calculation matched to an asymptotic expression for the image potential by silkin et al.@xcite ( states are labelled accordingly to their ordering in the image - potential series of unoccupied states lying above the fermi energy and below the vacuum level ) .
the effect of a repulsive barrier located at @xmath53 is presented under @xmath54 ( notice that in this case symmetry is broken and the parity is no longer a good quantum number ) .
for comparison , experimental values measured on gr / ir are quoted@xcite , and compared with the eigenvalues for the classical whittaker s problem@xcite .
finally , we show the effect of approaching the repulsive barrier to @xmath55 , @xmath56 , that can also be computed by introducing an appropriate quantum defect ( @xmath57 ) . to guide the eye , we highlight in bold face numbers that can be compared across different calculations or experiments and
have been given an accompanying interpretation in the text . [ cols=">,>,>,>,>,>,>,>,>",options="header " , ] this formalism predicts , for a free - standing graphene layer , the appearance of new states associated with the even parity ( e. g. , eigenvalues between @xmath58 and @xmath59 ev for @xmath60 , and between @xmath61 and @xmath62 ev for @xmath63 in figure [ fgr : autoe ] ) .
these new states have been obtained numerically and fit well into the classical scaling law proportional to @xmath64 for @xmath51 up to 7 .
obviously , the symmetric potential can be perturbed by external ones ( e. g. the cases of graphene on a support ) , and these states would be affected accordingly .
so far we have discussed a model that effectively represents a free - standing graphene layer . for those cases where
the graphene layer has been deposited on a metallic support the substrate is expected to manifest itself in two main physical ways : ( i ) the wave - functions may be constrained to be outside some spatial region where the substrate enforces an electronic gap , and ( ii ) as a consequence of the interaction between the layer and the support , some charge may be transferred to / from the slab , modifying the density of states at the fermi level , i. e. , the value of @xmath15 . to assess how sensitive the eigenvalues are to the penetration of wave functions into the material we have added to @xmath3 a repulsive term modeled as an exponential wall , @xmath65 .
since we are only interested in creating a decaying state inside the support , we fix the parameter @xmath66 to a large value , @xmath67 a.u . ,
akin to the infinite hard - wall limit ; its effective role is to expel states from the @xmath68 region , ensuring the exponential decay of wave functions inside an electronic band gap .
the resulting potential for the repulsive barrier located near the slab surface ( @xmath69 a.u . , green dashed line in figure [ fgr : lam28 ] ) is similar to the classical series with an image plane , @xmath70 , and can be easily solved by introducing a quantum defect in rydberg formula ( compare energies for the same number of nodes in the eigenfunctions for @xmath56 and whittaker series with quantum defect @xmath71 in table [ tbl : be ] ) .
the barrier , on the other hand , can be introduced below the surface , mimicking the effect of an electronic band gap due to a supporting substrate .
the evolution of the first few eigenvalues with the position of the barrier have been shown in the two limits in table [ tbl : be ] . as long as the barrier is located far away from the ultra - thin slab ( e. g. @xmath72 a.u . )
we get values reminiscent from the original members for the unperturbed symmetrical potential . on the other limit
, a barrier located just on the surface very much reminds of the classical solution . for large @xmath73
the eigenvalues are determined by the potential in the vacuum region and the existence of such a barrier distorts less and less the states as they approach the vacuum level .
table [ tbl : xm ] gives the expectation mean values in , @xmath74 , for whittaker wave functions compared with the ones obtained for the rpa or ft potentials ( e. g. ( [ vvacftscib ] ) ) .
these values compare well with each other , which reflect the manifest similarity between wave functions commented on figure [ fgr : autof ] , and show how the important region for the potential moves quickly away from the surface as @xmath73 grows .
the fact that @xmath75 for image - potential states with @xmath76 implies that wave functions are quite insensitive to the potential inside or near the layer and they are mostly influenced by the asymptotic region where ft and rpa are equivalent .
this suggests that higher @xmath9-corrections to the dielectric function arising from the random phase approximation are not very important , at least for @xmath76 states .
taking away the first level , largely affected by the details near the bottom of the potential , the rest of the series is only modified by a percentage comparable to differences found in table [ tbl : be ] between similar entries .
and 3 image - potential states as a function of work function , @xmath77 ( ev ) .
large open circles ( blue ) show experimental results@xcite .
for ru the data@xcite are plotted by filled circles ( green ) for two different work functions : @xmath78 and @xmath79 ev .
data for graphite have been taken from@xcite . ]
the effect of doping may be explored by looking at the @xmath15 dependence of the energies of the image series .
the important parameter of the dielectric model is the charge density of the graphene layer which can be related to the doping level and the work function .
this is summarized in the current formalism via a single parameter , the fermi - thomas wave - vector @xmath15 .
while @xmath15 can not be easily extracted in our formalism from doping levels of graphene , we can compare to experiments by exploiting the linear dependence predicted by this theory between the screening wave - vector and the work function of the thin slab . to compare with available experimental results it suffices to establish a connection between @xmath15 and a relevant energy scale , e. g. the distance between the vacuum level to the fermi level , i. e. the work function , @xmath80 .
to this end , we note that within our approach the work function for a semi - infinite surface , or a slab with @xmath81 , is @xmath82 in the fermi - thomas model , and @xmath83 in the rpa one .
we can see that rpa , so far our best approach , overestimates @xmath80 when compared with experimental values by about a factor @xmath84 . the linear dependence between @xmath80 and @xmath15 , on the other hand
, is solidly anchored in the theory , having its origin in the net attractive interaction between the external electrons and the polarization charges created inside the polarizable material .
therefore , @xmath80 should be proportional to @xmath85 , albeit the proportionality constant should be corrected to compare with the experimental value .
we take @xmath86 , that corresponds to the empirical work function for graphene , @xmath87 ev . therefore , the empirical dependence @xmath77 allows us to translate @xmath15 into an experimental energy scale , and to set up an energy origin at the value @xmath88 .
figure [ fgr : enwth ] shows the results for the first three members of the series by straight lines .
the experimental data will be discussed in section [ exp ] . computed for the first three image - potential states . ]
an interesting feature of this result is the fact that the higher member of the series are less and less affected by changes in @xmath15 ( @xmath80 and/or @xmath89 ) .
this is clearly seen in figure [ fgr : enwth ] where the slope of @xmath90 decreases for the higher @xmath51 values . in figure
[ fgr : denrhoth ] we plot this result , and show that it is well fitted by the empirical function @xmath91 .
since @xmath15 is linearly proportional to the density of states at the fermi level , @xmath89 , this gives us the empirical scaling law expected for the different terms of the image - potential - state series with doping .
the theoretical results of the preceding section can be tested experimentally by two - photon photoemission ( 2ppe ) .
this technique is able to measure image - potential states with high accuracy and results for various graphene - covered surfaces have been reported @xcite .
photoelectron spectroscopy can also provide precise values for the work function of the surfaces which is correlated to the doping level and charge density of the graphene layer .
we first discuss the work function before turning to the image - potential states .
the amount of charge transfer from the substrate to the graphene layer is determined by the work function difference between substrate and graphene .
this situation has been modeled with a simple capacitor model by giovanetti _
et al._and validated by comparison to results from calculations for various surfaces @xcite .
the curve plotted in figure [ fgr : workfun ] shows the calculated work function of the graphene - covered surfaces versus the work function of the clean substrates for the capacitor model .
the calculated values plotted by green open squares fit the curve quite well @xcite .
the available experimental data are shown by blue open circles @xcite .
the size of the circles represents approximately typical experimental error bars .
the curve was fitted to the experimental results for the noble - metal surfaces @xcite and monolayer graphene on sic @xcite .
the fit was done with the work function of graphene of 4.50 ev compared to 4.48 ev in the original work @xcite .
the chemical shift was reduced from 0.90 ev to 0.89 ev .
overall the fit of this work describes also the calculated values ( green open squares in figure [ fgr : workfun ] ) very well .
the surfaces included in the fit all have a graphene - metal distance around 3.4 @xcite .
the surfaces of ni , ru , and pd were not included in the fit and shown by blue dashed circles .
they have shorter graphene - metal distances @xcite and would require a larger chemical shift for a satisfactory description .
the capacitor model provides a meaningful measure to compare different surfaces with comparable graphene - metal distances via the work function .
in addition , the work function for graphene - covered surfaces can be estimated from the work function of the substrate at least for weakly coupled graphene .
the charges transferred between the substrate and the graphene layer originate from the dirac cone on the graphene side
. these might be electron or holes depending on the doping of the graphene .
the shift of the work function due to the charge transfer can therefore be directly related to the energy of the dirac point relative to the fermi energy .
zero is obtained for a work function of 4.5 ev .
the values are taken from the literature @xcite and are plotted in figure [ fgr : workfun ] as filled red circles .
the agreement is perfect for sic .
but for the other surfaces larger deviations are observed .
one has to keep in mind that for the noble - metal substrates the graphene layer is p - doped and the dirac point can not be observed directly in photoemission spectroscopy .
its energy is extrapolated from the dispersion of the dirac cone below the fermi energy .
the experimental results are in good agreement with the calculations , which show the constant difference between work function and energy of dirac point as predicted by calculations @xcite .
this has been found also for different modifications of graphene on sic @xcite .
the important parameter of the dielectric model is the charge density of the graphene layer which can be related to the doping level and work function .
the energies of the image - potential states as a function of work function have been plotted as red solid lines in figure [ fgr : enwth ] .
large open circles ( blue ) show the available experimental values @xcite .
the symbol size represents typical experimental uncertainties .
the data points for graphite were taken from@xcite which observed the lowest three image - potential states .
other groups reported binding energies of @xmath92 ev for the @xmath93 image - potential state@xcite .
the experimental values are slightly above the calculated lines , but the slope agrees fairly well
. the agreement could be improved by using a smaller value for the parameter @xmath94 relating work function and fermi - thomas wave vector @xmath95 .
the data point for the @xmath93 state of sic lies significantly too high .
this might be an effect of different screening properties of the dielectric substrate or residual binding to the buffer layer .
the accuracy and work function range of the experimental data for the noble - metal surfaces is not sufficient to determine the slope independently .
therefore we only conclude that the experimental data are in reasonable agreement with the predictions of the dielectric model calculations .
finally , we discuss the interpretation of experimental data measured on gr / ru @xcite .
three bound states have been measured with respect to the fermi energy ( all energies in ev ) : @xmath96 ( 1 ) , @xmath97 ( 1 ) and @xmath98 ( 2 ) .
the work function was measured as 4.24 ev . on the corrugated graphene on ru(0001 ) surface
also lower areas exist which have a work function of 4.00 ev . on all other surfaces only the @xmath93 and @xmath99
image - potential states have been found .
it is therefore worthwhile to check whether on ru the additional state might be the @xmath60 image - potential state .
the data are plotted by green solid circles with error bars in figure [ fgr : enwth ] for the two different values of the work function . for a work function of 4.00 ev
the experimental binding energies agree reasonably well with the calculated lines . this assignment would also be compatible with the observed monotonous decrease of the lifetime with binding energy @xcite .
however , we can not rule out the consistent interpretation based on different local work functions on the corrugated graphene on ru as proposed by armbrust _
et al . _
using standard models for the dielectric response and the reflection of electromagnetic waves at a surface we have computed the static self - energy for an ultra - thin slab mimicking a graphene layer .
the self - induced potential goes continuously from the exchange and correlation energy inside the material to the classical asymptotic image potential in the vacuum . for the purpose of obtaining image - potential states
binding energies we find that ft makes an excellent and convenient approximation to the accurate rpa .
eigenvalues and eigenfunctions have been compared with whittaker classical series and recent experiments on gr / ir .
a free standing graphene ultra - thin layer produces a spatially symmetric self - energy that induces an image potential series with even and odd states .
the odd members of the series show a remarkable resemblance to the solution of schrdinger equation for the classical image potential ( whittaker wave functions ) . on the other hand ,
even wave functions arise as new states that differ from whittaker in several key respects , e. g. their non - zero density probability at the origin .
while the qualitative aspects of the image series supported by a thin slab can be described in terms of quite general physical properties , the detailed quantitative values depend crucially on the penetration of wave functions into the substrate and the thin graphene film supported on it .
we have considered the limiting cases of full and none penetration . for
the case of films weakly interacting with a support and wave functions penetrating well inside the system some new states may consequently appear in between the classical ones , that can be traced back to the even states in a free - standing slab . in cases where the interaction is strong and
the surface electronic structure prevents the penetration of wave - functions a behavior more similar to a standard metallic surface is expected .
we notice that the formalism used here can be easily generalized to include the effect of surface plasmons via a frequency - dependent dielectric function @xcite .
the experimental results for graphene on various substrates compare well to theoretical predictions .
the measured work function change due to graphene agree with the capacitor model of giovanetti _
et al._@xcite .
this opens the possibility to predict the work function of graphene - covered surfaces from the substrate work function .
the work function difference from an isolated graphene sheet is related to the doping of the graphene layer and determines in turn the available screening charge .
the resulting change in the energy of the image - potential states calculated with the theoretical model is in agreement with the experimental data from two - photon photoemission .
this work has been financed by the governments of spain ( mat2011 - 26534 , and fis2010 - 19609-c01 - 01 ) , and the basque country ( it-756 - 13 ) .
computing resources provided by the cti - csic are gratefully acknowledged .
we are interested in the quasi - static self - energy potential by an external charge near a slab characterized by a dielectric function , @xmath101 . the dynamical problem , however , is solved exactly in the same way by introducing a k and w - dependent response function @xmath102 .
the problem is solved independently for two homogeneous systems ( pseudo - vacuum and pseudo - medium ) , with a set of fictitious charges , @xmath103 , to reproduce the real fields in the regions for the vacuum and the material respectively @xcite .
notice that the electromagnetic field propagator in vacuum , @xmath104 , corresponds to three spatial dimensions , @xmath105 .
symmetric and antisymmetric solutions are analyzed separately and combined to yield a solution for the most general case .
the extra fictitious charges are finally removed from the solutions by using matching conditions . since a free standing graphene slab is symmetric with respect to the middle plane , matching conditions on only one surface need to be considered . , region ii ) .
the problem is decomposed as a superposition of symmetric ( s , upper part ) and antisymmetric ( a , lower part ) configurations .
the slab polarization is characterized by @xmath8 , and its width , @xmath16 , and it is symmetric around o@xmath106 ( @xmath107 ) .
specular reflection conditions for the fields are imposed at one of the two equivalent surfaces ( o@xmath108 @xmath109 ) . ]
the self - energy of the external probe charge , @xmath29 is obtained by computing the total potential in each region , and subtracting the bare potential due to @xmath29 : @xmath110 where @xmath111 must be separately obtained for q inside / outside the slab .
the case where q is inside the material is schematically shown in figure [ fgr : qgrsa ] . an schematic distribution for the pseudo - charges when the external q is inside the slab is given in figure [ fgr : qant ] for the antisymmetric case .
the symmetric configuration follows easily by substituting the factors @xmath112 in the sums by @xmath113 .
continuity of the perpendicular component of the displacement field results in @xmath114 for the pseudo - vacuum , @xmath115 , and the pseudo - medium , @xmath116 , for both the symmetric and antisymmetric cases . inside the slab .
lower panel : @xmath29 in the vacuum region . ] inside the slab .
lower panel : @xmath29 in the vacuum region . ] therefore , the total potential for the symmetric ( antisymmetric ) cases , are : @xmath117 @xmath118 the surface charges , @xmath119 and @xmath120 , are separately obtained from the condition : @xmath121 .
convergence of the different series in these expressions is guaranteed by the zeros of the dielectric function , i. e. the normal modes dressed by the interaction in the medium @xcite . for the simple thomas - fermi dielectric function
we have , @xmath122 , and the integration over the perpendicular moment , @xmath123 , can be performed to obtain analytic or semi - analytic expressions that are given below .
a similar procedure yields for the case of q in the vacuum region : @xmath125 @xmath126 equation ( [ eq : selfv ] ) yields a numerical procedure that allows to obtain the self - induced potential . using the ft approximation , quasi - explicit expressions that only depend on a single numerical integration , have been obtained and given in ( [ eq : ftvi ] ) .
we remark this is an excellent approximation to the full rpa result for the purpose of computing binding energies of image - potential states .
the semi - infinite system ( @xmath127 ) , along with the use of fermi - thomas dielectric function , brings some simplifications to the expression for the potential that can be exploited to compute exact values at the surface , well inside the material , and in the asymptotic vacuum region . for convenience , in this appendix we move the origin to the surface ( @xmath109 ) . this asymptotic expression , corrected by the position of the image plane obtained above , makes an excellent approximation to the full potential either in the vacuum region outside the slab or in the middle of a vacuum gap , the case discussed below .
it is handy to apply the same techniques to the inverse problem : the potential in a vacuum gap between two semi - infinite media at @xmath145 .
the theoretical procedure proceeds along similar lines we have analyzed in this paper , and we simply give the fermi - thomas result here : ( @xmath151 is the digamma function , and @xmath152 is the euler constant ) , ( ii ) the classical value corrected by an image plane determined by an expansion to the full non - local potential calculated in the fermi - thomas approximation , and ( iii ) the independent non - local potential for each of the two surfaces delimiting the vacuum gap . , of a unit test charge at different positions in a vacuum gap of width @xmath153 a.u . , and inside the surrounding metals ( @xmath20 a.u . ) . continuous line ( black ) : non - local potential for the gap ( fermi - thomas ) . dashed line ( black ) : non - local potential for each surface in the gap considered as independent systems ( fermi - thomas ) . dashed line ( red ) :
classical result taking into account all the classical images and counter - images @xcite . dashed line ( blue ) : semi - classical result adding the contributions of all the classical images , but corrected by an image plane at @xmath154 ( valid in the vacuum region only ) . ] | in the framework of dielectric theory the static non - local self - energy of an electron near an ultra - thin polarizable layer has been calculated and applied to study binding energies of image - states near free - standing graphene .
the corresponding series of eigenvalues and eigenfunctions have been obtained by solving numerically the one - dimensional schrdinger equation .
image - potential - state wave functions accumulate most of their probability outside the slab .
we find that a random phase approximation ( rpa ) for the non - local dielectric function yields a superior description for the potential inside the slab , but a simple fermi - thomas theory can be used to get a reasonable quasi - analytical approximation to the full rpa result that can be computed very economically .
binding energies of the image - potential states follow a pattern close to the rydberg series for a perfect metal with the addition of intermediate states due to the added symmetry of the potential .
the formalism only requires a minimal set of free parameters ; the slab width and the electronic density .
the theoretical calculations are compared to experimental results for work function and image - potential states obtained by two - photon photoemission .
new journal of physics * 16 * ( 2014 ) 023012 ; doi : 10.1088/1367 - 2630/16/2/023012 _ keywords _ : graphene , ultra - thin slab , induced potential , self - energy , non - local dielectric response , random phase approximation , fermi - thomas approximation , image - potential states , binding energies , rydberg series , whittaker series , work function . |
the discovery of the first binary pulsar in 1974 by r. hulse and j.h .
taylor had opened fascinating new opportunities for testing alternative theories of gravitation in the radiative and in the strong - gravitational - field regimes outside the boundaries of the solar system . presently , two distinguished binary pulsars - psr b1913 + 16 and psr b1534 + 12 - serve as primary extra - solar astrophysical laboratories for gravitational physisists .
the radiative @xmath0 test of general relativity ( gr ) in the case of psr b1913 + 16 has been performed with a precision about 0.4% ( taylor & weisberg 1989 , damour & taylor 1991 ) .
in addition , the binary pulsars psr b1534 + 12 strong - field @xmath1 test of gr has been done with the precision 1% ( stairs et al .
1998 ) . what is worth noting
is that the binary pulsar tests of gr have been carried out with an accuracy being comparable to , or in some aspects , higher than the gr tests in the solar system which have been always performed in the near , weak - field zone of the solar system gravitating bodies .
reflecting about the future of gravitational experiments in binary pulsars one can ask about whether the accuracy of gr tests can be further improved and , if not , what physical influences can limit this accuracy .
it is well known there are two main obstacles for improving precision of any experiment - systematic and random errors .
systematic errors cause the bias in the mean values of estimated parameters and can lead to inconsistent conclusions on the validity of the theory of gravitation under investigation .
fortunately , continuous improvements being made in observational techniques and growing knowledge on dynamics of our galaxy can siginificantly reduce or even eliminate the influence of systematic errors ( damour & taylor 1991 ) .
on the other hand , random errors fully determine the magnitude of the numerical value of the dispersion of measured parameters which itself crucially depends on the nature of noise in the timing measurements .
usually , on rather short time spans the white noise of measurement errors dominates .
however , to improve the accuracy of the gr tests one has to observe binary pulsars over time intervals of about 10 years and longer when the white noise is strongly supressed and low - frequency ( or red ) noise dominates in observed times - of - arrivals ( toa ) of pulsar pulses .
the red noise worsens estimates of measured parameters and increases the errors in their mean values .
what kind of problems will one meet in processing such timing observations contaminated with the presence of the red noise ? before trying to answer this question let us briefly descibe the timing model of binary pulsars and the properties of red noise .
the conventional timing model is based on the damour - deruelle analytic parameterization of the two - body problem ( damour & deruelle 1985 , see also klioner & kopeikin 1994 ) .
schematically it reads as follows @xmath2 where @xmath3 is the number of observed pulse , @xmath4 is a constant , @xmath5 and @xmath6 are the pulsar s rotational frequency and its time derivative , @xmath7 is the pulsur s intrinsic noise , @xmath8 is the pulsar s proper time , which is related to the observer s proper time @xmath9 by the equation @xmath10 herein , @xmath11 and @xmath12 describe the classic and relativistic time delay corrections for the binary system and for the solar system respectively , @xmath13 are noises produced by a number of diverse perturbations during the time of propagation of pulses from the pulsar to observer , inaccuracies in ephemerides of the solar system bodies , imperfectness of atomic clocks used for the time metrology , and electronic equipment used for timing observations .
the equipment noise is performed to be white irrespective of the longivity of the observations . on the other hand , any other kind of noise present
has low - frequency components which show up when the measurements are carried out over a sufficiently long time span .
it is worth noting that standard procedures of data processing were worked out only for white noise which has a gaussian distribution of timing residuals .
if timing residuals are dominated by red noise , the standard statistical estimations give unrealistic ( overestimated ) numerical values of the measured parameters .
recently , we have begun to study the problem of pulsar data processing in the presence of red noise ( kopeikin 1997 , kopeikin 1999a , kopeikin & potapov 1998 ) . to model the red noise we have used a shot - noise approximation with a specific choice of step function in such a way that any rational spectrum of the red noise could be restored , including flicker noise of phase ( @xmath14 ) , random walk of phase ( @xmath15 ) , and so on .
the noise model includes both stationary and non - stationary parts of the autocovariance function which are well separated algebraically .
in addition , an exhaustive treatment of the polynomial drift of the noise was worked out in full detail . applying the model for processing fake data of a binary pulsar in a circular orbit we set upper limits on the numerical values of the parameter s variances .
these upper limits depend on the total span of observations @xmath16 and can either decrease or grow as @xmath16 increases . the time dependence some of the limits are shown in table 1 . .
dependence of variances of the pulsar s rotational frequency @xmath5 and some of the keplerian and post - keplerian parameters from the total span of observations @xmath16 in case of presence of red noise with the spectrum @xmath17 , where @xmath18 , @xmath19, ... ,@xmath20 are constants characterizing the intensity of the noise [ cols= " < , < , < , < , < , < , < " , ] one can see that in the case when the spectral index @xmath21 of the noise is big enough variances of some , or even all , parameters grow such that one can not get improvements in testing gr , which prevents better determination of masses of neutron stars and other physical characteristics of the binary system .
special methods of observations and/or data processing should be suggested to overcome this difficulty .
this problem can also be considered from a different point of view .
the fact is that some of the red noises have a specific astrophysical origin and their study would deliver extremely valuable information about physical processes generating such noises .
one particular example represents a low - frequency timing noise produced by the variable gravitational fields of binary stars .
precise calculations of the autocovariance function of the gravitational timing noise from binary stars requires having a mathematically complete solution to the problem of propagation of electromagnetic waves in variable gravitational fields of localized self - gravitating sources .
significant progress in solving this problem has been achieved recently by kopeikin et al .
( 1999 ) and kopeikin & schfer ( 1999 ) .
the relativistic time delay in an arbitrary time - dependent gravitational field has been presented as a function of the relative distances between observer , source of light , and localized source of the non - stationary gravitational field as well as the intrinsic characteristics of the source . using the exprerssion for the time delay
, the gravitational timing noise from an ensemble of binaries in our galaxy has been evaluated for the case of psr b1937 + 21 under some simplifying assumptions ( kopeikin 1999b ) .
using the same approach the gravitational timing noise from an ensemble of binaries in a globular cluster can also be parameterized and predicted .
long - term precise timing monitoring of the bunch of millisecond pulsars in 47 tuc ( or other globular cluster ) will be required to test the presence of the noise and its properties .
it will help to better understand spatial distribution , mass function , and other statistical properties of binaries in the globular cluster .
we thank d. moran for careful reading of the manuscript and useful comments .
kopeikin , s. m. & potapov , v. a. , 1998 , millisecond and binary pulsars as nature s frequency standards .
fourier analysis and spectral sensitivity of timing observations to low - frequency noise , e - print physics/9811014 , to be submitted to mnras kopeikin , s. m. , 1999b , e - print gr - qc 9903070 , to be published in the proceedings of the xxxivth rencontres de moriond meeting on gravitational waves and experimental gravity , les arcs , france , 23 - 30 january 1999 | the influence of the low - frequency timing noise on the precision of measurements of the keplerian and post - keplerian orbital parameters in binary pulsars is studied .
fundamental limits on the accuracy of tests of alternative theories of gravity in the strong - field regime are established . the gravitational low - frequency timing noise formed by an ensemble of binary stars
is briefly discussed . |
the earliest evolutionary phase of star formation , in which a deeply embedded protostar is known to exist , is the so - called class0 stage ( andr et al .
1993 ) . since protostars in this phase ( class0 sources )
are highly obscured by extended dusty envelopes , they can be observed mainly in the far - infrared to millimetre wavelength range .
specifically , a class0 source consists of a central protostellar object surrounded by an infalling dusty envelope and a flattened accretion disk .
many ( presumably all ) class0 sources are associated with strong molecular flows , usually more energetic than those from class1 sources ( davis & eislffel 1995 ) .
since class0 sources will determine most of the early evolutionary history of the protostars as they evolve toward main - sequence stars , it is crucial to determine the physical conditions within the envelope of class0 objects .
however , because of detection difficulty , constraining physical properties of class0 sources is a challenging task .
nevertheless , continuum submillimetre observations of these sources let us to detect dust emission of the massive circumstellar envelopes , and provide a powerful tool for constraining the distribution of matter ( adams 1991 ) .
independent methods of detection of class0 sources , and techniques and useful ways to help to the interpretation of properties of protostars have been developed and used during the last years ( e.g. comparation of submillimetre with near infrared surveys of a star forming region ( stanke et al .
2000 ) , interpretation from the dust emission ( shirley et al .
2000 ) , blackbody fitting ( chini et al . 2001 ) , radiative transfer codes ( wolf et al .
1999 ) , evolutionary models for protostars ( smith 2002 ) ) . in this work
, we combine them and develop an approach in order to constrain the structure ( geometry , radial intensity profile , spatial temperature , and spectral energy distributions ( seds ) ) , and physical properties ( mass , size , bolometric luminosity ( @xmath0 ) and temperature ( @xmath3 ) , and age ) of class0 sources .
specifically , we combine the information from the dust emission with a physical model of the extended envelope , the implementation of techniques like the blackbody fitting and the 1d continuum self - consistent mc3d radiative transfer code ( by wolf et al . 1999 ) , and the smith evolutionary model for protostars ( by smith 2002 ) . to describe and study stellar evolution , the hertzsprung -russell diagram ( hrd ) has become the main way .
however , to study protostellar evolution , the @xmath3@xmath0 diagram ( bltd , bolometric luminosity - temperature diagram ) was proposed as direct analogue to the hrd ( myers & ladd 1993 ; myers et al .
there , class0 sources are characterised by the lowest @xmath3 values of all protostars .
although the bltd provides a useful way to study the evolution of protostars for all evolutionary phases of relevance for star formation ( myers et al .
1998 ) , nevertheless , the smith scheme predicts the evolution of @xmath1 and @xmath0 as an envelope of initial mass @xmath5 and temperature @xmath6 dissipates on a timescale @xmath7 and forms a star .
how can we describe the protostellar emission from the envelope ? in order to facilitate the interpretation of submillimetre observations , we adopt the standard envelope model ( adams 1991 ) to keep the problem as simple as possible .
the circular symmetry of the observed emission ( see rengel 2004 ) , and the lack of significant internal structure justifies the simplicity of a spherical model case .
if the emission is optically thin , the observed intensity for a spherically symmetric protostellar envelope at an impact parameter @xmath8 is given by eq.1 assuming a single power - law index of the opacity @xmath9 .
@xmath10\rho(r)\frac{r}{\sqrt{r^{2}-b^{2}}}dr,\ ] ] @xmath11 is the outer radius , @xmath12 the density , @xmath13 the dust temperature , @xmath14 the opacity of the dust grains , and @xmath15 $ ] the planck function at dust temperature @xmath16 .
the current theory suggests that the density and temperature distributions are power - law indices ( @xmath4 and @xmath17 , respectively , see e.g. adams 1991 ) . if the emission is in the rayleigh - jeans limit and if @xmath11@xmath18 @xmath8 , eq.1 can be approximated to @xmath19 ( where @xmath20 is the power - law index of the observed intensity profile , and @xmath21 is the normalization factor to the peak emission ) .
to investigate the physical structure , processes , and properties of class0 sources , six star forming regions were observed in the perseus and orion molecular cloud complexes at 450 and 850 @xmath2 m with the submillimetre common - user bolometer array ( scuba ) camera at the james clerk maxwell telescope ( jcmt ) on mauna kea , hawaii .
the target regions were , , , , , and .
a distance of 300 pc for l1448 is assumed according to bachiller & cernicharo @xcite , 350 pc for l1455 , ngc1333 and hh211 ( chandler & richer 2000 , and herbig & jones 1983 ) , 460 pc for l1634 ( bohigas et al .
1993 ) , and 390 pc for l1641 ( anthony - twarog 1982 ) . as an example , the resulting combined scuba maps at 450 and 850 @xmath2 m of l1455 overlaid with isophotal contour plots are shown in fig.[fig1 ] .
+ from these continuum scuba maps , 36 submillimetre sources ( @xmath22 3@xmath23 ) are detected ( 26 in perseus and 10 in orion ) .
some of them are extended , and many contain multiple condensations , as well as extended diffuse features .
twelve objects are reported for the first time ( six in perseus and six in orion ) , and fifteen sources are analyzed here .
additionally , the first detailed submillimetre observations of the two regions ngc1333 south and l1641n have been obtained and are discussed in rengel ( 2004 ) , unveiling objects previously unknown .
the survey also reveals a large number of small groups ( rengel et al .
2003 ) .
physical parameters like the dust submillimetre spectral index @xmath24 ( this value is related directly to the dust emissivity exponent @xmath9 by @xmath9=@xmath25 - 2 ) , gas and dust masses , and sizes of the envelopes are derived .
the dust submillimetre spectral index @xmath26 is defined by @xmath27 the mean value of @xmath24 for these objects is 2.8 @xmath28 0.4 , implying that the regions investigated here are quite cold ( @xmath2910k ) .
such low dust temperatures have for example also been found for isolated globules in the perseus region ( see e.g. myers & benson 1983 ) , suggesting an exponent @xmath30 for the dust grain opacity , @xmath14 . the failure of the rayleigh - jeans approximation could increase the value of @xmath9 .
an aspect ratio was defined for each of the sources by dividing the semi - major and minor axis lengths as measured typically at the 97% contour .
we find that the observed aspect ratios of the protostellar emission maps lie in the range 1.0 - 2.0 with an average value of 1.3 @xmath28 0.1 , suggesting that the envelopes of class0 sources can be described approximately as spherically symmetric .
nevertheless , a departure of this symmetry , i.e. , from an aspect ratio of 1 , in some cases is found .
this mismatch could be produced by several factors like magnetic fields of energy densities sufficiently large to influence the object structure and cause flattening along field lines , the bipolar outflow , rapidly rotating structures , or high levels of angular momentum ( rengel 2004 ) .
the average value of the gas and dust mass of the sample , based on the thermal emission from the dust according to hildebrand ( 1983 ) , is 2.5 @xmath28 0.6 m@xmath31 ( typical aperture 45@xmath32 ) . in order to derive physical properties of the sources from the observed azimuthally
averaged radial intensity profiles of the thermal emission from the dust , like the power - law index of the temperature distribution as a function of the radius @xmath6(@xmath33 ) @xmath34 , @xmath17 , @xmath4 , and the envelope sizes , the standard envelope model ( adams 1991 ) was adopted .
assuming that the observed sources can be fitted by a power - law intensity distribution @xmath35 as class0/1 sources , the average value of @xmath36@xmath20@xmath22 is @xmath37 at 850 @xmath2 m and @xmath38 at 450 @xmath2 m . following the analytic estimate for the intensity ( adams 1991 ) , these values of @xmath36@xmath20@xmath22 correspond to the expected ones for very young objects : theoretical models as well as numerical simulations of the collapse of an isothermal sphere ( e.g. , shu 1977 ; larson 1969
; penston 1969 ) also predict power - law indices of the density distribution , @xmath12(@xmath33 ) @xmath39 , with @xmath40 2.0 . considering @xmath20=@xmath4 + @xmath17 -1 , the first two order terms of the equivalent power - law index @xmath2 of the observed intensity ( adams 1991 )
, the values @xmath17= @xmath41 as well as @xmath4= @xmath42 and [email protected] ( at 450 and 850 @xmath2 m , respectively ) are found . in fig.[fig2 ] , the normalized radial profiles [ @xmath43(@xmath8)/@xmath43(0 ) ] ( being @xmath43(0 ) the observed peak flux ) of rno15fir as an example are plotted versus @xmath8 . to estimate the source size , an outer diameter @xmath44 corresponding to the measured half power beam width ( hpbw ) size is adopted .
the observed sources are surrounded by extended envelopes , having typical sizes of 1500 - 6000 au ( at 450 @xmath2 m ) and 4000 - 9000 au ( at 850 @xmath2 m ) .
accurate seds are computed for nine sources combining existing multi - wavelength surveys with the submillimetre data obtained in sect.3 ( rengel 2004 ) , and fitting a modified blackbody curve to the fluxes ( froebrich et al .
this method computes a grid of different blackbody curves of the form ( [ formalsed ] ) in which the temperature , the optical depth at 100 @xmath2 m and @xmath9 are varied ( @xmath45k @xmath46@xmath6@xmath46 @xmath47k ; @xmath48 ; @xmath49 ) .
specifically , curves of the form ( [ formalsed ] ) are convolved with the filter curves of the used filter bands ( eq.[fluxfilter ] ) , and compared to the data - points . the model that best fits the observational data is determined .
we derived the bolometric temperature and luminosity , size of the envelope , and sub - millimetre slope for each source . in local thermodynamic equilibrium ,
the solution of the equation of transfer for an isothermal object that subtends an effective solid angle @xmath50 is given by the emergent flux density @xmath51 : @xmath52= \sigma \omega \cdot \left(1-e^{-\tau}\right ) \cdot b(\lambda , t_{d } ) .
\label{formalsed}\ ] ] @xmath53 ) denotes the planck function for a dust temperature @xmath13 , and @xmath54 the dust optical depth .
the optical depth can be expressed by @xmath55 @xmath56 is in @xmath2 m , the optical depth at 100 @xmath2 m ( @xmath57 ) is a free parameter , and @xmath9 is the submillimetre slope of the sed ( @xmath9=0 corresponds to a blackbody ) .
but we do not observe the emergent flux density @xmath58 directly , we observe @xmath59 , the emergent flux detected across a filter @xmath60 , which has a filter transmission curve @xmath61 .
@xmath59 is calculated by @xmath62 = \frac{\sigma \omega \ , \int\limits_{\lambda = 0}^{\infty } \left(1-e^{-\tau}\right ) \ , b(\lambda , t ) \ , t_f(\lambda ) \ ,
d\lambda}{\int\limits_{\lambda = 0}^{\infty } t_f(\lambda ) \ , d\lambda } \label{fluxfilter}\ ] ] for each filter @xmath60 separately using the filter transmission curves @xmath61 . for details about how the calculations of the maximum likelihood @xmath63 were performed ,
see froebrich et al .
( 2003 ) .
the seds revealed that the sample consists of cool objects ( @xmath1 ranges of @xmath2927 - 50 k ) which have @xmath0 of @xmath294 - 85 l@xmath64 .
thus , by means of the sub - millimetre to bolometric luminosity ratio , , , , , , , , , and are class0 sources . for detailed computations of the associated errors to each determined parameters ,
see rengel ( 2004 ) . in fig.[fig4 ]
( bottom , short dashed lines ) , best obtained fit to the data - points using our approach and overplotted the combined data of are shown as an example . because of the iras 12 and 25 @xmath2 m points are usually far above the fit to the sed ( e.g. chini et al .
2001 ; barsony et al .
1998 ) , and of the unclear quantification of a higher bolometric temperature value caused by filter leaks , outflow / dust interactions or envelope emission if these iras points are included on sed computations ( @xmath294 k froebrich et al .
2003 ) , these points are excluded on the target list and on the blackbody fits performed here . in fig.[fig4 ] , the blackbody curve and the data - points show significant deviations at the infrared astronomy satellite ( iras ) 12 and 25 @xmath2 m points for the class0 sample .
if these data - points had correct flux measurements , the sources would exhibit an excess of mid - infrared emission which could be caused by ongoing outflow / dust interactions .
because it is likely that these iras points are incorrect in some cases , further mid - infrared observations of these objects are necessary in order to investigate the nature of this possible excess .
results from the standard theory provide the first direct insights into observable estimations ( e.g. @xmath4 and @xmath17 , sect.3.1 ) .
nevertheless , the temperature profile will diverge from a single power - law @xmath17 as the envelope becomes optically thick at the primary wavelengths of energy transport ( inner portion of the envelope ) ( e.g. shirley et al .
2000 ) . here
it becomes necessary to calculate the temperature distribution self - consistently by implementation of a radiative transfer code .
detailed physical interpretation of the internal structure of nine objects is realized by intensive computer radiative transfer modelling . in particular ,
dust temperature distributions , seds , and intensity maps are derived self - consistently by the mc3d code ( wolf et al .
the basic model assumed is : a central heating source that is embedded within a spherical symmetric dust envelope of mass @xmath65 .
the central source is characterized by a stellar luminosity ( @xmath66 ) , and an effective temperature ( @xmath67 ) .
the dust envelope is specified by an outer radius ( @xmath68 ) , a density power - law index @xmath4 , an inner radius of the envelope @xmath69 , which is set by the dust destruction radius @xmath70 ( also called sublimation radius ) adopted to a temperature of 1500k , the dust destruction temperature of graphite ( e.g. adams & shu 1985 , which occurs at a temperature of 1500 k ; myers et al .
the central source luminosity @xmath66 is the sole source of luminosity for the envelope , so @xmath66=@xmath0 .
fig.[fig3 ] shows a scheme of the adopted model .
the validity of this model , and physical parameters of the sources ( e.g. envelope masses , density distributions , sizes , and sublimation radius ) are derived by finding the consistency between observed and modelled seds , and radial profiles ( sect4.3 ; rengel et al .
2004 ) .
the dominant opacity source in the radiative transfer models is dust grains .
the characteristics of grains in class0 sources are unknown .
however , graphite and silicate grains have been proposed as likely consistent of galactic dust .
the grain distribution adopted here consists of small interstellar grain size 0.005 - 0.25 @xmath2 m , with mixing of astronomical silicate and graphite ( relative abundances : 62.5 % & 37.5 % ( respectively ) ) , and the draine & lee ( 1984 ) optical constants to describe grain properties .
the adopted model assumes that the dust properties are not a function of radius in the envelope .
the gas - to - dust mass ratio is adopted to be 100:1 , whereby we use a dust grain density of 3.6 g @xmath71 . the wavelength grid ( 60 wavelengths )
are chosen to cover the relevant range . in order to derive the physical structure of the envelope ,
three steps are carried out : the first step constrains the spatial temperature structure of the circumstellar envelope , the second step determines the simulated sed , and the third step , derives the emission , which provides the intensity structure . these last observables lead to test the model . at this point
it is necessary to keep in mind that the values of the inferred parameters are valid only within the framework of the model assumptions considered here . in fig.[fig4 ] , the temperature profile , and best simulated and observed seds for rno15fir are shown as an example .
temperature profiles for rno15fir performed using the mc3d radiative transfer code ( wolf et al .
2003 ) ( top ) .
best simulated and observed seds ( bottom , short dashed and dotted lines , respectively ) .
the filled squares represent the measured 450 and 850 @xmath2 m fluxes from the scuba observations presented in sect.3 .
the iras 12 and 25 @xmath2 m points are excluded from the fit of the sed .
details regarding the summarized target list at several wavelengths from the literature are given in rengel ( 2004 ) .
for the model results , see table1.,title="fig:",scaledwidth=40.0% ] temperature profiles for rno15fir performed using the mc3d radiative transfer code ( wolf et al .
2003 ) ( top ) .
best simulated and observed seds ( bottom , short dashed and dotted lines , respectively ) .
the filled squares represent the measured 450 and 850 @xmath2 m fluxes from the scuba observations presented in sect.3 .
the iras 12 and 25 @xmath2 m points are excluded from the fit of the sed .
details regarding the summarized target list at several wavelengths from the literature are given in rengel ( 2004 ) .
for the model results , see table1.,title="fig:",scaledwidth=40.0% ] the central source should radiate from a specific radius @xmath33 , where the layer changes between the optically thick and the thin regime .
this radius @xmath33 is called photospheric radius " or effective photosphere " , which depends on the wavelength and the molecule species .
crudely , the observed emission arises from a characteristic " layer at @xmath72=1 ( myers et al .
1998 ) or 2/3 ( kenyon et al . 1993
; hartmann 1998 ) .
the temperature distribution changes its character inside the photospheric radius , where the envelope is optically thick ; trapping of radiation causes the temperature gradient to become steeper than in the optically - thin case . to calculate the photospheric radius , @xmath54=1 and the spherically symmetric optically thick envelope of myers et al .
( 1998 ) , with an inner radius @xmath73 , a density @xmath74 and an outer radius @xmath68 , are adopted here for simplicity .
following this model , the photospheric radius @xmath33 is given by @xmath75 @xmath76 being the optical depth from @xmath73 to @xmath68 .
following the myers et al .
( 1998 ) model , a sublimation radius of 3 - 5 au at 450 @xmath2 m , @xmath54=1 and @xmath4=2 ( @xmath2910 au for @xmath4=1.5 ) is estimated , which is roughly the same as the radius of the photosphere . as result of the modelling , it is found that the physical structure of a class0 envelope is characterized by the gas and dust temperature distribution @xmath6(@xmath33 ) @xmath34 , and the density distribution @xmath12(@xmath33 ) @xmath77 @xmath78 , with @xmath17=0.4 and @xmath4 in the range of 1.5 - 2 . in the inner part of the envelope ( 10 au ) ,
@xmath6(@xmath33 ) departs from the single - power law index of 0.4 , resulting in a steeper inner temperature profile .
this phenomenon can be produced probably by thermal convection in the inner envelope and causing zones of dust destruction by evaporation when the envelope becomes optically thick ( e.g. lin & papaloizou 1980 ) .
@xmath6 accounts for the effects of this heating .
we estimate the parameters for the class0 sources by comparing the observed seds obtained by blackbody fitting ( sect.[bb ] ) and the simulated seds from the envelope fitting procedure ( sect.[mc3d ] ) .
following the same technique as kenyon et al .
1993 , the best model for a given observational sed is determined by minimizing the normalized sum of the weighted squared residuals , @xmath79 where @xmath80 is the observed flux at wavelength @xmath81 , @xmath82 the scaled flux of the model , and @xmath83 a weighting factor .
given the uncertainties in estimating observational errors and the limitations of the models , a more complicated fitting treatment is unjustified .
because of the uncertainties in opacity , the difficulties in matching observational beam size with model results , and the possible contribution of disks to the submillimetre region ( c.f .
beckwith et al .
1990 ; butnet et al .
1991 ; keene & masson 1990 ) , we used weights @xmath84=1 for wavelengths 60 @xmath2 m @xmath46 @xmath56 @xmath46 300 @xmath2 m , and chose @xmath84=0.5 for 300 @xmath2 m @xmath36 @xmath56 @xmath36 1000 @xmath2 m .
data - points taken at wavelengths shorter than 40 @xmath2 m are excluded in the comparation calculations because they are expected to be optically thick and therefore very sensitive to asymmetric geometries ( e.g. outflow cavities and flattened envelopes ) , which are currently unable to be modelled . because of the spectral range considered here does not extend much over long wavelengths , data - points taken at wavelengths longer than 1000 @xmath2 m are rejected .
as it is claimed previously , a comparison between best simulated and observed seds , and between simulated and observed radial profiles led us to check the reliability of the model . in order to create the simulated radial profile , the envelope emission at 850 @xmath2 m is calculated as an image from the mc3d code for each source , and convolved by the scuba beam .
fig.[fig5 ] shows the observed and simulated radial profile of as an example .
we conclude from our comparison that the assumption of spherical symmetry in the envelope is a reasonable approximation to study the envelope of class0 sources .
nevertheless , deviations from this geometry between observed and modeled seds at short wavelengths , and between observed and simulated radial profiles of relatively spherical class0 sources become extremely prominent ( e.g. for rno15fir and ngc1333iras1 ) ( see rengel 2004 ) .
in addition , small differences between observed and simulated radial profiles are observed for hh211-mm and rno 15fir ( aspect ratios of 1.2 and 1.3 at 850 @xmath2 m , respectively ) .
this discrepancy could be due to the exclusion of a non - spherical geometry , of cavities produced by molecular outflows , or of a flattened disk in the model ; it could also be due to specific assumptions considered here , e.g.a single point - like nature of the sources ( e.g. it is uncertain that hh211-mm and rno15fir are point - like sources ( eislffel et al .
2003 ; davis et al .
1997 , respectively ) ) , and a constant dust opacity with radius in the model .
the observed properties of class0 sources can be understood more thoroughly if it is assumed that the objects evolve in time . using the smith protostellar evolutionary scheme (
smith 2002 ) , and constructing the @xmath0@xmath3 diagram for the sample , the age , the infall rate , and the envelope mass as a function of time are estimated for the detected class0 sources ( rengel 2004 ) . in particular , comparing the positions of the objects with an evolutionary model age and mass of the envelope of each source are derived ( fig.[fig6 ] , filled triangles ) .
the tracks in fig.[fig6 ] are derived by combining the unification scheme , as reviewed by smith ( 2000 , 2002 ) , with the framework for protostellar envelopes presented by myers et al .
( 1998 ) . for the sample ,
the resulting model - dependent ages of class0 sources are in a range from 1 to 3 @xmath85 @xmath86 yr . the evolution of these sources is traced by @xmath3 , which typically increases systematically between 25 - 50 k. the resulting bolometric luminosities span a range between 5 - 85 l@xmath31 . following the smith scheme ( with an accretion rate not constant in time , smith 2002 ) ,
the accretion rate @xmath87 at the earliest class0 stage increases exponentially for a short interval of time ( with an accretion peak in 1.5@xmath8510@xmath88 yr ) , and then declines later on ( rengel 2004 ) .
is there a correlation of the power - law index @xmath4 with time ?
we plot the estimated values of @xmath4 as a function of @xmath7 .
fig.[fig7 ] suggests a density structure of @xmath12(@xmath33 ) @xmath89 at younger ages , evolving to @xmath12(@xmath33 ) @xmath90 at later times . in order to investigate this with a larger spread ,
a class1 sample is mandatory .
table1 lists the characteristics of the best - fitting model parameters for the sample of sources analyzed in detail here .
the main results of this study are : 1 ) from the continuum scuba maps , 36 submillimetre sources ( @xmath22 3@xmath23 ) are detected . some of them are extended , and many contain multiple condensations , as well as extended diffuse features .
twelve of these objects are reported here for the first time , 15 sources are analyzed , and nine class0 sources are studied in detail .
2 ) an average value of the aspect ratio of the sample of 1.3 @xmath28 0.1 is found , suggesting that the envelopes of class0 sources can be described approximately as spherically symmetric .
nevertheless , a departure of this symmetry in some cases is found .
this mismatch could be produced by several factors like magnetic fields of energy densities sufficiently large to influence the object structure , the bipolar outflow , rapidly rotating structures , or high levels of angular momentum .
3 ) the sample consists of cool objects ( @xmath3 ranges of @xmath2927 - 50 k ) which have @xmath0 of @xmath294 - 85 l@xmath64 .
4 ) the physical structure of a class0 envelope is characterized by the gas and dust temperature distribution @xmath6(@xmath33 ) @xmath34 , and the density distribution @xmath12(@xmath33 ) @xmath77 @xmath78 , with @xmath17=0.4 and @xmath4 in the range of 1.5 - 2 ( @xmath4=2 at younger ages and @xmath4=1.5 at later times ) . in the inner part of the envelope ( 10 au ) , @xmath6(@xmath33 )
departs from the single - power law index of 0.4 , a phenomenon caused probably by thermal convection in the inner envelope .
5 ) from the thermal emission of the dust and following an emissivity model ( hildebrand 1983 ) , the average value of the gas and dust mass of a sample based on 15 submillimetre selected objects is 2.5 @xmath28 0.6 m@xmath31 .
examination of the radial profiles of the sample shows that the objects are surrounded by extended envelopes . typically , the sizes are @xmath291500 - 6000 au ( at 450 @xmath2 m ) and @xmath294000 - 9000 au ( at 850 @xmath2 m ) .
6 ) a sublimation radius of 3 - 5 au at 450 @xmath2 m , @xmath54=1 and @xmath4=2 ( @xmath2910 au for @xmath4=1.5 ) is estimated , which is roughly the same as the radius of the photosphere ( because the gas that is accreted close to the central object is optically thick , the central source should radiate from this specific radius , where the layer changes between the optically thick and the thin regime ) .
7 ) the blackbody curve and the data - points show significant deviations at the infrared astronomy satellite ( iras ) 12 @xmath2 m and 25 @xmath2 m points for the class0 sample .
if these data - points had correct flux measurements , the sources would exhibit an excess of mid - infrared emission which could be caused by ongoing outflow / dust interactions . because it is likely that these iras points are incorrect in some cases , further mid - infrared observations of these objects are necessary to investigate the nature of this possible excess .
8) the assumption of spherical symmetry in the envelope model is reasonable .
nevertheless , deviations from this geometry determined by comparations between observed and modeled seds at short wavelengths , and between observed and simulated radial profiles of relatively spherical class0 sources become extremely prominent .
this discrepancy could be due to the exclusion of a non - spherical geometry , of cavities produced by molecular outflows , or of a flattened disk in the model ; it could also be due to specific assumptions considered here , e.g.a single point - like nature of the sources , and a constant dust opacity with radius in the model .
9 ) for the sample of class0 sources , the model - dependent ages are in a range of 1 - 3@xmath8510@xmath91 yr .
detailed physical interpretation of the internal structure of nine observed class0 sources was carried out by intensive computer modelling . a simple spherically symmetric model envelope , and assumptions about density and dust distributions following the standard envelope model reproduced reasonably well the observed seds and the radial profiles of the sources .
@xmath6(@xmath33 ) @xmath92 is a good approximation for the sample .
the radial temperature distribution as function of distance , however , departed significantly from the optically - thin assumption , an observational derivation of a single - power law @xmath17 of 0.4 for radii @xmath36 10 au .
modelling results indicated a density profile well described by a power - law between @xmath4=1.5 and 2 , which is expected by lcccccccc + object & @xmath3 [ k ] & @xmath93 [ l@xmath94 & @xmath70 [ au ] & @xmath4 & @xmath68 [ au ] & age [ @xmath85 10@xmath95 yr ] & @xmath65[m@xmath31 ] & @xmath97 + + & 27 & 6 & 5 & 2.0 & 4000 & 13.8 & 2.8 & 6.4 + & 40 & 11 & 5 & 1.6 & 3000 & 19.5 & 1.7 & 2.6 + & 45 & 8 & 3 & 1.6 & 3500 & 20.7 & 1.0 & 1.8 + & 52 & 13 & 3 & 1.5 & 4500 & 21.9 & 1.5 & 8.6 + & 48 & 68 & 3 & 1.5 & 7000 & 26.1 & 3.0 & 0.1 + & 30 & 5 & 3 & 1.9 & 6000 & 13.8 & 4.1 & 4.5 + & 42 & 19 & 3 & 1.6 & 6000 & 21.9 & 2.8 & 1.1 + & 45 & 85 & 3 & 1.5 & 8500 & 25.2 & 7.0 & 1.0 + l1641smsiii & 50 & 80 & 3 & 1.5 & 10000 & 27.6 & 6.3 & 0.6 + + sed and radial profile fits constrained physical parameters of the sources such as envelope masses , density distributions , sizes and sublimation radii .
nevertheless , mid - infrared observations , and the inclusion of outflow and disk , and other geometries in the model are desirable to investigate in order to decrease the differences between the model and the data , or prove that data and model are different under several configurations . | we report on a study of the thermal dust emission of the circumstellar envelopes of a sample of class0 sources .
the physical structure ( geometry , radial intensity profile , spatial temperature and spectral energy distribution ) and properties ( mass , size , bolometric luminosity ( @xmath0 ) and temperature ( @xmath1 ) , and age ) of class0 sources are derived here in an evolutionary context .
this is done by combining scuba imaging at 450 and 850 @xmath2 m of the thermal dust emission of envelopes of class0 sources in the perseus and orion molecular cloud complexes with a model of the envelope , with the implementation of techniques like the blackbody fitting and radiative transfer calculations of dusty envelopes , and with the smith evolutionary model for protostars .
the modelling results obtained here confirm the validity of a simple spherical symmetric model envelope , and the assumptions about density and dust distributions following the standard envelope model .
the spherically model reproduces reasonably well the observed seds and the radial profiles of the sources .
the implications of the derived properties for protostellar evolution are illustrated by analysis of the @xmath0 , the @xmath3 , and the power - law index @xmath4 of the density distribution for a sample of class0 sources . |
we are developing a model which aims to reproduce self - consistently observational data of many kinds related to cosmic - ray origin and propagation : direct measurements of nuclei , electrons and positrons , gamma rays , and synchrotron radiation .
these data provide many independent constraints on any model and our approach is able to take advantage of this since it must be consistent with all types of observation .
we emphasize also the use of realistic astrophysical input ( e.g. , for the gas distribution ) as well as theoretical developments ( e.g. , reacceleration ) .
the code is sufficiently general that new physical effects can be introduced as required .
the basic procedure is first to obtain a set of propagation parameters which reproduce the cosmic ray b / c ratio , and the spectrum of secondary positrons ; the same propagation conditions are then applied to primary electrons .
gamma - ray and synchrotron emission are then evaluated .
models both with and without reacceleration are considered .
the models are three dimensional with cylindrical symmetry in the galaxy , the basic coordinates being @xmath0 where @xmath1 is galactocentric radius , @xmath2 is the distance from the galactic plane and @xmath3 is the total particle momentum .
the numerical solution of the transport equation is based on a crank - nicholson implicit second - order scheme . in the models the propagation region
is bounded by @xmath4 beyond which free escape is assumed .
a value @xmath5 kpc has been adopted since this is within the range which is consistent with studies of @xmath6be / be and synchrotron radiation . for a given @xmath7 the diffusion coefficient as a function of momentum is determined by b / c for the case of no reacceleration ; with reacceleration on the other hand it is the reacceleration strength ( related to the alfvn speed @xmath8 ) which is determined by b / c .
reacceleration provides a natural mechanism to reproduce the b / c ratio without an ad - hoc form for the diffusion coefficient @xcite . the spatial diffusion coefficient for the case _ without _ reacceleration is @xmath9 below rigidity @xmath10 , @xmath11 above rigidity @xmath10 , where @xmath12 is the particle speed .
the spatial diffusion coefficient _ with _ reacceleration assumes a kolmogorov spectrum of weak mhd turbulence so @xmath13 with @xmath14 for all rigidities . for this case
the momentum - space diffusion coefficient is related to the spatial one @xcite .
the injection spectrum of nucleons is assumed to be a power law in momentum .
the interstellar hydrogen distribution uses hi and co surveys and information on the ionized component ; the helium fraction of the gas is taken as 0.11 by number .
the interstellar radiation field for inverse compton losses is based on stellar population models and iras and cobe data , plus the cosmic microwave background .
energy losses for electrons by ionization , coulomb , bremsstrahlung , inverse compton and synchrotron are included , and for nucleons by ionization and coulomb interactions .
the distribution of cosmic - ray sources is chosen to reproduce the cosmic - ray distribution determined by analysis of egret gamma - ray data @xcite .
the bremsstrahlung and inverse compton gamma rays are computed self - consistently from the gas and radiation fields used for the propagation .
the @xmath15-decay gamma rays are calculated explicitly from the proton and helium spectra using @xcite .
the secondary nucleon and secondary @xmath16 source functions are computed from the propagated primary distribution and the gas distribution , and the anisotropic distributions of @xmath16 in the @xmath17 system was taken into account @xcite .
some results obtained are shown in the figures .
the energy dependence of the b / c ratio , and local _ proton and helium spectra _ are shown in figs .
[ fig1 ] and [ fig2 ] .
the spectrum of _ primary electrons _ is shown in fig .
the adopted electron injection spectrum has a power law index 2.1 up to 10 gev ; this is chosen using the constraints from synchrotron and from gamma rays .
the electron spectrum is consistent with the direct measurements around 10 gev where solar modulation is small and it also satisfies the constraints from @xmath18 . above 10 gev a break is required in the injection spectrum to at least 2.4 for agreement with direct measurements ( which may however not be necessary if local sources dominate the directly measured high - energy electron spectrum ) .
the synchrotron spectrum at high galactic latitudes ( fig . [ fig3 ] ) is important since its shape constrains the shape of the 110 gev electron spectrum .
an injection index 2.1 ( without reacceleration ) is the steepest which is allowed by the radio data over the range 38 to 1420 mhz .
as illustrated , an index 2.4 as often used ( e.g. , @xcite ) gives a synchrotron spectrum which is too steep .
the modelled _ gamma - ray _
spectrum for the inner galaxy , shown here for the case of no reacceleration ( fig .
[ fig4 ] ) , fits well the comptel @xcite and egret @xcite data between 1 mev and 1 gev beyond which there is the well - known excess not accounted for by @xmath15-decay with the standard nucleon spectrum @xcite .
the electron spectrum used here is increased by a factor 2 over that shown in fig .
[ fig6 ] ; this is required to reproduce well the observed gamma intensities . it is still within the range allowed by the positron fraction ( see below ) , and the shape accords well with the flatter electron injection spectrum required by synchrotron data ( fig .
[ fig3 ] ) .
inverse compton is dominant below 10 mev , bremsstrahlung becomes important for 3200 mev .
the lower bremsstrahlung combined with @xmath15-decay leads to a good fit to the flat spectrum observed in this range , in contrast to previous attempts to model the spectrum with a steeper bremsstrahlung spectrum @xcite .
the 1030 mev @xmath19-ray longitude profile ( fig .
[ fig5 ] ) at low latitudes from this model can be compared with that from comptel @xcite .
the @xmath20 _ spectrum _ ( fig .
[ fig6 ] ) and the _ positron fraction _
[ fig7 ] ) agree well with the most recent data compilation for 0.110 gev @xcite .
strong a.w . , youssefi g. , _ proc .
24th icrc _ * 3 * , 48 ( 1995 ) strong a.w . ,
moskalenko i.v .
, schnfelderv . , _
proc.25thicrc_,*4 * , 257 ( 1997 ) heinbach u. , simon m. , _ apj _ * 441 * , 209 ( 1995 ) seo e.s . ,
ptuskin v.s . ,
_ apj _ * 431 * , 705 ( 1994 ) strong a.w . , mattox j.r . , _
a&a _ * 308 * , l21 ( 1996 ) dermer c.d . ,
_ a&a _ * 157 * , 223 ( 1986 ) moskalenko i.v .
, strong a.w .
, _ apj _ , * 493 * , in press ( 1998 ) strong a.w .
et al . , _ a&as _ * 120 * , c381 ( 1996 ) strong a.w . , et al . , in _ aip proc . _ , 1997 , this symposium hunter s.d . , et al . , _ apj _ * 481 * , 205 ( 1997 ) mori m. , _ apj _ * 478 * , 225 ( 1997 ) webber w.r . , et al . , _ apj _ * 457 * , 435 ( 1996 ) seo e.s . , et al . , _ apj _ * 378 * , 763 ( 1991 ) engelmann j.j . , et al . ,
_ a&a _ * 233 * , 96 ( 1990 ) protheroe r.j .
, _ apj _ * 254 * , 391 ( 1982 ) barwick s.w . , et al . , _ apjl _ * 482 * , l191 ( 1997 ) | an extensive program for the calculation of galactic cosmic - ray propagation has been developed .
this is a continuation of the work described in @xcite . the main motivation for developing this code @xcite is the prediction of diffuse galactic gamma rays for comparison with data from the cgro instruments egret , comptel , and osse .
the basic spatial propagation mechanisms are ( momentum - dependent ) diffusion , convection , while in momentum space energy loss and diffusive reacceleration are treated .
primary and secondary nucleons , primary and secondary electrons , and secondary positrons are included .
fragmentation and energy losses are computed using realistic distributions for the interstellar gas and radiation fields . |
modelling particle acceleration at snrs is one of the main goal to explain the origin of galactic cosmic rays ( crs ) in the context of the so - called _ supernova paradigm _ ( see e.g. * ? ? ?
this paradigm requires that snrs are able to accelerate nuclei up to energies as high as a few times @xmath3 gev , converting a fraction @xmath4 of the snr kinetic energy into crs .
there is an increasing amount of evidence that shocks in young snrs can indeed reach the required efficiency .
unfortunately , all these evidence are indirect and when considered individually can be also explained by other mechanisms that do not require efficient acceleration .
a clear example of such an ambiguous situation is represented by @xmath2-ray emission . in the last few years
several snrs have been detected both in the gev and in the tev band but , in spite of this increasing amount of data , the question whether this emission is due to hadronic ( through the decay of neutral pions produced in nuclear interactions between accelerated nuclei and the background plasma ) or leptonic processes ( due to inverse compton and/or bremsstrahlung ) is still debated ( see e.g. * ? ? ?
* for a general discussion on this topic ) .
a possible way to discriminate between the _ leptonic _
( i.e. inefficient ) and the _ hadronic _ ( i.e. efficient ) scenario is studing the effects that particle acceleration produces at all observable wavelengths , rather that focus the attention only on the @xmath2-ray emission , simultaneously using multiple set of data to constrain the model . here
we apply the non - linear diffusive shock acceleration ( nldsa ) theory in order to explain the non - thermal emission and morphological properties of two similar snrs , namely tycho and kepler .
tycho , in particular , has been recently detected in @xmath2-rays by fermi - lat @xcite and veritas @xcite , and can be considered one of the most promising object where to test the shock acceleration theory and hence the cr
snr connection .
we presented a detailed model applyed to tycho in @xcite .
kepler , on the other hand , has not been detected in @xmath2-ray band , yet , but it is very similar to tycho in many respects , and we show that the predicted @xmath2-ray spectrum should be detectable by cta .
both tycho and kepler have been established to be remnants of a type ia sn , tycho thanks to observation of the scattered - light echo @xcite while kepler based on the o / fe ratio observed in the x - ray spectrum @xcite .
they have similar ages and similar values for the radio spectral index , i.e. 0.65 and 0.64 , respectively .
interestingly also the ratio of power emitted at 10 kev and at 1 ghz , i.e. @xmath5 , is very similar , being 56.9 for tycho and 52.9 for kepler .
finally , both remnants present very thin filaments in non - thermal x - ray emission . in order to make the model as simple as possible , in both cases we assume that the remnants expand into an uniform circumstellar medium ( csm ) with proton number density @xmath6 ( which we leave as a free parameter ) and temperature @xmath7 .
we model the remnant evolution by following the analytic prescriptions given by @xcite for type ia sne , assuming a sn explosion energy @xmath8 erg and one solar mass in the ejecta , whose structure function is taken as @xmath9 .
the radial structure of density and temperature profiles is then calculated by assuming that the shocked csm is roughly in pressure equilibrium . on top of this snr evolution , the spectrum of accelerated particles is calculated according to the semi - analytic kinetic formalism put forward in @xcite and references therein , which solves self - consistently the equations for conservation of mass , momentum and energy along with the diffusion - convection equation describing the transport of non - thermal particles for quasi - parallel , non - relativistic shocks .
the injection is regulated by a free parameter , @xmath10 , in such a way that all particles from the thermal plasma with momentum @xmath11 , with @xmath12 the typical thermal momentum downstream , start the acceleration process . a crucial role in our model
is played by the magnetic field amplification induced by the super - alfvnic streaming of relativistic particles upstream of the shock .
we model this magnetic amplification as in @xcite .
the large magnetic fields predicted by the resonant amplification have two main consequences : 1 ) the magnetic pressure upstream becomes comparable to , or even larger than , the thermal plasma pressure and , reducing the compressibility of the plasma , affects the shock compression factor .
2 ) when the magnetic field is amplified the velocity of the scattering centers , which is generally neglected with respect to the shock speed , may be significantly enhanced @xcite .
when this occurs , the total compression factor felt by accelerated particles may be appreciably reduced and , in turn , the spectra of accelerated particles may be considerably softer .
we explicitly include these effects assuming that the scattering centers moves with a speed equal to the alfvn speed in the amplified magnetic field .
once the magnetic field structure is known , we can compute the spectrum of accelerated electrons at the shock , which is assumed to be proportional to the proton spectrum @xmath13 , up to a maximum momentum @xmath14 determined by the synchrotron losses in the amplified magnetic field , where the spectrum presents a squared exponential cut - off @xmath15}$ ] .
the evolution of the electron and proton spectrum downstream of the shock is computed taking into account adiabatic losses for protons and adiabatic and synchrotron losses for electrons .
* results . *
we consider the following radiative processes : 1 ) synchrotron emission of relativistic electrons ; 2 ) thermal and non - thermal electron bremsstrahlung ; 3 ) inverse compton scattering ( ics ) of electrons on cmb radiation , local ir dust emission and galactic ir+optical light ; 4 ) emission due to the decay of @xmath16 produced in hadronic collisions . because the explosion energy and the mass of the ejecta are fixed _ a priori _ , in order to fit the observed spectra and the remnant angular size we can only vary three parameters : the number density of the upstream medium , @xmath17 , the injection efficiency of protons , @xmath10 , and the electron to proton normalization , @xmath18 . in fig .
[ fig : nuf_nu ] we show our best fit of the photon spectrum produced by the superposition of all the radiative processes outlined above , comparing it with the existing data .
the overall agreement is quite good . quite interestingly , for both tycho and kepler our best - fitting returns @xmath19 and @xmath20 which implies that the total amount of kinetic energy channeled into crs is @xmath21 .
the inferred values for the circumstellar medium implies that tycho and kepler have a distance of 3.3 and 7.4 kpc from the sun .
synchrotron radiation fits well the observed x - ray emission assuming an electron to proton ratio @xmath22 for tycho and @xmath23 for kepler .
the projected x - ray emission profile is shown in fig.[fig : xray ] , where it is compared with available chandra data . in the case of tycho
the predicted profile shows a very good agreement with the data while for kepler the model overpredicts the emission in the inner part of the rim .
this could be due to a deviation of the shock geometry from the perfectly spherical symmetry assumed in our model .
the sharp decrease of the emission behind the fs is due to the rapid synchrotron losses of the electrons in a magnetic field as large as @xmath1 g . in the @xmath2-ray band
the decay of @xmath16 produced in hadronic collision is the dominant process . in particular , we predict a slope for accelerated protons @xmath24 . in the case of tycho
this slope well accounts for fermi - lat and veritas data within the experimental errors .
the predicted proton spectrum shows a cut - off around @xmath25 tev / c .
ics of relativistic electrons can not explain the observed @xmath2-ray emission for two different reasons .
first , the strong magnetic field produced by the cr - induced streaming instability forces the number density of relativistic electrons to be too small to explain the @xmath2-ray emission as due to ics on the ambient photons .
second , even if we arbitrarily reduced the magnetic field strength , enhancing at the same time the electron number density in order to fit the tev @xmath2-rays with ics emission , we could not account for the gev @xmath2-rays because both the spectral slope and the flux would be incompatible with the fermi - lat data .
also non - thermal electron bremsstrahlung has to be ruled out , because it provides a flux two order of magnitudes lower than the fermi - lat detection , and can not be arbitrarily enhanced without over - predicting both the tev and the x - ray emission .
kepler , due to its larger distance , it is fainter than tycho and therefore not detected in the @xmath2-ray band by the current generation of telescopes .
however , the predicted tev flux should be observable by the cerenkov telescope array .
decay ( thick dashed ) and ics ( dotted ) as calculated within our model .
the total emission is showed by the solid curve .
the experimental data for tycho are , respectively : radio from @xcite ; x - rays from suzaku ( courtesy of toru tamagawa ) , gev @xmath2-rays from fermi - lat @xcite and tev @xmath2-rays from veritas @xcite . for kepler :
radio from @xcite and x - rays from @xcite.,title="fig : " ] decay ( thick dashed ) and ics ( dotted ) as calculated within our model .
the total emission is showed by the solid curve .
the experimental data for tycho are , respectively : radio from @xcite ; x - rays from suzaku ( courtesy of toru tamagawa ) , gev @xmath2-rays from fermi - lat @xcite and tev @xmath2-rays from veritas @xcite . for kepler :
radio from @xcite and x - rays from @xcite.,title="fig : " ] | we apply the non - linear diffusive shock acceleration theory in order to describe the properties of two supernova remnants , sn 1572 ( tycho ) and sn 1604 ( kepler ) . by analyzing the multi - wavelength spectra
, we infer that both tycho s and kepler s forward shocks ( fs ) are accelerating protons up to @xmath0 tev , channeling into cosmic rays more than 10 per cent of their kinetic energy .
we find that the streaming instability induced by cosmic rays is consistent with the x - ray morphology of the remnants , indicating a very efficient magnetic field amplification ( up to @xmath1 g ) . in the case of tycho
we explain the @xmath2-ray spectrum from the gev up to the tev band as due to pion decay produced in nuclear collisions by accelerated nuclei scattering against the background gas . on the other hand , due to the larger distance
, the @xmath2-ray emission from kepler is not detected , being below the sensitivity of the present detectors , but it should be detectable by cta .
address = inaf - osservatorio astrofisico di arcetri , l.go e. fermi 5 , i - 50125 firenze , italia address = department of astrophysical sciences , princeton university , princeton , nj 08544 , usa |
in 1940 kramers derived chemical reaction rates from certain limits in a fokker - planck equation that describes the probability density of a brownian particle in an energy landscape @xcite .
the limit of high activation energy has been revisited in a recent paper by peletier et al @xcite , where a spatially inhomogeneous extension of kramers formula is rigorously derived for unimolecular reactions between two chemical states @xmath0 and @xmath1 .
their derivation relies on passing to the limit in the @xmath2-gradient flow structure of the fokker - planck equation .
it is well - known by now that the fokker - planck equation has also an interpretation as a wasserstein gradient flow @xcite and the question was raised in @xcite whether kramers formula can also be derived and interpreted within this wasserstein gradient flow structure .
this concept has also been investigated on a formal level for more complicated reaction - diffusion systems in @xcite . a further motivation for studying the fokker - planck equation within the wasserstein framework comes from applications that additionally prescribe the time evolution of a moment @xcite .
in this note we present a rigorous derivation which is based on passing to the limit within the wasserstein gradient flow structure . to keep things simple
we restrict ourselves to the spatially homogeneous case and consider the simplest case of a unimolecular reaction between two chemical states @xmath0 and @xmath1 , which are represented as two wells of an enthalpy function @xmath3 . to avoid unimportant technicalities we assume that the enthalpy function @xmath4 is a ` typical ' double
- well potential .
specifically , we assume that @xmath4 is a smooth , nonnegative and even function that satisfies @xmath5 with @xmath6 the probability density of a molecule with chemical state @xmath7 is in the following denoted by @xmath8 . in kramers approach
the molecule performs a brownian motion in the energy landscape described by @xmath4 , so the evolution of @xmath8 is governed by the kramers - smoluchowski equation @xmath9 where @xmath10 is the so - called viscosity coefficient . in what follows we consider the high activation energy limit @xmath11 .
the leading order dynamics of can be derived by formal asymptotics and governs the evolution of @xmath12 which satisfy @xmath13 due to @xmath14 .
using wkb methods , for instance , one finds @xmath15 we emphasize that the constant @xmath16 depends on the details of the function @xmath4 near its two local minima and its local maximum , and that the time scale is exponentially slow in the height barrier @xmath17 between the two wells .
our goal is to derive rigorously by passing to the limit @xmath18 in . in order to derive a non - trivial limit
we have to rescale time accordingly .
thus we consider in the following the probability distribution @xmath19 that is a solution of @xmath20 an important role in the analysis will be played by the unique invariant measure @xmath21 which converges in the weak@xmath22 topology of probability measures to @xmath23 , where @xmath24 denotes the delta distributions in @xmath25 , as in @xcite it is often convenient to switch to the density of @xmath26 with respect to @xmath27 , that is @xmath28 .
heuristically we expect that @xmath29 is to leading order in @xmath10 piecewise constant for @xmath30 and @xmath31 , where the respective values correspond to @xmath32 and @xmath33 as introduced above . in what follows
we write @xmath34 instead of @xmath32 , so @xmath33 is given by @xmath35 . for the derivation of the limit equation
we assume that our data are well - prepared .
[ t.1 ] let @xmath36 and @xmath16 be as in , and for each @xmath10 let @xmath37 be a solution to with initial datum @xmath38 . moreover ,
suppose that the initial data @xmath38 are probability measures on @xmath39 that converge weakly@xmath22 as @xmath40 to some probability measure @xmath41 , and satisfy @xmath42 with constants @xmath43 and @xmath44 independent of @xmath10 .
then , for all @xmath45 we have that @xmath46 weakly@xmath22 in the space of probability measures , where the function @xmath47 satisfies @xmath48 where @xmath49 .
this result has already been derived in @xcite in the more general setting with spatial diffusion and under slightly weaker assumptions on the initial data .
our main contribution here is therefore not the result as such , but the method of proof .
we answer the question posed in @xcite , how the passage to the limit can be performed within a wasserstein gradient flow structure and we identify the corresponding structure for the limit .
we present the formal gradient flow structures of as well as in section [ sec : structure ] . in order to derive the limit equation we pass to the limit in the rayleigh principle that is associated to any gradient flow .
this strategy is inspired by the notion of @xmath50-convergence and has already been successfully employed in other singular limits of gradient flows ( e.g. in @xcite ) . in section [ s.compactness ]
we first obtain some basic a priori estimates as well as an approximation for @xmath29 that will be essential in the identification of the limit gradient flow structure .
it is somewhat unsatisfactory that we can not derive suitable estimates solely from the energy estimates associated to the wasserstein gradient flow .
instead we use the estimates that correspond to the @xmath2-gradient flow structure that is satisfied by @xmath29 .
it is not obvious to us how this can be avoided .
section [ s.proofs ] finally contains our main result , that is the novel proof of theorem [ t.1 ] .
we show that the limit of @xmath51 satisfies the rayleigh principle that one obtains as a limit of the rayleigh principle associated to the wasserstein gradient structure of . as a consequence
the limit is a solution of .
we briefly summarize the wasserstein gradient structure of the fokker - planck equation as well as the corresponding gradient flow structure of the limit problem . to point out the key ideas we give a formal exposition and
postpone some technical details to section [ s.compactness.1 ] . given an energy functional @xmath52 on a manifold @xmath53 ,
whose tangent space @xmath54 is endowed with a metric tensor @xmath55 , the @xmath56-gradient flow @xmath57 of @xmath52 is defined such that @xmath58 for all @xmath59 and for all @xmath60 . here
@xmath61 denotes the directional derivative of @xmath52 in direction @xmath62 .
our convergence result relies on the rayleigh principle , that amounts to the observation that a curve @xmath57 on @xmath63 solves if and only if for each @xmath64 the derivative @xmath65 minimizes @xmath66 among all @xmath59 .
slightly more general and more suitable for generalizations to abstract manifolds in function spaces is the time integrated version : for each @xmath67 the function @xmath68 minimizes @xmath69 among all functions @xmath62 with @xmath70 for all @xmath71 $ ] . in what follows
we pass to the limit @xmath40 in the time integrated rayleigh principle because only this one is compatible with the boundedness and compactness results derived below .
to describe the * wasserstein gradient structure * of the @xmath10problem we consider the formal manifold @xmath72 along with the metric tensor @xmath73 the energy of the fokker - planck equation is given by @xmath74 and has the directional derivative @xmath75 consequently , the direction of steepest descent @xmath76 is characterized by the requirement that @xmath77 this means @xmath78 and we conclude that is in fact the formal @xmath79-gradient flow of @xmath80 on @xmath81 .
the * gradient structure of the limit problem * is very simple .
the corresponding manifold @xmath82 is one - dimensional and equipped with the metric tensor @xmath83 notice that the metric tensor is continuous in @xmath34 with @xmath84 , and that @xmath85 is defined in .
the limit energy is given by @xmath86 and we easily check that is the @xmath56-gradient flow of @xmath52 on @xmath53 .
the density of @xmath26 with respect to @xmath27 , that is @xmath87 , is smooth and satisfies the equation @xmath88 and hence we readily justify that @xmath89 moreover , due to the assumption from theorem [ t.1 ] the maximum principle for implies that @xmath90the a priori estimates and are direct consequences of the @xmath91-gradient and the @xmath2-gradient flow structures of , see @xcite for details .
the wasserstein structure , however , implies the a priori estimate @xmath92 and conserves the mass via @xmath93 . as mentioned before
, our analysis does not make use of this estimate but employs and .
we now derive a rigorous setting for the time integrated rayleigh principle that corresponds to the wasserstein gradient structure of the fokker - planck equation . to this end
we suppose that @xmath67 is fixed and consider the weighted lebesgue space @xmath94\times{{\mathbb{r}}}\to{{\mathbb{r}}}\;:\ ; \int_0^t\int_{{\mathbb{r}}}\frac{{f_{{\varepsilon}}}^2}{{\gamma_{{{\varepsilon}}}}}\,dxdt<0\big\},\end{aligned}\ ] ] which is a hilbert space for each @xmath95 .
we also define the linear space @xmath96 and show that the rayleigh principle is a well - posed minimization problem on @xmath97 .
[ lem : tools ] for each @xmath98 with @xmath99 we have @xmath100 and @xmath101 as @xmath102 strongly in @xmath103\big)}}$ ] .
in particular , we have @xmath104 for almost all @xmath105\times{{\mathbb{r}}}$ ] .
clearly , holds for all smooth @xmath106 with compact support in @xmath107\times{{\mathbb{r}}}$ ] . by approximation in @xmath108 and since we have @xmath109 due to , , , and @xmath110 we then conclude that holds for all @xmath98 . moreover , using hlder s inequality and @xmath111 we find @xmath112 for all @xmath113 . therefore , and by assumption on @xmath114 , we know that @xmath115 converges strongly as @xmath102 to some limit functions in @xmath103\big)}}$ ] . from @xmath116 we then infer that these limit functions vanish .
[ cor : tools ] we have @xmath117 .
moreover , each @xmath118 satisfies @xmath119 for almost all @xmath64 , as well as @xmath120 with @xmath121 and @xmath122 as in and .
the first claim follows from @xmath123 and since the a priori estimates and guarantee that @xmath124 as well as @xmath125 .
now let @xmath126 be arbitrary . by definition and
, we have @xmath127 moreover , lemma [ lem : tools ] provides @xmath119 as well as @xmath128 which is . from corollary [ cor : tools ]
we finally conclude that @xmath129 is well defined for @xmath130 , and that @xmath131 is the unique minimizer .
we now exploit the a priori estimates and to derive suitable compactness results for @xmath26 , which then allow to extract convergent subsequences . to this end
we choose @xmath132 independent of @xmath10 and define the intervals @xmath133 as well as @xmath134 we proceed with summarizing some properties of @xmath27 .
[ l.0]we have @xmath135as well as @xmath136and@xmath137 from the definition and @xmath138 we readily derive .
thanks to our assumptions on @xmath4 we also have @xmath139 and this implies @xmath140 .
similarly , we find @xmath141 and hence @xmath142 and @xmath142 .
finally , a direct computation gives @xmath143 which is @xmath140 thanks to .
our compactness result for @xmath26 is illustrated in figure [ fig ] and reads as follows .
[ lem : compactness ] there exists a subsequence @xmath40 along with a function @xmath144}\right)}}}$ ] with weak derivative @xmath145 such that 1 . for each @xmath146 $ ]
, @xmath147 is a probability measure on @xmath39 that converges weakly@xmath22 to @xmath148 such that @xmath149 uniformly in @xmath146 $ ] , 2 .
@xmath131 is a measure on @xmath107\times{{\mathbb{r}}}$ ] that converges weakly@xmath22 to @xmath150 such that @xmath151 weakly and strongly in @xmath103\big)}}$ ] , respectively .
moreover , we have @xmath152 for some @xmath153 and all @xmath146 $ ] . the estimates and combined with imply @xmath154}{{\left({\int_{\bar{j}_{{\varepsilon}}}{\rho_{{{\varepsilon}}}}\,dx}\right)}}^2 & \leq\sup_{t\in[0,t]}{{\left({\int_{{{\mathbb{r}}}}{\gamma_{{{\varepsilon}}}}{u_{{{\varepsilon}}}}^2\,dx}\right ) } } { { \left({\int_{\bar{j}_{{\varepsilon}}}{\gamma_{{{\varepsilon}}}}\,dx}\right ) } } \quad\xrightarrow{{{\varepsilon}}\to0}\quad0\ , , \\\int_0^t{{\left({\int_{\bar{j}_{{\varepsilon}}}{\left|{\partial_t{\rho_{{{\varepsilon}}}}}\right|}dx}\right)}}^2\,dt & \leq\int_0^t{{\left({\int_{\bar{j}_{{\varepsilon } } } { \gamma_{{{\varepsilon}}}}{\left|{\partial_t{u_{{{\varepsilon}}}}}\right|}^2\,dx } \right ) } } { { \left({\int_{\bar{j}_{{\varepsilon}}}{\gamma_{{{\varepsilon}}}}\,dx}\right)}}\,dt \quad\xrightarrow{{{\varepsilon}}\to0}\quad0\,,\end{aligned}\ ] ] and hence we find @xmath142 and @xmath142 .
we now consider the functions @xmath155\to{{\mathbb{r}}}$ ] defined by @xmath156 and using , , and again we obtain @xmath157}\right)}}}&\leq\int_0^t{{\left({\int_{j^\pm_{{\varepsilon}}}{\gamma_{{{\varepsilon}}}}{u_{{{\varepsilon}}}}^2\ , dx}\right ) } } { { \left({\int_{j^\pm_{{\varepsilon}}}{\gamma_{{{\varepsilon}}}}\,dx}\right)}}\,dt\leq{c}\ , , \\\tfrac{1}{4}{{\|{\dot{u}^\pm_{{\varepsilon}}}\|}}^2_{l^2{{\left({[0,t]}\right)}}}&\leq\int_0^t{{\left({\int_{j^\pm_{{\varepsilon } } } { \gamma_{{{\varepsilon}}}}{\left|{\partial_t{u_{{{\varepsilon}}}}}\right|}^2\,dx } \right ) } } { { \left({\int_{j^\pm_{{\varepsilon}}}{\gamma_{{{\varepsilon}}}}\,dx}\right)}}\,dt\leq{c}\,.\end{aligned}\ ] ] by weak compactness we can therefore extract a subsequence such that @xmath158\big)}}\subset\subset{c{{\big([0,t]\big)}}}$},\quad\end{aligned}\ ] ] for some limit functions @xmath159 . setting @xmath160 ,
the remaining assertions hold either by construction , or thanks to and @xmath161 . from now on
we suppose that a subsequence as in lemma [ lem : compactness ] is chosen and prove the assertions of theorem [ t.1 ] for this subsequence .
since the function @xmath34 is uniquely determined by the limit problem , we then conclude afterwards that theorem [ t.1 ] holds for all subsequences . in order to pass to the limit @xmath162 in the minimization problem corresponding
to we show that the relative density @xmath163 is close to a step function but exhibits a narrow and smooth transition layer at @xmath164 whose shape is determined by @xmath27 , see figure [ fig ] .
specifically , we prove that @xmath163 can be approximated by @xmath165 notice that @xmath166 is an approximation of the sign function on the interval @xmath167 because lemma [ l.0 ] provides that @xmath168 [ l.1a ] we have @xmath154,x\in{i^\pm_{{\varepsilon } } } } { \big|{{u_{{{\varepsilon}}}}{{\left({t},\,{x}\right)}}-\tilde{u}_{{\varepsilon}}{{\left({t},\,{x}\right)}}}\big| } \;\xrightarrow{{{\varepsilon}}\to0}\;0\,,\qquad \int_0^{t } \sup _ { x \in j^0_{{{\varepsilon } } } } \big | { u_{{{\varepsilon}}}}(t , x ) - \tilde{u}_{{\varepsilon}}{{\left({t},\,{x}\right ) } } \big|^2 \,dt \;\xrightarrow{{{\varepsilon}}\to0}\;0\,.\end{aligned}\ ] ] from and @xmath142 we infer that @xmath154}\int_{i_{{\varepsilon}}}{\left|{\partial_x{u_{{{\varepsilon}}}}}\right|}^2\,dx \leq \sup_{t\in[0,t]}{{\left({\int_{i_{{\varepsilon}}}\frac{{\gamma_{{{\varepsilon}}}}{\left|{\partial_x{u_{{{\varepsilon}}}}}\right|}^2}{{\tau_{{{\varepsilon}}}}}\,dx}\right ) } } { { \left({\int_{i_{{\varepsilon}}}\frac{{\tau_{{{\varepsilon}}}}}{{\gamma_{{{\varepsilon}}}}}\,dx}\right ) } } \xrightarrow{{{\varepsilon}}\to0}0\,,\end{aligned}\ ] ] and hence @xmath169}\sup_{x , y\in{i^\pm_\eps}}\babs{\ueps\pair{t}{x}- \ueps\pair{t}{y}}\xrightarrow{\eps\to0}0.$ ] with we therefore conclude that @xmath170}\sup\limits_{x\in{i^\pm_{{\varepsilon}}}}{\left|{{u_{{{\varepsilon}}}}{{\left({t},\,{x}\right)}}-2\int_{j^\pm_{{\varepsilon } } } { \rho_{{{\varepsilon}}}}{{\left({t},\,{y}\right)}}\,dy}\right| } \quad\xrightarrow{{{\varepsilon}}\to0}\quad0\,,\end{aligned}\ ] ] so lemma [ lem : compactness ] gives @xmath170}\sup\limits_{x\in{i^\pm_{{\varepsilon}}}}{\big|{1\pm{{\left({u-1}\right)}}-{u_{{{\varepsilon}}}}{{\left({t},\,{x}\right)}}}\big| } \quad\xrightarrow{{{\varepsilon}}\to0}\quad0\,.\end{aligned}\ ] ] the first assertion now follows from . towards the second claim
we integrate twice with respect to @xmath7 .
this gives @xmath171 with @xmath172 where the two constants of integration can be computed by @xmath173 for @xmath174 the integral term in can be estimated by @xmath175 so and lemma [ l.0 ] imply @xmath176 in particular , we have @xmath177 in @xmath103\big)}}$ ] . from this , , and the definition of both @xmath178 and @xmath179 we finally conclude that @xmath180 and the proof is complete .
it follows from lemma [ l.0 ] that for @xmath40 the functions @xmath181 generate a delta distribution in @xmath164 with height @xmath182 .
the following result combined with lemma [ l.1a ] shows that @xmath183 has a similar property . for the proof we recall that the function @xmath184 is continous in @xmath185 and
uniformly positive for @xmath186 .
[ l.2 ] we have @xmath187 uniformly in @xmath146 $ ] . by definition
we have @xmath188 this implies @xmath189 so the desired result follows from and .
we are now able to prove our main result , which then implies theorem [ t.1 ] .
the limit of @xmath131 , that is @xmath145 , satisfies @xmath190 for all @xmath191\big)}}$ ] .
this theorem is implied by the following three lemmas , which ensure that the limit @xmath192 of the minimizers @xmath51 of the rayleigh principle associated to the @xmath10-problems is indeed a minimizer of the rayleigh principle associated to the limit problem .
notice that the assertions of lemmas [ l.3]-[l.5 ] are closely related , but not equivalent to the @xmath50-convergence of the rayleigh principles .
in fact , in lemma [ l.3 ] we prove lower semi - continuity of the metric tensor only for the minimizers @xmath51 .
this is , however , sufficient to conclude that @xmath192 is again a minimizer of the limit problem .
( lim - inf estimate for metric tensor ) [ l.3 ] we have @xmath193 recall that we have @xmath194 where @xmath195 satisfies the a priori estimate @xmath196$}\,.\end{aligned}\ ] ] consequently , lemma [ l.1a ] and ensure that @xmath197\big)}}$}\,,\end{aligned}\ ] ] and therefore it is sufficient to show that @xmath198 setting @xmath199 and using @xmath200 we find @xmath201 and hence @xmath202 integration with respect to @xmath64 , and employing as well as lemma [ l.0 ] , then gives @xmath203\big)}}$}\,.\end{aligned}\ ] ] applying jensen s inequality , the convergence , and lemma [ l.2 ] we estimate @xmath204 the desired result now follows from lemma [ l.2 ] and since lemma [ lem : tools ] , combined with corollary [ cor : tools ] and , implies that @xmath205 weakly in @xmath103\big)}}$ ] .
( lim - inf estimate for energy ) [ l.4 ] we have @xmath206 by , which implies that @xmath110 , and using , , and , we find @xmath207 moreover , the convergence results from lemma [ lem : compactness ] and lemma [ l.1a ] provide @xmath208 which is the desired result .
( existence of recovery sequence ) [ l.5 ] for all @xmath209\big)}}$ ] there exists a sequence @xmath210 such that @xmath211 for each @xmath95 we define @xmath212 , where @xmath213 is a standard dirac sequence with support in @xmath214 .
we also introduce functions @xmath215 by @xmath216 . by construction
, @xmath215 vanishes for @xmath217 , is equal to @xmath218 for @xmath219 , and satisfies @xmath220 thanks to and .
moreover , in view of lemma [ l.1a ] , lemma [ l.2 ] and we also have @xmath221 and hence @xmath222 finally , @xmath223 follows as in the proof of lemma [ l.4 ] .
the authors gratefully to alexander mielke for stimulating discussions and to the referee for his valuable comments .
this work was supported by the epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) . | we derive kramers formula as singular limit of the fokker - planck equation with double - well potential . the convergence proof is based on the rayleigh principle of the underlying wasserstein gradient structure and complements a recent result by peletier , savar and veneroni . + kramers formula , fokker - planck equation , wasserstein gradient flow , + rayleigh principle * 35q84 , 49s05 , 80a30 . * |
since the recent discovery of the strongly correlated qcd fluid , the phenomenology of quark - gluon plasma ( qgp ) has attracted a lot of attention @xcite .
many predictions have been made for the several possible phases of the qgp in which quarks and gluons are no longer confined to nucleons and pions @xcite . at extremely high temperature and low density , thermal excitations of the gluon field
are expected to screen interactions between color charges and ultimately , due to asymptotic freedom result in a weakly interacting quark gas @xcite . similarly , at low temperature but high density debye screening
is expected to reduce the range of strong interactions .
novel phases that correlate quark color and flavor are predicted to occur at asymptotic densities due to the attractive nature of quark - quark interactions in certain color - flavor locked combinations @xcite .
thus finite temperature and/or density are expected to reflect on various aspects of confinement .
furthermore any modification of the ground state influences symmetry properties and in particular restoration of chiral symmetry is expected . at finite density and low temperature
the precise relation between chiral symmetry restoration and deconfinement is not yet known .
a common wisdom is that in this regime there is a phase transition from hadronic ( confined ) matter at low density to the unconfined ( possibly superconducting ) phase at higher densities @xcite .
recently , however it has been observed that the confined and deconfined phases may be separated by a phase where quarks are confined but chiral symmetry is restored . in the limit of an infinite number of colors @xmath0
this so called quarkyonic phase would in fact extend to infinite density , since in this limit debye screening due to quark loop vanishes @xcite . in this paper
we examine the possible emergence of this new phase using a canonical formulation of the qcd many body problem in the coulomb gauge .
the coulomb gauge canonical formulation can describe both finite temperature and density . in this formulation manifestation of deconfinement can be inferred from the temperature and/or density dependence of the color coulomb interaction .
in contrast lattice simulations at finite density are still at their infancy @xcite .
one of the smoking gun signals of deconfinement is the possible dissociation of heavy quarkonia @xcite . by studying density dependence of the coulomb gauge heavy quark potential
we will be able to explore charmonium properties at finite density .
the paper is organized at follows . in the following section we discuss the coulomb gauge qcd and approximations relevant to the problem in hand .
zero and finite density properties are discussed in sections [ q0 ] and [ qf ] , respectively .
charmonium dissociation is studied in sec .
[ jp ] and followed by conclusions and outlook .
in this section we briefly discuss qcd in the coulomb gauge and the approximations appropriate for the high density and/or temperature systems @xcite . in the coulomb gauge gluons
are described by the transverse potentials , @xmath1 , @xmath2 and the conjugated , transverse momenta , @xmath3 , @xmath4 = i\bm{\delta}_t(\x - \y)\delta^{ab},\ ] ] where @xmath5 \delta^3(\x-\y)$ ] .
the canonical momentum @xmath3 is the negative of the transverse component of the chromo - electric field .
the quark and antiquark degrees of freedom will be defined below in terms of the canonical set of dirac fields , @xmath6 , @xmath7 , one for each flavor .
the hamiltonian is given by @xmath8 where @xmath9 contains the dirac kinetic energy and quark - transverse gluon interaction , @xmath10 is the yang - mills term , which contains the gluon kinetic energy and the three- and four - gluon interactions and finally @xmath11 is the coulomb potential given by , @xmath12 { \cal j } \rho^b(\y).\ ] ] it represents the non - abelian coulomb gauge interaction between color charge densities , @xmath13 , mediated by the coulomb kernel @xmath14 $ ] given by @xmath15 = \left [ { g \over { \bm{\nabla } \cdot { \cal d } } } ( -\bm{\nabla}^2 ) { g \over { \bm{\nabla } \cdot { \cal d } } } \right ] _ { ( \x , a;\y , b)}. \label{keff}\ ] ] finally @xmath16 $ ] is the determinant of the faddeev - popov operator ; @xmath17 is the covariant derivative in the adjoint representation , and @xmath18 is the chromo - magnetic field , @xmath19 .
at every space point-@xmath20 and color component @xmath21 the coulomb gauge potentials , @xmath22 are an analog of a curvilinear coordinate .
this is because their values are restricted to reside within the boundary of the gribov region which has a nontrivial metric determined by @xmath23 @xcite .
a confinement scenario in the coulomb gauge states that it is the field configurations near the boundary of the gribov region @xmath24 that dominate the qcd vacuum @xcite and it follows that fluctuations near the boundary lead to massive quasiparticle excitations . since the exact parametrization of the gribov horizon is not known the quantitative description of this confinement scenario varies depending on how restriction to the gribov horizon is implemented , but the general features seem to be robust @xcite . with this picture in mind
we approximate the ground state of a finite density quark plasma by the state with no quasiparticle gluon excitations and the large background fields concentrated near @xmath24 lead to enhancement in the long range behavior of the coulomb kernel @xcite .
that is we make the replacement @xmath15 \to \langle k(\x , a;\y , b)[\a ] \rangle = k(\x-\y)\delta_{ab}\ ] ] with the potential @xmath25 modified from its free ( @xmath26 ) form @xmath27 , for large @xmath28 due to the large fields @xmath29 .
in particular we approximate the kernel by the from @xmath30 with @xmath31 and @xmath32 being the short - range coulomb and long - range linear potentials , respectively which will be discussed in detail in the following section .
the final hamiltonian , describing massless quarks with energies below gluon quasiparticle excitations is given by @xmath33 note , that from eq .
( [ heff ] ) it follows that it is @xmath34 which is the instantaneous interaction in the color - singlet @xmath35 channel .
in presence of the effective density - density interaction mediated by the kernel @xmath36 , in eq .
( [ heff ] ) quarks and antiquarks acquire effective mass , which in the mean filed approximation can be described within the hartree - fock - bogolubov framework @xcite .
the single quark quasiparticle operators are defined by a canonical transformation to a plane wave representation of the dirac fields .
@xmath37 \nonumber \\\end{aligned}\ ] ] and similarly for @xmath38 .
here @xmath39 , @xmath40 is the quark ( antiquark ) spin projection and @xmath41 are the quark ( antiquark ) annihilation and creation operators , respectively .
these quasiparticle operators satisfy the standard fermion anti - commutation relations and define the vacuum state , by @xmath42 .
the single particle wave functions are given by @xmath43 where @xmath44 , @xmath45 and @xmath46 is the bcs angle , which determines the number density of quark - antiquark pairs in the bcs vacuum , _ i.e. _ the quark condensate and the extent of chiral symmetry breaking @xcite .
at zero temperature and density the bcs angle is determined by minimizing the vacuum energy density , @xmath47 , which leads to the gap equation ( @xmath48 ) @xmath49.\ ] ] here @xmath50 is the fourier transform of the effective potential from eq .
( [ keff ] ) .
the fourier transform of the linear potential , @xmath32 has to be taken with care , since naively , @xmath51 .
we introduce an infrared regulator , @xmath52 and define @xcite @xmath53 so that @xmath54 . the difference between the liner potential and the ir finite approximation is shown in fig .
[ kl ] . in momentum space
the ir finite kernel becomes @xmath55 it is clear that the @xmath56 term does not contribute the gap equation ( [ gap ] ) and the gap equation is well defined in the limit @xmath57 .
equivalently , the gap equation is invariant under a constant shift in the potential , @xmath58 since such a shift induces a contribution to the hamiltonian proportional to the square of the total charge operator @xmath59 , @xmath60 invariance under ( [ shift ] ) is an exclusive property of color singlet states and it is only matrix element invariant under global color rotations that are are physical .
in contrast , under this shift , the energy @xmath61 of a single quark state ( @xmath62 ) that is given by , @xmath63 \label{omega}\ ] ] transforms to @xmath64 so single quark states are clearly unphysical .
for @xmath65 the @xmath56-term in @xmath66 dominates the integrand in eq .
( [ omega ] ) and the quark self - energy becomes negative and tends to @xmath67 in the @xmath65 limit .
as pointed out in @xcite this is necessary in order for color - singlet @xmath35 excitations to have finite , non - negative energies .
this is because the potential energy in the @xmath35 bound state , given by @xmath68 , where @xmath69 is the relative momentum between the quark and the antiquark is large and positive for small @xmath52 ( and approached @xmath70 in the limit @xmath65 ) .
thus the infinities in the @xmath65 limit cancel between the self - energies and the residual interaction between the quark and the antiquark .
since @xmath71 is a lower energy state compared to @xmath72 , after cancelation of the ir divergencies , the finite energy of color single excitations is @xmath73-independent and non - negative .
this is , however not the case for color - nonsinglet states .
for example a single quark state with energy given by eq . ( [ omega ] ) has negative energy which becomes @xmath67 in the limit when the interaction is confining ( _ i.e. _ @xmath65 ) .
this is clearly unphysical as one would expect colored states to have positive , and ir diverging energies , @xmath74 in the confining limit .
since the shift in eq .
( [ shift ] ) is a symmetry of the physical sector we can choose @xmath73 to cancel the ir divergence in the quark self energy , and redefine @xmath32 accordingly thus instead of eq .
( [ k1 ] ) we should use @xmath75 which is positive , gives @xmath74 as @xmath65 and does not affect properties of color singlet states .
in fact and interaction without the ir divergent constant term constant term is obtained when computing the expectation value of @xmath76 $ ] ( _ cf . _ eq .
( [ keff ] ) ) in a mean filed ansatz for the gluon vacuum distribution @xcite . a similar argument for regulating the @xmath77 singularity of @xmath78
was proposed in @xcite .
there by explicitly restricting the spectrum of the hamiltonian to include only the color - singlet subspace an even stronger constraint on the ir momentum dependence of the kernel was derived , namely @xmath79 . with an ir regulated ( finite-@xmath52 ) kernel the minimal physical requirement
, however is that colored excitations have positive energies , ( becoming @xmath70 as @xmath65 ) .
thus we do consider kernels @xmath50 which have integrable , but in principle finite zero - modes . .
the proper potential for describing color - nonsinglet state , ( _ cf . _ eq .
( [ k0 ] ) ) corresponds to a downward - shift by two units .
[ kl],width=2 ]
with the interactions in the hamiltonian now well defined in the ir , the system at finite quark density can be described using standard many body techniques @xcite .
in particular the finite density gap equation becomes , @xmath80\ ] ] and the single quark energy is given by , @xmath81.\ ] ] here @xmath82 and @xmath83 is the quark ( antiquark ) occupation number at zero - temperature , @xmath84 with @xmath85 denoting the the fermi momentum . because of the pauli blocking factor @xmath86 interactions between quarks have small effect on quark levels inside the fermi sphere where quarks are effectively free @xmath87 .
the self - energy contributes mainly for states above the fermi surface and results in states which are similar to the confined quark states at zero - density . since any modification of the antiquark distribution from its vacuum value
is suppressed at finite quark density , it will be ignored in the following and we will only consider quark - hole excitations near the quark fermi surface . at finite density
the quark - quark potential @xmath88 is screened by particle - hole excitations near the fermi surface , resulting in the effective interaction @xmath89 which enters eqs .
( [ gapf]),([selfenergyf ] ) , @xmath90 or @xmath91 the vacuum polarization , @xmath92 is shown in fig .
[ vac : bubble ] and it describes the probability for creating a particle - hole excitation . since the pair is excited in the colored state in the confining limit ( @xmath65 ) it is expected that @xmath93 , since the amplitude is proportional the the inverse of the particle - hole excitation energy , which becomes infinite in the confining limit .
assuming phase transition does take place , however we can examine the effect of debye screening in the deconfined phase near or above the phase transition density . in this case vacuum polarization
is given by @xmath94 \nonumber \end{aligned}\ ] ] where @xmath95 is the number of light flavors .
the set of equations , ( [ gapf ] ) , ( [ selfenergyf ] ) , ( [ veff ] ) and ( [ vacpol ] ) with @xmath50 given by @xmath96 forms a set of coupled equations which we solve numerically and discuss below . . [ vac : bubble],width=1 ] we solved coupled equations ( [ gapf ] ) , ( [ selfenergyf ] ) , ( [ veff ] ) and ( [ vacpol ] ) at finite density and zero temperature with two light quarks flavors using @xmath97 with @xmath98 and @xmath99 determined from fitting the zero - density @xmath100 potential @xcite . at zero density
what makes quark energies finite is the ir regulator @xmath52 in eq .
( [ kfinal ] ) , at finite density , above the deconfinement phase transition we expect , however , that the self - consistent set of equations will admit nontrivial solutions in the limit @xmath65 , with the fermi momentum @xmath85 taking over the role of the ir regulator instead .
we have verified numerically that this indeed is the case .
the equations are solved by iterations . for given density , @xmath85
we start with a small , but finite @xmath52 , _ e.g. _
@xmath101 , so the self - energy at the first iteration has a sharp jump at the fermi surface ( see left panel in fig . [
fig : omega ] ) . at this initial state
the vacuum polarization is highly suppressed ( see left panel in fig . [
fig : pi ] ) , particularly for small momenta .
after a few iterations , however , the self - energy becomes regulated by the vacuum polarization itself and simultaneously the vacuum polarization increases at small momenta .
finally , we reduce the initial value of the ir regulator , @xmath52 and we verify that after several iterations solutions converge to the same value , regardless of the starting value of the regulator .
we repeat the calculations for several values of the fermi energy .
we show these finial results for the effective potential , bcs gap angle , quark single energy and vacuum polarization in figs .
[ fig : v ] , [ fig : gap ] , [ fig : omega ] , [ fig : pi ] , respectively .
it is often stated that chiral symmetry restoration and deconfinement occur simultaneously @xcite .
our calculation illustrates that this need not be the case . at finite density , effective potential is already deconfined , but the gap equation admits nontrivial solutions .
only when the fermi momentum increases above , approximately @xmath102 to @xmath103 , the effective potential at large distance is not strong enough to sustain spontaneous chiral symmetry breaking . ) .
[ fig : v],width=2 ] at finite density . for @xmath104
the solution of the gap equation is @xmath105 .
[ fig : gap],width=2 ] through iterations , for @xmath106 .
@xmath107 is the largest at first iteration .
right panel : @xmath107 at the end of the iterations .
[ fig : omega],width=2 ] for vacuum polarization @xmath108 . in the left panel
, @xmath92 is the smallest at first iteration .
[ fig : pi],width=2 ] another manifestation of the independence of chiral symmetry restoration and deconfinement is the appearance of confined , chiraly symmetric , quakyonic , matter @xcite .
since @xmath109 we can extract explicit @xmath110 dependence of the effective interaction , @xmath111 the @xmath35 potential being proportional to @xmath112 at large @xmath110 becomes @xmath113 since the vacuum polarization contribution is suppressed in the @xmath0 limit . in this limit
we thus find that the debye screening disappears and confinement is restored at any density .
the gap equation ( [ gapf ] ) , however is not affected by the large @xmath110 limit , and the pauli blocking remains in effect .
thus as density increases it will eventually prevent gap equation from developing nontrivial , charily broken solution @xcite .
modifications of charmonium properties , in particular binding energy and size may be strongly affected when the bound state propagates through the plasma . at finite temperature
such modifications have been studied by computing charmonium spectrum using a temperature - dependent static potential that is extracted from the lattice calculation @xcite . with such temperature dependent static potential , melting of charmonium
can be inferred from the temperature dependence of the spectral function @xcite or by directly solving the bound state shrdinger equation @xcite . in the former approach , melting of charmonium
can be seen when the bound state peak of spectral function collapses and as temperature increases it becomes buried under the continuous background . from shrdinger equation calculation melting
is inferred at a temperature when the bound state solution disappears .
melting temperature of the @xmath114 , from these two different approaches appears to be in a very good agreement at approximately @xmath115 .
here we use the shrdinger equation approach to calculate dissociation of the @xmath114 at finite density with two flavors of light quarks .
the mass @xmath116 of the @xmath114 is given by @xmath117 , \end{aligned}\ ] ] and @xmath118 with @xmath119 being the radial quantum number , @xmath120 stands for total quark spin @xmath121 and relative orbital angular momentum @xmath122 .
the heavy ( charm ) quark mass , @xmath123 absorbs any finite shift that has been removed from @xmath124 as discussed in sec .
[ q0 ] and is fixed , by fitting spin - averaged @xmath125 charmonium masses @xmath126 = 3.068\mbox { gev}$ ] .
the energy of @xmath114 and its first radial excitation as a function of the light quark density are given in table [ eigenenergy ] and the binding energy is plotted in fig .
[ psi ] . from our numerical evaluation
is follows that melting occurs at light quark density @xmath127 for @xmath114 and @xmath128 for @xmath129 . finally in fig .
[ wf ] , we show how the wave function of @xmath114 collapses with increasing density . [ eigenenergy ] .
@xmath114 and @xmath129 energy spectrum at finite density [ cols="^,^,^,^",options="header " , ] [ default ] and @xmath130 at finite density , where @xmath131 is defined in eq.([schrodinger ] ) .
[ psi],width=2 ] s wave function @xmath132 at finite density .
[ wf ] , width=2 ]
in this work we studied the screened quark - quark effective potential at finite temperature and presented numerical result for the effective potential .
we investigated the restoration of chiral symmetry using the many body framework of the coulomb gauge qcd and find that the transition densities of chiral symmetry restoration and de - confinement need not be related . in the deconfined phase , chiral symmetry can be broken at density @xmath102 to @xmath133 .
we also computed the melting density for @xmath114 and @xmath129 , which we find to be at @xmath127 and @xmath128 respectively .
similar to the situation at high temperature and low density @xcite , collision of gluons and quarks with heavy quarkonium will reduce the dissociation temperature of quarkonium .
we are expecting the same situation occurs at high density and low temperature .
as pointed out previously @xcite , production of quarkonium in heavy - ion collision may provide clear signal for quark - gluon plasma .
clearly , the collision effect of gluons and quarks with heavy quarkonium should be incorporated in this case , the detail discussion of collision dissociation of heavy quarkonium at the phase of high density and low temperature is under way . | in the framework of coulomb gauge qcd we explore dynamical breaking of chiral symmetry and screening of the confinement potential at finite density .
the screened potential is applied in study of charmonium dissociation . |
semiconductor superlattices ( ssl ) have attracted growing attention in view of their unique electronic properties , which can be used for generation , amplification and detection of a high - frequency electromagnetic radiation @xcite .
nonlinear transport properties of ssl allows a generation of thz radiation in conditions of negative differential conductance ( ndc ) @xcite .
however , the static ndc makes ssl unstable against formation of high - field electric domains @xcite .
these electric domains are believed to be destructive for the thz gain in ssls .
currently the main focus is on the possibilities to overcome this drawback within the scheme of dc - biased ssl @xcite .
however , schemes of thz superlatice devices with ac pump fields are also under discussion @xcite .
a strong microwave field @xmath0 pumps the ssl and a desirable signal field @xmath1 has a higher frequency @xmath2 .
because for typical ssls the characteristic scattering time @xmath3 at room temperature is of the order of @xmath4 fs , an interaction of the microwave fields with the miniband electrons is quasistatic ( @xmath5 ) .
importantly , we have showed recently that such quasistatic pump field can completely suppress domains in ssls @xcite .
two distinct possibilities exist for the signal field : it can be also quasistatic ( @xmath6 ) or it can not be described within the quasistatic approach if @xmath7 . the later situation , i.e. @xmath5 but
@xmath8 , can be called _
semiquasistatic interaction_. thus , the semiquasistatic approach is introduced to describe an amplification of thz field in ssl under the action of microwave field .
our main findings are following .
we consider the response of miniband electrons to an action of the total electric field @xmath9 , where @xmath10 is the dc bias and @xmath11 ( @xmath12 , @xmath13 ) is the weak signal ( probe ) field . in the real devices
@xmath14 is a mode of the resonator tuned to a desirable thz frequency .
we define the dimensionless absorption of the weak ac probe field in ssl as @xmath15 , where @xmath16 is the miniband electron velocity defined in the units of the maximal miniband velocity , averaging @xmath17 is performed over time . starting from the exact formal solution of the boltzmann transport equation
, we represent the absorption @xmath18 as a sum of three terms @xcite @xmath19 here @xmath20 , @xmath21 and @xmath22 describe the absorption ( gain ) seeded by generation of harmonics , the parametric amplification of the probe field due to a coherent interaction of the pump and the probe fields , and the nonparametric absorption , correspondingly .
the derivation of semiquasistatic formulas is based on the use of the asymptotic saddle - point method . using this method we have get the following results .
the term @xmath20 is just the expression for @xmath23th in - phase harmonic of the time - dependent current through ssl @xmath24 where @xmath25 is the dc voltage , @xmath26 is the length of ssl ( @xmath27 is the period of ssl and @xmath28 is the number spatial periods ) , @xmath29 is the amplitude of ac voltage created by the pump field across ssl , the current @xmath30 is normalized to the maximal current in ssl , @xmath31 , corresponding to the maximal miniband velocity .
note that @xmath20 gives the main contribution to the absorption of a weak probe @xcite .
the expression for @xmath20 is not specific for the semiquasistic limit because of its independence on @xmath32 .
however , _ our main finding is that the absorption components @xmath21 and @xmath22 can be represented within semiquasistatic approach using the specific terms of quantum derivatives as _
@xmath34 where @xmath35 is the amplitude of small - signal voltage and @xmath36 is the esaki - tsu voltage - current ( ui ) characteristic ( @xmath37 is the critical voltage and @xmath38 is normalized to the maximal current @xmath39 ) .
importantly , following eq .
( [ inc ] ) in order to find the incoherent absorption in ssl at arbitrary high frequency @xmath32 we need to know only @xmath40 that is , _ the time - averaged current induced by the quasistatic field ( voltage)_. for a given amplitude of ac voltage @xmath41 , the dc current @xmath42 is a function of only dc bias @xmath43 .
it is easy to calculate or to measure the modifications of ui characteristic caused by the action of microwave ( quasistatic ) field @xcite .
note that in the quasistatic case @xmath22 goes into a usual derivative of current .
the difference between these derivatives is shown on the figure [ fig:1 ] where the dashed line is a derivative of dc current and the dotted line is a quantum derivative depended on the number of harmonic @xmath23 .
the importance of finding of @xmath22 is stipulated by the following fact : @xmath22 plays an essential role in the stabilization of space - charge instability in ssl .
we need to know @xmath22 to determine the conditions of space - charge instabilities .
it is surprise that we can determine it only measuring the dc current as a function a dc bias . using only a finite difference we can select the working point of a generator that is optimal for its stability .
note that @xmath44 in the quasistatic case and the system is stable against small fluctuations of internal field . on the other hand , following eq .
( [ coh ] ) finding of the coherent component of absorption at the high frequency corresponding to @xmath23th harmonic of the pump frequency ( @xmath12 ) is reduced to _ the calculation of @xmath45th harmonic of the current within quasistatic approach_. that is , one needs to know @xmath46 ( note that in comparison with ( [ harm - gen ] ) we have in ( [ harm - vm ] ) : @xmath47 ) . in analogy with @xmath22 , the coherent absorption @xmath21 can be found if we know only the dependence on dc bias .
@xmath21 is proportional to the quantum derivative of @xmath20 in the semiquasistatic case , and @xmath21 goes into a common derivative in the quasistatic case .
these derivatives are shown in the figure [ fig:2 ] ( @xmath48 ) ) for the case of 7 harmonic . to the best of our knowledge
this paper is the first work , where the quantum derivatives naturally appeared in a description of microstructure s response to a bichromatic field .
it is the exciting problem to understand whether our semiquasistatic approach , developed for superlattices operating in the miniband transport regime , can be generalized to other tunnelling structure .
a. wacker , _ phys .
rep . _ * 357 * , 1 ( 2002 ) .
s. a. ktitorov , g. s. simin and v. ya .
sindalovskii , _ sov .
phys . solid state _ * 13 * , 1872 ( 1972 ) .
a. a. ignatov and v. i. shashkin , _ sov .
jetp _ * 66 * , 526 ( 1987 ) .
g. savvidis , _ phys
lett . _ * 92 * , 196802 ( 2004 ) .
h. kroemer , _ cond - mat/0009311_. k. n. alekseev , _ europhys .
( in print ) and _ cond - mat/0601056_. f. klappenberger , _ appl . phys .
lett . _ * 84 * , 3924 ( 2004 ) .
v. v. pavlovich , _ sov .
phys . solid state _ * 19 * , 54 ( 1977 ) f. klappenberger , _ eur .
j. b _ * 39 * , 483 ( 2004 ) .
s. winnerl , _ phys .
b _ * 56 * , 10303 ( 1997 ) . | we derive simple difference formulas describing a small - signal absorption of thz field in a semiconductor superlattice driven by a microwave pump .
we give a transparent geometric interpretation of these formulas that allows a search of optimum conditions for the gain employing only a simple qualitative analysis .
our theoretical approach provides a powerful tool for finding the correspondence between quasistatic and dynamic regimes in ac - driven semiconductor superlattices . |
despite an overwhelming evidence for the existence of dark matter ( dm ) in our universe @xcite , its real nature is still a great mystery as it has been only seen through gravitational interactions . from the particle physics point of view
, it is generally expected that dm is composed of some stable particles or the ones with the lifetime much longer than the age of the universe .
since these particles are distributed in our galaxy and the universe , it is widely believed that their annihilations and decays would give rise to the visible signals in terms of light stable particles , such as positrons / electrons , ( anti)protons , photons and neutrinos , which can be observed on the earth . as a result
, such an indirect search for dm @xcite is regarded as one of the most promising ways to detect dm .
recently , there has been a great advance in the dm indirect detections .
for example , the excesses of positrons / electrons were discovered by many experiments , such as ams-01 @xcite , atic @xcite , pamela @xcite , fermi - lat @xcite , and so on , and further confirmed last year by the ams-02 collaboration @xcite with the unprecedented precision .
these excesses can be interpreted as the possible signals as dm annihilations @xcite or decays @xcite although the astrophysical sources like pulsars @xcite would also provide the solution .
the previous studies on the dm annihilating / decaying explanations to the @xmath3 excesses showed many constraints and limitations .
one of the most stringent constraints comes from the measurement of the antiproton flux and the @xmath4 ratio by pamela @xcite , which agrees with the conventional astrophysical theory very well .
it clearly implies that dm annihilation / decay should not disturb the ( anti)proton flux spectrum much .
one simple way to achieve this is the so - called leptophilic dm framework which only allows dm to couple to the lepton sector of the standard model ( sm ) , rather than to quarks and gauge bosons .
however , even in this framework , the simplest scenario in which a single dm component annihilating or decaying into lepton pairs can not fit the precise spectra of the fermi - lat total @xmath5 flux and the ams-02 positron fraction simultaneously @xcite . in the literature , several scenarios have been proposed in order to reduce this tension , such as those with three / four - body decaying / annihilating @xcite or asymmetric decaying channels @xcite , and dynamical dms @xcite as well as other astrophysical solutions @xcite . in ref .
@xcite , we have given a multi - component dm scenario which can overcome the difficulty of accommodating the ams-02 positron fraction with the fermi - lat total flux data , while other aspects of multi - component dm models have been investigated in refs . @xcite .
in particular , we have shown that two dm components with the heavy dm decaying solely via the @xmath6-channel and the light one predominantly via the @xmath7-channel with the energy cutoff at @xmath8 gev could already give a reasonable fit to both spectra of the ams-02 positron fraction and the fermi - lat total @xmath1 flux .
another advantage for this two - component dm model is that the apparent substructure at around 100 gev in the above two spectra can have the simple explanation that the light dm drops at that energy .
we have also demonstrated that the predicted diffuse @xmath2-ray spectrum with the best - fit parameters in the @xmath3 spectrum does not exceed the recent fermi - lat bound @xcite , and is well consistent with the measured one by fermi - lat . in this report , we extend our discussions on the multi - component dm scenario in ref . @xcite . note that in our previous treatment of the two - dm case , we have fixed our heavy dm energy cutoff to be 1500 gev and only opened the @xmath6-channel with it .
although the choice of the parameter and channel is already enough to provide a good fit , it is not quite generic .
it is much better to relax such conditions by allowing the energy cutoff to vary and the three leptonic channels to be active in the fitting procedure .
one interesting result of this analysis is the change of the best - fitting flavor structure with the different values of the heavy dm cutoff
. we will demonstrate that the best - fitting parameters would give @xmath9 for the combined ams-02/fermi - lat data , while the predicted diffuse @xmath10ray spectra agree with the fermi - lat measurement well .
the three - component dm case is also explored in a specific example , in which on the basis of the previous two - dm case with the energy cutoffs 100 gev and 1500 gev , a third dm is added with the intermediate cutoff lying in the range between 500 gev and 900 gev .
we find that this additional dm contributes nothing to the final @xmath3 flux spectra in the best - fit point favored by the combined ams-02/fermi - lat data .
this paper is organized as follows . in sec .
2 , we introduce and analyze our multi - component decaying dm models . in sec . 3
, we discuss the diffuse @xmath2-ray flux spectra and compare our results with the fermi - lat measurement .
we give the conclusions in sec .
let us begin our discussions by reviewing the computation of the fluxes of electrons / positrons and diffuse @xmath2-rays in our multi - component dm framework .
in general , the total electron flux @xmath11 is composed of primary , secondary and dm - decay - induced electrons , while for the positrons in the cosmic rays ( crs ) , only secondary positrons and the ones from the dm decays contribute , which can be parametrized as follows : @xmath12 it is widely believed that the primary electrons come directly from the supernova remnants distributed in our galaxy @xcite , and their injection spectrum is usually assumed to be a broken power - law function with respect to the rigidity . note that in order to take into account the normalization uncertainty in the primary electrons , we insert a parameter @xmath13 which would be determined in the following fitting procedure .
secondary electron / positron fluxes @xmath14 are the final products of the collisions of the charged particles in the crs , such as protons and other nuclei , with the interstellar medium ( ism ) in the galaxy . in the present work ,
we use the galprop code @xcite to simulate the productions and propagations of these background electrons and positrons . for the details of the calculations , especially the choice of the astrophysical parameters
, we refer to our earlier work in ref .
the borderline of the calculations is that the spectra of both the positron fraction and the total @xmath5 flux are the decreasing power - law functions with respect to the energies . for the dm signals @xmath15 , we assume that all of the dm components dominantly decay via the two - body leptonic processes @xmath16 , where @xmath17 denotes the @xmath18-th dm component , @xmath19 , and @xmath20 is a charged particle with its mass taken to be @xmath21 gev . now , we can express the dm - decay induced source terms @xmath22 as : @xmath23 where @xmath24 , @xmath25 and @xmath26 denote the mass , lifetime and energy density distribution for the @xmath18-th dm component , respectively .
for simplicity , we assume that for the n - component dm models , each dm would carry the same fraction of the total energy density @xmath27 in our galaxy , implying that @xmath28 . moreover , we shall always take @xmath27 to be of the isothermal profile @xcite . in eq .
[ source ] , @xmath29 is the differential @xmath30 multiplicity per annihilation , which can be expressed as the linear combination of the aforementioned three leptonic channels : @xmath31\;,\end{aligned}\ ] ] where @xmath32 represent the corresponding branching ratios with the normalization condition @xmath33 and the factor @xmath34 accounts for the fact that @xmath35 and @xmath36 arise from two different channels . since the processes in our discussion are all two - body decays , the final spectrum can be determined solely by the kinematics , irrelevant to any other details of the underlying theory . for @xmath37- and @xmath6-channels ,
the normalized @xmath38 energy spectra @xmath39 can be expressed analytically as follows : @xmath40\theta(1-x),\end{aligned}\ ] ] with @xmath41 , while @xmath42 can only be obtained by the simulation of the @xmath7 decay with pythia @xcite . after being generated from the dm
decays , the positrons and electrons will propagate through a long way in our galaxy until they are detected by our satellites and telescopes .
the propagations of @xmath3 are very complex @xcite , during which they would be deflected by the galaxy magnetic fields and lose their energies via the inverse compton ( ic ) scattering , bremsstrahlung and synchrotron radiation . in the present work ,
these complicated propagations are solved with the galprop codes by dealing with all of these effects in a consistent way .
note that in our practical calculation we use the same parameter set for the @xmath3 diffuse propagations as those listed in table i in ref .
finally , it is generally expected that the positron and electron fluxes are suffered from the solar modulation , especially for those with energies below 10 gev . here
, we apply the simple force - field approximation @xcite to account for such effects with the potential @xmath43 gv . as already shown in ref .
@xcite , two dm components , denoted by @xmath44 , representing the light ( heavy ) dm , are enough to accommodate the ams-02 positron fraction and the fermi - lat total @xmath5 flux simultaneously .
note that in that analysis , the energy cutoffs of the two dm components were fixed to be 100 gev and 1500 gev , respectively , and for the heavy dm component only @xmath6-channel was opened , in order to simplify the calculation .
obviously , such a analysis is not very generic and it is more appropriate to investigate more possibilities .
this direction is pursued in the present study . here
, we still fix the cutoff @xmath45 of the light component @xmath46 to be 100 gev , which has the advantage that the substructure around 100 gev shown in both the ams-02 and fermi - lat data could be explained as the contribution to @xmath3 from the light dm terminating exactly there .
however , for the cutoff @xmath47 of the heavy component @xmath48 , it is taking to be free in the energy range from 1000 gev to 2000 gev since the fermi - lat total @xmath5 flux is extended to as high as 1000 gev .
furthermore , we let the entire 3 leptonic decay channels active , and pay a special attention to the variation of the flavor structure for the best - fitting points for each @xmath47 . for the fitting procedure ,
we apply the simple @xmath49-minimization method , and take the 26 data points of the total electron / positron flux from fermi - lat and the 42 data points of the positron fraction from ams-02 with the energy above 10 gev to reduce the influence of the solar modulation . in table
[ tab_2 dm ] and fig . [ fig_2 dm ] , we present the best - fit results with @xmath50=62.4 , 62.3 and 62.8 and @xmath51 all around 1.06 for the three typical examples with the cutoffs and the masses of the heavy dm at ( @xmath47,@xmath52)=(1200 , 2436 ) gev , ( 1500,3030 ) gev and ( 1800,3624 ) gev , respectively . @cccccc@ @xmath53 & @xmath13 & @xmath54 & @xmath55 & @xmath56 & @xmath57 + 1200 & 0.844 & 0.206,0.015 & 0.794,0 & 0,0.985 & 0.97,0.83 + 1500 & 0.844 & 0.058,0.020 & 0.942,0 & 0,0.980 & 0.78,0.82 + 1800 & 0.843 & 0,0.022 & 0.842,0 & 0.158,0.978 & 0.64,0.83 + [ tab_2 dm ] note that all three cases essentially give the same degree of the goodness of the fitting , indicating that the fitting results are very reasonable . however , what is interesting here is the modification of the flavor structure with the different choices of the cutoff for the heavy dm component .
specifically , when the heavy dm cutoff @xmath47 is chosen to be smaller than 1500 gev , quite a considerable portion of @xmath3 comes from the heavy dm decay via the electron channel . and
when @xmath47 becomes larger , the heavy dm contribution to @xmath3 from the @xmath37-channel would decrease , finally terminating at around 1500 gev , whereas the positrons / electrons from the @xmath58channel would arise and increase .
this feature shows the power of the precise measurement in the indirect dm search experiments , from which we can obtain a lot of information of the dm properties .
in general , it is expected that new dm components would result in better fittings for the @xmath3 spectra due to the more free parameters from the mixture of those of dm components . however
, we shall illustrate an interesting counterexample in this paper .
if we fix two energy cutoffs of the three dms to be 100 gev and 1500 gev as before and add a middle - mass dm with its energy cutoff lying in the range between 500 gev and 900 gev , the minimum @xmath49 with the ams-02/fermi - lat data yields that the middle - mass dm does not contribute any electrons and positrons to the final @xmath3 fluxes .
table [ tab_3 dm ] shows a typical example of this case , in which the cutoff and the mass of the extra dm are fixed to be ( @xmath59 , @xmath60)=(760 , 1577 ) gev .
@ccccccc@ @xmath13 & @xmath61 & @xmath62 & @xmath63 & @xmath64 & @xmath50 & @xmath51 + 0.844 & 0.058,0,0.020 & 0.942,0,0 & 0,0,0.980 & 0.52,@xmath65,0.55 & 62.3 & 1.06 + [ tab_3 dm ] it is seen that the best - fitting values prefer the three leptonic channels of the middle - mass dm to be all closed .
the only difference is the lifetimes of two remaining dms , which are just a factor 2/3 of those in the corresponding two - dm case in order to compensate the less energy density for each dm in eq .
( [ source ] ) .
thus , the resultant spectra of the positron fraction and the @xmath5 flux are exactly the same as the corresponding two - dm case with ( @xmath45 , @xmath66)=(100 , 416 ) gev and ( @xmath47 , @xmath52)=(1500 , 3030 ) gev .
the above discussion implies that in this specific range of the three - dm parameter space , the corresponding two - dm models are more statistical favored .
the positrons and electrons from dm decays or annihilations are inevitable to be followed with the emissions of high energy photons manifested in the measured diffuse @xmath2-ray spectrum @xcite .
however , the produced @xmath2-ray flux should not exceed the bounds obtained by the diffuse @xmath2-ray data by fermi - lat @xcite and egret @xcite .
thus , as pointed in refs .
@xcite , these diffuse @xmath2-ray constraints have already excluded a large portion of the parameter space of the decaying dm models to explain the pamela / fermi - lat @xmath3 excesses . nevertheless , in ref .
@xcite , we have demonstrated that our multi - component dm scenario survives even under the stringent @xmath2-ray constraints .
the diffuse @xmath2-ray flux in our present model has many components , including the usual astrophysical diffuse background @xmath2-ray radiation and dm contributions inside and outside our galaxy .
the final spectrum will be compared with the fermi - lat inclusive continuum photon spectrum @xcite .
the conventional astrophysical background inside the milky way includes three sources : pion decay , inverse compton ( ic ) scattering and bremsstrahlung .
all of them are calculated by the galprop code with the same astrophysical parameters already determined when we compute the electron / positron fluxes in the previous section . for the extragalactic @xmath2-ray background ( egb ) originated from the superposition of the unresolved extragalactic sources such as the active galactic nuclei ( agn ) , we take the form @xmath67 , obtained by fitting the low - energy data from egret @xcite . the dm contributions to the @xmath2-ray fluxes can also be divided into the parts inside and outside the galaxy . inside the galaxy ,
the extra electrons and positrons as the decay products of multiple dm components can induce the @xmath2-rays in the processes such as bremsstrahlung and ic .
moreover , the additional prompt decay processes , such as the final state radiations ( fsrs ) from the three leptons @xcite , the radiative muon decays @xcite and the pion decays from @xmath7 @xcite , could give rise to the continuous spectrum of the @xmath2-rays .
the dms outside the galaxy are also generally expected to lead to the @xmath2-rays via the prompt decay processes and the ic scattering with the cmb photons .
the only additional effect is the redshift of the spectra caused from the cosmic expansion . for the calculation details
, please refer to ref .
@xcite . here
, we only emphasize that in these calculations of dm diffuse @xmath2-ray spectra , all of the parameters are already fixed by fitting the ams-02/fermi - lat data , and thus the final spectra are the predictions of our dm models . by summing up the above backgrounds and dm contributions , we can obtain the total diffuse @xmath2-ray spectra in the multi - component decaying dm models .
[ fig_3 dm ] shows the diffuse @xmath2-ray spectra for the two - component dm models studied in section [ 2dmcase ] with the parameters listed in table [ tab_2 dm ] . from the table , it is clear that the the predictions of the @xmath2-ray fluxes in these three cases are consistent with the fermi - lat measurement in all the energy range . in order to further support this conclusion ,
we compute the usual @xmath49 with the 82 fermi - lat data points .
the final results are @xmath68 80.4 , 83.5 and 151.0 with @xmath69 for the two - component dm models , in which the cutoffs and masses of the heavy dm are fixed to be ( @xmath47 , @xmath52 ) = ( 1200 , 2436 ) gev , ( 1500 , 3030 ) gev and ( 1800 , 3625 ) gev , respectively .
we have explored the multi - component decaying dm scenario along the line described in ref .
the extension is in two aspects .
first , we allow the energy cutoff of the heavy dm to vary in the range between 1000 and 2000 gev , followed by the change of the heavy dm mass accordingly , and open all of 6 leptonic decay channels in the two - component dm case .
we have found an interesting change of the best - fit heavy dm flavor structure with the different choices of the cutoff and the mass of the heavy dm . with the increase of the heavy dm cutoff in the assigned range ,
the electron - channel branching ratio of the heavy dm originally with a relatively large size begins to decrease and stop at around 1500 gev , whereas from around 1500 gev its @xmath7-channel branching ratio arises and increases until a considerable portion at 2000 gev .
the points in the parameter space all give the reasonable fits to the ams-02 positron fraction and the fermi - lat total @xmath5 flux data , with the @xmath70 .
we have further shown that the predicted diffuse @xmath2-ray spectra for all these examples are consistent with the fermi - lat measurement .
we have also investigated the three - component dm model with the particular focus on a region of the parameter space . besides the two dms with the energy cutoffs 100 gev and 1500 gev as the two - component case
, we have incorporated the a third dm with the cutoff @xmath59 lying in the range between @xmath71 gev and @xmath72 gev .
this parameter space range is interesting because within it , the fittings of other parameters , especially the flavor branching ratios for the three dms , prefer the original two - component dm cases . for the best - fitting points
, the three leptonic channels related to the middle - mass dm do not contribute any @xmath3 to the final spectra .
this result contradicts our general expectation that the addition of more dms would lead to a better fitting due to a complicated mixture of the different channels .
it is clear that the present multi - component decaying dm scenario should be further studied .
one possible problem is related to the fact that we have only introduced a phenomenological decay coupling of dms with the effective dimensionless coupling being of o(@xmath73 ) , which is obviously quite unnatural . in order to examine this problem
, we should resort to some more fundamental particle models implemented with the full lagrangian and symmetries .
another question is about the compatibility of the ams-02 positron fraction and the fermi - lat total @xmath1 flux data .
more recently , the ams-02 collaboration has presented the preliminary measurements on the electron , the positron and the total @xmath1 fluxes .
it is more appropriate to use the data sets from a single experiment , say ams-02 .
we will study these issues elsewhere .
the work was supported in part by national center for theoretical science , national science council ( nsc-101 - 2112-m-007 - 006-my3 ) and national tsing hua university ( grant nos . 102n1087e1 and 102n2725e1 ) .
a. a. abdo _ et al .
_ [ fermi lat collaboration ] , phys . rev .
lett . * 102 * , 181101 ( 2009 ) .
m. ackermann _ et al .
_ [ fermi lat collaboration ] , phys . rev .
d * 82 * , 092004 ( 2010 ) .
m. ackermann _ et al .
_ [ fermi lat collaboration ] , phys . rev .
lett . * 108 * , 011103 ( 2012 ) .
m. aguilar _ et al . _ [ ams collaboration ] , phys .
lett . * 110 * , 141102 ( 2013 ) .
k. ishiwata , s. matsumoto and t. moroi , phys . lett .
b * 675 * , 446 ( 2009 ) ; l. bergstrom _ et al .
_ , phys .
111 * , 171101 ( 2013 ) ; d. gaggero and l. maccione , jcap * 1312 * , 011 ( 2013 )
. l. bergstrom , t. bringmann and j. edsjo , phys .
d * 78 * , 103520 ( 2008 ) ; m. cirelli and a. strumia , pos idm * 2008 * , 089 ( 2008 ) ; e. nezri , m. h. tytgat and g. vertongen , jcap * 0904 * , 014 ( 2009 ) ; x. j. bi _ et al .
_ , jhep * 0904 * , 103 ( 2009 ) ; y. .b .
zeldovich _ et al . _ ,
j. nucl .
* 31 * , 664 ( 1980 ) ; r. v. konoplich and m. yu . khlopov , phys .
nucl . * 57 * , 425 ( 1994 ) ; d. fargion _ et al .
_ , phys .
d52 * , 1828 ( 1995 ) ; astrodamus collaboration , in : proc . 1 int
. conf . on cosmoparticle physics
`` cosmion-94 '' , dedic . to 80 anniv . of ya.b .
zeldovich and 5 mem . of a.d.sakharov , moscow , dec . 5 - 14 , 1994 .
_ , editions frontieres , 1996
99 - 106 and 107 - 112 ; k. belotsky _ et al .
_ , phys . atom
. nucl . * 71 * , 147 ( 2008 ) .
k. cheung , p. y. tseng and t. c. yuan , phys .
b * 678 * , 293 ( 2009 ) . c. r. chen and f. takahashi , jcap
* 0902 * , 004 ( 2009 ) ; p. f. yin _ et al .
_ , phys .
d * 79 * , 023512 ( 2009 ) ; k. hamaguchi _ et al .
_ , phys .
b * 674 * , 299 ( 2009 ) ; a. ibarra and d. tran , jcap * 0902 * , 021 ( 2009 ) ; e. nardi , f. sannino and a. strumia , jcap * 0901 * , 043 ( 2009 ) ; i. gogoladze _ et al .
_ , phys .
d * 79 * , 055019 ( 2009 ) ; s. l. chen _ et al .
_ , phys .
b * 677 * , 311 ( 2009 ) ; a. arvanitaki _ et al .
_ , phys . rev .
d * 80 * , 055011 ( 2009 ) .
k. ishiwata , s. matsumoto and t. moroi , jhep * 0905 * , 110 ( 2009 ) .
a. ibarra , d. tran and c. weniger , int .
j. mod .
a * 28 * , 1330040 ( 2013 ) ; h. fukuoka , j. kubo and d. suematsu , phys .
b * 678 * , 401 ( 2009 ) ; a. ibarra , a. s. lamperstorfer and j. silk , phys .
d * 89 * , 063539 ( 2014 ) ; k. belotsky _ et al .
_ , adv . high energy phys . * 2014 * , 214258 ( 2014 ) . a. arvanitaki _ et al .
_ , phys .
d * 79 * , 105022 ( 2009 ) ; k. hamaguchi , s. shirai and t. t. yanagida , phys .
b * 673 * , 247 ( 2009 ) ; c. h. chen , c. q. geng and d. v. zhuridov , phys .
b * 675 * , 77 ( 2009 ) .
m. ibe _ et al .
_ , jhep * 1307 * , 063 ( 2013 ) ; k. kohri and n. sahu , phys .
d * 88 * , 103001 ( 2013 ) .
c. h. chen , c. q. geng and d. v. zhuridov , jcap * 0910 * , 001 ( 2009 ) .
v. barger _
et al . _ ,
b * 672 * , 141 ( 2009 ) ; m. cirelli _ et al .
_ , nucl .
b * 813 * , 1 ( 2009 ) ; c. r. chen , f. takahashi and t. t. yanagida , phys . lett .
b * 673 * , 255 ( 2009 ) ; c. r. chen _ et al .
_ , prog .
122 * , 553 ( 2009 ) ; j. liu , p. f. yin and s. h. zhu , phys .
d * 79 * , 063522 ( 2009 ) .
s. profumo , central eur .
j. phys .
* 10 * , 1 ( 2011 ) ; t. linden and s. profumo , astrophys .
j. * 772 * , 18 ( 2013 ) ; p.
f. yin _ et al .
_ , phys .
d * 88 * , 023001 ( 2013 ) ; d. gaggero _ et al .
_ , phys .
111 , * 021102 * ( 2013 ) .
o. adriani _ et al . _
[ pamela collaboration ] , phys .
105 * , 121101 ( 2010 ) .
l. feng _ et al .
_ , phys .
b * 728 * , 250 ( 2014 ) .
a. sharma , arxiv:1304.0831 [ astro-ph.co ] ; j. kopp , phys .
d * 88 * , 076013 ( 2013 ) ; a. de simone , a. riotto and w. xue , jcap * 1305 * , 003 ( 2013 ) ; i. cholis and d. hooper , phys . rev .
d * 88 * , 023013 ( 2013 ) ; q. yuan and x. j. bi , phys .
b * 727 * , 1 ( 2013 ) .
q. yuan _ et al .
_ , arxiv:1304.1482 [ astro-ph.he ] ; h. b. jin , y. l. wu and y. f. zhou , jcap * 1311 * , 026 ( 2013 ) . k. petraki and r. r. volkas , int
. j. mod .
a * 28 * , 1330028 ( 2013 ) ; l. feng and z. kang , jcap * 1310 * , 008 ( 2013 ) ; k. m. zurek , phys .
rept . * 537 * , 91 ( 2014 ) .
k. r. dienes , j. kumar and b. thomas , phys .
d * 88 * , 103509 ( 2013 ) .
c. q. geng , d. huang and l. h. tsai , phys .
d * 89 * , 055021 ( 2014 ) .
y. kajiyama , h. okada and t. toma , arxiv:1304.2680 [ hep - ph ] .
j. fan _ et al .
_ , phys .
dark univ .
* 2 * , 139 ( 2013 ) ; phys .
* 110 * , 211302 ( 2013 ) ; m. mccullough and l. randall , jcap * 1310 * ,
058 ( 2013 ) ; a. g. doroshkevich and m. y. .khlopov , yad . fiz .
* 39 * , 869 ( 1984 ) ; m.yu.khlopov , in : dark matter in cosmology , clocks and tests of fundamental laws .
eds.b.guiderdoni et al .
editions frontieres , pp .
133 - 138 ( 1995 ) ; m. y. khlopov and c. kouvaris , phys .
d * 78 * , 065040 ( 2008 ) .
m. ackermann _ et al . _
[ lat collaboration ] , phys .
d * 86 * , 022002 ( 2012 ) .
r. trotta _ et al .
_ , astrophys .
j. * 729 * , 106 ( 2011 ) .
a. w. strong and i. v. moskalenko , astrophys .
j. * 509 * , 212 ( 1998 ) . k. g. begeman _ et .
mon . not .
249 * , 523 ( 1991 ) .
r. essig , n. sehgal and l. e. strigari , phys .
d * 80 * , 023506 ( 2009 ) .
a. a. abdo _ et al .
_ [ fermi - lat collaboration ] , phys . rev .
lett . * 104 * , 101101 ( 2010 ) ; m. cirelli and p. panci , nucl .
b * 821 * , 399 ( 2009 ) ; s. matsumoto , k. ishiwata and t. moroi , phys .
b * 679 * , 1 ( 2009 ) .
a. ibarra , d. tran and c. weniger , jcap * 1001 * , 009 ( 2010 ) . c. r. chen , f. takahashi and t. t. yanagida , phys . lett .
b * 671 * , 71 ( 2009 ) .
a. ibarra and d. tran , phys .
lett . * 100 * , 061301 ( 2008 ) .
p. sreekumar _ et al .
_ [ egret collaboration ] , astrophys . j. * 494 * , 523 ( 1998 ) .
n. fornengo , l. pieri and s. scopel , phys . rev .
d * 70 * , 103529 ( 2004 ) . | we use multi - component decaying dark matter ( dm ) scenario to explain the possible cosmic ray excesses in the positron fraction recently confirmed by ams-02 and the total @xmath0 flux observed by fermi - lat . in the two - component dm models
, we find an interesting variation of the flavor structure along with the cutoff of the heavy dm .
for the three - component dm case , we focus on a particular parameter range in which the best fits prefer to open only 2 dm decay channels with a third dm contributing nothing to the electron and positron spectra .
we show that all models give the reasonable fits to both the ams-02 positron fraction and the fermi - lat total @xmath1 flux , which are also consistent with the measured diffuse @xmath2-ray flux by fermi - lat . |
one - dimensional ( 1d ) integer - spin heisenberg antiferromagnets ( haf ) are well - known for the haldane energy gap @xcite between the singlet spin - liquid ground state and the lowest excited state , which is an @xmath8 triplet .
the application of a magnetic field leads to zeeman splitting of the triplet and eventual vanishing of the gap @xmath9 at @xmath10 , where the energy of the lowest branch of the split triplet reaches the ground - state level . at this critical field ,
the expected quantum transition is to the tomonaga - luttinger spin liquid , @xcite in which spin correlations decay with characteristic power laws .
this scenario remains robust in an array of 1d integer - spin haf chains against the introduction of interchain exchange .
such a coupling reduces @xmath7 but , provided it is small , does not destroy the singlet ground state below @xmath7 . above @xmath7 , it leads to finite - temperature long - range order ( lro ) , which can be described as the three - dimensional bose - einstein condensation of the lower - branch triplets .
@xcite the nature of the ordered state , including the robustness of the tomonaga - luttinger spin liquid , is of strong current interest .
the first experimental evidence @xcite of a haldane gap was found in csnicl@xmath11 .
this material has a relatively large interchain coupling : @xmath12 , where @xmath13 and @xmath14 are interchain and in - chain exchanges , respectively . as a consequence ,
n@xmath15el ordering occurs at 4.85 k. ni(c@xmath3h@xmath16n@xmath3)@xmath3no@xmath3(clo@xmath17 ) ( nenp ) , with the small @xmath18 ratio of @xmath19 , was the first haldane - gap antiferromagnet with no n@xmath15el ordering .
@xcite magnetic susceptibility shows no anomaly indicative of ordering , at least down to 4 mk .
@xcite naturally , this material was a promising candidate for the field - induced lro at fields above @xmath7 .
however , it has been found that such an order is preempted by the staggered effective field arising from the staggered _ g _ tensors of the ni@xmath20 ions within each spin chain . to date , another spin-1 chain material ni(c@xmath1h@xmath2n@xmath3)@xmath3n@xmath4(pf@xmath5 ) ( ndmap ) remains the only laboratory model of a 1d haf array in which the nature of the field - induced lro states has been revealed by experiment .
@xcite this distinction owes to the unusually low @xmath7 , which is readily accessible to many experimental probes , as well as the absence of a staggered field .
ndmap has an orthorhombic structure with the lattice parameters @xmath21 , @xmath22 , and @xmath23 , with the antiferromagnetic spin chains running along the @xmath24 axis .
@xcite according to inelastic neutron scattering , @xcite the in - chain exchange is @xmath25 k , and the relative strengths of the interchain exchanges are @xmath26 and @xmath27 along the @xmath28 and @xmath29 axes , respectively .
easy - plane crystal - field anisotropy @xmath30 , where @xmath31 is the crystallographic @xmath24 axis and @xmath32 , @xcite is responsible for the strong anisotropy of both the gap @xmath9 and the magnetic phase diagram of the field - induced lro states , as has been measured by magnetic susceptibility , @xcite specific heat , @xcite magnetization , @xcite esr , @xcite and neutron scattering .
@xcite the gap energies of the triplet , whose degeneracy is lifted by the crystal - field anisotropy , have been measured by inelastic neutron scattering @xcite to be @xmath33 mev , @xmath34 mev , and @xmath35 mev for the @xmath36 excitations quantized along the @xmath29 , @xmath28 , and @xmath24 axes , respectively .
here @xmath37 is the quantization axis .
the small splitting of a doublet into @xmath38 and @xmath39 is due to a weak in - plane anisotropy @xmath40 . because of the gap anisotropy , @xmath7 depends on the crystal orientation and ranges from about 4 t to 6 t. @xcite above @xmath7 , the ordered states are commensurate with the crystal lattice for all four magnetic - field directions studied by neutron diffraction .
@xcite for field @xmath41 applied along the @xmath29 axis and along @xmath42 $ ] , the order is short - ranged and confined within each @xmath43 plane , whereas it is long - ranged and three - dimensional for @xmath44 and @xmath45 $ ] .
the neutron experiments have detected no incommensurate modulation of the magnetization component parallel to the field .
such a modulation has been predicted for the tomonaga - luttinger spin liquid , @xcite and its absence in ndmap can be understood as a consequence of a lack of axial symmetry , which is necessary for the existence of the tomonaga - luttinger spin liquid .
@xcite it is important to note that even @xmath44 is not an axisymmetric field configuration for ndmap because of its unique geometry . in this material , the crystal - field anisotropy of ni@xmath20
is determined by the local symmetry of the nin@xmath46 octahedron .
first , there is a weak in - plane anisotropy as described above .
second and probably more important , the principal axis of the octahedron is tilted from the @xmath24 axis by 15@xmath47 toward the @xmath29 axis , with the tilt direction alternating from chain to chain .
@xcite therefore , there is no magnetic - field direction that strictly satisfies axial symmetry . in this paper
, we explore the field - induced lro states to 32 t in field and 150 mk in temperature by means of specific heat measurements .
the present study greatly extends the @xmath41-@xmath48 parameter space for the phase diagram , which has been previously limited @xcite to 12 t in @xmath41 and 0.52 k in temperature @xmath48 except for a few isolated points obtained by neutron diffraction @xcite at lower temperatures with less precision .
preliminary results have been reported in ref . .
the single crystals of ndmap used in this work were grown from aqueous solutions by the method described in ref . .
fully deuterated crystals were used to eliminate the nuclear specific heat of protons in the set of measurements extending to the lowest temperature , whereas a hydrogenous sample was used when adequate .
the specific heat measurements to 18 t were performed in a superconducting magnet at temperatures down to 150 mk with a dilution refrigerator .
the relaxation calorimeter for this setup has been described in ref . .
for measurements from 20 t to 32 t , another relaxation calorimeter with a built - in @xmath49he refrigerator was used in a resistive bitter magnet . each sample for the @xmath44 configuration was attached with silver paint @xcite to the vertical face of a small silver bracket , whose horizontal face was in turn glued with a wakefield compound @xcite to the calorimeter platform .
the sample for the @xmath50 configuration was mounted with silver paint on a piece of 0.13 mm - thick sapphire substrate , which in turn was glued with a wakefield compound to the platform .
a total of three samples ranging in mass from 4 mg to 8 mg were used to cover the different field ranges and orientations . axis .
( a ) temperature dependence at constant fields .
the inset gives an expanded view of the region near the specific heat peak .
( b ) magnetic - field dependence at constant temperatures .
the lines are guides for the eye . ] the magnetic - field and temperature dependence of the specific heat of a deuterated sample is shown in fig . 1 for fields up to 18 t , applied along the @xmath24 axis of the crystal .
at each field , the peak in the specific heat seen in fig .
1(a ) clearly indicates a phase transition .
as can be seen in the inset , the transition temperature denoted by the peak position first increases with increasing field up to about 10 t and then decreases for fields up to 14 t , where it starts to increase again , making a shallow local minimum at around 14 t. furthermore , the peak height has a less pronounced minimum at roughly the same field .
these features reveal the existence of a new phase boundary , which separates two field - induced ordered states at around 14 t. for temperatures less than 0.85 k , specific heat is shown as a function of the magnetic field in fig .
again , the peak clearly indicates a phase transition .
we have confirmed that the transition temperatures determined by such magnetic - field scans agree with those determined by the temperature scans . axis .
the lines are guides for the eye . ] the specific heat for the same field direction at higher fields produced by the resistive magnet are shown in fig . 2 .
these data were obtained with the hydrogenous sample . at 26 t , the nuclear contribution of the protons is visible at temperatures below 0.7 k. however , the peak due to transition stands out , since it occurs at a higher temperature . in this field region extending from 20 t to 32 t , the transition temperature obtained from the peak position varies only monotonically , and so does the peak height . axis .
the anomaly at the lowest field of each curve is due to the proximity of the transition to the disordered spin - liquid phase .
the uninteresting proton contribution raises the curve at 0.260.29 k with respect to those at higher temperatures .
the lines are guides for the eye . ]
to further investigate the phase diagram near 14 t , we have measured the specific heat of the same hydrogenous sample as a function of the magnetic field , again applied along the @xmath24 axis , but to higher fields than in fig .
1(b ) . in these measurements ,
a constant electric current was fed to the heater of the thermal reservoir of the calorimeter , allowing the temperature to rise monotonically with an increasing field as dictated by the magnetoresistance of the heater . as seen in fig . 3 ,
a plateau - like anomaly in the specific heat occurs at around 14 t , clearly indicating a phase transition .
this anomaly was overlooked in our preliminary report , @xcite where the field range investigated was too narrow .
within the experimental resolution , no corresponding feature is found in magnetization , which has been measured at 80 mk and 1.3 k using a pulsed magnet and is featureless up to 60 t except for an anomaly at @xmath7 .
@xcite axis .
the lines are guides for the eye . ] for magnetic fields applied along the crystallographic @xmath29 axis , which is perpendicular to the spin chains , another deuterated sample was used to measure the specific heat at temperatures below 2.7 k as a function of field , as shown in fig .
, a transition to the field - induced ordered phase is clearly indicated by a peak at each temperature . for fields ranging from 6.2 t to 6.5 t
, these measurements were supplemented by temperature sweeps similar to those shown in fig .
no new phase boundary was found for this field direction up to 18 t. axis and the closed symbols for the field along the @xmath29 axis .
the circles are for the deuterated samples , and the squares and diamonds for the hydrogenous sample .
the lines are guides for the eye . ]
the magnetic phase diagram determined from the positions of the specific - heat peaks and the plateau - like anomalies is shown in fig . 5 for the two magnetic - field directions studied .
when the field is along the @xmath24 axis , the transition temperature exhibits a shallow but distinct local minimum at about @xmath6 t , and a new phase boundary extends nearly horizontally from this minimum . the small break in the phase boundary between the new high - field phase and the thermally disordered phase at around 20 t indicates that the hydrogenous sample has a slightly higher transition temperature than the deuterated sample .
as stated earlier , the calorimeters used for the two samples were different .
however , we have confirmed in another experiment the consistency of the temperature scales of the two calorimeters .
therefore , the discontinuity in the phase boundary is not an artifact and indicates that deuterated ndmap in fact has a somewhat weaker exchange .
when the field is along the @xmath29 axis , the transition temperature rises rapidly with increasing field , with no indication of a new phase .
this is the first time more than one field - induced ordered phase has been found in a quasi-1d antiferromagnet for a given magnetic - field orientation .
we believe it is significant that the field direction required for the new phase is along the crystalline @xmath24 axis .
this direction , parallel to the spin chains , is quite unlike the @xmath29 , @xmath28 , and @xmath51 $ ] directions in that the ordering field rises very rapidly with temperature in the overall @xmath41-@xmath48 phase diagram .
@xcite this feature probably exemplifies the underlying tendency of the spin chains in this field configuration to form a tomonaga - luttinger spin liquid , as has been presumed earlier .
@xcite hence it is likely that identifying the new phase above @xmath52 and understanding the mechanism of the transition will shed light on the physics of the tomonaga - luttinger spin liquid . at present , however , with the absence of experimental data in the field region near and above @xmath52 other than the specific heat and magnetization , which shows no anomaly at @xmath52 , we can at best speculate on the nature of the transition .
one possible scenario can be a transition involving spin rotation around the direction of the applied field , which is parallel to the spin chains .
such an exotic transition has been observed for instance in a spin-1/2 chain material bacu@xmath53si@xmath53o@xmath54 , where it is presumably driven by a competition between an off - diagonal exchange and the magnetic field .
@xcite although the dzyaloshinskii - moriya interaction , @xcite the likely candidate for the off - diagonal exchange , is usually quite small for ni@xmath20 and has not been detected in ndmap , a spin - rotation transition is an interesting possibility .
recently , wang @xcite has considered a spin-1 chain with broken symmetry using a fermionic field theory and has predicted that a magnetic field larger than @xmath7 will restore an approximate axial symmetry and lead to a well - defined second transition from a commensurate phase to an incommensurate phase . unlike in the tomonaga - luttinger spin liquid
, the excitations in the incommensurate phase will be gapped .
however , the gap will be small and roughly equal to @xmath55 , when the field is applied along the chain direction @xmath24 .
although wang has predicted an absence of anomaly in specific heat as well as magnetization at the second transition , which is other than first - order , it is worth exploring the possibility of the phase above @xmath52 being an incommensurate phase . and @xmath29 axes .
the closed symbols are for the deuterated samples of the present work , where the circles are from the temperature - scan data and the diamonds from the field - scan data .
the open circles are for hydrogenous samples from ref . .
the lines are low - temperature fits of the phase boundaries of the deuterated samples to the expression @xmath56 . ] the bose - einstein condensation of triplets is believed to be a valid description of field - induced lro in all gapped antiferromagnets , regardless of the type of disordered ground state below @xmath7 and the type of the ordered state above it , provided that the lowest excitations are bosons . according to the bose - einstein condensation theory , @xcite the phase boundary in the @xmath41-@xmath48 parameter space obeys a power law @xmath56 .
the field - induced ordering in the @xmath57 spin - dimer material tlcucl@xmath11 was the first case that was analyzed in terms of the bose - einstein condensation .
@xcite however , the exponent @xmath58 of 0.50 determined @xcite for this material is smaller than the theoretical value of 2/3 obtained by a hartree - fock approximation for a quadratic dispersion for the triplets . the exponent for ni(c@xmath1h@xmath2n@xmath3)@xmath3n@xmath4(clo@xmath17 ) ( ndmaz ) ,
a spin-1 chain material @xcite similar to ndmap , has been reported to be 0.45 for the magnetic field applied along the @xmath24 axis .
@xcite this value is also significantly smaller than the theoretical one . the four - fold improvement in the minimum temperature over the previous work @xcite for the field directions along the @xmath24 and @xmath29 axes allows us to determine the field dependence of the ordering temperature near @xmath7 with high accuracy .
the relevant portions of the phase boundaries from fig . 5 are reproduced in fig .
6 , separately for the two field directions .
these boundaries , shown with closed symbols , are for deuterated samples .
excellent fits of the data below 0.77 k for @xmath44 and below 1.9 k for @xmath50 are obtained with the power law , giving @xmath59 and @xmath60 , respectively .
the critical fields are @xmath61 t and @xmath62 t , respectively , along the @xmath24 and @xmath29 axes .
recently , specific heat measurements of the @xmath63 bond - alternating - chain antiferromagnet ni(c@xmath64h@xmath65n@xmath17)no@xmath3(clo@xmath17 ) ( ntenp ) have found @xmath66 and 0.52 , respectively , for the field parallel and perpendicular to the spin chains .
@xcite the similarity of these exponents to those for ndmap suggests that ntenp has similar field - induced ordered phases , although its low - field ground state is a spin - dimer singlet@xcite rather than a haldane spin liquid .
including ours , all the experimental values found to date for the exponent @xmath58 are smaller than 2/3 predicted by the theory .
however , the values for ndmap and ntenp for the magnetic field applied along the chain direction agree , within the combined uncertainties of the experiments and the calculation , with the quantum monte - carlo results by wessel @xmath67 .
, @xcite who have found @xmath68 for @xmath57 spin dimers with a weak three - dimensional inter - dimer coupling and @xmath69 for two - leg spin-1/2 ladders with a weak three - dimensional inter - ladder coupling .
their results support the bose - einstein condensation theory , according to the quantum monte - carlo calculation for three - dimensionally coupled @xmath57 spin dimers by nohadani @xmath67 .
, @xcite who have shown that @xmath58 deviates downward from 2/3 , as the temperature range of the power - law fit is widened .
furthermore , misguich and oshikawa @xcite have used a realistic dispersion for the triplets in tlcucl@xmath11 in their calculation of the field dependence of the bose - einstein condensation temperature @xmath70 and found good agreement with the experiment .
@xcite it remains to be seen , however , whether the bose - einstein condensation theory can explain the strongly anisotropic @xmath58 for ndmap and ntenp , particularly the extremely small @xmath58 when the field is applied along the spin chains . in fig .
6 , we have included the phase boundaries for hydrogenous samples from the previous work , as shown with open symbols .
@xcite again , the ordering temperature is somewhat higher for the hydrogenous samples except at the fields of 10 t and 12 t applied parallel to the @xmath24 axis , whereas @xmath7 is identical for a given field direction within our accuracy .
this indicates that the interchain exchange @xmath13 is slightly weaker in deuterated ndmap than in hydrogenous ndmap , since decreasing @xmath13 lowers the ordering temperature directly but has little effect on @xmath7 , which is primarily determined by the in - chain exchange @xmath14 . a previous study , @xcite which compared the phase diagrams of deuterated and hydrogenous samples for @xmath41 along the @xmath28 axis , saw a much smaller but qualitatively similar difference .
we propose that the smaller @xmath13 for deuterated ndmap is caused by the smaller zero - point motion of deuterons leading to less overlap of the electronic wavefunctions in the exchange paths involving hydrogens . in some classical antiferromagnets , such as the linear - chain compound ( ch@xmath11)@xmath71nmncl@xmath11 ( tmmc ) , deuteration has been reported to reduce the n@xmath15el temperature .
@xcite in the quasi - two - dimensional organic salt @xmath72-(bedt - ttf)@xmath53cu[n(cn)@xmath53]br , substituting deuterium for the hydrogen of the ethylene groups has been shown to drive the system from a superconductor to an antiferromagnetic mott insulator .
@xcite to our knowledge , however , the present work represents the first observation of an effect of deuteration in a haldane - gap antiferromagnet . in summary
, the magnetic phase diagram of ndmap has been extended by specific - heat measurements to 150 mk in temperature and 32 t in the magnetic field .
a new transition has been found at @xmath6 t , when the field is parallel to the spin chains .
further study using techniques other than specific heat is needed to investigate the nature of the new phase , including the possibility that it is an incommensurate phase . in addition , we have determined the power - law exponents for the transition temperatures of the field - induced ordering from the spin - liquid state and compared them with the exponents for tlcucl@xmath11 , ndmaz , and ntenp and with the predictions of the bose - einstein condensation theory .
finally , we have observed an effect of deuterium substitution on the ordering temperature .
the result indicates that the substitution slightly decreases the interchain exchange @xmath13 but hardly affects the in - chain exchange @xmath14 .
we thank s. haas , m. hagiwara , i. harada , m. w. meisel , d. a. micha , y. narumi , j. ribas , t. sakai , d. r. talham , and h. tanaka for useful discussions , and f. c. mcdonald , jr .
, t. p. murphy , e. c. palm , and c. r. rotundu for assistance .
this work was supported by the nsf through dmr-9802050 and the doe under grant no .
de - fg02 - 99er45748 .
a portion of it was performed at the national high magnetic field laboratory , which is supported by nsf cooperative agreement no
. dmr-0084173 and by the state of florida .
kk acknowledges support by a grant - in - aid for scientific research from the japan society for the promotion of science .
o. avenel , j. xu , j. s. xia , m .- f .
xu , b. andraka , t. lang , p. l. moyland , w. ni , p. j. c. signore , c. m. c. m. van woerkens , e. d. adams , g. g. ihas , m. w. meisel , s. e. nagler , n. s. sullivan , y. takano , d. r. talham , t. goto , and n. fujiwara , phys .
b * 46 * , 8655 ( 1992 ) . | we have determined the magnetic phase diagram of the quasi - one - dimensional @xmath0 1 heisenberg antiferromagnet ni(c@xmath1h@xmath2n@xmath3)@xmath3n@xmath4(pf@xmath5 ) by specific heat measurements to 150 mk in temperature and 32 t in the magnetic field . when the field is applied along the spin - chain direction
, a new phase appears at @xmath6 t. for the previously known phases of field - induced order , an accurate determination is made of the power - law exponents of the ordering temperature near the zero - temperature critical field @xmath7 , owing to the four - fold improvement of the minimum temperature over the previous work .
the results are compared with the predictions based on the bose - einstein condensation of triplet excitations . substituting deuterium for hydrogen
is found to slightly reduce the interchain exchange . |
the accelerated mode of expansion of the present day universe encourages many researchers to introduce different kind of sources that is able to explain this . among them
most popular is the dark energy given by a @xmath0 term @xcite , quintessence @xcite , chaplygin gas @xcite .
recently cosmological models with spinor field have been extensively studied by a number of authors in a series of papers @xcite .
the principal motive of the papers @xcite was to find out the regular solutions of the corresponding field equations . in some special cases ,
namely with a cosmological constant ( @xmath0 term ) that plays the role of an additional gravitation field , we indeed find singularity - free solutions .
it was also found that the introduction of nonlinear spinor field results in a rapid growth of the universe .
this allows us to consider the spinor field as a possible candidate to explain the accelerated mode of expansion .
note that similar attempt is made in a recent paper by kremer _ et .
_ @xcite . in this paper
we study the role of a spinor field in generating an accelerated mode of expansion of the universe .
since similar systems , though from different aspects were thoroughly studied in @xcite , to avoid lengthy calculations regarding spinor and scalar fields , we mainly confine ourselves to the study of master equation describing the evolution of bi universe .
we here give the solutions to the spinor and scalar field equations , details of these solutions can be found in @xcite .
we consider a self consistent system of nonlinear spinor and scalar fields within the scope of a bianchi type - i gravitational field filled with a perfect fluid .
the spinor and the scalar field is given by the lagrangian @xmath1 - m{\bar \psi}\psi + f + \frac{1}{2 } ( 1 + \lambda_1 f_1)\,{\varphi}_{,\alpha}{\varphi}^{,\alpha } , \label{lag}\ ] ] where @xmath2 is the coupling constant and @xmath3 and @xmath4 are some arbitrary functions of invariants generated from the real bilinear forms of a spinor field .
here we assume @xmath5 and @xmath6 with @xmath7 , @xmath8 , @xmath9 and @xmath10 .
the gravitational field is chosen in the form @xmath11 where @xmath12 are the functions of @xmath13 only and the speed of light is taken to be unity .
we also define @xmath14 we consider the spinor and scalar field to be space independent . in that case for the spinor and the scalar fields and metric functions we find the following expressions @xcite . for @xmath15
we find @xmath16 with @xmath17 being an integration constant .
the components of the spinor field in this case read @xmath18 with the integration constants obeying @xmath17 as @xmath19 here @xmath20 with @xmath21 .
for @xmath22 in case of massless spinor field we find @xmath23 .
the corresponding components of the spinor field in this case read : with @xmath24 with @xmath25 here @xmath26 with @xmath27 .
for the scalar field we find @xmath28 where @xmath29 and @xmath30 are the integration constants . solving the einstein equation for the metric functions we find @xmath31^{1/3 } \exp \bigl[x_i \int [ \tau ( t')]^{-1}dt ' \bigr ] , \label{a_i}\ ] ] with the integration constants @xmath32 and @xmath33 obeying @xmath34 and @xmath35 .
note that to evaluate the metric functions at any given time @xmath36 we should first integrate @xmath37 , and only then substitute @xmath13 by @xmath36 .
the theoretical arguments @xcite and recent experimental data which support the existence of an anisotropic phase that approaches an isotropic one , led us to consider the models of universe with anisotropic background .
on the other hand the isotropy of the present - day universe lead us to study how the initially anisotropic bi space - time can evolve into an isotropic friedman - robertson - walker ( frw ) one . since for the frw universe @xmath38 , for the bi universe to evolve into a frw one we should set @xmath39 .
moreover , the isotropic nature of the present universe leads to the fact that the three other constants @xmath33 should be close to zero as well , i.e. , @xmath40 , ( @xmath41 ) , so that @xmath42^{-1}dt \to 0 $ ] for @xmath43 ( for @xmath44 with @xmath45 the integral tends to zero as @xmath46 for any @xmath33 ) .
the rapid growth of the universe due to the introduction of the nonlinear spinor field to the system results in the earlier isotropization . as is seen from , , and , the spinor , scalar and metric functions are in some functional dependence of @xmath47 .
it should be noted that besides these , other physical quantities such as spin - current , charge etc . and invariant of space - time
are too expressed via @xmath47 @xcite .
it should be noted that at any space - time points where @xmath48 the spinor , scalar and gravitational fields become infinity , hence the space - time becomes singular at this point @xcite .
so it is very important to study the equation for @xmath47 ( which can be viewed as master equation ) in details , exactly what we shall do in the section to follow . in doing
so we analyze the role of spinor field in the character of evolution .
in this section we study the role of spinor field in the evolution of the universe . but first of all let me qualitatively show the differences that occur at the later stage of expansion depending on how the sources of the gravitational field were introduced in the system . in doing
so we write the einstein equation in the following form : [ ee ] @xmath49 here @xmath0 is the cosmological constant , @xmath50 is the energy - momentum tensor of the source field .
is thoroughly studied in @xcite .
after a little manipulation from one finds the equation for @xmath47 which is indeed the acceleration equation and has the following general form : @xmath51 note also that here a positive @xmath0 corresponds to the universal repulsive force which is often considered as a form of dark energy , while a negative one gives an additional gravitational force .
the bianchi identity @xmath52 in our case gives @xmath53 after a little manipulation from and one finds the following expression for @xmath54 : @xmath55 where the definition of the generalized hubble constant @xmath56 as @xmath57 let us analyze the relation in details .
consider the case when @xmath58 . at the moment when the expansion rate is zero ( it might be at a time prior to the `` big bang '' , or sometimes in the far future when the universe cease to expand we have @xmath59 . )
the nonnegativity of @xmath54 suggests that @xmath60 . before considering the case for large @xmath47
we should like to study the eq . in detail .
for the spinor and scalar fields chosen in this paper they are identically fulfilled .
if this is not the case , an additional equation , know as equation of state , is applied to connect pressure ( @xmath61 ) with energy density ( @xmath54 ) . in the long run from one
finds something like @xmath62 const . , where @xmath63 is some constant ( in case of perfect fluid @xmath64 ) .
thus we see that the energy density of the source field introduced into the system as above decreases with the growth of @xmath47 . now
if we consider the case when @xmath47 is big enough for @xmath54 to be neglected , from we find @xmath65 on account of from one finds @xmath66}. \label{tauhl}\ ] ] from and it follows that for @xmath47 to be infinitely large , @xmath67 in case of @xmath58 we find that beginning from some value of @xmath47 the rate of expansion of the universe becomes trivial , that is the universe does not expand with time . whereas , for @xmath68 the expansion process continues forever .
as far as negative @xmath0 is concerned , its presence imposes some restriction on @xmath47 , namely , there exists some upper limit for @xmath47 ( note that @xmath47 is essentially nonnegative , i.e. bound from below ) .
thus we see that a negative @xmath0 , depending on the choice of parameters can give rise to an oscillatory mode of expansion @xcite .
thus we can conclude the following : _ let @xmath50 be the source of the einstein field equation ; @xmath54 is the energy density and @xmath69 are the principal pressure and @xmath70
. an ever - expanding bi universe may be obtained if and only if the @xmath0 term is positive ( describes a repulsive force and can be viewed as a form of dark energy ) and is introduced into the system as in or if the source field introduced as a part of energy - momentum tensor behaves like a @xmath0 term as @xmath71 .
_ it should be noted that the sources of the gravitational field such as spinor , scalar and electromagnetic fields , perfect or imperfect fluids , as well as dark energy such as quintessence , chaplygin gas are introduced into the system as parts of the total energy - momentum tensor @xmath50 .
it is also known that the dark energy was introduced into the system to explain the late time acceleration of the universe . to show that though the dark energy is introduced into the system as a part of total energy - momentum tensor , it still behaves like a @xmath0 term as @xmath71
, we write them explicitly . the quintessence and chaplygin gas
are given by the following equation of states : [ qc ] @xmath72 , \label{quint}\\ p_{c } & = & -a/{\varepsilon}_{c } , \quad a > 0 .
\label{chap}\end{aligned}\ ] ] note that the energy densities of the quintessence and chaplygin gas are related to @xmath47 as @xcite [ qcs ] @xmath73 , \label{quints}\\ { \varepsilon}_c & = & \sqrt{{\varepsilon}_{0c}/\tau^2 + a } , \quad a > 0 .
\label{chaps}\end{aligned}\ ] ] from and follows that @xmath74 and @xmath75 as @xmath71 . in case of a quintessence , for @xmath76 , both energy density and pressure tend to zero as @xmath47 tends to infinity .
but for @xmath77 ( sometimes known as phantom matter ) we have @xmath78 and @xmath79 as @xmath71 .
it means a quintessence with @xmath77 and chaplygin gas behave like a @xmath80 term when @xmath71 and hence can give rise to an ever expanding universe . before solving the equation for @xmath47 we have to write the components of the energy - momentum tensor of the source fields in details : @xmath81 where , @xmath82 and @xmath83 in @xmath84 and @xmath85 are the energy density and pressure of the perfect fluid , respectively and related by the equation of state @xmath86 . \label{eos}\ ] ] let us now study the equation for @xmath47 in details and clarify the role of material field in the evolution of the universe . for simplicity
we consider the case when both @xmath3 and @xmath4 are the functions of @xmath87 only .
we also set @xmath88 and @xmath89 .
thanks to the spinor field equations and those for the invariants of the bilinear spinor form , the energy - momentum conservation law for the spinor field satisfied identically @xcite . as a result the eq . now reads @xcite @xmath90 in view of from for the energy density and pressure of the perfect fluid one finds @xmath91 further we set @xmath92 .
assume that @xmath93 and @xmath94 where @xmath80 is the self - coupling constant . as it was shown in @xcite , the spinor field equation , more precisely the equations for bilinear spinor forms , in this case
gives @xmath95 .
then setting @xmath89 for the energy density and the pressure from we find @xmath96 \tau^{r-2}}{2(\lambda_1 + \tau^r)^2 } - \frac{\zeta}{\tau^{1+\zeta } } \equiv p.\nonumber\end{aligned}\ ] ] taking into account that @xmath54 and @xmath61 are the functions of @xmath47 only , the eq . can now be presented as @xmath97 where we define @xmath98 where @xmath99 is the set of problem parameters .
the en . allows the following first integral : @xmath100 } \label{1stint}\ ] ] where we denote @xmath101 .
\label{poten}\ ] ] from a mechanical point of view eq . can be interpreted as an equation of motion of a single particle with unit mass under the force @xmath102 . in @xmath103
is the integration constant which can be treated as energy level , and @xmath104 is the potential of the force @xmath105 .
we solve the eq . numerically using runge - kutta method .
the initial value of @xmath47 is taken to be a reasonably small one , while the corresponding first derivative @xmath106 is evaluated from for a given @xmath103 .
let us go back to the eq . .
in view of one sees , @xmath107 as @xmath71 , i.e. , if @xmath108 is considered to be the acceleration of the bi universe , then the massive spinor field essentially can be viewed as a source for ever lasting acceleration .
note that it does not contradicts our previous statement about the role of energy - momentum tensor on ever expanding universe , since the spinor field satisfies the bianchi identity identically .
now a few words about considering @xmath108 as acceleration .
the einstein equations for the frw model read [ frw ] @xmath109 from one finds @xmath110 the equation is known as the acceleration equation . in analogy for the bi universe from we can write @xmath111 and declare it as acceleration equation . though setting @xmath112 we recover the original definition , hardly it will be helpful in our case .
so in bi universe we assume @xmath108 be the acceleration and eq .
be the acceleration equation .
let us now define the deceleration parameter .
in frw cosmology the deceleration parameter has the form @xmath113 = \frac{d}{dt}\bigl(\frac{1}{h_{\rm frw}}\bigr ) -1 , \label{decprfrw}\ ] ] where @xmath114 is the hubble parameter for frw model .
in analogy we can define a deceleration parameter as well .
if we define the generalized deceleration parameter in the following way : @xmath115 , \label{decpo}\ ] ] where @xmath116 , then the standard deceleration parameter is recovered at @xmath112 .
but is this case the definition for acceleration adopted here is no longer valid .
so we switch to the second choice and following belinchon and harko _ et .
@xcite define the generalized deceleration parameter as @xmath117 after a little manipulation in view of and the deceleration parameter can be presented as @xmath118 let us now go back to the equations , , and . as one sees , the positivity of the radical imposes some restriction on the value of @xmath47 , namely in case of @xmath119 and @xmath120 the value of @xmath47 can not be too close to zero at any space - time point .
it is clearly seen from the graphical view of the potential [ cf .
[ potp ] ] .
thus we can conclude that for some special choice of problem parameters the introduction of nonlinear spinor field given by a self - action provides singularity - free solutions . as it was shown in @xcite the regular solution
is obtained only at the expense of broken dominant - energy condition in the hawking - penrose theorem .
_ if , in an eigentetrad of @xmath121 , @xmath122 denotes the energy density and @xmath123 denote the three principal pressure , then the dominant energy condition can be written as _
@xcite : @xmath124 the dominant energy condition for the bi metric can be written in the form : @xmath125 in fig .
[ potm ] we plot the potential for a negative @xmath80 .
as one sees , in the vicinity of @xmath48 there exists a bottomless potential hole . as one sees ,
if in case of a self - action the initial value of @xmath47 is too close to zero and the constant @xmath103 is less than @xmath126 ( the maximum value of the potential in presence of a self - action ) , the universe will never come out of the hole . for numerical solutions we set @xmath127 , spinor mass @xmath128 , the power of nonlinearity we choose as @xmath129 , @xmath130 and for perfect fluid we set @xmath131 that corresponds to a radiation
. we also set @xmath132 and @xmath133 .
the initial value of @xmath47 is taken to be @xmath134 .
the coupling constant is chosen to be @xmath135 , while the self coupling constant is taken to be either @xmath136 or @xmath137 . here , in the figures we use the following notations
: + 1 corresponds to the case with self - action and interaction ; + 2 corresponds to the case with self - action only ; + 3 corresponds to the case with interaction only . 1 cm as one sees from fig .
[ potp ] , in presence of a self - action of the spinor field with a positive @xmath80 , there occurs an infinitely high barrier as @xmath138 , it means that in the case considered here @xmath47 can not be trivial [ if treated classically , the universe can not approach to a point unless it stays at an infinitely high energy level ] .
thus , the nonlinearity of the spinor field provided by the self - action generates singularity - free evolution of the universe .
but , as was already mentioned , this regularity can be achieved only at the expense of dominant energy condition in hawking - penrose theorem .
it is also clear that if the nonlinearity is induced by a scalar field , @xmath47 may be trivial as well , thus giving rise to space - time singularity .
note that cases in presence of a @xmath0 term are thoroughly studied in @xcite .
it was shown that introduction of a positive @xmath0 just accelerates the speed of expansion , whereas , a negative @xmath0 depending of the choice of @xmath103 generates oscillatory or non - periodic mode of evolution .
note also that the regular solution obtained my means of a negative @xmath0 in case of interaction does not result in broken dominant energy condition @xcite . in figs .
[ epp ] and [ epm ] we plot the corresponding energy density and pressure .
as one sees , in case of a positive @xmath80 the energy density is initially negative while the pressure is positive . in this case
though the solution is singularity - free , the violation of dominant energy takes place . in case of a negative @xmath80
the pressure is always negative .
the purpose of plotting the energy density and pressure is to show that the energy density of the source field indeed decreases with the increase of the universe .
this also shows that there exists an interval where the energy density of the system with spinor field nonlinearity generated by the self - action is negative .
this is in line with our earlier assumption .
moreover , we see the pressure of the source field becomes negative in course of evolution ( in case of self - action with a positive @xmath80 pressure is initially positive , but with the expansion of the universe it becomes negative , whereas , in case of of a negative @xmath80 as well as in case of interacting fields the pressure is always negative ) . recall that the dark energy ( e.g. quintessence , chaplygin gas ) , modelled to explain the late time acceleration of the universe , has the negative pressure .
so we argue that the models with nonlinear spinor field and interacting spinor and scalar fields to some extent can be considered as an alternative to dark energy which is able to explain the late time acceleration of the universe . in the figs .
[ acp ] and [ acm ] we illustrate the acceleration of the universe for positive and negative @xmath80 , respectively
. as one sees , in both cases we have decreasing acceleration that tends to @xmath139 as @xmath71 . in figs .
[ dpp ] and [ dpm ] we plot the deceleration parameter . the figs .
[ acp ] , [ acm ] , [ dpp ] and [ dpm ] show the accelerated mode of expansion of the universe .
as one sees , the acceleration is decreasing with time .
depending of the choice of nonlinearity it undergoes an initial deceleration phase .
it is also seen that the nonlinear term plays proactive role at the initial stage while at the later stage spinor mass is crucial for the accelerated mode of expansion .
we considered a system of interaction nonlinear spinor and scalar fields within the scope of a bi cosmological model filled with perfect fluid .
the spinor field nonlinearity gives rise to an effective negative pressure in the course of evolution .
comparison of the effective pressure of the nonlinear spinor field with that of a dark energy given by a quintessence or chaplygin gas leads us to conclude that the spinor field can be seen as an alternative to the dark energy able to explain the acceleration of the universe .
it was shown that the nonlinear spinor term is proactive at the early stage of the evolution and essentially accelerates the process of evolution , while at the later stage of evolution the spinor mass holds the key .
given the fact that _ neutrino _ is described by the spinor field equation and it too possesses mass ( though too small but nonzero ) , the presence of huge number of neutrino in the universe can be seen as one of the possible factor of the late time acceleration of the universe .
it was also shown that for some specific choice of parameters it is possible to construct singularity - free model of the universe , but this regularity results in the broken dominant energy condition of the hawking - penrose theorem . | a self - consistent system of interacting nonlinear spinor and scalar fields within the scope of a bianchi type - i cosmological model filled with perfect fluid is considered .
exact self - consistent solutions to the corresponding field equations are obtained .
the role of spinor field in the evolution of the universe is studied .
it is shown that the spinor field gives rise to an accelerated mode of expansion of the universe . at the early stage of evolution the spinor field nonlinearity
generates the acceleration while at the later stage it is done by the nonzero spinor mass . |
the discovery of new spin - orbit related phenomena and states of matter lead to an immense increase of interest in materials containing @xmath0 elements , ir and os in particular .
@xcite among the new effects , @xcite recently proposed a possibility of excitonic condensation in @xmath1 materials with cubic local symmetry .
the singlet atomic ground state promises no interesting low - energy physics .
however , if the energy of the lowest excitation , a magnetic moment carrying triplet , is sufficiently small , the inter - atomic exchange due to electron hopping may lead to formation of an ordered state - excitonic magnet .
thanks to the singlet - triplet exchange processes , the excitations can be viewed as mobile quasi - particles propagating on the singlet background with a non - trivial dispersion .
if the the bottom of this dispersion touches the singlet level a phase transition ( bose - einstein condensation ) takes place . in some systems pairwise creation and annihilation of s on the neighboring atoms
plays a role similar to the hopping .
realization of the s proposal @xcite is an interesting challenge since the spin - orbit coupling must be strong enough to enhance the correlation physics , but not too strong to allow sufficiently small singlet - triplet gap .
recently , a curious magnetic behavior was observed in ( syio ) , which was interpreted as excitonic magnetism .
@xcite in particular , it was argued that non - cubic crystal field splits the triplet state and thus reduces the first excitation energy of the ir ion .
subsequent band structure calculations @xcite found only moderate non - cubic crystal field on ir site and found that the physics of syio is similar to its cubic analog ( byio ) .
the gga+u calculations to a magnetically ordered ground state , which the authors interpreted as confirmation of the experimental data . in article we combine several methods to analyse the physics of byio and syio .
our results contradict the conclusions of refs . as we find byio and syio to be insulators with singlet local ground state .
we show that using the same gga+u approach as in ref .
a solution corresponding to the singlet ground state has a lower energy than the solution containing magnetic moments .
this result is supported by dynamical mean - field theory ( dmft ) calculations performed with full interaction vertex .
we conclude with a sufficient margin for uncertainties of various model parameters that the inter - atomic exchange processes in the studied double perovskites are too weak to overcome the singlet - triplet splitting due to the spin - orbit coupling . this result can be put to the perspective of the recent observation of signatures of excitonic magnetism in ca@xmath2ruo@xmath3 @xcite , a @xmath4 perovskite where spin - orbit coupling is weaker and
ru - ru hopping is stronger than in ir double perovskites .
the paper is organized as follows . after introducing the computational methods in sec .
[ sec : comp ] we discuss the bonding in double perovskite structure and construct an effective hubbard model on the basis of band structure calculations ( sec .
[ sec : gga ] ) . to compare with previous studies we have performed hartree - like gga+u calculations reported in sec .
[ sec : ldau ] . in the rest of the paper
we analyze the multi - band hubbard model .
first , we study the single - site problem with the focus on singlet - triplet splitting and van vleck susceptibility using exact diagonalization ( sec . [
sec : ed ] ) .
next , we calculate the one - particle properties of the lattice problem using dmft ( sec . [ sec : dmft ] ) and the dynamic of two - particle excitations in the strong - coupling expansion ( sec . [
sec : t2u ] ) .
in this work we combine several numerical and semi - analytical approaches . all the below described calculations
were performed for cubic byio .
the key steps were performed also for orthorhombic syio .
we start with gga @xcite density functional calculations employing the wien2k @xcite package .
all the results reported here were obtained with the spin - orbit coupling included @xcite .
the crystallographic as well as other parameters used in the calculations are summarized in supplemental material ( sm ) @xcite . to include the correlation effects on the ir site beyond the effective non - interacting picture of gga we follow two paths .
first , we use the static mean - field approach of gga+u . since this approach is known to often lead to multiple solutions we have run series of calculations for various initial conditions .
in particular , we used the spin - polarized gga solution and atomic ground state as starting points .
the calculations were performed for interaction parameters @xmath5 of 2 , 4 and 6 ev and @xmath6 from 0.1 to 0.7 ev . the so called fully localized limit double counting correction was used @xcite .
although it should be a standard for systems with strong spin - orbit coupling , we note explicitly that spin - off - diagonal terms of the +u orbital potential were included in the calculation .
second , we use the non - spin - polarized gga band structures to construct effective hubbard models . employing wien2wannier @xcite and wannier90 @xcite codes to construct the wannier representation of the gga bands we can extract the crystal - field @xcite and spin - orbit parameters as well as the inter - atomic hoppings . for the calculation of the partially screened coulomb interaction
, we used the density response code for elk @xcite . in the calculation of the constrained susceptibility , we excluded the contribution of particle - hole excitations within the target bands ( the @xmath7 bands for the @xmath8-model , and @xmath7 and o @xmath9 bands for the @xmath10 model ) and took 80 unoccupied bands .
we used 4@xmath114@xmath114 @xmath12 and @xmath13 meshes for the double fourier transform with the cutoff @xmath14=1 , 2 , 3.5 ( 1/a.u . ) , where @xmath15 is the reciprocal vector , and then extrapolated the result to @xmath16 .
we study two types of model hamiltonians : ( i ) @xmath8-model in the space of ir - like @xmath7 bands and ( ii ) @xmath10-model covering the space of ir-@xmath8 and o-@xmath9 bands .
the two models were expressed in two bases : ( a ) real cubic harmonics with sharp spin and ( b ) the eigenbasis of the local one - particle hamiltonian with doublet and quadruplet ( @xmath17-basis ) .
the two bases are equivalent and their use followed numerical convenience .
we use both @xmath8- and @xmath10-model primarily to demonstrate robustness of our results .
the questions of @xmath8 vs @xmath10 model was discussed by several authors * ? ? ? * in the context of oxides of @xmath18 elements .
the basic difference between the two models consists in how they describe low - energy one - particle excitations , whose orbital character is a mixture of atomic - like ir-@xmath8 and o-@xmath9 orbitals . in @xmath8-model
these excitations are molecular orbitals ( anti - bonding @xmath10 combinations ) whose structure is determined solely by the @xmath10 hybridization and does not depend on the interaction strength . in the @xmath10-model the orbital structure of the excitations
is affected by the interaction and can evolve from the non - interacting molecular orbital limit to the heitler - london limit where the double occupancy of the atomic orbitals is suppressed . while the @xmath10-model , which can describe richer physics , appears superior to the @xmath8-model , in practice it is not necessarily so .
larger hilbert space of the @xmath10-model involves more uncertainties in its construction , the most prominent of which is the so called double - counting correction which is directly related to the @xmath10 charge transfer energy . the basic information about the models is obtained by analyzing the local electronic structure .
we performed exact diagonalization of the single - site @xmath8-model where the six @xmath7 wannier orbitals are treated as the atomic @xmath8 orbitals , and iterative lanczos diagonalization of the iro@xmath19 cluster where the complete ir @xmath8 shell is hybridized with the o @xmath9 states .
this step served to determine the singlet - triplet splitting and to assess the differences between the two models , mainly the consequences of the explicit inclusion of the @xmath10 hybridization in the cluster model . to extend the many - body calculations on the lattice we have run dmft simulations for both the @xmath8 and the @xmath10 model with the interaction parameters obtained in the crpa calculations . to this end
we used our implementation of the self - consistent cycle @xcite and the triqs hybridization - expansion continuous - time quantum monte carlo solver @xcite ( based on the triqs library @xcite ) .
the calculations , with full coulomb vertex @xmath20 , were performed in - basis ( @xmath17-basis ) in which the on - site one - particle hamiltonian is diagonal .
this choice of basis helps to avoid the monte carlo sign problem .
@xcite the reported calculations were performed at the temperature of 290 k. the spectral functions were obtained by analytic continuation of the self - energy using the maximum entropy method @xcite .
while our dmft calculations were limited to studying one - particle dynamics , we use large-@xmath5 expansion of the @xmath8-model to investigate the two - particle excitations .
we have eliminated the charge fluctuations by the schrieffer - wolff transformation @xcite to the second order and constructed an effective hamiltonian on a local hilbert space spanned by the singlet and lowest triplet states .
the dynamics of this effective model was studied in the linear spin wave approximation . to check the consistency of our results
we have started from two different one - particle bases : the real cubic harmonics and @xmath17-basis .
first , we report the unpolarized gga solution for byio , which served as a starting point to construct the effective models .
the results for lower - symmetry syio are similar . unlike in simple perovskites , where the o@xmath21 octahedra around neighboring metal atoms share corners , in double perovskites the iro@xmath21 octahedra are not touching .
this geometry suggests that the double perovskites can be possibly viewed as molecular crystals built out of iro@xmath21 molecules plus some filler atoms . irrespective of whether this picture is useful for interpretation of the physics
, one can use molecular orbitals ( mos ) - linear combinations of atom - centered wannier functions ( wfs ) on the o@xmath21 octahedron - as an alternative to the atom - centered wf basis . in fig .
[ fig : mo ] we show the gga spectral density decomposed into the projections on the mos classified by their symmetry .
first , we note the magnitude of the @xmath10 hybridization .
nominally ir-@xmath8 orbitals around fermi level are superpositions of the atomic ir-@xmath8 orbital and its mo partner . for states
the mixture is essentially fifty - fifty , while for and @xmath22 states the ir-@xmath8 weight is somewhat larger .
we further observe that bands of distinct symmetry exhibit little hybridization .
the filled non - bonding mos therefore do not play an active role in the low - energy physics .
the gga+so spectral density of byio projected on the irreducible representations of the cubic symmetry in the space of @xmath10 wannier functions .
top panel : projection on wfs centered on the ir atom ( the @xmath7 functions were further resolved into the and states ) .
bottom panel : projection on wfs centered on the o atoms .
the @xmath23 , @xmath22 and @xmath24 mos are built from atomic o-@xmath9 orbitals pointing to the central ir atom , while the remaining mos are built from atomic o-@xmath9 orbitals perpendicular to the ir - o bond . ] in fig .
[ fig : bands ] it is demonstrated that they can be dropped from the model . we compare the gga band structure with the bands of a truncated @xmath10 model where only ir and o@xmath21 @xmath7 orbitals were kept , while the @xmath22 and non - bonding orbitals were simply left out without any adjustment of the @xmath7 subspace .
the model dispersion reproduces very well the anti - bonding bands around the fermi level and does a fairly good job also for the bonding bands which overlap with the @xmath22 and @xmath23 bands .
this suggests that the molecular crystal provides a useful picture of the studied double perovskites . the large bonding
anti - bonding splitting allows straightforward construction of the wannier representation of the anti - bonding bands only - the @xmath8 model . the gga+so band structure of byio ( black ) .
the bands of the @xmath10 model with @xmath7 orbitals described in the text ( red ) . ]
the crpa calculations were performed for @xmath8 and @xmath10 models .
the calculated direct and exchange interaction parameters in the low frequency limit can be found in sm . in our model calculations
we make the common assumption of atomic form of the interaction , which is then parametrized by @xmath5 and @xmath6 which are chosen to match the crpa data . for the @xmath8-model we can fairly accurately fit the crpa data with the values @xmath5=1.8 ev and @xmath6=0.4 ev , while we get @xmath5=4 ev and @xmath6=1.1 ev for the @xmath10-model .
the values of @xmath6 should be viewed as upper estimates obtained by the extrapolation of the fourier cut - off to infinity .
our main conclusions in this work hold for smaller @xmath6 , but eventually break for large @xmath6 .
therefore we try to use conservative ( upper limit ) values in our calculations . allowing ferromagnetic spin - polarized solutions we arrive at similar results as reported in ref .
. however , these solutions are metallic , contrary to experiment , suggesting that inclusion of electronic correlations beyond gga is desirable .
the gga+u approach provides a computationally cheap way to include the electronic correlations on a hatree - fock level .
the gga+u method is formally analogous to the weiss mean - field theory and as such it is prone to multiple self - consistent solutions that could be thermodynamically unstable .
the authors of ref . started from their spin polarized solutions for byio and syio and including the + u orbital potential they found these solutions to be modified , but qualitatively unchanged . while we were able to find similar spin - polarized solutions ( sm )
, we extended the search for possible stable solutions by starting from the occupation matrices corresponding to the atomic ground state reported below .
this way we obtained non - magnetic solutions with energies lower than their spin polarized counterparts .
the spectral density of the non - magnetic solution for byio , shown in fig .
[ fig : ldau ] , confirms the picture of a local singlet ground state with filled orbitals and empty ones . the ir @xmath7 spectral densities for the non - magnetic gga+u solutions obtained with @xmath25 ev and two values of @xmath5 .
] in the following , we present the results obtained for the @xmath8 and @xmath10 models of syio and byio .
the parameters of the models obtained using the wannier projection can be found in sm .
we start by discussing the physics of an isolated lattice site .
the key quantity of interest is the singlet - triplet excitation energy @xmath26 . in fig .
[ fig : spec ] we show the single - ion spectra of byio and syio of the @xmath8-model obtained by diagonalization of the single - site hamiltonian with 4 electrons . similar to ref .
we find that the non - cubic crystal - field in syio has only minor effect and the spectrum is controlled by the spin - orbit coupling and hund s exchange @xmath6 ( note that the isotropic part @xmath5 of the on - site interaction does not affect level splitting within a given valence state . ) since the singlet - triplet splitting decreases with @xmath6 , the excitation energy @xmath26 of 270 mev is a conservative lower estimate .
[ cols= " < , > " , ] besides the multiplet structure we have calculated the van vleck susceptibility of the ir atom in the @xmath8-model @xcite of syio .
the susceptibility increases with @xmath6 and it exhibits an anisotropy : for @xmath6=0.4 ev its values along a , b and c axes are 0.0011 , 0.0016 and 0.0009 emu / mol ( the corresponding values for @xmath6=0.2 ev are 0.0009 , 0.0009 and 0.0008 emu / mole ) .
these numbers correspond rather well to the experimental data of ref .
which exhibit only a weak temperature dependence above 50 k ( @xmath270.0015 emu / mole@50 k and @xmath270.0011 emu / mole@330 k ) .
next , we study the local physics of byio within the @xmath10-model , i.e. we diagonalize the iro@xmath21 cluster .
we treat the o-@xmath9 states as non - interacting and put the interaction on the central ir atom only . unlike for the isolated @xmath8-shell above , the isotropic part of the interaction @xmath5 affects the results strongly .
the dependence of the excitation spectrum on @xmath5 is , however , to a large extent compensated by the double - counting correction @xmath28 which renormalizes the splitting between the @xmath8 and @xmath9 orbital energies and which must be included to take into account the interaction energy present in gga .
since @xmath28 is not rigorously defined , we treat it as an adjustable parameter . in fig .
[ fig : spec ] we show the singlet - triplet excitation energy @xmath26 for the iro@xmath21 cluster . rather than @xmath28
, we plot our results as a function of the closely related ir-@xmath8 occupancy @xmath29 with clear physical meaning . for fixed @xmath29
the excitation energy @xmath26 is practically independent of @xmath5 .
we take the ir-@xmath8 occupancies between @xmath30 ( @xmath31 is the gga value ) to be physically reasonable .
the corresponding singlet - triplet splitting falls into the interval 190 - 280 mev and is consistent with 270 mev obtained from the @xmath8-model . as a side remark
we point out that a popular approximation of the interaction to the density - density terms in the @xmath17-basis leads to the substantially smaller values of singlet - triplet splitting of around 100 mev .
this can be traced back to the missing pair - hopping terms which contribute to lowering of the singlet energy . in the following ,
we extend our analysis to the lattice problem .
we address three questions .
is there a charge gap in the one - particle spectrum and what is its nature ?
it there a global ground state with broken symmetry predicted by ref . ?
what is the dynamics of two - particle excitations and can they condense in the sense proposed by @xcite ?
the first and , to some extent , the second questions are studied using dmft . the second and the third questions
are addressed using the effective strong coupling hamiltonian .
the dmft technique incorporates local quantum and thermal fluctuations and does not typically suffer from the problem of multiple solutions with distinct local states .
when several local states are quasi - degenerate the method can naturally describe their statistical mixture ( most common example being local moment fluctuations ) or even their condensate @xcite .
moreover , modern dmft implementations can treat full coulomb vertex without approximations . in fig .
[ fig : dmft ] we show the dmft one - particle spectra of the @xmath8 model .
the dominant correlation effect is the static hartree shift of the empty band with dominant character with respect to the occupied band dominated by the character .
the dynamical effects , arising from the pair - hopping terms in the interaction play only a minor role .
the picture provided by dmft is therefore consistent with the gga+u non - magnetic solution . the one - particle spectral density for the @xmath8-model of byio along high symmetry directions in the brillouin zone obtained with dmft .
the side panel shows the k - integrated density resolved into ( red ) and ( black ) orbital contributions .
the calculation were performed with @xmath321.8 ev , @xmath330.4 ev at the temperature of 290 k. ] in order to assess the possibility of excitonic condensation we have to investigate the two - particle dynamics . while it can , in principle , be studied by dmft , in practice
this is a difficult task which has been executed only for simple models or involving approximation .
@xcite moreover , the results so far suggest that byio ( and syio ) can be viewed as strongly coupled systems with suppressed charge fluctuations .
an effective strong coupling model provides a much more accessible description of the physics . in the following ,
we derive such a model and solve its dynamics using the linear spin - wave theory . the model in the space spanned by the singlet and triplet states is obtained by second - order schrieffer - wolff transformation of the @xmath8 model .
the classical ground state is a product of singlets .
the excitations in the limit of low -density are described by an effective hamiltonian describing propagation of a single @xmath34-excitation , i.e. , we neglect for example interactions between @xmath34s , @xmath35 here , @xmath36 is the singlet - triplet splitting @xmath26 renormalised by hopping processes of the type @xmath37 and @xmath38 , @xmath39 is the amplitude of the process @xmath40 and @xmath41 is the amplitude of the process @xmath42 . hopping to the first and second neighbors on the fcc lattice is considered . in fig .
[ fig : lsw ] we present the spectrum of ( [ eq : spw ] ) along the high symmetry directions of the cubic lattice .
the main observation is the large excitation gap . in the language of the s proposal the excitonic condensation takes place when the energy of the lowest triplet excitation touches the singlet ( vacuum ) level .
our results show that the width of the @xmath34-dispersion is at least an order of magnitude too small for that to happen .
since the parameters of ( [ eq : spw ] ) for syio are of similar magnitude as for byio we conclude that our result holds for syio as well .
dispersion of a triplet excitation in singlet background obtained from @xmath8-model with @xmath43 ev and @xmath44 ev . ]
the double perovskite structure with large ir - ir distance and small hopping amplitude @xmath34 is not very promising for observation of excitonic magnetism .
the electronic bandwidth @xmath45 does not have to be unusually small thanks to the high number of nearest neighbors @xmath46 .
the effect of small @xmath34 is much more pronounced in the dispersion of two - particle excitations with a bandwidth @xmath47 which scales as @xmath48 for fixed @xmath49 .
we have combined several theoretical approaches to study the low - energy properties of ir double perovskites byio and syio .
the two compounds are quite similar , in particular , the non - cubic crystal field of syio is found to be rather weak , a result that agrees with the conclusions of theoretical study of bhowal _ et al . _
our main conclusion is that excitonic magnetism in byio and syio is not favored .
small ir - ir hopping in double perovskite structure gives rise to valence and conduction bands that are sufficiently narrow to become orbitally polarized and gapped by a moderate on - site coulomb repulsion .
another consequence of the small electron hopping are small amplitudes for propagation of two - particle excitations .
the bandwidth of such excitations is an order of magnitude smaller than the singlet - triplet excitation energy determined by the spin - orbit coupling and hund s exchange @xmath6 , which prevents the possibility of their exciton condensation .
our conclusions contradict the conclusions of the experimental study @xcite . while our calculated van vleck susceptibility reproduces quite well the experimental susceptibility of ref . above 50 k
, we do not find any tendency towards a magnetic phase transition at low temperatures .
instead , we predict a large spin gap in the 200 mev range .
our results thus call for reexamination of the experimental data considering the possibility of an extrinsic origin of the observed low temperature behavior .
we acknowledge discussions with z. jirk , k. knek and j. hejtmnek .
we acknowledge support of the czech science foundation through the project 13 - 25251s ( jku and pn ) , german research foundation through for1346 ( vp ) and grants - in - aids for scientific research on innovative areas ( 15h05883 ) from jsps ( ra ) .
access to computing and storage facilities owned by parties and projects contributing to the national grid infrastructure metacentrum , provided under the programme `` projects of large infrastructure for research , development , and innovations '' ( lm2010005 ) , is greatly appreciated . | we combine several numerical and semi - analytical methods to study the @xmath0 double perovskites and which were recently proposed to exhibit excitonic magnetism . starting from the density functional theory and constrained random phase approximation we construct effective multi - band hubbard models .
these are analyzed by means of static and dynamical mean - field theories and strong coupling expansion .
we find both materials to be insulators , but , contrary to the experimental claims , with a large spin gap of several hundreds mev preventing formation of an ordered state at low temperature . |
several condensed matter systems have been discovered in recent years whose active electrons have their kinematics governed by the dirac equation @xcite , rather than by the schrdinger equation .
their dispersion relation , accordingly , has the same form as that of a relativistic particle .
the reason for this unusual behavior of electrons whose speed is at least two orders of magnitude less than the speed of light can be ascribed to a particular influence of the lattice background on the electronic properties . for many technological applications
, it would be interesting if those systems were a semiconductor , instead of a semimetal .
therefore , the possibility of an insulator state or the appearance of an excitonic gap due to short range electronic interactions in a system of massless dirac fermions has been the object of intense investigation @xcite .
nonetheless , early proposals that electron - electron interactions could generate an electronic gap were investigated @xcite , where the gap opening is an analogue of the `` chiral symmetry '' breaking process that occurs in massless quantum electrodynamics ( qed ) in two dimensions @xcite .
the excitonic interaction can be generated by different physical mechanisms .
the coulomb interaction , for instance , can generate it , precisely as it occurs in the case of the hubbard model , where the on - site density - density repulsion term can be derived from the static coulomb interaction .
this is the type of excitonic instability that we consider in the present paper .
the fact that some of those condensed matter systems become superconductors upon doping provides a vast phenomenological motivation for the investigation of the superconducting phase diagram of dirac electrons systems . condensed matter systems with such competing interactions display phenomena that are analogous to the onset of the chiral condensate and of color superconductivity in dense quark matter , which is an active subject of investigation @xcite .
the phase transitions obtained in such systems can be compared with the ones occurring in analog condensed matter systems with dirac fermions .
consider a system presenting a bipartite lattice formed of sublattices a and b , such as in graphene @xcite or in the pnictides @xcite . assuming that the interaction has two terms leading , respectively , to the superconducting and the excitonic instabilities , our model hamiltonian in the continuum limit reads @xcite , @xmath0 where @xmath1 and @xmath2 denote the dirac points and the fermi velocity respectively , @xmath3 , @xmath4 , which is written in terms of the electron creation operators of sublattices @xmath5 and @xmath6 , and the matrix @xmath7 is given by @xmath8 with @xmath9 and @xmath10 denoting the excitonic and the superconducting order parameters .
the effective potential ( free energy ) per dirac point reads @xmath11 } { \prod _ { j = 1 } ^ { 4 } \left [ i \omega_n - e_j ( \sigma = 0 , \delta = 0 ) \right ] } \right\ } \label{eqveff } , \end{aligned}\ ] ] where @xmath12 is the matsubara frequency for fermions ( with the boltzmann constant @xmath13 ) and the four @xmath14 s are @xmath15 ^ 2 } } \label{ej } \ , .\ ] ] notice that eq .
( [ ej ] ) is the typical dispersion relation observed in effective models for quantum chromodynamics , where @xmath16 corresponds to the energy required to create a particle or a hole above the fermi surface and @xmath17 is the corresponding antiparticle term @xcite .
moreover , the effective potential in eq .
( [ eqveff ] ) is the same obtained in @xcite , where the formation of chiral and diquark condensates involving two quark flavours in qcd was investigated . in their case , the values for the quark - antiquark and quark - quark interactions strengths were set equal by construction , via a fierz transformation applied to the nambu - jona - lasinio model , which is not the case in the present paper , since @xmath18 and @xmath19 are taken to be free parameters .
the conditions for the appearance of superconductivity or excitonic condensate are given by the nonzero solutions of @xmath10 or @xmath20 which minimize the free energy , those have been investigated already at zero temperature @xcite .
we have shown that , even if the excitonic interaction strength is larger than the superconducting interaction , as the chemical potential increases , superconductivity eventually suppresses the excitonic order parameter , which means that the system goes from an insulating state to a superconducting one as charge carriers are added to the system .
remarkably , similar results were obtained for a system for two - color and two - flavor qcd , where chiral and diquark condensates were calculated @xcite .
the model containing only the superconducting interaction at @xmath21 has been previously investigated .
we have shown that there is a threshold for the superconducting interaction strength which separates the normal and the superconducting states @xcite , which is quantum critical point at @xmath21 , @xmath22 .
this critical interaction is expressed in terms of the ultraviolet cutoff @xmath23 , namely , @xmath24
@xcite , where the fermi velocity should be known for the actual physical system under investigation .
therefore , the single free parameter of our model is @xmath23 and the numerical results presented below are given in terms of @xmath23 and the dimensionless strength @xmath25 .
nonetheless , notice that the cutoff might be adjusted _ a posteriori _ through experimental data , even if @xmath23 is always provided by the lattice in condensed matter systems . | we investigate the competition between the superconducting and the excitonic phases on dirac electrons on a bipartite planar lattice .
the conditions for the appearance of superconductivity or excitonic condensate are given by the minima of the free energy and our results are explained .
address = departamento de cincias naturais , universidade federal de so joo del rei , 36301 - 000 so joo del rei , mg , brazil address = departamento de cincias naturais , universidade federal de so joo del rei , 36301 - 000 so joo del rei , mg , brazil address = instituto de fsica , universidade federal do rio de janeiro , caixa postal 68528 , rio de janeiro , rj , 21941 - 972 , brazil |
today it is widely accepted that the formation of cosmic relaxed objects proceeds via gravitational instability from small to large scales .
this clustering process can be studied in detail , following accurately the dynamics of collapse by means of n - body simulations , and statistically , by approximate semi - analytical models which enable a swifter and more efficient exploration of the parameter space .
the most complete semi - analytical models developed to date are those by lacey & cole ( 1993 , 1994 , hereafter lc93 and lc94 ) , and by manrique & salvador - sol ( 1995 , 1996 , papers i and ii ) .
all the quantities in both models are derived from the statistics of the filtered random gaussian density field , @xmath0/\langle\rho\rangle$ ] , of cold matter encountered at an arbitrary initial epoch , @xmath1 , after recombination when fluctuations are still linear .
similarly , in each model the dynamics of dissipationless collapse is approximated by the simple spherical model .
these two models differ however in the following aspects .
lacey & cole s ( lc ) model is based on the press & schechter ( 1974 , ps ) heuristic prescription for the mass functions of halos , duly extended to obtain the conditional mass function of halos at a given epoch subject to having a larger mass at a later time ( bower 1991 ; bond et al .
@xmath2-body simulations show that this model gives highly satisfactory predictions ( lc94 ) . in spite of this remarkable success
, there remains the caveat that the seeds of halos are arbitrary points and their associated collapsing clouds undetermined ( possibly disconnected ) regions around them ( cf . lc93 ) . besides
, the formation and destruction of halos are arbitrarily and ambiguously defined , leading to poorly motivated formation and destruction times ( see paper ii ) .
the model developed by manrique & salvador - sol uses the new cusp ( confluent system of peak trajectories ) formalism , which enables one to follow the _ filtering _ evolution of peaks in the density contrast vs. filtering scale ( @xmath3 vs. @xmath4 ) diagram at @xmath1 , assumed to trace the mass evolution of halos .
this model is better founded than the lc model insofar as high peaks are good seeds of massive halos ( bond & myers 1996 ) and do follow spherical collapse ( bernardeau 1994 ) . on the other hand
, it provides a natural distinction between accretion and merger which makes the formation and destruction of dark halos well - defined and uniquely determined .
as shown in papers i and ii both models provide very similar results for the mass function , the mass increase rate , and the merger rate at large captured masses .
however , notable departures were observed in this latter rate at intermediate captured masses .
this discrepancy was suspected as being caused by the rather crude approximation used in the cusp model for the density of nested peaks , in connection with the cloud - in - cloud correction . in the present paper
we seek to verify this point .
we improve the nesting correction by taking into account the correlation among peaks and investigate the effects this has on the cusp model . in 2 we review the foundations of the cusp model developed in papers i and ii . in 3 we derive a more accurate approximation for the density of nested peaks and implement it in 4 .
our results are summarized in 5 .
comoving lengths are assumed throughout the paper .
based on the spherical collapse model , the peak ansatz states that there is a correspondence between non - nested peaks of fixed linear overdensity @xmath5 in the density field smoothed on scale @xmath4 at any arbitrary initial epoch @xmath1 and quasi - steady halos of mass @xmath6 at a time @xmath7 .
the mass @xmath6 of the spherical collapsing cloud , or equivalently , of the final halo associated with a peak is an increasing function of the filtering scale @xmath4 , whereas @xmath5 is a decreasing function of @xmath8 .
the mass function of halos at @xmath8 , @xmath9 , is therefore simply given by the density of peaks with @xmath5 on scales @xmath4 to @xmath10 , @xmath11 , properly corrected for nesting and transformed to the variables @xmath6 and @xmath8 through the relations @xmath12 and @xmath13 . the approximate expression for the nesting correction used in paper i was ( see next section for justification ) @xmath14 in equation ( [ 1 ] ) , @xmath15 is the mean density of the universe , @xmath16 is the volume , to 0th order in @xmath3 , of the collapsing cloud associated with a peak with @xmath3 on scale @xmath17 , and @xmath18 is the conditional density of peaks with @xmath3 on scales @xmath4 to @xmath10 subject to being located at a point with @xmath3 on scale @xmath19 . this equation is a volterra - type integral equation of the second kind for the scale function @xmath20 of non - nested peaks , which can be readily solved in the standard way from the known functions @xmath21 and @xmath22 derived in paper i ( eqs . [ 11 ] and [ 17 ] ) .
the functions @xmath12 and @xmath13 required to obtain @xmath23 from @xmath20 are fixed by simple consistency arguments ( see paper i ) @xmath24 in the above equation @xmath25 is the linear growth factor at the considered collapse time @xmath8 and @xmath26 is the density contrast for collapse at @xmath8 linearly extrapolated to @xmath8 , both functions being dependent on the particular cosmogony assumed .
constants @xmath27 and @xmath28 are therefore the only free parameters of the model . for each pair of values of these parameters
, the filtering process translates into a different mass evolution of halos , with only one couple of values ( approximately ) recovering the evolution encountered in real gravitational clustering .
the correct values of these two parameters can be obtained by adjusting the mass function predicted by the model to the mass function resulting from @xmath2-body simulations at some given time , for example at @xmath29 ( this automatically leads to a similarly good fit at any other time ; see 4 ) .
the remaining statistical quantities predicted by the model rely on the definitions of accretion and merger , and halo formation and destruction .
since in hierarchical clustering halos grow constantly by capturing other halos , the fact that they are quasi - steady systems implicitly presumes that some of the captures are tiny enough for the steady state of the capturing systems not to be essentially altered .
notable captures yielding a transient departure from steadiness establish then the frontier between the initial and final quasi - steady systems .
accretion is therefore defined as any apparently continuous and derivable mass increase along the temporal series of quasi - steady systems subtending at each step the mass of those which precede them .
in contrast , a merger is any appreciable discontinuity in the mass increase along such a temporal series causing a significant departure from the steadiness of the growing system at that step .
a halo is said to survive as long as it evolves by accretion , whereas when it merges it is said to be destroyed .
note that when a halo is captured by one that is more massive it merges . however , the capturing halo may just accrete it if the relative captured mass is small enough . only those events
in which _ all _ the initial halos merge and are destroyed give rise to the formation of new halos .
as shown in paper ii , the assumed correspondence between halos and non - nested peaks allows one to naturally identify the filtering processes at @xmath1 tracing all the preceding processes , as well as to derive their respective rates and characteristic times ( see 4 ) .
in the original version of the cusp model the density distribution around peaks was approximated by top - hat spheres .
in addition , the conditional density of peaks with @xmath3 on scales @xmath4 to @xmath10 , having @xmath30 on @xmath19 and being located within a peak on that larger scale , was approximated by the conditional density of peaks with @xmath3 on scales @xmath4 to @xmath10 simply having @xmath30 on scale @xmath17 . under these circumstances ,
the density of peaks with @xmath3 on scales @xmath4 to @xmath10 located within non - nested peaks with @xmath30 on scales @xmath17 to @xmath31 ( @xmath19 ) takes the simple form @xmath32 equation ( [ nnest ] ) leads , for @xmath33 , to the scale function of non - nested peaks ( eq . [ 1 ] ) and , for @xmath34 , to the density of peaks which become nested in an infinitesimal decrement of @xmath3 ( eq . [ nm ] below ) , respectively tracing the mass function of halos and the density of halos merging into more massive ones .
any inaccuracy in this expression , therefore , translates into these two quantities and any other that is related . to improve equation ( [ nnest ] )
, we would need to drop the top - hat approximation and take into account the peak - peak correlation .
approximate expressions for this latter quantity have been obtained by bardeen et al .
( 1986 ; hereafter bbks ) and regs & szalay ( 1995 ) . unfortunately , these are not useful for our purposes ; being only valid for large separations compared to the filtering scale , while what is needed here is the peak - peak correlation for separations up to @xmath17 .
an alternative , more accurate , expression for the density of nested peaks can be obtained , however , from the conditional probability function @xmath35 of finding the density contrast @xmath36 on scale @xmath4 at a separation @xmath37 from a peak with @xmath3 and curvature @xmath38 on that scale ( averaged over the orientations of the second order derivative tensor ) provided by bbks .
using this conditional probability function we can write @xmath39 with @xmath37 in units of @xmath40 , and @xmath41 the curvature probability function for non - nested peaks with @xmath30 on scales @xmath17 to @xmath31 . expression ( [ nnestini ] ) takes into account the real density distribution around peaks and , since this includes the distribution of neighboring peaks ( on every scale ) , it also accounts for the peak - peak correlation as desired .
it is not exact however , because the conditional mean density of peaks with @xmath3 on scales @xmath4 to @xmath10 subject to having @xmath36 on scale @xmath17 _ at a distance @xmath37 from a peak with @xmath30 and @xmath42 on @xmath17 _ is approximated by @xmath43 , the conditional mean density of peaks with @xmath3 on scales @xmath4 to @xmath10 subject to just having @xmath36 on scale @xmath17 . in the appendix
we detail the calculations that allow us to write equation ( [ nnestini ] ) in a compact form identical to equation ( [ nnest ] ) , but for the new function @xmath44 ( eq . [ [ npknest ] ] ) instead of the conditional density function @xmath45 . in figure 1
we plot these two estimates of the density of nested peaks . as can be seen
, they are very different for @xmath17 close to @xmath4 while they asymptotically coincide for @xmath46 .
this is well understood .
there is no correlation among peaks on very different scales .
thus the conditional mean density of peaks with @xmath3 on scales @xmath4 to @xmath10 subject to being located at a point with @xmath36 on scale @xmath17 at a distance @xmath37 from a peak with @xmath30 and @xmath42 on the same scale is very well approximated , in this case , by the simple conditional mean density @xmath43 used in equation ( [ nnestini ] ) leading to @xmath44 . on the other hand , the spherical top - hat approximation is also very good in the limit for @xmath46 .
hence , not only is the new function @xmath44 entirely accurate at this limit , but it also coincides with the former function @xmath45 .
all the quantities predicted by the new version of the cusp model are readily obtained , following the same derivations as in papers i and ii , by simply replacing the function @xmath45 by @xmath47 .
this leads to the following expressions .
the mass function of halos is @xmath48 with @xmath20 the scale function of non - nested peaks at @xmath3 , solution of the volterra equation @xmath49 in figure 2 we compare this mass function with that obtained in the lc model .
the parameters @xmath50 and @xmath27 entering in the cusp model were fixed by adjusting it to the ps mass function in the range @xmath51 m@xmath52 m@xmath53 at @xmath54 , where the latter gives a very good fit to the results of @xmath2-body simulations .
both functions are very similar for massive halos at any redshift . only for small masses
does the cusp mass function predict slightly fewer halos than that of the ps .
note also that the large differences between the new and old approximations for nesting at @xmath55 ( fig .
1 ) have no appreciable effect on the mass function : the original and new solutions being almost exactly superimposed .
this indicates that the mass function is particularly insensitive , in the mass range of figure 2 , to the peak - peak correlation .
the instantaneous mass accretion rate for halos of mass @xmath6 at @xmath8 is @xmath56 with @xmath57 the second order spectral moment ( eq . [ [ sigmai ] ] ) , and @xmath58 the average inverse curvature for the distribution function @xmath59 .
there is no similar prediction in the lc model which does not differentiate accretion from merger . only the total mass increase rates arising from the two models
can be compared ( see below ) . in figure 3
we plot the accretion rates obtained from the original and new versions of the cusp model . once again , both solutions are almost fully superimposed showing that this rate is also insensitive to the peak - peak correlation . the instantaneous specific merger rate for halos of mass @xmath6 at @xmath8 per infinitesimal range of the resulting masses @xmath60 is @xmath61 where @xmath62 is the density of peaks with @xmath5 on scales @xmath4 to @xmath10 becoming nested into non - nested peaks with @xmath63 on scales @xmath17 to @xmath31 @xmath64\big|_{\delta_f=\delta}.\label{nm}\ ] ] to compare this specific merger rate with that yielded by the lc model we must take into account the different merger definitions adopted in the two models . in the lc model , merger is any mass capture experienced by halos .
in contrast , in the cusp model only notable captures are considered true mergers , while small captures contribute to accretion . as shown in figure 4
, the cusp specific merger rate ( using both the original and new approximations for the density of nested peaks ) fully recovers that of the lc in the asymptotic regime at large @xmath65 , but notably deviates for small @xmath66 .
this departure is in the expected direction : for small mass captures the specific merger rate increases monotonically in the lc model , while it shows an abrupt cutoff in the cusp model . from figure 4
we also see that the original and new versions of the cusp model now give markedly different solutions .
they coincide at very large @xmath67 , owing to the similarity of the density of nested peaks at @xmath46 , but the old solution begins to decline , deviating from the lc solution , at @xmath67 much greater than unity .
this implies that some smooth distinction between tiny and notable captures is already operating at @xmath68 , which is meaningless .
this is not the case for the new solution which keeps closer to that of the lc model until @xmath69 , then rapidly falls off implying the existence of a sharp effective frontier between tiny and notable captures at @xmath70 .
nonetheless , the new solution is not fully satisfactory : it becomes slightly negative for @xmath67 smaller than the effective threshold for merger , and the latter tends to be unreasonably small or large for extreme values of @xmath6 .
this shows the shortcomings of the new approximation used for the density of nested peaks .
all the remaining quantities predicted by the cusp model are determined by the three preceding ones which consequently fix the behavior of the whole model .
the instantaneous total mass increase rate for halos of mass @xmath6 at @xmath8 , @xmath71 , is equal to the mass accretion rate ( eq . [ mar ] ) plus the mass merger rate , defined as @xmath72 the instantaneous destruction rate ( or global merger rate ) of halos of mass @xmath6 at @xmath8 is @xmath73 while their instantaneous formation rate take the form @xmath74 where @xmath75 is the density of non - nested peaks appearing from @xmath3 to @xmath76 , @xmath77 ( note the error in the expression quoted in paper ii ) , with @xmath78 and @xmath79 defined for a fixed @xmath4 instead of a fixed @xmath3 ( see paper i ) . to see that expressions ( [ fr ] ) and ( [ nfr ] ) can be written in terms of the first three quantities above
, one must simply substitute @xmath23 , @xmath80 , and @xmath81 ( eqs . [ mf ] , [ mar ] , and [ dr ] ) into the conservation equation @xmath82-r^{d}[m(t),t]-\partial_m r^a_{mass}(m , t)\bigl|_{m = m(t)},\label{conserv}\ ] ] for the number density of halos per unit mass along mean accretion tracks , @xmath83 , solutions of the differential equation @xmath84 . \label{massr}\ ] ] finally , the distributions ( and typical values ) of formation and destruction times can be readily inferred from the expression giving the cumulative spatial number density of halos at @xmath1 with masses from @xmath85 to @xmath86 surviving until @xmath87 @xmath88\,\delta m(t_i)\,\exp \biggl\{-\int_{t_i}^{t_f } r^{d}[m(t'),t']\,dt'\biggr\ } , \label{nsur}\ ] ] with @xmath83 the mass at @xmath8 of such halos calculated along their mean accretion tracks , and @xmath89 \label{deltam}\ ] ] the element of mass at @xmath1 evolving , through mean accretion tracks , into @xmath90 at @xmath87 . indeed , by taking @xmath91 and @xmath92 in equation ( [ nsur ] ) we obtain the distribution of destruction times of halos at @xmath29 with masses between @xmath93 and @xmath94 , with @xmath95 arbitrarily small , @xmath96\ , \exp\biggl\{-\int_{t_0}^t r^{d}[m(t'),t']\ , dt'\biggr\},\label{phid}\ ] ] while , by taking into account that the cumulative spatial number density of those halos at @xmath29 which preexist at @xmath97 , @xmath98 , coincides with the cumulative spatial number density of halos at @xmath99 surviving until @xmath100 , equations ( [ conserv ] ) and ( [ deltam ] ) lead to the distribution of formation times @xmath101\ , \exp\biggl\{-\int_t^{t_0 } r^{f}[m(t'),t']\,dt'\biggr\}.\label{phif}\ ] ]
we have investigated the effects that the inclusion of the spatial correlation among peaks has on the cusp model .
the predictions for the mass function of halos , their mass accretion rate , and their specific merger rate for large captured masses , are very similar to those arising from the original version of this model , which were already satisfactory . on the contrary ,
the specific merger rate at intermediate masses changes appreciably , now becoming more consistent with the corresponding prediction by the lc model .
we , therefore , confirm that the poor behavior shown by this latter quantity in the original version of the cusp model was not intrinsic to its foundations , but due to the rather crude approximation originally used for the density of nested peaks .
a more accurate expression for the peak - peak correlation than that used here is , however , required for the predicted specific merger rate to be fully satisfactory .
an interesting characteristic of the cusp model is that it distinguishes between accretion and merger , which allows one to unambiguously define the formation and destruction of halos respectively as the last and next merger they experience .
mergers are those mass captures restructuring the systems , and hence , establishing their morphological properties until a new merger takes place .
such a distinction is therefore especially well suited for investigating the origin of the morphology of halos and , by extension , that of galaxies and galaxy clusters .
since the ps formalism involves much simpler expressions than those appearing in the peak theory , implementing that distinction in the former would have the added advantage of simplicity .
the results of the present work suggest how it should be done .
such a modification of the lc model is used , in salvador - sol , solanes , & manrique ( 1997 ) , to study the origin of the correlation between halo concentration and halo mass recently found in cosmological @xmath2-body simulations ( navarro , frenk , & white 1996 ) .
the same reasoning leading to equation ( [ nnestini ] ) allows us to write its restriction to peaks with curvature @xmath38 to @xmath102 on scale @xmath4 , @xmath3 on scales @xmath4 to @xmath10 , and @xmath30 on scales @xmath17 to @xmath31 @xmath103 in the preceding expression we have taken the volume of the collapsing clouds associated with peaks delimited by a top - hat window with radius proportional to the filtering scale @xmath4 .
likewise , the density function of non - nested peaks with @xmath3 on scales @xmath4 to @xmath10 and curvature between @xmath38 and @xmath102 satisfies the relation @xmath104 similar to equation ( [ corr ] ) . from equations ( [ nnestini ] ) , ( [ b2 ] ) , and ( [ b3 ] ) we have @xmath105 .
\label{pxfull } \label{b4}\end{aligned}\ ] ] under the approximation @xmath106 , the curvature moments for this probability function are accurate to within a few percent .
since the function @xmath107 is bell - shaped , relatively symmetrical , and close to a gaussian it can be further approximated by a normal distribution with identical mean and variance , @xmath108 and @xmath109 . in taking this approximation
we will also extend the domain of integration over @xmath42 in equation ( [ b2 ] ) to the whole range @xmath110 .
this does not introduce any appreciable error since @xmath108 is considerably greater than @xmath111 for any reasonable values of @xmath3 and @xmath4 and any realistic spectrum .
the distribution function @xmath35 is a normal distribution with mean and variance equal to ( bbks ) @xmath112 and @xmath113 ^
2= \sigma_0 ^ 2\biggl\{1- { 1\over
( 1-\gamma^2)}\biggl[\psi^2 + \biggl(2\gamma\psi+{\nabla^2\psi\over u^2}\biggr){\nabla^2\psi\over u^2 } \biggr]-5\,\biggl({3\,\psi'\over u^2\,r}-{\nabla^2\psi\over u^2 } \biggr)^2-{3\bigl(\psi'\bigr)^2\over \gamma\,u^2}\biggr\}.\label{var}\ ] ] in the above expressions , @xmath114 , with @xmath37 in units of @xmath40 and @xmath115 the mass correlation function on scale @xmath4 , @xmath116 , @xmath117 , @xmath118 , and @xmath119 with @xmath120 the power spectrum of the density fluctuations .
given the normal character of both @xmath121 and @xmath122 , the integration over @xmath42 on the right hand - side of equation ( [ nnestini ] ) can be performed analytically , the result being @xmath123 the new function @xmath124 , equal to the probability function of finding the value @xmath36 at a separation @xmath37 from a peak with @xmath3 in the density field smoothed on scale @xmath4 , is a normal distribution with mean @xmath125 and variance @xmath126 ^ 2= \sigma_0 ^ 2\biggl\{1- { 1\over ( 1-\gamma^2)}\biggl[\psi^2 + \biggl(2\gamma\psi+{\nabla^2\psi\over u^2}\biggr){\nabla^2\psi\over u^2 } \biggr]-5\,\biggl({3\,\psi'\over u^2\,r}-{\nabla^2\psi\over u^2 } \biggr)^2-{3\bigl(\psi'\bigr)^2\over \gamma\,u^2}\nonumber\\ + { ( \delta x)^2\over ( 1-\gamma^2)^2 } \biggl(\gamma\psi+ { \nabla^2\psi\over u^2}\biggr)^2\biggr\}. \label{b7}\end{aligned}\ ] ] this enables us to perform the integration over @xmath36 on the right hand - side of equation ( [ b5 ] ) also analytically , obtaining @xmath127 where @xmath128 \over ( 2\pi)^2r_*^3 } { \displaystyle { \exp\biggl\ { -{[\nu-\epsilon(r)\nu'(r)]^2 \over { 2\,[1-\epsilon^2(r)]}}\biggr\ } } \over { [ 1-\epsilon^2(r)]^{1/2}}}\,{\sigma_2\over \sigma_0}\,r\,dr\label{npkr}\ ] ] is the conditional density of peaks with @xmath3 on scales @xmath4 to @xmath10 subject to being located at a separation @xmath37 from a peak with @xmath30 on scales @xmath17 to @xmath31 . in equation ( [ npkr ] )
@xmath129 , and @xmath130 $ ] is defined just as @xmath131 $ ] in paper i ( eq.[17 ] ) , but for the @xmath37 dependence in @xmath132 and @xmath133 introduced through @xmath134 ^ 2\over\sigma_0 ^ 2(r')}\biggr\}^{1/2 } , \qquad\qquad\nu'(r)={\overline{\delta'(r)}\over \sigma_0(r ' ) } \biggl\{1-{[\delta\delta'(r)]^2\over\sigma_0 ^ 2(r')}\biggr\}^{-1/2}. \label{nur}\ ] ] therefore , by defining @xmath135 equation ( [ b8 ] ) adopts the desired compact form similar to equation ( [ nnest ] ) @xmath136 bardeen , j.m . ,
bond , j.r . ,
kaiser , n. , & szalay , a.s .
1986 , , 304 , 15 ( bbks ) bernardeau , f. 1994 , , 427 , 51 bond , j.r . , cole , s. , efstathiou , g. , kaiser , n. 1991 , , 379 , 440 bond , j.r . , & myers , s.t .
1996 , , 103 , 41 bower , r.j . , 1991 , , 248 , 332 lacey , c. , & cole , s. 1993 , , 262 , 627 ( lc93 ) lacey , c. , & cole , s. 1994 , , 271 , 676 ( lc94 ) manrique , a. , & salvador - sol , e. 1995 , , 453 , 6 ( paper i ) manrique , a. , & salvador - sol , e. 1996 , , 467 , 504 ( paper ii ) navarro , j. , frenk , c. , & white , s.d.m .
1996 , , 462 , 563 press , w.h . , & schechter , p. 1974
, , 187 , 425 ( ps ) regs , e. , & szalay , a.s . , 1995 , , 272 , 447 salvador - sol , solanes , j.m . , & manrique , a. 1997 , , submitted | in two previous papers a semi - analytical model was presented for the hierarchical clustering of halos via gravitational instability from peaks in a random gaussian field of density fluctuations .
this model is better founded than the extended press - schechter model , which is known to agree with numerical simulations and to make similar predictions . the specific merger rate , however , shows a significant departure at intermediate captured masses .
the origin of this was suspected as being the rather crude approximation used for the density of nested peaks . here , we seek to verify this suspicion by implementing a more accurate expression for the latter quantity which accounts for the correlation among peaks .
we confirm that the inclusion of the peak - peak correlation improves the specific merger rate , while the good behavior of the remaining quantities is preserved .
# 1*#1 * |
the study of evolution of globular clusters has a long history
. it would be most desirable to use direct @xmath0-body calculations in order to follow the evolution of realistic globular clusters realistically , but we do not have enough computing power to do that for @xmath0 comparable to the number of stars in real globular clusters at this moment .
therefore we have to rely on approximate techniques to understand the dynamics of star clusters .
the fokker - planck equation has been widely used for the study of globular clusters , but it requires many simplifying assumptions .
the assumption of velocity isotropy is usually employed because it allows fast and accurate integration of the fokker - planck equation ( cohn 1980 ) . however , high degree of velocity anisotropy is known to be generated by two - body relaxation in the outer parts of isolated clusters ( e.g. , spitzer 1987 ) .
the observational properties of the clusters depend on the degree of anisotropy and correct treatment of anisotropy becomes an important issue .
the development of velocity anisotropy was investigated in some recent works which used various numerical methods : e.g. , direct @xmath0-body calculations by giersz & heggie ( 1994 ) , gaseous model calculations by louis & spurzem ( 1991 ) and spurzem ( 1996 ) , and fokker - planck calculations by cohn ( 1979 ) , drukier et al . ( 1999 ) and takahashi ( 1995 , 1996 , 1997 ) .
most of these calculations confirmed the earlier expectations of generation of radial anisotropy in the outer parts of the clusters , although detailed behaviour depends somewhat on the numerical methods .
actual clusters are embedded in the galactic tidal field which should impose a finite boundary to the cluster .
the introduction of tidal boundaries should have significant effects on the velocity anisotropy .
recently takahashi , lee & inagaki ( 1997 ; abbreviated as tli hereafter ) performed anisotropic fokker - planck simulations for single mass clusters and demonstrated that radial anisotropy grows up to the time of core collapse , but the degree of anisotropy begins to decrease to smaller values as the cluster expands and loses significant amount of mass .
they showed that there is even _
anisotropy very close to the tidal boundary in the late phase of the cluster s evolution .
the density profiles of the clusters can be reasonably fitted by isotropic king models for most of the time except for the brief phase around the core collapse , when the degree of radial anisotropy nearly reaches its maximum .
the behaviour of the anisotropy for isolated clusters with mass spectra has some interesting features .
takahashi ( 1997 ) showed that the degree of anisotropy for different mass components behaves differently .
there is always radial anisotropy for low mass components , but tangential anisotropy can develop for high mass components at early evolutionary stages when the temperature of heavy stars decreases rapidly toward the equipartition . during this ` cooling ' phase , the radial velocity dispersion decreases more rapidly than the tangential one .
this leads to the development of tangential anisotropy among high mass stars . as in the case of single mass models , extension of these studies into tidally limited clusters
should provide useful insight on the evolution of realistic clusters . based on the experience of single mass models
, one obviously expects that the velocity anisotropy will be greatly affected by the presence of the tidal field .
some of the essential points have been clarified by the analyses of @xmath0-body results ( vesperini & heggie 1997 ) , but one should remember that the @xmath0-body calculations are done for small @xmath0 cases . many of the behaviour of self - gravitating systems does not allow simple scaling ( e.g. , portegies zwart et al .
1998 , takahashi & portegies zwart 1998 ) .
we make detailed calculations for the evolution of globular clusters with these effects in order to make quantitative assessments for the long term evolution of star clusters with the tidal cut - off and mass spectrum .
the study of the evolution of tidally limited clusters has important areas of application .
the mass function of globular clusters is difficult to obtain because of large dynamic range in brightness of individual stars as well as large variation of stellar densities .
it is very difficult to determine the _ global _ mass function observationally , but there have been several studies to determine the global mass function based on _ locally _ measured mass functions in limited areas of globular clusters ( richer et al . 1990 , 1991 and references therein ) .
it is absolute necessity to obtain the global mass function if one wants to study factors determining the mass function of clusters .
the conversion of the local mass function to the global one is not a trivial task .
the general practice employed widely is to fit the observation to multi - mass king models ( e.g. , meylan 1988 ) or to assume that the mass function in low mass stars in the outer parts are close to the global one ( richer et al .
1991 ) . however , there is no guarantee that these processes give a right answer .
the detailed studies of evolving clusters with the tidal boundary can provide an important check of the above mentioned practices .
the mass function changes with time because the rate of escape of stars depends on the mass of the star .
the high mass stars tend to spiral into the central parts by dynamical friction , and consequently low mass stars are preferentially removed from the tidally limited clusters .
the slope of mass function thus becomes less steep as the cluster loses mass .
even if one gets the global mass function through careful observations and correct conversion process , the present day mass function ( pdmf ) may be significantly different from the initial mass function ( imf ) .
the detailed studies of evolving star clusters could provide an important framework of estimating the relation between pdmf and imf .
in the present paper , we make detailed studies of the the evolution of tidally limited multi - mass clusters including the effects of velocity anisotropy .
we pay special attention to the behaviour of mass function during the course of dynamical evolution .
we examine the adequacy of the process of converting the local mass function to the global mass function using multi - mass king model fitting .
we then analyze how the mass function changes with time and location .
this paper is organized as follows . in
[ sec : fpmodels ] , we describe our models for the evolution of globular clusters . in
[ sec : ch ] , we introduce the idea of half - mass concentration and discuss its relation to the development of anisotropy and to mass loss rates . in [ sec : mmmodels ] we describe initial conditions of multi - mass models . in [ sec : results ] we present the results of the model calculations . the relationship between local and global mass function and the temporal evolution of mass function
are examined in [ sec : mf ] .
we discuss the velocity dispersion profiles in [ sec : vd ] .
the final section summarizes basic results .
we use the anisotropic fokker - planck code developed by takahashi ( 1995 ) .
it is based on the direct integration method of the fokker - planck equation in @xmath1 space ( cohn 1979 ) , where @xmath2 is the energy of a star per unit mass and @xmath3 is the angular momentum per unit mass .
the extension to multi - mass clusters is described in takahashi ( 1997 ) and the treatment of tidal boundary can be found from tli .
tli used two different criteria for the escape of a star from the cluster : apocenter criterion and energy criterion .
the _ apocenter _ criterion assumes that a star escapes if the apocenter distance of a star with @xmath1 becomes greater than tidal radius @xmath4 .
this criterion preferentially removes the stars with low @xmath5 and thus suppresses the development of radial anisotropy .
the @xmath6 criterion assumes that a star escapes if @xmath7 where @xmath8 .
the radial anisotropy is also suppressed in this criterion and tangential anisotropy can develop ( see tli ) .
the general behaviour of the cluster evolution is not much dependent on the choice of escape criteria for the models of tli , but details of the clusters vary with the treatment of stellar evaporation .
takahashi & portegies zwart ( 1998 ) have shown that the apocenter criterion gives better results when compared with realistic @xmath0-body calculations . in the present study
, we examine the results of our calculations with both of the criteria , but more emphasis is given to the calculations with the apocenter criterion .
hereafter , we denote anisotropic models with the apocenter criterion as _ aa _ models and anisotropic models with the energy criterion as _ ae _ models .
the capital _ a _ stands for anisotropic models , and the lower cases _ a _ and _ e _ stand for apocenter and energy criteria , respectively . similarly isotropic models are denoted as _ ie _ models , but _
ia _ models do not exist .
before going into descriptions of the results of multi - mass models , we examine the effects of the degree of the concentration of a cluster on the development of velocity anisotropy and on the cluster mass loss rate .
we use single - mass models in this section because these models are simple , but the conclusions of this section are also true for multi - mass models , as we show later .
tli used the initial model of a single - mass plummer model with the cutoff radius of 32 in plummer model units .
the results of their simulations show that the _ aa _ and _ ae _ models lose mass faster than the _ ie _ model ( see fig . 2a of tli , though the _ ae _ model is not shown in it ) .
the evaporation time of the _ ie _ model is about 12% longer than that of the _ aa _ model , and the latter is about 10% longer than that of the _ ae _ model .
the biggest difference in the mass loss rate between the anisotropic models and the isotropic model appears in the pre - collapse and the early post - collapse phases .
the mass loss rates of the different models are similar in the late post - collapse phase .
in [ sec : results ] we show the results of simulations which start with multi - mass king models . as shown in fig .
[ massfig ] , the _ ie _ model and the _ ae_model are very similar in the evolution of the total mass .
the _ aa _
model loses mass _ slower _ than the _ ie _ model .
this trend is opposite to that seen in the simulations of tli . while fig .
[ massfig ] shows the results for a @xmath9 king model , the same trend is seen for king models with other @xmath10 .
why does the opposite trend appear ?
the change from the single - mass model to the multi - mass model is not the reason of the change of the trend .
rather , it is found that the trend changes with the concentration of a cluster to the tidal cutoff radius .
first we discuss what parameter is useful in measuring the degree of the concentration of a cluster . for king models ,
the concentration is usually measured in terms of the concentration parameter @xmath11 , where @xmath12 is the king core radius ( king 1966 ) .
the parameter @xmath13 is useful in measuring the concentration of the core .
however , when discussing the variation of the total mass , @xmath13 is not always an appropriate parameter , because the fraction of the core mass is very small in large @xmath13 clusters
. it may be more appropriate to use the half - mass radius @xmath14 instead of the core radius @xmath12 .
thus we define half - mass concentration @xmath15 ( note the logarithm is not taken here ) .
king models form a one - parameter family parameterized by the concentration @xmath13 or equivalently by the dimensionless central potential depth @xmath10 ; @xmath13 monotonically increases with @xmath10 ( king 1966 ) . on the other hand
, @xmath16 does vary monotonically with @xmath10 .
the relation among @xmath10 , @xmath13 , and @xmath16 is given in fig .
[ fig : king ] for @xmath17 .
in contrast to that the core concentration @xmath13 varies by nearly three orders of magnitude , the variation of the half - mass concentration @xmath16 is within a factor of three for this range of @xmath10 .
the half - mass concentration @xmath16 increases with @xmath10 for @xmath18 to 8 , then it decreases until @xmath10 reaches and increases again . for the range of @xmath17
, @xmath16 reaches the maximum of about 10 at @xmath19 .
the initial model of tli has @xmath20 , therefore it is far more concentrated than any king models in terms of the half - mass concentration . in order to investigate the effects of the half - mass concentration on the cluster evolution , we carry out simulations which start with single - mass plummer models with different cutoff radii which correspond to @xmath2110 , 20 , and 30 . in these simulations
the number of particle is set to @xmath22 , which is specified to include the effects of three - body binary heating . fig .
[ fig : plmass ] shows the evolution of the total mass for the _ ae _ and _ ie _ models . for the large @xmath16 initial models , the _ ae
_ model loses mass much faster than the _ ie _ model at the early evolutionary phase as in tli s simulations .
( the core collapse occurs at @xmath231618@xmath24 in all the models . ) as @xmath16 decreases , the difference between the _ ae _ and _ ie _ models decreases and finally almost diminishes for @xmath25 . fig .
[ fig : plbeta ] shows the evolution of the anisotropy @xmath26 at the 90%-mass radius .
the anisotropy parameter @xmath26 is defined as follows : @xmath27 where @xmath28 and @xmath29 are one - dimensional tangential and radial velocity dispersions , respectively .
the development of the radial anisotropy during the pre - collapse phase becomes strongly suppressed as @xmath16 decreases .
this is simply because radial - orbit stars can easily escape from a small tidal radius cluster . as tli discussed , the emergence of a large number of radial - orbit stars is the reason for faster mass loss in the anisotropic models . for low @xmath16 clusters ,
since there is almost no room for the radial anisotropy to develop , the _ ae _ model and the _ ie _ model do not differ very much in the evolution of the total mass .
since king models have such low @xmath16 ( @xmath30 ) , the difference among the three models shown in fig .
[ massfig ] can be understood from this viewpoint .
the mass loss of the _ aa _ model should be always slower than that of the _ ae _ model because of the apocenter criterion ( see tli ) .
note that we have assumed that the initial tidal radius is equal to the limiting radius of king models .
this is a conventional way for setting up initial conditions of simulations , but not the only one ; these two radii can be different .
finally in this section , we comment on the relation between the mass loss rate and the half - mass concentration for king models .
johnstone ( 1993 , figure 2 ) and gnedin & ostriker ( 1997 , figure 6 ) found that for king models the mass loss rate per half - mass relaxation time ( exactly speaking , the inverse of the evaporation time in units of the initial half - mass relaxation time in gnedin & ostriker ) has a minimum around @xmath311.52 .
johnstone ( 1993 ) calculated the evaporation rate for king models using the local fokker - planck equation , and gnedin & ostriker ( 1997 ) performed isotropic orbit - averaged fokker - planck simulations with the initial conditions of king models .
their findings are qualitatively consistent with a simple estimation of the mass loss rate based on only the half - mass concentration @xmath16
. naively one may think that the mass loss rate decreases as @xmath16 increases .
in fact @xmath16 has the maximum at @xmath19 ( @xmath32 ) , where johnstone ( 1993 ) and gnedin & ostriker ( 1997 ) found the minimum evaporation rate .
in addition johnstone ( 1993 ) found a local maximum of the mass loss rate at @xmath33 .
this may correspond to the fact that @xmath16 has a local minimum at @xmath34 ( @xmath35 ) as shown in fig .
[ fig : king ] . as an additional reference , we show the mass loss rate per half - mass relaxation time @xmath36 calculated from the modified ambartsumian - spitzer formula ( spitzer 1987 , p.57 ; see also tli ) for king models in fig . [
fig : king ] .
this estimation is even quantitatively not very different from the result of more detailed calculations for @xmath37 by johnstone ( 1993 ) .
there are growing evidences that the mass function in clusters as well as in the field is not a simple power - law , but should be approximated as composite power - laws ( e.g. , kroupa , tout & gilmore 1993 ) .
however , we use simple power - laws in the present study of general evolution of clusters , for simplicity .
the adoption of multiple composite power - laws is trivial task , but requires specification of more model parameters .
the assumed initial mass function ( imf ) has the following form : @xmath38 where @xmath39 and @xmath40 are constants . as in takahashi ( 1997 )
, we use the discrete mass components .
the @xmath41-th mass bin for equal @xmath42 interval can be obtained by @xmath43 where @xmath44 is the number of components .
the parameters determining the mass function are @xmath40 and @xmath45 .
we use a few values of @xmath40 and @xmath46 for our model calculations .
the issue of the effects of stellar evolution naturally comes out for multi - mass models if one extends the mass function to high mass stars .
obviously the stellar evolution should have been the most important process in determining the early phase of the dynamical evolution of globular clusters .
however , we will ignore the stellar evolution in this paper because it adds one more complexity to the models . without stellar evolution
, the clusters usually undergo core - collapse before losing significant amount of stars unless the initial cluster has very low central concentration , because the time scale for the core collapse is generally much shorter than the evaporation time .
note that the core collapse time is of order of @xmath47 while the evaporation time is typically of order of @xmath48 , where @xmath49 is the initial half - mass relaxation time for single mass models .
the core collapse is very much accelerated for multi - mass clusters , and the discrepancy between the evaporation time and the core collapse time could become even larger .
therefore one needs to supply the mechanism that allows the evolution of cluster beyond the core collapse .
we assumed that the heating is provided by binaries formed by three - body processes because it is easiest to implement in the framework of fokker - planck method .
the evolution of parameters for the central parts such as density and velocity dispersion should depend on the detailed mechanism for driving the post core - collapse evolution , but most of the global properties do not sensitively depend on the heating mechanism . for our purposes ,
three - body binary heating is sufficiently simple and easy to be used as heating mechanism .
the heating formula for multi - mass clusters is taken from lee , fahlman & richer ( 1991 ) .
we have used various initial models .
the characteristics of the initial models are summarized in table 1 . in all models
, we have assumed that the shape of density profiles of different mass components are the same with the different amplitudes dictated by the assumed mass functions .
the velocity dispersions are assumed to be the same for all components .
are assumed to have the same density and velocity profiles .
the behaviour of the cluster up to the core collapse does not depend on the initial total mass or size of the cluster .
but the amount of binary heating depends on the number of stars in the cluster .
we have assumed that the mass of the cluster is @xmath50 for the model @xmath51 , and we have fixed the total number of stars at @xmath52 for other models . but
this number only controls the central density during the postcollapse phase .
most of our results does not really depend on the mass of the cluster , except for the surface density profiles .
if we used higher initial mass , the core radii at a given epoch ( measured by the @xmath53 ) would be smaller .
the minimum mass @xmath54 of the mass function was assumed to be 0.1 @xmath55 in all models . throughout this paper
, we use @xmath56 to represent the distance from the center of the cluster , and @xmath51 to represent the _ projected _ distance .
we first describe the general behaviour of _ reference model _ ( model r )
whose initial parameters are given in table 1 .
this set of parameters is used by heggie et al .
( 1998 ) to compare the outcomes of various computational methods .
we will consider this initial condition as a reference model .
the number of mass component was 8 . [ cols="<,<,^,^,<",options="header " , ] [ tab : ref ] the evolution of the total mass
is shown in fig .
[ massfig ] with different criteria for ejection of stars .
unlike single component models , the mass loss rate is not a constant , but increases with time .
the increase of @xmath57 is due to the fact that the mass function becomes flatter as the cluster loses mass .
( the mass loss rate becomes larger for clusters with flatter mass function . ) as discussed in 3 , the apocenter criterion ( @xmath58 ) gives longer lifetime than the energy criterion ( @xmath59 ) , but it is not a significant effect . for comparison , we have also shown the mass evolution for the isotropic model ( @xmath60 ) with the same initial conditions .
the time of maximum collapse is shown as dots in this figure .
the behaviours of the degree of velocity anisotropy @xmath26 for the apocenter criterion are shown in figure [ aniso - apo ] .
the upper , middle and lower panels show @xmath61 at the time @xmath53=0.7 , 0.5 and 0.2 , respectively .
the same figure as fig .
[ aniso - apo ] for the energy criterion is shown in fig .
[ aniso - en ] .
note that the sign of @xmath26 can be both positive ( radial anisotropy ) and negative ( tangential anisotropy ) .
if a cluster is isolated , strong radial anisotropy develops in the halo for all components , as shown by takahashi ( 1997 ) .
however the present model cluster is in a rather strong tidal field : the initial king model with @xmath9 has @xmath62 .
therefore , from the discussion in
[ sec : ch ] , we may expect that the development of such radial anisotropy is strongly suppressed in the present case . from the profiles shown in fig .
[ aniso - apo ] ( _ aa _ model ) , in fact , only weak radial anisotropy is seen for the high mass components at intermediate radii .
the low mass components tend to have tangential anisotropy at all radii .
a similar trend is seen in fig .
[ aniso - en ] ( _ ae _ model ) .
this is in sharp contrast with isolated clusters where strong radial anisotropy develops among low mass components ( takahashi 1997 ) . in the top panel of fig .
[ aniso - en ] , the high mass components also have tangential anisotropy . this is due to the initial cooling of the high mass components ( takahashi 1997 ) .
the velocity distribution becomes mostly tangential near the tidal boundary regardless of the mass .
the tangential anisotropy was also noted by oh & lin ( 1992 ) , and kim & oh ( 1999 ) from their hybrid calculation of n - body and fokker - planck .
the behaviour of @xmath61 becomes different for apocenter and energy criteria in the outer parts , because the energy criterion forces @xmath26 at tidal boundary to become zero ( i.e. , isotropic ) .
the profiles of surface densities for different mass components are shown in fig .
[ density_a ] for a model with the apocenter condition at the epochs of @xmath63 , 0.5 and 0.2 .
[ density_e ] shows the density profiles for a model with the energy condition .
the density profiles at @xmath53=0.5 and 0.2 are similar for the apocenter condition and the energy condition .
the big difference at @xmath53=0.7 is purely due to the fact that the core - collapse occurs when @xmath64 0.7 for the _ aa _ model while it occurs somewhat later for the _ ae _ model .
therefore , the difference in the degree of anisotropy at outer radii does not play important role in determining the density distribution .
the profiles at @xmath53=0.5 and 0.8 are compared with the isotropic model in fig .
[ compare1 ] .
the radial profiles of the total density of isotropic and anisotropic models have very little difference , but the detailed distribution of individual mass components could be significantly different for these two models . from these comparisons , we expect that there will be some difference in the behaviour of the mass function in the lower mass parts at late phase of the evolution . in fig .
[ compare2a ] , we tried to compare the surface density profiles of the @xmath65 model at @xmath53=0.5 with the best - fitting multi - mass king models having the same mass function .
multi - mass king models have the following distribution function ( e.g. , binney & tremaine 1987 ) : @xmath66 , \quad ( i=1 , 2 , ... , k),\ ] ] where @xmath67 satisfies the relation @xmath68 for all @xmath41 and @xmath69 .
the constants @xmath70 should be determined by the given mass function . as in fig .
[ compare1 ] , the profiles of density of low mass stars do not fit king models : if the distribution of combined density is to be reproduced , the central densities of low mass components of king model are higher than that generated by fokker - planck results .
this means that the lower mass stars are more distributed at the outer parts than expected from king models .
the mass segregation effect is greater than that predicted by multi - mass king models .
this is the same trend that was seen from the comparisons between anisotropic and isotropic models .
for the purpose of application to more realistic cases , we also attempted to fit the surface density profiles of _ visible _ components only , assuming that the highest mass bin ( @xmath71 ) represents the remnant stars such as neutron stars or massive white dwarfs .
this is still far from more realistic situation because the extension of simple power - law to massive remnant will give results in too much contribution from these stars .
nonetheless , such a comparison could be worthwhile in understanding the observed data .
the result is shown in fig .
[ compare2b ] .
the fitting model is the same as shown in fig .
[ compare2a ] , but the difference between the king model and the fokker - planck model is more pronounced . choosing other parameters for king models would not help to improve the fitting :
the difference simply is due to the fact that the distribution of stars are substantially different between king models and fokker - planck results . because of stronger mass segregation in fokker - planck results ,
the fitting of profiles in the intermediate region gives significant differences in outer parts .
the fitting does not improve when one attempts to use anisotropic king models ( king - michie models ) , because the king - michie models always have @xmath72 anisotropy regardless of mass , while the fokker - planck results give both radial and tangential anisotropy among different mass bins .
the isotropic king models are usually better in reproducing the fokker - planck results .
this is somewhat contradictory with the single component models : during the early phase of the evolution , king - michie models better represent the anisotropic fokker - planck results of tli than isotropic king models . as described in
[ sec : ch ] for single - mass models , the development of radial velocity anisotropy in the outer halo is strongly suppressed in low half - mass concentration ( @xmath16 ) clusters . to confirm this for multi - mass clusters we now examine the results for different initial @xmath16 .
[ aniso - apo - king8 ] shows the radial profiles of anisotropy @xmath26 at selected epochs for a reference initial condition except for @xmath73 .
initially this model has a larger @xmath74 than the model discussed in [ sec : heggie ] ( @xmath62 ) .
comparing figs .
[ aniso - apo ] and [ aniso - apo - king8 ] , however , we find that the behaviours of anisotropy are not very different between the two models .
for example , the maximum value of @xmath26 is about 0.3 at the epoch of @xmath75 for both models .
this is not surprising because we may well expect that @xmath16 for the king model of @xmath73 is still not large enough for strong radial anisotropy to develop ( @xmath30 ) , if we remember the results presented in
[ sec : ch ] .
[ aniso - apo - plummer20 ] shows @xmath26 profiles for an _ aa _ model for the initial conditions of a plummer model with @xmath76 , @xmath77 , and @xmath78 ( model @xmath79 ) . in this case , as expected , stronger radial anisotropy ( maximum @xmath80 ) appears and is kept for a longer time than in the previous two cases .
it is observationally challenging task to obtain the mass function of globular clusters .
globular clusters are known as oldest galactic objects and should possess very important memory of the early phase of the universe .
for example , the imf of globular clusters could provide a glimpse of the star formation process during the phase of very low heavy elements .
the loss of mass occurs near the tidal boundary .
the high mass stars gradually spiral into the central parts via dynamical friction and low mass stars are preferentially removed from the cluster .
therefore the shape of the mass function changes with time .
this phenomenon has been observed from many numerical simulations ( e. g. , chernoff & weinberg 1990 ; lee , fahlman & richer 1991 ; lee & goodman 1995 ; vesperini & heggie 1997 ) .
the evolution of the mass function at several different epochs are shown in fig .
[ mfheggie ] .
the epochs are chosen according to the mass of the cluster : @xmath81=0.8 , 0.7 , 0.6 , 0.5 , 0.4 , 0.3 , 0.2 and 0.1 .
this figure clearly shows the flattening of the mass function as the cluster loses mass .
also shown in this figure are the mass functions within the half - mass radius @xmath82 as dotted lines .
the shape of the mass function deviates from the simple power - laws as a result of the selective mass loss : the average slope of the mass function decreases with time , and the slope of the mass function is steeper at lower mass than at higher mass .
although the mass loss rate is larger for the lower mass , it is not sensitive enough to flatten toward the lower mass .
such a behaviour ( i.e. , steeper slope at lower mass parts ) appears to be inconsistent with the observational data for a number of globular clusters ( e.g. , chabrier & mra 1997 ) . the initial mass function of the globular clusters should not be simple power - laws since the dynamical evolution only makes the situation worse , but it is likely to flatyten ( or even decreasing ) toward the lower mass
. the evolution of mass function as a function of time can be measured by the variation of the power - law index .
suppose that the mass function can be approximated as a power - law with index @xmath40 as shown in equation ( [ eq : pmf ] ) . as seen from fig .
[ mfheggie ] , it is clear that the mass function does not follow a single power law as the cluster loses mass .
thus we have computed instantaneous @xmath40 at two different masses : @xmath83 and 0.55 @xmath55 .
the behaviour of these indices is shown in fig [ alpha_hg ] as a function of @xmath84 .
the solid lines represent @xmath85 ) and the dotted lines are for @xmath86 ) .
since the mass function depends on the location as well , we have computed @xmath40 s within the entire cluster ( thick lines ) and within the half - mass radius ( thin lines ) .
the trends toward the flatter mass function with time are clearly shown in this figure .
the global mass function changes slowly in the early phase of the evolution , and rather abruptly in the late phase .
the mass function within the half - mass radius changes more rapidly than the global mass function . the model with smaller @xmath46 gives more rapid variation in @xmath40 with time than that with higher @xmath46 , as shown in figs . [ alpha_hg ] and [ alpha_low ] .
we also note here that the behaviour of @xmath40 for the isotropic fokker - planck model is very close to that of the anisotropic model .
the variation of mass function observed by richer et al .
( 1991 ) could be most attributed by the dynamical evolution since the mass function tends to be _
flatter _ for clusters with short relaxation time .
clusters with short relaxation time are likely to have lost significant fraction of the initial mass , mostly in the form of low mass stars .
richer et al ( 1991 ) claim that there exists a tight relation between the disruption time @xmath87 and the power - law index of the mass function .
although the amount of uncertainties is large , the trend of flattening of mass function for clusters with significant mass loss appears to be real .
the accurate photometry over any entire globular cluster is very difficult .
the luminosity function ( and thus mass function ) is usually measured in limited range of the cluster . since the mass function varies with the location if the cluster has undergone significant evolution , it is important to understand the relationship between the locally determined mass function and the ` global ' mass function .
we have shown the variation of @xmath40 as a function of projected radius for three different epochs in fig .
[ alpha_s ] for 0.2@xmath55 ( dotted lines ) and 0.55@xmath55 ( solid lines ) . also shown as horizontal lines are the power - law indices of the global mass function at that epoch .
we have seen from fig .
[ alpha_low ] that the global mass functions measured at 0.2 and 0.55 @xmath55 are rather similar .
however , the local mass function changes by a great amount . the 0.2@xmath55 mass function varies slowly over radius , but 0.55 @xmath55 one varies a lot .
this is due to the fact that the high mass stars are concentrated toward the central parts .
thus one has to be very careful in determining the mass function from the observations .
the power - law index of the global mass function at 0.2@xmath55 roughly coincides with the local value near the half - mass radius .
the mass function near the half - mass radius appears to represent the global mass function , when measured at the lower mass .
this is also seen in n - body calculations by vesperini & heggie ( 1997 ) .
the power - law index at @xmath82 for 0.55 @xmath55 , however , deviates significantly from the global value .
therefore , the determination of global mass function remains observationally challenging task .
so far we have examined density profiles and mass functions .
we should be able to obtain more detailed dynamical information by measuring the velocity dispersion profiles of the stars in the cluster . in principle , both projected and transverse velocities can be measured if the astrometric accuracy becomes of order of @xmath88 arcseconds , as the astrometric project of gaia is targeting . however , only the projected velocity information is available for most of the clusters in the galaxy .
we now discuss the prediction of projected velocity dispersions of the cluster stars by our fokker - planck models . in fig .
[ vproj ] , we have plotted the projected velocity dispersion profiles for the model @xmath65 at @xmath53=0.7 , 0.5 and 0.2 for four different mass components , together with the corresponding @xmath60 at the same epoch ( measured by @xmath53 ) . the maximum collapse occurs near @xmath53=0.7 .
the unit of velocity in this plot is @xmath89 .
the behaviour of velocity profiles for @xmath90 model was not shown here , but we note that the velocity profiles of such a model behave similarly to those of isotropic model
. this may be due to the similarity in tidal boundary condition .
the @xmath58 model predicts the velocity profiles _ flatter _ than the isotropic model .
since the cluster masses and the tidal radii are kept constant for different models , the difference in velocity profiles is related to the difference in cluster structure .
for example , the @xmath58 model generally exhibits lower central velocity dispersion than the isotropic model .
this means that the anisotropic model has _ larger _
@xmath82 than the isotropic model . near the tidal boundary ,
the projected velocity profiles of the model @xmath58 deviate significantly from the isotropic model .
there are even mild bumps at around 0.5 @xmath4 for low mass stars in late phase ( i.e. , @xmath91 ) .
such a behaviour is only seen for low mass stars in @xmath58 models .
closer examination of tangential and radial velocity dispersions revealed that the bump is caused by the tangential component .
the radial component monotonically decreases like the isotropic model .
the bumps appear in almost all models with apocenter criterion for low mass stars during late postcollapse phase , but not in other models .
therefore , this phenomena is clearly related to the boundary conditions .
we note here that the detailed observations by drukier et al .
( 1998 ) has discovered nearly flat or slightly rising velocity dispersions from m15 at the outer parts .
this seems to be the indication of @xmath92 anisotropic velocity distribution in the outer parts of this cluster
. it would be very desirable to carry out velocity measurements for other clusters .
we have studied the evolution of multi - mass star clusters in the galactic tidal field including the effects of velocity anisotropy .
the radial anisotropy developed by dynamical relaxation tends to be suppressed by the presence of the tidal boundary . except for very early epochs ,
high mass stars show more radial ( larger @xmath26 ) anisotropy than low mass stars in general .
as the cluster loses a large amount of mass from the tidal boundary , @xmath26 decreases , i.e. , anisotropy becomes more tangential for all mass species . as a result , at late epochs of the cluster s life
, low mass stars generally have tangential ( negative @xmath26 ) anisotropy throughout the cluster , while high mass stars show weak radial anisotropy in the inner parts and tangential anisotropy in the outer parts .
the overall degree of anisotropy depends on the half - mass concentration @xmath16 of the cluster .
larger @xmath16 clusters allow the development of stronger radial anisotropy in the halo .
depending on the treatment of the tidal boundary condition , the detailed behaviour of the degree of anisotropy over radius changes near the tidal radius .
the apocenter criterion gives nearly tangential motion for the stars near the tidal radius but the energy criterion forces isotropy there .
the density profiles of tidally limited clusters are compared with the king models . because of complex behaviour of the degree of anisotropy , finding the best - fitting king models becomes very difficult .
the isotropic king models tend to give flatter profiles for low mass components compared to the fokker - planck results when the overall density profiles are matched .
introduction of radial anisotropy gives only worse fits , because most of the anisotropy is _
tangential_. we discussed the adequacy of converting the observed local mass function to the global mass function assuming that the cluster can be approximated by isotropic multi - mass king models . because of the discrepancy between king models and fokker - planck results
, this process gives somewhat erroneous results of mass function for low mass components .
we also looked for the best place where the local mass function is close to the global mass function and found that the half - mass radius should be a reasonable place .
the mass function inevitably changes with time because of the selective loss of mass .
the power - law indices of the mass function is found to be well correlated with the fraction of mass evaporated _ via tidal overflow_. the amount of fractional mass loss can be determined by observed value of @xmath93 and assuming the constant escape probability and coeval formation of all globular clusters . the observational data for the mass function are found to be consistent with the notion that the mass function is strongly influenced by the dynamical processes ( e.g. , richer et al .
1990 ; piotto & zoccali 1999 ) , but imf itself may have been quite different from cluster to cluster : some clusters may have formed with very steep imf .
the projected velocity profiles of anisotropic models with apocenter criterion show significant deviation from isotropic models or anisotropic models with energy condition in the sense that the velocity profiles are much flatter for @xmath58 models in the outer parts during the postcollapse phase .
this phenomena appears to be consistent with the observational data for m15 .
this research was supported by the research for the future program of japan society for the promotion of science ( jsps - rftp97p01102 ) to kt and by kosef grant no .
95 - 0702 - 01 - 01 - 3 to hml . | we study the evolution of globular clusters with mass spectra under the influence of the steady galactic tidal field , including the effects of velocity anisotropy . similar to single - mass models , velocity anisotropy develops as the cluster evolves , but the degree of anisotropy is much smaller than isolated clusters . except for very early epochs of the cluster evolution ,
nearly all mass components become tangentially anisotropic at the outer parts .
we have compared our results with multi - mass , king - michie models . the isotropic king model better fits to the fokker - planck results because of tangential anisotropy . however , it is almost impossible to fit the computed density profiles to the multi - mass king models for all mass components
. thus if one attempts to derive global mass function based on the observed mass function in limited radial range using multi - mass king models , one may get somewhat erratic results , especially for low mass stars .
we have examined how the mass function changes in time . specifically , we find that the power - law index of the mass function decreases monotonically with the total mass of the cluster .
this appears to be consistent with the behaviour of the observed slopes of mass functions for a limited number of clusters , although it is premature to compare quantitatively because there are other mechanisms in contributing the evaporation of stars from the clusters .
the projected velocity profiles for anisotropic models with the apocenter criterion for evaporation show significant flattening toward the tidal radius compared to isotropic model or anisotropic model with the energy criterion .
such a behaviour of velocity profile appears to be consistent with the observed profiles of collapsed cluster m15 . #
1to 0pt#1 # 1to 0pt#1 [ firstpage ] celestial mechanics , stellar dynamics
globular clusters : general |
the nrqcd factorization formalism provides a systematic framework for analyzing the inclusive production of heavy quarkonium @xcite .
long - distance effects involving the binding of a heavy quark - antiquark pair into quarkonium are factored into parameters called nrqcd matrix elements .
these nonperturbative parameters are universal , so values extracted from one high energy physics experiment can be used to predict the production rate in others .
the nrqcd matrix elements scale as definite powers of @xmath11 , where @xmath11 is the typical relative velocity of the heavy quark .
the nrqcd factorization approach becomes phenomenologically useful upon truncating the expansion in @xmath11 so as to reduce the independent nrqcd matrix elements to a manageable number .
the truncation is most reliable for the heaviest quarkonium states , namely the @xmath12 system for which @xmath13 is roughly 1/10 .
the most abundant source of data on bottomonium production is the tevatron @xmath0 collider . in run
ia of the tevatron , the cdf collaboration was able to resolve the individual s - wave bottomonium states @xmath1 , @xmath2 , and @xmath3 and measure their production cross sections @xcite .
an analysis of the cdf data within the nrqcd factorization formalism was carried out by cho and leibovich @xcite .
the analysis is complicated by the production of p - wave bottomonium states that subsequently make transitions to s - wave states .
cho and leibovich found that the cdf data was insufficient to determine all the important nrqcd matrix elements and they had to make educated guesses for some of them .
the cdf collaboration has recently analyzed the data on bottomonium production from run ib at the tevatron .
in addition to much higher statistics on the cross sections for @xmath1 , @xmath2 , and @xmath3 @xcite , they also have results on the production of the p - wave states @xmath5 and @xmath6 @xcite .
the high quality of the new cdf data justifies an updated theoretical analysis , with careful attention to the experimental and theoretical errors . in this paper
, we present a quantitative analysis of the new cdf data on bottomonium production within the nrqcd factorization formalism .
the color - singlet nrqcd matrix elements for s - wave states are determined from their electromagnetic decays , while those for p - wave states are estimated from potential models .
the color - octet nrqcd matrix elements are determined by fitting the cdf data , taking full account of the feeddown from transitions between bottomonium states .
the resulting values of the matrix elements are used to predict the cross sections for the spin - triplet and spin - singlet bottomonium states in run ii of the tevatron .
the nrqcd factorization approach provides a model - independent framework for analyzing the inclusive production of heavy quarkonium @xcite . the factorization formula for the differential cross section for the inclusive production of a bottomonium state @xmath14 of momentum
@xmath15 has the schematic form @xmath16 \;=\ ; \sum_n
d \sigma[b \bar b(n , p ) ] \langle o^h(n ) \rangle , \label{sig - nrqcd}\ ] ] where the sum extends over both color - singlet and color - octet and over all angular momentum channels for the @xmath12 pair .
the @xmath12 cross sections , which are independent of the bottomonium state @xmath14 , can be calculated using perturbative qcd .
all dependence on the state @xmath14 is factored into parameters @xmath17 called _ nrqcd matrix elements_. these phenomenological parameters can be expressed as matrix elements in an effective field theory called nonrelativistic qcd ( nrqcd ) .
a nonperturbative analysis of nrqcd reveals how the various matrix elements scale with the typical relative velocity @xmath11 of the heavy quark in quarkonium .
spin symmetry , which is an approximate symmetry of qcd , also gives relations between various matrix elements .
the relative importance of the terms in the factorization formula ( [ sig - nrqcd ] ) depends on the size of the @xmath12 cross sections and on the size of the matrix elements . according to the velocity - scaling rules ,
the most important matrix element for direct @xmath1 production is the color - singlet parameter @xmath18 .
the spin - symmetry relations can be used to reduce the next most important matrix elements to three color - octet parameters : @xmath19 , @xmath20 , and @xmath21 .
these color - octet matrix elements are important , because the cross sections for producing color - octet @xmath12 pairs can be much larger than for color - singlet @xmath12 pairs .
there are analogous matrix elements that describe the direct production of @xmath2 and @xmath3 .
the nrqcd factorization formula ( [ sig - nrqcd ] ) for direct @xmath22 production reduces to @xmath23 & = & d \sigma[b \bar b_1(^3s_1 ) ] \langle o^{\upsilon(ns)}_1(^3s_1 ) \rangle + d \sigma[b \bar b_8(^3s_1 ) ] \langle o^{\upsilon(ns)}_8(^3s_1 ) \rangle \nonumber \\ & & + d \sigma[b \bar b_8(^1s_0 ) ] \langle o^{\upsilon(ns)}_8(^1s_0 ) \rangle + \left ( \sum_j ( 2j+1 ) d \sigma[b \bar b_8(^3p_j ) ] \right ) \langle o^{\upsilon(ns)}_8(^3p_0 ) \rangle .
\label{sig - ups}\end{aligned}\ ] ] the factor of @xmath24 in the last term comes from using a spin - symmetry relation to eliminate @xmath25 in favor of @xmath26 . the most important matrix elements for the direct production of the p - wave states @xmath27 , @xmath28 , can be reduced to a color - singlet parameter @xmath29 and a single color - octet parameter @xmath30 .
there are analogous matrix elements that describe the direct production of @xmath31 and @xmath32 .
the nrqcd factorization formula ( [ sig - nrqcd ] ) for direct @xmath33 production reduces to @xmath34 & = & d \sigma[b \bar b_1(^3p_j ) ] \langle o^{\chi_{bj}(np)}_1(^3p_j ) \rangle + ( 2j+1 ) d \sigma[b \bar b_8(^3s_1 ) ] \langle o^{\chi_{b0}(np)}_8(^3s_1 ) \rangle .
\label{sig - chi}\end{aligned}\ ] ] in the last term , the factor of @xmath24 comes from using a spin - symmetry relation to eliminate @xmath35 in favor of @xmath36 .
we can also use a spin - symmetry relation to replace @xmath37 in the first term by @xmath38 .
the matrix elements for @xmath22 and @xmath33 enumerated above should be sufficient for a quantitative description of the production of s - wave and p - wave bottomonium states .
the nrqcd factorization formula gives the cross section for the _ direct _ production of a given bottomonium state .
the cross sections that are most easily measured in experiments are _ inclusive _ cross sections that include contributions from the direct production of higher bottomonium states which subsequently decay into the given state .
for example , the feeddown from @xmath5 , @xmath2 , and @xmath6 accounts for roughly 27% , 11% , and 11% of the @xmath1 cross section , respectively @xcite .
taking into account the feeddown from higher @xmath39 and @xmath40 states , the cross section for inclusive @xmath22 production can be written @xmath23_{\rm inc } & = & d \sigma[b \bar b_1(^3s_1 ) ] \langle o_1(^3s_1 ) \rangle^{\upsilon(ns)}_{\rm inc } + \sum_j d \sigma[b \bar b_1(^3p_j ) ] \langle o_1(^3p_j ) \rangle^{\upsilon(ns)}_{\rm inc } \nonumber \\ & & + d \sigma[b \bar b_8(^3s_1 ) ] \langle o_8(^3s_1 ) \rangle^{\upsilon(ns)}_{\rm inc } + d \sigma[b \bar b_8(^1s_0 ) ] \langle o_8(^1s_0 ) \rangle^{\upsilon(ns)}_{\rm inc } \nonumber \\ & & + \left ( \sum_j ( 2j+1 ) d \sigma[b \bar b_8(^3p_j ) ] \right ) \langle
o_8(^3p_0 ) \rangle^{\upsilon(ns)}_{\rm inc } , \label{sig - total}\end{aligned}\ ] ] where the inclusive nrqcd matrix elements " are @xmath41 \rangle^{\upsilon(ns)}_{\rm inc } \;=\ ; \sum_h b_{h \to \upsilon(ns ) } \langle o^h[n ] \rangle .
\label{o - total}\ ] ] the sum over @xmath14 includes @xmath22 and all higher bottomonium states that can make transitions to @xmath22 .
the coefficient @xmath42 is the inclusive branching fraction for @xmath14 to decay into @xmath43 . by convention
, we define @xmath44 . the inclusive branching fraction for the observed bottomonium states
are collected in table [ tab : branch ] .
these numbers were obtained by combining the measured branching fractions for the exclusive decays @xmath45 , @xmath46 , and @xmath47 , with the exception of @xmath48 , which is a direct measurement @xcite . . [
tab : branch ] inclusive branching fractions @xmath42 ( in % ) for transitions between spin - triplet bottomonium states .
the entries `` 0 ? '' in the first row indicate that the feeddown from @xmath32 is neglected in our analysis . [
cols="<,^,^,^,^,^,^,^,^,^",options="header " , ] we can make a rough estimate of the cross sections integrated over all @xmath4 by assuming that the spin - singlet cross sections have the same shape at small @xmath4 as the @xmath1 cross section .
the measured inclusive @xmath1 cross section at central rapidity integrated over all @xmath4 up to 20 gev satisfies @xmath49 pb , where @xmath50 is the branching fraction of @xmath1 into @xmath51 . the cross section integrated only over @xmath52 gev satisfies @xmath53 pb . the ratio of these cross sections is @xmath54 . multiplying the inclusive @xmath1 cross section @xmath55 nb by the factor of 6.5 to take into account the small @xmath4 region and by the ratio 4.3 from table [ tab : rincl ] , we find that the cross section for @xmath9 integrated over all @xmath4 should be approximately @xmath56 nb .
the cross section for @xmath9 indicates that this state must have been produced in abundance in run i of the tevatron .
however the @xmath9 can be observed only if it has a large enough branching fraction into a decay mode that can be triggered upon .
one possibility is the double-@xmath57 decay @xmath58 , followed by the decays @xmath59 .
the decay @xmath58 has essentially the same kinematics as the decay @xmath60 , except that all masses are scaled up by a factor of 3 .
thus the branching fraction for @xmath58 could be as large as that for @xmath60 , which is approximately @xmath61 .
we can obtain a lower bound on the branching fraction by using the fact that in the limit @xmath62 with @xmath63 fixed , the branching fraction for @xmath58 scales like @xmath64 @xcite .
if @xmath63 and @xmath65 were both in this scaling region , then the branching fraction into light @xmath66 mesons would be smaller for @xmath9 than for @xmath67 by a factor of @xmath68 , which is about @xmath69 . since we are not deep into this scaling region , the suppression should be smaller than this .
thus the branching fraction for @xmath58 should be in the range between @xmath70 and @xmath61 . multiplying by the branching fractions of 6% for each of the decays
@xmath59 , our estimate for the branching fraction for @xmath71 is @xmath72 .
the cross section for producing this particular decay mode of @xmath9 is therefore @xmath73 pb , give or take a factor of 10 .
multiplying by the rapidity interval 0.8 and by the integrated luminosity of about 100 pb@xmath74 in run i of the tevatron , we obtain between 16 and 1600 produced events .
we must also take into account the acceptances and efficiencies for observing the decays @xmath59 .
these can be estimated using the cdf data on the production of prompt @xmath57 in run ia of the tevatron @xcite .
based on the observation of about 22,000 @xmath59 candidates with @xmath75 gev and pseudorapidity @xmath76 in a 15 pb@xmath74 data sample , they measured the cross section in that region of @xmath4 and @xmath77 to be @xmath78 nb .
we infer that the product of the acceptance and the efficiency is roughly @xmath79 . multiplying the number of events that are produced by @xmath80
, we get between 0.13 and 13 observed events . thus this cross section may be large enough to be observed in run i of the tevatron . in run
ii , the integrated luminosity will be larger by a factor of 20 and there will be significant improvements in the acceptances and efficiencies for observing muons in both the cdf and d0 detectors .
thus the @xmath9 should certainly be observed in run ii through the decay @xmath10 .
we have carried out an updated nrqcd analysis of the cdf data on the production of spin - triplet bottomonium states from run i of the tevatron . in spite of using only the data from @xmath52 gev
, we were able to extract all the relevant color - octet matrix elements directly from the data .
only one of the 8 color - octet matrix elements comes out with a negative central value , but several others are also consistent with zero to within errors . in our analysis
, we distinguished between the inclusive color - octet matrix elements that can be used to compute inclusive @xmath22 cross sections and the direct color - octet matrix elements required to compute direct @xmath22 cross sections and , by spin symmetry , direct @xmath7 cross sections .
the most serious deficiency in our analysis was our failure to take into account the effects of soft - gluon radiation that are needed to give a smooth @xmath4 distribution at small @xmath4 .
this forced us to use only the small fraction of the data from @xmath52 gev to fit the color - octet matrix elements , which led to large errors in these matrix elements .
if these matrix elements are used to predict bottomonium cross sections in other high energy processes , the predictions will probably have large error bars .
our theoretical cross sections also diverge from the cdf data below @xmath81 gev , which gives us another reason to be cautious in applying our matrix elements to other high energy processes .
an analysis that deals properly with the small @xmath4 region could take full advantage of the cdf data and therefore determine the color - octet matrix elements more accurately .
such an analysis requires a prescription for combining the leading - order cross sections for @xmath82 with the next - to - leading order cross sections for @xmath83 recently calculated by petrelli et al .
@xcite to get a smooth @xmath4 distribution near @xmath84 .
the matrix elements extracted from such an analysis should give reliable predictions for observables at the tevatron that are dominated by low @xmath4 .
they should also give reliable predictions for bottomonium production in other high energy processes .
our analysis should give reliable predictions for the cross sections of the spin - singlet states @xmath7 and @xmath8 at the tevatron .
we find that the direct cross section for @xmath9 at @xmath52 gev should be greater than the inclusive cross section for @xmath1 by a factor of about 4.3 .
if we assume that the @xmath4 distributions have the same shapes at smaller @xmath4 , we can estimate the cross section for @xmath9 integrated over all @xmath4 .
the resulting cross section is large enough that it should be possible to discover the @xmath9 at the tevatron if there is a decay mode with a large enough branching fraction that can be used as a trigger .
we argued that the decay @xmath10 should allow the discovery of the @xmath9 , if not in the data from run i , then certainly in run ii of the tevatron .
this work was supported in part by the u.s .
department of energy under grants de - fg02 - 91-er40690 and doe - er-40682 - 143 and by the natural sciences and engineering research council of canada .
we thank f. maltoni for valuable comments .
penin and a.a pivovarov , nucl . phys . *
b549 * , 217 ( 1999 ) ; phys .
lett * b435 * , 413 ( 1998 ) ; a. hoang , phys . rev .
d * 61 * , 034005 ( 2000 ) ; k. melnikov and a. yelkhovsky , phys . rev .
d * 59 * , 114009 ( 1999 ) ; m. beneke and a. signer , phys .
lett * b471 * , 233 ( 1999 ) ; m. beneke , a. signer , and v.a .
smirnov , hep - ph/9906476 . | recent data from the cdf collaboration on the production of spin - triplet bottomonium states at the tevatron @xmath0 collider are analyzed within the nrqcd factorization formalism .
the color - singlet matrix elements are determined from electromagnetic decays and from potential models .
the color - octet matrix elements are determined by fitting the cdf data on the cross sections for @xmath1 , @xmath2 , and @xmath3 at large @xmath4 and the fractions of @xmath1 coming from @xmath5 and @xmath6 .
we use the resulting matrix elements to predict the cross sections at the tevatron for the spin - singlet states @xmath7 and @xmath8 .
we argue that @xmath9 should be observable in run ii through the decay @xmath10 . |
the cause of the accelerated cosmic expansion @xcite remains elusive .
one possibility is quintessence , in which the acceleration is driven by the potential energy of a scalar field displaced from the minimum of its potential .
the idea that late - time acceleration could be connected to a new fundamental scalar was suggested in refs .
@xcite ( see ref .
@xcite for a comprehensive review ) and developed into viable models in refs .
alternative gravity solutions to cosmic acceleration or models based upon large extra dimensions also generally require some parameter to be tuned to be extremely small @xcite .
string landscape @xcite and/or anthropic arguments @xcite explain the small value of the cosmological constant by supposing a universe with a cosmological constant of this value just happens to be the one , of @xmath14 , that allows intelligent observers .
quintessence models do not generically address the `` why now ? ''
i.e. , why the universe transitions from decelerated expansion to accelerated expansion only fairly recently , after the universe has cooled 30 orders of magnitude below the planck temperature .
there are so called tracker models , " @xcite in which the functional form of the quintessence potential causes its energy density to scale as the dominant component until late times , when its equation - of - state parameter @xmath15 and it comes to dominate the cosmic expansion and drive late - time acceleration .
these models obviate the need for finely tuned initial conditions but still require a finely tuned overall energy scale for quintessence @xcite .
other quintessence models typically require a potential that is extremely flat .
this latter problem may be solved @xcite if the quintessence field is an axion - like field , in which case the shift symmetry of the axion potential protects the extraordinary flatness required of the quintessence potential .
this solution , though , still requires either that the axion decay constant have a super - planckian value , something that violates the gravity - as - the - weakest force conjecture @xcite , or that the initial axion misalignment angle is extremely close to the value @xmath3 that maximizes the potential .
this latter option is considered unappealing as it replaces the fine - tuning of a cosmological constant with the fine tuning of some other parameter .
the hypothesis that quintessence is an axion - like field also requires that the dark - energy density today must still be put in by hand . in recent work @xcite ,
several of us suggested that some of the remaining problems with axion - like quintessence may be addressed in the string axiverse @xcite .
string theories may lead to a family of @xmath16 axion - like fields with masses that span a huge number of decades @xcite . in ref .
@xcite , a scenario was proposed in which these axions have masses that are distributed roughly evenly per decade of axion mass . in such models
, ultralight axions can furnish dark matter _ and _ dark energy candidates .
the theoretical motivation for and phenomenology of ultralight axions is reviewed at length in ref .
@xcite . in this scenario , in each decade of the cosmic expansion , one of these axions becomes dynamical and has some small chance to drive an accelerated expansion .
this chance is determined by the initial value , assumed to be selected at random , of the axion misalignment angle .
it was shown that with reasonable parameter selections , there is a one in @xmath17100 chance that the universe will expand by @xmath18 decades in scale factor before it undergoes accelerated expansion .
the scenario thus explains quintessence in terms of an axion - like field with dimensionless parameters that never differ from unity by more than an order of magnitude and also addresses the `` why now ? ''
problem .
while the scenario may lead to a variety of observational consequences , the most basic prediction is that dark energy is an axion - like quintessence field with a decay constant very close to , but just below , the planck mass .
the purpose of this paper is to explore the observational consequences of this type of quintessence theory , to impose constraints to the parameter space from current measurements , and to forecast the sensitivity of future measurements to the remaining regions of the model parameter space .
it is found that current cosmic microwave background measurements reduce the allowed parameter space to a fraction @xmath10 of the prior volume , while future measurements of the baryon acoustic oscillation scale by galaxy surveys could distinguish all but a fraction of @xmath12 of the parameter - space volume ( allowed by simple priors ) from the @xmath13cdm model .
on the other hand , it is found that if constraints are projected onto the two - dimensional space of axion mass @xmath1 and initial misalignment angle @xmath4 , the sensitivity of future experiments improves only modestly upon current constraints , assuming that the correct underlying dark - energy model is in fact the fiducial @xmath13cdm scenario .
we begin in sec .
[ sec : model ] by introducing the axiverse inspired quintessence model that is the focus of this paper . in sec .
[ sec : dynamics ] we obtain the relevant equations of motion , and discuss the resulting expansion histories . in sec .
[ sec : conditions ] , we identify axion model parameter choices that give appropriate values of the dark energy density today . in sec .
[ sec : predictions ] , we explore the probability distribution of dark energy density @xmath19 and present - day equation - of - state parameter @xmath20 under flat priors for the axion mass @xmath1 ( in units of @xmath21 ) and initial axion displacement @xmath22 . in sec .
[ sec : currentconstraints ] , we obtain constraints on @xmath1 and @xmath23 for different values of the dimensionless peccei - quinn energy scale @xmath24 using measurements of the cosmic expansion history from the baryon acoustic oscillation scale and the cmb .
we conclude in sec .
[ sec : conclusions ] .
in the appendix , we discuss in detail our treatment of cmb anisotropies in axiverse - inspired quintessence models .
the scenario we consider is one in which cosmic acceleration is due to the slow rolling of a quintessence field @xmath25 down toward the minimum of its potential @xmath26 .
the standard axion - like potential is @xmath27 , \label{eqn : standardpotential}\ ] ] where @xmath0 is the axion decay constant , and @xmath28 is a vacuum - energy density comparable to the dark - energy density today .
below we refer to this standard axion model as model a. in the language of ref .
@xcite , this is a thawing " potential for dark energy @xcite . as noted in ref .
@xcite , in this scenario , the many axions with mass @xmath29 ( which do not contribute to dark energy today ) might dominate the energy density of the universe , overclosing it in the process .
one way around this problem is to invoke the decay of such axions into standard - model particles .
another is to consider a different potential : @xmath27 ^ 3 .
\label{eqn : alteredpotential}\ ] ] we refer to this potential as model b ; it has the added advantage that after beginning to coherently oscillate , higher mass axions become energetically subdominant , leaving the standard cosmological expansion history in place ( up to the time evolution of lighter axion fields , which contribute to the dark energy ) . in either case
, the field @xmath25 takes on values @xmath30 .
we also define the axion mass to be @xmath31 , although the field in the latter potential is in fact massless .
we define the misalignment angle to be @xmath32 and often use it as a proxy for @xmath25 .
the model is also specified by an initial misalignment angle @xmath2 which , as we will see , must be very close to @xmath3 .
we thus define @xmath33 .
the time derivative of @xmath34 is assumed to be zero at time @xmath35 .
for reasons discussed in ref .
@xcite ( and as seen below ) , the decay constant can be written as @xmath36 , where @xmath37 gev is the reduced planck mass , with values @xmath38 . in summary ,
the model is parametrized ( for either potential ) by three quantities , which we take to be ( 1 ) the axion mass parameter @xmath1 , ( 2 ) the decay - constant @xmath39 , and ( 3 ) the initial displacement @xmath40 of the misalignment angle .
the evolution of the field is determined by its equation of motion , @xmath41 where the dot denotes a derivative with respect to time and the prime a derivative with respect to @xmath25 .
the hubble rate @xmath42 is given by the friedmann equation , @xmath43where @xmath44 is the cosmological scale factor . here ,
@xmath45 is the matter ( radiation ) density and the energy density of the axion is given by , @xmath46 the pressure @xmath47 is given by the same equation but an alternate sign in front @xmath48 .
the equation - of - state parameter for this quintessence field is given by , @xmath49 the dependence with redshift @xmath50 is @xmath51 and @xmath52 where @xmath50 is related to the scale factor by @xmath53 .
the radiation energy density @xmath54 includes the contribution of photons and massless neutrinos , while @xmath55 is the energy density of massive neutrinos .
the dependence of their energy density with redshift transitions from photon - like to matter - like when the temperature @xmath56 is comparable to the neutrino mass @xmath57 with numerical solutions obtained using standard expressions for a fermi - dirac distribution ( e.g. refs .
@xcite ) .
it is then convenient to introduce the density parameters @xmath58 where @xmath59 is the critical energy density with present - day value @xmath60 where @xmath61 , and we assume the best - fit planck @xmath13cdm values for the dimensionless hubble constant @xmath62 @xcite .
assuming that the dark matter density and the energy density in the quintessence field are unrelated quantities , we fix the matter content to the cmb - inferred physical baryon and cold dark matter densities using the `` planck tt + lowp '' values of table 3 of ref .
@xcite , @xmath63 and @xmath64 , respectively ; @xmath65 .
massless and one massive neutrino with @xmath66 , for consistency with the choices made in ref .
] finally , we further assume a flat universe and a present day photon temperature @xmath67 k @xcite .
the equations of motion ( [ eqn : eom ] ) together with the scale factor @xmath44 are integrated numerically .
we show the scalar field value @xmath25 using its instantaneous displacement from its maximum - energy value , given by the variable @xmath68 .
we use a high initial redshift @xmath69 whose precise value is numerically inconsequential , but sufficiently large that the quintessence energy density is initially negligible and @xmath70 can be assumed .
figure [ fig : evol ] shows some representative solutions for model a ( left panel ) and model b ( right panel ) as a function of redshift @xmath50 for the model parameters as labeled .
the top row gives the evolution of @xmath23 and the hubble rate in comparison to the @xmath71case .
integration stops when @xmath72 .
the second row shows the matter and axion density parameters , and the third row shows the equation - of - state parameter @xmath20 as well as the overall ratio of total pressure to total energy density of all components .
finally , the angular diameter distance ( for a flat universe ) is given by , @xmath73 } , \end{aligned}\ ] ] where @xmath74 is the cosmic time today ( at redshift @xmath75 ) .
the lines in the bottom row of fig .
[ fig : evol ] show @xmath76 normalized to the angular diameter distance of a flat @xmath77 universe with constant @xmath78 @xcite .
the scenarios depicted by the thick line are visually indistinguishable from @xmath13cdm , and as we see in sec .
[ sec : currentconstraints ] and the appendix , are consistent with observational constraints . at the @xmath23 values shown , for lower @xmath39 values or higher @xmath1 values ,
field evolution is fast enough to distinguish by eye from @xmath13cdm . the deviation of @xmath5 and @xmath6 hint that observations sensitive to the cosmological expansion rate and specific angular - diameter distances ( like measurements of baryon acoustic oscillations and cmb anisotropies ) will test the axiverse - inspired quintessence model . for axion masses
@xmath79 , the field rolls quickly enough to enter the oscillatory regime ( unless @xmath23 is finely tuned ) , spoiling the utility of the axion as a dark - energy candidate , as we see more quantitatively in sec .
[ sec : currentconstraints ] .
the first issue we address is the range of values of @xmath80 that give rise to dark energy domination today .
we begin with a rough analytic argument . in ref .
@xcite , it was argued that @xmath81 is achieved if the slow - roll condition @xmath82 is satisfied .
writing @xmath83 , for @xmath84 , this suggests that @xmath85 will be achieved if @xmath86 , or @xmath87 ; for @xmath88 , this is amended to @xmath89 .
however , evolution of the field from the intial time @xmath90 to the redshifts @xmath91 at which dark energy domination begins implies that the value of @xmath23 at @xmath91 is not necessarily the same as the initial value @xmath80 at @xmath92 .
this evolution becomes significant for @xmath93 , and thus indicates a quintessence field that begins to roll before the present day .
[ tab : one ] .the range of values of the maximal initial misalignment angle @xmath94 and axion mass @xmath1 consistent , for different values of the decay - constant @xmath39 , with a current dark energy equation - of - state parameter @xmath95 and dark energy density @xmath96 . [ cols="^,^,^,^,^ " , ] these arguments can be made more quantitative .
we numerically evolve eqs .
( [ eqn : eom ] ) and ( [ eqn : hubble ] ) , using the values @xmath97 , @xmath98 , @xmath99 , @xmath100 .
we determine a range of values for the dimensionless axion mass @xmath101 by imposing the condition @xmath102 under the no - roll approximation .
we then use the numerical solutions to determine the maximum initial value @xmath103 of the field displacement needed to satisfy the constraint @xmath95 .
the results are shown in table i. the results are relevant for the string axiverse scenario described in ref .
there it was argued that the fraction of the @xmath104 range for the initial misalignment angle that would give rise to cosmic acceleration was @xmath105 .
this _ ansatz _ then led , through further rough calculations , to an estimate for the @xmath106 probability that , for @xmath107 , a given universe would have a dark - energy density like that we observe .
that estimate , though , neglected the early evolution of the scalar field .
we now see , with our more precise calculation , that the real probability drops several orders of magnitude below @xmath108 for @xmath109 ( for model a ) and for @xmath110 ( for model b ) . we thus conclude that the string - axiverse solution to the `` why now '' problem described in ref .
@xcite requires @xmath111 for model a and @xmath112 for model b. our numerical results for @xmath113 also suggest that for these allowable ranges of @xmath39 , the probability that a given universe has a dark - energy density like that we observe ( parameterized by the allowed range of @xmath23 ) is reduced by roughly an order of magnitude relative to the estimates given in ref . @xcite .
of course the conditions @xmath102 and @xmath95 ( which describe the properties of dark energy today ) only crudely approximate the precise observational constraints to dark energy from actual data ( like the galaxy correlation function and cmb power spectra ) , which are more precisely evaluated in sec .
[ sec : currentconstraints ] and appendix [ sec : cmb ] . for
now , we move on to make predictions for cosmological observables using the parameter - space regions identified in table i.
suppose now that dark energy is indeed a quintessence field of the form we consider here .
the string axiverse motivation requires , as we just learned , @xmath111 .
an obvious question is what histories @xmath114 , @xmath5 , and @xmath115 are predicted by the parameter values roughly consistent with this scenario for quintessence , especially in light of the fact that a whole suite of recent / ongoing cosmological measurements @xcite target observables related to the recent history of these variables .
we now address this question . in order to address this question , we fix a value of @xmath39 and then assume that models are distributed uniformly in axion mass @xmath1 and uniformly in the initial misalignment angle @xmath80 , _
i.e. _ , the joint probability distribution of models in the @xmath1-@xmath80 space is @xmath116constant . in certain regions of parameter space there
exists a one - to - one mapping of the @xmath117 parameter space to @xmath118 , where @xmath119 . the distribution of models in the @xmath120 parameter space is then given by the jacobian of the transformation , _ i.e. _ , @xmath121 , which we obtain numerically .
the probability distribution function ( pdf ) @xmath122 that arises for @xmath20 , given a family of models uniformly distributed in @xmath1 and @xmath80 is then , @xmath123 where we begin from a number of @xmath124 prospective dark energy solutions @xmath125 that satisfy the restrictions @xmath126 and @xmath127 following sec .
[ sec : conditions ] .
the expression for @xmath128 is obtained by the replacement @xmath129 in ( [ eqn : wdistribution ] ) .
we also wish to infer the pdf of other quantities , such as distance measures @xmath76 at @xmath130 or @xmath131 ( where @xmath132 is the redshift of decoupling ) .
starting from the same set of generated dark energy solutions , the pdf of an observable @xmath133 is numerically straightforward to obtain by generalizing eq .
( [ eqn : wdistribution ] ) . in the top row of fig .
[ fig : pdf ] we show the pdfs for @xmath20 at redshift @xmath75 that arise for both models a and b for several different values of @xmath39 .
the results indicate that the distribution in @xmath20 covers all values relatively evenly , but with an increase as @xmath134 which is due to an integrable square - root singularity at @xmath134 .
the gray lines show the cumulative probability distribution ( @xmath135 ) for @xmath136 that quantifies the probability of having @xmath20 near @xmath137 among the dark energy candidates ; for example , @xmath138 . we thus conclude that if the empirically allowed window for @xmath20 continues to converge to an ever smaller range around @xmath139 , the string axiverse scenario becomes less attractive in the sense that the scenario does not uniquely predict that @xmath140 . the second row of fig .
[ fig : pdf ] shows the pdfs for @xmath141 .
they are flat , and do not favor a particular value of @xmath141 .
some qualitative understanding might be gained as follows : for a given @xmath39 , all dark energy candidate solutions have approximately the same mass @xmath1 that is small enough that little evolution of the field has yet occurred , but big enough to yield @xmath142 .
for example , in model a and in the limit of neglecting the field evolution altogether , the energy density is proportional to @xmath143 . for constant @xmath144 and @xmath39
, @xmath145 is hence distributed as @xmath146 ( the pdf of @xmath147 ) and a relatively flat distribution is not surprising .
if @xmath23 and @xmath1 are both drawn from a uniform distribution , @xmath145 still has a relatively flat distribution ( @xmath148 away from the edges of the prior on @xmath23 and @xmath1 ) .
the same argument holds roughly for model b , although the field evolution is more dramatic for this scenario . ratios of the hubble parameter @xmath149 to @xmath13cdm values are shown in fig .
[ fig : pdf_hubble ] , while the analogous plots for @xmath150 are shown in the upper row of fig .
[ fig : pdf_dang ] , along with the equivalent plots for @xmath151 in the lowest row .
common to all pdfs is that they are clustered around the fiducial @xmath71 case ( by construction ) and deviate from it by no more than @xmath152 .
such deviations fall within the reach of completed surveys like boss @xcite and the projected sensitivities of galaxy redshift surveys by wfirst @xcite , ska @xcite , and spherex @xcite . in the next section
we explore this sensitivity and derive constraints from existing observations .
in the analysis above , we roughly ( and conservatively ) identified the dark energy parameter space as @xmath95 and @xmath153 , and derived predictions for the distribution of prospective observables in that region .
the actual allowed region of parameter space , however , can be constrained more precisely , using quantities [ like @xmath115 and @xmath5 ] which are more straightforwardly related to observables . when this is done , the allowed region in @xmath120 parameter space is not rectangular , as we have approximated so far . in this section ,
we impose constraints to the model and find the implications for the @xmath155 parameter space using precise cosmological data .
we treat the @xmath13cdm cosmology with planck fiducial values as our null hypothesis @xcite or fiducial cosmology . in particular , the cosmic sound horizon at decoupling ( of the baryon - photon plasma )
sets the scale of the bao feature in the galaxy power spectrum , as well as the angular scale of the cmb acoustic peaks .
below , we use both quantities to test the possibility of axiverse - inspired quintessence models .
constraints to the model a case were also obtained in ref .
@xcite . in that work
the effect of quintessence perturbations was not considered , more restrictive priors on @xmath141 were used , and the degree of fine tuning in @xmath156 is not explicitly stated .
our work also goes beyond that work in exploring the power of future experiments . given a parameter selection ,
the model makes specific predictions for the time evolution of @xmath20 that can be constrained . if there is no energy transfer between the quintessence field and another component ( as assumed in this work )
, energy conservation @xmath157 implies , @xmath158 .
\end{aligned}\ ] ] in the simplest case , @xmath20=const , the right - hand side reduces to @xmath159 and @xmath160 corresponds to a cosmological constant .
we could impose constraints to our axiverse - inspired quintessence models by determining which @xmath23 , @xmath1 , and @xmath39 values are consistent with constraints to @xmath20 .
this approach , however , neglects the ( model - dependent ) dynamics of @xmath20 and would lead to biases in the results , as well as an underestimate of the error bars .
one could also consider a simple two - parameter model , with @xmath161 . here
@xmath162 is a parameter that encodes the time dependence of the quintessence field @xcite , but this parameterization is not accurate for dark energy models in general or axion models in particular . instead , we compute @xmath115 and @xmath5 ( which are more directly related to survey observables ) at various redshifts for a grid of models in the @xmath163 parameter space and compare with survey data . with the mapping of vast volumes of our universe , measurements of galaxy clustering and the bao feature embedded in them have proved to be a powerful tool for constraining the cosmic expansion history and various cosmological parameters ( see e.g. refs .
the characteristic length scale that is imprinted by sound waves in the early universe can be computed as the comoving sound horizon @xmath164 at the drag epoch with redshift @xmath165 , not to be confused with the sound horizon @xmath166 at decoupling . by measuring the sound horizon in the galaxy two - point correlation function , one
can then infer @xmath115 and @xmath5 at low redshift @xmath50 . roughly speaking ( for details ,
see the detailed discussion of refs . @xcite and references therein ) , galaxy separations along the line of sight depend on @xmath167 ( corresponding to differences in redshift ) and separations transverse to the line of sight depend on @xmath168 ( corresponding to differences in angular position ) .
we also compute the sound horizon @xmath169 at the redshift of decoupling , @xmath131 .
this quantity can be related to the angular sound horizon at decoupling @xcite , @xmath170 this quantity sets the angular scale of acoustic peaks in the cmb acoustic peaks , and is one of the most precisely measured numbers in cosmology . by altering the late - time cosmic expansion history ,
axion dark energy will also change @xmath151 , the angular - diameter distance to the surface of last scattering .
this will alter the angular scale of the cmb acoustic peaks , yielding an additional cmb constraint to axion dark energy .
for some model parameter values , the physical sound horizon @xmath166 at decoupling changes ( very modestly ) , altering the dependence of the angular acoustic scale on axion model parameters .
as we show in appendix [ sec : cmb ] , computing the effect of axiverse - inspired quintessence models on @xmath171 is a very accurate ( and much faster ) proxy for a full computation of cmb power spectra , and can thus be efficiently used to infer cmb constraints to such models .
we now quantify the discriminating power of the various cosmological observables for the axiverse - inspired models that satisfy the basic requirements @xmath172 and @xmath173 .
we obtain constraints from current data , and then assess the sensitivity of future experiments .
figure [ fig : hist ] shows histograms [ probability distribution functions ( pdfs ) and cumulative distribution functions ( cdfs ) ] of the predictions of the models and how they fare in comparison to boss observed values of @xmath76 and @xmath42 at redshift @xmath174 and to the planck 2015 inferred angular scale of the sound horizon at decoupling . for the central values and errors of the bao observables we use measurements of the angular diameter distance to and hubble parameter at @xmath175 from the measurement of the bao peak in the correlation of galaxies from the sloan digital sky survey
iii baryon oscillation spectroscopic survey @xcite , @xmath176 where @xmath177 @xcite and @xmath178 for our assumed @xmath13cdm fiducial model .
for other models in the axion parameter space @xmath179 takes on different values and we compute those self - consistently .
we include the covariance between these quantities @xcite when determining if a given axion model is inside or outside the allowed region at the observable level . for planck ,
we are conservative and use the value @xmath180 stated in ref .
@xcite , based on temperature data at all scales and low @xmath181 polarization data ( as opposed to the full polarization data set ) . for model a , we swept the parameter space on a logarithmic grid satisfying @xmath182 and @xmath183 , with 81 @xmath40 values and @xmath184 @xmath39 values .
we then obtain central @xmath185 value for the axion mass by demanding that @xmath186 in the no - roll limit , which implies that @xmath187 .
we then define a scan range of @xmath188 @xmath1 values , linearly spaced with @xmath189 . for model
b , the same density condition implies that @xmath190 , with @xmath191 .
here we use the range @xmath192 , with the same number of grid points as for model a , in all dimensions .
we check that these values ensure a full sweep of the allowed parameter space .
we ran a full numerical integration of the equations of motion [ eqs .
( [ eqn : eom ] ) and ( [ eqn : hubble ] ) ] for all @xmath193 models in this grid and then compare the values of @xmath6 , @xmath5 and @xmath171 with the boss and planck confidence regions described above .
we also ran this integration within a modified version of the camb code , described in appendix [ sec : cmb ] , in order to self - consistently compute the ( small ) change in @xmath171 in our model that results from modified early - time evolution of the ( highly sub - dominant ) dark - energy component , including small corrections to the value of @xmath132 .
we found that our results are insensitive to this correction .
constraints from current measurements and predictions of future sensitivity are shown in fig .
[ fig : hist ] . for the duration of this ( parameter - space volume ) discussion
, @xmath6 are @xmath5 are treated simply , as uncorrelated independent observables .
correlations are considered more carefully below , when detailed parameter constraints to @xmath163 are obtained . for model
a ( b ) , at the @xmath194 confidence level , @xmath195 ( @xmath196 ) of the models ( in our grid ) consistent with dark energy priors are allowed by the @xmath197 measurement , while only @xmath198 ( @xmath199 ) of the models are allowed by the @xmath200 measurement .
this is driven by the well known @xmath194 tension between boss results and planck fiducial @xmath13cdm values @xcite . for model a ( b ) , at the @xmath201 confidence level ,
@xmath202 ( all ) of the models consistent with the dark energy prior ( @xmath203 and @xmath204 ) are allowed by the @xmath197 measurement , while @xmath205 ( @xmath206 ) of the models are consistent with the @xmath200 measurement .
finally , cmb measurements are more discriminating . for model a ( b ) , at the @xmath201 confidence level ,
only @xmath207 ( @xmath208 ) of the models are consistent with planck constraints .
in addition to the high precision bao measurements at @xmath174 from galaxy clustering , there are boss lyman-@xmath39 forest ( lyaf ) data that test the expansion history at higher redshift as well .
the latter data show a @xmath209 tension with a cmb - supported standard cosmology , with lyaf results favoring a lower hubble rate and smaller angular diameter distance at @xmath210 @xcite .
it is worthwhile to consider whether or not axiverse - inspired quintessence models can relieve that tension , while preserving agreement with bao data at z = @xmath211 .
it turns out that they can not , as we obtain no data points ( for model a or b ) that agree with lyaf ( fig .
15 of ref .
@xcite ) constraints and measurements of @xmath197 and @xmath200 .
of course this tension could be a statistical fluke , or the result of some unrecognized systematic error @xcite . assuming a fiducial planck @xmath13cdm cosmology
, we also determine how future galaxy surveys could improve on those constraints , by performing a standard fisher forecast @xcite of the errors on the observed galaxy power spectrum ( at @xmath174 ) for the proposed spherex satellite @xcite and the square kilometre array ( ska ) @xcite ( at @xmath130 ) .
these results are obtained from a standard bao forecast prescription @xcite and are shown in the bottom panel of fig .
[ fig : hist ] . for model
a and b , at the @xmath201 confidence level , we see that future @xmath197 measurements from ska will be comparably sensitive to current cmb constraints , potentially distinguishing all but @xmath212 of the parameter - space volume ( allowed by our prior ) from the @xmath13cdm model .
the next major improvement in sensitivity will come from spherex constraints to @xmath200 , which for model a ( b ) will distinguish all but @xmath213 ( @xmath214 ) of the parameter - space volume ( allowed by our prior ) from the @xmath13cdm model
. a future cosmic - variance limited cmb experiment ( like cmb stage-4 @xcite ) with fisher - level sensitivity of @xmath215 could observationally distinguish ( from the @xmath13cdm model ) a comparable parameter - space volume ( at @xmath201 confidence - level ) to spherex measurements of @xmath200 .
the analogous measurement using the ska could empirically distinguish ( from the @xmath13cdm model ) the whole parameter - space volume allowed by our prior , up to sampling uncertainties , which allow @xmath216 of the parameter - space volume to potentially evade constraints .
forthcoming cosmological observations could thus either validate axiverse - inspired quintessence models or show that they are extremely fine - tuned at the @xmath216 level ( or worse ) in terms of parameter - space volume .
we now go beyond raw volume measurements to examine the detailed constraints in @xmath155 parameter space . here
we show the allowed parameter space at @xmath217 confidence level .
results here are shown without the @xmath102 , @xmath218 , ` dark energy ' prior , to see how much the data alone constrain the quintessence parameter space .
the allowed region for @xmath1 sits well within the scan range described above , so even though this scan range amounts to a nontrivial ( but linearly flat ) prior , our conclusions do not depend on this prior .
we then resort the constraints as a function @xmath219 and use a third - order polynomial fit to densely interpolate between points to get ( approximately ) continuous physical allowed regions .
we consider the same set of current / future experiments , with the addition of the wfirst satellite @xcite ( considering a measurement of the bao feature at @xmath220 ) . for model a , we show the allowed region of @xmath155 in fig .
[ fig : d_a - h_z057-cmb_bossfid ] , as constrained by boss ( top left panel ) and planck ( bottom left panel ) measurements of @xmath221 at @xmath174 and @xmath171 respectively . predicted constraints ( under the null hypothesis ) from the spherex satellite @xcite are shown in the top right panel .
the bottom right panel illustrates a comparison between the allowed regions from boss , planck , and spherex ( predicted ) measurements .
we perform the same analysis for model b , with results shown in fig .
[ fig : d_a - h_z057-cmb_bossfid_b ] . the main difference between models a and
b is that at the same mass and peccei - quinn scale , model b is considerably more finely tuned in @xmath23 .
this trend subsists in the forecasted constraints from future data , discussed below ( for the null hypothesis ) .
we can see that there is a common general behavior , that as @xmath39 decreases , the allowed mass increases and the allowed range of misalignment angles decreases . for sub - planckian peccei - quinn scale @xmath222 , the initial misalignment angle must be finely tuned to lie near the top of the quintessence potential , @xmath223 , and the axion mass @xmath224 .
qualitatively , these results arise from the requirement that the quintessence field have sufficient energy density to explain current cosmic acceleration , without spoiling the agreement of the standard cosmology at higher redshifts , where nonrelativistic matter must dominate the cosmic expansion history .
the area of the parameter space that is allowed from observational constraints is very similar when comparing current bao and cmb measurements .
increasing the precision in bao measurements shrinks the projection of the allowed region onto two dimensions , but not dramatically .
we then investigated if having measurements of the bao peak at higher redshift can help reduce the area of the allowed region .
we predicted values of @xmath221 at @xmath130 and @xmath220 using our full grid of axion models , and then compared with a fisher forecast of the error bars of future galaxy surveys produced by the spherex satellite , the square kilometre array ( ska ) ( for a specific ska - oriented bao measurement prediction , see ref .
@xcite ) , and the wfirst satellite mission .
we predict the expected @xmath217 confidence constraints to the @xmath225 parameter set .
the results are shown in fig .
[ fig : future ] for model a , and in fig .
[ fig : future_b ] for model b. qualitatively , we see that if future experiments are consistent with the null hypothesis , the allowed region ( as parameterized by the angle @xmath40 ) will become increasingly finely tuned . on the other hand ,
future experiments will not be _ dramatically _ more sensitive to any @xmath226 of @xmath227 than current measurements , in spite of the increasing precision in @xmath221 measurements and the dramatic decrease in allowed parameter - space _ volume_. we can see why this happens in fig .
[ fig : ellipse ] , where we show the @xmath217 confidence level ellipse ( in @xmath76 and @xmath42 , divided by their fiducial values ) , for a futuristic experiment probing @xmath76 and @xmath42 at @xmath228 precision . the points in fig .
[ fig : ellipse ] show predictions for different choices of parameters for the axion model . for each point ,
the color indicates the value of @xmath23 , the symbol indicates the value of @xmath1 , while the size of the symbol is proportional to the value of @xmath39 .
we see that it is possible to get very close to the central values of @xmath229 and @xmath230 and still find allowed models with no dramatic change in the @xmath231 range consistent with the data . here
@xmath232 denotes values computed at fiducial ( planck ) @xmath13cdm parameter values .
an investigation of other possible observables affected by axiverse models is left for future work .
so far , we have treated cmb probes of axiverse - inspired quintessence models using a simple constraint from the angular acoustic horizon @xmath171 . in appendix
[ sec : cmb ] , we verify that this simple treatment is equivalent to the result that would have been obtained through a full computation of cmb anisotropies in axiverse - inspired quintessence models . +
by positing the existence of many light fields , the axiverse scenario raises the possibility that today s epoch of accelerated cosmic expansion results from a moderate fine tuning , in which today s acceleration occurs in a fraction @xmath233 of cosmological draws of the axion misalignment angle @xmath23 , as long as the dimensionless peccei - quinn scale @xmath234 .
this is a vast improvement with respect to the required fine tuning of quintessence models , which do not generically address the
" problem ( tracker " models offer some improvement by eliminating the need for finely tuned initial conditions @xcite , but they still require a finely tuned overall quintessence energy scale @xcite ) .
the axiverse scenario motivates a family of quintessence models in which the field displacement from maxima of anharmonic potentials must be relatively small . after discussing the dynamics of these models
, we showed that the dark energy equation - of - state , hubble parameter at a variety of redshifts , angular - diameter distances , and cmb angular sound horizon deviate sufficiently from their fiducial values to allow precision cosmological probes of axiverse - inspired quintessence .
we proceed to test these scenarios using two very different measurements of the baryon - photon sound horizon , the bao feature measured in galaxy surveys and the angular scale of the cmb acoustic peaks . in terms of parameter space volume , the cmb is dramatically more constraining , eliminating a fraction @xmath235 of models that pass a well motivated dark energy prior . in the projected parameter space of @xmath1 and @xmath23 ( for sub - planckian @xmath236 ) , we find that the cmb and bao measurements constrain axiverse - inspired quintessence models to comparable _ areas _ in parameter space . in the future
, cosmic - variance limited measurements of cmb polarization and very sensitive bao surveys will be dramatically more sensitive , empirically distinguishing ( from the @xmath13cdm model ) all but @xmath12 of the parameter space allowed by dark energy priors .
the constraints presented in this work were derived simply , varying the quintessence parameters of interest without performing a full cosmological monte carlo markov chain . in future work
, we will pursue such techniques to explicitly account for priors on all parameters , and consider a variety of other observables , such as the galaxy - cmb cross - correlation ( which may be more sensitive to the quintessence - induced isw effect than the cmb temperature power spectrum ) , measurements of the cosmic expansion history from type ia supernovae , direct kinematic tests of cosmic acceleration , and gravitational lensing ( of the cmb and background galaxies ) .
this work was supported at jhu by nsf grant no.0244990 , nasa nnx15ab18 g , the john templeton foundation , and the simons foundation .
r .. e acknowledges the support of the new college oxford - johns hopkins centre for cosmological studies as well as a grant from research grants council of the hong kong special administrative region , china ( hkust4/crf/13 g ) .
is funded at the university of chicago by a national science foundation astronomy and astrophysics postdoctoral fellowship under award no .
this work was supported in part by the kavli institute for cosmological physics at the university of chicago through frant nsf phy-1125897 and an endowment from the kavli foundation and its founder fred kavli .
j.p . is supported by the new frontiers program of the austrian academy of sciences .
dark energy contributes to the expansion history of the universe but clusters very weakly ( and indeed , is completely smooth in the limit of a pure cosmological constant ) . as a result ,
gravitational potential wells decay on scales that enter the horizon during dark energy domination .
this late - time integrated sachs - wolfe ( isw ) effect imprints additional temperature anisotropies on the cmb at low multipole index @xmath181
. the detailed evolution of @xmath20 in axion models will thus affect the precise shape of both the cmb temperature anisotropy power spectrum @xmath237 and galaxy - cmb temperature cross power spectrum @xmath238 @xcite at low @xmath181 .
although the statistical significance of measurements of these quantities at low @xmath181 is limited by cosmic variance , there could be axion dark - energy models allowed by the constraints of sec .
[ sec : currentconstraints ] that are further constrained by the imprint of the late - time isw effect on @xmath237 and @xmath239 .
the quintessence - induced change to the angular sound horizon @xmath171 drives the cmb constraints / forecasts of sec .
[ sec : currentconstraints ] , but here we explore if this simple geometric constraint captures the full power of the cmb to probe axiverse inspired quintessence models .
we do this by computing in greater detail the effect of axiverse - inspired quintessence on cmb anisotropies . the quintessence field
@xmath25 may depend on spatial location as well as time , and must be permitted to carry spatial perturbations in a self - consistent computation of observable quantities like @xmath237 for a quintessence cosmology .
the relevant equation of motion ( the perturbed klein - gordon equation ) can be rewritten as a pair of fluid equations in synchronous gauge @xcite , to ease their inclusion in the cosmological boltzmann code camb @xcite : @xmath240\dot{h}/2 - 3\mathcal{h}\left[1-w(a)\right]\delta_{\phi}\nonumber \\-&9\mathcal{h}^{2}\left(1-c_{\rm ad}^{2}\right)u_{\phi}/k,\label{eq : eoma}\\ u'_{\phi}=&~2\mathcal{h}u_{\phi}+k \delta_{\phi}+3\mathcal{h}\left(w - c_{\rm ad}^{2}\right)u_{\phi}\label{eq : eomb},\end{aligned}\]]where @xmath241 is the fractional axion energy density perturbation , and the adiabatic sound speed is @xmath242 the dimensionless axion heat flux is @xmath243 where @xmath244 is the scalar axion velocity perturbation .
gravitational potentials enter through the synchronous gauge metric perturbation @xmath245 .
the conformal hubble parameter is given by @xmath246 , where @xmath247 here denotes a derivative with respect to conformal time @xmath248 , defined by @xmath249 .
the gravitational imprint of axions is in turn included through the synchronous gauge einstein equations , with axionic pressure , energy density , and momentum flux contributions @xmath250\label{dpdef},\\ \delta
\rho_{\phi}&=&\rho_{\phi}\delta_{\phi},\\ \left(\rho_\phi+p_{\phi}\right)v_{\phi}&=&\rho_{\phi}u_{\phi}.\label{eq : dqdef}\end{aligned}\]]in recent work by one of us and other collaborators @xcite , these equations were self - consistently implemented in the cmb boltzmann code camb @xcite for a quadratic potential ( in that case , allowing for both slowly rolling axion dark energy and coherently oscillating axion dark matter ) , along with the accompanying equations for the mean axion energy - density and pressure , which affect the recombination and decoupling history of the universe . here
, we have generalized this modified camb @xcite using the full model a and model b anharmonic potentials in eqs .
( [ eqn : standardpotential])-([eqn : alteredpotential ] ) , but restrict ourself to the dark - energy like " part of parameter space in the language of ref . @xcite .
we solve the quintessence perturbation equations of motion numerically .
the axion potential enters only through the evolution of the equation - of - state @xmath251 and axion adiabatic sound speed @xmath252 .
we thus solve eq .
( [ eqn : eom ] ) numerically , evaluating @xmath251 and @xmath253 as described in detail in ref .
these functions are then tabulated and interpolated by our modified version of camb to obtain cosmological observables like @xmath237 . unlike the treatment in ref .
@xcite , we do not solve eqs .
( [ eq : eoma])-([eq : eomb ] ) with a wkb approximation . for both models
a and b , we generate 790 uniformly sampled ( in @xmath254 , @xmath1 , and @xmath255 ) subsets of the full grid of models used to generate constraints in sec . [
sec : currentconstraints ] and generate the resulting cmb power spectra .
one example ( for model a ) is shown in fig . [
fig : cltt ] , which shows the @xmath13cdm @xmath237 power spectrum , along with the results of the full quintessence - modified camb computation , after applying the mapping @xmath256 , where @xmath257 is the angular sound horizon in our fiducial cosmology ( with no axiverse - inspired quintessence ) .
this mapping should undo the change to @xmath237 induced by quintessence - induced changes to the angular sound horizon and isolate the effect induced by the late - time isw effect .
we see that @xmath237 looks almost indistinguishable from the @xmath13cdm case after remapping the sound horizon , with some visibly discernible deviation from the fiducial case for @xmath258 .
we find that the fractional change to @xmath237 is negligible , compared both with cosmic variance and the benchmark for noise - free parameter fitting , @xmath259 @xcite .
we can see that even for this model ( which shows the most dramatic deviation from @xmath13cdm in our subsample after remapping the angular sound horizon ) that the modifications to the late - time isw effect induced by the quintessence field are well below cosmic variance at all scales and hence undetectable .
in other words , the constraints imposed in sec . [ sec : currentconstraints ] using cmb measurements of @xmath171 are robust for axiverse - inspired quintessence models .
we find that remapping of @xmath171 is able to absorb @xmath260 of the change in @xmath261 between the quintessence and @xmath13cdm model , showing that additional changes to @xmath237 from the late - time isw effect are negligible .
the much larger change in the isw effect between axion and @xmath13cdm models seen in ref .
@xcite results from the much wider range of axion masses considered in that work , where axions were considered as both dark matter _ and _ dark energy candidates .
as camb includes many different physical effects , not just the isw effect , it is interesting to consider a simpler calculation that isolates the effect of dark energy on the isw effect . to this end
, we implement the equations of ref .
@xcite for the isw effect sourced by cosmological perturbation modes with a variety of comoving wave numbers @xmath262 .
we assume an arbitrary initial perturbation normalization ( with no additional power spectrum induced scale dependence ) .
we use the same @xmath263 models as above but impose the dark energy " prior discussed above .
we then compare the results to the null hypothesis , with results shown in fig .
[ fig : isw ] for modes roughly corresponding to the multipoles @xmath264 and @xmath265 . the correspondence between wave number
is treated approximately , via @xmath266 .
we see from fig .
[ fig : isw ] that the @xmath267 change in cmb anisotropies at low @xmath181 is much smaller than cosmic variance , confirming our conclusion that the late - time isw induced change to cmb anisotropies from axiverse - inspired quintessence is undetectable .
77ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ]
+ 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1086/300499 [ * * , ( ) ] , link:\doibase 10.1086/307221 [ * * , ( ) ] , link:\doibase 10.1007/bf00759587 [ * * , ( ) ] link:\doibase 10.1007/bf00756088 [ * * , ( ) ] link:\doibase 10.1103/physrevd.26.2580 [ * * , ( ) ] in @noop _ _ , ( , ) p. in @noop _
_ , ( , ) p. in @noop _ _ ( ) link:\doibase 10.1002/prop.19860341204 [ * * ,
( ) ] link:\doibase 10.1016/0375 - 9601(86)90121 - 0 [ * * , ( ) ] link:\doibase 10.1103/physrevd.35.2339 [ * * , ( ) ] link:\doibase 10.1016/0370 - 2693(87)90338 - 8 [ * * , ( ) ] link:\doibase 10.1016/0550 - 3213(88)90192 - 7 [ * * , ( ) ] link:\doibase 10.1016/0550 - 3213(88)90193 - 9 [ * * , ( ) ] link:\doibase 10.1016/0370 - 2693(88)90805 - 2 [ * * , ( ) ] link:\doibase 10.1142/s0217751x88000692 [ * * , ( ) ] link:\doibase 10.1103/physrevd.42.361 [ * * , ( ) ] link:\doibase 10.1016/0550 - 3213(90)90272-f [ * * , ( ) ] link:\doibase 10.1002/asna.2113110307 [ * * , ( ) ] link:\doibase 10.1142/s0218271800000542 [ * * , ( ) ] , link:\doibase 10.1103/revmodphys.75.559 [ * * , ( ) ] , link:\doibase 10.1146/annurev - nucl-010709 - 151330 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.37.3406 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.80.1582 [ * * , ( ) ] , @noop * * , ( ) , link:\doibase 10.1103/physrevd.55.1851 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.56.r4439 [ * * , ( ) ] , link:\doibase 10.1142/s021827180600942x [ * * , ( ) ] , link:\doibase 10.1088/0034 - 4885/72/9/096901 [ * * , ( ) ] , link:\doibase 10.1088/1126 - 6708/2000/06/006 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.59.2607 [ * * , ( ) ] link:\doibase 10.1086/185100 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.75.2077 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.81.3067 [ * * , ( ) ] , link:\doibase 10.1088/1126 - 6708/2007/06/060 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.113.251302 [ * * , ( ) ] , link:\doibase 10.1088/1126 - 6708/2006/06/051 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.81.123530 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1103/physrevlett.95.141301 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.90.105023 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.81.043517 [ * * , ( ) ] , @noop ( ) @noop ( ) , link:\doibase 10.1088/1674 - 1137/38/9/090001 [ * * , ( ) ] @noop ( ) , link:\doibase 10.1088/0034 - 4885/71/5/056901 [ * * , ( ) ] , link:\doibase 10.1146/annurev.astro.46.060407.145243 [ * * , ( ) ] , link:\doibase 10.1016/j.physrep.2013.05.001 [ * * , ( ) ] , link:\doibase 10.1093/mnras / stu523 [ * * , ( ) ] , @noop ( ) , @noop * * , ( ) , @noop ( ) , @noop ( ) , @noop ( ) , link:\doibase 10.1103/physrevlett.90.091301 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1111/j.1365 - 2966.2004.08146.x [ * * , ( ) ] , link:\doibase 10.1038/nature06555 [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2966.2011.19592.x [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2966.2011.20169.x [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2966.2012.21779.x [ * * , ( ) ] , link:\doibase 10.1093/mnras / stu197 [ * * , ( ) ] , link:\doibase 10.1093/mnras / stt1517 [ ( ) , 10.1093/mnras / stt1517 ] , @noop ( ) , link:\doibase 10.1103/physrevd.92.123516 [ * * , ( ) ] , link:\doibase 10.1086/303939 [ * * , ( ) ] , link:\doibase 10.1086/379122 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2014.05.014 [ * * , ( ) ] , in http://inspirehep.net/record/1339611/files/arxiv:1501.04088.pdf[__ ] ( ) link:\doibase 10.1086/306274 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.69.083503 [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2966.2003.07144.x [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1103/physrevd.91.103512 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.68.083507 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.82.083522 [ * * , ( ) ] | inspired by the string axiverse idea , it has been suggested that the recent transition from decelerated to accelerated cosmic expansion is driven by an axion - like quintessence field with a sub - planckian decay constant .
the scenario requires that the axion field be rather near the maximum of its potential , but is less finely tuned than other explanations of cosmic acceleration .
the model is parametrized by an axion decay constant @xmath0 , the axion mass @xmath1 , and an initial misalignment angle @xmath2 which is close to @xmath3 . in order to determine the @xmath1 and @xmath4 values consistent with observations ,
these parameters are mapped onto observables : the hubble parameter @xmath5 at an angular diameter distance @xmath6 to redshift @xmath7 , as well as the angular sound horizon of the cosmic microwave background ( cmb ) .
measurements of the baryon acoustic oscillation ( bao ) scale at @xmath8 by the boss survey and planck measurements of cmb temperature anisotropies are then used to probe the @xmath9 parameter space . with current data ,
cmb constraints are the most powerful , allowing a fraction of only @xmath10 of the parameter - space volume .
measurements of the bao scale made using the spherex or ska experiments could go further , observationally distinguishing all but @xmath11 or @xmath12 of the parameter - space volume ( allowed by simple priors ) from the @xmath13cdm model . |
the bifurcation of limit cycles by perturbing a planar system which has a continuous family of _ cycles _ , i.e. periodic orbits , has been an intensively studied phenomenon ; see for instance @xcite and references therein .
the simplest planar system having a continuous family of cycles is the linear center , and a special family of its perturbations is given by the generalized polynomial linard systems : @xmath0 where @xmath1 , @xmath2 , @xmath3 , @xmath4 and @xmath5 are polynomials for @xmath6 , and @xmath7 is a small parameter .
the classical and generalized linard systems appear very often in several branches of science and engineering , as biology , chemistry , mechanics , electronics , etc .
see for instance @xcite and references therein .
in particular linard systems are frequent specially in physiological processes , see for instance @xcite .
in addition , the family of generalized polynomial linard systems is one of the most considered families in the study of limit cycles , see @xcite .
we assume that @xmath8 , @xmath9 , @xmath10 , and @xmath11 . for a small enough @xmath7 ,
let @xmath12 be the maximum number of limit cycles of that bifurcate from cycles of the _ linear center _
@xmath13 , i.e. the maximum number of _ medium amplitude limit cycles _ which can bifurcate from @xmath13 under the perturbation .
if @xmath14 , then @xmath15 does not depend on @xmath16 ; hence we only write @xmath17 .
the main problem concerning @xmath12 is finding its exact value . we know from @xcite that @xmath18 $ ] , where @xmath19 $ ] denotes the integer part function .
moreover , by following ( * ? ? ?
* theorem 3.1 ) we can prove that @xmath20 $ ] for @xmath21 ; theorem [ mth1 ] ( below ) is a generalization of this result . also , we know from @xcite that @xmath22 $ ] , @xmath23 , \left[m/2\right]+\left[n/2\right]-1\right\}$ ] , and @xmath24 - 1 $ ] .
however , the exact values of @xmath25 , @xmath26 , and @xmath27 were not reported there . in this paper
we give the exact value of @xmath12 for two subfamilies of .
more precisely , we will give the exact value of @xmath28 and @xmath29 , where @xmath28 is the value of @xmath12 by assuming that @xmath4 is odd for @xmath30 , and @xmath29 is the value of @xmath12 by assuming that @xmath5 is even for @xmath31 , where @xmath32 with @xmath33 is the smallest integer such that @xmath34 .
of course , if @xmath35 , then @xmath36 .
our main result is the following : [ mth1 ] @xmath37 @xmath38 $ ] .
@xmath39 @xmath29 is either @xmath40 $ ] if @xmath41 is odd or @xmath42 + \left[\frac{n}{2}\right]-1 $ ] if @xmath41 is even .
the assumptions on @xmath4 and @xmath5 in definitions of @xmath28 and @xmath29 , respectively , are necessary .
otherwise , we can construct systems having more medium amplitude limit cycles , see remark 1 in section [ sec3 ] .
theorem [ mth1](@xmath43 ) is a generalization of theorem 1.1 in @xcite , where the case @xmath44 was considered .
we note that in such a case @xmath45 . hence theorem 1@xmath39 ( ( * ? ? ?
* theorem 1.1 ) ) gives the exact value of @xmath46 .
the proof of theorem [ mth1 ] is based on computing the maximum number of isolated zeros of the first non - vanishing poincar pontryagin melnikov function of the displacement function of , by taking into account the restrictions : @xmath4 odd for @xmath30 and @xmath5 even for @xmath47 , respectively .
the paper is organized as follows . in section [ ppmf ]
we recall the definition of the displacement function of , as well as the algorithm to compute the poincar pontryagin melnikov functions .
preliminary results that allow us to provide elementary proofs of the main results are given in section [ sec3 ] .
finally , in section [ sec4 ] we will prove theorem [ mth1 ] .
the linear center @xmath13 is the hamiltonian system associated to the polynomial @xmath48 ; hence its cycles are the circles @xmath49 with @xmath50 . by using @xmath51 as a parameter , the first return map of can be expressed in terms of @xmath7 and @xmath51 : @xmath52 .
therefore the corresponding _ displacement function _
@xmath53 is analytic for small enough @xmath7 and can be written as the power series in @xmath7 @xmath54 where @xmath55 with @xmath6 is the _ poincar pontryagin melnikov function _ of order @xmath56 , which is defined for @xmath57 .
let @xmath58 with @xmath59 be the first non - vanishing coefficient in .
the zeros of @xmath58 are important in the study of medium amplitude limit cycles of because of the _ poincar pontryagin andronov criterion _ : the maximum number of isolated zeros , counting multiplicities , of @xmath58 is an upper bound for @xmath12 .
furthermore each simple zero @xmath60 of @xmath58 corresponds to one and only one limit cycle of with @xmath7 small enough bifurcating from the cycle @xmath61 . we know from @xcite that @xmath58 has at most @xmath62 $ ] positive zeros , counting multiplicities
. however , this result does not give the value of @xmath12 because the upper bound for @xmath63 depending on @xmath64 , @xmath65 , @xmath41 , and @xmath16 is unknown .
now , we will recall the algorithm to compute the functions @xmath55 .
system can be written as @xmath66 where @xmath67 , or equivalently as @xmath68 as we know , @xmath69 is given by the classical poincar
pontryagin formula @xmath70 the result for computing the higher order poincar pontryagin
melnikov functions is the following : [ th3 ] * ( yakovenko franoise iliev algorithm @xcite , @xcite , @xcite)*. if @xmath71 and @xmath72 , then there are polynomials @xmath73 and @xmath74 such that @xmath75 , and @xmath76 where @xmath77 the proof of this result easily follows from the poincar pontryagin formula , and the ilyashenko gavrilov theorem ( @xcite , @xcite ) : if @xmath78 for all @xmath57 , then @xmath79 , where @xmath80 and @xmath81 are polynomials , and by applying an induction argument . for a detailed proof ,
see for instance @xcite , @xcite . to simplify the computation of the poincar pontryagin
melnikov functions , we will give some properties of @xmath82 .
for computing @xmath83 for we will use the following two elementary lemmas whose proof is omitted .
[ lem1 ] let @xmath84 be a polynomial in the ring @xmath85 $ ] .
we define @xmath86 to be the degree of @xmath84 in @xmath87 $ ] . * for @xmath88 there are homogeneous polynomials @xmath89 $ ] with @xmath90 and @xmath91 , such that @xmath92 or @xmath93 . if @xmath94 , then @xmath95 . * for @xmath96 there are homogeneous polynomials @xmath97 $ ] with @xmath98 and @xmath99 , such that @xmath100 . * for @xmath96 we have @xmath101 [ lem2 ] if @xmath102\right\}$ ] and @xmath103\right\}$ ] , then @xmath104 .
the next two results are straightforward consequences of these two previous lemmas .
[ cor1 ] if @xmath105 , then @xmath106 , @xmath107 with @xmath108 , and @xmath109 .
[ cor2 ] if @xmath110 $ ] , then @xmath111 the following two lemmas will be important in the proof of theorem [ mth1 ] .
[ lem3 ] suppose theorem [ th3 ] . if @xmath112 for @xmath113 , then @xmath114 for @xmath113 , and @xmath115 we proceed by induction on @xmath63 . if @xmath116 , then @xmath117 .
hence @xmath118 , @xmath119 , and @xmath120 by corollary [ cor1 ] .
since @xmath121 because of theorem [ th3 ] , @xmath122 .
we assume that the lemma is true for @xmath123 , and we will prove it for @xmath63 . by assumption ,
@xmath112 for @xmath124 .
then , by corollary [ cor1 ] , @xmath125 with @xmath126 for all @xmath124 .
in addition , by the induction hypothesis , @xmath127 for @xmath128 . thus , @xmath129 with @xmath130 is an element of @xmath131 following lemma [ lem2 ] . since @xmath132 , @xmath133 .
hence it is clear that @xmath134 with @xmath130 is an element of @xmath131 , which implies that @xmath135 by corollary [ cor1 ] .
finally , from theorem [ th3 ] we have @xmath136
. therefore @xmath137 .
before announce next lemma , we note that each polynomial @xmath138 of degree @xmath139 can be written as @xmath140}a_{2r+1}x^{2r } , \quad \mbox{and } \quad \tilde{h}{\left(x^{2}\right)}= \sum_{r=0}^{\left[\frac{m-2}{2}\right]}a_{2r+2}x^{2r}.\ ] ] [ lem4 ] let @xmath141 , where @xmath142 and @xmath143 . *
@xmath144}{2(r+1 ) \choose r+1}\frac { a_{2r+1}}{2^{r}(2r+1)}{c^{r}}\right ) .
$ ] * if @xmath145 , then @xmath146 with @xmath147 $ ] of degree @xmath148 $ ] , and @xmath149 } \left(\sum_{r=0}^{\left[\frac{m-2}{2}\right]}{2(s+r+1 ) \choose s+r+1 } \frac{{\left(b_{2s}\right ) } ( a_{2r+2})}{2^{s+r}(2s+1)}{c^{s+r } } \right).\ ] ] * @xmath150 if and only if @xmath151 , or @xmath152 .
( @xmath153 ) . by ( @xmath153 ) and ( @xmath43 ) of lemma [ lem1 ] , @xmath154 .
finally , the statement follows from corollary [ cor2 ] .
( @xmath43 ) .
if @xmath155 , then @xmath156 by statement ( @xmath153 ) .
this property implies that @xmath157}a_{2r+2}x^{2r+1}y dx$ ] by . from lemma [ lem1](@xmath43 )
we obtain @xmath158 , thus @xmath159}a_{2r+2 } \overline{q}_{r}\right)\right)+\left(y\left(\sum_{r=0}^{\left[\frac{m-2}{2 } \right]}a_{2r+2}\overline{q}_{r}\right)\right ) dh = dq+\left(y\overline{q}\right ) dh,\ ] ] where @xmath160}$ ] are homogeneous and @xmath161 $ ] .
moreover , a simple computation shows that @xmath162 as @xmath163 and @xmath164 , it follows that @xmath165 because of statements @xmath37 and @xmath39 of lemma [ lem1 ] .
hence we obtain @xmath166}a_{2r+2}\overline{q}_{r}\right)\left ( \sum_{s=0}^{\left[n/2\right]}b_{2s}x^{2s}\right)y dx.\ ] ] by using expression of @xmath167 , a straightforward computation , and lemma [ lem1](@xmath51 ) we obtain the formula given in the statement . finally , statement ( @xmath51 )
follows from the formula given in statement ( @xmath43 ) .
system with @xmath44 , @xmath168 , and @xmath169 does not satisfy the hypothesis in definition of @xmath28 because @xmath170 is not an odd function . here
@xmath171 and from theorem [ mth1](@xmath153 ) it follows that @xmath172 ; however , for @xmath7 small enough , this system has one medium amplitude limit cycle .
indeed , we need only to prove that the first non - vanishing coefficient of the displacement function , associated to the system , has a simple positive zero .
the system can be written in the form as @xmath173 with @xmath174 . by lemma [ lem4]@xmath37 , @xmath175 , and by theorem [ ppmf ] and lemma [ lem4]@xmath39 , @xmath176 .
now , system with @xmath177 , @xmath178 , @xmath179 , @xmath180 , and @xmath181 does not satisfy the hypothesis in definition of @xmath29 because @xmath182 is not an even function . in this case
@xmath183 and by theorem [ mth1](@xmath43 ) , @xmath184 ; however , for @xmath7 small enough , the resulting system has two medium amplitude limit cycles .
indeed , following previous ideas , and using theorem [ ppmf ] and lemma [ lem4 ] it is easy to see that @xmath175 , @xmath185 , and @xmath186 .
we can assume , after a linear change of variables if necessary , that @xmath187 for all @xmath188 .
suppose that @xmath189 and @xmath190 .
thus , @xmath191 and @xmath4 can be written as @xmath192 and @xmath193 , respectively , according to .
( @xmath153 ) . by hypothesis
, @xmath4 is odd for @xmath30 , which means that @xmath194 for @xmath30 .
let @xmath83 be the first non - vanishing poincar pontryagin melnikov function in .
if @xmath195 , then the theorem is true .
indeed , we have @xmath196 , and as @xmath197 , and @xmath198 by corollary [ cor1 ] , we obtain @xmath199 .
from we have @xmath200 $ ] ; hence @xmath69 has at most @xmath201 $ ] positive zeros because of corollary [ cor2 ] .
moreover , we can choose suitable coefficients of @xmath202 in such a way that @xmath69 has exactly @xmath201 $ ] simple positive zeros .
therefore , by applying the poincar pontryagin
andronov criterion it follows that @xmath203 $ ] .
suppose then that @xmath71 and we are therefore in the hypothesis of theorem [ th3 ] .
if @xmath112 for @xmath204 , then @xmath205 by lemma [ lem3 ] , and by applying the same idea as in previous paragraph , we obtain @xmath203 $ ] . accordingly , it remains to prove that @xmath112 for @xmath124 .
we proceed by induction on @xmath63 . if @xmath116 , then @xmath206 , which implies that @xmath207 .
we now assume that the assertion is true for @xmath208 , and we will prove it for @xmath123 . by induction hypothesis ,
@xmath209 for @xmath210 , which implies that @xmath211 with @xmath212 for @xmath213 by corollary [ cor1 ] .
furthermore , by lemma [ lem3 ] , @xmath214 for @xmath215 .
hence @xmath129 with @xmath216 is an element of @xmath131 because of lemma [ lem2 ] . since @xmath217 , @xmath218 , whence @xmath219 .
therefore @xmath220 , which completes the proof of statement ( @xmath153 ) . for @xmath222
the 1-form @xmath223 is exact , that is , @xmath224 with @xmath225 .
hence , by theorem [ th3 ] , @xmath226 and @xmath227 for @xmath222 , and @xmath228 . on the other hand , since @xmath5 is even for @xmath229 , @xmath230 for @xmath31 .
thus , @xmath231 for @xmath232 , and as @xmath233 because of lemma [ lem1](@xmath43 ) , we conclude that @xmath234 ; of course @xmath235 for @xmath236 .
therefore @xmath237 for all @xmath238 .
_ case @xmath41 odd .
_ if @xmath41 is odd , then @xmath239 because @xmath5 is an even polynomial for @xmath240 . since @xmath241 has an even degree , @xmath242 .
hence , from lemma [ lem4](@xmath153 ) it follows that @xmath243 , and it has at most @xmath201 $ ] positive zeros , counting multiplicities ; moreover , we can choose suitable coefficients of @xmath244 in such a way that @xmath245 has exactly @xmath201 $ ] simple positive zeros . therefore by the poincar
pontryagin andronov criterion ,
@xmath246 $ ] .
_ case @xmath41 even .
_ let @xmath83 be the first non - vanishing poincar pontryagin melnikov function of .
if @xmath247 , then @xmath245 has at most @xmath201 $ ] positive zeros , counting multiplicities , because of lemma [ lem4](@xmath153 ) .
since @xmath41 is even , @xmath201\leq \left[m/2\right]+\left[n/2\right]-1 $ ]
. hence @xmath245 has at most @xmath248+\left[n/2\right]-1 $ ] positive zeros , counting multiplicities .
we claim that if @xmath249 , then @xmath250 , @xmath251 with @xmath212 for @xmath252 , and @xmath253 . by assuming that this assertion is true and by applying lemma [ lem4](@xmath43 )
we conclude that @xmath83 has at most @xmath248+\left[n/2\right]-1 $ ] positive zeros , counting multiplicities ; moreover , we can choose suitable coefficients of @xmath254 and @xmath255 in such a way that @xmath83 has exactly @xmath248+\left[n/2\right]-1 $ ] simple positive zeros .
thus , by the poincar pontryagin andronov criterion ,
@xmath256+\left[n/2\right]-1 $ ] . therefore , to finish the proof of statement ( @xmath43 ) we need only to confirm the assertion , which we prove next by proceeding by induction on @xmath63 .
if @xmath257 , then we will prove that @xmath258 with @xmath259 , and that @xmath260 .
we know that @xmath261 , and from lemma [ lem4](@xmath43 ) it follows that @xmath262 , where @xmath263 . on the other hand , by theorem [ th3 ] , @xmath264 , where @xmath265 . since @xmath225 for @xmath266 , @xmath267
. moreover , since @xmath268 , @xmath269 . if @xmath270 , then @xmath271 . since @xmath272 , @xmath273 by lemma [ lem4](@xmath51 ) .
this implies that @xmath274 , and by corollary [ cor1 ] , @xmath275 with @xmath276 .
moreover , we know that @xmath277 with @xmath278 , and @xmath279 . thus , @xmath280 with @xmath281 because of corollary [ cor1 ] .
on the other hand , from theorem [ th3 ] we have @xmath282 as @xmath283 and @xmath281 , then we have @xmath284 following lemma [ lem2 ] and @xmath285 by corollary [ cor1 ] .
in addition , we know that @xmath225 for @xmath266 and @xmath286 . hence @xmath287 . on the other hand , from the induction hypothesis it follows that @xmath291 , @xmath251 with @xmath212 for @xmath292 , and @xmath293 .
since @xmath294 , @xmath295 because of lemma [ lem4](@xmath51 ) , which implies that @xmath296
. therefore , @xmath297 by lemma [ lem2 ] .
moreover , we have @xmath298 , and by applying corollary [ cor1 ] we obtain @xmath299 hence @xmath300 by lemma [ lem2 ] . in addition , @xmath301 .
thus , we obtain @xmath302 part of the results of this work come from the author s postdoctoral stay at the departament de matemtiques of the universitat autnoma de barcelona .
the author would like to thank the centre de recerca matemtica for their support and hospitality during the period in which this paper was written .
j. llibre , _ a survey on the limit cycles of the generalization polynomial linard differential equations .
_ mathematical models in engineering , biology and medicine , 224233 .
, 1124 , amer .
melville , ny , 2009 . | we will consider two special families of polynomial perturbations of the linear center .
for the resulting perturbed systems , which are generalized linard systems , we provide the exact upper bound for the number of limit cycles that bifurcate from the periodic orbits of the linear center .
limit cycle , linard system , periodic orbit |
the phenomenon of gamma ray burst ( grb ) has raised many unresolved issues .
for example , the nature of the compact object in the progenitor responsible for the jet emission is not yet understood .
the widely accepted scenario for the grb emission , supported by strong observational evidences ( van paradijs et al .
2000 ) , is that of a catastrophic energy release from a highly massive , rapidly rotating , low metallicity star towards the end of its life .
this catastrophic process for the long grb class ( t@xmath0 @xmath1 2 s , where t@xmath0 is the time taken by the burst to accumulate between 5% to 95% of its prompt emission
see kouveliotou et al .
1993 ; norris et al .
2005 ) is attributed to the collapsar or hypernova scenario ( woosley 1993 ; paczynski 1998 ; fryer et al .
1999 ; mszros 2006 ) while the short grbs ( t@xmath0 @xmath2 2 s ) are thought of as the outcome of merging neutron star binaries or neutron star - black hole binaries ( paczynski 1986 ; meszaros 1992;1997 ; rosswog et al .
2003a ; b ) .
though the collapsar model supports the formation of a stellar mass black hole , it is also suggested that the compact object could be a magnetar ( metzger et al .
2007 ; metzger 2010 ) .
the prompt emission of long grbs shows a wide range of structures in their light curve , though the time integrated energy spectra are adequately fit by a four parameter band model ( band et al . 1993 ) .
attempts have been made to fit the light curve of the prompt emission with empirical models such as the fast rise exponential decay ( fred ) model ( kocevski et al .
2003 ) and the exponential model ( norris et al . 2005 ) .
these models assume an underlying pulse structure of the prompt emission .
the curve fitting shows that the pulses are nearly simultaneous in a wide range of energy bands ( norris et al .
2005 ) and have self similar shapes ( nemiroff 2000 ) .
hakkila et al .
( 2011 ) showed that the observed properties of the short pulses of long grbs correlate the same way as typical pulses of short grbs .
all these properties of pulse emission indicate that despite being diverse in their many features , the prompt emission of grb has a simple underlying mechanism of pulse emission which can be characterized by empirical models irrespective of the progenitor or the environment of its formation ( hakkila et al .
2011 ) . the properties of the prompt emission of grbs like @xmath3 ( the energy at which the emission peaks amati et al .
2002 ) , the lag between the light curves at two energies ( norris et al .
2000 ) , the average variability seen during the prompt emission ( fenimore & ramirez - ruiz 2000 ) , the rise time of the pulses ( schaefer 2002 ) are found to be correlated to the luminosity of the bursts and hence they are used as distance indicators ( schaefer 2007 ; gehrels et al . 2009 ) .
most of these derived relations are empirical in nature and the analyses that are done to derive these parameters , however , are done separately in the time or the energy domain .
for example , to derive @xmath3 the time - integrated spectrum is used , ignoring the time variability and the pulse characteristics . similarly , to derive lags ,
the time profile is examined in a few broad energy bands .
attempts have been made to investigate the variations of these derived parameters in the other domain .
for example , the variation of @xmath3 with time has been investigated by kocevski & liang ( 2003 ) .
a complete empirical description of grb prompt emission which describes the time as well as energy distribution , however , is lacking .
this is probably due to the method of spectral parameter determination which is forward in nature : that is one assumes a possible spectral model , convolves with the detector response and verifies the validity of the model by comparing with the data ( arnaud 1996 ) .
hence it is difficult to incorporate time variability parameters in this scheme .
grbs often have overlapping pulses which further complicates the extracting of model parameters for any unified empirical description . in this paper
we develop a technique to give a complete pulse - wise empirical description of the prompt emission of a grb using the time - integrated spectrum and the energy integrated light curve .
we apply this method to the _ fermi_/gbm and the _ swift_/bat data of grb 090618 , which has the highest fluence among all 438 gbm grbs until march 2010 ( nava et al . 2011 ) .
we exploit the large area of _ swift_/bat to get accurate light curves and the large band - width of _
fermi_/gbm to get accurate energy spectra .
we use the band description for the time integrated spectral model ( band 2003 ) and the exponential model ( norris et al .
2005 ) for the energy integrated timing description for the pulses .
we use the expression for the variation of @xmath3 with time given by kocevski & liang ( 2003 ) and develop a method to derive the model parameters ( see section 4.1.2 ) .
we demonstrate that these model parameters correctly describe the variation of pulse width with energy and the variation of the lag with energy . in
2 we give a brief description of grb 090618 , and in 3 we give the data analysis techniques including spectral analysis ( 3.1 ) , timing analysis ( 3.2 ) and a pulse - wise joint ( _ swift_/bat and _ fermi_/gbm ) spectral analysis ( 3.3 ) . in
4 we describe our method in detail , which gives a complete three dimensional pulse - wise description of the grb . the model predicted timing results are compared with the observations in 4.2 and the implications for global correlations are touched upon in 4.3 .
the results are discussed and the major conclusions are given in the last section ( 5 ) .
the brightest gamma ray burst ( grb ) in the _ fermi _ era ( as of march 2010 ) , grb 090618 , was simultaneously observed with _ swift_/bat ( schady et al . 2009a ; 2009b ; 2009c ) , _ agile _ ( longo et al . 2009 ) , _ fermi_/gbm ( mcbreen et al . 2009 ) , _ suzaku_/wam ( kono et al . ) , _ wind_/konus - wind , _ coronas - photon_/konus - rf ( golenetskii et al . 2009 ) , _ coronas - photon_/rt-2 ( rao et al . 2009 ; 2011 ) .
this is a long grb ( t@xmath0@xmath4113 s ) with a fluence of @xmath5 erg/@xmath6 ( integrated over @xmath7=182.27 s ) .
the detection time reported by _
swift_/bat is 2009 june 18 at 08:28:29.85 ut ( schady et al .
2009a ) whereas the _ fermi_/gbm detection time is 08:28:26.66 ut ( mcbreen et al .
this burst showed structures in the light curve with four peaks along with a precursor .
a significant spectral evolution was also observed in the prompt emission which lasted for about 130 s from the trigger time ( for a summary see rao et al .
2011 ) .
the grb afterglow was observed in the softer x - rays and optical regions .
the optical afterglow was tracked by palomar 60-inch telescope ( cenko 2009 ) , katzman automatic imaging telescope ( perley et al .
2009 ) followed by various other optical , infrared , and radio observations .
observations using the kast spectrograph on the 3-m shane telescope at lick observatory ( cenko et al . 2009 ) revealed a redshift of z = 0.54 ( cenko et al .
the x - ray afterglow was measured by the _ swift_/xrt ( schady et al .
initially very bright in the x - rays , the flux rapidly decayed with a slope of @xmath46 before breaking after 310 s from the trigger time when the burst entered a shallower decay phase ( slope [email protected] beardmore et al .
2009 ) .
rao et al . ( 2011 ) analyzed the prompt emission of grb 090618 using _ coronas - photon_/rt-2 and _ swift_/bat data .
the light curves in various energy bands were fitted with a fred profile to identify pulse characteristics such as width , peak position etc . a joint rt-2 and bat spectral analysis of individual pulses
was done using the spectral model of band .
the reported time integrated low energy photon index @xmath9 , the high energy photon index @xmath10 , and the peak energy @xmath11 are -1.40 , -2.50 , and 164 kev , respectively .
ghirlanda et al . (
2010 ) performed a time resolved spectral analysis of _ fermi _ grbs of known redshifts ( grb 090618 being one of the twelve in the list ) to show that the time - resolved correlation @xmath12 within individual grbs is consistent with time - integrated correlation @xmath13 indicating that there must be a physical origin for the correlation and may not be an outcome of instrumental selection effects .
we used _ fermi_/gbm and _ swift_/bat data for our analysis .
the gbm consists of twelve sodium iodide ( nai : nx where x runs from 0 to 11 ) and two bismuth germanate ( bgo : by , where y runs from 0 to 1 ) detectors .
the nai covers the energy range @xmath1 8 kev to @xmath4 1 mev while the bgo covers a wider range of @xmath4 200 kev to @xmath4 40 mev .
data are saved in three types of packets ( meegan et .
al . 2009 ) : ( 1 ) tte ( time tag events ) in which time and energy information ( in 128 energy channels ) of individual photons are stored , ( 2 ) cspec : binned data in 1.024 s bins and 128 energy channels and ( 3 ) ctime : binned data in 0.064 s bins and 8 energy channels ( which is not suitable for spectroscopy ) .
the triggered detectors for the grb 090618 were n4 , n7 , b0 , and b1 . in the following analysis
we have used n4 , n7 and either b0 or b1 , one at a time . on the other hand , bat has an array of cdznte detector modules located behind a coded aperture mask .
this covers an energy range of 15 kev to 150 kev ( barthelmy et al . 2005 ; gehrels et al .
the large area of bat ( 5200 @xmath14 ) gives very good sensitivity for timing analysis , compared to that of nai / bgo ( 126 @xmath14 ) detectors of gbm .
ghirlanda et al .
( 2010 ) have presented the time resolved spectral analysis results for grb 090618 in 14 time segments .
we essentially followed the same methodology , but have also used the simultaneous _ swift_/bat data .
grb 090618 triggered the _
swift_/bat instrument on 2009 - 06 - 18 at 08:28:29.851 or 267006514.688 met ( s ) .
fermi_/gbm trigger time is @xmath15(gbm ) = 2009 - 06 - 18 08:28:26.659 ut or 267006508.659 met ( s ) .
hence there is a delay time in the bat data file which must be subtracted during the joint analysis .
the calculated time delay is -3.192 s and when we convert it to met it gives 267006511.496 met ( s ) for _ swift_/bat .
in all the analysis this time is subtracted from the trigger time unless specified otherwise . for the gbm we chose the tte ( time tag event ) files of the triggered nai detectors , namely n4 and n7 .
the bgo detector was excluded as it is not suitable for time resolved spectroscopy due to its lower sensitivity ( ghirlanda et al .
the energy range was chosen to be @xmath1 8 kev to @xmath4 900 kev .
package provided by the user contributions of fermi science support center ( _ fssc _ ) to select the desired time cuts and for generating the background files .
the background was taken before and after the burst , avoiding burst contamination ( ghirlanda et al . 2010 ) and the exposure time for background was optimized for different detectors .
the energy channels were binned so as to get a minimum of 40 counts per channel . for the analysis of the bat data we used standard tasks prescribed in _ the swift bat software guide , version 6.3 _ of _ heasoft 6.9 _ employing a time delay of -3.192 s. these spectra
were then fitted with a cut - off power law of the form @xmath16 where f is the photon flux .
the constant factor , _ cons _ , was kept free while fitting essentially to take care of the different systematic errors in the area calibration of the two instruments .
the derived parameter values ( using gbm alone ) were found to be consistent with the values presented in ghirlanda et al .
figure 1 shows the comparison results of the spectral analysis . in figure 1 ( left panel )
the indices @xmath17 obtained from the joint analysis are plotted against those obtained from an analysis of the gbm data for the fourteen different time segments .
similar data for the parameter @xmath18 are shown in the right panel .
a deviation is noted for the parameter @xmath17 in the 0 to 3 s region . this may be attributed to lower count rate of bat in the precursor region .
we note a systematic increment in the value of @xmath17 and @xmath18 after the inclusion of bat data .
further , the constant factor is always lower by 10 - 20% for the bat .
sakamoto et al .
( 2011 ) have made a joint spectral analysis of 14 grbs using data from _ swift_/bat , _ konus_/wind and _ suzaku_/wam and they have noted that for the bat data , the constant factor is systematically lower by 10 20% , the photon index is steeper by 0.1 0.2 and the @xmath3 becomes higher by 10 20% .
the results presented here are in general agreement with these conclusions . in figure 2
we have plotted the ratio of fractional errors as a function of the measured parameters .
we find that there is a marginal improvement for the parameter @xmath18 ( figure 2 , right panel ) .
the parameter @xmath17 ( figure 2 , left panel ) , however is measured always with a greater accuracy from the joint analysis .
the large effective area of bat improves the accuracy of the spectral index whereas its low high energy threshold ( @xmath4150 kev ) makes relatively low impact on the measurement of the high energy cut - off value .
the joint analysis always gives confidence on the measured parameters and ensures that the effects of instrumental artefacts are minimized .
hence we use , in the later analysis , a joint fit for measuring the global spec- tral parameters and use only the bat data for a joint timing and spectral analysis .
we identified four pulses ( other than the precursor ) in the @xmath460 s to @xmath4130 s period after the trigger .
both the _ swift_/bat and _ fermi_/gbm ( n4 ) were used for the timing analysis .
we subtracted background and shifted the time origin to the trigger time ( trigtime is the keyword in the header ) of the respective satellite . as we noted previously , there is a time lag of -3.192 s for the bat trigger .
light curves were extracted for different energy bands for both the data sets .
the energy bands for the _ swift_/bat were : 15 - 25 kev , 25 - 50 kev , 50 - 100 kev , and 100 - 200 kev .
for the _ fermi_/gbm ( n4 ) we chose 8 - 15 kev , 15 - 25 kev , 25 - 50 kev , 50 - 100 kev , 100 - 200 kev , and 200 - 500 kev .
greater than 500 kev region of nai and the bgo detector of gbm were excluded because the third and the fourth pulses were not detected in these bands .
all the light curves were fitted with the four - parameter model of norris et al .
( 2005 ) @xmath19 for @xmath20 , where @xmath21 and @xmath22 .
@xmath23 is the pulse amplitude , @xmath24 is the pulse start time and @xmath25 , @xmath26 are time constants characterizing the rise and decay part of the pulse . from these parameters ,
various derived quantities like the peak position , the pulse width ( measured as the separation between the two 1/e intensity points ) , and the pulse asymmetry can be calculated ( see norris et al .
2005 for details ) .
errors are calculated for the parameters as nominal 90% confidence level errors ( @xmath27=2.7 ) and these errors are propagated for the derived quantities , as described in norris et al .
( 2005 ) .
figure 3 shows the pulse width ( w ) as a function of energy ( e ) for all the pulses .
apart from the fact that the width broadens in the lower energy bands , we note a tentative evidence for a decrement ( or constancy within error bar ) of the width from 50 - 100 kev through 8 - 15 kev energy bands for the third and the fourth pulses ( lower panels of figure 3 ) .
we fitted these data for the individual pulses with ( i ) constant ( @xmath28 ) and ( ii ) linear ( @xmath29 ) models .
for the first two pulses there was a significant improvement in the @xmath30 for a linear fit ( the @xmath30 varied from 16.2(9 ) to 5.1(8 ) and 140.9(9 ) to 7.9(8 ) respectively for the two pulses ; the numbers in the brackets are the degrees of freedom ) .
the derived slopes are @xmath31 s @xmath32 and @xmath33 s @xmath32 , respectively for the two pulses .
the third and the fourth pulses , however , showed negligible improvement in 10 - 70 kev band ) for the other two pulses ( see section 3.2 ) . ] @xmath30 ( from 8.7(9 ) to 6.0(8 ) and 2.1(8 ) to 2.1(7 ) , respectively for the two pulses ) for the linear fit .
for these two pulses , when the data below 70 kev is considered , there is a marginal improvement in the @xmath30 for a linear fit with a positive slope ( from 8.7(9 ) to 3.8(5 ) and 2.1(8 ) to 1.5(5 ) , respectively for the two pulses ) .
the derived slopes are @xmath34 s @xmath32 and @xmath35 s @xmath32 respectively .
though the slopes for these two pulses are positive , the error bars are quite large and hence the evidence of this reverse variation is _
tentative_. in figure 3 ( lower two panels ) the linear fit for pulse 3 and 4 are shown by solid lines ( in the low energy ) and dot - dashed lines ( in full energy band ) .
we have ignored the fourth pulse in the 200 - 500 kev band of fermi for the fitting as it was not prominent in this energy band .
the peak position and the width gave a clue for the time cut for each pulse and facilitated the selection of integration time for pulse spectral analysis of the individual pulses .
the pulse spectral analysis was done with a band function fit and including the bgo detector of gbm .
we included the precursor pulse in this analysis .
we divided the grb event into four time segments namely , part 1 ( @xmath15 to @xmath36 ) , part 2 ( @xmath36 to @xmath37 ) , part 3 ( @xmath37 to @xmath38 ) and part 4 ( @xmath38 to @xmath39 ) .
the first and second pulses are considered together as there is too much overlap between them .
all the spectra were fitted in the energy range @xmath4 8 kev - @xmath4 1 mev with the four parameter model of band ( band et al .
1993 ) : @xmath40^{\alpha}exp\left[\frac{-(2+\alpha)e}{e_{peak}}\right ] \\ \\
$ \rm if$ ~~e\le [ ( \alpha-\beta)/(2+\alpha)]e_{peak}\\ \\ a_{b}\left[\frac{e}{100}\right]^{\beta}exp\left[\beta-\alpha\right]\left[\frac{\left(\alpha-\beta\right)e_{peak } } { 100\left(2+\alpha\right)}\right]^{(\alpha-\beta ) } \\ \\ otherwise \end{array } \right.\ ] ] the lower and the higher energy parts are characterized by the parameters @xmath17 and @xmath41 respectively and they meet smoothly at the spectral break energy @xmath42e_{peak}$ ] . the peak energy , @xmath3 or equivalently the characteristic energy , defined as @xmath43 , may be considered as the third parameter of the model . the best fit model parameters along with the nominal 90% confidence level error ( @xmath27=2.7 ) for the joint fit of _ fermi_/gbm ( n4 , n7 , b1 ) and _ swift_/bat are reported in table 1 ( b0 in place of b1 gives similar results ) .
the high energy photon index @xmath41 is constrained primarily by the gbm data and hence we have kept its value fixed at -2.5 to determine precise values of @xmath3 and @xmath17 . only in one case namely for the full range ( @xmath15 to @xmath39 sec ) of the grb
, @xmath41 was made to vary and a value of -2.96 was obtained .
figure 4 shows the joint band spectral fitting of this range along with the @xmath44 deviation of the data .
an examination of the residuals in figure 4 reveals that there are some systematic differences in the response in the 20 - 40 kev region between bat and gbm data and the resultant @xmath44 > p0.5inccccc & @xmath17 & @xmath41 & @xmath3(kev ) & @xmath44(dof ) & @xmath18 ( kev ) + & -1.35@xmath45 & -2.5 ( fixed ) & 154@xmath46 & 428(299 ) & 236@xmath47 + & & -2.5(fixed ) & @xmath48 & 369(299 ) & @xmath49 + & -1.11@xmath45 & -2.5(fixed ) & 212@xmath50 & 460(299 ) & 238@xmath51 + & -1.15@xmath52 & -2.5(fixed ) & 114@xmath53 & 334(259 ) & 134@xmath50 + & @xmath54 & -2.5(fixed ) & 33@xmath55 & 431(299 ) & @xmath56 + & -1.36@xmath45 & [email protected] & @xmath57 & 418(288 ) & @xmath58 + 8 kev to @xmath4 1 mev .
the range of the bgo detector starts above 200 kev , while bat covers up to @xmath4 150 kev .
the other two detectors namely n4 and n7 have the spectrum above 8 kev to @xmath4 800 kev .
the extreme channels and specifically the 25th channel of n4 are ignored due to high background rates .
the unfolded spectrum of bat shows a lower normalization ( see text ) . ]
values generally are quite high .
the resultant spectral parameters , however , agree with the results reported for this grb ( ghirlanda et al .
, 2010 ; nava et al . 2011 ; rao et al . 2011 ) .
the errors are significantly better compared to the joint bat and rt-2 analysis ( rao et al .
2011 ) , but comparable to that obtained from a gbm analysis alone ( nava et al .
since a joint analysis using data from different satellites pins down the systematic errors in each instruments , we believe that the joint analysis results , reported here , are devoid of instrumental artefacts , particularly for the parameters @xmath17 and @xmath3 .
the light curve of each pulse of a grb can be described in terms of the four - parameter model of norris ( equation 1 ) and the spectrum for each pulse can be fitted with the band model ( equation 2 ) .
liang and kargatis ( 1996 ) showed for clean , separable pulses that the peak energy of the @xmath59 spectrum , @xmath3 , decays with the photon fluence as , @xmath60 where @xmath61 is the peak energy at zero fluence , @xmath62 is the decay constant and @xmath63 .
the quantity @xmath64 is the fluence up to time t from the start of the pulse ( @xmath65 ) . for convenience of description
we shift our origin to @xmath65 and hence @xmath66 for each pulse analysis .
equation 3 quantifies the overall softening of the burst with time .
kocevski and liang ( 2003 ) have integrated the light curves of a sample of grbs up to various times t and found the fluence , @xmath67 , at those times .
the peak energy at the same times , @xmath68 , were found by fitting the spectra with a band model .
they determined the @xmath61 and @xmath62 from a semi - log plot of @xmath68 versus @xmath67 and fitting a straight line .
this procedure was adopted for clean , bright , and separable fred pulses . here
we explore a method to determine the timing and spectral parameters by _ assuming that @xmath68 and @xmath67 are related according to equation 3 in each pulse . _
essentially we assume that for each pulse , the variation of flux with time is given by the exponential model ( equation 1 ) , and at each time the spectra is described by a band function ( equation 2 ) , and , for a given pulse , only the peak energy varies with time governed by equation 3 .
we develop a method to derive the model parameters from a time integrated spectrum and energy integrated light curve .
the flux of a grb pulse is a function of time and energy .
if there exist empirical formulae for the temporal and spectral beahvior then the flux can be described in terms of all the model parameters .
for example , the variation of flux with time is given by norris exponential model , with four parameters ( @xmath69 , the normalization , @xmath65 , @xmath25 , and @xmath26 ) .
the spectral behaviour is parametrized by a band function , with four parameters ( @xmath70 , the normalization , @xmath17 , @xmath41 , and @xmath3 ) . for any timing model ,
the time parameters change with energy e.g. , the derived parameter , width , changes in various energy bands ( see section 3.2 ) .
similarly , the spectral parameters of any spectral model change with time e.g. , @xmath17 and @xmath71 of cut - off power law change with time in the same pulse ( see section 3.1 ) . the norris exponential model ( equation 1 ) describes the evolution of a pulse with time . this description can be used for counts in any energy band and the variation of the model parameters with energy can be studied . instead
, we assume that the model parameters @xmath25 and @xmath26 derived for a pulse integrated over all energy bands are global parameters describing the evolution of total flux with time .
since the variation of spectrum is best described by the peak energy evolution with fluence ( equation 3 ) , we further assume that @xmath17 and @xmath41 do not change with time . the normalization factors ( @xmath69 and @xmath70 )
are related by the fact that the fluence calculated either by time integration of the pulse light curve or by the integration over the total energy should be the same . with these assumptions ,
the parameters for a given pulse are determined as follows .
> p0.45in > p0.5in > p0.6in > p0.75in > p0.8in > p0.7in >
p0.7 in pulse & @xmath61 & @xmath62 & norm & @xmath44 ( dof ) & @xmath25(s ) & @xmath26(s ) & @xmath24(s ) + 1 & @xmath72 & @xmath73 & @xmath74 & 40.12 ( 75 ) & @xmath75 & @xmath76 & + 2 & & & & & @xmath77 & @xmath78 & @xmath79 + 3 & @xmath80 & @xmath81 & @xmath82 & 31.41 ( 75 ) & @xmath83 & & @xmath84 + 4 & @xmath85 & @xmath86 & @xmath87 & 56.42 ( 75 ) & @xmath88 & & + for a given pulse , we use the energy integrated light curve from bat and use equation 1 for the fitting .
this best fit light curve is used to determine the global parameters namely , @xmath69 , @xmath25 , and @xmath26 .
since we can arbitrarily choose the pulse starting time by shifting the time origin , the parameter @xmath65 is not very relevant ( also see section 4.1.1 ) .
this function is then integrated up to a time t to determine @xmath67 . from the individual spectral fitting for each pulse using the band model , @xmath17 and @xmath41 are determined ( see section 3.3 ) .
now , if the constant parameters namely , @xmath61 and @xmath62 are determined , we have a complete spectral and timing description of a grb pulse .
for this purpose , we use a two - dimensional grid of guess values for these two parameters and then determine the best fit values by a spectral fitting method . for a particular combination of @xmath61 and @xmath62 values , @xmath68 at any time t can be determined from equation 3 . on the other hand ,
equation 2 can be rewritten as @xmath89 where @xmath90 is the normalization and f(e ) is the rest of the functional form valid in the two energy bands .
given the global parameters ( @xmath17 and @xmath41 ) and the time dependent parameter ( @xmath68 ) , f(e ) can be determined .
the normalization @xmath91 can be found by dividing the fluence ( @xmath92(t ) ) of the light curve by the f(e ) integrated over the whole energy range .
hence photon spectrum is generated for the time instance t. choosing a time and energy grid , counts can be plotted as a function of time and energy for a particular combination of @xmath61 and @xmath62 values , and the resulting surface can be referred to as a three dimensional pulse model .
hence for a two - dimensional grid of guess values of @xmath61 and @xmath62 we can generate three dimensional pulse models for each of the grid points .
these model pulses can be used in two ways : ( 1 ) integration over time gives the model spectrum of that pulse , ( 2 ) integration over various energy bands gives model light curves of that pulse in those energy bands . in our analysis , we first generate a two parameter xspec table model using the guess values of @xmath61 and @xmath62 and then use this table model to determine the best fit values of these two parameters by fitting the time integrated spectrum ( see section 4.1.3 ) . with this method
, we get a complete description of the pulse with model parameters determined from the data . and @xmath93 for each pulse .
counts in arbitrary unit is plotted in the vertical axis against time and energy .
the pulses are shifted to original pulse start time ( see table 2 ) . ] using this model we can reconstruct the three dimensional pulse profiles and can generate light curves in any energy band . various timing parameters ( width and lag )
are derived in these energy bands ( see section 4.2 ) and these model predicted values can be compared with the observed values to get a validation of the whole procedure .
for all the pulses we use a time grid resolution of 0.5 s and energy grid resolution of 2.0 kev .
each pulse has a separate xspec table model , though the first and second pulses are combined together as the data overlap .
a typical table model contains a spectrum in the 2 - 200 kev energy range for 200 values of @xmath61 from 100.0 kev to 1100.0 kev and 50 values of @xmath62 from 2.0 photon @xmath94 to 77.0 photon @xmath94 , energy binning information , parameter values and the required header keywords standardized as per the requirement by xspec described in ogip memo ogip/92 - 009 .
the additive table model is used to determine the best fit values of @xmath61 and @xmath62 by the standard @xmath44 minimization procedure .
the best fit values with nominal 90% confidence errors are reported in table 2 .
the parameters , @xmath61 and @xmath62 of the first two pulses are assumed to be the same .
it should be noted here that we have used the normalization @xmath23 as a global parameter to calculate @xmath92(t ) .
the normalization of the subsequent pulse profile is derived from the integration of f(e ) over the total energy band ( see section 4.1.2 ) .
effectively this leads to a spectrum with normalization as a free parameter .
this normalization is determined during the spectral fitting using xspec , and is also reported in table 2 .
three dimensional model of each pulse is generated for the best fit parameter values ( as described in sections 4.1.2 and 4.1.3 ) .
the three dimensional model of the whole grb obtained from the sum of these models of the pulses with each pulse shifted to its proper starting time is shown in figure 5 .
this figure is used only for display purposes . in actual analysis ,
each pulse is normalized properly , before calculating various parameters as the overall normalization may vary for each pulse .
this normalization is achieved for spectrum and light curve separately .
once we get the best fit values of @xmath61 and @xmath62 for each pulse we can generate the best fit model light curve by integrating the three dimensional pulse model for the best fit value over the desired energy range . we generate three dimensional pulse models , shift them over appropriate starting time , co - add them with proper normalization factors and integrate over energy ranges to generate the combined light curves in the energy bands : 15 - 25 kev , 25 - 50 kev , 50 - 100 kev and 100 - 200 kev of bat . in figure 6 ( right panel ) , we show the observed light curve along with the model predicted light curve for the 25 - 50 kev energy band . in the left panel
norris exponential model fit for the total energy range is shown .
parameters from this fit ( figure 6 , left panel ) were used for characterizing the timing parameters in our model .
the grb light curve shows rapid time variable components over and above the smooth pulse structure . in the pulse description
we consider only this smooth variation .
any small rapid time variability in the data gives large value of @xmath44 when we use the smooth pulse model for fitting .
hence , the normalization factor could not be determined by @xmath44 minimization .
instead we physically examined the heights and determined the normalization .
the co - added pulses reproduce the light curve quite well .
hence , we can conclude that the total light curve and the integrated spectrum , with the assumption of @xmath68 varying with @xmath67 in a predetermined way ( equation ( 3 ) ) , correctly reproduces the energy resolved light curves .
this can be taken as a confirmation of the assumption of the @xmath3 variation .
it should , however , be noted that c > p1inccc pulse & & & & + 1 & @xmath95 ( @xmath96 ) & & & + 2 & @xmath97 ( @xmath98 ) & & & + 3 & @xmath99 ( @xmath100 ) & & & + 4 & @xmath101 ( @xmath102 ) & & & + a single joint spectral fit is done for pulses 1 and 2 and hence the derived values of @xmath61 and @xmath62 for the first two pulses are only the average values .
pulse width of individual pulses in the various energy bands can be derived by fitting the light curve using equation ( 1 ) and using the best fit values of @xmath103 and @xmath104 ( see section 3.2 ) . for overlapping pulses this method is erroneous e.g. , rising part of one pulse can fall on top of the falling part of the previous pulse affecting the width .
also , pulse delay characteristics are affected by these overlaps . in such cases
the method described here is very useful .
it takes as input some global parameters ( @xmath23 , @xmath103 , @xmath104 , @xmath17 and @xmath41 ) which are average quantities characterizing the pulse and some constant parameters ( @xmath61 , @xmath62 ) , determines the variable parameter ( @xmath68 ) over the time grids and generates the three dimensional model of the pulse .
note that the timing parameters @xmath103 and @xmath104 are determined from the energy integrated light curve and are fixed in our model .
the timing properties of the pulse ( e.g. , width in various energy bands , spectral delay ) are now determined crucially by the evolution of the peak energy .
integration of the three dimensional model in various energy bands generates the light curves in those energy bands .
hence physically we can determine the full width at the 1/e intensity and the peak positions .
we define the spectral delay as the shift in the peak position at each energy band compared to the lowest energy band ( 15 - 25 kev ) .
the derived pulse width and delay are reported in table 3 and table 4 , respectively . in table 3
the numbers in the brackets are the width determined from norris exponential fitting for individual light curves .
figure 7 shows a comparison between the measured width following the new method with that determined by fitting the light curve .
we note a systematic decrement of the width measurement by our method for the first two pulses .
this is simply due to the fact that individual pulses rather than co - added pulses are taken into consideration .
hence , measured widths are more accurate and devoid of contamination effects from other pulses .
one of the pulse properties of grb is broadening at lower energies ( norris et al .
1996 , hakkila 2011 ) .
figure 8 ( left panel ) shows the pulse width [ cols="^,^,^,^",options="header " , ] variation with energy .
while the first two pulses follow the normal trend , we notice that the width for the third and fourth pulses broadens with higher energy .
this can not arise from the contamination of the previous pulse as could have been the argument for fitting the entire light curve . in our case , each pulse is generated and analyzed separately for the width determination .
the energy - width plot of individual pulses is fitted with a linear model .
the slopes are negative for the first two pulses ( @xmath105 s @xmath32 and @xmath106 s @xmath32 respectively ; @xmath30 = 0.13 ( 2 ) and 0.11 ( 2 ) respectively ) and positive for the other two ( @xmath107 s @xmath32 and @xmath108 s @xmath32 respectively ; @xmath30 = 0.0046 ( 2 ) and 0.0021 ( 2 ) respectively ) .
figure 8 ( right panel ) shows a comparison of the slopes meaured by the linear fit of data determined by the two methods ( compare slopes determined in section 3.2 ) .
we note again that the slopes of the first two pulses are convincingly negative ( low error bar ) , but slope of the other two pulses have large error bars . hence
, the evidence of reverse pulse broadening is tentative , though individual pulse analysis in the new method shows a better evidence for the reverse pulse broadening effect ( see figure 8 - left panel ) .
this deviation from the canonical pulse width broadening with energy decrement may have a physical significance . in our analysis physical process of pulse emission
is not considered .
but phenomenologically , this occurs depending on the values of @xmath103 and @xmath104 .
a detailed comparison between the first and the third pulse reveals the fact that they have different pulse width variation albeit having nearly similar parameters except for @xmath25 and @xmath26 .
these values are as follows ( the numbers within parentheses are those of the third pulse ) : @xmath17= -1.11 ( -1.15 ) , @xmath41 = -2.5 ( -2.5 ) , @xmath61=359 ( 324 ) , @xmath62=12.2 ( 18.8 ) , @xmath25=795.4 ( 353.4 ) , @xmath26 = 0.54 ( 2.47 ) .
it is apparent that the main contributing factors for the different width - energy relation are the global values of @xmath25 and @xmath26 .
hence we examined for a generic pulse keeping all other parameters fixed at the values of the first pulse and varying @xmath25 for different values of @xmath26 .
figure 9 shows the regions in which pulse broadening or its reverse phenomenon can occur .
this figure is generated by fixing all the parameters other than @xmath25 and @xmath26 to those for the first pulse . for four values of @xmath26
we vary @xmath25 and note the ratio of the width in 100 - 200 kev region ( @xmath109 ) to that in the 15 - 25 kev ( @xmath110 ) region .
the solid line parallel to @xmath25 with @xmath111/@xmath112=1 divides the plot into two regions : ( i ) _ normal width broadening _ region for which @xmath111 < @xmath112 and ( ii ) _ anomalous width broadening _ region for which @xmath111 > @xmath112 .
global values of @xmath25 and @xmath26 falling in the @xmath111 < @xmath112 region will lead to normal pulse broadening ( i.e. , pulse broadens with lower energy ) and vice versa .
we identify from the figure that the @xmath25 and @xmath26 values ( 795.4 and 0.54 respectively ) of the first pulse lie well within the region of _ normal width broadening_. we also note that the values of @xmath25 and @xmath26 for the third pulse ( 353.4 and 2.47 respectively ) lie within the region of _ anomalous width broadening _ despite the fact that the parameter values used , other than @xmath25 and @xmath26 , are those of the first pulse .
this shows how insensitive is the pulse broadening effect with all other parameters and crucially determined by the global values of @xmath25 and @xmath26 .
another pulse property is the hard - to - soft spectral evolution . in figure 10
we have plotted the model predicted delay as a function of energy ( left panel ) . it can be seen that the spectral time delay always shows a decrement with energy .
the hard x ray always precedes the softer one .
the values for the first three peaks can be compared with the observed delay given in rao et al .
( 2011 ) and this is shown in the right panel of figure 10 .
the time intervals taken by rao et al .
( 2011 ) are @xmath15 to @xmath15 + 50 s , @xmath15 + 50 s to @xmath15 + 77 s , @xmath15 + 77 s to @xmath15 + 100 s , @xmath15 + 100 s to @xmath15 + 180 s. among these the first time bin contains the precursor burst which is not taken in our calculation .
the first and the second pulses are combined in the second time bin . the third pulse is covered in the third time bin .
data of the fourth pulse can not be compared with the time bin as they differ by @xmath4 55 s ( 80 s in time bin as compared to @xmath4 25 s in pulse analysis ) .
the deviation of the model predicted delay from the observed delay can be explained by the fact that the delay in the second time bin ( @xmath15 + 50 to @xmath15 + 77 ) is the combined effect of the first two pulses .
the data for the third time bin agrees within error with the model predicted delay . also , the observed delays are calculated using peak position measurement and cross correlation of the pulses whereas model predicted delays are calculated from peak position measurement alone .
one of the important properties of the prompt emission is that the various measured quantities ) to the width in 15 - 25 kev ( @xmath112 ) , as predicted by the pulse model , is plotted against @xmath25 for different values of @xmath26 . in the figure maximum as well as the minimum of the curves
move to lower @xmath25 values for higher @xmath26 .
the values of @xmath26 are 0.54 , 1.20 , 1.90 , and 2.47 , respectively from left to right .
the @xmath25 values determined for the second and the third pulses of the grb ( 795.4 s and 353.4 s , respectively ) are shown by dash - dotted lines .
the solid line divides the region into two parts : the normal width broadening region and the anomalous width broadening region ( see section 4.2.2 ) . ]
correlate with luminosity .
particularly , the peak energy of the @xmath59 spectrum , @xmath3 correlates with isotropic energy @xmath113 ( amati et al .
2002 ) and also with the peak isotropic luminosity @xmath114 ( yonetoku et al .
ghirlanda et al . (
2004 ) showed that the correlation is tighter with collimation - corrected energy , @xmath115 .
schaefer ( 2007 ) used a set of 69 grb sample ( up till 7 june 2006 ) with measured redshift to calculate various correlations viz .
lag - luminosity ( @xmath116 ) , variability - luminosity ( @xmath117 ) , @xmath118 , @xmath119 , minimum rise time - luminosity ( @xmath120 ) , number of peaks - luminosity ( @xmath121 ) to get distance moduli for each .
weighted average of them gave the distance modulus ( @xmath122 ) which is plotted against the measured redshift ( z ) extending the hubble diagram ( hd ) to z > 6 .
therefore , these correlations can be used to standardize the grb energy budget making grbs a class of cosmic distance indicators .
also , these correlations might point to a underlying fundamental physical process ( yamazaki et al . 2004 ; rees & mszros 2005 ; xu et al . 2005 ; firmani et al . 2005 ; 2006 ; 2007 ; ghirlanda et al .
2006a;b ; wang et al .
2006 ; thompson 2006 ; thompson et al . 2007 ; liang et al . 2005 ; 2008 and the references therein ; li et al . 2008 ; qi et al .
2008 ) . in order to establish the reality of these correlations one must show that all of these correlations hold just as the same way within a single grb as for a sample of many grbs .
ghirlanda et al . (
2010 ) showed for a set of grbs that the _ time - resolved _ correlations of a single grb are similar to the _ time integrated _ correlations among different grbs .
in the present work we have used the underlying pulse structure to investigate the correlations : @xmath123 , @xmath124 , @xmath125 and @xmath126 within the pulses of a single grb .
the correlations are shown in figure 11 .
the mean and the 3@xmath127 scatter of @xmath128 ( top left panel ) and @xmath129 ( top right panel ; ghirlanda et al .
2010 ) are shown for reference . as expected
, the @xmath61 values always lie above the mean of @xmath3 values .
it seems that the correlations of @xmath61 with @xmath114 and @xmath130 are rather tighter in our case , though with only three points we can not draw any definite conclusions . while the parameter @xmath61 is a constant
, @xmath3 is an average quantity .
hence , we expect that @xmath61 provides a better pulse description than @xmath3 .
this can be tested in future for correlations of @xmath61 rather than @xmath3 with other quantities in different grbs .
but if the same correlations hold for different grbs then we might conclude ( a ) the correlations are rather physical and not an outcome of the bias in sample selection , ( b ) in a single multi - peaked grb each pulse is generated independently .
we also note that the calculated @xmath114 and @xmath130 are somewhat lower than those calculated by ghirlanda et al .
this must have come from the instrumental effect as noticed earlier .
the normalization of bat is always less than that of _
this leads to a lower flux calculated from the bat spectrum .
we have attempted to explain the timing and spectral variations within a pulse of grb by a simple empirical model and have given a prescription to measure the parameters of this model .
we use simultaneous data from _ swift_/bat and _ fermi_/gbm to determine the average pulse spectral characteristics using the band model .
then we use the integrated light curve from _ swift_/bat to get a simple description of the time evolution using norris exponential model . using these parameters ,
we develop a method to fit the spectrum using xspec table model by assuming that @xmath68 decreases exponentially with fluence in the pulse .
we determine the parameters of the model and then find that these parameters can correctly describe the energy dependence of the pulse characteristics like width and spectral lags .
we also find , in the data , a tentative evidence of _ anomalous width broadening with energy _ for some pulses .
the joint spectral and timing analysis developed here confirms this phenomena for the same pulses at least in the lower energy bands .
the method developed here is applicable for both single pulse and structured grbs .
most of the long grbs are structured which makes them difficult for pulse analysis .
as far as the correlations are concerned , pulse analysis is more indicative than intensity guided time - resolved spectroscopy . hence to draw conclusion regarding these correlations within a sample of grbs and within the pulses of a single grb
we must include these structured grbs with multiple pulses .
kocevski et al .
( 2003 ) have prescribed a method for short , clean pulses .
our method extends the domain to all kinds of grbs .
also the present method is much more physical than the time - resolved one in the sense that the individual pulses are treated separately .
hence there wo nt be contamination effects .
the l@xmath131-e@xmath132 looks tighter because correlations within pulses may be more physical than the time - resolved correlation or the time integrated correlations . in future the method developed here
can be used for a large set of grbs ( onto their individual pulses ) to investigate the improvement in the correlations .
it is very interesting to note that the two distinct pulses ( pulse 3 and pulse 4 ) in grb 090618 have e@xmath131 values within a factor of 2 , but widely different e@xmath133 values ( 114 kev and 33 kev , respectively ) .
they will show a huge scatter if used individually for a global correlation .
the e@xmath132 values , on the other hand , are closer to each other and lead to a lower scatter in the correlation ( see figure 11 , top right panel ) .
hence we suggest that the empirical relations are intrinsic to the emission mechanism of grbs and e@xmath132 values , instead of e@xmath133 , should be the correct parameter to be used for correlation studies .
the method described in this paper is quite generic in the sense that any empirical description of the spectrum and its evolution with time can be tested against the data and the derived parameters can be used for global correlation studies .
for example the band spectrum is derived from an extensive study of data from _ cgro_/batse , which is primarily sensitive above 50 kev ( band et al .
extension of the data to lower energies in fact requires additional features .
for example a time resolved spectral study of grb 041006 using _
hete-2 _ data covering the energy range of 2 kev 400 kev , indicated the presence of soft components , each of them having distinctive time evolution ( shirasaki et al . 2008 ) .
we emphasize again here that the use of different instruments can help in pinning down the systematics , identifying additional spectral components and developing an empirical spectral formalism . the simultaneous data from _ swift_/bat and
_ fermi_/gbm have the potential to explore the existence of additional components because the highly sensitive bat has an energy range ( @xmath415 kev 200 kev ) which is a subset of the gbm energy range .
our joint analysis of bat and gbm data has shown that in the 20 40 kev region there is a distinct discrepancy .
this could be due to a ) systematic errors in each of the detectors , b ) presence of additional components or c ) an interplay of both these .
since the spectral fitting method used in xspec is forward in nature ( that is fold the assumed spectrum and compare with the data - see arnaud 1996 ) , the derived spectral parameters will have strong bias of the bandwidth of the detectors , particularly for such low resolution spectral data .
hence a joint fitting of data with good overlap will result in a better estimate of the parameters . in conclusion ,
we have developed a method to test empirical descriptions of the energy spectra and their evolution with time in grbs and to derive the underlying physical parameters .
this is the first attempt in this different approach to study various parameters and predict quantities like spectral delay and pulse width of individual pulses in a grb .
this research has made use of data obtained through the heasarc online service , provided by the nasa / gsfc , in support of nasa high energy astrophysics programs .
we thank the referee for the valuable suggestions in making the text more readable and quantifying some of the results .
amati , l. , et al .
2002 , , 390 , 81 band , d. , et al .
1993 , , 413 , 281 barthelmy , s. d. , et al .
2005 , , 120 , 143 beardmore , a. p. , & schady , p. 2009 , grb coordinates network , 9528 , 1 cenko , s. b. 2009 , grb coordinates network , 9513 , 1 cenko , s. b. , perley , d. a. , junkkarinen , v. , burbidge , m. , diego , u. s. , & miller , k. 2009 , grb coordinates network , 9518 , 1 fenimore , e. e. , & ramirez - ruiz , e. 2000 , arxiv : astro - ph/0004176 firmani , c. , avila - reese , v. , ghisellini , g. , & ghirlanda , g. 2007 , , 43 , 203 firmani , c. , avila - reese , v. , ghisellini , g. , & ghirlanda , g. 2006 , , 372 , l28 firmani , c. , ghisellini , g. , ghirlanda , g. , & avila - reese , v. 2005 , , 360 , l1 fryer , c. l. , woosley , s. e. , & hartmann , d. h. 1999 , , 526 , 152 gehrels , n. , et al .
2004 , , 611 , 1005 ghirlanda , g. , ghisellini , g. , & firmani , c. 2006 , new journal of physics , 8 , 123 ghirlanda , g. , ghisellini , g. , firmani , c. , nava , l. , tavecchio , f. , & lazzati , d. 2006 , , 452 , 839 ghirlanda , g. , nava , l. , & ghisellini , g. 2010 , , 511 , a43 ghirlanda , g. , ghisellini , g. , & lazzati , d. 2004 , , 616 , 331 golenetskii , s. , et al .
2009 , grb coordinates network , 9553 , 1 hakkila , j. , & preece , r. d. 2011 , arxiv:1103.5434 kocevski , d. , & liang , e. 2003 , , 594 , 385 kocevski , d. , ryde , f. , & liang , e. 2003 , , 596 , 389 kono , k. , et al .
2009 , grb coordinates network , 9568 , 1 kouveliotou , c. , meegan , c. a. , fishman , g. j. , bhat , n. p. , briggs , m. s. , koshut , t. m. , paciesas , w. s. , & pendleton , g. n. 1993 , , 413 , l101 li , h. , xia , j .- q . ,
liu , j. , zhao , g .- b . , fan , z .- h . , & zhang , x. 2008 , , 680 , 92 liang , e. , & kargatis , v. 1996 , , 381 , 49 liang , e. , & zhang , b. 2005 , , 633 , 611 liang , n. , xiao , w. k. , liu , y. , & zhang , s. n. 2008 , , 685 , 354 longo , f. , et al .
2009 , grb coordinates network , 9524 , 1 mszros , p. 2006 , reports on progress in physics , 69 , 2259 mszros , p. 2006 , kitp program : the supernova gamma - ray burst connection , mcbreen , s. 2009 , grb coordinates network , 9535 , 1 meegan , c. 2009 , aps meeting abstracts , 4001 meszaros , p. , & rees , m. j. 1997 , , 482 , l29 meszaros , p. , & rees , m. j. 1992 , , 397 , 570 metzger , b. d. 2010 , new horizons in astronomy : frank n. bash symposium 2009 , 432 , 81 metzger , b. d. , thompson , t. a. , & quataert , e. 2007 , supernova 1987a : 20 years after : supernovae and gamma - ray bursters , 937 , 521 nava , l. , ghirlanda , g. , ghisellini , g. , & celotti , a. 2011 , , 530 , a21 nemiroff , r. j. 2000 , , 544 , 805 norris , j. p. , bonnell , j. t. , kazanas , d. , scargle , j. d. , hakkila , j. , & giblin , t. w. 2005 , , 627 , 324 norris , j. p. , nemiroff , r. j. , bonnell , j. t. , scargle , j. d. , kouveliotou , c. , paciesas , w. s. , meegan , c. a. , & fishman , g. j. 1996 , , 459 , 393 norris , j. p. , marani , g. f. , & bonnell , j. t. 2000 , , 534 , 248 paczynski , b. 1986 , , 308 , l43 paczynski , b. 1998 , , 494 , l45 perley , d. a. 2009 , grb coordinates network , 9514 , 1 qi , s. , wang , f .- y . , & lu , t. 2008 , , 483 , 49 rao , a. r. , et al .
2009 , grb coordinates network , 9665 , 1 rao , a. r. , et al .
2011 , , 728 , 42 rees , m. j. , & mszros , p. 2005
, , 628 , 847 rosswog , s. , & ramirez - ruiz , e. 2003 , , 343 , l36 rosswog , s. , ramirez - ruiz , e. , & davies , m. b. 2003 , , 345 , 1077 sakamoto , t. , et al . 2011 , , 63 , 215 schady , p. 2009 , grb coordinates network , 9527 , 1 schady , p. , baumgartner , w. h. , & beardmore , a. p. 2009 , gcn report , 232 , 1 schady , p. , et al .
2009 , grb coordinates network , 9512 , 1 schaefer , b. 2002 , 34th cospar scientific assembly , 34 , schaefer , b. e. 2007 , , 660 , 16 shirasaki , y. , et al .
2008 , , 60 , 919 thompson , c. , mszros , p. , & rees , m. j. 2007 , , 666 , 1012 thompson , c. 2006 , , 651 , 333 van paradijs , j. , kouveliotou , c. , & wijers , r. a. m. j. 2000 , , 38 , 379 wang , f. y. , & dai , z. g. 2006 , , 368 , 371 woosley , s. e. 1993 , , 405 , 273 xu , d. , dai , z. g. , & liang , e. w. 2005 , , 633 , 603 yamazaki , r. , ioka , k. , & nakamura , t. 2004 , , 606 , l33 yonetoku , d. , murakami , t. , nakamura , t. , yamazaki , r. , inoue , a. k. , & ioka , k. 2004 , , 609 , 935 | we develop a new method for simultaneous timing and spectral studies of gamma ray burst ( grb ) prompt emission and apply it to make a pulse - wise description of the prompt emission of grb 090618 , the brightest grb detected in the fermi era .
we exploit the large area ( and sensitivity ) of _ swift_/bat and the wide band width of _ fermi_/gbm to derive the parameters for a complete spectral and timing description of the individual pulses of this grb , based on the various empirical relations suggested in the literature .
we demonstrate that this empirical model correctly describes the other observed properties of the burst like the variation of the lag with energy and the pulse width with energy .
the measurements also show an indication of an _ increase _ in pulse width as a function of energy at low energies for some of the pulses , which is naturally explained as an off - shoot of some particular combination of the model parameters .
we argue that these model parameters , particularly the peak energy at the beginning of the pulse , are the natural choices to be used for correlation with luminosity .
the implications of these results for the use of grbs as standard candles are briefly described . |
since the early days of quantum electronics , non - linear optics has played a basic role both for its relevance as a fundamental chapter of modern science and for its technological applications @xcite .
nonlinear parametric processes , due to the peculiar correlation properties of the generated photons , have been adopted to investigate the quantum properties of electromagnetic fields . in the last few years
it has been proposed to exploit entangled quantum states in order to increase the resolution in quantum interferometry , specifically , for quantum lithography @xcite and to achieve heisenberg limited resolution @xcite . in such framework , particular attention has been devoted to the generation of noon states , path entangled states of the form @xmath0 , of fundamental relevance since a single - photon phase shift @xmath1 induces a relative shift between the two components equal to @xmath2 .
this feature can be exploited to enhance phase resolution in interferometric measurements , leading to a sub - rayleigh resolution scaling as @xmath3 ; @xmath4 being the wavelength of the field @xcite .
the generation of photonic noon states has been the subject of intense theoretical research @xcite , but up to now the actual experimental implementation has been limited to a posteriori generation of two , three and four photons states @xcite and to the conditional generation of a noon state with @xmath5 @xcite .
the weak value of the generated number of photons strongly limits the potential applications to quantum lithography and quantum metrology . as alternative approach to emulate the quantum resolution ,
it has been proposed to adopt classical , coherent light @xcite effectively exploiting the non - linearity of the absorbing material .
first experimental results have been recently reported : yablonovitch et al .
proposed to use an interference technique with multiple - frequency beams achieving an experimental visibility of @xmath6 @xcite while boyd et al .
achieved a two - fold enhancement of the resolution over the standard rayleigh limit adopting a uv lithographic material excited by multi - photon aborption @xcite .
recently it has been proposed to exploit a high gain optical parametric amplifier acting on the vacuum field to generate fields with a high number of photons still exhibiting sub - rayleigh periods : fig.1-a @xcite .
the quantum lithography architecture involves the generation of correlated beams over the modes @xmath7 and @xmath8 , the mixing over a beamsplitter ( bs ) , the phase shifting of one mode and then the recombination of the two modes over a second .
finally the output state is detected via a two - photon absorption over the outgoing modes . in the typical low - gain regime
the quantum state @xmath9 is generated by spontaneous parametric down - conversion ( spdc ) and a hong - ou - mandel interferometer adopting a post - selection technique @xcite . in ref .
@xcite it has recently been theoretically shown that , for any gain of the parametric amplifier , the two - photon excitation rate presents a fringe pattern of the form @xmath10 which never falls below a visibility of @xmath11 .
such an approach could lead to high resolution with a high number of photons overcoming the difficulty connected with the adoption of noon state with a low number of photons . at variance with the scheme based on classical radiation , the one here demonstrated exploits quantum features of the opa field .
in the present paper we experimentally investigate the properties of the optical parametric amplifier ( opa ) operating in the high gain regime . instead of dealing with path entangled states
, we consider the generation of entangled states over the same mode but with orthogonal polarizations , respectively , horizontal ( @xmath12 ) and vertical ( @xmath13 ) . the scheme is shown in fig.1-b .
the non - linear crystal ( bbo ) is pumped by a high - power laser . the first order contribution to the output field
is the twin photons state over the same mode : @xmath14 , with @xmath15 .
such polarization - entangled photons can be easily converted to path entangled via an additional polarizing .
the of fig1.b mixes the two polarization components @xmath16 and @xmath17 as the second of fig1.a mixes the two different path modes , similar arrangements of polarization noon were adopted in ref.@xcite . :
count rates @xmath18 versus the phase @xmath19 .
( b ) @xmath20 : coincidence counts @xmath21 $ ] versus the phase @xmath22 let us introduce the generic quantum states generated by the opa acting on the vacuum fields @xmath23 .
the interaction hamiltonian of the optical parametric amplification @xmath24 acts on the single spatial mode @xmath25 a fundamental physical property of @xmath26 consists of its expression for any polarization basis belonging to the equatorial basis .
indeed @xmath26 can be written as @xmath27 where @xmath28 are the creation operators for the @xmath29 polarization modes , respectively .
the output state over the mode @xmath30 of the unseeded optical parametric amplifier is found to be : @xmath31 with @xmath32 , @xmath33 , being @xmath34 the non linear ( nl ) gain @xcite .
this state is usually called a two - mode squeezed state .
the average photon number created per polarization mode is equal to @xmath35 . in the interferometric setup
the output state is shifted by a phase @xmath1 in the basis @xmath36 .
hence the output state is detected in the basis @xmath37 adopting a polarizing beam splitter ( pbs ) . for different values of the phase @xmath1 ,
the quantum state @xmath38 is analyzed through two different second - order correlation functions @xmath39 and @xmath40 where @xmath41 are the output modes of the pbs . by tuning @xmath1 it is found @xmath42 and @xmath43 @xcite .
the corresponding visibilities of the obtained fringe patterns read @xmath44 and @xmath45 .
we observe that a non - vanishing visibility is found for any value of @xmath46 @xmath47 and @xmath48 .
the two fringe patterns exhibit a dependence on @xmath49 and hence a period equal to @xmath50 . this feature can be exploited to carry out interferometry with sub - rayleigh resolution , i.e. , with fringe period lower than @xmath4 , in a higher flux regime compared to the two photon configurations . : two - photon coincidence counts @xmath51 $ ] versus the phase @xmath19 introduced by the soleil - babinet compensator @xmath52 .
circle data : expected accidental ( without correlations ) coincidence rates .
( b ) @xmath53 two - photon coincidence counts @xmath21 $ ] versus the phase @xmath54 @xmath52 .
variations of the detectors signal @xmath55 due to different couplings of polarizations @xmath56 and @xmath57 with fiber have been corrected by normalizing coincidence counts with signal rates . ]
we now briefly describe the experimental configuration : fig.2 .
the excitation source was a ti : sa coherent mira mode - locked laser further amplified by a ti : sa regenerative rega device operating with pulse duration @xmath58 at a repetition rate of @xmath59 .
the output beam , frequency - doubled by second harmonic generation , provided the excitation beam of uv wavelength @xmath60 and power @xmath61 .
the horizontally polarized uv beam was then adopted to pump a non - linear bbo crystal , which generated pairs of photons with polarization @xmath62 and @xmath63 over the modes @xmath7 with same wavelength @xmath64 .
this source allows us to obtain a high value of the gain @xmath65 which depends on the pumping power : @xmath66 compared to conventional pulsed sources used to pump spdc process , which achieve @xmath67 , the energy per pulse is enhanced by a factor @xmath68 , leading to an increase of the gain value in the range of @xmath69 depending on the focal length of the uv beam .
the pumping power could be tuned adopting a half - wave plate and a polarizing beam splitter ( @xmath70 ) .
the output state of bbo crystal with wavelength @xmath4 was spatially separated by the fundamental uv beam through a dichroic mirror ( ) , then spectrally filtered adopting an interferential filter ( ) with bandwidth equal to @xmath71 and then coupled to a single mode fiber in order to select spatially a single mode of emission . a @xmath72 waveplate and a bbo with thickness of @xmath73 provided the compensation of walk - off effects . at the output of the fiber , after compensation ( ) of the polarization rotation induced by the fiber , a phase shifting @xmath1
was introduced adopting a soleil - babinet compensator ( ) .
the output radiation was then analyzed through a polarizing beam splitter ( ) and detected adopting single photon detectors spcm - aqr14 ( @xmath74 ) . to characterize the detection apparatus , a coherent state with wavelength @xmath4 and
polarization @xmath62 was fed into the mode @xmath75 the count rates @xmath18 and the coincidence rates @xmath21 $ ] were measured versus the phase @xmath1 : fig.3 .
high visibility patterns have been observed with a period equal to @xmath4 .
( b ) excitation rate versus non - linear gain .
the continuous line corresponds to the function @xmath76 where the parameter
@xmath77 has been optimized by fitting the data and reads @xmath78 ] as a first experimental step we have characterized the two - photon state generated by spdc in the low gain regime .
the visibilities have been found @xmath79 and @xmath80 .
the discrepancy with the expected values @xmath81 is attributed to double - pair emission and to experimental imperfections .
the same measurement has been carried out increasing the uv pump beam in order to measure the fringe patterns for different gain values .
the gain has been estimated with the method introduced in ref .
@xcite . fig .
4 refers to the configuration @xmath82 .
the visibilities have been found to be @xmath83 and @xmath84 .
the sub rayleigh resolution is clearly shown by the experimental data of fig.4 .
for the sake of completeness the value of @xmath85 has been measured for different gains : fig.5-a .
the continuous line shows the expected theoretical function @xmath86 multiplied by the extrapolated visibilities for @xmath87 @xmath88 .
we attribute the discrepancy between experimental and theoretical visibilities to partial multimode operation of the optical parametric amplifier @xcite .
( two - photon coincidences @xmath21 $ ] ) versus the phase @xmath19 ; square data : @xmath89 ( three - photon coincidences @xmath90 $ ] ) versus the phase @xmath19 @xmath91 ( b ) excitation rate versus non - linear gain .
the continuous line corresponds to the function @xmath92 .
inset : visibility @xmath93 versus non - linear gain . ] to estimate the dependence of the excitation rate from the nl gain @xmath34 , the maxima and the minima of the fringes have been measured versus the pumping power .
the average data are reported in fig.5-b .
our experiment validates the result found by agarwal et al .
@xcite showing an exponential dependence on the parameter @xmath34 .
the expected two - photon excitation rate reads @xmath94 where @xmath95 is a generalized two - photon excitation cross section @xcite .
fig.5-b shows how the excitation efficiency scales quadratically with the light intensity , in contrast with the two - photon spdc regime ; which leads to a linear dependence @xcite . as further demonstration of the potentialities of the present approach
, the simultaneous detection of three photons over the same mode has been investigated .
the average three photon excitation rate reads @xmath96 where @xmath97 is a generalized three - photon excitation cross section and the visibility of the fringes is theoretically found as @xmath98 @xcite@xmath99 again a non - vanishing value of @xmath93 is found for any value of @xmath46 @xmath100 and the patterns exhibit a period equal to @xmath50 .
furthermore an increase of visibility is expected @xmath101 . to demonstrate such a feature the three - photon coincidence rate
@xmath89 has been measured versus the phase @xmath1 : fig.6-a@xmath99 an increase of the visibility has been found @xmath102 , the experimental dependencies of the three photon absorption rates and visibilities are shown in fig.6-b . in order to investigate the connection between the quantum feature of the state and the visibility of the fringe pattern , we have introduced a decoherence between the two polarization components @xmath103 or @xmath104 with a quartz crystal ( ) with a length equal to @xmath105@xmath99 this device introduces a temporal delay higher than the coherence time of the multiphoton fields .
the @xmath106 has been first measured without the quartz : fig.7 . when the decoherence affects the @xmath103 components
, we observe a reduction of visibility down to @xmath107 ; on the other hand when the decoherence involves the @xmath104 we observe the disappearance of the fringe patterns .
the inversion of the maxima and minima of the square data compared to the circle data , a phenomenon not expected in the single - pair regime , is due to the reduction of the bunching effect among photons detected on the same polarization mode when decoherence is introduced . without introducing decoherence .
circle data : decoherence introduced between the polarization components@xmath108 .
triangle data : decoherence introduced between @xmath109 . ]
in the present paper we have experimentally demonstrated that the output state of a high - gain optical parametric amplifier can be intense ; yet exhibits quantum features .
we have extended the previous implementations @xcite , limited to the a - posteriori generation of a two photon state to a large number of photons .
the reduced visibility has been increased by means of three - photon absorption as suggested by @xcite . in this situation
a higher visibility has been achieved but without an increase of the fringe density .
the present article opens interesting perspectives for the implementation of a quantum interferometric setup . by adopting homodyne detection on the output fields , the heisenberg limit for the phase resolution
can be achieved @xcite .
recently it has been proposed to combine high - gain spdc with coherent states to create general noon states with a fidelity of about @xmath110 @xcite .
this work was supported by the prin 2005 of miur and progetto innesco 2006 ( cnism ) , aro ; dto and darpa . | quantum lithography proposes to adopt entangled quantum states in order to increase resolution in interferometry . in the present paper
we experimentally demonstrate that the output of a high - gain optical parametric amplifier can be intense yet exhibits quantum features , namely , sub - rayleigh fringes , as proposed by agarwal et al .
( phys .
rev .
lett . * 86 * , 1389 ( 2001 ) ) .
we investigate multiphoton states generated by a high - gain optical parametric amplifier operating with a quantum vacuum input for a gain values up to 2.5 .
the visibility has then been increased by means of three - photon absorption .
the present article opens interesting perspectives for the implementation of such an advanced interferometrical setup .
pacs : 03.67.-a , 03.67.hk , 42.65.lm |
we have calculated the cross section of radiative capture process @xmath10 .
we applied pionless eft to find numerical results for the m1 contributions for this capture process for incident neutron energies relevant for bbn , @xmath12 mev . at these energy our calculation is dominated by s - wave state and magnetic transition m1 contribution .
the error estimate in the cross section in comparison with evaluated nuclear data file endf @xcite is shown in fig .
[ error ] and table 1 .
errors estimate are 20 - 30 percent at leading order , below 10 percent up to nlo and by insertion of three - body force at n@xmath2lo , this error is reduced to below 1@xmath13 percent .
specially , our calculation shows minimum error for energy 60 - 70 kev up to at n@xmath2lo .
comparison of the lo with the nlo and n@xmath2lo results demonstrate convergence of the effective field theory .
finally , three - body forces will enter at higher orders of the eft approach and reduce the theoretical uncertainty .
the authors would like to thanks u. van kolck for helpful discussions .
we would like to thanks p.f .
bedaque and harald w. griehammer for useful comments and valuable mathematica code . | the cross section of neutron - deuteron radiative capture @xmath0 is calculated at energies relevant to big - bang nucleosynthesis ( @xmath1 kev ) with pionless effective field theory . at these energies ,
magnetic transition m1 gives the dominant contribution .
the m1 amplitude is calculated up to next - to - next - to leading order(n@xmath2lo ) with insertion of three - body force .
results are in good agreement within few percent theoretical uncertainty in comparison with available calculated data below e=200 kev . * h. sadeghi * and * s. bayegan * + _ department of physics , university of tehran , p.o.box 14395 - 547 , tehran , iran . _ + 1.0 cm 26.35.+c , 21.30.fe , 25.40.lw , 11.80.jy , 27.10.+h [ cols= " > , < " , ] at lo and nlo , this is the only three - body force entering , but at n@xmath2lo , where we saw that @xmath3 is required , it is determined by the triton binding energy @xmath4 .
we solve integral equation by expansion in order of @xmath5 and properly iterating the kernel .
then , the resulted @xmath6 will be folded to electromagnetic interaction order by order and properly integrated on the involving momentum . in kev .
the short dashed , long dashed and solid line correspond to the error up to lo , nlo and n@xmath2lo , respectively . ] for our calculation source of error due to low cutoffs and low momentums for very low energy calculation is neglected .
the cutoff variation decreases steadily as we increase the order of the calculation and is of the order of @xmath7 , where @xmath8 is the order of the calculation .
we used @xmath9 mev , for the smallest cutoff .
the cross section calculation for neutron radiative capture by deuteron as function of the center - of - mass energy at lo , nlo and n@xmath2lo is shown in fig . [ crosssection ] .
table 1 shows numerical results for the eft @xmath10 cross section for various nucleon center of mass energies e up to n@xmath2lo and errors estimate at every order in comparison with the last column .
the corresponding values for the cross section from the online evaluated nuclear data file endf / b - vi @xcite are shown in the last column .
the eft results for this cross section are presented up to only two significant digits(the third digit is shown for better comparison with endf data ) . in fig
. [ error ] , we show the error due to available evaluated data endf @xcite in percentage versus nucleon center - of - mass kinetic energy @xmath11 in kev . |
six kilometer - scale laser interferometers designed to observe gravitational waves ( gw ) with unprecedented sensitivities should complete or approach the end of their commissioning in the year 2003 .
three are operated in north america by the ligo laboratory @xcite , two in europe by the virgo @xcite and the geo600 @xcite projects , on one in asia by the tama300 @xcite project .
a collaborative analysis of the data collected by these instruments provides the best prospects for detecting and analyzing gw events of astronomical origin .
the focus of this article will be on `` bursts '' of gravitational radiation , loosely defined as gw of duration of the order of a few seconds at most , and present in a frequency range overlapping at least partially with the bandwidth of the interferometers ( 10 hz @xmath0 1 khz ) .
other types of signals that will not be discussed here include continuous gw from rotating neutron stars , and a stochastic background of gw of cosmological origin .
arguments based on the astrophysics and on the dynamics of the sources of gw bursts show that the detection of these signals will be challenging , as the signals will be deeply buried in the instrumental noise @xcite .
consequently , a significant research effort is currently on - going to develop and study efficient algorithms for the detection and the characterization of the elusive gw signals .
an important fraction of the literature on the subject concerns signals with a precisely known form @xcite .
the knowledge of the signal allows the construction of a phase coherent filter ( the wiener or matched filter ) which is known to be optimal for signal detection .
only the coalescence of compact binaries and possibly the ringdown of excited black holes should be detectable using matched filtering .
for the particular case of compact binaries , it is known that a coherent analysis using data from all the interferometers of the international network will improve the detection prospects noticeably @xcite , although the computational cost of such an analysis might be prohibitive @xcite .
in addition , it was shown in @xcite that the use of the advanced ligo detectors and of the virgo interferometer cooperatively might allow the localization of the gw source with enough accuracy to permit its observation with electromagnetic instruments , thus complementing with information about the thermodynamics of the source the information on its dynamics provided by the gw .
the remainder of this article will be concerned only with gw signals that are not known with enough precision to allow matched filtering .
the algorithms that have been proposed in the literature to detect these signals fall into two general categories : time - domain filters , and power detectors .
time - domain filters @xcite rely on the development of a small bank of linear filters which are expected to cover relatively well the space of possible gw signals .
they offer the advantages of speed , simplicity , and possibly ease of interpretation , but might lack the robustness and efficacy of power detectors .
only the latter will be discussed here .
the power detectors threshold on some non - linear measure of the data , often constructed from a time - frequency representation of the signal @xcite .
they have been shown to be optimal for the detection of signals with especially poor waveform descriptions .
all power detectors were explicitly designed and implemented to process data from the different interferometers of the world - wide network independently . under this mode of operation , it is expected that event lists are generated individually from the data stream provided by each interferometer , and are later compared to form coincidences based on temporal , frequency , or more general information . this
_ incoherent _
approach should not yield the maximum efficacy , in part because gw bursts in individual interferometers have to be rather loud to register with the power detector and to give accurate estimates of their start time , duration , frequency band , amplitude , etc .
, all of which might be needed by the coincidence gate .
the alternative is to combine all data streams first , and then run a burst detector on the _ synthetic _ data stream so produced .
i implement this _ coherent _ approach as a generalization of the power detectors developed to date for single interferometers , by calculating the optimal way to combine any number of interferometer data streams into a single time - series , such that when this time - series is fed to a single interferometer power detector , a larger signal - to - noise ratio is obtained than for any other combination of the data streams .
this brings to the already implemented and well - characterized power detectors the benefits of a network coherent analysis , which include improved sensitivity , and the ability to precisely locate the source position on the sky .
the coherent power filter ( cpf ) algorithm presented in this paper involves the following steps : 1 . a point on a grid in parameter space
is chosen , where the parameters are the source angular position , and two numbers describing the plus and cross polarizations of the gw signal .
one of these numbers is the ratio of the power in the cross polarization to the power in the plus polarization , and the other is the amount of overlap between the two polarizations , approximately measured by integrating the product of the two waveforms over time . 2 .
given the source position and the network topology , the data streams from all interferometers are time - shifted to align the gw signals to a common origin in time .
every data stream is multiplied by a scale factor , and all the data streams are added together to form a synthetic time - series .
the choice of the scale factors depends on the network topology and on the four parameters chosen in step 1 .
the synthetic time - series is processed by a power detector , and the power measurement is recorded .
if all points of the grid in parameter space have been visited , the algorithm exits .
if the maximum of the power measurements exceeds a certain threshold , a detection is announced , and the parameters that gave the largest power measurement are returned as an estimate of the source parameters .
to step 1 .
a more detailed discussion of this algorithm is presented in section [ algorithm ] .
the scale factors in step 3 are chosen so that the value of the signal - to - noise ratio ( snr ) is maximized .
it is quite significant that only two parameters in addition to the source position are required to perform this maximization .
geometrically , this can be understood by realizing that the signals in all interferometers , after being properly time - shifted , are linear combinations of the two polarization waveforms , and therefore lie in a hyperplane spanned by these two polarization waveforms .
consequently , a knowledge of the ratio of the lengths of the two polarization waveforms and of their angle with respect to each other , together with a knowledge of the beam - patterns of all interferometers , is sufficient to determine the signals in all interferometers , up to an overall scale factor , and up to the orientation of the hyperplane .
however , neither of these two pieces of information are needed to calculate how to linearly combine the signals to get maximum power in the synthetic time - series .
the validity of the geometrical picture and the actual conclusion that only four parameters are required to perform the snr maximization depends critically on the right choice of the position dependent time - shifts .
as shown in section [ algorithm ] , in the specific case where the cross - correlation function of the plus and the cross polarizations does not have an extremum at zero time lag , there are no formal guarantees that the cpf algorithm will converge to the right source parameters .
physically , this results from possible interactions ( or cross - terms ) between the plus and the cross polarizations which can not be properly handled by the coherent algorithm .
this does not affect the detection performances of cpf , but in some cases is significant for source position estimations . as discussed in section
[ algorithm ] , a number of canonical sources do not satisfy this condition exactly , so that a careful study of the position systematic errors is needed . for the difficult case where the two polarizations are long monochromatic signals with a phase difference of a quarter of a cycle , it is shown in section [ systematics ] that a correction for this systematic error can be implemented such that for @xmath1 of the sky the systematic error is negligible , while that for about @xmath2 of the sky it is too large to allow any position estimation .
this correction procedure only requires the additional knowledge of a quantity which is closely related to the characteristic frequency of the signal . the performances of the cpf algorithm are explored empirically in section [ simulations ] through numerical simulations .
all the experiments are limited to the three interferometer network ( the hlv network ) consisting of the ligo interferometer near hanford , washington , the ligo interferometer near livingston , louisiana , and the virgo interferometer in italy .
the signal is short ( 1/16 s ) and narrow band ( 25 hz ) , and is assumed to originate from a position along the northern hemisphere normal to the hlv plane .
all experiments are performed with the tfclusters @xcite algorithm as the single interferometer power detector . a realistic , computationally efficient hierarchical implementation of the cpf algorithm
is shown to offer better detection performances than a incoherent approach which uses only coincidences between events generated by independent tfclusters operating on the three interferometers .
it is also shown that the cpf algorithm can be used to estimate the position of the source of gw .
when the gw signal has four times more power in its plus polarization than in its cross polarization , roughly one quarter of all trials lead to a position estimate that is within one degree from the true source position , for signals with reasonable amplitudes .
the ability to pinpoint the source location is debilitated by the misalignment of virgo with respect to the ligo detectors , the reduction of the signal - to - noise ratio , and the reduction of the ratio of the power in the plus and in the cross polarizations .
let bold characters denote time - series ; whether these time - series are continuous in time or discretely sampled will be immaterial in the following discussion .
it is assumed that a gw is observed with a network of @xmath3 independent detectors .
calibrated data corresponding to measurements of the gw strain in all detectors are denoted @xmath4 , @xmath5 .
the noises @xmath6 are assumed to be additive , so @xmath7 { \bm{s}}_+ + f^\times_i(\theta , \phi , \psi ) t[\delta_i(\theta , \phi ) ] { \bm{s}}_\times + { \bm{n}}_i , \label{eq : sigmodel}\ ] ] where @xmath8 and @xmath9 are the two polarizations of the gw signal , @xmath10 , @xmath11 are the beam - pattern functions@xcite of the @xmath12 detector , and @xmath13 denotes the time - shift operator ; for time - series with continuous time , for instance , @xmath14 .
the beam - pattern functions depend on the two angles describing the source position ( the right ascension and the declination , denoted @xmath15 and @xmath16 respectively ) , and on the polarization angle @xmath17 .
the time - shift at the @xmath12 detector , denoted @xmath18 , is the same for the two polarizations , and depends only on the source position on the sky .
the frame in which @xmath8 and @xmath9 are defined is irrelevant since the waveforms are not assumed to be known a priori ; a rotation of that frame is equivalent to a change in @xmath8 , @xmath9 and @xmath17 .
the parameters @xmath19 , and @xmath9 , and the derived quantities @xmath20 , @xmath21 , and @xmath22 , will be used below to describe the parameters of a real source which is assumed to be present in the data , and which we are trying to detect .
the scalar product between two time - series is denoted @xmath23 . for time - series with continuous time
, it is defined as @xmath24 and similarly for time - series with discrete time .
the kernel @xmath25 can be viewed as a filter applied to the time - series in order to detect more efficiently a particular signal or to modify the character of the noise , for instance .
the square of the norm of a time - series , also called its power , is denoted @xmath26 , and is defined by @xmath27 .
the noise in each of the @xmath3 interferometers is only assumed to be wide - sense stationary @xcite , i.e. , it does not have to be gaussian or white .
the noises can always be made zero mean and independent by linear filtering @xcite , so @xmath28 = r_i(\delta)$ ] if @xmath29 and is zero otherwise , for @xmath30 the autocorrelation of the noise , and @xmath31 $ ] denoting the expectation value of its argument . as usual , the fourier transform of the noise autocorrelation function is the noise power spectral density .
the _ synthetic response _ of the network is denoted @xmath32 , and is a simple linear combination of the time - shifted individual detector responses : @xmath33 for some set of real coefficients @xmath34 and time shifts @xmath35 , @xmath36 , which are arranged in two vectors , @xmath37 and @xmath38 , respectively .
note that the @xmath35 are the trial time delays used in the data analysis , and the algorithm defined below is used to estimate these delays so that they are close from the real delays @xmath22 corresponding to a source located at position @xmath39 .
the _ network power _
@xmath40 is the estimate of the power in the gw signal , according to our norm definition , i.e. , @xmath41 .
the motivation behind this particular design is that it is the simplest generalization of the numerous power detectors for single interferometers described in the literature , which in many cases have already been implemented and characterized . in practical terms , a software code can be designed to compute the synthetic response for a network of interferometers , and these data can be fed to a power detector , as if they were data from a single interferometer , in order to measure the network power .
the kernel of the dot product used for the computation of the network power is then determined by the single interferometer power detector used to process the synthetic response .
different power detectors are efficient for detecting different types of signals , so this generality of the synthetic response approach is very economical in terms of code development .
some single interferometer power detectors provide a non - linear measure of the power ; this is not a serious limitation given the algorithm structure defined above , for the power measurements are all very nearly linear for detectable signals .
let @xmath42 denotes the error on the estimated time - of - flight between detectors @xmath43 and @xmath44 .
the network power can be expanded as @xmath45 where the signal term is given by @xmath46 , \label{eq : zeta}\end{aligned}\ ] ] and where the noise term is given by @xmath47.\ ] ] the signal correlation functions are given by @xmath48 for @xmath49 or @xmath50 .
the signal - to - noise ratio @xmath51 is defined as @xmath52},\ ] ] where the expectation of the noise can be rewritten as @xmath53 = \sum_{i = 1}^n a_i^2 \sigma^2_i,\ ] ] where the noise variance is @xmath54 . for fixed noises and signals
, the signal - to - noise ratio depends only on @xmath37 and @xmath38 .
it can therefore be maximized for a given choice of the source parameters ( @xmath55 ) by varying @xmath37 and @xmath38 .
let @xmath56 and @xmath57 denote those values of @xmath37 and @xmath38 which maximize @xmath58 for some source parameters identified by primes to differentiate them from the true source parameters .
the following algorithm is then defined : 1 .
pick a set of trial source parameters @xmath59 .
2 . compute @xmath56 and @xmath57 .
form the synthetic response @xmath32 from @xmath60 and @xmath61 .
estimate @xmath62 using a single detector algorithm .
retain the source parameters @xmath59 if @xmath63/e[\eta]$ ] is the largest to date .
go back to step 1 .
the expectation of @xmath63/e[\eta]$ ] is just @xmath58 , so on average this algorithm will converge to the true parameters of the source .
the maximization problem for @xmath58 can be recast as the maximization of @xmath64 subjected to the constraint that @xmath65 $ ] is constant .
the normal equations are @xmath66 = 0 \label{eq : normdelta}\\ \lambda
a_i \sigma_i^2 + a_i [ f^+_i f^+_i { \bm{s}}_+ \cdot { \bm{s}}_+ + ( f^+_i f^\times_i + f^\times_i f^+_i ) { \bm{s}}_+ \cdot { \bm{s}}_\times + f^\times_i f^\times_i { \bm{s}}_\times \cdot { \bm{s}}_\times ] + \nonumber \\ \sum_{j=1 , j\neq i}^n a_j [ f^+_i f^+_j r_{++}(\delta_{ij } ) + f^+_i f^\times_j r_{+\times}(\delta_{ij } ) + \nonumber \\ f^\times_i f^+_j r_{\times+}(\delta_{ij } ) + f^\times_i f^\times_j r_{\times\times}(\delta_{ij } ) ] = 0 , \label{eq : normalpha } \end{aligned}\ ] ] where @xmath67 is the lagrange parameter for the constraint , @xmath68 , and @xmath69 . eq . ( [ eq : normdelta ] ) can be simplified using the identity @xmath70 : @xmath71 = 0 .
\label{eq : normdeltai}\ ] ] if @xmath72 has an extremum at @xmath73 , i.e. if @xmath74 , then a solution to eq .
( [ eq : normdeltai ] ) is @xmath75 for @xmath76 , since @xmath77 and @xmath78 are maximal at @xmath73 . back into eq .
( [ eq : normalpha ] ) , this solution gives @xmath79 = 0 .
\label{eq : normalphaii}\end{aligned}\ ] ] the choice of @xmath37 to maximize the signal - to - noise ratio then depends on the angles @xmath80 as before , but now only on the two numbers @xmath81 and @xmath82 instead of the full waveforms for the two polarizations . these numbers are denoted @xmath83 and @xmath84 , respectively .
( [ eq : normalphaii ] ) is an eigenvalue problem ( with the lagrange parameter @xmath67 playing the role of the eigenvalue , and the weight vector @xmath37 that of the eigenvector ) , and is straightforward to solve numerically .
the matrix of the eigenvalue problem is hermitian , so all eigenvalues are real ; the eigenvector which maximizes eq .
( [ eq : zeta ] ) with @xmath85 is picked to form the synthetic response . to recapitulate ,
if @xmath86 has an extremum at zero , a synthetic waveform can be constructed from the data of all interferometers such that this waveform is optimal for its processing by power detectors , and the maximization of the detection statistic for parameter estimation can be performed over only four parameters : @xmath87 and @xmath84 .
i do not include @xmath17 in this list of parameters because it is completely degenerated with @xmath83 and @xmath84 .
in particular , changing the basis where the two polarizations are defined by a rotation ( i.e. , a redefinition of the polarization angle @xmath17 ) does not change the value of @xmath88 . for
the reminder of this paper , i will take @xmath89 ; this fixes the definition of @xmath8 and @xmath9 with respect to the frame of reference of the detectors network .
the cpf algorithm is then defined as followed : 1 .
pick a set of trial source parameters @xmath90 , @xmath91 , @xmath92 , and @xmath93 .
2 . compute @xmath37 from eq.([eq : normalphaii ] ) , and set @xmath94 .
3 . form the synthetic response @xmath32 using eq.([eq : syntheticresponse ] ) .
4 . use a single interferometer power detector to calculate @xmath41 .
retain the source parameters @xmath90 , @xmath91 , @xmath92 , and @xmath93 if @xmath40 is the largest to date .
go back to step 1 .
if the signal is linearly polarized , it can be written as @xmath95 and @xmath96 , for some polarization angle @xmath17 and some waveform @xmath97 .
the eigenmatrix in eq .
( [ eq : normalphaii ] ) is then of rank one , and consequently only has a single non - trivial eigenvalue .
the corresponding eigenvector is given by @xmath98 / \sigma_i^2 = f^+_i(\theta , \phi , \psi ) / \sigma_i^2 $ ] for @xmath99 .
hence , when the gw signal is linearly polarized , the signals from the different interferometers are optimally combined by weighting them with the ratio of their noise variance to the value of the beam pattern functions , weighted appropriately by the polarization angle .
the signals from the different interferometers of the network are therefore emphasized linearly in the observed power of the gw signal , and inversely in the power of the noise .
another interesting subcase involves _ directed _ searches : in that case , the position @xmath100 of a potential gw source is known precisely ( @xmath101 ) , and the goal is to be maximally sensitive to gravitational radiation from that source . for @xmath102 ,
the maximization of the signal - to - noise ratio can be rewritten as the maximization of the quadratic form @xmath103 subjected to @xmath104 where the matrix @xmath105 has elements @xmath106 given by @xmath107.\ ] ] it is well known from rayleigh s principle @xcite that the maximum value of @xmath58 is given by the largest eigenvalue of the matrix @xmath105 .
this maximum signal - to - noise ratio can be compared to the signal - to - noise ratio that can be obtained using the best interferometer in the network , which is typical of the signal - to - noise ratio of an incoherent search .
it is given by @xmath108 considering the hlv network simplified so that all interferometers have the same noise level ( @xmath109constant ) , and fixing the beam - pattern functions by selecting a source along the northern hemisphere normal of the hlv plane , the ratio @xmath110 varies between 1.03 and 1.57 , depending on the value of @xmath83 and @xmath84 .
figure [ fig : snrratios ] shows that variation of the ratio @xmath110 with @xmath83 and @xmath84 : when the signal has a large degree of linear polarization , the improvement is fairly large , with @xmath111 . when both polarizations roughly contain the same power , the improvement can be large ( @xmath112 ) or very small ( @xmath113 ) , depending on the structure of the signal ( i.e. , on @xmath84 ) .
if we add a detector at tama s location to the hlv network , with noise similar to the noise of the other interferometers , the ratio @xmath110 varies between 1.15 and 1.79 for a source at the same location as before .
the four interferometer network works better than the three interferometer hlv network , although the improvement is somewhat limited by the unavoidable misalignment between instruments located on different continents . as a function of @xmath83 and @xmath84 , for a source along the northern hemisphere normal of the hlv plane.,title="fig:",height=288 ] if @xmath86 does not have an extremum at zero , the normal equations are not necessarily satisfied at @xmath75 .
they become non - linear and rather complex to solve , but , more importantly , a full knowledge of the waveforms for the two polarizations is then necessary to obtain a solution . physically , this is a result of the fact that the two polarizations may interfere together constructively when shifted by a non - zero lag ; this lag gets added to the estimated time delay between the interferometers in the network , and a systematic error in the source position estimate appears .
it is important to realize that this systematic error does not significantly reduce the detection capabilities of cpf , but only its positioning ability .
essentially , the convergence to a secondary maximum provides an alternative way to cross the detection threshold when the global maximum of the signal power is plagued by noise , such that the measured power at its position is not a global maximum of the measured network power .
it is nevertheless interesting to estimate the size of the position systematic error when the cpf algorithm is used on a signal for which @xmath114 does not hold , i.e. when eq .
( [ eq : normalphaii ] ) is imposed as a solution .
this is the subject of the following subsection . before this question
is addressed , however , it is worth examining the physical meaning of the condition @xmath74 . as it was pointed out by (
2.30d , 4.3 , and 5.18a ) , the gw strain for slow - motion , weak gravity sources radiating mostly due to variations of their mass moments ( by contrast to current moments ) can be written as @xmath115 where stt means `` symmetric transverse - traceless '' , @xmath116 are the spherical harmonics , and @xmath51 is the mass density of the source .
a number of possible gw sources consist in anisotropic mass distributions that are rapidly rotating about a well - defined axis , and consequently radiate principally with @xmath117 , @xmath118 ( e.g. , binaries , bar or fragmentation instabilities , longest live mode of a perturbed kerr black hole , as noted by @xcite ) .
the imaginary part of @xmath119 is rotated by @xmath120 about the polar axis ( the rotation axis ) with respect to its imaginary part .
this results in the mass distribution being sampled at any given time according to two spatial patterns rotated by @xmath120 with respect to each other . since the gw are emitted at twice the rotation frequency of the source , this rotation angle produces a @xmath121 phase difference between the real and imaginary parts of the time dependent integral in eq .
( [ eq : hsph ] ) .
as noted by @xcite , the dependence of the polarization on the inclination angle comes from the pure - spin tensor harmonics @xmath122 , which for @xmath117 can be found in @xcite . for @xmath118 , @xmath123 where @xmath124 is the inclination angle of the rotation axis with respect to the line of sight ( @xmath125 along the polar axis ) , and where @xmath126 and @xmath127 are the unit linear - polarization tensors for the gw .
consequently , the plus and cross polarization waveforms can be written as @xmath128 where @xmath129 is some phase function . by definition
, equations ( [ eq : pluspolarizations ] ) and ( [ eq : crosspolarizations ] ) give @xmath130 so that @xmath131 in general , the phase @xmath129 will have at least a linear component [ @xmath132 , so that @xmath133 as the system s polar axis becomes aligned with the line - of - sight , the assumption that @xmath74 becomes progressively worst , and significant systematic position errors might appear , and will have to be accounted for , as discussed in the next section .
some signals might also satisfy the condition @xmath74 .
this is the case , for instance , if @xmath129 is an even function of @xmath134 .
sources which radiate predominantly in a @xmath117 , @xmath135 mode will show a similar correlation between their plus and cross polarizations . in that case , the real and imaginary parts of @xmath136 differ by a @xmath121 rotation about the polar axis , but have a @xmath135 symmetry , so that the waves are radiated at the spin frequency , and consequently the phase shift between the plus and cross polarizations is again @xmath121
. it might be that the dominating population of sources to be observed will not be dominated by rotation about a principal axis ; as a result , the correlation between the plus and cross polarizations might be quite arbitrary . in an axisymmetric core collapse , for instance , the @xmath117 , @xmath137 mode dominates @xcite .
the angular response is @xmath138 , so the waves are linearly polarized .
( [ eq : normdeltai ] ) can be used to check the error on @xmath15 and @xmath16 by solving it for @xmath38 , with @xmath37 obtained from the cpf algorithm , i.e. , from the solution of eq .
( [ eq : normalphaii ] ) . for every trial choice of @xmath139 , the algorithm returns a set of weights @xmath37 ( by definition of cpf , @xmath140 ) .
however , the maximum of the signal - to - noise ratio occurs when the normal equations are satisfied ; assuming that we let @xmath83 and @xmath84 be varied freely in the maximization of @xmath40 , the values of @xmath15 and @xmath16 returned by this maximization will be those that solve eq .
( [ eq : normdeltai ] ) .
consider the following parameterization of the signal cross - correlation functions : @xmath141 for some parameters @xmath142 , @xmath143 , and @xmath144 .
if @xmath8 and @xmath9 were long , nearly monochromatic signals of angular frequency @xmath145 , with the same amplitude but a phase difference of @xmath121 , for instance , the parameters could be chosen to be @xmath146 . in order to get an analytical solution ,
i linearize eq .
( [ eq : normdeltai ] ) to obtain the first order equation @xmath147 = 0,\ ] ] where @xmath148 represent first order errors between the true time delays and the delays returned by the algorithm , and where @xmath149 and @xmath150 .
the solutions to this linear system of equations are degenerate ; for @xmath151 , they are @xmath152 and @xmath153 the times - of - flight between two pairs of detectors are sufficient to triangulate the position of the source on the sky and to obtain its position @xmath39 , up to a reflection with respect to the plane containing the three detectors . the magnitude @xmath154 of the systematic error due to @xmath155
is given by the arclength of the portion of the great circle connecting the true source position and the position obtained by triangulation from @xmath156 and @xmath157 in eqs .
( [ eq : delta13 ] ) and ( [ eq : delta23 ] ) . to give a representative example ,
@xmath154 is computed for the long monochromatic signal described above , for the hlv network with @xmath158 , i.e. under the simplifying assumption of instruments with identical noises at all sites .
only a very small fraction of the sky ( about 0.7% ) has a negligible systematic error [ @xmath159 rad ] .
it is therefore plain that position estimations will be grossly off target if the assumption that @xmath114 is wrongly made . when @xmath155 , more information about the waveforms is required to estimate the source position .
one possible approach is to use the cpf algorithm , which assumes @xmath114 , and then to correct the source position estimate using eqs .
( [ eq : delta13 ] ) and ( [ eq : delta23 ] ) .
this requires only a knowledge of the slope of @xmath86 with respect to the slope of @xmath160 at zero lag ( i.e. , @xmath161 ) , which is closely related to the characteristic frequency of the waveforms . a map from estimated position ( with systematic error ) to actual position
must be constructed for every choice of @xmath161 .
the remaining systematic error is given by the higher order terms not included in the correction .
for @xmath162 the second order errors on the time delays , the second order equation derived from the linearization of eq .
( [ eq : normdeltai ] ) is @xmath163 = 0.\ ] ] this linear system of equations can be solved to obtain : @xmath164 + a_3 \ { a_3 \delta^{(1)}_{12 } \delta^{(1)}_{13 } g_{13 } [ f_{23 } + 2\frac{\omega_2 ^ 3}{\omega_1 ^ 2 } \delta^{(1)}_{23 } g_{23 } ] + \nonumber \\ a_2 [ \delta^{(1)2}_{23 } f_{12 } g_{23 } + \delta^{(1)}_{12 } g_{12 } ( \delta^{(1)}_{12 } f_{23 } + 2\frac{\omega_2 ^ 3}{\omega_1 ^ 2 } \delta^{(1)}_{23 } ( \delta^{(1)}_{12 } + \delta^{(1)}_{23 } ) g_{23 } ) ] \ } \end{aligned}\ ] ] and @xmath165 + a_1 \ { 2 a_2 \delta^{(1)}_{12 } \delta^{(1)}_{13 } f_{12 } g_{12 } + \nonumber \\ a_3 [ \delta^{(1)2}_{13 } f_{12 } g_{13 } + \delta^{(1)}_{12 } g_{12 } ( \delta^{(1)}_{12 } f_{13 } + 2\frac{\omega_2 ^ 3}{\omega_1 ^ 2 } \delta^{(1)}_{13 } ( \delta^{(1)}_{12 } - \delta^{(1)}_{13 } ) g_{13 } ) ] \},\end{aligned}\ ] ] where @xmath166 \}.\end{aligned}\ ] ] for @xmath167 the arclength of the portion of the great circle connecting the true source position and the position obtained by triangulation using @xmath162 , i.e. , for @xmath167 the systematic error after correction using the first order expansion , and for the same example as above ( with @xmath168 rad / s ) , one finds that 23.2% of the sky has a negligible systematic error ( @xmath169 rad ) .
figure [ fig : syserr ] shows the fraction of the sky with a systematic error smaller than a certain value , for @xmath154 and @xmath167 .
it should be noted that for only about 50% of the sky is it meaningful to use an expansion in the small parameters @xmath148 and @xmath162 .
all in all , these numbers show that it is possible to use the cpf algorithm for any signals , at the cost of limiting the sky coverage to @xmath1 , and of requiring one additional piece of information about the signal , the value of the ratio @xmath161 , which describes the behavior of the signal cross - correlation functions near zero lag .
shorter signals , or signals with less overlap between their two polarizations , are likely to offer smaller systematic errors on position estimates , so that the correction suggested above becomes unnecessary .
rad / s , with two polarizations of the same amplitude , but at a phase offset of @xmath121 .
the rightmost curve is for @xmath154 , the leftmost one for @xmath167.,title="fig:",height=288 ]
all the numerical simulations were performed using the cpf implementation for the ligo data analysis system ( ldas ) , which uses the tfclusters algorithm @xcite to measure the signal power in a given time - series .
the hlv network was analyzed , except that the noises of all three interferometers were assumed identical , for simplicity .
for every realization of the simulation procedure , a 10 s long segment of white noise of unit variance , sampled at 16384 hz , was generated independently for all three interferometers .
the use of white instead of colored noise is not an important loss of generality , since tfclusters is quite robust with respect to the presence of correlations in the background noise .
simultaneously , two 1/8 s long segments of unit variance white noise were generated , and filtered by a 6@xmath170 order elliptical bandpass filter with 3 db cutoff frequencies at 125 hz and 150 hz , so that the amount of power outside this band was negligible after filtering .
these two segments were then truncated to their central 1/16 s long portion , and were used as the plus and cross polarizations of the gravitational wave signal incident on the network of detectors . by construction , the average value of @xmath83 was 1 , and that of @xmath84 was zero . in some simulations , the plus polarization waveform was multiplied by @xmath171 and the cross polarization waveform was divided by @xmath171 , so that on average @xmath83 was equal to 2 .
all gravitational wave signals were injected from the position corresponding to the northern hemisphere normal of the hlv plane , so that they arrived in phase at the three interferometers .
the two polarizations were combined at each interferometer using the beam - pattern functions @xmath10 and @xmath11 , and were added directly to the background white noise .
a scale factor @xmath172 was multiplying the gw signal injected in the simulated data for all interferometers .
if the frequency band occupied by the signal , its duration , and its arrival time had all been known exactly , it would have been possible to filter the data optimally in that band . in that case , the signal - to - noise ratio @xmath173 for a certain scale factor @xmath172 would have been given by @xmath174 of course , it can not be assumed that this information is available , but i will nevertheless use eq.([eq : snropt ] ) as a measure of the strength of the injected signals , instead of using @xmath172 which is less intuitive .
numerically , @xmath175 for a source at the northern hemisphere normal of the hlv plane .
the simulations were performed using a realistic hierarchical implementation of cpf .
the simulated data from the three interferometers were first processed separately by tfclusters to produce three lists of events . in terms of the notation developed in @xcite ,
the settings of tfclusters were @xmath176 , @xmath177 , and @xmath178 $ ] .
only the frequencies below 1024 hz were considered , and the time resolution of the time - frequency decomposition was @xmath179 s. the number of events in this first stage was controlled by the black pixel probability threshold , @xmath180 ( note that larger values of @xmath180 give _ larger _ false alarm rates ) .
the events produced by tfclusters are in the form of rectangles in the time - frequency plane , with information about the power present in each pixel of these rectangles . for a given cluster identified by tfclusters ,
the rectangle is defined as the smallest rectangle containing all the pixels of the cluster .
a triple - coincidence condition was therefore applied as in @xcite : to be coincident , the time - frequency rectangles corresponding to the events from all three interferometers had to be overlapping .
this coincidence condition selects events that are close in time and in frequency , and can be understood as a standard time and frequency coincidence gate with varying windows that are fixed by the events under consideration .
all the coincident events that were present in a given 10 s segment were then considered in turn .
their start time , duration , central frequency and bandwidth were estimated from the smallest rectangle in the time - frequency plane that could contain the union of the rectangles from the individual events . the cpf algorithm was then ran on the data , once for each coincident event .
the implementation employed used tfclusters to process the synthetic data , and the measure of the power was the sum of the power in all the pixels identified by tfclusters that were inside the time - frequency rectangle identified from the triple coincidence .
the parameters for tfclusters were the same as those mentioned above , except for the black pixel probability , which was set to a value @xmath181 .
the threshold defined by @xmath181 was chosen so that only loud enough signals were detected by tfclusters , and their estimated power was linearly related to their actual power .
the power measured by tfclusters was maximized over the source position ( two angles ) and over the parameter @xmath84 .
it was assumed that the value of @xmath83 was known beforehand , in order to keep the size of the parameter space small enough for simulations . in a first time , the sky was covered by picking 100 points uniformly distributed in the range @xmath182 for the right ascension , and 100 points in @xmath183 , uniform in the sine of the declination angle .
in addition , 10 points were used to cover the range @xmath183 uniformly for the parameter @xmath84 .
consequently , @xmath184 was ran @xmath185 times on every 10 s long simulation . including the overhead from the ldas system , this part of the search ran in @xmath186 s on 31 , 2 ghz pentium iv computers , with 512 mb of ram . if none of the triple coincidence events registered above the @xmath181 threshold of tfclusters when analyzed by cpf , a non - detection was reported and the analysis was stopped
. otherwise , a detection was announced , and cpf produced a scan of the parameter space for every triple coincidence event above threshold . the triple coincidence event with the largest maximum power
was then selected as a possible gw candidate , and was analyzed in more details .
the point in parameter space where the power was maximum defined the parameters for a second run of cpf , used to obtain a refined position estimate not limited by the coarseness of the grid covering the parameter space .
the value of @xmath84 was fixed to the value estimated in the first run , and a square search window of size 0.2 rad in right ascension and in declination , with 50 steps in both angles , was centered on the value of the position obtained in the first run . the position with the maximum power in this second run
was taken as the final estimate of the source position . for the sole purpose of detecting the presence of a signal in the data , only the first run of cpf
was required .
data were ran through tfclusters separately , the events were combined in the triple coincidence gate , and the data were fetch through cpf for the parameters defined by each triple coincidence event . if at least one of the triple coincidence event lead to a detection by cpf , the 10 s long time interval under scrutiny was assumed to contain a signal . by design , this hierarchical scheme required a fairly permissive threshold in the first stage where tfclusters was ran independently on every interferometer , so that a given signal was very likely to make it to the second stage where cpf was ran .
the limit on this threshold was determined by the availability of computational resources , and by the confusion that resulted from the proliferation of events at low threshold .
most of the rejection of accidental coincidences occurred at the second stage , where cpf was operated with a reasonably strict threshold .
a numerical experiment was performed by running this simulation a large number of times , with and without signal injection . when injected , the signal had @xmath187 and @xmath188 .
the thresholds were chosen to be @xmath189 and @xmath190 . in the first stage ( triple coincidence ) , the probability to detect a signal was @xmath191 , and the probability of a false alarm when no signal was present was @xmath192 . in the second stage ( cpf ) , it was measured that @xmath193 .
overall , it was measured with both staged combined that @xmath194 .
table [ tab : detprob ] gives the detailed results from the simulations .
the errors quoted here come from 68.3% confidence intervals ( `` @xmath195 '' ) for a bernoulli process , built using the prescription of @xcite .
the 6.4% probability of false alarm is larger than typical values for gw searches , as it would give a false alarm rate around 7 mhz .
it was chosen , however , in order to provide enough detections for small errors on the measured probabilities . in a more realistic setting , @xmath180 would be similar to the value used here , while @xmath181 might be smaller by one or two orders of magnitude .
it should be noted that the threshold settings were found by trial and error ; there is an infinity of points along a curve in the @xmath196-plane that give a 6.4% probability of false alarm , and a choice different than the one above may give a larger probability of detection .
.details of the simulations to measure the efficacy of the hierarchical implementation of the cpf .
numbers in brackets show the 68.3% ( `` @xmath195 '' ) confidence interval for the fraction of the number of trials to the total number of trials . [
cols="^,^,^ " , ] nevertheless , it seems that the choice i made for the @xmath180 and @xmath181 thresholds is sufficient to show the superiority of the coherent approach over the incoherent one for detection .
i repeated the experiment above , but using only the first stage to detect events .
a detection was announced when at least one triple coincidence was observed between the outputs of the three tfclusters runs on the interferometers data . in that case , a threshold of @xmath197 gave @xmath198 , in an experiment with 742 trials for the measurement of @xmath199 ( 519 detections ) , and 742 trials for @xmath200 ( 48 detections ) . at a similar false alarm probability ,
the probability of detection is significantly smaller in the incoherent case than in the coherent case . for the particular signal and false alarm probability under consideration
, the signal - to - noise ratio would have to be increased to @xmath201 in order for the incoherent approach to be as efficient as the coherent one .
for a homogeneous distribution of sources in space , this corresponds to a factor of @xmath202 improvement in detection rate , assuming no significant degradation of the cpf algorithm performances with respect to the incoherent algorithm as the position of the source is varied away from the northern hemisphere normal to the hlv plane .
the performances of the cpf search were mostly limited by the quality of the estimation of the time - frequency rectangle containing the burst , in the first stage of the analysis ( the triple coincidence ) . with the signal injection for @xmath187 ,
the first stage gave an estimated rectangle that overlapped with the one containing the signal in 96% of the 578 cases where the signal was discovered , but only in 3% of the 61 missed detections was this the case . without signal injection , only 1% of the 611 triple coincidences had time - frequency rectangles overlapping with the signal rectangle .
these numbers show that if an oracle were available to provide the rectangle containing the signal without error every time the search is ran , a probability of detection @xmath203 would be possible for a probability of false alarm @xmath204 . stated differently , cpf is extremely efficient at detecting a burst when it receives the right parameters describing that burst ; the triple coincidence incoherent search provides many candidates ; when one such candidate corresponds to the signal , cpf picks it out of the others very efficiently . a better approximation to this oracle than the one used here might be to tile the time - frequency plane with a variety of rectangles , and to run cpf on each rectangle .
suppose that it is known that the signal has a bandwidth of 25 hz .
one can cover the time - frequency plane with non - overlapping rectangles of duration 1/16 s , and bandwidth of 25 hz , so that the 10 s long , 1024 hz bandwidth data segment in my simulations is covered by @xmath205 tiles .
the false alarm probability for each run of cpf must be reduced to @xmath206 so that the global search has @xmath207 .
the efficiency of the search would then approximately be given by the probability of detection of cpf with an oracle , for a threshold giving @xmath208 .
i have not measured this probability of detection , but it is plausible that it is larger than the 79% efficiency measured for the hierarchical implementation used in the simulations . however , running this search in real time would be prohibitively expensive , as it would require @xmath209 teraflops of computational power . in the present hierarchical implementation ,
the first stage took @xmath210 gigaflops to run in real time , and the second stage , @xmath211 gigaflops , where @xmath200 is the false alarm probability in the first stage .
these numbers should be taken as upper bounds on the required computational power , because the codes were not optimized to minimize overhead , to make an optimal usage of the parallel resources , or to do an optimal scan of the parameter space .
optimized codes should be able to run at least @xmath212 times faster . in order to get a precise idea of the magnitude of the position estimation errors , the analysis scheme described above
was simplified by removing the first incoherent step involving the three different instances of tfclusters running at each site . instead , the cpf algorithm was instructed to compute the power according to the output of tfclusters in a rectangle of duration 1/8 s located at the right position in the time series , with a lower frequency of 50 hz and an upper frequency of 150 hz . the black pixel probability was set to @xmath213 , so that the number of clusters unrelated to the signal and produced only by the noise was small .
figure [ fig : posu10 ] presents a scatter plot of the position estimates obtained for 240 realizations of the simulation , when @xmath188 .
the estimates tend to cluster along the curve corresponding to the locus of positions having equal delays at the hanford and livingston interferometers .
this is a direct consequence of the good alignment between these two interferometers , and the relatively poor alignment of the virgo detector with them .
estimates tend to fall on that curve , but also to cluster at different places along it , where the signals at hanford and at livingston are delayed by an integer number of the characteristic periods of the signal with respect to the signal at virgo . and @xmath187 .
horizontal axis is right ascension , vertical axis is the sine of the declination .
the curves represent loci of equal time delay for the three independent interferometer pairs .
the point where they intersect in the upper left corner of the figure is where the signal was injected.,title="fig:",height=288 ] figure [ fig : posu210 ] shows a similar plot as fig.[fig : posu10 ] , except that @xmath214 .
this corresponds to a gw signal which has more structure in its polarizations than the one for the case @xmath188 , i.e. which is closer to a linearly polarized signal .
since linearly polarized signals are the easiest ones to analyze with a network of interferometers , it is expected that the position estimates will be better .
this is indeed was is observed in fig.[fig : posu210 ] ; the estimates still hug the hanford - livingston equal - delay curve , but now present less scatter around the points where the signals are in phase with the virgo signal along that curve . , but with @xmath214.,title="fig:",height=288 ] the position error can be quantified as in fig.[fig : syserr ] : it is taken to be the length of the shortest portion of the great circle joining the estimated position and the true source position or its mirror image with respect to the hlv plane .
the error can be defined with respect to the mirror image because there is a natural ambivalence in the estimation of the position when only three detectors are used .
figure [ fig : poserr ] shows the cumulative distribution of the position error from the simulations , for @xmath188 and for @xmath214 , and for two different values of the signal - to - noise ratio of the injected signals . as expected , signals with larger values of the signal - to - noise ratio or of @xmath83 lead to smaller position errors . the curves in fig.[fig : poserr ]
present a number of `` steps '' , which are produced by the clustering along the hanford - livingston equal - delay curve at positions in phase with virgo . , and the two dotted lines to @xmath214 . in both cases ,
the rightmost curve is for @xmath187 , and the leftmost one is for @xmath215 .
each curve is built from 240 realizations of the simulation .
the error on the curves is estimated to be @xmath216 of the fraction of trials.,title="fig:",height=288 ] roughly 50% of the trials lead to unusable position estimates ( errors @xmath217 10 degrees ) when @xmath214 , while this number reaches @xmath218 when @xmath188 . however , with @xmath214 , approximately 25% of the trials have errors smaller than one degree . moreover , at least in the regime of signal - to - noise ratios under consideration here , the scaling of the position error with the signal - to - noise ratio is rather weak .
a method was presented for the optimal generalization of the power detectors developed for single interferometers so that they can process coherently data from a network of interferometers .
the coherent method , as compared to an incoherent approach where event lists independently generated at all interferometers are searched for coincidences , offers the advantage of better detection efficiency , and the possibility to accurately estimate the position of the source .
a few systematic effects affect the performances of the cpf algorithm for position estimation , including : cross terms between the plus and cross polarization waveforms of the gw signal , lack of differences between the characteristics of the two polarization waveforms , and misalignment of the interferometers of the network .
the three effects are related , and reflect the obvious fact that gw signals incident on a network of misaligned detectors will only show entangled versions of their two polarization waveforms , different in each interferometer . if one of the two polarizations is significantly stronger than the other ( i.e. , the gw signal has a stronger degree of linear polarization ) , the problem is drastically simplified .
similarly , aligned interferometers are much less sensitive to this problem
. if the two polarization waveforms are fairly coherent with each other and of similar amplitudes , the cross terms between the two polarizations in different interferometers may show significant maxima when the time shifts imposed on the different data streams do not correspond to the differences in time - of - arrival of the gw signals at each interferometer .
these maxima can not be distinguished from the maximum resulting from the product of the waveforms of the same polarization in different interferometers , and systematic position errors may result .
it should be noted , however , that the same effect leads to similar position errors when the differences in arrival time of the gw signal in different interferometers are estimated by maximizing the cross - correlation between pairs of interferometers , and these time differences are used to triangulate the source position .
the study of the cpf algorithm presented here was based on a three interferometer network .
however , the structure of the algorithm allows for any number of interferometers .
more complex networks will most likely reduce the effects of the systematic errors , by increasing the number of linear combinations of the two polarization waveforms that are being sampled , or , equivalently , by increasing the area of the sky where at least three interferometers are measuring similar combinations of the two polarizations . numerical simulations on the network formed by the ligo hanford , ligo livingston , and virgo interferometers ( the hlv network ) with a short random signal showed that the cpf algorithm could be used to accurately measure the position of the source a significant fraction ( @xmath1 ) of the time , for reasonably strong sources located at the normal of the hlv plane . in about @xmath219 of the trials ,
the cpf algorithm correctly placed the source position at a point where the signals at the hanford detector and at the livingston detector were in phase .
however , due to its misalignment with respect to the ligo detectors , the information provided by the virgo detector was often insufficient to pinpoint correctly the position of the source . the average position error was a rather weak function of the strength of the signal , but a stronger function of the amount of difference in the structure of the two polarization waveforms .
it was also shown that the cpf algorithm offers better detection efficiencies than its incoherent equivalent , both for directed and for all - sky blind searches , and independently of the systematic errors affecting the position estimations . in the former case ,
improvements in the detection signal - to - noise ratio of 40% or better are expected , excepted for a few values of the source parameters . in the latter case ,
a 25% improvement in signal - to - noise ratio was measured for typical source parameters .
this improvement for the all - sky blind search comes at the cost of increasing the computational power required to perform the data analysis in real time by a factor of @xmath220 .
this may be significant , especially since lengthy simulations probably have to be completed in order to estimate the false alarm and detection statistics of a real search .
however , the signals studied in the numerical simulations , which could be viewed as very rough approximations to the signals that could be emitted in the collapse of the core of a star in supernova explosions , would have been detected by the cpf algorithm at a rate @xmath202 times larger than the rate for its incoherent equivalent , assuming a homogeneous spatial distribution of the sources , and that the performances of the cpf algorithm for a source injected at the normal of the hlv plane are characteristic of its performances at other injection positions .
finally , an important advantage of the design of the cpf algorithm is that it inherits the robustness , efficacy , and computational efficiency of the single interferometer power detectors . in other words
, the cpf algorithm should not be more sensitive to non - gaussian or non - stationary noises than incoherent analyzes , it should be searching more efficiently for gw signals sharing similar morphologies , and , while more computationally intensive , its speed should scales the same way with improvements in the implementation of the single interferometer power detectors , or with computing hardware ameliorations .
this work was supported by the national science foundation under cooperative agreement phy-9210038 and the award phy-9801158 .
this document has been assigned ligo laboratory document number ligo - p030007 - 01-d .
see , for instance , n. arnaud , _ et al .
d * 59 * , 082002 ( 1999 ) ; t. pradier , _ et al .
d * 9 * , 309 ( 2000 ) ; t. pradier , _ et al .
_ , phys . rev .
d * 63 * , 042002 ( 2001 ) ; n. arnaud , _ et al .
_ , gr - qc/0210098 ( 2003 ) .
w. g. anderson , p. r. brady , j. d. e. creighton , .
flanagan , int .
d * 9 * , 303 ( 2000 ) ; w. g. anderson , p. r. brady , j. d. e. creighton , .
flanagan , phys .
d * 63 * , 042003 ( 2001 ) . a stochastic process @xmath221 is _ wide - sense stationary _ if its mean value is time - independent , @xmath222 = \eta$ ] , and if its autocorrelation satisfies @xmath223 = e[x(t+t_1-t_2)x(t ) ] = r(t_1-t_2)$ ] .
the karhunen - love expansion can always be used to expand an arbitrary stochastic process , stationary or not , into a series with orthogonal coefficients .
a. papoulis , _ probability , random variables , and stochastic processes _ ( mcgraw - hill ed .
, usa , 1991 ) , 3rd edition , p. 413 . | searches for gravitational wave signals which do not have a precise model describing the shape of their waveforms are often performed using power detectors based on a quadratic form of the data .
a new , optimal method of generalizing these power detectors so that they operate coherently over a network of interferometers is presented .
such a mode of operation is useful in obtaining better detection efficiencies , and better estimates of the position of the source of the gravitational wave signal .
numerical simulations based on a realistic , computationally efficient hierarchical implementation of the method are used to characterize its efficiency , for detection and for position estimation .
the method is shown to be more efficient at detecting signals than an incoherent approach based on coincidences between lists of events .
it is also shown to be capable of locating the position of the source . |
determining how well a particular graph is connected is a problem that arises often in various applications .
for example , as defined in @xcite , _ resilience _ is the ability of a network to provide and maintain an acceptable level of service in the face of various faults and challenges to normal operation .
examples of such networks include telecommunication networks , electrical grids , and commodity supply networks .
one of the scientific disciplines that forms the basis of resilience is _ disruption tolerance _ the ability of a system to tolerate disruptions in connectivity among its components ; such disruptions could consist of environmental challenges such as weak and episodic channel connectivity , mobility , and unpredictably - long delay , as well as tolerance of energy challenges @xcite . the most common way to interpret such networks is as connected graphs , where the resilience of a network as a function of its disruption tolerance can be studied by analyzing the connectivity properties of the associated graph .
graph connectivity has also been shown to play an important role in data security ; see @xcite and the references therein for more information . in 1973 , fiedler related the vertex - connectivity of a graph to the second - smallest eigenvalue of its laplacian matrix in the following way ( see section [ preliminaries ] for relevant definitions ) : [ thm1.1 ] @xcite if @xmath0 is a simple , non - complete graph , then @xmath1 .
note that theorem [ thm1.1 ] also implies @xmath2 , since @xmath3 . this seminal result provided researchers with another parameter that quantitatively measures the connectivity of a graph
; hence , @xmath4 is known as the _ algebraic connectivity _ of @xmath0 .
fiedler s discovery ignited interest in studying the connectivity of graphs by analyzing the spectral properties of their associated matrices .
akin to other connectivity measures such as vertex - connectivity , edge - connectivity , and isoperimetric number , the algebraic connectivity of a graph has applications in the design of reliable communication networks @xcite and in analyzing the robustness of complex networks @xcite .
recall that for a @xmath5-regular multigraph @xmath0 on @xmath6 vertices , @xmath7 for @xmath8 .
thus , for regular multigraphs , spectral bounds related to connectivity are often expressed in terms of the second - largest eigenvalue , instead of the second - smallest laplacian eigenvalue .
we now discuss several results relating the vertex- and edge - connectivity of graphs to their second - largest eigenvalues , which are similar in nature to the main results of the present paper .
the following result of chandran @xcite relates the second - largest eigenvalue of an @xmath6-vertex @xmath5-regular graph to its edge - connectivity .
@xcite let @xmath0 be an @xmath6-vertex @xmath5-regular simple graph with @xmath9 . then @xmath10 . in 2006 ,
krivelevich and sudakov improved the above result as follows .
let @xmath0 be a @xmath5-regular simple graph with @xmath11
. then @xmath10 .
the next result was shown in 2010 by ciob @xcite .
[ cioba ] @xcite let @xmath12 be a nonnegative integer less than @xmath5 , and let @xmath0 be a @xmath5-regular , simple graph with @xmath13
. then @xmath14 . in the same paper ,
ciob also gave improvements of theorem [ cioba ] for the following two particular cases .
[ cioba improvement for t=1 ] @xcite let @xmath15 be an odd integer and let @xmath16 denote the largest root of @xmath17 . if @xmath0 is a @xmath5-regular , simple graph such that @xmath18 , then @xmath19 .
the value of @xmath16 above is approximately @xmath20 . for @xmath21 , it is still an open problem to find the best upper bound for @xmath22 in a @xmath5-regular simple graph @xmath0 to guarantee that @xmath23 .
[ cioba improvement for t=2 ] @xcite let @xmath15 be any integer .
let @xmath0 be a @xmath5-regular , simple graph with @xmath24 then @xmath25 .
the value of @xmath26 above is approximately @xmath27 . in 2016 , o @xcite generalized fiedler s result to multigraphs , and established similar bounds as those above .
[ o , bound 1 for multigraphs ] @xcite let @xmath0 be a connected , @xmath5-regular multigraph with @xmath28 then @xmath19 .
[ o , bound 2 for multigraphs ] @xcite let @xmath29 and let @xmath0 be a connected , @xmath5-regular multigraph . if @xmath30 , then @xmath14 .
furthermore , if @xmath12 is odd and @xmath31 , then @xmath14 .
moreover , for every positive integer @xmath12 less than @xmath5 , o @xcite found the best upper bound for @xmath22 to guarantee that @xmath32 .
note that the bounds for @xmath22 in theorems [ cioba improvement for t=1 ] and [ cioba improvement for t=2 ] are larger than those in theorems [ o , bound 1 for multigraphs ] and [ o , bound 2 for multigraphs ] ; this suggests that if multiple edges are not allowed , then a bound for @xmath22 that will guarantee a desired edge - connectivity may be larger .
hence , we see that a @xmath5-regular multigraph @xmath0 may have high edge - connectivity while @xmath22 is small ; this supports the implicit meaning of cheeger s inequality ( see @xcite ) for a @xmath5-regular multigraph @xmath0 , which asserts that @xmath33 where @xmath34}{|s|}.\ ] ] the cheeger constant @xmath35 ( also known as the isoperimetric number ) can be considered as an approximation to the edge - connectivity of @xmath0 .
the results above make assertions about the edge - connectivity of a graph based on its eigenvalues . in a more recent paper , o @xcite also established analogous results for vertex - connectivity .
@xcite let @xmath0 be a @xmath5-regular multigraph that is not the @xmath36-vertex @xmath5-regular multigraph .
if @xmath37 , then @xmath38 .
algebraic connectivity has also been studied in the context of hypergraphs @xcite and directed graphs @xcite .
see @xcite and the bibliographies therein for other recent results on algebraic connectivity , more general structural results on vertex- and edge - connectivity , and other algebraic measures of connectivity .
the aim of the present paper is to investigate what upper bounds on the second - largest eigenvalues of regular simple graphs and multigraphs guarantee a desired vertex- or edge - connectivity .
in other words , we address the following question : for a @xmath5-regular simple graph or multigraph @xmath0 and for @xmath39 , what is the best upper bound for @xmath22 which guarantees that @xmath14 or that @xmath40 ?
the majority of the related results listed above were derived using a variety of combinatorial , linear algebraic , and analytic techniques ; moreover , they feature upper bounds for @xmath22 which do not depend on the order of the graph . in contrast , the results derived in the present paper feature bounds for @xmath22 which depend on both the degree and the order of the graphs , and as such are tight for infinite families of graphs .
furthermore , the derivations of these results combine analytic techniques with computer - aided _ symbolic _ algebra ; this proves to be a powerful approach , easily establishing the desired results in all but finitely - many cases .
the remaining cases are verified through a brute - force approach which relies on enumerating all multigraphs with certain properties . in order to avoid enumeration and post - hoc elimination of the exponential number of multigraphs without the desired properties , our approach required the development of novel combinatorial and graph theoretic techniques . while the problem of generating all non - isomorphic simple graphs having a certain degree sequence and other properties is well - studied ( cf .
@xcite ) , there are not as many efficiently - implemented algorithms for constrained enumeration of multigraphs ( see @xcite for some results in this direction ) .
thus , the developed enumeration procedure may also be of independent interest .
the paper is organized as follows . in the next section ,
we recall some graph theoretic and linear algebraic notions , specifically those related to eigenvalue interlacing . in sections 3 and 4 , respectively , we give spectral bounds which guarantee a certain vertex- and edge - connectivity .
we conclude with some final remarks and open questions in section 5 .
the appendix includes further details and computer code for symbolic computations used in some of the proofs .
in this paper , a _ multigraph _ refers to a graph with multiple edges but no loops ; a _ simple graph _ refers to a graph with no multiple edges or loops .
the _ order _ and _ size _ of a multigraph @xmath0 are denoted by @xmath41 and @xmath42 , respectively .
a _ double edge _
( respectively _ triple edge _ ) in a multigraph is an edge of multiplicity two ( respectively three ) .
the _ degree _ of a vertex @xmath43 of @xmath0 , denoted @xmath44 , is the number of edges incident to @xmath43 . the _ degree sequence _ of @xmath0 is a list @xmath45 of the vertex degrees of @xmath0 .
we may abbreviate the degree sequence of @xmath0 by only writing distinct degrees , with the number of vertices realizing each degree in superscript .
for example , if @xmath0 is the star graph on @xmath6 vertices , the degree sequence of @xmath0 may be written as @xmath46 . a _ vertex cut _
( respectively _ edge cut _ ) of @xmath0 is a set of vertices ( respectively edges ) which , when removed , increases the number of connected components in @xmath0 .
a multigraph @xmath0 with more than @xmath47 vertices is said to be _
@xmath47-vertex - connected _ if there is no vertex cut of size @xmath48 .
the _ vertex - connectivity _ of @xmath0 , denoted @xmath49 , is the maximum @xmath47 such that @xmath0 is @xmath47-vertex - connected .
similarly , @xmath0 is _ @xmath47-edge - connected _ if there is no edge cut of size @xmath48 ; the _ edge - connectivity _ of @xmath0 , denoted @xmath50 , is the maximum @xmath47 such that @xmath0 is @xmath47-edge - connected .
cut - vertex _
( respectively _ cut - edge _ ) is a vertex cut ( respectively edge cut ) of size one .
given sets @xmath51 , @xmath52 $ ] denotes the number of edges with one endpoint in @xmath53 and the other in @xmath54 .
the _ induced subgraph _
$ ] is the subgraph of @xmath0 whose vertex set is @xmath53 and whose edge set consists of all edges of @xmath0 which have both endpoints in @xmath53 .
a _ matching _ is a set of edges of @xmath0 which have no common endpoints ; a @xmath47-matching is a matching containing @xmath47 edges .
@xmath56 denotes the graph @xmath57 , and @xmath58 denotes the graph @xmath59 . the _ complete graph _ on @xmath6 vertices
is denoted @xmath60 .
an _ odd path _
( respectively _ even path _ ) in a graph is a connected component which is a path with an odd ( respectively even ) number of vertices . for other graph theoretic terminology and definitions ,
we refer the reader to @xcite .
the _ adjacency matrix _ of @xmath0 will be denoted by @xmath61 ; recall that in a multigraph , the entry @xmath62 is the number of edges between vertices @xmath63 and @xmath64 .
the _ eigenvalues _ of @xmath0 are the eigenvalues of its adjacency matrix , and are denoted by @xmath65 .
the _ laplacian matrix _ of @xmath0 is equal to @xmath66 , where @xmath67 is the diagonal matrix whose entry @xmath68 is the degree of vertex @xmath63 . the
_ laplacian eigenvalues _ of @xmath0 are the eigenvalues of its laplacian matrix and are denoted by @xmath69 .
the dependence of these parameters on @xmath0 may be omitted when it is clear from the context .
a technical tool used in this paper is _ eigenvalue interlacing _ ( for more details see section 2.5 of @xcite ) .
given two sequences of real numbers @xmath70 and @xmath71 with @xmath72 , we say that the second sequence _
interlaces _ the first sequence whenever @xmath73 for @xmath74 .
if @xmath75 is a real symmetric @xmath76 matrix and @xmath77 is a principal submatrix of @xmath75 of order @xmath78 with @xmath72 , then for @xmath79 , @xmath80 , i.e. , the eigenvalues of @xmath77 interlace the eigenvalues of @xmath75 .
let @xmath81 be a partition of the vertex set of a multigraph @xmath0 into @xmath82 non - empty subsets .
the _ quotient matrix _ @xmath83 corresponding to @xmath84 is the @xmath85 matrix whose entry @xmath86 ( @xmath87 ) is the average number of incident edges in @xmath88 of the vertices in @xmath89 .
more precisely , @xmath90}{|v_i|}$ ] if @xmath91 , and @xmath92)|}{|v_i|}$ ] .
note that for a simple graph , @xmath86 is just the average number of neighbors between vertices in @xmath88 and vertices in @xmath89 .
[ cor ] the eigenvalues of any quotient matrix @xmath83 interlace the eigenvalues of @xmath0 .
in this section , we establish an upper bound for the second - largest eigenvalue of an @xmath6-vertex @xmath5-regular simple graph or multigraph which guarantees a certain vertex - connectivity . to our knowledge , this is the first spectral bound on the vertex - connectivity of a regular graph which depends on both the degree and the order of the graph .
[ theorem1 ] let @xmath0 be an @xmath6-vertex @xmath5-regular simple graph or multigraph , which is not obtained by duplicating edges in a complete graph on at most @xmath94 vertices ; let @xmath95 where @xmath96 . if @xmath97 , then @xmath98 .
assume to the contrary that @xmath99 . if @xmath0 is disconnected , then @xmath100 , a contradiction .
now , assume that @xmath101 .
hence , there exists a vertex cut @xmath102 of @xmath0 with @xmath103 .
let @xmath104 be a union of some components of @xmath105 such that @xmath106=[c,\overline{s } ] \le \frac{cd}{2 } \le \frac{td}{2}$ ] , where @xmath107 and @xmath108 .
see figure [ fig_ptd ] for an illustration of this partition . into @xmath109 and @xmath110 . ] let @xmath111=p$ ] , and @xmath112 ; then , we have @xmath113=d(s_1+c)-p$ ] , and @xmath114=d(n - s_1-c)-p$ ] , so the quotient matrix for the partition @xmath115 is @xmath116 and the characteristic polynomial of @xmath83 with respect to @xmath117 is @xmath118 .
then by corollary [ cor ] , we have @xmath119 we now consider two cases based on whether @xmath0 is a simple graph or a multigraph . _ _
@xmath0 is a simple graph .
if @xmath120 , since the degree of each vertex in @xmath104 is @xmath5 , it holds that @xmath121 .
if @xmath122 , @xmath123 $ ] is a complete subgraph of @xmath0 , so the vertex in @xmath102 has degree greater than @xmath5 because @xmath124 ; this is a contradiction .
thus @xmath125 , @xmath126 , and so @xmath127 , as desired .
if @xmath128 , by the same argument as above , we must have @xmath129 , @xmath130 , and so @xmath131 , as desired . _
@xmath0 is a multigraph . if @xmath120 , then @xmath132 , @xmath133 , and so @xmath134 , as desired .
if @xmath128 , then @xmath135 , @xmath136 , and so @xmath137 . consider the function @xmath138 ; then , @xmath139<0 $ ] .
thus , @xmath140 is decreasing with respect to @xmath141 for @xmath142 , whence it follows that @xmath143 , as desired .
we now improve the result of theorem [ theorem1 ] in the case when @xmath0 is a multigraph and @xmath120 . recall that in this case , theorem [ theorem1 ] states that if @xmath145 , then @xmath144 .
[ theo23 ] let @xmath0 be an @xmath6-vertex @xmath5-regular multigraph with @xmath146 and @xmath147 .
if @xmath148 , then @xmath144 . assume to the contrary that @xmath149 .
if @xmath150 , then @xmath151 , a contradiction .
thus , we can assume henceforth that @xmath152 .
let @xmath43 be a cut - vertex of @xmath0 , and @xmath104 and @xmath153 be two components of @xmath154 with @xmath112 and @xmath155 .
let @xmath156 $ ] and @xmath157 $ ] ; without loss of generality , we can assume that @xmath158 , and hence that @xmath159 ( otherwise the roles of @xmath104 and @xmath153 can be reversed ) ; note that since @xmath147 , we must have @xmath160 ; moreover , @xmath161
. see figure [ fig2 ] for an illustration of this partition in the case when @xmath162 . into @xmath104 , @xmath163 and @xmath153 ,
when @xmath164 .
] the quotient matrix for the partition @xmath165 is @xmath166 and its characteristic polynomial with respect to @xmath117 is @xmath167.\ ] ] then by corollary [ cor ] , we have @xmath168 , where @xmath169 is the second - largest root of the characteristic polynomial of @xmath83 ; it can be verified that @xmath169 can be expressed as follows : @xmath170.\ ] ] if we set the derivative of @xmath169 with respect to @xmath171 equal to zero and solve for @xmath171 , we obtain @xmath172 substituting @xmath173 for @xmath174 , and the right hand side of ( [ eq2 ] ) for @xmath171 in ( [ eq1 ] ) , and simplifying , we obtain @xmath175 finally , when we substitute @xmath176 for @xmath177 , the resulting expression has a minimum at @xmath162 , for @xmath146 , @xmath147 , and @xmath160 , with minimal value @xmath178 .
this minimization and some of the algebraic manipulations described above were carried out using symbolic computation in mathematica ; for details , see the appendix .
let @xmath0 be a multigraph with the following adjacency matrix : @xmath179 then @xmath180 .
moreover , @xmath0 is a @xmath5-regular multigraph with 5 vertices , @xmath181 , @xmath182 , and @xmath152 .
thus , the bound in theorem [ theo23 ] is the best possible for this infinite family of multigraphs .
@xmath184 in this section , we first give an upper bound for @xmath22 in an @xmath6-vertex @xmath5-regular multigraph which guarantees that @xmath185 ; its proof is omitted , since it is similar to that of theorem [ theorem1 ] .
theorem [ theorem4 ] extends a result of ciob @xcite to multigraphs .
[ theorem4 ] let @xmath0 be an @xmath6-vertex @xmath5-regular multigraph , which is not obtained by duplicating edges in a complete graph on at most @xmath94 vertices .
let @xmath186 where @xmath96 .
if @xmath187 , then @xmath185 . now , we will improve the bound in theorem [ theorem4 ] for the case of @xmath120 ; see observation [ obs_improvement ] for an explanation of why theorem [ theo31 ] is an improvement .
[ theo31 ] let @xmath0 be an @xmath6-vertex @xmath5-regular multigraph with @xmath188 , where @xmath189 is the second - largest eigenvalue of the following matrix : @xmath190 then @xmath191 .
assume to the contrary that @xmath192 .
if @xmath193 , then since the largest eigenvalue of @xmath83 equals @xmath5 , we have that @xmath194 , a contradiction .
now , assume that @xmath195 . for any graph @xmath196 , define @xmath197 to be the number of vertices in the smallest connected component of @xmath196 .
let @xmath198 be a cut - edge of @xmath0 such that @xmath199 .
in other words , @xmath200 is a cut - edge such that one of the components of @xmath201 has minimum size among all subgraphs of @xmath0 which can be separated by removing a cut - edge of @xmath0 .
let @xmath202 and @xmath203 be the two components of @xmath201 , where @xmath204 , @xmath205 , and @xmath206 . for @xmath207 , let @xmath208 and @xmath209 . by the degree - sum formula , @xmath210 , whence it follows that @xmath211 is odd .
thus , both @xmath5 and @xmath212 are odd , and hence @xmath6 is even ; moreover , @xmath213 , and hence @xmath214 .
see figure [ partition ] for an illustration .
-regular multigraph with @xmath195 . ]
we now consider three cases based on the cardinality of @xmath215 .
case 1 : @xmath216 .
: : in this case , the structure of the graph is determined uniquely , and the vertex partition @xmath217 corresponds to the quotient matrix @xmath83 defined in the statement of the theorem ; see figure [ case_n3 ] for an illustration .
therefore , the inequality @xmath218 holds for all @xmath5 and @xmath6 .
+ [ case_n3 ] + -regular multigraph with @xmath195 and @xmath162 . ]
case 2 : @xmath219 .
: : consider the partition @xmath217 and the corresponding quotient matrix : + @xmath220 + let @xmath221 . by corollary [ cor ] , @xmath222 .
note that @xmath5 is odd , and that due to the partition structure , @xmath223 .
thus , to show that @xmath224 holds for all @xmath5 and @xmath6 , we will show that @xmath224 holds when @xmath225 and @xmath226 , and that + @xmath227 + holds for all other values of @xmath5 and @xmath6 . to verify that @xmath224 holds when @xmath225 and @xmath226 , we compute the second - largest eigenvalues of all possible multigraphs which have these parameters , and compare them to @xmath228 and @xmath229 , respectively ; the enumeration procedure is described in the appendix .
for all other values of @xmath5 and @xmath6 , we verify ( [ ineq_rho_pp ] ) by separating it into the following cases and using symbolic computation in mathematica ; see the appendix for details . see also case 3 below for a more detailed explanation of why this computation is sufficient to establish the claim . + 1 .
@xmath225 , @xmath230 .
fix @xmath225 and @xmath231 .
then , @xmath232 and @xmath233 hold for all @xmath230 . 2 .
@xmath234 , @xmath226 .
fix @xmath234 and @xmath235 .
then , @xmath232 and @xmath233 hold for @xmath236 and @xmath237 .
@xmath238 , @xmath236 .
fix @xmath238 and @xmath239 . then , @xmath232 and @xmath233 hold for @xmath236 .
@xmath234 , @xmath230 ; @xmath238 , @xmath240 ; @xmath241 , @xmath223 .
fix @xmath242 .
then , @xmath232 and @xmath233 hold for all values of @xmath5 and @xmath6 described in this case .
case 3 : @xmath243 .
: : in this case , we consider the vertex partition of @xmath0 with the sets @xmath244 and @xmath245 ; see figure [ case_n7 ] for an illustration . + into @xmath244 and @xmath245 . ]
+ the second - largest eigenvalue of the quotient matrix @xmath246 corresponding to this vertex partition is equal to @xmath247 . by corollary [ cor ] , @xmath248 , where the last inequality follows from the fact that @xmath249 .
note that @xmath6 is even , @xmath5 is odd , @xmath147 , and due to the partition structure , @xmath230 .
thus , to show that @xmath224 holds for all @xmath5 and @xmath6 , we will show that @xmath224 holds when @xmath225 and @xmath250 , and that + @xmath251 + holds for all other values of @xmath5 and @xmath6 .
to verify that @xmath224 holds when @xmath225 and @xmath250 , we compute the second - largest eigenvalues of all possible multigraphs which have these parameters , and compare them to @xmath252 , @xmath253 , and @xmath253 , respectively ; the enumeration procedure is described in the appendix .
for all other values of @xmath5 and @xmath6 , we verify ( [ ineq_rho_7 ] ) as follows .
+ note that @xmath254 is a monic polynomial of degree 4 , with roots @xmath255 , @xmath169 , @xmath256 , and @xmath257 ; all roots are real , since they interlace the eigenvalues of @xmath0
. moreover , @xmath258 and @xmath259 , which implies that @xmath260 . by theorem [ theorem4 ] , @xmath261
; thus , @xmath262 for all @xmath147 and @xmath230 . since @xmath263
, it follows that @xmath264 . finally , note that @xmath265 thus , showing that ( [ ineq_rho_7 ] ) holds is equivalent to showing that + * @xmath266 for @xmath267 , and * @xmath233 for @xmath268 , + whence it follows that @xmath269 . using symbolic computation in mathematica
, we can verify that a ) holds for all @xmath147 and @xmath230 , and b ) holds when @xmath225 and @xmath270 , and when @xmath271 and @xmath230 ; for details , see the appendix . since the case @xmath225 , @xmath272 was verified by enumeration
, this completes the proof .
[ obs_improvement ] when @xmath120 , theorem [ theorem4 ] states that if @xmath273 , then @xmath191 .
case 3 of the proof of theorem [ theo31 ] guarantees that @xmath274 , which means that @xmath189 is a better bound than @xmath275 .
let @xmath0 be the @xmath5-regular multigraph on 6 vertices with @xmath147 and @xmath195 .
then @xmath276 , where @xmath189 is defined as in the statement of theorem [ theo31 ] .
thus , the bound in theorem [ theo31 ] is the best possible for this infinite family of multigraphs .
in this paper , we presented upper bounds for the second - largest eigenvalues of regular graphs and multigraphs , which guarantee a desired vertex- or edge - connectivity .
the given bounds improve on several previous results , and hold with equality for infinite families of graphs . in deriving these bounds , we used computer - aided symbolic algebra , which synergizes well with the technique of eigenvalue interlacing ; this combination gives a viable approach to investigating spectral bounds guaranteeing graph theoretic properties , which differs from the typical analytic strategies used in similar derivations .
as part of our proof of theorem [ theo31 ] , we developed an approach to enumerate all multigraphs with certain properties , by adding edges to certain simple graphs .
since the theory and available software for enumerating simple graphs are generally better - developed than their analogues for multigraphs , it may be worth investigating this strategy further and adapting it to other applications .
another problem of interest is to obtain bounds on the second - largest eigenvalues of a graph which guarantee a desired connectivity , and depend on other graph invariants such as girth or circuit rank . in particular
, one could explore the utility of symbolic computer algebra used together with eigenvalue interlation in deriving such bounds .
@xmath184 the authors gratefully acknowledge financial support for this research from the following grants and organizations : nsf - dms grants 1604458 , 1604773 , 1604697 and 1603823 ( all authors ) , the combinatorics foundation ( a. abiad , s. o ) , nsf 1450681 ( b. brimkov ) , institute of mathematics and its applications ( x. martnez - rivera ) , krf - msip grant 2016r1a5a1008055 ( s. o ) .
xx n.m.m .
de abreu , old and new results on algebraic connectivity of graphs . _ linear algebra and its applications _ , 423 : 5373 ( 2007 ) .
brouwer and w.h .
haemers , spectra of graphs , springer , new york , 2011 .
chandran , minimum cuts , girth and spectral threshold . _ information processing letters _ , 89(3 ) : 105110 ( 2004 ) .
j. chen and i. safro , algebraic distance on graphs .
_ siam journal on scientific computing _ , 33(6 ) : 34683490 ( 2011 ) .
cioaba , eigenvalues and edge - connectivity of regular graphs . _ linear algebra and its applications _ , 432(1 ) : 458470 ( 2010 ) .
m. fiedler , algebraic connectivity of graphs .
_ czechoslovak mathematical journal _ , 23(98 ) : 298305 ( 1973 ) .
a. frank , augmenting graphs to meet edge - connectivity requirements .
_ siam journal on discrete mathematics _ , 5(1 ) : 2553 ( 1992 ) .
hakimi , on the realizability of a set of integers as degrees of the vertices of a simple graph .
_ siam journal on applied mathematics _
, 10 : 496506 ( 1962 ) .
p. hanlon , enumeration of graphs by degree sequence .
_ journal of graph theory _ , 3(3 ) : 295299 ( 1979 ) .
s. hoory , n. linial , and a. wigderson , expander graphs and their applications .
_ bulletin of the american mathematical society _
, 43 : 439561 ( 2006 ) .
s. hu and l. qi , algebraic connectivity of an even uniform hypergraph .
_ journal of combinatorial optimization _ , 24(4 ) : 564579 ( 2012 ) .
a. jamakovic and p. van mieghem , on the robustness of complex networks by using the algebraic connectivity .
_ international conference on research in networking _ , springer berlin heidelberg , 2008 .
a. jamakovic and s. uhlig , on the relationship between the algebraic connectivity and graph s robustness to node and link failures .
_ 3rd eurongi conference on next generation internet networks _ , ieee , 2007 .
kao , data security equals graph connectivity . _
siam journal on discrete mathematics _ , 9(1 ) : 87100 ( 1996 ) .
s. kirkland , j.j .
molitierno , m. neumann , and b.l .
shader , on graphs with equal algebraic and vertex connectivity .
_ linear algebra and its applications _ , 341 : 4556 ( 2002 ) .
m. krivelevich and b. sudakov , pseudo - random graphs . _ more sets , graphs and numbers _ , 15 : 199262 ( 2006 ) .
w. liu , h. sirisena , k. pawlikowski , and a. mcinnes , utility of algebraic connectivity metric in topology design of survivable networks .
_ 7th international workshop on design of reliable communication networks _ , ieee pp .
131138 , 2009 .
s. o , algebraic connectivity of multigraphs .
_ arxiv_:1603.03960 ( 2016 ) .
s. o , the vertex - connectivity and second largest eigenvalue in regular multigraphs .
_ linear algebra and its applications _ , 491 : 414 ( 2016 ) .
rad , m. jalili , and m. hasler , a lower bound for algebraic connectivity based on the connection - graph - stability method .
_ linear algebra and its applications _ , 435(1 ) , 186192 ( 2011 ) .
f. ruskey , r. cohen , p. eades , and a. scott , alley cats in search of good homes .
_ congressus numerantium _
, 97110 ( 1994 ) .
sterbenz , d. hutchison , e.k .
etinkaya , a. jabbar , j.p .
rohrer , m. schller , and p. smith , resilience and survivability in communication networks : strategies , principles , and survey of disciplines .
_ computer networks _ , 54 : 12451265 ( 2010 ) .
taqqu , and j.b .
goldberg , regular multigraphs and their application to the monte carlo evaluation of moments of non - linear functions of gaussian random variables .
_ stochastic processes and their applications _ , 13(2
) : 121138 ( 1982 ) .
west , _ introduction to graph theory _ , prentice hall , inc .
, upper saddle river , nj , 2001 .
wu , algebraic connectivity of directed graphs . _ linear and multilinear algebra _ , 53(3 ) : 203223 ( 2005 ) .
below we provide the mathematica code used to calculate the minimum of the second root of the characteristic polynomial in the proof of theorem [ theo23 ] , and to check some cases in the proof of theorem [ theo31 ] .
the version of mathematica used is 10.0.0.0 for 64-bit microsoft windows .
we also include additional details about the procedure of enumerating certain multigraphs in the proof of theorem [ theo31 ] .
case 2d : note that both inequalities hold for @xmath234 and @xmath230 , @xmath238 and @xmath240 , and @xmath241 and @xmath223 .
+ @xmath292;x = d-1/5 - 1/(n-5);}\\ \pmb{\text{reduce}[\text{polyq}<0\&\&n\geq 10\&\&d\geq 3,\text{integers}]}\\ \pmb{\text{reduce}[\text{polyqprim}>0\&\&n\geq 10\&\&d\geq 3,\text{integers}]}$ ] case 3 : note that both inequalities hold for @xmath225 and @xmath270 , and @xmath271 and @xmath230 .
the case @xmath225 , @xmath272 is verified by enumeration in the next section .
+ @xmath295}\\ \pmb{\text{clear}[x];x = d-1/3 - 1/(n-3);}\\ \pmb{\text{reduce}[\text{polyq}>0\&\&n\geq 14\&\&d\geq 3,\text{integers}]}$ ] let @xmath298 and @xmath299 respectively be the sets of 3-regular multigraphs of order @xmath300 and @xmath301 with edge - connectivity 1 , such that the removal of any cut - edge of these graphs produces components of order at least 5 .
let @xmath302 , @xmath303 , and @xmath304 respectively be the sets of 3-regular multigraphs of order @xmath305 , @xmath306 , and @xmath307 with edge - connectivity 1 , such that the removal of any cut - edge of these graphs produces components of order at least 7 .
these constraints imply that a graph in @xmath298 or @xmath302 must have exactly one cut - edge , a graph in @xmath299 or @xmath303 can have one or two cut - edges , and a graph in @xmath304 can have one , two , or three cut - edges . for @xmath308 ,
let @xmath309 be the set of all connected multigraphs which have degree sequence @xmath310 and have no cut - edges .
for any graph @xmath311 , @xmath308 , define @xmath312 to be the degree 2 vertex of @xmath196 .
let @xmath313 be the graph consisting of two vertices joined by a double edge , let @xmath314 be the graph obtained by joining two copies of @xmath313 by one edge , and let @xmath315 be a complete graph on four vertices with one edge removed . for @xmath316 , define @xmath317 to be one of the degree 2 vertices of @xmath318 , and @xmath319 to be the other degree 2 vertex of @xmath318 .
for any @xmath320 , define @xmath321 to be the set @xmath322 ( where @xmath323 denotes disjoint union ) . for
any @xmath320 and @xmath316 , define @xmath324 to be the set @xmath325 .
in other words , `` @xmath326 '' denotes the set obtained by joining all possible pairs of graphs from the indicated families by a cut - edge incident to their degree 2 vertices . with this in mind , it is easy to see that @xmath327 thus , to find the graphs in @xmath328 , @xmath329 , it suffices to find the graphs in @xmath330 , @xmath331 .
since the graphs in @xmath330 are 3-regular and connected , they can not have triple edges ; moreover , they can have at most @xmath332 double edges .
let @xmath333 be the set of multigraphs in @xmath330 which have @xmath334 double edges . then , @xmath335 .
we will now describe a procedure for enumerating the graphs in @xmath333 .
if the double edges of the graphs in @xmath333 are replaced by single edges , the resulting graphs will be simple , 2-vertex - connected , and have degree sequence @xmath336 .
there are well - known algorithms for generating all nonisomorphic simple graphs with a given degree sequence ( cf .
@xcite ) ; a practical algorithm is implemented in the software system sagemath .
let @xmath337 be the set of nonisomorphic simple graphs with degree sequence @xmath336 .
then , by adding double edges in all feasible ways to the simple graphs in @xmath337 , we can recover the multigraphs in @xmath333 .
specifically , a double edge can be added to a graph in @xmath337 only where a single edge with two degree 2 endpoints already exists .
moreover , not every graph in @xmath337 can have @xmath334 double edges added to it in a way that the resulting multigraph is in @xmath333 ; similarly , it may be possible to add @xmath334 double edges to a graph in @xmath337 in multiple ways so that the resulting multigraphs are in @xmath333 .
let @xmath196 be a graph in @xmath337 and let @xmath338 be the subgraph induced by the degree 2 vertices of @xmath196 . since the maximum degree of @xmath338 is 2 , @xmath338 is the disjoint union of some paths and cycles .
however , if @xmath338 contains a cycle with less than @xmath339 vertices , a multigraph in @xmath333 can not be obtained by doubling single edges of @xmath196 with two degree 2 endpoints ( since any resulting multigraph with degree sequence @xmath340 will be disconnected ) .
similarly , if @xmath338 contains more than one odd path , a multigraph in @xmath333 can not be obtained by doubling single edges of @xmath196 with two degree 2 endpoints ( since any resulting multigraph with degree sequence @xmath340 will not have @xmath334 multiple edges ) .
thus , let @xmath341 is either a cycle @xmath342 , or contains exactly one odd path@xmath343 . for any graph @xmath196 in @xmath344 , the different maximum matchings ( i.e. @xmath334-matchings ) of @xmath338 correspond to different ways to add double edges to @xmath196 .
let @xmath345 be the set of multigraphs obtained by adding double edges to @xmath196 corresponding to the different @xmath334-matchings of @xmath338 .
then , @xmath346 , @xmath347 , and @xmath328 can be obtained by joining pairs of graphs in @xmath330 as described earlier .
note that the set of distinct maximum matchings of a graph whose components are paths ( one of which is odd ) can be found in linear time .
see figure [ enumeration2 ] for an illustration of this enumeration for @xmath348 ; the other sets of multigraphs @xmath333 are handled analogously , and combined to obtain the graphs in @xmath328 .
finally , for each multigraph in @xmath328 , we can easily compute and compare the second - largest eigenvalue to @xmath349 ; we have found that all of these eigenvalues are greater than or equal to @xmath349 , as desired . .
_ top row _ :
the graphs in @xmath350 ; the three graphs on the right are not 2-vertex - connected , so they are not considered further . _
second row _ : @xmath338 for the remaining graphs @xmath196 ; the graph on the left has multiple odd paths , so it is not considered further .
_ third row _ :
all possible 2-matchings of the remaining graphs in the second row . _ bottom row _ : adding double edges specified by the matchings to obtain the graphs in @xmath348 ; the two matchings of the graph on the right happen to result in isomorphic multigraphs . ] | the second - largest eigenvalue and second - smallest laplacian eigenvalue of a graph are measures of the graph s connectivity .
these parameters can be used to analyze the robustness , resilience , and synchronizability of networks , and are related to other connectivity attributes such as the vertex- and edge - connectivity , isoperimetric number , and characteristic path length . in this paper
, we give upper bounds for the second - largest eigenvalues of regular graphs and multigraphs which guarantee a desired vertex- or edge - connectivity .
the given bounds are in terms of the order and degree of the graphs , and hold with equality for infinite families of graphs .
second - largest eigenvalue ; vertex - connectivity ; edge - connectivity ; regular multigraph ; algebraic connectivity .
+ 05c50 , 05c40 . |
cataclysmic variables ( cvs ) are interacting binary stars consisting of white dwarfs accreting from low - mass , roche lobe filling companions ( see @xcite for a review ) .
angular momentum is continuously lost from the binary orbit , driving mass transfer , as well as evolution in orbital period ( @xmath10 ) .
there are still many uncertainties in the theoretical description of cv formation and evolution ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) , as well as several serious discrepancies between the predictions of theory and the properties of the observed cv population ( e.g. @xcite ; @xcite ; @xcite ; pretorius & knigge 2008a , b ; @xcite ; @xcite ) . in order to constrain evolution models , more and better observational constraints on the properties of the galactic cv population are needed .
a fundamental parameter predicted by cv evolution theory , that is expected to be more easily measured than most properties of the intrinsic cv population , is the space density , @xmath0 . the luminosity function ( @xmath1 )
is a closely related property ; although more challenging to constrain than @xmath0 , @xmath1 contains information on the mass transfer rate ( @xmath11 ) distribution of cvs and is therefore potentially much more valuable than the space density alone .
some theoretically predicted values of the cv space density are @xmath12 @xcite , @xmath13 @xcite , @xmath14 @xcite .
observational estimates are typically lower , but have a large range ; values from @xmath15 to @xmath16 have been reported ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) .
fewer estimates of the cv luminosity function are available , but x - ray ( @xcite ; @xcite ) and hard x - ray @xcite luminosity functions have been published ( see also @xcite for near - ir and optical luminosity functions ) .
the work of @xcite is based on the sample of non - magnetic cvs with parallax distances less than 200 pc ; although this sample is certainly not complete , it gives a firm lower limit on the luminosity function .
the cv samples used by @xcite and @xcite consist mostly of magnetic cvs .
uncertainty in @xmath0 ( and @xmath1 ) measurements is in part caused by statistical errors , arising from uncertain distances and small number statistics .
however , the dominant source of uncertainty , as well as the cause of inconsistencies between some estimates , is most likely systematic errors caused by selection effects .
the selection effects acting on cv samples are most easily accounted for in samples with simple , well - defined selection criteria .
whereas optical cv samples always include selection criteria based on , e.g. , colour , variability , or emission lines , there are x - ray selected cv samples that are purely flux - limited .
the completeness of these samples is easier to model than that of any existing optically selected sample .
furthermore , there is a well - known empirical correlation between the ratio of optical to x - ray flux ( @xmath17 ) and optical luminosity , implying that an x - ray flux limit does not introduce as strong a bias against intrinsically faint , short - period cvs as an optical flux limit ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) .
the _ rosat _ bright survey ( rbs ) consists of all bright ( count rate @xmath18 ) , high galactic latitude ( @xmath19 ) sources in the _ rosat _ all - sky survey ( rass ; see @xcite and @xcite ) .
optical counterparts have been identified for all sources in the rbs @xcite , and it includes 46 cvs ( of which 11 were previously unknown ; see @xcite ) .
@xcite already used this survey to estimate @xmath20 for non - magnetic cvs ; they note that this measurement is dominated by two systems with very small distance estimates , and would be an over - estimate if these two distances where significantly under - estimated ( as has turned out to be the case ; @xcite ) . omitting those two sources , @xcite
obtain @xmath21 . in a previous paper @xcite ,
we have considered non - magnetic cvs from the _ rosat _ north ecliptic pole ( nep ) survey ; this survey is 2 orders of magnitude deeper than the rbs , but covers a much smaller area ( @xmath2281 sq.deg .
, down to roughly @xmath23 ; see e.g. @xcite ; @xcite ) , and also has complete optical follow - up .
the rbs and nep survey thus complement each other in terms of depth and angle .
here we combine these 2 surveys to provide new observational constraints on the space density , as well as the x - ray luminosity function of non - magnetic cvs .
the combined survey is still purely x - ray flux limited ( although the flux limit is variable over the survey area ) , and yields a sample of 20 non - magnetic cvs . in a future paper
we will consider magnetic cvs detected in the rbs .
we describe the non - magnetic cv sample in section [ sec : sample ] , where we also present distance and @xmath24 estimates for these 20 systems . in section [ sec : rho ] we describe the calculation of @xmath0 and @xmath1 , and in sections [ sec : results ] and [ sec : fluxl ] we present the results .
finally , the results are discussed in section [ sec : disc ] and the conclusions listed in section [ sec : concl ] .
the rbs covers half the sky down to a flux limit of @xmath25 ( assuming a 10 kev thermal bremsstrahlung spectrum ) , and includes 16 non - magnetic cvs . with the same assumed spectrum ,
the nep flux limit varies from roughly @xmath26 to @xmath27 over the 81 sq.deg . survey area .
only 4 cvs where detected in the nep , all of them non - magnetic ( rx j1715 , sdss j1730 , rx j1831 , and ex dra ) . the 2 surveys have no cvs in common , although they overlap slightly none of the rbs systems are in the area covered by the nep survey , and the nep cvs are all fainter than the rbs flux limit . the complete sample of 20 non - magnetic cvs is presented in table [ tab : distances ] .
the rbs sample differs from that used by @xcite in that we include tw pic and exclude ei uma .
tw pic was at first thought to be an intermediate polar ( @xcite ; @xcite ) , but more recently @xcite has argued convincingly that the data favour a non - magnetic nature .
ei uma , on the other hand , has now been shown to be magnetic ( @xcite ; @xcite ; @xcite ) .
we assume that none of the systems in our sample are period bouncers ( cvs that have evolved through the observed minimum @xmath10 near 80 min , and that are now evolving towards longer @xmath10 ; e.g. @xcite ) . in the case of the long - period cvs , tt ari , ef tuc , rx j1831.7 + 6511 , ww cet , v405 peg , and ex dra , this needs no justification .
eight of the short - period systems ( sw uma , t leo , bz uma , vw hyi , wx hyi , su uma , sdss j173008.38 + 624754.7 , and ty psa ) have mass ratios implying that they are not period bouncers ( as indicated by fractional superhump period excess ; see @xcite ; @xcite ) . for cc
scl , rx j1715.6 + 6856 , iq eri , rbs490 , tw pic , and rbs1411 there is no information on mass ratio ( in fact , the orbital periods of several of these systems are not known ) , and the simplest assumption is that they are normal , pre - period bounce cvs . in order to measure @xmath0 and @xmath1 , we need distance estimates for all cvs in the sample . although good distance measurements are available for some of these systems , in many cases the distance estimates that we can derive are quite imprecise .
t leo , bz uma , rbs490 , sw uma , and v405 peg have high - quality parallax distance measurements ( @xcite ; @xcite ; @xcite ; @xcite ) .
other reliable estimates of cv distances are found by photometric parallax , in cases where the white dwarf or the donor star is detected in a way that makes it possible to disentangle its flux contribution from that of other light sources .
we use such estimates for tt ari and ex dra .
tt ari has a distance measurement based on the donor star spectrum ( @xcite ; this is also consistent with the distance derived from the white dwarf spectrum ) .
based on the photometric parallax of the secondary , the distance to ex dra is @xmath28 ( see @xcite ; @xcite ; @xcite ; we have revised this to be slightly less conservative than the estimate we used previously ) .
there are a few well - known , less direct ( and less reliable ) methods of estimating distances to cvs .
we use methods based on dwarf nova ( dn ) outburst maximum ( @xcite ; see also @xcite and @xcite ) , the near - ir apparent brightness for systems in which the donor star is not directly detected ( the method of @xcite , but as prescribed by @xcite and @xcite ) , and the strength of h@xmath29 emission lines ( @xcite ; see also @xcite ) .
the relation between the absolute magnitude at dn outburst maximum and @xmath10 was most recently studied by @xcite , who confirms that the scatter is relatively small , and that there are no large outliers .
we therefore use this relation as far as possible ( for cc scl , vw hyi , wx hyi , su uma , ty psa , ww cet , sdss j1730 , and ef tuc ) .
it should be noted that su uma has a parallax distance estimate that agrees very well with the distance based on outburst maximum , but which is less precise ; we choose in this case to use the estimate based on outburst , rather than the parallax ( see @xcite and @xcite ) .
two systems , sdss j1730 and ef tuc , do not have orbital inclination measurements ( although it is known that they are not eclipsing ) , and have less well determined maximum apparent magnitudes , leading to more imprecise distance estimates from this method . for rx j1831 we use the prescription of @xcite ( as updated by @xcite ) .
this is based on a semi - empirical donor sequence for cvs , and the offsets between this sequence and the absolute @xmath30 magnitudes of a sample of cvs with parallax distances .
apparent near - ir magnitudes for rx j1831 were obtained from the two micron all sky survey ( 2mass ; @xcite ) .
finally , although there clearly exists an empirical relation between @xmath31 and the absolute magnitude of the disc , this relation contains large scatter ( see @xcite for an updated plot ) .
we therefore use it as a last resort , in those four cases where the data required by the other two methods are not available ( tw pic , iq eri , rx j1715 , and rbs1411 ) . for rx j1715 and rbs1411 , we have no data to allow us to check whether the absolute magnitudes we find are reasonable ( other than that rx j1715 is known to be a short - period cv , and that rbs1411 has an optical spectrum resembling that of a short - period cv ) . for the remaining 2 systems , we find absolute magnitudes that are in reasonable agreement with what we would find from outburst ( in the case of iq eri ) , and the method of knigge ( in the case of tw pic ) , if we made reasonable assumptions about orbital period .
tw pic has spectroscopic and photometric periods near 2 and 6 hours ( the orbital period was initially assumed to correspond to the 6-h signal ; @xcite discuss the alternative interpretation of the more stable 2-h period being the orbital period ) . assuming the period near 2 hours is @xmath10 , the @xcite sequence would give a distance in agreement with what we find from @xmath31 .
however , if the orbital period is as long as 6 h , the observed @xmath32 would imply an improbably bright absolute @xmath33 magnitude ( @xmath34 ; although note that at such long @xmath10 , the system would probably have an evolved donor , for which the donor sequence is not appropriate ) . ] .
interstellar extinction is expected to be low for our systems , since they are at high galactic latitude , and since most of them are quite nearby . for those systems with distances below 200 pc , we neglect interstellar extinction . for the more distant objects , we use @xmath35 estimates from @xcite , where available , and for a few more
, we find estimates in @xcite . in the cases where no more direct estimate of extinction is available
, we use the model of @xcite , with a few iterations , so that the value we finally adopt in the distance calculation is that given by the model at the estimated distance to the object . to convert from visual extinction to extinction in the 2mass bands , we use @xmath36 , @xmath37 , and @xmath38 @xcite .
we conservatively assume errors of 50% in the extinction values .
we then find the probability distribution for the distance to each source , assuming gaussian errors in all the input parameters ( apparent magnitudes , inclination , @xmath31 , extinction ) .
the distances estimates , listed in column 3 of table [ tab : distances ] , are in all cases the median , together with the 1-@xmath39 confidence interval corresponding to the 16th and 84th percentile points .
[ cols="<,<,<,<,<,<,<,<",options="header " , ] most non - magnetic cvs have x - ray emission that can be described as thermal bremsstrahlung from the boundary layer ( the inner part of the disc where the flow is no longer keplerian , but is slowing down to match the rotation of the white dwarf ; @xcite , but see also e.g. @xcite ) .
the x - ray spectrum should then be the sum of emission from material with temperatures ranging from the temperature of the shock at the outer edge of the boundary layer to the temperature of the white dwarf photosphere ( e.g. @xcite ) .
x - ray observations of many systems can be fit by a single temperature thermal bremsstrahlung spectrum with @xmath40 between roughly 5 and 20 kev ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . in other cases , multi - temperature ( or cooling flow )
models are needed to provide satisfactory fits to observations ( e.g. @xcite ; @xcite ; @xcite ; @xcite @xcite ; @xcite ; @xcite ) .
high-@xmath11 systems ( nova - like cvs in a high state and dne in outburst ) are expected to have optically thick boundary layers , and therefore much softer spectra ; however , part of the boundary layer remains optically thin so that the soft component does not dominate the spectrum above @xmath41 kev ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) .
we assume a @xmath42 thermal bremsstrahlung spectrum for all cvs in our sample , and quote x - ray fluxes and luminosities in the 0.52.0 kev band .
although there is no good physical justification for this simple approach , it is acceptable for our purposes , because the energy band we are using is narrow ( decreasing the sensitivity of @xmath24 to the assumed spectrum ) , and because our distances are for the most part quite imprecise ( implying that error arising from the assumed spectral shape is unlikely to contribute significantly to the total error in @xmath24 ; see below ) .
we use the dust to gas ratio of @xcite to convert the @xmath35 estimates to @xmath43 for each system .
the @xmath43 estimates are low ( the highest being @xmath44 , for sdss j1730 ; @xcite ) .
we list unabsorbed @xmath45 and @xmath24 in table [ tab : distances ] . in order to provided further justification for our assumptions regarding the x - ray spectrum , we can consider the error in @xmath24 using standard error propagation : @xmath46 where @xmath47 is the error in distance ( in the best cases about 15% , but larger for most systems ) , @xmath48 is the observational error ( typically @xmath2210% ) , and @xmath49 is the error caused by uncertainty in @xmath43 , and by an incorrect assumption of x - ray spectrum . in order for @xmath49 to dominate @xmath50 , we must have @xmath51 .
this seems unlikely , because in the narrow energy band ( 0.52.0 kev ) , for moderate @xmath43 and allowing for a 50% error in @xmath43 , the difference in @xmath45 between single - temperature bremsstrahlung spectra with @xmath52 and @xmath53 is only @xmath54 .
high-@xmath11 cvs are naturally expected to have brighter @xmath24 ; therefore , @xmath24 should increase with @xmath10 .
this is certainly not apparent in the luminosities we find ( see the @xmath10 and @xmath24 values in table [ tab : distances ] ) .
@xcite show that , for low-@xmath11 systems , @xmath24 indeed increases with @xmath11 , but that at roughly @xmath55 the relation between @xmath24 and @xmath11 flattens off .
binary inclination may also be expected to add noise to this relation .
@xcite find only a weak correlation between @xmath24 and @xmath10 , while @xcite find none .
given the large scatter or weakness of any correlation between @xmath24 and @xmath10 , @xmath24 is not a good indicator of period ( or , by implication , age ) ; in the calculations described in section [ sec : simphi ] , @xmath24 is therefore the only physical parameter of the simulated cvs .
here we describe the calculation of the space density and x - ray luminosity function from the sample of 20 non - magnetic cvs detected in the combined rbs and nep survey .
the results of this calculation are presented in the next section .
the most important assumption of the method we use is that the observed cv sample is representative of the underlying population . whether this is a reasonable assumption
is addressed further in section [ sec : fluxl ] and [ sec : disc ] .
the number of cvs per unit volume is of course a function of position in the galaxy .
we ignore the ( weak ) radial dependence of @xmath0 , and assume that the vertical density profile is exponential @xmath56 with @xmath57 the distance from the galactic plane ( i.e. @xmath58 ) .
the local space density is defined as @xmath59 , the mid - plane value of @xmath0 .
the x - ray luminosity function , @xmath60 is defined so that @xmath61 where @xmath62 is the local space density of cvs in the luminosity bin of width @xmath63 centred on @xmath64 . in other words , @xmath65 is the integral of @xmath1 over @xmath64 . in calculating @xmath65
, we will compare two assumptions regarding galactic scale height .
first , as in @xcite , we take @xmath66 for long - period cvs , and @xmath67 for short - period systems , as a ( quite crude ) approximation of the fact that these are young and old populations , respectively .
the second @xmath65 calculation uses a single scale height of @xmath67 for all systems , and gives results that are not significantly different ( see section [ sec : rhoresult ] ) .
this implies that the simpler approach of using only a single scale height is justified , and that is what we will assume in calculating @xmath1 ( the reason we require the simpler assumption for @xmath1 will be explained in section [ sec : simphi ] ) .
interstellar extinction is computed by integrating the density of the interstellar medium along each line of sight to obtain the neutral hydrogen column density , @xmath43 .
we again assume only vertical dependence in the density of gas ; i.e. , @xmath68 we take @xmath69 ( e.g. @xcite ; @xcite ) , and we find @xmath70 by assuming @xmath71 for @xmath72 deg @xcite , while the models of @xcite and @xcite give lower average values still .
however , the results that will be presented in the next two sections are not sensitive to this assumption , and do not change significantly if we take the midplane extinction to be as low as @xmath73 .
this is probably because the two surveys are restricted to quite high galactic latitudes . ] , and @xmath74 @xcite . as discussed in section [ sec : xlum ] , we will assume a @xmath42 thermal bremsstrahlung spectrum for all cvs in our sample .
we will explain below that the @xmath0 calculation uses the maximum distance at which a given cv could have been detected .
since this maximum distance depends on the ratio of @xmath45 to the flux limit , which may as well be expressed as a ratio of count rate to limiting countrate , the assumed x - ray spectral shape has very little influence on the @xmath0 calculation ( only the interstellar absorption depends on the x - ray spectrum ) .
since the combined survey is complete ( up to the variable flux limit ) , we simply need to count the systems detected inside the observed volume to calculate @xmath0 ( and do the same in a set of luminosity bins to find @xmath1 ) . however , the dependence of @xmath0 on @xmath57 , as well as the flux- rather than volume - limited nature of the sample need to be accounted for .
both these complications are taken care of by the well - known @xmath75 method ( see e.g. @xcite ; @xcite ; @xcite ; @xcite ) .
our aim is to find not only a best estimate of @xmath65 and @xmath1 , but also to estimate the errors on both , accounting for all sources of statistical error ( including distance errors , small number statistics , and observational error on @xmath45 or count rate ) .
we describe in the following sections first how we implement the @xmath75 method , and then the monte carlo simulation to find the uncertainties on our measurements , as well as the numerical tests we use to verify that the error estimate for @xmath65 is correct .
the uncertainty in @xmath1 proves to be harder to estimate , and we will return to this issue in sections [ sec : obsphi ] and [ sec : simphi ] .
the @xmath75 method essentially allows the ` volume limit ' of the survey to vary according to the luminosities of the systems in the sample . for each observed system , we find the ` generalized ' ( i.e. , taking into account the exponential vertical density profile ) maximum volume , @xmath76 \label{eq : v}\ ] ] ( e.g. @xcite ) .
the index @xmath77 represents the cvs in our sample ; @xmath78 is the solid angle covered by the survey and @xmath79 , with @xmath80 the maximum distance at which cv @xmath77 could have been detected ( given its luminosity and the survey flux limit ) .
equation [ eq : v ] assumes the spatial dependence of @xmath0 as given by equation [ eq : z ] .
@xmath81 is then the volume probed by the survey for sources with @xmath24 the same as the observed system @xmath77 . because @xmath82 ( and in the case of the nep survey , also the flux limit ) is variable over @xmath78 , we compute each @xmath81 as a sum over smaller solid angles , @xmath83 . for the rbs ,
each @xmath83 is a slice subtended by @xmath84 in @xmath82 ( excluding the small area that is also covered by the nep survey ) , while the nep area is divided into a @xmath85 pixel grid . for both surveys ,
the @xmath83 are sufficiently small that the error introduced by @xmath82 varying over @xmath83 is negligible .
the space density @xmath65 is then the sum of the space densities represented by each cv , i.e. @xmath86 the luminosity function @xmath1 in a given luminosity bin is calculated similarly , by restricting the sum to systems in that luminosity bin .
the @xmath87 and @xmath88 values for the systems in our sample , obtained with the maximum distance found by using the best @xmath89 and @xmath45 estimates , are given in table [ tab : distances ] .
these values thus neglect for the moment all uncertainties .
we list them only as an indication of the contribution that each system makes to the total @xmath65 . in order to find the error on @xmath65 , we compute its probability distribution function using a monte carlo simulation that finds @xmath65 ( as described above ) for a large number of mock samples with properties that fairly sample the parameter space allowed by the data .
this also produces a best - estimate @xmath1 .
the appropriate mock samples are generated as described in @xcite .
briefly , we treat each observed cv as a representative of a population of similar systems , and allow each population to contribute to each mock sample . for a given mock sample ,
we generate 20 mock cvs by drawing @xmath45 and @xmath89 for a mock cv from the probability distribution functions ( gaussian errors in @xmath45 and @xmath89 distributions as computed in section [ sec : dist ] ) of a real observed system ; thus each mock cv is the counterpart of one of the observed cvs .
this takes care of the uncertainties in @xmath45 and @xmath89 ( or , combining those two parameters , @xmath24 ) . the poisson uncertainty associated with the small sample size
is accounted for by a weighting factor , @xmath90 , drawn from the probability distribution of the number of sources belonging to the population ( corresponding to a particular observed system ) that one expects to detect in the combined survey .
using bayes theorem , this distribution is related to the poisson distribution @xmath91 by @xmath92 where the actual number observed in the rbs and nep survey is @xmath93 , since we take each observed cv as representing a population of similar systems . as in @xcite , we use the uninformative prior @xmath94
. then for each mock sample , @xmath95 , and @xmath96 is found by summing only over the @xmath77 corresponding to mock cvs with luminosities placing them in that @xmath64 bin . we generate a large number of mock samples , and calculate @xmath65 and @xmath1 for each of them .
we then add the @xmath1 values from all the mock samples in a given bin , and divide by the number of mock samples .
the distribution of @xmath65 is normalized to give a probability distribution function . in our @xmath65 calculation ,
the largest part of the error budget comes from uncertainty in distance ( or , equivalently , @xmath24 ) .
the calculation naturally yields the error on @xmath65 , but the situation for @xmath1 is more complicated ( see section [ sec : test ] ) .
we carried out several tests to confirm that the bayesian procedure outlined above produces reliable error estimates for @xmath65 .
the input for each test is a galaxy model ( the same as assumed by equation [ eq : z ] and [ eq : rho_ism ] ) , survey area and flux limit , and a cv luminosity function .
we experimented with several different luminosity functions , as well as with including simulated errors in luminosity for the `` observed '' cvs .
a given test first simulates an `` observed '' sample , and then calculates the @xmath65 distribution based on that sample in the same way as described above .
this is repeated to generate many samples with their @xmath65 distributions ( note that for a given input , the samples do not all have the same size ) . for certain input luminosity functions
, we found that for @xmath2268% of these simulated samples , the true input @xmath65 is contained a 1-@xmath39 interval of the calculated @xmath65 distribution .
this holds when the combination of input flux limit and luminosity function produces an average of 20 sources per simulated sample , but also when the simulated samples are much larger , and the @xmath65 distributions become narrower . the luminosity functions for which the tests fail ( in other words , for which we can not recover the input @xmath65 ) are those in which a significant contribution to @xmath0 is made up by systems so faint that none of them are included in the `` detected '' samples
this is then what it means for the observed sample to be representative of the intrinsic population at least one cv belonging to any intrinsically large population is detected .
we perform similar tests to see if we can recover the luminosity function used as input .
the result is that we can , but only when the errors in luminosity we assign to `` observed '' cvs are small compared to the size of the luminosity bins . in this case , we can always reliably place a given cv in a single @xmath24 bin , and the error on each @xmath1 bin can be found in the same way as the error on @xmath65 . however , clearly also here , we can not recover any faint end of the luminosity function beyond the faintest `` detected '' @xmath24 .
when we simulate a survey where the errors on @xmath24 of the `` observed '' cvs are large , we recover a luminosity function that is broader and flatter than the input .
furthermore , we can no longer find the error on a given @xmath1 bin as before , from the distribution of values in that bin .
the reason for this is the correlation between @xmath97 and @xmath24 for that system .
a fainter system can be detected only to a smaller distance , and hence yields a larger @xmath87 ; this is illustrated in fig .
[ fig : rholxcorr ] . to understand why the calculation does not give the error on @xmath1
, one can consider constructing a luminosity function with bins sufficiently small that , for each mock sample , either 1 or 0 systems fall in a given @xmath24 bin .
for some bin of this luminosity function , the value of @xmath1 , if it is not 0 , will be determined by the correlation shown in fig .
[ fig : rholxcorr ] , with relatively small scatter , determined by the weighting factor @xmath90 .
the most important source of error will be gone , since @xmath24 is known , in a given bin . for our real sample ,
the uncertainties in @xmath24 are large enough that a given mock system is not always in the same @xmath24 bin ( even for very coarse bins ) , and in part because we compute @xmath1 over a large range in @xmath64 , a typical mock sample leads to many empty bins .
this implies , in a given bin , many points ( corresponding to many mock samples ) at @xmath98 .
the distribution of the non - zero values is determined by the correlation , with perhaps a factor of 2 or so spread ( 1 or 2 systems in that bin ) and with some additional scatter from the distribution of the weights .
therefore , the distribution of values in a given bin can not tell us the error on @xmath1 in that bin , and more work is needed ( see section [ sec : simphi ] ) . and @xmath24 for the rbs ( fine curve ) , the nep survey ( dotted curve ) , and the combined survey ( bold curve ) .
this shows the @xmath97 one would find for a given system , as a function of its @xmath24 .
the correlation becomes flat at the bright @xmath24 end , because a sufficiently bright system can be detected out to the edge of the galaxy .
, width=317 ]
here we present the results of the calculations described in section [ sec : rho ] above .
we will elaborate on the interpretation and limitations of these results in section [ sec : fluxl ] .
the probability distribution function of @xmath65 resulting from the calculation described above is shown in fig .
[ fig : rhopdf ] . as already noted , we find that the distances errors dominate the total uncertainty in @xmath65 , although the small sample size also contributes significantly .
the mode , median , and mean of the distribution are @xmath99 , @xmath100 , and @xmath101 , and are marked by solid lines .
the dashed lines at @xmath102 and @xmath103 show a 1-@xmath39 confidence interval ( the 16th and 84th percentile points of the distribution ) . considering the large errors ,
our @xmath0 estimate of @xmath104 resulting from the combined rbs and nep survey is in reasonable agreement with the value found by @xcite when omitting the two systems for which they estimated very small distances .
we will compare the new estimate with the higher value we found from the nep survey alone in some detail in section [ sec : consistent ] . considering short- and long - period cvs separately ( even assuming that tw pic , rbs490 , rbs1411 , and iq eri are all short - period systems )
, we find that the space density estimate is dominated by long - period systems ; @xmath65 of long - period cvs is roughly 1.2 times that of short - period systems ( this is discussed further in section [ sec : disc ] ) . if we use a single galactic scale - height of 260 pc for all systems , we find @xmath105 ; this is not significantly different from the result when using different scale heights for long- and short - period cvs .
similarly , if we correct for the possible distance biases listed in table [ tab : bias ] , the resulting @xmath65 is insignificantly different , namely @xmath106 .
distribution resulting from our simulation . for this distribution ,
we used scale heights of @xmath67 and @xmath107 for short- and long - period systems , respectively .
solid lines mark the mode , median , and mean at @xmath99 , @xmath100 , and @xmath101 .
dashed lines show a 1-@xmath39 interval from @xmath102 to @xmath103 .
, width=317 ] the same calculation that gives @xmath65 also yields @xmath1 , as explained in section [ sec:1overvmax ] . the x - ray luminosity function obtained from this calculation ( assuming @xmath108 for all cvs ) is shown in fig .
[ fig : obsphi ] .
we also plot there the @xmath109 estimates of the 20 detected cvs , in order to show the size of the errors in these values .
the uncertainty in @xmath24 has the effect of smoothing features in the luminosity function .
this explains , for example , why the inferred luminosity function has a gradual turn - over at the faint @xmath24 end , rather than a sharp cut - off at the faintest observed @xmath24 value .
if we integrate this histogram , we find @xmath110 , a number between the median and mean of the @xmath65 distribution . as explained in section [ sec : test ] , this calculation does not provide the uncertainty on @xmath1 , hence we do not show error bars on the histogram in fig . [
fig : obsphi ] .
furthermore , although the result does not change significantly if we correct for the possible distance biases discussed in section [ sec : dbias ] , the luminosity function probably still suffers from malmquist - type biases , and while the turnover at the faint end might represent a real cutoff , it might also just reflect the limited depth of the observations .
we will return to these problems in section [ sec : simphi ] . , the calculation that yields this histogram does not produce an estimate of the errors on it .
the points with error bars show the estimates of @xmath24 for the 20 non - magnetic cvs that the calculation is based on ( these points are arbitrarily offset in the vertical direction , for display purposes ) .
, width=317 ]
whether a given survey produces a cv sample that is representative of the underlying population is determined by its depth and area .
no evidence of even a large population of cvs is expected to show up in a a flux - limited survey , if such a population is sufficiently faint .
we can , however , still place a limit on the size of a hypothetical faint population of cvs that may have gone undetected in these two surveys ; this is done in section [ sec : limits ] below . in detail , the effect of the flux limit on the @xmath0 estimate that a given survey produces depends on how steeply the luminosity function rises towards the faint end and on where it cuts off or drops . in section [ sec : simphi ]
we return to the matter of errors on @xmath1 , and attempt to constrain the shape of the true luminosity function that , after being subjected to the appropriate observational biases and errors , would give rise to the ( smoothed ) luminosity function we actually observe . in section
[ sec : consistent ] , we also compare the two individual surveys that were combined to construct our sample , to verify that they are consistent with each other . in order to calculate how large a faint cv population could have gone undetected
, we use another monte carlo simulation .
we assume that there is a population of cvs , all with the same @xmath24 , and find the upper limit on its space density , based on detecting no member of this population .
while the assumption of a single-@xmath24 population is unphysical , it gives useful results that are easily expressed in terms of @xmath24 .
we again model the galaxy as having an exponential vertical density profile and no radial dependence in the number density of stars ; we also include extinction , in the same way as described in section [ sec : assump ] .
we assume a single scale height of @xmath111 pc for the hidden population ( as we did in section [ sec : obsphi ] when calculating @xmath1 ) .
we then find the value of @xmath65 for which the predicted number of detected systems is 3 ( detecting 0 systems such systems is then a 2-@xmath39 result ) .
[ fig : limit ] shows the maximum allowed @xmath65 as a function of @xmath24 for cvs that make up the hypothetical hidden population , from the rbs and nep survey separately ( the middle , and top - most fine histograms , respectively ) , as well as for the combination of the two surveys .
the limit from the simulation for the combined survey is plotted as a bold histogram , and the fine curve is a fit to the data , given by @xmath112 thus for @xmath113 , we have @xmath114 .
this limit is stronger than we previously found for the nep survey alone ( @xmath65 of @xmath8 implying @xmath115 ; @xcite ) ; this is a result of the much larger volume probed by the rbs ( see section [ sec : consistent ] ) .
note that here @xmath65 refers only to a possible undetected population , and does not include the small contribution from the observed systems .
as noted earlier , the luminosity function presented in section [ sec : obsphi ] is expected to suffer from malmquist - type biases , and we were not able to estimate the uncertainty on it .
biases in luminosity functions computed by the @xmath75 method are discussed in detail by e.g. @xcite , @xcite , and @xcite .
@xcite show that even when unbiased distance estimates are used , @xmath1 is still biased . in order to correct for these effects ,
we have to assume a functional form for the true luminosity function .
we will model it as a power law , and then determine the power law index that best reproduces the observations we have . in the same way as described in section [ sec :
test ] , we populate a model galaxy with cvs from an input luminosity function ; here the form is @xmath116 for the range @xmath117 , and @xmath98 elsewhere ( in practice , of course the input @xmath1 is discrete ) .
using the flux limit and area of the combined rbs and nep survey , we then find the sample that would be detected and calculate an output @xmath1 exactly as in section [ sec : obsphi ] .
errors in @xmath24 similar to those of the real cv sample are assigned to the `` detected '' cvs ( these errors are gaussian in @xmath118 , which is not quite the case for all the real systems ; see table [ tab : distances ] ) . for a given input @xmath1 ,
we repeat this many times ( note that the `` detected '' cv samples do not all have the same size ) .
then we compare the distribution of the output in every @xmath24 bin with the observed @xmath1 presented in section [ sec : obsphi ] .
we vary the power - law index @xmath119 , to find the best input @xmath1 .
the normalization constant is fixed ( for a given @xmath119 ) so that the integral of @xmath1 over the range @xmath120 is the same for the assumed input as for the observed @xmath1 , plotted in fig . [
fig : obsphi ] .
the position of the faint cutoff in the input @xmath1 is not important we would obtain the same result if it extended to fainter @xmath24 .
however , allowing the input @xmath1 to be non - zero to @xmath121 results in output values much higher than the data in the brightest few @xmath24 bins ( the brightest position of the cutoff consistent with the data depends on the the power law index , but we performed only a single - parameter fit , with both bright and faint cutoff @xmath24 fixed , because the calculations are computationally expensive ) .
we use a single scale height in these simulations ( 260 pc ) , because @xmath24 is the only property that a simulated cv has ( we can not assign an age , because , as mentioned before in section [ sec : xlum ] , @xmath24 is not a clean indicator of @xmath10 ) . since the `` detected '' cv samples in these simulations are subject to the same selection criteria as the real sample
, they are affected by malmquist - type biases in the same way as the data . therefore , to the extent that our simple galaxy model is suitable , the best - estimate power - law index , found by comparing these simulations to the observations , is unbiased .
the assumed @xmath1 resulting in the best - fit model output is shown in fig .
[ fig : obssimphi ] , together with the output from the simulation , as well the observed @xmath1 from section [ sec : obsphi ] .
we find @xmath122 , where the error on @xmath119 is based on @xmath123 increasing by 1 .
for this best - fit , the reduced @xmath123 is 1.1 .
note that the output is lower than input @xmath1 at the faint end ; this is the expected effect of the flux limit .
also , at the bright end , the sharp cutoff in the input is not recovered , because of error in @xmath24 .
we expect that the errors on the output from these simulations give a reliable indication of the uncertainty on the observed @xmath1 .
( crosses ) , and the output resulting from the calculation described in section [ sec : simphicalc ] ( points with error bars ) , plotted over the observed @xmath1 ( histogram ) .
the data shown here are the input that gives the best match between the simulated output and the observed @xmath1 .
the points and error bars are the median and a 1-@xmath39 interval of the the distribution of values in each bin , resulting from many `` detected '' samples .
, width=317 ] this calculation also allows us to check the distribution in height above the galactic plane of the cv sample . since it is a high galactic latitude , flux - limited sample , it is not expected to have the same @xmath57-distribution as the underlying population .
however , we can check that the observed sample is consistent with the assumed galaxy model . using the best - fit power law input @xmath1
, we construct a smooth cumulative probability distribution function for @xmath57 , from many `` detected '' cv samples .
we then use a kolmogorov - smirnov ( ks ) test to compare this to the @xmath57-distribution of the real sample of 20 systems ; this is done many times , in order to sample the large errors in @xmath57 of the real cvs .
we find that the probability that the model and observed distributions are drawn from the same parent population is 0.56 , for our model scale - height of 260 pc .
given that the observed sample is small , this is not a very sensitive test the sample is also consistent with an assumed scale height of 120 pc ( in this case , the probability that the model and observed distributions are drawn from the same population is 0.14 ) . in @xcite
, we found @xmath124 from the nep sample of only 4 cvs .
given the large errors in both this previous measurement and the one based on the combined survey presented here , they are not too different . however , if we calculate @xmath65 based on the rbs alone , the result is almost an order of magnitude less than a measurement based only on the nep survey .
the @xmath125 probability distribution functions for the two separate samples are shown in fig .
[ fig : comparepdf ] and are at first sight inconsistent . from the rbs ( upper panel ) and
nep data ( lower panel ) .
the vertical dashed lines mark 1-@xmath39 intervals .
, width=317 ] this would be easy to understand if the nep survey were simply more powerful than the rbs , i.e. if it reached a fainter population ( represented by the faint system ex dra ) than the rbs was capable of finding
. however , this is not the case the rbs is in fact more powerful than the nep survey . despite the much brighter flux limit of the rbs
, it reached a larger volume ( at all @xmath24 ) than the nep survey , because of its wider angle ( see fig .
[ fig : survey_vol ] ) .
it is then a fluke that the faintest cv in our combined sample was detected in the nep survey , rather than in the rbs .
based on the detection of ex dra in the nep survey , we would expect to detect around 2 such systems in the rbs ( since the volume of the rbs for cvs as bright as ex dra is roughly twice that of the nep survey ) ; the non - detection of any cv at that @xmath24 in the rbs is therefore unlucky , but has less than 2-@xmath39 significance . .
this is basically the inverse of what is plotted in fig . 1 .
the inset shows volume on a linear scale , for a small range in @xmath24 ; at the luminosity of ex dra , @xmath126 , the rbs volume is roughly twice that of the nep survey .
, width=317 ] for the rbs ( bold histogram and large points ) and nep ( fine histogram and smaller points ) samples separately , where the simulated @xmath1 in both cases uses the best - fit input power law @xmath1 , found for the combined survey ( i.e. , @xmath127 ) .
although the nep and rbs observed luminosity functions are quite different , each is satisfactorily fit by the corresponding simulated `` observed '' @xmath1 , for this assumed underlying luminosity function .
, width=317 ] beyond this one dominant system , to determine whether the results of the two surveys can be reconciled , we need to consider the luminosities of all the observed systems .
[ fig : rbsnepconsistent ] shows the observed luminosity functions ( constructed in the same way as for the combined survey in section [ sec : obsphi ] ) of the rbs and nep separately .
over - plotted are the output distributions , found as is section [ sec : simphicalc ] , but again treating the two surveys separately .
both simulations use as input the best - estimate power law @xmath1 found in the previous section .
least - squares fits of the output from the models to the observed luminosity functions give reduced @xmath123 of 0.8 and 0.9 for the rbs and nep survey , respectively ; both acceptable values . therefore , despite the large difference in @xmath65 when the two surveys are considered separately , they are consistent to within their uncertainties .
a basic assumption of the @xmath75 method is that the detected objects are representative of the luminosity function of the true underlying cv population .
note that ` representative ' here does not imply that faint systems are as common in the observed sample as intrinsically , but only that some ( or even one ) are detected ( see section [ sec : test ] ) . for an indication of whether our sample is likely to be representative of the underlying cv population , and thus gives reliable @xmath0 and @xmath1 estimates , it is important to know how faint cvs can be in x - rays .
theory predicts that the vast majority of cvs should be intrinsically faint ( e.g. @xcite ; @xcite ) ; @xcite and @xcite find that , although an as yet undetected faint cv population can not dominate the overall population to the extent predicted , observed cv samples are nevertheless strongly biased against faint systems .
the faintest secular average @xmath24 expected of cvs can be estimated from the gravitational radiation - driven @xmath11 .
we find that the @xcite relation between @xmath24 and @xmath11 predicts that the majority of cvs in the theoretical population of @xcite should have time - averaged x - ray luminosities of a few times @xmath128 and higher . in intrinsically faint cvs
, however , the rate of transfer of material onto the white dwarf surface ( which determines @xmath24 ) , is not the the same as the secular @xmath11 , since these systems are dwarf novae .
it is possible that , in the faintest cvs , hardly any material reaches the white dwarf surface during quiescence , so that they may spend most of their time at very faint @xmath24 ( perhaps with x - ray emission from the donor star being brighter than from the accretion flow ) .
the faintest short - period cvs in our sample have luminosities of a few times @xmath129 .
several intrinsically faint , short - period systems are now known to have @xmath130 ( see @xcite , as well as peter wheatley , public communication ) .
it is not known how intrinsically common such systems are , but they may well dominate the population .
the standard theory of cv evolution predicts that @xmath2270% of cvs are period bouncers ( e.g. @xcite ) . using the observed mass - radius relationship of cv donors ,
@xcite predict an even larger fraction of period bouncers .
observations are also now indicating a large population of intrinsically faint cvs ( probably both normal short - period cvs and period bouncers ) .
first , @xcite find a large number of intrinsically faint cvs at the shortest orbital periods .
furthermore , several period bouncers and good candidate period bouncers are now known ( e.g. @xcite ; @xcite ; @xcite ) , and @xcite argue that these systems may be common enough to make up most of the intrinsic population .
two ( related ) properties of the sample used here show that it probably does not fairly represent the underlying population : it contains no faint short - period cvs , and it probably contains no period bouncers ( see section [ sec : sample ] ) .
the lack of period bouncers alone likely means that it has missed at least half the intrinsic population .
furthermore , it is disconcerting that the faintest member of our sample , and therefore the system that dominates our @xmath0 measurement , is a long - period cv , ex dra , and is almost entirely responsible for the faint end of the luminosity function .
this system has @xmath131 , which is unusually low for a long-@xmath10 cv , and is probably caused by very high orbital inclination . ] .
population synthesis models predict that at most a few percent of all cvs are above the period gap ( @xcite finds less than 1% , while @xcite predict 3% ) .
although we find that long - period systems account for slightly more than 50% of our total space density , the data do not rule out these theoretical predictions .
for example , using the @xcite fraction of long - period systems , and assuming that we have not significantly under - estimated the space density of long - period cvs , the space density of short - period cvs is @xmath132 . using the upper limit on @xmath0 from section [ sec : limits ] , we find that a short - period cv population of this size could escape detection in the two surveys , provided that the systems have @xmath133 ( for the simple case of a hypothetical single-@xmath24 population of faint , undetected cvs ) .
clearly , if cvs are arbitrarily faint in x - rays , the data allow for an arbitrarily large population to escape detection .
however , we can also choose to place a restriction in terms of what we might consider reasonable x - ray luminosities for active cvs .
for example , if we integrate the best - fit power law luminosity function over the range @xmath7 ( this is a luminosity range where cvs are known to exist , but where we detect none ) , we find @xmath134 for systems at those luminosities , a factor of almost 3 larger than our @xmath65 estimate from the detected cvs .
this again indicates that it is reasonable to think that our @xmath65 estimate is low by a factor of more than 2 .
we find that a power law x - ray luminosity function extending beyond @xmath135 is inconsistent with the observed sample .
this is in agreement with figure 7 of @xcite , which shows that non - magnetic cvs should not be found at x - ray luminosities above @xmath136 .
the main difficulty in constructing a luminosity function from the rbs and nep sample is that the distance estimates are very imprecise ( with errors @xmath137% in some cases , implying that @xmath24 is only very poorly constrained ; see section [ sec : sample ] ) .
since parallax measurements are at the moment only available for a fairly small number of relatively bright cvs , this problem will persist until results from a mission such as _ gaia _ are available .
we finally note that we have not considered the impact of variability on the x - ray luminosity function .
the rass data were taken over a period of about 6 months , during which the ecliptic poles were observed many times , whereas lower ecliptic latitudes were covered only once @xcite .
frequently outbursting dne in the nep area were therefore probably observed both in quiescence and in outburst , while most dne in the rbs sample were likely observed only in quiescence .
to summarize , we have constructed a complete , purely x - ray flux - limited sample of 20 non - magnetic cvs , and have used it to place constraints on the space density and x - ray luminosity function of cvs .
our main conclusions are listed below . 1 . with the assumption that the combined non - magnetic cv sample from the rbs and nep surveys is representative of the intrinsic population , we find @xmath138 .
it is likely that this @xmath65 estimate excludes a large population of faint cvs ( consisting of both normal short - period systems and period bouncers ) , and that it is low by at least a factor of @xmath222 , as a result . in other words ,
the data are consistent with more than half of all cvs having @xmath7 , and escaping detection .
3 . to reach @xmath139 ( at the high end of the predicted range ) , we require that the majority of cvs have @xmath140 .
the precision with which @xmath0 and @xmath1 can be measured is mainly limited by poorly constrained distances to cvs .
we find it impossible to correct for bias in a measurement of the x - ray luminosity function , without assuming a functional form for @xmath1 .
if the x - ray luminosity function of non - magnetic cvs is a truncated power law , @xmath141 , the power law index that best reproduces our data is @xmath142 .
we thank the anonymous referee for comments that improved this paper . ak t. , bilir s. , ak s. , eker z. , 2008 , newa , 13 , 133 ak t. , bilir s. , ak s. , cokunolu k. b. , eker z. , 2010 , newa , 15 , 491 allen c. w. , astrophysical quantities , 3rd edn .
atholne press , london amres e. b. , lpine j. r. d. , 2005 , aj , 130 , 659 araujo - betancor s. , gnsicke b.t . , long k.s . , beuermann k. , de martino d. , sion e.m . , szkody p. , 2005 ,
apj , 622 , 589 aungwerojwit a. , et al . , 2006 , a&a , 455 , 659 bailey j. , 1981 , mnras , 197 , 31 baptista r. , cataln m. s. , costa l. , 2000 , mnras , 316 , 529 barwig h. , kudritzki r. p. , vogt n. , hunger k. , 1982 , a&a , 114 , l11 baskill d. s. , wheatley p. j. , osborne j. p. , 2005 , mnras , 357 , 626 bruch a. , engel a. , 1994 , a&as , 104 , 79 buckley d. a. h. , tuohy i. r. , 1990 , apj , 349 , 296 byckling k. , mukai k. , thorstensen j. r. , osborne j. p. , 2010
, mnras , 408 , 2298 cambrsy l. , beichman c.a . , jarrett t.h . , cutri r.m . , 2002 ,
aj , 123 , 2559 chen a. , odonoghue d. , stobie r. s. , kilkenny d. , warner b. , 2001 , mnras , 325 , 89 cieslinski d. , diaz m.p . , mennickent r.e . , pietrzyski g. , 2003 , pasp , 115 , 193 de kool m. , 1992 , a&a , 261 , 188 drimmel r. , spergel d. n. , 2001 , apj , 556 , 181 drimmel r. , cabrera - lavers a. , lpez - corredoira m. , 2003 , a&a , 409 , 205 felten j. e. , 1976 , apj , 207 , 700 gnsicke b. t. , sion e. m. , beuermann k. , fabian d. , cheng f. h. , krautter j. , 1999 , a&a , 347 , 178 gnsicke b. t. , et al . , 2009 , mnras , 397 , 2170 geijo e. m. , torres s. , isern j. , garca - berro e. , 2006 , mnras , 369 , 1654 gioia i.m . , henry j.p . , mullis c.r . , bhringer h. , briel u.g . , voges w. , huchra j.p . , 2003 , apjs , 149 , 29 gonzalez a. h. , faber s. m. , 1997 , apj , 485 , 80 harrison t. e. , johnson j. j. , mcarthur b. e. , benedict g. f. , szkody p. , howell s. b. , gelino d. m. , 2004 , aj , 127 , 460 hamilton r. t. , harrison t. e. , tappert c. , howell s. b. , 2011 , apj , 728 , 16 hawkins n. a. , smith r. c. , jones d. h. p. , 1990
mauche , ed . , accretion - powered compact binaries .
cambridge university press , cambridge , p. 113
henry j.p . , mullis c.r . , voges w. , bhringer h. , briel u.g . , gioia i.m . ,
huchra j.p . , 2006 , apjs , 162 , 304 hertz p. , bailyn c.d . , grindlay j.e . , garcia m.r . , cohn h. , lugger p.m. , 1990 , apj , 364 , 251 hilton e. j. , szkody p. , mukadam a. , mukai k. , hellier c. , van zyl l. , homer l. , 2007 , aj , 134 , 1503 hoard d. w. , et al . , 2010 ,
aj , 140 , 1313 howell s.b . , rappaport s. , politano m. , 1997 , mnras , 287 , 929 godon p. , sion e. m. , cheng f. h. , szkody p. , long k. s. , froning c. s. , 2004 , apj , 612 , 429 ivanova n. , taam r.e . , 2003 , apj , 599 , 516 jones m. h. , watson m. g. , 1992 , mnras , 257 , 633 kato t. , et al . , 2009 ,
pasj , 61 , 395 knigge c. , 2006 , mnras , 373 , 484 knigge c. , baraffe i. , patterson j. , 2011 , apjs , in press ( arxiv:1102.2440 ) kolb u. , 1993 , a&a , 271 , 149 kolb u. , 2002 , in gnsicke b.t .
, beuermann k. , reinsch k. , eds , asp conf .
261 , the physics of cataclysmic variables and related objects .
soc . pac .
, san francisco , p. 180
kolb u. , stehle r. , 1996 , mnras , 282 , 1454 kozhevnikov v. p. , 2010 ,
astl , 36 , 554 littlefair s.p . , dhillon v.s .
, marsh t.r . ,
gnsicke b.t . ,
southworth j. , watson c.a . , 2006 , science , 314 , 1578 littlefair s. p. , dhillon v. s. , marsh t. r. , gnsicke b. t. , southworth j. , baraffe i. , watson c. a. , copperwheat c. , 2008 , mnras , 388 , 1582 lutz t. e. , kelker d. h. , 1973 , pasp , 85 , 573 malmquist , k.g . 1924 , medd .
lund astron .
obs . , 2(32 ) , 64 mennickent r. e. , diaz m. p. , tappert c. , 2004 , mnras , 347 , 1180 mohanty p. , schlegel e. m. , 1995 , apj , 449 , 330 mouchet m. , bonnet - bidaud j. m. , buckley d. a. h. , tuohy i. r. , 1991 , a&a , 250 , 99 mukai k. , wood j. h. , naylor t. , schlegel e. m. , swank j. h. , 1997 , apj , 475 , 812 mukai k. , kinkhabwala a. , peterson j. r. , kahn s. m. , paerels f. , 2003 , apj , 586 , l77 mukai k. , zietsman e. , still m. , 2009 , apj , 707 , 652 nelemans g. , tout c. a. , 2005 , mnras , 356 , 753 north r. c. , marsh t. r. , kolb u. , dhillon v. s. , moran c. k. j. , 2002 , mnras , 337 , 1215 norton a. j. , beardmore a. p. , retter a. , buckley d. a. h. , 2000 , mnras , 312 , 362 odonoghue d. , soltynski m. g. , 1992 , mnras , 254 , 9 paczyski b. , 1981 , aca , 31 , 1 pandel d. , crdova f. a. , mason k. o. , priedhorsky w. c. , 2005 , apj , 626 , 396 pandel d. , crdova f. a. , howell s. b. , 2003 , mnras , 346 , 1231 patterson j. , 1984 , apjs , 54 , 443 patterson j. , 1998 , pasp , 110 , 1132 patterson j. , 2011 , mnras , 411,2695 patterson j. , moulden m. , 1993 , pasp , 105 , 779 patterson j. , raymond j.c .
, 1985a , apj , 292 , 535 patterson j. , raymond j.c . , 1985b , apj , 292 , 550 patterson j. , et al . , 2005 , pasp , 117 , 1204 perna r. , mcdowell j. , menou k. , raymond j. , medvedev m. v. , 2003 , apj , 598 , 545 politano m. , 1996 , apj , 465 , 338 predehl p. , schmitt j.h.m.m . , 1995 , a&a , 293 , 889 pretorius m. l. , knigge c. , 2008a , mnras , 385 , 1471 pretorius m. l. , knigge c. , 2008b , mnras , 385 , 1485 pretorius m. l. , knigge c. , kolb u. , 2007a , mnras , 374 , 1495 pretorius m. l. , knigge c. , odonoghue d. , henry j. p. , gioia i. m. , mullis c. r. , 2007b , mnras , 382 , 1279 ramsay g. , wheatley p. j. , norton a. j. , hakala p. , baskill d. , 2008 , mnras , 387 , 1157 reimer t. w. , welsh w. f. , mukai k. , ringwald f. a. , 2008 , apj , 678 , 376 revnivtsev m. , sazonov s. , krivonos r. , ritter h. , sunyaev r. , 2008 , a&a , 489 , 1121 richman h.r . , 1996 , apj , 462 , 404 ringwald f. a. , thorstensen j. r. , honeycutt r. k. , smith r. c. , 1996 , aj , 111 , 2077 ritter h. , burkert a. , 1986 , a&a , 158 , 161 ritter h. , zkan m.t . , 1986 , a&a , 167 , 260 robin a.c .
, reyl c. , derrire s. , picaud s. , 2003 , a&a , 409 , 523 rogel a.b . , cohn h.n . , lugger p.m. , 2008 , apj , 675 , 373 sazonov s. , revnivtsev m. , gilfanov m. , churazov e. , sunyaev r. , 2006 , a&a , 450 , 117 schmidt m. , 1968 , apj , 151 , 393 schoembs r. , vogt n. , 1981 , a&a , 97 , 185 schreiber m.r .
, gnsicke b.t . , 2003 ,
a&a , 406 , 305 schwope a. , et al . , 2000 , an , 321 , 1 schwope a.d .
, brunner h. , buckley d. , greiner j. , heyden k.v.d .
, neizvestny s. , potter s. , schwarz r. , 2002 , a&a , 396 , 895 shafter a. w. , holland j. n. , 2003 , pasp , 115 , 1105 shara m. , moffat a. , potter m. , bode m. , stephenson f.r . , 1993 , in regev o. , shaviv g. , eds , annals of the israel physical society , vol . 10 , cataclysmic variables and related physics .
israel physical society , jerusalem , p. 84
sion e. m. , szkody p. , cheng f .- h .
, huang m. , 1995 , apj , 444 , l97
smith a. j. , haswell c. a. , hynes r. i. , 2006 , mnras , 369 , 1537 smith h. , 2003 , mnras , 338 , 891 skrutskie m.f . , et al .
, 2006 , aj , 131 , 1163 stobie r. s. , ishida k. , peacock j. a. , 1989 , mnras , 238 , 709 szkody p. , howell s.b . , 1992 ,
apjs , 78 , 537 szkody p. , nishikida k. , liller w. , 2000 , pasp , 112 , 1607 szkody p. , et al . , 2002 ,
aj , 123 , 430 tappert c. , augusteijn t. , maza j. , 2004 , mnras , 354 , 321 thorstensen j. r. , 2003 , aj , 126 , 3017 thorstensen j.r . , lpine s. , shara m. , 2006 , pasp , 118 , 1238 thorstensen j. r. , lpine s. , shara m. , 2008 , aj , 136 , 2107 thorstensen j. r. , schwarz r. , schwope a. d. , staude a. , vogel j. , krumpe m. , kohnert j. , nebot gmez - morn a. , 2009 , pasp , 121 , 465 tinney c. g. , reid i. n. , mould j. r. , 1993 , apj , 414 , 254 torres s. , garca - berro e. , isern j. , 2007 , mnras , 378 , 1461 van paradijs j. , augusteijn t. , stehle r. , 1996 , a&a , 312 , 93 van teeseling a. , verbunt f. , 1994 , a&a , 292 , 519 van teeseling a. , beuermann k. , verbunt f. , 1996 , a&a , 315 , 467 vergely j .-
, ferrero r. f. , egret d. , koeppen j. , 1998 , a&a , 340 , 543 voges w. , et al . , 1999 , a&a , 349 , 389 voges w. , et al . , 2000 , iauc , 7432 , 3 vrtilek s.d . , silber a. , raymond j.c . , patterson j. , 1994 , apj , 425 , 787 warner b. , 1987 , mnras , 227 , 23 warner b. , 1995 , cataclysmic variable stars .
cambridge univ . press , cambridge warner b. , 2001 , in paczynski b. , chen w .- p .
, lemme c. , eds , asp conf
. ser .
246 , small telescope astronomy on global scales .
, san francisco , p. 159
wheatley p. j. , mauche c. w. , mattei j. a. , 2003 , mnras , 345 , 49 willems b. , taam r.e . , kolb u. , dubus g. , sandquist e.l . , 2007 , apj , 657 , 465 | we combine two complete , x - ray flux - limited surveys , the _ rosat _ bright survey ( rbs ) and the _ rosat _ north ecliptic pole ( nep ) survey , to measure the space density ( @xmath0 ) and x - ray luminosity function ( @xmath1 ) of non - magnetic cvs .
the combined survey has a flux limit of @xmath2 over most of its solid angle of just over @xmath3 , but is as deep as @xmath4 over a small area .
the cv sample that we construct from these two surveys contains 20 non - magnetic systems .
we carefully include all sources of statistical error in calculating @xmath0 and @xmath1 by using monte carlo simulations ; the most important uncertainty proves to be the often large errors in distances estimates .
if we assume that the 20 cvs in the combined rbs and nep survey sample are representative of the intrinsic population , the space density of non - magnetic cvs is @xmath5 .
we discuss the difficulty in measuring @xmath1 in some detail in order to account for biases in the measurement , we have to adopt a functional form for @xmath1 . assuming that the x - ray luminosity function of non - magnetic cvs is a truncated power law , we constrain the power law index to @xmath6 .
it seems likely that the two surveys have failed to detect a large , faint population of short - period cvs , and that the true space density may well be a factor of 2 or 3 larger than what we have measured ; this is possible , even if we only allow for undetected cvs to have x - ray luminosities in the narrow range @xmath7 .
however , @xmath0 as high as @xmath8 would require that the majority of cvs has x - ray luminosities below @xmath9 in the 0.52.0 kev band .
[ firstpage ] binaries stars : dwarf novae novae , cataclysmic variables x - rays : binaries methods : observational , statistical . |
solar cycle full - shape prediction methods aim to determine the approximate whole cycle curve , not only its peak magnitude and timing .
this task is particularly useful for numerous scientific and technological applications .
the main involved areas are the electric power transmission systems , airline and satellite communications , gps signals , and extra - vehicular - activities of astronauts during space missions and , more in general , all the solar - terrestrial interactions .
numerous techniques for time series forecasting are developed to accurately predict phase and amplitude of future solar cycles , but with limited success .
depending on the nature of the prediction methods we can distinguish five principal classes : 1 ) curve fitting ; 2 ) precursor ; 3 ) spectral ; 4 ) neural networks ; 5 ) climatology .
apart from precursor methods , the main limitation is the short time interval of reliable extrapolations , as the case of the mcnish - lincoln curve fitting method . in the climatological method we predict the behaviour of the future cycle by a weighted average from the past n cycles , based on the assumption of a some degree of correlation of the phenomenon .
modern classes of solar activity prediction methods appear to be reasonably successful in predicting ` long range ' behavior , the so - called precursor methods .
precursors are early signs of the magnitude of future solar activity that manifest before clear evidence of the next solar cycle .
there are two main kinds of precursor methods : geomagnetic and solar .
the basic idea is that if these methods work , they must be based on solar physics , in particular dynamo theory .
the precursor methods are correlated with a solar dynamo mechanism , where we suppose that the polar field in the descending phase and minimum of a cycle gives an indication of future fully developed toroidal fields inside the sun that will drive the next solar activity .
although the dynamo methods and , in general , the precursor methods seems to work quite well , they might be affected by some severe drawbacks , mainly depending on various unknown physical parameters . however , as pointed out by different authors , we need again a better scientific basis .
thus , at present , the statistical - numerical approaches , based on reliable characterization and prediction of complex time series behavior , without any intermediate physical model , still appear as valuable techniques to provide at least the basic elements for new insight for future advanced physical prediction methods .
many different techniques proposed in literature appear quite reliable and accurate to predict , well in advance , the peak and timing values of future solar cycles . on the other hand
, it is well known the difficulties in predicting the full evolution of future solar cycles , due to highly complex non - linear dynamical processes involved , mainly related to interaction of different components , deterministic and stochastic ones , of the internal magnetic fields .
we recall here some brief introductive considerations of the argument under study , as well reported in the reference article by sello , 2012a .
the predictions of the `` anomalous '' current solar cycle 24 allow us to deeply test our current knowledge on the complex dynamics of solar activity evolution and to further refine both our current solar dynamo models and new mathematical / numerical methods .
the main detected anomaly of cycle 24 is related to its extended period of minimal solar activity , with very few small and weak sunspots and flares
. there is now a general consensus about a clear diminished level of solar activity , suggesting a current transition phase between a long period of maximum and a next long period of moderately or weak activity ( see : de jager and duhau , 2009 ) . in the modern era there is no precedent for such a protracted activity minimum , but there are historical records from a century ago of a similar pattern .
based on the flux f10.7 activity index , there seems to be little question that the new - cycle activity 24 , started late in 2008 or in early 2009 .
the successive progression of the solar cycle was very slow but constant and a first peak of strong activity is well documented during year 2011 , with the detection of strong x - flares activity and large sunspot areas with complex magnetic field configuration ( see fig.1 showing the big region ar11302 ) . taken through a line centered @xmath0 filter .
beta - gamma - delta magnetic configuration and area : @xmath1 msh and located at @xmath2 ( h.a.s.o .
september 29 , 2011 15:30 ut ) . ]
the unusual character of the current solar cycle is also evidenced by the unexpected results of the prediction methods documented in the literature .
a comprehensive collection of predictions for solar cycle 24 is described by pesnell and regularly released by noaa / sec international panel ( pesnell , noaa / sec panel ) . the considered prediction methods include different classes : precursors , spectral , climatology , neural network , physics based , statistical , etc . in the panel forecast for sunspot numbers released in 2009 , half of the members predicts a moderately strong cycle of @xmath3 expected to peak in october 2011 ; the other half predicts a moderately weak cycle of @xmath4 peaking in august 2012 . if we consider a small but representative sample of different forecasts we clearly obtain the uncertainty of the whole set of predictions included in the noaa / sec panel document and explains the evenly split between a moderately strong and a moderately weak solar cycle . moreover , as the cycle progresses there is an increasing consensus towards a weak solar activity . for more details on the prediction results from different methods , including the nonlinear one
, we refer to paper by sello , 2012a . in order to quantitatively evaluate the performances of forecasts for solar cycle 24 ,
we need to compute a global error index , updated on a monthly basis , which gives a performance evaluation on the whole shape predicted curve as the cycle progresses .
unfortunately , it is not very common to find works in the literature specifically devoted to verify the performances a posteriori of previously proposed forecasting methods , also because this process requires a long term analysis using the observed data for a period of , at least , five or six years after the given prediction .
in fact , many authors , after the publication of their predictions during the increasing phase of the solar cycle , forget to evaluate a posteriori , several years later , the quality of the suggested methods . here we compared the nonlinear method prediction with the noaa / sec panel prediction which is the better estimate coming from many reliable and accurate classes of prediction methods .
a preliminary list of predictions from different methods can be found in a report by pesnell , nasa , goddard space flight center , may 2007 .
more updated reviews have been published in 2008 and 2012 . since waldmeier work ( 1935 ) , several authors have proposed to represent the shape of the solar cycle using families of analytical models with few free parameters , such as the start time and amplitude of the cycle , in order to estimate the mean trend of both the ascending and descending phases of the cycle .
these parametric fully deterministic analytical models have , as main objective , to provide a mean behavior of the cycle or a general curve shape fitting .
we note that some of the free parameters must be carefully tuned in these models and it is not always easy to accurately estimate their values near the minima .
however , here we consider the result shown by wilson , 2011 . using the hwr two free parameters planck - like general shape fitting function with the observed first 18-months sunspot numbers ( 12-month moving average ) of cycle 24 ( end 2010 ) , the author obtains a best - fit prediction for monthly smoothed sunspot numbers peak of 70 at august 2013 .
this fairly good result is consistent with the `` low '' panel consensus prediction ( pesnell , 2008 ) .
the global error evaluation for the whole shape ( amplitude and phase ) of the solar cycle was already suggested in sello , 2001 for previous solar cycle 23 .
here we recall shorthly the main definition only .
the accuracy of a full - shape prediction model is estimated through the computation of the `` global average prediction error '' defined as : @xmath5 where @xmath6 and @xmath7 are the predicted and observed monthly smoothed sunspot numbers , respectively and @xmath8 is the number of observations related to time t. here @xmath9 is the standard deviation of observed data . the global error can be updated on a monthly basis in order to see the behavior of the error as the cycle progresses .
of course , the optimal model corresponds to a minimum value of @xmath10 as a function of t. however , the final comparison of different prediction methods is given through the global error evaluation for the last t value which contains a more comprehensive cycle information .
for further details the reader may refer to : sello , 2001 .
figure 2 shows the monthly smoothed predictions released in the original article ( see : sello , 2012a ) , comparing nonlinear method with noaa / sec prediction , using observed data from sidc , rwc belgium up to october 2010 , when the initial increasing phase of activity is already well developed . of course
, successive updated and revised predictions are certainly more accurate but here we are interested to show the full forecast potential of a given full - shape prediction method , where we are again well distant from the next maximum phase .
however , in order to make a correct evaluation of different forecasting methods it is necessary to take into account only predictions made in the same period , with the same available information . at the beginning
we see a clear delay in the early ascent period of the cycle than predicted , determining a significant initial error of predictions both on amplitude and phase . in the following figure 3
we show the results of the global error computation as a function of time , on a monthly basis , as the solar cycle progresses . here
we can directly compare the global error with the corresponding solar cycle phase .
the observed monthly smoothed sunspot numbers are provided by wdc - silso , royal observatory of belgium , brussels . as stated in the related web site , on july 1st , 2015 , the sunspot number series has been replaced by a new improved version ( version nb .
2.0 ) that includes several corrections of past inhomogeneities in the time series . the most prominent change in the sunspot number values is the choice of a new reference observer , a.wolfer ( pilot observer from 1876 to 1928 ) instead of r. wolf himself .
this means dropping the conventional 0.6 zurich scale factor , thus raising the scale of the entire sunspot number time series to the level of modern sunspot counts .
more information about the various diagnostics and corrections applied to the sunspot number and group number series in version 2.0 can be found in : clette et al.(2014 ) .
the residual differences from the two time series are mainly located after 1950 with an irregular pattern that mimics the solar cycle behavior . in order to consistently perform our analysis even past july 1st 2015
, we reconstructed the original older values using a combined method , observed / predicted , that uses both the new observed values and a predicted set of differences , through a properly designed neural network predictor .
this allowed us to use , as a reference and without interruption , a consistent monthly smoothed sunspot number time series up to january 2016 ( available in june 2016 ) , where the global average prediction error value is well stabilized .
a further unexpected delay in the first increasing phase of the cycle has most affected both the prediction methods ; however , the subsequent recovery is quite rapid . the nonlinear method starts better with values less distant from the observed data .
however , as the cycle progresses the two methods tend to reach an equivalent error also due to some overestimated increasing values from nonlinear method .
it is already noted that there are clear evidences that the solar maximum , given by the monthly smoothed sunspot numbers , in the northern hemisphere occurred at @xmath11 ( with the value : @xmath12 ) , as we can see from the quite constant behavior of the observed data ( see also : sello , 2012b ) . after a short increasing phase ,
the observed data start to decrease since 2012.3 suggesting a monthly smoothed maximum reached at 2012.2 with the value : @xmath13 ( monthly mean maximum reached at 2011.9 with the value : @xmath14 ) , well in advance with respect to predictions .
this caused an increase of the updated global error for both the methods .
we recall here that the nonlinear method predicted a monthly smoothed maximum at around 2012.5 with the value : @xmath15 ; whereas the noaa / sec predicted a maximum value of @xmath16 at around 2013.4 .
after 2012.45 the two prediction methods start to deviate : in particular the nonlinear method starts the decreasing phase of the monthly smoothed values , whereas the noaa prediction confirms a stable increasing phase of the amplitude values .
this fact leads to significant effects on the global average prediction error .
the increasing rate of the first derivative is quite clear from the figure . in particular , after 2012.83 an inflection point reduces the increase of nonlinear global error curve , whereas the noaa global error curve maintains a nearly constant value of its first derivative .
a second smaller peak for monthly smoothed values is visibile near 2012.9 with a value around @xmath17 .
it is well known the multi - peaked structure of many solar cycle maxima and the related gnevyshev gap ( see : sello , 2003 ) .
a new strong increasing trend is visible after 2013.38 towards a greater monthly smoothed maximum reached at 2014.3 with the value : 81.9 ( monthly mean maximum reached at 2014.1 with the value : 102.3 ) .
this is the updated main peak of solar cycle 24 .
the multipeaked structure of the maximum phase of solar cycle 24 is now clearly visible .
this feature is very difficult to predict in detail using the current prediction methods .
in fact , after 2014.1 the global error of nonlinear method starts to increase whereas the global error of noaa / sec quickly reduces to smaller values .
the predicted amplitude of the monthly smoothed maximum is quite good for nonlinear method : @xmath15 vs @xmath18 ( only @xmath19 higher ) but unfortunately the predicted phase is quite wrong ( see figure 3 ) .
one possible explanation can be the times of reversals of the polar poloidal field that occur near the maxima of sunspot activity .
in fact , the current solar cycle 24 is characterized by a relatively high north - south asymmetry of polar field reversal : the northern polar field reversed its sign more than one year before the southern field ( 13.68 months ) .
it is well known that fluctuations are inherent to the babcock - leighton @xmath20-effect and the times of polar field reversals on both hemispheres are quite difficult to predict ( see : kitchatinov and khlystova , 2014 and references therein ) . a high north - south asymmetry in the reversal of poloidal fields produces a delay in the evolution of solar maximum phase , with a longer duration of the gnevyshev gap .
the consequence is an unpredictable delay in the developing of the maximum phase peaks .
thus solar cycle 24 proves to be @xmath21 weaker than the previous solar cycle and thus belongs to the class of moderate cycles , like cycles 12 to 16 , which were usual in the late 19th and early 20th century .
in fact , compared to recent strong cycles , such moderate cycles typically feature a broader maximum , with a multi year plateau on top of which two or more surges of activity can produce multiple peaks of similar height . as a comparison , for the previous solar cycle 23 , the nonlinear method predicted a monthly smoothed peak of @xmath22 higher than the observed value , well inside the predicted uncertainty ( sello , 2001 ) .
the last value of the global average prediction error computed up to : @xmath23 ( i.e. excluding the common first rising and the last descending phases of the cycle , outside the so - called extended maximum phase of a solar cycle ) is : @xmath24 and @xmath25 for nonlinear and noaa / sec methods , respectively .
these values quantify the global prediction error on the whole curves in the considered time interval ( about @xmath26 years ) .
the normalized mean values related to a unit time interval ( i.e. the slopes of the equivalent linear behaviors ) are : 0.19 @xmath27 and : 0.3 @xmath27 for nonlinear and noaa / sec respectively .
these values suggest that solar cycle @xmath28 was a quite difficult task for solar cycle predictors . as a comparison , for previous solar cycle 23 , the global average prediction error of nonlinear method reached a normalized mean value of 0.14 @xmath27 ( sello , 2001 ) . to complete the solar cycle full - shape prediction evaluation , we add to the global average prediction error , eq.(1 ) , a curve similarity index derived from spectroscopy comparison methods . more precisely , taking a lead from the spectral contrast angle methods used in hplc - uv spectroscopy , wan et al .
2001 , implemented vector based representations of tandem mass spectra .
a curve or spectrum is a n - dimensional vector constructed with n different predicted or observed values at the same time or frequency values .
here we use the euclidean distance to evaluate the lenght of vectors . the lengths ( r ) of two vectors a and
b @xmath29 are simply determined by : @xmath30 @xmath31 the two curves ( vectors ) can be quantitatively compared by the derived spectral contrast angle ( @xmath32 ) .
the angle , @xmath32 , is defined as : @xmath33 where @xmath34 and @xmath35 are the intensities or amplitudes of the corresponding curves a and b. an angle of zero degrees means there are no discernible curve differences . curves that resemble each other have vectors that point in the same direction in the space .
a @xmath36 degree angle indicates a maximal curve differentiation .
the similarity index is a real absolute number between 0 ( maximum difference ) and 1 ( maximum similarity ) and it can be expressend as a percent similarity . here
we computed a percent cycle similarity degree index based on the above spectral contrast angle comparing the shape of a given prediction curve ( a ) with the corresponding curve shape of observed values ( b ) .
considering the current whole set of @xmath37 data in fig.3 , the percent cycle similarity degree ( pcsd ) is : @xmath38 and @xmath39 for nonlinear and noaa / sec prediction curves , respectively .
the very low value of similarity index obtained for nonlinear method is mainly due to the presence of the second and dominant peak in the solar cycle activity , producing an opposite phase behavior in the observed and predicted curves .
another interesting full - shape cycle prediction method , based on a similar cycle concept developed by du and wang , 2011 ( using both the rising rate and solar minimum value to select the similar cycles ) , predicted monthly smoothed sunspot numbers for cycle 24 , using sidc data up to november 2010 , obtaining a peak around january @xmath40 ( months ) with a size of about @xmath41 .
if we compute the percent cycle similarity degree for the whole curve ( see fig .
5c in the cited paper ) , we found the value : @xmath19 . on the other hand , the cited hwr general shape fitting method ( wilson , 2011 ) ,
gives a percent cycle similarity degree for the whole curve of @xmath42 , a fairly good relative result . in order to better evaluate the shape similarity of the different prediction methods
, we computed the time behavior of index pcsd , starting from 12 months after the epoch : october 2010 ( 2010.788 ) .
figure 4 shows the pcsd time evolutions for the above four shape prediction methods : nonlinear , noaa , wang and hwr . after an initial high pcsd value for all methods due to the first initial phase of the cycle
, there is a sharp decreasing trend of the similarity index for all methods up to 20-months after the selected epoch . in the interval 20 - 33
months after the epoch the nonlinear method shows the better performance , with a pcsd value around 10% .
but after 34-months , the similarity index behavior completely changes : we note a sharp decrease of pcsd for nonlinear method and a fast increasing trend for noaa , hwr and wang predictions with peaks : 25% , 58% and 68% , respectively .
this phase corresponds to the second increase trend of the solar activity towards the second and higher peak of the cycle .
after the local peak the behavior of pcsd index changes again with a new fast and deep decreasing trend for all methods and after about 54-months from the epoch only the hwr prediction reaches an acceptable far from zero stable similarity degree index , around 4% .
although the hwr method has performed better in predicting the global shape of the solar cycle curve than the other methods considered here , the low value of the pcsd found ( @xmath42 ) , suggests a quite different full shape , as in fact we can observe , and that we are still too far from having a method for predicting the full cycle curve that is sufficiently accurate .
it will need further deep work on the basis of both precursors and numerical - statistical methods , as well as an upgrade of the current physical models based on both solar dynamo and stochastic mechanisms ( e.g. fluctuations of the meridional flow , see : schssler , 2007 ) in order to obtain quite reliable and accurate prediction methods not only for the amplitude and phase of the maximum peak but also for the solar cycle full - shape curve .
the degree of predictability of solar cycles is still a great subject of an extensive discussion inside the solar physics community and it is not yet clear whether it will be possible to model the overall multi - scale components involved in the physical processes , both deterministic and stochastic , responsible for the cyclical solar activity .
( see : bushby and tobias , 2007 ; abreu et al .
, 2008 ; for a recent review see : balogh et al . , 2014 ) .
it is quite significant a recent work by nandy and karak , 2013 , where , considering the turbulent pumping as the dominant mechanism for flux transport , it is possible to show that the solar cycle memory ( here defined as the finite time necessary for the coupling of the spatially segregated source layers of the dynamo mechanism ) is short , lasting less than a complete 11 year cycle and this implies that solar cycle predictions for the maxima of cycles are best achieved near the preceding solar minimum , about 4 - 5 years in advance , whereas long - term predictions are unlikely to be accurate .
this result is quite consistent with previous estimations of the predictability horizon ( tp = 4.85 years ) based on nonlinear dynamics concepts ( sello , 2001 ) .
a method of performance evaluation of a given full - shape solar cycle curve prediction is proposed , computing a running global average prediction error and a specific percent cycle similarity degree .
the unusual character of the current solar cycle 24 has put a strain on even the best consolidated forecasting methods . here
we compared the precursor non - linear method and the noaa / sec panel prediction using monthly sunspot number observed data up to october 2010 .
the evaluation of the full - shape cycle predictions is here performed considering the time interval : 2010.8 - 2015.85 the performance evaluation can be summarized by two indices : the last value of the global average prediction error computed on the whole curve and the corresponding percent cycle similarity degree . the first index was : 0.99 and 1.5 for nonlinear and noaa / sec prediction methods respectively .
the second index was : 0.01 and 0.19 for nonlinear and noaa / sec prediction methods respectively .
the best relative performance has been reached by hwr method with a similarity degree index of about 4 .
although none of the considered methods produced a sufficiently high value of the similarity degree , the small amplitude of the monthly smoothed peak is quite well predicted by various prediction methods , but due to the strong multipeak shape , with a significant wrong phase .
in fact , this is the main reason for a very low value of the similarity index obtained using the considered methods .
the second strong peak in the solar cycle curve is the main factor for a very low value of similarity index for the nonlinear method .
further deep work is needed in order to obtain quite reliable and accurate prediction methods of the solar cycle full - shape curve , both amplitudes and phases : for this task we suggest the use of the above two performance indices , the progressive values of the global average prediction error computed on the whole curve and the corresponding percent cycle similarity degree , to better evaluate and to quantitatively compare the `` a posteriori '' performances obtained from different methods developed for future solar cycles predictions .
abreu , j. a. , j. beer , f. steinhilber , s. m. tobias , and n. o. weiss , 2008 : for how long will the current grand maximum of solar activity persist ?
, geophys .
lett . , 35 , l20109 .
pesnell , w.d . ,
2007 : predictions of solar cycle 24 , solar cycle 24 , noaa / sec panel report , may 24 , 2007 ; predictions of solar cycle 24 , 2008 , solar phys , 252 : 209 - 220 ; solar cycle predictions .
281 , 507532 ( 2012 ) . sidc - team , world data center for the sunspot index , royal observatory of belgium , monthly report on the international sunspot number , online catalogue of the sunspot index : http://www.sidc.be / sunspot - data/,year(s)-of - data. wilson , r.m . ,
2011 : an estimate of the size and shape of sunspot cycle 24 based on its early cycle behavior using the hathaway - wilson - reichmann shape - fitting function , nasa technical paper , 216461 , marshall space flight center , al , usa , march , 2011 . | there are many proposed prediction methods for solar cycles behavior . in a previous paper
we updated the full - shape curve prediction of the current solar cycle 24 using a non - linear dynamics method and we compared the results with the predictions collected by the noaa / sec prediction panel , using observed data up to october 2010 .
the aim of the present paper is to give a quantitative evaluation , a posteriori , of the performances of these prediction methods using a specific global error , updated on a monthly basis , which is a measure of the global performance on the predicted shape ( both amplitude and phase ) of the solar cycle .
we suggest also the use of a percent cycle similarity degree , to better evaluate the predicted shape of the solar cycle curve . |
a central problem in the study of deterministic dynamical systems is to identify different types of asymptotic behavior of a given system and to understand how the behavior changes as the system parameter changes @xcite .
the asymptotic behavior can be , for instance , a periodic oscillation , a quasiperiodic motion , a sna , or a random or a chaotic motion @xcite .
there has been considerable work in the past addressing how dynamical systems develop chaos from periodic or quasiperiodic motions and how they can be characterized by various diagnostic tools such as lyapunov exponents , fractal dimemsion and recurrence analysis .
some among these like the lyapunov exponents and fractal dimension also require substantial amounts of data , free of noise , in order to perform well .
recently , a much simpler test for the presence of deterministic chaos was proposed by gottwald and melbourne @xcite .
their 0 - 1 test for chaos takes as input a time series of measurements , and returns a single scalar value either 0 for periodic attractors or 1 for chaotic attractors .
the test does not require phase space recontruction .
the dimension of the dynamical system and the form of the underlying equations are in general irrelevant , and the input is just the time series data
. the test returns a unit value in the presence of deterministic chaos and zero otherwise .
recently , many works have been carried out in this connection , including reliability of the 0 - 1 test @xcite , characterization of noisy symbolic time series @xcite and efficient time series detection in fpu lattices @xcite .
however all these studies focused only on distinguishing between regular and chaotic motions . in the present work we investigate and characterize the performance of the 0 - 1 test when applied to quasiperiodically forced dynamical systems exhibiting snas .
+ snas which are commonly found in quasiperiodically forced systems were first described by grebogi et al @xcite .
these attractors are geometrically strange as they are properly described by a fractal dimension but the largest nontrivial lyapunov exponent is negative , implying nonchaotic dynamics . briefly speaking , a sna has a complicated ( fractal ) structure but does not have a positive lyapunov exponent .
trajectories separate at rates that are slower than an exponential divergence @xcite .
snas have been reported in many physical systems such as the quasiperiodically forced pendulum @xcite , biological oscillators @xcite , and quantum particles in quasiperiodic potentials @xcite , which are also related to the anderson localization in the schr@xmath0dinger equation @xcite . also , these exotic attractors were confirmed by an experiment consisting of a quasiperiodically forced , buckled , magnetoelastic ribbon @xcite , in analog simulations of a multistable potential @xcite , neon glow discharge experiment @xcite , chemical oscillations @xcite and various electronic circuits @xcite .
snas can be quantitatively characterized by a variety of methods , including the estimation of lyapunov exponents and fractal dimension @xcite and spectral properties @xcite .
the geometric strangeness of the attractor can be measured through indices such as the phase - sensitivity exponent @xcite , while the chaoticity properties can be studied by examining the finite - time lyapunov exponents @xcite .
the transition from quasiperiodic to chaotic motion via snas can be studied by both the variance of finite time lyapunov exponents and the recurrence plots @xcite .
the quasiperiodic ( torus ) attractor and snas are characterized by the lyapunov exponents with zero and negative values , respectively . however , for quasiperiodically forced dynamical systems , the values of the lyapunov exponents are negative for both quasiperiodic attractors(tori ) and snas . in this case
the lyapunov exponents fail to detect transitions from quasiperiodic dynamics to snas . moreover , analytical methods such as the approximation of quasiperiodic forcing by a series of rational approximations , or the analysis of the phase sensitivity of the dynamics demand a knowledge of the underlying map or differential equations which is often not available .
however in the present work for the first time it is reported that the 0 - 1 test clearly distinguishes tori , snas and chaos , from the time series alone .
in particular this test indicates a value 0 for torus , a value in between 0 and 1 for sna , and for chaos a value that tends to 1. + the mechanisms and routes for the birth of snas , their characterization in different dynamical systems , and tools for distinguishing different dynamical transitions to snas were reported in @xcite . among the different routes to snas ,
heagy - hammel , fractalization and intermittency transitions are ubiquitous in quasiperiodically driven nonlinear systems . in the present paper
, we employ the 0 - 1 test to characterize different dynamical transitions to snas , in particular the heagy - hammel , fractalization and intermittent transitions .
specifically , the 0 - 1 test characterization for transition to snas is examined in quasiperiodically driven logistic map , cubic map and parametrically driven duffing oscillator to validate our conclusions . + the organization of the paper is as follows . in sec.ii
we briefly review the 0 - 1 test .
in sec.iii we report the applicability of the 0 - 1 test to quasiperiodically driven systems .
sec.iv examines the characterization using the 0 - 1 test for different routes of creation of snas , while in sec.v we discuss the transition from snas to chaos .
the final section gives our conclusions . in appendix
a we briefly describe our procedure to identify an optimal value for the parameter @xmath1 required for the 0 - 1 test . in appendix b we briefly analyze the behaviors of the translation variables @xmath2 and @xmath3 , and the asymptotic growth rate @xmath4 as a function of the length of the time series @xmath5 . in appendix c
we present the nature of attractors in the quasiperiodically forced cubic map using the 0 - 1 test .
finally , in appendix d we analyze the distribution of local asymptotic growth rate @xmath6 along with finite time lyapunov exponents(ftles ) for a statistical evaluation of the results .
consider an n - dimensional dynamical system @xmath7 , @xmath8 and denote a solution of the underlying system as @xmath9 .
let @xmath10 be an observable of the underlying system .
the 0 - 1 test uses the observable to drive the dynamics on a well chosen euclidean extension , and then exploits a theorem from nicol et al @xcite , which states that the dynamics on the group extension is bounded if the underlying dynamics is nonchaotic , but it behaves like a brownian motion if the dynamics is chaotic .
the 0 - 1 test for chaos first appeared in @xcite .
it is designed to distinguish chaotic behavior from regular behavior in deterministic systems .
the test results in a value 1 if the system is chaotic and a 0 if the system is regular .
theoretical justification was given to the 0 - 1 test in @xcite .
an implementation technique was also proposed for the 0 - 1 test in @xcite , which has been very useful for the present work . here
we present a brief outline of the 0 - 1 test followed by a modified version of the 0 - 1 test illustrated in @xcite .
consider a time series of length @xmath5 .
the 0 - 1 test begins with a computation of two variables @xmath11 and @xmath12 which are the so - called translation variables : @xmath13 @xmath14 where @xmath15 + here @xmath16 is an observable constructed from the time series .
the value of @xmath1 may be chosen randomly in the interval ( 0,2@xmath17 ) for nonquasiperiodic systems @xcite .
one may note that it is the behavior of these translation variables that determines whether the dynamics of the system is regular or chaotic .
the translation variables also exhibit a brownian motion as a function of @xmath18 , when the dynamics is chaotic . for brownian motion
we expect the average of [ @xmath19-@xmath20 and [ @xmath21-@xmath22 to grow like @xmath23 for @xmath24 therefore the square of [ @xmath19-@xmath20 and [ @xmath21-@xmath25 should asymptotically approach a linear growth proportional to @xmath18 for sufficiently large values .
hence , to find the behavior of the translation variables the 0 - 1 test calculates their mean square displacement @xmath26 , @xmath27^{2 } + \left [ q(j+n ) - q(j ) \right]^{2}\ ] ] however , in order to obtain a better convergence of mean square displacement , for numerical purpose , we have used a modified mean square displacement @xmath28(which exhibits the same asymptotic growth rate as @xmath29 ) of the form as suggested in @xcite @xmath30 where @xmath31 the theory behind the 0 - 1 test shows that if the dynamics of the system is regular , then ( 4 ) is bounded in time .
however if the dynamics of the system is chaotic then ( 4 ) scales linearly in time . for numerical purpose , with the limit in ( 4 ) , we must use @xmath32 . in practice , the limit @xmath28 exists when @xmath33 , where one expects @xmath28 to scale linearly .
the dynamics is further characterized through the computation of the asymptotic growth rate @xmath4 which can be calculated directly from @xmath28 by using a linear regression for the log - log plot of @xmath28 given by the defnition , @xmath34 it has been pointed out in @xcite that the linear regression method leads to a distortion for small values of @xmath18 .
hence , it has been recommended @xcite to estimate the asymptotic growth rate @xmath4 by using a correlation method which works better than the linear regression method . in the present work
we employ the correlation method for the estimation of @xmath4 . in this procedure
, @xmath4 is the correlation coefficient of the vectors @xmath35 now one can define the correlation coefficient @xmath36 \label{k - eq}.\ ] ] here @xmath37 where @xmath38 finally a value of @xmath39 indicates a non - chaotic data set while a value of @xmath40 indicates a chaotic data set .
we use the above analysis to characterize snas from quasiperiodic and chaotic attractors by properly choosing the value of c in eqs.(1)-(2 ) for quasiperiodic dynamical systems .
+ the functions @xmath11 and @xmath12 in eq.(1 ) can be viewed as a component of the solution to the skew product system , @xmath41
@xmath42 @xmath43 + driven by the dynamics of the observable @xmath44 . here ( @xmath45,@xmath46,@xmath47 ) represent the coordinates on the euclidean group rotation @xmath45 and translation @xmath48 in the plane .
it was argued that an inspection of the dynamics of the ( @xmath2,@xmath3 ) trajectories provides a quick and simple visual test of whether the underlying dynamics is regular or chaotic @xcite .
the above 0 - 1 test , when applied in the standard way to quasiperiodically forced systems , exhibiting snas , does not clearly distinguish snas from torus and chaotic attractors .
the 0 - 1 test returns either a value 0 or 1 for the snas depending on the route by which they are created , as long as the parameter @xmath1 in eqs.((1)-(2 ) ) is chosen as an arbitray number in the interval ( 0,2@xmath17 ) .
in fact , in a recent study dawes and freeland @xcite have attempted to distinguish between quasiperiodic attractors and snas with a modification of the 0 - 1 test by the addition of noise with the mean square displacement .
in particular , by choosing the value of @xmath1 to be equally spaced in the interval ( 0.1,@xmath49 ) , they showed that for a quasiperiodic attractor @xmath4 takes a value zero and for a sna it takes a value 1 with large deviations for certain values of @xmath1 . however
this method does not distinguish snas from chaotic attractors . in the present study
, we have chosen the value of @xmath1 in terms of the golden mean number like @xmath50 or @xmath51 , where @xmath52 to characterize snas in quasiperiodically forced dynamical systems , so as to avoid resonance with the translation variables .
details are given in appendix a. with such a choice , we find that we can clearly distinguish torus , sna and chaos as well as different dynamical transitions involving snas . these are discussed in detail in the following .
+ in our numerical analysis the value of the asymptotic growth rate @xmath4 is computed by taking the average of it over a number of segments in the given trajectory .
that is , a given trajectory is divided into a number of segments each of length @xmath53 and , in each of these , the corresponding asymptotic growth rate @xmath54 ( where @xmath55 denotes the segment ) is evaluated .
finally the average is taken to compute the value of @xmath4 .
specifically , we consider a total time series of length @xmath56 for both quasiperiodically forced logistic map and cubic map while for quasiperiodically forced double well duffing oscillator we take @xmath57 ( for continuous systems the data points are considered after sampling the time series data ) . in the total time series
@xmath58 data points is considered as the length of each segment .
+ we also note here that the qualitative nature of the attractors , the trajectories in the translation variables @xmath48 space , and the mean square displacement essentially remain unchanged as a function of @xmath5 ( as discussed in appendix b ) .
so whenever these quantities are displayed in the following , we use @xmath59 data points , but for the calculation of @xmath4 we use a much larger time series as mentioned above .
in this section we consider the applicability of the 0 - 1 test for quasiperiodically forced dynamical systems to identify snas and distinguish them from periodic , quasiperiodic and chaotic attractors .
we choose the following models of quasiperiodically forced dynamical systems for our present study ; + ( i)*_quasiperiodically forced logistic map _ * : + the dynamics is described by the coupled map @xcite , @xmath60 x_{n}(1-x_{n } ) , \\ & & \theta_{n+1}=\theta_{n}+\omega\,(mod\ , 1),\end{aligned}\ ] ] where @xmath52 is the irrational driving frequency , @xmath61 represents the forcing amplitude , and @xmath62 is a driving parameter . the different dynamical regimes can be characterized by using the rescaled parameter of the forcing amplitude @xmath63 versus @xmath62 . in this map the transition from torus to snas is characterized by the behavior of the largest lyapunov exponent as well as the characteristic distribution of the finite time lyapunov exponents .
further , snas are created at a saddle node bifurcation , when the dynamics shows type - i intermittency and other transitions .
+ ( ii)*_quasiperiodically forced cubic map _ * : + the system is given by @xmath64 where @xmath52 is the irrational driving frequency . by varying the values of @xmath65 and @xmath66 , the creation of snas has been shown to occur by means of four distinct routes due to the truncation of torus doubling bifurcations . in particular ,
the formation of snas through heagy - hammel , fractalization , type - iii intermittency and crises induced intermittency are described in @xcite .
the snas created through different mechanisms are characterized in terms of the lyapunov exponents , phase sensitivity and finite time lyapunov exponents .
+ ( iii)*_damped two frequency parametrically driven double well duffing oscillator _ * : + the damped two frequency parametrically driven double well duffing oscillator is described by the nonlinear differential equation @xmath67x+x^{3}=0 , \hspace*{0.5 cm } ( .=\frac{d}{dt})\ ] ] which is a well suited model for buckled beam oscillations . here
the driving frequency @xmath68 is chosen to be the irrational number @xmath69 .
as the parameters @xmath70 and @xmath71 are varied one can locate different regions in the parameter space which exhibit different dynamical states .
there are several mechanisms through which snas are formed in this system , two of which appear to be quasiperiodic analogies of intermittency transition in unforced systems .
in addition to these , torus collision route to sna as well as fractalization route have been identified in @xcite . to start with
we characterize periodic and chaotic attractors in eqs.(11 ) and ( 12 ) in the absence of quasiperiodic forcing . for this purpose
the value of @xmath1 in eqs.(1 ) and ( 2 ) is randomly chosen in the interval ( 0,2@xmath17 ) , and the observable is @xmath72 .
first , the periodic and chaotic behaviors are investigated in the logistic map by using the new coordinates of translation variables @xmath73 and @xmath74 . to analyze the behavior of the translation variables ,
let us fix the parameters as @xmath75 ( @xmath76 ) in ( 11 ) .
then the system admits regular dynamics . for @xmath77 ( @xmath76 )
the system exhibits chaotic dynamics . for regular / chaotic dynamics the trajectories in the ( @xmath78 , @xmath79 )
space indicate bounded / unbounded ( brownian ) motion [ see fig .
1a(i ) , fig .
1b(i ) ] , respectively . the bounded and unbounded behaviors in the ( @xmath78 , @xmath79 ) space is further investigated in terms of the mean square displacement . figs .
1a(ii ) and b(ii ) indicate mean square displacement versus @xmath18 which is oscillatory for periodic dynamics , and scales linearly with time for chaotic dynamics . a similar behavior has also been confirmed for the cubic map with @xmath80 in eq.(12 ) .
+ next , we consider the full dynamics of the logistic map given by eq.(11 ) and cubic map ( eq.12 ) in the absence of quasiperiodic forcing . while the logistic map exhibits period - doubling bifurcations as shown in fig .
2a(i ) , the cubic map exhibits period doubling route to chaos followed by period halving phenomenon as a function of the system parameter @xmath81 which is shown in fig .
in order to confirm the applicability of the 0 - 1 test , we evaluate the asymptotic growth rate @xmath4 of the mean square displacement given by eq.(7 ) in figs .
2a(ii ) and b(ii ) , respectively , for the logistic map and cubic map in the absence of quasiperiodic forcing . as expected the 0 - 1 test returns a value 0 for periodic motion and 1 for chaotic motion .
a similar analysis has been performed for the duffing oscillator ( 13 ) with single periodic force , but we do not present the results here for brevity . ( a ) variation of the largest lyapunov exponent ( b ) variation of the asymptotic growth rate ( @xmath4 ) from 0 - 1 test as a function of @xmath62 for fixed @xmath82 indicating torus@xmath83sna@xmath83chaos transitions in the quasiperiodically forced logistic map .
the vertical arrows in this and the following figures indicate transition points . here , @xmath84 with @xmath85 as the length of each segment . ] for quasiperiodically forced discrete driven systems as well as flows , the dynamics can be chaotic , nonchaotic or strange nonchaotic depending on the values of the system parameters .
previous studies show that in quasiperiodically driven dynamical systems the regions of snas can be quite complicated and also the boundaries between chaotic attractors and snas are convoluted , with regions of torus attractors interspersed among the regions of snas @xcite .
we now wish to characterize snas in these quasiperiodically forced systems by using the 0 - 1 test to distinguish snas with quasiperiodicity and chaos .
again , we take the time series data ( observable ) as @xmath72 and the data always contains a component of the quasiperiodic forcing .
+ in the present study , as noted earlier we fix a single value for @xmath1 in eqs.(1 ) and ( 2 ) in terms of an irrational number like @xmath50 or @xmath86 , where @xmath52 to characterize snas in quasiperiodically forced dynamical systems .
we zero down on a most suitable value of @xmath1 by working on the attractors in the space of the translational variables for a set of values of @xmath1 in terms of the golden mean for typical cases of torus , snas and chaos in the quasiperiodically forced systems ( 11)-(13 ) .
we pick up that value of @xmath1 which most closely replicates the expected nature of the attractors : torus like structure for quasiperiodic attractors , bounded motion for snas and unbounded motion for chaotic attractors .
the details are given in appendix a. by this procedure we find that a choice @xmath87 , where @xmath52 , is the most suitable value for the present study .
+ the different possible dynamical behaviors are characterized in the logistic map , by varying the values of @xmath63 and @xmath62 .
for instance , for @xmath88 , @xmath89 the system admits torus like motion ; for @xmath90 , @xmath91 it is a sna , while for @xmath92 , @xmath93 , it is a chaotic attractor @xcite .
figures [ fig3 ] a(i),b(i ) , and c(i ) clearly indicate the projection of the dynamical attractors in the ( @xmath94 , @xmath45 ) space .
then , the dynamics of translation variables in the ( @xmath78,@xmath79 ) space indicates that the points are uniformly distributed over a circle for quasiperiodic ( torus ) regime ( fig.3a(ii ) , brownian like behavior for chaotic regime ( fig.3c(ii ) ) , and for snas the ( @xmath78,@xmath79 ) space shows that the points are uniformly distributed and bounded with minimal brownian like behavior ( fig.3b(ii ) ) .
we characterize the behavior of the translation variables by using the mean square displacement@xmath95 as a function of @xmath18 , and find that @xmath28 is oscillatory for quasiperiodic regimes ( fig.3a(iii ) ) ; grows linearly with time for chaotic attractors ( fig.3c(iii ) ) , and for snas @xmath28 grows linearly initially with oscillatory motion , but tends to get bounded for large @xmath18 ( fig.3b(iii ) ) .
the aforementioned analysis clearly shows that the 0 - 1 test allows one to distinguish snas from quasiperiodic regime and chaos .
a similar analysis has also distinguished tori , snas and chaotic attractors in the quasiperiodically forced cubic map represented by eq.(12 ) . for details , see appendix c. + we next study the characterization of different dynamical attractors as a function of the parameter @xmath62 in the quasiperiodically forced logistic map ( eq.11 ) in the parameter regime @xmath62 @xmath96 by using the 0 - 1 test .
the logistic map ( eq.11 ) has been already investigated and various transitions to snas have been identified using tools such as lyapunov exponents and their variance , finite time lyapunov exponents and power spectral measures @xcite . here , we consider the above parameter regime investigated in ref .
@xcite for our 0 - 1 test analysis also . to compare our results using the 0 - 1 test with the already reported results on torus @xmath83sna@xmath83 chaos transitions , here we present the nature of the largest lyapunov exponent along with the asymptotic growth rate @xmath4 using the 0 - 1 test as shown in fig.4 . as noted above our present study of the 0 - 1 test using the value of @xmath1 = @xmath51 , where @xmath52 , indicates that the asymptotic growth rate returns a value 0 for torus regime , 1 for chaotic regime , while in the sna regime the value of @xmath4 varies in between 0 and 1 non - monotonically .
it may be also noted that there is an abrupt transition in the value of @xmath4 from a small value to a larger one ( indicated by a vertical arrow below @xmath97 ) during the transition from torus to sna . between the value of @xmath62 , [email protected]
, the torus gets wrinkled before transition to a sna which is indicated by a nonzero value of @xmath4 .
one also notes that while the lyapunov exponent ( @xmath99 ) as a function of @xmath62 ( fig.4(a ) ) does not clearly distinguish between torus and sna , where it takes negative value in both the regimes , the asymptotic growth rate @xmath4 clearly distinguishes the torus and sna regimes , as well as the chaotic regime .
+ similar analysis has been carried out for the quasiperiodically forced cubic map and duffing oscillator to confirm that with the choice of @xmath1 = @xmath51 , where @xmath52 , as an irrational number , the asymptotic growth rate @xmath4 varies between 0 and 1 .
one of our main concerns in this work is whether the 0 - 1 test can distinguish different dynamical transition scenarios involving snas through appropriate characterizations . in this section
we show clearly that the formation of snas through different routes exhibits distinctly clear signatures in terms of changes in the growth rate of @xmath4 .
in particular we demonstrate this for the ubiquitous routes , namely heagy - hammel route on torus doubling to chaos via snas @xcite , fractalization of torus leading to snas @xcite , and the appearance of snas through type -i intermittent phenomenon @xcite . for other routes also
, one can identify similar signatures .
we also note here that each one of the routes has specific significance as discussed in the literature @xcite : ( i ) heagy - hammel ( hh ) route : collision of doubled torus with its unstable parent leading to sna before the onset of chaos .
( ii ) fractalization route : gratual fractalization of torus giving rise to sna before the onset of chaos . (
iii ) intermittency route : analog of saddle node bifurcation leading to the birth of sna before the onset of chaos .
the first route we consider is the heagy - hammel(hh ) route . in this route ,
a sna occurs when the period doubled torus collides with its unstable parent .
that is , the period-@xmath100 torus gets wrinkled and upon collision with its unstable parent , a period-@xmath101 band sna is formed . to exemplify the nature of this transition ,
let us consider the quasiperiodically forced logistic map ( eq .
11 ) , where we fix the parameter @xmath102 , while @xmath62 is varied . in this case
, the hh transition takes place at @xmath103 @xcite .
when we examine the neighborhood of @xmath104 , the transition from torus to snas is characterized by the lyapunov exponent(@xmath99 ) as shown in fig .
5a(i ) . particularly in the torus region , @xmath105
, @xmath99 varies smoothly whereas in the sna region , @xmath106 , the variation in @xmath99 is irregular . to validate the applicability of the 0 - 1 test for snas
, we consider the aforementioned parameter region and analyze the asymptotic growth rate @xmath4 versus @xmath62 .
one can see that in fig .
5a(ii ) @xmath4 becomes minimal , that is 0 or nearly 0 , in the torus region .
when @xmath106 , the growth rate @xmath4 is varying between 0.2 and 0.7 non - monotonically .
the behavior of @xmath4 versus @xmath62 shows a transition from regular dynamical regime to a strange nonchaotic one with a threshold value @xmath104 , separating both regimes .
+ similarly , in the quasiperiodically forced cubic map @xcite given by eq .
( 12 ) , we fix the system parameter at @xmath107 and vary @xmath65 in the range 1.8865 to 1.8875 .
initially for @xmath108 the system has a period doubled torus .
then as @xmath65 increases to the value 1.8868 , the torus begins to wrinkle and approaches its unstable parent .
after this the torus collides with the parent and gives birth to a fractal attractor ( sna ) , when @xmath65 reaches the value 1.88697 . for this value of @xmath65 ,
the attractor possesses a geometrically strange and nonchaotic behavior @xcite .
it is noticed that these dynamical changes are clearly distinguished in the 0 - 1 test as shown in fig .
5b(ii ) , while the corresponding lyapunov exponent structure is shown in fig .
for the torus region the value of @xmath4 is close to 0 , and in the sna region it oscillates between 0.2 and 0.4 .
we can again note an abrupt transition in the value of @xmath4 at the transition point @xmath109 when the torus gets transferred to a sna .
+ we conclude that the results of 0 - 1 test indicate in both the quasiperiodically driven logistic and cubic maps that the value of @xmath4 is smooth in the torus region and , at the critical value , there is a sudden jump after which @xmath4 oscillates with irregular behavior .
+ the hh analysis has also been carried out for the parametrically driven duffing oscillator ( eq .
( 13 ) ) with the system parameter fixed as @xmath110 , while the value of @xmath71 is varied between 0.181 to 0.186 .
here one finds that the transition from a sna to a torus takes place at @xmath111 @xcite .
the lyapunov exponent @xmath99 does not show any distinct change in its behavior during this transition .
figure 6(a ) is a plot of @xmath99 as a function of @xmath71 , in the neighborhood of @xmath112 , where the variation of @xmath99 is small and smooth , but in the sna phase the variation of @xmath99 is large . on the other hand , the plot between the asymptotic growth rate @xmath4 and the parameter @xmath71 indicates as shown in fig .
6(b ) that the value of @xmath4 sharply ( but smoothly ) decreases from 0.78 to 0.00 within a narrow range of the parameter @xmath71 with respect to the transition from sna to the torus regime , again giving a clear distinction between regions of snas and quasiperiodic attractors .
we may also note that the hh transition is quite narrow in the parameter range @xmath71 as seen in fig.6 .
transition from sna to torus through hh mechanism in the duffing oscillator ( a ) behavior of the lyapunov exponent(@xmath99 ) ( b ) behavior of the asymptotic growth rate ( @xmath4 ) from 0 - 1 test . here ,
@xmath113 ( after sampling the data points ) and @xmath85 . ]
the fractalization route is a common and rapidly changing intriguing transition to snas as pointed out by nishikawa and kaneko @xcite . here
a period @xmath18 torus attractor gets wrinkled and eventually forms a @xmath18-band sna .
now we examine the fractalization route in the quasiperiodically forced logistic map ( eq . ( 11 ) ) , where we fix @xmath114 and the value of @xmath62 is varied from 2.64 to 2.66 . here , the variation of @xmath99 as a function of the control parameter @xmath62 as shown in fig .
7a(i ) describes the transition from a torus to a sna ( through a wrinkled torus ( wt ) ) which occurs at @xmath115 @xcite .
the lyapunov exponent @xmath99 is smooth and linear with no particular signature observed during this transition .
however , as shown in fig.[fig7]a(ii ) the results of the 0 - 1 test indicate that the transition from tori to snas via wt is characterized by the asymptotic growth rate @xmath4 which grows linearly from torus to wt , and becomes irregular at the transition for sna at @xmath115 .
+ transition from sna to torus ( a ) behavior of the lyapunov exponent(@xmath99 ) ( b ) behavior of the asymptotic growth rate ( @xmath4 ) from 0 - 1 test during the fractalization route in the parametrically driven double well duffing oscillator . here
@xmath116 ( after sampling the data points ) with @xmath85 . ] to study the fractalization mechanism in the cubic map(eq.(12 ) )
, we fix the parameter at @xmath117 , and vary @xmath65 in the range 2.14 to 2.17 . for @xmath118 the system admits a torus like motion ; for @xmath119 the system exhibits quasiperiodic oscillations of the doubled torus and when it is increased to @xmath120 , the doubled torus becomes a wt , which persists up to @xmath121 , where the oscillatory structure(wrinkled ) gradually approaches a fractal structure(sna ) @xcite .
this region is examined by both the lyapunov exponent(@xmath99 ) and the 0 - 1 test . again
the lyapunov exponent @xmath99 does not show any specific signature when the transition takes place [ see fig .
7b(i ) ] . but in the case of the 0 - 1 test as shown in fig .
7b(ii ) the value of asymptotic growth rate @xmath4 is nearly 0 in the torus region and is linearly increasing during transition from torus to wt , and exhibits an irregular behavior at the transition to snas .
+ the fractalization route occurs in the parametrically driven duffing oscillator with @xmath110 , and @xmath71 in the range @xmath122 .
the transition from a sna to a torus occurs at @xmath123 @xcite .
we find that at this transition value of @xmath71 the asymptotic growth rate @xmath4 decreases to zero from the sna region to the torus one [ see fig.8(b ) ] , whose lyapunov exponent is as shown in fig.8(a ) . again during the transition there is an irregular behavior in @xmath4 during transition from snas to wt , while it is fairly smooth and linear ( with small fluctuations ) in @xmath4 from wt to torus .
one can conclude that the asymptotic growth rate @xmath4 behaves differently in the fractalization route compared to the hh mechanism .
transition from intermittent sna to torus motion in the two frequency parametrically driven double well duffing oscillator:(a ) behavior of the lyapunov exponent(@xmath99 ) ( b ) behavior of the asymptotic growth rate ( @xmath4 ) from 0 - 1 test . here ,
@xmath113 ( after sampling the data points ) with @xmath85 as the length of each segment . ]
finally we discuss the intermittency route for sna , which differs from those formed through other mechanisms . in @xcite
the mechanism for the route in the logistic map has been studied .
it has been shown that in this route a sna disappears and is replaced by a one - frequency torus through an analogy with the saddle node bifurcation . in the logistic map [ eq . ( 11 ) ]
the intermittent transition from a sna to a torus occurs at @xmath124 with @xmath114 @xcite .
the intermittent transition from a sna to a torus in the cubic map(eq .
( 12 ) ) is studied in@xcite . for the forcing amplitude @xmath107
, a sna occurs at @xmath125 ; on increasing the value of @xmath65 , an intermittent transition occurs at @xmath126 which indicates a torus motion . + in fig.9(a )
we present the nature of the lyapunov exponent in the neighborhood of the transition point in the system parameter both for the logistic map ( fig.9a(i ) ) and cubic map ( fig.9b(i ) ) . at the intermittent transition
, the lyapunov exponent does not show any significant variation during the transition from snas to torus . on the other hand the 0 - 1
test detects intermittent snas and torus clearly in figs .
9a(ii ) and b(ii ) , where the asymptotic growth rate @xmath4 fluctuates in the intermittent sna region while it smoothly varies in the torus region .
the value of @xmath4 exhibits a sharp transition from a positive value ( nearly 0.8 ) to zero during the transition from snas to torus .
+ further we consider the intermittent transition in the parametrically driven double well duffing oscillator [ eq .
( 13 ) ] with @xmath127 , while the value of @xmath71 is varied . for @xmath128 ,
the system admits a chaotic attractor ; a sna appears at @xmath129 .
further when the value of @xmath71 is increasing , an intermittent transition from a sna to a torus occurs at @xmath130 @xcite . at this transition point
, the lyapunov exponent [ fig.10(a ) ] is of decreasing nature , where for both intermittent sna and torus the value of @xmath99 is negative .
the 0 - 1 test clearly detects changes in the growth rate as shown in fig.10(b ) .
( a ) variation of the lyapunov exponent @xmath99 and ( b ) variation of the asymptotic growth rate ( @xmath4 ) from 0 - 1 test as a function of @xmath62 indicating transition from snas to chaotic attractors in the quasiperiodically forced logistic map with @xmath131 .
here @xmath84 and @xmath85 . ] ( a ) varition of the lyapunov exponent @xmath99 and ( b ) variation of the asymptotic growth rate @xmath4 as a function of @xmath71 indicating transition from chaos to snas in the parametrically driven double well duffing oscillator with r=0.47 . here
@xmath113 ( after sampling the data points ) with @xmath85 . ]
in the previous sections we have characterized different routes for the transition between tori and snas , which are qualitatively distinguished in terms of the asymptotic growth rate @xmath4 by the 0 - 1 test . it has been shown that the value of @xmath4 shows an abrupt change followed by an irregular behavior in general after the birth of snas . in this section
we mainly concentrate on important dynamical characteristics of transition from snas to chaotic attractors or vice versa , usually characterized by the largest lyapunov exponent which changes from negative to positive values @xcite . in general snas occur in narrow parameter ranges , and finally get transferred into chaotic attractors .
the transition is not usually accompanied by any major change in the form or shape of the attractor .
we now study this type of transition in the quasiperiodically forced logistic map [ eq.(11 ) ] with parameter @xmath102 .
negi @xmath132
@xcite have found that the transition from snas to chaos occurs at @xmath133 .
11(a ) presents the lyapunov exponent(@xmath99 ) for the transition from snas to chaos which clearly shows that @xmath99 changes from negative to positive values . in the case of the 0 - 1 test
, the value of the asymptotic growth rate @xmath4 oscillates between 0.5 and 0.8 in the sna regime , and suddenly goes to 1 in the chaotic regime [ see fig.11(b ) ] .
similarly we have also carried out a study of the transition from chaos to snas in the parametrically driven duffing oscillator [ eq.(13 ) ] with @xmath134 while varying @xmath71 . for @xmath135 ,
the attractor is chaotic . as @xmath71
is increased to 0.0558 , one obtains a sna .
the 0 - 1 test shows a specific signature of transition from chaos to snas as shown in fig .
12(b ) whose lyapunov exponent is shown in fig .
in this work we have studied different routes to snas and transitions from tori to snas and then to chaotic attractors in prototypical examples , namely the quasiperiodically forced logistic map , cubic map and parametrically driven duffing oscillator . in these systems
we have studied the applicability of 0 - 1 test to distinguish snas from quasiperiodic motion and chaos , and have shown that the test clearly distinguishes snas from other attractors .
the present study has also been able to detect the strange nonchaotic dynamics in the heagy - hammel , fractalization and intermittecy routes , which are in general not easily detected using the lyapunov exponent .
we have pointed out that in each of the routes , the 0 - 1 test clearly detects an abrupt change in the asymptotic growth rate @xmath4 at critical parameter values during transitions . + in particular for the hh route
the value of @xmath4 is smooth in the torus region and at a critical value there is a sudden jump after which it exhibits an irregular behavior . in the fractalization route
, the value of @xmath4 grows linearly from torus to wt and becomes irregular at the transition to sna . for intermittency route
the value of @xmath4 exhibits a sharp transition from a fluctuating value ( nearly 0.8 ) to zero during the transition from snas to torus .
finally , our analysis shows that for quasiperiodic attractors the value of @xmath4 is zero , and if chaotic it is 1 , whereas for snas it takes values in between 0 and 1. + thus the tools of 0 - 1 test helps one to distinguish snas from other attractors in a clear manner .
further , we note that the advantages of 0 - 1 test are that it requires minimal time series data , does not require any equation of motion and is fairly easy to compute .
this test has some disadvantages in that it depends upon a properly chosen value of @xmath1 in eq.(2 ) . in this paper
we have used @xmath1 in terms of suitable irrational numbers to characterize quasiperiodically forced dynamical systems . finally , as it requires minimal time series , we believe that this test is more suitable for experimental situations in order to distinguish snas from other regular and irregular behaviors .
the work of r.g and m.l . has been supported by the department of science and technology ( dst ) , government of india sponsored irhpa research project .
is also supported by a department of atomic energy raja ramanna fellowship and a dst ramanna program .
acknowledges the support from ugc minor research project .
in this appendix we illustrate the method by which a suitable choice for the parameter @xmath1 in eqs.(1)-(2 ) of the 0 - 1 test can be made using the quasiperiodically forced logistic map ( 11 ) as a specific example .
the same analysis has been performed for the quasiperiodically driven cubic map and duffing oscillator . here
we present the nature of different dynamical attractors in the quasiperiodically forced logistic map , which has been studied in sec.iii.b , and examine the dynamics in terms of the translation variables @xmath48 for tori , snas and chaotic attractors for different values of @xmath1 chosen in terms of irrational numbers .
the results indicate that while many values of @xmath1 fail to distinguish quasiperiodic / sna / chaotic attractors , there exists a set of values which clearly distinguishes them as shown in figs.13 .
particularly , we find that for the choice @xmath1 = @xmath51 , where @xmath52 , ( see figs.13e ) very clear distinction between different attractors can be made .
note that this choice does not produce any resonance with the @xmath48 variables .
the same value works quite well in the cases of quasiperiodically driven cubic map and duffing oscillator .
so in our analysis we fix @xmath1 as @xmath1 = @xmath51 , where @xmath52 .
a second choice is @xmath136 , corresponding to fig.13(a ) , which we find to be reasonably satisfactory , though not as good as @xmath137 , where @xmath52 .
in order to investigate how our results behave as a function of the length @xmath5 of the time series , we discuss the dynamics of the translation variables @xmath48 as well as variation of the asymptotic growth rate @xmath4 as a function of @xmath5 for the quasiperiodically forced logistic map . we have plotted the variables @xmath48 in figs.14 for different types of attractors ( eq.(11 ) ) for different lengths @xmath5 of time series data . from the figure
, one can see first that the @xmath48 trajectories of the torus motion are bounded for @xmath138 ( fig.14.a(i ) ) . then , upon increasing @xmath5 the @xmath48 trajectories continue to be bounded ( figs .
14a(ii)-14.a(iv ) ) even upto @xmath84 .
next , we demonstrate in fig.15 that the asymptotic growth rate @xmath4 takes a value close to zero irrespective of the length of time series ( fig.15(a ) ) .
we also include the corresponding plot of the lyapunov exponent in fig.15(b ) for comparision .
we have also carried out similar analysis for the snas and chaotic attractors which are reported in figs .
14(b ) and 14(c ) respectively .
the corresponding asymptotic growth rate @xmath4 and lyapunov exponents are depicted in figs .
15(a ) and ( b ) respectively ( note that in fig.15(a ) , the value of @xmath4 is shown without taking averages over segments in order to depict its behavior as a function of @xmath5 ) .
the above results confirm that in the cases of sna and chaotic attractors also the dynamics remains qualitatively unchanged irrespective of the length of the time series .
finally , the results have also been confirmed for the quasiperiodically forced cubic map and duffing oscillator .
in this appendix , we present the characterization of different dynamical attractors in the quasiperiodically forced cubic map ( eq.(12 ) ) in terms of the translation variables @xmath48 and the mean square displacement @xmath28 using the 0 - 1 test with @xmath1 = @xmath51 , and @xmath52 .
the results are given in fig.16 .
again this analysis indicates that the 0 - 1 test allows one to distinguish snas from quasiperiodic and chaotic attractors as in the case of the quasiperiodically forced logistic map ( sec .
in this section we wish to present a statistical evaluation of our results by analyzing the distribution @xmath6 of local asymptotic growth rate of @xmath4 and comparing the same with the distribution @xmath139 of ftles for each one of the transitions from torus to sna . the local distribution of @xmath4 is calculated over short segements of the trajectories of the given dynamical system .
then , the distribution of local asymptotic growth rate of @xmath4 is defined similar to ftles[21 ] as we have calculated both the distributions of the local asymptotic growth rate @xmath142 and ftles@xmath143 in order to compare the effectiveness of the two descriptions . for each one of the routes to snas
, we have investigated the distributions of local @xmath4 and ftles , ( i ) for a torus prior to the transition to snas and ( ii ) for a sna just after the transition .
we first present the results for the quasiperiodically forced logistic map in fig.17 .
the local distribution of @xmath4 is given in figs.17(a ) , 17(c ) , and 17(e ) and the distribution of ftles is given in figs .
17(b ) , 17(d ) and 17(f ) corresponding to hh , fractalization , and intermittency routes , respectively .
+ from figures 17(a),(c ) and ( e ) , one can note that for all the transitions @xmath144 ) tends strongly to zero when the attractor is a torus but on the sna the distribution spreads between 0.1 and 0.9 .
in particular , the value of @xmath4 is distributed from 0.1 to 0.35 for the birth of sna through hh route .
the distribution of @xmath4 extends from 0.05 to 0.18 for the logistic map ( or 0.2 to 0.3 for the cubic map in fig .
18 ) for the genesis of sna through fractalization route . in the intermittent sna
the distribution spreads across the range of @xmath4 from 0.1 to 0.9 .
the distribution of different ranges of values of @xmath4 helps to distinguish the different transitions to snas .
it is instructive to compare the above results with the distribution of finite time lyapunov exponents as shown in figs.17(b),(d),(f ) .
one can note that the distribution of finite time lyapunov exponents have the same shape for hh and fractalization route and a slightly different one for the intermittent sna .
moreover , the distribution of @xmath139 for the both cases of torus and sna are almost identical except for the tails on the positive side of @xmath99 .
however , the distribution of local @xmath4 clearly distinguishes torus and sna as well as the three transitions . +
a similar behavior of the distribution of both local @xmath4 and ftles has been observed for the cubic map as shown in fig.18 . for the case of
quasiperiodically forced duffing oscillator also a similar picture arises . | we show that the recently introduced 0 - 1 test can successfully distinguish between strange nonchaotic attractors(snas ) and periodic / quasiperiodic / chaotic attractors , by suitably choosing the arbitrary parameter associated with the translation variables in terms of the golden mean number which avoids resonance with the quasiperiodic force .
we further characterize the transition from quasiperiodic to chaotic motion via snas in terms of the 0 - 1 test .
we demonstrate that the test helps to detect different dynamical transitions to snas from quasiperiodic attractor or the transitions from snas to chaos .
we illustrate the performance of the 0 - 1 test in detecting transitions to snas in quasiperiodically forced logistic map , cubic map , and duffing oscillator .
* a strange nonchaotic attractor(sna ) is considered as a complicated structure in phase space .
such a complex structure is a typical property usually associated with a chaotic attractor .
however , snas are nonchaotic in a dynamical sense because they do not exhibit sensitive dependence on initial conditions ( as evidenced by negative lyapunov exponents )
. these attractors are ubiquitous in different quasiperiodically driven nonlinear systems . while the existence of snas has been firmly established , a question of intense interest is how do these attractors morphologically differ from other attractors and how they can be characterized and quantified . two specific characterizations , namely finite time lyapunov exponents and recurrence plots were proposed earlier in the literature .
however , in the present work we show that the recently introduced measure of 0 - 1 test clearly distinguishes quasiperiodic motion , snas and chaotic attractors and also allows one to detect different dynamical transitions from quasiperiodic motion to snas and snas to chaotic attractors . * |
during the spectroscopic search for rotation in white dwarfs we noticed that all known zz ceti stars showed peculiar h@xmath0 line profiles ( koester et al .
the lines could be fitted using a rotationally broadened profile , but the derived rotation rates were incompatible with other data , e.g. from asteroseismology . using the caii k line for such a study , berger et al .
( 2005 ) found rotation velocities varying from 11 - 28 km / s from seven high - resolution keck and vlt spectra of the variable g29 - 38 a clear indication that the broadening could not be due to rotation .
while the spectra of almost all stars could be fitted with zero rotation , for g29 - 38 this was not true .
the observed spectra ( fig .
1 ) are much flatter than the non - rotating model .
since the rotation can certainly not change by a factor of three within a few years , a different mechanism has to be found .
another obvious possibility are surface motions due to the non - radial pulsations . indeed , velocity shifts with the same periods as the light variations have been observed in g29 - 38 ( van kerkwijk , clemens , & wu 2000 ; thompson et al . 2003 ) . in the latter study
, however , these were dismissed as a reason for the peculiar line profiles , since already the time - resolved spectra seemed to show the same profile as the time integrated spectrum .
this argument neglects the fact , that the velocity dispersion over the disk leads to a significant broadening already at a single phase , compared to a non - pulsating star . in this study we will test this conclusion by brute - force numerical integration of time - dependent spectra over the visible disk of the variable , taking into account temperature variations , accurate limb - darkening , doppler shifts from the pulsation , and the geometry of the star in the observers frame .
the radial displacement of a surface element is described by a spherical harmonic with mode indices @xmath1 and @xmath2 . the corresponding velocity vector is given by dziembowski ( 1977 ) .
for white dwarfs the radial component is negligible and the motions are nearly horizontal in the stellar system . as our test case we have chosen a @xmath3 , @xmath4 mode with a period of 600 s , and a pole - on view of the observer .
after transformation of the stellar surface velocity into the observers frame we obtain the visual light variation , the mean velocity over the visible disk ( which shifts the line ) , and the velocity dispersion ( which broadens the line at each phase ) .
2 shows these results .
figure 3 shows the broadening and shift of the ca line in three phases during the cycle .
position , width and depth of the profile change significantly .
this is the effect which would be observed with time - resolved spectroscopy .
figure 4 shows the comparison of three observed spectra with a profile of a non - pulsating star and with the integrated profile ( over one period , the exposure time of keck spectra is 1800 s ) for a pulsating star . with the assumed values for the observable velocity amplitude ( @xmath5 km / s )
the least extreme observation can be explained .
the other spectra would need slightly larger velocities .
the amplitudes we have used in this simulation for the velocity and light variations are about a factor of 2 higher than found for a single strong mode by thompson et al .
( 2003 ) . however , due to several modes present they find velocity amplitudes easily exceeding 10 km / s in the direct measurement of the total pulsation .
we are thus convinced that these motions are at the origin of the observed peculiar line profiles .
the study of these profiles - unfortunately requiring very large telescopes - is thus another tool for the study of the variations .
one recent example is our prediction , based on the line profile , of the variability of wd1150@xmath6153 , which has been confirmed photometrically in the meantime .
berger , l. , koester , d. , napiwotzki , r. , reid , i. n. , & zuckerman , b. 2005 , a&a , 444 , 565 dziembowski , w. 1977 , act .
astr . , 27 , 203 koester , d. , dreizler , s. , weidemann , v. , & allard , n. f. 1998 , a&a , 338 , 612 thompson , s. e. , clemens , j. c. , van kerkwijk , m. h. , & koester , d. 2003 , apj , 589 , 921 van kerkwijk , m. h. , clemens , j. c. , & wu , y. 2000 , mnras , 314 , 209 | we demonstrate that the peculiar line profiles observed in da white dwarfs in the temperature range of the zz ceti variables can be explained by the surface velocity fields associated with the pulsations . |
inclusive semileptonic decays of @xmath1 mesons to charmed final states provide an avenue for measuring the cabibbo - kobayashi - maskawa ( ckm ) matrix element @xmath5 @xcite and for determining non - perturbative hadronic properties of the @xmath1 meson . in particular ,
the moments of the hadronic mass in @xmath6 decays calculated in the framework of the operator product expansion ( ope ) and the heavy quark effective theory ( hqet ) @xcite depend on the @xmath7-quark mass ( @xmath8 ) and a few non - perturbative matrix elements that also appear in the expression of the total semileptonic width .
thus , measurements of the hadronic invariant mass moments @xcite allow the determination of these non - perturbative parameters from the data and reduce the theoretical uncertainty in the extraction of @xmath5 from measurements of the semileptonic branching fraction .
an improved knowledge of @xmath8 also results in a more precise determination of @xmath9 from inclusive charmless semileptonic @xmath1 decays .
this analysis uses @xmath10 events in which the hadronic decay of one @xmath1 meson is fully reconstructed .
the semileptonic decay of the other @xmath1 is inferred from the presence of an identified lepton ( electron or muon ) amongst the remaining particles in the event .
we calculate the first two moments of the hadronic invariant mass squared ( @xmath0 ) distribution @xcite directly from the measured spectrum after the effects of finite detector resolution have been removed using the singular value decomposition algorithm @xcite .
the measurement described in this paper improves the results previously reported by the babar and cleo collaborations @xcite .
the sensitivity to @xmath8 and other non - perturbative parameters is increased by lowering the minimum lepton energy threshold to 0.7 gev .
finally , this analysis minimizes the dependence on particular @xmath6 model assumptions by calculating the moments directly from the unfolded @xmath0 spectrum .
the data used in this analysis were taken with the belle detector @xcite at the kekb asymmetric energy @xmath11 collider @xcite .
belle is a large - solid - angle magnetic spectrometer that consists of a three - layer silicon vertex detector , a 50-layer central drift chamber ( cdc ) , an array of aerogel threshold erenkov counters ( acc ) , a barrel - like arrangement of time - of - flight scintillation counters ( tof ) , and an electromagnetic calorimeter comprised of csi(tl ) crystals ( ecl ) located inside a super - conducting solenoid coil that provides a 1.5 t magnetic field .
an iron flux - return located outside of the coil is instrumented to detect @xmath12 mesons and to identify muons ( klm ) .
the data sample consists of 140 fb@xmath3 taken near the @xmath4 resonance , or
@xmath13 @xmath14 events .
another 15 fb@xmath3 taken at 60 mev below the resonance are used to estimate the non-@xmath14 ( continuum ) background .
the off - resonance data is scaled by the integrated on- to off - resonance luminosity ratio corrected for the @xmath15 dependence of the @xmath16 cross - section .
a generic @xmath14 monte carlo ( mc ) sample equivalent to about three times the integrated luminosity is used in this analysis .
mc - simulated events are generated with evtgen @xcite and full detector simulation based on geant3 @xcite is applied .
the decays @xmath17 and @xmath18 are generated using an hqet inspired form factor parameterization @xcite .
the decays @xmath19 @xcite are simulated according to the leibovich - ligeti - stewart - wise ( llsw ) model @xcite ( both relative abundance and form factor shape ) .
the @xmath20 model also includes non - resonant @xmath21 decays which are generated using the goity - roberts model @xcite .
the model for the @xmath22 background is a hybrid mixture of exclusive modes and an inclusive component described by the de fazio - neubert model @xcite .
light - cone sum rule form factors @xcite are used for @xmath23 , @xmath24 and @xmath25 .
other exclusive modes are simulated according to the isgw2 model @xcite .
qed bremsstrahlung in @xmath26 decays is included using the photos package @xcite .
hadronic events are selected based on the charged track multiplicity and the visible energy in the calorimeter .
the selection is described in detail elsewhere @xcite .
we fully reconstruct the hadronic decay of one @xmath1 meson ( @xmath27 ) using the decay modes @xmath28 and @xmath29 @xcite .
pairs of photons satisfying @xmath30 mev in the laboratory - frame and 118 mev/@xmath31 mev/@xmath32 ( @xmath33 around the @xmath34 mass ) are combined to form @xmath34 candidates .
@xmath35 mesons are reconstructed from pairs of oppositely charged tracks with invariant mass within @xmath36 mev/@xmath32 ( @xmath37 ) of the nominal @xmath35 mass and a decay vertex displaced from the interaction point . candidate @xmath38 and @xmath39 mesons are reconstructed in the @xmath40 and @xmath41 decay modes , requiring their invariant masses to be within @xmath42 mev/@xmath32 of the nominal @xmath43 mass . candidate @xmath44 mesons are obtained by combining a @xmath39 candidate with a charged pion and requiring an invariant mass between 1.0 and 1.6 gev/@xmath32 . @xmath45
candidates are searched for in the @xmath46 , @xmath47 , @xmath48 , @xmath49 and @xmath50 decay modes .
the @xmath51 and @xmath52 modes are used to reconstruct @xmath53 mesons .
charmed mesons are selected in a window corresponding to @xmath54 times the mass resolution in the respective decay mode .
mesons are reconstructed by pairing a charmed meson with a low momentum pion , @xmath56 .
the decay modes @xmath57 and @xmath58 are used to search for neutral charmed vector mesons . for each @xmath27 candidate , the beam - energy constrained mass @xmath59 and the energy difference @xmath60 are calculated , @xmath61 where @xmath62 , @xmath63 and @xmath64 are the beam energy , the 3-momentum and the energy of the @xmath1 candidate in the @xmath4 frame . in @xmath59 and @xmath60 , the signal peaks at the nominal @xmath1 mass and zero , respectively .
we define the signal region by the selections @xmath65 gev/@xmath32 and @xmath66 gev . if multiple candidates are found in a single event , the best candidate is chosen based on the proximity of @xmath67 , @xmath68 and @xmath69 to their nominal values , where @xmath68 is the reconstructed @xmath70 meson mass and @xmath69 is the difference between the reconstructed @xmath71 and @xmath70 meson masses . without making any requirement on the decay of the other @xmath1 meson , the number of @xmath72 ( @xmath73 ) tags in this region , after subtraction of continuum and combinatorial backgrounds , is @xmath74 ( @xmath75 ) , fig . [ fig:1 ] .
distributions for charged and neutral @xmath27 candidates after requiring @xmath76 gev .
no constraints are made on the signal side .
the points with error bars are on - resonance data after subtraction of the scaled off - resonance data .
the combinatorial background ( cross - hatched histogram ) is estimated using mc simulation .
] semileptonic decays of the other @xmath1 meson ( @xmath77 ) are selected by searching for an identified charged lepton ( electron or muon ) within the remaining particles in the event .
electron candidates are identified using the ratio of the energy detected in the ecl to the track momentum , the ecl shower shape , position matching between track and ecl cluster , the energy loss in the cdc and the response of the acc counters .
muons are identified based on their penetration range and transverse scattering in the klm detector . in the momentum region
relevant to this analysis , charged leptons are identified with an efficiency of about 90% and the probability to misidentify a pion as an electron ( muon ) is 0.25% ( 1.4% ) @xcite .
we further require electron ( muon ) candidates to originate from near the interaction vertex , have a laboratory - frame momentum greater than 0.3 gev/@xmath78 ( 0.6 gev/@xmath78 ) and satisfy @xmath79 ( @xmath80 ) , where @xmath81 is the polar angle in the laboratory - frame relative to the beam direction .
if more than one charged lepton candidate is found in the event , we only keep the one with the highest momentum in the @xmath1 rest frame .
electrons from photon conversion are vetoed by rejecting the event if the invariant mass of the electron candidate and another oppositely charged particle in the event is below 0.04 gev/@xmath32 and secondary vertex criteria are satisfied .
if the charged lepton candidate is consistent with the decay @xmath82 ( _ i.e. _ , the invariant mass of the lepton candidate and another oppositely charged lepton in the event is between 3 gev/@xmath32 and 3.15 gev/@xmath32 ) , the event is also rejected . in @xmath72
tagged events , we require the lepton charge to be consistent with a prompt semileptonic decay of @xmath77 . in @xmath73 events , we make no requirement on the lepton charge . in electron events ,
we partially recover the effect of bremsstrahlung by searching for a photon with laboratory - frame energy @xmath83 gev within a 5@xmath84 cone around the electron direction at the interaction point .
if such a photon is found , it is merged with the electron and removed from the event .
the 4-momentum @xmath85 of the hadronic system @xmath86 recoiling against @xmath87 is determined by summing the 4-momenta of the remaining charged tracks and unmatched clusters in the event .
we exclude tracks passing very far away from the interaction point or compatible with a multiply reconstructed track generated by a low - momentum particle spiraling in the central drift chamber .
unmatched clusters in the barrel region must have an energy greater than 50 mev .
higher thresholds are applied in the endcap regions . to improve the resolution in @xmath0
, we reject events with a missing mass larger than 3 gev@xmath88/@xmath89 .
further improvement is obtained by recalculating the 4-momentum of the @xmath86 system , @xmath90 taking the neutrino 4-momentum @xmath91 to be @xmath92 , where @xmath93 is the missing 3-momentum . defined as the half width at half maximum
, the resolution in @xmath0 obtained from @xmath94 is about 0.8 gev@xmath95 , compared to 1.4 gev@xmath95 in @xmath0 from @xmath85 .
we consider the following contributions to the background in the @xmath0 spectrum : non-@xmath14 ( continuum ) background , combinatorial background , background from secondary or fake leptons and @xmath22 background .
combinatorial background are true @xmath96 events for which reconstruction or flavor assignment of the tagged @xmath1 meson is not correct .
the shapes of these background components in @xmath0 are determined from the mc simulation , except for the continuum where off - resonance data is used .
the shape of the fake muon background is corrected by the ratio of the pion fake rate in the experimental data over the same quantity in the mc simulation , as measured using kinematically identified pions in @xmath97 decays .
we derive the shape of the combinatorial background from the generic @xmath14 simulation by selecting events in which the reconstruction of @xmath27 does not correspond precisely to what was generated in the simulation .
the continuum background is scaled by the integrated on- to off - resonance luminosity ratio , taking into account the cross - section difference .
the mc - prediction of the combinatorial background is normalized to the data using the side - band region ( @xmath65 gev/@xmath32 and @xmath98 gev ) .
the normalization of the secondary or fake lepton background is found from the data by fitting the electron ( muon ) momentum distribution @xmath99 @xcite in the @xmath1 meson rest frame in the range from 0.3 to 2.4 gev/@xmath78 ( 0.6 to 2.4 gev/@xmath78 ) .
the @xmath100 component is normalized to the number of @xmath72 ( @xmath73 ) tags , assuming a branching fraction of @xmath101 ( @xmath102 ) for @xmath103 ( @xmath104 ) @xcite .
the background in the @xmath0 spectrum is estimated separately in the four sub - samples , defined by the charge of @xmath27 ( @xmath72 , @xmath73 ) and the lepton type ( electron , muon ) .
the purity of the @xmath6 signal depends on the sub - sample and the lepton energy threshold , typical values being around 75% .
table [ tab:1 ] shows the numbers of signal events and purities for each combination of @xmath27 charge , lepton type and lepton energy threshold .
c@cccc ' '' '' @xmath105 & @xmath72 electron & @xmath72 muon & @xmath73 electron & @xmath73 muon + ' '' '' 0.7 & @xmath106 ( 70.5% ) & @xmath107 ( 61.5% ) & @xmath108 ( 65.9% ) & @xmath109 ( 60.3% ) + 0.9 & @xmath110 ( 73.2% ) & @xmath111 ( 64.8% ) & @xmath112 ( 73.4% ) & @xmath113 ( 67.3% ) + 1.1 & @xmath114 ( 74.9% ) & @xmath115 ( 68.3% ) & @xmath116 ( 77.1% ) & @xmath117 ( 74.2% ) + 1.3 & @xmath118 ( 75.8% ) & @xmath119 ( 70.6% ) & @xmath120 ( 80.4% ) & @xmath121 ( 78.0% ) + 1.5 & @xmath122 ( 74.6% ) & @xmath123 ( 72.3% ) & @xmath124 ( 84.2% ) & @xmath125 ( 79.7% ) + 1.7 & @xmath126 ( 77.2% ) & @xmath127 ( 72.4% ) & @xmath128 ( 83.7% ) & @xmath129 ( 80.7% ) + ' '' '' 1.9 & @xmath130 ( 73.8% ) & @xmath131 ( 74.0% ) & @xmath132 ( 84.3% ) & @xmath133 ( 76.7% ) + we measure the @xmath0 spectrum in 45 bins in the range from 0 to 15 gev@xmath88/@xmath89 ( bin width 0.333 gev@xmath88/@xmath89 ) , which is shown in fig .
[ fig:2 ] , and unfold the finite detector resolution in this distribution using the singular value decomposition ( svd ) algorithm @xcite .
the unfolded @xmath0 spectrum has 15 bins in the range from @xmath134 to about 15 gev@xmath95 .
the bin width is 1 gev@xmath95 , except around the narrow states
@xmath70 , @xmath71 , @xmath135 and @xmath136 where smaller bin sizes are chosen .
signal and the different background components , explained in more detail in the text . ]
the unfolding is done separately in each sub - sample ( @xmath72 electron , @xmath72 muon , @xmath73 electron and @xmath73 muon ) . from the unfolded spectrum , we calculate the first moment and its statistical uncertainty squared , @xmath137 here , @xmath138 is the unfolded spectrum corrected for slightly different bin - to - bin efficiencies and @xmath86 is its covariance matrix , also determined by the svd algorithm .
@xmath139 is the central value of the @xmath140-th bin of the unfolded spectrum .
the second central and non - central moments , @xmath141 and @xmath142 are calculated from the same spectrum , substituting @xmath0 by @xmath143 and @xmath144 in eq .
[ eq:1 ] , respectively .
as the hadron mass moments are not expected to depend on the @xmath1 meson charge or the lepton type @xcite , we take the average over the four sub - sample results .
we have tested the entire measurement procedure including event reconstruction , unfolding and moment calculation on mc simulated events and no significant bias has been observed over the full range of lepton energy thresholds .
our measurements of @xmath145 , @xmath146 and @xmath142 for different lepton energy thresholds are shown in table [ tab:2 ] and fig .
[ fig:3 ] .
the sub - sample results for a given charge of @xmath27 ( @xmath72 , @xmath73 ) or lepton type ( electron , muon ) are compatible within their statistical uncertainty only
. c@ccc ' '' '' @xmath105 ( gev ) & @xmath147 ( gev@xmath88/@xmath89 ) & @xmath148 ( gev@xmath149/@xmath150 ) & @xmath142 ( gev@xmath149/@xmath150 ) + ' '' '' 0.7 & @xmath151 & @xmath152 & @xmath153 + 0.9 & @xmath154 & @xmath155 & @xmath156 + 1.1 & @xmath157 & @xmath158 & @xmath159 + 1.3 & @xmath160 & @xmath161 & @xmath162 + 1.5 & @xmath163 & @xmath164 & @xmath165 + 1.7 & @xmath166 & @xmath167 & @xmath168 + ' '' '' 1.9 & @xmath169 & @xmath170 & @xmath171 + .
the error bars indicate the statistical and total experimental errors . ]
the different contributions to the systematic error are shown in tables [ tab:3][tab:5 ] .
the total systematic error in table [ tab:2 ] corresponds to the quadratic sum of these components .
l@ccccccc ' '' '' & + ' '' '' @xmath105 ( gev ) & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' secondary / fake leptons & 0.033 & 0.023 & 0.013 & 0.007 & 0.004 & 0.002 &
0.000 + combinatorial background & 0.006 & 0.004 & 0.003 & 0.002 & 0.002 & 0.002 & 0.000 + continuum & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 + ' '' '' @xmath172 background & 0.004 & 0.004 & 0.004 & 0.004 & 0.006 & 0.007 & 0.009 + ' '' ''
@xmath173 & 0.008 & 0.007 & 0.007 & 0.007 & 0.006 & 0.005 & 0.003 + @xmath174 & 0.022 & 0.014 & 0.006 & 0.000 & 0.000 & 0.008 & 0.006 + @xmath175 & 0.024 & 0.017 & 0.007 & 0.004 & 0.004 & 0.004 & 0.004 + @xmath176 form factors & 0.013 & 0.013 & 0.012 & 0.011 & 0.010 & 0.008 & 0.006 + ' '' ''
@xmath177 form factors & 0.003 & 0.002 & 0.002 & 0.001 & 0.001 & 0.001 & 0.004 + ' '' '' unfolding & 0.015 & 0.015 & 0.015 & 0.015 & 0.015 & 0.015 & 0.015 + binning & 0.001 & 0.001 & 0.001 & 0.001 & 0.001 & 0.000 & 0.001 + ' '' '' efficiency & 0.008 & 0.011 & 0.012 & 0.009 & 0.008 & 0.005 & 0.004 + ' '' '' total & 0.052 & 0.041 & 0.029 & 0.024 & 0.022 & 0.022 & 0.021 + l@ccccccc ' '' '' & + ' '' '' @xmath105 ( gev ) & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' secondary / fake leptons & 0.167 & 0.109 & 0.050 & 0.023 & 0.009 & 0.005 &
0.002 + combinatorial background & 0.028 & 0.018 & 0.009 & 0.005 & 0.003 & 0.002 & 0.001 + continuum & 0.000 & 0.000 & 0.000 & 0.000 & 0.000 & 0.001 & 0.000 + ' '' '' @xmath172 background & 0.004 & 0.004 & 0.004 & 0.003 & 0.003 & 0.005 & 0.005 + ' '' ''
@xmath173 & 0.013 & 0.010 & 0.007 & 0.004 & 0.002 & 0.002 & 0.003 + @xmath174 & 0.216 & 0.169 & 0.102 & 0.049 & 0.042 & 0.011 & 0.009 + @xmath175 & 0.168 & 0.125 & 0.058 & 0.041 & 0.024 & 0.004 & 0.004 + @xmath176 form factors & 0.029 & 0.028 & 0.024 & 0.019 & 0.017 & 0.016 & 0.007 + ' '' ''
@xmath177 form factors & 0.013 & 0.009 & 0.006 & 0.003 & 0.004 & 0.001 & 0.004 + ' '' '' unfolding & 0.035 & 0.035 & 0.035 & 0.035 & 0.035 & 0.035 & 0.035 + binning & 0.001 & 0.001 & 0.000 & 0.001 & 0.001 & 0.001 & 0.002 + ' '' '' efficiency & 0.025 & 0.032 & 0.027 & 0.014 & 0.013 & 0.005 & 0.002 + ' '' '' total & 0.327 & 0.244 & 0.137 & 0.080 & 0.064 & 0.040 & 0.038 + l@ccccccc ' '' '' & + ' '' '' @xmath105 ( gev ) & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' secondary / fake leptons & 0.46 & 0.31 & 0.16 & 0.09 & 0.04 & 0.02 & 0.00 + combinatorial background & 0.08 & 0.05 & 0.04 & 0.03 & 0.01 & 0.01 & 0.00 + continuum & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 + ' '' '' @xmath172 background & 0.04 & 0.04 & 0.04 & 0.04 & 0.05 & 0.06 & 0.07 + ' '' '' @xmath173 & 0.07 & 0.07 & 0.06 & 0.05 & 0.04 & 0.03 & 0.02 + @xmath174 & 0.41 & 0.30 & 0.14 & 0.04 & 0.04 & 0.05 & 0.06 + @xmath175 & 0.38 & 0.28 & 0.12 & 0.08 & 0.06 & 0.04 & 0.04 + @xmath176 form factors & 0.15 & 0.14 & 0.13 & 0.12 & 0.10 & 0.09 & 0.04 + ' '' '' @xmath177 form factors & 0.04 & 0.03 & 0.02 & 0.01 & 0.01 & 0.00 & 0.03 + ' '' '' unfolding & 0.16 & 0.16 & 0.16 & 0.16 & 0.16 & 0.16 & 0.16 + binning & 0.01 & 0.01 & 0.01 & 0.00 & 0.00 & 0.01 & 0.00 + ' '' '' efficiency & 0.09 & 0.13 & 0.13 & 0.09 & 0.08 & 0.04 & 0.03 + ' '' '' total & 0.77 & 0.58 & 0.36 & 0.26 & 0.23 & 0.21 & 0.20 + the uncertainties related to the different background components in @xmath0 are estimated by varying the respective background normalization factors within @xmath178 standard deviation .
we consider both variations of the @xmath179 branching fractions and form factor shapes .
for the former , the ranges of variation are taken from ref .
for the latter , the curvature @xmath180 in the form factor parametrization @xcite is varied within @xmath181 ( @xmath182 ) for @xmath17 ( @xmath183 ) @xcite . for @xmath17
, we also vary the form factor ratios @xmath184 and @xmath185 @xcite .
the llsw model @xcite predicts the relative abundance and the form factor shape of the different components in @xmath19 only . to obtain the absolute branching fractions of the @xmath186 components and of @xmath187
, we use @xmath188 @xcite , the recent belle measurement of @xmath189 @xcite and the total semileptonic branching fraction @xcite . the uncertainty assigned to the @xmath186 branching fractions in tables [ tab:3][tab:5 ]
reflects the uncertainty in these measurements and the change in the @xmath186 composition when varying the llsw parameters within their allowed range . the svd algorithm used to unfold the measured @xmath0 distribution requires the detector response matrix , _ i.e. _ , the distribution of measured versus true values of @xmath0 .
we determine this matrix from the mc simulation .
to study the systematics related to unfolding and a possible mismodeling of the detector response , we change the amount of bin - to - bin migration by varying the effective rank of the detector response matrix , the main tunable parameter of the svd algorithm .
we have further studied a change of the binning of the unfolded distribution and the effect of disabling the bin - to - bin efficiency correction . due to overlapping events , the moment measurements corresponding to different lepton energy thresholds
are highly correlated .
systematic uncertainties are another source of correlation .
we have estimated the correlations due to both sources using a toy mc approach based on 50,000 simulated measurements .
the results for the self- and cross - correlation coefficients are given in tables [ tab:6][tab:10 ] .
c@c|ccccccc & + & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' & 0.7 & 1.000 & 0.932 & 0.786 & 0.615 & 0.481 & 0.168 & 0.071 + & 0.9 & & 1.000 & 0.888 & 0.715 & 0.573 & 0.241 & 0.116 + & 1.1 & & & 1.000 & 0.849 & 0.693 & 0.363 & 0.194 + @xmath145 & 1.3 & & & & 1.000 & 0.804 & 0.470 & 0.254 + & 1.5 & & & & & 1.000 & 0.591 & 0.308 + & 1.7 & & & & & & 1.000 & 0.363 + ' '' '' & 1.9 & & & & & & & 1.000 + c@c|ccccccc & + & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' & 0.7 & 0.897 & 0.847 & 0.788 & 0.713 & 0.576 & 0.306 & 0.102 + & 0.9 & 0.777 & 0.843 & 0.804 & 0.726 & 0.608 & 0.356 & 0.144 + & 1.1 & 0.548 & 0.615 & 0.757 & 0.690 & 0.606 & 0.426 & 0.211 + @xmath145 & 1.3 & 0.328 & 0.371 & 0.483 & 0.718 & 0.599 & 0.476 & 0.260 + & 1.5 & 0.223 & 0.263 & 0.346 & 0.481 & 0.702 & 0.559 & 0.280 + & 1.7 & @xmath190 & @xmath191 & 0.035 & 0.126 & 0.237 & 0.846 & 0.296 + ' '' '' & 1.9 & @xmath192 & @xmath193 & @xmath194 & 0.040 & 0.075 & 0.228 & 0.865 + c@c|ccccccc & + & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' & 0.7 & 0.983 & 0.933 & 0.830 & 0.683 & 0.523 & 0.194 & 0.073 + & 0.9 & 0.890 & 0.974 & 0.910 & 0.765 & 0.606 & 0.264 & 0.117 + & 1.1 & 0.704 & 0.810 & 0.976 & 0.857 & 0.707 & 0.380 & 0.196 + @xmath145 & 1.3 & 0.508 & 0.601 & 0.774 & 0.980 & 0.800 & 0.479 & 0.258 + & 1.5 & 0.383 & 0.469 & 0.614 & 0.764 & 0.985 & 0.597 & 0.305 + & 1.7 & 0.079 & 0.137 & 0.273 & 0.403 & 0.539 & 0.994 & 0.357 + ' '' '' & 1.9 & 0.017 & 0.052 & 0.136 & 0.208 & 0.271 & 0.348 & 0.995 + c@c|ccccccc & + & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' & 0.7 & 1.000 & 0.939 & 0.838 & 0.698 & 0.534 & 0.167 & @xmath195 + & 0.9 & & 1.000 & 0.901 & 0.732 & 0.586 & 0.195 & @xmath196 + @xmath197 & 1.1 & & & 1.000 & 0.793 & 0.638 & 0.262 & 0.034 + @xmath198 & 1.3 & & & & 1.000 & 0.731 & 0.340 & 0.102 + & 1.5 & & & & & 1.000 & 0.484 & 0.146 + & 1.7 & & & & & & 1.000 & 0.296
+ ' '' '' & 1.9 & & & & & & & 1.000 + c@c|ccccccc & + & 0.7 & 0.9 & 1.1 & 1.3 & 1.5 & 1.7 & 1.9 + ' '' '' & 0.7 & 1.000 & 0.932 & 0.784 & 0.601 & 0.442 & 0.111 & 0.017 + & 0.9 & & 1.000 & 0.877 & 0.684 & 0.524 & 0.168 & 0.051 + & 1.1 & & & 1.000 & 0.817 & 0.651 & 0.297 & 0.137 + @xmath142 & 1.3 & & & & 1.000 & 0.780 & 0.421 & 0.212 + & 1.5 & & & & & 1.000 & 0.557 & 0.270 + & 1.7 & & & & & & 1.000 & 0.346 + ' '' '' & 1.9 & & & & & & & 1.000 +
we have measured the first , @xmath145 , and the second central and non - central moments , @xmath199 and @xmath142 , of the hadronic mass squared spectrum in @xmath6 decays for lepton energy thresholds ranging from 0.7 to 1.9 gev . using a toy mc approach ,
we have also evaluated the full covariance matrix for this set of measurements .
it is expected that this measurement , combined with measurements of the semileptonic branching fraction , moments of the lepton energy spectrum in @xmath6 decays and possibly other moments , will lead to an improved determination of @xmath7-quark mass @xmath8 and the ckm matrix element @xmath5 @xcite .
we thank the kekb group for the excellent operation of the accelerator , the kek cryogenics group for the efficient operation of the solenoid , and the kek computer group and the national institute of informatics for valuable computing and super - sinet network support .
we acknowledge support from the ministry of education , culture , sports , science , and technology of japan and the japan society for the promotion of science ; the australian research council and the australian department of education , science and training ; the national science foundation of china and the knowledge innovation program of the chinese academy of sciences under contract no .
10575109 and ihep - u-503 ; the department of science and technology of india ; the bk21 program of the ministry of education of korea , the chep src program and basic research program ( grant no . r01 - 2005 - 000 - 10089 - 0 ) of the korea science and engineering foundation , the pure basic research group program of the korea research foundation , and the sbs foundation ; the polish state committee for scientific research ; the ministry of science and technology of the russian federation ; the slovenian research agency ; the swiss national science foundation ; the national science council and the ministry of education of taiwan ; and the u.s.department of energy .
n. cabibbo , phys .
* 10 * , 531 ( 1963 ) ; m. kobayashi and t. maskawa , prog .
phys . * 49 * , 652 ( 1973 ) .
d. benson , i. i. bigi , t. mannel and n. uraltsev , nucl .
b * 665 * , 367 ( 2003 ) [ hep - ph/0302262 ] .
gambino and n. uraltsev , eur .
j. c * 34 * , 181 ( 2004 ) [ hep - ph/0401063 ] .
d. benson , i. i. bigi and n. uraltsev , nucl .
b * 710 * , 371 ( 2005 ) [ hep - ph/0410080 ] . c. w. bauer , z. ligeti , m. luke and a. v. manohar , phys .
d * 67 * , 054012 ( 2003 ) [ hep - ph/0210027 ] .
b. aubert _ et al .
_ [ babar collaboration ] , phys .
d * 69 * , 111103 ( 2004 ) [ hep - ex/0403031 ] .
s. e. csorna _ et al .
_ [ cleo collaboration ] , phys . rev .
d * 70 * , 032002 ( 2004 ) [ hep - ex/0403052 ] .
d. acosta _ et al .
_ [ cdf collaboration ] , phys . rev .
d * 71 * , 051103 ( 2005 ) [ hep - ex/0502003 ] .
j. abdallah _ et al .
_ [ delphi collaboration ] , eur .
j. c * 45 * , 35 ( 2006 ) [ hep - ex/0510024 ] .
we provide numbers for both definitions of the second moment , @xmath141 and @xmath142 , used in literature @xcite for the convenience of further applications though one can be derived from the other .
a. hcker and v. kartvelishvili , nucl .
instrum .
a * 372 * , 469 ( 1996 ) [ hep - ph/9509307 ] .
a. abashian _ et al . _ [ belle collaboration ] , nucl .
instrum .
meth . a * 479 * , 117 ( 2002 ) .
s. kurokawa , nucl .
instrum .
a * 499 * , 1 ( 2003 ) , and other papers included in this volume .
d. j. lange , nucl .
instrum .
a * 462 * , 152 ( 2001 ) .
r. brun , f. bruyant , m. maire , a. c. mcpherson and p. zanarini , cern - dd / ee/84 - 1 .
i. caprini , l. lellouch and m. neubert , nucl .
b * 530 * , 153 ( 1998 ) [ hep - ph/9712417 ] . in this paper ,
the symbol @xmath200 refers collectively to the @xmath135 , @xmath136 , @xmath201 and @xmath202 states .
a. k. leibovich , z. ligeti , i. w. stewart and m. b. wise , phys .
d * 57 * , 308 ( 1998 ) [ hep - ph/9705467 ] .
j. l. goity and w. roberts , phys .
d * 51 * , 3459 ( 1995 ) [ hep - ph/9406236 ] .
f. de fazio and m. neubert , jhep * 9906 * , 017 ( 1999 ) [ hep - ph/9905351 ] .
p. ball and v. m. braun , phys .
d * 58 * , 094016 ( 1998 ) [ hep - ph/9805422 ] .
p. ball and r. zwicky , jhep * 0110 * , 019 ( 2001 ) [ hep - ph/0110115 ] .
d. scora and n. isgur , phys .
d * 52 * , 2783 ( 1995 ) [ hep - ph/9503486 ] .
e. barberio and z. was , comput .
commun .
* 79 * , 291 ( 1994 ) .
k. abe _ et al . _ [ belle collaboration ] , phys .
d * 64 * , 072001 ( 2001 ) [ hep - ex/0103041 ] . in this paper , the inclusion of the charge conjugate decay is implied .
k. hanagaki , h. kakuno , h. ikeda , t. iijima and t. tsukamoto , nucl . instrum .
a * 485 * , 490 ( 2002 ) [ hep - ex/0108044 ] .
a. abashian _ et al .
_ , nucl .
instrum .
meth . a * 491 * , 69 ( 2002 ) . in this paper , quantities calculated in the @xmath1 meson rest frame are denoted by an asterisk .
w. m. yao _ et al .
_ [ particle data group ] , j. phys .
g * 33 * , 1 ( 2006 ) .
e. barberio _ et al .
_ [ heavy flavor averaging group ( hfag ) ] , hep - ex/0603003 . b. aubert _ et al . _ [ babar collaboration ] , hep - ex/0602023 .
d. liventsev _ et al .
_ , phys .
d * 72 * , 051109 ( 2005 ) . | we present a measurement of the hadronic invariant mass squared ( @xmath0 ) spectrum in charmed semileptonic @xmath1 meson decays @xmath2 based on 140 fb@xmath3 of belle data collected near the @xmath4 resonance .
we determine the first , the second central and the second non - central moments of this spectrum for lepton energy thresholds ranging between 0.7 and 1.9 gev .
full correlations between these measurements are evaluated . |
in this data driven area , the amount and complexity of the available data grows at an almost incredible speed .
therefore , there is a high need to develop novel tools to cope with such complex data structures .
whereas the first statistical techniques were designed only to manage either quantitative or qualitative data , we can now find statistical procedures to handle functional data ( see for instance arribas - gil and mller @xcite ; febrero - bande and gonzlez - manteiga @xcite ; jacques and preda @xcite ) , fuzzy - valued data ( see , for instance , ferraro and giordani @xcite ; gonzlez - rodrguez _ et al . _
@xcite ; coppi _ et al . _
@xcite ) ; incomplete / missing data ( see , for instance , bianco _ et al . _
@xcite ; ferraty _ et al . _
@xcite ; lin @xcite ; zhao _ et al . _
@xcite ) , and several other types of data .
interval - valued data are a type of complex data that requires specific statistical techniques to analyze them .
interval - valued data may arise for different reasons . in some cases
the underlying random variable is intrinsically interval - valued , e.g. the daily fluctuation of the systolic blood pressure . in other cases , there is an underlying real - valued but to preserve a level of confidentiality respondents are only asked to indicate the interval containing their value , e.g. their salary .
it may also happen that the real - valued measurement is only partially known due to certain limitations , such as is the case for interval censored data .
finally , aggregation of a typically large dataset may lead to e.g. interval - valued symbolic data which include interval variation and structure .
the @xmath0-median considered here does not make any assumption about the source of the interval - valued data .
in particular , it does not matter whether the random experiment that generated the data involves an underlying observable real - valued random variable or not .
an important remark is that the space of intervals is only a semilinear space , but not a linear space due to the lack of the opposite of an interval .
therefore , although intervals can be identified with two - dimensional vectors ( with first component the mid - point / centre and second component the nonnegative spread / radius ) , it is not advisable to treat them as regular bivariate data . indeed ,
common assumptions for multivariate techniques do not hold in this case .
statistical procedures for random interval - valued data have already been proposed in the literature for different purposes , such as regression analysis ( e.g gil _ et al . _
@xcite ; gonzlez - rodrguez _ et al . _
@xcite ; blanco - fernndez _ et al . _
@xcite ; lima neto _
et al . _
@xcite ; fagundes _ et al . _
@xcite ; giordani @xcite ) ; testing hypotheses ( e.g. montenegro _ et al . _
@xcite ; nakama _ et al . _
@xcite ; gonzlez - rodrguez _ et al . _
@xcite ) , clustering ( e.g. de carvalho _ et al . _
@xcite ; durso _ et al . _ @xcite ; giusti and grassini @xcite ; da costa _ et al . _
@xcite , etc . ) , principal component analysis ( e.g. billard and diday @xcite ; durso and giordani @xcite ; makosso - kallyth and diday @xcite , etc . ) , modelling distributions ( see brito and duarte silva @xcite ; sun and ralescu @xcite ) .
one of the most commonly used location measures is the aumann - type mean ( see aumann @xcite ) .
it is indeed supported by numerous valuable properties , including laws of large numbers , and is also coherent with the interval arithmetic .
the main disadvantage is that it is strongly influenced by outliers and data changes , which makes this measure not always suitable as a summary measure of the distribution of a random interval .
this drawback is in fact inherited from the standard real / vectorial - valued case . in the real case , the most popular alternative is the median . in the real case , the most popular robust alternative to the mean is the median
. for multivariate data the spatial median ( also called the @xmath1-median , as introduced by weber @xcite ) is a popular robust alternative to estimate the center of the multivariate data .
the spatial median is defined as the point in multivariate space with minimal average euclidean distance to the observations . for more details and extensions ,
see for instance gower ( @xcite ) , brown ( @xcite ) , milasevic and ducharme ( @xcite ) , ( cadre @xcite ) , roelant and van aelst ( @xcite ) , debruyne _ et al .
_ ( @xcite ) , fritz _ et al . _ ( @xcite ) , zuo ( @xcite ) .
sinova and van aelst ( 2014 ) adapted the spatial median to interval - valued data by using a suitable @xmath2 metric on this space ( see also sinova et al .
they used the versatile generalized metric introduced by bertoluzza _
( @xcite , see also gil _ et al . _
@xcite ; trutschnig _ et al . _
@xcite ) the resulting @xmath0-median estimator has been shown to be robust with high breakdown point and good finite - sample properties . in this paper
we show another important property of the estimator , which is its strong consistency . the rest of this paper is organized as follows : in section 2
the basic concepts related to the interval - valued space , interval arithmetic and metric for intervals will be introduced , as well as the usual location measure . in section 3 ,
the @xmath0-median for random intervals and its main properties are recalled .
the strong consistency of the @xmath0-median is proven in section 4 .
finally , some concluding remarks are presented in section 5 .
let @xmath3 denote the class of nonempty compact intervals .
any interval @xmath4 in the space @xmath5 can be characterized in terms of either its infimum and supremum , @xmath6 $ ] , or its mid - point and spread or radius , @xmath7 $ ] , where @xmath8 the usual interval arithmetic provides the addition , i.e. @xmath9 $ ] with @xmath10 and the product by a scalar , i.e. @xmath11 $ ] with @xmath12 and @xmath13 . with these two operations the space @xmath3 is semilinear , but
not linear due to the lack of a difference of intervals . therefore
, statistical techniques for interval - valued data are based on distances . to measure the distance between two interval - valued observations , we consider the _
@xmath0 metric _ introduced by bertoluzza _
et al . _ ( @xcite ) , which can be defined as ( see gil _ et al . _
@xcite ) : @xmath14 where @xmath15 and @xmath16 . following the general random
set approach , a _ random interval _ can usually defined as a borel measurable mapping @xmath17 , where @xmath18 is a probability space with respect to @xmath19 and on @xmath5 the borel @xmath20-field generated by the topology induced by the @xmath0 metric . as a consequence from the borel measurability , crucial concepts in probabilistic and inferential developments , such as the ( induced ) distribution of a random interval or the stochastic independence of random intervals , are well - defined .
one of the most used location measures is the _ aumann - type mean value_. it is defined , if it exists , as the interval
@xmath21=[e(\inf x),e(\sup x)]$ ] or @xmath21=[e({\hbox { \rm mid}}\ , x)-e({\hbox { \rm spr}}\ , x),e({\hbox { \rm mid}}\ , x)+e({\hbox { \rm spr}}\ , x)]$ ] ( both expressions are equivalent ) .
moreover , it is the frchet expectation with respect to the @xmath0 metric , i.e. , it is the unique interval that minimizes , over @xmath12 , the expression @xmath22 $ ] . as a robust alternative to the aumann - type mean , sinova and van aelst ( 2014 ) proposed the @xmath0-median as measure of location , which is defined as follows .
@xmath23-median(s ) _ of a random interval @xmath24 is(are ) the interval(s ) @xmath25\in \mathcal k_c(\mathbb r)$ ] such that @xmath26 ) ) = \min_{k\in\mathcal k_c(\mathbb r)}e(d_{\theta}(x , k)),\ ] ] whenever the involved expectations exist .
analogously , the sample @xmath0-median statistic is defined as follows .
let @xmath27 be a simple random sample from a random interval @xmath24 with realizations @xmath28 .
the _ sample @xmath23-median _ ( or medians ) @xmath29}_n$ ] is ( are ) the random interval that takes , for @xmath30 , the interval value(s ) @xmath31}$ ] that is ( are ) the solution(s ) of the following optimization problem : @xmath32 where @xmath4 , @xmath33 and @xmath34 depend on @xmath30 ( which has been omitted from the notation for the sake of simplicity ) and the fixed value @xmath35 .
sinova and van aelst ( 2014 ) showed the existence of the sample @xmath0-median estimator and its uniqueness whenever not all the two - dimensional sample points @xmath36 are collinear .
moreover , the robustness was shown by its finite sample breakdown point ( donoho and huber @xcite ) which is given by @xmath37}_n,\mathbf{x}_n , d_\theta\big)=\frac{1}{n}\cdot\lfloor\frac{n+1}{2}\rfloor,\vspace{-0.25cm}\ ] ] where @xmath38 denotes the floor function .
in this section we investigate the strong consistency of the sample @xmath0-median under general conditions . [ consistency ]
let @xmath39 be a random interval associated with a probability space @xmath40 such that the @xmath0-median exists and is unique .
then , the sample @xmath0-median is a strongly consistent estimator of the @xmath0-median , that is , @xmath41_n},m_\theta[x])=0
\quad \text{a.s .
} [ p].\ ] ] _ proof .
_ sufficient conditions for the strong consistency of an estimator are given in huber ( @xcite ) .
we will check that these conditions , detailed below , are satisfied in our case : * the parameter set ( @xmath42 in our case , with the topology induced by the @xmath0-metric ) is a locally compact space with a countable base and @xmath18 is a probability space .
let @xmath43 be the following real - valued function on @xmath44 : @xmath45)}. \end{array}\ ] ] * assuming that @xmath46 are independent @xmath47-valued random elements with + common probability distribution @xmath48 , the sequence of functions @xmath49 , defined as @xmath50}_n$ ] , satisfies that almost surely ( obviously because of the definition of the sample @xmath0-median ) .
_ assumption ( a-1 ) _ for each fixed @xmath51 , the function @xmath52)}\\[0.8ex ] & & & & \displaystyle{=\sqrt{({\rm mid}\ , x(\omega)-y_0)^2+\theta \cdot ( { \rm spr}\ , x(\omega)-z_0)^2}}\vspace{-0.1 cm } \end{array}\ ] ] is @xmath19-measurable and separable in doob s sense : there is a p - null set @xmath53 and a countable subset @xmath54 such that for every open set @xmath55 and every closed interval @xmath56 , the sets @xmath57 @xmath58 differ by at most a subset of @xmath53 .
_ assumption ( a-2 ) _ the function @xmath59 is a.s .
lower semicontinuous in @xmath60 , that is , @xmath61 as the neighborhood @xmath62 of @xmath60 shrinks to @xmath63 .
_ assumption ( a-3 ) _ there is a measurable function @xmath64 such that @xmath65 ^ -<\infty \quad \text { for all } ( y , z)\in \mathbb r \times [ 0,\infty),\vspace{-0.2cm}\ ] ] @xmath65^+<\infty \quad \text { for some } ( y , z)\in \mathbb r \times [ 0,\infty).\ ] ] thus , @xmath66 $ ] is well - defined for all @xmath67 . _ assumption ( a-4 ) _ there is a @xmath68 such that @xmath69 @xmath70 for all @xmath71 _ assumption ( a-5 ) _ there is a continuous function @xmath72 such that * for some integrable @xmath73 , @xmath74 * the following condition is satisfied : @xmath75 * it is also fulfilled that : @xmath76\geq 1.\ ] ] we now verify these conditions of huber : _ ( a-1 ) _ for each fixed @xmath51 , the function @xmath77 is @xmath19-measurable ( because @xmath78 and @xmath79 are measurable functions since @xmath39 is a random interval ) and separable in doob s sense : choosing @xmath80 as countable subset , for every open set @xmath81 and every closed interval a , it will be seen that the sets @xmath82 coincide . obviously , @xmath83 . by _
reductio ad absurdum _ , it is now supposed that @xmath84 .
let @xmath85 : * since @xmath86 , @xmath87 for all @xmath88 ; * since @xmath89 , there exists @xmath90 such that @xmath91 .
@xmath92 is an open set , so there exists a ball of radius @xmath93 such that @xmath94 notice now that , for a fixed @xmath95 , the function @xmath96 is continuous .
therefore , @xmath97 is an open set of @xmath42 and @xmath98 too .
@xmath99 is a dense set of @xmath42 , so @xmath100 let @xmath101 . then , @xmath102 , so @xmath103 . but also , @xmath104 this is a contradiction , so the conclusion is that @xmath105 . _
( a-2 ) _ indeed , it will be proved for all @xmath95 .
let @xmath106 be any element of @xmath47 and let @xmath60 be any ( fixed ) point of @xmath42 .
first , notice that it is fulfilled for a sequence of neighborhoods @xmath107 of @xmath60 when @xmath108 for all @xmath109 that @xmath110)\right\}_{n\in \mathbb n}\ ] ] is a monotonically increasing sequence .
furthermore , this sequence is bounded since @xmath111 ) \leq d_\theta(x(\omega),[y_0-z_0,y_0+z_0])\ ] ] for all @xmath112 because @xmath113 .
therefore , the sequence converges to its supremum , which will be @xmath114)$ ] . by _ reductio ad absurdum _
, suppose that there is a smaller upper bound @xmath115)-\varepsilon,\ ] ] for an arbitrary @xmath116 .
let s denote by @xmath117 a neighborhood of @xmath60 satisfying that @xmath118 .
then , it can be seen that @xmath119),\ ] ] so @xmath120 can not be the supremum .
thus , using the triangular inequality , @xmath121)\geq \underset{(y , z)\in b((y_0,z_0),\frac{\varepsilon}{2})}{\inf}d_\theta(x(\omega),[y - z , y+z])\ ] ] @xmath122)-d_\theta([y - z , y+z],[y_0-z_0,y_0+z_0])\right]\ ] ] @xmath123)-\underset{(y , z)\in b((y_0,z_0),\frac{\varepsilon}{2})}{\sup}d_\theta([y - z , y+z],[y_0-z_0,y_0+z_0])\ ] ] @xmath124)-\varepsilon = c.\ ] ] now this result will be extended to general sequences @xmath125 .
consider the suprema and the infima radii reached in every neighborhood , namely , @xmath126,[y - z , y+z]),\ ] ] @xmath127,[y - z , y+z]).\ ] ] it is known that @xmath128 , since @xmath125 shrinks to @xmath63 .
moreover , @xmath129 as @xmath130 for all @xmath112 .
let @xmath131 be any nonnegative number .
as @xmath132 , there exists @xmath133 such that for all @xmath134 , @xmath135 .
then , @xmath136 and @xmath111)\geq \underset{(y , z)\in b((y_0,z_0),r_n)}{\inf}d_\theta(x(\omega),[y - z , y+z])\ ] ] @xmath137)-\underset{(y , z)\in b((y_0,z_0),r_n)}{\sup}d_\theta([y_0-z_0,y_0+z_0],[y - z , y+z])\ ] ] @xmath138)-\varepsilon.\ ] ] analogously , as @xmath139 , there exists @xmath140 such that for all @xmath141 , @xmath142 .
therefore , @xmath143 and @xmath111 ) \leq \underset{(y , z)\in b((y_0,z_0),s_n)}{\inf}d_\theta(x(\omega),[y - z , y+z])\ ] ] @xmath144)+\underset{(y , z)\in b((y_0,z_0),s_n)}{\inf}d_\theta([y - z , y+z],[y_0-z_0,y_0+z_0])\ ] ] @xmath145)+\varepsilon.\ ] ] so for any @xmath116 , there exists @xmath146 , such that for all @xmath147 , @xmath148)-\varepsilon < \underset{(y , z)\in u_n}{\inf}d_\theta(x(\omega),[y - z , y+z])\ ] ] @xmath145)+\varepsilon,\ ] ] that is to say , @xmath149)-d_\theta(x(\omega),[y_0-z_0,y_0+z_0])\right|<\varepsilon,\ ] ] so the sequence @xmath150)\right\}_{n\in \mathbb n}$ ] converges to @xmath151).$ ] _ ( a-3 ) _ let @xmath152 be the measurable function ( see ( a-1 ) ) : @xmath153)=\sqrt{({\rm mid}\ , x(\omega))^2+\theta \cdot ( { \rm spr}\ , x(\omega))^2.}}\vspace{-0.1 cm } \end{array}\ ] ] fixed any arbitrary @xmath154 , @xmath65
^ -\]]@xmath155)-d_\theta(x(\omega),[0,0]),0\}\ , dp(\omega)\ ] ] @xmath156)\\>d_\theta(x(\omega),[y - z , y+z])\}\end{array}}}\big[d_\theta(x(\omega),[0,0])-d_\theta(x(\omega),[y - z , y+z])\big ] dp(\omega).\ ] ] by the triangular inequality , @xmath157)\\>d_\theta(x(\omega),[y - z , y+z])\}\end{array}}}\big[d_\theta(x(\omega),[y - z , y+z])+d_\theta([y - z , y+z],[0,0])\]]@xmath158)\big ] dp(\omega)\ ] ] @xmath159,[0,0])\cdot p\big(\omega : d_\theta(x(\omega),[0,0])>d_\theta(x(\omega),[y - z , y+z])\big ) < \infty.\ ] ] analogously , @xmath65^+\]]@xmath160)-d_\theta(x(\omega),[0,0]),0\}\ , dp(\omega)\ ] ] @xmath156)\\\leq d_\theta(x(\omega),[y - z , y+z])\}\end{array}}}\big[d_\theta(x(\omega),[y - z , y+z])-d_\theta(x(\omega),[0,0])\big ] \ , dp(\omega).\ ] ] by the triangular inequality , @xmath157)\\ \leq d_\theta(x(\omega),[y - z , y+z])\}\end{array}}}\big[d_\theta(x(\omega),[0,0])+d_\theta([0,0],[y - z , y+z])\]]@xmath161)\big ] dp(\omega)\ ] ] @xmath162,[y - z , y+z])\cdot p\big(\omega : d_\theta(x(\omega),[0,0])\leq d_\theta(x(\omega),[y - z , y+z])\big ) < \infty.\ ] ] so the second inequality also holds for all @xmath163 in this case . _
( a-4 ) _ the @xmath0-median exists and is unique , so that @xmath164 , \hbox { \rm spr}\ , m_\theta[x])=\arg \underset{(y , z)\in \mathbb r\times [ 0,\infty)}{\min}e\left[d_\theta(x(\omega),[y - z , y+z])\right]\ ] ] @xmath165)\right]-e\left[d_\theta(x(\omega),[0,0])\right]\]]@xmath165)-d_\theta(x(\omega),[0,0])\right]\]]@xmath166 and @xmath167 , { \hbox { \rm spr}}\ , m_\theta[x])$ ] fulfills this assumption .
there is a continuous function @xmath72 @xmath168,[0,0])+1}\vspace{-0.1 cm } \end{array}\ ] ] such that * for the integrable function @xmath169 , @xmath170)-d_\theta(x(\omega),[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\geq -1\ ] ] because using the triangular inequality , @xmath170)-d_\theta(x(\omega),[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\ ] ] @xmath171)-d_\theta([y - z , y+z],[0,0])-d_\theta(x(\omega),[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\ ] ] @xmath172,[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\geq -1.\ ] ] * the following condition is satisfied : @xmath75 let @xmath173 be any sequence with @xmath174 ( i.e. , @xmath175,[0,0])\underset{n}{\longrightarrow}\infty$ ] ) and @xmath176)-d_\theta(x(\omega),[0,0])\right]=\gamma((y_0,z_0))\in \mathbb r,\ ] ] where @xmath60 represents the minimum found in ( a-4 ) .
then , there exists @xmath177 such that for all @xmath178 , @xmath179,[0,0])>m.\ ] ] so , for all @xmath178 , @xmath180,[0,0])+1\right)\geq m+1.\ ] ] finally , @xmath181 * it is also fulfilled that : @xmath182)-d_\theta(x(\omega),[0,0])}{b((y , z))}\right]\geq 1.\ ] ] let s see that @xmath183)-d_\theta(x(\omega),[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\geq 1,\ ] ] so the result follows .
@xmath183)-d_\theta(x(\omega),[0,0])}{d_\theta([y - z , y+z],[0,0])+1}\ ] ] @xmath184)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right)\ ] ] for any fixed @xmath185 .
the sequence @xmath186)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right\}_{n\in\mathbb n}\ ] ] is monotonically increasing and is upper bounded by @xmath187 : for all @xmath188 , using the triangular inequality , @xmath189)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\ ] ] @xmath190,[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\leq 1.\ ] ] + so it converges to its supremum : @xmath191)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right)\ ] ] @xmath192)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right)\ ] ] let s finally see that this supremum is at least equal to @xmath187 . by
_ reductio ad absurdum _ ,
let s suppose that @xmath193)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right)=1-\varepsilon,\ ] ] for some @xmath116 .
one gets then a contradiction because one finds an @xmath194 such that @xmath195)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}>1-\varepsilon\ ] ] since for all @xmath196 , @xmath189)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\geq 1-\frac{\varepsilon}{2}>1-\varepsilon\ ] ] as we will show now .
recall that @xmath174 , so for all @xmath197 , there exists @xmath194 such that for all @xmath198 , @xmath175,[0,0])>m$ ] .
therefore , @xmath179,x(\omega))\geq d_\theta([y_n - z_n , y_n+z_n],[0,0])-d_\theta(x(\omega),[0,0])\]]@xmath199).\ ] ] taking @xmath200)\in \mathbb r$ ] ( for the fixed arbitrary @xmath185 ) , we can easily check that @xmath201 is a lower bound of the sequence @xmath202)-d_\theta(x(\omega),[0,0])}{d_\theta([y_k - z_k , y_k+z_k],[0,0])+1}\right\}_{k\geq n^*}.\ ] ] + for any @xmath196 , @xmath203)-d_\theta(x(\omega),[0,0])\ ] ] @xmath204)+\frac{\varepsilon}{2}d_\theta(x(\omega),[y_k - z_k , y_k+z_k])\ ] ] @xmath161)\ ] ] @xmath205,[0,0])-\left(1-\frac{\varepsilon}{2}\right)d_\theta(x(\omega),[0,0])\ ] ] @xmath206)-d_\theta(x(\omega),[0,0])\ ] ] @xmath207,[0,0])+\frac{\varepsilon}{2}d_\theta(x(\omega),[y_k - z_k , y_k+z_k])\ ] ] @xmath208)\ ] ] @xmath209,[0,0])+\frac{\varepsilon}{2}\left(\frac{2}{\varepsilon}-1+\big(\frac{4}{\varepsilon}-1\big)d_\theta(x(\omega),[0,0])\right)\ ] ] @xmath208)=\left(1-\frac{\varepsilon}{2}\right)d_\theta([y_k - z_k , y_k+z_k],[0,0])+1-\frac{\varepsilon}{2}\ ] ] @xmath210,[0,0])+1\big).\hspace{2.4cm}\square\ ] ]
this paper complements the study of the properties of the @xmath23-median as a robust estimator of the center of a random interval by showing its strong consistency which is one of the most important basic properties of an estimator .
we obtained this result by showing that all the sufficient conditions of huber ( @xcite ) are fulfilled .
these results open the door to further develop robust statistical inference for random intervals based on the @xmath23-median such as the development of hypotheses testing procedures .
authors are grateful to mara ngeles gil for her helpful suggestions to improve this paper .
the research by beatriz sinova was partially supported by / benefited from the spanish ministry of science and innovation grant mtm2009 - 09440-c02 - 01 .
she has been also granted with the ayuda del programa de fpu ap2009 - 1197 from the spanish ministry of education and the ayuda para estancias breves del programa fpu est12/00344 , an ayuda de investigacin 2011 from the fundacin banco herrero and three short term scientific missions associated with the cost action ic0702 .
the research by stefan van aelst was supported by a grant of the fund for scientific research - flanders ( fwo - vlaanderen ) and by iap research network grant nr .
p7/06 of the belgian government ( belgian science policy ) .
their financial support is gratefully acknowledged .
arribas - gil a , mller h - g ( 2014 ) pairwise dynamic time warping for event data .
comput stat data anal 69:255268 aumann rj ( 1965 ) integrals of set - valued functions .
j math anal appl 12:112 bertoluzza c , corral n , salas a ( 1995 ) on a new class of distances between fuzzy numbers .
math & soft comput 2:7184 bianco am , boente g , rodrigues i m ( 2013 ) robust tests in generalized linear models with missing responses .
comput stat data anal 65:8097 billard l , diday e ( 2003 ) from the statistics of data to the statistics of knowledge : symbolic data analysis .
j am stat ass 98:470487 blanco - fernndez a , corral n , gonzlez - rodrguez g ( 2011 ) estimation of a flexible simple linear model for interval data based on set arithmetic .
comput stat data anal 55(9):25682578 blanco - fernndez a , colubi a , garca - brzana m ( 2013 ) a set arithmetic - based linear regression model for modelling interval - valued responses through real - valued variables . inform sci 247:109122 brito p , duarte silva ap ( 2012 ) modelling interval data with normal and skew - normal distributions .
j appl stat 39(1):320 brown bm ( 1983 ) statistical uses of the spatial median .
j royal stat soc ser b 45(1):2530 cadre b ( 2001 ) convergent estimators for the @xmath1-median of a banach valued random variable .
statistics 35:509521 de carvalho fdat , brito p , bock hh ( 2006 ) dynamic clustering for interval data based on @xmath211 distance .
comput stat 21(2):231250 coppi r , durso p , giordani p ( 2012 ) fuzzy and possibilistic clustering for fuzzy data .
comput stat data anal 56(4):915927 da costa afbf , pimentel ba , de souza rmcr ( 2013 ) clustering interval data through kernel - induced feature space .
j intell inf syst 40(1):101140 debruyne m , hubert m , van horebeek j ( 2010 ) detecting influential observations in kernel pca .
comput stat data anal 54(12):30073019 donoho dl , huber pj ( 1983 ) the notion of breakdown point . in : bickel pj , doksum k , hodges jr jl ( eds . ) a festschrift for erich l. lehmann .
wadsworth , belmont .
milasevic p , ducharme gr ( 1987 ) uniqueness of the spatial median .
ann statist 15:13321333 durso p , giordani p ( 2004 ) a least squares approach to principal component analysis for interval valued data .
chemometr intell lab 70(2):179192 durso p , giordani p ( 2006 ) a robust fuzzy k - means clustering model for interval valued data .
comput stat 21(2):251269 durso p , de giovanni l ( 2011 ) midpoint radius self - organizing maps for interval - valued data with telecommunications application .
appl soft comput 11(5):38773886 durso p , de giovanni l , massari r ( 2014 ) trimmed fuzzy clustering for interval - valued data .
adv data anal classif , in press ( doi:10.1007/s11634 - 014 - 0169 - 3 ) fagundes raa , de souza rmcr , cysneiros fja ( 2013 ) robust regression with application to symbolic interval data .
eng appl art intel 26(1):564573 febrero - bande m , gonzlez - manteiga w ( 2013 ) generalized additive models for functional data .
test 22(2):278292 ferraro mb , giordani p ( 2013 ) on possibilistic clustering with repulsion constraints for imprecise data .
inform sci 245:6375 ferraty f , sued m , vieu p ( 2013 ) mean estimation with data missing at random for functional covariables .
statistics 47(4):688706 fritz h , filzmoser p , croux c ( 2012 ) a comparison of algorithms for the multivariate @xmath1-median .
comput stat 27(3):393410 gil ma , lubiano ma , montenegro m , lpez - garca mt ( 2002 ) least squares fitting of an affine function and strength of association for interval data .
metrika 56:97111 gil ma , gonzlez - rodrguez g , colubi a , montenegro m ( 2007 ) testing linear independence in linear models with interval - valued data .
comput stat data anal 51(6):30023015 giordani p ( 2014 ) lasso - constrained regression analysis for interval - valued data .
adv data anal classif , in press ( doi:10.1007/s11634 - 014 - 0164 - 8 ) giusti a , grassini l ( 2008 ) cluster analysis of census data using the symbolic data approach .
adv data anal classif 2(2):163176 gonzlez - rodrguez g , blanco a , corral n , colubi a ( 2007 ) least squares estimation of linear regression models for convex compact random sets . adv data anal classif 1(1):6781 gonzlez - rodrguez g , colubi a , gil ma ( 2012 ) fuzzy data treated as functional data : a one - way anova test approach .
comput stat data anal 56:943955 gower jc ( 1974 ) algorithm as 78 : the mediancentre .
appl statist 23:466470 huber pj ( 1967 ) the behavior of maximum likelihood estimates under nonstandard conditions . in : proceedings of the fifth berkeley symposium on mathematical statistics and probability 1 , pp .
221233 ishibuchi h , tanaka h ( 1990 ) multiobjective programming in optimization of the interval objective function .
europ j oper res 48:219225 jacques j , preda c ( 2014 ) model - based clustering for multivariate functional data .
comput stat data anal 71:92106 lima neto ea , cordeiro gm , de carvalho fat ( 2011 ) bivariate symbolic regression models for interval - valued variables .
j stat comput simul 81(11):17271744 lin th ( 2014 ) model selection information criteria in latent class models with missing data and contingency question .
j stat comput simul 84(1):159170 liu ry ( 1990 ) . on a notion of data depth based on random simplices .
ann statist 18:405414 makosso - kallyth s , diday e ( 2012 ) adaptation of interval pca to symbolic histogram variables .
adv data anal classif 6(2):147159 montenegro m , casals mr , colubi a , gil ma ( 2008 ) testing two - sided hypothesis about the mean of an interval - valued random set . in : dubois d , lubiano ma , prade h , gil ma , grzegorzewski p , hryniewicz o ( eds . ) soft methods for handling variability and imprecision .
springer , heidelberg , pp .
133139 nakama t , colubi a , lubiano ma ( 2010 ) two - way analysis of variance for interval - valued data . in : borgelt c , gonzlez - rodrguez g , trutschnig w , lubiano ma , gil ma , grzegorzewski p , hryniewicz o ( eds . ) combining soft computing and statistical methods in data analysis .
springer , heidelberg , pp .
475482 roelant e , van aelst s ( 2007 ) an l1-type estimator of multivariate location .
stat meth & appl 15:381393 rousseeuw pj , ruts i ( 1996 ) bivariate location depth
c 45:516526 rousseeuw pj , ruts i ( 1998 ) constructing the bivariate tukey median .
statistica sinica 8:827839 sinova b , gil ma , colubi a , van aelst s ( 2012 ) the median of a random fuzzy number . the 1-norm distance approach .
fuzzy sets syst 200:99115 sinova b , gonzlez - rodrguez g , van aelst s ( 2013 ) an alternative approach to the median of a random interval using an @xmath2 metric . in : kruse r
, berthold mr , moewes c , gil ma , grzegorzewski p , hryniewicz o ( eds . ) sinergies of soft computing and statistics for intelligent data analysis .
springer , heidelberg , pp .
273281 sun y , ralescu da ( 2014 ) a normal hierarchical model and minimum contrast estimation for random intervals .
ann inst stat math , in press ( doi:10.1007/s10463 - 014 - 0453 - 1 ) trutschnig w , gonzlez - rodrguez g , colubi a , gil ma ( 2009 ) a new family of metrics for compact , convex ( fuzzy ) sets based on a generalized concept of mids and spread .
inf sci 179:39643972 tukey jw ( 1975 ) mathematics and the picturing of data . in : proc .
international congress of mathematicians , vancouver , 1994 , 2 , pp .
523531 vitale ra ( 1985 ) @xmath212 metrics for compact , convex sets .
j approx theory 45:280287 weber a ( 1909 ) ber den standort der industrien .
mohr , tbingen .
zhao p - y , tang
m - l , tang
n - s ( 2013 ) robust estimation of distribution functions and quantiles with non - ignorable missing data .
canad j stat 41(4):575595 zuo y ( 2013 ) multidimensional medians and uniqueness .
comput stat data anal 66:8288 | the sample @xmath0-median is a robust estimator of the central tendency or location of an interval - valued random variable .
while the interval - valued sample mean can be highly influenced by outliers , this spatial - type interval - valued median remains much more reliable . in this paper , we show that under general conditions the sample @xmath0-median is a strongly consistent estimator of the @xmath0-median of an interval - valued random variable . |
the theory of mixed type equations is one the principal parts of the general theory of partial differential equations .
the interest for these kinds of equations arises intensively because of both theoretical and practical of their applications .
many mathematical models of applied problems require investigations of this type of equations .
the actuality of the consideration of mixed type equations has been mentioned , for the first time , by s. a. chaplygin in 1902 in his famous work `` on gas streams '' @xcite .
the first fundamental results in this direction was obtained in 1920 - 1930 by f. tricomi @xcite and s. gellerstedt @xcite .
the works of m. a. lavrentev @xcite , a. v. bitsadze @xcite , @xcite , f. i. frankl @xcite , m. protter @xcite and c. morawetz @xcite , have had a great impact in this theory , where outstanding theoretical results were obtained and pointed out important practical values of them .
bibliography of the main fundamental results on this direction can be found , among others , in the monographs of a. v. bitsadze @xcite , y. m. berezansky @xcite , l. bers @xcite , m. s. salakhitdinov and a. k. urinov @xcite and a. m. nakhushev @xcite . in most of the works devoted to the study of mixed type equations , the object of study was mixed elliptic - hyperbolic type equations .
comparatively , few results have been obtained on the study of mixed parabolic - hyperbolic type equations .
however , this last type of equations have also numerous applications in the real life processes ( see @xcite for an interesting example in mechanics ) .
the reader can found a nice example given , for the first time , by gelfand in @xcite , and connected with the movement of the gas in a channel surrounded by a porous environment . inside the channel the movement of gas was described by the wave equation and outside by the diffusion one .
mathematic models of this kind of problems arise in the study of electromagnetic fields , in heterogeneous environment , consisting of dielectric and conductive environment for modeling the movement of a little compressible fluid in a channel surrounded by a porous medium @xcite . here
the wave equation describes the hydrodynamic pressure of the fluid in the channel , and the equation of filtration - pressure fluid in a porous medium .
similar problems arise in the study of the magnetic intensity of the electromagnetic field @xcite . in the last few years
, the investigations on local boundary value problems , for mixed equations in domains with non - characteristic boundary data , were intensively increased .
we point out that the studies made on boundary value problems for equations of mixed type , in domains with deviation from the characteristics ( with non - characteristic boundary ) , have been originated with the fundamental works of bitsadze @xcite , where the generalized tricomi problem ( problem m ) for an equation of mixed type is discussed . in the works @xcite and @xcite ,
the analog to the tricomi problem for a modeled parabolic - hyperbolic equation , was investigated in a domain with non - characteristic boundary in a hyperbolic part . moreover ,
the uniqueness of solution and the volterra property of the formulated problem was proved .
we also refer to the recent works devoted to the study of parabolic - hyperbolic equations @xcite-@xcite . in the last years
, the interest for considering boundary value problems of parabolic - hyperbolic type , with integral gluing condition on the line of type changing , is increasing @xcite , @xcite . in the present work we study the analog to the generalized tricomi problem with integral gluing condition on the line of type changing .
we prove that the formulated problem has the volterra property .
the obtained result generalizes some previous ones from m. a. sadybekov and g. d. tajzhanova given in @xcite .
let @xmath0 be a domain , bounded at @xmath1 by segments @xmath2 of straight lines @xmath3 respectively , and at @xmath4 by a monotone smooth curve @xmath5 and by the segment @xmath6 , which is the characteristic curve of the equation @xmath7 where @xmath8 now we state the problem that we will consider along the paper : * problem b. * to find a solution of the eq.(1 ) , satisfying boundary conditions @xmath9 @xmath10 and gluing conditions @xmath11 where @xmath12 is a given function such that @xmath13 \times \left [ { 0,1 } \right ] } \right)$ ] , and @xmath14 satisfy @xmath15 .
when the curve @xmath16 coincides with the characteristic one @xmath17 , @xmath18 and @xmath19 , the problem b is just the tricomi problem for parabolic - hyperbolic equation with non - characteristic line of type changing , which has been studied in @xcite .
regular solvability of the problem b with continuous gluing conditions ( @xmath20 ) have been proved , for the first time , in @xcite , and strong solvability of this problem was proved in the work @xcite .
several properties , including the volterra property of boundary problems for mixed parabolic - hyperbolic equations , have been studied in the works @xcite-@xcite .
we denote the parabolic part of the mixed domain @xmath21 as @xmath22 and the hyperbolic part by @xmath23 .
a regular solution of the problem b in the domain @xmath21 will be a function @xmath24 that satisfies eq.(1 ) in the domains @xmath22 and @xmath23 , the boundary conditions ( 3)-(4 ) , and the gluing condition ( 5 ) . regarding the curve @xmath16 , we assume that @xmath25 is monotonically increasing . then , rewriting it by using the characteristic variables @xmath26 and
@xmath27 , we have that the equation of the curve @xmath16 can be expressed as @xmath28 .
* theorem 1 . *
let @xmath29 $ ] and @xmath13 \times \left [ { 0,1 } \right ] } \right)$ ] .
then for any function @xmath30 , there exists a unique regular solution of the problem b. ` proof : ` by a regular solution of the problem b in the domain @xmath31 we look for a function that fulfills the following expression @xmath32 - \int\limits_\xi ^\eta { d{\xi _ 1}\int\limits_{{\xi _ 1}}^\eta { { f_1}({\xi _ 1},\,{\eta _ 1 } ) } d{\eta _ 1}},\eqno ( 6)\ ] ] where @xmath33 based on ( 4 ) from ( 6 ) , using the expressions on ( 7 ) , we deduce that @xmath34 by virtue of the unique solvability of the first boundary problem for the heat equation ( 1 ) satisfying condition ( 3 ) , and the fact that @xmath35 , its solution can be represented as @xmath36 where @xmath37 and @xmath38 is the green s function related to the first boundary problem , for the heat equation in a rectangle @xmath39 , which has the form @xcite @xmath40}.\eqno ( 10)\ ] ] calculating the derivative @xmath41 in ( 9 ) and passing to the limit at @xmath42 we get @xmath43 where @xmath44 and @xmath45 thus , the main functional relation between @xmath46 and @xmath47 , reduced to the segment @xmath48 from the parabolic part of the domain , imply that @xmath49 suppose , in a first moment , that @xmath50 . from ( 8) and ( 13 ) , considering the gluing condition ( 5 ) , we obtain the following integral equation regarding the function @xmath46 : @xmath51 here @xmath52,\eqno ( 15)\ ] ] and @xmath53 hence , the problem b is equivalent , in the sense of unique solvability , to the second kind volterra integral equation ( 14 ) . the restrictions imposed on the functions @xmath54 , @xmath12 , and the right hand of the eq.(1 ) guarantee that , by virtue of ( 11 ) and ( 15 ) , the kernel @xmath55 is a kernel with weak singularity .
so , we have that eq.(14 ) has a unique solution and @xmath56 . since @xmath57
, we deduce the uniqueness of the function @xmath58 .
eq.(8 ) gives us the uniqueness of function @xmath59 and , as consequence , we deduce , from eq.(6 ) , the uniqueness of solution of problem b when @xmath50 .
consider now the other case , i.e. @xmath60 and @xmath61 . from functional relations ( 8) and ( 13 ) , and taking gluing condition ( 5 ) into account at @xmath60 , we have @xmath62q(x , t)dt}\ ] ] or , which is the same , @xmath63dt } = { f_0}(x ) + 2\beta \int\limits_0^x { dt\int\limits_{\lambda ( t)}^t { q(x , t){f_1}({\xi _ 1},t)d{\xi _ 1 } } .}\ ] ] considering the representation of @xmath64 , the previous equation can be rewritten as follows @xmath65.\eqno ( 17)\ ] ] since eq.(17 )
is the abel s equation , it can be solved and so we arrive at the following identity : @xmath66dz } } \right\}. \hfill \\\end{aligned}\ ] ] considering @xmath67 , after some simplifications we get @xmath68 where @xmath69dt } \right\},\ ] ] @xmath70f({x_1},{y_1})d{y_1 } } \hfill \\ & & \hspace{11 cm } ( 19)\\ & & + \frac{{2\beta } } { { \sqrt \pi } } \int\limits_0^x { d{\eta _ 1 } } \int\limits_{\lambda ( { \eta _ 1})}^{{\eta _ 1 } } { \left [ { \frac{{q({\eta _ 1},{\eta _ 1})}}{{\sqrt { x - { \eta _ 1 } } } } + \int\limits_0^{x - { \eta _ 1 } } { \frac{{{q_t}(t,{\eta _ 1})}}{{\sqrt { x - { \eta _ 1 } - t } } } dt } } \right ] } { f_1}({\xi _ 1},{\eta _ 1})d{\xi _ 1}. \hfill \\\end{aligned}\ ] ] since the kernel @xmath71 has a weak singularity , then eq.(18 ) has a unique solution , and it can be represented as @xmath72 where @xmath73 is the resolvent kernel of ( 18 ) . as consequence , arguing as in the case @xmath74 , we deduce , from eq.(6 ) , the uniqueness of solution of problem b for @xmath75 and @xmath76 , and the result is proved . in the sequel
, we will deduce the exact expression of the integral kernel related to the unique solution of problem b. to this end , we suppose , at the beginning , that @xmath74 .
note that the unique solution of eq.(14 ) can be represented as @xmath77 where @xmath78 is the resolvent kernel of the eq.(14 ) , and it is given by the recurrence formula : @xmath79 from ( 21 ) , taking @xmath37 into account , we have that @xmath80 where @xmath81 . from the formula ( 6 ) , and considering ( 8) , one can easily deduce that @xmath82 substituting the representation of @xmath83 into ( 22 ) and considering ( 12 ) and ( 16 ) , after some evaluations we get @xmath84 where @xmath85 and @xmath86 in an analogous way , substituting the representation of @xmath83 into ( 9 ) , we have @xmath87 where @xmath88 and @xmath89 from ( 23 ) and ( 24 ) , we arrive at the following expression @xmath90 where @xmath91 } \right\ } + \theta ( - y)\left\ { { \theta ( { y_1})\theta ( \xi - { x_1}){g_1}(\xi - { x_1},{y_1 } ) } \right . \hfill \\ & & + \left .
\theta ( - { y_1})\left [ { \frac{1}{2}\theta ( \eta - { \eta _ 1})\theta ( { \eta _ 1 } - \xi ) \theta ( \xi - { \xi _ 1 } ) } \right . { + \left .
{ \theta ( \xi - { \eta _ 1})\left [ { { \gamma _ 1}(\xi , { \eta _ 1 } ) + \frac{\beta } { \alpha } { g_0}(\xi - { \eta _ 1},{\eta _ 1 } ) } \right ] } \right ] } \right\}. \hfill\end{aligned}\ ] ] here @xmath92 when @xmath75 and @xmath76 , by using a similar algorithm , we conclude that @xmath93 where @xmath94 } \right .
\hfill \\ & & \left .
{ + \theta ( - { y_1})\theta ( x - { \eta _ 1}){g_5}(x - { \eta _ 1},y,{\eta _ 1 } ) } \right\ } + \theta ( - y)\left\ { { \frac{{}}{{}}\theta ( { y_1})\theta ( \xi - { x_1}){g_3}(\xi , { x_1},{y_1 } ) } \right .
\hfill \\ & & \left .
{ + \theta ( - { y_1})\left [ { \theta ( \xi - { \eta _ 1}){q_1}(\xi , \,{\eta _ 1 } ) + \frac{1}{2}\theta ( \eta - { \eta _ 1})\theta ( { \eta _ 1 } - \xi ) \theta ( \xi - { \xi _ 1 } ) } \right ] } \right\ } , \hfill\end{aligned}\ ] ] with @xmath95 } \right\}ds } \right]dz , \hfill\end{aligned}\ ] ] @xmath96 } \right\}ds } \right]dz , \hfill \\\end{aligned}\ ] ] @xmath97 and @xmath98 thus we have partially proved the following lemma .
* lemma 1 . * the unique regular solution of problem b can be represented as follows @xmath99 where @xmath100 and @xmath101 @xmath102 ` proof ` : expression ( 25 ) has been proved before .
let s see that @xmath103 .
note that in the kernel defined in ( 25 ) , all the items are bounded except the first one .
so , we only need to prove that @xmath104 from the representation of the green s function @xmath105 given in ( 10 ) , it follows that , for the aforementioned aim , it is enough to prove that ( for @xmath106 ) : @xmath107\ ] ] is bounded .
first , note that @xmath108 using this fact , we deduce that @xmath109 by means of the change of variables @xmath110 , we get that this last expression is less than or equals to the following one @xmath111 as consequence , @xmath103 and lemma 1 is completely proved .
define now @xmath112 we have the following regularity result for this function : * lemma 2 . * if @xmath113 and @xmath13 \times \left [ { 0,1 } \right ] } \right)$ ] , then @xmath114 $ ] and @xmath115 . `
proof : ` using the explicit form of the green s function given in ( 10 ) , it is not complicated to prove that function @xmath116 , defined by formulas ( 16 ) and ( 19 ) , belongs to the class of functions @xmath117 $ ] and @xmath115 . + lemma 2 is proved .
* lemma 3 .
* suppose that @xmath13 \times \left [ { 0,1 } \right ] } \right)$ ] and @xmath118 , then @xmath119 and @xmath120 ` proof : ` consider the following problem in @xmath121 : @xmath122 obviously , we have that @xmath123 . first , note that it is known @xcite that problem ( 27 ) has a unique solution @xmath124 , and it satisfies the following inequality @xmath125 using now the obvious equality @xmath126 we have that @xmath127.\eqno ( 29)\ ] ] from ( 28 ) and ( 29 ) we obtain @xmath128 now , by virtue of the conditions of lemma 3 and the representations ( 16 ) and ( 19 ) , from expression ( 30 ) and the cauchy - bunjakovskii inequalities , we get the estimate ( 26 ) and conclude the proof .
denote now @xmath129 as the norm of the sobolev space @xmath130 with @xmath131 .
* lemma 4 .
* let @xmath132 be the unique regular solution of problem b. then the following estimate holds : @xmath133 here @xmath134 is a positive constant that does not depend on @xmath132 . `
proof : ` by virtue of lemma 3 , and from ( 20 ) and ( 21 ) , we deduce that @xmath135 the result follows from expression ( 22 ) . *
definition 1*. we define the set @xmath136 as the set of all the regular solutions of problem b. a function @xmath137 is said to be a strong solution of problem b , if there exists a functional sequence @xmath138 , such that @xmath139 and @xmath140 converge in @xmath141 to @xmath132 and @xmath142 respectively .
define @xmath143 as the closure of the differential operator @xmath144 , given by expression ( 2 ) .
note that , according to the definition of the strong solution , the function @xmath132 will be a strong solution of problem b if and only if @xmath145 .
now we are in a position to prove the following uniqueness result for strong solutions .
* theorem 2 . * for any function @xmath13 \times \left [ { 0,1 } \right ] } \right)$ ] and @xmath118 , there exists a unique strong solution @xmath132 of problem b. moreover @xmath146 , satisfies inequality ( 31 ) and it is given by the expression ( 25 ) . `
proof : ` let @xmath147 be the set of the @xmath148 functions that vanish in a neighborhood of @xmath149 ( @xmath149 is a boundary of the domain @xmath21 ) . since @xmath147 is dense in @xmath141 , we have that for any function @xmath118 , there exist a functional sequence @xmath150 , such that @xmath151 , as @xmath152 .
it is not difficult to verify that if @xmath153 then @xmath154)$ ] ( with obvious notation )
. therefore equations ( 14 ) and ( 18 ) can be considered as a second kind volterra integral equations in the space @xmath155)$ ] .
consequently , we have that @xmath156 $ ] . due to the properties of the solutions of the boundary value problem for the heat equation in @xmath22 and the darboux problem , by using the representations ( 6 ) and ( 9 )
, we conclude that @xmath157 for all @xmath158 . by virtue of the inequality ( 31 )
we get @xmath159 consequently , @xmath160 is a sequence of strong solutions , hence , problem b is strongly solvable for all right hand @xmath161 , and the strong solution belongs to the space @xmath162 .
thus , theorem 2 is proved .
consider now , for all @xmath163 , the sequence of kernels given by the recurrence formula @xmath164 with @xmath165 and @xmath166 defined in lemma 1 .
* lemma 5 . * for the iterated kernels @xmath167 we have the following estimate : @xmath168 where @xmath169 and @xmath170 is the gamma - function of euler . `
proof : ` the proof will be done by induction in @xmath171 . taking the representation of the green s function given in ( 10 ) into account , and from the representation of the kernel @xmath172 at @xmath173 , the inequality ( 32 ) @xmath174 is automatically deduced .
let ( 32 ) be valid for @xmath175 .
we will prove the validity of this formula for @xmath176 . to this end , by using inequality ( 32 ) , at @xmath173 and @xmath175 , we have that @xmath177 evaluating the previous integrals we have that @xmath178 which proves lemma 5
. now we are in a position to prove the final result of this paper , which gives us the volterra property for the inverse of operator @xmath143 .
* theorem 3 .
* the integral operator defined in the right hand of ( 25 ) , i.e. @xmath179 has the volterra property ( it is almost continuous and quasi - nilpotent ) in @xmath180(@xmath181 ) . `
proof : ` since the continuity of this operator follows from the fact that @xmath182 . to prove this theorem
, we only need to verify that operator @xmath183 , defined by ( 33 ) , is quasi - nilpotent , i.e. @xmath184 where @xmath185\,\,\,\,\,\,,\,\,\,\,n = 1,2,3 , \ldots\ ] ] from ( 33 ) , and by direct calculations , one can easily arrive at the following expression : @xmath186 consequently , using the inequality of schwarz and expression ( 32 ) , from the representation ( 35 ) we obtain that @xmath187}^2}dxdy } \hfill \\ & \leq & \iint\limits_\omega \left [ { \left ( { \iint\limits_\omega { { { \left| { f\left ( { { x_1},{y_1 } } \right ) } \right|}^2}d{x_1}d{y_1 } } } \right)\,\left ( { \iint\limits_\omega { { { \left| { { k_n}\left ( { x , y;\,{x_1},{y_1 } } \right ) } \right|}^2}d{x_1}d{y_1 } } } \right ) } \right]dxdy \\ & \leq & \left ( { \frac{3}{2}\sqrt \pi m } \right)^{2n}\frac{1}{{n\left ( { n - 1 } \right){\gamma ^2}\left ( { \frac{n}{2 } } \right)}}\left\| f \right\|_0
^ 2.\end{aligned}\ ] ] from here we get @xmath188 from the last equality one can state the validity of the equality ( 34 ) and theorem 3 is proved .
* consequence 1 .
* problem b has the volterra property .
* consequence 2 .
* for any complex number @xmath189 , the equation @xmath190 is uniquely solvable for all @xmath191 . due to the invertibility of the operator @xmath143
, the unique solvability of the eq.(36 ) is equivalent to the uniqueness of solution of the equation @xmath192 which is a second kind volterra equation .
this proves consequence 2 of theorem 3 .
the authors declare that they have no competing interests .
the four authors have participated into the obtained results .
the collaboration of each one can not be separated in different parts of the paper .
all of them have made substantial contributions to the theoretical results .
the four authors have been involved in drafting the manuscript and revising it critically for important intellectual content .
all authors have given final approval of the version to be published .
this research was partially supported by ministerio de ciencia e innovacin - spain , and feder , project mtm2010 - 15314 and kaznpu rector s grant for 2013 .
ostertagova e. , ostertag o. , bocko j. : * problems of mechanics described by parabolic and hyperbolic differential equations ( of second order)*. _ proceedings of the 4th international conference `` modelling of mechanical and mechatronic systems 2011 '' _ 2011:385391 .
berdyshev a.s .
, karimov e.t . : * some non - local problems for the parabolic - hyperbolic type eguation with non - characteristic line of changing type*. _ central european journal of mathematics cejm _ 2006 ,
_ volume 4 _ , * 2*:183193 .
kapustin , n. yu . ,
moiseev , e. i. : * on spectral problems with a spectral parameter in the boundary condition*. ( russian ) _ differ .
volume 33 _ , * 1*:115119 ; translation in _
differential equations _ 1997 , _
volume 33 _ , * 1*:116120 .
berdyshev a.s .
, rakhmatullaeva n.a . :
* nonlocal problems with special gluing for a parabolic - hyperbolic equation*. _ proceedings of the 6th international isaac congress `` further progress in analysis '' _ 2007:727734 .
ankara , turkey , 13 - 18 august 2007 .
berdyshev a.s .
, cabada a. , kadirkulov b.j .
: * the samarskii - ionkin type problem for the fourth order parabolic equation with fractional differential operator*. _ computers and mathematics with applications ( camwa ) _ 2011 , _ volume 62_:38843893 .
berdyshev a. s. , karimov e. t. , akhtaeva n.s .
: * boundary value problems with integral gluing conditions for fractional - order mixed - type equation*. _ international journal of differential equations _ 2011 , _
volume 2011 _ , article i d 268465 , 10 pages
. doi:10.1155/2011/268465 .
berdyshev a.s .
, cabada a. , karimov e.t .
: * on a non - local boundary problem for a parabolic - hyperbolic equation involving a riemann - liouville fractional differential operator*. _ _ nonlinear analysis : theory , methods and applications ( natma)__2012,_volume 75 _ , * 6*:32683273 . | in the present work we consider a boundary value problem with gluing conditions of integral form for parabolic - hyperbolic type equation . we prove that the considered problem has the volterra property .
the main tools used in the work are related to the method of the integral equations and functional analysis .
[ 1995/12/01 ] |
* quantum fisher information . *
we start with a brief introduction of qfi and give a useful form of qfi for a special kind of mixed qudit states , which usually represents the output states of qudit cloning .
recall that qfi of parameter @xmath3 encoded in @xmath0-dimensional quantum state @xmath4 is generally defined as^@xcite^ @xmath5 where @xmath6 is the so - called symmetric logarithmic derivative , which is defined by @xmath7 with @xmath8 . by diagonalizing the matrix as @xmath9
, one can rewritten the qfi as^@xcite^ @xmath10 where @xmath11 is the qfi for pure state @xmath12 with the form @xmath13.\ ] ] note that eq .
( [ e2 ] ) suggests the qfi of a non - full - rank state is only determined by the subset of @xmath14 with nonzero eigenvalues .
physically , the qfi can be divided into three parts^@xcite^. the first term is just the classical fisher information determined by the probability distribution ; the second term is a weighted average over the qfi for all the nonzero eigenstates ; the last term stemming from the mixture of pure states reduces the qfi and hence the estimation precision below the pure - state case . though the eq .
( [ e2 ] ) is powerful , there is no explicit expression for an arbitrary @xmath0-dimensional mixed state .
however , it is worth noting that the output reduced states of uqcm and pqcm all have a form as @xmath15 which is completely characterized by a parameter independent shrinking factor @xmath16 and the dimensionality @xmath0
. here , @xmath17 and @xmath18 are the input state and @xmath0-dimensional identity matrix , respectively .
although @xmath19 has such a simple form , an analytical expression of qfi is still difficult to achieve .
fortunately , if we restrict our discussions to the special case of input states in the form @xmath20 could be given by ( see methods ) @xmath21 the above equation is a key mathematical tool for our analysis of this paper .
although this expression only holds for the combination of eqs .
( [ e3 ] ) and ( [ e4 ] ) , it is powerful since the scaling form of @xmath19 is usually satisfied in quantum cloning machines or in the case of a pure state under white noises . on the other hand ,
the equatorial states are widely employed in the physical implementations of quantum communication ideas ( such as bb84 protocol^@xcite^ ) as well as in the demonstration of fundamental questions in quantum information processing . after a simple calculation
, we find that @xmath22 is a monotonically increasing function of the shrinking factor @xmath16 .
this is to be expected because the larger @xmath16 indicates more information the reduced output state @xmath19 contains about the relevant parameter .
* distributability . * before moving to the discussion of distribution , a proper measure that quantify the distributability of cloning machines should be well defined .
note that one can define the distributability in different ways which depends on what is distributed in the procedure .
for example , it can be defined from the perspectives of fidelity which quantifies the total information of the state , and qfi which quantifies the information of particular parameters in the state .
it is well known that , unlike the symmetric cloning in which all the copies are the same , the outputs of asymmetric cloning are nonidentical .
hence , the optimality of asymmetric cloning can be judged by maximizing the sum of all copies fidelities , as discussed in refs .
@xcite . intuitively , when we consider the distribution of quantum states , the distributability of asymmetric cloning can be defined as @xmath23 where @xmath24 is the fidelity between the original and the @xmath25th copy .
namely , the larger @xmath26 indicates the better capability of distribution on the quantum states . in a seminal work^@xcite^
, the authors pointed that both fidelity and qfi are highly related to the distinguishability of the states , which is measured by bures distance^@xcite^. therefore , from the perspective of qfi , the measure @xmath27 also qualifies the capability of distributing information of relevant parameter encoded in the input state , where @xmath28 denotes the qfi of parameter @xmath3 in the @xmath25th copy . in the following discussions , we will use the definitions ( [ e06 ] ) and ( [ e07 ] ) to quantify the distributabilities of fidelity and qfi respectively in asymmetric cloning machines . * qfi distribution for 2-dimensional cloning . *
as we mentioned above , one is particularly interested in the asymmetric cloning machines which produce two copies with different qualities within the framework of quantum cryptography .
two typical asymmetric cloning machines are asymmetric uqcm^@xcite^ , which clones all input states equally well , and asymmetric pqcm^@xcite^ , which works equally well only for equatorial input states with the form of eq .
( [ e4 ] ) .
we first discuss the 2-dimensional cloning by obtaining the analytical results , and then generalize it to _
d_-dimensional cloning by the assistance of numerical simulations .
* asymmetric 2-dimensional uqcm . *
the @xmath29 optimal asymmetric uqcm was independently proposed by niu and griffiths^@xcite^ , buek _ _ et al__^@xcite^ and cerf^@xcite^. though their formalisms are slightly different , the results are exactly the same . for the sake of convenience ,
we adopt the quantum circuit approach developed by buek .
the transformation of asymmetric uqcm can be written in the following form @xmath30 where @xmath31 and @xmath32 .
the parameters @xmath33 and @xmath34 are real , and satisfy the normalization condition @xmath35 . with the transformation ( [ e6 ] ) in mind , it is now an easy exercise to verify that the two different copies of the original state @xmath36 are @xmath37 from the point of geometry , the asymmetric uqcm shrinks the original bloch vector by two different shrinking factors ( @xmath38 , @xmath39 ) regardless of its direction . as special cases , we can see that if @xmath40 , then no information has been transferred from the original system , while , if @xmath41 , then all of the information in system a has been transferred to system b. in addition , if @xmath42 ( i.e. , @xmath43 ) , then it reduces to the symmetric uqcm case .
obviously , the two shrinking factors @xmath44 and @xmath45 are related , and should satisfy the no - cloning inequality^@xcite^ @xmath46 which is an ellipse in the @xmath47 space .
an optimal asymmetric uqcm is characterized by a point @xmath48 which lies on the boundary of this ellipse . according to eq .
( [ e5 ] ) , when @xmath49 , the qfi of parameter @xmath3 is proportional to @xmath50 .
immediately , the qfis are @xmath51 here and henceforth we omit the subscript @xmath52 for brevity since we restrict our discussions to the single - parameter scenario .
remarkably , a trade - off relation exists for the two qfis : if one qfi is large , correspondingly another qfi will become small . combining eqs .
( [ e9 ] ) and ( [ e10 ] ) , the trade - off relation of qfi is expressed as @xmath53 this trade - off tells us that even we only concern cloning the information of a particular parameter encoded in quantum states , two close - to - perfect copies can not be achieved simultaneously , imposed by quantum mechanics .
an intuitive presentation of this trade - off relation is shown in fig .
[ fig2]b ( dashed line ) . in the following ,
we consider the distribution of qfi in the asymmetric uqcm . as we defined in eq .
( [ e07 ] ) .
the distributability of qfi is measured by @xmath54 therefore , the larger @xmath55 is , the more qfi of the relevant parameter has been distributed to the two copies .
we find that the asymmetric uqcm always performs better than symmetric uqcm in distributing qfi , which means @xmath56 this can be proved by the method of lagrange multiplier .
one will find three extreme points : @xmath57 , @xmath58 and @xmath59 .
it is easy to verify that @xmath60 is the minimum value . in order to show the difference of distributability between qfi and fidelity
, we also write the corresponding fidelities defined as @xmath61 @xmath62 according to eq .
( [ e06 ] ) , we adopt @xmath63 to qualify the capability of asymmetric uqcm in distributing the entire quantum state . fig .
[ fig1 ] shows the results of asymmetric 2-dimensional uqcm .
it is remarkable that , from the perspective of qfi , the asymmetric uqcm is always works better than symmetric uqcm , which is a sharp contrast to the result of fidelity^@xcite^ , where the symmetric uqcm is always optimal .
* asymmetric 2-dimensional pqcm . * in the context of quantum cryptography^@xcite^ , the uqcm studied in the previous subsection might be optimal if the detail setup of qkd protocol is not specified .
but it may not be optimal for the quantum states involved in a special qkd protocol . practically , it is possible that we already know a priori information of the input states .
thus , a state - dependent cloning machine would perform better than uqcm .
the best - known example of state - dependent cloning machine is the so - called pqcm .
the symmetric pqcm was firstly proposed by bru _
et al _ for the equatorial qubit state^@xcite^ and then an asymmetric version was demonstrated by niu and griffiths^@xcite^. recently , the asymmetric pqcm has been experimentally realized using nmr^@xcite^ and fiber optics^@xcite^. previous studies suggest that the equatorial qubit state pqcm can be realized by both economic^@xcite^ and non - economic^@xcite^ transformations .
one can check that both economic and non - economic methods achieve the same distributability of qfi . in the text
we discuss the non - economic case as it can be directly generalized to _
d_-dimensional pqcm .
the transformation of asymmetric pqcm can be written in the following form @xmath64 with @xmath65 and @xmath66 .
@xmath67 is the normalization condition .
@xmath68 is a set of orthogonal normalized ancillary state . for the equatorial qubit - state @xmath36 ,
the output states have the form as @xmath69 then , the shrinking factors are @xmath70 . using the normalization condition ,
the shrinking factors can be simplified as @xmath71 as is seen , in the scenario of asymmetric pqcm , there are two free parameters to be optimized .
therefore , an optimal asymmetric pqcm is defined as the following : if we fix the quality of one copy , then the other copy is optimal with the highest quality . from the eqs .
( [ e18 ] ) and ( [ e19 ] ) , one can eliminate @xmath34 and obtain the trade - off relation between @xmath44 and @xmath45 @xmath72 assuming @xmath44 is constant , the optimal value of @xmath73 can be found , and the optimal trade - off relation reduces to @xmath74 the corresponding qfis are @xmath75 in the symmetric case , we have @xmath76 . remarkably , according to eq .
( [ e11 ] ) , one can immediately find the inequality @xmath77 the meanings of above equation are twofold .
on one hand , it indicates that asymmetric pqcm performs better than asymmetric uqcm in distributing qfi by virtue of the known information .
this result is essentially in agreement with that of fidelity . to be clear , we plot the trade - off relations between a s fidelity ( qfi ) and b s fidelity ( qfi ) for both asymmetric uqcm and pqcm in fig .
it is evident that the lines of asymmetric pqcm are always above those of asymmetric uqcm , except for the start points and end points . on the other hand
, it should be noted that , for the 2-dimensional pqcm , the asymmetric case is as good as the symmetric case on the capability of distributing qfi , while the later always performs better than the former with the measure of fidelity .
the reason is that the fidelity @xmath78 as a function of @xmath79 is strictly concave , but the qfi @xmath80 as a function of @xmath81 is convex .
this results in the sum of two fidelities and qfis achieving its maximal and minimal value , respectively , in the symmetric case .
* qfi distribution for _ d_-dimensional cloning .
* until now , we have restricted our discussions to the 2-dimensional cloning .
although all quantum information tasks can be performed by using only two - level systems , it has been recently recognized that higher - dimensional quantum states ( i.e. , qudits ) can offer significant advantages for improving the security of quantum cryptographic protocols^@xcite^ , achieving higher information - density coding^@xcite^ and reducing the required resources for quantum computation and simulation^@xcite^. based on these considerations , it would be essential to extend above discussions to _ d_-dimensional cloning .
one may think the results will be trivial and analogous conclusions will be obtained as well as the 2-dimensional cloning . however , as we will show below , there are some similarities between them , but more importantly , significant differences will appear with increasing dimensionality @xmath0 . *
asymmetric @xmath0-dimensional uqcm .
* the optimal asymmetric @xmath0-dimensional uqcm was proposed by cerf^@xcite^ and braunstein _ _ et al__^@xcite^. for a @xmath0-dimensional quantum system , the corresponding asymmetric uqcm can be generalized directly from the transformation ( [ e6 ] ) with @xmath82 and @xmath83 instead defined in higher - dimension , @xmath84 , and @xmath85 respectively .
hence , the normalization condition now reads @xmath86 .
the output reduced density matrices are written in the form of eq . ( [ e3 ] )
@xmath87 with shrinking factors @xmath38 and @xmath39 .
in particular , if @xmath88 , we recover the results of symmetric uqcm .
similarly , we can obtain a trade - off relation between two shrinking factors @xmath44 and @xmath45 .
@xmath89 which corresponds to a set of ellipses in the space of shrinking factors that their eccentricities vary with dimensionality .
it should be noted that in the infinite dimensional case , the corresponding ellipse shrinks to the line @xmath90 .
now we turn to the calculation of qfi . assuming the input state is a @xmath0-dimensional equatorial state , then the qfis of output states ( [ e23 ] ) and ( [ e24 ] ) are obtained directly by ( [ e5 ] ) . @xmath91 the tradeoff relation between @xmath92 and @xmath93 can be derived by substituting @xmath94 into ( [ e25 ] ) with @xmath95 , but it is too complicated to present in the text .
nevertheless , there is no doubt that one can not gain , at the same time , two copies whose qfis are above values allowed by the trade - off relation .
for @xmath96 , respectively from top to down .
the red circles denote the corresponding results of symmetric uqcm , and the insertion is magnified plot of @xmath97.,title="fig:",scaledwidth=40.0% ] + for @xmath96 , respectively from top to down .
the red circles denote the corresponding results of symmetric uqcm , and the insertion is magnified plot of @xmath97.,title="fig:",scaledwidth=40.0% ] + we are concerned with whether the @xmath0-dimensional asymmetric uqcm still performs better than symmetric uqcm in distributing qfi . against all expectations , the results become subtle with increasing @xmath0 . unlike the 2-dimensional case where asymmetric uqcm is always better than symmetric uqcm , the @xmath0-dimensional asymmetric uqcm may be worse than symmetric uqcm under certain conditions . as shown in fig .
[ fig3]a , we find the sum of two fidelities still reaches its largest value at the point of @xmath98 , which means , by the measure of fidelity , the symmetric uqcm will optimally copy the state regardless of the dimension^@xcite^. however , fig .
[ fig3]b shows that the distributability of qfi achieves its smallest value at the point of @xmath99 when @xmath100 , while it is interesting to note that the point of @xmath99 becomes a local maximum point when @xmath97 .
namely , the asymmetric uqcm is no longer always better than symmetric uqcm with increasing @xmath0 .
the reason is that even though qfi is a monotonically function of the shrinking factor , it is not a linear function of it .
this sophisticated relation between @xmath55 and @xmath16 reveals above interesting results . naturally , we start wondering when the asymmetric uqcm may become worse than symmetric uqcm .
the numerical simulation shows that when @xmath1 , the @xmath44 of global minimal @xmath55 is equal to the symmetric case . while a bifurcation appears at @xmath101 , which means @xmath102 is no longer the global point of minimal @xmath55 , as shown in fig .
mathematically , we can understand the bifurcation as follows : @xmath55 as a function of @xmath44 has three extreme points which are physically allowed when @xmath1 , and @xmath99 is the point of global minima . when @xmath2 , it has five extreme points , as seen from the insertion in fig .
moreover , @xmath99 is no longer the global minima but a local maximum point .
thus , the symmetric uqcm may outperform the asymmetric case .
however , it is hard to understand why the critical point is @xmath101 in physical , we conjecture this critical point is related to the hilbert space structure of qudits . we leave this as an open question and the further study is underway .
* asymmetric @xmath0-dimensional pqcm . *
the generalization of asymmetric pqcm to @xmath0-dimensional is much more difficult , and in particular , it is too complicated to present an analytical trade - off relation between two copies .
however , with the aid of numerical simulations , we can confirm two main results about the distribution of qfi in asymmetric @xmath0-dimensional pqcm : ( i ) pqcm gains an advantage over uqcm by utilizing the priori information , and ( ii ) a sudden change of the point of minimum also exists in asymmetric pqcm with increasing @xmath0 .
the cloning transformation of asymmetric @xmath0-dimensional pqcm can be introduced^@xcite^ @xmath104 with the normalization condition @xmath105 . given the input state in the form of ( [ e4 ] ) , then , the shrinking factors of the output copies read as @xmath106 where we have use the normalization condition to eliminate the parameter @xmath107 . similar to the 2-dimensional case , here we again need to optimize two free tuning parameters .
therefore , in the same way , an optimal asymmetric pqcm is defined by optimizing @xmath45 as large as possible when @xmath44 is fixed , and vice versa . by eliminating the parameter @xmath34 ,
we can obtain the trade - off relation ( not the optimal one ) @xmath108\nonumber\\ & + \frac{\eta_{a}-(d-2)a^2}{a}\sqrt{\frac{1-\big(\frac{\eta_{a}-(d-2)a^2}{2a}\big)^2}{d-1}-a^2}.\end{aligned}\ ] ] the optimal trade - off relation need to be further optimized by choosing a proper value of @xmath109 to make the largest @xmath45 .
unfortunately , there is not a closed analytical form of @xmath109 for any dimensionality @xmath0 .
however , by simple numerical simulations , we find that asymmetric pqcm indeed always performs better than asymmetric uqcm in distributing qfi as shown in fig .
when the dimensionality _ d _ is large ( e.g. , @xmath110 ) , it should be noted that the advantage of pqcm over uqcm almost disappears .
moreover , similar to the case of asymmetric @xmath0-dimensional uqcm , the asymmetric @xmath0-dimensional pqcm is not always better than the symmetric case for any dimensionality @xmath0 .
[ fig5]b shows that a bifurcation of the point of global minimum also occurs at @xmath101 .
this phenomena stresses that the critical point appearing at @xmath101 is not in any sense accidental .
the physical reason behind this is worth further study . and
@xmath80 for _ d_-dimensional asymmetric uqcm ( dashed lines ) and pqcm ( solid lines ) with @xmath111 , respectively from top to bottom .
the insertion is magnified plot of @xmath112 .
note that when @xmath110 and 30 , the two lines overlap greatly .
( b ) @xmath44 of minimal @xmath113 as a function of dimensionality @xmath0 .
, title="fig:",scaledwidth=40.0% ] + and @xmath80 for _ d_-dimensional asymmetric uqcm ( dashed lines ) and pqcm ( solid lines ) with @xmath111 , respectively from top to bottom . the insertion is magnified plot of @xmath112 .
note that when @xmath110 and 30 , the two lines overlap greatly .
( b ) @xmath44 of minimal @xmath113 as a function of dimensionality @xmath0 .
, title="fig:",scaledwidth=40.0% ] + * discussions . * in this paper , we have investigated the distribution of qfi in asymmetric cloning machines which produce two nonidentical copies . in particular , we have elucidated four questions as we mentioned before . here , we summarize our results by replying these questions .
( i ) the answer is yes . it is definite that improving the qfi in one copy results in decreasing the qfi of the other copy , and the trade - off relation can be obtained analytically except for the asymmetric @xmath0-dimensional pqcm .
( ii ) the answer is also yes . thanks to a priori knowledge of the input states , pqcm always performs better than uqcm in distributing qfi .
( iii ) the answer is not so straightforward .
it should be divided into two categories : for 2-dimensional cloning , asymmetric cloning always outperforms symmetric cloning on the distribution of qfi ; while for the _ d_-dimensional cloning case , the above conclusion only holds when @xmath1 and becomes invalid when @xmath2 , i.e. , the asymmetric cloning is not always better than symmetric cloning for any dimensionality .
( iv ) the most significant difference between fidelity and qfi is that fidelity is a linear function of the shrinking factor while qfi is nonlinear .
this leads to the counterintuitive result that symmetric cloning is always optimal from the perspective of fidelity , but asymmetric cloning usually works better than symmetric cloning on the distribution of qfi , except for some particular situations ( e.g. , when @xmath97 and @xmath114 , see the insertion in fig . [
fig3]b ) . in view of these findings
, we note that there are some problems in need of further clarifications .
the first important issue is to understand why does the critical point appear at @xmath101 , not other numbers .
secondly , we should realize that we have confined our discussion to the distributability of single parameter in asymmetric cloning machines .
however , from both theoretical and practical points of view , it seems to be interesting to examine the problem of multi - parameter distribution in asymmetric quantum cloning .
intuitively , there would be a trade - off relation of the quantum fisher information matrices between two nonidentical copies .
these would be very intriguing topics that need further studies . here , we give the details of the derivation of eq.([e5 ] ) from eqs .
( [ e3 ] ) and ( [ e4 ] ) . to be clear
, we can rewrite the ( [ e3 ] ) as @xmath115 note that the eigenvalues of @xmath116 consists of only two categories : @xmath117/d$ ] , and @xmath118 with @xmath119 .
obviously , @xmath120 is an eigenstate of @xmath116 .
thus the problem is converted to construct a complete orthogonal set ( containing @xmath121 bases ) of the operator @xmath122 which is also orthogonal to @xmath123 at the same time .
the procedure can be divided into three steps : ( i ) finding @xmath121 bases of @xmath124 which are orthogonal to @xmath123 ; ( ii ) using the gram - schmidt procedure to orthogonalize them and ( iii ) the normalization . by this time
, we have diagonalized the state ( [ e3 ] ) in the bases @xmath129 , and then the qfi can be calculated by eq .
( [ e2 ] ) .
note that all the parameters @xmath52 is equally weighted due to the symmetry of @xmath123 , thus the qfi of any parameter is the same .
furthermore , we observe that the part of classical fisher information vanishes since the probability distribution is independent of parameters @xmath52 , if measured in this bases . the remaining work is to determine the last two terms in eq .
( [ e2 ] ) .
after lots of complicated but straightforward calculations , we obtain the qfi of any parameter @xmath52 @xmath130 the authors are supported by the national natural science foundation of china under grants no .
11247006 , 11025527 and 10935010 , the national 973 program under grants no .
2012cb921602 , and the china postdoctoral science foundation under grant no .
2014m550598 .
bennett , c. h. & brassard , g. quantum cryptography : public key distribution and coin tossing . _ in proceedings of ieee international conference on computers , systems and signal processing _ pp.175 - 179 ( 1984 ) .
ghiu , i. asymmetric quantum telecloning of _
d_-level systems and broadcasting of entanglement to different locations using the `` many - to - many '' communication protocol .
a _ * 67 * , 012323 ( 2003 ) .
braunstein , s. l. , buek , v. & hillery , m. quantum - information distributors : quantum network for symmetric and asymmetric cloning in arbitrary dimension and continuous limit .
a _ * 63 * , 052313 ( 2001 ) . | an unknown quantum state can not be copied on demand and broadcast freely due to the famous no - cloning theorem .
approximate cloning schemes have been proposed to achieve the optimal cloning characterized by the maximal fidelity between the original and its copies . here , from the perspective of quantum fisher information ( qfi ) , we investigate the distribution of qfi in asymmetric cloning machines which produce two nonidentical copies . as one might expect , improving the qfi of one copy results in decreasing the qfi of the other copy , roughly the same as that of fidelity .
it is perhaps also unsurprising that asymmetric phase - covariant cloning machine outperforms universal cloning machine in distributing qfi since a priori information of the input state has been utilized .
however , interesting results appear when we compare the distributabilities of fidelity ( which quantifies the full information of quantum states ) , and qfi ( which only captures the information of relevant parameters ) in asymmetric cloning machines .
in contrast to the results of fidelity , where the distributability of symmetric cloning is always optimal for any @xmath0-dimensional cloning , we find that asymmetric cloning performs always better than symmetric cloning on the distribution of qfi for @xmath1 , but this conclusion becomes invalid when @xmath2 .
classical information can be replicated perfectly and broadcast without fundamental limitations .
however , information encoded in quantum states is subject to several intrinsic restrictions of quantum mechanics , such as heisenberg s uncertainty relations^@xcite^ and quantum no - cloning theorem^@xcite^. the no - cloning theorem tells us that an unknown quantum state can not be perfectly replicated because of the linearity of the time evolution in quantum physics , which is the essential prerequisite for the absolute security of quantum cryptography^@xcite^. nevertheless , it is still possible to clone a quantum state approximately , or instead , clone it perfectly with certain probability^@xcite^. therefore , various types of quantum cloning machines have been designed for different quantum information tasks , including universal quantum cloning machine ( uqcm)^@xcite^ , state - dependent cloning machines^@xcite^ and phase - covariant quantum cloning machine ( pqcm)^@xcite^. so far , the optimality of the approximate cloning machine is judged generally by whether the obtained fidelity between the cloning output state and the ideal state achieves its optimal bound .
although the fidelity may have qualified the complete information of the quantum states , in most scenarios , only the information of certain parameters which are physically encoded in quantum states is our practical concern .
for example , the relative phase estimation is an extremely important issue in the field of quantum metrology^@xcite^. thus it is not necessary to gain complete information of the whole quantum states themselves , but rather the relevant parameter information .
qfi is a natural candidate to quantify the physical information about the involved parameters^@xcite^. in ref .
@xcite , the authors pointed that the qfi of relevant parameter encoded in quantum states also can not be cloned perfectly , while it might be broadcast even in some non - commuting quantum states .
furthermore , from the perspective of qfi , song _ et al .
_ showed that wootters - zurek cloning performs better than universal cloning for the symmetric cloning cases^@xcite^. in our recent work , the multiple phase estimation problem was investigated in the framework of symmetric quantum cloning machines^@xcite^. on the other hand , we note that quantum cloning machines not only provide a good platform for investigating distribution of quantum information , but also have been proved to be very efficient eavesdropping attacks on the quantum key distribution ( qkd ) protocols^@xcite^. in this context , asymmetric quantum cloning machines would be of particular interest since the eavesdropper can adjust the trade - off between the information gained from a quantum communication channel and the error rate of information transmitted to the authorized receiver . motivated by these considerations , we investigate the problem of distributing qfi in asymmetric quantum cloning machines for any dimension .
we focus on the following four questions : ( i ) is it possible to improve the qfi of one copy by decreasing that of the other copy ?
if yes , what s the trade - off relation between them ?
( ii ) does asymmetric pqcm always perform better than asymmetric uqcm on the capability of distributing qfi ?
( iii ) does asymmetric cloning always outperform symmetric cloning in distributing qfi for any dimensionality ?
( iv ) what s the difference between fidelity and qfi on the characterization of distributability in asymmetric cloning ? except for the fourth question need to be clarified in detail , we can briefly answer the first two questions in the affirmative but the third in the negative .
our results shed an alternative light on quantum cloning and may be exploited for quantum phase estimation . |
in applications of duality theory for crossed products one often has to know how common constructions such as induction and restriction of representations behave under duality .
thus , after piecemeal results by several authors , echterhoff was led to prove that , for systems involving abelian groups , induction and restriction are dual to one another @xcite .
he later used these results to great effect in his analysis of crossed products with continuous trace @xcite . to state echterhoff s theorem precisely , we fix an action @xmath10 of a locally compact abelian group @xmath1 on a @xmath2-algebra @xmath3 , and a closed subgroup @xmath4 of @xmath1 .
green s imprimitivity theorem gives an imprimitivity bimodule @xmath11 implementing a morita equivalence between @xmath12 and the imprimitivity algebra @xmath13 ; the latter algebra is naturally isomorphic to the crossed product @xmath14 by the dual action @xmath15 of the subgroup @xmath16 of @xmath17 .
let @xmath18 denote the map on equivalence classes of representations given by induction of representations in the sense of green : if @xmath19 is a representation of @xmath12 on @xmath20 , then @xmath21 is by definition the natural left action on ( a completion of ) @xmath22 .
similarly , let @xmath23 denote the map given by restricting a covariant representation @xmath24 of @xmath25 to the representation @xmath26 of @xmath27 . then echterhoff s theorem says that we have commutative diagrams @xmath28 and @xmath29 in which the horizontal arrows on the bottom are the bijections induced by the imprimitivity bimodule @xmath11 .
as it stands , this theorem only makes sense for abelian groups , because otherwise there is no dual action .
however , as echterhoff himself observed , this is an obvious case of a theorem about abelian groups which should extend to an arbitrary locally compact group @xmath1 and a closed normal subgroup @xmath4 , replacing dual actions by dual coactions .
( another such theorem is ( * ? ? ?
* theorem 2.4 ) , which was extended to non - abelian groups in ( * ? ? ? * theorem 4.4 ) . )
this project has been carried through by echterhoff and the first and third authors @xcite , relating induction and restriction for @xmath30 to , respectively , the restriction and induction processes of mansfield @xcite for crossed products by coactions . here
we study the dual situation in which we start with a coaction @xmath0 of a locally compact group @xmath1 on @xmath3 .
our main theorems give ( in special cases ) commutative diagrams @xmath31 and @xmath32 similar to and .
this time , the horizontal arrows are the bijections given by mansfield s morita equivalence of @xmath33 with @xmath34 .
these theorems are potentially very useful , since they reduce questions about mansfield induction and restriction to questions about the much more familiar and manageable theory of green @xcite . indeed , in the special case @xmath35 , diagrams and reduce to results used by gootman and lazar @xcite to study induced ideals and the primitive ideal space of crossed products by coactions .
we begin with a preliminary section in which we discuss our conventions about hilbert modules , basic facts about induction of representations , and a version of green s imprimitivity theorem for reduced crossed products .
this is necessary because we want to avoid unnecessary assumptions of amenability ; for the same reason , we shall use the version of mansfield induction given in @xcite .
we stress , however , that for amenable groups , the only part of @xcite needed here is the extension of mansfield s theorem to full coactions and twisted systems .
a special case of this extension is in @xcite , and the general case is hinted at in @xcite .
our first main results in section [ res - ind - sec ] concern the `` @xmath36-@xmath37 '' diagram .
the result is more general than that stated above in three respects .
first , we consider two normal subgroups @xmath38 .
second , we consider also the twisted crossed products of @xcite , which introduces another normal subgroup @xmath39 containing @xmath40 ; our strategy , both here and in section [ ind - res - sec ] , is to prove the untwisted version , and then check that the constructions respect the twists .
third , we prove that the diagram commutes in a very strong sense : all the maps in the diagram are implemented by hilbert modules , so their compositions are given by tensor products of modules , and we prove that the tensor products corresponding to the two alternative routes are naturally isomorphic .
this not only gives a stronger theorem than the corresponding result of @xcite , but also a distinctly shorter proof .
thus we can claim that this is another example of a theorem about abelian groups which becomes much cleaner when seen from the point of view of coactions .
( well , we would , would nt we . ) in section [ ind - res - sec ] we turn to the `` @xmath37-@xmath36 '' diagram . as in section [ res - ind - sec ] , our main results are much more general than stated above .
the overall pattern of this section is like that of section [ res - ind - sec ] , but this time passing to twisted crossed products is harder . in echterhoff
s paper , the second diagram amounts to the induction in stages of green @xcite .
mansfield did not prove such a theorem for his induction process , so we are forced to do it here ( corollary [ mansfield - stages - thm ] ) . in section [ me - infl - stab - sec ]
we give some applications of our theorems .
we prove that restriction and induction are compatible with morita equivalence in general , and in particular with the stabilization trick of @xcite , which allows us to replace twisted crossed products by ordinary ones .
( this result for abelian groups was a fundamental tool in @xcite . ) along the way we deduce , from the corresponding properties of @xmath37 and @xmath36 for crossed products by actions , that both @xmath36 and @xmath37 are respected by morita equivalence of coactions .
we feel that this is a good illustration of how our results might be used to deduce information about coactions from known properties of actions .
finally , in the last section , we use our main theorems to study the maps @xmath36 , @xmath37 , @xmath41 and @xmath42 on ideals of crossed products by coactions , obtaining generalizations to the case of nonamenable @xmath1 of many of the results in @xcite . here
we see the power not only of @xmath36-@xmath37 ( and @xmath42-@xmath41 ) duality , but also that of the hilbert module techniques , since our maps on spaces of representations automatically give rise , via the rieffel correspondence , to maps on spaces of ideals .
this research was carried out while the second author was visiting the university of newcastle in 1994 and 1995 , and while the first author was visiting arizona state university in 1995 .
the various visitors are grateful to their respective hosts for their hospitality .
[ prelim - sec ] throughout , @xmath1 will be a locally compact group with modular function @xmath43 .
we use left haar measure .
the group @xmath2-algebra of @xmath1 is denoted @xmath44 ; a subscript @xmath45 , as in @xmath46 or @xmath47 , always indicates a reduced object .
nondegenerate homomorphisms of @xmath2-algebras extend to homomorphisms of their multiplier algebras , and this will be done implicitly .
we use the conventions of @xcite , @xcite , ( * ? ? ?
* section 7 ) , and @xcite , although the latter uses maximal tensor products .
our coactions use minimal tensor products , are injective , and are full , i.e. , use @xmath44 . let @xmath48 be a coaction .
we let @xmath49 denote the crossed product , @xmath50 the dual action of @xmath1 on @xmath8 , and @xmath51 the normalization of @xmath0 . if @xmath4 is a closed normal subgroup of @xmath1 , we let @xmath52 denote the restricted coaction of @xmath53 on @xmath3 and @xmath7 the restricted crossed product
. moreover we let @xmath54 denote the restriction to @xmath55 of a nondegenerate homomorphism @xmath56 of @xmath57 . in @xcite
the first two authors generalized mansfield s imprimitivity machine @xcite .
when @xmath48 is a nondegenerate coaction and @xmath4 is a closed normal subgroup of @xmath1 , there are dense @xmath58-subalgebras @xmath59 and @xmath60 of @xmath8 and @xmath61 , respectively , such that @xmath59 is a ( full ) pre - hilbert @xmath60-module under right multiplication and inner product @xmath62 the integral converging strictly in @xmath63 .
let @xmath64 denote the completion of the pre - hilbert module @xmath59 .
then @xmath64 has a left @xmath9-module action determined by left multiplication on @xmath59 and @xmath65 when @xmath4 is amenable and @xmath0 is a reduced coaction , mansfield ( * ? ? ?
* theorem 27 ) proves that @xmath66 is morita equivalent to @xmath7 . for nonamenable subgroups and full coactions ,
the corresponding result is : ( * ? ? ?
* corollary 3.4)[manimp ] let @xmath48 be a nondegenerate coaction and @xmath4 a closed normal subgroup of @xmath1 such that @xmath67 is faithful .
then @xmath64 is an @xmath9 @xmath7 imprimitivity bimodule . in view of the above theorem ,
if @xmath0 is a nondegenerate coaction and @xmath4 is a closed normal subgroup of @xmath1 , we say _
mansfield imprimitivity works _ for @xmath4 and @xmath0 whenever @xmath68 is faithful ( * ? ? ?
* definition 3.5 ) .
when mansfield imprimitivity works we let @xmath69 denote the extension to @xmath64 of the inner product @xmath70 on @xmath59 .
mansfield s computations show that the left inner product @xmath71 for @xmath72 can be identified with the element @xmath73 of @xmath74 .
when @xmath0 is nondegenerate , mansfield imprimitivity works if either @xmath4 is amenable or @xmath0 is normal ( * ? ? ?
* lemma 3.2 ) .
if mansfield imprimitivity works for @xmath4 and @xmath0 , then it also works for any closed subgroup contained in @xmath4 ( * ? ? ?
* theorem 5.2 ) and any coaction morita equivalent to @xmath0 ( * ? ? ?
* theorem 5.3 ) .
if @xmath39 is a closed normal subgroup of @xmath1 containing @xmath4 and @xmath75 is a coaction , then mansfield imprimitivity works for @xmath4 and @xmath76 if and only if it works for @xmath4 and the inflated coaction @xmath77 ( * ? ? ?
* theorem 5.4 ) .
when we say @xmath78 is a twisted coaction , we mean @xmath39 is a closed normal subgroup of @xmath1 and @xmath79 is a twist for @xmath0 over @xmath80 @xcite .
we let @xmath81 denote the twisting ideal of @xmath8 , @xmath82 the twisted crossed product , and @xmath83 the dual action of @xmath39 on @xmath84 .
if further @xmath4 is a closed normal subgroup of @xmath1 contained in @xmath39 , then @xmath85 is also a twist for the restricted coaction @xmath86 over the quotient @xmath87 .
let @xmath88 denote the twisting ideal of @xmath89 .
then there is a restricted twisted crossed product @xmath90 the following result generalizes ( * ? ? ?
* theorem 4.1 ) : ( * ? ? ?
* theorem 4.4)[mprimp ] let @xmath78 be a nondegenerate twisted coaction and @xmath4 a closed normal subgroup of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath4 and @xmath0 .
then the quotient @xmath91 is an @xmath92 @xmath93 imprimitivity bimodule .
if @xmath78 is a nondegenerate twisted coaction , then mansfield imprimitivity works for @xmath39 and @xmath0 if and only if @xmath0 is normal ( * ? ? ?
* lemma 3.6 ) .
if @xmath78 is a nondegenerate normal twisted coaction and @xmath4 is a closed normal subgroup of @xmath1 contained in @xmath39 , then mansfield imprimitivity works for @xmath4 and @xmath0 since @xmath0 is normal , and also for @xmath4 and the morita equivalent stabilized coaction @xmath94 ( * ? ? ?
* theorem 3.1 ) , ( * ? ? ? * theorem 5.5 ) .
everything in this paper revolves around rieffel s induction process , so we should make our conventions explicit .
for more detailed treatments of this material we refer the reader to @xcite , @xcite , @xcite .
all our hilbert modules will be full , i.e. , the closed span of the inner product generates the @xmath2-algebra .
if @xmath95 is a right hilbert @xmath96-module and @xmath3 acts nondegenerately on @xmath95 by adjointable @xmath96-module maps ( so there is a homomorphism @xmath97 such that @xmath98 ) , we say @xmath95 is a right - hilbert @xmath3 @xmath96 bimodule .
( this terminology first appears in @xcite . )
if @xmath95 is also a left hilbert @xmath3-module such that @xmath99 for @xmath100 , then of course @xmath95 is an @xmath3
@xmath96 imprimitivity bimodule .
we denote the reverse bimodule by @xmath101 , with elements @xmath102 . when @xmath95 is a right - hilbert @xmath3
@xmath96 bimodule , rieffel induction gives a functor @xmath103 and we leave out parts of the notation if confusion seems unlikely .
actually , @xmath104 can be factored as @xmath105 followed by `` restriction '' from @xmath106 to @xmath107 .
since we will need it a lot , we abstract this latter bit : if @xmath108 is a nondegenerate homomorphism , composition with @xmath19 gives a `` restriction '' map @xmath109 we view this as a rieffel induction process : @xmath96 becomes a right - hilbert @xmath3
@xmath96 bimodule via @xmath110 for @xmath111 , @xmath112 . now suppose @xmath108 is a nondegenerate homomorphism and @xmath95 is a right - hilbert @xmath96 @xmath113
then @xmath95 becomes a right - hilbert @xmath3
@xmath113 bimodule via @xmath114 on the other hand , we can regard @xmath96 as a right - hilbert @xmath3
@xmath96 bimodule , and the map @xmath115 induces an isomorphism @xmath116 of right - hilbert @xmath3 @xmath113 bimodules .
in particular , we get a commutative diagram @xmath117{{\operatorname{rep}}a } \end{diagram } \ ] ] where we use prescripts and postscripts to indicate the coefficient algebras when necessary . in general
, we will omit parts of the notation , so that when we say @xmath118{{\operatorname{rep}}a } \end{diagram } \ ] ] is a commutative diagram , we mean @xmath95 is a right - hilbert @xmath3 @xmath96 bimodule , and similarly for @xmath119 and @xmath120 , and the equation @xmath121 holds in the strong sense that @xmath122 as right - hilbert @xmath3 @xmath113 bimodules . recall that rieffel induction gives rise to maps between ideals , so
if @xmath123 is an ideal of @xmath96 and if @xmath19 is any nondegenerate representation of @xmath96 with kernel @xmath123 , then @xmath124 is the kernel of @xmath125 .
when we have a commutative diagram of hilbert modules as in , we of course get @xmath126 as maps from ideals of @xmath113 to ideals of @xmath3
. we will often want to pass from a commutative diagram of hilbert modules to quotients .
there is a subtle point that needs checking : let @xmath95 and @xmath119 be right - hilbert @xmath3 @xmath96 and right - hilbert @xmath96
@xmath113 bimodules , respectively , and let @xmath39 be an ideal of @xmath113 .
further , let @xmath127 and @xmath128 be the corresponding induced ideals of @xmath96 and @xmath3 , respectively , and let @xmath129 and @xmath130 be the quotient maps .
then @xmath131 induces an isomorphism between the right - hilbert @xmath132
@xmath133 bimodules @xmath134 and @xmath135 .
straightforward ; a slightly different version was given in ( * ? ? ?
* lemma 1.10 ) . [ quotient ]
suppose the diagram @xmath136{{\operatorname{rep}}a } \end{diagram}\ ] ] commutes in the usual strong sense that @xmath137 .
further suppose that @xmath39 is an ideal of @xmath113 , and set @xmath127 and @xmath128 .
then the diagram @xmath138{{\operatorname{rep}}a / i } \end{diagram}\ ] ] also commutes in the usual strong sense . since @xmath137 ,
certainly @xmath139 so the above lemma immediately gives the corollary .
similarly , commutative diagrams with any number of vertices pass to quotients
. we will need an induction process for reduced crossed products by actions
. we could deduce it from green s version by applying @xcite , but we give a direct argument since we need the explicit imprimitivity bimodule .
let @xmath140 be an action and @xmath40 a closed subgroup of @xmath1 .
recall that green s inducing process starts with the right - pre - hilbert @xmath141 @xmath142 bimodule @xmath141 , where the operations for @xmath143 , @xmath144 are given by @xmath145 the particular arrangement of modular functions comes from @xcite ; green uses unorthodox conventions .
let @xmath146 be the completion of the pre - hilbert module @xmath141 , so @xmath146 is a right - hilbert @xmath147 @xmath148 bimodule .
we need to know that the kernels of the regular representations match up , so that @xmath146 passes to a right - hilbert @xmath149
@xmath150 bimodule .
[ reduced ] let @xmath140 be an action , @xmath40 a closed subgroup of @xmath1 , @xmath146 green s right - hilbert @xmath147
@xmath148 bimodule , and @xmath123 the kernel of the regular representation of @xmath148 .
then the induced ideal @xmath151 is the kernel of the regular representation of @xmath147 .
consequently , the quotient @xmath152 is a right - hilbert @xmath149 @xmath150
bimodule .
this follows from induction in stages : the proof of ( * ? ? ?
* proposition 8) shows that the diagram @xmath153{{\operatorname{rep}}b\times g } \end{diagram } \ ] ] commutes in the usual strong sense .
if @xmath19 is any faithful representation of @xmath96 , then @xmath154 and @xmath155 are the regular representations of @xmath148 and @xmath147 , respectively .
so the ideals @xmath156 and @xmath157 are the kernels of the regular representations . by commutativity of diagram ,
@xmath158 so indeed @xmath146 induces the kernel @xmath123 of @xmath159 to the kernel of @xmath160 .
[ reduced - stages ] let @xmath140 be an action , @xmath40 a closed subgroup of @xmath1 , and @xmath161 the right - hilbert @xmath149 @xmath150 bimodule obtained in lemma [ reduced ] .
then the diagram @xmath162{{\operatorname{rep}}b\times_r g } \end{diagram } \ ] ] commutes in the usual strong sense .
this follows from the above lemma and corollary [ quotient ] .
we emphasize that @xmath163 may be viewed as the completion of @xmath141 with respect to the norm induced by the @xmath150-valued pre - inner product .
in particular , the actions of @xmath150 and @xmath149 on @xmath163 are determined by the covariant representations of @xmath164 and @xmath140 on @xmath141 .
suppose we have a twisted coaction @xmath165 , and closed normal subgroups @xmath166 of @xmath1 contained in @xmath39 . in this section , we show that when mansfield imprimitivity works , the following diagram commutes in the usual strong sense : @xmath167 we will do this in two steps , first showing that the analogous untwisted diagram commutes , and then showing that the twisting ideals in the various crossed products match up properly , so that commutativity is preserved on taking quotients by these ideals . for @xmath168 ,
a weak form of the following theorem was proven in ( * ? ? ?
* theorem 4.1 ) .
[ plain - res - ind - thm ] let @xmath48 be a nondegenerate coaction , and let @xmath166 be closed normal subgroups of @xmath1 such that mansfield imprimitivity works for @xmath40 ( which is automatic if @xmath40 is amenable ) .
then the diagram @xmath169 commutes in the usual strong sense . first note that by ( * ? ? ?
* theorem 5.2 ) , mansfield imprimitivity also works for @xmath4 and @xmath0 , so @xmath64 is indeed an @xmath9 @xmath7
imprimitivity bimodule , and the above diagram makes sense .
we shall actually prove that the hilbert module tensor product @xmath170 of mansfield bimodules is isomorphic to the reduction @xmath171 of green s bimodule , as a right - hilbert @xmath172
@xmath9 bimodule .
this suffices , because then @xmath173 as a right - hilbert @xmath172
@xmath7 bimodule , and this is exactly what it means for the above diagram to commute in the strong sense .
both bimodules in the tensor product are completions of mansfield s dense subalgebra @xmath59 of @xmath8 for the appropriate inner products .
our isomorphism will be the extension to @xmath174 of the map @xmath175 defined by @xmath176 note that , up to a modular function , @xmath177 is just mansfield s left @xmath178-valued inner product @xmath179 , and hence does indeed give an element of @xmath178 .
in fact , if we define @xmath180 , then the map @xmath181 is a homeomorphism of @xmath178 ( with the inductive limit topology ) onto itself , which takes @xmath182 to @xmath183 .
this latter set is dense in @xmath184 for the inductive limit topology ( ( * ? ? ?
* lemma 25 ) ) ; it follows that the range of @xmath185 is also inductive limit dense in @xmath184 , and therefore in @xmath171 .
it only remains to show that @xmath185 preserves the hilbert module structure .
for the left action of @xmath172 , fix @xmath186 and @xmath187 .
then : @xmath188 and @xmath189 for the right action of @xmath9 , fix @xmath186 , @xmath190 and @xmath191 ; then one similarly verifies that @xmath192 and @xmath193 for the right @xmath9-valued inner product , fix @xmath191 and compute : @xmath194 it now follows that @xmath185 is a right - hilbert @xmath172 @xmath9 bimodule isomorphism of @xmath195 onto @xmath171 .
[ twisted - res - ind - cor ] let @xmath78 be a nondegenerate twisted coaction , and let @xmath166 be closed normal subgroups of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath40 and @xmath0 ( which is automatic if @xmath40 is amenable ) .
then the diagram @xmath196 commutes in the usual strong sense .
we show the appropriate ideals in diagram match up and appeal to corollary [ quotient ] ; because the diagram commutes , and the top and bottom maps are morita equivalences , we need only match up the ideals along three sides .
let @xmath81 , @xmath197 , and @xmath198 be the twisting ideals of @xmath8 , @xmath7 , and @xmath199 , respectively .
the reduced crossed product @xmath92 is by definition @xmath200 .
if @xmath19 is a representation of @xmath8 with kernel @xmath81 ( for example , if @xmath201 ) , then @xmath19 induces a faithful representation @xmath202 of @xmath203 ; thus @xmath204 it is part of the content of ( * ? ? ?
* theorem 4.4 ) that @xmath205 ( see ( * ? ? ?
* equation 4.2 ) ) .
hence the appropriate ideals match up along the top of diagram , and similarly along the bottom .
since @xmath206 by corollary [ reduced - stages ] , the ideals also match up along the right side of diagram .
in this section we prove analogs of theorem [ plain - res - ind - thm ] and corollary [ twisted - res - ind - cor ] , where now we use mansfield induction on the left sides of the diagrams and restriction on the right . as in the previous section ,
we first prove an untwisted version , and then show that the twisting ideals match up properly .
[ plain - ind - res - thm ] let @xmath48 be a nondegenerate coaction , and let @xmath166 be closed normal subgroups of @xmath1 such that mansfield imprimitivity works for @xmath40 ( which is automatic if @xmath40 is amenable ) .
then the diagram @xmath207 commutes in the usual strong sense .
let us denote the hilbert module for mansfield induction from @xmath199 to @xmath7 by @xmath208 ; here we are identifying @xmath209 with @xmath210 , so @xmath208 is a completion of mansfield s dense subalgebra @xmath60 of @xmath7 .
we shall prove that the hilbert module tensor product @xmath211 is isomorphic to @xmath212 as a right - hilbert @xmath9 @xmath199 bimodule .
this suffices , because then @xmath213 as a right - hilbert @xmath9
@xmath199 bimodule , and this is exactly what it means for the above diagram to commute in the strong sense
. our map will be the extension to @xmath214 of the map @xmath215 defined by @xmath216 here the product @xmath217 makes sense in @xmath63 because @xmath218 belongs to @xmath219 .
in other words , this product is given by the right action of @xmath60 on mansfield s bimodule @xmath59 , and hence @xmath220 does indeed belong to @xmath59 .
we now show that @xmath221 preserves the right - hilbert bimodule structure .
since @xmath8 on the left and @xmath199 on the right act by multiplication in @xmath63 , it is immediate that @xmath221 preserves these actions . to see that @xmath221 preserves the left @xmath4-action is a straightforward calculation , using the fact that each @xmath222 is fixed by @xmath223 for @xmath191 : @xmath224 to see that @xmath221 preserves the right @xmath199-valued inner products , note that @xmath225 on @xmath89 , and compute : @xmath226 it only remains to show that the range of @xmath221 is dense in @xmath212 . for this , note that @xmath227 , since if @xmath228 is @xmath229 , we may choose @xmath230 such that @xmath231 is identically @xmath232 on @xmath233 , and then @xmath234 .
hence , @xmath235 is dense in @xmath212 .
since @xmath212 is a nondegenerate right @xmath3-module , and @xmath236 is dense in @xmath3 , we therefore have @xmath237 it now follows that @xmath221 is a right - hilbert @xmath9
@xmath199 bimodule isomorphism of @xmath214 onto @xmath212 .
[ mansfield - stages - thm ] ( mansfield induction in stages . ) let @xmath48 be a nondegenerate coaction , and let @xmath166 be closed normal subgroups of @xmath1 such that mansfield imprimitivity works for @xmath40 ( which is automatic if @xmath40 is amenable ) .
then the diagram @xmath238 commutes in the usual strong sense .
the corollary requires that the hilbert module tensor product @xmath211 be isomorphic to @xmath212 as right - hilbert @xmath8
@xmath199 bimodules ; we showed slightly more than this in the proof of theorem [ plain - ind - res - thm ] by including the left @xmath4-action .
[ twisted - ind - res - cor ] let @xmath78 be a nondegenerate twisted coaction , and let @xmath166 be closed normal subgroups of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath40 and @xmath0 ( which is automatic if @xmath40 is amenable ) .
then the diagram @xmath239 commutes in the usual strong sense . as with corollary [ twisted - res - ind - cor ] , we need only show the appropriate ideals in the diagram match up , and appeal to corollary [ quotient ] . as before , let @xmath81 , @xmath197 , and @xmath198 be the twisting ideals of @xmath8 , @xmath7 , and @xmath199 , respectively .
the reduced crossed product @xmath92 is by definition @xmath200 .
if @xmath19 is a representation of @xmath8 with kernel @xmath81 ( for example , if @xmath201 ) , then @xmath19 induces a faithful representation @xmath202 of @xmath203 ; thus @xmath240 it is part of the content of ( * ? ? ?
* theorem 4.4 ) that @xmath205 ( see ( * ? ? ?
* equation 4.2 ) ) .
hence the appropriate ideals match up along the top of diagram , and similarly along the bottom .
it only remains to see that the ideals match up along the right side of .
we need @xmath241 now , our reduced version corollary [ reduced - stages ] of induction in stages gives @xmath242 so it will suffice to show two things : 1 .
the ideal @xmath243 of @xmath244 is @xmath40-invariant ; 2 .
if @xmath245 is an @xmath40-invariant ideal of @xmath9 , then @xmath246 these are shown abstractly in the next two lemmas , which although at most partly new , may have independent interest .
let @xmath164 be an action , @xmath4 a closed normal subgroup of @xmath40 , and @xmath123 an @xmath40-invariant ideal of @xmath96 .
then the ideal @xmath247 of @xmath248 is @xmath40-invariant for the decomposition action .
the decomposition action of @xmath40 on @xmath249 leaves the kernel of the regular representation invariant , hence it indeed induces an action @xmath250 of @xmath40 on @xmath248 .
explicitly , for @xmath191 , @xmath251 , and @xmath252 , @xmath253 where @xmath254 is the modular function of conjugation of @xmath40 on @xmath4 : @xmath255 let @xmath256 be the usual right - hilbert @xmath248
@xmath96 bimodule , so @xmath95 is a completion of @xmath257 .
we will show @xmath95 is @xmath40-equivariant , which will imply that @xmath40-invariant ideals of @xmath96 induce to @xmath40-invariant ideals of @xmath248 .
more precisely , we will construct a strongly continuous banach representation @xmath258 of @xmath40 on @xmath95 such that for @xmath251 , @xmath259 , @xmath260 , and @xmath261 we have @xmath262 a straightforward calculation then shows that if @xmath123 is @xmath40-invariant , so is @xmath263 .
for @xmath251 , @xmath264 define @xmath265 by @xmath266 to show , take @xmath267 and compute : @xmath268 it is now clear that @xmath258 is a homomorphism of @xmath40 into the isometric automorphisms of the normed space @xmath257 , hence determines by continuity a homomorphism of @xmath40 into the isometric automorphisms of the banach space @xmath95 , and follows , again by continuity . for , it suffices to take @xmath264 : for @xmath269 we have @xmath270 we next show @xmath258 is strongly continuous . by @xcite , it suffices to show that @xmath258 is strongly continuous for the inductive limit topology on @xmath257 .
fix @xmath271 with compact support @xmath233 , and suppose @xmath272 in @xmath1 .
since @xmath273 uniformly ( by a standard compactness argument ) , we need only find a compact set @xmath274 in @xmath4 and @xmath275 such that the support of @xmath276 is in @xmath274 for @xmath277 . to do this , choose a neighborhood @xmath278 of @xmath279 in @xmath1 with compact closure , and let @xmath275 be such that @xmath280 for @xmath277 .
then the compact set @xmath281 will do . finally , for , it suffices to take @xmath282 : for @xmath191 we have @xmath283 [ res - ind - invt - ideal - lem ] if @xmath164 is an action , @xmath4 is a closed normal subgroup of @xmath40 , and @xmath245 is an @xmath40-invariant ideal of @xmath248 , then @xmath284 we first show the analogous equality @xmath285 for full crossed products , assuming @xmath39 is an @xmath40-invariant ideal of @xmath249 . by (
* proposition 1 ) , the decomposition action of @xmath40 on @xmath249 is twisted over @xmath4 , and @xmath286 so follows from ( * ? ? ?
* proposition 11 ) .
now we have an @xmath40-equivariant commutative diagram @xmath287 where @xmath288 denotes the regular representation of @xmath249 .
this gives an @xmath40-equivariant commutative diagram @xmath289 also , we have a diagram @xmath290 which commutes by our construction of the bottom arrow in lemma [ reduced ] . since @xmath245 is an @xmath40-invariant ideal of @xmath248 , @xmath291 is an @xmath40-invariant ideal of @xmath249 , and we have @xmath292 by . since @xmath293 is surjective , @xmath294 is injective on ideals , and the lemma follows .
in this section we show that our res - ind duality is compatible with certain standard constructions . before discussing morita equivalence of coactions , we recall the concept of multiplier bimodules introduced in @xcite .
a _ multiplier _
@xmath295 of @xmath296 consists of an @xmath3-linear map @xmath297 and a @xmath96-linear map @xmath298 such that @xmath299 for @xmath111 , @xmath260 . the _ multiplier bimodule _
@xmath300 consists of all multipliers of @xmath95 .
an _ imprimitivity bimodule homomorphism _
@xmath301 consists of homomorphisms @xmath302 and @xmath303 and a bimodule map @xmath304 preserving the inner products : @xmath305 @xmath306 is called _ nondegenerate _ if @xmath307 and @xmath308 are nondegenerate .
significantly , this implies an ostensibly stronger form of nondegeneracy : [ iains - ib - hom - lem ] if @xmath309 is a nondegenerate imprimitivity bimodule homomorphism , then @xmath310 by symmetry , it suffices to show the second equality .
since @xmath311 , and @xmath307 is nondegenerate , we have @xmath312 .
now , since the range of the inner product @xmath313 spans @xmath3 , we have @xmath314 because we can factor @xmath311 , and @xmath315 , the pairing @xmath316 takes @xmath317 to @xmath318 , and we can deduce from our calculation that @xmath319 .
the other inclusion is trivial , so this establishes the lemma .
[ lem : nondegenerate ] let @xmath296 and @xmath320 be imprimitivity bimodules , and let @xmath321 be a nondegenerate homomorphism such that @xmath322 where @xmath323 is the linking algebra of @xmath95 , and similarly for @xmath119 .
then @xmath306 restricts on the corners to give a nondegenerate imprimitivity bimodule homomorphism @xmath324 .
since @xmath306 is nondegenerate , we have @xmath325 .
since @xmath326 we deduce that there are unique linear maps @xmath327 such that @xmath328 the algebraic properties of @xmath306 show that @xmath329 is an imprimitivity bimodule homomorphism of @xmath296 to @xmath330 , and nondegeneracy of @xmath331 and @xmath332 is inherited from nondegeneracy of @xmath306 .
we will need the following elementary fact about imprimitivity bimodule homomorphisms .
[ surjection ] if @xmath333 is a nondegenerate imprimitivity bimodule homomorphism , then the diagram @xmath334 commutes in the usual strong sense .
we must show @xmath335 as a right - hilbert @xmath3 @xmath318 bimodule .
of course , @xmath336 is isomorphic to @xmath337 .
define @xmath338 by @xmath339 the required properties of @xmath221 follow from the nondegeneracy of @xmath306 , and straightforward calculations showing that @xmath340 and @xmath341 for @xmath111 , @xmath259 , and @xmath342 . following ( * ? ? ?
* definition 3.3 ) ( see also @xcite , @xcite , @xcite ) , a _
coaction _
@xmath0 of @xmath1 on an imprimitivity bimodule @xmath296 is an imprimitivity bimodule homomorphism @xmath343 such that @xmath344 and @xmath345 are @xmath2-coactions , and satisfying @xmath346 as a consequence of the definition , we automatically have @xmath347 and @xmath348 for @xmath349 , @xmath350 . also , since by assumption @xmath351 and @xmath352 are nondegenerate @xmath2-homomorphisms , @xmath0 is automatically nondegenerate as an imprimitivity bimodule homomorphism . hence , by lemma [ iains - ib - hom - lem ] , we have @xmath353 when such a @xmath0 exists we say @xmath344 and @xmath345 are _ morita equivalent _ , and we call @xmath354 a _ morita equivalence _ of @xmath351 and @xmath352 . if @xmath4 is a closed normal subgroup of @xmath1 , then @xmath355 is a coaction of @xmath53 on @xmath95 , where @xmath356 is the canonical quotient map . a
_ morita equivalence _ between twisted coactions @xmath357 and @xmath358 is an @xmath344
@xmath345 morita equivalence @xmath354 such that @xmath359 in this case , for any closed normal subgroup @xmath4 of @xmath1 contained in @xmath39 , there are @xmath84 @xmath360 and @xmath93 @xmath361 imprimitivity bimodules @xmath362 and @xmath363 , respectively .
moreover , there is an action of @xmath39 on @xmath362 inducing a morita equivalence between the dual actions @xmath364 and @xmath365 , so there is an @xmath366 @xmath367 imprimitivity bimodule @xmath368 .
the next result shows corollary [ twisted - res - ind - cor ] is compatible with morita equivalence : [ me - res - ind - cube - thm ] if @xmath354 is a morita equivalence between nondegenerate twisted coactions @xmath357 and @xmath358 , and @xmath166 are closed normal subgroups of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath40 and one of the coactions , then the cube @xmath369{{{\operatorname{rep}}}b\times_{g / k}g / n } \arrow[2]{e } \arrow{s,- } \arrow{sw , t}{x\times_{g / k}g / n } \node[2]{{{\operatorname{rep}}}b\times_{g
/ k}g\times_r n } \arrow[2]{s , r}{{{\operatorname{ind } } } } \arrow{sw , t}{x\times_{g / k}g\times_r
n}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / n } \arrow[2]{e } \arrow[2]{s , l}{{{\operatorname{res } } } } \node { } \arrow{s , l}{{{\operatorname{res } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n } \arrow[2]{s , r,3}{{{\operatorname{ind}}}}\\ \node[2]{{{\operatorname{rep}}}b\times_{g / k}g / h } \arrow{e,- } \arrow{sw , b}{x\times_{g / k}g / h } \node { } \arrow{e } \node{{{\operatorname{rep}}}b\times_{g / k}g\times_r h } \arrow{sw , b}{x\times_{g / k}g\times_r h}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \end{diagram}\ ] ] commutes in the usual strong sense . first note that by ( * ? ? ?
* theorem 5.3 ) , if mansfield imprimitivity works for @xmath40 and one of the coactions , it works for the other , so the above cube makes sense
. because all the horizontal arrows are bijections , we need only show commutativity of three of the vertical faces , as well as the top and bottom .
the front and back commute by corollary [ twisted - res - ind - cor ] .
the bottom face is just the top face with @xmath4 replaced by @xmath40 ; we show the top commutes .
the untwisted version ( i.e. , with @xmath370 ) @xmath371 is ( * ? ? ?
* proposition 4.5 ) . even though they use reduced coactions and require @xmath4 to be amenable , their arguments carry over to our setting since we assume that mansfield imprimitivity works , as pointed out in @xcite .
so , it only remains to show the ideals in diagram match up and appeal to corollary [ quotient ] .
the ideals match up along the top and bottom by ( * ? ? ?
* theorem 4.4 ) , and along the left side by ( * ? ? ?
* corollary 3.3 ) , and this is enough , since the top and bottom are morita equivalences . for the left face , we use the following general lemma : [ co - res - res - lem ] let @xmath354 be a morita equivalence between coactions @xmath344 and @xmath345 , and let @xmath4 be a closed normal subgroup of @xmath1 . then the diagram @xmath372 commutes .
we aim to apply lemma [ surjection ] , so we need a nondegenerate imprimitivity bimodule homomorphism @xmath373 with @xmath374 let @xmath375 be the linking algebra for @xmath95 ; then there is a coaction @xmath376 of @xmath1 on @xmath375 . by [ er , appendix ] , we have @xmath377 and @xmath378 by lemma [ lem : nondegenerate ] , the nondegenerate homomorphism @xmath379 restricts on the corners to a nondegenerate imprimitivity bimodule homomorphism @xmath380 since the restrictions of @xmath381 to the diagonal corners @xmath7 and @xmath382 agree with @xmath383 and @xmath384 , respectively , we are done . returning to the proof of theorem [ me - res - ind - cube - thm ] , the untwisted version of the left - hand face of diagram follows from lemma [ co - res - res - lem ] .
we need only show that the twisting ideals in each crossed product in diagram match up , and apply corollary [ quotient ] . that the ideals match up across the horizontal arrows
is shown in ( * ? ? ?
* corollary 3.3 ) . that the ideals match up along the vertical arrows follows from the proof of corollary [ twisted - res - ind - cor ] .
this completes the proof of theorem [ me - res - ind - cube - thm ] in the proof of theorem [ me - res - ind - cube - thm ] , we were able to deduce commutativity of the right face of from the other five faces .
this is a special case of a general `` @xmath37-@xmath37 '' diagram for actions which is related to ( * ? ? ?
* theorem 3 ) and ( * ? ? ? * proposition 4.1.2 ) .
we now show that corollary [ twisted - ind - res - cor ] is compatible with morita equivalence : [ me - ind - res - cube - thm ] if @xmath354 is a morita equivalence between nondegenerate twisted coactions @xmath357 and @xmath358 , and @xmath166 are closed normal subgroups of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath40 and one of the coactions , then the cube @xmath385{{{\operatorname{rep}}}b\times_{g / k}g / n } \arrow[2]{e } \arrow{sw , t}{x\times_{g / k}g /
n } \node[2]{{{\operatorname{rep}}}b\times_{g / k}g\times_r n } \arrow{sw , t}{x\times_{g / k}g\times_r
n}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / n } \arrow[2]{e } \node { } \arrow{n , l}{{{\operatorname{ind } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n}\\ \node[2]{{{\operatorname{rep}}}b\times_{g / k}g / h } \arrow{n,- } \arrow{e,- } \arrow{sw , b}{x\times_{g / k}g / h } \node { } \arrow{e } \node{{{\operatorname{rep}}}b\times_{g / k}g\times_r h } \arrow[2]{n , r}{{{\operatorname{res } } } } \arrow{sw , b}{x\times_{g / k}g\times_r h}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow[2]{n , l}{{{\operatorname{ind } } } } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \arrow[2]{n , r,3}{{{\operatorname{res } } } } \end{diagram}\ ] ] commutes in the usual strong sense .
the front and back faces commute by corollary [ twisted - ind - res - cor ] , and the top and bottom faces are the same as in , so it suffices to show commutativity of the right face . for this we use the following lemma : [ rt - face - res - lem ] let @xmath386 be a morita equivalence between actions @xmath387 and @xmath388 , and let @xmath4 be a closed subgroup of @xmath40
. then the diagram @xmath389 commutes .
we aim to apply lemma [ surjection ] , so we need a nondegenerate imprimitivity bimodule homomorphism @xmath390 with @xmath391 let @xmath392 be the linking algebra for @xmath119 ; then there is an action @xmath393 of @xmath40 on @xmath392 . by [ er , appendix ]
, we have @xmath394 and @xmath395 by lemma [ lem : nondegenerate ] , the nondegenerate homomorphism @xmath396 restricts on the corners to a nondegenerate imprimitivity bimodule homomorphism @xmath397 since the restrictions of @xmath398 to the diagonal corners @xmath399 and @xmath400 agree with @xmath401 and @xmath402 , respectively , we are done . returning to the proof of the right - hand face of diagram , we need only show that the kernels of the regular representations of each crossed product in diagram match up , and apply corollary [ quotient ] . that the kernels match up across the horizontal arrows is shown in @xcite .
the kernels match up along the vertical arrows because the regular representation of @xmath4 is quasi - equivalent to the regular representation of @xmath40 restricted to @xmath4 .
alternatively , we can argue as follows : we must show that @xmath403 by green induction in stages , @xmath404 since for @xmath4 normal , @xmath405 is @xmath40-invariant , equation gives us @xmath406 this completes the proof of theorem [ me - ind - res - cube - thm ] . by ( * ? ? ?
* proposition 2.3 ) , morita equivalence respects nondegeneracy of coactions .
hence , in theorems [ me - res - ind - cube - thm]and [ me - ind - res - cube - thm ] , one coaction is in fact nondegenerate if and only if the other one is .
this time , we were able to deduce commutativity of the left face of from the other five faces . written in terms of tensor products , @xmath407 where we have used self - explanatory notation to distinguish the bimodules for @xmath3 and @xmath96
the special case @xmath168 is ( * ? ? ?
* theorem 4.4 ) .
a direct proof is presumably possible , although probably quite tedious .
we turn to inflation of coactions : if @xmath39 is a closed normal subgroup of @xmath1 and @xmath75 is a coaction , composing with the natural embedding of @xmath408 in @xmath409 gives a coaction @xmath410 , called an _ inflated _ coaction .
@xmath411 is trivially twisted over @xmath80 by @xmath412 , and ( * ? ? ?
* example 2.14 ) gives a natural isomorphism of @xmath413 onto @xmath414 which takes @xmath415 to @xmath416 .
the next two theorems show corollaries [ twisted - res - ind - cor ] and [ twisted - ind - res - cor ] are compatible with inflation : [ inf - res - ind - cube - thm ] if @xmath166 are closed normal subgroups of @xmath1 contained in @xmath39 , and @xmath75 is a nondegenerate coaction such that mansfield imprimitivity works for @xmath40 and @xmath76 , then the cube @xmath417{{{\operatorname{rep}}}a\times
k / n } \arrow[2]{e } \arrow{s,- } \node[2]{{{\operatorname{rep}}}a\times k\times_r n } \arrow[2]{s , r}{{{\operatorname{ind}}}}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g
/ n } \arrow{ne , t}{\cong } \arrow[2]{e } \arrow[2]{s , l}{{{\operatorname{res } } } } \node { } \arrow{s , l}{{{\operatorname{res } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n } \arrow{ne , t}{\cong } \arrow[2]{s , r,3}{{{\operatorname{ind}}}}\\ \node[2]{{{\operatorname{rep}}}a\times k / h } \arrow{e,- } \node { } \arrow{e } \node{{{\operatorname{rep}}}a\times k\times_r h}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow{ne , b}{\cong } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \arrow{ne , b}{\cong } \end{diagram}\ ] ] commutes in the usual strong sense . first , by ( * ? ? ?
* theorem 5.4 ) , mansfield imprimitivity works for @xmath40 and the ( nondegenerate ) inflated coaction @xmath77 , so the above diagram makes sense .
the front and back faces commute by theorem [ plain - res - ind - thm ] .
we show the top commutes ; the bottom is essentially the same .
* proposition 4.8 and remark 4.9 ) give an imprimitivity bimodule homomorphism @xmath418 hence , lemma [ surjection ] gives a commutative diagram @xmath419 to pass from to the top of , we need to show the appropriate ideals @xmath197 in @xmath7 , @xmath420 in @xmath9 , and @xmath421 in both @xmath422 and @xmath423 match up .
the ideals match up along the top ( by ( * ? ? ?
* theorem 4.4 ) ) , bottom ( trivially ) , and left ( since @xmath424 ) , and this is enough . finally , for the left face of , the diagram @xmath425 of nondegenerate homomorphisms commutes , since the diagram @xmath426 does .
hence , the required diagram @xmath427 of right - hilbert bimodules commutes as well .
[ inf - ind - res - cube - thm ] if @xmath166 are closed normal subgroups of @xmath1 contained in @xmath39 , and @xmath75 is a nondegenerate coaction such that mansfield imprimitivity works for @xmath40 and @xmath76 , then the cube @xmath428{{{\operatorname{rep}}}a\times k / n } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times k\times_r n}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / n } \arrow{ne , t}{\cong } \arrow[2]{e } \node { } \arrow{n , l}{{{\operatorname{ind } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n } \arrow{ne , t}{\cong}\\ \node[2]{{{\operatorname{rep}}}a\times k / h } \arrow{n,- } \arrow{e,- } \node { } \arrow{e } \node{{{\operatorname{rep}}}a\times k\times_r h } \arrow[2]{n , r}{{{\operatorname{res}}}}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow[2]{n , l}{{{\operatorname{ind } } } } \arrow{ne , b}{\cong } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \arrow[2]{n , r,3}{{{\operatorname{res } } } } \arrow{ne , b}{\cong } \end{diagram}\ ] ] commutes in the usual strong sense .
the front and back faces commute by theorem [ plain - ind - res - thm ] , and the top and bottom faces are the same as in . since the horizontal maps in the right face come from equivariant isomorphisms , the commutativity of this face follows from lemma [ rt - face - res - lem ] as in the proof of theorem [ me - ind - res - cube - thm ] . by (
* proposition 2.1 ) , in theorems [ inf - res - ind - cube - thm]and [ inf - ind - res - cube - thm ] , @xmath411 is in fact nondegenerate if and only if @xmath76 is .
moreover , when the coactions are nondegenerate , mansfield imprimitivity works for @xmath40 and @xmath411 if and only if it works for @xmath40 and @xmath76 .
as before , we get commutativity of a sixth face of diagrams and from the other five faces . the left face of is new ; the special case @xmath168 follows from ( * ? ? ?
* theorem 4.7 ) . finally , we show that our res - ind duality is compatible with the stabilization trick of @xcite , as adapted to full coactions in @xcite .
let @xmath78 be a nondegenerate twisted coaction such that mansfield imprimitivity works for @xmath0 and @xmath39 itself ; equivalently , such that @xmath0 is normal ( * ? ? ?
* lemma 3.6 )
. then mansfield imprimitivity works for @xmath0 and any closed normal subgroup of @xmath1 contained in @xmath39 by ( * ? ? ?
* theorem 5.2 ) ( or by ( * ? ? ?
* lemma 3.2 ) , since @xmath0 is normal ) .
* theorem 3.1)shows that the twisted coaction @xmath78 is morita equivalent to the ( nondegenerate ) inflated twisted coaction @xmath429 , so by ( * ? ? ?
* theorems 5.3 and 5.4 ) , mansfield imprimitivity works for @xmath430 and any closed normal subgroup of @xmath1 contained in @xmath39 .
thus we can chain together theorems [ me - res - ind - cube - thm ] and [ inf - res - ind - cube - thm ] , and similarly theorems [ me - ind - res - cube - thm ] and [ inf - ind - res - cube - thm ] , to obtain : [ stab - cube - thm ] let @xmath78 be a nondegenerate normal twisted coaction , and let @xmath166 be closed normal subgroups of @xmath1 contained in @xmath39 .
further let @xmath431 , which carries the double dual coaction of @xmath39 .
then the cubes @xmath432{{{\operatorname{rep}}}b \times k / n } \arrow[2]{e } \arrow{s,- } \node[2]{{{\operatorname{rep}}}b\times k\times_r
n } \arrow[2]{s , r}{{{\operatorname{ind}}}}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / n } \arrow{ne } \arrow[2]{e } \arrow[2]{s , l}{{{\operatorname{res } } } } \node { } \arrow{s , l}{{{\operatorname{res } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n } \arrow{ne } \arrow[2]{s , r,3}{{{\operatorname{ind}}}}\\ \node[2]{{{\operatorname{rep}}}b\times k / h } \arrow{e,- } \node { } \arrow{e } \node{{{\operatorname{rep}}}b\times k\times_r h}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow{ne } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \arrow{ne } \end{diagram}\ ] ] and @xmath433{{{\operatorname{rep}}}b\times k / n } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}b\times k\times_r n}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / n } \arrow{ne } \arrow[2]{e } \node { } \arrow{n , l}{{{\operatorname{ind } } } } \node{{{\operatorname{rep}}}a\times_{g / k}g\times_r n } \arrow{ne}\\ \node[2]{{{\operatorname{rep}}}b\times k / h } \arrow{n,- } \arrow{e,- } \node { } \arrow{e } \node{{{\operatorname{rep}}}b\times k\times_r h } \arrow[2]{n , r}{{{\operatorname{res}}}}\\ \node{{{\operatorname{rep}}}a\times_{g / k}g / h } \arrow[2]{n , l}{{{\operatorname{ind } } } } \arrow{ne } \arrow[2]{e } \node[2]{{{\operatorname{rep}}}a\times_{g / k}g\times_r h } \arrow[2]{n , r,3}{{{\operatorname{res } } } } \arrow{ne } \end{diagram}\ ] ] both commute in the usual strong sense .
again , the left face of is new ; when @xmath168 it reduces to ( * ? ? ?
* theorem 4.7 ) .
in this section , as a sample application of our @xmath36-@xmath37 duality , we generalize some results of gootman and lazar @xciteconcerning restriction and induction of ideals in crossed products by coactions of amenable groups , to nonamenable groups .
nilsen @xcite has recently proved similar results , using different , representation - theoretic techniques .
our methods , based on our res - ind duality results , appear to be more efficient than those of gootman and lazar . with some additional effort
, we could further generalize to the setting of intermediate twisted crossed products by coactions of nonamenable groups , but we feel that to do so at this point would only muddy the waters .
first , recall from ( * ? ? ?
* proposition 9 ) that a nondegenerate homomorphism @xmath434 gives rise to maps @xmath435 and @xmath436 between spaces of ideals . by definition ,
@xmath437 , and @xmath438 is the ideal generated by @xmath439 . if in addition @xmath95 is a @xmath96 @xmath113 imprimitivity bimodule , then by definition we have the following commutative diagram : @xmath440 green defines a map @xmath441 by requiring the diagram @xmath442 to commute .
( actually , we have abstracted green s definition a bit in order to bring the properties of the maps into high relief . )
so , @xmath42 is to @xmath37 as @xmath41 is to @xmath36 , and in fact @xmath41 is a special case of @xmath42 , just as @xmath36 is a special case of @xmath37 . * @xmath443 ; @xmath444 ; * @xmath445 ; @xmath446 ; * @xmath37 , @xmath36 , @xmath42 and @xmath41 are order - preserving * @xmath37 and @xmath36 preserve arbitrary intersections ; * @xmath42 and @xmath41 preserve arbitrary sups ; * @xmath447 ; @xmath448 ; * @xmath449 ; @xmath450 ; * @xmath451 is the smallest ideal @xmath452 of @xmath96 such that @xmath453 ; * @xmath454 is the largest ideal @xmath455 of @xmath3 such that @xmath456 ; * @xmath457 is the smallest ideal @xmath458 of @xmath113 such that @xmath459 ; * @xmath460 is the largest ideal @xmath455 of @xmath3 such that @xmath461 now consider an action @xmath140 .
green applies the above machinery to the canonical map @xmath462 to get maps @xmath463 and @xmath464 he then applies the same machinery to the canonical map @xmath465 and the @xmath466 @xmath96 imprimitivity bimodule @xmath95 from his imprimitivity theorem to get maps @xmath467 and @xmath468 ( green uses @xmath469 as the imprimitivity algebra , rather than the isomorphic co - crossed product @xmath466 . ) * @xmath454 and @xmath474 are @xmath1-invariant ; * @xmath475 is the largest @xmath1-invariant ideal of @xmath96 contained in @xmath245 ; * @xmath476 is the smallest @xmath1-invariant ideal of @xmath96 containing @xmath245 ; * @xmath245 is @xmath1-invariant if and only if @xmath477 .
we will have to adapt green s machinery to reduced crossed products .
let @xmath140 be an action , and let @xmath478 be the regular representation .
the commutative diagram @xmath479{m(b\times_{r}g ) } \end{diagram}\ ] ] gives in turn commutative diagrams @xmath480{{\mathcal i}(b\times_{r}g ) } \end{diagram}\ ] ] and @xmath481{{\mathcal i}(b\times_{r}g ) , } \arrow{nw , b}{{{\operatorname{res}}}_{\rho\circ i_b } } \arrow{n , r}{\rho^ * } \end{diagram}\ ] ] the horizontal @xmath41 and @xmath36 are what green uses ; we will need the diagonal ones , together with their associated maps @xmath482 and @xmath483 .
similarly , the commutative diagram @xmath484 gives in turn commutative diagrams @xmath485{s , l}{\cong } \arrow{se , b}{\cong } \node[2]{{\mathcal i}(b\times g ) } \arrow[2]{w , t}{{\operatorname{ex } } } \arrow{sw , b}{{\operatorname{sub } } } \arrow[2]{s , r}{\rho_*}\\ \node[2]{{\mathcal i}(b ) } \arrow{sw , t}{\cong}\\ \node{{\mathcal i}(b\times_{r}g\times g ) } \node[2]{{\mathcal i}(b\times_{r}g ) } \arrow[2]{w , b}{{\operatorname{ex } } } \arrow{nw , t}{{\operatorname{sub } } } \end{diagram}\ ] ] and @xmath486{s , l}{\cong } \arrow{se , b}{\cong } \arrow[2]{e , t}{{{\operatorname{res } } } } \node[2]{{\mathcal i}(b\times
g)}\\ \node[2]{{\mathcal i}(b ) } \arrow{sw , t}{\cong } \arrow{ne , b}{{{\operatorname{ind } } } } \arrow{se , t}{{{\operatorname{ind}}}}\\ \node{{\mathcal i}(b\times_{r}g\times g ) } \arrow[2]{e , b}{{{\operatorname{res } } } } \node[2]{{\mathcal i}(b\times_{r}g ) . } \arrow[2]{n , r}{\rho^ * } \end{diagram}\ ] ] diagrams and , together with the @xmath1-equivariance and injectivity of @xmath487 , imply that all of green s results in ( * ? ? ?
* lemma 10(ii ) and proposition 11)carry over to reduced crossed products . in particular , those results summarized in proposition [ green - sum - prop ] carry over ; we will cite these as simply @xcite without further comment .
we now consider a nondegenerate normal coaction @xmath48 .
we would like to apply the general abstract nonsense of the beginning of this section to obtain maps @xmath42 and @xmath41 among the spaces @xmath488 and @xmath489 to go along with the maps @xmath36 and @xmath37 .
we use the canonical map @xmath490 to get a map @xmath491 ( in fact , we ve already used this map in diagram in the case @xmath492 . )
similarly , we use the canonical map @xmath493 and the @xmath494 @xmath3 imprimitivity bimodule @xmath119 from mansfield s imprimitivity theorem to get a map @xmath495 in this situation , theorem [ plain - res - ind - thm ] in the case @xmath168 and @xmath496 gives us the commutativity of the following diagram : @xmath497 by the same token , theorem [ plain - ind - res - thm ] reduces to the commutativity of @xmath498 which is just the definition of @xmath37 .
[ sub - ex - thm ] let @xmath78 be a nondegenerate twisted coaction , and let @xmath166 be closed normal subgroups of @xmath1 contained in @xmath39 such that mansfield imprimitivity works for @xmath40 and @xmath0 .
then the following diagrams commute : @xmath499 @xmath500 the theorem follows immediately from corollaries [ twisted - res - ind - cor ] and [ twisted - ind - res - cor ] , together with elementary properties of @xmath36 , @xmath41 , @xmath42 and @xmath37 .
for example , by @xcite , corollary [ twisted - ind - res - cor ] gives a commutative diagram @xmath501 now fix @xmath502 . by proposition [ gan - prop ]
, @xmath457 is the smallest ideal @xmath245 of @xmath503 such that @xmath504 ; hence @xmath505 is the smallest ideal @xmath39 of @xmath506 such that @xmath507 .
again using proposition [ gan - prop ] , this implies that @xmath508 so diagram commutes . for the rest of this section , we will only need the special case @xmath168 and @xmath509 of theorem [ sub - ex - thm ] . for reference ,
this gives the commutativity of the following triangles : @xmath510 @xmath511 ( again , diagram has simply reduced to the definition of @xmath42 . )
we are now able to generalize some of the coaction results of @xcite .
let @xmath48 be a nondegenerate normal coaction .
adapting ( * ? ? ?
* definition 2.4 ) to full coactions , we say that an ideal @xmath123 of @xmath3 is _ @xmath1-invariant _ if @xmath512 where @xmath513 is the quotient map and @xmath514 is the left regular representation .
we do not know if this implies that @xmath0 restricts to a coaction on @xmath123 ( unless @xmath1 is amenable ) , but we do get a coaction @xmath515 of @xmath1 on @xmath132 . since we will not need this fact here , we omit the proof
. green shows ( * ? ? ?
* proposition 13 ) that for an action @xmath140 with @xmath1 amenable , @xmath519 on @xmath473 .
this is definitely not true for non - amenable @xmath1 : @xmath520 , while @xmath521 if and only if @xmath522 .
for all we know , even if we pass to reduced crossed products , @xmath41 and @xmath37 can be different on @xmath473 . even for trivial actions , the question reduces to the unsolved problem of whether there exists a locally compact group @xmath1 for which @xmath46 is not exact .
we do not know how to prove the companion result for @xmath42 and @xmath36 ( generalizing ( * ? ? ?
* proposition 3.14(iv ) ) ) ; it certainly does nt follow from @xcite and duality as in the proof of lemma [ ex = ind - lem ] .
* @xmath543 is the smallest @xmath1-invariant ideal of @xmath3 containing @xmath123 ; * @xmath544 is the largest @xmath1-invariant ideal of @xmath8 contained in @xmath245 ; * @xmath545 is the smallest @xmath1-invariant ideal of @xmath8 containing @xmath245 .
\(i ) let @xmath39 be the intersection of all invariant ideals of @xmath3 containing @xmath123 ; since @xmath516 is closed under intersections , @xmath39 is invariant and hence is the smallest such ideal of @xmath3 .
now @xmath543 is invariant , and @xmath443 because @xmath526 is increasing ; so we get @xmath546 taking @xmath41 and using proposition [ gan - prop](vi ) , we have @xmath547 so equality holds throughout .
in particular , @xmath548 since @xmath41 is injective on invariant ideals , we get @xmath549 | consider a coaction @xmath0 of a locally compact group @xmath1 on a @xmath2-algebra @xmath3 , and a closed normal subgroup @xmath4 of @xmath1 .
we prove , following results of echterhoff for abelian @xmath1 , that mansfield s imprimitivity between @xmath5 and @xmath6 implements equivalences between mansfield induction of representations from @xmath7 to @xmath8 and restriction of representations from @xmath9 to @xmath8 , and between restriction of representations from @xmath8 to @xmath7 and green induction of representations from @xmath8 to @xmath9 .
this allows us to deduce properties of mansfield induction from the known theory of ordinary crossed products . |
the mesic nuclear bound states have been considered to be interesting systems and studied both theoretically and experimentally @xcite .
so far , one - nucleon transfer reactions have been mainly considered and successfully used to produce the nuclear bound states of meson which is lighter than nucleon in the recoilless kinematics @xcite .
recently , ikeno _ et al .
studied the two nucleon pick - up reaction on the @xmath2 target for the formation of heavy meson bound states with the @xmath3 particle @xcite .
we develop this study in this article and improve the theoretical method in the following four points ; + ( i ) the distortion effects of the projectile and ejectile are taken into account , + ( ii ) the elementary cross section is evaluated phenomenologically and the absolute value of the formation cross section is reported , + ( iii ) the realistic @xmath3 density distribution is used to calculate the meson-@xmath3 bound states @xcite , + ( iv ) the recoil effects of the formation reaction are considered . + we show some of the numerical results here . _
we apply the effective number approach to evaluate the formation rate of the bound systems in the two nucleon pick - up @xmath4 reactions as in ref . @xcite . in the effective number approach ,
the formation cross section by the @xmath4 reaction can be written as , @xmath5 where @xmath6 is the elementary cross section of the @xmath7 meson production , and @xmath8 the width of the @xmath7 bound states .
@xmath9 is the effective deuteron number .
the all combinations of the final states with @xmath10 and @xmath7 are indicated by @xmath11 and are summed up to evaluate the inclusive cross section .
the energy transfer of the reaction @xmath12 in the laboratory frame is defined as , @xmath13 where @xmath14 is the emitted deuteron kinetic energy , @xmath15 the incident photon momentum , and @xmath16 the @xmath7 meson mass .
the @xmath7 meson binding energy @xmath17 and the deuteron separation energy @xmath18 are determined for each combination of the bound level of @xmath7 meson and the excited level of the daughter nucleus . here
, we neglect the recoil energy of the nucleus in this expression since we consider the kinematics close to the recoilless condition .
the details of the theoretical formula will be reported in ref .
we show in fig . [
fig : neff - e1 ] the calculated effective numbers with @xmath19}$ ] case as an example . for each state
, the recoilless condition is satisfied at @xmath20}$ ] for the @xmath21 state , @xmath22}$ ] for the @xmath23 state , and @xmath24}$ ] for the @xmath25 state , respectively .
we can see from the figure that the effective number for the @xmath23 state formation takes the maximum value at the recoilless energy as expected . on the other hand
the effective number for the @xmath25 state takes the smallest value at the recoilless kinematics because of the quasi - orthogonal condition between bound @xmath7 and deuteron wave functions . the effective number for the @xmath21 state does not show the clear @xmath26 dependence in this energy region because of the compact wave function of the deeply bound states . here
, the bound @xmath7 meson wave functions in the final state are calculated by solving the klein - gordon equation with the optical potential @xmath27 written as , @xmath28 where @xmath29 and @xmath30 are the parameters which determine the real and imaginary potential strength , respectively .
the normal nuclear density @xmath31 is fixed to be @xmath32}$ ] . }
the quantum numbers of the @xmath7 bound states are indicated in the figure.,width=302 ] we show the calculated formation spectra of the @xmath33 bound states in fig . [
fig : cs - v0_150 ] for the @xmath34}$ ] case for four incident photon energies .
the potential parameter of the absorptive part is assumed to be @xmath35}$ ] .
the horizontal axis indicates the excitation energy of the @xmath7 mesic nucleus and the peaks in @xmath36 region mean the formation of the bound states .
reaction for the formation of the @xmath7 bound state in @xmath3 are plotted as functions of the excitation energy of the @xmath7 mesic nucleus for four incident photon energies as indicated in the figure .
the parameter of the real part of the optical potential is assumed to be @xmath34}$ ] and that of the imaginary part to be @xmath35}$ ] for the solid lines .
the contribution from the quasi - free @xmath7 production is not included in these spectra . ]
we have reported the first calculated results of the formation of the @xmath1 mesic nucleus in the @xmath0 reaction .
we show the results for @xmath19}$ ] case as an example .
the strength of the @xmath7-nucleus potential is still controversial and has not been determined well . in refs .
@xcite , where the formation of the @xmath7 mesic nucleus was considered for the first time , the strength of the @xmath7-nucleus potential was evaluated by njl model to be around @xmath37}$ ] .
actually the recent evaluation based on the chiral symmetry restoration @xcite also indicates the strong attractive and less absorptive potential . the potential strength considered in this article
is based on these studies .
another calculation based on the chiral unitary model @xcite reveals the sensitivity of the potential to the coupling strength of the singlet @xmath38 to the octet baryons . on the other hand ,
the latest experimental data indicate the small @xmath39 scattering length @xcite and the shallow @xmath7-nucleus potential @xcite .
theoretical evaluation in ref .
@xcite also indicates weak attractive potential .
therefore , we believe that the further studies reported in this article in addition to those proposed in refs . @xcite are much important to develop this field further .
the comprehensive results of the @xmath4 reaction including those with the shallow @xmath7-nucleus potential will be reported in ref .
@xcite soon .
we acknowledge the fruitful discussions with h. fujioka and t. ishikawa from the beginning of this research .
we would like to thank e. hiyama for providing us the realistic density of the @xmath3 particle .
this work is partly supported by the grants - in - aid for scientific research no .
24540274 ( s.h . ) and no .
15h06413 ( n.i . ) in japan . | we have investigated the @xmath0 reaction theoretically for the formation of the @xmath1 mesic nucleus .
we have reported the numerical results in this article . |
efforts to design scalable quantum computers ( qcs ) @xcite and other new quantum technologies @xcite are rekindling interest in the phenomenology of open quantum system dynamics .
while effects of atom - radiation field interactions @xcite have been extensively studied , the consequences of interaction with a complex environment are just beginning to be explored .
solid and condensed phase environments bequeath a diversity of controllable interactions for scalable implementation of new quantum technologies , and thus understanding the open system dynamics induced by such complex environments is of particular interest .
recent studies have reported a number of interesting effects such as suppression of decoherence with increasing bath chaos @xcite and large coherent shifting @xcite .
manifestation of chaos @xcite , or at least non - integrability , is a defining and unavoidable feature of complex environments .
evidence suggests that condensed phase dynamics is generally chaotic : dynamics of a colloidal particle in water @xcite , and both vibrational dynamics @xcite of si and its electronic structure @xcite have been shown to be chaotic .
anharmonic corrections are known to be important in the study of phonons and essential for an understanding of heat transport @xcite .
thus , the chaotic bath model provides a better representation for the condensed or solid state environment than standard uncoupled oscillator bath models .
chaotic bath models are substantially more difficult to simulate than the radiation field . at high temperatures random matrix theories @xcite , or classical trajectory simulations such as the wigner method @xcite , can be used .
redfield theory @xcite and its generalizations @xcite can also be employed .
but at the very low temperatures relevant to quantum computing technologies the options are more limited .
there is a recently developed approximate master equation @xcite which appears to be accurate in this low temperature regime .
however , as with any approximate theory its predictions need to be otherwise confirmed . in this manuscript
we derive a simple formula for the reduced density which captures all the important features of open system dynamics induced by chaotic environments in the low temperature regime .
the equation is of the kraus operator sum representation ( osr ) @xcite form which automatically satisfies all the required conservation laws for the reduced density , i.e. hermiticity , complete positivity and norm conservation .
our kraus osr@xcite provides a formally exact equation for the reduced density in terms of explicit representations of the kraus operators . aside from chaos ,
the only required property of the bath hamiltonian is that it be of large thermodynamic dimension . as a first test
, we compare the predictions of the chaotic kraus osr@xcite to exact calculations for a model of a single qubit interacting with an isolated qc with static internal flaws . while our derivation assumes a bath of thermodynamic dimensions , our numerical test shows that very accurate results can be obtained even for a chaotic bath of just ten qubits .
organization of this manuscript is as follows .
a derivation of the kraus osr@xcite , in three steps , is given in section 2 . in section 3
we discuss our chaotic spin - bath model .
in section 4 we test the osr against exact numerical results . in section 5
we discuss our results .
consider the following total hamiltonian for the bipartite closed system @xmath1 where @xmath2 is the system hamiltonian and @xmath3 is the bath hamiltonian .
@xmath4 is the interaction hamiltonian where @xmath5 and @xmath6 are the system and bath coupling operators , respectively .
our derivation consists of three parts .
first we review the construction of the osr assuming a complete bath eigenbasis ( i.e @xmath7 and @xmath8 ) .
next we show that the off - diagonal ( @xmath9 ) matrix elements of bath coupling operator @xmath10 vanish for a chaotic bath in the thermodynamic limit . finally , we use this fact to show that the resulting kraus osr takes a particularly simple and potentially useful form .
we consider an uncorrelated initial state for the whole system @xmath11 where @xmath12 is an arbitrary initial state of the system and @xmath13 is an initial bath state of canonical form , i.e. @xmath14 the subsequent states will be given by the exact solutions @xmath15 of the liouville - von neumann equation @xmath16.\ ] ] here @xmath17 .
the exact reduced density of the system is then formally obtained by tracing over the bath degrees of freedom , i.e , @xmath18 since the bath eigenbasis is complete , and the initial canonical bath density is already diagonal in eigenstates of @xmath3 , performing the partial trace in the same basis yields @xmath19 which in turn can be written in more compact kraus osr form @xmath20 where the kraus operators are defined by @xmath21 here @xmath22 are the initial populations of the bath eigenstates .
note that @xmath23 .
this form of the kraus osr is of formal interest only .
although it is exact and so satisfies all the required conservation laws ( i.e. the hermiticity , complete positivity , and norm ) for the subsystem density it is not practical for computational purposes .
first , the explicit forms of the kraus operators ( [ kraus ] ) are not known in general , because the kraus osr ( [ kraus ] ) is not unique . and
secondly , the double sum in ( [ osr ] ) would make use of the osr impractical even if simple forms for ( [ kraus ] ) were known .
we will now show that in the limit of a chaotic bath of thermodynamic dimension , explicit forms for ( [ kraus ] ) can be supplied and that the double sum reduces to a unique single sum . using the bath eigenvalues and eigenstates , the total hamiltonian ( [ ham ] )
can alternatively be written as @xmath24 where @xmath25 are the bath coupling operator matrix elements defined above in eq .
( [ bjk ] ) , and @xmath26 and @xmath27 are again the bath eigenenergies and eigenstates .
decoherence and open system dynamical effects are mediated by the matrix elements @xmath25 and by operators @xmath5 and @xmath28 .
we will focus on the properties of @xmath25 .
now we assume that @xmath3 is chaotic ( both quantally and classically ) and that the wigner function@xcite of @xmath6 has a classical limit .
coordinate and momentum bath coupling operators , for example , have wigner functions with classical limits .
let @xmath29 denote the number of bath degrees of freedom .
we will eventually send @xmath30 to attain the thermodynamic limit .
quantum chaos has a number of important consequences for matrix elements and spectra@xcite .
we will now show that off - diagonal matrix elements vanish for thermodynamic baths .
we begin with an exact formula for the off - diagonal matrix elements of an arbitrary quantum system : @xmath31 this is obtained by combining the exact equations ( 50 ) and ( 119 ) of @xcite . here
@xmath32 is the wigner function@xcite of operator @xmath6 and @xmath33 is a scaled wigner function for eigenstate @xmath34 .
it is well known@xcite that @xmath35 has a well defined classical limit if the classical hamiltonian @xmath36 is chaotic .
[ the wigner function of @xmath37 may differ from the classical hamiltonian by small corrections which vanish with planck s constant , but we ignore this distinction in our discussion . ]
we employ a notation where @xmath38 , where @xmath39 are the momenta and @xmath40 the coordinates .
@xmath41 is the @xmath42 dimensional symplectic matrix . here
@xmath43 and @xmath44 has the dimensions of action .
now , note that @xmath45 where @xmath46 is the microcanonical entropy and @xmath47 is the boltzmann constant .
now since entropy is extensive , @xmath46 scales linearly with @xmath29 for sufficiently large @xmath29 .
consequently , by lhpital s rule , it follows that @xmath48 scales exponentially with @xmath29 .
finally , it follows that @xmath49 is constant , or at most grows weakly with @xmath29 since the energies @xmath26 also grow with @xmath29 .
the wigner functions @xmath35 and @xmath32 and the associated integrals in ( [ bij2 ] ) have well defined classical@xcite and therefore thermodynamic limits . in the classical limit ,
for example , these terms reduce to @xmath50 which increases slowly with @xmath29 .
( here @xmath51 is the classical canonical average of the squared classical limit of the wigner function for @xmath6 .
note that @xmath52 and so @xmath51 would normally scale linearly with @xmath29 .
@xmath53 is only weakly dependent on @xmath29 .
note also that @xmath54 should be understood as a distribution with a small but nonzero width . )
thus , by lhpital s rule , the magnitude of @xmath55 in the thermodynamic limit is determined by the factor @xmath56 .
consequently , if @xmath57 , which will always be true for sufficiently large @xmath29 , then @xmath58 exponentially fast as @xmath59 . note that the requirement is @xmath57 , not the stronger semiclassical @xmath60 .
hence , the off - diagonal matrix elements of @xmath6 vanish for a chaotic bath of thermodynamic dimension .
they also decay exponentially fast with increasing @xmath29 , and so they may be neglected even for quite small baths . in the semi - classical limit , the scaling , @xmath61 has long been known in the quantum chaos literature@xcite , but its consequences in the thermodynamic limit , and for the kraus osr in particular , have not been appreciated .
as we have shown , the semi - classical limit @xmath62 is not a necessary condition for the vanishing of the off - diagonals , @xmath63 , for a chaotic bath of thermodynamic dimension .
it is also worth noting that similar formulas arise in other contexts from completely different arguments@xcite .
now , we will show that the vanishing of the off - diagonal matrix elements of the bath coupling operator has important consequences for the kraus osr ( i.e. eq .
( [ osr ] ) ) . to begin with , hamiltonian ( [ ham2 ] )
reduces the simpler form @xmath64 for integer powers of @xmath65 then , it is true that @xmath66 for all @xmath34 and @xmath67 . using eq .
( [ ham4 ] ) for @xmath68 one can readily show that the kraus operators ( [ kraus ] ) take the explicit form @xmath69 finally , substituting this back into ( [ osr ] ) gives the desired form of the chaotic kraus decomposition @xmath70 equation ( [ ctk ] ) is an exact equation for the reduced density of an arbitrary quantum system interacting with a chaotic bath of thermodynamic dimension .
while the arguments of the previous section were cast in terms of a phase space representation , similar arguments could be devised for spin baths using generalized coherent state representations@xcite . in the next section
we actually employ a spin bath to test the formula ( [ ctk ] ) numerically .
we will show that ( [ ctk ] ) is very accurate even for a bath of just ten qubits .
in our recent investigation @xcite we showed that internal decoherence , dissipation and coherent shifting , caused by static one and two - body internal imperfections , are potential error sources for flawed qcs@xcite .
we also showed that a single detector qubit can be configured to probe internal bath dynamics @xcite and thus one can obtain valuable information about the bath self - interactions . here
we extend our previous exact numerical results to a longer time limit and test the chaotic kraus decomposition against these exact results . in what follows
we briefly review our flawed qc model and our exact numerical approach .
details of this study are given in @xcite .
our chaotic spin - bath model represents a flawed qc core@xcite .
the total hamiltonian takes the form @xmath71 where the first term represents a central qubit ( i.e. the detector qubit ) .
the second term is the system - bath coupling operator , and the last two terms constitute the chaotic bath hamiltonian which represents the rest of the flawed qc .
our numerical simulations are based on the experimentally relevant parameter regime for a josephson charge - qubit qc proposal @xcite .
we set the detector qubit energy to @xmath72 in units of @xmath73 mk . we assume that all other qubits other than the detector qubit have one - body flaws , i.e. @xmath74 $ ] and similarly @xmath75 $ ] where the detuning is set to @xmath76 .
two - body flaws , i.e. residual system - bath couplings , and chaos generating intra - bath interactions are modeled via @xmath77 $ ] and @xmath78 $ ] , respectively .
we kept the bath temperature constant at @xmath79 in all our simulations . keeping the value of the system - bath coupling constant at @xmath80 we considered three different values of intra - bath couplings ; @xmath81 in units of @xmath82 .
by @xmath83 the bath is fully chaotic . exploiting the experimentally relevant low temperature regime @xcite for the charge - qubit qc proposal @xcite we calculated the exact reduced density @xmath84 via the formula @xmath85 , \label{tden2}\ ] ] where the populations of bath are defined by @xmath86 and @xmath87 are the solutions to the schrdinger equation for all initial states @xmath88 .
the initial system state was chosen as a superposition state @xmath89 .
@xmath90 bath eigenstates sufficed for all our calculations at the given low temperature .
we performed numerical integrations by an eighth order variable stepsize runge - kutta method@xcite and exact diagonalization of the bath hamiltonian is achieved by a lanczos algorithm@xcite for @xmath91 bath qubits .
the explicit form of kraus operators for the hamiltonian of our model is @xmath92 the chaotic kraus decomposition for the given temperature @xmath93 and initial density of the form @xmath94 then has the form @xmath95^{\ast } \\
c_{0}^{n}(t)[c_{1}^{n}(t)]^{\ast } & |c_{0}^{n}(t)|^{2 } \end{array}\right)\end{aligned}\ ] ] where @xmath96 \\
c_{0}^{n}(t ) = \frac{\sqrt{2}}{2 } \left [ \cos{a_{n } t } - i \frac{(b_{z}+2b_{n , n})}{b_{n } } \sin{a_{n } t } \right]\end{aligned}\ ] ] @xmath97 we calculated the diagonal matrix elements of the bath coupling operator by using the exact eigenstates of the bath hamiltonian . in order to confirm an integrable to chaotic transition in the bath hamiltonian @xmath3 as a function of @xmath98
, we analyzed the nearest neighbor level spacing distribution@xcite @xmath99 for the 200 unfolded lowest eigenvalues .
the unfolded spectrum @xmath100 is obtained from the actual eigenspectrum @xmath101 via smoothed staircase functions @xmath102 , i.e. @xmath103 . smoothed staircase functions are polynomial best fits to the actual staircase functions @xmath104 where @xmath105 is a heaviside step function .
the onset of chaos is indicated by the functional form of @xmath99 changing from the poisson form @xmath106 characteristic of integrable systems to the wigner
dyson form @xmath107 associated with chaos .
our results are given in figure [ level ] .
the onset of chaos is observed above @xmath108 . for stronger couplings
the eigenstatistics are consistent with wigner
dyson distribution . while this level spacing statistic is considered to be a universal indicator of quantum chaos
, it does not provide information on the degree of chaoticity .
we wish to know how the accuracy of the kraus decomposition changes with increasing degree of chaos .
therefore , we alternatively refer to the loschmidt echo @xmath109@xcite , which is also widely believed to be an efficient indicator of quantum chaos@xcite .
we calculate @xmath109 for the bath hamiltonian with the following formula @xmath110 where @xmath111 is an initial state chosen as the ground state of @xmath112 .
@xmath113 is the integrable bath hamiltonian ( i.e. @xmath37 for @xmath114 ) and @xmath115 is the chaos generating perturbation hamiltonian for @xmath116 .
a summary of @xmath109 calculations is presented in fig .
( [ echo ] ) .
( black solid line ) , @xmath117 ( red dotted line ) , and @xmath118 ( green dashed line ) . ] it is clear from fig . ( [ echo ] ) that increasing the magnitude of intra - bath coupling @xmath98 results in faster exponential decay of @xmath109 , which can be interpreted as increasing the degree of chaoticity .
we use purity , @xmath119^{2}$ ] , to determine the extent of non - unitary evolution in the system ( i.e. decoherence and dissipation ) .
the ideal value of purity is unity for pure initial system states .
the fidelity is defined by @xmath120 $ ] where @xmath121 is ideal system evolution in the absence of system - bath interactions .
the ideal value of fidelity is also 1 .
while the purity is insensitive to the unitary effects of system - bath interactions , the fidelity is sensitive to both unitary and non - unitary effects .
therefore , large deviations between the magnitudes of purity and fidelity are a good indicator of the presence of unitary shifts .
we tested the accuracy of the chaotic kraus osr for several values of intra - bath couplings in the chaotic bath regime . here
we compare the prediction of the chaotic kraus decomposition with exact numerical results by calculating the purity @xmath122 and fidelity @xmath123 of the reduced density .
in figure ( [ pur ] ) we plot @xmath122 for three different values of intra - bath coupling @xmath98 .
( time is in units of @xmath124 . )
the dashed lines are the kraus predictions while the solid curve represents the exact dynamics . even for @xmath125 ,
shown in part ( a ) , the decoherence predicted by the kraus decomposition is of the correct order of magnitude . clearly , however it is not quantitatively accurate .
there is also a faster time scale to the exact dynamics which is not captured at all by the kraus decomposition . in part ( b ) , for @xmath126 and hence stronger chaos , we see much better agreement . in part ( c ) , for @xmath127 , the kraus decomposition is in very good agreement with the exact results .
this is remarkable given that the bath contains only ten qubits .
moreover , it suggests that the kraus decomposition could actually be employed as a useful computational method for low temperature systems .
in figure ( [ fidel ] ) we plot @xmath128 for both exact and kraus decomposition results for the same three different values of intra - bath coupling @xmath98 .
notice that the magnitude of the errors in @xmath128 is much larger than those in the purity @xmath122 .
the purity measures only non - unitary errors .
@xmath123 is also sensitive to unitary errors .
the large deviation of @xmath122 from unity therefore indicates the presence of a large coherent shift .
the origin of this shift has been explained in detail elsewhere@xcite .
the agreement between the kraus and exact results is quite good even for @xmath125 ( shown in part ( a ) ) . for @xmath126 ( part ( b ) ) and @xmath127 ( part ( c ) ) ,
the agreement is excellent . : for @xmath129 in ( a ) , @xmath117 in ( b ) , @xmath118 in ( c).,width=288,height=336 ] : for @xmath129 in ( a ) , @xmath117 in ( b ) , @xmath118 in ( c).,width=288,height=336 ]
the kraus decomposition captures the correct quantitative behavior for both unitary and non - unitary types of contributions from chaotic environmental dynamics .
the accuracy of the kraus decomposition even for our small bath model is quite impressive and thus it it may prove a useful numerical simulation method . the suggestive form of the decomposition ( [ ctk ] ) is the result of primary interest .
for example , it suggests that all non - unitary effects vanish when all diagonal coupling matrix elements are identical , i.e. @xmath130 . in this case
the dynamics reduces to @xmath131 and the only effect of the bath is the coherent shift @xmath132 .
moreover , it is also clear that the greater the variance in the coupling matrix elements , the more decoherence will be observed .
this prediction is in agreement with that of an approximate master equation@xcite which has been shown to be quantitatively accurate for a number of systems@xcite .
it also predicts that a chaotic bath of thermodynamic dimension with uncorrelated initial state can not cause decoherence or dissipation at absolute zero temperature .
such a bath can still cause coherent shifting , however .
it was previously known that strongly chaotic finite baths cause little decoherence at low temperature@xcite .
the fact that no decoherence is possible at zero temperature for large chaotic baths is in strong contrast to the behavior of integrable baths . in the spin - boson model
strong decoherence is possible even when the bath has zero temperature .
we derived an exact and explicit kraus decomposition for systems interacting with chaotic baths of thermodynamic dimension .
we showed that the decomposition is very accurate even for a small finite qubit bath , and it predicts all important features of the open system dynamics induced by chaotic environments .
it could therefore prove useful as a numerical simulation method .
see for example : c. anastopoulos , b.l .
hu phys.rev . a 62 ( 2000 ) 033821
caldeira , a.j .
leggett , ann .
( 1983 ) 374 ; r.p .
feynman , f.l .
vernon , jr .
( 1963 ) 118 ; f. haake , r. reibold , phys . rev .
a 32 ( 1985 ) 2462 ; a.o .
caldeira , a.j .
leggett , physica a 121 ( 1983 ) 587 . c.m .
dawson , a.p .
hines , r.h .
mckenzie , g.j .
milburn , phys .
a 71 ( 2005 ) 052321 ; s. paganelli , f. de pasquale , s.m .
giampaolo , phys .
a 66 ( 2002 ) 052317 ; m. lucamarini , s. paganelli , s. mancini , phys . rev . a 69 ( 2004 ) 062308 ; r. alicki , open sys . & inf . dyn . 11 ( 2004 ) 53 ; x.s .
ma , a.m. wang , x.d .
yang , h. you , j. phys .
a 38 ( 2005 ) 2761 ; x .- z .
yuan , k .-
zhu , europhys .
69 ( 2005 ) 868 ; x .- z .
yuan , k .-
zhu , z .- j .
j. d 33 ( 2005 ) 129 ; t. prosen , t.h .
seligman , j. phys .
a. 35 ( 2002 ) 4707 ; m. nidari , t. prosen , j.phys .
a 36 ( 2003 ) 2463 .
gaspard and m. nagaoka , j. chem .
phys . , 111 ( 1999 ) 5668 ; a. surez , r. silbey and i. oppenheim , j. chem .
phys . , 97 ( 1992 ) 5101 ; v. romero - rochin and i. oppenheim , j. stat .
phys . , 53 ( 1988 ) 307 ; physica a , 155 ( 1989 ) 52 ; v. romero - rochin , a. orsky and i. oppenheim , _ ibid .
_ , 156 ( 1989 ) 244 .
b. georgeot and d.l .
shepelyansky , phys .
e , 62 ( 2000 ) 3504 ; b. georgeot and d.l .
shepelyansky , _ ibid .
_ ( 2000 ) 6366 ; b. georgeot and d.l .
shepelyansky , phys .
lett . , 81 ( 1998 ) 5129 ; g. benenti , g. casati , and d.l .
shepelyansky , euro .
j. d , 17 ( 2001 ) 265 .
e. hairer , s.p .
norsett , and g. wanner , solving ordinary differential equations i. nonstiff problems , 2nd ed .
, springer series in computational mathematics , vol . 8 , springer - verlag , berlin ; new york , 1993 . | we consider a system interacting with a chaotic thermodynamic bath .
we derive an explicit and exact kraus operator sum representation ( osr ) for the open system reduced density .
the osr preserves the hermiticity , complete positivity and norm .
we show that it is useful as a numerical tool by testing it against exact results for a qubit interacting with an isolated flawed quantum computer .
we also discuss some interesting qualitative aspects of the osr . , kraus decomposition , decoherence , coherent shift , chaotic baths 03.65.@xmath0w , 05.30.@xmath0d , 03.67.lx , 03.65.yz , 05.45.mt |
since the surprising discovery of the @xmath0 in 2003 @xcite , there has been a steadily growing list of new @xmath9 mesons discovered at the @xmath10 factories @xcite .
these discoveries have revealed that the spectrum of @xmath9 mesons is richer than the charmonium states predicted by quark potential models .
the candidates for some of the new @xmath9 mesons include charm meson molecules , tetraquark states , and charmonium hybrid states @xcite .
an ideal experiment for studying some of the new @xmath9 mesons would be @xmath1 collisions at resonance .
the effectiveness of resonant @xmath1 collisions for studying conventional charmonium states was demonstrated by the e760 and e835 experiments at fermilab @xcite .
the `` @xmath9 mesons '' section of the review of particle physics @xcite was largely rewritten by these experiments .
there will be future opportunities to study @xmath9 mesons through resonant @xmath1 collisions .
the panda experiment at gsi is expected to begin taking data on @xmath1 collisions at energies in the charmonium region around 2014 @xcite .
a recently proposed @xmath1 resonance experiment at fermilab could begin even earlier @xcite .
the rate at which a resonance is produced in @xmath1 collisions is proportional to its partial width into @xmath1 .
thus a resonant @xmath1 annihilation experiment would be useful for studying the new @xmath9 mesons provided their @xmath1 partial widths are sufficiently large . to estimate the @xmath1 partial widths of conventional charmonium states , one can take advantage of the measured @xmath1 partial widths of the @xmath11 , @xmath12 , @xmath13 , @xmath8 , @xmath14 , and @xmath15 .
it is difficult to estimate the @xmath1 partial widths for most of the more exotic candidates for the new @xmath9 mesons .
a weakly - bound s - wave charm meson molecule is an exception .
its @xmath1 partial width is related in a simple way to the cross section near threshold for @xmath1 annihilation into the charm mesons that are its constituents .
the energy dependence of that cross section near threshold is determined by the mass and width of the resonance .
if the s - wave contribution to the cross section for @xmath1 annihilation into the charm mesons well above the threshold was known , a reasonable extrapolation to the threshold could be made in terms of the position and width of the resonance .
cross sections for @xmath1 annihilation into pairs of charm mesons have not been measured . however at energies well above threshold , those cross sections can be estimated by scaling the measured cross sections for @xmath1 annihilation into the corresponding strange mesons . in this paper , we will use this strategy to estimate the partial width of the @xmath0 into @xmath1 .
we will asume in this paper that the @xmath0 is a weakly - bound hadronic molecule whose constituents are a superposition of @xmath16 and @xmath17 .
we first summarize the evidence for this identification .
measurements of the mass of the @xmath0 by the belle , cdf , babar , and d0 collaborations @xcite , combined with a new measurement of the @xmath18 mass by the cleo collaboration @xcite , imply that the mass is extremely close to the @xmath16 threshold : @xmath19 the observation of the decay of @xmath4 into @xmath20 @xcite implies that the charge conjugation quantum number is @xmath21 .
studies of the decays of @xmath4 into @xmath22 @xcite strongly favor the spin and parity @xmath23 , although @xmath24 is not excluded .
the observation of decays into @xmath25 @xcite excludes @xmath26 , because the available phase space is so small .
thus the quantum numbers of the @xmath0 have been determined to be @xmath27 .
these quantum numbers imply that the @xmath4 has an s - wave coupling to the charm meson channel @xmath28 .
the mass measurement in eq . ( [ mx ] ) implies that it is a resonant coupling .
this is sufficient to conclude that the @xmath0 is either a charm meson molecule with particle content @xmath28 or it is a virtual state of the charm mesons .
the reason such an unambiguous statement can be made is that nonrelativistic systems with s - wave threshold resonances have universal properties that depend on the large scattering length of the constituents but are otherwise insensitive to their interactions at shorter distances @xcite .
these universal properties have been exploited to describe the decays of @xmath4 into @xmath25 and @xmath29 @xcite , the production process @xmath30 @xcite , the line shapes of the @xmath4 @xcite , and decays of @xmath4 into @xmath12 and pions @xcite .
these applications rely on factorization formulas that separate the length scale @xmath31 from all the shorter distance scales of qcd @xcite .
hanhart et al .
have analyzed the data from the belle and babar collaborations on the decays @xmath32 in the decay channels @xmath22 and @xmath25 and concluded that the @xmath4 must be a virtual state of charm meson @xcite . a more recent analysis that took into account consistently the effects of the @xmath33 width concluded that a charm meson molecule was preferred by the data ,
although a virtual state could not be excluded @xcite . in this paper ,
we exploit the identification of the @xmath0 as a weakly - bound charm meson molecule to give an order - of - magnitude estimate of its partial width into @xmath1 . in section [ sec : hi - energy ] , we estimate the cross section for @xmath34 well above the threshold by scaling measured cross sections for @xmath6 . in section [ sec :
threshold ] , we extrapolate the cross sections for @xmath34 to the threshold under the assumption that there is an s - wave threshold resonance in the @xmath7 channel . in section [ sec : width ] , we deduce the partial width for @xmath4 into @xmath1 from the extrapolated cross section for @xmath35 .
we discuss the results in section [ sec : disc ] .
given that @xmath0 is a weakly - bound s - wave hadronic molecule composed of @xmath2 or @xmath3 , its partial width into @xmath1 is proportional to the cross section for @xmath5 near the @xmath2 threshold . in this section , we estimate that cross section at energies well above the threshold by scaling the measured cross section for @xmath6 . at the quark level ,
the process @xmath34 is @xmath37 .
this process can proceed through the short - distance annihilation process @xmath38 followed by a long - distance process in which the remaining constituents @xmath39 and @xmath40 of the @xmath41 and @xmath42 bind to the @xmath43 and @xmath44 to form the charm mesons .
the @xmath39 and @xmath40 carry only a small fraction of the momenta of the outgoing charm mesons .
this process can proceed most easily via the feynman process in which the @xmath40 and @xmath39 that do not annihilate carry only a small fraction of the momenta of the colliding @xmath42 and @xmath41 .
the momenta transfered to the @xmath40 and to the @xmath39 are therefore small .
we assume that the cross section for this process can be factored into the rate for the short - distance annihilation process @xmath38 and a long - distance probability factor associated with the @xmath40 and @xmath39 that do not annihilate .
we first consider the short - distance factor .
the short - distance process involves the annihilation of @xmath45 into two virtual gluons which then create a @xmath9 pair .
since it involves 6 external partons , the dimensional counting rules for this process imply that its rate scales with the center - of - mass energy @xmath46 like @xmath47 @xcite .
this should be contrasted with the rate for @xmath1 annihilation at resonance into charmonium . in this case , the short - distance process is @xmath48 .
the dimensional counting rules imply that its rate scales like @xmath49 , where @xmath50 is the charmonium mass @xcite . in the case of @xmath34 ,
the rate for the short - distance subprocess scales with two fewer powers of the center - of - mass energy , because one of the three quarks in the proton is not required to annihilate .
we now consider the long - distance factor .
this factor is the probability for the surviving @xmath40 and @xmath39 from the colliding @xmath42 and @xmath41 to become constituents of the outgoing charm mesons . in the initial state , the @xmath51 and the @xmath52 that annihilate act as light - like colored sources for the remaining low - momentum @xmath40 and @xmath39 . in the final state , the @xmath43 and @xmath44 act as colored sources for the low - momentum @xmath39 and @xmath40 . in the @xmath1 rest frame , these sources have equal and opposite velocities @xmath53^{1/2}$ ] .
the effect of the short - distance process is to suddenly replace the light - like colored sources @xmath51 and @xmath52 by colored sources @xmath43 and @xmath44 with equal and opposite velocities @xmath54 .
the long - distance factor @xmath55 is the square of the amplitude for the low - momentum @xmath40 and @xmath39 to evolve from constituents of the @xmath42 and @xmath41 into constituents of the @xmath56 and @xmath33 after the sudden change in color sources .
the long - distance factor @xmath55 depends on the velocity @xmath54 of the colored sources in the final state , but it does not depend on the flavor of the particles that serve as the colored sources
. the amplitude would be the same if the charm quarks @xmath43 and @xmath44 were replaced by strange quarks @xmath57 and @xmath58 with the same velocities @xmath54 .
thus the rate for producing the charm mesons @xmath59 , whose quark content is @xmath60 , can be related to the rate for producing strange mesons @xmath61 , whose quark content is @xmath62 .
if we take into account the masses of the mesons , a @xmath63 pair with center - of - mass energy @xmath64 has the same relative velocity @xmath65 in the center - of - mass frame as a @xmath16 pair with center - of - mass energy @xmath66 if the center - of - mass energies satisfy @xmath67 where @xmath68 .
the cross section for @xmath34 can be expressed as the product of a flux factor @xmath69^{-1/2}$ ] , a phase space factor @xmath70 , and a matrix element factor . under our factorization assumption ,
the matrix element factor is the product of a short - distance factor that scales asymptotically like @xmath71 and a long - distance factor @xmath55 that is a function of the velocities @xmath54 of the @xmath43 and @xmath44 created by the short - distance process .
if we use the asymptotic scaling behavior of the short - distance factor in the matrix element , the energy dependence of the cross section is given by @xmath72 & \sim & \frac{1}{[s ( s - 4 m_p^2)]^{1/2 } } \ , \frac{p(\beta)}{s^2 } \ , \frac{\lambda^{1/2}(s^{1/2 } , m_{d^ * } , m_d)}{s } .
\label{sig - e}\end{aligned}\ ] ] in the limit @xmath73 , the right side of eq .
( [ sig - e ] ) approaches @xmath74 in accord with the dimensional counting rules .
there is an expression analogous to eq .
( [ sig - e ] ) for the cross section for @xmath6 with the same long - distance factor @xmath55 , but with @xmath75 and @xmath76 replaced by @xmath77 and @xmath78 . using eq .
( [ sd - sk ] ) to eliminated that factor , we obtain a scaling relation between the two cross sections : @xmath79 & \approx & \sigma [ p \bar p \to k^{*- } k^+ ; s_k ] \ , \frac { [ s_k ( s_k - 4 m_p^2)]^{1/2 } } { [ s_d ( s_d - 4 m_p^2)]^{1/2 } } \nonumber \\ & & \hspace{1 cm } \times \left ( \frac { s_k}{s_d } \right)^3 \frac { \lambda^{1/2}(s_d^{1/2 } , m_{d^ * } , m_d ) } { \lambda^{1/2}(s_k^{1/2 } , m_{k^ * } , m_k ) } , \label{sig - scale}\end{aligned}\ ] ] where @xmath80 is the function of @xmath81 obtained by solving eq .
( [ sd - sk ] ) . as a function of the center - of - mass energy .
the open dots are the data from refs .
the two solid dots are data from the crystal barrel collaboration @xcite .
the curves passing through the crystal barrel data points are extrapolations whose energy dependence is given by the analog of eq .
( [ sig - e ] ) for @xmath63 with the probability factor @xmath55 approximated by a constant .
[ fig : sigmak],width=453 ] the cross sections for @xmath6 have been measured for antiproton momenta in the lab frame ranging from 702 mev to 1642 mev @xcite .
this corresponds to center - of - mass energies @xmath82 ranging from 1990 mev to 2304 mev .
the only recent measurements were from the crystal barrel collaboration @xcite .
the cross section was measured to be @xmath83b at @xmath84 mev and @xmath85b at 2304 mev @xcite .
the relative velocities @xmath54 given by eq .
( [ sd - sk ] ) are 0.74 and 0.80 , respectively .
the data points are shown in figure [ fig : sigmak ] as a function of the center - of - mass energy .
also shown in the figure are curves that pass through the crystal barrel data points and whose dependence on the energy is given by the analog of eq .
( [ sig - e ] ) for @xmath63 with the probability factor @xmath55 approximated by a constant .
notice that the two curves do not differ dramatically .
this suggests that at these energies , the matrix element factor in the @xmath63 cross section may already be close to its asymptotic behavior proportional to @xmath74 . to estimate the cross section for @xmath34 , we insert the crystal barrel data points for the @xmath63 cross section into eq .
( [ sig - scale ] ) , where @xmath80 is the function of @xmath86 obtained by solving eq .
( [ sd - sk ] ) .
the center - of - mass energies @xmath64 for the two crystal barrel data points are 2050 mev and 2304 mev .
the corresponding center - of - mass energies @xmath66 for the process @xmath34 are 5714 mev and 6391 mev .
if we use the higher - energy crystal barrel data point as the input , our estimate for the cross section for @xmath16 is @xmath87 \approx \left ( \frac{m_{d^*}+m_d}{\sqrt{s } } \right)^6 \ , \frac{\lambda^{1/2}(s^{1/2 } , m_{d^ * } , m_d ) } { [ s ( s - 4 m_p^2)]^{1/2 } } \left ( 4800 ~ { \rm nb } \right ) .
\label{sig - hi}\ ] ] if we use the lower - energy crystal barrel data point as the input , the factor of 4800 nb is replaced by 5600 nb .
the relatively small difference implies that the probability factor @xmath55 in eq .
( [ sig - e ] ) does not depend dramatically on @xmath54 in the region @xmath88 . if we were to change the assumed scaling behavior of the short - distance factor in the square of the matrix element from @xmath89 to @xmath90 , the prediction in eq .
( [ sig - hi ] ) would change by a multiplicative factor of @xmath91 .
given the estimate in eq .
( [ sig - hi ] ) of the cross section for @xmath34 at energies far above threshold , we would like to obtain an estimate of the cross section near threshold .
we can not simply use eq .
( [ sig - hi ] ) to extrapolate to the threshold region , because this expression ignores the suppression of higher partial waves near the threshold , which can be taken into account through the probability factor @xmath55 in eq .
( [ sig - e ] ) , and because it also ignores the threshold resonance associated with the @xmath0 .
we will assume that the threshold resonance is in the s - wave @xmath21 channel , and that this is the only effect that gives dramatic dependence of the s - wave phase shift on the center - of - mass energy @xmath82 .
as the center - of - mass energy approaches the threshold at @xmath92 , the terms in the matrix element with nonzero orbital angular momentum quantum number @xmath93 are suppressed by a factor of @xmath94 .
only the s - wave @xmath95 term survives in the limit . if the s - wave threshold resonance could be ignored , a crude extrapolation of the cross section for @xmath16 in eq .
( [ sig - hi ] ) towards the threshold could be obtained simply by multiplying it by the fraction @xmath96 of the cross section that comes from s - wave scattering .
the determination of the s - wave fraction requires measurements of the angular distribution .
there is some information on the angular distribution for the related process @xmath6 . in the crystal barrel experiment @xcite
, it was estimated that angular momenta up to @xmath97 contribute to the cross section at center - of - mass energy 2304 mev . in ref .
@xcite , the angular distribution for @xmath6 was measured at center - of - mass energy 2006 mev .
the ratios @xmath98 of the coefficients of the legendre polynomials in the partial wave expansion were given for @xmath93 up to 5 .
from those results , we can infer that the fraction of the cross section at 2006 mev that comes from s - wave scattering is about 9.3% .
since no experimental information is available on the s - wave fraction @xmath99 for @xmath16 , we will use the s - wave fraction for @xmath63 at 2006 mev as an estimate : @xmath100 .
if the threshold resonance could be ignored , we could estimate the cross section for @xmath16 near threshold obtained simply by extrapolating eq .
( [ sig - hi ] ) to the threshold and multiplying it by the fraction @xmath96 .
however the existence of the @xmath0 with quantum numbers @xmath7 implies that there is a resonant enhancement of the cross section for @xmath16 near the threshold due to scattering in the s - wave @xmath21 channel .
the enhancement applies only to the fraction @xmath101 of the cross section that comes from @xmath1 scattering into channels that are even under charge conjugation .
experiments on @xmath1 annihilation at very low @xmath41 momentum have revealed a suppression of the production of final states that are even under charge conjugation . in ref .
@xcite , the suppression of @xmath102 final states was also observed in final states containing pairs of strange mesons at the center - of - mass energy 2006 mev .
the suppression factor in the case of @xmath63 can be deduced from measurements of @xmath103 , which is related by isospin symmetry .
the cross section for @xmath104 followed by the decay @xmath105 , which gives a final state with @xmath102 , is only @xmath106 @xmath107b .
the cross section for @xmath108 with no restriction on the decays of @xmath109 is @xmath110 @xmath107b .
the ratio of these two cross sections gives an even - charge - conjugation fraction of about 13.1% .
since no experimental information is available on the even - charge - conjugation fraction @xmath111 for @xmath16 , we will use the even - charge - conjugation fraction for @xmath112 at 2006 mev as an estimate : @xmath113 . to extrapolate the cross section for @xmath34 to the threshold region
, we will exploit the universal features of s - wave threshold resonances in nonrelativistic 2-body systems @xcite .
these universal features can be conveniently expressed in terms of the scattering length @xmath31 in the resonant s - wave channel .
if the resonance is sufficiently close to the threshold , @xmath114 will be much larger than all other length scales associated with the structure of the particles and their interactions .
the scattering amplitude in the resonant channel has the universal form @xmath115 where @xmath116 is the energy relative to the 2-body threshold and @xmath50 is the reduced mass . if @xmath31 is real and positive , the amplitude in eq .
( [ f - e ] ) has a pole at @xmath117 , where @xmath118 the pole is associated with a bound state with binding energy @xmath119 .
if @xmath31 is real and negative , there is no pole on the physical sheet of the complex energy @xmath116 , so there is no bound state .
there is however a pole on the unphysical sheet of @xmath116 and it is conventionally referred to as a `` virtual state '' . the scattering length @xmath31 is complex if the 2-body system has inelastic scattering channels . if @xmath31 is complex , its real part can still be used to distinguish between a bound state ( @xmath120 ) and a virtual state ( @xmath121 ) . in the case
@xmath120 , the bound state has a nonzero width determined by the imaginary part of @xmath31 .
the existence of the @xmath0 with quantum numbers @xmath7 and mass given by eq .
( [ mx ] ) implies that there is an s - wave threshold resonance in the channel @xmath122 the scattering amplitude in this channel is given by the universal expression in eq .
( [ f - e ] ) , where @xmath123 and @xmath50 is the reduced mass @xmath124 . since @xmath0 decays , its constituents have inelastic scattering channels , so the scattering length @xmath31 must be complex . in ref .
@xcite , the authors analyzed the data from the belle and babar collaborations on the @xmath0 resonance produced via @xmath32 and decaying through the channels @xmath22 and @xmath25 .
they concluded that @xmath0 must be a virtual state with @xmath121 . in a subsequent analysis of the same data , the nonzero width of the constituent @xmath33 or @xmath125 of the @xmath4
was taken into account @xcite .
the conclusion of that analysis was that the data preferred a bound state corresponding to @xmath120 , although a virtual state was not excluded .
we will assume that the @xmath0 is indeed a bound state .
we also assume for simplicity that the imaginary part of @xmath31 is small compared to its real part , so the binding energy of the @xmath4 can be approximated by the simple expression in eq .
( [ ex ] ) with @xmath50 replaced by @xmath126 .
we proceed to use the universality of s - wave threshold resonances to extrapolate the estimated cross section for @xmath34 to the threshold region .
matrix elements corresponding to the projection of the charm mesons onto the channel @xmath127 defined in eq .
( [ dstardbar ] ) will be enhanced at energies near the @xmath16 threshold by the resonance associated with the @xmath0 . for the process @xmath34
, the resonance enhances the matrix element near the threshold by a dimensionless factor @xmath128 where @xmath129 and @xmath130 is the momentum scale above which the effects of the nonzero range of the interaction become important . if the s - wave effective range @xmath131 for the scattering of the charm mesons was known , we could use @xmath132 as a quantitative estimate .
a natural order - of - magnitude estimate for @xmath130 is @xmath133 , which is the next smallest relevant momentum scale after @xmath134 .
this is a relevant momentum scale , because the low energy scattering of @xmath16 is dominated by pion exchange .
our estimate of the cross section for @xmath34 near the threshold is obtained by extrapolating the estimated cross section in eq .
( [ sig - hi ] ) to the threshold region and multiplying it by the s - wave fraction @xmath100 , the even - charge - conjugation fraction @xmath113 , and the resonance factor @xmath135 : @xmath136 \approx \left ( 1.9 ~ { \rm nb } \right ) \frac{\lambda^2 k_{\rm cm}}{m_\pi ( 1/a^2 + k_{\rm cm}^2 ) } .
\label{sigd*dbar0}\ ] ] the factor of @xmath133 in the denominator in eq .
( [ sigd*dbar0 ] ) was introduced to make the last factor dimensionless . if the lower - energy crystal barrel data point was used as the input for eq .
( [ sig - hi ] ) , the prefactor in eq .
( [ sigd*dbar0 ] ) would have been 5.5 nb .
to relate the cross sections for @xmath1 annihilation into @xmath16 and @xmath0 , we begin by considering the forward scattering amplitude for @xmath138 near the @xmath16 threshold .
this amplitude can be expressed as the product of three factors : a short - distance factor for the transition from @xmath1 to the resonant channel @xmath127 , a long - distance resonance factor @xmath139 given by eq .
( [ f - e ] ) , and a short - distance factor for the transition from @xmath127 to @xmath1 . by the optical theorem ,
the total @xmath1 cross section is proportional to the imaginary part of the forward scattering amplitude .
the resonant contribution to the imaginary part associated with @xmath16 or @xmath17 in the final state comes from the imaginary part of the resonance factor : @xmath140 where @xmath141 and @xmath142 is the binding energy of the @xmath4 given in eq .
( [ ex ] ) .
the first term on the right side corresponds to the production of @xmath16 and @xmath17 above the threshold .
the second term corresponds to the production of the @xmath0 resonance .
the cross sections for producing @xmath16 near threshold , @xmath17 near threshold , and @xmath0 are @xmath143 & = & \mbox{$\frac12 $ } \sigma_{\rm sd } \frac{m_\pi k_{\rm cm}}{1/a^2 + k_{\rm cm}^2 } \theta(e ) , \label{sigd*dbar - fact } \\ \sigma [ p \bar p \to d^0 \bar d^{*0};e ] & = & \mbox{$\frac12 $ } \sigma_{\rm sd } \frac{m_\pi k_{\rm cm}}{1/a^2 + k_{\rm cm}^2 } \theta(e ) , \label{sigddbar*-fact } \\ \sigma [ p \bar p \to x;e ] & = & \sigma_{\rm sd } \frac{\pi m_\pi ( 2 m_{d^ * \bar d } e_x)^{1/2}}{m_{d^ * \bar d } } \delta(e + e_x ) , \label{sigxfact}\end{aligned}\ ] ] where @xmath144 is a short - distance factor . in eq .
( [ sigxfact ] ) , we have used eq .
( [ ex ] ) to eliminate the scattering length @xmath31 in favor of the binding energy @xmath119
. a factor of @xmath133 has been inserted into the long distance factor to ensure that @xmath144 has the dimensions of a cross section .
the delta function of the energy in eq .
( [ sigxfact ] ) arises from neglecting the width of the @xmath0 resonance .
an estimate for the short - distance factor @xmath144 can be obtained by comparing the cross section in eq .
( [ sigd*dbar - fact ] ) with the estimated cross section in eq .
( [ sigd*dbar0 ] ) : @xmath145 the cross section for @xmath146 in the narrow resonance limit is related to the partial width for @xmath147 by simple kinematics .
if we denote the t - matrix element for @xmath147 by @xmath148 , the partial width into @xmath1 is @xmath149 = \frac{1}{2 m_x } \ , \frac{1}{3 } \sum_{\rm spins }
| { \cal m } |^2 \ , \frac{(m_x^2 - 4 m_p^2)^{1/2}}{8 \pi m_x } .
\label{gamma - t}\ ] ] the cross section for @xmath146 in the narrow resonance limit averaged over the spins of the proton and antiproton is @xmath150 = \frac{1}{2 m_x ( m_x^2 - 4 m_p^2)^{1/2 } } \ , \frac{1}{4 } \sum_{\rm spins } | { \cal m } |^2 \ , \frac{\pi}{m_x } \delta(e + e_x ) .
\label{sigma - t}\ ] ] eliminating the t - matrix element from eqs .
( [ gamma - t ] ) and ( [ sigma - t ] ) , we obtain @xmath150 = \frac{6 \pi^2 \gamma[x \to p \bar p]}{m_x^2 - 4 m_p^2 } \delta(e + e_x ) .
\label{sigma - gamma}\ ] ] comparing with eq .
( [ sigxfact ] ) and using the estimate of the short - distance factor @xmath144 in eq .
( [ sigma - sd ] ) , we obtain an estimate of the partial width of @xmath4 into @xmath1 : @xmath149 = \frac{(m_x^2 - 4 m_p^2 ) \lambda^2 ( 2 m_{d^ * \bar d } e_x)^{1/2 } } { 3 \pi m_\pi m_{d^ { * } \bar d } } \left ( 1.9 \ { \rm nb } \right ) .
\label{gamxppbar}\ ] ] the partial width is proportional to the square root of the binding energy @xmath119 .
the current measurement of the binding energy is given by eq .
( [ mx ] ) : @xmath151 mev . the expression in eq .
( [ gamxppbar ] ) can be reduced to @xmath149 = \left ( \frac{\lambda}{m_\pi } \right)^2 \left ( \frac{e_x}{0.6 \ { \rm mev } } \right)^{1/2 } \left ( 28 \ { \rm ev } \right ) .
\label{gamxppbar : est}\ ] ] the total width of the @xmath4 must be greater than the width of its constituent @xmath33 , which is about 70 kev .
thus the estimate in eq .
( [ gamxppbar : est ] ) implies that the branching fraction for @xmath147 is less than about @xmath152 .
we have estimated the partial width of @xmath0 into @xmath1 .
the estimate was obtained by a sequence of well - motivated steps .
the first step was estimating the cross section for @xmath34 well above the threshold by scaling measured cross sections for @xmath6 .
the second step was estimating the contribution from the s - wave @xmath7 channel using suppression factors @xmath99 and @xmath111 for the @xmath63 process .
the third step was extrapolating the cross section for @xmath34 to the @xmath16 threshold under the assumption that there is an s - wave threshold resonance in the @xmath7 channel .
the cross section for @xmath146 was obtained from the cross section for @xmath34 near threshold by using the universal properties of an s - wave threshold resonance . the partial width for @xmath147
is related to the cross section for @xmath146 by simple kinematic factors .
our estimate for the partial width is given in eq .
( [ gamxppbar : est ] ) .
it is widely believed that the production of a loosely - bound molecular state in a process involving energies much greater than the binding energy should be strongly suppressed .
there is some suppression , but it is not as strong as one might expect . in the estimate in eq .
( [ gamxppbar : est ] ) , the suppression is taken into account by the factor proportional to @xmath153 .
this factor goes to @xmath154 as @xmath155 , suggesting that there would be complete suppression in this limit .
however the suppression factor has this simple form proportional to @xmath153 only if @xmath119 is greater than @xmath156 , where @xmath157 is the total width of the @xmath4 .
there is no further suppression once @xmath119 becomes comparable to @xmath156 .
a lower bound on @xmath157 is provided by the width of the constituent @xmath33 or @xmath125 : @xmath158 = 66 \pm 15 $ ] kev .
thus even if the binding energy is much smaller than 0.6 mev , the associated suppression factor can not decrease the estimate in eq .
( [ gamxppbar : est ] ) by more than about a factor of 0.2 .
the estimate for the partial width for @xmath159 in eq .
( [ gamxppbar : est ] ) is proportional to @xmath160 , where @xmath130 is the momentum scale above which the effects of the nonzero range of the interaction between the charm mesons becomes important .
thus the rate is very sensitive to @xmath130 .
the scale @xmath130 can be identified with @xmath161 , where @xmath131 is the effective range for the scattering of the charm mesons , which is not known .
we have suggested that a natural scale for @xmath130 is @xmath133 , since low - energy scattering of charm mesons is dominated by pion exchange .
the estimates of the rate for @xmath162 in refs .
@xcite were also proportional to @xmath160 . in this case , the choice @xmath163 may underestimate the rate by about an order of magnitude . thus setting @xmath163 in eq .
( [ gamxppbar : est ] ) may also underestimate the partial width of @xmath147 by an order of magnitude . since the p - wave charmonium state @xmath8 has the same quantum numbers @xmath7 as the @xmath0 , it is useful to compare the @xmath1 partial widths of the @xmath0 with that of @xmath8 , which is @xmath164 ev . if we set @xmath163 and @xmath165 mev in eq .
( [ gamxppbar : est ] ) , our estimate of the @xmath1 partial width of @xmath0 is within a factor of 2 of that of @xmath8 .
the e760 experiment on charmonium production at resonance in @xmath1 collisions measured the total width of the @xmath8 with an uncertainty of 0.14 mev @xcite .
our estimate for the partial width of @xmath0 suggests that a new resonant @xmath1 experiment at gsi @xcite or at fermilab @xcite should be able to measure the properties of the @xmath0 with comparable or higher accuracy than those of the @xmath8 .
it is also useful to compare the production rates of @xmath0 and @xmath8 in other processes .
the production rate of @xmath4 by the exclusive decay @xmath166 has been measured by the belle and babar collaborations @xcite .
dividing the measured product of the branching fractions for @xmath166 and @xmath167 by the branching fraction for @xmath168 @xcite , we obtain @xmath169 } { \gamma [ b^+ \to k^+ + \chi_{c1 } ] } = \frac { 0.023 \pm 0.005}{br[x \to j/\psi \ , \pi^+ \pi^- ] } .
\label{rbk}\ ] ] some information on the branching fraction for @xmath4 into @xmath22 is provided by a measurement of the relative production rates for @xmath170 and @xmath25 in the @xmath0 resonance region by the belle collaboration @xcite : @xmath171}{br[x \to j/\psi \ , \pi^+ \pi^- ] } = 8.8^{+3.1}_{-3.6 } .
\label{rbrx}\ ] ] this measurement suggests that the branching fraction for @xmath4 into @xmath170 is less than about 1/8.8 , which implies that the production ratio in eq .
( [ rbk ] ) is greater than about 0.2 .
the production rates of @xmath0 and @xmath8 are subject to various suppression factors . in the case of @xmath0 , they include the factor @xmath153 associated with the large mean separation of its constituents . in the case of @xmath8 ,
they include a factor of @xmath172 associated with the small relative momentum @xmath173 of the @xmath43 and @xmath44 .
these suppression factors apply equally well to all production processes .
however there is one suppression factor in resonant @xmath1 production that does not apply to exclusive @xmath10 meson decay .
this is the suppression factor associated with the annihilation of all three quarks in the proton in the case of @xmath8 and the annihilation of only two of the quarks in the case of @xmath0 .
because of this suppression factor , the production ratio for @xmath0 and @xmath8 in resonant @xmath1 collisions should be larger than the production ratio in eq . ( [ rbk ] ) .
this is consistent with the estimated production ratio in resonant @xmath1 collisions obtained by dividing eq .
( [ gamxppbar : est ] ) by the measured partial width for @xmath174 . in summary , our estimate of the partial width for @xmath0 into @xmath1 in eq .
( [ gamxppbar : est ] ) indicates that it should be comparable to and perhaps even larger than that of the @xmath8 .
this implies that a future experiment on @xmath1 collisions at the @xmath0 resonance will allow the properties of this remarkable meson to be studied in great detail . | we present an estimate of the partial width of @xmath0 into @xmath1 under the assumption that it is a weakly - bound hadronic molecule whose constituents are a superposition of the charm mesons @xmath2 and @xmath3 . the @xmath1 partial width of @xmath4 is therefore related to the cross section for @xmath5 near the threshold . that cross section at an energy well above the threshold is estimated by scaling the measured cross section for @xmath6 .
it is extrapolated to the @xmath2 threshold by taking into account the threshold resonance in the @xmath7 channel .
the resulting prediction for the @xmath1 partial width of @xmath0 is proportional to the square root of its binding energy . for the current central value of the binding energy ,
the estimated partial width into @xmath1 is comparable to that of the p - wave charmonium state @xmath8 . |
the nature of dark energy , which is believed to be responsible for the present phase of accelerated expansion of the universe @xcite , still remains unknown . despite its phenomenological success ,
the simplest description in terms of a cosmological constant ( @xmath0cdm model ) suffers from an important naturalness problem , since the measured value of @xmath0 , corresponding to @xmath1 ev)@xmath2 , finds no natural explanation in the context of known physics .
moreover , the fact that today matter and dark energy have comparable contributions to the energy density , turns out to be difficult to understand if dark energy is a true cosmological constant .
thus , the energy density of a cosmological constant remains constant throughout the history of the universe , whereas those of the rest of components ( matter or radiation ) grow as we go back in time .
then the question arises as to whether it is a _ coincidence _ ( or not ) that they have comparable values today when they have differed by many orders of magnitude in the past .
notice also that if @xmath0 is a fundamental constant of nature , its scale ( around @xmath3 ev ) is more than 30 orders of magnitude smaller than the natural scale of gravitation , @xmath4 with @xmath5 gev . on the other hand ,
if @xmath0 is just an effective parametrization of dark energy , still a proper understanding of the underlying physics would be needed in order to explain the measured value .
alternative models have been proposed in which dark energy is a dynamical component rather than a cosmological constant .
such models are usually based on new physics , either in the form of new cosmological fields or modifications of einstein s gravity @xcite . however , they are generically plagued by classical or quantum instabilities , fine tuning problems or inconsistencies with local gravity constraints . in this paper
we explore the possibility of understanding dark energy from the standard electromagnetic field , without the need of introducing new physics ( previous works on models of dark energy based on vector fields can be found in @xcite ) .
we will show that the behavior of electromagnetic fields on very large ( super - hubble ) scales differs from the well studied short - distance ( sub - hubble ) behavior .
thus , on super - hubble scales , the time component of the electromagnetic field grows linearly in time in the matter and radiation eras , giving rise to a cosmological constant contribution in the electromagnetic energy - momentum tensor ( the potential gravitational effects of longitudinal electromagnetic fields were considered in a different context in @xcite ) . at late times
this contribution becomes dominant giving rise to the accelerated phase . as a possible generating mechanism
, we calculate the spectrum of super - hubble electromagnetic modes produced during inflation from quantum fluctuations and find that the correct value of the dark energy density can be naturally obtained in the case in which inflation took place at the electroweak scale .
we start by writing the electromagnetic action including a gauge - fixing term in the presence of gravity : @xmath6 \label{action}\end{aligned}\ ] ] the gauge - fixing term is required in order to define a consistent quantum theory for the electromagnetic field @xcite , and we will see that it plays a fundamental role on large scales .
still this action preserves a residual gauge symmetry @xmath7 with @xmath8 .
einstein s and electromagnetic equations derived from this action can be written as : @xmath9 where @xmath10 is the energy - momentum tensor for matter and radiation and @xmath11 is the energy - momentum tensor of the electromagnetic field .
notice that since we will be using the covariant gupta - bleuler formalism , we do not a priori impose the lorentz condition .
the effect of the high conductivity of the universe in the matter and radiation eras will be discussed below .
we shall first focus on the simplest case of a homogeneous electromagnetic field ( zero mode ) in a flat robertson - walker background , whose metric is given by : @xmath12 in this space - time , equations ( [ fieldeq ] ) read : @xmath13 with @xmath14 the hubble parameter . notice
that ( [ eqa0 t ] ) implies that the gauge - fixing term exactly behaves as a cosmological constant throughout the history of the universe , irrespective of the background evolution .
indeed , for homogeneous fields we have : @xmath15 we can solve ( [ eqa0 t ] ) and ( [ feqrw ] ) during the radiation and matter dominated epochs when the hubble parameter is given by @xmath16 with @xmath17 for radiation and @xmath18 for matter .
in such a case the solutions for ( [ feqrw ] ) are : @xmath19 where @xmath20 and @xmath21 are constants of integration .
hence , the growing mode of the temporal component does not depend on the epoch being always proportional to the cosmic time @xmath22 , whereas the growing mode of the spatial component evolves as @xmath23 during radiation and as @xmath24 during matter , i.e. at late times the temporal component will dominate over the spatial ones . on the other hand ,
the @xmath25 component of einstein s equations adopts the following form : @xmath26 \label{friedmann}\end{aligned}\ ] ] where @xmath27 stands for radiation and matter respectively and : @xmath28 notice that we need @xmath29 in order to have positive energy density for @xmath30 . besides
, when inserting the solutions ( [ a0sol ] ) and ( [ azsol ] ) into these expressions we obtain that @xmath31 , @xmath32 and @xmath33 const as commented before .
thus , the field behaves as a cosmological constant throughout the evolution of the universe since its temporal component gives rise to a constant energy density whereas the energy density corresponding to @xmath34 always decays as radiation .
moreover , this fact prevents the generation of a non - negligible anisotropy which could spoil the highly isotropic cmb radiation ( see @xcite for a more general discussion ) .
finally , when the universe is dominated by the cosmological constant arising from the gauge - fixing term , both the hubble parameter and @xmath30 become constant leading therefore to a future de sitter universe .
let us emphasize that according to ( [ const ] ) , @xmath35 always contributes as a cosmological constant .
as the observed fraction of energy density associated to a cosmological constant today is @xmath36 , we obtain that the field value today must be @xmath37 .
the effects of the high electric conductivity @xmath38 can be introduced using the magneto - hydrodynamical approximation and including on the r.h.s . of maxwell s equations the corresponding current term , which is given by @xmath39 with @xmath40 the velocity associated to the comoving observers .
notice that the strict neutrality of the plasma , which is consistent with a vanishing electric field , implies @xmath41 , and finally , the current can be written as : @xmath42 .
notice that electric neutrality also implies that conductivity does not affect the evolution of @xmath43 .
the infinite conductivity limit simply eliminates the growing mode of @xmath44 in ( [ azsol ] ) .
the inhomogeneous case , corresponding to @xmath45 modes , will be discussed in next section . we still need to understand which are the appropriate initial conditions leading to the present value of @xmath30 . in order to avoid the cosmic coincidence problem ,
such initial conditions should have been set in a natural way in the early universe . in a very interesting work @xcite
, it was suggested that the present value of the dark energy density could be related to physics at the electroweak scale since @xmath46 , where @xmath47 gev .
this relation offers a hint on the possible mechanism generating the initial amplitude of the electromagnetic fluctuations .
indeed , we see that if such amplitude is set by the size of the hubble horizon at the electroweak era , i.e. @xmath48 , then the correct scale for the dark energy density is obtained .
thus , using the friedmann equation , we find @xmath49 , but according to ( [ rhoa0 ] ) , @xmath50const .
, so that @xmath51 as commented before .
a possible implementation of this mechanism can take place during inflation .
notice that the typical scale of the dispersion of quantum field fluctuations on super - hubble scales generated in an inflationary period is precisely set by the almost constant hubble parameter during such period @xmath52 , i.e. @xmath53 @xcite .
the correct dark energy density can then be naturally obtained if initial conditions for the electromagnetic fluctuations are set during an inflationary epoch at the scale @xmath54 .
let us make these arguments more precise .
we shall look at the electromagnetic perturbations generated during inflation in order to determine its primordial power spectrum . in this case
it is more convenient to use conformal time @xmath55 defined by means of @xmath56 and to introduce the conformal components of the field @xmath57 . besides we shall focus on a single fourier mode of the vector field with wave vector @xmath58 and decompose the field in temporal , transverse and longitudinal components with respect to @xmath58 . in this frame , equations ( [ fieldeq ] )
read : @xmath59{{\mathcal{a}}}_{0k } -2ik\left[\frac{1+\lambda}{2\lambda}{{\mathcal{a}}}_{\parallel k } ' -{\mathcal{h}}{{\mathcal{a}}}_{\parallel k}\right]=0 \label{modes}\\
\vec{{{\mathcal{a}}}}_{\perp k}''&+&k^2\vec{{{\mathcal{a}}}}_{\perp k } = a\sigma\vec{{{\mathcal{a}}}}_{\perp k } ' \nonumber\\ { { \mathcal{a}}}_{\parallel k}''&-&k^2\lambda{{\mathcal{a}}}_{\parallel k}-2ik\lambda\left[\frac{1+\lambda}{2\lambda}{{\mathcal{a}}}_{0k } ' + { \mathcal{h}}{{\mathcal{a}}}_{0k}\right]=a\sigma({{{\mathcal{a}}}}_{\parallel k}'-ik{{\mathcal{a}}}_{0k})\nonumber\end{aligned}\ ] ] with @xmath60 and @xmath61 is the hubble parameter in conformal time .
he have included for completeness the current term on the right - hand side as commented before .
notice that once again the electric neutrality of the universe implies that the evolution equation for the temporal component is not modified . during inflation the electric conductivity of the universe is negligible and this term can be safely neglected so that in the following we shall set @xmath62 .
it is easy to see from equations ( [ modes ] ) that the transverse modes are just plane waves irrespective of the expansion rate . on the other hand , the components @xmath63 and @xmath64
are coupled to each other even in the absence of gravity .
this is due to the fact that we are working with arbitrary @xmath65 and not using the simple feynman gauge @xmath66 .
let us first consider quantization in minkowski space - time , with @xmath67 .
the decomposition in fourier modes can be written as follows : @xmath68\nonumber\\ { { \mathcal{a}}}_{\parallel}&=&\int\frac{d^3\vec{k}}{2k_0(2\pi)^3}\left[\left(i\frac{1 + \lambda}{1-\lambda}({\bf a}_0+{\bf a}_\parallel)k\eta+{\bf a}_\parallel\right ) e^{-ikx}\right.\nonumber \\ & + & \left.\left(-i\frac{1+\lambda}{1-\lambda}({\bf a}_0^++{\bf a}_\parallel^+ ) k\eta+{\bf a}_\parallel^+\right)e^{ikx}\right]\label{solsmink}\end{aligned}\ ] ] with @xmath69 .
now , in order to have the canonical commutation rules @xmath70= i\delta^\nu_\mu\delta^{(3)}(\vec{x}-\vec{y})$ ] , the creation and annihilation operators appearing in ( [ solsmink ] ) should satisfy : @xmath71= & & \frac{1-\lambda}{\lambda}k_0(2\pi)^3\delta^{(3)}(\vec{k}-\vec{k'})\nonumber\\ \left[{\bf a}_\parallel(\vec{k}),{\bf a}_\parallel^+(\vec{k'})\right]= & -&\frac{1-\lambda}{\lambda}k_0(2\pi)^3\delta^{(3)}(\vec{k}-\vec{k'})\end{aligned}\ ] ] for simplicity in the following we will take @xmath72 so that we use canonically normalized operators with positive sign for the temporal component .
notice that this is just the opposite situation to the usual feynman gauge .
in fact , @xmath72 and @xmath66 are the only two possible choices with canonical normalizations .
as is well - known @xcite , in order to recover maxwell s theory , we need to eliminate the negative norm states by defining the corresponding restricted hilbert space . following the standard gupta - bleuler formalism
, the physical states @xmath73 will be those annihilated by the combination @xmath74 , that is : @xmath75 since the contributions from longitudinal and temporal modes cancel each other .
thus , as expected , the theory is free from ghosts .
notice that in minkowski space - time , we also get @xmath76 .
now we can proceed to the quantization in the inflationary epoch . in order to present the calculational method explicitly
, we assume an exact de sitter phase .
the general quasi - de sitter results will be given below .
thus in de - sitter : @xmath77 and @xmath78 with @xmath79 .
the classical solutions of the corresponding equations are a bit more complicated , although it is still possible to obtain analytic expressions : @xmath80\nonumber\\ \nonumber\\ { { \mathcal{a}}}_{\parallel k}&=&ic_1(1+ik\eta)e^{-ik\eta } -ic_2\left[\frac{3}{2}e^{-ik\eta}+(1-ik\eta)e^{ik\eta}e_1(2ik\eta)\right]\end{aligned}\ ] ] where @xmath81 is the exponential integral function .
note that the mode @xmath82 can be gauged away by means of a residual gauge transformation .
the sub - hubble limit ( @xmath83 ) of these solutions reads : @xmath84 the choice of adiabatic vacuum @xcite is made by matching these solutions with those obtained in the minkowski case ( [ solsmink ] ) ( up to sub - leading terms in the sub - hubble limit ) . to do this
, we choose the modes @xmath82 and @xmath85 in the following way : @xmath86 on the other hand , on super - hubble scales ( @xmath87 ) we have : @xmath88 with @xmath89 the euler s constant .
we see that @xmath90 , which means that @xmath91 is ( almost ) constant during inflation , once the mode leaves the horizon . in the case of quasi - de sitter slow - roll inflation
, the hubble parameter reads @xmath92 , where the slow - roll parameter is defined as @xmath93 , with @xmath94 the inflaton potential .
following the same steps as before , we obtain the power spectrum for @xmath30 on super - hubble scales : @xmath95^{n_{a_0}}\end{aligned}\ ] ] which is almost scale - invariant ( as in the scalar field case ) since for the electromagnetic spectral index we obtain @xmath96 . in a similar way it is possible to obtain the primordial power spectrum of longitudinal modes on super - hubble scales : @xmath97^{-4\varepsilon}\end{aligned}\ ] ]
if we now compare the power spectra for the conformal fields @xmath98 and @xmath99 we find that : @xmath100 which is negligible on super - hubble scales , and allows us to safely ignore the longitudinal modes on such scales after inflation . notice that since @xmath101 , @xmath102 is a red - tilted spectrum which means that the contribution to @xmath103 from long wavelenghts dominates over small scales . in particular , provided inflation lasted for a sufficiently large number of e - folds , this allows to decompose the fluctuations field at any given time into a large homogeneous contribution ( with scales @xmath104 ) and a small inhomogeneous perturbation ( @xmath105 ) , and therefore we can use standard perturbation theory around the homogeneous background .
thus , for the homogeneous part we get : @xmath106 where @xmath107 , @xmath108 and @xmath109 is set by the hubble horizon at the beginning of inflation @xcite . here
@xmath110 is the total number of e - folds of inflation which should not be confused with @xmath111 which is the number of e - folds since the time when the scale @xmath112 left the horizon .
typical values for @xmath111 are around 50 , whereas generically there is no upper limit to @xmath110 .
thus as expected , up to tilt corrections , @xmath52 sets the scale for the field dispersion .
= 10 cm once the fluctuations generated during inflation enter the radiation dominated era , it would be in principle possible that the high conductivity of the universe could modify their evolution , spoiling the growing behavior of super - hubble models found in ( [ a0sol ] ) .
however since conductivity does not affect the temporal equation , the @xmath30 modes are not modified on super - hubble scales . in fig.1
we show the evolution of super - hubble temporal and longitudinal modes , both for vanishing and infinite conductivity .
we see that the evolution exactly corresponds to @xmath113 as expected from ( [ a0sol ] ) in the radiation era , even in the infinite conductivity case .
we see that @xmath64 is sub - dominant compared to @xmath114 , until the modes re - enter the hubble radius for @xmath115 .
the result is not sensitive to the change of initial conditions .
thus , the only effect of the high conductivity is the damping of the electric field,(which is consistent with the strict neutrality of the plasma ) .
in particular , this implies @xmath116 , which corresponds to the field evolution shown in fig .
1 . let us emphasize that the vanishing of the electric field does not imply the vanishing of the temporal component @xmath63 . in fact , it is straightforward to show that the value of @xmath117 giving rise to the effective cosmological constant is not affected by the presence of conductivity .
indeed , maxwell s equations in the presence of conserved currents read : @xmath118 taking the four - divergence of the equation we get : @xmath119 where we have used current conservation @xmath120 .
thus , we see that the field @xmath117 evolves as a free scalar field , and it is therefore constant on super - hubble scales , independently of the presence of external currents .
a remarkable consequence of the covariant quantization formalism in the context of inflationary cosmology ( which had not been considered previously ) is the breaking of gauge invariance on cosmological scales . indeed , in this formalism , the classical energy - momentum tensor depends on the gauge - fixing term .
however , in minkowski space - time , when one takes the expectation value of this object in a physical state ( that belonging to the restricted hilbert space ) the gauge dependence disappears because the contributions from the temporal and longitudinal degrees of freedom cancel each other .
this is so just because in minkowski space - time , in the restricted hilbert space , the amplitudes of temporal and longitudinal mode solutions are exactly the same ( see @xcite ) .
however when considering the quantization in an expanding universe important differences arise . at short distances , i.e. for sub - hubble modes , it is easy to see from ( [ sub ] ) that the same cancellation takes place , as it should be , and the theory is exactly the same as in minkowski space - time
. therefore the energy density does not depend on the gauge - fixing term .
nevertheless , when the modes become super - hubble , it can be seen from ( [ super ] ) , that the amplitude of the temporal modes grows in time faster than that of the longitudinal ones .
this spoils the mentioned cancellation and a net energy density results from the @xmath65 term .
notice that in the covariant formalism the four polarizations are always present . in minkowski space - time
( or for sub - hubble modes ) only two of them contribute to the energy density , however on cosmological scales also the temporal one can have observational consequences . as @xmath30 is also a propagating degree of freedom ,
the gauge - fixing term can be seen in the gupta - bleuler formalism as a kinetic term for it , and therefore the coefficient @xmath65 can be fixed by the standard normalization of the creation and annihilation operators .
this effect could not be studied in other ( non - covariant ) formalisms , such as coulomb gauge quantization , since only transverse polarizations would be present in that case . to summarize , the presence of the background cosmological electromagnetic field breaks @xmath121 symmetry on large scales while preserving local ( small - scales ) invariance .
this is analogous to the situation with lorentz symmetry , where the presence of matter or radiation in the universe breaks global lorentz invariance , but respecting local transformations . in other words ,
the presence of a non - vanishing cosmological constant could be signalling the breakdown of gauge invariance on cosmological scales .
let us emphasize that this effect is a consequence of the quantization of electromagnetic theory in the covariant formalism and , as discussed above , it does not modify any of the physical predictions of maxwell s theory for laboratory experiments or astrophysical observations . as a matter of fact , the electromagnetic interaction has not been tested on distance scales larger than 1.3 au @xcite .
despite the fact that the background evolution in the present case is the same as in @xmath0cdm , the evolution of metric perturbations could be different , thus offering an observational way of discriminating between the two models . with this purpose , we have calculated the evolution of metric , matter density and electromagnetic perturbations in the longitudinal gauge with @xmath122 , @xmath123 , @xmath124 , @xmath125 and taking @xmath126 , where as commented before the main contribution to @xmath127 comes from the temporal component .
the propagation speeds of scalar , vector and tensor perturbations are found to be real and equal to the speed of light , so that the theory is classically stable .
we have also checked that the theory does not contain ghosts and it is therefore stable at the quantum level . on the other hand , using the explicit expressions in @xcite for the vector - tensor theory of gravity corresponding to the action in ( [ action ] ) , it is possible to see that all the parametrized post - newtonian ( ppn ) parameters agree with those of general relativity , i.e. the theory is compatible with all the local gravity constraints for any value of the homogeneous background electromagnetic field .
= 9.5 cm = 10 cm in figs 2 - 3 we plot the evolution of scalar perturbations , satisfying @xmath128 in this theory , and matter density contrast @xmath129 , in both , vanishing and infinite conductivity limits .
we find that the only relevant deviations with respect to @xmath0cdm appear on large scales @xmath130 and that they depend on the primordial spectrum of electromagnetic fluctuations .
however , for @xmath131 , the results on the cmb temperature power spectrum and evolution of density perturbation are compatible with observations . taking @xmath132 gev in this case , with @xmath133 the reheating temperature , we obtain the correct value of the dark energy density today .
in addition , the different evolution of @xmath134 with respect to the @xmath0cdm model gives rise to a possible discriminating contribution to the late - time integrated sachs - wolfe effect @xcite .
the presence of large scale electric fields generated by inhomogeneities in the @xmath30 field opens also the possibility for the generation of large scale currents which in turn could contribute to the presence of magnetic fields with large coherence scales .
this could shed light on the problem of explaining the origin of cosmological magnetic fields .
work is in progress in this direction .
we have shown that the present phase of accelerated expansion of the universe can be explained by the presence of a cosmological electromagnetic field generated during inflation .
this result not only offers a solution to the problem of establishing the true nature of dark energy , but also explains the value of the cosmological constant without resorting to new physics . in this scenario
the fact that matter and dark energy densities _ coincide _ today is just a consequence of inflation taking place at the electroweak scale .
such a relatively low inflation scale implies also that no cosmological gravity wave background is expected to be measurable in future cmb polarization observations .
notice also that any vector - tensor theory ( not necessarily electromagnetism ) whose low - energy effective action is given by ( [ action ] ) and in which the vector field only interacts gravitationally with the rest of particles would provide a natural model for dark energy .
in fact all the previous models trying to account for the cosmic acceleration are plagued by classical or quantum instabilities , fine tuning problems or inconsistencies with solar system experiments .
however , in this work we present , for the first time , an explanation to the cosmic acceleration with none of the aforementioned problems . _
acknowledgments : _ we would like to thank j.d .
barrow , j.a.r .
cembranos and c. tamarit for useful comments and suggestions .
this work has been supported by dgicyt ( spain ) project numbers fpa 2004 - 02602 and fpa 2005 - 02327 , ucm - santander pr34/07 - 15875 , cam / ucm 910309 and mec grant bes-2006 - 12059 .
99 s. perlmutter et al . , _ astrophys .
j. _ * 517 * , 565 ( 1999 ) a.g .
riess et al . , _ astron .
j. _ * 116 * , 1009 ( 1998 ) and * 117 * , 707 ( 1999 ) d. n. spergel et al .
_ astrophys .
j. suppl . _
* 148 * , 175 , ( 2003 ) and astro - ph/0603449 m. tegmark et al .
_ * d69 * : 103501 , ( 2004 ) . c. wetterich , _ nucl .
phys . _ * b302 * , 668 ( 1988 ) ; r.r .
caldwell , r. dave and p.j .
steinhardt , _ phys .
lett . _ * 80 * , 1582 ( 1998 ) c. armendariz - picon , t. damour and v. mukhanov , _ phys .
lett . _ * b458 * , 209 ( 1999 ) s.m .
carroll , v. duvvuri , m. trodden , m.s .
turner , _ phys .
_ * d70 * : 043528 , ( 2004 ) g. dvali , g. gabadadze and m. porrati , _ phys .
lett . _ * b485 * , 208 ( 2000 ) v.v .
kiselev , _ class .
* 21 * : 3323 , ( 2004 ) c. armendariz - picon , _ jcap _ * 0407 * : 007 , ( 2004 ) c.g .
boehmer and t. harko , _ eur .
phys . j. _
* c50 * : 423 , ( 2007 ) m. novello , et al .
_ * d69 * : 127301 , ( 2004 ) t. koivisto and d.f .
mota , arxiv:0805.4229 [ astro - ph ] h.s .
zhao , _ astrophys .
j. _ , * 671 * , l1-l4 ( 2007 ) .
j. beltrn jimnez and a.l .
maroto , _ phys .
rev . _ * d78 * , 063005 ( 2008 ) and arxiv:0807.2528 [ astro - ph ] k. bamba , s. nojiri and s. d. odintsov , phys .
rev . * d77 * , 123532 ( 2008 ) s. deser , _ ann .
henri poincar _ ,
* 16 * : 79 ( 1972 ) c. itzykson and j.b .
zuber , _ quantum field theory _ , mcgraw - hill ( 1980 ) j.d
. barrow , _ phys .
* d55 * , 7451 ( 1997 ) ; j.d . barrow and r.
maartens , _ phys
_ * d59 * : 043502 ( 1999 ) n. arkani - hamed , l. j. hall , c. f. kolda and h. murayama , _ phys .
lett . _ * 85 * ( 2000 ) 4434 a. linde , _ particle physics and inflationary cosmology _ , harwood academic press ( 1996 ) n.d .
birrell and p.c.w .
davies , _ quantum fields in curved space _ , cambridge ( 1982 ) a.r .
liddle and d.h .
cosmological inflation and large - scale structure _ , cambridge ( 2000 ) a.s .
goldhaber and m.m .
nieto , arxiv:0809.1003 [ hep - ph ] c. will , _ theory and experiment in gravitational physics _ , cambridge university press , ( 1993 ) r.g .
crittenden and n. turok , _ phys .
_ * 76 * ( 1996 ) 575 | we show that the presence of a temporal electromagnetic field on cosmological scales generates an effective cosmological constant which can account for the accelerated expansion of the universe .
primordial electromagnetic quantum fluctuations produced during electroweak scale inflation could naturally explain the presence of this field and also the measured value of the dark energy density .
the behavior of the electromagnetic field on cosmological scales is found to differ from the well studied short - distance behavior and , in fact , the presence of a non - vanishing cosmological constant could be signalling the breakdown of gauge invariance on cosmological scales .
the theory is compatible with all the local gravity tests , and is free from classical or quantum instabilities .
thus we see that , not only the true nature of dark energy can be established without resorting to new physics , but also the value of the cosmological constant finds a natural explanation in the context of standard inflationary cosmology .
this mechanism could be discriminated from a true cosmological constant by upcoming observations of cmb anisotropies and large scale structure . |
in the next few years , the new generation of cmb experiments ( map and planck ) will provide all - sky maps of the cosmic microwave background ( cmb ) with high resolutions and sensitivities .
theoretical investigations of different kinds of foregrounds , instrumental noise properties , scan strategies and related topics for the future planck mission are now in progress .
the planck lfi and hfi instruments will be able to measure the cmb anisotropy and polarization using 10 frequency channels that cover the frequency range 30 - 857 ghz ( mandolesi et al.,1998 ; maino et al.,1999 ) .
the maps provided by planck will contain contributions from various physical components , including the primary and secondary cmb signals and foregrounds such as free - free and synchrotron emission , the sunyaev - zeldovich effect , galactic dust and extragalactic point sources .
one of the important goals of this experiment , apart from the separation of the cmb from the remaining parts of the signal , is the creation of a point source catalogue .
the aim of this paper is to show how one may optimize the strategy for producing the point source catalogue by using the multi - frequency properties of the radio / submillimetre sky observations .
the effect of point sources on satellite observations has been widely discussed in the literature ( see the review by vielva et al .
( 2001 ) , and references therein ) .
many authors have developed various linear and non - linear methods for point source extraction .
hobson et al . ( 1999 ) suggested the combination of the maximum- entropy method ( mem ) with mexican hat wavelet filtering in order to separate all the physical components of the microwave ski including extragalactic point sources and using different frequency channels .
tegmark and de oliveira - costa ( 1998 ; hereafter to-98 ) described the linear filtering of point sources in the form of a band - pass filter in order to render point sources on each frequency map more visible relative to the background of other physical components .
cayon ( 2000 ) , sanz et al .
( 2000 ) and vielva et al .
( 2000 ) applied the mexican hat wavelet method to denoising the cmb maps and for production of the planck point source catalogue .
naselsky et al .
( 2000 ) , chiang et al . ( 2001 ) proposed the approach of amplitude - phase analysis and applied it to simulated maps of the cmb with foregrounds , point sources and pixel noise included for the accurate determination of the locations of bright point sources and separation of these sources from the rest of the signal by an iteration technique .
all of the methods mentioned above can complement each other in the framework of the future highly sensitive cmb experiments because of the different sensitivities of each method to different properties of the signal . in this paper
we focus our attention on the combination of maps with different frequencies in order to maximize the signal ( @xmath0- point sources ) to noise ( @xmath1- rest of the signal ) ratio .
we show that co - adding of harmonics from different maps with specific weights leads to an increase in the accuracy of the point source subtraction .
the aim of our paper is to produce the best possible combination of all frequency channels in order to construct a map in which the presence of point sources of some particular population is most visible .
this procedure allows us to obtain a more complete and accurate catalogue of point sources than can be obtained by independently investigating each frequency channel ( to-98 ) .
the layout of our paper is as follows . in section 2
we introduce the definition of the best map combination , derive the analytical solution for the maximum possible signal - to - noise ratio and make some numerical predictions for the planck experiment .
we also present a possible special combinations of the multifrequency maps .
we briefly summarize our results in section 3 .
in this section , we derive the optimal way for using the different frequency channels to produce the best possible map for detection and subsequent subtraction of extra - galactic point sources . the idea behind our approach is quite simple and transparent .
we define the best map as the combination of different frequency maps which maximizes the signal - to - noise ratio . in our case
, the signal is the signal from point sources .
the noise consists of other physical components including pixel noise , cmb , dust , synchrotron and free - free emission .
all of these components limit our capability to detect point sources . _
optimization _ suppose that we have @xmath1 different maps of the same region of the sky made using @xmath1 different channels with frequencies @xmath2 .
the flux of each point source @xmath3 is a function of frequency and , therefore , differs from one channel to another , while the position of this source is , obviously , the same for all maps . in this subsection
, we consider for simplicity one particular population of point sources with the same dependence of fluxes on frequency .
the contribution of point sources to the the temperature on the cmb map in the j - th frequency channel can be written as a set of dirac delta functions with coefficients @xmath4 proportional to the flux of the source at position @xmath5 : @xmath6 here , @xmath7 is the total number of point sources .
the temperature fluctuations on the map convolved with the antenna beam are therefore : @xmath8 where @xmath9 is the beam for the j - th channel , the amplitude of the point source @xmath10 is proportional to the flux and depends on the antenna width : @xmath11 . the coefficients @xmath12 have zero mean and variance @xmath13 ( pixel noise is not convolved with the beam ) .
the set of harmonics @xmath14 can be considered as the n - dimensional vector : @xmath15 .
the aim of our paper is to combine @xmath1 given maps with known values of the expected noise to construct the map with the best signal - to - noise ratio . in order to produce such a map ,
we introduce the vector - filter @xmath16 with components @xmath17 and consider the combination of n maps in the following form : @xmath18 where @xmath19 .
we consider a symmetric antenna beam and subsequently drop the subscript m because @xmath20 .
thus the filter that we use should depend on l only : @xmath21 .
our approach can readily be generalized to the case of an anisotropic beam , but this is beyond the scope of the present paper .
our problem now is to find vectors @xmath22 for each @xmath23 so that the ratio @xmath24 is maximized : @xmath25 here @xmath26 is the vector with components @xmath27 ( @xmath28 is the beam ) and @xmath29 $ ] is the covariance matrix : @xmath30 . in our calculations ,
we use : @xmath31 where @xmath32 is the kronecker symbol . the ratio @xmath33 from equation ( 4 ) is maximized
if we choose the filter to be : @xmath34 consequently , one can conclude that point sources with some particular dependency of flux on frequency are most clearly seen in the map that is the combination of @xmath1 maps taken with these specific weights of harmonics .
if only one channel is under consideration , then this formula becomes especially simple and is exactly the same as the one obtained by tegmark and de oliveira - costa ( to-98 ) : @xmath35 using equations ( 4 ) and ( 6 ) , one can get the final result for the maximum possible signal - to - noise ratio @xmath33 , which is achieved in the combined map : @xmath36 _ estimates for the planck experiment _ a multi - frequency analysis gives us the possibility of detecting point sources in coincidence with different frequency maps . in our calculations , we assume the frequency dependence of the intensity for some particular point source population in the form of a simple power law : @xmath37 . multiplying this intensity by the conversion factor between surface brightness and temperature ( tegmark and efstathiou , 1996 ) and taking into account the resolution @xmath38 , one can obtain the dependence of the point source temperature on frequency : @xmath39 where @xmath40 denotes the planck channel number .
channels are numbered in order of increasing frequency ( see table 1,2 ) .
two different populations have been chosen : radio and far - ir sources . up to about 200 ghz
, we can expect mostly radio sources to be detected , while at higher frequenies far - ir sources dominate .
for the radio source population , we took the spectral index @xmath41 from ( vielva et al . , 2001 )
to be close to zero up to 100 - 200 ghz ( table 1 ) and increasing to @xmath42 at higher frequency . for the far - ir population ,
this spectral index has the opposite behavior and we choose @xmath41 to be @xmath43 for 857 - 352 ghz with practically no contribution from these sources to the signal at lower frequencies .
of course , our estimates are applicable only for these two specific populations .
each particular population of point sources should be investigated separately .
this means that we should perform the combination of channels with specific filters @xmath22 which correspond to the appropriate spectral index @xmath41 for this population . using the results of the previous subsection
, one can calculate the signal - to - noise ratio for each channel separately @xmath44 ( if only the j - th channel is under consideration ( to-98 ) ) and for the combination of n channels . from eq .
[ 7,8 ] one finds : @xmath45 the contribution to @xmath46 in each @xmath23 is obviously higher for the combination of maps than for each map separately and consequently @xmath47 . in fig .
( 1,2 ) we show this ratio for two populations ( radio and fir ) as described above .
a significant improvement can be achieved by considering both populations , @xmath48 $ ] , j=1, .. ,10 .
for radio source , s we found @xmath49 and for fir @xmath50 is about 1.2 .
therefore , a more complete catalogue can be produced for this population of sources .
_ special cases and a possible combinations _ let us for simplicity consider a special case by dealing with only two channels with different frequencies : @xmath51 .
since the spectral behavior of different populations of point sources is not well known , we introduce the parameter @xmath52 .
this parameter contains information about amplitudes of temperature fluctuations due to the same point sources in different channels : @xmath53 in fig .
3 we show the signal - to - noise ratio that can be found applying to-98 filters for the 143 and 217 ghz channels separately and for the combination of these two maps with different possible ratios of amplitudes @xmath54 and @xmath55 .
we choose these two channels because significant steepening or even a break in the point source spectrum due to synchrotron emission is expected over this range of frequencies , so it is hard to predict the exact value of @xmath52 . for smaller values of @xmath56
quite a large number of point sources are expected to be found in coincidence for different channels .
it is interesting to investigate how does our filter look like if we combine for example the 70 ghz and 100 ghz ( lfi ) maps . according to formula ( 6 )
one can obtain the filter @xmath57 with components @xmath58 : @xmath59 where @xmath60 .
using eq [ 5 ] we can investigate the behaviour of these two components at different scales . for small @xmath23 s that correspond to scales much larger than the antenna resolution ,
the instrumental noise is negligible ( at least for the low frequency channels : @xmath61 ) and : @xmath62 substituting [ 13 ] in [ 12 ] we obtain : @xmath63 therefore , harmonics with small @xmath23 from different channels are taken with the opposite sign .
thus the contributions from the first and second maps to the combined map compensate each other .
finally the total signal in the combined map turns to be close to zero at large scales .
this means that our filter removes contributions from the cmb and other physical components at scales larger that the typical scale of the point sources . for @xmath23 values that correspond to scales comparable with the antenna beam ,
the pixel noise is dominant and : @xmath64 therefore the components of the filter @xmath22 are close to the filters found in to-98 for each channel separately ( see fig.4 ) : @xmath65 these components are both positive and add to form peaks that correspond to point sources in the combined map .
we have presented an analysis of microwave sky observations based on the combination of channels with different frequencies in order to separate extragalactic point sources for subsequent production of a point source catalogue .
we made some numerical estimates for the future planck experiment and showed that a multi - frequency analysis allows us to increase the signal - to - noise ratio by a factor of 1.2 - 4.4 to create a more complete and precise catalog of point sources .
.results of combination of 10 planck channels for the population of radio sources with given spectral behavior .
the frequency of each channel is indicated in column 2 for both lfi and hfi .
the 3-rd column gives spectral indices as calculated by vielva et al .
@xmath42 denotes spectral break .
fourth column shows the initial signal - to - noise ratio in units of @xmath66 under the condition that the point source in the 1-st channel has @xmath67 level .
the fifth column gives this ratio after applying the to-98 filter to each channel separately . in the last column
we show this ratio for the best combination of channels from 1 to j. this means that the 2-nd row gives the result for the combination of the 1-st and 2-nd channels , 3-d means the combination of 1,2,3 channels and so on .
[ cols="^,^,^,^,^,^ " , ] | a method is proposed for combining multi - frequency maps in order to produce a catalogue of extragalactic point sources using data from future high - precision satellite experiments .
we have found the optimal way for combining maps at different frequencies in order to maximize the signal ( point sources ) to noise ( rest of the signal ) ratio . our approach is a natural multi - frequency generalization of the band - pass filter introduced by tegmark and de oliveira - costa ( 1998 ) .
we show that combination of different frequency maps gives us the possibility of creating a more complete catalogue of point sources . |
understanding the inner workings of active galactic nuclei ( agns ) is important for understanding the role of supermassive black holes in galaxy evolution .
the mid - to - far infrared ( mfir ) is a key wavelength regime for testing the agn unified scheme and probing the nature of the nuclear obscuring medium .
the _ spitzer space telescope _ allows studies to be made at higher sensitivity and spatial resolution that previous ir missions .
we are conducting a large observing program with all three _ spitzer _ instruments to determine the mfir spectral energy distributions ( seds ) of 87 seyfert galaxies ( pi : j. gallimore , pid : 3269 ) . the sample of seyfert galaxies used for this study comprises all seyfert galaxies from the extended 12 sample of @xcite that have @xmath0 kms@xmath1 . here
we discuss the 51 objects in the sample for which the low - resolution , 5 35 irs spectra are currently available .
the analysis of these spectra and a more detailed discussion of the results are presented in @xcite .
we find four distinct types of continuum shapes and spectral features among the irs spectra of the seyfert nuclei .
figure [ fig : egspec ] shows typical spectra in each group .
we group the spectra according to the following main properties : + @xmath2 twenty - four objects ( 47% of the observed objects ) have spectra dominated by strong pah emission features and very red continuum suggestive of cool dust ; mrk938 is the archetype .
many ( 16 ) of these objects show clear or weak silicate absorption at 10 .
+ @xmath2 sixteen objects ( 31% ) have continua that can be described by a broken power - law ; they show a flattening in the continuum slope at @xmath320 ; ngc4151 is the brightest of this class .
this flattening may be due to a warm ( @xmath3170 k ) dust component , peaking at @xmath320 , dominating the mid - ir emission ( e.g. , @xcite ) , however , a simple model of an emissivity - weighted blackbody function does not fit the continuum of these spectra well .
two of these objects , mrk6 and mrk335 , show clear silicate emission features .
a further 9 objects in this group , including ngc4151 , show weak excesses at 10 and 18 that may be due to silicate emission .
+ @xmath2 eight objects ( 16% ) have power - law continuous spectra .
ngc 3516 is representative of this class .
these spectra show no strong dust emission or absorption features , though most have weak excesses at 10 and 18 that may be due to silicate emission .
+ @xmath2 two objects ( ngc1194 & f04385@xmath40828 ) show a broad absorption feature at 9.7 due to silicate dust .
+ one object ( ngc7603 ) appears to show both pah and silicate emission features and otherwise fails to meet any of the above classifications .
the spectral shapes may be quantitatively characterized by the ir colors , or spectral indices , derived from the spectra .
the spectral index between 20 and 30 quantifies the slope of the spectrum ( above the @xmath320 ` break ' for broken power - law spectra ) , while the spectral index between 8 and 10 indicates the presence or absence of the 9.7 silicate feature and pah features .
the resulting color - color diagrams are given in figure [ fig : cols ] .
one sy 2 ( ngc3079 ) with extreme colors , @xmath5 and @xmath6 , is not shown .
the sy 1 s ( seyfert types 1.0 , 1.2 and 1.5 ) lie in the lower - right part of the diagram , occupied by unbroken and broken power - law spectra , specifically avoiding the region of the reddest pah emitters .
sy 2 s are found in both regions .
earlier type seyferts ( i.e. 1 1.5 ) have the bluest colors , while seyfert 2 s with hidden broad line regions ( hblrs ; type 2h ) span the range of colors .
the fact that sy 1 s consistently show the bluest infrared colors and the sy 2 s are the reddest sources indicates the ir spectra are changing systematically with seyfert type .
we performed a principal component analysis ( pca ) on the spectra to determine the component spectral shapes ( eigenvectors ) producing the variety of spectra seen in the sample ( see , e.g. , @xcite ) .
we find that the first eigenvector is the most dominant and contributes 91% of the variance in the spectra .
we find that the relative contribution of eigenvector 1 to each spectrum relates strongly to the shape of the sed , as the different shapes have significantly different contributions from this eigenvector ( figure [ fig : evonehist ] ) .
the similarity in the features between the red objects like mrk 938 and the ir spectra of starburst galaxies such as m 82 strongly suggests these objects are dominated by the starburst contribution to the dust heating at mid - infrared wavelengths ( figure [ fig : sbfigs ] ) .
we find a significant difference between the contribution of the first eigenvector to the spectra of the sy 1 s and sy 2 s ( figure [ fig : evonesy ] ) .
our results confirm the previous findings of @xcite and @xcite , based on ground - based photometry , that sy 2 s show more star formation than sy 1 s .
we present the results of spitzer irs nuclear spectra for the first 51 of a complete sample of 12 selected seyfert galaxies .
we find the following main conclusions : + @xmath2 the spectra clearly divide into groups based on their continuum shapes and spectral features .
the largest group ( 47% of the sample of 51 ) shows very red continuum suggestive of cool dust and strong emission features attributed to pahs .
sixteen objects ( 31% ) are well described by a broken power - law continuum .
a further 16% of the sample show power - law continua with unchanging slopes .
two objects are dominated by a broad silicate absorption feature .
one object in the sample shows an unusual spectrum dominated by emission features , that is unlike any of the other spectra .
some features are clearly related to the starburst contribution to the ir spectrum , while the mechanisms producing the power - law continuum attributed to the agn component are not yet clear .
the ir seds are related to the seyfert types but the precise correspondence between the ir types and the optical spectral types is unclear .
+ @xmath2 principal component analysis suggests that the relative contribution of starburst - heated dust emission to the sed is the dominant cause of variance in the observed spectra .
+ @xmath2 we find that the sy 2 s typically show stronger starburst contributions in their ir spectra than the sy 1 s , confirming previous results found using photometric data .
+ detailed modeling of the continuum emission is underway to separate in detail the starburst and agn contributions to the ir spectrum in order to place constraints on the opacity and geometry of the nuclear obscuring material , and to compare the relative starburst contributions of seyfert types 1 and 2 .
antonucci , r. 1993 , , 31 , 473 buchanan , c. l. , gallimore , j. f. , odea , c. p. , baum , s. a. , axon , d. j. , robinson , a. , elitzur , m. , & elvis , m. 2006 , , submitted francis , p. j. , & wills , b. j. 1999 , asp conf .
ser . 162 : quasars and cosmology , 162 , 363 gorjian , et al . 2004 , , 605 , 156 maiolino , r. , ruiz , m. , rieke , g. h. , keller , l. d. 1995 , , 446 , 561 perez garcia , a. m. , rodriguez espinosa , j. m. 2001 , , 557 , 39 rodriguez espinosa , j. m. , perez garcia , a. m. , lemke , d. , meisenheimer , k. 1996 , , 315 , l129 rush , b. , malkan , m. a. , & spinoglio , l.1993 , , 89 , 1 shang , z. , & wills , b. 2004 , asp conf .
ser . 311 : agn physics with the sloan digital sky survey , 311 , 13 urry , c. m. , & padovani , p. 1995
, , 107 , 803 weedman , d. w. , et al .
2005 , , 633 , 706 | we present the results of irs low - resolution spectroscopy of 51 seyfert galaxies , part of a large _ spitzer _ observing program to determine the mid - to - far infrared spectral energy distributions of a well - defined sample of 87 nearby , 12 -selected seyferts .
we find that the spectra clearly divide into groups based on their continuum shapes and spectral features .
the infrared spectral types appear to be related to the seyfert types .
some features are clearly related to a starburst contribution to the ir spectrum , while the observed power - law continuum shapes , attributed to the agn , may be dust or non - thermal emission .
principal component analysis results suggest that the relative contribution of starburst emission is the dominant cause of variance in the spectra .
we find that the sy 2 s show on average stronger starburst contributions than the sy 1 s . |
this paper addresses the efficient evaluation of finite - temperature response functions @xmath1 for quantum many - particle systems on one - dimensional ( 1d ) lattices in the framework of the _ density matrix renormalization group _ ( dmrg ) @xcite . from the theoretical perspective
, such quantities occur for example in the context of linear response theory @xcite and characterize the effect of a perturbation @xmath2 of the system at time zero on the expectation value of an observable @xmath3 at time @xmath4 .
initially , the system is in thermal equilibrium , where @xmath5 is the inverse temperature and @xmath6 .
the units are chosen such that boltzmann s and planck s constants are @xmath7 and @xmath8 .
such response functions contain important information on the many - body physics and are addressed in many experimental setups .
for example , recent advances in neutron - scattering techniques make very precise measurements possible .
see for example refs .
it is hence very important to have numerical tools at hand that allow for an efficient and precise evaluation of thermal response functions for ( strongly - correlated ) condensed matter models in order to match theoretical models to actual materials and to gain an understanding of the underlying physical processes . as discussed and demonstrated in several works @xcite , finite - temperature response functions for strongly - correlated 1d systems
can be evaluated up to some maximum reachable time by using the time - dependent density matrix renormalization group ( tdmrg ) @xcite applied to a _ purification
_ @xcite of the density matrix . in the dmrg approach ,
those purifications are approximated with a controllable precision by _ matrix product states _ ( mps ) @xcite . to address finite - temperature states with tdmrg
was suggested in refs .
@xcite with first applications in refs .
a difficulty in the simulations of time - evolved states is the growth of entanglement with time @xcite . in tdmrg calculations
, this leads to a corresponding , typically exponential , increase of the computation cost with time and a strong limitation of the maximum reachable times , depending on the available computational resources and the desired accuracy of the simulation .
the effect is much more drastic for mixed states .
the focus of this paper is hence to study the computation cost of the different tdmrg schemes for the evaluation of thermal response functions and to suggest novel more efficient approaches that allow for a substantial increase in the maximum reachable times .
the results of such simulations can be fourier transformed to study the spectral properties of the response . in order to avoid ringing artifacts from the fourier transformation of the data on a restricted time interval , one can either use filters , which result in an artificial broadening , or use linear prediction @xcite .
this was employed first in ref .
@xcite for @xmath9 and in ref .
@xcite for @xmath10 .
see also refs .
@xcite , for further applications of linear prediction in dmrg calculations at @xmath9 .
earlier alternative dmrg approaches for finite - temperature response functions have built on the _ transfer matrix renormalization group _
( tmrg ) @xcite . in a first step ,
analytic continuation techniques as in _ quantum monte carlo _ ( qmc ) were employed @xcite . in a second development ,
transfer matrices comprising the imaginary- and the real - time evolution were used to access autocorrelation functions @xcite . in finite - temperature qmc calculations ( e.g. positive - definite path integral @xcite or stochastic series expansion with directed loops @xcite ) , real - time or real - frequency response functions have to be extracted by analytic continuation from imaginary - time results @xcite which is ill - conditioned and numerically challenging .
finite temperatures have also been addressed by a hybrid algorithm based on monte carlo and tdmrg @xcite . in this paper ,
i discuss how the simulation based on the evolution of a certain mps purification of the thermal density matrix as described in @xcite can actually be understood as evolving a _ matrix product operator _
( mpo ) representation of the square root of the thermal density matrix .
the description in terms of purifications is in this sense superfluous .
a decisive observation is now that there are considerable degrees of freedom for the choice of a specific evaluation scheme .
for the two specific schemes , corresponding to refs .
@xcite ( _ scheme a _ ) and ref .
@xcite ( _ scheme b _ ) , respectively , i examine the scaling of the computation cost with time .
it turns out that _ scheme a _ has some advantage at low temperatures and _ scheme b _ is advantageous at higher temperatures , for which i give an explanation on the basis of quasi - locality @xcite .
the maximum reachable times @xmath11 and @xmath12 differ , for the same computation cost and accuracy , by a factor of order one .
it is then shown that an optimized evaluation scheme , for which one optimizes over the aforementioned degrees of freedom , yields maximum reachable times that are at least twice as large as for _ scheme b _ , specifically @xmath13 . in all cases
one finds that @xmath14 as @xmath15 varies slowly as a function of @xmath16 .
we can devise a new _ scheme c _ that does not require any optimization , but outperforms _ scheme b _ by a factor of two in the maximum reachable time .
it is only outdone by _
scheme a _ at very low temperatures . to go beyond _ scheme c _ , one can study the computation cost for calculating mpo approximations of operators @xmath17 to determine optimized evolution schemes . finally , the connection between _ scheme a _ and an earlier tmrg approach @xcite is explained . for each of the tdmrg schemes
, one can devise a corresponding tmrg variant . due to the substantially increased reachable times and its simplicity ( no need for optimization ) , the novel _ scheme c _ , which was introduced recently in ref .
@xcite and is explained here in detail , can be expected to be the method of choice for future applications . in ref .
@xcite , it allowed us to show that the thermal spectral functions of 1d bosons in the quantum critical regime with dynamic critical exponent @xmath18 follows a universal scaling form and to compute the corresponding scaling function precisely .
as described for example in refs .
@xcite , the ( unnormalized ) thermal density matrix @xmath19 for a chain of @xmath20 sites can be obtained in the form of an mps purification [ eq : mps_purify ] @xmath21 where the @xmath22 label orthonormal site basis states for @xmath23 , the auxiliary hilbert space @xmath24 is isomorphic to @xmath23 , and @xmath25 label orthonormal site basis states for @xmath24 .
the mps is constructed from @xmath26 matrices @xmath27 , where @xmath28 . the matrix sizes @xmath29 are also called _ bond dimensions _ ; see for example the review @xcite . in order for @xmath30 to be a purification of @xmath31
, it needs to be constructed such that @xmath32 this can be achieved by choosing the infinite - temperature purification to be @xmath33 which can be written as an mps with bond dimensions @xmath34 . in this so - called ancilla approach , finite - temperature purifications can be calculated by imaginary - time evolution @xmath35 and response functions are evaluated after a subsequent real - time evolution @xmath36{\hat{b}}\big[e^{-i{\hat{h}}t}{\hat{a}}|{\rho}_\beta{\rangle}\big].\ ] ] the evolution of the mps can for example be implemented by decomposing the propagators into circuits of local gates by a trotter - suzuki decomposition @xcite as described in section [ sec : tdmrg_vs_tmrg ] .
the square brackets in eq .
indicate which parts of the expression are represented as mps after the evolution .
note that @xmath37 , which is required in eq .
, can be determined as @xmath38 .
equivalently , one can work with purifications @xmath30 that are normalized to one . in each step of the tdmrg , the evolved states @xmath39 are approximated by an mps with bond dimensions @xmath40 that are as small as possible for a given constraint on the desired precision of the approximation @xcite .
this is achieved through _
truncations_. for every splitting of the system into a left and a right part , one does a schmidt decomposition @xmath41 of the state @xcite which boils down to doing singular value decompositions of the tensors @xmath42 .
the corresponding reduced density matrices are @xmath43 and @xmath44 .
the bond dimension is then reduced from @xmath45 to some value @xmath46 by retaining only the @xmath47 largest schmidt coefficients and truncating all smaller ones . @xmath48
the precision is in each step of the algorithm controlled by bounding the _ truncation weight _
@xmath49 an mps purification @xmath50 and the corresponding mpo @xmath51 , according to the isomorphism . in the graphical representation ,
boxes represent the tensors @xmath42 that define the mps or mpo , and lines represent partial tensor contractions .
the diagram for @xmath52 represents the tensor network that has to be contracted in order to evaluate the trace of two mpos @xmath51 and @xmath53 , and a local observable @xmath54 as occurring in the tdmrg schemes for the evaluation of thermal response functions .
see for example eq . .
the computational cost is determined by the bond dimensions of the mpos as in eq .
.,scaledwidth=90.0% ] as a matter of fact , the notation of this procedure as an algorithm on purifications is in a sense superfluous . due to the fact that @xmath55 , the space of linear maps on @xmath23 , and the tensor product space @xmath56 are isomorphic , everything can be rewritten in terms of mpos in @xmath55 instead of mps purifications in @xmath56 , @xmath57 examples for this one - to - one relation are @xmath58 where the transposition is to be executed in the @xmath59 basis . in particular , the counterpart of the mps purification is the mpo @xmath60 see for example refs .
@xcite for discussions of mpos .
applying evolution operators @xmath61 or @xmath62 to the left or right of an mpo @xmath51 can be done by tdmrg in exactly the same way as for mps , for example , by a trotter - suzuki decomposition as described in section [ sec : tdmrg_vs_tmrg ] .
the truncation weight , which is kept below a certain bound in order to control the precision , is then given by @xmath63 where @xmath64 is the schatten 2-norm .
this is the natural norm within the dmrg framework .
evaluation _ scheme a _ as described and employed in refs .
@xcite corresponds to eq . .
in the mpo notation it reads @xmath65{\hat{b}}\big[e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}/2}\big]\right).\ ] ] the square brackets indicate which parts of the expression are represented as mpos after the evolution .
figure [ fig : schemesab ] shows the algorithm diagrammatically .
the partition function @xmath37 can be obtained from the schatten 2-norm @xmath66\vert}_{2})^2 $ ] . in the dmrg method , multiplications of the matrices @xmath67 and singular value decompositions , required for the truncations of the mpos , dominate the computation costs .
hence , the computation costs for a single evolution step for the first mpo ( bond dimensions @xmath68 ) and the second mpo ( bond dimensions @xmath69 ) , and for the final evaluation of eq . scale as @xmath70 respectively @xcite ; see figure [ fig : mpo ] .
_ schemes a _ and _ b _ for the evaluation of the response function according to equations and . in the tdmrg simulations ,
the operators in square brackets are approximated as mpos .
the computation costs for given @xmath4 and @xmath71 depend on the chosen scheme and the desired accuracy of the mpo approximations.,scaledwidth=80.0% ] for a nonsingular operator @xmath72 , the value of is invariant under transformations [ eq : trafo ] @xmath73 & \quad\longrightarrow\quad & & \big[\hat t e^{-\beta{\hat{h}}/2 } e^{i{\hat{h}}t}\big]\\ \big[e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}/2}\big ] & \quad\longrightarrow\quad & & \big[e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}/2}\hat t^{-1}\big]\end{aligned}\ ] ] of the involved mpos @xcite .
this is a considerable degree of freedom that can be exploited to reduce , for given @xmath71 , @xmath4 , and @xmath74 , the bond dimensions @xmath68 of the mpos and , hence , to reduce the computation cost . in cases where such a reduction is possible , one can extend the simulation to longer times @xmath4 .
some numerical experiments showed that the corresponding computation cost for the optimization of @xmath75 scales exponentially with the system size .
so a search for globally optimal @xmath75 seems to be a hopeless exercise .
it is hence reasonable to constrain ourselves to certain classes of transformations , e.g. , @xmath76 and to optimize with respect to the degrees of freedom of the class @xmath77 and @xmath78 in that case .
see section [ sec : opt ] . the modified evaluation _
scheme b _ ( see fig . [
fig : schemesab ] ) employed in refs .
@xcite corresponds to the choice @xmath79 and reads in the operator notation @xmath80{\hat{b}}\big[e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}/2}e^{i{\hat{h}}t}\big]\right)\nonumber\\ \label{eq : chi_schemeb } & = \frac{1}{z_\beta}{\operatorname{tr}}\left(\big[e^{-\beta{\hat{h}}/2}\big]{\hat{b}}\big[e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}/2}e^{i{\hat{h}}t}\big]\right).\end{aligned}\ ] ]
zero - temperature phase diagram of the spin-@xmath0 xxz heisenberg model in a magnetic field .
the ground state is fully polarized in the ferromagnetic phase . in the nel phase
, it has a finite staggered magnetization .
the system is critical ( vanishing energy for excitations ) in the xy , or spin - liquid , phase ( gray ) .
the phase boundaries can be obtained from the bethe ansatz @xcite . in this work ,
the properties of different tdmrg schemes for the evaluation of thermal response functions are exemplified with this model at the three points @xmath81 with @xmath82.,scaledwidth=55.0% ] computation cost per time step ( @xmath83 ) for _ scheme a _ ( top ) and _ scheme b _ ( bottom ) with @xmath84 and @xmath85 , @xmath86 @xcite , and truncation weights per time step @xmath87 , @xmath88 .
the contour lines correspond to maximum reachable times for different computational resources ( per time step ) and the predefined precision of the simulations ( @xmath89 , @xmath90 ) .
, scaledwidth=95.0% ] maximum bond dimensions @xmath91 for _ scheme a _ ( top ) and _ scheme b _
( bottom ) with @xmath84 and @xmath85 , @xmath86 , and truncation weights per time step @xmath87 , @xmath88.,scaledwidth=95.0% ] in order to study the computational costs of evaluation _ schemes a _ and _ b _ [ eqs . and ] , as functions of time and temperature , let us choose as an example the spin-@xmath0 xxz heisenberg model @xmath92 the phase diagram @xcite , as derived by the bethe ansatz , is shown in figure [ fig : xxz ] .
the simulations are carried out for vanishing magnetic field @xmath82 , system size @xmath93 , and coupling constants @xmath94 and @xmath95 , where the model is gapless , and at @xmath96 , where the model is in its gapped antiferromagnetic phase . for the time evolution , a fourth order trotter - suzuki decomposition with step sizes @xmath97
the truncation weights in the imaginary - time and real - time evolutions were fixed to values @xmath89 and @xmath90 which are specified in the captions of the corresponding figures .
figure [ fig : ab_sp63_cost ] shows the computation cost per time step for the operators @xmath86 in eqs . and
, i.e. , spin - flips in the middle of the system @xcite .
the density plots show , as a function of time and temperature , the number of operations needed for a single time step as measured by @xmath83 . for _ scheme a _ ,
the cost is dominated by the cost for the computation of the mpo @xmath98.\ ] ] for _ scheme b _ , the computation of the mpo @xmath99\ ] ] dominates the numerical costs .
hence , the bond dimensions @xmath100 of those operators were used in the diagrams . please notice the logarithmic scale for @xmath5 and the cost , i.e. , the color coding and the distances of the contour lines @xcite .
the contour lines correspond to maximum reachable times for certain computational resources and a predefined precision of the simulation , determined by @xmath89 and @xmath90 .
figure [ fig : ab_sp63_dim ] shows , for the same simulations , the maximum bond dimensions @xmath101}m_i(\beta , t)$ ] .
first of all , the results show that the computation costs increase , for fixed temperature , exponentially with time @xmath4 .
this corresponds to a linear increase of the entanglement entropy in the evolution of pure states after a quench @xcite . for the noncritical case , @xmath96 , the costs per time step become @xmath71-independent for temperatures that are sufficiently below the excitation gap . at higher temperatures , the cost for _ scheme b _
is systematically smaller than that for _ scheme a_. the effect is strongest for the non - interacting system at @xmath94 . for @xmath95 and @xmath96 ,
the increase in the maximum reachable times is more moderate with a factor of @xmath102 . at lower temperatures , _ scheme
a _ is more efficient than _ scheme b_. this trend strengthens when the accuracy of the simulation is reduced ( higher truncation weights @xmath89 and @xmath90 ) as documented in the left part of figure [ fig : ab_sp63sk0_cost ]
. figure [ fig : ab_sp63_dim ] shows that the maximum bond dimensions @xmath91 evolve almost in the same way as the computation costs . for the noncritical case @xmath96 ,
one sees however that the maximum bond dimensions occurring in _ scheme a _ are for all temperatures smaller than those occurring in _ scheme b _ , whereas the computation cost for _ scheme b _ is lower at higher temperatures .
also , for @xmath95 , the maximum bond dimensions occurring in the two schemes are quite similar .
computation cost per time step ( @xmath83 ) for _ schemes a _ and _ b_. left : @xmath95 , and @xmath86 with ( increased ) truncation weights @xmath103 , @xmath104 .
right : @xmath95 and @xmath85 , and @xmath105 with the usual truncation weights @xmath87 , @xmath88.,scaledwidth=98.5% ] evolution of the mpos and as occurring in _ schemes a _ and _ b _ , respectively .
the plots show the bond dimensions @xmath40 for @xmath106 , high as well as low temperatures , and the critical point @xmath95 as well as the gapped @xmath96 . the truncation weights were @xmath87 and @xmath88 . unlike in _ scheme a _ ,
the mpo for _ scheme b _ remains unchanged outside a certain space - time cone also at high temperatures .
this is due to the quasi - locality ( right ) ; see eq .
.,scaledwidth=98.5% ] these findings can be explained as follows . in ref . @xcite it was pointed out that the operator of _ scheme b _ is time - independent for the simple case @xmath107 , whereas the computation cost for the mpo , occurring in _ scheme a _ , can increase with time , even in this trivial case .
more generally , the following argument applies for all operators @xmath2 with finite spatial support @xmath108 .
the typical condensed matter systems are _ quasi - local _
@xcite , i.e. , the spatial support of operators like @xmath109 , occurring in _ scheme b _ , grows only linearly with time .
more precisely , outside a certain space - time cone originating from @xmath108 , the evolved operator acts almost like the identity and does hence not change the entanglement in that region ; see figure [ fig : ab_sp63_slices ] . in mathematical terms
, quasi - locality means @xmath110 where all terms @xmath111 in the hamiltonian @xmath112 are required to be short - ranged and norm - bounded , @xmath113 is the hamiltonian truncated to a vicinity @xmath114 of the spatial support of the operator @xmath2 ( @xmath115 ) , @xmath116 is the boundary of @xmath114 , and @xmath117 is the lieb - robinson velocity @xcite .
so the norm - distance between the exactly evolved operator and the operator , evolved on the subsystem @xmath114 only , decays exponentially with increasing distance of @xmath116 to a space - time cone defined by the lieb - robinson velocity @xmath118 , and originating from @xmath108 at time zero .
as the evolved operator @xmath109 , hence , behaves up to exponentially small corrections like the identity outside the specified space - time cone , it does not alter the operator @xmath119 $ ] in that region in particular , not the mpo bond dimensions @xmath68 for the product occurring in _ scheme b_. this effect is very well visible in the plots for @xmath95 and @xmath120 shown in figure [ fig : ab_sp63_slices ] .
for both tdmrg schemes , the maximum bond dimensions occur in the middle of the system , where @xmath106 acts , and grow ( exponentially ) with time @xmath4 .
the maxima have comparable values .
whereas , in _ scheme b _ , @xmath68 remains unchanged outside the lieb - robinson space - time cone due to the quasi - locality , they grow for all bonds in _ scheme a_. this implies according to eq . an increased cost for _ scheme a _ at those temperatures .
nevertheless , as figures [ fig : ab_sp63_cost][fig : ab_sp63_slices ] exemplify , _ scheme a _ is often advantageous at low temperatures , especially for noncritical systems . for low temperatures ,
the quasi - locality is not essential as @xmath121 limits the effect of the real - time propagators @xmath122 to a low - energy subspace . sufficiently far away from the support of @xmath2 ,
the resulting dynamics is then very restricted and can not lead to a big increase of the bond dimensions .
this is true for both schemes .
furthermore , in the middle of the system , where @xmath106 acts , the growth of the bond dimensions is slower for _ scheme a _ which can be attributed to the fact that one acts with one propagator in eq . instead of acting with two propagators in eq . .
as the plots for @xmath123 and @xmath124 in figure [ fig : ab_sp63_slices ] exemplify , the effect is much more pronounced for gapped systems .
the described properties have also been found for other local operators @xmath2 and @xmath3 like @xmath125 etc
. the behavior for non - local operators like @xmath126 is also quite similar as exemplified in figure [ fig : ab_sp63sk0_cost ] for @xmath127 .
please note that @xmath128=0 $ ] in the isotropic case @xmath95 for which the evolution of @xmath129 is hence trivial .
the computation cost per time step @xmath83 ( top and bottom ) and the maximum bond dimension @xmath91 ( middle ) for the optimized evaluation scheme with @xmath84 and @xmath85 , respectively . with @xmath130
, the schemes have been optimized with respect to @xmath77 and @xmath78 .
the corresponding efficiency measures were chosen to be the computation cost per time step for the top and bottom rows of diagrams , and the maximum bond dimension was chosen for the middle row of diagrams . in all computations ,
the truncation weights were chosen as @xmath87 , @xmath88.,scaledwidth=95.0% ] _ schemes a _ and _ b _ , eqs . and , are typically far from optimal .
one has a lot of freedom in designing a scheme that is as efficient as possible .
the choice for a specific measure for the efficiency of a scheme can depend on the available computational resources . in this paper , the computation cost per time step as quantified by @xmath83 [ eq . ] and , as an alternative , the maximum bond dimension @xmath91 are considered as useful measures . with a nonsingular operator @xmath75 , the most generic scheme involving two mpos is @xmath131 \big[\hat t^{-1}e^{-i{\hat{h}}t}{\hat{a}}e^{-\beta{\hat{h}}}\big]\big)$ ] . as pointed out in section [ sec : schemesab ] , to optimize efficiently with respect to generic @xmath75 is in general not possible .
thus , it is reasonable to constrain ourselves to a certain subclass of schemes and to optimize the efficiency over the remaining degrees of freedom of that class .
let us consider the class of schemes @xmath132 \big[e^{-i{\hat{h}}(t - t'')}{\hat{a}}e^{-(\beta-\beta'){\hat{h}}}e^{i{\hat{h}}(t - t')}\big]\right).\ ] ] as before , the two factors in square brackets are to be approximated by mpos @xcite . for given inverse temperature @xmath71 , time @xmath4 , and accuracy as quantified by the truncation weight @xmath74 [ eq .
] one can now optimize the chosen efficiency measure with respect to @xmath77 , @xmath78 , and @xmath133 .
_ scheme a _ corresponds to the choice @xmath134 , _ scheme b _ to @xmath135 , and the choice @xmath136 for example to the heisenberg picture . in actual applications
, one can first study the computation cost , for obtaining operators @xmath137 $ ] and @xmath138 $ ] . from the results ,
one can then conclude on optimal values for @xmath77 , @xmath78 , and @xmath133 in the evaluation of the response function according to eq . .
for the demonstrational purposes of this paper , it is sufficient to not explore the full three - dimensional parameter space , but to constrain ourselves in the following to the subspace with @xmath139 , i.e. , to the schemes @xmath140 \big[{\hat{a}}(t'-t ) e^{-(\beta-\beta'){\hat{h}}}\big]\right)\\ \label{eq : chi_schemefull2 } & = \frac{1}{z_\beta}{\operatorname{tr}}\left(\big[e^{i{\hat{h}}t'}e^{-\beta'{\hat{h}}}{\hat{b}}e^{-i{\hat{h}}t'}\big ] \big[e^{-i{\hat{h}}(t - t')}{\hat{a}}e^{-(\beta-\beta'){\hat{h}}}e^{i{\hat{h}}(t - t')}\big]\right).\end{aligned}\ ] ] figure [ fig : opt_sp63spk0 ] shows , with @xmath141 @xcite , the computation cost per time step and the maximum bond dimensions for the optimized scheme .
the maximum reachable times @xmath142 of the optimized scheme are at least twice as large as the reachable times @xmath143 for _ scheme b _ , specifically @xmath144 .
see also figure [ fig : cmp_sp63 ] .
the factor @xmath0 in the temperature argument is inessential as one finds that @xmath145 .
this is due to the fact that @xmath15 varies slowly as a function of @xmath16 for all response functions considered in this paper .
let us discuss the possible scenarios for the optimum values of @xmath77 and @xmath78 for the case @xmath141 : the computation cost for _ scheme b _ is dominated by the tdmrg computation of the operator @xmath146 $ ] in eq . .
the corresponding maximum time , reachable for some given computational resources , truncation weight @xmath74 , and inverse temperature @xmath77 , is denoted by @xmath147 .
the maximum reachable time for the scheme as a function of @xmath71 is then given by @xmath148 there are two possible scenarios , as depicted in figure [ fig : schemeopt ] . if @xmath147 is concave on the interval @xmath149 $ ] , the optimal scheme corresponds to @xmath150 .
otherwise , there will be an optimum @xmath151 . in the examples studied here ,
@xmath147 was found to be almost concave in the relevant temperature ranges . as this direct comparison for @xmath95 and @xmath86 shows ( all time axes have the same scales ) ,
the maximum reachable times in the optimized scheme exceed those of the earlier schemes roughly by a factor of two . the efficiency measure for the optimization
was chosen to be the computation cost per time step ( left ) and the maximum bond dimension ( right ) , respectively .
the truncation weights were chosen as @xmath87 , @xmath88.,scaledwidth=100.0% ] left : two possible scenarios for the maximum times @xmath142 [ eq . ]
reachable with the optimizable tdmrg scheme for the case @xmath152 and given computational resources .
if @xmath147 is non - concave on the interval @xmath149 $ ] , the optimal scheme corresponds to some nontrivial @xmath77 .
this can for example occur for systems with dynamics on different energy scales .
if @xmath147 is concave , the optimal scheme is given by @xmath150 , i.e. , _ scheme c_. in the examples studied here , @xmath147 was found to be almost concave in the relevant temperature ranges .
right : _ scheme c _ for the evaluation of the response function according to eq .
.,scaledwidth=100.0% ] one can hence devise a corresponding _ scheme c _ that is near - optimal among the schemes ( at least for @xmath141 ) , does not require any optimization , but outperforms _ scheme b _ by a factor of two in the maximum reachable times .
it is only outdone by _ scheme a _ at very low temperatures .
the scheme corresponds to the choice @xmath150 in eq .
@xmath153 \big[{\hat{a}}(-t_a ) e^{-\beta){\hat{h}}/2}\big]\right)\\ \label{eq : chi_schemec } & = \frac{1}{z_\beta}{\operatorname{tr}}\left(\big[e^{i{\hat{h}}t_b}e^{-\beta{\hat{h}}/2}{\hat{b}}e^{-i{\hat{h}}t_b}\big ] \big[e^{-i{\hat{h}}t_a}{\hat{a}}e^{-\beta{\hat{h}}/2}e^{i{\hat{h}}t_a}\big]\right).\end{aligned}\ ] ] after the imaginary - time evolution that yields @xmath154 $ ] , one runs two real - time tdmrg simulations to obtain mpos @xmath155 $ ] and @xmath156 $ ] . with eq .
one then obtains @xmath157 ; see figure [ fig : schemeopt ] .
the accuracy of the mpos should be kept under control during the whole simulation , for example , as described in section [ sec : dmrg ] by bounding the truncation error @xcite . if this is done properly ,
it is of minor importance what specific @xmath158 and @xmath159 are chosen to evaluate @xmath160 for a given time @xmath4 .
the maximum reachable time @xmath4 is of course given by the sum of the maximum reachable @xmath158 and @xmath161 . for the case
@xmath152 ( or cases where @xmath162 is for example simply a translate of @xmath2 @xcite ) , the maximum reachable times for @xmath158 and @xmath161 are equal , and the total maximum reachable time with this scheme is then twice as large as the maximum time of _ scheme b_. this makes many more physical applications accessible . as pointed out above , _
scheme c _ is optimal among the classes of schemes corresponding to eq . if @xmath147 is a convex function . for the cases studied here it was indeed found to be convex or almost convex in the considered temperature ranges .
one can expect a different behavior for example for systems with dynamics on different energy scales .
in such cases one can achieve considerable performance improvements by optimizing among the class of schemes or . in exceptional cases
it also occurs that evolved operators @xmath163 have an mpo representation with a time - independent bond dimension .
this implies @xmath164 and one can compute the response function for arbitrary times by using @xmath165 and @xmath166 in eq . .
one such case is the operator @xmath167 for the xy model ( @xmath168 ) which can for all times be written as an mpo of bond dimension @xmath169 @xcite .
the time evolution of matrix product states or density operators , as employed in the presented evaluation schemes for thermal response functions , can be implemented within the tdmrg framework in several different ways .
the results for the computation costs of the different evaluation schemes are essentially independent of the chosen time evolution algorithm . the currently most common choice @xcite employs a trotter - suzuki decomposition @xcite .
alternatively , one can use for example the arnoldi method , runge - kutta method or other krylov subspace approaches @xcite .
let us consider in the following a hamiltonian @xmath170 with nearest - neighbor interactions @xmath171 , operators @xmath2 and @xmath3 with finite spatial support ( for simplicity single sites ) , and implementing the time evolution with a trotter - suzuki decomposition .
this case allows for an alternative evaluation of the thermal response functions on the basis of the transfer matrix renormalization group ( tmrg ) @xcite .
the trotter - suzuki decompositions of real- or imaginary - time propagators read @xmath172 where @xmath173 or @xmath174 , respectively , @xmath175 is the time step , and @xmath176 contain the hamiltonian terms on even and odd bonds , respectively .
the coefficients @xmath177 and @xmath178 sum up to one ( @xmath179 ) and define different decompositions of order @xmath180 and number of stages @xmath181 per time step .
a common choice is the leapfrog algorithm which has @xmath182 stages and order @xmath183 with @xmath184 and @xmath185 . in this work ,
a decomposition with @xmath186 stages and order @xmath187 was used . when the operator exponentials occurring in the formula @xmath188 for the response function in _ scheme a _ , eq .
, are decomposed by such a trotter - suzuki decomposition , one ends up with the task of contracting a 2d tensor network ( quantum cellular automaton ) of local gates as displayed in figure [ fig : tdmrg_vs_dmrg ] with the space direction being horizontal and the time direction being vertical .
similar tensor networks result for the other evaluation schemes .
visualization of the 2d tensor network that has to be contracted for the evaluation of the thermal response function @xmath189 on the basis of _ scheme a _ when using a trotter - suzuki decomposition of the operator exponentials ( nearest - neighbor interactions ) .
this can be done either using tdmrg or using tmrg . in the tdmrg method , one contracts one row of local gates after another ( with intermediate truncation steps ) , evolving two mpos that are oriented horizontally .
the final evaluation of the response function according to eq . corresponds to taking the hilbert - schmidt scalar product of the two mpos .
alternatively , one can contract this tensor network with tmrg . in this case
the meaning of states and dual states is changed as indicated on the right side of the figure .
one starts on the left and on the right with vertically oriented mps .
one column of local gates after another , the transfer matrices , is applied to those mps , again with intermediate truncation steps to reduce the bond dimensions .
finally , the expectation value is obtained by evaluating the scalar product of the two mps.,scaledwidth=80.0% ] the tdmrg approach _
scheme a _ corresponds to starting with a trivial mpo @xmath190 $ ] that represents the identity with bond dimensions @xmath34 @xmath191}$ ] .
beginning at the bottom , one then contracts one row of local gates after another to that mpo until reaching the row that contains operator @xmath3 , yielding @xmath192 $ ] . multiplying , in every step of this iteration , one row of gates with the current mpo in an exact manner , gives a resulting mpo with increased bond dimensions .
a subsequent truncation step reduces the bond dimensions again . the degree
to which the bond dimensions are reduced in the truncation steps is controlled by the predefined truncation weight which determines the accuracy of the computation . similarly , starting from the top , one contracts rows of gates to obtain @xmath193 $ ] . finally , one computes the response function by evaluating the hilbert - schmidt scalar product as visualized in figure [ fig : mpo ] . in the tmrg approach ,
the columns of local gates in figure [ fig : tdmrg_vs_dmrg ] are interpreted as transfer matrices @xmath194 ( also @xmath195 and @xmath196 ) and operate on a hilbert space of dimension @xmath197 , where @xmath198 denotes the dimension of the single - site hilbert space .
the last column corresponds to a pure state @xmath199 and the first column to a dual state @xmath200 .
those transfer matrices and states have matrix product representations with bond dimensions @xmath201 @xmath202 .
starting from the left and the right with the states @xmath203 or @xmath204 , respectively , one transfer matrix after another is applied with intermediate truncation steps , for example , until one reaches the column containing operator @xmath3 . the response function
is then given by the overlap of the two resulting mps , @xmath205[\hat t_{{\hat{b}}}\hat t\dotsb\hat t|\psi_l{\rangle}]$ ] .
the advantages of the tdmrg approaches , pursued in this work , are that , ( i ) , the number of matrices in the mpos is fixed by the lattice size @xmath20 instead of growing linearly with the time @xmath4 and inverse temperature @xmath71 in the tmrg approach ,
( ii ) , the operators @xmath2 and @xmath3 can have non - local support , ( iii ) , one can evaluate the spectral function for all times @xmath4 , up to the maximum reachable time , by two tdmrg runs whereas , in the tmrg approach , one has to do a new calculation for every time @xmath4 of interest .
concerning ( ii ) , it should be pointed out that non - local operators @xmath2 and @xmath3 would also be accessible in the tmrg approach as long as they are mpos with sufficiently small bond dimensions . in refs .
@xcite the focus was on autocorrelation functions of local operators in the thermodynamic limit @xmath206 . in this case , one does not need to contract one column after another , but can evaluate the response function directly from the left and right transfer matrix eigenvectors with maximum eigenvalue . when one uses the so - called infinite - system dmrg @xcite to obtain those eigenvectors ( a single build - up sweep for the finite imaginary - time lattice ; not to be mistaken with the algorithm presented in ref .
@xcite which generates translationally invariant mps for infinite systems ) , problem ( iii ) does not occur and autocorrelation functions can be evaluated for all accessible times
@xmath4 in a single run .
one should however be careful in applying infinite - system dmrg only , as this algorithm has several pitfalls @xcite .
at least for larger times , finite - system dmrg ( multiple sweeps ) should be necessary .
in this paper , i have studied and explained the computation costs of different tdmrg schemes for the efficient and precise evaluation of finite - temperature response functions of strongly correlated quantum systems . simplifying the notation to some extent by formulating everything in terms of mpos , elucidated the effects of quasi - locality on the costs . the new class of optimizable evaluation schemes , eqs . and
, typically outperforms the earlier schemes from the literature in terms of the maximum reachable times by a factor of @xmath207 .
this gives access to many more physical applications .
the novel _ scheme c _ , that requires no additional optimization and is near - optimal in many typical examples , can be expected to be the method of choice for future applications . for more complex models , like systems with dynamics on different energy scales , _ scheme c
_ still outperforms the older tdmrg methods , but one can achieve further substantial performance gains by first studying the costs for the computation of operators @xmath137 $ ] , and then using this information to determine the optimal parameters for the scheme . finally , it has been argued that the tdmrg schemes are in some respects favorable to corresponding tmrg variants . as a first application , _
scheme c _ has been used in ref .
@xcite to calculate , for 1d bosons in the quantum critical regime with dynamic critical exponent @xmath18 , the universal scaling function for the thermal spectral function
. discussions with j. barthel , c. karrasch , a. kolezhuk , i. p. mcculloch , t. nishino , s. sachdev , j. sirker , and u. schollwck as well as financial support through dfg for 801 are gratefully acknowledged .
b. nachtergaele and r. sims , in _ new trends in mathematical physics . selected contributions of the xvth international congress on mathematical physics _ , edited by v. sidoravicius ( springer , heidelberg , 2009 ) , pp
. 591614 .
please note that bond dimensions @xmath68 , as studied in this paper , depend to some extent on the particular implementation .
the logarithmic maximum bond dimensions @xmath208 and computation costs @xmath209 , shown in the density plots , are however robust measures .
particularly , @xmath210 is the so - called hartley entropy @xcite , and the @xmath211 limit of the rnyi entropy .
please note that the expressions , , , and for the response function in the different evaluation schemes give the same values , the exact @xmath212 of eq .
, only if the mpo representations of the operators in square brackets are exact corresponding to exponentially big bond dimensions with @xmath213 .
a guiding principle of the dmrg approach is that in typical condensed matter applications such exponentially big @xmath68 are not required .
corresponding truncations of the mpos result in slight deviations of the values from the different evaluation schemes .
it is essential to keep the errors in these mpo truncations controlled and small at all times .
it might seem restrictive to only provide data for the choice @xmath214 .
however , the cases where @xmath3 is a translate of @xmath215 are very similar .
this covers the typical cases of interest like @xmath216 , @xmath217 etc . if neither site @xmath218 nor site @xmath219 are close to the boundaries of the system , the resulting computation costs and maximum bond dimensions are very similar to those occurring for @xmath220 .
the peaks in the bond dimensions simply occur at different locations in the lattice , but have similar shape and height ; compare to figure [ fig : ab_sp63_slices ] .
mikeska and a. k. kolezhuk , in _ quantum magnetism _ ,
645 of _ lecture notes in physics _ , edited by u. schollwck , j. richter , d. j. j. farnell , and r. f. bishop ( springer , berlin , 2004 ) , pp . | this paper provides a study and discussion of earlier as well as novel more efficient schemes for the precise evaluation of finite - temperature response functions of strongly correlated quantum systems in the framework of the time - dependent density matrix renormalization group ( tdmrg ) .
the computational costs and bond dimensions as functions of time and temperature are examined for the example of the spin-@xmath0 xxz heisenberg chain in the critical xy phase and the gapped nel phase .
the matrix product state purifications occurring in the algorithms are in one - to - one relation with corresponding matrix product operators .
this notational simplification elucidates implications of quasi - locality on the computational costs . based on the observation that there is considerable freedom in designing efficient tdmrg schemes for the calculation of dynamical correlators at finite temperatures , a new class of optimizable schemes , as recently suggested in arxiv:1212.3570 ,
is explained and analyzed numerically . a specific novel near - optimal scheme that requires no additional optimization reaches maximum times that are typically increased by a factor of two , when compared against earlier approaches .
these increased reachable times make many more physical applications accessible . for each of the described tdmrg schemes
, one can devise a corresponding transfer matrix renormalization group ( tmrg ) variant . 0.4 |
as far as one - baryon processes are concerned , baryons have always been approximated as static objects at leading order in heavy - baryon chiral perturbation theory ( hbchpt ) , and recoil corrections are treated as subleading perturbations @xcite .
however , the phenomenological successes of covariant approaches , in which recoil corrections are in effect resummed to all orders , challenged this point of view @xcite .
the initial theoretical rationale for covariant treatment of baryons comes from ref .
@xcite ( also touched upon in refs .
@xcite ) , in which the pion - baryon triangle diagram was examined . figure [ fig_trangle ] shows the said diagram , where @xmath3 is the incoming 4-momentum of the baryon and @xmath4 the momentum transfer . with different vertexes
, this diagram contributes to various processes . in particular ,
when both baryonic external lines are on - shell the diagram contributes to most nucleon form factors , and the loop integral is a function of @xmath5 .
if the baryon is represented by the dirac field , the triangle diagram can be shown to have a branch - point singularity in the second riemann sheet of @xmath6 , an example of so - called anomalous threshold @xcite .
reference @xcite argued that the static - limit approximation is not aware of this second - sheet singularity , so a lorentz - invariant treatment , or at least resummation of the recoil correction in hbchpt , is necessary .
however , from a more puristic point of view toward effective field theory ( eft ) , symmetries and degrees of freedom are the only constraints one must stand by . along this line of thinking , if the anomalous threshold in the triangle diagram does turn out to account for important physics , one would like such an analytic structure to come out of the power counting , not the other way around . we will show that a peculiarity of kinematics in the region centered around the two - pion cut @xmath7 , where @xmath8 is the pion mass , allows the baryon propagator to be unexpectedly close to its mass shell .
this eventually leads to the loop integral being enhanced by @xmath2 , compared with the standard chpt counting .
the enhancement of this sort resembles that of two - baryon reducible diagrams @xcite , and it demands change of power counting just like in the two - baryon sector . the manuscript is structured as follows .
we first give a qualitative analysis on the diagram and point out the configurations of loop momenta that enhance the integral .
numerical results are shown in sec .
[ sec_numerics ] , followed by a short summary in sec .
[ sec_summary ] .
we choose the rest frame of the incoming baryon , so that its four - momentum is @xmath9 where @xmath10 is the baryon mass .
the outgoing baryon has four - momentum @xmath11 , with @xmath12 and @xmath13 as functions of @xmath6 : @xmath14 in the physical region , @xmath12 and @xmath13 are both positive real numbers ; hence , @xmath15 .
chpt is expected to work in certain regions in the complex @xmath6-plane where @xmath16 is sufficiently small .
note that the incoming baryon line can be taken to mean an outgoing antibaryon .
if the antibaryon is on - shell @xmath9 , the outgoing baryon is far off shell @xmath17 , unless @xmath6 is large and comparable to @xmath18 .
although both kinematic configurations baryon form factor and baryon - antibaryon pair production are very plausibly related by crossing symmetry , the latter is in principle out of the realm of chpt .
the loop integral being investigated is given by @xmath19 where @xmath4 and @xmath20 label momenta as depicted in fig .
[ fig_trangle ] .
we have so far used relativistic kinematics , even though the framework is not covariant .
the only assumption we have made related to hbchpt is that the internal nucleon line always propagates forward , so the intermediate states do not include any baryon - antibaryon pairs .
power counting la weinberg @xcite tries to capture long - range physics represented by a loop diagram , by inspecting contributions from virtual three - momenta that push at least a subset of the propagators near their mass shell .
when both pion propagators are on - shell , @xmath4 and @xmath20 satisfy the following equations : @xmath21 since @xmath22 , @xmath23 is found to a first approximation to be collinear with @xmath24 , plus corrections in powers of @xmath25 : @xmath26 and @xmath27 \ , .\label{eqn_k0}\ ] ] reasoning of the standard chpt counting starts by assuming external three - momenta to be of the same order as the pion mass : @xmath28 , where we have used @xmath29 to denote generically the size of external three - momenta . since the integral has dimension @xmath30^{-1}$ ] , loop momenta with @xmath31 do not contribute appreciably .
therefore , the meaningful integration volume of @xmath32 must be of order @xmath33 . in the physics region where @xmath34 and @xmath35 , the volume of @xmath36 and @xmath37 itself are too of order @xmath29 , according to eq . .
it follows immediately that the denominator of the baryon propagator is controlled by @xmath38 , whereas the recoil correction @xmath39 is subleading .
therefore , the static limit is justified . with the baryon propagator and the pion propagators
being counted respectively as @xmath40 and @xmath41 , we arrive at the standard counting that the loop integral is of order @xmath40 .
while the above argument covers the generic case of @xmath42 , it , however , needs modification for peculiar cases defined by special values of @xmath13 .
pertinent to our concern in the present paper is the unphysical region where @xmath43 , with @xmath44 ; therefore , @xmath6 is near the two - pion cut @xmath45 .
the condition still holds , that the most important @xmath23-modes are those around @xmath46 .
now that @xmath24 may take complex values , @xmath23 may too , i.e. , @xmath47 may be negative .
so it is not surprising that there exists a small region of @xmath23 centered around @xmath46 in which the pion energy takes a value smaller than @xmath8 by @xmath48 : @xmath49 where we have used @xmath50 .
although this is a small region of @xmath20 , with volume @xmath51 , all of the three propagators have large values because they are very close to their mass shell .
especially , @xmath37 in the denominator of the baryon propagator is now of order @xmath52 , so the recoil correction @xmath53 can no longer be considered subleading .
it follows that the baryon propagator scales as @xmath54 , and the pion propagators @xmath55 .
putting together these elements , we conclude that when @xmath6 is within a small window around @xmath0 , @xmath56 , the loop integral is @xmath57 , enhanced by a factor of @xmath58 as opposed to the aforementioned generic counting . here
the static limit is not a legitimate approximation taken by the baryon propagator : the leading recoil term @xmath59 must be retained in eq . .
the resummation of @xmath59 immediately reproduces the anomalous threshold @xcite , as it was rather anticipated .
we evaluate numerically the integral with only the first recoil term kept in the baryon propagator : @xmath60 .
a feynman parameter is first used to combine the pion propagators and then the following identity is used to incorporate the baryon propagator : @xmath61 we then proceed by integrating out @xmath20 after completing the square in @xmath20 , eventually arriving at @xmath62 we wish to compare the above integral with its relativistic version : @xmath63 which has a well - known branch point at @xmath64 .
the comparison is facilitated if the branch point of @xmath65 coincides that of @xmath66 .
this can be achieved by retaining @xmath67 and @xmath68 as defined in eq . .
if we had chosen to conform to nonrelativistic kinematics for the outgoing nucleon , @xmath69 , @xmath67 and @xmath68 would have been @xmath70 however , the discrepancy between eqs . and is not crucial for our qualitative statement regarding power counting . to see how the branch point arises at @xmath64 , we first notice that when @xmath71 the integrand has two poles in the @xmath72 plane , and they will pinch at @xmath73 when @xmath64 .
because the pinching in the @xmath72 plane happens at one of the end points of the @xmath74 integration , the corresponding singularity of the integrand is inevitable no matter how we deform the contours of @xmath74 and/or @xmath72 .
therefore , @xmath65 has a branch cut starting from @xmath64 , running toward @xmath75 along the positive real axis . for @xmath76
, the contours in both @xmath72 and @xmath74 planes need to be deformed so as not to be crossed by the poles of the integrand .
the chosen contours are shown by the solid lines in fig .
[ fig_xypoles ] .
in particular , the @xmath74 contour is a straight line from the origin to infinity , @xmath77 off the positive real axis .
when @xmath74 takes a value on that contour , the integrand in eq .
has corresponding poles in the @xmath72 plane .
the trajectories of these poles as @xmath74 moving along its contour are represented by the dashed lines in fig .
[ fig_xypoles ] ( a ) , with @xmath78 .
a similar story goes to fig .
[ fig_xypoles ] ( b ) , in which the @xmath74 contour is illustrated by the solid line and the poles associated with the @xmath72 contour are marked out by the dashed lines . on a side note
, there are other ways to evaluate the integral . for example
, one can first calculate the imaginary part , i.e. , the discontinuity along the branch cut , and then the real part by way of dispersion integral .
in fact , this is the way integral is evaluated . and
@xmath74 integrations , where @xmath79 .
the dashed lines represent the poles of the integrand in eq .
when @xmath78 ( see the text for more detailed explanation).,title="fig : " ] and @xmath74 integrations , where @xmath79 .
the dashed lines represent the poles of the integrand in eq . when @xmath78 ( see the text for more detailed explanation).,title="fig : " ] the numerical results for the real and imaginary parts of @xmath65 are plotted in fig .
[ fig_gamma_smallregion ] , with @xmath80 the ratio between the physical pion and nucleon masses .
the solid lines correspond to evaluation of @xmath65 according to eq .
[ eqn_doubleintegral ] . for comparison ,
we have also plotted the static - limit approximation ( dashed ) , which diverges at @xmath64 , and the relativistic results ( dot - dashed ) according to eq . .
the enhancement of @xmath65 around @xmath64 is clearly demonstrated by a peak in its real part and a rapid rise in its imaginary part .
the difference between the recoil and relativistic representations of the diagram is very small .
this indicates that the higher - order recoil corrections like @xmath81 are indeed subleading . in neighborhood of @xmath64 .
the dashed , dot - dashed , and solid lines correspond respectively to calculations in the static limit , with the relativistic baryon propagator [ eq . ] , and with the leading recoil term [ eq . ] .,title="fig : " ] in neighborhood of @xmath64 .
the dashed , dot - dashed , and solid lines correspond respectively to calculations in the static limit , with the relativistic baryon propagator [ eq . ] , and with the leading recoil term [ eq . ] .,title="fig : " ] the infrared enhancement is essentially the statement that within a small window centered on @xmath64 , @xmath65 scales like @xmath82 , rather than @xmath83 .
the lack of analytic expression for @xmath65 understandably obscures this point .
we wish to demonstrate in this section how the scaling changes near @xmath64 . for this purpose , we rescale both low - energy mass scales @xmath6 and @xmath84 by the same factor , while keeping the only large mass @xmath10 fixed , and then look at how @xmath65 varies with respect to the rescaling factor . more precisely , we define the following function to quantify the scaling behavior , @xmath85 with @xmath86 intentionally set up to correspond to the physical pion mass @xmath87 : @xmath88 when @xmath89 , where @xmath90 , we expect @xmath91 . when @xmath92 is far outside of the enhancement window , which will always happen for a fixed @xmath92 if @xmath93 is small enough , the standard chpt counting recovers and @xmath94 .
figure [ fig_scaling ] confirms the expected change of scaling .
shown in fig .
[ fig_scaling ] is the log - log plot of @xmath95 versus @xmath93 , for @xmath96 and @xmath97 .
when @xmath93 is of @xmath98 , @xmath99 clearly scales as @xmath100 , whereas the enhancement of the real part @xmath101 is milder , with a slope between @xmath102 and @xmath103 . owing to its qualitative nature , however , it should not be utterly surprising that the power - counting analysis in sec .
[ sec_enhancement ] could not capture this difference between the real and imaginary parts of @xmath104 .
in the range of @xmath93 we showed , @xmath105 is always outside of the enhancement window ; therefore , both @xmath106 and @xmath107 show a slope of @xmath108 in the log - log plots .
when @xmath93 is small enough , the enhancement window will be so narrow that for both values of @xmath92 , @xmath109 is outside of it ; therefore , @xmath110 . as functions of the scaling factor @xmath93 , with various values for @xmath92
[ see the text for the definition of @xmath95 ] .
the solid ( dot - dashed ) line has @xmath96 ( @xmath105 ) .
the dashed lines are there to assist gauging the slope of the solid lines : the steeper dashed line has slope @xmath103 , and the other has slope @xmath102.,title="fig : " ] as functions of the scaling factor @xmath93 , with various values for @xmath92 [ see the text for the definition of @xmath95 ]
. the solid ( dot - dashed ) line has @xmath96 ( @xmath105 ) .
the dashed lines are there to assist gauging the slope of the solid lines : the steeper dashed line has slope @xmath103 , and the other has slope @xmath102.,title="fig : " ]
we have shown an example of one - baryon processes with an infrared enhancement .
more specifically , when @xmath6 is inside a small window centered on the two - pion cut @xmath112 , the triangle diagram is enhanced by a factor of @xmath113 , compared with the standard chpt counting .
similar to the two - baryon sector , the infrared enhancement calls for resumming baryon recoil corrections .
the main reason for this change of power counting is that the pion three - momenta in the loop can take complex values in the unphysical region , which in turn diminishes the pion energy flowing into the baryon internal line .
therefore , there exists an enhancement window in which the loop integral is dominated by modes that take all propagators to be close to their mass shell .
we have evaluated numerically the loop integral to demonstrate the enhancement near @xmath64 .
it was also found that there is very small discrepancy between the resummation of recoil terms and the fully relativistic calculation .
more numerical evidence was shown to prove that the loop integral scales as @xmath41 inside the enhancement window .
this was done by rescaling small mass parameters @xmath24 and @xmath8 by the same factor @xmath93 , and investigating how the integral varies accordingly .
bwl thanks the nuclear theory group at beihang university and the institute for nuclear theory at the university of washington for hospitality when part of the work was carried out there .
this work was supported in part by the national natural science foundation of china ( nsfc ) under grant no .
11375120 .
e. e. jenkins and a. v. manohar , phys .
b * 255 * , 558 ( 1991 ) .
t. becher and h. leutwyler , eur .
j. c * 9 * , 643 ( 1999 ) [ hep - ph/9901384 ] .
t. fuchs , j. gegelia , g. japaridze , and s. scherer , phys .
d * 68 * , 056005 ( 2003 ) [ hep - ph/0302117 ] . v. pascalutsa and d. r. phillips , phys .
c * 67 * , 055202 ( 2003 ) doi:10.1103/physrevc.67.055202 [ nucl - th/0212024 ] .
v. lensky , j. m. alarcn , and v. pascalutsa , phys .
c * 90 * , no . 5 , 055202 ( 2014 ) [ arxiv:1407.2574 [ hep - ph ] ] .
j. m. alarcn , j. martin camalich , and j. a. oller , phys .
d * 85 * , 051503 ( 2012 ) [ arxiv:1110.3797 [ hep - ph ] ] . y. h. chen , d. l. yao , and h. q. zheng , phys .
d * 87 * , 054019 ( 2013 ) [ arxiv:1212.1893 [ hep - ph ] ] .
l. s. geng , j. martin camalich , l. alvarez - ruso , and m. j. vicente vacas , phys . rev
. lett .
* 101 * , 222002 ( 2008 ) [ arxiv:0805.1419 [ hep - ph ] ] .
j. martin camalich , l. s. geng , and m. j. vicente vacas , phys .
d * 82 * , 074504 ( 2010 ) [ arxiv:1003.1929 [ hep - lat ] ] .
ren , l. s. geng , j. martin camalich , j. meng , and h. toki , jhep * 1212 * , 073 ( 2012 ) [ arxiv:1209.3641 [ nucl - th ] ] .
l. geng , front .
china * 8 * , 328 ( 2013 ) [ arxiv:1301.6815 [ nucl - th ] ] .
j. gasser , m. e. sainio , and a. svarc , nucl .
b * 307 * , 779 ( 1988 ) .
v. bernard , n. kaiser , and u. g. meiner , nucl .
a * 611 * , 429 ( 1996 ) [ hep - ph/9607428 ] .
y. nambu , nuovo cimento , * 6 * , 1064 ( 1957 ) . | the pion - baryon triangle diagram is inspected for the special kinematic region in which the squared momentum transfer is close to @xmath0 .
the pion propagators can have very small energies , as opposed to @xmath1 in the physical region , which allows the nucleon propagator to be near its mass shell .
this observation leads us to conclude that in this particular domain the triangle diagram is augmented by @xmath2 compared with the standard counting of chiral perturbation theory , hence an infrared enhancement in the single - baryon sector . |
two - dimensional planar billiards are nonlinear systems with rich and interesting dynamical properties .
a point particle , moving with constant velocity within a closed boundary and exhibiting specular reflections on the walls , can have regular , mixed or fully chaotic dynamics , in strong dependence on details of the boundary shape . in physics ,
two - dimensional billiards offer good examples of coexistence of regular , mixed and chaotic dynamics in hamiltonian systems .
this type of behavior , illustrated by the standard map and explained by means of the kam - theorem , is present in many realistic phenomena , such as planetary systems and various types of coupled oscillators@xcite .
chaotic billiards were first introduced by sinai@xcite who considered the defocusing effects of circular scatterers in the two - dimensional lorentz gas .
after the important discovery by bunimovich@xcite that also the focusing circular arcs can lead to a fully chaotic behavior , many investigations were devoted to billiards with circular arcs and , in a smaller extent , to other types of curved boundaries
. the systematic mathematical description of chaotic billiards and an extended list of references can be found in the book by chernov and markarian@xcite .
rigorous investigations were concentrating on methods for producing fully chaotic billiards and on specific properties ( bernoulli , k - property , mixing and hyperbolicity ) expressing differences between chaotic systems@xcite .
various aspects of billiard dynamics have been extensively examined during last decades@xcite . in recent years
, properties of classical billiards and their quantum - mechanical counterparts were used to explain and improve performances of devices in microelectronics and nanotechnology , especially of optical microresonators in dielectrical and polymer lasers@xcite .
we are stressing the fact that notable regions of full chaos have been discovered in billiards with elliptical arcs and piecewise flat boundaries , indicating that such billiards deserve further attention@xcite . in our previous work we analyzed several types of billiards with noncircular arcs ( parabolic , hyperbolic , elliptical and generalized power - law ) , exhibiting mixed dynamics@xcite .
next we investigated , in the full parameter space@xcite , the elliptical stadium billiards ( esb ) , first introduced by donnay@xcite . here
we extend the same type of analysis to the truncated elliptical billiards ( teb ) , which although similar in appearance , have different dynamical properties .
the truncated elliptical billiard ( teb ) is defined by a two - parameter planar domain constructed by truncating an ellipse on opposite sides ( fig .
1 ) . a symmetrical stadium - like shape thus obtained consists of a rectangle with two elliptical arcs added at its opposite ends .
the corresponding billiard has been introduced by del magno@xcite who , investigating a restricted part of the parameter space and applying the mathematical method of invariant cones , determined the region of hyperbolic behavior and presented an estimate of the region where such billiard could be ergodic . in the present paper
we investigate numerically and analitically the truncated elliptical billiard ( teb ) in the full parameter space , by using two shape parameters @xmath0 and @xmath1 .
this description of the billiard geometry and dynamics is consistent with our previous analysis of the elliptical stadium billiard ( esb)@xcite , which is a two - parameter generalization of the bunimovich stadium billiard@xcite and is a special case of the mushroom billiard@xcite .
it has been confirmed by analysis and numerical computation@xcite that this billiard is fully chaotic ( ergodic ) for a sizeable but strictly limited region in the parameter space , defined by the stable two - bounce horizontal periodic orbit on one and the pantografic orbits on the other side .
our investigations of the esb and teb billiards confirm the suggestion by del magno@xcite that in spite of apparently similar stadium - like shapes , these two billiards have essentially different dynamical properties . in the present paper
we describe our analytical and numerical investigation of the truncated elliptical billiard and compare the obtained results with those for the elliptical stadium billiard . in section
ii we define the teb billiard boundary and describe its geometrical properties . in section
iii the existence and stability of selected orbits are discussed and illustrated by poincar ' e plots and orbit diagrams . in section iv the poincar ' e sections are used to estimate , by means of the box - counting numerical method , the degree of chaoticity for a given boundary shape .
the results are shown in the parameter - space diagram and compared with the same type of diagram for the elliptical stadium billiard . in section
v we briefly discuss the possible generalization of the truncated elliptical billiard providing a transition between two types of the stadium - like elliptical billiards . finally , in section vi we summarize the obtained results and propose further investigations .
in our parametrization the truncated elliptical billiard ( teb ) is defined in the @xmath2 plane by means of the two parameters @xmath0 and @xmath1 , satisfying conditions @xmath3 and @xmath4 . the billiard boundary is described as @xmath5 the horizontal diameter is normalized to 2 , so that the horizontal semiaxis of the ellipse is 1 .
the vertical semiaxis of the ellipse is @xmath6 , and the possible height @xmath7 of the billiard extends from @xmath8 to @xmath9 .
the horizontal length of the central rectangle is @xmath10 . in special cases , for @xmath11
the shape is a full ellipse , for @xmath12 it is rectangular , for @xmath13 it is a square and for @xmath11 and @xmath14 a full circle . for @xmath15
one obtains a set of truncated circle billiards , which separates two distinct billiard classes , one for @xmath16 with elongated elliptical arcs and the other with @xmath17 and flattened elliptical arcs .
1 shows three typical shapes of the truncated elliptical billiard with circular , flattened and elongated elliptical arcs .
the coordinates of the focal points are @xmath18\ ] ] for @xmath16 , and @xmath19\ ] ] for @xmath17 .
they contain the important term @xmath20 which is negative for @xmath16 , positive for @xmath17 , and zero for @xmath15 ( circular arcs ) .
this limit is shown as the thick circular line in fig .
2 presenting the structure of the @xmath21 parameter space . for @xmath22
the curvature radius is @xmath23^{3/2 } } { \gamma(1-\delta^2)}\ ] ] for @xmath24 the boundary is flat and the curvature radius is @xmath25 , but for @xmath26 has a discontinuity and drops to @xmath27^{3/2 } } { \gamma(1-\delta^2)}\ ] ] at the endpoints of the horizontal axis of the ellipse ( @xmath28 ) the curvature radius is @xmath29 which reduces to @xmath30 for circular arcs . for full ellipses with @xmath11
the curvature radius at @xmath31 is @xmath32 .
as explained in @xcite , the symbols @xmath33 , @xmath34 and @xmath35 , respectively , denote the angles which the normal , the incoming path and the outcoming path make with the x - axis .
the angle between the incoming ( or outcoming ) path and the normal to the boundary is @xmath36 .
the angle between the tangent to the boundary at the point t@xmath37 of impact and the incoming ( or outcoming ) path , needed in the computation of the orbit stability , is @xmath38 .
the angles @xmath33 , @xmath34 and @xmath35 are connected by the relation @xmath39 the expression ( 7 ) is the basis for finding the existence criteria for particular periodic orbits@xcite . in our further description
we refer to the impact points t@xmath37 in the first quadrant , with no loss of generality for the obtained results . in the poincar ' e sections the points p@xmath40 are obtained by plotting the slope of the velocity direction @xmath41 versus the x - coordinate of the intersection point with the x - axis , as explained in @xcite .
the poincar ' e diagrams obtained in this way are area preserving .
as described in @xcite , the stability of a periodic orbit is assured if the absolute value of the trace of the stability matrix @xmath42 is smaller than 2 , thus if @xmath43 such orbits are elliptic , and those with @xmath44 are neutral ( parabolic ) . the stability matrix of the closed orbit of period @xmath45 can be written as @xmath46 , where the @xmath47 matrix @xmath48 for two subsequent impact points t@xmath49 and t@xmath50 , connected by a rectilinear chord of the length @xmath51 , is @xmath52
this subfamily of truncated elliptical billiards has elongated elliptical arcs . in fig .
3(a - d ) we show poincar ' e sections for @xmath53 and different values of @xmath54 .
similar results for @xmath55 and @xmath56 are shown in fig .
4(a - d ) .
these pictures reveal a highly chaotic behavior .
there are no elliptic islands , however , flights of points typical for neutral orbits can be discerned .
this is remarkably different from the corresponding results for the elliptical stadium billiards@xcite , where in the same parameter region there were many fixed points and elliptic islands due to stable pantographic and other orbits .
we investigate the existence and stability of the bow - tie orbit ( the lowest pantographic orbit ) , shown in fig .
this orbit exists if the coordinates @xmath57 and @xmath58 of the impact point and the derivative @xmath59 of the boundary at this point satisfy the equation ( 7 ) , which now reads @xmath60 giving as solution the coordinates of the point of impact @xmath61 and @xmath62 the condition @xmath63 that this point should lie on the elliptical part of the boundary leads to the requirement @xmath64 this limit is shown in fig . 2 and is denoted with the letter a.
if we denote the points with positive @xmath57 by 1 and the points on the negative side by -1 , the deviation matrix can be calculated as @xmath65 the corresponding angle @xmath66 needed in the matrix ( 9 ) is given by @xmath67 the chords are @xmath68 and @xmath69 the curvature radius at the impact point is obtained by substituting ( 11 ) into ( 4 ) and reads @xmath70 if we define @xmath71 the trace of the deviation matrix is @xmath72\ ] ] the left - hand side of the stability condition ( 8) is valid automatically , but the right - hand side is fulfilled only if @xmath73 by substituting ( 15 ) , ( 16 ) , ( 17 ) and ( 18 ) into ( 19 ) , one obtains @xmath74 for all allowed cases .
the conclusion is that the bow - tie orbit is neutral for all parameter values satisfying the existence condition .
further we investigate properties of the rectangular orbit shown in fig .
according to ( 7 ) , this orbit exists if the derivative on the boundary is @xmath75 .
corresponding solutions for the impact point are @xmath76 and @xmath77 the condition @xmath63 leads to the requirement @xmath78 this limit is shown in fig . 2 denoted by letter f. stability is calculated with equation ( 9 ) and the matrix ( 14 ) , where the angle @xmath66 is given by @xmath79 .
the chords are @xmath80 and @xmath81 and the curvature radius is @xmath82^{3/2}\ ] ] again , the trace is given by ( 20 ) , and for this case one obtains @xmath83 .
the conclusion is that also this orbit is neutral for all shapes , both flattened and elongated , allowed by ( 24 ) .
the elongated truncated elliptical billiards were discussed in @xcite .
their boundary shapes were described by means of two parameters @xmath84 and @xmath85 , related to our parameters @xmath0 and @xmath1 as follows : @xmath86 in @xcite the billiards with @xmath87 and @xmath88 have been analysed and the hyperbolic behavior has been identified in the region @xmath89 . in our parameters , this corresponds to the quasi - triangular region in the parameter space , denoted by a and b in fig .
2 , delimited by curves @xmath90 ( denoted in fig . 2 by letter e ) and
@xmath15 and by the straight line @xmath91 .
the region @xmath92 is rigorously proved to be ergodic@xcite . written with our parameters , it obeys the conditions @xmath93 @xmath94 the corresponding part of the parameter space in fig .
2 is the one denoted by a. the comparison with our results shows that the limit ( 27 ) or ( 28 ) is identical to the limit ( 24 ) in the parameter space , where the parabolic rectangular orbits emerge .
this part of the parameter space , with flattened elliptical arcs , had not been investigated previously . in fig .
3(e - h ) we show poincar ' e sections for @xmath53 and different values of @xmath95 . in this parameter region dynamics
is following the kam scenario .
similar behavior is noticed for values @xmath55 and @xmath96 ( fig .
4(e - h ) ) .
elliptic islands corresponding to the horizontal two - bounce and some other orbits are visible , similarly to the corresponding results for the elliptical stadium billiards@xcite .
we investigate the existence and stability criteria for these orbits . the horizontal two - bounce orbit ( fig .
5(c ) ) obviously exists for all combinations of @xmath0 and @xmath1 , but according to @xcite the stability condition @xmath97 takes the form @xmath98 so that bifurcations giving birth to stable diametral orbits appear at the values @xmath15 , corresponding to circular arcs . in the poincar ' e diagrams this orbit and the surrounding quasiperiodic orbits are visible as two large bands near @xmath99 .
according to ( 7 ) a tilted two - bounce orbit ( fig .
5(d ) ) exists at the point t@xmath37 on the billiard boundary with derivative @xmath59 if @xmath100 this is realized for any @xmath63 provided that @xmath101 thus only for the truncated circle . since in this case
the chord in ( 29 ) is @xmath102 and the radius is @xmath103 , these orbits are neutral . the diamond orbit of period four , shown in fig .
5(e ) , exists for any parameter choice .
it has two bouncing points at the ends of the horizontal semiaxis , and the other two on the flat parts on the boundary . to assess its stability
, one should calculate the stability matrix @xmath104 .
the angles contained in the matrix are given as @xmath105 where @xmath106 the curvature radius at @xmath107 is given by ( 6 ) .
this leads to the trace @xmath108\ ] ] and to the condition @xmath109 or @xmath110 , thus the stable diamond orbit appears when @xmath111 this limit is shown in fig .
2 as the line denoted by letter c. the multidiamond orbit of order @xmath112 is the orbit of period @xmath113 , which has two bouncing points at the ends of the horizontal axis and @xmath114 bouncing points on the flat parts of the boundary ( fig . 4 ( f ) ) .
such orbit exists if @xmath115 as explained for a similar case in @xcite , the chord @xmath116 in ( 35 ) should be replaced by @xmath117 where , for the truncated elliptical billiard , @xmath118 the trace of the stability matrix is then @xmath119\ ] ] with @xmath120 given by ( 6 ) .
the resulting condition for the stability of the multidiamond orbit is @xmath121 the limiting curves ( 41 ) are plotted in fig .
2 . the line with @xmath122 is denoted by letter d , and above it
there are several lines for @xmath123 . for @xmath124 the minimal values of @xmath0 above which the multidiamond orbits appear are @xmath125 the emergence of multidiamond orbits can be followed by observing the poincar ' e sections for a set of shapes with @xmath126 ( fig .
the values of this parameter for which an orbit of new @xmath112 appears are given as intersections of the straight line @xmath126 with curves ( 41 ) , and obey the equation @xmath127 for the diamond orbit ( @xmath128 ) this equation reads @xmath129 and the orbit appears for @xmath130 for the same type of boundary the stable two - bounce orbit appeared at @xmath131 the hour - glass orbit ( fig .
5(g ) ) looks like the bow - tie orbit rotated by @xmath132 .
it exists if the coordinates @xmath57 and @xmath58 of the impact point and the derivative @xmath59 of the boundary at this point satisfy the equation @xmath133 giving as solution the coordinates of the impact point @xmath134 and @xmath135 the condition @xmath63 that this point should lie on the elliptical part of the boundary leads to the requirement @xmath136 @xmath137 these limits define the region shown in fig . 2 denoted by letter b.
if we denote the points with positive @xmath58 by 1 and the points on the negative side by -1 , the deviation matrix can be calculated from ( 14 ) .
the angle @xmath66 needed in the calculation is given as @xmath138 the curvature radius at this point is given as @xmath139 the chords are @xmath140 and @xmath141 if we define @xmath142 as in ( 19 ) , the trace of the deviation matrix is again given by ( 20 ) and the orbit is stable if @xmath143 .
when we substitute the calculated values of @xmath120 , @xmath116 , @xmath144 and @xmath145 into ( 19 ) , we obtain @xmath74 for all allowed shapes and conclude that the hour - glass orbit is neutral .
this means that in the truncated elliptical billiards ( teb ) there is no stable hour - glass orbit , at variance with the elliptical stadium billiard ( esb ) , where such an orbit having interesting properties was stable in a large fraction of the parameter space@xcite . besides the diamond and multidiamond orbits , in fig .
6 one discerns the presence of another , `` 8-shaped '' , stable orbit , shown in fig .
in this section we return to the question of limits within which the truncated elliptical billiard is fully chaotic . here
we test these limits numerically , with the help of the box - counting method@xcite .
we calculate the poincar ' e sections for a chosen pair of shape parameters , starting with @xmath146 randomly chosen sets of initial conditions and iterating each orbit for @xmath147 intersections with the x - axis , thus obtaining @xmath148 points in the poincar ' e diagram .
then we divide the first quadrant of the phase plane into a grid of @xmath149 squares ( boxes ) , count the number of boxes which have points in them and calculate the ratio of this number to the total number of boxes .
the obtained ratio is denoted by @xmath150 . in this way
also certain points belonging to invariant curves within the regular islands are included .
but since our main aim is to examine the onset of full chaos , this method gives satisfactory results , providing that the appropriate values of @xmath146 , @xmath147 and @xmath112 are used .
detailed testing has shown that reliable results are obtained for values @xmath151 , @xmath152 and @xmath153 used in our present calculation@xcite . in fig .
7 we plot in the @xmath154 plane the points representing the pairs of shape parameters .
points are plotted in different colors , depending on the corresponding value of @xmath155 .
the full chaos , corresponding to @xmath156 , is depicted by black points .
colored points denote shapes within intervals between 0 and 0.99 .
this diagram confirms that for the truncated elliptical billiards ( teb ) , in the region below the onset of the stable two - bounce horizontal orbit , dynamics is practically completely chaotic .
this is in strong contrast with the behavior of the elliptical stadium billiard for which the similar diagram is shown in fig .
8 . for the esb
billiards the region of chaos was strictly bounded also from the lower side and determined by emergence of stable pantographic orbits . to examine the possible mechanism for this difference
, we assume that the teb and the esb billiards are two extreme cases and search for a possible transition between them .
in this section we propose a new large class of stadium - like billiards which we call generalized truncated elliptical stadium - like billiards ( gtesb ) .
such a billiard depends on three shape parameters @xmath0 , @xmath1 and @xmath157 .
the allowed values of the shape parameters are @xmath158 for the limiting values of @xmath157 we obtain the two billiard families considered before : for @xmath159 gtesb reduces to the elliptical stadium billiard ( esb ) , and for @xmath160 gtesb becomes the truncated elliptical billiard ( teb ) .
the new gtesb billiard boundary is obtained by adding elliptical arcs symmetrically at the two opposite ends of a rectangle with sides @xmath10 and @xmath7 .
elliptical arcs are cut out from the two identical but generally detached ellipses by two horizontal straight lines at @xmath161 ( fig .
the two ellipses have centers at the points @xmath162 the distance between the two centers is @xmath163 .
the horizontal and vertical semiaxis are given , respectively , as @xmath164 and @xmath165\ ] ] and the equation of the two ellipses reads @xmath166 the horizontal diameter of the billiard is 2 . for @xmath159 and @xmath167
the gtesb becomes the bunimovich stadium billiard . in fig .
10 the poincar ' e sections are shown for @xmath168 and @xmath169 with @xmath157 assuming different values between -1 and 1 .
the four islands typical for the bow - tie orbit are present for all @xmath157 except for @xmath160 ( teb ) , where this orbit becomes neutral and the island reduces to a caracteristical flight of points .
the limits separating chaotic from mixed behavior are determined by the onset of the stable horizontal 2-bounce orbit and are given by ( 29 ) . since the curvature radius at @xmath28 is @xmath170}{2(1-\delta)(1+\delta\kappa)}\ ] ] the upper limit of chaos is determined by the condition @xmath171 in fig .
11 the chaotic fraction @xmath155 is shown for the special case @xmath126 for different @xmath157 , in dependence on @xmath1 .
it is noticed that in the case @xmath159 ( esb ) there is a narrow , strictly limited region of full chaos , outside of which the values of @xmath155 are low . for @xmath160 ( teb )
the fully chaotic region is much larger and extends practically over all values @xmath0 and @xmath1 below the chaotic limit . between these two limits , the regions of full chaos ( @xmath172 ) are shorter and limited , but there are many shapes with chaotic parameter close to 1 ( between 0.90 and 0.99 ) .
this corresponds to a selection of narrow islands in the poincar ' e plots , as seen in fig .
in conclusion , our investigation of the elliptical stadium - like billiards has revealed a rich variety of integrable , mixed and chaotic behavior , which is connected with the character of the two elliptical arcs and with their mutual position .
this strong dependence on parameters @xmath0 and @xmath1 is confirmed for the truncated elliptical billiards , but is even more enhanced when a third shape parameter @xmath157 is added .
analysis shows , however , that among all considered shapes the truncated elliptical billiard ( teb ) , created by cutting a single ellipse with two parallel straight lines , has exceptional properties , notably that it is chaotic practically in the whole region of elongated elliptical arcs .
notable is the presence of many neutral orbits in this region , consistent with the fact that these orbits actually can be identified as orbits in an ellipse . for the flattened arcs ,
the stable islands due to the two - bounce horizontal orbit and to the diamond and multidiamond orbits occupy an important part of the phase plane
. our investigations can be useful for the experimental application of billiards in the laser technology , where properties and directional intensities of the optical microresonators depend strongly on the boundary shape .
they can also be applied in designing the semiconducting optical devices and in the technology of microwave and acoustic resonant cavities . with this purpose in mind ,
we propose further analysis of the stadium - like billiards with elliptical arcs and an extension of the present investigation to different types of open billiards .
authors are thankful to a. b " acker , m. lebental , n. pavin , t. prosen , m. robnik and t. tanaka for useful discussions and comments and to v. danani ' c and d. radi ' c for help with numerical methods and computation . | chaotic properties of symmetrical two - dimensional stadium - like billiards with elliptical arcs are studied numerically and analytically . for the two - parameter truncated elliptical billiard the existence and linear stability of several lowest - order periodic orbits are investigated in the full parameter space .
poincar ' e plots are computed and used for evaluation of the degree of chaoticity with the box - counting method .
the limit of the fully chaotic behavior is identified with circular arcs . above this limit , for flattened elliptical arcs , mixed dynamics with numerous stable elliptic islands
is present , similarly as in the elliptical stadium billiards . below this limit
the full chaos extends over the whole region of elongated shapes and the existing orbits are either unstable or neutral .
this is conspicuously different from the behavior in the elliptical stadium billiards , where the chaotic region is strictly bounded from both sides . to examine the mechanism of this difference , a generalization to a novel three - parameter family of boundary shapes
is proposed and suggested for further evaluation . |
the capacitated vehicle routing problem ( @xmath3 ) is an extensively studied combinatorial optimization problem ( see e.g. , the book @xcite and references therein ) . @xmath3is defined on a metric space @xmath4 , where @xmath5 is a finite set of locations / vertices and @xmath6 a distance function that is symmetric and satisfies triangle inequality .
there is a depot vertex @xmath7 that contains an infinite supply of an identical item , and each vertex @xmath8 demands some units @xmath9 of this item .
a single vehicle of capacity @xmath10 is used to distribute the items .
the objective is to find a minimum length tour of the vehicle that satisfies all demands subject to the constraint that the vehicle carries at most @xmath11 units at any time .
@xmath3is closely related to the traveling salesman problem ( @xmath12 ) .
it is clear that @xmath3reduces to @xmath12 in the absence of capacity constraint .
more interestingly , a reverse relation is also known essentially the best known approximation algorithm for @xmath3 @xcite achieves a guarantee of @xmath13 , where @xmath14 is the best approximation ratio for @xmath12 . in practice
, it is natural to have a fleet of _ multiple _ vehicles that can run in parallel .
the objective can then be to either minimize the sum of completion times of all the vehicles or to minimize the maximum completion time over all vehicles ( or the makespan of the routing ) .
furthermore the vehicles can all be identical ( same speed ) or heterogeneous ( have different speeds ) . in either case , it is not hard to see that the total completion time objective reduces to the usual @xmath3on a single maximum - speed vehicle , and constant - factor approximation algorithms readily follow . when the objective is to minimize the makespan with identical vehicles , ideas for approximating the regular @xmath3problem using a tour - splitting heuristic introduced by frederickson et al .
@xcite can be easily adapted to derive a constant - factor approximation algorithm ( see below ) .
this motivates the _ heterogenous capacitated vehicle routing problem _ ( @xmath15 ) that we consider . in this problem , a fleet of @xmath1 vehicles with
non - uniform speeds _ and uniform capacities is initially located at the depot vertex @xmath0 .
the objective is to satisfy the demands subject to the capacity constraints while minimizing the makespan .
our main result is a constant - factor approximation algorithm for @xmath15 .
most of our algorithmic ideas lie in solving the special case of @xmath15when there is no capacity constraint .
this problem , which we call @xmath16 , is a generalization of @xmath12that might be of independent interest . for most of this paper ,
we will focus on obtaining a constant - factor approximation for @xmath16 .
* tour - splitting solutions : * to illustrate the use of known techniques , we outline how to obtain a constant - factor approximation algorithm for @xmath16with uniform speeds @xcite .
first , notice that the union of the tours of @xmath17 connects all vertices , and hence a minimum spanning tree has length at most @xmath18 .
then consider an @xmath19 , duplicate its edge and take an euler tour @xmath20 , which is of length @xmath21 .
now split @xmath20 into @xmath1 segments of lengths at most @xmath22 by removing edges .
finally , the tour for the @xmath23 vehicle is obtained by connecting both endpoints of the @xmath23 segment of @xmath20 to the depot .
since twice the distance from the depot to any vertex is a lower bound on @xmath17 , the length of each tour is at most @xmath24 and hence this solution is a 3-approximation .
we remark that this can be extended to obtain an @xmath25-approximation for @xmath15with uniform speeds ( e.g. , using theorem [ th : vrptotsp ] in section [ sec : hvrp ] ) . at a very high level
, this strategy has two main components : ( 1 ) partitioning an @xmath19into manageable - sized connected parts ; ( 2 ) assigning these parts to vehicles .
this simple idea which was already present in the 70s is the central piece of many heuristics and approximations for vehicle routing problems ( e.g. , @xcite ) .
however , it is not clear how to employ this technique in the presence of vehicles with multiple speeds .
this is because the two main components now need some correlation : a small part of the @xmath19 , which should be assigned to a slower vehicle , must also be relatively closer to the depot in order to be reachable by this vehicle .
* set - cover based solutions : * for @xmath16with non - uniform speeds , previous approaches seem to give only a logarithmic approximation , as follows .
guess the optimal makespan @xmath26(within a constant factor ) .
if each vehicle of speed @xmath27 is given a length budget of @xmath28 , then the vehicles can collectively cover all vertices .
using an approximation algorithm for @xmath1-@xmath19 @xcite ( or the related orienteering problem @xcite ) within a maximum - coverage framework ( see e.g ..
@xcite ) , we can obtain tours of length @xmath26that cover a constant fraction of all vertices .
repeating this coverage step until all vertices are covered gives a solution to @xmath16of makespan @xmath29 .
the intrinsic problem of this approach is that it is too general in fact , the above algorithm also yields a logarithmic approximation even in the setting where the metric faced by each vehicle is arbitrary ( instead of just scaling by its speed ) , and this generalization of @xmath16can be shown to be set - cover hard .
it is unclear whether the approximation of this set - covering based approach can be improved for @xmath16 .
we extend the first tour - splitting approach described above to obtain the following result .
there are constant - factor approximation algorithms for @xmath16and @xmath15 . in order to obtain the approximation for @xmath16
, we abstract the requirements of the two components in the tour - splitting strategy . as a preprocessing step
, we round the speeds of vehicles to powers of two and guess the optimum makespan @xmath30 .
first , we specify conditions which guarantee that a collection of @xmath0-rooted trees is `` assignable '' , that is , each vehicle can visit the nodes of the trees assigned to it within time @xmath31 ( definition [ def : assignable ] ) .
the conditions in definition [ def : assignable ] are based on the lp to obtain a 2-approximation for _ scheduling in unrelated parallel machines _ by lenstra et al .
@xcite .
secondly , instead of partitioning an @xmath19as in the previous section , we consider more structured spanning trees which we call level - primtrees .
consider grouping the vertices ordered according to their distance from @xmath0 into levels , where the @xmath32th level includes all vertices within distance @xmath33 .
can only be served by vehicles of speed @xmath34 or higher given the makespan bound @xmath30 .
] the level - primtree is simply the tree resulting from running prim s algorithm with the restriction that all nodes in a level are spanned before starting to pull in nodes from the next .
a level - primtree has two important properties : ( i ) the vertices along every root - leaf path are monotonically nondecreasing in level and ( ii ) for every suffix of levels , the subgraph induced on it costs at most @xmath25 times its induced @xmath19 .
the first condition , which is the departing point from @xmath19s , greatly simplifies the decomposition procedure carried in the next step .
the second property is related to the assignability conditions in definition [ def : assignable ] and guarantees that the we can decompose a level - primtree into an assignable collection .
these properties are formalized in theorem [ thm : lprim ] .
the level - primconstruction combine both @xmath19and shortest - path distances from a root , so it is not surprising that this structure is related to _
light approximate shortest - path trees _ ( last ) introduced by khuller et al .
indeed , we use the existence of a suitably defined lastin proving theorem [ thm : lprim ] .
we remark , however , that the properties guaranteed by lasts are not enough for our purposes ( see section [ sec : levelprim ] ) . the third main component of our approximation for @xmath16is decomposing level - priminto an assignable collection of @xmath0-rooted trees . roughly , we partition the edges of level - priminto subtrees while ensuring that each subtree consisting of vertices in levels up to @xmath32 ( and hence is at a distance of about @xmath33 from the root ) also has length approximately @xmath33 , and thus can be assigned to a vehicle of speed about @xmath34 .
this partition , which relies on the two properties of level - prim , gives a collection of _ unrooted _ trees which is assignable .
due to the length of these trees , the extra distance to connect them to the root @xmath0 can be charged to their edges , hence this collection can be turned into a @xmath0-rooted assignable collection . in order to obtain an approximation to @xmath15 , we reduce this problem to approximating @xmath16 in a suitably modified metric space .
this new distance function encodes any additional trips to and from the root that a vehicle has to make if it runs out of capacity .
the exact transformation is presented in section [ sec : hvrp ] . for the @xmath3 , the best known approximation ratio @xcite is essentially @xmath13 where @xmath14 is the best guarantee for @xmath12 .
the current best values for @xmath14 are @xmath35 for general metrics @xcite , and @xmath36 ( for any constant @xmath37 ) for constant dimensional euclidean metrics @xcite .
this has been improved slightly to @xmath38 when @xmath39 @xcite .
recently , das and mathieu @xcite gave a quasi - polynomial time approximation scheme for @xmath3on the euclidean plane .
several variants of @xmath12have been studied , most of which have a min - sum objective .
one related problem with min - max objective is _ nurse station location _
@xcite , where the goal is to obtain a collection of trees ( each rooted at a distinct depot ) such that all vertices are covered and the maximum tree length is minimized .
even et al .
@xcite gave a 4-approximation algorithm for this problem .
this is based on partitioning the @xmath19and assigning to trees along the lines of section [ sec : previoustech ] ; their second step , however , involves a non - trivial bipartite matching subproblem .
in proving the properties of level - prim , we use _ light approximate shortest - path trees _ introduced by khuller , raghavachari and young @xcite , building on the work on shallow - light trees of awerbuch , baratz and peleg @xcite .
an @xmath40-lastis a rooted tree that has ( a ) length at most @xmath41 times the @xmath19and ( b ) the distance from any vertex to the root ( along the tree ) is at most @xmath42 times the distance in the original metric .
khuller et al .
@xcite showed that every metric has an @xmath43-last(for any @xmath44 ) and this is best possible .
one phase of our algorithm uses some ideas from scheduling on parallel machines @xcite , which also has a min - max objective . in this problem
, job @xmath45 has processing time @xmath46 on machine @xmath32 and the goal is to assign jobs to machines while minimizing the maximum completion time .
lenstra et al .
@xcite gave an lp - based 2-approximation algorithm for this problem .
the input to the _ heterogenous tsp _ ( @xmath16 ) consists of a metric @xmath4 denoting distances between vertices , a depot @xmath7 and @xmath1 vehicles with speeds @xmath2 greater than or equal to 1 .
the goal is to find tours @xmath47 ( starting and ending at @xmath0 ) for each vehicle so that every vertex is covered in some tour and which minimize the _ maximum completion time _ @xmath48 . at the loss of a factor of two in the approximation
, we assume that the @xmath49 s are all ( non - negative integral ) powers of @xmath50 .
then , for each integer @xmath51 we use @xmath52 to denote the number of vehicles with speed @xmath34 .
we let @xmath26denote the optimal value of this modified instance of @xmath16 .
we let @xmath53 be the complete graph on vertices @xmath5 with edge - weights corresponding to the distance function @xmath54 .
for any set @xmath55 of edges , we set @xmath56 . given any ( multi)graph @xmath57 and a subset @xmath58 of its vertices , @xmath59 $ ] denotes the subgraph induced on @xmath58 and @xmath60 denotes the graph obtained by contracting vertices @xmath58 to a single vertex ( we retain parallel edges ) .
moreover , for any pair of vertices @xmath61 in @xmath57 , we use @xmath62 to denote the length of the shortest path in @xmath57 between @xmath63 and @xmath64 .
assume that we have correctly guessed a value @xmath30 such that @xmath65 .
( this value can be found via binary search and we address this in the end of section [ sec : decomp ] . )
we partition the set of vertices @xmath5 according to their distance to @xmath0 : @xmath66 @xmath67\ } , \mbox { for all } i\ge 0.\ ] ] the vertices in @xmath68 are referred to as _ level @xmath32 _ vertices . for any @xmath51
, we use @xmath69 as a shorthand for @xmath70 and similarly @xmath71 .
we also define the _ level of an edge _ @xmath72 as the larger of the levels of @xmath63 and @xmath64 . for each @xmath51
, @xmath73 denotes the edges in @xmath74 of level @xmath32 .
note that @xmath75 for all @xmath76 , since both end - points of @xmath77 are in @xmath69 and the triangle inequality bounds its distance by the two - hop path via the root .
we use the notation @xmath78 and @xmath79 .
we start by studying collections of trees that can be assigned to vehicles in a way that each vehicle takes time @xmath31 to visit all of its assigned trees .
[ def : assignable ] a collection of @xmath0-rooted trees @xmath80 covering all vertices @xmath5 is called @xmath40-_assignable _ if it satisfies the following properties . 1 . for each @xmath51 and every @xmath81 , @xmath82 .
2 . for each @xmath51 , @xmath83 . intuitively , the trees in @xmath84 can be assigned to vehicles with speed @xmath34 so as to complete in time @xmath85 .
condition ( 2 ) guarantees that the trees @xmath86 targeted by vehicles of speed @xmath34 and above stand a chance of being handled by them within makespan @xmath87 .
interestingly , these minimal conditions are enough to eventually assign all trees in collection to vehicles while guaranteeing makespan @xmath88 . [
lem : assignable ] given an assignable collection @xmath80 of @xmath0-rooted trees , we can obtain in polynomial time an @xmath89-approximation for @xmath16 . to prove this lemma is also possible , but
the route we take reveals more properties of the requirement at hand and could potentially be useful in tackling generalizations of @xmath16 . ] , we show that condition ( 2 ) guarantees the existence of a fractional assignment of trees where each vehicle incurs load at most @xmath90 . then using condition ( 1 ) and a result on scheduling on parallel machines @xcite , we round this assignment into an integral one while increasing the load on each vehicle by at most @xmath91 .
we loose an extra factor of 2 to convert the trees into routes . [
[ fractional - assignment . ]
] fractional assignment .
+ + + + + + + + + + + + + + + + + + + + + + consider the bipartite graph @xmath57 whose left side contains one node for each tree in @xmath92 and whose right side contains one node for each vehicle .
( we identify the nodes with their respective trees / vehicles . )
there is an arc between the tree @xmath81 and a vehicle of speed @xmath93 if @xmath94 .
consider the following @xmath95-matching problem in @xmath57 : for each tree @xmath81 , we set @xmath96 and for each vehicle @xmath63 of speed @xmath93 we set @xmath97 .
a ( left - saturating ) @xmath95-matching is one which fractionally assigns all @xmath98 units of each tree @xmath99 such that no vehicle @xmath63 is assigned more than @xmath100 units .
notice that a feasible @xmath95-matching gives a fractional assignment of trees where each vehicle incurs load at most @xmath90 .
then our goal is to show the existence of a @xmath95-matching in @xmath57 . using a standard generalization of hall s theorem ( e.g. ,
see page 54 of @xcite ) , we see that @xmath57 has a feasible @xmath95-matching iff for every set @xmath5 of trees , @xmath101 is at most @xmath102 , where @xmath103 is the neighborhood of @xmath5 .
however , the structure of @xmath57 allows us to focus only on sets @xmath5 which are equal to @xmath104 for some @xmath32 . contains a tree in @xmath84 then @xmath103 already contains all vehicles of speed @xmath93 for @xmath94 . then adding to @xmath5 extra trees in @xmath104 does not change its neighborhood and thus leads to a dominating inequality . ] using this revised condition , @xmath57 has a @xmath95-matching iff for all @xmath32 , @xmath105 . since this is exactly condition ( 2 ) in definition
[ def : assignable ] , it follows that @xmath57 indeed has a @xmath95-matching ( which can be obtained in polynomial time using any maximum flow algorithm @xcite ) .
[ [ scheduling - parallel - machines . ] ] scheduling parallel machines .
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + we show how to round the fractional assignment obtained in the previous section .
we consider each tree as a `` job '' and each vehicle as a machine " , where the `` processing time '' @xmath106 of a tree @xmath99 in a vehicle @xmath63 of speed @xmath93 is @xmath107 ; then the `` makespan '' of a vehicle is exactly equal to the sum of the processing times of the trees assigned to it .
let @xmath108 denote the fraction of tree @xmath99 assigned to vehicle @xmath63 given by scaling down a @xmath95-matching in @xmath57 ( i.e. , if the matching assigns @xmath54 units of @xmath99 to vehicle @xmath63 , we have @xmath109 ) .
the feasibility of the matching gives @xmath110 for all @xmath63 .
moreover , by construction of the edges of @xmath57 , @xmath111 for @xmath81 implies that @xmath63 has speed at least @xmath112 .
then using property ( 1 ) of assignable trees we get that @xmath111 implies @xmath113 .
these two properties guarantee that @xmath114 is a feasible solution for the natural lp formulation for the scheduling problem with a feasible makespan value of @xmath90 and the maximum processing time @xmath115 set to @xmath116 .
theorem 1 of @xcite then asserts that @xmath114 can be rounded into an integral assignment of trees to vehicles such that the load on any vehicle is at most @xmath117 .
as in section [ sec : previoustech ] , we can transform each tree in @xmath118 into a cycle while at most doubling its length , which then gives a @xmath119 approximation for @xmath16 .
this concludes the proof of lemma [ lem : assignable ] . in order to obtain an assignable collection of @xmath0-rooted trees for our instance , we formally introduce level - primtrees .
these are the trees obtained by the the following procedure .
[ alg : stagei ] for each @xmath120 , let @xmath121 be an @xmath19for @xmath122 / v _ { < i}$ ] . @xmath123 .
note that level - primtrees can alternately be defined by modifying prim s algorithm such that nodes in level @xmath32 are only considered to be added to the tree after all nodes in levels below @xmath32 have already been added .
[ thm : lprim ] a level - primtree @xmath124 satisfies the following : the vertex - levels along every root - leaf path are non - decreasing . for each @xmath51 , @xmath125 .
note that the second property in theorem [ thm : lprim ] mirrors the second property in definition [ def : assignable ] .
a formal connection between the two is established via the following lemma that uses an optimal vehicle routing solution to derive a feasible spanning tree connecting a suffix of the level sets .
[ lem : lb ] for each level @xmath126 , @xmath127 .
consider an optimal solution for @xmath16and let @xmath128 be the set of edges traversed by vehicles in this solution ; label each edge in @xmath128 by the vehicle that traversed it .
clearly @xmath128 connects all vertices to the root @xmath0 .
observe that only vehicles having speed at least @xmath129 can even reach any vertex in @xmath130 ( since a vehicle of speed @xmath27 travels distance at most @xmath131 ) .
thus every edge in @xmath132 must be labeled by some vehicle of speed at least @xmath129 .
this implies that @xmath133 , since the right hand side is a bound on the total length traversed by vehicles having speed at least @xmath129 . on the other hand ,
since @xmath128 connects all vertices , @xmath132 contains a spanning tree of @xmath134 .
thus we have @xmath135 .
we then get the following corollary of theorem [ thm : lprim ] .
[ cor : lprim ] a level - primtree @xmath124 satisfies the following : the vertex - levels along every root - leaf path are non - decreasing . for each @xmath51 , @xmath136 .
in the rest of this section , we prove theorem [ thm : lprim ] .
it is easy to see that for every @xmath137 , @xmath138 spans @xmath139 $ ] , hence the procedure produces a spanning tree for @xmath140 .
moreover , by construction we obtain that every root - leaf path in @xmath141 traverses the levels in non - decreasing order as desired .
thus , we focus on proving the second property in the theorem . instead of comparing the length of the edges in @xmath141 with an @xmath142
, it turns out to be much easier to use a specific lasttree as proxy for the latter .
the following lastis implicit in the construction given in @xcite ; for completeness we outline a proof in appendix [ app : spider ] . recall that a _ spider _ is a tree with at most one vertex ( the center ) having degree greater than two .
[ thm : last ] given any metric @xmath4 with root @xmath0 , there exists a spanning spider @xmath143 with center @xmath0 such that : for each @xmath8 , the distance from @xmath0 to @xmath63 in @xmath143 is at most @xmath144 .
the length of @xmath143 is at most four times the @xmath19 in @xmath4 , i.e. @xmath145 .
we remark that we can not use a lastdirectly instead of level - primsince the former does not need to have the properties asserted by theorem [ thm : lprim ] ; it is easy to find a lastwhich does not satisfy the first property , while figure [ fig : last ] also shows that the second can also be violated by an arbitrary amount . using these spider
lasts we can obtain the main lemma needed to complete the proof of theorem [ thm : lprim ] . [
lem : lprim - length ] for any graph @xmath140 and any level - primtree @xmath141 on @xmath140 , we have @xmath146 .
consider a spider last@xmath143 for @xmath140 and let @xmath147denote the set of all root - leaf paths in @xmath143 ; note that @xmath148 is edge - disjoint .
consider any root - leaf path @xmath149 in @xmath147 .
we claim that @xmath150 crosses levels almost in an increasing order .
specifically , there does not exist a pair of nodes @xmath151 with @xmath152 , @xmath153 and @xmath154 .
suppose ( for a contradiction ) that this were the case ; then we would have that @xmath155 where the last inequality uses @xmath156 . on the other hand , @xmath157 since @xmath154
; so we obtain @xmath158 , which contradicts the definition of @xmath143 ( see theorem [ thm : last ] ) .
now we transform @xmath143 into another spider @xmath159 which traverses levels in non - decreasing order as follows . for each root - leaf path @xmath160 , perform the following modification .
let @xmath161 be the subsequence of @xmath150 consisting of the vertices in _ even numbered _ levels , i.e. each @xmath162 for some @xmath163 .
similarly , let @xmath164 be the subsequence of @xmath150 consisting of the vertices in _ odd numbered _ levels . define two paths @xmath165 ( shortcutting @xmath150 over nodes @xmath166 s ) and @xmath167 ( shortcutting @xmath150 over @xmath168 s ) .
observe that both @xmath169 and @xmath170 cross levels _ monotonically _ : if not then there must be some @xmath152 in @xmath150 with @xmath153 and @xmath171 , contrary to the previous claim .
also , by employing the triangle inequality we have that @xmath172 .
finally define the spider @xmath159 as the union of the paths @xmath173 over all root - leaf paths @xmath174 . by construction ,
the vertex levels along each root - leaf path of @xmath159 are non - decreasing .
additionally @xmath175 , by theorem [ thm : last ] .
now partition the edges of @xmath159 as : @xmath176 & \mbox { if } \ell=0,\\ l'[v_{\le \ell}]\setminus l'[v_{\le \ell-1 } ] & \mbox { if } \ell\ge 1 . \end{array } \right.\ ] ] by the monotone property of paths in @xmath159 , it follows that @xmath177 $ ] is connected for every @xmath126 .
thus @xmath178 is a _ spanning tree _ in graph @xmath139/v_{<\ell}$ ] . since @xmath179 in the level - primconstruction ,
is chosen to be an @xmath19 in @xmath139/v_{<\ell}$ ] , we obtain @xmath180 .
so , @xmath181 . this completes the proof of the lemma . [ [ completing - proof - of - theoremthmlprim . ] ] completing proof of theorem [ thm : lprim ] .
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we now prove the second property in theorem [ thm : lprim ] .
lemma [ lem : lprim - length ] directly implies this property for @xmath182 .
for any level @xmath183 consider the graph @xmath184 ; observe that @xmath185 is a level - primtree for @xmath186 ( due to the iterative construction of @xmath187 ) .
thus applying lemma [ lem : lprim - length ] to graph @xmath186 and its level - prim@xmath185 , we have @xmath188 . in this section
we decompose a level - primtree into an assignable collection @xmath118 of @xmath0-rooted trees . motivated by corollary [ cor : lprim ] ,
the idea is to essentially break each subgraph @xmath121 into many pieces and connect them to @xmath0 in order to form the set of trees @xmath84 .
more specifically , assume for now that each connected component in @xmath121 is large enough , i.e. has length at least @xmath33 .
then for each @xmath120 , break the connected components of @xmath121 into trees of length approximately @xmath33 ; add to each tree the shortest edge connecting them to @xmath0 and set @xmath84 as the collection of @xmath0-rooted trees obtained . by construction
we get that @xmath118 satisfies the first property of an assignable collection .
moreover , notice that each edge added to connect a tree to the root has approximately the same length as the tree itself ; this guarantees that @xmath189 .
it then follows that the collection @xmath118 is assignable .
notice that it was crucial to break @xmath121 into trees of length at least approximately @xmath34 .
but this is problematic when @xmath121 has a small connected component . in this case
we show that such a small component is always attached to ( `` dangling '' from ) a large enough component in @xmath190 ( otherwise the dangling edge to a much earlier level will already make this component heavy enough not to be small ) ; then we simply treat the small component as an integral part of the latter .
now we formally describe the proposed decomposition procedure
. * step 1 .
* let @xmath191 contain the subtree @xmath192 .
for each level @xmath193 : partition edges @xmath194 into a collection @xmath195 of ( unrooted ) subtrees such that each subtree contains exactly one edge from @xmath196 to @xmath68 . for any @xmath197 call the unique edge from @xmath196 to @xmath68 its _ head - edge _ @xmath198 .
note that such a partition is indeed possible since @xmath199/v_{<i}$ ] is connected .
any subtree in @xmath195 ( for @xmath51 ) is referred to as a level @xmath32 subtree .
note that head - edges are defined only for subtrees in level @xmath200 and above
. * step 2 .
* for each level @xmath51 : _ mark _ those @xmath197 that have @xmath201 .
in addition , _ mark _ the tree @xmath202 in @xmath191 .
let @xmath203 and @xmath204 denote the marked and unmarked subtrees in @xmath195 .
* for each level @xmath193 and unmarked @xmath205 : define @xmath206 as the subtree in @xmath207 containing the other end - point of @xmath208 .
[ cl : decomp1 ] for @xmath193 and unmarked @xmath205 , @xmath209 .
moreover , @xmath206 is marked . since @xmath210 is unmarked in level @xmath193 , @xmath211 .
so the end - point @xmath64 of @xmath208 in @xmath206 satisfies @xmath212 , otherwise @xmath213 .
in particular @xmath214 and thus @xmath215 . for
the second part of the claim , notice that if @xmath216 then @xmath217 , which is always marked .
so suppose @xmath218 . from the above
, @xmath206 is in level @xmath219 and hence contains a head - edge .
this implies that @xmath206 contains some vertex @xmath220 , namely an end - point of @xmath221 .
but then @xmath222 , where we used @xmath223 since @xmath220 and @xmath224 from above . thus @xmath206 must be marked .
* for each level @xmath51 and marked @xmath225 : define @xmath226 as the set of all unmarked @xmath227 having @xmath228 .
clearly @xmath229 for all @xmath230 .
* for each level @xmath51 and marked @xmath225 : partition the tree @xmath231 into subtrees @xmath232 such that the first @xmath233 trees have length in the range @xmath234 $ ] and @xmath235 has length at most @xmath236 . notice that this is possible since all edges of @xmath231 belong to @xmath237 and hence have length at most @xmath238 .
finally , add the shortest edge from @xmath0 to each of these new subtrees to obtain a collection @xmath239 of @xmath0-rooted trees .
[ cl : decomp2 ] for any @xmath240 , we have @xmath241 .
notice that every @xmath242 consists of a @xmath243 ( for some @xmath244 ) and an edge from @xmath0 to a node in @xmath245 .
since the former has length at most @xmath246 and the latter has length at most @xmath247 , it follows that @xmath248 .
@xmath249 $ ] .
[ cl : decomp3 ] we break the analysis into two cases depending of @xmath250 .
suppose @xmath251 , namely @xmath239 consists of a single tree @xmath99 . in this case
@xmath252 , where @xmath77 is an edge to @xmath0 .
if @xmath253 then @xmath254 and the result holds directly . if @xmath183 then @xmath255 has a node in @xmath256 and hence @xmath257 . because @xmath255 is marked and different than @xmath202 , the lower bound on its length implies that @xmath258 .
the result follows by adding the length of @xmath231 to both sides .
now suppose @xmath259 .
since all trees in @xmath239 lie in @xmath260 , each edge from the root in @xmath239 has length at most @xmath261 .
so the left hand side is at most @xmath262 .
but for @xmath263 we have @xmath264 , so the last term of the previous expression can be upper bounded by @xmath265 .
this bound is smallest when @xmath266 , which then gives @xmath267 .
this concludes the proof of the claim .
* for each level @xmath51 : define @xmath268 .
the following lemma summarizes the main property of our decomposition procedure .
[ lem : isassignable ] the collection @xmath269 obtained from the above procedure is @xmath270-assignable . by claim [ cl : decomp2 ] , each tree in @xmath84 has length at most @xmath271 . so the collection satisfies condition ( 1 ) of definition [ def : assignable ] .
fix any @xmath51 for condition ( 2 ) in definition [ def : assignable ] . due to corollary [ cor : lprim ]
, it suffices to prove that @xmath272 . using claim [ cl : decomp3 ]
we obtain that @xmath273 = 5\cdot d(\subt^m_j ) + 5\cdot d(\subt^u_{j+1}).\ ] ] the last equality above uses the fact that that @xmath274 is a partition of @xmath275 .
thus : @xmath276 this concludes the proof of lemma [ lem : isassignable ] .
[ [ summary - of - the - algorithm . ] ] summary of the algorithm .
+ + + + + + + + + + + + + + + + + + + + + + + + + our algorithm starts with an initial low guess of @xmath30 and runs the level - primprocedure .
if the second condition in corollary [ cor : lprim ] does not hold for this run , we double the guess for @xmath30 and repeat until it is satisfied ( this happens the first time that @xmath30 reaches the condition for the correct guess : @xmath65 ) .
we use the decomposition in this section summarized in lemma [ lem : isassignable ] to obtain a ( 6,40)-assignable collection of trees .
using lemma [ lem : assignable ] on this collection gives us the desired constant approximation ratio by observing that the guess @xmath30 in this step obeys @xmath277 .
the input to the _ heterogenous cvrp _ ( @xmath15 ) consists of a metric @xmath4 denoting distances between vertices , depot @xmath7 ( containing an infinite supply of items ) , demands @xmath278 and @xmath1 vehicles with speeds @xmath2 , each having capacity @xmath11 . a solution to @xmath15consists of tours @xmath279 ( starting and ending at @xmath0 ) for each vehicle so that all demands are satisfied and each vehicle carries at most @xmath11 items at any point in time .
the objective is to minimize the _ maximum completion time _ ,
we study the `` split - delivery '' version of @xmath3here , where demand at a vertex may be served by multiple visits to it ; however , our result easily extends to the `` unsplit - delivery '' @xmath15 .
we show that the @xmath15problem can be reduced to @xmath16 in an approximation preserving way ; so we also obtain an @xmath25-approximation for @xmath15 . the idea in this reduction
is to modify the input metric based on lower - bounds for @xmath3 @xcite . in order to avoid ambiguity
, we use @xmath281 to denote the optimum for @xmath282 and @xmath283 to denote the optimum for @xmath284 .
[ th : vrptotsp ] consider an instance @xmath285of @xmath15 . there is a poly - time constructible instance @xmath286of @xmath16such that @xmath287 .
moreover , a solution to @xmath286of makespan @xmath30 can be converted in poly - time to a solution to @xmath285with makespan @xmath31 .
let @xmath285be an instance of @xmath15as specified above .
standard scaling arguments can be used to ensure that @xmath11 is polynomial in @xmath288 and @xmath289 for all @xmath290 ( details in the full version ) .
let @xmath53 denote the complete graph on vertices @xmath5 with edge - weights equal to distances @xmath54 .
augment @xmath140 to a new graph @xmath57 by adding vertices @xmath291\}$ ] , and edges @xmath292\}$ ] ; each edge @xmath293 has weight @xmath294 . for any vertex @xmath290 , the vertices @xmath295\}$ ]
are referred to as copies of @xmath64 .
let @xmath296 denote the metric induced on vertices @xmath297 where @xmath137 denotes the shortest - path distances in graph @xmath57 .
we let @xmath286be the instance of @xmath16on metric @xmath296 with depot @xmath0 and @xmath1 vehicles having speeds @xmath2 . since @xmath298 this reduction runs in polynomial time . for any graph @xmath143 and subset @xmath299 of vertices ,
let @xmath300 denote the minimum length steiner tree connecting @xmath299 .
for any subset @xmath301 and @xmath290 let @xmath302 denote the number of @xmath64-copies in @xmath99 ; also define @xmath303 .
observe that for any @xmath301 we have @xmath304 by the definition of graph @xmath57 .
we first show that @xmath305 .
consider an optimal solution @xmath279 to @xmath285 .
for each @xmath306 $ ] , let @xmath307 $ ] denote the units of demand at vertex @xmath290 served by vehicle @xmath32 , and let @xmath308 .
note that @xmath309 for all @xmath290 ; hence we can choose @xmath310 for each @xmath311 $ ] such that @xmath312 and @xmath313 for all @xmath314 $ ] . since @xmath315 is a capacitated tour in @xmath140 serving demands @xmath316 , we have @xmath317 using the ( connectivity and capacitated routing ) lower - bounds for @xmath3 @xcite .
thus @xmath318 .
now consider the solution to @xmath286where the @xmath23 vehicle visits vertices @xmath319 along the minimum tsp tour on @xmath320 , for all @xmath311 $ ] ; the distance traversed by the @xmath23 vehicle is at most @xmath321 .
so the @xmath16objective value of this solution is at most @xmath322 } \frac{4\cdot d(\sigma_i)}{\lambda_i}=4\cdot { \ensuremath{{\sf opt_{vrp}}}\xspace}({\ensuremath{\mathcal{i}}\xspace})$ ] .
now consider a solution @xmath47 to @xmath286with makespan @xmath30 .
let @xmath323 denote the vertices that are served by each vehicle @xmath306 $ ] . since @xmath324 is a tsp tour on @xmath325 , we have @xmath326 . now fix @xmath327 $ ] and consider the instance of @xmath3on vertices @xmath328 with demands @xmath329 . as mentioned in the previous paragraph , @xmath330 is a lower - bound for this instance , and the algorithm from @xcite returns a solution @xmath315 within a @xmath331 factor of this lower - bound .
it readily follows that @xmath332 is a feasible solution to @xmath285with makespan at most @xmath322 } \frac{\rho\cdot d(\tau_i)}{\lambda_i}= o(m)$ ] .
we note that this algorithm returns a _
non - preemptive _ @xmath15solution , i.e. , each item once picked up at the depot stays in its vehicle until delivered to its destination .
moreover , the lower - bounds used by the @xmath15algorithm also hold for the ( less restrictive ) _ preemptive _ version , where items might be left temporarily at different vertices while being moved from the depot to their final destination .
thus our algorithm also bounds the `` preemption gap '' ( ratio of optimal non - preemptive to preemptive solutions ) in @xmath15by a constant .
one interesting open question regards the approximability of @xmath284 and @xmath282 when vehicles are located in multiple different depots across the space .
the current definition of an assignable collection and the definition of level - primcrucially depend on the assumption of a unique depot , hence an extension to the multi - depot case is likely to require new ideas .
another interesting direction is to consider @xmath15with non - uniform capacities , where the ideas presented in section [ sec : hvrp ] do not seem to generalize directly .
we will show that the following algorithm produces an @xmath333-spider for @xmath140 .
consider an @xmath19for @xmath140 and traverse it in preoder to obtain a path @xmath334 such that @xmath335 .
[ alg : shortcutif ] @xmath336 to @xmath299 and * mark * @xmath337 .
[ alg : shortcut ] marked node @xmath337 , remove @xmath338 from @xmath299 .
[ alg : delete ] @xmath339 . since the algorithm keeps adding edges to the path @xmath341 , it is clear that before step [ alg : delete ] only the root @xmath342 and marked nodes have degree larger than 2 .
moreover , each marked node has degree exactly 3 .
thus , after step [ alg : delete ] we have that only the root has degree larger then 2 , and the lemma follows .
first we prove that @xmath343 for all @xmath32 .
to see this , consider @xmath299 right before step [ alg : delete ] .
it follows from step [ alg : shortcut ] that @xmath344 for all @xmath32 . noticing that @xmath340 is a shortest path tree of @xmath299 from node @xmath342 implies the desired result .
now we prove that @xmath340 satisfies the second property of @xmath333-spider .
define @xmath345 and let @xmath346 be the @xmath32th node marked by the algorithm .
it is clear that @xmath347 ; so our goal is to upper bound the last summation .
fix a node @xmath346 . consider the beginning of the iteration where @xmath346 is marked .
notice that at this point @xmath348 , since edge @xmath349 was already added to @xmath299 ; since @xmath341 is subgraph of @xmath299 , it is also clear that the right hand side is at most @xmath350 .
however , since @xmath346 was marked , we have that @xmath351 , and then using the previous bounds we obtain that @xmath352 . adding the previous inequality over all @xmath346
s we get that @xmath353 noticing that @xmath354 and reorganizing leads to @xmath355 .
finally , notice that @xmath356 : this follows from traversing the path @xmath341 and using the triangle inequality .
this gives the bound @xmath357 . | the _ capacitated vehicle routing problem _ ( cvrp ) @xcite involves distributing ( identical ) items from a depot to a set of demand locations , using a single capacitated vehicle .
we study a generalization of this problem to the setting of multiple vehicles having non - uniform speeds ( that we call _ heterogenous cvrp _ ) , and present a constant - factor approximation algorithm .
the technical heart of our result lies in achieving a constant approximation to the following tsp variant ( called _ heterogenous tsp _ ) .
given a metric denoting distances between vertices , a depot @xmath0 containing @xmath1 vehicles having respective speeds @xmath2 , the goal is to find a tour for each vehicle ( starting and ending at @xmath0 ) , so that every vertex is covered in some tour and the maximum completion time is minimized .
this problem is precisely heterogenous cvrp when vehicles are uncapacitated .
the presence of non - uniform speeds introduces difficulties for employing standard tour - splitting techniques . in order to get a better understanding of this technique in our context
, we appeal to ideas from the 2-approximation for scheduling in parallel machine of lenstra et al .
@xcite .
this motivates the introduction of a new approximate mst construction called _ level - prim _ , which is related to _
light approximate shortest - path trees _ @xcite .
the last component of our algorithm involves partitioning the level - prim tree and matching the resulting parts to vehicles .
this decomposition is more subtle than usual since now we need to enforce correlation between the size of the parts and their distances to the depot . |
the recent advance in radioactive nuclear beam experiments provides us with novel information on unstable nuclei far away from the stability line ( tanihata et al . 1985 ) .
the symmetry energy becomes more essential to understand the nuclear structure as we go further away from stability .
the symmetry energy is also important to provide the equation of state ( eos ) for neutron stars and supernovae .
the properties of dense matter under neutron rich environment determine the structure and chemical composition of stars and may change their evolution drastically .
therefore , it is very interesting to study the influence of the symmetry energy on astrophysical problems .
the effects of the symmetry energy on the evolution of neutron stars and supernovae have been studied by several authors .
the possibility of the rapid cooling of neutron stars due to the direct urca process has been discussed in the case of the large proton fraction of neutron star matter , which is sensitive to the density dependence of the symmetry energy ( boguta 1981 ; lattimer et al .
1991 ) . bruenn ( 1989 ) studied the effects on supernova explosions systematically by doing numerical simulations of gravitational core collapse with the parameterized eos .
swesty et al .
( 1994 ) also studied the role in the prompt phase of the supernova explosion .
however , there has been no systematic study on the effect of the symmetry energy on the birth of neutron stars and supernova neutrinos as far as we know .
moreover , in most of previous studies , the parameterized formula have been used to provide the eos of dense matter beyond the normal nuclear matter density .
recently , there has been a great progress in the study of nuclei and dense matter within the relativistic many body framework ( serot & walecka 1986 ) .
it was demonstrated that the relativistic brueckner hartree fock ( rbhf ) theory is capable of reproducing the saturation property of nuclear matter starting from the nucleon - nucleon interaction determined by the scattering experiments ( brockmann & machleidt 1990 ) . the relativistic mean field
( rmf ) theory has been shown to be very successful as an effective theory to describe the ground state properties of nuclei in the wide mass range of the periodic table ( gambhir et al .
1990 ) and has been applied to the eos for neutron stars ( serot & walecka 1986 ) .
it is amazing that the rmf theory describes also the properties of unstable nuclei away from stability extremely well ( hirata et al . 1991 ; sugahara & toki 1994 ) . having the framework constrained by unstable nuclei , sumiyoshi and toki applied the same rmf theory to provide the data table of the eos for neutron stars and supernovae ( sumiyoshi & toki 1994 ; sumiyoshi et al . 1995 ) , which enables us to do numerical simulations of the thermal evolution of neutron stars and supernovae quantitatively while taking care of the experimental data of unstable nuclei .
the properties of unstable nuclei have been shown to be very sensitive to the symmetry energy in the rmf theory ( sumiyoshi et al .
hence , it would be nice if we could see how sensitive are the properties of neutron stars to the symmetry energy and what is the influence of the symmetry energy on the birth of neutron stars and supernova neutrinos .
the numerical simulations of the cooling of protoneutron stars and neutrino burst have been done by several groups ( burrows & lattimer 1986 ; burrows 1988 ; suzuki 1993 ) .
the influence on the numerical simulation of the birth of neutron stars due to the difference of eos was studied by adopting the eos tables in the two different many body frameworks ( sumiyoshi et al .
the effects of various thermodynamical properties were pointed out there besides the stiffness of eos , which has been mainly studied so far ( burrows 1988 ) . here , we focus on the influence of the symmetry energy by changing solely the strength of the isovector interaction in the rmf theory in order to see its effect clearly .
the purpose of this paper is to explore the influence of the symmetry energy on the cooling of protoneutron stars just born in supernova explosions .
we construct the tables of eos for supernova simulations in the rmf theory and make comparisons of eos s when we change the symmetry energy drastically .
then , we perform numerical simulations of the birth of neutron stars and supernova neutrinos adopting the two eos s with different symmetry energy .
we investigate the influence of the symmetry energy on the thermal evolution of protoneutron stars and the properties of neutrino burst emitted during the cooling stage .
this paper is arranged as follows . in section [ s :
eos ] , we briefly describe the relativistic eos for neutron stars and supernovae . in section [ s : results ] , after a short introduction on the birth of neutron stars , we present the results of our numerical simulations .
we discuss the effects of the symmetry energy on the deleptonization , the thermal evolution and the supernova neutrinos in the subsections .
we summarize this paper in section [ s : summary ] .
we start with a brief explanation of the tables of eos for numerical simulations .
we calculate all the physical quantities of dense matter within the relativistic mean field theory .
we refer the review article by serot & walecka ( 1986 ) as for the relativistic many body framework for nuclei and dense matter .
all the details on the relativistic eos for neutron stars and supernovae in the rmf theory has been reported in the recent papers .
( sumiyoshi & toki 1994 ; sumiyoshi et al .
1995 ) .
we adopt the phenomenological lagrangian with the non - linear @xmath0 and @xmath1 terms , which is motivated by the recent success of the rbhf theory ( brockmann & machleidt 1990 ) and has been shown to be very successful both for nuclear properties and dense matter ( sugahara & toki 1994 ) . the best parameter set for the lagrangian , named tm1 , was determined by the least square fitting to a set of nuclei including unstable ones , which is important to constrain the isovector interaction in the theory .
it is remarkable that the rmf theory with the parameter set tm1 has been demonstrated to reproduce successfully the properties of unstable nuclei other than the ones used in the fitting .
the properties of nuclear matter in the rmf theory with tm1 thus constrained has been shown to be quite similar to the properties of nuclear matter derived in the rbhf theory ( sugahara & toki 1994 ) . extending the rmf theory to the case at finite temperature , the table of the numerical data of physical quantities under various conditions of chemical composition , temperature and density , which are required for the numerical simulations ,
was constructed for the parameter set tm1 ( sumiyoshi et al .
we use this table of eos as a standard one . in order to explore the influence of the symmetry energy ,
we newly construct the table of eos with a reduced value of the symmetry energy in the rmf theory .
we reduce the value of the coupling constant @xmath2 between isovector - vector @xmath3 meson and nucleon in the lagrangian , which is essential to determine the symmetry energy , while keeping other parameters of tm1 unchanged .
hereafter , we call the modified parameter set as tms . the symmetry energy is @xmath4mev for tm1 and @xmath5mev for tms : the corresponding coupling constant is @xmath6 for tm1 and @xmath7 for tms .
we note that the symmetry energy in tm1 has been checked by unstable nuclei and this modification in tms is to explore the effect of the symmetry energy on astrophysical applications .
we display in fig .
[ f : epb ] the energy per baryon of symmetric nuclear matter and pure neutron matter for the cases of tm1 and tms .
the incompressibility k at the normal nuclear matter density is 281mev for both cases , since the isovector meson does not contribute to symmetric nuclear matter .
we comment here on the density dependence of the symmetry energy .
the symmetry energy for both cases of tm1 and tms has a monotonically increasing feature with density , which is common for all the relativistic many body calculations due to the contribution of the isovector meson .
this feature is demonstrated by the microscopic calculation with the use of the rbhf theory ( li et al .
the reduction of the symmetry energy at the saturation density by changing the strength of the isovector interaction clearly corresponds to the reduction of the symmetry energy at high density .
in contrast , the symmetry energy in the non - relativistic many body calculations such as the one by wiringa et al .
( 1988 ) has generally a weak density dependence and has sometimes a decreasing feature at high density .
a variety of the density dependence depends on the choice of the density dependent potential , which is introduced to reproduce the saturation of nuclear matter .
therefore , the relation between the symmetry energy at the saturation density and its behavior at high density is ambiguous in non - relativistic many body calculations .
when we apply the eos s in the two cases to neutron stars , we found that the chemical compositions are very different while the hydrostatic structures are quite similar . in fig .
[ f : eos]-a , we display the proton fraction of neutron star matter , which is the ratio between the proton density and the baryon density , as a function of the baryon mass density . the proton fraction in the case of tms is smaller than the case of tm1 because of the reduced symmetry energy .
we show in fig .
[ f : eos]-b the neutron star mass as a function of the central baryon mass density .
in contrast to the difference in the chemical composition , the neutron star masses in the two cases are found very similar .
the maximum mass turns out to be almost the same value of 2.2m_. the difference in the central baryon mass density is only about 1% for the case of neutron stars with the gravitational mass of 1.4m_.
first of all , we describe briefly the supernova explosions and the birth of neutron stars ( suzuki 1994 ) .
gravitational collapse of the core of a massive star leads to explosion of the envelope ( supernova explosion ) and formation of the neutron star . we shall divide the series of stages starting from the onset of the core collapse to the birth of neutron star into two phases .
the first phase is the dynamical phase .
the core of a massive star becomes unstable when it grows to the point near the chandrasekhar mass ( @xmath8 1.4 m _ ) , and it begins to collapse .
the collapse never ceases until the central density exceeds the nuclear density .
sudden stiffening of the eos above the nuclear density stops the inner core ( 0.5 0.8 m _ ) and the bounce of the inner core launches a shock wave into the falling outer core .
the matter of the falling outer core is swept and decelerated by the shock wave and , then , accretes onto the unshocked inner core which has been in hydrostatic equilibrium in its dynamical time scale of milliseconds .
the shock wave expels the envelope and we identify it as supernova explosion .
the remnant at the center which consists of the unshocked inner core and the shocked outer core is called as a protoneutron star .
it contains so many leptons and protons ( @xmath8 30% ) that we can not call it a neutron star at this stage .
it takes only about 1 second for this dynamical process to take place . the second phase , which is the quasistatic phase of protoneutron star cooling , follows this dynamical phase .
after the shock wave breaks out of the core surface and the accretion onto the inner core ceases , the protoneutron star evolves quasistatically keeping hydrostatic configuration .
neutrinos which are trapped in the core diffuse out of the protoneutron star in a time scale of neutrino diffusion , which is of order of 10 seconds .
they drive the evolution of the protoneutron star into the cold neutron star by carrying out thermal energy and lepton number from the protoneutron star . concerning supernova neutrinos , about a half of the total energy is emitted during the dynamical phase and the rest is emitted during the quasistatic phase .
in this paper , we study the influence of the nuclear symmetry energy on the second stage : the evolution of the hot protoneutron star into the normal neutron star .
the quasistatic cooling of the protoneutron star is simulated numerically by solving the general relativistic equations for hydrostatic structure ( oppenheimer - volkoff equation ) and for neutrino transport with the deleptonization and the entropy change of the matter simultaneously using the henyey - type method . as for the neutrino transfer
, we adopt the multigroup flux limited diffusion scheme ( bruenn 1985 ) .
we take into account the energy dependence of neutrino transport coefficients in the multigroup scheme .
the flux limiter should be introduced in order to express the neutrino flux in the transparent regime in terms of the diffusion flux .
we adopt mayle and wilson s flux limiter ( mayle et al .
1987 ) in this work .
we include the general relativistic effects such as the time dilation and the red shift of the neutrino energy .
we treat explicitly , _
e , |_e and @xmath9 , where @xmath9 represents the average of _ , | _ , _ and |_. this is a good approximation in the case where we can neglect the existence of @xmath10 and @xmath11 leptons because of the low temperature ( @xmath12100mev ) .
the following neutrino interactions are included as opacity sources or collision terms in the neutrino transfer equations .
[ cols= " < , < , < , < , < , < " , ] where @xmath13 represents all species of neutrinos , a is a representative heavy nucleus , and n is either a proton or a neutron . most of the interaction rates are taken from bruenn ( 1985 ) with some modifications . at present , many body effects on neutrino opacity are not included except for the multiple scattering suppression effects on nucleon bremsstrahlung process ( raffelt & seckel 1991 ) .
the evolution of the protoneutron star is driven by the exchange of energy and lepton number between the matter and neutrinos due to the above interactions and the neutrino transport .
descriptions of our numerical code are also given in suzuki ( 1993 ) .
we perform numerical simulations of the protoneutron star evolution using the tables of eos in the rmf theory with tm1 and tms .
as for the eos of the low density matter ( @xmath14 ) , wolff s eos ( hillebrandt & wolff 1985 ) is used for both cases .
we construct the initial models for our numerical simulation by referring to the numerical results of mayle and wilson at 0.4sec after the core bounce in the hydrodynamical simulations of the supernova explosion ( mayle & wilson 1989 ; wilson 1990 ) .
initial models for the simulations using the two eos s are constructed using the same entropy profile and the same electron fraction profile in order to study the effects of the difference in the symmetry energy . of course , in principle , these profiles should differ from each other because the difference of eos should also affect on the dynamical phase . to extract the direct influence of the nuclear symmetry energy on the protoneutron star evolution , we neglect its influence on the dynamical phase .
the calculated initial hydrostatic structures of the two protoneutron stars are found similar ; the differences in the initial densities and the temperature profiles are smaller than 3% .
on the contrary , there is a large difference in @xmath15 which is the chemical potential of _
e in the @xmath16-equilibrium with the matter because @xmath17 and @xmath18 are directly affected by the symmetry energy . in figs .
[ f : entropy][f : mu ] , we present the profiles of the two initial models .
@xmath19 at the center of the initial models are 142mev ( @xmath20 ) for tm1 and 170mev for tms ( @xmath21 ) , respectively .
the resultant lepton fractions , @xmath22 are 0.330 for tms and 0.315 for tm1 .
starting from these initial models we simulate the evolution of the protoneutron stars for the following 15 seconds with the same numerical code . figs .
[ f : entropy][f : mu ] also show the profiles of the protoneutron stars at the end of the calculation ( @xmath23 ) for the two models .
while the density profiles are still nearly identical , the distributions of the temperature and the electron fraction differ largely from each other . especially , the lepton fraction at the center for tm1 ( 0.179 ) is larger than that for tms ( 0.155 ) , while the former is smaller than the latter at the initial stage .
this means that the deleptonization proceeds faster for tms than for tm1 .
we find that this is caused by the difference of @xmath19 . in the central region of protoneutron stars ,
electron type neutrinos are degenerate and their diffusion fluxes are roughly proportional to @xmath24 , where @xmath25 is the mean free path of _ e and @xmath26 is the number density of _
e. @xmath25 is roughly inversely proportional to the neutrino energy squared , @xmath27 , and @xmath28 in the degenerate limit .
consequently , the diffusion flux in the degenerate limit is proportional to @xmath29 ; the smaller symmetry energy results in the larger @xmath19 , the larger _ e flux , and therefore the faster deleptonization .
the influence of the symmetry energy on the evolution of the temperature is complicated in our models .
the central temperature at @xmath30 are 23.2mev for tm1 and 26.3mev for tms .
higher temperature for tms is due to the higher total ( matter + neutrinos ) entropy per baryon ( 1.29 for tms and 1.20 for tm1 ) and the lower lepton fraction ( 0.155 for tms and 0.179 for tm1 ) .
the lower lepton fraction which is the consequence of fast deleptonization means the smaller lepton number density and the temperature corresponding to a given total entropy becomes higher .
as for the difference in the central entropy , we analyze the results of numerical simulations in detail and find the following reason for the present case . in the first 10 seconds of the protoneutron star cooling , the matter entropy of the central region increases .
this entropy change has three main origins ; inward flux of @xmath9 , the downscattering of _
e and the emission of _ e due to electron capture . since , during the first 10 seconds , the temperature profile has its peak in the middle region of the protoneutron star , not at the center , there is the negative gradient in the number density of @xmath9 at the central region .
the flux of @xmath9 is proportional to the gradient .
therefore , at the early stage , @xmath9 flow inwards in the central region and transport the heat into the central region from the middle region .
the pair annihilations of @xmath9 ( @xmath31 ) increase the matter entropy .
in addition , in the central region where the electrons are strongly degenerate , _ e - electron scattering ( @xmath32 ) leads to the increase of the matter entropy .
neutrinos lose their energy at the scattering ( downscattering ) because the scattered electrons which were within the fermi sea should have energy greater than the fermi energy . on the other hand , emission of _
e ( @xmath33 ) decreases the matter entropy .
the two heating processes and the one cooling process result in the net heating of the central region .
detailed analysis of the numerical simulations reveals that , among the above three processes which alter the central entropy , the largest difference of heating / cooling rate ( @xmath34 ) due to each process between the two models is the difference of the cooling rate due to the _ e emission .
furthermore it is found that the difference in @xmath34 due to the _
e emission is caused mainly by the difference of @xmath19 as the case of the deleptonization rate .
@xmath35 due to @xmath36 can be expressed as @xmath37 where @xmath38 is the energy density of _
e. since , in the central region at the early stage where the electron capture proceeds , @xmath39 is positive , it can be seen that the larger @xmath19 for the smaller symmetry energy results in the larger @xmath35 due to _ eemission , that is , the less entropy loss of the matter because @xmath40 is negative there .
this small cooling rate due to the _
e emission leads to the higher entropy and higher temperature for the case of smaller symmetry energy .
we note , however , that the temperature profile is affected by many factors .
the difference in the evolutions of the protoneutron star due to the symmetry energy is reflected with the properties of supernova neutrinos emitted during the cooling stage of the star .
we show in fig .
[ f : lepton : flux ] the net flux of the electron type lepton number from the protoneutron star ; the flux of _ e minus |_e .
the net flux is larger for tms than for tm1 during 15 seconds because of the difference in the diffusion fluxes as we have discussed in section [ s : deleptonization ] .
this fact corresponds to the higher deleptonization rate in tms having the smaller symmetry energy .
since the final electron fraction of the neutron star at the end of the deleptonization is smaller for tms , the total net neutrino number emitted during the cooling is larger for tms than for tm1 . it may be possible to extract the information of the final electron fraction of the neutron star from this quantity
we show in fig .
[ f : mean : energy ] the calculated time profile of the mean energy of |_e .
the mean energy for tms becomes higher than that for tm1 at a later stage .
this is due to the difference of the temperature profile of the protoneutron stars , as shown in fig .
[ f : temperature ] .
the mean energy of neutrinos depends mainly on the temperature at the neutrinosphere .
the difference of the mean energies is larger for @xmath9 and smaller for _ e depending on the position of its neutrinosphere .
the neutrinosphere for _ ewhich interact most strongly with the matter locates in the outermost region , where both the temperature and the density are low and the difference between the temperature in the two cases is small .
we show in fig .
[ f : luminosity ] the luminosity of |_e , which is higher for tms than for tm1 .
the general feature is quite similar for the other types of neutrinos .
these results correspond to the larger flux and the higher mean energy of neutrinos .
the total energy carried out by neutrinos is larger for tms than for tm1 , since the gravitational mass of the neutron star at zero temperature is smaller for tms .
we note that the total baryon mass of the protoneutron stars is fixed to be 1.62 m _ ( the gravitational mass of the initial protoneutron star is 1.5479 m _ for tms and 1.5482 m _ for tm1 ) in the present study and the gravitational mass of the cold neutron star turns out to be 1.475 m _ for tms and 1.485 m _ for tm1 .
therefore , the total energy released from the protoneutron star amounts to 0.073 m _ ( @xmath41 ) for tms and 0.063m_(@xmath42 ) for tm1 , which means the larger luminosity for tms , provided that the time duration of neutrino emission is similar .
the main influence of the symmetry energy of the eos on the birth of neutron stars comes from the change of the chemical composition rather than the stiffness .
a change in the chemical potential for neutrinos affects the rates of the interaction with the matter even if the matter density is unchanged .
it affects the deleptonization rate through the change of the diffusion flux , and the temperature inside stars through the change of the heating rate .
those changes influence the properties of the neutrino burst such as the time profile of the mean energy while the density profile remains unchanged .
the difference in the properties of the neutrino burst is emphasized moreover by the difference in the proton fraction and the binding energy of the cold neutron stars at the end of the birth stage .
the above results come from our study on the evolution of protoneutron stars by assuming the same initial conditions following the result of mayle & wilson . here
we comment on expected difference in the initial protoneutron star configuration when we simulate also the dynamical phase , the birth of the protoneutron stars , using the two eos s .
since tms has a smaller coupling constant @xmath2 , it is expected that the symmetry energy and hence the difference between the chemical potential of neutrons and that of protons are smaller for tms than for tm1 even at the density of the collapsing core .
this means a larger fraction of free protons for tms , which will result in a larger electron - capture rate during the collapse and a smaller trapped lepton fraction at the core bounce ( bruenn 1989 ) . throughout the birth stage of neutron stars including both the dynamical phase and the quasistatic phase
, deleptonization will proceed faster for tms than for tm1 and electron - type lepton number flux will be larger for tms . the work in this direction to simulate
also the dynamical phase is in progress .
further studies are required to predict precisely the profile of the neutrino burst and to extract the information on the symmetry energy from the observational signals of the next supernova explosion in our galaxy or close - by .
the influence of the thermodynamical properties of eos such as the symmetry energy should be carefully examined besides the one by the stiffness , which has been mainly focused .
the symmetry energy is one of the most interesting keys to clarify the current problems in nuclear physics and astrophysics
. it would be interesting to apply the many body framework such as the rmf theory , in which the symmetry energy is being checked by the properties of unstable nuclei , to solve consistently the current issues on the cooling of the neutron stars and the supernova explosion together with the ones in nuclear physics .
we would like to thank k. oyamatsu , s. nishizaki , t. takatsuka and k. sato for comments and discussions , and r. mayle , j. r. wilson and r. g. wolff for providing us with their data .
we are grateful to y. sugahara , d. hirata and i. tanihata for providing us with valuable information and for fruitful discussions on the aspect of unstable nuclei .
k. s. acknowledges the support by the special researchers basic science program .
boguta j. , 1981 , phys .
b106 , 255 brockmann r. , machleidt r. , 1990 , phys .
c42 , 1965 bruenn s. w. , 1985 , apjs 58 , 771 bruenn s. w. , 1989 , apj 340 , 955 burrows a. , lattimer j. m. , 1986 , apj 307 , 178 burrows a. , 1988 , apj 334 , 891 gambhir y. k. , ring p. , thimet a. , 1990 , ann . of phys .
198 , 132 hillebrandt w. , wolff r. g. , 1985 , in : arnett w. d. , truran j. w. , ( eds . ) nucleosynthesis : challenges and new developments , the university of chicago press , chicago , p.131 hirata d. , toki h. , watabe t. , tanihata i. , carlson b. v. , 1991 , phys .
c44 , 1467 lattimer j. m. , pethick c. j. , prakash m. , haensel p. , 1991
66 , 2701 li g. q. , machleidt r. , brockmann r. , 1992 , phys .
c45 , 2782 mayle r. , wilson j. r. , 1989 , private communication mayle r. , wilson j. r. , schramm d. n. , 1987 , apj 318 , 288 raffelt g. , seckel d. , 1991 , phys .
67 , 2605 serot b. d. , walecka j. d. , 1986 , adv .
16 , 1 sugahara y. , toki h. , 1994 , nucl .
a579 , 557 sumiyoshi k. , hirata d. , toki h. , sagawa h. , 1993a , nucl .
a552 , 437 sumiyoshi k. , suzuki h. , toki h. , 1993b , in : morrissey d. j. ( ed . ) proceedings of the third international conference on radioactive nuclear beams , michigan , usa , 1993 , editions frontieres , gif - sur - yvette , p.501 sumiyoshi k. , toki h. , 1994 , apj 422 , 700 sumiyoshi k. , kuwabara h. , toki h. , 1995 , nucl .
a581 , 725 suzuki h. , 1993 , in : suzuki y. , nakamura k. ( eds . ) proceedings of the international symposium on frontiers of neutrino astrophysics , takayama , japan , 1992 , universal academy press inc . , tokyo , p.219 suzuki h. , 1994 , in : fukugita m. , suzuki a. ( eds . )
physics and astrophysics of neutrinos , springer - verlag , tokyo , p.763 , and references therein swesty f. d. , lattimer j. m. , myra e. s. , 1994 , apj 425 , 195 tanihata i. , hamagaki h. , hashimoto o. , et al . , 1985 ,
55 , 2676 wilson j. , 1990 , nucl .
b.(proc . suppl . ) 13 , 380 wiringa r. b. , fiks v. , fabrocini a. , 1988 , phys . | we study the influence of the symmetry energy of the equation of state on the thermal evolution of protoneutron stars and the properties of supernova neutrinos by the numerical simulations after the protoneutron stars are formed .
as for the equation of state ( eos ) of nuclear matter , we take two eos s with different symmetry energies obtained by the relativistic mean field theory .
we find the symmetry energy plays the essential role on the evolution of lepton profiles and the neutrino fluxes .
@referee = 1.0 in |
ultra cold electrons in two dimensions in a strong perpendicular magnetic field , the quantum hall ( qh ) system @xcite , exhibits a fascinating phase diagram including phases with fractionalized excitations and topological order @xcite . ever since its discovery three decades ago , the qh system has inspired a huge amount of experimental and theoretical effort , not least due to its richness in phenomenology and mathematical structure .
new developments include the observation of the fractional quantum hall effect in graphene @xcite and ideas of applications in the context of topological quantum computing @xcite .
moreover , it has been realized that the theoretical description of a system of rapidly rotating bosons is formally very similar to that of an electron gas in a magnetic field @xcite .
a key property of fractional quantum hall ( fqh ) states is their topological order @xcite .
one consequence thereof is that their physical properties are insensitive to smooth deformations of the manifold on which we choose to study them .
this fact has been exploited in a series of recent studies of the interacting many - body problem in the limit geometry of a thin torus , referred to as the tao - thouless ( tt ) limit @xcite ( see also refs .
@xcite for precursory studies and ref .
@xcite for some recent related approaches ) . in this proceeding
we study the fqh system beyond the tt limit and argue that some characteristics that are not readily understood in the tt limit states are however manifest in the leading quantum fluctuations away from this limit .
while several of the ideas presented here have been published earlier @xcite , we corroborate these results and also present a number of new results .
we study the structure of excitation spectra of two - dimensional electrons in a magnetic field with coulomb interactions as a function of the torus circumference , @xmath8 , and contrast the cases of even and odd denominator filling fractions .
for @xmath6 we interpret these results in terms of a deforming fermi sea @xcite .
then , by extending the results of refs .
@xcite for the @xmath9 fqh state , we further establish that fqh systems in a jain sequence@xcite , which has filling factor @xmath0 in the lowest landau level can be described by @xmath2 spin chains with @xmath3 sites unit cell .
the remainder of this proceeding is organized as follows . in section [ model ]
we review the lattice describing interacting particles in a landau level on the torus and present numerical spectra thereof as a function of the circumference , @xmath8 . in section [ beyond ] , we briefly motivate and restate known results about the energy spectra in the tt limit and then introduce spin models that provide an intuitive understanding of the numerical data away from this limit .
in particular , these considerations shed light on why odd denominator fractions remain gapped while even denominator states tend to be gapless .
we consider a model of @xmath10 interacting electrons on a torus pierced by @xmath11 flux quanta . in the landau gauge , a complete basis of @xmath11
degenerate single - particle states in the lowest landau level , labeled by @xmath12 , can be chosen as @xmath13 where @xmath14 are the circumferences of the torus , @xmath15 the corresponding coordinates , and @xmath16 the momentum along the @xmath8-cycle .
we have set the magnetic length @xmath17 equal to unity . in this basis
, any translation - invariant two - dimensional two - body interaction hamiltonian assumes the following model on a one - dimensional discrete lattice , @xmath18 where the matrix - element @xmath19 specifies the amplitude for a process where particles with separation @xmath20 hop @xmath21 steps to a separation @xmath22 ( note that @xmath21 can be negative ) .
the number of the lattice sites is fixed by the area , @xmath23 . at the filling @xmath24 the hamiltonian commutes with the center - of - mass magnetic translations @xcite @xmath25 and @xmath26 along the cycles , which implies , in particular , that the total momentum @xmath27 along the @xmath8-cycle is conserved modulo @xmath11 in this gauge . in this system ( [ 1d_model ] ) , two conservation numbers are given as @xmath28 where @xmath29 and @xmath30 . we have calculated energy spectra of this system for @xmath31 using exact diagonalization in subspaces labeled by @xmath32 .
using these symmetries , the hilbert space splits into @xmath33 sectors that are all of comparable size .
this enable us to treat larger systems .
the matrix elements of @xmath19 are calculated assuming coulomb interaction between the electrons .
for the half - filled landau level @xmath6 , we have calculated the energies up to @xmath34 which is larger than the former analysis @xcite ( see fig . [ spectra12 ] ) .
increasing @xmath8 from the tt limit ( @xmath35 ) , the first level - crossing in the ground state energy takes place at @xmath36 which corresponds to a phase transition from a charge - density - wave ( cdw ) to a gapless phase .
further level crossings occur among different gapless states , and finally @xmath8 arrives at the duality point @xmath37 ( @xmath38 for @xmath39 , @xmath40 for @xmath34 and @xmath41 for @xmath42 ) . the spectrum is invariant under @xmath43 ( although quantum numbers change ) and hence we only present data for @xmath44 .
we have calculated the energies of the @xmath9 and @xmath45 cases , and confirmed that energy gaps do not close as @xmath8 changes , and level - crossings occur only in excited states .
this is consistent with the fact that these states have energy gaps in the bulk systems as is evident in the thin - torus limit ( see below ) . ) for ( a ) @xmath9 with @xmath46 and ( b ) @xmath45 with @xmath39 .
the energy gaps never close as functions of the circumference @xmath8 of the torus.,title="fig:",width=302 ] ) for ( a ) @xmath9 with @xmath46 and ( b ) @xmath45 with @xmath39 .
the energy gaps never close as functions of the circumference @xmath8 of the torus.,title="fig:",width=264 ] ) for ( a ) @xmath9 with @xmath46 and ( b ) @xmath45 with @xmath39 .
the energy gaps never close as functions of the circumference @xmath8 of the torus.,title="fig:",width=283 ] ) for ( a ) @xmath9 with @xmath46 and ( b ) @xmath45 with @xmath39 .
the energy gaps never close as functions of the circumference @xmath8 of the torus.,title="fig:",width=283 ]
from the shape of the single particle states ( [ psik ] ) it follows that the distance between adjacent states , the lattice constant @xmath47 , depends on @xmath8 while the extent of the wave functions is constant . for small
@xmath8 the overlap between different single particle states decreases rapidly and the matrix elements @xmath19 simplify considerably . as @xmath5 one finds that @xmath48 thus the @xmath49 terms are exponentially suppressed for generic interactions in this limit .
the remaining ( @xmath50 ) problem is exactly solvable : ground states at any @xmath24 are gapped periodic crystals ( with a unit cell of @xmath3 electrons on @xmath51 sites ) and the fractionally charged excitations appear as domain walls between degenerate ground states . the fractal structure ( a.k.a .
devils staircase ) of the abelian haldane - halperin hierarchy @xcite is manifest for generic convex two - body interactions and provides an explicit estimate for the relative stability of odd denominator states in excellent agreement with experiment @xcite .
the leading fluctuations are the @xmath52 terms describing correlated hopping of two particles one site each in opposite directions . in a series of earlier investigations ,
some of us have shown that these fluctuations ultimately lead to a melting of the gapped state for @xmath6 resulting in a gapless state @xcite , while for @xmath9 the excitation structure and gap is essentially unaltered , even for very strong fluctuations @xcite .
below , we discuss how the numerical results at finite @xmath8 can be interpreted in terms of effective spin chains and extend this picture to include the most prominent ( @xmath52 ) jain series , @xmath0 and @xmath53 .
the strategy is as follows .
first , we truncate the interaction to only include the leading @xmath52 term , @xmath54 .
this is especially sensible for a pseudo potential interaction where @xmath55 with @xmath56 being a polynomial in @xmath57 and @xmath21 . in this case
our expansion in @xmath4 is well controlled , and we expect it to capture the physics also for more general interactions .
second , we restrict the local hilbert space by including only local configurations that correspond to the tt ground state and local applications of @xmath54 .
this naturally leads to a spin chain representation of the hamiltonian and enables us to deform it to connect to well studied spin chain models and obtain physical insights by drawing analogies between the physics of spin chains and fqh states .
the electron system on a torus can be described in terms of @xmath7 variables in the following way @xcite : in the vicinity of the tt limit , the relevant physics is captured by a model containing the three dominant terms as @xmath58 , \label{truncated_ham}\ ] ] where @xmath59 . in the tt limit , the ground state is the charge - density - wave ( cdw ) state where electrons are located on every two sites
@xmath60 , since the matrix elements of the electrostatic term @xmath61 is dominant .
away from the tt limit , the competition between @xmath61 , @xmath62 and @xmath63 can be included as an interaction of the local spin states @xmath64 and @xmath65 which means that @xmath66 , @xmath67 , @xmath68 , @xmath69 , where @xmath70 is the index of the unit cell .
the effective spin hamiltonian is then given by the @xmath7 xxz chain , @xmath71 where @xmath72 .
it is well known that this model has a phase transition from a ferromagnetic ( cdw ) phase to a gapless ( tomonaga - luttinger liquid ) phase at @xmath73 .
this describes the phase transition from the tt state to the gapless phase for the half - filled landau level @xmath6 .
the obtained solution corresponds to the ( first ) elongated fermi seas in fig .
[ spectra12 ] appearing for @xmath74 .
next we consider spin - mapping for @xmath9 state . in the tt limit
, the ground state of the system is the three - fold degenerate cdw state @xmath75 where the underlines denote unit cells . to capture the effect of the leading quantum fluctuations , we introduce @xmath2 spin variables as @xmath76 , where the operators are related as @xmath77 in this mapping ,
the three - fold degeneracy of the ground state is hidden .
the degenerate states have different center - of - mass quantum numbers and there are no nonvanishing matrix elements between states in different momentum sectors .
it follows that there are three distinct sectors which are described by the same effective hamiltonian , @xmath78[1-(s^z_{n+1})^2]+\mbox{h.c . }
\label{s=1_chain_for_nu=1/3}\ ] ] this model is given by four spin interactions or the xy model with projectors to the state @xmath79 .
note that this hamiltonian does not have the space inversion and spin reversal symmetries : the exchange process @xmath80 exists , but @xmath81 does not . the spin rotational symmetry is also broken .
actually , numerical analysis shows that this hamiltonian well describes the fqh state around @xmath82 , by checking the overlap between the truncated hamiltonian and full hamiltonian with coulomb interaction and the trugmann - kivelson type potential @xcite which provides an exact parent hamiltonian for the laughlin state .
there is also a term given by a function @xmath83 which stems from the electrostatic terms , but this term does not alter any essential features of the present model ( [ s=1_chain_for_nu=1/3 ] ) @xcite . in refs .
@xcite , this model with extra parameters @xmath84 has been analyzed numerically to study adiabatic continuity from a conventional spin chain ( @xmath85 ) . for @xmath85 , it is known that there is a phase transition at @xmath86 between the haldane ( @xmath87 ) and large-@xmath88 ( @xmath89 ) phases @xcite . in refs . @xcite it was shown that the ground state of the model ( [ s=1_chain_for_nu=1/3 ] ) is smoothly connected both to the haldane phase and the large-@xmath88 phase without closing the energy gap under the change of parameters .
this indicates that the spin model describing the @xmath9 fqh state has the nature of these two phases simultaneously , and there are indeed some similarities between the @xmath9 fqh effect and the @xmath2 haldane - gap state as noticed early on @xcite .
the reason for the absence of a phase transition is due to the breaking of the discrete symmetries . according to ref .
@xcite , haldane and large-@xmath88 phases are smoothly connected if the dihedral group , time reversal and parity symmetries are broken . since the present model ( [ s1extended ] ) with @xmath90 breaks these three symmetries , our result is consistent with this general argument .
most properties of a finite chain of this model ( [ s1extended ] ) are similar to those of a usual large-@xmath88 phase where no edge spins appear , however this is not inconsistent with the absence of a phase transition @xcite .
let us now consider extentions of the above @xmath2 mapping to other filling factors .
as discussed by jain @xcite , @xmath91 fqh states may be described by the composite fermion picture , where @xmath92 quantum flux are attached to non - interacting electrons of the @xmath3-th landau level , and projection onto the lowest landau level . for definiteness
, we will consider the positive @xmath52 jain sequence @xmath0 , @xmath93 .
the results are easily transfered to the negative jain series by noting that our formulation is particle - hole symmetric , hence the @xmath0 is equivalent to the @xmath94 case .
this is in contrast to the wave function based approaches and the composite fermion phenomenology , which lack this symmetry . before giving the general result ,
let us consider the @xmath45 state ( @xmath95 ) as the simplest example . in the tt limit
, the ground state is given by the cdw state with the configuration @xmath96 .
when @xmath54 and @xmath97 are applied to this cdw state several times , one finds that the number of electrons in each unit cell is always conserved , and the configuration @xmath98 never appears .
therefore , configurations such as @xmath99 , @xmath100 , @xmath101 , @xmath102 are absent in the truncated hilbert space .
this means that the @xmath45 states can be mapped to two @xmath2 variables by inserting @xmath103 appropriately ( between the two @xmath104 s ) in each unit cell : @xmath10510 \rangle \rightarrow |00 \rangle,\\ & | 00110 \rangle \rightarrow | 00[10]10 \rangle \rightarrow |\mathsf{-}0 \rangle,\\ & | 01100 \rangle \rightarrow | 01[01]00 \rangle \rightarrow |0\mathsf{+ } \rangle.\end{aligned}\ ] ] the relationship between fermion and spin-@xmath104 operators ( @xmath106 , @xmath107 ) for the @xmath45 state is summarized in table [ stmapping ] .
we now obtain an effective spin chain as @xmath108 + \mbox{h.c.}\end{aligned}\ ] ] .relationship between fermion and spin-@xmath104 operators for the @xmath45 state . [ cols="^,^,^,^ " , ] similarly , we can extend the @xmath2 spin mapping for general @xmath3 . for @xmath93 cases , the configuration in the tt limit is given by @xmath109 , and we obtain @xmath2 chains with @xmath3-sites unit cell ( see also fig . [ s1_chain_p_unit ] ) , @xmath110+\mbox{h.c . }
\label{s=1mapping}\ ] ] where @xmath111 means @xmath2 operators of @xmath112-th site in @xmath70-th unit cell .
we have again ignored the contribution from the electrostatic terms , for simplicity . note that this model includes the @xmath9 model in eq .
( [ s=1_chain_for_nu=1/3 ] ) , but is not appropriate for the @xmath6 state .
since this model also breaks the three discrete symmetries mentioned above , we expect that it has similar low - energy properties as @xmath113 quantum spin chains with haldane gaps . for @xmath114 cases
, effective spin chains can be obtained by replacing @xmath3 in eq .
( [ s=1mapping ] ) by @xmath115 .
the detailed derivation and further analysis of this model will be published elsewhere @xcite .
( 0,-2)(27,1 ) ( 0,0 ) ( 1.5,-1.5)@xmath116 ( 0,0)0.3(1.5,0)0.3(3,0)0.3 ( 0.3,0)(1.2,0 ) ( 1.8,0)(2.7,0 ) ( -0.75,0)(-0.3,0 ) ( 3.3,0)(3.75,0 ) ( 6,0 ) ( 1.75,-1.5)@xmath117 ( 0,0)(2.5,0)2 ( 0.5,0)(1.0,0.8 ) ( 0,0)0.3(1,0)0.3 ( 0.3,0)(0.7,0 ) ( -0.75,0)(-0.3,0 ) ( 1.3,0)(2.2,0 ) ( 3.8,0)(4.25,0 ) ( 13,0 ) ( 2.75,-1.5)@xmath118 ( 0,0)(3.5,0)2 ( 1.0,0)(1.5,0.8 ) ( 0,0)0.3(1,0)0.3(2,0)0.3 ( 0.3,0)(0.7,0 ) ( 1.3,0)(1.7,0 ) ( -0.75,0)(-0.3,0 ) ( 2.3,0)(3.2,0 ) ( 5.8,0)(6.25,0 ) ( 23,0 ) ( 1.5,-1.5)@xmath119 ( 0,0)(4,0)1 ( 1.5,0)(2.0,0.8 ) ( 0,0)0.3(1,0)0.3 ( 2,0)0.3(3,0)0.3 ( 0.3,0)(0.7,0 ) ( 1.3,0)(1.7,0 ) ( 2.3,0)(2.7,0 ) ( -0.75,0)(-0.3,0 ) ( 3.3,0)(3.75,0 )
we have studied the fractional quantum hall ( fqh ) effect based on the one - dimensional description with torus boundary conditions and illustrated how energy spectra of interacting electrons in a magnetic field changes as functions of the circumference of the torus for various filing factors using exact diagonalization . in an effort to understand these results microscopically , we discussed spin chain mappings of the fqh states . by extending earlier results for the @xmath6 and the @xmath9 cases , we mapped the jain sequences @xmath0 and @xmath120 to @xmath2 chains with @xmath3 spins in the unit cell that we conjecture to be gapped for all @xmath3 .
this further strengthens the known analogies between the fqh physics and the quantum spin chains .
we have also studied the gapless state at @xmath6 numerically . by going to slightly larger system sizes than before ( from @xmath121 to @xmath122 particles )
, we confirmed that the evolution from the thin torus to the bulk is , after an initial first order transition that is accurately describe by an @xmath7 spin chain model , smooth and described by a natural deformation of the quasi - particle fermi sea .
we thank a. karlhede and j. suorsa for related collaborations , and k. okamoto and m. oshikawa for discussions .
m. n. acknowledges support from global center of excellence program `` nanoscience and quantum physics '' of the tokyo institute of technology by mext .
9 wilkin n k , gunn j m f and smith r a 1998 _ phys .
lett . _ * 80 * 2265 ; cooper n r and wilkin n k 1999 _ phys rev .
_ b * 60 * 16279(r ) ; regnault n and jolicoeur th 2003 _ phys . rev .
lett . _ * 91 * 030402 ; for a recent review see viefers s 2008 _ j. phys . :
condens . matter _ * 20 * 123202 .
bergholtz e j , kailasvuori j , wikberg e , hansson t h and karlhede a 2006 _ phys .
_ b * 74 * 081308(r ) ; seidel a and lee d h 2006 _ phys .
lett . _ * 97 * 056804 ; read n 2006 _ phys .
_ b * 73 * 245334 ; ardonne e , bergholtz e j , kailasvuori j and wikberg e 2008 _ j. stat .
_ p04016 ; ardonne e 2009 _ phys .
lett . _ * 102 * , 180401 ; seidel a 2010 _ phys .
lett._*105 * , 026802 .
wen x -g and wang z , 2008 _ phys .
_ b * 78 * 155109 ; barkeshli m and wen x -g 2009 _ phys .
_ b * 79 * 195132 ; bernevig b a , gurarie v and simon s h 2009 _ j. phys . a : math .
theor . _ * 42 * 245206 ; bernevig b a and haldane f d m 2009 _ phys .
lett . _ * 102 * 066802 ; hermanns m 2010 _ phys .
lett . _ * 104 * , 056803 ; thomale r , estienne b , regnault n and bernevig b a arxiv:1010.4837 .
pan w , strmer h l , tsui d c , pfeiffer l n , baldwin k w and west k w 2003 _ phys .
lett . _ * 90 * 016801 ; pan w , xia j s , strmer h l , tsui d c , vicente c , adams e d , sullivan n s , pfeiffer l n , baldwin k w and west k w 2008 _ phys .
_ b * 77 * 075307 .
pollmann f , berg e , turner a and oshikawa m , arxiv:0909.4059 ( unpublished ) ; pollmann f , turner a , berg e and oshikawa m , 2010 _ phys .
_ b * 81 * 064439 .
wang z y , nakamura m and bergholtz e j 2011 _ in preparation . | we discuss the relationship between the fractional quantum hall effect in the vicinity of the thin - torus , a.k.a .
tao - thouless ( tt ) , limit and quantum spin chains .
we argue that the energetics of fractional quantum hall states in jain sequence at filling fraction @xmath0 ( and @xmath1 ) in the lowest landau level is captured by @xmath2 spin chains with @xmath3 spins in the unit cell .
these spin chains naturally arise at sub - leading order in @xmath4 which serves as an expansion parameter away from the tt limit ( @xmath5 ) .
we also corroborate earlier results on the smooth fermi surface deformation of the gapless state at @xmath6 , interpolating between a state described by a critical @xmath7 chain and the bulk . |
the vast energy budget required by active galactic nuclei ( agn ) is probably provided by the accretion of matter onto a supermassive black hole ( bh ) . evidence that such objects can be found in agn is provided by the keplerian motions of the mega - masers observed in some sources ( e.g. miyoshi et al .
1995 ) , the broad fek@xmath4 emission line observed in the x - ray spectrum of some sources ( e.g. mcg-6 - 30 - 15 ; see tanaka et al .
1995 ; wilms et al .
2001 ) and the reverberation mapping experiments ( e.g. blanford & mckee 1982 ; netzer & peterson 1997 ; wandel , peterson & malkan 1999 ; kaspi et al .
2000 , hereafter k00 ; peterson et al .
the gas in the broad line region ( blr ) resides close to the central continuum source and can therefore be used as a probe of the central mass . in particular ,
if the blr is gravitationally bound and in near - keplerian orbits , it should be possible to estimate the central mass , @xmath5 , from the mean radius , @xmath6 , and the velocity dispersion , @xmath7 , of the emitting gas .
evidence that the velocity dispersions of several emission lines , in particular , are indeed dominated by keplerian motions is presented by peterson & wandel ( 1999 ) and onken & peterson ( 2002 ) .
velocity widths measured either from the rms spectra or the mean spectrum , combined with reverberation mapping measurements of the blr radius , are now available for more than 34 sources ( k00 , and references therein ; onken & peterson 2002 ; woo & urry 2002 , and references therein ) .
although there are some problems with this method ( see krolick 2001 for a detailed discussion ) , comparison of masses obtained in this way with those obtained from the bulge stellar velocity dispersion have been found to be in good agreement ( ferrarese et al .
2001 , gebhardt et al .
2000 , woo & urry 2002 ) .
reverberation mapping is extremely time consuming , requiring that source light curves are well sampled over a number of years and that the sources themselves are variable . as a result of these constraints ,
reverberation mapping has only been used to estimate the virial mass in a small number ( @xmath840 ) of relatively low redshift objects .
it is not practical to expand these studies to obtain masses for a larger number of sources spanning a larger range in redshift and luminosity .
the available reverberation mapping results show a clear correlation between the radius of the emitting region , @xmath9 , and the monochromatic luminosity at 5100 such that @xmath10 @xcite where @xmath11 .
the uncertainty on the slope @xmath12 reflects the large scatter in the data as well as the statistical method used for the linear regression .
these issues have been discussed , in detail , by vestergaard ( 2002 ) , mclure and jarvis ( 2002 ) , maoz ( 2002 ) and others .
the most recent paper on this topic , by netzer ( 2003 ) , also addresses these issues and comments on the use of the so called `` photoionization radius '' ( see wandel et al .
regardless of the exact value of @xmath12 , such a relationship can be used to estimate @xmath9 in _ every source _ of known luminosity and to obtain the bh mass by combining this size with the observed line width .
this method has been referred to as the _ single epoch mass estimate_. reverberation mapping experiments have concentrated on the emission line .
this has restricted the use of the `` photionization method '' to the measurement of bh mass in relatively low redshifts ( @xmath130.9 ) and low luminosity objects , although there are some cases of high - z quasars where the line has been observed in the infrared ( shields et al .
2002 and references therein ) .
the obvious extension to high redshift sources is to use the uv luminosity as an estimate of @xmath6 and the width of a certain uv line ( e.g. , or ) as a measure of the gas velocity .
unlike the case of , where the location of the line emission ( @xmath9 ) is known , there is no direct measure of the location of the uv lines except in a handful of sources .
thus , the only solution is to calibrate the known relationship for against other lines assuming the two must result in the same derived bh mass .
this must be done for sources where @xmath9 is known and where the velocity width of the uv line and the uv continuum luminosity can be measured .
such intercalibration has been performed , successfully , for by vestergaard ( 2002 ) and for by mclure & jarvis ( 2002 ) .
thus , both and can be used to obtain single - epoch bh masses .
although the results from both vestergaard ( 2002 ) and mclure & jarvis ( 2002 ) are encouraging , their correlations display a large amount of scatter .
this may well be intrinsic , rather than due to poor signal to noise .
one possibility ( young et al . 1999 ; mclure & dunlop 2002 ; smith et al .
2002 ) is that some ( or even most ) blrs display a flattened disc - like geometry . in this case
, the emission line widths would be a function of the viewing angle , with larger widths observed at larger inclinations , resulting in the velocity dispersion ( and hence mass ) of a face - on system being underestimated .
in addition , the k00 original relationship which is the basis of all further work , is limited to @xmath14 ergs s@xmath15 . thus , mass estimates for the highest luminosity sources are uncertain because of the necessary extrapolation . in this paper , we make use of high signal - to - noise ratio ( @xmath16 ) composite spectra from the 2df qso redshift survey 2qz ( 2qz ; see croom et al .
2002 ) to test the velocity - luminosity relationship in qsos and to obtain new bh mass estimates .
the use of composite spectra minimizes the scatter in the relation between luminosity and velocity width caused by effects such as inclination angle .
it also minimizes the influence of objects with extremely small or extremely large line widths .
the broad range in luminosity and redshift allows us to investigate correlations of velocity width and bh mass with redshift and luminosity . in section 2
we briefly describe the primary spectral database and outline the procedure used to obtain the composite spectra .
we discuss the method used to measure the emission line widths in section 3 and derive relations between velocity width and luminosity for a number of emission lines in section 4 . in section 5
we extend the work to estimate black hole masses from the uv broad emission line widths and the qso luminosity , calibrating them against the reverberation mapping results for , and discuss the implications of these results for both qso evolution and blr structure .
our conclusions are summarized in section 6 .
the data used in our analysis is taken from the 2df and 6df qso redshift surveys ( 2qz , croom et al . 2001 ; 6qz croom et al .
2003 , in preparation ) .
qso candidates were selected from uk schmidt photographic material over the range @xmath17 using their @xmath18 colours .
photometric errors in the @xmath19 band were typically 0.1 mag .
objects brighter than @xmath20 were observed with the 6df facility on the uk schmidt over the range 39007600 with a spectral resolution of 11 .
those fainter than @xmath20 were observed with 2df on the aat with a spectral range of 37007900and a spectral resolution of 9 .
typical redshift errors are @xmath21 .
the median snr was 15 for the 6df spectra and 5 for the 2df spectra .
in total 22041 independent qso spectra from the 2df and 6df observations were included in the dataset .
absolute magnitudes ( @xmath22 ) were computed from the observed photographic @xmath19 magnitude , after correction for galactic extinction @xcite , using the k - corrections found by cristiani & vio ( 1990 ) . throughout
we assume a flat cosmological world model with @xmath23 , @xmath24 and @xmath25 mpc@xmath15 .
full details of the creation of the composite spectra and line profile fitting is given by croom et al .
briefly , composite qso spectra ( excluding bal qsos ) were created in discrete absolute magnitude ( @xmath26mag ) and redshift ( @xmath27 ) bins .
a further series of composite spectra were obtained from the qsos binned in absolute magnitude only , as well as a composite contained all qsos .
the numbers of qsos contributing to each composite are given in table 1 in croom et al .
( 2002 ) , typically ranging from 100 to 1000 objects .
the individual spectra were not flux calibrated and each individual spectrum was normalised by a continuum fitted between known emission features . due to the limited wavelength coverage of these spectra it was necessary to treat the large features underlying the 2798 emission line as continuum .
after continuum normalization the spectra were shifted to the rest frame , interpolating linearly onto a uniform scale of 1bin@xmath15 .
finally the composite spectrum in each @xmath28 bin was produced by taking the median value in each wavelength element .
the median @xmath29 and @xmath22 of the contributing qsos was also determined for each wavelength element the composite formed from all 22041 qsos is shown in figure 1 .
a number of emission features can be clearly identified , including 3 narrow ( forbidden ) lines , 9 broad ( permitted ) emission lines and one semi - forbidden line ( 1909 ) .
the features identified in figure 1 with solid lines are suitable for detailed study because they exhibit large equivalent widths ( e.g. , 1909 ) , are relatively free from contamination by other emission lines and are visible over a large redshift range . or ( fig.1 ; dotted lines ) were not included in our study as , due to the redshift distribution of the sample , too few spectra contain these emission lines for meaningful conclusions to be drawn .
although the use of composite spectra provides a good picture of the ensemble average properities of the qso population , there may be small subsamples of the population e.g. weak - lined qsos for which the any general trends identified below do not apply .
the use of composite spectra may also hide magnitude - dependent biases against qsos with , in particular , weak broad emission lines which would prove more difficult to identify in the original low snr spectra .
fortunately , although the spectroscopic identification of qsos in the 2qz is a function of apparent magnitude , the identification rate is generally high .
croom et al . ( 2003 ) provide an empirical fit to the overall spectroscopic identification rate ( @xmath30 ) as a function of apparent @xmath19 magnitude : @xmath31 for the 6qz survey , the spectroscopic identification rate is more - or - less constant at @xmath32 .
thus the identification rate for the survey remains above 90 per cent for all but the faintest @xmath33mag range of the combined sample .
even for the faintest magnitude , the identification rate does not go below 80 per cent .
finally , we note that as redshift increases , the observed spectrum corresponds to a different portion of the rest - frame spectrum .
the continuum normalization of the individual spectra that make up the composite will therefore be defined by different regions of rest - wavelength .
this has been largely circumvented by fixing continuum bands in rest - wavelength near the lines of interest and interpolating between them to define the continuum beneath emission lines .
comparison of composite spectra generated for different redshift bins within the same luminosity range do not reveal any obvious evidence of such systematic redshift effects .
it is possible , however , that low level systematic redshift biases are present in our data , but we do not believe that they would be sufficiently strong to alter our main conclusions .
the majority of the strong emission lines in the composite spectra used in this study are blended with other weaker emission lines , usually from different elements .
additionally , permitted emission lines such as and often exhibit both a broad and a narrow component which are emitted from physically distinct regions and we therefore attempted to separate these components . to measure line widths , the local `` pseudo - continuum '' on either side of each feature was fitted with a straight line and subtracted .
we modeled the overlapping lines contributing to each spectral feature using multi - component gaussian fits .
we note , however , that assuming a gaussian form for the features in our spectra is a gross over - simplification , and thus the gaussian fits were only used to remove contaminating emission ( which represented only a fraction of the flux in the main line ) with non - parametric methods used to determine the line equivalent width , central wavelength and velocity widths . for the purposes of this fitting process
the narrow emission lines and the broad balmer lines ( , and ) were modeled as single gaussians . when possible the number of independent parameters in the model was reduced by linking some together . for example , since the 5007,4959 and narrow emission arises from the same region of the qso ( the narrow line region , nlr ) it seems reasonable to assume that the emitting gas will have similar velocity shifts and dispersions .
the broad ultra - violet lines , i.e. from 2798 blue - ward , display emission line profiles with very broad bases which can not be adequately modeled by a single gaussian ( figure 1 ) .
they were therefore fitted with two components ; a very broad gaussian ( full width at half maximum , fwhm@xmath34 ) and a narrower component ( fwhm @xmath35 ) .
these components were constrained to have the same central wavelength , with the exception of as absorption to the blue side of the line results in an asymmetric profile and it was therefore necessary to allow a velocity shift between the two components to fit the line profile . in all cases the best fit to the spectral feature
was found using @xmath36 minimization techniques .
see croom et al .
( 2002 ) for further details on the line fitting procedure .
no attempt was made to model and remove the broad emission underlying the feature .
our normalisation procedure treats much of this feature as continuum emission , since there is no true continuum region close to the line , and it is therefore not possible to model and remove it from the normalised spectrum . the small feature at @xmath37 just to the red of line , was modeled and removed .
it was also not possible to deblend the emission lines underlying the spectral feature , the ] 1402 emission lines from the 1398 emission line and the ] 1892 emission from the 1909 feature ( although the 1857 emission was modeled and removed from ) .
finally , the broad balmer emission line proved difficult to deblend as it is contaminated by emission from both [ ] 4358 and 4364 as well as narrow emission .
since the [ ] and emission are within 6 of each other they are not resolved in the 2df spectra and were therefore modeled as single narrow component centered between the two lines .
the fit was further constrained by fixing the velocity width of the narrow and the combined [ ] and lines to that obtained for the 5007 emission .
once we had modeled a spectral feature , the fits to the contaminating line emission were subtracted , leaving only the line of interest .
non - parametric methods ( rather than the gaussian fit ) were used to measure the flux , velocity widths and central wavelengths of the emission line .
these measurements were obtained for each feature both before and after the contaminating emission was subtracted in order to check for consistency .
the emission lines widths were measured for the composites binned by magnitude and redshift ( 76 spectra in total ; @xmath38 composites hereinafter ) .
line widths have been measured using several different methods .
these methods can be divided into two broad categories .
the most common method ( e.g. boroson & green 1992 ) is based on the full width of the line at some fraction of its maximum flux , e.g. the full width at half maximum ( fwhm ) or full width at zero intensity ( fwzi ) , whereas the second method measures the velocity width outside which some fraction of the line flux falls , e.g.inter-percentile velocity widths ( ipvs , e.g. whittle 1985 ; stirpe 1991 ; stirpe , robinson & axon 1999 ) .
the most obvious difference between these two methods is that in the former method the emphasis is placed on the velocity width of the gas contributing to the core of the line emission whereas the latter method takes into account the total flux in the line .
the relative merits of the two methods are discussed elsewhere ( see , e.g. robinson 1995 ; stirpe 1991 ) and it is not clear which gives the most consistent estimate of the emitting gas velocity . for this study we measured the emission line width of our sample using both methods and
comparison of the velocity widths measured for the same line using the two sytems shows they give consistent results .
[ cols= " < , > , > , > , > , > , > " , ] we derived the spearman rank correlation coefficients for the correlation of line width ( as measured by each method ) with the source luminosity for each line ( table 1a and b ) .
the emission lines widths were corrected for the instrumental resolution ( @xmath39=9for @xmath40=5500 in the observed spectrum ) by assuming that both the emission line and the instrumental resolution were gaussians .
the small number of measurements with widths narrower than the instrumental resolution were excluded as were velocity measurements with fractional errors of more than 20 per cent .
we measured the correlation of the line width both before and after the contaminating emission was subtracted in order to determine whether our fitting procedure could be introducing spurious correlations into the data . _ a priori _ we chose a significant correlation to be one with a formal probability @xmath41 per cent of being due to chance . for the @xmath38 spectra
significant correlations of line width with luminosity were found for the broad emission lines , , and the balmer lines and ( table 1 ) . with the exception of the emission line ,
these correlations were independent of the method used to measure the line width and whether the contaminating emission had been subtracted . for only the ipv of the whole feature ( plus the two lines ) failed to show a significant correlation between luminosity and line width , most likely due to the contribution of the lines to the red side of the line .
a highly significant correlation ( @xmath42 per cent ) was also found between the width of the combined feature and the source luminosity but not in the -subtracted line .
this probably reflects the difficulties of accurately fitting the components in the feature rather than a lack of a correlation between the width of and luminosity .
the blue side of emission line is subject to a considerable amount of absorption and it is probable that is not well modeled using a combination of symmetrical line profiles , such as a gaussians .
any errors in the fit to the component would also cause errors in the fit to the component and the -subtracted is therefore probably not a good representation of the true line .
additionally , any variation in the relative strength of the and lines , for example if became relatively stronger with luminosity , would also alter the fwhm of the feature . aside from , the only broad lines which do not exhibit a significant correlation between line width and source luminosity in the @xmath38 spectra are and .
is a relatively weak emission line and it is difficult to measure its width accurately . on the other hand
does not appear to vary significantly at all , either in flux or velocity ( croom et al . 2002 )
. of the narrow emission lines , is the only line which exhibits a significant correlation with luminosity ( table 1 ) .
lrrrrrrrr ( a ) median fwhm & & + line & @xmath43 & @xmath44 & @xmath45 & @xmath46 & @xmath43 & @xmath44 & @xmath45 & @xmath46 + ly@xmath4 & @xmath4719.43 & 4.50 & 0.511 & 0.099 & @xmath473.64 & 2.13 & 0.161 & 0.047 + siiv & 2.08 & 1.06 & 0.036 & 0.023 & 2.08 & 1.06 & 0.036 & 0.023 + civ & @xmath470.78 & 1.01 & 0.099 & 0.022 & @xmath470.72 & 0.95 & 0.098 & 0.021 + ciii@xmath48+$]siiii@xmath48 $ ] & 0.39 & 0.61 & 0.075 & 0.014 & 1.86 & 1.58 & 0.042 & 0.013 + mgii & 1.43 & 0.40 & 0.048 & 0.009 & 1.41 & 0.40 & 0.048 &
0.009 + @xmath49nev ] & 0.75 & 1.48 & 0.053 & 0.034 & 0.75 & 1.48 & 0.053 &
0.034 + @xmath49oii ] & 1.30 & 1.06 & 0.034 & 0.024 & 1.30 & 1.06 & 0.034 &
0.024 + @xmath49neiii ] & @xmath471.96 & 3.08 & 0.112 & 0.070 & 0.24 & 2.97 & 0.061 & 0.067 + h@xmath50 & 0.77 & 1.30 & 0.062 & 0.029 & 2.79 & 1.93 & 0.017 & 0.044 + h@xmath12 & @xmath470.83 & 1.72 & 0.099 & 0.039 & @xmath473.49 & 1.77 & 0.162 & 0.040 + h@xmath51 & @xmath474.03 & 1.66 & 0.169 & 0.038 & @xmath475.75 & 1.41 & 0.214 & 0.032 + @xmath49oiii@xmath48 $ ] 5007&@xmath472.48 & 1.87 & 0.120 & 0.043 & @xmath472.45 & 1.89 & 0.120 & 0.043 + ( b)ipv & & + line & @xmath43 & @xmath44 & @xmath45 & @xmath46 & @xmath43 & @xmath44 & @xmath45 & @xmath46 + ly@xmath4 & 1.62 & 0.90 & 0.048 & 0.020 & @xmath471.93 & 2.34 & 0.123 & 0.051 + siiv & 2.84 & 0.86 & 0.016 & 0.019 & 2.84 & 0.86 & 0.016 & 0.019 + civ & 0.92 & 0.67 & 0.063 & 0.015 & 0.40 & 0.77 & 0.072 & 0.017 + ciii@xmath48]+$]siiii@xmath48 $ ] & 1.09 & 0.58 & 0.057 & 0.013 & 2.93 & 0.36 & 0.015 & 0.011 + mgii & @xmath470.59 & 1.55 & 0.099 & 0.035 & 1.34 & 1.01 & 0.053 & 0.023 + @xmath49nev ] & @xmath471.92 & 0.90 & 0.110 & 0.020 & @xmath471.92 & 0.89 & 0.110 & 0.020 + @xmath49oii ] & @xmath471.03 & 0.94 & 0.036 & 0.021 & 1.03 & 0.92 & 0.036 & 0.021 + @xmath49neiii ] & @xmath475.48 & 2.41 & 0.188 & 0.054 & @xmath471.44 & 1.99 & 0.095 & 0.045 + h@xmath50 & 1.57 & 1.87 & 0.041 & 0.042 & 2.58 & 1.49 & 0.018 & 0.034 + h@xmath12 & @xmath473.16 & 1.11 & 0.149 & 0.025 & @xmath473.42 & 1.50 & 0.156 & 0.034 + h@xmath51 & @xmath474.81 & 3.05 & 0.193 & 0.069 & 0.16 & 0.95 & 0.076 & 0.022 + @xmath49oiii@xmath48 $ ] 5007&@xmath477.44 & 1.31 & 0.231 & 0.030 & @xmath476.77 & 1.27 & 0.216 & 0.029 + to investigate whether there is any evidence for evolution in the agn line width ( i.e. narrower or broader lines observed at higher redshifts ) , we derived the spearman rank correlation coefficients for a correlation of line width with redshift .
significant positive correlations between redshift and velocity were found for some methods of measuring the velocity of the emission lines ( e.g. the fwhm velocity of , , and ; the ipv of ) but unlike the luminosity correlations , no line showed a significant correlation between redshift and velocity for every method .
a rank spearman correlation test of our data reveals highly significant correlations ( @xmath53 ; generally @xmath54 ) between luminosity and redshift for all the emission lines except and and it is entirely possible that the handful of significant correlations we find between redshift and velocity are due to a combination of the redshift - luminosity ( @xmath55 ) correlation and that seen between luminosity and velocity ( @xmath56 correlation ) .
we therefore derived the partial spearman correlation coefficients ( see croom et al .
2002 for a detailed discussion of this analysis method ) for the data . in this analysis
we test the hypothesis that any observed correlation between velocity and redshift is due to the @xmath55 correlation and the @xmath56 correlation with the coefficients and their associated probabilities shown in table 2 . for comparison we repeated this analysis for the @xmath56 correlation , testing the hypothesis that the @xmath56 correlation is due to a @xmath57 correlation and the @xmath55 correlation ( again shown in table 2 ) .
we find that while the velocity is clearly correlated most strongly with luminosity for the , and emission lines , there is some evidence for a weak anti - correlation between velocity and redshift , significant only for the emission line . if there is a weak trend for the emission lines to become narrower as redshift increases , this could easily be masked by the strong positive @xmath56 and @xmath58 correlations , particularly as the range of redshifts over which the different emission lines are observed is limited .
the possible dependence of line velocity with redshift is discussed further in section 5.2 . none of the narrow emission lines exhibit any evidence for a correlation between redshift and velocity .
we note that neither the or lines exhibit a significant @xmath56 correlation in the partial spearman correlation analysis although both lines show strong positive correlations using the standard spearman correlation .
they are also the only lines which exhibit a weak positive @xmath57 correlation .
this may be because the they are not observed over a sufficiently large redshift range for the effects of the @xmath55 , @xmath56 and @xmath57 correlations to be disentangled using this analysis ( , for example , is only observed in three redshift bins ) . in any case
it is clear from figure 2 and the regression line analysis ( section 4.3 ) that both line widths of both and are correlated positively with luminosity .
the interpretation of the partial spearman correlation analysis for the emission line is also complex .
our results indicate the line exhibits a weak correlation between the fwhm velocity width and luminosity and a much stronger correlation between the ipv width and luminosity .
the ipv widths also show a significant correlation with redshift .
we have been unable to correct for the emission underlying the line and if the relative strength of the features is a function of luminosity or redshift , this could cause the behaviour seen below in the line .
for example , an increase in qso metallicity with redshift would increase the strength of the emission relative to that of the line and hence increase the width of the broad base ( see figure 1 and section 3.1 ) . for our composite spectra
, the @xmath19 luminosity l@xmath59 was estimated from the mean absolute @xmath19 magnitude of the spectra contributing to the emission line in each composite .
the best fit to the relationship @xmath60 was obtained from a standard least - squares fit to the velocity width of the emission lines and qso luminosity dependence .
the parameters @xmath43 and @xmath45 fitted to each line are given in table 3 and a representative data - set is shown in figure 2 ( the fwhm velocity width of the contamination subtracted data ) .
it is likely that the errors in the velocity widths have been underestimated in our analysis due to the method used to normalize the spectra , underlying features errors in the subtraction of the contaminating emission .
the estimated errors on the emission line widths do not take these factors into account and hence the errors on @xmath43 and @xmath45 derived using the standard method are probably an underestimate of the true error on the fit .
we therefore used @xmath36 from the fit to rescale the errors on the velocity so that @xmath36 is equal to the number of degrees of freedom and hence achieve a more realistic estimate of the errors on @xmath43 and @xmath45 .
as a further check of the correlation analysis performed above , we refit the data , this time fitting a line of the form @xmath61 . for
a perfect correlation between velocity width and luminosity ( and no errors in either parameter ) the best fit lines of @xmath62 upon @xmath63 will coincide with those of @xmath63 upon @xmath62 . for
no correlation between @xmath7 and l@xmath59 the lines of best fit will be perpendicular .
the fwhm of the broad hydrogen lines , , and show a slightly steeper dependence on luminosity than the other emission lines ( table 3 ) .
the best fit slopes to , and lie in the range @xmath64 whereas the slope of the hydrogen lines is in the range @xmath65 .
previous investigations into a possible correlation between velocity width and qso luminosity have produced ambiguous results . from observations of the emission line in individual sources
k00 report an anticorrelation between velocity and luminosity with a slope ( in log - log space ) of -0.27 . on the other hand , several authors ( e.g. wandel & yahil 1985 ; joly et al .
1985 ; boroson & green 1992 ; stirpe , robinson & axon 1999 ) have found the width of the balmer lines ( either fwhm , full width at zero intensity or full widths at some percentile of the maximum flux ) to be weakly correlated luminosity .
these studies were all based on samples containing @xmath8150 sources and it is possible that the inconclusive results may be due to the intrinsic scatter in the @xmath56 relationship as well as selection effects .
in reverberation mapping studies ( e.g. wandel et al . 1999 , onken & peterson 2002 ) the rms spectrum of the emission line is often used to estimate the velocity dispersion of the gas .
the rms spectra , formed from the rms difference between the individual spectra obtained during the monitoring programme and the mean spectrum , measures only the components which vary and , since the emission from the nlr does not vary significantly on short time scales it should not be present in the rms spectrum . at first sight it therefore seems reasonable that the velocity dispersion of the blr gas would be best represented by the velocity width measured from the line after subtraction of the narrow emission line component .
however , k00 found that there was little difference between the line width obtained from the rms spectrum and that measured from the mean spectrum without correcting for the narrow line emission .
vestergaard ( 2002 ) also found that the width of the line in the rms spectrum was more similar to that of a single - epoch observation before any correction for nlr emission .
this implies that much of the narrow core of the emission line ( figure 1 ) is from gas associated with the blr ( rather than the nlr ) and we therefore use the line width from the composite spectra before correction for any narrow line emission in the analysis which follows .
for the non - balmer lines , we use the fwhm line widths measured after contaminating emission has been subtracted .
a secondary issue is the underlying emission . as discussed in section 3.1 , many of the emission features have been treated as continuum , particularly the large feature underlying , and it is therefore not possible for us to correct the emission lines for the underlying .
this may impact on our measurement of the fwhm . in their study ,
mclure & jarvis ( 2002 ) used fwhm measurements from spectra taken with the international ultraviolet explorer ( iue ) which had been corrected for the underlying emission .
they found a linear almost one - to - one , correlation between the fwhm measured for the and lines whereas we find that the emission is generally broader than the emission . on the other hand ,
some of the fwhm measurements used by mclure & jarvis ( 2002 ) have been corrected for narrow emission and in these cases it is reasonable to asume that their fwhm would be broader than ours .
there is also an emission line to the red of ( @xmath66 4930 , see e.g. vanden berk et al .
2002 ) which we have not removed before measuring the fwhm .
vestergaard ( 2002 ) found that it was not necessary to correct for the feii emission unless the feii emission from the object was unusually strong , and as there is no evidence for strong emission in our composite spectra , we do not believe this feature will have a significant impact on the fwhm . in any case , we note that k00 did not correct for emission when measuring the fwhm from their mean spectra . having determined which measurements of the line widths most accurately represent the velocity dispersion of the gas , we need to obtain an estimate of the radius of the blr . from reverberation mapping experiments , k00 found the blr radius , @xmath67 @xmath68 where @xmath69 is the continuum luminosity at 5100 ( defined as @xmath70 in k00 and @xmath71 for our data ) .
this result depends on the cosmology assumed and the analytic method used to determine the fit .
k00 assumed @xmath72 , @xmath73 and @xmath74 mpc@xmath15 and used the _ fitexy _ linear regression method ( press et al .
1992 ) , which takes into account errors in both @xmath75 and @xmath6 , to fit their data .
re - analysing the k00 data , assuming the cosmology adopted here ( @xmath23 , @xmath24 and @xmath25 mpc@xmath15 ) , netzer ( 2003 ) finds a similar relationship with @xmath76 using the bces bisector method ( akritas & bershady 1996 ) to fit the same data , netzer ( 2003 ) obtains a somewhat flatter relationship , @xmath77 , although the two fits are formally consistent .
similar results are obtained by mclure & jarvis ( 2002 ) who find @xmath78 for our cosmology and the bces bisector method .
here we have adopted @xmath79 ( equation [ eq : rl ] ) but we note that a flatter @xmath80 relationship will yield smaller black hole masses at higher luminosities .
the black hole mass is then given by ( k00 ) : @xmath81 where @xmath7 is the fwhm of the emission line .
the blr clouds are assumed to be in random orbits about the black hole and thus in equation [ eq : bhmass ] the velocity dispersion is assumed to be @xmath82 ( netzer 1990 ) . if the orbits of the blr clouds are not random but confined to disks , as suggested by e.g. , young et al .
1999 , mclure & dunlop ( 2002 ) , the velocity dispersion is @xmath83 , and the black hole masses are a factor of 3 larger than we derive .
the black hole masses for selected lines were estimated using equations [ eq : rl ] and [ eq : bhmass ] and are shown in figure [ vel_bm_noscal ] plotted against @xmath69 . as expected , a strong correlation can be seen between bh mass ( @xmath5 ) and the luminosity , @xmath69 , for each line with remarkably little scatter , generally less than 0.3 dex .
the best fit of the form @xmath84 for each line is shown in figures [ vel_bm_noscal ] and [ fig_allmass ] with the parameters for each fit given in table [ table_mslope ] .
the data were fitted using both the standard linear regression method ( yonx ) and the _ bces _ method . from table [ table_mslope ] it is clear that the two methods agree to within the errors ( see also figure [ vel_bm_noscal ] ) . for the linear regression method
the errors in the fit parameters were evaluated as described in the previous section , using the @xmath36 of the fit to derive the errors in the mass estimate and hence @xmath43 and @xmath51 .
we note that the linear relationship between @xmath85 and @xmath69 derived for the line is similar to that obtained for the emission line in this analysis .
fits to the other emission lines are slightly offset from , and flatter than , the fit to the hydrogen lines ( figure [ fig_allmass ] ) .
nevertheless , for a given continuum luminosity , the black hole masses obtained from the different emission lines typically lie within @xmath86 dex of each other .
lccccccr line & method & @xmath43 & @xmath44 & @xmath51 & @xmath87 & @xmath50 + & yonx&@xmath4736.79 & 4.32 & 1.00 & 0.09 & @xmath470.06 + & bces&@xmath4735.82 & 5.08 & 0.98 & 0.11 & + & yonx&@xmath4725.49 & 1.82 & 0.76 & 0.04 & @xmath470.24 + & bces&@xmath4725.38 & 1.75 & 1.02 & 0.12 & + & yonx&@xmath4730.85 & 1.93 & 0.87 & 0.04 & @xmath470.22 + & bces&@xmath4731.90 & 1.86 & 0.90 & 0.04 & + & yonx&@xmath4725.65 & 1.17 & 0.76 & 0.03 & @xmath470.38 + & bces&@xmath4727.09 & 1.45 & 0.80 & 0.03 & + & yonx&@xmath4726.54 & 0.80 & 0.78 & 0.02 & @xmath470.15 + & bces&@xmath4725.71 & 0.88 & 0.76 & 0.02 & + & yonx & @xmath4737.12 & 3.32 & 1.01 & 0.08 & 0.0 + & bces&@xmath4736.90 & 2.81 & 1.00 & 0.06 & + all data & yonx&@xmath4732.05&0.08 & 0.90 & 0.00 + & bces&@xmath4733.53&0.78 & 0.93 & 0.02 + and & yonx&@xmath4736.68&0.08 & 1.00 & 0.00 + only & bces&@xmath4736.77&0.74&1.00 & 0.02 + reverberation mapping experiments ( e.g. , peterson & wandel 1999 , 2000 ; onken & peterson 2002 ) find that different emission lines respond to changes in the uv continuum with different time lags .
this indicates that the blr is stratified , with the broader emission lines ( such as ) originating closer to the central engine than the balmer lines . the empirical @xmath88 relationship ( equation [ eq : rl ] )
was obtained by k00 from the and emission lines and thus its use to determine black hole masses via equation [ eq : bhmass ] will lead to an overestimate of the black hole mass for . to obtain more reliable estimates of @xmath5 for the uv lines
it is necessary to introduce a `` calibration '' factor to equation [ eq : rl ] to correct for the difference between the radius of the gas emitting the broad uv lines , such as , and that emitting the lines .
this has already been attempted by vestergaard ( 2002 ) who compared the bh masses estimated from reverberation mapping of the emission line with those estimated from the fwhm in the same object assuming that , for the line , @xmath89 where @xmath90 is the continuum luminosity at 1350 ( i.e. @xmath40l@xmath91 in vestergaard 2002 ) and @xmath43 is a constant .
she found @xmath92 which implies that the radius of the emission , @xmath93 ( assuming that the qso spectra have a spectral index of @xmath94 ) although the errors in @xmath43 , and hence this result , are large . to determine the calibration factor for our data we re - normalised the mass - luminosity relationship so that the weighted mean of the mass derived from each line lies on the ( yonx ) relation , noting that an extrapolation of the mass - luminosity relation to higher luminosities is entirely consistent with that observed for .
however , because our estimates of black hole mass are based on the underlying assumption that the k00 radius - luminosity relation @xmath95 holds for all emission lines , we chose not to adjust the slopes of the observed mass - luminosity relation for each emission line
. the mass - luminosity relationships were therefore normalised by the addition of a constant offset , @xmath50 ( table 4 ) .
assuming that this offset is due to the emission lines being emitted from gas situated at different radii , @xmath50 can be converted to provide an estimate of the radius at which the uv lines are emitted relative to that of the emission ( radius = @xmath96 ) .
we find that the @xmath97 ( table 4 ) which is consistent with that found by vestergaard ( 2002 ) . from their reverberation mapping study of ngc 3783 ,
onken & peterson ( 2002 ) also find that the emission line lag is approximately half that of the line although the errors on this measurement are also large .
the calibration factor @xmath50 does not address the variation in the slopes of the mass luminosity relationships for the different lines , which , if significant , must result from a difference in the measured line width - luminosity relation for different species . since it is difficult to see how different emission lines could give rise to a different mass luminosity relations ( a line - independent quantity ) based on the same set of qso composite spectra ,
the most straightforward interpretation of this observation is that the different slopes are caused by the contaminating emission blended with the lines .
as we have already discussed in this and previous sections , the and emission lines are blended with emission from other elements ( and respectively ) which would effectively broaden the measured fwhm of the lines , making our calculation of @xmath50 unreliable .
the blended emission could also alter the slope of the @xmath98 relation , particularly if the relative contribution of the lines vary with luminosity or redshift .
we note that the slope of the @xmath99 relationship for and are statistically identical given their errors ( table 4 ) and that these lines also suffer the least amount of contamination .
we can not , however , rule out the possibility that the radius - luminosity relation may differ slightly from line to line .
the re - normalised mass - luminosity relations for all lines are plotted in figure [ fig_allmass_scaled ] and we fit an overall mass - luminosity relation of the form : @xmath100 to all emission lines ( solid line , fig 5 ) .
however , mindful the problems associated with contaminating emission , we also fit a mass - luminosity relation to the and emission lines only ( fig.5 ; dotted line ) since these lines are the least affected by contaminating emission and should therefore give the most reliable estimate .
for the and lines alone , we find @xmath101 which is slightly steeper than that found for all the emission lines combined .
the errors on the @xmath102 relations are larger than those quoted in table 4 . in estimating the errors on these relations we have taken into account the errors on the initial fit to which , since it is used to normalise the mass estimates from the other emission lines , contributes to the overall errors on the @xmath102 relation .
the error in the fit to the and lines is larger than that obtained for all the emission lines as we have fewer data points .
note the errors on the gradient and the intercept in equation [ eq : ml1 ] and [ eq : ml2 ] are not independent and have a correlation of nearly -1 .
the relations we obtain are in reasonable agreement with the statistical analysis of netzer ( 2003 ) , who finds @xmath103 for @xmath104 ( where @xmath90 is the luminosity at 1350 ) for a sample of over 700 objects ( table 1 ; netzer 2003 ) . an important caveat on the @xmath102 relationship given in equations [ eq : ml1 ] and [ eq : ml2 ]
is that throughout our analysis we have used the relationship @xmath105 discovered by k00 and later refined by mclure & jarvis ( 2002 ) and netzer ( 2003 ) . as discussed previously ,
the value of @xmath12 is uncertain and depends on the statistical method used to fit the data , with the _ fitexy _ method and the bces method giving values of @xmath106 and @xmath107 respectively for the same data .
although there is a correlation between velocity and luminosity ( section 4 ) the slope of the @xmath108 relationship is quite flat for most of the emission lines ( e.g , @xmath109 for , 0.1 for , 0.05 for ; table 3 ) and the slope of the @xmath110 relation is therefore dominated by the value assumed for @xmath12 which is currently known to an accuracy of only @xmath111 .
for example , if we adopt @xmath112 in equation [ eq : rl ] , we find @xmath113 @xmath114 for all lines and @xmath113 @xmath115 for the and lines alone , again consistent with netzer ( 2003 ) .
additionally , and perhaps more importantly , it is not known whether @xmath12 is constant across all luminosity ranges .
the sources in k00 s sample all have @xmath116 erg s@xmath15 , with only 4 ( out of a total sample of 34 ) having @xmath117 erg s@xmath15 .
our mass estimates for the uv lines are therefore based on the assumption that the @xmath118 relation can be safely extrapolated to higher luminosities .
netzer ( 2003 ) found that eliminating the 3 objects with @xmath119 erg s@xmath15 from k00 s sample resulted in a @xmath120 and @xmath121 whereas eliminating the seven objects with @xmath122 erg s@xmath15 resulted in @xmath123 and @xmath124 .
reverberation mapping of multiple line species in the same object and extending the luminosity range of reverberation mapped sources would clearly be of great benefit in eliminating some of uncertainties with this analysis .
a number of studies have found that the stellar velocity dispersion of the host galaxy bulge is related to the bulge luminosity which is , in turn , related to the mass of the black hole ( e.g. magorrian et al .
1998 ; ferrarese et al .
2001 ; tremaine et al .
recently , shields et al .
( 2002 ) and boroson ( 2003 ) investigated the possibility of using the velocity width of as a substitute for the stellar velocity dispersion when determining the black hole mass .
they both found that the velocity width of is correlated with black hole mass , assuming that the black hole mass @xmath125 @xmath126 where @xmath75 is the qso luminosity at 5100 , @xmath7 is the velocity width of the emission line and @xmath127 and 0.7 for shields et al .
( 2002 ) and boroson ( 2003 ) respectively .
we find that the velocity width of is correlated with the luminosity of the qso ( table 1 ) with @xmath128 ( table 3 ; fwhm values as the ipv values are probably contaminated by emission either side of the lines ) , similar to that found for .
this raises the intriguing possibility first , that the nlr in qsos is sufficiently close to the black hole for its motion to be dominated by the gravitational potential of the black hole , and secondly , that the narrow emission lines may be used to derive the black hole mass .
however , none of the other narrow emission lines in our study ( , and ) display the same strength of correlation between line width and qso luminosity as and given the anti - correlation between narrow line strength and continuum luminosity ( croom et al . 2002 ) the use of narrow lines in estimating the black hole mass may be limited to low luminosity and hence low mass objects .
figure [ vel_bm_noscal ] shows the mass luminosity relation for the individual broad lines separated into redshift bins and appears to demonstrate a lack of evolution in the relation between black hole mass and luminosity with redshift .
however , the line does show a significant anti - correlation between velocity and redshift in our partial correlation analysis ( table 2 ) which may be evidence of a weak redshift evolution of line width .
no other lines show significant correlations with redshift . in order to make a quantitative estimate of the amount of redshift evolution we fit a simple power law model to the @xmath58 data points for each emission line .
we first assume that black hole mass is related to luminosity and redshift by independent power laws such that @xmath129 and that secondly , the radius - luminosity relation is also described by a power law : @xmath130 we assume that the radius - luminosity relation ( equation [ eq : rlevol ] ) does not to evolve with redshift as it should be primarily driven by the physics of photoionization , which we would not expect to change with redshift ( although this may not be the case if qso metallicity was found to evolve ) . assuming keplerian motion , we have @xmath131 this relation is then fitted to the observed velocity - luminosity data for each emission line individually ( as we do not wish to assume that the radius - luminosity relation is the same for each emission line ) . to obtain a more realistic estimate of the errors on the fitted parameters given a large intrinsic scatter in the data , we re - scale the errors on individual points such that @xmath132 is equal to the number of degrees of freedom .
the results of this model fitting are shown in table [ table_evol ] .
the only line which shows clear evidence ( @xmath133 ) for evolution is .
however , and also show weak evolution in the same sense .
all of these lines have best fit values of @xmath51 which are negative , implying that for a given luminosity , the measured velocities are lower at higher redshift . by contrast , the value of @xmath51 for the line is entirely consistent with no evolution at all and the balmer lines ( and ) all show positive values of @xmath51 , although these are not significant .
we do not attempt to model the redshift evolution of the line , as the contribution of forest absorption and emission makes any interpretation highly uncertain . to show more clearly the evolution in the velocity width , we plot @xmath63 vs. @xmath134 for this line , first renormalizing the velocity points to a constant absolute magnitude ( based on the above fits ) of @xmath135 ( fig .
[ fig_civ_evol ] ) .
the error bars in fig .
[ fig_civ_evol ] have been rescaled such that @xmath132 is equal to the number of degrees of freedom .
a weak , but significant , decline in velocity can be seen , particularly at @xmath136 .
the evolution seen in the higher redshift lines is in the sense of higher redshift qsos having smaller velocities for the same luminosities , which might be expected if qso luminosity evolution is driven by a decline in fueling rate at lower redshift .
however , the amount of evolution is small compared to the strong luminosity evolution shown by qsos ( typically @xmath3 ) . at @xmath137 the qso evolution slows down and appears to turn over .
assuming pure luminosity evolution ( ple ) , croom et al .
( 2003 ) finds that for the full 2qz catalogue the evolution of @xmath138 , the characteristic luminosity of the qso population , is well described by @xmath139 with @xmath140 and @xmath141 at redshifts less that @xmath142 .
this model implies that @xmath138 reaches an extrapolated maximum at @xmath143 .
if black hole masses do not evolve , and the luminosity evolution is purely due to other influences ( e.g. fueling rate ) , then at a given luminosity @xmath144 .
thus , using the ple fit from boyle et al ( 2003 ) and assuming that @xmath145 we find @xmath146 in figure [ fig_evol_slopes ] we compare this model to the evolutionary slopes found in our fits . our ple model ( solid line ) is steep at low redshift , but flattens off at @xmath137 .
here we assume that @xmath147 , as found in section 5.1 for all broad emission lines , but the shape of the model curve changes little if @xmath148 , as found for the and lines alone ( section 5.1 ) .
we plot , as shaded regions , the range of allowable ( @xmath149 ) slopes for each of five lines ( , , , and ) , over the redshift ranges that they are measured .
each is normalized to the @xmath150 ratio at the mean redshift of the line . at low redshift
, the line has a positive slope which is consistent with @xmath151 , but not with the strong evolution of the model ( solid line ) . at higher redshift the and lines are close to @xmath151 and although this is inconsistent with our model , we note that both these lines are contaminated by blended emission which could be distorting their behaviour . only the highest redshift lines , and
show evolution which is comparible to that shown by the model , partly as they have higher values of @xmath51 and partly as the model starts to turn over at high redshift .
lrr line & @xmath51 & @xmath152 + & @xmath470.76 & 0.36 + & @xmath470.86 & 0.25 + & @xmath470.23 & 0.12 + & 0.00 & 0.18 + & 0.84 & 1.85 + & 0.94 & 2.23 + it appears that only at @xmath153 does the measured evolution in the luminosity - velocity relation compare to the above model . over
the majority of the redshift range covered in this investigation evolution in the luminosity - velocity relation is insignificant .
this conclusion has significant implications for qso evolution , implying little evolution in the mass - luminosity relation for super - massive black holes over a wide range in redshift . assuming no evolution in the mass - luminosity relation we can transform the qso luminosity function ( lf ) and its evolution with redshift into a mass function ( mf ) . in fig .
[ bhmf ] we show the evolution of the qso mass function based on the lf derived from the 2qz+6qz samples ( croom et al .
we immediately note that the luminosity ( mass ) evolution of the qso lf can not be explained by decline in the fueling rate either in continuous model ( long - lived qsos ) or in a duty cycle ( short - lived qsos ) , since this would imply that a qso at a given relative position on the lf ( mf ) evolving from high black hole mass at high redshift to low black hole mass at low redshift .
rather , the evolution of the black hole mass function suggests that qsos can only be active once , possibly associated with formation of the black hole itself during spheroid formation . under this interpretation ,
the evolution of the lf is driven by the efficiency of black hole formation which drops rapidly towards low redshift .
this picture is qualitatively consistent with the recent theoretical work of kauffmann & haehnelt ( 2000 ) , although it places tight constraints on the fraction of qsos which exhibit a duty cycle of activity in such models .
it also implies that the gebhardt et al .
( 2002 ) relation between black hole mass and galaxy velocity dispersion will evolve toward higher ratios at higher redshifts .
the black hole mass will be frozen when black hole formation and subsequent qso activity ceases while the mass / velocity dispersion of the host galaxies will increase due to hierarchical merging to give the relationship seen at the present day . at @xmath154 and greater ,
there does seem to be some evolution of the velocity - luminosity relation , as shown by the ( fig .
[ fig_civ_evol ] ) and lines .
it appears that there is a general decline in velocity for a given luminosity at @xmath136 ( @xmath155 ) .
the measured rate of evolution is no different if we remove the highest redshift bin ( @xmath156 ) and re - fit the data .
the fact that the strongest evolution in the velocity - luminosity relationship is found in the redshift regime corresponding to the peak of qso activity suggests that as @xmath138 declines at high redshift ( @xmath157 ) , the black hole mass - luminosity relation may be changing more rapidly than at lower redshift .
we have used composite spectra generated from more than 22000 qsos observed in the course of the 2df and 6df qso redshift surveys to investigate the relationship between the profiles of emission lines and luminosity .
we find that the velocity width of the broad emission lines , , , and show a positive correlation with the continuum luminosity ( as measured in the @xmath19 filter ) but the slope of this dependence is much flatter for the uv emission lines ( , and ) than for the hydrogen lines .
this correlation is statistically significant with a confidence level @xmath158 per cent and is independent of the method used to measure the emission line widths . from partial spearman correlation analysis
we find only weak evidence for an anti - correlation between redshift and velocity . of the narrow emission lines , , , and ,
only the @xmath405007 line exhibits a correlation between line width and luminosity .
assuming that the gas in the blr is in near - keplerian orbits and that the radius of the blr , @xmath159 ( k00;netzer 2003 ) , where @xmath75 is the continuum monochromatic luminosity at 5100 , we have used our measurements of the emission line velocity widths to derive estimates of the average black hole mass in the composite spectra .
we find that for a given luminosity the black hole masses obtained from the different the emission lines agree to within 0.6dex .
however the scatter in the mass - luminosity relationships for the individual lines is much smaller ( @xmath160 dex ) and given that the different emission lines may be emitted from different radii , producing an offset between the masses estimated from different emission lines , we use the @xmath102 relationship to calibrate the other emission lines .
we find an overall black hole mass - luminosity relationship of the form @xmath161 .
it is possible that the fwhm measurements of several of the emission lines are overestimated as they are blended with weaker lines from different elements which are difficult to remove reliably .
we have therefore also derived the mass luminosity relationship for the and emission lines only , as these are the lines least effected by contaminating emission , and find @xmath162 .
our@xmath102 relationships depend heavily on the slope of the @xmath163 relationship which is currently known to an accuracy of only @xmath111 . using the midpoint of the derived @xmath102 powerlaws and taking into account the error in the slope of the @xmath163 relationship , we conclude that @xmath164 .
we find only weak evidence of evolution in the velocity - luminosity correlation , which places strong constraints on the evolution of black hole masses for a fixed qso luminosity ; @xmath165 , @xmath166 .
this is much smaller than that required for the luminosity ( mass ) function of qsos to be due to a single population of long - lived sources .
the 2df qso redshift survey was based on observations made with the anglo - australian telescope and the uk schmidt telescope .
we warmly thank all the present and former staff of the anglo - australian observatory for their work in building and operating the 2df facility .
stirpe g. , 1991 , a&a , 247 , 3 stirpe g. , robinson a. , axon d. j , 1999 , in _ structure and kinematics of quasar broad line regions _ , ed .
c. m. gaskell , w. brandt , m. dietrich , d. dultzin - hacyan , eracleous , 175 , p. 249 | we have used composite spectra generated from more than 22000 qsos observed in the course of the 2df and 6df qso redshift surveys to investigate the relationship between the velocity width of emission lines and qso luminosity .
we find that the velocity width of the broad emission lines , , , and are correlated with the continuum luminosity , with a significance of more than 99 per cent .
of the major narrow emission lines ( 5007 , 3727 , 3870 and 3426 ) only exhibits a significant correlation between line width and luminosity . assuming
that the gas is moving in keplerian orbits and that the radius of the broad line region is related to the qso continuum luminosity , we use the velocity widths of the broad lines to derive average black hole masses for the qsos contributing to the composite spectra .
the resultant qso mass - luminosity relationship is consistent with @xmath0 .
we find that the correlation between line width and redshift , if present , must be weak , and only shows significant evidence of evolution .
this enables us to constrain the redshift evolution of the black hole mass - luminosity ratio to be @xmath1 with @xmath2 , much less than the @xmath3 evolution seen in qso luminosity evolution . assuming that the motion of the broad line region gas is keplerian and that its radius depends on the qso luminosity , our models indicate that the observed weak redshift dependence is too small for the observed qso luminosity function to be due to the evolution of a single long - lived population of sources . = = = = = = = = # 1 # 1 # 1 # 1 @mathgroup@group @mathgroup@normal@groupeurmn @mathgroup@bold@groupeurbn @mathgroup@group
@mathgroup@normal@groupmsamn @mathgroup@bold@groupmsamn = `` 019 = ' ' 016 = `` 040 = ' ' 336 = " 33e = = = = = = = = # 1 # 1 # 1 # 1 = = = = = = = = galaxies : active quasars : general quasars : emission lines galaxies : stellar content |
the tip - based scanning force microscopes of the atomic force microscope ( afm ) family constitute perhaps the single most important tool bag in nanotechnology .
the substrate topography is extracted from a map of the oscillation frequency of a tip , hovering a short distance above the surface . besides the frequency shift however , the tip also develops a damping , reflecting a position dependent mechanical dissipation .
et al . _
@xcite showed recently that afm dissipation whose general occurrence has been widely discussed by several groups a decade ago @xcite but whose potential importance was still underestimated is able to map exquisitely delicate features such as the moir superstructure pattern formed by misfit dislocations ( `` solitons '' ) of incommensurate kbr adsorbate islands ( fig .
[ model : fig]a ) on nacl(100 ) .
surprisingly , the experimental dissipation map , fig .
[ model : fig]b , showed similar or better contrast than the corresponding topographic map , with a characteristic reversed contrast ( higher dissipation at the soliton , where topographic height is minimal @xcite ) .
given also the great importance of achieving newer routes toward high - quality imaging , this is more than a mere curiosity , and deserves a proper understanding .
existing linear - response theory and other approaches to afm dissipation @xcite and to general frictional dissipation @xcite suggest a larger tip damping above softer substrates , and that provides an initial and valuable clue .
local tip dissipation can effectively reveal the underlying superstructure , since the local mechanical compliance is higher for example at surface soliton lines , where atoms sit at metastable positions .
however at the relatively low afm oscillation frequencies , the current understanding rules out linear response as the chief dissipation mechanism . a typical energy dissipation as large as @xmath1 ev per oscillation can only be accounted for by a hysteretic response of the interacting tip - substrate system , as was understood by theoretical analysis @xcite , and demonstrated experimentally @xcite .
such nonlinear effects of hysteresis are most likely involved in the surprisingly large afm dissipation contrast of adsorbate superstructures too .
yet , it is unclear how inert systems ( such as for example alkali halide overlayers ) could give rise to hysteretic phenomena , and in particular how they would be connected to the presence of misfit superstructures ( `` solitons '' ) .
this is the issue which we address here by means of dynamical simulations of the simplest one - dimensional model .
our main result is the identification of an unexpected soliton - related hysteretic mechanism . during the first part of its swing
, the tip can locally drag or push an underlying defect here a soliton portion causing it to jump across a ( peierls - nabarro ) energy barrier . during the return journey ,
the defect follows only sluggishly , and remains trapped somewhat longer on the wrong side of the energy barrier , thus opening a hysteresis loop .
the area enclosed in the hysteretic force - displacement diagram represents a large tip energy dissipation , one that can survive down to realistically low afm vibration frequencies , a regime where the linear - response dissipation is quantitatively irrelevant .
this mechanism is likely to play a significant role every time a `` softness pattern '' is present , and should be easier to observe for horizontal than for vertical tip oscillations .
( a ) 2 mm ( c )
to emphasize the basic and general aspects of the phenomena , rather than a realistic model targeted on the chemical detail of a specific tip / sample configuration @xcite , we use the simplest possible model a tip potential oscillating over a one - dimensional harmonic chain ( the overlayer ) moving in a rigid incommensurate periodic potential ( the substrate ) , fig .
[ model : fig]c .
the hamiltonian of the mobile overlayer atoms is @xmath2 where @xmath3 is the kinetic energy , @xmath4 is the mutual ( harmonic ) interaction potential , and @xmath5 is the substrate potential , which we take of a pure cosine form : @xmath6 here @xmath7 is the mean spacing between adatoms , @xmath8 is their mutual spring constant , @xmath9 , and @xmath10 is the period of the substrate potential @xcite .
the two periodicities @xmath7 and @xmath10 define the coverage @xmath11 . for convenience
we take @xmath10 as the unit length , @xmath12 as the force unit , and the mass @xmath13 of the particles as the mass unit .
( to get a feeling for quantities , the frequency units @xmath14 should be typical of an atomic vibration or a debye frequency @xmath15 , typically 1 thz or more ) . the general lack of commensuration between adsorbate and substrate periodicities gives rise to two - dimensional misfit dislocations , sometimes called solitons , which form a regular superstructure with the beat periodicity between the two .
fixed boundary conditions ( bcs ) are chosen in order to prevent the advancing tip to drag the entire pattern along , that would occur if , e.g. , periodic bcs were used instead .
@xmath16 is the time - dependent oscillating potential describing the tip action on the overlayer .
we represent the afm tip as a gaussian - shaped oscillating potential , with @xmath17 , @xmath18 , and @xmath19 here @xmath20 represents the repulsive ( contact afm , @xmath21 ) or attractive ( noncontact afm , @xmath22 ) tip - atom interaction strength , @xmath23 is the tip width , @xmath24 and @xmath25 are the tip oscillation amplitude and angular frequency around its central position @xmath26 .
the equation of motion for the @xmath27-th overlayer atom is @xmath28 where @xmath29 , and the tip force @xmath30 is given by a straightforward analytical expression .
a damping force term @xmath31 is introduced to represent all dissipation phenomena which remove energy and allow the attainment of a stationary frictional state .
we integrate the equations of motion ( [ motionequation ] ) by means of a standard adaptive fourth - order runge - kutta routine @xcite starting each simulation from a stationary fully relaxed overlayer , as obtained by a preliminary relaxation of equally - spaced adatoms @xmath32 and @xmath33 .
all simulations are carried out at a nearly commensurate coverage @xmath34 , realized by means of a chain of @xmath35 particles in a region of length @xmath36 . a finite temperature @xmath37 is implemented by adding a standard langevin random force to eq .
( [ motionequation ] ) , and averaging over a long simulation time , usually at least 100 tip - oscillation periods .
the extreme simplicity of the model allows us to extend simulations down to the realistic afm frequency in the mhz range , which requires exceedingly long integration times .
the instantaneous power drained away by the damping term amounts to @xmath38 and is thus proportional to the total kinetic energy of the overlayer .
the power pumped by the tip into the chain is @xmath39 while these two quantities fluctuate separately , they must of course coincide on average over a period @xmath40 in the dynamical steady state @xmath41 also indicating how the work done by the tip oscillation is eventually dissipated entirely by the viscous friction term .
figure [ linear_regime : fig ] displays the dissipation results obtained in simulation for a weakly attractive tip potential , very high oscillation frequency , and general parameters that fall well inside the linear - response regime @xcite .
the linear response results show ( i ) strong dissipation enhancement at solitons , with @xmath42 several orders of magnitude stronger than in a terrace between two of them , closely mirroring the phonon local density of states ( ldos ) ; ( ii ) dissipated power which is proportional to @xmath43 , independent of the attractive / repulsive sign of the tip - overlayer interaction , i.e. of the noncontact or contact mode of the afm ; ( iii ) absolute dissipation values that are very weak everywhere , and dropping with decreasing afm frequency as @xmath44 . summing up
, the predicted relative contrast of the soliton pattern in linear response dissipation is indeed very large .
however , the exceedingly low value of realistic afm frequencies ( @xmath45mhz ) relative to microscopic frequencies ( @xmath46thz ) renders this linear - dissipation mechanism entirely academic . we reach a more realistic regime by enhancing the tip - overlayer interaction strength , while still remaining in a moderate - interaction regime representing noncontact afm .
this new regime is dominated by nonlinear effects , where dissipation no longer drops as @xmath47 , but at most linearly in @xmath48 ( apart from logarithmic corrections ) .
comparison of fig .
[ nonlinear : fig]a with fig .
[ linear_regime : fig]c shows that , nonlinear dissipation is again much larger near the solitons than in between them . a two - order of magnitude increase in @xmath20
would in linear regime imply a dissipation increase by a factor @xmath49 , whereas we find a much larger factor of about @xmath50 already at this large frequency ( @xmath25 is here @xmath51 , corresponding to the gigahertz range ) .
decreasing frequency down to realistic afm values , the increase will become gigantic , because the nonlinear dissipation lacks the extra power of @xmath25 appearing in the linear - regime dissipation .
the new element brought in by nonlinearity is mechanical .
a strongly interacting tip is now able to drag , or to push , the soliton a mobile entity forward or backward during the oscillation cycle . as the soliton must overcome the ( peierls - nabarro ) barrier in order to move , its motion is sluggish , and can follow the tip only with hysteresis and , as anticipated , hysteresis entails a large dissipation . as shown in detail in fig .
[ nonlinear : fig]a , b , the higher dissipation point a is found to corresponds to two successive peierls - nabarro barriers being overcome in the oscillatory process , the smaller dissipation point b to a single barrier .
the dissipation at point c , where the tip potential is unable to `` grab '' the soliton , is negligible by comparison .
the onset of this large - dissipation region , dominated by hysteresis , is rather sharp .
[ nonlinearity : fig ] illustrates this point , by showing the average power @xmath52 at location a , divided by the linear - regime factor @xmath53 , as a function of the tip amplitude @xmath54 . beyond a value of @xmath54 of order @xmath55 , the linear regime behavior
is abruptly abandoned , and the dissipation increases rapidly by several orders of magnitude .
our simulated example strongly suggests that a large hysteretic component should be present in the existing dissipation maps @xcite of moir patterns .
more generally , hysteretic defect dragging should dominate the afm dissipation maps .
what are the predicted signatures of this mechanism ?
our model study suggests two main signatures .
\(i ) abruptness of afm friction onset with increasing strength of tip - surface interaction . as suggested by figs .
[ nonlinear : fig]-[nonlinearity : fig ] , dragging sets in abruptly only above a certain threshold , which means below a certain tip - surface or tip - soliton distance .
\(ii ) anomalously mild ( linear with logarithmic corrections ) frequency dependence of afm friction at finite temperature .
it is a rather general property of all hysteretic friction phenomena to heal away at sufficiently low frequencies , where adiabatic motion allows sufficient time to jump thermally over barriers .
for instance , thermolubricity experiments @xcite and detailed calculations within the tomlison model @xcite show an average friction force @xmath56 , where @xmath57 is the driving velocity , and @xmath58 and @xmath59 are system - dependent dimensional constants . in incommensurate overlayers , the soliton nearest to
the tip behaves similarly to a tomlinson particle , as is driven across a peierls - nabarro barrier . in fig .
[ powervsfrequency : fig]a we do observe a thermal reduction of dissipation , due to a shrinkage of the hysteretic loop , illustrated in fig .
[ powervsfrequency : fig]b .
when driving is oscillatory as in afm dissipation , the role of @xmath57 is taken by @xmath60 .
we find our data to be compatible with a similar relation @xmath61
. the parameters @xmath62 and @xmath63 are here related to the effective soliton properties ( mass , damping , barrier height ... ) , and are nontrivial functions of the `` bare '' model parameters . we conclude that afm dissipation maps of incommensurate overlayer superstructures can in principle achieve an extremely high contrast resolution of soliton defects relative to commensurate terraces .
the most important theoretically predicted dissipation mechanism is the nonlinear dragging or pushing of some local portion of the defect , where the large tip damping is associated with hysteresis of defect motion . besides a sharp threshold in the tip - surface interaction and oscillation magnitudes ,
this mechanism predicts a very characteristic logarithmic dependence ( eventually turning to linear at extremely low frequencies ) of dissipation upon frequency and temperature .
more generally , the nonlinear dragging of soft defects or features ( e.g. a floppy residue in a biomolecule ) should give rise to a strong visibility in afm dissipation topography , of considerable potential impact for applications . | in the atomic force microscope , the nanoscale force topography of even complex surface superstructures is extracted by the changing vibration frequency of a scanning tip . an alternative dissipation topography with similar or even better
contrast has been demonstrated recently by mapping the @xmath0-dependent tip _ damping _ : but the detailed damping mechanism is still unknown . here
we identify two different tip dissipation mechanisms : local mechanical softness , and hysteresis .
motivated by recent data , we describe both of them in a one - dimensional model of moir superstructures of incommensurate overlayers .
local softness at `` soliton '' defects yields a dissipation contrast that can be much larger than the corresponding density or corrugation contrast . at realistically low vibration frequencies ,
however , a much stronger and more effective dissipation is caused by the tip - induced nonlinear jumping of the soliton , naturally developing bi - stability and hysteresis .
signatures of this mechanism are proposed for experimental identification . |
narrow line seyfert 1 galaxies ( nls1s ) are a special type of active galactic nuclei ( agn ) whose optical spectra are similar to seyfert1 type , but with much narrower balmer line width ; they also show weak [ oiii ] lines and strong feii emission in optical and uv ( osterbrock & pogge 1985 ) .
the importance of nls1s lies in that their properties are at the extremity among seyfert1 family , including steep soft x - ray spectra ( @xmath0 ) and rapid x - ray variability , strong blue wings of emission lines ( e.g. boroson & green 1992 ; laor et al . 1994 ; wang 1996 , boller et al .
1996 , grupe et al . 1999 , 2001 , xu , et al .
observational data show that nls1 tend to have small black hole masses and the high eddington ratios @xmath1 ( e.g. boroson 2002 ) .
in fact , they were found to locate at one extreme end of eigenvector1 of the correlation matrix ( boroson & green 1992 , sulentic et al .
2000 ) , which is believed to be driven primarily by @xmath2 .
radio - loud nls1s were also found to exist , though their number is still small ( e.g. komossa et al . 2006 ) .
yet there are quite a few aspects about nls1s where controversies remain or little study has been given to .
the reasons can be ascribed partly to the relatively small number of nls1 known ( a few hundreds ) before this work , as well as the heterogeneity of the objects as a whole .
for example , little work has been done regarding broad band sed , associated outflows , radio - loud nls1s and their jets , host galaxies , optical variability , mean spectrum and dispersion , the @xmath3 relation , as well as their cosmic evolution . a much larger and homogeneously selected nls1 sample is required . with its large sky coverage and unprecedented sensitivity and uniformity , the spectroscopic survey of the sloan digital sky survey ( sdss ; york et al .
2000 ) provides us with the most suitable database for discovering nls1s . as a pioneer work , williams et al .
( 2002 ) have compiled a sample of 150 nls1 from the sdss early data release .
we carried out an extensive and systematic search for nls1s by making use of the sdss dr3 data , which yielded 2011 nls1s , mostly previously unknown .
the sample and preliminary statistical results are presented in zhou et al .
( 2006 ) . in this proceedings contribution
, we summarize the sample properties and some of the significant results , with some new , preliminary development .
our nls1s were drawn from the sdss dr3 spectroscopic databases of both galaxies and qsos . the data reduction procedure , as described in detail in zhou et al . ( 2006 ) , consists of mainly two steps .
firstly , star light contribution to fiber spectra from host galaxies is carefully removed by performing starlight - nuclear spectral decomposition using our algorithm ( lu et al .
2006 ) developed at university of science of technology of china .
our technique has proved to be accurate and robust when performing extensive testing against results from independent observations and measurement , including hst spectroscopic measurements ( wang t. , et al . in prep . ) .
secondly , nuclear emission - line spectra with the star - spectrum and continuum subtracted are fitted to derive spectral parameters using the code as described in dong et al .
in this process , the balmer emission lines are de - blended into a broad and a narrow component this is essential as by doing so we could really measure the physical parameters of the broad line region
. this process accounts for the difference in the results for some of the objects between our work and that of williams et al .
( 2002 ) .
we selected nls1s as those having fwhm of the _ broad _ component of h@xmath4 or h@xmath5 ( detected at the @xmath6 confidence level ) narrow than @xmath7 .
the 10% relaxed margin in the linewidth cutoff compared to the conventional @xmath8 ( goodrich 1989 ) is to recover those nls1s dropped out due to the uncertainty in linewidth measurement .
our criteria resulted in 2011 nls1 candidates .
using a stricter criterion of @xmath9 would reduce the sample to 1885 objects .
all but one fulfill @xmath10/h\beta<3$]the second item of the conventional definition .
we gave , along with the source list , measured spectral parameters and their error estimates .
below we summarize some of the preliminary results directly based on statistical analysis of observed and derived parameters ; for details of these analysis please refer to zhou et al .
( 2006 ) .
\1 ) nls1 fraction
+ we found a strong dependence of the nls1 fraction among all seyfert1 agns . with increasing optical luminosity , the fraction of nls1s first increases and peaks at @xmath11 , and then drops again rapidly .
the nls1 fraction also depends on radio - loudness , dropping from @xmath1215% in radio - quiet to only 5% in radio - loud objects .
the nls1 fraction in optically selected agn sample is much lower than that in soft x - ray ( rosat ) selected agn samples .
\2 ) emission line properties + on average the relative feii emission , @xmath13 , in nls1s is about twice that in normal agns , and is anti - correlated with the broad component width of the balmer emission lines .
the equivalent width of h@xmath4 and feii emission lines are strongly positively correlated with the h@xmath4 and continuum luminosities ( the inverse baldwin effect ) .
we do not find any difference between nls1s and normal agns in regard to the narrow line region .
\3 ) @xmath14@xmath15 relation + we have examined the black hole mass vs.stellar velocity dispersion ( @xmath16 ) relation for a subsample of 308 nls1s for which @xmath16 could be measured directly from fitting the starlight in the sdss spectra with our stellar spectral templates . a significant correlation between @xmath17 and @xmath16 is found , but with the bulk of black hole masses falling below the values expected from the @xmath18 relation for normal galaxies and normal agns ( figure1 , left panel ) .
this result indicates that nls1s are underage agns , where the growth of the smbh lags behind the formation of the galactic bulge ( see also , e.g. mathur et al .
2001 , grupe et al .
2004 , bian & zhao 2004 , botte et al .
we also find that the width of [ nii ] is well correlated with @xmath16 .
\4 ) x - ray properties + 635 nls1s in the sample have x - ray counterparts detected in the rass .
the well - known anti - correlation between the width of broad low - ionization lines and the soft x - ray spectral slope for broad line agns extends down to @xmath19 in nls1s , but the trend appears to reverse at still smaller line widths .
there are 28 nls1s being observed by xmm - newton either as targets or serendipitously , for which their spectra were analyzed . for most of the objects
the 0.310kev x - ray continuum can be fitted with a power law plus a soft x - ray excess , mostly well parameterized by a blackbody component .
no significant fe k@xmath5 line was detected . in figure1
( right panel ) we plot the distribution of the photon indices of the underlying continuum derived from xmm - newton data .
the spectral slopes are systematically steeper than those for broad line seyfert 1 galaxies , confirming previous results ( e.g. leighly 2000 ) .
\5 ) radio - loud nls1s + of the sample 142 ( 7% ) nls1s were detected in radio in the first survey .
about two dozen are very radio loud ( radio - loudness @xmath20100 , zhou et al . in prep . ) .
of particular interest , some objects show hybrid properties of nls1 and blazar with 2masxj0324 + 3410 being an extreme example ( zhou et al .
2007 , apj in press ) .
a detailed investigation of radio - loud fraction of nls1s is underway , by taking into account various selection effects in the sdss spectroscopic survey .
we are currently working on updating the nls1 agn sample using the latest sdss dr5 data . other work currently undertaken
includes : analyzing the x - ray spectral and temporal data for those having xmm - newton and rosat observations , examining optical variability of nls1 using the sdss supernovae survey data ( ai et al .
this proceedings ) , radio - loud nls1s , broad - band sed .
as for the future , we also plan to work on , as example , nls1s host galaxies and their stellar populations and comparisons with those of broad line seyfert agns , high - redshift nls1s . | this proceedings contribution summarizes first results obtained from a systematic study of 2011 nls1 galaxies , as presented in zhou et al .
( 2006 , apjs 166 , 128 ) .
the sample was compiled by examining the spectral parameters of galaxies and qsos derived from sdss dr3 data .
we discuss some preliminary results on the statistic properties of the sample , such as the fraction of nls1 , the properties of broad and narrow emission lines , and emission in other wavebands .
the black hole mass velocity dispersion relation for nls1s was re - examined using the velocity dispersion values estimated from the stellar absorption spectra of the host galaxies .
preliminary result from an x - ray study for a small subset using data obtained by xmm - newton is briefly discussed . |
we consider the cauchy problem for the ( nonlinear ) schrdinger equation @xmath1 where the potential @xmath2 is smooth and sub - quadratic ( see below ) , the nonlinearity @xmath3 is sufficiently smooth , and the initial data @xmath4 may or may not belong to weighted @xmath5 spaces @xmath6 ( sometimes denoted @xmath7 ) , where @xmath8 stands for the fourier transform .
note that we consider only propagation in the future ; this choice is made only to simplify some statements .
we show that if the potential @xmath2 is sub - linear , then is locally well - posed in @xmath9 , upon suitable assumptions on @xmath3 . on the other hand ,
if @xmath2 is super - linear ( e.g. harmonic potential ) , then is ill - posed in all sobolev spaces of positive order ; this is not a nonlinear result , since it holds even when @xmath10 .
this is heuristically reasonable , at least in the case of the harmonic oscillator : the potential rotates the phase space , so the natural space for the initial data is of the form @xmath11 . if @xmath12 , for @xmath13 , then @xmath14 for arbitrarily small @xmath15 . for the linear equation , this can be seen via the fourier integral representation ( mehler s formula in the case of the harmonic potential ) .
the proof we present treats both linear and nonlinear cases . before going further into details ,
we clarify our assumptions .
we define the fourier transform as @xmath16 denote @xmath17 . for @xmath18
, we define @xmath19 in particular , @xmath20 is just the weighted @xmath5 space : @xmath21 we assume that the potential is smooth , real - valued and sub - quadratic : @xmath22 , and @xmath23 for all @xmath24 .
we say that @xmath2 is _ sub - linear _
if @xmath23 as soon as @xmath25 .
we say that @xmath2 is _ super - linear _ if @xmath26 is unbounded . for super - quadratic potentials , the theory must be modified .
first , if @xmath2 is super - quadratic and negative , then @xmath27 is not essentially self - adjoint on @xmath28 ( @xcite ) .
if @xmath2 is super - quadratic and positive , then even the local existence results are different .
we refer to @xcite for very interesting results in this direction .
the construction of the parametrix for the propagator of @xmath29 provided by d. fujiwara @xcite shows that @xmath30 , which is @xmath5-unitary , satisfies the local dispersion estimate : there exists @xmath31 such that @xmath32 one can infer local and global existence results for if @xmath33 when @xmath34 , under suitable assumptions on the nonlinearity @xmath3 , as proved initially by oh @xcite .
the assumption @xmath34 is actually not necessary , and one can prove the local existence results of oh in weighted sobolev spaces of the form @xmath11 thanks to strichartz estimates ( see e.g. @xcite , and @xcite where global existence results are recalled for potentials @xmath2 which are not necessarily non - negative ) . in all this paper
, @xmath35 is assumed to be a mild solution to , that is , to solve @xmath36 in proposition [ prop : local ] though , we construct a classical solution for .
when @xmath34 and latexmath:[$f(|u|^2)=\mu for the solution @xmath35 of under suitable assumptions on @xmath38 and @xmath39 , thanks to the following conservations : @xmath40 the question we ask is : what remains when we do not assume @xmath41 ?
roughly speaking , the local existence results remain when @xmath2 is sub - linear , but fail when @xmath2 is super - linear ( we prove the latter under slightly more restrictive assumptions on @xmath2 , see th .
[ theo : superlinear ] ) . note that in the above example , if we assume @xmath42 , then one can prove the existence of a global solution , with an @xmath5 regularity , as in @xcite .
our goal is to understand better the relevance of sobolev spaces with positive index , when no extra decay of the initial datum is assumed .
we recall a particular case of ( * ? ? ? * lemma 1 ) : [ lem : hj ] there exist @xmath43 and a unique solution @xmath44\times\r^n)$ ] to : @xmath45 this solution is sub - quadratic : @xmath46\times\r^n)$ ] as soon as @xmath24 .
if @xmath47 with @xmath48 , then @xmath49 this shows that in general , the above result is really local in time , due to the formation of caustics .
if @xmath50 , with @xmath51 , then we can see that caustics appear in finite time even if the potential @xmath2 is sub - linear .
[ prop : local ] let @xmath52 .
+ @xmath53 if @xmath10 ( linear equation ) , assume that @xmath54 for some @xmath18
. then has a unique solution @xmath35 such that @xmath55;h^s)$ ] , where @xmath56 and @xmath57 are given by lemma [ lem : hj ] .
+ @xmath58 for the nonlinear equation , assume that @xmath3 is smooth , @xmath59 , and that @xmath54 for some @xmath60 .
then has a unique solution @xmath35 such that @xmath55;h^s)$ ] , where @xmath56 and @xmath57 are given by lemma [ lem : hj ] .
[ prop : sublinear ] let @xmath52 , @xmath61 , and assume that @xmath2 is sub - linear and that the nonlinearity @xmath3 is of the form @xmath62 then there exists @xmath63 such that has a unique solution @xmath64;h^1)\cap l^{\frac{4\si+4}{d\si}}\([0,\tau ] ; w^{1,2\si+2}\)$ ] . + if moreover @xmath65 or @xmath66 , then this solution is global in time : @xmath67 even the local result is not a consequence of proposition [ prop : local ] : the regularity required on the initial data is not the same .
the reason is that proposition [ prop : local ] is established without dispersive or strichartz estimates , while the local existence result in proposition [ prop : sublinear ] is proven thanks to ( local in time ) strichartz estimates .
we also discuss the local cauchy problem in @xmath68 , @xmath69 , in section [ sec : hs ] .
the main point consists in showing that in the presence of a sub - linear potential , local strichartz estimates are available in sobolev and besov spaces .
we prove : [ prop : hs ] let @xmath2 be sub - linear , @xmath70 and @xmath71 .
if @xmath39 is not an integer , assume that @xmath72<2\si$ ] .
then there exist @xmath43 and a unique solution @xmath73;h^s)\cap l^\gamma ( [ 0,t];b_{\rho,2}^s)\ ] ] to , where @xmath74 we now come to the non - existence result : [ theo : superlinear ] let @xmath52 , and @xmath3 be smooth , @xmath75 .
assume that @xmath2 is super - linear , and that there exist @xmath76 and @xmath77 such that @xmath78 and @xmath79 such that @xmath80 then there exists @xmath81 such that for arbitrarily small @xmath15 and all @xmath69 , the solution @xmath82 to provided by proposition [ prop : local ] fails to be in @xmath83 .
as a potential @xmath2 , we may consider any non - trivial quadratic form , or @xmath84 , with @xmath85 , for some decomposition @xmath86 . note that no assumption is made on the growth of the nonlinearity at infinity : the above result reveals a _ geometric phenomenon _ , and not an ill - posedness result like for super - critical nonlinearities without a potential ( @xcite ) . in section [ sec : prelim ]
, we outline the proof of proposition [ prop : local ] , which is a particular case of ( * ? ? ?
* proposition 3 ) .
we establish proposition [ prop : sublinear ] in section [ sec : sublinear ] .
we extend the local theory to all the spaces @xmath68 for @xmath69 in section [ sec : hs ] , where we prove proposition [ prop : hs ] .
finally , theorem [ theo : superlinear ] is proved in section [ sec : general ] .
in this section , we outline the proof of proposition [ prop : local ] , which is a straightforward consequence of the analysis in @xcite , with the choice @xmath87 .
this will also guide us for the proof of theorem [ theo : superlinear ] .
first , lemma [ lem : hj ] is a straightforward consequence of the local hamilton - jacobi theory , gronwall lemma , and a global inversion theorem , which can be found for instance in ( * ? ? ?
* th . 1.22 ) or ( * ? ? ?
a.7.1 ) . to prepare the proof of theorem [ theo : superlinear ] , we recall some details .
let @xmath88 and @xmath89 solve @xmath90 the local hamilton - jacobi theory provides a solution to in the neighborhood of every point where @xmath91 is invertible .
the theory is global in space ( not in time , in general ) thanks to the global inversion theorem mentioned above , and to gronwall lemma .
the gradient of @xmath56 is given by @xmath92 where @xmath93 is the inverse mapping of @xmath91 .
introduce the jacobi determinant @xmath94 the global inversion theorem can be applied since there exists @xmath77 such that @xmath95\times \r^d.\ ] ] the change of unknown function @xmath96 turns into the equivalent cauchy problem : @xmath97 the major difference with is that the potential @xmath2 is no longer present in the equation .
the idea is to view the left hand side as a transport operator with velocity @xmath98 and a renormalization factor along the characteristics , @xmath99 .
we can then reduce the problem of existence of solutions of , to the existence of _ a priori _ estimates , thanks to a mollification procedure . since we seek @xmath100;h^s)$ ] , we note that the term @xmath101 on the right hand side is skew - symmetric , and has no contribution in the energy estimates . to take advantage of this property ,
we do not rewrite along the characteristics , but notice that from lemma [ lem : hj ] , @xmath102 . for the convective term , we use lemma [ lem : hj ] , and an integration by parts : if @xmath103 is such that @xmath104 , we write @xmath105 if @xmath106 is not an integer , we can use interpolation .
proposition [ prop : local ] follows easily , since @xmath60 ensures that @xmath68 is an algebra . [ rk : op ] let @xmath107 $ ] be a compact time interval .
the approach of @xcite recalled above shows that the map @xmath108 , where @xmath109 sends @xmath110 to @xmath111 continuously : @xmath112 where @xmath113 depends only on @xmath114 and @xmath115 .
to prove the first part of proposition [ prop : sublinear ] , the idea is to keep the same proof as without potential .
the gradient does not commute with @xmath27 , but we have : @xmath116 the new term is @xmath117 , that is , @xmath35 multiplied by a bounded term .
recall that @xmath118 .
we show that for @xmath119 sufficiently small , there exists @xmath35 such that : @xmath120 we see that @xmath121 recall that @xmath122 is schrdinger - admissible in @xmath123 if @xmath124 it follows from @xcite that strichartz estimates are available for @xmath125 ( see e.g. @xcite ) : for all admissible pairs @xmath122 , @xmath126 and @xmath127 , there exist @xmath128 and @xmath129 such that for any compact interval @xmath130 and any @xmath131 , @xmath132 , @xmath133 where @xmath134 stands for the hlder conjugate exponent of @xmath135 .
note that the powers of @xmath136 on the right hand sides are sharp in general , for @xmath27 may have eigenvalues . for @xmath122 an admissible pair , define @xmath137 introduce the following lebesgue exponents : @xmath138 then @xmath122 is the ( admissible ) pair of the proposition , and @xmath139 for @xmath140 and any pair @xmath141 , we use the notation @xmath142;l^b ) } .\ ] ] we first prove that there exists @xmath119 such that the set @xmath143 ) ; \
\|u\|_{l^\infty_\tau l^2}\le 2\|a_0\|_{l^2},\ \|u\|_{l^q_\tau l^r}\le 2c_{r}\|a_0\|_{l^2 } \ , \\ & \|\nabla u\|_{l^\infty_\tau l^2}\le 2\|\nabla a_0\|_{l^2}\ , \quad \|\nabla u\|_{l^q_\tau l^r}\le 2 c_{r}\|\nabla a_0\|_{l^2}\big\ } \end{aligned}\ ] ] is stable under the map @xmath144 , where @xmath128 is the constant of the homogeneous strichartz inequality . then choosing @xmath145 even smaller , @xmath144 is a contraction on @xmath146;l^r)$ ] .
let @xmath147 . for @xmath148 , yields : @xmath149 sobolev embedding yields : @xmath150 it follows that @xmath151 the same computations yield : @xmath152 similarly , @xmath153 and @xmath154 therefore @xmath144 leaves @xmath155 stable for @xmath156 to complete the proof of the first part of the proposition , it is enough to prove contraction for small @xmath145 in the weaker metric @xmath146;l^r)$ ] .
we have : @xmath157 as above , we have the estimate @xmath158 therefore , contraction follows for @xmath145 sufficiently small , according to . if @xmath2 is sub - linear and unbounded , then the energy @xmath159 may not be defined initially , if we simply require @xmath160 . to complete the proof of proposition [ prop : sublinear ] ,
the idea is to notice that the time derivative of the `` bad '' term in the energy is controlled by the @xmath0 norm of the solution .
we present the computations at a formal level only , and refer to @xcite for a justification method which uses the multiplication by gaussians .
we have @xmath161 we infer , thanks to the conservation of mass : @xmath162 when @xmath163 , this yields the estimate @xmath164 hence @xmath165 grows at most exponentially .
if @xmath65 and @xmath166 , gagliardo nirenberg inequality and the conservation of mass yield : @xmath167 using young inequality @xmath168 and choosing @xmath169 sufficiently small , we conclude as before .
this completes the proof of proposition [ prop : sublinear ] .
when @xmath54 with @xmath69 not necessarily equal to one , and @xmath2 is sub - linear , it is still possible to establish a local in time theory . without potential ,
@xmath171 , proposition [ prop : hs ] was proved by t. cazenave and f. weissler ( * ? ? ?
* theorem 1.1 , ( i)(ii ) ) . as in this paper
, we shall not define besov spaces by using a dyadic decomposition , but rather use their characterization in terms of interpolation between sobolev spaces .
we first recall the argument when @xmath171 , and then show how it can be adapted to infer proposition [ prop : hs ] .
the idea is to apply a fixed point argument , as in section [ sec : local ] .
however , when @xmath172 is not an integer , it becomes delicate to estimate the @xmath170 norm of the nonlinearity .
this is why in @xcite , the authors work in besov spaces .
when @xmath106 is an integer , the above result can be refined .
we shall not recall this aspect more precisely , and simply refer to @xcite .
the proof proceeds in three steps .
the authors first establish strichartz estimates for the free group @xmath173 in ( homogeneous ) besov spaces ( * ? ? ?
next , they prove estimates for the nonlinear term , in homogeneous besov spaces as well ( * ? ?
finally , these tools , along with strichartz estimates , make it possible to apply a fixed point argument , and prove proposition [ prop : hs ] when @xmath171
. denote @xmath174 the first step yields , for @xmath69 , and @xmath122 , @xmath175 admissible pairs : @xmath176 where @xmath129 does not depend on the time interval @xmath130 .
next , under the assumptions of proposition [ prop : hs ] , we have @xmath177 proposition [ prop : hs ] follows from , , hlder s inequality and a fixed point argument . note that and still hold if we replace homogeneous besov spaces with inhomogeneous ones .
this remark simplifies the generalization to the case when @xmath2 is sub - linear .
we show that when @xmath2 is sub - linear , still holds , up to two modifications : * the strichartz inequalities hold on _ finite _ time intervals only . *
we replace the homogeneous besov spaces with inhomogeneous ones .
the first point is unavoidable , as recalled in section [ sec : local ] .
since we shall prove a local in time result , in the rest of this section we consider time intervals of length at most one .
the second point is here to consider pseudo - differential operators with smooth symbols which do not contain @xmath178-variable .
if @xmath179 is a pseudo - differential operator with smooth symbol , we have : @xmath180\varphi & = -i\int_0^t u(t - s)[p , v]u(s)\varphi ds= -i { \mathcal i}(t ) \ ( [ p , v]u(\cdot)\varphi\),\\ [ p,{\mathcal i}(t)]f & = -i\int_0^t u(t - s)[p , v]{\mathcal i}(s)f ds =- i { \mathcal i}(t ) \ ( [ p , v]{\mathcal i}(\cdot)\varphi\).\end{aligned}\ ] ] first , assume @xmath181 .
for @xmath130 a time interval with @xmath182 , yields : @xmath183u(\cdot)\varphi\ ) \right\|_{l^q(i;l^r)}\\ & \lesssim \|p\varphi\|_{l^2 } + \left\| [ p , v]u(\cdot)\varphi \right\|_{l^1(i;l^2)}\\ & \lesssim \|p\varphi\|_{l^2 } + \left\| [ p , v]u(\cdot)\varphi \right\|_{l^\infty(i;l^2)}.\end{aligned}\ ] ] similarly , @xmath184{\mathcal i}(\cdot)f \right\|_{l^\infty(i;l^2)}.\ ] ] for @xmath69 , let @xmath185 . by (
2 ) ( see also @xcite ) , we know that if in addition @xmath186 , then @xmath187 $ ] is bounded from @xmath5 to @xmath5 , with norm controlled by @xmath188 for some universal constant @xmath113 . we infer , when @xmath186 , @xmath189 where we have used strichartz estimates .
this means : @xmath190 similarly , when @xmath186 , @xmath191 for @xmath192 , replace @xmath193 with the family @xmath194 , where @xmath195 $ ] .
reasoning as above , we see that since @xmath196 for all @xmath197 , and hold for all @xmath69 . interpolating ( as in @xcite , up to replacing homogeneous spaces by their inhomogeneous counterparts ) ,
we conclude : @xmath198 where the constants @xmath128 and @xmath129 do not depend on @xmath130 , provided that @xmath182 . * conclusion . *
since holds with homogeneous besov spaces replaced by their inhomogeneous counterparts , the fixed point argument used in @xcite can be transported here .
this completes the proof of proposition [ prop : hs ] .
the idea is to resume the approach of weakly nonlinear geometric optics recalled in section [ sec : prelim ] .
we consider an intermediary function defined by leaving out the term @xmath199 in : without this term , is an ordinary differential equation along the characteristics of the transport operator with velocity @xmath98 ( _ i.e. _ the bicharacteristics associated to @xmath27 ) .
recall that @xmath200 solves , and define @xmath201 as the solution on @xmath202 $ ] to : @xmath203 to see that @xmath201 solves an ordinary differential equation along the rays of geometric optics ( the projections of the hamilton flow on the physical space ) , introduce @xmath204 where @xmath88 is given by and the jacobi determinant is defined by .
this change of unknown function makes sense for @xmath205 $ ] , where @xmath91 is a global diffeomorphism .
then is equivalent to @xmath206 since in theorem [ theo : superlinear ] , we assume that @xmath3 is real - valued , we note that @xmath207 so that is just a _ linear _ ordinary differential equation : @xmath208 we infer @xmath209 the main observation is that implies that @xmath210;h^s(\r^d))$ ] for all @xmath18 .
let @xmath211 : for every @xmath212 $ ] , @xmath213 . for @xmath214 , @xmath215 solves : @xmath216 notice that the fundamental theorem of calculus yields : @xmath217 where @xmath218 . in particular , we know that @xmath219 reasoning as in remark [ rk : op ] , we see that : @xmath220;l^2)}\le c\(1 + \|x\delta b\|_{l^1([0,t];l^2 ) } \ ) .\ ] ] we must make sure that the last term is , or can be chosen , finite
. we shall demand @xmath221;l^2)$ ] . in view of ,
this requirement is met as soon as @xmath222 is such that @xmath223 .
we then have : @xmath224;h^s ) \ \forall s\ge 0 , \text { and } xr \in c([0,t];l^2 ) .
\end{aligned}\ ] ] we infer from and lemma [ lem : hj ] that @xmath231 from and , we also have @xmath232 we claim that @xmath233 indeed , we have from , @xmath234 and follows from gronwall lemma .
we infer that for @xmath15 sufficiently small , @xmath235 and therefore , @xmath236 then yields @xmath237 lemma [ lem : dl ] follows by integration in time .
consider @xmath239 as is easily checked , @xmath4 meets the requirements of the first line of .
denote @xmath240 obviously , @xmath241 .
we see from and that @xmath242 for @xmath15 sufficiently small , under the assumptions of theorem [ theo : superlinear ] . on the other hand , @xmath243 for all
@xmath205 $ ] , hence @xmath244 for @xmath245 . to complete the proof of theorem [ theo : superlinear ] , we now just have to see that the same holds if we replace @xmath9 with @xmath68 for @xmath181 .
we use the following characterization of @xmath68 ( see e.g. @xcite ) : for @xmath131 and @xmath181 , @xmath246 since @xmath247 for all @xmath205 $ ] , we shall prove that @xmath248 for @xmath249 sufficiently small . let @xmath181 . we prove that for @xmath245 , @xmath250 to apply a fractional leibnitz rule , write @xmath251 in view of the inequality @xmath252 , we have : @xmath253 we can leave out the last term , since @xmath254 for @xmath205 $ ] : @xmath255.\ ] ] we now want to prove @xmath256 lemma [ lem : hj ] yields : @xmath257\times\r^d\)^{d^2}\quad ; \quad \nabla^2 \phi_{{\rm eik } \mid t=0}=0.\ ] ] therefore , @xmath258 we infer : @xmath259 and @xmath260 the second term is @xmath261 . using the estimate @xmath262 again
, we see that the integral corresponding to the second term is finite , and can be left out . to prove that @xmath263 we can localize @xmath264 in a small conic neighborhood of @xmath265 : @xmath266 , yields : @xmath267 introduce a conic localization for @xmath178 close to @xmath268 , excluding the origin : @xmath269 change the variable in the @xmath264-integral : for @xmath249 and @xmath270 sufficiently small , and @xmath271 , set @xmath272 this change of variable is admissible , from and .
for @xmath273 , we have : @xmath274 the assumption , the expression of @xmath201 and the choice for @xmath4 then show that for @xmath245 , @xmath275 .
this completes the proof of theorem [ theo : superlinear ] .
n. burq , p. grard , and n. tzvetkov , _ multilinear eigenfunction estimates and global existence for the three dimensional nonlinear schrdinger equations _ ,
sci . cole norm .
( 4 ) * 38 * ( 2005 ) , no . 2 , 255301 .
chemin , _ perfect incompressible fluids _ , oxford lecture series in mathematics and its applications , vol .
14 , the clarendon press oxford university press , new york , 1998 , translated from the 1995 french original by i. gallagher and d. iftimie .
n. dunford and j. t. schwartz , _ linear operators .
part ii : spectral theory .
self adjoint operators in hilbert space _
, with the assistance of william g. bade and robert g. bartle , interscience publishers john wiley & sons new york - london , 1963 .
j. t. schwartz , _ nonlinear functional analysis _
, gordon and breach science publishers , new york , 1969 , notes by h. fattorini , r. nirenberg and h. porta , with an additional chapter by hermann karcher , notes on mathematics and its applications . | we consider the cauchy problem for nonlinear schrdinger equations in the presence of a smooth , possibly unbounded , potential .
no assumption is made on the sign of the potential .
if the potential grows at most linearly at infinity , we construct solutions in sobolev spaces ( without weight ) , locally in time . under some natural assumptions , we prove that the @xmath0-solutions are global in time . on the other hand ,
if the potential has a super - linear growth , then the sobolev regularity of positive order is lost instantly , not matter how large it is , unless the initial datum decays sufficiently fast at infinity . |
schnirelman s theorem @xcite as well as berry s physical reasoning predict that quantum energy eigenfunctions ( wigner or husimi representation @xcite ) of systems whose classical counterpart is chaotic are uniformly distributed on the energy shell .
heller , however , has shown that there exist quantum eigenfunctions which are strongly localized ( scarred ) on hyperbolic periodic orbits @xcite . at present , the general opinion is that scars are exceptional , while the majority consits of uniformly ( phase - space ) distributed eigenstates @xcite . on the other hand
, it is well known that there also exist weakly localized eigenfunctions , which continuously fill the `` gap '' between uniformly distributed and strongly localized eigenfunctions .
conveniently , instead of single exceptions we here consider localization properties of the whole set of eigenfunctions of the system ; restricting our studies on finite dimensional hilbert spaces . to that purpose ,
we introduce squared @xmath0 norms of husimi eigenfunctions as a measure for phase - space localization @xcite .
as it will be shown in the sequel this measure proves amenable to semiclassical considerations .
the classical dynamics of chaotic systems can be described by the time evolution of phase - space density functions .
the corresponding propagator is the frobenius - perron operator @xcite , where the poles of its resolvent are called ruelle - pollicott resonances which coincide with decay rates of classical correlation functions @xcite .
the quantum - classical correspondence , in particular the influence of ruelle - pollicott resonances on the quantum energy spectrum , is still a point of interest @xcite .
we show that phase - space overlaps of energy eigenstates turn out lorentz distributed with respect to the differences of their eigenenergies .
the lorentzians are determined by ruelle - pollicott resonances . in other words ,
the probability that two eigenstates strongly overlap becomes large if the difference of their eigenenergies coincides with the position where the lorentzian is peaked . on the other hand ,
if a pair of eigenstates strongly overlaps ( much more than random - matrix theory predicts ) each of them must be localized , i.e. scarred , in the same phase - space regions .
we here consider systems with a compact two - dimensional phase space , in particular the unit sphere , whereby the hilbert - space dimension of the quantum counterpart becomes finite .
periodically driving destroyes integrability in general .
moreover , a stroboscopic description leads to a hamilton map or a floquet operator in the classical or quantum case , respectively .
a well known representative of such a dynamics is the kicked top @xcite .
the dynamics of the kicked top is described by a stroboscopic map of an angular momentum vector whose length is conserved , for such a dynamics the phase spaces is the unit sphere . the classical time evolution is usually described in the `` hamilton picture '' , whereby the stroboscopic consideration leads to a hamilton map which describes a trajectory after each period , @xmath1 here the primes denote the final position @xmath2 and momentum @xmath3 coordinates . on the sphere
the canonical phase - space coordinates are given by the azimuthal and polar angle as @xmath4 and @xmath5 .
the quantum time evolution in the schrdinger picture is generated by a floquet operator @xmath6 which is built by the components of an angular momentum operator @xmath7 ; we choose it as a sequence of rotations about the @xmath8 and @xmath9-axis followed by a nonlinear torsion about the @xmath9-axis , @xmath10 where @xmath11 is called the torsion strength and @xmath12 and @xmath13 are rotation angles .
the dimension of the quantum hilbert space is @xmath14 , where @xmath15 is the quantum angular momentum formally replacing the inverse of planck s constant , @xmath16 ( @xmath17 in this article ) . since @xmath6 is unitary ,
it has @xmath18 orthogonal eigenstates with unimodular eigenvalues characterized by eigenphases ( quasi - eigenenergies ) as @xmath19 .
the classical counterpart has in general a mixed phase space . by the choice of the parameters as @xmath20 and @xmath21 the dynamics
becomes strongly chaotic .
for @xmath22 which we use for numerical results stable island are not resolved by the planck cell of size @xmath23 , whereby the dynamics looks , from a quantum point of view , effectively hyperbolic . in order to compare the results of the kicked top with those of random - matrix theory ( rmt ) we here discuss the symmetries of the system .
the dynamics proves invariant under nonconventional time reversal . in terms of random - matrix theory
the floquet matrix belongs to the circular orthogonal ensemble ( coe ) , where the coefficients of the eigenvectors can be chosen real in a suitable basis @xcite .
we here expand the floquet operator in the basis of eigenstates of the @xmath9-component of the angular momentum operator , @xmath24 . by a unitary transformation given by a simple rotation , @xmath25 ,
the floquet matrix becomes symmetric , @xmath26 , where the eigenvectors become real . here
@xmath27 denotes transposition .
this is an important property , since the @xmath0 norm of a husimi function which we will use as a measure for phase - space localization is invariant under rotations .
therefore , eigenvectors of the kicked top must be compared to real random vectors .
the rotation is of form @xmath28 , whereby the transformed floquet matrix becomes @xmath29 after commutation of the rotation @xmath30 with the torsion the product @xmath31 is a rotation about the @xmath32-axis . using the latter relation
the transformed floquet matrix finally becomes @xmath33 which is obviously symmetric .
another way to describe classical time evolution is the `` liouville picture '' , as the propagation of density in phase space .
the corresponding propagator , the frobenius - perron operator @xmath34 , is defined through the hamilton map as @xmath35 where @xmath36 denotes an arbitrary phase - space density function . note that the hamilton map is invertible and area preserving .
an expectation value of a classical observable is given by the phase - space integral @xmath37 where we have introduced the dirac notation . to avoid confusion with quantum wave functions we here use round brackets .
note that this notation is generally to read as a linear functional , where the density function @xmath38 belongs to the banach space @xmath39 and the observable @xmath40 to the dual space @xmath41 .
furthermore , we suppose that both functions are real , otherwise @xmath40 is to complex conjugate in the integral notation ( [ average ] ) . for classically chaotic systems ( we assume purely hyperbolic dynamics ) correlations of observables
decay exponentially in time . due to ergodicity the time correlation can be written by a phase - space integral , @xmath42 where the time dependence of the observables must be read as @xmath43 . here
@xmath44 denotes the stationary ( invariant ) density with @xmath45 , i.e. the constant on the sphere .
the associated stationary eigenvalue is @xmath46 and ensures that no probability gets lost , i.e. it preserves the @xmath39 norm of a density function .
we may replace the observable @xmath47 by an initial density function @xmath38 and further we assume that @xmath48 , then the correlation function can be written in terms of the frobenius - perron operator .
finally , we introduce the ruelle - pollicott resonances @xmath49 which are to identify as decay rates , @xmath50 where the @xmath51 denote the coefficients of the resonance expansion .
we here assume the simplest case that resonances appear with multiplicity @xmath46 , otherwise the spectral decomposition of the frobenius - perron operator would be given by a so - called jordan block structure , whereby the expansion on the rhs would become more complicate .
it should be remarked that the decay rate is precisely given by the logarithm of @xmath49 which is more convenient to consider in continuous - time dynamics .
while the @xmath49 are located inside the unit circle on the complex plane , the logarithm of @xmath49 are customarily chosen to be in the lower half plane . due to the fact that the frobenius - perron operator preserves the positivity of density functions the resonances are real or appear as complex pairs .
the trace of the frobenius - perron operator defined through @xmath52 , i.e. setting image and original points in ( [ fpoper ] ) equal , becomes @xmath53 in terms of the resonances .
we here have separated the stationary eigenvalue @xmath46 from the summation of the resonances .
we should carefully distinguish between forward and backward time evolution , since we do nt expect an increase of correlations for the backward time propagation .
it has been shown that the backward time frobenius - perron operator has the same resonances .
ruelle - pollicott resonances are defined as the poles of the resolvent of @xmath34 , @xmath54 the corresponding `` eigenfunctions '' are not square - integrable functions like those of the quantum propagator , but distributions .
it is known that unstable manifolds of periodic orbits function as supports of these singular eigenfunctions . for the backward time evolution
stable and unstable manifolds exchange their role .
supported by arguments of pertubation theory , weber et al . @xcite
have shown that classical ruelle - pollicott resonances of the frobenius - perron operator can be found by investigating the propagator restricted on different phase - space resolutions .
moreover , one finds approximate eigenfunctions which scar along unstable manifolds
. this important result will be useful for the comparison of classical and quantum eigenfunctions .
we restrict our considerations on the hilbert space of square integrable functions @xmath0 . for the system considered the phase space is the unit sphere , where it becomes convenient to use the basis of spherical harmonics .
the frobenius - perron operator becomes an infinite unitary matrix whose unimodular spectrum can be separated into a discrete part for integrable components and into a continuous part for hyperbolic components of the dynamics .
truncation to an @xmath55 matrix corresponding to a restricted phase - space resolution destroys unitarity .
the spectrum becomes discrete and the eigenvalues are inside or on the unit circle .
as @xmath56 increases some eigenvalues prove @xmath56-independent ; these are said to be stabilized .
stabilized eigenvalues reflect spectral properties of the frobenius - perron operator .
they are ( almost ) unimodular for integrable components or stable islands .
in contrast , eigenvalues which are stabilized inside the unit circle reflect ruelle - pollicott resonances .
non - stabilized eigenvalues have typically smaller moduli than the stabilized ones .
they change their positions as @xmath56 increases till they reach positions of resonances where they can settle for good .
we also know that the eigenfunctions corresponding to stabilized eigenvalues are localized either on tori for unimodular eigenvalues or around unstable manifolds for non - unimodular eigenvalues .
we expect that the latter eigenfunctions converge weakly to singular resonance `` eigenfunctions '' .
we will call them approximate resonance eigenfunctions in the sequel . fig .
[ resdens ] ( a ) shows the eigenvalues of the truncated frobenius - perron matrix in the complex plane . here
the dimension is @xmath57 , where @xmath58 is the maximal total angular momentum of the spherical harmonics in which the density functions are expanded . in ( b ) we see a grey - scale shaded plot of the eigenvalue density calculated from truncated matrices ( @xmath59 ) .
dark spots corresponding to large amplitudes of the density indicate resonance positions , since resolution independent eigenvalues highly increase the density through accumulation . a comparison of both shows that some eigenvalues in ( a ) reflect resonance positions ( see also tab
. 1 ) . in fig .
[ ef4 ] ( a ) the modulus of the approximate eigenfunction is plotted in phase space , where the dark regions belong to large moduli of the complex - valued function .
the comparison with the unstable manifolds ( b ) of a weakly unstable period-4 orbit shows that the approximate eigenfunction scars along these manifolds . in ( c ) and
( d ) we see the scarring of the eigenfunction of backward time propagation ( same resonance as in ( a ) ) along the stable manifolds .
to represent quantum operators in a way close the corresponding classical observables it is convenient to start from coherent states .
for the @xmath60 group coherent states can be generated by a rotation of the state @xmath61 as @xmath62 . in the @xmath63 basis
these coherent states are given by @xmath64 since the set of the coherent states is overcomplete , there are several ways describing quantum operators with coherent states @xcite .
on the one hand , one can use the so - called @xmath65 function , defined as the weight of coherent - state projectors in the continuous mixture @xmath66 the integral is over the unit sphere , where @xmath67 . on the other hand
, we have the so - called @xmath68 function , the coherent state expectation value of an operator , @xmath69 .
it is important that @xmath68 is a smooth function on phase space , while @xmath65 can strongly oscillate , particularly in the shortest wave lengths .
however , in contrast to the coherent states of weyl groups , @xmath65 functions always exist .
both functions can be expanded in terms of spherical harmonics @xcite , @xmath70 where the summations break off at the total angular momentum @xmath71 .
the hilbert space of phase - space functions must not be confused with the hilbert space of quantum wave functions ; we therefore use round brackets for the scalar product already introduced in ( [ average ] ) , @xmath72 .
although the @xmath65 functions tend to oscillate more strongly than the @xmath68 functions , the expansion coefficients of the @xmath65 and @xmath68 functions corresponding to the same operator converge to one another in the classical limit , @xmath73 for fixed @xmath74 .
it is easy to see that the trace of an operator product can be written as the scalar product of @xmath65 and @xmath68 as @xmath75 . in particular ,
if the set @xmath76 of wave functions form an orthogonal basis of the quantum hilbert space , then the @xmath65 and @xmath68 functions of ket - bras @xmath77 and @xmath78 generate biorthonormal sets in the hilbert space of phase - space functions @xmath79 the @xmath68 function of a density operator @xmath80 is also called husimi function .
if the density operator is a projector of form @xmath81 the corresponding husimi function is a phase - space representation of a quantum wave function .
we call the @xmath78 husimi eigenfunction if the @xmath82 denote floquet eigenstates .
the latter notation becomes obvious in the next section .
we further distinguish between diagonal husimi eigenfunctions @xmath83 and skew ones @xmath84 with @xmath85 .
the husimi propagator @xmath86 is defined through the time evolution of a quantum density operator @xmath87 as @xmath88 using the floquet eigenstates , @xmath89 , the husimi eigenfunctions are easily calculated as @xmath90 .
the husimi propagator thus has @xmath91 unimodular eigenvalues whose phases are differences of the floquet eigenphases .
there is an @xmath18-fold degeneracy of the eigenvalue @xmath46 corresponding to the diagonal husimi eigenfunctions @xmath83 which are real and normalized as @xmath92 .
all other ( skew ) husimi eigenfunctions with @xmath85 are complex and their phase space - integral vanishes . in basis of spherical harmonics
the husimi propagator becomes an @xmath93 matrix ( husimi matrix ) .
the diagonal representation of the husimi propagator is simply given as @xmath94 the husimi spectral density is identified as the density - density correlation function with respect to the floquet eigenphases , @xmath95 where @xmath96 . some authors prefer the normalized density @xmath97 .
the squared @xmath0 norm of a diagonal husimi eigenfunction , @xmath98 is the inverse participation ratio ( ipr ) with respect to coherent states ( phase - space distribution ) .
it becomes large if the husimi function is strongly localized in phase space , say scarred on periodic orbits . on the other hand ,
the squared @xmath0 norm of a skew husimi eigenfunction can be understood as the overlap of two diagonal husimi functions on phase space , @xmath99 from the schwarz inequality , @xmath100 it becomes obvious that for large values of @xmath101 both diagonal husimi eigenfunctions must be localized in the same phase - space regions .
we illustrate two examples .
a constant function on the sphere is of course a uniformly distributed function .
but note that it is not a husimi eigenfunction , since the corresponding density operator is of form @xmath102 .
using the normalization @xmath103 the husimi function becomes @xmath104 .
for the squared @xmath0 norm we find @xmath105 for uniform function . in comparison ,
the random - matrix averaged squared @xmath0 norms of diagonal husimi eigenfunctions is larger by a factor 2 ( see sec . [ rmt ] ) which can be explained by quantum fluctuations .
a most strongly localized husimi function , in contrast , corresponds to a density oparator which is a coherent - state projector . due to the invariance under rotations
the @xmath0 norm is the same for all coherent - state husimi functions . using the coherent - state projector @xmath106 and ( [ exaktl2 ] )
one easily finds for the squared @xmath0 norm @xmath107 for strongly localized husimi function .
it becomes obvious that squared @xmath0 norms of most strongly localized and uniformly distributed husimi functions differ by a factor of order @xmath18 .
we here show the calculation of the @xmath0 norm from vector coefficients in the @xmath63 basis ( @xmath108 ) . introducing the decomposition of unity in terms of the angular momentum states , @xmath109 ,
the @xmath0 norm of a skew husimi eigenfunction becomes a four - fold sum , @xmath110 from the definition ( [ cs ] ) of the coherent states we find , using the relations @xmath111 for @xmath112 , @xmath113 after a few further steps the phase - space integral @xmath114 leads to a kronecker @xmath115 and to a beta function @xcite for the @xmath116- and @xmath117-dependent part , respectively .
we thus obtain for the coefficients of the summation ( [ summation ] ) @xmath118 we make use of the argument of the kronecker @xmath115 and get for the @xmath0 norm @xmath119 the latter formular will be used for the calculation of husimi @xmath0 norms of random vectors in the next section .
in order to compare the results of the kicked top with those of random - matrix theory we now consider diagonal husimi eigenfunctions and replace the coefficients by real or complex random numbers due to the orthogonal ( coe ) or unitary ensemble ( cue ) , respectively @xcite . it is easy to see from ( [ l2normdiag],[l2normskew ] ) that the @xmath0 norm of a husimi function is invariant under rotations . as it has been shown in sec .
[ kickedtop ] the eigenvectors of the kicked top must be compared to real random vectors .
the only correlation between the coefficients is the normalization of the vector , @xmath120 ( we drop the upper index in the following ) .
therefore we neglegt all terms containing random phases and keep the contributing terms with @xmath121 or @xmath122 . by the choice of real coefficients one has the further possibility @xmath123 .
we may abbreviate the coefficients of the summation ( [ sumcoeff ] ) as @xmath124 for a moment to find out the contributing terms , @xmath125 \left\langle|c_{m}|^2|c_n|^2\right\rangle + \delta_{m_im_k}^3\left\langle|c_m|^4\right\rangle\!\bigg ) \ , .
\nonumber\\\end{aligned}\ ] ] the diagonal part @xmath126 is separated , since there appears the average of @xmath127 . due to the symmetry of @xmath124 the first two terms in the bracket
give the same contribution ; to that end , one executes the summations over @xmath128 and @xmath129 . the contribution coming from real coefficients @xmath130 vanishes ; here one summates over @xmath131 and @xmath129 .
finally , one finds for the averaged @xmath0 norm @xmath132 \ , .
\label{sumbinom}\end{aligned}\ ] ] next we need the averages of products , @xmath133 the averages are easily calculated from the probability distribution given in @xcite .
it should be remarked here that semiclassical corrections are of next to leading order of @xmath134 .
therefore we do not neglect this order in our rmt results .
let us consider first the cue .
since @xmath135 , we can complete the summations over the diagonal @xmath136 and off - diagonal @xmath137 part .
thus we have for the cue @xmath138 it might not be easy to see that this results to @xmath139 . to that end we present an alternative way to calculate the averaged @xmath0 norm which is unfortunately not applicable for the coe @xcite .
an arbitrary complex random vector can be written as @xmath140 , where @xmath141 is a unitary random matrix .
a coherent state is generated through a rotation operator as @xmath62 .
we thus have @xmath142 .
the product @xmath143 defines a new unitary random matrix , where the haar measure keeps unchanged @xmath144 .
therefore the averaged @xmath0 norm becomes @xmath145 note that for the coe , in contrast , one deals with real vectors which generally become complex after rotating .
for the coe we again complete the diagonal and off - diagonal summations in ( [ sumbinom ] ) , whereby a further contribution remains , because of the factor @xmath146 of the fourth moment ( [ avipr ] ) , @xmath147 the latter equation can be calculated as follows @xcite .
we rewrite @xmath148 with @xmath149 .
we now show that @xmath150 the generating function of @xmath151 is @xmath152 which is easily seen from taylor expansion , @xmath153 squaring the generating function gives @xmath154 where @xmath155 . comparing the latter equation with the geometric series @xmath156 finishes the proof .
the aforementioned contribution ( [ forcontrib ] ) becomes more familiar by the approximation @xmath157 the averaged @xmath0 norms finally become @xmath158 as an interesting point to remark here the coe eigenstates are somewhat more localized than the cue eigenstates , since their averaged @xmath0 norm contains a further contribution of order @xmath159 .
but in contrast to the fourth moment ( [ avipr ] ) - it might be understood as the averaged ipr with respect to the @xmath63 basis - the difference vanishes in the classical limit @xmath160 .
however , for both ensembles the averaged squared @xmath0 norms are roughly twice larger than for a constant distribution on the sphere . for the skew husimi eigenfunctions
the averaged @xmath0 norms can be calculated from the relation @xmath161 which leads to @xmath162 introducing the averages of the diagonal ones we find for cue @xmath163 for the coe one obtaines the same result up to an order of @xmath164 , @xmath165
it has been shown in @xcite that quantum quasiprobability propagation looks classical if phase - space resolution is blurred such that planck cells are far from being resolved . as a result ,
if we truncate the husimi matrix to coarse phase - space resolution , then it becomes almost equal to the truncated frobenius - perron operator , @xmath166 , where @xmath167 denotes the truncation projector restricting the `` classical '' hilbert space dimension to @xmath57 . in the following we choose @xmath168 , and
thus have @xmath169 .
again we consider the classical propagator .
let @xmath170 be an approximate resonance eigenfunction with stabilized eigenvalue @xmath171 of @xmath172 .
since the stabilized eigenvalues reflect spectral properties of the non - truncated frobenius - perron operator , @xmath173 must be a stabilized eigenvalue of @xmath174 , at least if @xmath56 is chosen large enough .
this property does not hold for non - stabilized eigenvalues , since @xmath175 . choosing the approximate resonance eigenfunction @xmath0 normalized
, the return probability becomes @xmath176 for small @xmath177 ( see fig .
[ lambda1r ] ) .
note that the latter property makes sense only if we consider approximate resonance eigenfunctions , because the singular resonance eigenfunctions are not square integrable .
[ lambda1r ] shows the return probability of the approximate eigenfunction from stabilized eigenvalue @xmath178 ( no 4 in tab .
[ restab ] ) .
as well the moduli as the phases coincide with @xmath179 for about 20 iterations . beyond this time quantum fluctuations become visible . since @xmath169
, we expect @xmath180 for small @xmath177 .
now we can replace the husimi propagator by its diagonal representation ( [ diagrep ] ) , @xmath181 the return probability becomes a double sum of overlaps of quantum and classical eigenfunctions . for large @xmath18 the husimi eigenphases are quite dense in the interval @xmath182 . outside an interval around the husimi eigenphase @xmath183 wherein level repulsion of floquet eigenphases
become perceptible the spectral density of differences @xmath184 is almost constant , @xmath185 .
it is convenient to replace the sum by a continuous integral over the husimi eigenphases , where the overlaps can be replaced by a continuous function as a smoothed distribution of overlaps , @xmath186 here @xmath187 denotes a smoothing interval , @xmath188 in the classical limit we let first @xmath189 and then @xmath190 . in this limit
the foregoing result becomes valid for all @xmath177 .
the smoothed overlaps are given by the inverse fourier transform of @xmath173 as @xmath191 which is a periodic `` lorentzian '' displaced by the phase of @xmath171 and of width @xmath192 @xcite ( see fig .
[ lambda1f ] ) .
the smoothing is done by a convolution with a sinc function which naturally results from a truncated fourier transform . here
the time restriction is @xmath193 , where @xmath194 corresponding to the validity of the return probability of the approximate eigenfunction ( fig .
[ lambda1f ] ) . for simplification the integral is restricted on the interval between the first zeros of the sinc function , @xmath195 a simple argument connects quantum scars with ruelle - pollicott resonances .
classical resonance eigenfunctions are scarred along unstable manifolds ( or stable manifolds for backward time propagation ) .
quantum eigenfunctions which strongly overlap with resonance eigenfunctions have to be scarred as well .
this is an important result , because it explains scarring of quantum eigenfunctions not only on periodic orbits , but also along stable and unstable manifolds .
this has been observed for the kicked top @xcite .
for instance , we consider the skew husimi eigenfunction which shows the largest overlap with the classical approximate eigenfunction considered before .
the corresponding diagonal husimi eigenfunctions ( no 11 and no 30 in tab .
[ hustab ] ) plotted in fig .
[ husp4 ] are scarred in the same phase - space regions , where the the approximate resonance eigenfunctions are scarred ( fig .
[ ef4 ] ) , while the difference of their floquet eigenphases ( @xmath196 ) is close to the phase of the resonance ( @xmath197 ) .
in the foregoing section a qualitative explenation of the connection between resonances and quantum eigenfunctions was given , but now we are interested in more quantitative results . to that end , we consider transition rates of coherent states in the classical limit @xmath198 the latter relation becomes obvious if one suggests that coherent states are wave functions most strongly localized on phase - space points .
we now consider the return probability and integrate over the phase space , @xmath199 where the rhs is the trace of the frobenius - perron operator @xcite .
we remark here that the integral on the rhs leads to a sum of contributions from periodic orbits , which is an important connection between scars on periodic orbits and the results of our paper .
on the one hand , periodic orbits , in particular the weakly unstable ones , contribute to the trace of the frobenius - perron operator , i.e. influence the resonances . on the other hand , scars typically appear around weakly unstable periodic orbits . on the lhs of ( [ trace ] )
we introduce the diagonal representation of the floquet operator and get @xmath200 fourier transformation of the latter expression leads to a sum of @xmath115 functions weighted by @xmath0 norms , @xmath201 due to the arguments of the foregoing section we expect that for finite @xmath18 the relation ( [ semicl ] ) is valid for finite times @xmath202
. validity of semiclassical methods is guarantied for times up to the ehrenfest time , where the number of fixpoints coincides with the number of planck cells .
thus we identify @xmath203 as ehrenfest time .
the truncated fourier transform leads to a sum of smoothed @xmath115 functions ( [ smooth ] ) . using ( [ trace ] )
we get @xmath204 which we may call smoothed @xmath0 norms .
the next step is to drop the stationary eigenvalue @xmath46 in the traces of the frobenius - perron operator .
the fourier transform of this eigenvalue leads to a @xmath115 function in the limit @xmath205 which is not a point of interest here . in the husimi representation
we identify the eigenvalue @xmath46 as the sum of squared @xmath39 norms of husimi eigenfunctions .
this is easily seen from the husimi matrix in basis of spherical harmonics . in the first row and column
is only one non - vanishing matrix element @xmath206 .
note that @xmath207 is a constant function on the sphere and therefore proportional to the stationary density , i.e. it is the eigenfunction from eigenvalue @xmath46 . introducing the diagonal representation of the husimi propagator ( [ diagrep ] ) one identifies @xmath208 .
note that @xmath209 , where @xmath210 . for @xmath211
the trace of the frobenius - perron operator is not defined .
the integral on the lhs of ( [ trace ] ) , however , is defined and gives the leading order contribution @xmath18 .
we replace the traces by sums of the ruelle - pollicott resonances ( [ trres ] ) and make use of the symmetry @xmath212 , @xmath213 note that the eigenvalue @xmath46 is also dropped in the leading order term . to get from the smoothed @xmath0 norms @xmath214 to a mean value @xmath215 we have to divide it by the smoothed level density of the husimi spectrum .
the density of the husimi spectrum is identified as the density - density correlation function with respect to the floquet eigenphases , @xmath216 this spectral density can be calculated from the time fourier transformation of the form factor .
the smoothed density is given by a truncated fourier transformation as @xmath217 where we have separated the leading order term ( @xmath211 ) .
it is known that the form factor is small as the time @xmath177 is small ( @xmath218 or @xmath219 for cue or coe , respectively ) .
thus , the summation up to the ehrenfest time which is much smaller than @xmath18 becomes negligible and the smoothed spectral density is nearly constant .
before we write down the final result , we consider the truncated fourier transformation of the resonances . since the moduli of the resonances are smaller @xmath46 , the summations in ( [ sumres ] ) converge quickly such that we can replace the ehrenfest time @xmath203 by @xmath220 , @xmath221 the constant term coincides with the rmt result ( [ skewl2coe ] ) , where @xmath222 . the resonances lead to an @xmath104 ( alias @xmath223 ) correction in form of overlapping lorentz distributions ( see ( [ lorentz ] ) ) . in the classical limit the resonance corrections vanish as @xmath18 goes to infinity .
however , for finite dimension we see non - universal corrections which are related to the chaoticity of the system .
if , for instance , the classical dynamics is strongly chaotic such that all correlations vanish after one iteration , the resonances are close to the origin and the averaged @xmath0 norms show no deviations from rmt result .
before we come to numerical results we should discuss some prelimaries .
the kicked top is known to have a mixed phase space .
although elliptic islands of stable periodic orbits are much smaller than the planck cell , bifurcations can be responsible for further localization phenomena which are sometimes called super scars @xcite .
but in contrast to the resonances , bifurcations strongly influence spectral correlations .
that means that peaks resulting from bifurcating orbits are higher for the smoothed @xmath0 norms than for the averaged @xmath0 norms .
thus , we are able to distinguish between localization phenomena of resonances and bifurcations
. the smoothing is again done by a convolution with a sinc function ( [ smooth ] ) . from the contributions of the diagonal husimi eigenfunctions
we have neglegted the squared @xmath39 norms which correspond to the stationary eigenvalue .
[ mitqd ] shows ( a ) the smoothed @xmath0 norms , ( b ) the mean @xmath0 norms , and ( c ) the smoothed spectral density . in ( a )
we see a couple of peaks at the husimi eigenphases @xmath183 , @xmath224 , @xmath225 , and @xmath226 . in ( c )
are remarkable peaks at the positions @xmath227 and after division we see in ( b ) that these peaks are suppressed , while the other peaks still have the same magnitude .
comparison with the semiclassical prediction is shown in fig .
[ qmithsp ] . due to the fact that stabilized eigenvalues representing the resonances
are of largest moduli we expect that the semiclassical prediction is almost independent of the set of eigenvalues as well as all stabilized ones are taken into account . in ( a )
the semiclassical prediction is computed from ( i ) eight stabilized eigenvalues , ( ii ) eigenvalues of modulus larger @xmath228 , and ( iii ) almost all eigenvalues of the truncated frobenius - perron matrix .
comparison of ( i)(iii ) shows that the semiclassical prediction is mainly influenced from a few classical eigenvalues of large moduli which represent the resonances . in ( b )
we compare the semiclassical prediction ( iii ) with the quantum result . in particular for the peaks we find a very good agreement .
this result shows that the probability to find strongly overlapping eigenfunctions becomes large if the differences of their eigenphases coincides with the phase of a leading resonance , i.e. resonance of large modulus . in comparison with the background the peaks are small ( a few percent ) .
however , we show in the next section that strongly scarred eigenfunctions are mainly responsible for the peaks .
the eigenvalues of the frobenius - perron matrix ( c ) are plotted logarithmically ( @xmath229 versus @xmath230 ) .
these are easily associated with the peaks in ( a ) .
it should be remarked that there is no eigenvalue of large modulus which corresponds to the small peaks at @xmath231 in fig .
[ mitqd ] ( a ) .
in this section we consider single eigenfunctions and verify the statement that phase differences of strongly overlapping eigenfunctions coincide with phases of leading resonances . due to the schwarz inequality we first check
that skew husimi eigenfunctions with large @xmath0 norms are composed by scarred diagonal husimi eigenfunctions . in tab .
[ scartab ] we find 25 @xmath0 norms of most strongly localized skew husimi eigenfunctions , their eigenphases , and the corresponding diagonal husimi eigenfunctions ( the numbers correspond to the enumeration in tab . [ hustab ] ) . due to the symmetry
@xmath232 we have restricted the eigenphases as @xmath233 .
the 32 most strongly localized diagonal husimi eigenfunctions are presented in tab .
[ hustab ] .
comparison of both tables proves that all skew husimi eigenfunctions considered are composed from at least one scarred diagonal eigenfunction . in fig .
[ ql2dgn ] ( a ) @xmath0 norms of all diagonal husimi eigenfunctions are shown , while in ( b ) we see 530 ( of total @xmath234 ) @xmath0 norms of most strongly localized skew husimi eigenfunctions .
due to our foregoing results localized skew husimi eigenfunctions appear frequently around the resonance phases .
interestingly , there are only two remarkable eigenfunctions ( no 11 and 12 in tab .
[ scartab ] ) around the husimi eigenphases @xmath231 . moreover , in both cases one of the underlying diagonal husimi eigenfunctions is no 16 in tab .
[ hustab ] . further investigation of this eigenfunction has shown that it is strongly scarred on two bifurcating orbits of periods @xmath46 and @xmath146 .
it seems that we found a super scar corresponding to a period - tripling bifurcation .
in conclusion , phase - space localization of quantum eigenenergy functions , say scarring , is not only explained by periodic orbits , but also by ruelle - pollicott resonances and their corresponding resonance eigenfunctions .
in particular , we found the interesting result that quantum floquet eigenfunctions are pairwise localized in the same phase - space regions if the difference of their eigenenergies coincides with the phase of a leading resonance , i.e. resonance close to the unit circle .
but note that this is a statistical statement which does not make a prediction for individual eigenstates .
moreover , we can not determine if there are either a few strongly scarred or many weakly localized eigenfunctions .
however , the semiclassical prediction of the averaged @xmath0 norms is in a good agreement with numerical results .
the correspondence between scars around periodic orbits described by heller and the results of this paper might be understood as follows : resonances can be computed by a so - called cycle expansion , where resonances appear as roots of a polynomial whose coefficients are calculated from contributions of short periodic orbits ( pseudo orbits ) @xcite .
on the one hand , scars typically appear around weakly unstable periodic orbits . on the other hand , these weakly unstable
orbits mainly induce the cycle expansion .
although the kicked top has a mixed phase space , localization effects of stable orbits or bifurcations can be neglected if such phase - space structures are not resolved by the planck cell .
further investigations are needed for the understanding of the so - called super scars which are related to bifurcations .
99 a. schnirelman , usp .
29 , 181 ( 1974 ) e. p. wigner , phys .
40 * , 749 ( 1932 ) k. husimi , proc .
phys . math .
55 , 762 ( 1940 ) e. j. heller , phys .
53 , 1515 ( 1984 ) s. nonnenmacher and a. voros , j. stat .
. vol . * 92 * , 431 ( 1998 ) s. gnutzmann and k. yczkowski , j. phys .
a * 34 * , 10123 ( 2001 ) h. h. hasegawa and w. c. saphir , phys .
a * 46 * , 7401 ( 1992 ) a. lasota and m. c. mackey , _ chaos , fractals , and noise _ , ( springer , new york , 1994 ) p. gaspard , _ chaos , scattering and statistical mechanics _ , ( cambridge university press , new york , 1998 ) m. pollicott , invent . math . * 81 * , 415 ( 1985 ) d. ruelle , phys .
lett . * 56 * , 405 ( 1986 ) d. ruelle , j. stat . phys .
* 44 * , 281 ( 1986 ) v. baladi , j .-
eckemann und d. ruelle , nonlinearity * 2 * , 119 ( 1989 ) a. v. andreev and b. l. altshuler , phys .
* 75 * , 902 ( 1995 ) o. agam , b. l. altshuler , and a. v. andreev , phys .
lett . * 75 * , 4389 ( 1995 ) a. v. andreev , o .
agam , b. d. simons , and b. l. altshuler , phys .
lett . * 76 * , 1 ( 1996 ) a. v. andreev , b. d. simons , o. agam , and b. l. altshuler , nuclear physics b , * 482 * , 536 ( 1996 ) f. haake , m. ku , and r. scharf , z. phys .
b * 65 * , 381 ( 1987 ) p. a. braun , p. gerwinski , f. haake , and h. schomerus , z. phys .
b * 100 * , 115 ( 1996 ) m. ku , f. haake , and b. eckhardt , z. phys .
b * 92 * , 221 ( 1993 ) f. j. dyson , j. math .
phys . * 3 * , 140 , 166 ( 1962 ) m. l. mehta , _ random matrices _ , ( academic press , new york , 1991 ) j. weber , f. haake , and p. eba , phys .
lett . * 85 * , 3620 ( 2000 ) j. weber , f. haake , p. a. braun , c. manderfeld , and p. eba , j. phys .
a * 34 * , 7195 ( 2001 ) f. t. arecchi , e. courtens , r. gilmore , and h. thomas , phys .
rev . a * 6 * , 2211 ( 1972 ) r. glauber and f. haake , phys .
a * 13 * , 357 ( 1976 ) a. m. perelomov , _ generalized coherent states and their applications _
( springer , new york , 1986 ) c. manderfeld , _ coherent state representation of the @xmath60 group _ , http://www.theo-phys.uni-essen.de/tp/u/chris/ ( 2001 ) i. s. gradshteyn and i. m. ryzhik , _ table of series , integrals , and products _ ( academic press , san diego , 1994 ) f. haake , _ quantum signatures of chaos _
( springer , berlin 2nd edition 2001 ) c. manderfeld , j. weber , and f. haake , j. phys .
a * 34 * , 9893 ( 2001 ) the author thanks petr braun for showing this identity j. p. keating , nonlinearity 4 , 309 ( 1991 ) g. m. dariano , l. r. evangelista und m. saraceno , phys .
a * 45 * , 3646 ( 1992 ) j. p.
keating and s. d. prado , proc .
a * 457 * , 1855 ( 2001 ) f. christiansen , g. paladin and h. h. rugh , phys .
rev . lett . * 65 * , 2087(1990 ) p. cvitanovi and b. eckhardt , j. phys .
a * 24 * , l237 ( 1991 ) @xmath236 & \frac{n^2}{4\pi } ||q_{ik}||^2 & { \rm no}&{\rm no } \\ \hline 1 & 0.02293 & 4.87904 & 9 & 8 \\ \hline 2 & 0.02995 & 2.17252 & 4 & 2 \\ \hline 3 & 0.03690 & 2.12795 & & 8 \\ \hline 4 & 0.12504 & 2.26266 & 4 & 1 \\ \hline 5 & 0.66008 & 2.53247 & 8 & 4 \\ \hline 6 & 0.69002 & 2.14943 & 8 & 2 \\ \hline 7 & 0.95140 & 2.07455 & 32 & 8 \\ \hline 8 & 1.58566 & 3.28086 & 30 & 11\\ \hline 9 & 1.61324 & 2.07295 & 30 & \\
\hline 10 & 1.63071 & 2.31385 & 30 & \\ \hline 11 & 1.93314 & 2.77164 & 16 & 9 \\ \hline 12 & 1.95607 & 3.47930 & 16 & 8 \\ \hline 13 & 2.12224 & 2.03752 & 16 & \\
\hline \end{array } \quad \begin{array}{|r|r|r|r|r| } \hline { \rm no } & \big[\phi_i-\phi_k\big ] & \frac{n^2}{4\pi } ||q_{ik}||^2 & { \rm no}&{\rm no } \\
\hline 14 & 2.41853 & 2.01226 & 21 & \\ \hline
15 & 2.61615 & 2.08038 & 16 & 4 \\ \hline 16 & 2.64610 & 2.04506 & 16 & 2 \\ \hline 17 & 2.92433 & 2.26189 & 21 & 5 \\ \hline 18 & 2.97684 & 2.21163 & & 8 \\ \hline 19 & 2.99690 & 2.05725 & & 9 \\ \hline 20 & 3.01983 & 2.73944 & & 8 \\ \hline 21 & 3.10287 & 2.10537 & 23 & 9 \\ \hline 22 & 3.11230 & 3.05349 & 24 & 8 \\ \hline 23 & 3.11857 & 2.86021 & 17 & 30\\ \hline 24 & 3.12580 & 2.55565 & 23 & 8 \\ \hline 25 & 3.13523 & 2.48017 & 24 & 9 \\ \hline \end{array}\ ] ] | we study the correspondence between phase - space localization of quantum ( quasi-)energy eigenstates and classical correlation decay , given by ruelle - pollicott resonances of the frobenius - perron operator .
it will be shown that scarred energy eigenstates are correlated : pairs of eigenstates strongly overlap in phase space ( scar in same phase - space regions ) if the difference of their eigenenergies is close to the phase of a leading classical resonance .
phase - space localization of quantum states will be measured by @xmath0 norms of their husimi functions . |
it is long known that the forward compton scattering ( cs ) amplitudes can , by unitarity , causality and crossing , be expressed through integrals of the photoabsorption cross sections @xcite .
the low - energy expansions of these expressions lead to a number of useful sum rules , most notably those of baldin @xcite , and gerasimov , drell and hearn ( gdh ) @xcite . given the photoabsorption cross sections
, one can thus provide a reliable assessment of some of the static electromagnetic properties of the nucleon and nuclei , as well as of the forward cs amplitudes in general .
for the proton , the first such assessments was performed in the early 1970s @xcite . since then
, the knowledge of the photoabsorption cross sections appreciably improved , and yet for the unpolarized case only the baldin sum rule has been updated @xcite . in this work , we provide a re - assessment of the forward spin - independent amplitude of proton cs , and evaluate the associated sum rules involving the _ dipole and quadrupole polarizabilities of the proton_. sum rules are essentially the only way to gain empirical knowledge of the forward cs amplitudes .
it is impossible to access the forward kinematics in real cs experiments .
the measurement of the forward spin - independent cs amplitude can be done indirectly through the process of dilepton photoproduction ( @xmath1 ) @xcite .
the timelike cs , involved in the process of dilepton photoproduction , yields access to real cs given the small virtuality of the outgoing photon , or equivalently , the nearly vanishing invariant mass of the produced pair .
the experimental result @xcite compared well with the aforementioned evaluations @xcite . despite the substantial additions to the database of total photoabsorption cross sections , the works of damashek and gilman ( dg ) @xcite as well as armstrong _
et al . _
@xcite remained to be , until now , the only evaluations of the full amplitude .
the newer data were used , however , in the most recent evaluations of the baldin sum rule @xcite , which yields the sum of the electric and magnetic dipole polarizabilities , .
these recent analyses obtained somewhat lower value for the sum than dg , cf .
table [ sumruletest ] . in this work
we find that the difference between the early and the recent evaluations arises from systematic inconsistencies in the experimental database .
we also obtain the sum rule value for a combination of higher - order quadrupole polarizabilities and compare it with several theoretical predictions .
this paper is organized as follows . in sect .
[ sec : overview ] we give a brief overview of the kramers - kronig relation and sum rules for polarizabilities . in sect .
[ sec : fitting ] we discuss the fitting procedure for the unpolarized total proton photoabsorption cross section data .
the sum rule evaluations of scalar polarizabilities and of the spin - independent forward cs amplitude are presented in sect .
[ sec : sumrules ] .
conclusions are given in sect .
[ sec : summary ] .
the appendix demonstrates the elastic - channel contribution to the sum rules and polarizabilities on the example of one - loop scalar qed .
for a spin-1/2 target , such as the proton , the forward cs amplitude is given by t_fi = f ( ) ^ + g ( ) i ( ^ ) , where @xmath2 and @xmath3 are scalar functions of the photon lab energy @xmath4 ; vectors @xmath5 and @xmath6 represent the photon and proton polarizations , respectively .
the crossing symmetry implies that the spin - independent amplitude @xmath2 is an even and the spin - dependent amplitude @xmath3 is an odd function of @xmath4 .
the optical theorem ( unitarity ) relates the imaginary part of the amplitudes to the total photoabsorption cross sections : @xmath7 , \\
\operatorname{im}g(\nu ) & = & \frac{\nu}{8\pi } \left[\sigma_{1/2}(\nu)-\sigma_{3/2}(\nu)\right ] . \eqlab{unitarity}\end{aligned}\ ] ] here @xmath8 is the doubly - polarized cross section with @xmath9 representing the combined helicity of the initial @xmath10 state .
averaging over the polarization of initial particles gives the unpolarized photoabsorption cross section , @xmath11 . in the present article we focus on relations involving the spin - independent amplitude @xmath2 , and the unpolarized cross section @xmath12 .
the kramers - kronig relation between these quantities exploits the optical theorem , causality and crossing symmetry , to yield for the proton @xcite : f ( ) = - + _ 0^ ( ),where @xmath13 is the fine - structure constant and @xmath14 is the proton mass ; the slashed integral denotes the principal - value integration .
we next would like to consider the low - energy expansion of @xmath2 . at this point
it is important to note that the elastic scattering , i.e. the cs process itself , is one of the photoabsorption processes .
the total cs cross section does not vanish for @xmath15 but goes to a constant the thomson cross section : ( 0 ) = .
this means does not admit a taylor - series expansion around @xmath16 ( each coefficient in that expansion is infrared divergent , cf .
appendix ) .
such expansion is nonetheless important for establishing the polarizability sum rules .
we hence prefer to take the cs out of the total cross section , i.e. : ( ) = _ ( ) + _ ( ) , where @xmath17 can be assumed to be dominated by hadron - production processes , for which there is a threshold at some @xmath18 .
the amplitude @xmath2 can be decomposed accordingly into the elastic and inelastic terms , f ( ) & = & f _ ( ) + f _
( ) , + f _ ( ) & = & - + _ 0^ _ ( ) , + f _
( ) & = & _ _ 0^ _ ( ) .
the details on dealing with @xmath19 can be found in appendix [ sec : scalarqedexample ] . in what follows
, however , we neglect the contribution from @xmath20 , as it is suppressed by an extra order of @xmath21 .
hence we set @xmath22 , as is usually done . considering @xmath23 , the low - energy expansion of both sides of leads to the sum rules for polarizabilities . at the leading order [ @xmath24 , one obtains the baldin sum rule @xcite for the sum of electric ( @xmath25 ) and magnetic ( @xmath26 ) dipole polarizabilities : _ e1 + _ m1 = _ _ 0^. at @xmath27 we obtain ` the 4^th^-order sum rule ' : _ e +
_ m + ( _ e2 + _ m2 ) = _ _ 0^ , + which involves the quadrupole polarizabilities @xmath28 , @xmath29 , as well as the leading dispersive contribution to the dipole polarizabilities denoted as @xmath30 , @xmath31 , see @xcite for more details .
our aim here is to provide an empirical fit of the available data for @xmath17 and evaluate the various sum rules .
the presently available experimental data , together with the results of the empirical analyses maid and said , as well as our fits are displayed in . in our fitting we distinguish the following three regions : * _ low - energy _ , @xmath32 ; * _ medium - energy _ , @xmath33 ; * _ high - energy _
, @xmath34 ; where @xmath35 ( @xmath36 gev ) and @xmath37 ( @xmath38 gev ) are respectively the thresholds for the single- and double - pion photoproduction on the proton . in the _ low - energy _
region we use the pion production ( @xmath39 and @xmath40 ) cross sections from the maid @xcite and said @xcite partial - wave analyses . in our error estimate we assign a 2% uncertainty on these values . in the _ medium - energy _
region we fit the actual experimental data , using a sum of breit - wigner resonances and a background .
following @xcite , we take six breit - wigner resonances , each parameterized as [ res_fit ] _
r(w)=a , where @xmath41 is the total energy of the @xmath42 system .
the background function is from @xcite : [ bgd_fit ] _
b(w)=_k=-2 ^ 2 c_k ( w - w_0)^k , where @xmath43 corresponds with the pion photoproduction threshold . .fitting parameters for the resonances ( [ res_fit ] ) obtained for fit i and ii .
[ cols="^,^,^,^",options="header " , ] the corresponding full results ( sum of the three regions ) are given in table [ sumruletest ] , and compared with the results of previous works . in this table we also give the result for the 6th - order integral , and for the full amplitude @xmath2 at @xmath44 gev .
the real part of @xmath2 is plotted in over a broad energy range and compared with previous evaluations and the experimental number from the 1973 desy experiment at 2.2 gev .
although none of the evaluations really contradicts the experiment , there is a clear tendency to a higher central value .
the new dilepton photoproduction experiments planned at the mainz microtron ( mami ) could perhaps provide experimental values in the lower energy range .
obviously , the regions of the extrema ( e.g. , the @xmath45 region or the interval between 0.6 and 0.7 gev ) are most interesting as the different evaluations seem to differ there the most . in the region around 0.6 gev , for example , one of our evaluations ( fit i ) is nearly identical with armstrong s @xcite , while the other one ( fit ii ) is aligning with dg @xcite and a2 coll . @xcite .
an appropriately precise experiment could tell which of the groups is correct , if any .
compared to the @xmath46pt calculation of ref .
@xcite . ] and @xmath47 combinations of polarizabilities , compared to results from dispersion relation approaches ( dr ) @xcite , baryon chiral perturbation theory ( b@xmath46pt ) @xcite , and heavy baryon chiral perturbation theory ( hb@xmath46pt ) @xcite .
the errors of the @xmath46pt derive from our crude estimate of the next - order corrections . ]
figure [ fig : reim_f ] shows both the real and imaginary part of @xmath2 at lower energies , where it can be compared with a calculation done within chiral perturbation theory ( @xmath46pt ) @xcite .
a rather nice agreement between theory and empirical evaluations is observed for energies up to about the pion - production threshold . for very low energy
this comparison can be made more quantitative by looking at the polarizabilities . while for the baldin sum rule the situation was extensively discussed in the literature ( cf .
@xcite for a recent review ) , the 4^th^-order sum rule was not studied at all .
it can , however , be very useful in unraveling the higher - order polarizabilities , as illustrated by .
this is the plot of a combination of proton magnetic polarizabilities versus electric , where the various theory predictions are compared with our 4^th^-order sum rule evaluation .
the band representing the sum rule covers the interval between the two values given in table [ sumruletest ] ( rows ` fit i ' and ` fit ii ' ) .
the sum rule clearly provides a model - independent constraint on these polarizabilities and a rather stringent test for the theoretical approaches .
the fundamental relation between the photon absorption and scattering , encompassed in the kramers - kronig type of relations , allows us to evaluate the forward compton scattering off protons using the empirical knowledge of the total photoabsorption cross sections .
the present database of the unpolarized photoabsorption cross section is not entirely consistent and so as to reflect that we obtain two distinct fits to it .
the two fits yield slightly different results for the spin - independent amplitude @xmath48 , and hence for its low - energy expansion characterized by the scalar polarizabilities of the proton .
our two results for the sum of dipole polarizabilities ( or , baldin sum rule ) correspond nicely with the results of previous evaluations , which too can be separated into two groups : the _ old _
@xcite , with the value slightly above 14 ( in units of @xmath49 fm^3^ ) , and the _ new _
@xcite , with the value slightly below 14 .
the 1996 daphne@mami experiment @xcite , superseding the 1972 experiment of armstrong _ et al . _
@xcite , is clearly responsible for this difference .
neglecting the older data in favor of the newer ones , yields the lower value of the baldin sum rule , and vice versa . while one can take a preference in one of the two fits and corresponding results , we prefer to think of their difference as a systematic uncertainty in the present evaluation of the polarizabilities and of the forward spin - independent amplitude of the proton .
as far as polarizabilities are concerned , only the baldin sum rule is appreciably affected by the inconsistency in the photoabsorption database .
nevertheless , the two results ( fit i and ii in table [ sumruletest ] ) are not in conflict with each other , given the overlapping error bars .
it is customary to take a statistical average in such cases .
taking a weighted average , over a set @xmath50 , we use @xcite : @xmath51 ] over our two values for the baldin sum rule we obtain : @xmath52 the error bar here does directly not include the aforementioned systematic uncertainty of the cross section database .
however , since the two results are fairly well surmised by the weighted average , the latter should be less prone to the systematic uncertainty of the database .
we have presented a first study of the sum rule involving the quadrupole polarizabilities , , here referred to as the 4^th^-order sum rule. our weighted average value for this sum rule , in the proton case , is @xmath53 .
it agrees very nicely with the state - of - the - art calculations of these polarizabilities based on fixed-@xmath54 dispersion relations and chiral perturbation theory , see .
we note that , while the calculations demonstrate significant differences in the values of individual higher - order polarizabilities , these differences apparently cancel out from the forward combination of these polarizabilities which enters the sum rule . in the subsequent paper we will discuss the evaluation of the forward _ spin - dependent _ amplitude @xmath55 and related sum rules for the forward spin polarizabilities of the proton .
the knowledge of the two amplitudes will allow us to reconstruct the observables for the proton compton scattering at zero angle .
we thank jrgen ahrens for kindly supplying us with a database of total photoabsorption cross sections .
this work was supported by the deutsche forschungsgemeinschaft ( dfg ) through the collaborative research center sfb 1044 [ the low - energy frontier of the standard model ] , and the graduate school dfg / grk 1581 [ symmetry breaking in fundamental interactions ] .
consider the elastic forward scattering of a photon with momentum @xmath56 from a charged spinless particle with four - momentum @xmath57 and mass @xmath58 . in the forward direction ( @xmath59 )
this process is completely described by a single amplitude @xmath48 .
the tree - level qed calculation ( ) yields immediately @xmath60 , where we have chosen the normalization of this amplitude to coincide with the analogous amplitude for the spin-1/2 case [ see ] ; the superscript indicates the order of @xmath21 .
next we consider the one - loop corrections .
figure [ imdiagrams ] shows the one - particle - irreducible ( 1pi ) diagrams appearing in scalar qed .
the corresponding one - particle - reducible ( 1pr ) diagrams vanish in forward direction , due to the transversality of the photon polarization vector @xmath61 with respect to any of the four momenta , i.e. : @xmath62 .
[ eq : scalarscatteringamp ] f^(2 ) ( ) & = & \ { + + ( 1 + ) + & + & } + ^(2 ) ( ) , where li@xmath64 is the dilogarithm , and @xmath65 is the total cs cross section arising at the tree level ( cf . ):
@xmath66 we note that in the low - energy limit it reproduces the thomson cross section : @xmath67 , a result that is unaltered by loop corrections , i.e. @xmath68 . as the total photoabsorption cross section , to this order in @xmath21 ,
is given entirely by the tree - level cs cross section , the fact that @xmath69 coincides in this case with the statement of the optical theorem .
we have also checked that the one - loop amplitudes satisfies the once - subtracted dispersion relation : @xmath70 and hence the full amplitude , @xmath71 , indeed enjoys the kramers - kronig relation given in .
now , the whole point of this exercise is to understand the low - energy expansion , and thus the polarizability sum rules , in the case when the photoabsorption cross section is not vanishing at @xmath16 . expanding the real part of around @xmath16
, we find : ( ^2 + ^4 + ^6 + ) & = & _ n=1^^2n _ 0^ .
hence , the coefficients diverge in the infrared .
however , there is an apparent mismatch : they are logarithmically divergent on one side and power - divergent on the other . to match the sides exactly at each order of @xmath4 , thus defining the sum rules for `` quasi - static '' polarizabilities , we subtract all the power divergences on the right - hand side ( rhs ) and regularize both sides with the same infrared cutoff ( equal to @xmath4 ) : ( ^2 + ^4 + ) & = & _ n=1^^2n _ ^ .
+ both sides are now identical at each order of @xmath4 .
this is nontrivial , at least for the analytic terms ; the logs are fairly easily obtained from the non - regularized right - hand side ( rhs ) in , cf .
@xcite . extending these arguments to all orders in @xmath21
, we find that the proper low - energy expansion for the ` elastic ' part of the amplitude [ see ] reads as : f_()= - + _ n=1^^2n _ ^ , where @xmath72 is the total cross - section of compton scattering and @xmath73 are the infrared subtractions : |_n( ) _
k=0 ^ 2(n-1)|_=0^ k. now we can , for instance , formulate the baldin sum rule for the elastic contribution to the dipole polarizabilities . by definition
f _ ( ) = -/m + ( _ e1+_m1)_^2 + o(^4 ) , and hence , matching it with the rhs of , we obtain : @xmath74 in our scalar qed example , where @xmath12 is the tree - level cross section @xmath75 , we obtain @xmath76 which of course reproduces the one - loop result [ cf . the first term in the expansion of @xmath77 in ] .
v. lensky and v. pascalutsa , eur .
j. c * 65 * , 195 ( 2010 ) .
h. w. griesshammer , j. a. mcgovern , d. r. phillips and g. feldman , prog . part . nucl .
* 67 * ( 2012 ) 841 .
d. drechsel , b. pasquini and m. vanderhaeghen , phys .
* 378 * ( 2003 ) 99 .
v. lensky , j. a. mcgovern and v. pascalutsa , preprint mitp/15 - 017 , to be published .
b. r. holstein , d. drechsel , b. pasquini and m. vanderhaeghen , phys .
c * 61 * ( 2000 ) 034316 [ hep - ph/9910427 ] .
l. g. parratt , _ probability and experimental errors in science _ , ( john wiley & sons , new york , 1961 ) . | we evaluate the forward compton scattering off the proton , based on kramers - kronig kind of relations which express the compton amplitudes in terms of integrals of total photoabsorption cross sections .
we obtain two distinct fits to the world data on the unpolarized total photoabsorption cross section , and evaluate the various spin - independent sum rules using these fits . for the sum of proton electric and magnetic dipole polarizabilities , governed by the baldin sum rule
, we obtain the following average ( between the two fits ) : @xmath0 .
an analogous sum rule involving the quadrupole polarizabilities of the proton is evaluated too .
the spin - independent forward amplitude of proton compton scattering is evaluated in a broad energy range .
the results are compared with previous evaluations and the only experimental data point for this amplitude ( at 2.2 gev ) .
we remark on sum rules for the elastic component of polarizabilities . |
calorimetric studies have long been valuable tools for rigorous tests of physical law , ranging from joseph black s early work on latent heat , to measurements of the metallic specific heat showing the inadequacy of drude s classical model .
more recently , the high degree of control available in cold atomic gases has opened up exciting avenues for experimental verification of finite temperature theories of bose and fermi gases @xcite . to date
, however , there have been few works experimentally investigating the energy - temperature relationship of a harmonically trapped bose gas . pioneering work performed by ensher _ et al . _
relied on extracting both release energy and temperature information from time - of - flight images at different evaporation points @xcite .
this work was extended by gerbier _ et al .
_ , whose measurements of the release energy were found to be in good agreement with hartree - fock theory for an interacting gas @xcite .
gati _ et al .
_ have measured temperature dependent phase fluctuations of an ideal bose gas and revealed qualitatively that the system deviates from a classical gas @xcite . to further study the energy dependence of bose - einstein condensate ( bec ) thermodynamics , two options present themselves .
firstly , an improvement in the ability to extract thermodynamic information from time - of - flight images could extend the results presented in @xcite .
secondly , an alternative method to study the energy - temperature relationship could be envisioned , allowing the total internal energy of the system to be measured , rather than just the release energy .
new methods to study this relationship are emerging , motivated by the recent characterization of the heat capacity of a strongly interacting fermi gas @xcite .
an attempt to measure the specific heat of an ultracold bose gas using a time - dependent trapping potential , as well as by heating using laser pulses , has been performed @xcite , although obtaining accurate data was found to be impractical in their system .
a recent experiment has extracted information regarding the heat capacity of a bose gas using global variables @xcite . in this paper
we follow the theoretical proposal of blakie _ et al .
_ to transfer a known quantity of irreversible work to a bec and measure the resulting temperature @xcite . by utilizing two independent methods
we perform known , precise amounts of work on a@xmath0rb condensate .
the resulting temperature is measured after a period of thermalization , giving the transferred energy as a function of temperature .
this provides a rigorous test of the energy dependence of the thermodynamics of our system , as energy and temperature measurements are performed independently .
our approach contrasts that of ensher _ et al .
_ , who measure their system at differing evaporation points , extracting both energy and temperature information from time - of - flight images .
furthermore , our approach is not sensitive to the ground state energy of the system present at @xmath1 , allowing a more direct comparison with the specific heat .
our results from the two methods compare well , both with each other and with hartree - fock numerical calculations for an interacting gas .
the paper is organised as follows : section [ sec : exp ] reviews the experimental parameters and methods .
[ sec : temp ] details the temperature measurements .
the two methods of energy transfer are discussed in sec .
[ sec : grav ] and sec .
[ sec : kick ] .
we close with a discussion and conclusion in sec .
[ sec : dis ] and sec .
[ sec : con ] respectively .
[ sec : exp ] our experiment involves a bec of atoms prepared in the ground state , and held in an optical dipole trap @xcite . the trap is formed at the intersection of two focused co@xmath2 laser beams , with wavelength 10.6 @xmath3 m , and each with a @xmath4 radius of 33 @xmath3 m .
the co@xmath2 laser power is stabilised using a closed - loop feedback system to ensure long term reproducibility of the trap depth , bec atom number , and temperature . after loading atoms into the dipole trap from a magneto - optical trap operating on the 780.2 nm @xmath5 transition ,
a 6 second evaporative cooling sequence is used to produce a bec .
we then execute the experimental sequence shown in figure [ fig:1 ] . the laser power is adiabatically ramped to a higher value over 100 ms using an exponential profile .
the deeper potential resulting from this ramp prevents atom loss during the heating process .
the adiabaticity of this ramp has been confirmed by ensuring that negligible non - condensed fraction exists following the ramp and a 100 ms hold time at the final laser power .
we then transfer a precise amount of energy to the system , using one of two methods , before allowing the system to rethermalize for 100 ms
. - @xmath6 .
work is then done on the condensate in one of two ways : ( a ) the release of the atoms for a time @xmath7 0 - 1000 @xmath3s leads to falling expansion of the cloud , resulting in increased kinetic and potential energy when the trap is subsequently reinstated ; ( b ) a 300 ns pulse of an off - resonant standing wave leads to diffraction of a fraction of the atoms . following
either of these is a 100 ms period of thermalization , before the condensate is left to expand for 10 ms to allow the momentum distribution to be imaged via time - of - flight.,width=313 ] we approximate our optical dipole trap as a harmonic potential characterized by a set of frequencies @xmath8 that define the potential in three dimensions .
these frequencies are measured through a parametric heating process @xcite , where the trap depth is modulated sinusoidally for a period of 200 ms with an amplitude of @xmath9% of the total trap depth .
parametric excitation along dimension @xmath10 occurs for , for integer @xmath11 .
measurements of the excitation frequencies allow us to characterize our trap and calculate the critical condensation temperature for an ideal bose gas @xmath12 , given by @xmath13^{1/3},\ ] ] where @xmath14 is the geometric mean of the trapping frequencies , @xmath15 is the boltzmann constant , @xmath16 is the number of atoms , and is the riemann zeta function ( e.g. see @xcite ) . we ensure that the initial system is at zero temperature by evaporating to a point where the thermal fraction can no longer be observed .
we find that any further lowering of the trapping potential only leads to strong depletion of the condensate , and conclude that the zero initial temperature condition is satisfied . in our system , the interaction energy of the initial bec far outweighs the kinetic energy , as , where @xmath17 is the s - wave scattering length , and is the characteristic harmonic oscillator length @xcite , where @xmath18 is the mass of an atom .
typically we find that , and can assume the thomas - fermi approximation applies .
the temperature and number of atoms are measured using time - of - flight imaging with resonant absorption , and the properties of the atomic clouds are inferred from these images . following an experimental sequence , the dipole trap containing the atoms
is rapidly switched off using an acousto - optic modulator , and the atoms are allowed to freely expand for 10 ms . after a repumping pulse
, the atoms are probed with a @xmath19 pulse on resonance with the transition .
the probe light has an intensity of , which is less than the saturation intensity of . when processing time - of - flight images we first subtract the average background of all images and apply a fringe - removal algorithm @xcite , improving our signal - to - noise ratio and ability to detect low density components of the expanded atomic cloud .
the images are then integrated along the @xmath20- and @xmath21-dimensions to obtain two one - dimensional density profiles , from which we extract both the temperature and the number of atoms of our sample . above the critical temperature
, the expansion of an ideal bose gas evolves according to a simple scaling relation , where we can define the effective temperature after an expansion time @xmath22 in dimension @xmath23 as @xmath24 here , @xmath25 is the variance of the resulting distribution as a function of the expansion time . far above the experimentally observed critical temperature @xmath26 ,
this can be determined by a fit to a gaussian function .
close to @xmath26 and below , the density distribution of the thermal cloud becomes predominantly the bose - distribution , and by setting the chemical potential to zero , it can be described by a bose - enhanced gaussian @xcite . in the hydrodynamic regime , there is the possibility of anisotropic expansion for a very elongated trap , which occurs when the mean free path of the atoms is less than the dimension of the trap ( e.g. see @xcite ) .
for the experiments presented in this paper our trap is nearly isotropic , as we have @xmath27 , with @xmath28 and @xmath29 for the two experiments .
in addition , the 10 ms expansion time used sets @xmath30 , and hence we can assume that the expansion of the ideal thermal component to be isotropic , as observed experimentally .
we therefore are able to assume that @xmath31 . below the critical temperature
there exists a non - negligible condensed fraction , and as such the assumption of a ballistic expansion for the entire cloud is no longer valid .
we therefore extract the temperature using the method presented in @xcite .
the central interacting region is systematically excluded from our measurements by performing multiple fits of a bose - enhanced gaussian to the wings of the profile , with a varying cut - off width for the excluded central region . for each fit
, the width of this region is determined by a scaling factor @xmath32 , such that the region @xmath33 is excluded from the fit , as shown in figure [ fig : splot](a ) . here
@xmath34 is the thomas - fermi radius of the condensed fraction in dimension @xmath35 , with @xmath36 .
@xmath32 is chosen such that the region excluded for the fit is larger than the width of the condensed fraction , as sampling this region would cause us to systematically underestimate the temperature .
on the other hand , if @xmath32 is too large , we are limited by the signal - to - noise ratio of our images
. there exists an intermediate region where the measured temperature depends only weakly on the width of the excluded region , which we typically find to be @xmath37 .
we infer the temperature from this region , as illustrated in figure [ fig : splot](b ) . .
( a ) the bimodal fit applied to a sample along the @xmath21-dimension , where we have inferred the temperature from a fit to the grey shaded region , here with @xmath38 . here the position @xmath21 has been scaled by the thomas - fermi radius @xmath39 .
( b ) the effect of scanning @xmath32 on the temperature measurement for a sample image .
the grey shaded region indicates the typical values of @xmath32 used to infer the temperature.,width=313 ] this technique is only valid when the extent of the thermal profile is much larger than that of the condensed fraction , and thus fails at very low temperatures . in our experiment
this is more pronounced in one direction , due to higher oscillator frequency in that dimension ( @xmath27 ) .
we therefore choose to extract temperature information from the @xmath40dimension , where the resulting momentum distribution of the condensate fraction is narrower , and apply a temperature cut - off at @xmath41 .
an investigation of the very low temperature region ( @xmath42 ) would require a new thermometry technique to be used , such as recently presented by olf _ et al .
_ this technique involves analyzing the decoherence of a quantum superposition of spin states , which allows them to measure temperatures as low as @xmath43 @xcite . to check the accuracy of our temperature measurement well below @xmath26
, we have simulated the mean - field effect of the condensate on the thermal cloud as the gas expands .
we model the expansion of the condensate using hydrodynamic scaling @xcite . for the thermal cloud we use two approaches , monte carlo with @xmath44 test particles @xcite , and a scaling approach after @xcite . in both cases
we find that the effect on the temperature measurement is less than 10% . for higher temperatures , still below @xmath26 , the magnitude of this effect
is decreased due to the smaller condensed fraction .
we estimate the uncertainty on @xmath45 here to be 5% , accounting for the uncertainty in length calibration , possible collisional effects during expansion , and any variance of the temperature measurement depending on the specific choice of @xmath32 .
the atom number is obtained from a bimodal fit to the absorption profile @xcite . for temperatures below @xmath26
, we perform a bimodal fit of a bose - enhanced gaussian to the thermal fraction , and a thomas - fermi profile to the condensed fraction using the method presented in @xcite .
integration over the entire bimodal profile gives the total optical density @xmath46 , which is directly proportional to the number of atoms @xmath16 , given by @xmath47 .
here @xmath48 is the area of a pixel , and @xmath49 is the experimental absorption cross - section .
here @xmath50 , with @xmath51 being the wavelength of the probe laser beam used in the imaging process , and @xmath52 is a scaling factor that allows us to account for the clebsch - gordan coefficients combined with the experimental distribution of magnetic substates following repumping . for an even distribution across the magnetic substates
, as would be the case after repumping , @xmath53 . experimentally , we independently determine @xmath52 by observing experimental images having a temperature close to @xmath26 .
we then scale our measured number of atoms such that the measured temperature agrees with the theoretical critical temperature @xmath54 , which we have determined to be from hartree - fock numerical simulations . due to the nature of the interacting transition
, the condensed fraction does not go abruptly to zero as the temperature crosses @xmath54 , as in the ideal case . using this method
we determine that @xmath55 .
we utilize two separate methods for transferring energy to our system .
the first method was proposed by . here
we consider an irreversible work process : a bose - einstein condensate is released from the trapping potential and allowed to expand under the influence of gravity . after a time @xmath56 ( typically @xmath57 ) , the atoms are recaptured , and allowed to rethermalize .
there are three contributions to the amount of work done on the atoms during this process .
one , the atoms fall under gravity ( acceleration @xmath58 ) and gain kinetic energy ; two , the displacement @xmath59 from the fall leads to a potential energy gain when the trap is reinstated ; three , the larger cloud size after the expansion results in greater potential energy when the trap potential is restored .
energy from the first two contributions will be coupled to a center - of - mass oscillation in the potential in the @xmath60-direction , known as the `` kohn '' mode @xcite .
although this mode will theoretically persist in a harmonic trap , we observe that these oscillations are damped in the thermalization process after @xmath61 ms .
we attribute this to anharmonicities in the gaussian laser trap profile @xcite . due to the observed damping of the kohn mode
, we consider this energy to be completely available for rethermalization , and to have the form @xmath62 where @xmath63 is the trap frequency parallel to the direction of gravity . to calculate the energy acquired from expansion of the cloud , we assume that the atoms will undergo a self - similar expansion , with widths evolving according to @xcite . here
@xmath64 are the thomas - fermi profile condensate widths in directions @xmath65 , and the evolution of @xmath66 is given by @xmath67 with @xmath68 .
et al . _
@xcite have shown that the energy transferred to the system due to a symmetric expansion about the trap minimum is given by @xmath69 where @xmath70 is the thomas - fermi chemical potential , is the ratio of trapping frequencies before and after @xmath56 , and . in our experiment
we constrain the trapping frequencies both before and after @xmath56 to be identical , such that @xmath71 , reducing this expression to @xmath72 the various contributions to @xmath73 are shown in figure [ fig : simtest](a ) .
the total amount of energy transferred to the system and available for rethermalization is then the sum of the contributions @xmath74 and @xmath73 .
we wish to reduce the dependence of our energy calculation on the absolute number of atoms @xmath16 , which has an uncertainty of 15% due to the error in @xmath52 .
we therefore choose to calculate the energy per particle , rather than the total energy transferred to the atoms . in this scenario , only @xmath75 maintains a dependence on @xmath16 , with @xmath70 proportional to @xmath76 .
this term accounts for less than 20% of the total energy in our experiment , as shown by a numerical simulation in figure [ fig : simtest ] .
here we have calculated the ground state of our harmonic potential in three dimensions by solving for the ground state of the gross - pitaevskii equation for our trap parameters .
this initial condition is then allowed to expand under gravity for up to 1000 @xmath3s using a split - step fourier method .
the uncertainty in @xmath77 is mainly due to the uncertainty in the measurement of the trap frequency @xmath63 , used for determining the potential energy contribution in equation [ eq : edrop ] , giving an uncertainty of 4% for this component .
the accuracy of the kinetic component of @xmath77 is due to our accuracy in determining the local value of @xmath58 , as well as the timing accuracy in our experiment .
this component accounts for more than half of the total energy transferred to the system , and has an uncertainty .
experimental results are shown in figure [ fig:2 ] , and have been plotted with a hartree - fock calculation for this system .
as we are only interested in the energy transferred to the system , the ground state energy present at @xmath1 is subtracted from the theoretical curve .
this allows us to make close comparisons with the specific heat of the system , which is defined as the temperature derivative of the energy per particle .
the theoretical curve for an ideal gas with finite - size effects includes a shaded region representing the uncertainty in our determination of the critical temperature @xmath12 , which is mainly due to the uncertainty in the absolute measurement of @xmath16 . notwithstanding the 15% error in @xmath16 ,
the hartree - fock numerical simulation gives a better description of the behavior of our data than an ideal gas having finite - size effects .
to support our previous evidence , which relies on a calculation of the work done on the bec , we utilize a separate method which allows for a direct measurement of the transferred energy . here , we transfer energy to the system using a single pulse of an optical standing wave , as indicated in figure [ fig:1](b ) , before allowing the system to rethermalize . using a setup analogous to
the atom - optics kicked rotor pioneered by raizen and co - workers @xcite , we apply a short 300 ns pulse to the atoms , diffracting the system into quantized momentum orders .
an additional advantage of this method is that the kohn mode is naturally not present , as the diffraction is symmetric about zero momentum .
we use a pair of counter - propagating laser beams to form our standing wave , with the beams red - detuned by @xmath78 ghz from the resonant transition , such that the effects of spontaneous emission are negligible .
the atoms are diffracted into quantized momentum states , with the @xmath11th momentum state having momenta @xmath79 , where @xmath80 is the wavenumber of the laser and @xmath11 is integer . in a separate calibration experiment
, we quantitatively measure the amount of energy transferred as a function of kick - strength , given by @xmath81 , where @xmath82 is the pulse length , @xmath83 is the rabi frequency of a single beam and @xmath84 is the detuning .
experimentally , @xmath85 is varied by controlling the intensity of the laser for each pulse using an acousto - optic modulator , and we utilise @xmath86 .
we measure the resulting momentum distribution by turning off the trap immediately after the kick and observing the atoms after a 10 ms expansion time , as shown in figure [ fig : kicktest ] .
the total energy transferred to the system is computed as @xmath87 where @xmath88 is the fraction of atoms in the @xmath11th momentum state , for integer @xmath11 .
we again scale this energy measurement by @xmath16 to remove the dependence of our energy measurement on our determination of the absolute number of atoms .
the uncertainty in @xmath89 is then due to shot - to - shot variation in intensity of the kicking laser and the temporal width of the pulse .
we perform multiple calibration runs to obtain an average transferred energy , and allow the variation to be experimentally represented as a variation in the resulting temperature after thermalization .
we find that the shot - to - shot variation in @xmath89 can be up to 10% .
once @xmath89 has been calibrated we repeat the experiment , but allow the atoms to rethermalize in the trap for 100 ms following the kick , before imaging the system via time - of - flight .
the experimental data are shown in figure [ fig:3 ] .
figure [ fig:3 ] shows both sets of experimental data on a logarithmic axis . here
we scale the temperature by the ideal gas critical temperature @xmath12 , and the energy per particle by the characteristic energy of the transition @xmath90 .
the data clearly deviates from the classical prediction of @xmath91 , where we would expect @xmath92 in each of the three potential and three kinetic degrees of freedom from the equipartition theorem .
there is also a deviation from the prediction for an ideal bose gas .
this deviation is beyond finite - size corrections @xcite , and is consistent with hartree - fock simulations of an interacting gas .
corrections to the hartree - fock approximation , such as in hartree - fock - bogoliubov - popov theory @xcite , are very small for our system , and impractical to measure .
theoretical studies have shown that the hartree - fock approximation can accurately reproduce the thermodynamic properties of a trapped bose gas @xcite , and we have confirmed through simulation that these two theories give very similar results for our system .
the specific heat is defined as the temperature derivative of the energy per particle , in our case with the external potential held constant . taking numerical derivatives of our experimental data is impractical .
we can instead make comparisons with the specific heat extracted from derivatives of the theoretical curves , shown in figure [ fig : hc ] .
we find that our experiments support the notion that the presence of interactions will tend to increase the specific heat at low temperatures when compared to an ideal bose gas .
this can can be understood as a consequence of the repulsion of the thermal atoms by a large condensate fraction .
the effective potential seen by the thermal atoms is modified to a `` mexican hat '' type potential @xcite , increasing the volume occupied by the thermal atoms , thereby increasing the density of states .
consequently , this allows the otherwise `` saturated '' thermal cloud to hold more atoms , and hence more energy .
we have directly measured the energy - temperature relationship of an interacting , harmonically trapped , ultracold bose gas .
two separate calorimetric techniques have produced similar results ; namely , that interactions lead to an increased specific heat from the ideal gas prediction , which is proportional to @xmath93 below @xmath12 .
we have performed quantitative measurements , utilising independent determinations of the energy and temperature , that are well described by hartree - fock theory .
future research could involve a thorough investigation of the effect of interactions on the specific heat by employing feshbach resonances , an investigation into ways to reduce error in the experiment , and a detailed investigation of the thermalization process .
28ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ]
+ 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) _ _ , @noop , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , ) link:\doibase 10.1063/1.3125051 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physreva.66.033606 [ * * , ( ) ] link:\doibase 10.1103/physreva.67.063614 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop , * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) | we experimentally study the energy - temperature relationship of a harmonically trapped bose - einstein condensate by transferring a known quantity of energy to the condensate and measuring the resulting temperature change .
we consider two methods of heat transfer , the first using a free expansion under gravity and the second using an optical standing wave to diffract the atoms in the potential .
we investigate the effect of interactions on the thermodynamics and compare our results to various finite temperature theories . |
quantum information has recently entered the domain of continuous variable ( cv ) systems @xcite and quantum error correction theory in particular has been extended to the cv framework @xcite .
in such a framework , the main damage induced by decoherence is typically a small diffusion in the two canonical coordinates characterizing the cv system , which may represent the position and momentum of a particle or two orthogonal field quadratures of an optical mode .
et al . _ have shown that these effects are best fought by encoding quantum information within appropriate _ shift - resistant _ quantum codes of the cv system @xcite . the relevant aspect of the proposal of ref .
@xcite is that , using qubits encoded in such a way , a universal set of fault - tolerant quantum gates can be implemented using only `` cheap '' standard devices which , in the case of a cv optical mode , correspond to beam splitters , phase shifters , squeezers , homodyne and photodetectors . in such a proposal ,
all the difficulties are confined to the preparation of the encoded states which , in the ideal limit , are superpositions of an infinite number of infinitely squeezed states and require the use of particular nonlinear interactions for their generation .
the preparation of these shift - resistant quantum codes represents the main drawback of the gottesman _
proposal for universal fault - tolerant quantum computation in a cv system and therefore it is crucial to find an experimentally viable way to generate the quantum codewords employing optical cv modes . a first attempt has been based on a conditional scheme in which the cv system is coupled to a single qubit and which could be implemented on a single trapped ion @xcite . here ,
elaborating on a suggestion in @xcite of using a nonlinear hamiltonian of ponderomotive type , we present an all - optical scheme involving two modes interacting in a crossed - kerr medium .
@xcite shows that a single qubit , whose generic state can be written as @xmath0 , can be encoded in an oscillator with canonical coordinates @xmath1 ( with @xmath2=i$ ] ) so that the corresponding logical state @xmath3 is easily correctable with respect to small diffusion errors in both position and momentum .
the codewords @xmath4 and @xmath5 are the simultaneous eigenstates with eigenvalue @xmath6 of the two displacement operators @xmath7 and @xmath8 ( the so called _ stabilizer generators _ of the code @xcite ) with arbitrary @xmath9 .
the two codewords are a coherent superposition of infinitely squeezed states both in position and momentum , that is , @xmath10 they have a comb - like structure with spacing @xmath11 in the @xmath12-space and @xmath13 in the @xmath14-space and are displaced with respect to each other by @xmath15 along the @xmath12-axis ( this is also responsible for the alternating signs in the p - representation , see fig .
[ ideal ] ) . in a cv system ,
errors and decoherence manifest themselves through random shifts of the two quadratures @xmath16 and @xmath17 . the present coding is designed for the case when these shifts are never too large , i.e. , the error is correctable only if the shift is below a certain threshold . in fact , the recovery process is realized by measuring the stabilizer generators @xmath18 , which is equivalent to measure @xmath17 ( @xmath19 ) and therefore detects momentum shifts , and @xmath20 , which is equivalent to measure @xmath16 ( @xmath21 ) and detects position shifts .
then the error is corrected by displacing the state to the nearest peak , that is , to the nearest multiple of @xmath22 along the @xmath17-axis and to the nearest multiple of @xmath15 along the @xmath16-axis .
therefore the recovery is successful only when the momentum shift @xmath23 satisfies @xmath24 and the position shift @xmath25 satisfies @xmath26 , because otherwise the displacement would recover the wrong codeword , corrupting therefore the encoded information .
an alternative choice for the codewords is the basis formed by the two superposition states @xmath27 , which are used in @xcite for the error recovery in the @xmath14-space because they have the same displaced comb - like structure of @xmath28 and @xmath29 in the @xmath14-space rather than in the @xmath12-space ( see fig . 1 ) .
practically , the two sets of codewords @xmath30 , @xmath31 are needed to measure the stabilizers @xmath18 and @xmath20 respectively @xcite . , @xmath32 , @xmath33 and @xmath34.,scaledwidth=70.0% ] in real experiments one can only create finite squeezing and , therefore , only approximate codewords , @xmath35 , @xmath36 and @xmath37 with @xmath38)\}^{1/2}$ ] . in this practical situation ,
even if we had only small position and momentum shifts satisfying the above conditions for a correct recovery , one would still have errors due to the use of approximate codewords possessing a nonzero overlap @xmath39 and @xmath40 .
these _ intrinsic errors _ manifest themselves once the measurement of a stabilizer is performed .
the quality of an experimental implementation of the error correction scheme of @xcite is therefore established by an _ intrinsic error probability _
, which can be considered as the largest error probability among those associated with the four approximate codewords @xmath41 , @xmath42 , @xmath43 .
our proposal for encoding a qubit in a cv optical mode is schematically described in fig . [ setup ] .
we consider two radiation modes @xmath44 and @xmath45 ( @xmath46=[\hat{b},\hat{b}^{\dagger}]=1 $ ] ) impinging on a nonlinear medium where they interact through the cross - kerr effect , described by the interaction hamiltonian @xcite @xmath47 where @xmath48 is the cross - kerr susceptibility of the medium .
the two modes are both prepared in coherent states , @xmath49 and @xmath50 , with @xmath51 ( we take @xmath52 and @xmath53 real from now on ) .
we are interested in the fluctuations of the @xmath45 mode around the classical intense coherent state @xmath50 , and therefore , redefining the @xmath54 mode through the displacement @xmath55 and neglecting higher order contributions , the interaction hamiltonian of eq .
( [ ham1 ] ) can be rewritten as @xmath56 if the losses within the kerr medium are negligible the system dynamics is unitary and one has that the two - mode state after an interaction time @xmath57 within the kerr medium is given by @xmath58\}\left| n\right\rangle _ { a}\left| \gamma\right\rangle _ { b}\label{evol2}\ ] ] where @xmath59 and @xmath60 respectively represent a number state for @xmath44 and a coherent state for @xmath45 . just at the exit of the kerr medium , mode
@xmath44 is subject to a homodyne measurement of the field quadrature @xmath61 with measurement result @xmath62 . as a consequence ,
the two modes are disentangled , with mode @xmath44 in the corresponding eigenstate @xmath63 of @xmath64 and mode @xmath45 in the state conditioned to the measurement result @xmath62 , which , introducing the quadratures @xmath65 and @xmath66 , can be described by the spatial and momentum wave functions @xmath67 \label{waveq } \\
\psi(p;\tau,\alpha , x)&= & \pi^{-1/2}\mathcal{n}\sum_{n=0}^{\infty}\mu_{n } ( \alpha , x)\exp[-\tfrac{1}{2}(p-\pi n\tau)^{2 } ] , \label{wavep}\end{aligned}\ ] ] where @xmath68 , @xmath69 , @xmath70 , with @xmath71 the hermite polynomials , and we have introduced a dimensionless , scaled , interaction time @xmath72 .
our proposal consists in taking this conditional state as the approximate codeword @xmath73 , i.e. , @xmath74 and @xmath75 , with the following correspondence of parameters @xmath76 .
the other approximate codeword state @xmath77 is then obtained by applying a displacement , i.e. , @xmath78 , while @xmath79 are obtained using the fault - tolerant schemes discussed in @xcite .
from the wave functions of the encoded states shown in fig . 1 it is easy to define and calculate the intrinsic error probabilities . in fact
, an intrinsic error occurs when : @xmath80 given @xmath81 , the measurement of @xmath82 ( mod @xmath15 ) gives a result inside one of the error regions @xmath83 $ ] for integer @xmath84 ; @xmath85 given @xmath86 , the measurement of @xmath16 ( mod @xmath15 ) gives a result inside one of the error regions @xmath87 $ ] ; @xmath88 given @xmath89 , the measurement of @xmath90 ( mod @xmath91 ) gives a result inside one of the error regions @xmath92 $ ] ; @xmath93 given @xmath94 , the measurement of @xmath90 ( mod @xmath91 ) gives a result inside one of the error regions @xmath95 $ ] .
therefore , the corresponding intrinsic error probabilities are given by @xmath96 , for both cases @xmath80 and @xmath85 and by @xmath97 for cases @xmath88 and @xmath93 .
as discussed above , we can take as the overall intrinsic error probability of the proposed approximate encoding the largest error probability , that is , @xmath98 $ ] .
it can be seen that this quantity is a decreasing function of the scaled interaction time @xmath99 , at least for sufficiently large values of @xmath99 .
it is therefore convenient to choose a large @xmath99 ( which means using a large enough value of the coherent amplitude @xmath53 ) and simply consider the asymptotic value @xmath100 .
the threshold value @xmath101 of the scaled interaction time beyond which the asymptotic values of the error probabilities are essentially reached can be easily evaluated ; @xmath101 depends upon @xmath52 and for @xmath102 it is @xmath103 and therefore easily achievable with a moderate value of @xmath53 .
the asymptotic error probabilities @xmath104 and @xmath105 can be evaluated analytically and are given by @xmath106 and their behavior as a function of the homodyne measurement result @xmath62 is shown in fig .
[ probs15]a .
( solid line ) , @xmath107 ( dashed line),@xmath108 ( dotted line ) versus the homodyne measurement result @xmath62 and for @xmath109 . the error probability @xmath110 is given by the largest value between these three functions and its behavior for three different values of @xmath52 is shown ( solid line ) in ( b ) ( @xmath111 ) , ( c ) ( @xmath112 ) and ( d ) ( @xmath113 ) .
the dashed line in ( b ) , ( c ) and ( d ) corresponds to the asymptotic probability density , @xmath114 , of getting the result @xmath62.,scaledwidth=90.0% ] it is evident from fig .
[ probs15 ] that the lowest intrinsic error is achieved for measurement results @xmath62 as close as possible to @xmath115 , which is the measurement result for which @xmath110 is minimum .
the probability densities in @xmath12 and @xmath14 for the four approximate codewords @xmath116 , @xmath81 , and @xmath117 corresponding to the particular measurement result @xmath115 are shown in fig .
[ fonda ] , where the parameter values @xmath118 have been chosen . a first qualitative comparison with the ideal codewords of fig . 1 , shows that the present scheme is able to generate well approximated codewords . , @xmath119 , @xmath120 for @xmath118 ( @xmath121 ) .
in ( a ) the probability densities @xmath122 ( solid line ) and @xmath123 ( dashed line ) are shifted by @xmath121 as expected , while in ( b ) the corresponding probability densities @xmath124 and @xmath125 ( solid line ) coincide as expected , with spikes at @xmath126 . in ( c ) the probability densities @xmath127 ( solid line ) and @xmath128 ( dashed line ) differ only slightly , with spikes at @xmath129 while in ( d ) their momentum probability densities @xmath130 ( solid line ) and @xmath131 ( dashed line ) are approximately shifted by @xmath132.,scaledwidth=80.0% ] it is also evident from fig .
[ probs15 ] that for values of @xmath62 of the order or larger than @xmath52 the error probability becomes unacceptably large .
therefore the coding scheme proceeds by fixing an acceptable interval centered around @xmath115 for the homodyne result , say @xmath133 , with @xmath134 , and accepting the resulting approximate codeword @xmath81 only if @xmath62 lies within this interval .
in such a case one has two relevant quantities : @xmath80 the _ success probability _ of the coding scheme @xmath135 , where @xmath136 is the probability density of getting the result @xmath62 in the homodyne measurement ( referred to the asymptotic limit of large @xmath99 ) ; @xmath85 the corresponding _ mean intrinsic error probability _ @xmath137 .
the success probability @xmath138 gives the fraction of times a satisfactory approximate codeword is generated .
the mean intrinsic error probability @xmath139 is the most relevant quantity since it measures the effective quality of the generated codewords .
the behavior of these two probabilities is plotted as a function of @xmath140 for different values of @xmath52 in fig .
[ duobis ] .
it is easy to see that for @xmath141 , which means accepting all measurement results , @xmath142 , i.e. the encoding is too bad and can not be used for quantum information processing .
it is also evident that the best results are achieved for vary large @xmath140 , i.e. , when only values very close to @xmath115 are accepted , where however the success probability becomes lower .
in general the value of the coherent amplitude @xmath52 has not to be either too large or too small .
in fact , if it is too small , the intrinsic error probability is always too large ( see fig . [ duobis ] for @xmath143 ) , while if it is too large , @xmath139 stays around the value @xmath144 for a large interval of @xmath140 ( see fig .
[ duobis ] for @xmath145 ) due to the oscillating behavior of @xmath110 ( see figs . [
probs15](c ) and [ probs15](d ) ) .
this fact implies that one has to use very large values of @xmath140 to achieve an acceptable intrinsic error , corresponding however to a very low success probability .
we see from fig .
[ duobis ] that one has a good compromise between success probability and intrinsic error probability for intermediate @xmath52 , such as , for example , @xmath146 .
as an example , when @xmath147 and @xmath148 one has a mean intrinsic error probability @xmath149 and a success probability @xmath150 .
( dashed line ) and mean intrinsic error probability @xmath139 ( solid line ) versus @xmath140 for @xmath151.,scaledwidth=90.0% ]
we have presented an all - optical conditional scheme able to generate the quantum codewords of the cv encoding scheme proposed in ref .
@xcite for fault tolerant quantum computation .
the generation of the encoded states is the most difficult part of that proposal and necessarily involves a nonlinear interaction .
the present scheme involves two travelling beams initially prepared in coherent states interacting in a nonlinear cross - kerr medium and is feasible with present technologies . | we show how a qubit can be fault - tolerantly encoded in the infinite - dimensional hilbert space of an optical mode .
the scheme is efficient and realizable with present technologies .
in fact , it involves two travelling optical modes coupled by a cross - kerr interaction , initially prepared in coherent states , one of which is much more intense than the other . at the exit of the kerr medium ,
the weak mode is subject to a homodyne measurement and a quantum codeword is conditionally generated in the quantum fluctuations of the intense mode . |
in the last several decades many new materials with intriguing properties were discovered and synthesized .
these materials range from the high - temperature superconductors to magnetic materials , and the latter have already found real applications in electronic industry .
typically , most of these materials contain elements from the 3@xmath0 or 4@xmath3 rows and their electronic structure is characterized by the presence of a partially filled narrow band across the fermi level .
the fact that the narrow band is partially filled implies that there are many configurations with approximately equal weight rendering a one - particle description of the electronic structure problematic .
indeed , it has been recognized for a long time that many of the intriguing properties of these materials originate from correlations among the electrons residing in the partially filled narrow band .
the electrons are neither fully localized , like core electrons , nor itinerant , like @xmath4 or @xmath5 electrons in alkalis or conventional semiconductors such as silicon or diamond .
this hybrid property poses a tremendous theoretical difficulty for an accurate description of the electronic structure , because due to the electrons partially itinerant character the problem can neither be treated as a purely atomic problem nor within a pure band picture .
moreover , the interaction with other electrons can be very important .
a large amount of work has been directed to solving the correlation problem of the above materials .
the usual approach is to consider only the narrow bands near the fermi level and eliminate the degrees of freedom for the rest of the bands by the downfolding procedure , resulting in the well - known hubbard model which contains an effective on - site coulomb interaction , the hubbard @xmath6 . in general , the models represent multi - band systems containing interorbital as well as long - ranged part of coulomb interaction .
the models can then be solved with various sophisticated low - energy solvers such as dynamical mean - field theory ( dmft ) @xcite or solvers for lattice models @xcite .
an important issue in mapping the real system to a model hamiltonian is how to determine the one - particle kinetic energy term and the effective interaction or the hubbard _ u _ in the model . unlike the one - particle parameters that can be downfolded from the band structure , the hubbard _ u _ is much more elusive to determine and it is often treated as an adjustable parameter . a widely used scheme to calculate the hubbard _ u _ from first - principles is the constrained lda ( clda ) method @xcite .
the clda method , however , is known from early on to yield values of @xmath6 , which are too large in some cases ( e.g. late transition metals ) .
it has been argued that this arises from technical difficulty in including transitions of electrons between the @xmath0 and @xmath1 space contributing the screening processes .
this oversight leads in some cases to a larger value of _ u _ @xcite .
recent extensions of the clda method may be found in refs.@xcite and @xcite .
another method for determining the effective interaction is a scheme based on the random phase approximation ( rpa ) .
early attempts along this direction can be found in refs.@xcite . a combined clda and rpa method to circumvent the difficulty was also proposed @xcite .
some years ago a scheme for calculating the hubbard _ u _ , called the constrained rpa ( crpa ) scheme @xcite , was proposed .
the main merit of the crpa method over currently available methods is that it allows for a precise elimination of screening channels , which are instead to be included in a more sophisticated treatment of the model hamiltonian .
this is a controlled approximation without any ambiguity , expected to become asymptotically exact if the @xmath1-space becomes well separated from the @xmath0 space .
moreover , the effective screened interaction can be calculated as a function of @xmath7 and @xmath8 , i.e. , @xmath9 , independent of the basis functions .
this allows easy access to obtaining not only on - site matrix elements but also off - site matrix elements as well as screened exchange matrix elements , which are usually taken to be the atomic value .
another merit is the possibility of obtaining the frequency - dependent hubbard _ u _ , which may prove to be important .
the crpa method has now been applied to a number of systems with success @xcite .
although the crpa method is rather general , its applications to real systems have revealed a serious technical problem .
the problem arises when the narrow band is entangled with other bands , i.e. , it is not completely isolated from the rest of the bands .
in many materials , the narrow band of interest is entangled . even in simple materials such as the 3@xmath0 transition metals ,
the 3@xmath0 bands mix with the 4@xmath4 and 4@xmath5 bands .
similarly , the 4@xmath3 bands of the 4@xmath3 metals hybridize with the more extended @xmath4 and @xmath5 bands . for such cases ,
it is not clear anymore which part of the polarization should be eliminated when calculating the hubbard @xmath6 using the crpa method .
some procedures to overcome the problem of determining _ u _ for entangled bands have been attempted .
one of these is to choose a set of band indices and define the bands of hubbard model as those bands corresponding to the chosen indices .
another alternative is to introduce an energy window and define the hubbard bands to be those that have energy within the energy window .
yet another alternative is to have a combination of energy window and band indices .
these procedures , however , suffer from a number of difficulties . when choosing band indices it is inevitable that some of the states will have a character very different from that of the intended model .
for example , in the case of 3@xmath0 transition metals , choosing five 3@xmath0 " bands will include at some @xmath10-points states which have little 3@xmath0 character , with a considerable 4@xmath4 component instead .
moreover , the chosen bands will be awkward to model since they do not form smoothly connected bands .
a similar problem is encountered when choosing an energy window . a combination of band indices and energy window proposed in ref .
@xcite partially solves the problem but it suffers from arbitrariness .
another procedure is , as we will discuss in detail later , to project the polarization to the orbitals of interest , e.g. , 3@xmath0 orbitals , but this procedure has been found to yield an unphysical result of negative static _
u_. in this work , we offer a solution to the problem of determining the hubbard _ u _ for entangled bands .
the basic idea is to disentangle the narrow bands of interest from the rest and carry out the crpa calculation for the disentangled band structure , not using the original band structure .
the disentangling procedure is described in sec.[sec : method ] .
we apply the method to 3@xmath0 transition metals in sec.[sec : result ] and show that the method is numerically stable and yields reasonable values of @xmath6 .
finally the paper is summarized in sec.[sec : summary ] .
in the crpa method we first choose a one - particle subspace @xmath11 , which defines the model hamiltonian , and label the rest of the hilbert space by @xmath12 .
we define @xmath13 to be the polarization within the @xmath0 subspace and the total polarization is written as @xmath14 .
it is important to realize that the rest of the polarization @xmath15 is _ not _ * * the same as the polarization of the @xmath1 subspace because it contains polarization arising from transitions between the @xmath0 and @xmath1 subspaces .
since @xmath13 is the polarization of the model hamiltonian , this polarization should be subtracted out from the total polarization when the effective parameter of the model is determined .
the effective coulomb interaction @xmath16 should be calculated with the rest of the polarization @xmath17 : @xmath18^{-1}v\label{wr } \;,%\ ] ] where @xmath19 is the bare coulomb interaction .
it can indeed be shown @xcite that the fully screened interaction is given by @xmath20^{-1}w_{r } \;.\ ] ] this mathematical identity ensures that @xmath16 can be interpreted as the effective interaction among the electrons residing in the @xmath0 subspace since the screening of @xmath16 by @xmath13 leads to the fully screened interaction .
the matrix elements of @xmath16 in some localized functions can then be regarded as the frequency - dependent hubbard _ u_. it has been shown that the formula in eq.([wr ] ) is formally exact , provided @xmath17 is the difference between the exact polarization @xmath14 and the exact polarization of the @xmath0 subspace @xmath13 . in the crpa method , @xmath17 is calculated within the random - phase approximation .
if the @xmath0 subspace forms an isolated set of bands , as for example in the case of the @xmath21 bands in srvo@xmath22 , the crpa method can be straightforwardly applied .
however , in practical applications , the @xmath0 subspace may not always be readily identified .
an example of these is provided by the 3@xmath0 transition metal series where the 3@xmath0 bands , which are usually taken to be the @xmath0 subspace , do not form an isolated set of bands but rather they are entangled with the 4@xmath4 and 4@xmath5 bands . to handle these cases
we propose the following procedure .
we first construct a set of localized wannier orbitals from a given set of bands defined within a certain energy window .
these wannier orbitals may be generated by following the post - processing procedure of souza , marzari and vanderbilt @xcite or other methods , such as the preprocessing scheme proposed by andersen _
et al . _ within the nth - order muffin - tin orbital ( nmto ) method @xcite .
we then choose this set of wannier orbitals as the generators of the @xmath0 subspace and use them as a basis for diagonalizing the one - particle hamiltonian , which is usually the kohn - sham hamiltonian in the local density approximation ( lda ) or generalized gradient approximation ( gga ) .
the so obtained set of bands , which equivalently define the @xmath0 subspace , may be slightly different from the original bands defined within the chosen energy window , because hybridization effects between the @xmath0 and @xmath1 spaces are neglected .
however , it is important to confirm that the dispersions near the fermi level well reproduces the original kohn - sham bands . from these bands
we calculate the polarization @xmath23 as @xmath24 \ ; , \label{eq : pd}\ ] ] where @xmath25 , @xmath26 @xmath27 are the wavefunctions and eigenvalues obtained from diagonalizing the one - particle hamiltonian in the wannier basis .
it would seem sensible to define the rest of the polarization as @xmath28 where @xmath14 is the full polarization calculated using the _ original _ ( kohn - sham ) wavefunctions and eigenvalues @xmath29 , @xmath30 @xmath31 , and calculate @xmath16 according to eq.([wr ] ) .
we have found , however , that this procedure is numerically very unstable , resulting in some cases to unphysically negative static @xmath6 and a large oscillation as a function of frequency .
this is understandable given that @xmath14 and @xmath32 are obtained from two different band structures , so that low energy screening channels associated with the @xmath0-@xmath0 transitions are not excluded from @xmath33 completely .
due to the singular nature of the expression in eq.([wr ] ) these low - energy excitations can cause a large fluctuation in @xmath16 .
another way of calculating @xmath33 is to project the wavefunctions to the @xmath0 space , @xmath34 where the projection operator @xmath35 is defined as @xmath36 the effective @xmath0 polarization may be expressed as @xmath37 \;,\ ] ] and @xmath38 can be used to calculate @xmath39 .
we found that this procedure does not work either and is again unstable .
this problem may be related to the fact that @xmath40 s are not orthogonal with each other and transitions between the states do not correspond to single particle - hole excitations .
based on these observations we propose the following procedure .
we define the @xmath1 subspace by @xmath41 which is orthogonal to the @xmath0 subspace constructed from the wannier orbitals . in practice
it is convenient to orthonormalize @xmath42 and prepare @xmath43 basis functions . by diagonalizing the hamiltonian in this subspace a new set of wavefunctions @xmath44 and eigenvalues @xmath45
@xmath46 are obtained . as a consequence of orthogonalizing @xmath47 and @xmath48 , the set of @xmath1 bands @xmath45 are completely disentangled from those of the @xmath0 space @xmath49 , and they are slightly different from the original band structure @xmath50 .
as we will see later , however , the numerical tests show that the disentangled band structure is close to the original one .
the hubbard @xmath6 is calculated according to eq.([wr ] ) with @xmath51 , where @xmath52 is the full polarization calculated for the _ disentangled _ band structure .
it is important to realize that the screening processes between the @xmath0 space and the @xmath1 space are included in @xmath6 , although the @xmath0-@xmath1 coupling is cut off in the construction of the wavefunctions and eigenvalues .
as an illustration we apply the method to 3@xmath0 transition metals .
the electronic structure calculations are done in the local density approximation @xcite of density functional theory @xcite with the full - potential lmto implementation @xcite .
the wavefunctions are expanded by @xmath53 mtos and a @xmath54 * k*-mesh is used for the brillouin zone summation .
spin polarization is neglected .
more technical details are found elsewhere @xcite .
figure [ fig : band](a ) shows the kohn - sham band structure of nickel .
there are five orbitals having strong 3@xmath0 character at [ -5 ev:1 ev ] , crossed by a dispersive state which is of mainly 4@xmath4 character . using the maximally localized wannier function prescription with the energy window of [ -7 ev:3 ev ] , interpolated `` @xmath0 ''
bands are obtained .
the subsequent disentangling procedure gives the associated `` @xmath1 '' bands .
comparing fig.[fig : band](b ) with ( a ) we can see that there is no anti - crossing between the @xmath0 bands and the @xmath1 bands in ( b ) .
otherwise the two band structures are nearly identical . in order to see the impact of the disentanglement on the screening effects ,
we perform the full rpa calculation using the disentangled band structure .
the fully screened coulomb interaction is compared with that for the original band structure in fig.[fig : w ] where the average of the five diagonal terms in the wannier basis @xmath55 is plotted , @xmath56 the two methods yield similar results the static values agree with each other within 0.2 ev , and the frequency dependence is weak at low frequencies . as frequency increases
there is a sharp increase at @xmath5720 ev , where screening by plasmons becomes ineffective .
these results assure that screening effects can be treated accurately with the disentangled band structure .
the hubbard @xmath6 is calculated by the constrained rpa , namely , by replacing @xmath58 in eq.([eq : w ] ) with @xmath39 .
the results are shown in fig.[fig : u ] .
there is no large fluctuation against frequency , in contrast to the methods described in sec.[sec : method ] , and @xmath59 shows a stable behavior .
as is expected , @xmath6 is significantly larger than @xmath58 at low frequencies .
this implies proper elimination of @xmath60 screening processes is crucial .
comparing with the previous results using a combined energy and band window @xcite , the agreement is reasonably good .
a small difference between these two results at low frequency may be due to a small portion of @xmath0-@xmath0 screening presumably contained in the previous method @xcite , although it should be excluded from the crpa calculations .
we carried out the calculations for a series of other 3@xmath0 metals as well and found in all the cases that ( i ) the present scheme is numerically stable and does not result in unphysical frequency dependence of @xmath6 , and ( ii ) the value of @xmath61 is close to that from the original band structure .
the latter is confirmed in fig.[fig : static ] where the static values ( @xmath62 limit ) are summarized . concerning @xmath6 , the present method gives larger values compared to the previous results , particularly for early transition metals .
since @xmath58 is nearly equal to each other in the two methods , the discrepancy is ascribed to the different treatment between @xmath2 and @xmath63 .
we should note that @xmath2 in the previous method depends on the choice of the window . for a wider window ,
obviously we would obtain a better agreement .
also , some states have a mixed character of 3@xmath0 and 4@xmath4 near the anti - crossing points .
this makes elimination of the screening process difficult in the original band structure . the present scheme , on the other hand , enables us to determine @xmath6 without ambiguity .
the @xmath0 bands are disentangled from the @xmath1 bands . consequently , the polarization in the @xmath0 space is well - defined and can be removed completely in @xmath64 . in the present formulation ,
small off - diagonal matrix elements of the kohn - sham hamiltonian between the @xmath0 space wavefunction @xmath65 constructed from the wannier orbitals and the @xmath1 space @xmath66 are ignored .
this is the reason why the anti - crossing is avoided .
if the energy of this hybridization point in the band dispersion is smaller than the screened coulomb energy and the energy scale of the interest , strictly speaking , one has to keep all of these hybridizing bands in the effective model , because the hybridization effects are non - perturbative . in the present case of transition metals , the energy crossing point of 4@xmath4 and 3@xmath0 bands
are relatively larger than the screened coulomb energy scale and the low energy models constructed only from the 3@xmath0 wannier orbitals may give at least qualitatively reasonable description of the low energy physics .
we have proposed a method to calculate the effective interaction parameters for the effective low - energy models of real materials when bands are entangled .
the key point is to first properly orthogonalize the low - energy subspace contained in the models and the complementary high - energy subspace to each other .
this orthogonalization by the projection technique enables the disentanglement of the bands .
once the disentangled band structure is obtained , the constraint rpa method can be used to determine the partially screened coulomb interaction uniquely .
numerical tests for 3@xmath0 metals show that the method is stable and yields reasonable results .
the method is applicable to any system .
applications to more complicated systems , such as interfaces of transition metal oxides are now under way .
we thank s. biermann for fruitful discussions .
this work was supported by grants - in - aid for scientific research from mext , japan under grant numbers 17064004 and 19019013 .
tm acknowledges support from next generation supercomputing project , nanoscience program , mext , japan .
for example , the path - integral renormalization group method has been applied by y. imai , i. solovyev and m. imada , phys .
95 * , 176405 ( 2005 ) ; y. imai and m. imada , j. phys .
. jpn . * 75 * , 094713 ( 2006 ) ; y.
otsuka and m. imada , j. phys .
. jpn . * 75 * , 124707 ( 2006 ) .
-@xmath1 hybridization switched off .
the red lines show the @xmath0 states obtained by the maximally localized wannier scheme , while the blue lines are disentangled @xmath1 states .
energy is measured from the fermi level .
, title="fig:",width=340 ] -@xmath1 hybridization switched off .
the red lines show the @xmath0 states obtained by the maximally localized wannier scheme , while the blue lines are disentangled @xmath1 states .
energy is measured from the fermi level .
, title="fig:",width=340 ] of nickel as a function of frequency .
the diagonal term of the partially screened interaction in the wannier basis is calculated by the present method and compared with the published data of ref.@xcite .
, width=340 ] | we present a first - principles method for deriving effective low - energy models of electrons in solids having entangled band structure .
the procedure starts with dividing the hilbert space into two subspaces , the low - energy part ( `` @xmath0 space '' ) and the rest of the space ( `` @xmath1 space '' ) . the low - energy model is constructed for the @xmath0 space by eliminating the degrees of freedom of the @xmath1 space .
the thus derived model contains the strength of electron correlation expressed by a partially screened coulomb interaction , calculated in the constrained random - phase - approximation ( crpa ) where screening channels within the @xmath0 space , @xmath2 , are subtracted .
one conceptual problem of this established downfolding method is that for entangled bands it is not clear how to cut out the @xmath0 space and how to distinguish @xmath2 from the total polarization .
here , we propose a simple procedure to overcome this difficulty . in our scheme , the @xmath0 subspace is cut out from the hilbert space of the kohn sham eigenfunctions with the help of a procedure to construct a localized wannier basis .
the @xmath1 subspace is constructed as the complementary space orthogonal to the @xmath0 subspace .
after this disentanglement , @xmath2 becomes well defined . using the disentangled bands , the effective parameters are uniquely determined in the crpa .
the method is successfully applied to 3@xmath0 transition metals . |
nucleation of bubbles or droplets of the stable phase is a very common phenomenon which initiates relaxation of metastable states near the first order phase transition in many systems of condensed matter physics @xcite , quantum fields @xcite , and cosmology @xcite .
the main quantity of interest for the theory and applications is the nucleation time , which characterizes the time required by the system to leave the metastable state . in lack of a consistent theory of the metastability
, the nucleation problem has been subjected to extensive study in computer simulations .
numerical data are usually interpreted in terms of the phenomenological classical nucleation theory , or in terms of its field - theoretical version proposed by langer @xcite .
a great deal of this work has been carried out on the ising model supplied with some appropriate dynamics .
it seems reasonable , however , to extend the numerical study of the nucleation to more complex and rich models , taking into account a great variety of physical realizations of this phenomenon . in this paper , we have studied nucleation in the two - dimensional @xmath0-state potts model by use of the monte - carlo simulation . due to a rich equilibrium phase structure
, this model provides a lot of possibilities to form the initial metastable phase . for @xmath2
, the zero - field potts model has the first order phase transition point driven by the temperature @xcite .
near this point , the metastable state can be prepared by a quenched cooling from the high to low temperature phase .
relaxation of such a state in different regimes has been studied numerically by safran _ et .
@xcite , and by arkin _ et .
al _ @xcite .
meunier and morel @xcite analysed metastability in the thermally quenched potts model within the fisher s droplet picture approach @xcite .
al _ confirmed @xcite , that in a case of slight quench ( small @xmath3 ) , the equilibrium proceeds through the usual nucleation mechanism .
however , lack of the macroscopic order parameter ( like the average magnetization in the ising system below @xmath4 ) in the both initial and final states hinders precise measurement of the nucleation time in this regime .
we simulate relaxation of the metastable states , which are induced in the potts ferromagnet near the first order phase transition driven by the external magnetic field @xmath5 .
the heat - bath dynamics is used to thermalize spins .
nucleation time is measured as the function of the applied field at @xmath1 for @xmath6 .
one can say , that for small @xmath0 , @xmath7 , the line @xmath8 in the potts model represents the liquid - vapour phase transition , and it simulates the liquid - solid transition for large @xmath0 , @xmath2 .
it should be noted , that the langer s theory of nucleation @xcite based on the coarse - grained free - energy functional can not be applied in the latter case @xcite .
so , our aim is to identify in the potts model the nucleation regimes well - known in the ising model , and to study how they are modified by increase of @xmath0 .
we study the @xmath0-state ferromagnetic potts model on the square lattice having @xmath9 rows and @xmath9 columns .
the discrete spin variable @xmath10 is associated with each lattice site @xmath11 .
helical boundary conditions are used @xcite .
the model hamiltonian is defined as @xmath12 here the first summation is over the nearest neighbour pairs , @xmath13 is the coupling constant , @xmath5 is the external magnetic field applied along the @xmath14-direction , and @xmath15 we use also dimensionless parameters @xmath16 , and @xmath17 , where @xmath18 is the inverse temperature .
the macroscopic state of the lattice can be described by @xmath0 real positive order parameters @xmath19 @xmath20 the model undergoes ferromagnetic phase transition at the critical temperature @xmath21 the ferromagnetic low - temperature phase at zero field is @xmath0-times degenerated .
different ferromagnetic phases are enumerated by the discrete parameter @xmath22 .
this implies that in the @xmath22-phase , the orientation @xmath22 dominates in the lattice . for a review of many others known properties of the potts model see @xcite
the classical nucleation theory ( cnt ) considers thermally induced generation of the stable phase droplets in the metastable surrounding .
a droplet of radius @xmath23 can appear with probability @xmath24 proportional to the boltzmann factor @xmath25,\ ] ] where @xmath26 is the droplet free energy of the droplet has an additional factor 2 due to a different definition of the magnetic field @xmath5 . ] : @xmath27 here @xmath28 is the surface tension on the droplet boundary , @xmath29 is the zero - field spontaneous magnetization .
the critical droplet radious @xmath30 maximizes the free energy ( [ fnuc ] ) .
it is postulated in cnt , that the nucleation time @xmath31 is inversely proportional to the probability of generation of the critical droplet : @xmath32= l^{-2 } \exp [ \pi(\beta s)^2/(m\ , h ) ] .
\label{istime}\ ] ] the classical nucleation theory considers only spherical droplets .
this is reasonable , if the surface tension of the domain wall does not depend on its is orientation . if the surface tension is anisotropic , the critical droplet are supposed to have the equilibrium shape given by the wulff s construction @xcite . in this case equation ( [ istime ] ) should be modified to @xmath33 , \label{antime}\ ] ] where the amplitude @xmath34 is determined by the equilibrium droplet shape . in the ising model , this shape is given by the equation @xcite @xmath35 where @xmath36 denote descartes coordinates on the droplet boundary .
it varies from circle at @xmath37 to square at @xmath38 .
the amplitude @xmath34 can be written as @xmath39 , where @xmath40 denotes the area bounded by the curve ( [ shape ] ) , and @xmath41^{-4}\}^{1/8}.\ ] ] numerically we have for the @xmath42 state potts ( ising ) model at @xmath1 : @xmath43 it should be noted , that equation ( [ antime ] ) gives only the leading @xmath44 asymptotics of the nucleation time .
the most important correction is expected to have the form of the prefactor with a power - law @xmath45-dependence @xcite @xmath46 .
\label{tcor}\ ] ] the exponent @xmath47 here arises in the field theoretical nucleation theory from the contribution of the capillary waves on the critical droplet surface @xcite , @xmath48 in the two - dimensional ising model .
the exponent @xmath49 gives the @xmath45 variation of the so - called kinetic prefactor , which depends on the chosen dynamics . for dynamics which can be described by the fokker - plank equation , the value @xmath50 is expected @xcite . the exact equilibrium ( wulff s ) shape of the droplet is not known for the potts model with @xmath51 .
so , we had to use more simple and less accurate estimates for the amplitude @xmath34 .
we have used square and circle approximations for the critical droplet shape .
the surface tension in the both cases was replaced by its zero - temperature value @xmath52 .
this yields for the amplitude @xmath34 in the nucleation time exponent : @xmath53 numerical values for the amplitude @xmath34 calculated from ( [ a ] ) for different @xmath0 at @xmath1 are given in table [ ta ] . .amplitude
@xmath34 at @xmath1 for square and circle droplets . [ cols="^,^,^",options="header " , ] the size - dependent behaviour ( [ antime ] ) of the nucleation time is observed in not very large lattices , in the so - called nucleation regime . in this regime , only one droplet of the stable phase appears in the system , and then expands to the whole lattice . at large @xmath9 ,
the coalescence regime is realized . in this regime
the relaxation time does not depend on @xmath9 .
many critical droplets appear in the system , expand and coalesce .
it is expected from the kolmogorow - johnson - mehl - avrami theory @xcite , that @xmath54 \label{coa}\ ] ] in the coalescence regime in two - dimensional systems .
in the potts model , the applied field and the initial metastable configuration of spins in the lattice can be chosen in many different ways .
we have studied only two of them . * * first regime*. a positive magnetic field @xmath55 is applied along the @xmath14-direction , whereas all the spins are initially oriented along the @xmath56-directions : @xmath57 .
such a field removes completely degeneracy of the low temperature phase , leaving only one equilibrium ferromagnetic state : @xmath58 . * * second regime*. a negative magnetic field @xmath59 is applied along 2-direction , and all spins are initially oriented in the same direction @xmath60 .
the ferromagnetic phase still remains @xmath61-times degenerated ( if @xmath62 ) in such a field . in this regime , the system moves to some bifurcation point , and then relaxes to one of the equivalent @xmath61 equilibrium states .
the heat - bath technique ( glauber kinetics ) has been used to orient the spins . to produce a sequence of pseudorandom numbers , we have used the standard generator of pseudorandom numbers with multiplication by 16807 to give 64-bit integers .
the candidate spins to flip have been chosen sequentially : site by site in the row , and row by row in the lattice .
we define the nucleation time @xmath63 as the number of sweeps through the lattice after which the number of spins in the initial ( metastable ) phase reduces to @xmath64 .
the @xmath45-dependence of the nucleation time @xmath63 has been measured at @xmath65 , and @xmath66 . in the nucleation regime
, the nucleation time fluctuates very much .
so , we had to average the nucleation time in this regime over 100 or even 1000 runs .
it is important to note , that for @xmath51 the relation between the nucleation time @xmath63 and the total relaxation time @xmath67 is not straightforward and is different in different regimes .
a typical time dependence of the order parameters @xmath19 in the first nucleation regime for @xmath68 is shown in fig .
1 . since the fall of the magnetization @xmath69 is very rapid compared with the time which the system has spent in the metastable state , the nucleation time @xmath63 and the total relaxation time @xmath67 are essentially the same in this regime .
this is true also in the second nucleation regime . in the latter case , however , the system can fall with equal probabilities to one of the equilibrium states , either @xmath70 , or @xmath71 .
2 shows a typical time evolution of the magnetization in the first coalescence regime . up to the saturation at 1 ,
the magnetization @xmath72 increases approximately linearly with time due to the steady expansion of many stable phase droplets .
thus , the system reaches equilibrium in about two nucleation times : @xmath73 .
the situation is quite different in the second coalescence regime , see fig .
3 . in this case
many critical droplets of the both equilibrium phases @xmath74 and @xmath70 are initially generated in the lattice .
they expand and expel the metastable @xmath58 phase from the system in about two nucleation times .
the resulting state is still far from the equilibrium , since the lattice is divided into domains of two stable phases @xmath70 and @xmath71 . during the final ( very slow at low temperatures ) stage of the relaxation ,
these two phases compete , until one of them will expel another from the system .
so , the nucleation time @xmath63 and the total relaxation time @xmath67 may be of different order of magnitude in the second coalescence regime .
our simulations were restricted to the temperature @xmath1 and the lattice sizes @xmath65 and @xmath66 .
we have started from the @xmath42 potts model , which is equivalent to the ising model .
this provides a good possibility to verify efficiency of the applied simulation scheme , since nucleation in the ising model has been thoroughly studied @xcite .
figure 4 shows the nucleation time in the logscale for the @xmath42 potts ( ising ) model .
pluses correspond to the nucleation time averaged over 1000 runs in the @xmath65 lattice .
stars relate to the nucleation time in the @xmath66 lattice .
crosses indicate the dispersion @xmath75 $ ] of the nucleation time distribution in the @xmath65 lattice : @xmath76=(\langle t^2\rangle - \langle t\rangle^2)^{1/2},\ ] ] where @xmath77 denotes averaging over different runs .
three relaxation regimes are clearly identified : the strong field regime ( sfr ) , the coalescence regime ( cr ) , and the nucleation regime ( nr ) .
increase of the lattice size does not change the nucleation time in the strong field and coalescence regimes .
the nucleation regime was not observed in the @xmath66 lattice in the studied region of fields @xmath45 . in the nucleation regime
, the nucleation time fluctuates very much , and the distribution @xmath78 of these fluctuations is described by the simple exponential law : @xmath79 for this distribution , the averaged time equals to the dispersion : @xmath80= \tau $ ] .
this is indeed observed in the @xmath65 lattice at @xmath81 indicating well established nucleation regime in this field region .
the theoretical prediction for the averaged nucleation time in the nucleation regime with the wulff s shape droplet is given by equation ( [ tcor ] ) , where @xmath82 , and @xmath83 .
the upper straight line in fig .
4 shows the nucleation theory exponent , and the second ( curved ) line plots the the same exponent corrected by the prefactor @xmath84 , i.e. equation ( [ tcor ] ) .
though the latter curve contains only one free parameter - the constant factor @xmath85 - it fits well the numerical simulation points for the @xmath65 lattice in the nucleation regime , much better then the first ( straight ) line .
the curved line gives also a good fit for the dispersion points in the small @xmath65 lattice not only in the nucleation region , but also at crossover to the coalescence regime down to @xmath86 .
the possible explanation of this result may be the following .
measured nucleation time @xmath63 consists of two terms @xmath87 , which have the same order of magnitude in the crossover region .
the first term relates to the time of nucleation of the first critical droplet in the system , and the second term denotes the time of the expansion of this droplet to @xmath88 of the whole lattice .
the first term should strongly fluctuate according to the exponential law ( [ texp ] ) .
if the second term @xmath89 fluctuates much weaker , one should have for the total time dispersion : @xmath76=d[t_n+t_e ] \approx d[t_n]=\langle t_n \rangle = \tau,\ ] ] where the averaged time @xmath90 of the first droplet nucleation is still given by equation ( [ tcor ] ) .
thus , in the crossover region , dependence ( [ tcor ] ) should fit better the time dispersion points @xmath75 $ ] , than the averaged time @xmath91 , which is indeed observed .
the first numerical verification of the magnetic field dependence of the prefactor in ( [ tcor ] ) for the two - dimensional ising model has been reported by rikvold _
al _ @xcite . in this work ,
the value @xmath82 was confirmed at @xmath92 for the metropolis dynamics with randomly chosen sites .
however , a smaller value @xmath93 was observed in @xcite , when the sites in the lattice were chosen sequentially . though we have also used the sequential chose of sites in the lattice ,
our data for the nucleation time in the nucleation region are better fitted by ( [ tcor ] ) with the prefactor exponent @xmath82 , rather then @xmath94 , see fig .
this is more clear for the dispersion @xmath75 $ ] ( crosses in fig .
5 ) , since the latter is still described by ( [ tcor ] ) at the crossover to the coalescence regime .
the lower straight line is the best linear fit for the nucleation time in the larger @xmath66 lattice in the coalescence regime ( cr ) .
the ratio of slopes @xmath95 of two straight lines in fig . 4 corresponding to the nucleation and coalescence regimes is much smaller than @xmath96 , expected from the theory in the small @xmath45 limit .
however , we are still far from the well - established small @xmath45 limit , as it is evident from the considerable difference between first and second lines in fig .
if one takes this into account , this disagreement of the observed @xmath95 value with @xmath96 expected in the @xmath97 limit becomes not so surprising .
figures 6 - 10 show the observed dependence of the nucleation time on the inverse field @xmath98 in the @xmath99 potts models , @xmath100 in the first @xmath101 and second @xmath102 regimes .
pluses correspond to the nucleation time averaged over 100 runs in the smaller lattice .
crosses indicate nucleation time in the @xmath66 lattice . only the second regime @xmath102 has been studied in the @xmath103 potts model , since the nucleation time in the first regime @xmath101 becomes too long in this case , see equation ( [ tratio ] ) below .
qualitatively , the magnetic field dependence of the nucleation time in all cases is similar to that in the ising model . the same three field regions ( srf , nr , cr ) have been observed . as in the ising case ,
the nucleation regime was realized only in the smaller lattice , @xmath65 . in the latter regime
, the relaxation time strongly fluctuates . in all cases , @xmath104 behaves linearly with @xmath105 in the nucleation regime ( nr ) , in agreement with the classical nucleation theory . in the @xmath68 and @xmath106 systems the observed slopes ( @xmath107 ) in nr agree with the circle shape of the critical droplets
however , in the @xmath103 model our data indicate the square critical droplet shape , see fig .
10 , and table [ ta ] .
the ratio of slopes in nucleation and coalescence regimes varies from @xmath108 in the @xmath68 , @xmath101 case , down to @xmath109 in the @xmath106 , @xmath102 case . as in the @xmath42 model , these values are lower than @xmath96 predicted by the theory . in the first nucleation regime ( @xmath101 ) , the nucleating critical droplet should be of the single equilibrium phase @xmath58 , whereas in the second nucleation regime ( @xmath102 ) , a critical droplet of either of @xmath61 equilibrium phases @xmath110 provides relaxation .
so , it is naturally to expect , that nucleation time @xmath111 in the first nucleation regime will be longer , that the nucleation time @xmath112 in the second nucleation regime , and @xmath113 the observed ratios @xmath114 determined from the circle droplet fits in fig .
6 - 9 @xmath115 are in a good agreement with ( [ tratio ] ) .
we have simulated nucleation in the @xmath0-state potts model on the square lattice for @xmath6 .
the magnetic field dependence of the nucleation time have been measured in two regimes ( @xmath101 and @xmath102 ) at @xmath116 for two lattice sizes @xmath65 and @xmath66 .
the strong field regime , coalescence regime , and nucleation regime are clearly identified .
the observed magnetic field dependencies of the nucleation time is in agreement with the classical nucleation theory , if one takes into account the anisotropy of the critical droplet shape .
our numerical data indicate , that the shape of the critical droplet changes from ( roughly ) circle towards square with increasing @xmath0 at the fixed temperature @xmath1 .
a similar result was observed by bikker _
@xcite in numerical calculations of the equilibrium crystal shape in the three - dimensional potts model .
so , increase of @xmath0 has a similar effect on the equilibrium droplet shape , as decrease of temperature .
this is natural , since the both reduce the correlation length , which should increase anisotropy in the system .
the ratio of ( @xmath117 ) slopes in nucleation and coalescence regimes does not approach 3 , predicted by the theory in the @xmath118 limit .
we associate this discrepancy with the finite-@xmath45 corrections , which are still considerable in the studied region of the magnetic field . though @xmath119 and @xmath120 state potts models have different type phase transitions at @xmath4 ( continuous for @xmath119 and first order for @xmath120 ) , metastable phases in these models relax to the equilibrium in a similar way .
rikvold and b.m .
gorman , in _ annual review of computational physics i _
, d. stauffer ed . , ( world scientific , singapore , 1994 ) . h.e .
stanley , _ introduction to phase transitions and critical phenomena , _
( clarendon press , oxford , 1971 ) .
guth , _ the inflationary universe _ , ( addison wesley , reading , mass . , 1997 ) .
kastrup , phys .
letters b * 419 * , 40 ( 1998 ) .
langer , ann .
( n.y . ) * 41 * , 108 ( 1967 ) .
newman and g.t .
barkema , _ monte - carlo methods in statistical mechanics _
( clarendon press , oxford , 1999 ) . r.j .
baxter , j. phys .
c * 6 * , l445 ( 1973 ) .
safran , p.s .
sahni , and g.s .
grest , phys .
b * 28 * , 2693 ( 1983 ) .
sahni , d.j .
srolovitz , g.s .
anderson , and s.a .
safran , phys .
b * 28 * , 2705 ( 1983 ) .
h. arkin , t. celik , b.a .
berg , and h.meyer-ortmanns , int .
c * 10 * , 1261 ( 1999 ) .
h. arkin , t. celik , int . j. mod .
c * 11 * , 1313 ( 2000 ) .
meunier , and a.morel , eur .
j. b * 13 * , 341 ( 2000 ) .
fisher , physics ( n.y . ) * 3 * , 255 ( 1967 ) .
gunton , m.san miguel , and p.s .
sahni _ the dynamics of first order phase transitions _ in _ phase transitions and critical phenomena _ , vol.8 , c.domb and j.l .
lebowitz eds . , ( academic press , london , 1983 ) . f.y .
phys . * 54 * , 235 ( 1982 ) .
k. sekimoto , int . j. mod .
b * 5 * , 1843 ( 1991 ) .
zia and j.e .
avron , phys .
b * 25 * , 2042 ( 1982 ) . n.j .
gnther , d.a .
nicole , and d.j .
wallace , j. phys .
a * 13 * , 1755 ( 1980 ) .
langer , phys .
* 21 * , 973 ( 1968 ) .
langer , ann .
( n.y . ) * 54 * , 258 ( 1969 ) .
d. stauffer , a. coniglio , d.w .
heermann , phys .
lett . * 49 * , 1299 ( 1982 ) .
rikvold , h. tomita , s. miyashita , and s.w .
sides , phys .
e * 49 * , 5080 ( 1994 ) .
s.wonczak , r. strey , d. stauffer , j. chem . phys . * 113 * , 1976 ( 2000 ) .
m. acharyya , d. stauffer , eur .
j. b * 5 * , 571 ( 1998 ) .
bikker , g.t .
barkema , and h. van beijeren , cond - mat/0011336 ( 2000 ) . | nucleation in the two - dimensional @xmath0-state potts model has been studied by means of monte - carlo simulations using the heat - bath dynamics .
the initial metastable state has been prepared by magnetic quench of the ordered low - temperature phase .
the magnetic field dependence of the nucleation time has been measured as the function of the magnetic field for different @xmath0 and lattice sizes at @xmath1 . a size - dependent crossover from the coalescence to nucleation region
is observed at all @xmath0 .
the magnetic field dependence of the nucleation time is roughly described by the classical nucleation theory .
our data show increase of the anisotropy in the shape of the critical droplets with increase of @xmath0 .
* key words : * potts model , nucleation , metastable state , critical droplet , monte - carlo simulations . |
the atlas experiment@xcite is one of two general purpose collider detectors for the large hadron collider ( lhc ) at cern .
the atlas detector consists of an inner detector employing silicon pixel , strip , and transition radiation tracking detectors , all in a solenoidal magnetic field of 2 tesla ; electromagnetic and hadronic calorimeters using liquid argon and scintillator tile detectors ; and a muon spectrometer .
the muon spectrometer consists of a large air - core barrel and endcap toroid magnets with a @xmath1 between 2 - 6 @xmath2 , and four types of trigger and precision tracking detectors , described below .
the muon spectrometer is designed to measure the transverse momentum ( @xmath3 ) of muons with @xmath4 gev with a resolution of 4% up to @xmath3 of 100 gev and increasing to 10% @ 1 tev .
the atlas muon spectrometer consists of monitored drift tubes ( mdts ) for precision tracking in the spectrometer bending plane , resistive plate chambers ( rpcs ) and thin gap chambers ( tgcs ) for triggering in barrel and endcap , respectively , and cathode strip chambers ( cscs ) for precision measurements in the high - rate endcap inner layer where mdts would have occupancy problems .
the magnet system consists of 3 sets of air - core toroids , each with 8 coils , 1 for the barrel , and 1 for each endcap .
the barrel toroids coils are each 25 m @xmath5 7 m and the endcap coils are 9 m @xmath5 4 m .
the magnetic field provides an approximately 1 t field at the center of each coils , but is non - uniform , especially in the barrel - endcap transition region . for track reconstruction ,
the field is mapped using a computer model of the field which is normalized to measurements from 1850 hall sensors mounted on spectrometer chambers .
alignment measurements of the spectrometer are also critical for momentum determination and are accomplished with an optical alignment system of 12k sensors .
measurements from these sensors allow a 3-dimensional reconstruction of chamber positions accurate to better than 50 @xmath0 m .
in addition , the optical alignment system is complemented by alignment done with tracks .
table [ table : muon_spectrometer ] gives a summary of the muon spectrometer detector components and fig .
[ fig : layout ] shows the layout .
.atlas muon spectrometer . [ cols="<,^,^,^,^",options="header " , ] [ table : muon_spectrometer ] the spectrometer is designed so that muons cross three layers of mdt chambers for the sagitta measurement .
the track coordinate in the bending plane of the spectrometer is measured by the precision chambers with a resolution of 40 @xmath0 m . in comparison
, the sagitta of a 1 tev muon will be about 500 @xmath0 m .
the trigger chambers are placed on opposite sides of the middle mdt layer .
the trigger chambers provide a trigger based on muon momentum in addition to identifying the bunch crossing time of the muon .
the also provide the second coordinate measurement ( non - bending plane ) accurate to 5 - 10 cm .
figure [ fig : layout ] shows the expected resolution of the muon spectrometer . for @xmath6
gev / c multiple scattering is the dominant contributor to the resolution . above 100 gev / c calibration and alignment of the spectrometer
become the most significant factors in momentum resolution .
atlas muon reconstruction is done using momentum measurements from both the inner detector spectrometer and the muon spectrometer .
the two spectrometers nicely complement each other as inner detector measurements are better below 100 gev / c above which the muon spectrometer resolution is superior .
high @xmath3 measurements with the muon spectrometer require very accurate mdt and alignment calibrations which will become particularly important in a few years when lhc reaches 7 tev energy per beam and higher luminosities .
this paper shows a mix of results from 2010 and 2011 using the most up - to - date plots whenever possible .
not all plots are available for 2011 so in some cases 2010 plots are used .
there are three calibrations required for the mdts : timing offsets ( @xmath7 ) ; time - space ( @xmath8 ) functions ; and drift tube resolution functions@xcite . in order to obtain high quality calibrations for the mdts a special high statistics calibration data stream
is extracted second - level trigger processors and sent for processing at three calibration centers at michigan , rome , and munich .
this calibration stream provides 10 - 100x the rate of single muon tracks compared to regular atlas data . with this stream
it is possible to do daily calibrations of the monitored drift tubes as well as detailed data quality monitoring .
a timing offset represents the minimum measured drift time .
i.e. the time of a muon passing at the wire of the drift tube .
this time is not zero due to cables and other delays in the data acquisition system .
figure [ fig : drifttime ] shows a typical time spectrum from an mdt .
the @xmath7 fit is shown in blue , and the @xmath7 is defined as the half - way point of the rising edge .
the falling edge represents hits at the tube wall .
figure [ fig : drifttime ] shows statistical error on @xmath7 fits as a function of number hits in the time spectrum .
we require at least 10000 hits for the fits yielding a typical error of 0.5 ns .
the average drift speed is 20 @xmath0m / ns so this error corresponds to a 10 @xmath0 m error due to the @xmath7 measurement .
fit shown in blue ; right : the statistical error on @xmath7 fits as a function of number hits in the time spectrum . , title="fig:",width=302 ] fit shown in blue ; right : the statistical error on @xmath7 fits as a function of number hits in the time spectrum .
, title="fig:",width=302 ] s for all mdt chambers over a 2-month period in 2010 .
right : tracking residuals from the middle layer of mdt chambers .
, title="fig:",width=336 ] s for all mdt chambers over a 2-month period in 2010 .
right : tracking residuals from the middle layer of mdt chambers . , title="fig:",width=268 ] figure [ fig : t0diff ] shows the distribution of the change in @xmath7 for all mdt chambers over a 2 month period from 2010 .
there is small global drift of a fraction of a nanosecond , but the overall width is close to the typical statistical error of 0.5 ns .
hence , the t0s are quite stable .
the other main calibration is the time - to - space or @xmath8 function .
this function gives the drift radius ( impact parameter ) of the hit based on the drift time .
an example of an @xmath8 function is shown in fig .
[ fig : rtfun ] .
the function is non - linear since atlas uses a non - linear drift gas , @xmath9 due to its better aging characteristics in high - radiation environments .
the @xmath8 function is determined by an iterative procedure looping over tubes hits and minimizing the tracking residuals of track segments reconstructed with hits within a single mdt chamber which have either 6 or 8 layers of mdt tubes .
tracking residuals are the differences between the drift radii from the drift time and the radius from the track fit .
typically 10000 tracks segments are used in the calibration of a single chamber .
figure [ fig : rtfun ] shows the difference between @xmath8 functions for several chambers in the barrel .
the differences in @xmath8 functions are due primarily to the temperature gradient within the atlas cavern ( about @xmath10 from top to bottom ) , as well as due differences in magnetic field within chambers .
function ; right : differences between @xmath8 functions for several barrel chambers.,title="fig:",width=245 ] function ; right : differences between @xmath8 functions for several barrel chambers.,title="fig:",width=374 ] the precision of the @xmath8 function is shown by fig .
[ fig : resvr ] which shows the mean of the tracking residuals as a function of the mdt tube radius .
except for the region close to the wire , the mean residuals are within 20 @xmath0 m .
tracking residuals from the middle layer of endcap chambers are shown in fig .
[ fig : t0diff ] .
the residual width of 96 @xmath0 m is typical for all chambers in atlas .
figure [ fig : resvr ] shows the single tube resolution as a function of drift tube radius .
the resolution is determined from the tracking residuals width with the fit errors subtracted .
we see that the resolution is close to the 80 @xmath0 m for large radii .
near the wire the resolution degrades to faster drift speed and fewer drift electrons .
this plot was made with 2010 data .
we expect improvements in the future by applying some addition timing corrections such as a hit - level magnetic field correction and by using tube - level @xmath7s .
the alignment system system is designed to track chamber positions with a 40 @xmath0 m precision .
the monitoring is done with optical sensors which is cross - checked by doing alignment with straight tracks from magnet - off runs .
the barrel and endcap have separate alignment systems .
figure [ fig : alignment ] shows measurements from the mean value of the `` false '' sagitta measured with straight tracks from magnet - off runs .
straight tracks should have a sagitta of zero and hence this sagitta measurement gives the precision of the alignment system .
the plot shows the sagitta as a function of @xmath11 with the black points for the barrel and the red and blue corresponding with to the endcap .
the barrel achieves a resolution of 50 @xmath0 m , close to the design goal , whereas the endcap gives a resolution of around 110 @xmath0 m indicting that further improvements are necessary .
sector 7 of the muon spectrometer.,width=302 ]
figure [ fig : triggerocc ] shows the an occupancy plot for the barrel from collision data . as can be seen the coverage is quite uniform except for dead regions due to the support feet of the atlas detector .
the geometric acceptance is about 80% .
figure [ fig : triggereff ] shows the trigger efficiency as a function of @xmath3 and @xmath11 , respectively .
the triggers are very efficient within geometric coverage of the trigger .
the data show a slightly higher efficiency than monte carlo due to some analysis improvements in the data analysis which have not yet been introduced to the monte carlo code . and @xmath11.,width=302 ] ; right : trigger efficiency as a function @xmath11.,title="fig:",width=302 ] ; right : trigger efficiency as a function @xmath11.,title="fig:",width=302 ]
figure [ fig : momresb ] shows the momentum resolution as a function of muon @xmath3 in the barrel region of atlas .
the momentum resolution is derived from the width of the reconstructed z mass as well as by comparing single muons reconstructed by both the inner detector and muon spectrometer .
monte carlo and inner detector measurements are used to derive the contributions of momentum resolution from energy loss in calorimeters , multiple scattering , and the intrinsic resolution of the spectrometer .
the measured resolution is somewhat worse than the simulation .
this result is from the preliminary calibrations for 2010 data , so we expect improvement from more refined calibrations in the future .
the atlas muon spectrometer is working well with a high trigger efficiency and tracking resolution near design specifications .
calibrations of the drift tubes are done daily using a high statistics data stream from the level-2 trigger processors .
alignment is working well in barrel , but needs some improvement in the endcap .
momentum resolution is near design specifications .
we expect improvements from better calibrations and statistics from 2011 data .
9 `` the atlas experiment at the cern large hadron collider '' , the atlas collaboration , g aad _ et al _ 2008 jinst * 3 * s08003 , http://www.iop.org/ej/abstract/1748-0221/3/08/s08003 `` calibration model for the mdt chambers of the atlas muon spectrometer '' , p. bagnaia _
et al _ , atlas note atl - muon - pub-2008 - 004 , http://cdsweb.cern.ch/record/1089868 | the atlas muon spectrometer is designed to measure muon momenta with a resolution of 4% @ 100 gev / c rising to 10% @ 1 tev / c track momentum .
the spectrometer consists of precision tracking and trigger chambers embedded in a 2 t magnetic field generated by three large air - core superconducting toroids .
the precision detectors provide 50 @xmath0 m tracking resolution to a pseudo - rapidity of 2.7 .
the system also includes an optical monitoring system which measures detector positions with 40 @xmath0 m precision .
this paper reports on the calibration and performance of the atlas muon spectrometer . |
the most metal - deficient stars of the local universe provide an unparallelled opportunity to study material representative of a much earlier epoch , perhaps predating the formation of the prominent components of the highest redshift galaxies being discovered today .
while the massive globular star clusters ( gcs ) offer distinct , well - known advantages in such research , individual stellar chemistry is a powerful tool in the quest to understand early galactic history .
unusually comprehensive recent review volumes are those of martnez roger , prez fournn & snchez ( 1999 ) , carney ( 2001 ) and harris ( 2001 ) .
as is well known , the metallicity distribution of the 137 milky way gcs with [ fe / h ] determinations is well described by two gaussian curves centered on [ fe / h]@xmath1@xmath2 and @xmath3 ( harris 2001 ) on the popular zinn & west ( 1984 ) metallicity scale . of particular interest in this context
is that the most metal - deficient clusters have [ fe / h]@xmath4@xmath5 whereas the most metal - deficient halo stars extend to metallicity values more than an order of magnitude lower ( c.f .
, carney et al . 1996 ) . while not entirely a proper comparison , the most metal - poor gcs and halo stars exhibit compositions well below those of damped lyman-@xmath6 forest systems at @xmath7 ( c.f .
, prochaska & wolfe 2000 ) , yet ironically many of the extremely metal - dificient galactic objects likely to have formed at high redshift can be seen with binoculars .
other properties of the gc system offer important clues to its origin . at galactocentric radii ( r@xmath8 ) beyond 10 kpc , the average m@xmath9 of those clusters with blue hb morphology is @xmath10 , while clusters exhibiting red horizontal - branch ( hb ) morphology have a corresponding value of @xmath11 ( van den bergh 1996 ) .
half - light radii for galactic gcs show a dependence on galactocentric radii , with the farthest ones generally being larger in diameter .
while exhibiting similar trends , the clusters of the fornax dsph or the lmc have a different distribution ( van den bergh 2000a ) .
van den bergh ( 2000a , b ) interprets the present data as evidence that the galaxy , fornax and the lmc were distinct systems when they started to form their gcs : it appears that somehow the gcs `` knew '' they were going to end up in their respective parent galaxy . by correlating the spatial distributions and kinematics of galactic halo tracers , hartwick ( 2000 )
argues that the spatial distribution of milky way companion galaxies define a real triaxial structure reflecting the initial conditions of the galaxy s formation epoch .
the outer halo gcs form a nearly oblate , flattened system whose minor axis is highly inclined to the present rotation axis , and thus do not seem part of the same triaxial system .
how long did the initial burst of cluster formation last for the milky way ?
the answer , from relative ages of the most metal - deficient halo stars , can help distinguish among conceivable formation scenarios ( e.g. , fast collapse , leisurely buildup through mergers , etc . ) . while relative measurements are usually more reliable to make and more straightforward to interpret , controversy about relative cluster ages persists . morphological properties of hbs have been extensively explored for their potential as a powerful tool for sensitive age discrimination among clusters . from plots of metallicity against the ratio of the difference in the number of blue and red hb stars to the total hb population , compared with isochrones from synthetic hb models for gcs , y .- w .
lee , demarque , & zinn ( 1994 , _ et seq . _ ; hereafter ldz ) found evidence for an age range of about 4 gyrs , or some 2535@xmath12 of the age of a 1215 gyr old galaxy .
the age range so inferred is present at all metallicities , and an age-@xmath13 relation was suggested .
the concepts which emerged from this important work have exerted wide influence in thinking about how long it took to form or assemble the galactic halo .
many accept that evidence from gcs strongly favors a relatively drawn out formation , which fits nicely with growing support for the importance of hierarchical mergers during the formation of large galaxies .
however , this compelling interpretation of hb morphologies has been the subject of debate over the past seven or so years . in part
this arises because stellar evolutionary models indicate that hb morphology is sensitive to differences in parameters that are well below present observational uncertainties .
specific counterexamples to the ldz analysis noted by , e.g. , stetson , vandenberg , & bolte ( 1997 ) , are disconcerting . what seems to be the current situation ?
harris et al . ( 1997 ) compared comparably deep ( v , v@xmath14i ) color - magnitude diagrams ( cmds ) for the massive gcs m92 and ngc@xmath15 , which are among the most metal - deficient gcs in the galaxy .
they were unable to detect any age difference in excess of 1 gyr .
importantly , m92 resides in the inner part of the galactic halo , while ngc@xmath15 s r@xmath8@xmath4@xmath16 kpc places it in the outermost halo ( roughly twice as far as the lmc ) .
moreover , its large core and tidal radii are a strong indication that it has existed in the outer halo throughout its life , for otherwise dynamical forces would have modified its appearance .
( note that shetrone , ct , & sargent 2001 have recently determined , for the first time , a metallicity for ngc@xmath15 from high - dispersion spectra of giants which confirms the validity of pairing it with m92 ) .
this single comparison among archtypical metal - deficient clusters from different regions of the halo already provides stringent constraints on the duration of the formation of the most massive clusters early in galactic ( pre)history .
( 2001 ) comparisons of deep cmds for m92 , ngc@xmath17 and ngc@xmath18 in the strmgren photometric system similarly found no evidence of age differences in excess of @xmath41 gyr .
an extensive study by rosenberg et al . ( 1999 ) applied both techniques of relative age determinations , the color - difference ( between the lower giant branch and the turnoff ) and luminosity - difference ( between the turnoff and the hb ) , to a homogeneous data set for 35 gcs .
they found all the clusters with [ fe / h]@xmath19 to be old and coeval .
some clusters in the range @xmath20 [ fe / h ] @xmath21 appear to be up to 25% younger than the otherwise co - eval oldest clusters , while the most metal - rich clusters studied appear to be co - eval .
they infer that the gc formation process started at the same time throughout the halo to a distance of @xmath4@xmath22 kpc from the galactic center , with the small population of significantly younger halo clusters found at r@xmath8@xmath23@xmath24 kpc likely the result of later mergers .
they further note that their evidence for the co - eval metal - rich clusters being @xmath417% younger than other gcs ( that is , for a mild age - metallicity relation ) is particularly model dependent .
lee et al . (
2001 ) have extended the rosenberg et al .
analysis with three extremely metal - poor clusters located in the innermost galactic halo , further strengthening the evidence against an age range or of a trend in ages with r@xmath8 among the most metal - deficient gcs in the galaxy . on the question of the existence of an age - metallicity relation , vandenberg ( 2000 ) analyzed in a systematic manner high - quality cmd data from the literature for 26 gcs .
he infers that unless [ o / fe ] or [ @xmath6/fe ] are up significantly at the lowest metallicities , then a mild age - metallicity relation results . at any given metallicity
, however , he found the data do not permit an age range @xmath25 10 - 15% .
the famous `` second - parameter '' gcs at intermediate metallicities and large r@xmath8 figure prominently in efforts to understand galactic halo formation .
their great distances , and low masses ( hence sparse cmds ) have challenged observations from the ground .
stetson et al .
( 1999 ) reported relative age comparisons from hst data by both cmd techniques of three such clusters , pal 3,4 and eridanus , with the inner halo clusters m3 and m5 , of similar metallicities .
they found no evidence for an age spread among the three outer - halo clusters . at face value
, their data suggest that those clusters could be younger by @xmath4@xmath26 gyr than their comparison clusters ; a caveat in their result is that errors of 0.2 dex ( which can not be ruled out with present data ) in [ fe / h ] or [ @xmath6/h ] could reduce or eliminate the inferred small age difference between the two groups . finally , two independent studies that compared cmds for seven old , metal - poor gcs in the large magellanic cloud with those of comparable clusters in the galaxy failed to detect any age differences between them in excess of @xmath41 gyr or
so ( olsen et al . 1998 , johnson et al . 1999 ) ; constraints on the age range from such studies are also potentially limited by the quality of available spectroscopic abundance determinations for the lmc clusters . in summary
, there is evidence that a fraction of the intermediate - metallicity gcs in the galactic halo may be younger than the bulk of the system .
however , in recent years evidence from careful differential cmd comparisons suggests any age dispersion among the bulk of the galactic gcs ( or those in our distant halo ) remains below the observational detection limit at a given metallicity , while the evidence for a mild age - metallicity relation remains a more debated question .
it thus appears that the bulk of the galactic gc system , and especially the most metal - poor ones , formed quickly during the earliest phases of galactic history .
we now turn to the more difficult question of absolute ages or , at what redshift did metal - poor star formation begin in the galaxy ?
recent work by bergbusch & vandenberg ( 2001 ) interpreting the cmd of m92 with their new isochrones favors an age of @xmath41516 gyr .
this would be reduced by @xmath4@xmath27% if he or heavy - element diffusion were incorporated into their models .
selection of the appropriate mixing - length for the comparison continues to require judgement .
three field halo subgiants they concurrently studied appeared also to have ages @xmath25@xmath28 gyr .
grundahl et al . ( 2000 ) found an m92 age of @xmath415 gyr using strmgren color - color ( c , c@xmath14y ) diagrams , which have the powerful advantage of being distance independent .
however , their interpretation requires accurate knowledge of the t@xmath29 scale , which they ( and bergbusch & vandenberg 2001 ) argue is well - established .
they also compared m92 strmgren color - color data with those of the field subdwarfs having [ fe / h]@xmath30@xmath31 and concluded that the subdwarfs are the same age , @xmath4@xmath32 gyr , as m92 before taking diffusion into account .
carretta et al .
( 2000 ) studied nine clusters along with a large sample of subdwarfs with good parallax measurements , and concluded , after deciding that distances from the subdwarf fitting method are @xmath4@xmath33 too long , that the mean age of their sample was [email protected] gyrs ( 95% confidence range ) .
further analysis in terms of cosmological models led them to conclude that the initial epoch of gc formation occurred around z@xmath43 .
until now , the quality both internal and external of photometry has been a limiting factor in the quest for reliable absolute ages for gcs .
1 is a remarkable example of work peter stetson has been doing in the past year or so combining data from many observers and telescopes .
this particular cmd is based upon 197 ( ! ) ccd images from four telescopes obtained during 15 nights and eight observing runs .
the data all came from the original observers . enabling such research constitutes a powerful argument for cataloguing and archiving data from ground - based telescopes , as most images taken since the introduction of ccds are tragically languishing on deteriorating tapes in observers offices worldwide . with photometric zero points now known to better than 1% on the new standard system ( stetson 2000 ) , and with internal photometric scatter reduced to insignificance by combining many data ,
photometric quality will no longer limit the ability to determine accurate absolute and relative ages .
as photometric precision ceases to be a determining factor in the accuracy of age determinations , what factors remain important ?
accurate knowledge of interstellar reddening for individual clusters and stars therein remains crucial , and nontrivial .
similarly , reducing uncertainties in distances from parallaxes remains a major challenge .
uncertainties in chemical composition determinations are a fundamental yet extremely challenging subject ripe for renewed efforts , as we now discuss .
for instance , there are still controversies about the basic scale of gc abundances ( zinn & west 1984 , carretta & gratton 1997 ; rutledge , hesser , & steton 1997 ) . that highly experienced teams often find differences in overall abundances well beyond observational errors indicate that systematic errors still limit accuracy in high spectral resolution studies .
finally , abundances for the most distant objects often are inferred from broad - band photometry , or low - resolution spectra of @xmath6-elements , like ca , in the most luminous giants , rather than from high - dispersion spectra of less evolved stars .
a question of considerable interest is whether inferences from more luminous cluster stars apply to the main - sequence and turnoff stars weighted so heavily in age determinations . until the advent of 810-m telescopes hb and red giant branch ( rgb ) stars have been the only ones for which abundances based upon higher spectral resolution were possible .
hb stars anchor distance moduli , underpin the ldz interpretation of halo formation , and play many other important roles in our understanding of the galactic halo .
an interesting example of modern abundance work is behr , cohen , & mccarthy s ( 2000 ) study of the abundances of hb stars in m15 , with comparisons to m13 ( behr et al .
1999 ) and ngc6752 ( moehler et al . 1999 ) . among the slowly rotating hot stars
two were found that exhibit the same extreme metal - deficiency that characterizes other m15 stars , but the other six stars studied show solar ( ! ) [ fe / h ] values due to the radiative levitation of metals predicted by michaud , vauclair , & vauclair ( 1983 ) .
many other interesting trends in detailed element abundances emerging from their study remind us how sensitively the hb reflects many parameters of stellar astrophysics .
the recognition of the phenomenon of abundance spreads among giant stars within an individual gc , but not among open cluster or field stars ( e.g. , hesser , hartwick , & mcclure 1976 ) , sparked studies aimed at clarifying the percentage of the dispersion attributable to internal mixing during evolution up the giant branch , differences in initial abundances during the main - sequence phase , accretion , etc .
ivans et al .
( 1999 ) reported the range of individual chemical elements in the [ fe / h]@xmath1@xmath35@xmath34@xmath36 cluster m4 determined from r=30,000 or 60,000 spectral data with s / n@xmath23100 for 36 giants .
most of the scatter seen for elements heavier than al is observational , but they found evidence for a wide range of [ o / fe ] , [ na / fe ] and [ al / fe ] within this cluster . since both [ fe / h ] and [ @xmath6/fe ] play important roles in the age dating process , we need to understand how to characterize them for a given cluster when selecting appropriate isochrones . work with keck and the vlt at higher resolution for fainter apparent magnitudes than heretofore possible using 4-m class telescopes ( e.g. , cannon , et al .
1998 ) allowed gratton et al .
( 2001 ) to compare abundances derived from rgb and turnoff ( to ) stars .
encouragingly , they found the same @xmath30[fe / h]@xmath23 values from stars in these different evolutionary states in the clusters ngc6752 and 6397 .
they also infer an o - na anticorrelation among the main sequence stars , as previously seen among the giants in these clusters . as the third most abundant element
, oxygen plays an important opacity role in stellar atmospheres .
uncertainty in the [ o / fe ] ratio is a factor in determining the absolute ages of gcs via theoretical isochrone fits .
vandenberg ( 1992 ) determined that a 0.25 dex change in [ o / fe ] will change the resulting cluster age by 1 gyr . while modern abundance analyses can obtain internal relative abundance uncertainties of less than 0.25 dex
, the systematic offsets between the different measurement methods of determining oxygen abundances can be much larger .
figure 2 plots [ o / fe ] abundances determined by three methods : the near - uv oh lines , the forbidden resonant [ o i ] lines in the optical , and the far red o i permitted triplet .
each method has strengths and drawbacks , as described in king ( 2000 ) .
as can be seen in figure 2 the oxygen abundance ratio measures determined from the [ o i ] lines are significantly lower than the determinations derived from the oh and permitted lines . for the most metal - poor gcs , the difference is about 0.5 dex , which translates into a @xmath42 gyr systematic uncertainty in the ages of these gcs .
the gc oxygen abundances are determined from the forbidden line , which is relatively stronger in cool giants .
the permitted and oh lines are stronger in hotter stars , and the sample of very metal - poor stars with both forbidden- and permitted - line oxygen abundance analyses is small .
fulbright & kraft ( 1999 ) , for example , found that the [ o / fe ] ratio derived from [ o i ] line in two very metal - poor subgiants does not match the [ o / fe ] ratio derived from the oh or permitted lines in the same stars .
recent attempts to resolve this discrepancy ( king 2000 , israelian et al .
2001 ) have involved a careful examination of the atmospheric parameters used in the abundance analysis and the inclusion of non - lte corrections .
asplund & garca prez ( 2001 ) suggests that full 3-d atmospheres may be necessary to model the oxygen lines correctly .
silicon is the lightest @xmath6-element that does not show large star - to - star abundance variations within individual metal - poor gcs .
however there does seem to be a cluster - to - cluster scatter in [ si / fe ] . as first pointed out by kraft ( 2000 ) , studies of m4 ( ivans et al .
1999 ; 36 stars ) and m5 ( sneden et al . 1992
; 12 stars ) found both to have similar values of [ fe / h ] ( @xmath35 and @xmath37 , respectively ) , but m4 is twice as si - rich as m5 ( [ si / fe ] ratios of + 0.55 and + 0.20 , respectively ) .
both studies were conducted by the same group using similar techniques , and the star - to - star scatter within the individual clusters is smaller than the cluster - to - cluster difference .
this is evidence that there was a significant chemical inhomogeneity between the material that formed these clusters .
since the clusters are massive and on the metal - rich end of the halo metallicity distribution , it is unlikely that a random enrichment event would have had a sizable affect on the cluster s abundances .
the lack of [ si / fe ] variations within a cluster suggest that the cluster - forming material was well - mixed prior to formation , so whatever led to the cluster - to - cluster variations in [ si / fe ] most likely took place before the clusters formed . in figure 3 we plot the distribution of [ si / fe ] vs [ fe / h ] for several globular clusters ( mean values ) and field stars . aside from ru 106 and
pal 12 , there is general agreement in the range of [ si / fe ] values seen in three samples .
two separate , parallel , tracks seem to be defined by m4 and the more metal - rich gcs and another by m5 and the metal - poor gcs .
whether this structure is truly there or is just a sampling effect is hard to determine from the data in hand .
nissen & schuster ( 1997 ) found that for field stars with [ fe / h]@xmath4@xmath38 , there are halo stars that show lower [ @xmath6/fe ] enhancements than disk and other halo stars at a similar metallicity .
fulbright ( 2001 ) shows evidence that for field halo stars with @xmath38@xmath23[fe / h]@xmath23@xmath31 , the stars having higher kinetic energy ( e.g. , on more extreme orbits ) on average , show lower values of [ x / fe ] for several elements , including [ si / fe ] .
dinescu et al .
( 1999 ) computed orbital parameters for 38 gcs and found m4 s apogalactic radius to be @xmath46 kpc , while m5 s is @xmath440 kpc .
thus connection between kinematics and observed abundances for halo objects extends to m4 and m5 , as well .
the origin of such a relationship , however , is unclear .
the age information we have suggests that all the metal - poor gcs formed at about the same time , so both of these processes must have occurred simultaneously , although one or both of the clusters could have formed in a separate fragment that was accreted later .
an exciting new development in the study of gcs is the possibility that the long - lived radioisotopes of the r - process elements th ( 14 gyr half - life ) and u ( 4.5 gyr half - life ) can be used to obtain ages for individual stars .
cowan et al .
( 1999 ) obtained a mean age of @xmath39 gyr for two field stars based on th line strengths .
sneden et al .
( 2000 ) obtained a th - based age for m15 of @xmath40 gyr .
johnson & bolte ( 2001 ) also measured th - based ages for several field stars plus the m92 giant vii-18 . for vii-18 ,
johnson & bolte obtain an age of @xmath41@xmath34@xmath42 gyr .
a major source of the large uncertainties is calculating the initial th abundance for the stars . in these studies
the models for the initial r - process production were scaled to the observer ed stable rare earth elemental abundances ( eu , for example ) .
the mass differences between the rare earths and th leaves room for sizable error . cayrel et al .
( 2001 ) successfully observed the 3895 @xmath0 line of u ii in the metal - poor field giant cs31082 - 001 and derived an age of @xmath43 gyr for it .
this star is special in that while the fe and other light element abundances are about 1/800 of solar , the r - process abundances are about 1/9 of solar .
having two radioactive species creates the advantageous situation where the observed u / th ratio can be compared to a theoretical initial u / th ratio without having to scale to other r - process elements . due to their proximity in atomic mass , the theoretical initial u /
th ratio should be more precise than scaling either to the lighter r - process elements ( goriely & clerbaux 1999 ) .
besides the dependence on quality theoretical r - process calculations , the radioactive age - dating technique could be improved by work in two areas .
first , our knowledge of the oscillator strengths of th ii and u ii need improvement .
cayrel et al . (
2001 ) quote a 0.12 dex uncertainty in the u ii log(gf ) value they used . that systematic error can change the resulting age by 2.6 gyr .
it is hoped that the potential of this method to provide an independent age for the galaxy will induce atomic physicists to improve the data .
second is the need for near uv / blue - sensitive very high resolution spectrographs on 10-m or larger telescopes .
while the present status of the field would not be possible without the excellent hires and uves spectrographs , the near uv region of even these very metal - poor stars are still plagued with line blending at resolutions @xmath45000070000 .
the situation would be vastly improved at resolutions @xmath23150000 with similar s / n levels .
the reduction of blending problems and the increased contrast against the continuum would help make measurements of the th and u lines possible in gc stars , which have not been found to be as enhanced in the r - process elements as cs31081 - 001 .
true to neruda s poetry , the pale , metal - deficient objects throughout the galaxy s extended environs remain enigmatic , but they continue to yield important insights into how the milky way might have formed , and how that might in turn constrain understanding of galaxy formation in general .
the salient points from our attempt to review the current status are : \1 .
the initial epoch of massive , metal - poor star cluster formation proceeded quickly ( @xmath301.5 - 2.0 gyr ) ; there are , however , younger clusters at intermediate - metallicities , which may have been accreted later . in particular , the most metal - deficient clusters ( and field stars ) show no discernible age dispersion throughout their 1 - 100 kpc extent in r@xmath8 . \2
. the most distant second - parameter gcs do not appear to be more than @xmath4@xmath262 gyr younger than comparable metallicity , inner - halo clusters , but uncertain knowledge of their respective chemistry presents a caveat .
absolute ages of the most metal - deficient clusters remain controversial with recent determinations falling in the range 12 to 16 gyrs with uncertainties of @xmath423 gyrs ; recent consistency checks using abundances of radioactive elements ( th , u ) are promising . \4
. existence of a putative age - metallicity relation also remains controversial ; it presently seems likely to be 10% .
photometric precision will soon not be a limiting observational factor on the accuracy of cluster ages ; rather , uncertainties in reddenings , distances and chemical composition will dominate .
the time is ripe for major efforts on abundance determinations with 8-m and larger telescopes , but the most interesting problems may require telescopes with 20 - 30-m aperture .
there is growing evidence that dispersion of light elements within individual gcs reflect the epoch of star formation , rather than of mixing during stellar evolution ; such differences combine with uncertain [ o / fe ] determinations to complicate accurate age determinations .
it is encouraging that the average compositions determined from evolved and turnoff region stars seem to be the same within the limits of the data quality in the few clusters yet examined .
cluster - to - cluster differences for ratios of light elements imply substantially different enrichment histories .
field halo stars exhibit wide ranges of [ x / fe ] ; such differences provide detailed insight into early star formation . \9 . over @xmath44 orders of magnitude in total mass ( from gcs to dsphs to the lmc ) ,
the most metal - poor local group constituents appear to have formed at the same epoch , yet to also have been aware of their ultimate association with a articular parent .
we are grateful to frank grundahl , jennifer johnson , j .- w .
lee and peter stetson for sharing material that we used when preparing this paper , and to eva grebel and doug geisler for organizing an outstanding conference in a beautiful location .
asplund , m. & garca prez , a.e .
2001 , , 372 , 601 behr , b.b .
, cohen , j.g . & mccarthy , j.k .
2000 , , 531 , l37 behr , b.b .
. 1999 , , 517 , l135 bergbusch , p.a . &
vandenberg , d.a .
2001 , , 556 , 322 bosegaard , a. et al .
1999 , , 117 , 492 brown , j.a .
, wallerstein , g. , & zucker , d. 1997 , , 114 , 180 brown , j.a . , wallerstein , g. , & gonzalez , g. 1999 , , 118 , 1245 cannon , r.d .
1998 , , 298 , 601 carney , b.w .
1996 , , 108 , 900 carney , b.w .
1996 , , 112 , 668 carney , b.w .
2001 , in saas - fee advanced course 28 , star clusters , ed .
l. labhardt & b. binggeli , springer ( berlin ) , 1 carretta , e. et al .
2000 , , 533 , 215 caretta , e. & gratton , r.g .
1997 , , 121 , 95 cavallo , r.m . , pilachowski , c.a . , & rebolo , r. 1997 , , 109 , 226 cayrel , r. et al .
2001 , nature , 409 , 691 cowan , j.j .
et al . 1999 , , 521 , 194 dinescu , d.i .
1999 , , 117 , 1792 fulbright , j.p .
2000 , , 120 , 1841 fulbright , j.p .
2001 , , submitted fulbright , j.p . & kraft , r.p . 1999 , , 118 , 527 gonzalez , g. & wallerstein , g. 1998 , , 116 , 765 goriely , s. & clerbaux , b. 1999 , , 346 , 798 gratton , r. et al .
2001 , , 369 , 87 grundahl , f. 2001 , private communication grundahl , f. et al .
2000 , , 120 , 1884 harris , w.e .
2001 , in saas - fee advanced course 28 , star clusters , ed .
l. labhardt & b. binggeli , springer ( berlin ) , 223 harris , w.e .
1997 , , 114 , 1030 hartwick , f.d.a .
2000 , , 119 , 2248 hesser , j.e .
, hartwick , f.d.a . & mcclure , r.d .
1976 , , 207 , l113 israelian , g. , garca lpez , r. j. & rebolo , r. 1998 , , 507 , 805 israelian , g. et al .
2001 , , 551 , 833 ivans , i.i .
1999 , , 118 , 1273 johnson , j.a . 1999 ,
phd thesis , university of california , santa cruz johnson , j.a .
et al . 1999 , , 527 , 199 johnson , j.a . &
bolte 2001 , , 554 , 888 king , j.r .
1993 , , 106 , 1206 king , j.r .
2000 , , 122 , 1056 kraft , r.p .
2000 , in iau joint discussion 5 , mixing and diffusion in stars : theoretical predictions and observational constraints , in press .
kraft , r.p .
1992 , , 104 , 645 lee , j .- w .
2001 , private communication lee , y .- w . , demarque , p. , & zinn , r. 1994 , , 423 , 248 mart nez roger , c. , prez fournn , i. , & snchez , f. ( eds . ) , globular clusters , cambridge university press , 1999 mcwilliam , a. et al .
1995 , , 109 , 2757 michaud , g. , vauclair , g. & vauclair , s. 1983 , , 267 , 256 moehler , s. et al .
1999 , , 346 , l1 nissen , p.e . &
schuster , w.j .
1997 , , 326 , 751 olsen , k.a.g .
1998 , , 300 , 665 prochaska , j.x . , &
wolfe , a.m. 2000 , , 553 , l5 rosenberg , a. et al .
1999 , , 118 , 2306 ryan , s. g. , norris , j. e. & bessell , m. s. 1991 , , 102 , 303 rutledge , g.a .
, hesser , j.e . & stetson , p.b .
1997 , , 109 , 907 shetrone , m.d .
1996 , , 112 , 1517 shetrone , m.d . , ct , p. , & sargent , w.l.w .
2001 , , 548 , 592 sneden c. et al .
1991 , , 102 , 2001 sneden c. et al .
1992 , , 104 , 2121 sneden c. et al .
2000 , , 536 , 85 stetson , p.b .
2000 , , 112 , 925 stetson , p.b .
et al . 1999 , , 117 , 247 stetson , p.b . , vandenberg , d.a . ,
& bolte , m. 1996 , , 108 , 560 vandenberg , d.a .
1992 , , 391 , 685 vandenberg , d.a .
2000 , , 129 , 315 van den bergh , s. 1996 , , 108 , 986 van den bergh , s. 2000a , , 530 , 777 van den bergh , s. 2000b , , 112 , 932 zinn , r. , & west , j.j .
1984 , , 55 , 45
_ dantona : _ [ comment made after demarque s remarks ] + recently , gratton et al . have found that the iron and oxygen abundances in ngc 6397 turnoff stars are equal to subgiant branch abundances , although models including diffusion predict differences , also by a factor of two , for this metal - poor cluster . of course , this does not give information on helium diffusion , which affects the age determinations , but is an indicator that the models are not adequate for the diffusion of metals . | we review properties of the most metal - deficient globular clusters and halo stars in the local universe as of march 2001 , with goals of examining present evidence for the duration of the initial burst of massive cluster formation in the milky way , determining when that burst occurred , and elucidating what appears to be limiting our present understanding .
such issues bear upon many topics to arise later in the symposium .
# 1_#1 _ # 1_#1 _ = # 1 1.25 in .125 in .25 in _ `` he aqu@xmath0 los astros plidos todos llenos de enigma '' _ + pablo neruda , _ el hondero entusiasta _ + |
potassium sulfate k@xmath4so@xmath1 crystallizes at room temperature in the orthorhombic olivine type structure and has four formula units per d@xmath5 unit cell , with lattice constants @xmath6 , @xmath7 , and @xmath8 @xcite . the mineral olivine ( mg , fe)@xmath4sio@xmath1 is a major component of the earth s crust .
( mg , fe)@xmath4sio@xmath1 and its isomorphs a@xmath4bo@xmath1 ( where a = k , nh@xmath1 , rb , cs , b = s , se ) exhibit many interesting physical properties and phase equilibria that were studied repeatedly @xcite . upon heating k@xmath4so@xmath1 undergoes a first order transformation at @xmath9c to a hexagonal structure d@xmath10 with @xmath11 and @xmath12 ( measured at @xmath13c @xcite ) . in this high-@xmath14 structure the oxygen positions of the so@xmath15 tetrahedra are only partially occupied as a result of rotational disorder .
k@xmath4so@xmath1 crystals that are grown at room temperature from aqueous solution incorporate oh@xmath16 ions .
the oh@xmath16 concentration decays in the temperature region from @xmath17c to @xmath18c and usually crystals are destroyed by this process @xcite .
el - kabbany @xcite has reported a thermal hysteresis for the solid phase transformation with @xmath19c at heating and @xmath20c at cooling .
electrical conductivity measurements on single crystals have been carried out by choi et al .
@xcite who found @xmath21c on heating and @xmath22c on cooling , with a thermal hysteresis of 5.4k .
most of the samples crack near @xmath23c and show an abrupt drop in electrical conductivity .
the hysteresis phenomena , cracking , and electrical `` pretransition phenomena '' were attributed to rotational disorder of so@xmath24 ions @xcite and inclusions of oh@xmath25 @xcite .
the present authors reported recently , that the addition of cd@xmath0 to the nutrient solution during k@xmath4so@xmath1 crystal growth leads to an improvement of the crystalline quality @xcite . in this paper , the influence of cd@xmath0 on the growth and on thermal and optical properties of k@xmath4so@xmath1 crystal is studied in more detail .
colorless and transparent crystals of pure potassium sulfate could be obtained by slow evaporation of the solvent from saturated aqueous solutions .
k@xmath4so@xmath1 usually crystallized in prismatic crystals , the @xmath26-axis was found along the long axis of the prism and the @xmath27-axis was along one edge of the quasi - triangular basal plane . k@xmath4so@xmath1:cd@xmath0 and
k@xmath4so@xmath1:fe@xmath0 crystals were grown by the same method from solutions containing 4wt.% of cdso@xmath1 or feso@xmath1 , respectively .
the growth process was performed in a multi - jar crystallizer ( fig [ fig1:exmp ] , top ) to ensure identical growth conditions . in a period of 60 days
, we were able to grow colorless , transparent k@xmath4so@xmath1 single crystals with well - developed faces having dimensions of @xmath28@xmath29 as shown in fig .
[ fig1:exmp ] , bottom .
contrary , the growth rate of colorless k@xmath4so@xmath1:cd@xmath0 crystals was much lower : in a period of @xmath30 days crystals up to 3 mm in diameter could be grown .
so@xmath1 crystals .
bottom : as grown undoped k@xmath4so@xmath1
crystal.,title="fig:",height=181 ] so@xmath1 crystals .
bottom : as grown undoped k@xmath4so@xmath1
crystal.,title="fig:",height=181 ] the crystals were characterized by inductively coupled plasma
optical emission spectroscopy ( icp oes ) , by x - ray powder diffractometry , and by simultaneous differential thermal analysis / thermogravimetry ( dta / tg ) , as described recently @xcite . a netzsch sta409cd dta
/ tg / qms ( qms = quadrupole mass spectrometer ) system with skimmer coupling of the qms was used for the simultaneous analysis of gaseous species that are evaporating from the sample during heating @xcite .
the shape and position of melting peaks was additionally investigated with a dsc / tg sample carrier and pt / rh crucibles with lid .
the optical transmission was measured in the region @xmath31 nm ( vuv spectrometer made by laser centre hannover / lzh ) .
for the undoped k@xmath4so@xmath1 crystals only impurities on the ppm level could be found by icp oes .
if crystal growth was performed from solutions containing 4wt.% cdso@xmath1 , the resulting crystals contained only 0.014wt.% of the dopant . from a k@xmath4so@xmath1 nutrient solution with 4wt.% feso@xmath1 , crystals with 0.076wt.%
fe concentration were harvested .
the small value of the cd@xmath0 distribution coefficient @xmath32 is in agreement with literature data for the system k@xmath4so@xmath1cdso@xmath1 @xcite that report an almost vanishing solubility . for iron one
calculates a larger distribution coefficient @xmath33 . but this value is still small , and agrees with christov s observation @xcite that almost pure k@xmath4so@xmath1 or na@xmath4so@xmath1 are crystallizing from aqueous solutions containing fe@xmath4(so@xmath1)@xmath34 or feso@xmath1 , respectively .
the small values of @xmath35 and especially @xmath36 must lead to high dopant concentrations in the liquid interface layer and slows down the crystallization process , as compared to the growth of undoped crystals . from the x - ray diffraction powder patterns of @xmath37-values and unit cell parameters
were calculated with xpowder computer software @xcite .
the cell parameters that are reported in table [ tab : xray ] match very well with the standard values for k@xmath4so@xmath1 as found in the literature ( pdf card no .
01 - 070 - 1488 ) @xcite .
.lattice parameters for pure and doped k@xmath4so@xmath1 crystals .
[ cols="<,^,^,^,^ " , ] [ tab : xray ] so@xmath1 and k@xmath4so@xmath1:cd@xmath0 crystals.,scaledwidth=40.0% ] fig .
[ fig3:exmp ] compares results of dta measurements with heating rates of + 10k / min around @xmath38 of pure k@xmath4so@xmath1 and of k@xmath4so@xmath1:cd@xmath0 crystals .
the curves reveal the expected endothermic peaks around @xmath3c related to the phase transition of pure and cd@xmath0 doped k@xmath4so@xmath1 . in the first heating run of pure k@xmath4so@xmath1 the phase transition peak is superimposed by other effects while in the second heating run the phase transition peak occurs without superimposed features .
it will be shown later that traces of water are incorporated in the ( formally anhydrous ) crystal structure .
this water evaporates during the first heating run , and the second heating curve of pure k@xmath4so@xmath1 shows the phase transition peak without additional effects .
it is surprising , that k@xmath4so@xmath1 doped with cd@xmath0 already in the first heating run did not show the superimposed features .
an explanation of this phenomenon can be found using the tg ( thermogravimetry ) and qmid ( quasi multiple ion detection ) curves of k@xmath4so@xmath1 and k@xmath4so@xmath1:cd@xmath0 crystals that are shown in fig .
[ fig4:exmp ] .
so@xmath1 and k@xmath4so@xmath1:cd@xmath0 crystals with 10k / min and qmid for @xmath39 ( water).,scaledwidth=40.0% ] during the tg measurement the gas flowing around the k@xmath4so@xmath1 sample was analyzed by a qms system with 298 scans for relative mass numbers @xmath40 ( @xmath41 molecule mass , @xmath42 charge ) , this means that one mass spectrum is available every 2.5k . from these mass spectra qmid curves
can be calculated that show the ion current for every individual @xmath43 within the mass detection range given above .
[ fig4:exmp ] shows such qmid curve for the signal @xmath39 ( h@xmath4o ) of undoped k@xmath4so@xmath1 together with the tg curves .
the peak around @xmath38 proves that evaporating water is the origin of the mass loss .
the monotonous slope of the qmid background is typical for skimmer qms coupling systems and plays no role in this discussion .
the tg curves can be divided into two temperature regions : the first one ranges from @xmath44c up to the onset temperature of the dta peak ( @xmath45c ) and the second region is around the phase transition ( @xmath46 ) . in region
i the pure crystal looses about @xmath47% of its initial mass , while for cd@xmath0 doped crystals the mass loss is @xmath48% .
the smaller mass loss of k@xmath4so@xmath1 becomes remarkable for @xmath49c were the tg curve for this sample is bent upward . in region
ii the mass loss of k@xmath4so@xmath1:cd@xmath0 is smaller than the mass loss of pure k@xmath4so@xmath1 by 0.25% .
it is obvious , that the small cd@xmath0 content encourages the evaporation of water already before the onset of the phase transformation .
an explanation if this behaviour can not be given on the basis of the current data .
it should be added here that the influence of the @xmath50 times larger iron content in k@xmath4so@xmath1:fe@xmath0 is just the opposite : here the tg step at @xmath38 is larger compared with undoped k@xmath4so@xmath1 ( not shown in fig .
[ fig4:exmp ] ) .
arnold et al .
@xcite reported that crystals of k@xmath4so@xmath1 grown at room temperature from aqueous solution incorporate oh@xmath51 ions .
they decay in the temperature region from @xmath17c to @xmath18c .
if k@xmath4so@xmath1 is crystallized above @xmath52c , the content of water in crystals can be minimized @xcite .
therefore investigation of the effect of crystal growth temperature in comparison with different concentrations of cd@xmath0 on this phenomena of k@xmath4so@xmath1 crystals will be considered in future studies .
so@xmath1:cd@xmath0 and k@xmath4so@xmath1 crystals.,scaledwidth=40.0% ] dta heating and cooling curves around the phase transition peak for k@xmath4so@xmath1 and k@xmath4so@xmath1:cd@xmath0 crystals are shown in fig .
[ fig5:exmp ] .
the transition temperature @xmath38 can be determined from the intersection of the extended basis line with the tangent at the inflection point ( extrapolated onset ) .
it is remarkable that for pure k@xmath4so@xmath1 , @xmath38 as obtained from the first heating or cooling runs , respectively , differ considerably by a hysteresis of about 8k . for k@xmath4so@xmath1:cd@xmath0
this hysteresis is only 4.5k .
it was reported recently that the hysteresis is reduced considerably for both k@xmath4so@xmath1 and k@xmath4so@xmath1:cd@xmath0 in subsequent ( starting from the second ) heating / cooling runs @xcite .
hence it can be suggested that the effect of cd@xmath0 on hysteresis is similar to the effect of annealing for k@xmath4so@xmath1 crystal , if the first heating / cooling run is considered as annealing for the undoped crystal .
[ fig : melt ] compares the melting behavior of pure k@xmath4so@xmath1 with that of k@xmath4so@xmath1:cd@xmath0 and , for reference , with that of a higher doped k@xmath4so@xmath1:fe@xmath0 . for k@xmath4so@xmath1 as well k@xmath4so@xmath1:cd@xmath0 the melting temperature @xmath53c and the crystallization temperature are almost identical with negligible hysteresis .
in contrast , the higher dopant concentration in k@xmath4so@xmath1:fe@xmath0 leads to remarkable hysteresis of 0.4k and shifts the melting point to @xmath54c .
in analogy to fig . 6 of @xcite which represents a model for the @xmath55 phase transformation , an extended two phase region @xmath56/_liquid _ can explain the shift of @xmath57 and the hysteresis .
k / min ) near the melting temperature for undoped k@xmath4so@xmath1 and for crystals doped with cd or fe , respectively.,scaledwidth=40.0% ] for k@xmath4so@xmath1 and k@xmath4so@xmath1:cd@xmath0 crystals.,scaledwidth=40.0% ] the widths of the k@xmath4so@xmath1:cd@xmath0 melting and crystallization peaks are slightly larger compared to the undoped k@xmath4so@xmath1 , as a result of the minor cd@xmath0 dopant concentration .
it is remarkable , however , that this impurity leads in the end to improved optical properties .
[ fig6:exmp ] shows that the optical transmission of k@xmath4so@xmath1:cd@xmath0 is higher then the transmission of k@xmath4so@xmath1 for all wavelengths shown here ( @xmath58 nm ) .
this may partially be a result of scattering centers in the crystals that were grown from solutions containing cd@xmath0 , as cd@xmath0 doped samples can already with the naked eye be recognized to possess a higher optical clarity .
nevertheless it is remarkable that the absorption edge is shifted toward the vuv region , resulting in an increased energy gap of k@xmath4so@xmath1:cd@xmath0 compared with undoped k@xmath4so@xmath1 .
pure k@xmath4so@xmath1 and cd doped k@xmath4so@xmath1 crystals can be grown from aqueous solution by the slow evaporation technique .
the incorporation of cd is weak : from solutions with 4wt.% cdso@xmath1 one obtains cd@xmath0:k@xmath4so@xmath1 crystals with only 0.014wt.% of the dopant .
nevertheless , the influence of cadmium is remarkable : cd doped crystals contain less water , and the remaining water is loosely bond compared with water in undoped crystals .
the optical transparency of cd@xmath0:k@xmath4so@xmath1 in the uv region is larger . | single crystals of pure and cd@xmath0 doped potassium sulfate were grown from aqueous solutions by the slow evaporation technique . from nutrient solutions with a cdso@xmath1 concentration of 4wt.% crystals containing 0.014wt.%
dopant concentration could be obtained .
the x - ray diffraction patterns of powdered crystals confirmed their crystal structures for both cases .
thermal analysis of pure crystals shows that the @xmath2 phase transformation peak around @xmath3c is superimposed with spurious effects , while for cd@xmath0 doped crystals this is not the case .
the thermal hysteresis of the phase transition is 8k for undoped k@xmath4so@xmath1 and is reduced to 3.5k for k@xmath4so@xmath1:cd@xmath0 .
compared to undoped crystals , the optical transmittance of cd@xmath0 doped crystals is higher . , , , , and a. inorganic compounds , b. crystal growth , c. differential scanning calorimetry ( dsc ) , c. thermogravimetric analysis ( tga ) , d. phase transitions 61.10.nz , 61.50.ah , 67.80.gb |
within our solar system , earth and smaller bodies are primarily mixtures of refractories , rock and metals . in the outer solar system ,
bodies that are too small to retain deep atmospheres contain rock and ices . in the larger planets , including uranus and neptune , the light elements h and he dominate the volume .
there no local examples of bodies intermediate in size or mass between earth ( @xmath0 , @xmath1 ) and uranus / neptune , both of which are larger than 3.8 @xmath2 and more massive than 14 m@xmath3 . mass determinations of transiting exoplanets are beginning to allow the characterization of planets in this size range . as more planetary masses and radii
are measured , their bulk densities will provide more constraints on their compositions . to derive meaningful planetary densities
requires both accurate mass and radius determinations .
the ratio of the planetary radius to the stellar radius is a direct measurement from transit light curves , where the fraction of starlight blocked during transit is a simple measure of the projected area of the planet .
the uncertainty in this measurement typically rests on the accuracy to which the stellar radius can be constrained .
the star 55 cancri is unique as the host of a transiting sub - neptune exoplanet in having a direct measurement of its radius via interferometry @xcite .
high resolution spectral classification of the atmosphere of _ kepler _ host stars gives a model dependent measurement of the stellar radius with an uncertainty of typically 10% .
the measurement of stellar radius from spectral classification and modelling can be improved upon with additional information .
the gold standard for this purpose is where asteroseismic oscillations are detected in the photometric light curve . amongst these stars , uncertainties in mass and radius
can be reduced to @xmath4 , although detections are only available for giant stars and the brightest dwarf stars @xcite .
another constraint on the stellar density , and hence its radius , can be gleaned from the orbital constraints of exoplanets . the scaled semi - major axis ,
@xmath5/@xmath6 , ( where @xmath7 is the stellar radius ) can be estimated roughly from the transit and ingress durations , but accurate measurement of @xmath5/@xmath6 requires additional information about the orbital eccentricity and alignment ( @xmath8 ) , the longitude of pericenter from the passage of the orbiting planet through the sky - plane towards the earth . for a measured fractional transit depth @xmath9 , transit duration @xmath10 , and ingress or egress duration @xmath11 , @xmath12 @xcite .
for the purposes of measuring the stellar radius , information about the the orbital eccentricity from dynamical fits , used in eq .
[ arstar ] , can provide an independent constraint on the stellar bulk density following kepler s third law ( @xcite ) : @xmath13 here @xmath14 is the bulk density of the star , @xmath15 is the gravitational constant , and @xmath16 is the orbital period of the transiting planet . in their study of kepler-11 , @xcite used dynamical solutions for the orbital eccentricities , alongside high resolution spectra to reduce the uncertainty on the stellar radius to 2% .
the planetary radii were measured with nearly that precision . of the transiting exoplanets ,
the majority of mass determinations to date are the result of radial velocity ( rv ) spectroscopy , and amongst these , most are the so - called hot jupiters , planets with substantial envelopes that orbit very close to their host star .
the transiting neptunes and sub - neptunes with measured rv masses all have short orbital periods , the longest being kepler-68 b , @xcite at 5.4 days .
note that kepler-18 s planets have had their masses measured by the combined constraints of rv and transit timing variations ( ttvs ) , the farthest one orbiting every 14.9 days @xcite .
ttvs exploit the high degree of accuracy in measuring the transit times of transiting exoplanets , with transit time uncertainties as low as a few minutes in some cases .
these probe interplanetary perturbations , and in general are sensitive to the mass ratio of perturbing neighboring planets to the host star .
the strongest signals in ttvs occur when planets are near ( but not trapped in ) mean motion resonances , and the resonant argument cycles with a periodicity that is well sampled over a baseline of transit timing measurements . near first order resonances , the coherence time is long enough for perturbations to build constructively to an easily detectable amplitude . too far from resonance ,
the perturbations lose their coherence rapidly , and the ttv amplitude is reduced . too near resonance , _ kepler
_ s four years of observations do not cover a complete cycle .
nevertheless , since ttvs are sensitive to interplanetary perturbations and not the effect of the planet on the star , planetary masses can be determined to far greater orbital periods than rv , as long as enough transit times have been measured to detect timing variations .
the super - neptune kepler-30 d , with an orbit period of 143 days @xcite has the longest orbital period for an exoplanet with a mass determination on the mass - radius diagram .
the difference in orbital periods probes by these two separate techniques highlight their respective biases .
rv planets orbit close to their star , are hotter , and several , particularly the earths ; including kepler-10 b @xcite , corot-7 b @xcite , koi-94 b @xcite , kepler-20 b @xcite and potentially kepler-18 b @xcite seem to lack deep atmospheres .
the ttv mass determinations , such as kepler-11 ( @xcite ) , kepler-18 @xcite , kepler-30 @xcite and kepler-36 @xcite are all either compact multiplanet systems with a small period ratio between neighboring planets , or near resonance .
although detecting a ttv signal certainly favors larger masses , the orbital stability of compact multiplanet systems like kepler-11 require smaller masses and/or eccentricies given their neptune - like planetary radii . on a practical note , a compact configuration with high multiplicity
reduces the risk of intermediate , non - transiting planets confusing ttv dynamical models .
koi-152 ( with a _ kepler _ magnitude 13.9 , located at ra = @xmath17 , dec = @xmath18 @xcite ) is an excellent candidate for ttv modeling .
it has four planetary candidates near first order mean motion resonances , with clear evidence of transit timing variations ( ttvs ) .
@xcite first noted ttvs at koi-152 following 6 quarters of data , though only three candidates were known at the time .
these three are near the 1:2:4 resonance suggesting that multi - planet resonances are reasonably common @xcite .
@xcite noted the ttvs at koi-152 , and also found evidence that the inner two planetary candidates have significant eccentricities .
this suggests that circular fits to the transit times would be hampered by the mass - eccentricity degeneracy ( @xcite ) , casting uncertainty on measured masses . in section 2
, we introduce our methodology for measuring transit times , and in section 3 , we examine the transit timing variations , evaluating the applicability of zero - eccentricity analytical solutions for the planetary masses . in section 4
we describe our numerical models and fits to the transit timing variations , and present our results for the planetary masses and orbital parameters , and stellar parameters .
our analysis confirms that the candidates of koi-152 are planets , and we refer to them as such for the remainder of the paper . in section 5 we consider the potential effects of non - transiting perturbers on our four - planet model . in section 6 we characterize the planets masses , radii and by bulk densities , and in section 7 , we compare our results for koi-152 s planets with other known plamets of masses less than 30 @xmath19 on the mass - radius diagram .
variations in the brightness of koi-152 were monitored with an effective duty cycle exceeding 90% starting at barycentric julian date ( bjd ) 2454964.512 , with all data returned to earth at a cadence of 29.426 minutes ( long cadence , lc ) ; data were also returned at a cadence of 58.85 seconds ( short cadence , sc ) beginning from bjd 2455093.216 . here and throughout we base our timeline for transit data from t = jd-2,454,900 .
our analysis uses short cadence data where available , augmented by the long cadence dataset primarily during the epoch prior to t@xmath20 days , for which no sc data were returned to earth .
we obtained these data from the publicly - accessible mast archive at http://archive.stsci.edu/kepler/ .
to measure the transit times from the light curve , we adopt the procedure explained in detail in appendix 7.1 of @xcite . here and throughout we refer to the planets as ` b ' ( koi-152.03 ) , ` c ' ( koi-152.02 ) , ` d ' ( koi-152.01 ) , and ` e ' ( koi-152.04 ) in order of their orbital periods .
we begin with the orbital periods based on a linear fit to the observed transit times , summarized in table [ tbl - intro ] .
we shall solve for the orbital parameters of these planets at t = 780.0 days , an epoch chosen to be near the middle of our dataset .
for each candidate , the first transit time after this chosen epoch , calculated from a linear ephemeris to the set of transit times is at time @xmath21 .
.a linear fit to sixteen quarters of _ kepler _ s observed transit times for the planets at koi-152 , specified as orbital periods and the first calculated transit time after t = 780 days .
[ cols="^,^,^",options="header " , ] [ tbl - planet ] table [ tbl - planet ] shows our measured masses , radii , densities and incidence fluxes for each planet based on our best fit solutions , with uncertainties extended to account for all four sets of transit times .
the planetary bulk densities follow from the constraint in stellar density : @xmath22 each of the terms in parentheses in equation [ rhoplanet ] is an independent source of uncertainty that we add in quadrature .
we use this formula because it provides a more appropriate accounting of the uncertainties than the uncertainties in the planetary masses and radii . due to the tight constraints on stellar and planetary radii ,
the dominant source of uncertainty here is planet - to - star mass ratios from our dynamical models .
nevertheless , all four candidates can be characterized as having low bulk density , due to the retention of volatiles .
koi-152 b has the highest bulk density ( albeit with large uncertainties ) , despite having a similar radius to koi-152 c and e. the nominal density of koi-152 b is slightly less than a pure ice world of the same mass , suggesting either a substantial mass fraction of water and/or a relatively thin h / he envelope , with heavy elements in the interior .
koi-152 c and e have similar characteristics to the planets of kepler-11 d and e ( @xcite ) , and likely have envelopes that are far more significant by volume than by mass @xcite .
the bulk density of koi-152 d is the lowest for a _ kepler _ planet measured to date .
although the temperature of koi-152 d is unknown , its place on the mass - radius diagram appears to give its envelope a total mass ( as a fraction of the planetary mass ) of around 50@xmath23 at t = 500k , roughly 10@xmath23 at 1000 k , @xcite .
the equilibrium temperature of a planet assumes zero albedo and no internal heat source , and depends solely on the temperature of the host star and the orbital distance . for low eccentricities ,
@xmath24 given its distance from the star , the equilibrium temperature of koi-152 d is 634@xmath25 16 k. hence , it appears reasonable that koi-152 d has an h / he envelope that contributes significantly more than 10@xmath23 of the planetary mass , but less than 50@xmath23 of the mass .
here we plot neptunes and sub - neptunes on the mass - radius diagram , including all planets with measured radii and masses , with nominal mass determinations up to 30 @xmath19 .
these include mass determinations by radial velocity spectroscopy ( rv ) as well as ttvs .
we adopt the data from published studies that include the most recent stellar spectral classification and mass determinations for : hat - p-11 b ( kepler-3 b ) @xcite hat - p-26 b @xcite , 55 cancri e ( @xcite ) , gj 3470 b @xcite , gj 436 b ( @xcite ) , gj 1214 b ( @xcite ) , corot-7 b ( @xcite ) , kepler-4 b , @xcite , kepler-10 b @xcite , kepler-11 b - f @xcite , kepler-18 b - d @xcite , kepler-30 b and d @xcite , kepler-36 b and c @xcite , kepler-68 b @xcite , and koi-94b @xcite .
we exclude koi-94 c ( @xmath26 ) and koi-94 e ( @xmath27 ) from the mass - radius diagram because their masses are poorly constrained , and in the case of koi-94 e , its nominal mass is beyond 30 @xmath19 .
for the solar system , we adopt data given in @xcite . the planets in fig .
[ fig : m - r ] are all compared to theoretical models of pure water ice , silicate rock , or iron planets .
model curves for planetary radii of planets made from pure ice , rock or iron follow the results of @xcite .
several features stand out from fig .
[ fig : m - r ] .
firstly , amongst the sub - neptunes , ttv planets are systematically larger and hence less dense than rv planets in the same mass range .
this is most likely due to the biases of both techniques .
since rv detectability declines rapidly with orbital distance , few rv masses have been measured beyond a few days orbital period , and several of the planets in the rv sample may well have suffered mass loss from the intense stellar radiation upon their atmospheres . on the other hand ,
the ttv systems are all dynamically compact multiplanet systems , where for a given planetary size , smaller masses are more likely to permit stable orbits ( even though a larger mass makes the ttvs more detectable ) .
furthermore , for planets in multiplanets systems of a particular mass , the larger ( and hence lower density ) planets can have their transit times measured more accurately for ttv anlysis .
the bottom panel of fig .
[ fig : m - r ] highlights the two orders of magnitude in the range in planetary densities that are observed in the mass range up to 30 @xmath19 .
the rv density determinations appear more tightly correlated than the ttv determinations , although the uncertainty in density for the low mass , low density ttv planets are certainly larger .
note that in this plot we show planetary radii from a mass limited sample .
for given masses , and thus given detectability considerations , larger planets have more precise transit timing measurements .
there is thus a bias in ttv detections to large , low density planets , and not massive , smaller planets .
this selection effect is part of the reason that the @xmath28 curve declines to the right .
nominally , koi-152 d has the lowest planetary bulk density of exoplanets measured to date , although kepler-12 b , with a mass of 137 @xmath19 , and radius of 19 @xmath2 , has a very similar bulk density within uncertainties .
koi-152 d is twenty times less massive than kepler-12 b , and also receives roughly 30 times less insolation than kepler-12 b @xcite .
it is significantly less dense than other characterized sub - saturn mass planets .
if we parametrize the ratio of the planetary radius to the radius of a planet of the same mass composed purely of ice , or rock , we have a simple test of whether the planet has retained a substantial gaseous envelope .
planets that are less dense than water ice , ( above the blue line in the top panel of fig .
[ fig : iceline ] ) , are likely to have a volumetrically substantial h / he envelope .
planets that are more dense than pure silicate rock , ( below the brown line in the bottom panel of fig .
[ fig : iceline ] ) , are less likely to retain a substantial amount of volatiles . between these two limits , the atmosphere of a planet can not be determined without more information , since it is unknown in what proportions volatiles are likely to be present in the form of ices and gases .
nevertheless , we surmise from fig .
[ fig : iceline ] that planets with @xmath29 appear unlikely to retain significant atmosphere , and from the bottom panel that planets with @xmath30 are very likely to retain deep atmospheres .
all the characterized planets in our chosen mass range show remarkably little scatter in fig .
[ fig : iceline ] . in fig .
[ fig : rfrockline ] , we show the equilibrium temperature of the planet against their radii relative to pure rock or ice .
we see an abrupt transition around 1050 k. at higher temperatures , there is no evidence of planets harboring substantial gases , and below this temperature , a wide range of densities appears permissible .
one possible exception is kepler-4b , with a equilibrium temperature of 1614 k , @xmath31 , and @xmath32 .
@xcite estimated that kepler-4b retains a deep h / he atmosphere of about 4 - 6@xmath23 by mass .
of the hot " planets below 30 @xmath19 , kepler-4b is alone in requiring a deep atmosphere .
nevertheless , we still see a much wider range in bulk densities amongst the cool sub - neptunes .
kepler-36 stands out as having two planets that are close in orbital periods ( 13.84 and 16.24 days respectively , @xcite ) , with remarkably different densities .
these planets appear to straddle the transition in equilibrium temperature at @xmath33 k , with the outer planet retaining an atmosphere , and the inner one denser than pure silicate rock @xcite .
koi-152 b lies close to kepler-36 c on this plot , with a blackbody temperature of 1004 @xmath25 24 k , although it is marginally denser . amongst the `` cool '' sub - neptunes , no exoplanets that are denser than pure rock have yet been discovered
assuming a four - planet model for the ttvs in the four transiting candidates of koi-152 , we have added four sub - uranus masses to the mass - radius diagram , bringing the total to twenty exoplanets .
we confirm the planetary nature of these candidates and that they are planets in the same system .
the planets of koi-152 appear to follow the trend of the planets orbiting kepler-11 , being of low density with significant envelopes .
nevertheless , the planets show remarkable variety in their bulk masses .
koi-152 b , c and e all have radii between 3.5 and 4.0 @xmath2 , and yet their masses range from @xmath343 to @xmath3413 @xmath19 , and their bulk densities range from 0.3 to 1.6 g @xmath35 .
the largest planet , koi-152 d , ( @xmath36 ) , has a remarkably low mass given its size , and most likely has the lowest nominal bulk density amongst known planets with sub - saturn masses . | koi-152 is among the first known systems of multiple transiting planetary candidates @xcite ranging in size from 3.5 to 7 times the size of the earth , in a compact configuration with orbital periods near a 1:2:4:6 chain of commensurability , from 13.5 days to 81.1 days .
all four planets exhibit transit timing variations with periods that are consistent with the distance of each planet to resonance with its neighbors .
we perform a dynamical analysis of the system based on transit timing measurements over 1282 days of _ kepler _ photometry .
stellar parameters are obtained with a combination of spectral classification and the stellar density constraints provided by light curve analysis and orbital eccentricity solutions from our dynamical study .
our models provide tight constraints on the masses of all four transiting bodies , demonstrating that they are planets and that they orbit the same star .
all four of koi-152 s transiting planets have low densities given their sizes , consistent with other studies of compact multiplanet transiting systems .
the largest of the four , koi-152.01 , has the lowest bulk density yet determined amongst sub - saturn mass planets . |
hierarchical models of structure formation suggest that dwarf galaxies are the building blocks of larger galaxies , merging at high redshift to form the distribution of galaxies we see today .
dwarfs , and even smaller stellar systems , are also responsible for the origin of some of the lyman @xmath9 forest absorption features observed in the spectra of distant quasars ( fransson & epstein 1982 , wang 1995 , ciardi & ferrara 1997 ) .
it is therefore crucial to understand both their formation and evolution . in turn , it has recently become clear that supernova - driven winds play a crucial role in the evolution of such objects since they regulate the mass , metal enrichment , and energy balance of the interstellar medium ( ism ) in these galaxies .
these aspects have been investigated in a number of theoretical papers ( larson 1974 ; dekel & silk 1986 ; silk 1987 ; vader 1986 , 1987 ; and ferrara & tolstoy 1998 , hereafter ft ) both in local and cosmological contexts .
the observational evidence in support of the existence of outflows from dwarf galaxies has grown rapidly in recent years .
meurer ( 1992 ) found an expanding bubble of ionized gas with @xmath10 kpc size and velocity of order @xmath11 km s@xmath1 in the active dwarf galaxy ngc1705 , and martin ( 1996 ) found evidence for superbubbles in i zw 18 in which the ionized gas expands with velocities of several tens of km @xmath12 .
similar results have been obtained by della ceca ( 1996 ; see also heckman 1995 ) from asca observations of ngc 1569 , a star - forming dwarf galaxy in which the authors have been able to demonstrate the presence of diffuse hot gas ( @xmath13 kev ) very likely heated by supernova explosions .
bomans , chu , & hopp ( 1997 ) found x - ray emission from hot gas within a supergiant shell in the dwarf irregular ngc 4449 , which also has a large - scale radio synchrotron halo ( klein 1996 ) . in their search for outflows in dwarf galaxies marlowe ( 1995 ) conclude that the outflow phenomenona are relatively frequent in centrally star - forming objects ; in addition they point out that these outflows tend to be generally oriented along the galaxy minor axis . on a smaller scale ,
puche ( 1992 ) in their vla hi study of dwarf galaxies , have detected low - density bubbles surrounded by a shell expanding with velocity @xmath14 km s@xmath1 .
an absorption line study ( bowen 1996 ) from low column density gas in the dsph galaxy leo 1 has given upper limits on the amount of hydrogen present in the halo of such galaxy ( @xmath15 ) .
this low value seems to be in contradiction with the expectations from simple blowaway models in which the gas content of the galaxy has been ejected into the halo .
however , as the authors stress , the possibility that a significant fraction of _ hot _ gas exists along the line of sight can not be ruled out .
( alternatively , even if the galactic gas had indeed been ejected , it could have been stripped either by ram pressure against the intergalactic medium ( igm ) or by a galactic encounter . ) in many cases these observed gas velocities clearly exceed the escape velocity from the potential well of the parent galaxy , and therefore this material will be injected into the igm and lost to the galaxy .
however , it is difficult to determine what fraction of the ambient medium is indeed brought to such high speeds .
mass can be ejected from a galactic disk through the `` blowout '' of a superbubble driven by multiple supernova remnants , carving a hole through the galactic disk ( mac low & mccray 1988 ) .
gas accelerated above the escape velocity is then `` blown away '' , and escapes the potential of the galaxy entirely ( de young & heckman 1994 ) . determining the conditions under which blowout occurs , and under which most of the ism can be blown away are necessary steps towards the understanding of the evolution of such objects , and also of the possible relationships between early and late type dwarfs . in particular ,
for blowout there is yet no clear answer to the question of what fraction of the gas swept up in the shell is accelerated above the escape velocity .
the reason for this is that most previous studies in which blowout has been simulated numerically ( tomisaka & ikeuchi 1986 , 1988 ; mac low 1989 ; tenorio - tagle , rozyczka , & bodenheimer 1990 ; silich 1992 ; mineshige , shibata , & shapiro 1993 , suchkov 1994 ) have not been followed for long enough evolutionary times .
moreover , these models have been almost totally concerned with superbubbles and starbursts in massive galaxies ; thus it is not straightforward to extrapolate their results to dwarfs , which have rather different properties , both in terms of their ism and , perhaps more important , of their dark matter content .
de young & gallagher ( 1990 ) did model a dwarf galaxy undergoing a starburst , and concluded that about @xmath16 of the mass of the swept - up shell can escape the galaxy .
however , their study neglected the presence of dark matter in the galaxy , and assumed a thin gaseous disk with a scale height of @xmath17 pc , as well as suffering from insufficient numerical resolution , leaving their conclusions questionable .
ideally , one would also like to find how the fraction of mass that can escape depends on the metallicity of the gas ; the cooling provided by metals can be strongly inhibited in the relatively unpolluted gas of dwarfs , whose metallicity is typically more than one order of magnitude below solar ( skillman , kennicutt , & hodge 1989 ) . a related important point concerns the fate of metals ejected by supernovae exploding in the burst . the perceptive idea introduced by vader ( 1986 ) that outflows can be `` metal enhanced '' ,
although appealing , needs to be quantitatively examined . de young & gallagher ( 1990 ) did try to investigate the fate of metals in their model .
they concluded that most of the metal - rich supernova ejecta are indeed ejected , but that as much as 1/3 of the ejecta end up within the swept - up shell surrounding the hot bubble .
however , this appears to have been due to unphysical diffusion of their tracer particles from the hot interior gas of the bubble into the surrounding cold shell , due to the low numerical resolution of their model .
if this result were valid , then the concept of metal - enhanced winds would be much less viable , since the metal - to - gas ratio in the wind could not achieve values much higher than in the ambient ism of the galaxy . given the apparent problems with the computation , however , it seems worthwhile to reexamine this issue . in summary ,
the present study concentrates on dwarf galaxies , and attempts to answer the following questions : 1 . what are the conditions for either blowout or blowaway to occur ( see our definitions in
3 ) ? 2 .
what fraction of gas escapes the galaxy when they do ? 3 . what is the fate of metals ejected by the central supernovae ?
we investigate these problems taking into account the full structure of dwarf galaxies , by making physically meaningful models for the structural properties of dwarf galaxies in order to determine the characteristics of the ambient medium in which the multiple supernova explosions take place .
we combine analytic calculations with numerical hydrodynamic simulations using zeus-3d , a code using the algorithms described by stone & norman ( 1992 ) .
the plan of the paper is as follows . in
2 we describe our model for the structure of dwarf galaxies , based on recent data on their dark matter and gas content . in
3 we describe how we perform our numerical computations . before presenting the results of these experiments in 5 , we derive in 4 simple analytical conditions for the occurrence of blowout and blowaway as a function of the mass of the galaxy and the mechanical luminosity of the central supernovae .
6 contains discussion and a brief summary of the paper .
for our purposes , we model a dwarf galaxy as a system made of a visible disk with combined gaseous and stellar mass @xmath18 , and a dark matter halo of mass @xmath19 .
we assume that the gas has a density distribution given by @xmath20 , where @xmath21 , @xmath22 is the galactocentric radius , @xmath23 is the vertical coordinate , and @xmath24 is a function determined , as described below , by imposing hydrostatic equilibrium on the gas in the total gravitational potential of the dwarf galaxy , @xmath25 which consists of halo and disk components . the behavior of the dark - to - visible mass ratio @xmath26 in galaxies has been explored in great detail in a key study by persic ( 1996 ) .
these authors find that @xmath27 is a function of the galactic mass ; using their relations one can easily derive the dependence of this ratio on the visible mass of the galaxy : @xmath28 where @xmath29 . from this equation
it is clear that the gravitational potential of dwarf galaxies with @xmath30 is dominated by the dark matter halo ; therefore we will neglect the potential due to visible mass @xmath31 in all the following calculations involving the total potential @xmath32 .
the general trend of increasing dark matter fraction with decreasing galaxy mass is indeed consistent with other kinematical studies of dwarf spheroidal galaxies ( mateo 1997 ) for which velocity dispersions of a significant number of stars have been derived .
however , we caution that equation ( [ phi ] ) was derived by persic ( 1996 ) from a sample of larger galaxies with @xmath33 or @xmath34 . lacking more suitable data
, we are extrapolating those results to lower masses . to make this assumption clear throughout the paper we will keep @xmath27 indicated in a general form as long as possible . the density profiles of the dark matter haloes remains uncertain .
cold dark matter cosmological models predict that halo density profiles are essentially self - similar with a weak mass dependence ( lacey & cole 1994 ; navarro 1997 ) ; specifically the dark matter density should decrease as @xmath35 at small radii , where @xmath36 .
the detailed form of @xmath37 is found to be ( navarro 1997 ) @xmath38 where @xmath39 is the central density and @xmath40 is the characteristic radius within which the mean dark matter density is 200 times the present critical density , @xmath41 g @xmath42 , where @xmath43 km s@xmath1 mpc@xmath1 is the hubble constant ; throughout the paper we will use @xmath44 .
equation ( [ rhocdm ] ) holds down to @xmath45 kpc , where the resolution of the simulations becomes inadequate .
recent observational work , however , appears to disagree with this prediction .
salucci & persic ( 1997 ) have demonstrated convincingly that dark matter halos have a core , a central region of almost constant density , whose size increases with galaxy luminosity both in absolute units and as a fraction of the optical radius .
the structure of very small galaxies with mass @xmath46 is even less well understood . in view of the uncertain state of the art , we calculate the halo gravitational potential by assuming that the density distribution of the halo can be approximated by a modified isothermal sphere ( binney & tremaine 1987 ) , which is general enough to be appropriate for an idealized situation such as the one presented here , and does reproduce the observed central core .
it follows that @xmath47 the halo mass as a function of radius is then @xmath48 where @xmath49 .
thus , if @xmath50 , which we will show later on , @xmath51 where @xmath52 is an appropriately defined halo radius . following a common _ ansatz _ we take @xmath53 as obtained using the definition of @xmath40 given above and equation ( [ mh ] ) . to proceed
further we need a relation between the scale radius , @xmath54 , the central dark matter density @xmath39 , and the mass of the halo @xmath19 .
burkert ( 1995 ) has shown that the total dark matter inside @xmath54 , given by @xmath55 ( which implies @xmath56 ) , is related to @xmath54 and @xmath39 through the relations @xmath57 substituting for @xmath58 into equations ( [ grelsa ] ) and ( [ grelsb ] ) we get @xmath59 substituting the value of @xmath52 from equation ( [ rh ] ) into these expressions we obtain the dependence of the halo radius on @xmath19 , @xmath60 and can substitute back into equations ( [ r0 ] ) and ( [ rhoc ] ) , to find @xmath61 with these assumptions the gravitational potential of the halo is @xmath62 thus the galaxy has a circular velocity @xmath63 the rotation curve increases rapidly in the inner parts of the galaxy , already being practically flat at @xmath64 .
the asymptotic value of the rotation velocity at @xmath65 can then be found by using equations ( [ r01 ] ) and ( [ rhoc1 ] ) to be @xmath66 we tabulate the circular velocities for galaxies of various masses in table [ pars ] , taking @xmath67 and using equation ( [ phi ] ) for the mass ratio @xmath27 .
we then compute the gas density distribution by solving the steady state momentum equation , @xmath68 where @xmath69 , @xmath70 is the gas density , @xmath71 is an effective sound speed , @xmath72 is the gas sound speed , and @xmath73 its turbulent velocity . to compute @xmath54
, we again assume @xmath67 . in the horizontal direction ,
the gas is supported against the gravitational pull by both thermal pressure and rotation . in analogy with the dark matter component
, we suppose that the gas extends out to a disk cut - off radius , @xmath74 .
the latter quantity is obtained from a fit to a sample of dwarfs by ft .
they find that the radius - hi mass relation is well approximated by the law @xmath75 defining the constants @xmath76 and @xmath9 .
cutting the disk at an arbitrary radius does introduce a region out of hydrostatic equilibrium ; as a result the disk tends to expand at its own effective sound speed during the simulations . as this is rather a low velocity , it only results in a slight rounding of the sharply cut off outer edges of the disks .
we will use equation ( [ rc ] ) thoughout the paper , although one should be aware of the presence of some degree of uncertainty . for the sake of clarity
, it might be useful to give the explicit expressions for the relations among the three characteristic radii so far introduced in this section .
these are : @xmath77 the last equation justifies the approximation @xmath50 used in eq .
[ mh ] .
outside @xmath74 we postulate the existence of an igm whose present density , @xmath78 has been estimated by assuming @xmath79 and @xmath80 , giving @xmath81 g @xmath42 .
when @xmath82 has decreased to @xmath83 , we set it constant for the remainder of the grid .
as this gas is not in hydrostatic equilibrium , it tends to accrete onto the galaxy .
this efffect is negligible : if the entire grid were filled with gas of this density , it would have a mass of only @xmath84 , and if it accreted at a constant rate over a typical 200 myr run , it would give an accretion rate of only @xmath85 yr@xmath1 . with the above assumptions , the solution of equation ( [ hydroeq ] ) provides a reasonably self - consistent initial distribution for the gas in which supernova explosions are simulated .
the derived distribution is uniform to 1% in the equatorial plane of the galaxy .
a cut along the vertical axis through the galactic center shows that a gaussian distribution matches the actual distribution very well .
however , in order to provide in the simplest possible manner a tractable analytic estimate for the order of magnitude of the scale height @xmath86 , @xmath87 , and @xmath88 , the gas column density , we make the assumption that all the mass is concentrated in a thin disk , producing a constant gravitational acceleration @xmath89 , where @xmath90 is the total matter surface density , using the prescription for the ratio of dark to visible matter @xmath27 given by equation ( [ phi ] ) .
such a potential would generate an exponential gas distribution with parameters given by the following formulae : @xmath91 where , @xmath92 , @xmath76 is defined by equation ( [ rc ] ) in kpc , and @xmath93 is the mean molecular weight .
we tabulate in table [ pars ] these values as a function of @xmath18 for the given values of @xmath94 .
we use only @xmath95 from these values in the simulations , but they are useful for analytic estimates . we note that in this simple approximation , @xmath96 is independent of both @xmath27 and @xmath94 and , given that @xmath97 , only weakly dependent on the visible mass of the galaxy .
before we present the results of the numerical simulations , it is instructive to derive simple analytical relations that approximately give the final fate of the shocked gas and of the galaxy itself .
we distinguish between two potential results of a central starburst .
first , in a `` blowout '' , the central supernova explosions blow a hole through the galactic gas distribution , parallel to the steepest density gradient ( usually along the rotation axis ) , accelerating some fraction of the gas and releasing the energy of subsequent explosions without major effects on the remaining gas .
second , in a `` blowaway '' , all , or nearly all , of the ambient ism is accelerated above the escape velocity and is lost to the galactic potential well . in the following we derive in brief the conditions under which the two processes can occur , as a function of the mechanical luminosity of the starburst , and of the mass of the galaxy .
a more extended discussion can be found in ft .
the blowout condition can be derived by requiring that the blowout velocity @xmath98 exceeds the escape velocity of the galaxy @xmath99 .
the blowout velocity can be defined as the velocity at a height @xmath100 above the galactic midplane , where @xmath86 is the exponential scale height . since the velocity of a shock produced by an explosion in a stratified medium decreases down to a minimum occurring at @xmath101 before being reaccelerated in case of blowout
, the above definition of @xmath98 corresponds to the minimum of such a curve .
the blowout condition follows from an analysis of the kompaneets ( 1957 ) solution for an explosion in a stratified medium .
the explicit expression for @xmath98 has been obtained by ft : @xmath102 we recall that the gas mass for a vertically exponential , horizontally constant density distribution is @xmath103 using equations ( [ nhn ] ) , it follows that @xmath104 in general the effective sound speed @xmath94 has a contribution from a turbulent velocity , but we make no attempt to model its value , merely assuming that it is constant .
. [ vb1 ] can now be cast in the convenient form @xmath105 we note that the value of the blowout velocity does not depend on the radius @xmath74 , but only on the the mechanical luminosity of the explosion and the dark - to - visible mass ratio .
we now derive the necessary condition for a starburst to blow away ( completely unbind ) the ambient gas of the galaxy .
we have seen that blowout takes place when the shell velocity at @xmath100 exceeds the escape velocity . following the blowout the pressure inside the cavity drops suddenly due to the inward propagation of a rarefaction wave .
the lateral walls of the shell , moving along the major axis of the galaxy , will continue to be accelerated by the interior pressure until they are reached by the rarefaction wave at @xmath115 , corresponding to a time @xmath116 elapsed from the blowout .
after that moment , the shell enters the momentum - conserving phase , since the driving pressure has been dissipated by the blowout .
the requirement for the blowaway to take place is then that the momentum of the shell ( of mass @xmath117 at @xmath118 ) is larger than the momentum necessary to accelerate the gas outside @xmath118 ( of mass @xmath119)at a velocity larger than the escape velocity : @xmath120 with some simple assumptions and algebra outlined in ft , one can write the condition for the blowaway to occur : @xmath121 where the radius of the shell in the plane when the pressure is released is @xmath122 , and @xmath123 is the ratio of the major to the minor axis of the galaxy . using the expressions given in 2 for @xmath74 and @xmath86 , we find @xmath124 flatter galaxies ( larger @xmath125 values ) preferentially undergo blowout , whereas rounder ones are more likely to be blown away .
the condition for the blow away to occur , obtained expanding eq .
[ cond ] assuming that @xmath126 , is then @xmath127 we point out the strong dependence of blowaway on the amount of turbulence present in the galactic ism : as the value of @xmath128 is increased , less powerful explosions can lead to complete disruption of the galaxy .
this is because virialized objects with higher effective sound speed tend to be rounder , and hence more fragile with respect to blowaway .
the same type of situation might occur if for some reason the distribution of the gas is stretched and becomes less concentrated , as , for example , after the first episode of star formation , or following a galactic encounter .
thus after a first stage of the evolution in which blowout dominates , blow away might occur if the gas distribution can be puffed up enough . from the previous results we can predict the final fate of the galactic gas after a starburst episode . in figure 1 we show the different regions of the @xmath129 plane in which blowout or blowaway occurs . in this figure
we have used equation ( [ rc ] ) for @xmath74 , equation ( [ phi ] ) for @xmath27 , and equation ( [ eps ] ) for @xmath125 .
galaxies with gas masses below @xmath130 undergo blowaway almost independently of the values of the mechanical luminosity of the starburst in the range @xmath131 .
the blowout has a somewhat steeper dependence on @xmath132 : for the luminosities @xmath131 , blowout can take place up to masses of @xmath133 , respectively .
this , as we will see in the next sections , is in excellent agreement with the results of our numerical simulations .
for our numerical models , we use zeus-3d , a second - order , eulerian , astrophysical , gas dynamics code ( stone & norman 1992 ) using van leer ( 1977 ) monotonic advection .
we have implemented equilibrium radiative cooling using the cooling curve given by macdonald & bailey ( 1981 ) , which is a raymond & smith ( 1977 ) cooling curve extended to lower temperatures .
we use an implicit energy equation , implemented with a newton - raphson root finder , supplemented by a binary search algorithm for occasional zones where the tabular nature of the cooling curve prevents the newton - raphson algorithm from converging .
we also include an empirical heating function tuned to balance the cooling in the background atmosphere , but linearly proportional to density , so that it is overwhelmed by cooling in compressed gas ( mac low 1989 ) .
this is to prevent the background atmosphere from spontaneously cooling , and may be thought of physically as a crude model for the stellar energy input into the background atmosphere .
the gas distribution of the galaxy is set up in hydrostatic equilibrium with the dark halo potential given in equation ( [ grav ] , where @xmath39 is given by equation ( [ grelsa ] ) .
to find @xmath54 , we use equation ( [ r01 ] ) , taking @xmath67 , and proceeding as described following those equations . the central energy source is set up as a constant luminosity wind driven by a thermal energy source .
we note that strickland & stevens ( 1998 ) have shown that a single central energy source will more efficiently transfer energy to the ambient gas than multiple smaller clusters with the same total luminosity , as we describe in more detail in 6 .
a source region of radius five zones and volume @xmath139 is set up initially in pressure equilibrium with the background atmosphere , but at a temperature a factor @xmath140 larger than that of the background atmosphere , and density correspondingly lower by a factor of @xmath141 .
every timestep we then added energy to the source region at a rate @xmath142 , and mass at a rate @xmath143 , where @xmath144 is the midplane energy density of the background atmosphere , maintaining roughly constant specific energy in the source region , and generating a supersonically expanding wind .
the amount of mass in this wind is significantly greater than that expected from supernova ejecta alone , as we are trying to also account for mass that would have evaporated off the shell walls had we included thermal conduction in our model .
this is only a very rough approximation ; as we discuss below , the resulting temperatures are , in fact , somewhat cooler than they would be realistically , which gives us a good lower bound on the metal ejection rates that we compute .
we chose this energy input phase to last 50 myr in our models , the typical lifetime of the least massive star able to become a supernova ( e.g. , mccray & kafatos 1987 ) , effectively assuming simultaneous formation of all the stars in the starburst .
we assume azimuthal symmetry and so use in each of our models a two - dimensional grid with the center of the galaxy at the origin , with reflecting boundary conditions along the symmetry axis and along the galaxy mid - plane , and outflow boundary conditions on the other two axes .
we have performed a parameter study at a resolution of 25 pc per zone , varying the gas mass @xmath18 of the dwarf galaxies from @xmath145 m@xmath146 to @xmath147 m@xmath146 ( see table [ pars ] for equivalent rotational velocities and halo masses ) , and varying the mechanical luminosity of winds and supernovae from the central starbusts from @xmath148 ergs s@xmath1 to @xmath149 erg s@xmath1 .
these mechanical luminosities are equivalent to a supernova with energy of @xmath150 erg exploding every 3 million yr to every 30,000 yr , respectively . in this section
we report the results of these models , describing their overall development , computing the efficiency of mass and metal ejection , and making some rough approximations to their observable emission .
the results in this section on the occurrence of blowout and blowaway agree very well with the analytic results described in [ analytic ] and summarized in figure 1 .
we begin by showing in figure [ den ] a time history of the density distribution of all the models in the parameter study .
the first part of the figure shows the state of the gas at a time of 50 myr , at the end of the energy input phase . in the higher luminosity models ,
the internal termination shock can be clearly seen , surrounded by a hot , pressurized region of shocked wind and supernova ejecta .
this is in turn surrounded by a colder , radiatively cooled shell of swept - up ambient gas from the galaxy and surrounding igm .
( another minor effect that can be observed is that the ambient gas at the edge of each galaxy , which is not pressure confined , also begins to expand into the surrounding medium at its sound speed of 10 km / s , producing the rounded ends of the disks . )
the larger bubbles have begun to accelerate , and show strong rayleigh - taylor instabilities .
our assumption of azimuthal symmetry has the consequences described in mac low _
( 1989 ) for the rayleigh - taylor instabilities . in three dimensions ,
these instabilities produce a characteristic spike and bubble morphology . in azimuthal symmetry ,
only the central spike can grow , while the other modes are forced into rings . as a result ,
the central spike is more pronounced than it would be in three dimensions , and the other spikes are less numerous .
however , mac low
_ et al . _
( 1989 ) showed that changing to an assumption of slab symmetry ( _ e .
g. _ using cartesian coordinates ) did not prevent the appearance of a central spike .
although the instabilities do not fully develop because of these effects , they do function to effectively allow hot gas to accelerate beyond the colder , denser , swept - up shells . after energy input ceases , the holes in the planes of the galaxies recollapse under the influence of gravity and the pressure of the disk gas , except in the extreme cases of low mass and high mechanical luminosity , where the disk gas escapes the potential of the dark matter halo , and is swept completely off the grid at late times . at the same time , the hot bubbles rising above the disks expand out into the surrounding igm , leaving most of the swept - up shell material behind .
after 200 myr , at the last time shown , the galactic disks have almost completely recovered unless blow - away occurred .
meanwhile , low density gas originating in the central winds and supernovae spreads over regions of tens of kpc . in figure
[ col ] we show the distribution of metal - enriched wind and supernova ejecta material after 200 myr for the more massive models . in all cases in which blow - out occurs , this material escapes from the disks of the galaxies ; much of it is travelling at high enough velocity to escape from the halo potential as well .
this result should be fairly independent of the details of how the hot gas is traced , so long as hot gas is not allowed to unphysically cool or diffuse into the cold shell .
cooling can happen by numerical diffusion transferring too much mass into the hot interior from the cold shell , as occurred in the models of tomisaka & ikeuchi ( 1986 ) , for example , while diffusion can happen if tracer particles diverge away from the gas they should be following , as occurred in de young & gallagher ( 1990 ) .
physically , the hot gas will not radiatively cool within a dynamical time , and its sound speed is significantly higher than the escape velocity from any of our galaxies , so most of it should indeed escape as is seen in our models .
we can directly compute the efficiency of mass ejection and retention of metal - enriched gas from our models .
the question we ask is how much of each is travelling at speeds higher than the local escape velocity due to the potential of the dark matter halo . in principle , the efficiency of mass ejection is @xmath151 where @xmath152 is the mass on the grid moving at higher than the escape velocity , @xmath153 is the mass lost off the edge of the grid , @xmath154 is the mass initially on the grid at @xmath155 , and @xmath156 is the mass injected at the center of the grid by winds and supernovae during the first 50 myr . to get a good value for @xmath157
we must measure it after most of the gas has been accelerated to its final velocity but before it has left the grid .
we chose a value of 100 myr as a compromise .
although the shells from the lowest mass galaxies have already left the grid , they were moving at above the escape velocity as they went off , so it is reasonable to assume that all mass lost from the grid escaped .
we also need to make special provision for the tendency of the relatively large disks in the most massive models to expand sideways off the grid at the sound speed .
as no high - speed gas has left the grid in these models , we do this by neglecting @xmath153 for these models . in table
[ mass ] we show the ejection efficiency @xmath157 for each of our models . only in our most extreme models , with masses of @xmath145 m@xmath146 , is most of the mass ejected . in more massive objects , less than 7% of the mass
is ejected , usually far less .
the tracer field allows us to separately determine the fate of the metal - enriched gas ejected by the massive stellar winds and sne of the starburst .
we compute the efficiency of ejection of this material @xmath158 by applying equation ( [ effeq ] ) only to the gas carrying the tracer field . in table
[ metal ] we give @xmath158 for each model . in the more massive galaxies and at lower supernova rates , a significant fraction of the metal enriched gas is retained in the gravitational well of the dark halo , and will eventually fall back on to the galaxy , while at lower masses and higher luminosities , virtually all of the metals escape the grasp of the halo and travel freely into the surrounding intergalactic space .
when estimating the metal ejection efficiency @xmath158 , we have assumed that the metals ejected by massive stars and supernovae are well mixed within the hot interior of the bubble , and are therefore ejected uniformly with this gas .
two questions can be raised here . first ,
will metals ejected by later supernovae mix with the hot gas of the already existing bubble ?
a simple order of magnitude calculation ( tenorio - tagle 1996 ) shows that the metal diffusion time in the hot bubble gas ( mostly consisting of evaporated shell material ) is @xmath159 yr , where @xmath160 is the electron density , @xmath161 pc is the radius of the bubble , and @xmath162 k is the temperature of the hot gas . this time is comparable with , but shorter than the evolutionary timescale of the superbubbles studied in the previous section .
second , how will the mixing proceed if sn ejecta are strongly clumped ?
this might well be the case if rayleigh - taylor instabilities act during expansion of the shock through the outer layers of the star ( fryxell 1991 ) .
such high - density , high - metallicity clumps could travel almost unimpeded inside the hot , rarefied interior of the bubble , but would be rapidly thermalized when they hit the external shell . in this case
the metals would be mixed into the shell , rather than into the escaping hot gas .
the conditions under which the clumps can reach the shell , rather than being destroyed in the superbubble interior , have been derived by franco ( 1993 ) .
these authors found that for reasonable density contrasts in the clumps , only the clumps produced by the first 510 sns in the starburst can reach the shell and pollute it with their metal content .
the values of @xmath163 0.110 and the duration of the energy input phase considered in this paper of 50 myr imply @xmath10 15150 sns .
thus , ejecta clumping can play a role in the low - luminosity starburts , where we expect that a significant fraction of the metals contained in ejecta clumps ends up not in the hot gas , but rather mixed with the cool shell gas . for these objects
our @xmath158 might represent an overestimate ; this error should become fairly small for higher luminosity starbursts .
it has to be kept in mind , though , that the clumping factor of the ejecta is still far from well - established .
we have checked that our results are reasonably well converged numerically by rerunning one of our models with @xmath164 and @xmath165 erg s@xmath1 at a resolution of 40 zones per scale height , twice that of our standard models .
we find that , although details of the shell structure differ slightly , the overall morphology remains almost constant , and the computed efficiencies shift by no more than 10 - 20% , little enough for our conclusions to remain valid .
finally , we discuss some predictions of observable properties that we can draw from our models . as our assumptions about the microphysics of the gas are quite crude , we are only in a position to make qualitative predictions ; however we believe these can nevertheless be useful .
furthermore , we find ourselves in basic agreement with the more careful work performed by suchkov et al .
( 1994 ) for massive starbursts , adding credibility to our approximations .
we begin by computing the temperatures on our grid from the specific energies computed by zeus . to do this ,
we need to assume a composition for the gas , so we take it to be singly ionized with one atom of he for every 10 protons .
figure [ temp ] shows the resulting temperatures ( for the more massive models ) .
the hot gas on the grid appears somewhat cooler than physically reasonable .
this is most likely due to our replacement of the effects of conduction from the cold shell with the injection of an arbitrary amount of gas at the center of the superbubble as discussed in
these temperatures can then be used to make rough approximations to the emissivity in x - rays and in optical lines such as h@xmath9 . for optical emissivity
, we take @xmath166 while for x - ray emission , we take @xmath167 we chose @xmath168 to have an unrealistically low value of @xmath169 k in order to more clearly delineate the hot gas emissivity .
this choice may be justified for the reasons outlined above , but it should be remembered that these are only qualitative images of the distribution of emissivity . in figure [ opt ]
we show the optical emissivity , while in figure [ xray ] we show the x - ray emissivity .
note that these are two - dimensional slices across the galaxy , so they would have to be rotated around the axis and tilted to the line of sight before being directly compared to observations .
however , they do clearly indicate the regions expected to be bright in line emission and in x - ray emission .
probably the most important thing to note in this figure is the clear separation between the relatively dense filaments and the bulk of the hot gas .
deep imaging of optical lines often shows filamentary or bubble - like structures in these dwarf galaxies ( e.g. marlowe et al .
1995 ; della ceca 1996 ) ; as can be seen from figure [ opt ] these usually do not trace the outer edges of bubbles but often represent instead fragments of cold shell left behind during the blowout of hot gas to much greater distances .
we have explored the effects of stellar winds and sn explosions from starbursts with mechanical luminosities ranging from @xmath170 to @xmath149 ergs s@xmath1 on the ism of dwarf galaxies with mass @xmath145 to @xmath8 .
specifically , we have addressed in detail the following three questions : what are the conditions for blowout and blowaway to occur , as defined in
3 ? what fraction of gas escapes the galaxy in a blowout episode ? what is the fate of metals ejected by the massive stars during their lives and in their terminal sn explosions ?
we have studied the problem by means of both analytical and numerical techniques , checking where possible the agreement between the two types of results .
we have found that dwarf galaxies in the above mass range undergo blowout for moderate - to - high luminosity values ( @xmath171 ) whereas blowout is inhibited for galaxies with @xmath172 if @xmath173 , as shown in figure [ den ] and described in tables [ mass ] and [ metal ] . however , the fraction , @xmath157 , of the gas mass of the galaxy lost in such an outflow is surprisingly low in all studied cases : we find that more than a few percent of the mass is lost only in objects with masses as low as @xmath174 . in larger objects , the fractions drop to as low as @xmath175 ( table [ mass ] ) . in the lowest mass objects ,
the blowaway occurs virtually independently of @xmath6 .
thus it appears that explosive mass loss is a process involving a threshold determined by the geometry of the gas distribution as well as the depth of the potential well .
the coupling between the energy input and the disk gas becomes better for the smaller galaxies as they are more spherical and they need less energy input to begin with for blowaway to occur .
it is worth stressing that our results are dependant on the total mass of the galaxy , or the gravitational potential in which the gas is embedded .
although we have tried to model the dark halo carefully , using the best available observational and theoretical results , the precise value of the dark - to - visible mass ratio and the details of dark matter distribution remain quite uncertain .
our general results should be rather solid , but the specific values we quote for efficiency and the mass threshold for blowaway can only be taken as representative . on the other hand , metal - enriched material from stellar winds and sn ejecta is far more easily accelerated than the ambient gas to velocities above the escape speed .
we find that for @xmath7 less than 3% of the metal - enriched material is ejected from a galaxy with @xmath176 of visible mass , but this fraction increases to 60% for @xmath177 , and to unity for @xmath178 ( see table [ metal ] ) .
that is , the smaller galaxies inject virtually all metals produced by the massive stars of the starburst into the intracluster medium or igm .
stronger starbursts with higher mechanical luminosities eject metal - enriched material even more easily ; only for lower starburst luminosities do galaxies retain a large fraction of the heavy elements produced by massive stars .
our assumption that all energy injection occurs in the central hundred parsecs of the galaxy can questioned , as real starburst galaxies have multiple clusters scattered across the disk , which one might think could be more effective at blowing away the ism than a single , central energy source .
however , strickland & stevens ( 1998 ) have shown with a simple but elegant argument that our assumption is actually the one leading to the most effective blowaway for a given total luminosity .
the gist of their argument can be seen by considering a superbubble in a homogenous medium ( weaver et al .
1977 ) , whose radius @xmath179 , but whose kinetic energy @xmath180 . the coupling of a superbubble to the background ism can be characterized by the energy per unit volume supplied by the superbubble to the ism @xmath181 , which drops with decreasing luminosity .
thus , several small bubbles will be less effective at transferring energy to the ambient gas than one large one , and so multiple clusters will be less effective at blowing away the ism than our assumed single central cluster .
these results have several implications for the evolution of dwarfs .
first , outflows from dwarf galaxies should be strongly metal enriched . tables [ mass ] and [ metal ] show that the typical metal fraction of the outflow greatly exceeds that of a homogeneously mixed galactic disk .
this enrichment occurs because the metals produced by the massive stars in the starburst remain within the hot , shocked cavity gas , which has a high sound speed and so can easily escape the galaxy . on the other hand , it is much more difficult to accelerate the denser , cooler shell of swept - up ambient gas , so that gas remains bound .
de young & gallagher ( 1990 ) found that about 2/3 of the sn ejecta in the chimney escape from the galaxy for a case in which @xmath182 and @xmath183 . for similar parameters , table [ metal ]
shows that the blowout is prevented , due to the presence of the dark matter halo and a realistic hydrostatic distribution of the gas resulting in a larger scale height ( about three times larger than theirs ) .
this suggests that dark matter is key to the evolution of dwarf galaxies and any future models must include it properly .
our simulations show that after blowout has occurred , a substantial fraction of the hot , metal - rich gas escapes from the galaxy .
metals are then dispersed into the igm ( subject to the caveats discussed in the previous section ) . as a consequence
, it appears that dwarfs could be the major pollutors of the igm , and certainly have major effects on the environment in which they live .
we note the result by renzini ( 1997 ) , who finds that most of the iron in clusters appears to reside in the intracluster gas rather than inside galaxies , a clear sign that outflow phenomena are at work in that environment .
it is out of the scope of this paper to review the vast literature concerning these issues ; for a detailed discussion see ft .
we only mention that several authors have already argued that dwarf galaxies might be the most important objects for the igm enrichment : our calculations provide a solid basis for these previous works and should allow the development of more detailed scenarios .
starbursts in dwarf galaxies may be traced and studied in detail by x - ray and optical emission - line studies .
the results presented here show that huge gaseous halos , with sizes of dozens of kpc are produced , with regions of high x - ray emissivity close to the galactic disk .
relatively cool , dense filaments also occur near the galaxy , well within the external shock , due to shell fragmentation .
high spatial resolution spectra of the filaments may be useful to investigate the radiation field in the halo of dwarfs and the escape fraction of ionizing photons from massive stars in the disk , since this cool gas should be predominantly photoionized and hence show up in optical emission lines .
since these filaments are surrounded by the hot gas , the x - ray emission may actually be strongest in regions close to the filaments , where evaporation takes place .
thus , the bulk of the observed x - ray emission may come from these conductive interfaces , in which the gas is far out of ionization equilibrium , and hence emitting strongly .
( these effects are not included in figure [ xray ] ) . the filling factor and the nonequilibrium state of such regions
have to be taken into account properly in order to avoid overestimates of the mass and energy involved in the x - ray emission .
the temperature of the hot gas ( @xmath184 kev ) detected by della ceca ( 1996 ) in their observations of ngc 1569 is in qualitative agreement with the ones shown in figure [ temp ] .
we have not tried to apply our models to a specific object ; an interpretation of the soft x - ray emission from the dwarf galaxy ngc 5253 in terms of a superbubble model has been attempted by martin & kennicutt ( 1995 ) .
they concluded that a simple superbubble model with @xmath7 , exploding in a smooth ambient medium would underproduce the observed x - ray luminosity , @xmath185 ergs s@xmath1 , by a factor @xmath186 .
as an alternative , they suggest that a clumpy ambient medium generating extra mass of hot gas via cloud evaporation , may help reconcile the above discrepancy .
we emphasize that the required hot material could come from the evaporation of the shell fragments seen in our simulations , without the need for a multi - phase medium , whose existence in dwarf galaxies has still to be proven . due to their large number , dwarf
starbursting galaxies may significantly contribute to the x - ray background ( persic 1989 ) , although a precise estimate crucially depends on their x - ray luminosity function .
the halos created by the starburst energy injection might also be detectable via absorption line studies towards background objects such as qsos .
of course this type of experiment is limited by the low column densities in the large , rarefied halo , as well as the availability of suitably placed background objects , which could be a rare occurence for the small dwarf galaxies .
a particularly suitable case has been investigated by bowen ( 1996 ) .
the dsph leo 1 has three qsos in the line of sight though the halo ; however , the authors were only able to put upper limits on the presence of ionized species like civ , siv and mgii , but not to exclude the presence of a hotter diffuse component .
similar studies would be particularly important for dirr and bcd galaxies which are much more active and gas rich and , in general , for normal galaxies . finally , we recall that the blowout and blowaway episodes discussed here might be able to carry magnetic field lines along into the halo .
it is reasonable to expect that , if a sufficiently strong magnetic field does exist in dwarf galaxies , extended radio - halos should be found around such objects .
radio continuum observations of dwarf galaxies are still scarce ; probably the most complete survey has been presented in the pioneering work by klein 1984 , who concluded that , on average , blue compact dwarf galaxies exhibit about 10 times higher radio - to - optical luminosity than normal spirals and they have a flatter radio spectrum indicating a weak synchrotron radiation ( in turn possibly due to a weak magnetic field ) .
these findings have been largely confirmed by subsequent works devoted to dirrs ( klein 1986 ) and larger samples ( klein 1991 ) .
this idea will also apply to larger starburst galaxies .
we are grateful to e. tolstoy , d. bomans , r .- j .
dettmar , g. golla , and l. van zee for discussions of the observations .
d. strickland & i. stevens kindly allowed us to quote their results prior to publication .
computations were performed at the rechenzentrum garching of the max - planck - gesellschaft .
each author thanks the other s institute for hospitality during work on this paper . | we model the effects of repeated supernova explosions from starbursts in dwarf galaxies on the interstellar medium of these galaxies , taking into account the gravitational potential of their dominant dark matter haloes .
we explore supernova rates from one every 30,000 yr to one every 3 million yr , equivalent to steady mechanical luminosities of @xmath0 ergs s@xmath1 , occurring in dwarf galaxies with gas masses @xmath2 .
we address in detail , both analytically and numerically , the following three questions : 1 .
when do the supernova ejecta blow out of the disk of the galaxy ? 2 .
when blowout occurs , what fraction of the interstellar gas is blown away , escaping the potential of the galactic halo ?
3 . what happens to the metals ejected from the massive stars of the starburst ?
are they retained or blown away ?
we give quantitative results for when blowout will or will not occur in galaxies with @xmath3 .
surprisingly , we find that the mass ejection efficiency is very low for galaxies with mass @xmath4
. only galaxies with @xmath5 have their interstellar gas blown away , and then virtually independently of @xmath6 . on the other hand , metals from the supernova ejecta
are accelerated to velocities larger than the escape speed from the galaxy far more easily than the gas .
we find that for @xmath7 , only about 30% of the metals are retained by a @xmath8 galaxy , and virtually none by smaller galaxies .
we discuss the implications of our results for the evolution , metallicity and observational properties of dwarf galaxies . |
according to the experimentally realizable phases that support topological objects , the fractional quantum hall ( fqh ) effect , since its discovery @xcite , has appealed to tremendous theoretical and experimental attractions and achieved us a collection of methods to study the strongly correlated electron systems .
the quasiparticle excitations in the fqh liquids can have fractional charges and obey fractional statistics @xcite . within a serious of filling factors of fqh states ,
some of them may support more exotic excitations with non - abelian statistics , which have potential applications in the topological protected quantum computation @xcite .
the fqh with an even denominator on the first landau level at filling factor @xmath2 @xcite is the most studied state that belongs to the family of non - abelian fqh states .
since the seminal work of moore and read @xcite , a connection between the wavefunction of the fqh state and the conformal field theory ( cft ) has been established .
thereafter , a series of non - abelian fqh states have been proposed which are described by the @xmath3 topological quantum field theory @xcite .
the index @xmath4 describes the clustering properties in the model wavefunction and has a connection with its filling factor @xmath5 .
they are addressed read - rezayi @xmath1 quantum hall states since the corresponding wavefunctions can be calculated from the correlation functions in the @xmath1 parafermionic conformal field theory @xcite .
notablely , it was found @xcite that the @xmath1 fqh states are the exact ground state of certain hamiltonian with @xmath6-body interaction for all integers @xmath7 .
further more , bernevig et.al .
@xcite recently found that the homogeneous polynomial part of the @xmath1 fqh wavefunction can be obtained recursively from the jack polynomials which is one of the polynomial solutions for calogero - sutherland hamiltonian @xcite .
the @xmath1 fqh wavefunctions in this language are labelled by a negative parameter @xmath8 and a root configuration ( or partition ) .
because of its computation advantages , we thereafter use the jacks to produce the model wavefunctions for the ground state and their quasiparticle excitations in the following calculation .
the measurement of the transport properties of the quasiparticles propagating along the edge of the fqh states is crucial for identification of the topological nature of the systems . as standard practice in the noise and interference experiments @xcite , quantum point contacts
are introduced to allow quasiparticles propagating on one edge to tunnel to another .
this motivated us to study the quasiparticle tunneling amplitudes in fqh liquids in the disk geometry @xcite . on the disk
, we considered a tunneling potential along a specific direction @xmath9 . the tunneling amplitudes exhibit interesting scaling behavior , whose exponent is related to the conformal dimension and the charge of the tunneling quasiparticles .
specificaly , from an effective field theory analysis , we found @xcite that the bare quasiparticle tunneling matrix element satisfies a scaling function @xmath10 where @xmath11 and @xmath12 are the ground state and quasihole wavefunction respectively and @xmath13 is the conformal dimension of the quasiparticle operators .
@xmath14 $ ] is the edge - edge tunneling hamiltonian .
the function @xmath15 reveals the dependence on the tunneling distance @xmath16 .
we analyzed the tunneling amplitude of the abelian quasiparticle , such as the @xmath17 and @xmath18 quasiholes in the laughlin state at @xmath19 and @xmath20 quasihole in the moore - read state at @xmath2 .
as shown in table 1 of ref .
, an excellent agreement with the above relation @xmath21 was found .
however , for non - abelian excitations , such as @xmath22 quasihole excitation in the moore - read state , it is likely that there is a correction on the scaling parameter , i.e. , @xmath23 incidentally , the anomalous term can be written as @xmath24 where @xmath25 is the conformal dimension for the neutral part of the non - abelian quasihole operator .
we therefore , in the case of disk geometry , had an argument that this anomalous term may origin from the unconstrained tunneling of the neutral parafermions . however , because of the curvature the disk geometry , for a finite size system , the tunneling distance is limited by the radius of the system which is proportional to @xmath26 . a natural way to overcome
this problem is putting the electrons on cylinder or torus . comparing to the disk
, the cylinder geometry has advantages that there is no curvature difference between landau orbitals and the edge - edge distance is linearly proportional to the system size @xmath27 .
moreover , the edge - edge distance can be easily tuned via varying the aspect ratio , by comparison , in a longer range .
our recent work on the quasihole tunneling and entanglement entropy in laughlin state on cylinder @xcite found that a critial length scale of the edge - edge distance exists .
it can be explained as a threshold value that the two edges of the cylinder can be treated independently , or the effects of the edge - edge interaction can only be neglected while @xmath28 where @xmath29 .
therefore if we want to consider the scaling behavior of the quasiparticle tunneling between two independent edges with a longer tunneling distance , the cylinder geometry is a better choice than disk . in this paper
, we study the physical properties of the fqh liquids and reconsider the quasiparticle tunneling amplitudes scaling on cylinder , especially focus on the scaling of the non - abelian quasiparticles tunneling .
we aim to address the question of the existence of the anomalous term in the scaling parameter for non - abelian quasiholes in the whole region of tunneling distance .
the remainder of this paper is organized as follows .
section gives a brief review of the mode and previous results in the disk geometry .
the model of a quasihole tunneling on cylinder is introduced in section .
section is devoted to the scaling analysis for the abelian and non - abelian quasiholes in @xmath30 state .
in section we focus on the scaling analysis in other @xmath1 states such as the @xmath31 and @xmath32 cases .
section provides a conclusion and discussion of the paper .
in the disk geometry @xcite , we considered a single - particle tunneling potential @xmath33 , which breaks the rotational symmetry .
it defines a tunneling path for fqh quasiparticles under the gate influence at a quantum point contact .
the tunneling amplitude for a quasihole to the fqh droplet edge is @xmath34 . the most efficient way of getting these wavefunction is by the help of jack polynomial approachment @xcite .
generally , jacks belongs to a family of symmetric multivariate polynomials of the complex particle coordinates .
potentially , they can be the bosonic version of fqh wavefunction ( appending the ubiquitous gaussian factor @xmath35 on disk ) , or fermionic version with an extra vandermonde determinant @xmath36 .
a jack @xmath37 is parameterized by a negative rational number @xmath8 , which is related to the clustering properties of the wavefunction , and a root configuration @xmath38 , which satisfies a generalized pauli exclusion principle and from which one can derive a set of monomials that form a basis for the wavefunction . for the read - rezayi @xmath1 parafermion state , the @xmath39 and the corresponding root configuration can be expressed by a binary format `` @xmath40 '' in the occupation representation of landau orbitals .
taking the moore - read state with @xmath41 as an example , the root configuration for ground state is `` @xmath42 '' where the leftmost orbital represents the innermost landau orbital with the symmetric gauge . in order to vary the tunneling distance , a large number of abelian quasihole with charge @xmath20
were inserted at the center , namely a bunch of zeros are attached in front of the root configurations . therefore , the tunneling distance for a @xmath27-electron @xmath1 parafermion state is @xmath43 in which @xmath44 is the number of quasiholes inserted at the center .
the system evolves from disk to annulus and finally to a ring shape while increasing @xmath44 .
the tunneling distance @xmath45 in the one dimensional ring limit while @xmath46 . in this limit
, we had a conjecture that the tunneling amplitudes for abelian quasihole with charge @xmath47 for @xmath1 parafermion state is @xmath48 where we introduce the @xmath49 function @xmath50 .
unfortunately , we did not find a universal formula for the non - abelian quasihole tunneling amplitudes for @xmath1 fqh states except for the @xmath22 quasihole in moore - read state at @xmath41 : @xmath51 the unrescaled ( a ) and rescaled tunneling amplitude @xmath52 with @xmath53 ( b ) for e/2 quasihole in the mr state as a function of the tunneling distance @xmath16 on disk .
, width=302 ] before discussing the scaling behavior of the tunneling amplitudes of the @xmath1 parafermion states , we firstly write down the formula of the scaling dimension for the quasiholes . in @xmath1
parafermionic cft , the abelian quasihole operator is @xmath54 where @xmath55 is the charge bosonic field .
the abelian quasihole has charge @xmath47 and scaling dimension @xmath56 . on the other hand ,
the operator for the smallest charged non - abelian quasihole is @xmath57 in which the @xmath58 is the neutral spin fields which has scaling dimension @xmath59 .
therefore , the smallest charged non - abelian quasihole has charge @xmath60 and scaling dimension @xmath61 .
eq.([abelianformula ] ) gives the asymptotic scaling behavior for abelian quasiholes in the ring limit @xmath62 .
combining with the long distance behavior with a gaussian decay , we conjectured a scaling function as @xmath63 where @xmath64 is the scaling parameter .
[ diskhalf ] shows the bare and rescaled tunneling amplitudes for the abelian @xmath20 quasihole as a function of @xmath16 on a disk . with the scaling parameter @xmath65 , the data for systems from @xmath66 electrons collapse to the same value while @xmath45 which are in agreement with the scaling conjecture of eq.([scalingconj ] ) .
the rescaled tunneling amplitude @xmath67 for e/4 quasihole in the @xmath30 parafermion , or moore - read state as a function of the tunneling distance @xmath16 on disk .
, width=302 ] the more interesting is for the tunneling amplitudes of the non - abelian quasihole . in fig.[diskquarter ] , we present two rescaled data for @xmath22 quasihole in moore - read state .
it is shown that the data from different system sizes in the ring limit collapse well at @xmath68 instead of at @xmath69 . since the correction of the scaling parameter for @xmath1 with @xmath70 can be approximated by @xmath71 , we speculated @xcite that the anomalous scaling behavior for the non - abelian quasihole may origin from the effect of non - independent edges for neutral component , or the edge - edge interactions while in small @xmath16 . on the other hand , if we look carefully at the fig.[diskquarter](b ) , for the range @xmath72 , although few data points limited by the geometry and system size , there is still a hint that the scaling behavior works well in large @xmath16 regime without the neutral part correction .
therefore , a natural question is that whether the scaling conjection of eq .
( [ scalingconj ] ) works for large tunneling distance ? with this motivation , we reconsider the scaling behavior of the tunneling amplitudes for @xmath1 parafermion states on cylinder in the following section .
the sketch of a cylinder we used to make the quasihole tunnel from the left edge to the right under a tunneling potential @xmath73 .
@xmath74 is the circumference of the edge and @xmath75 is the length of the finite cylinder .
, width=302 ] as modelled in fig .
[ cylinder ] , for a cylinder with circumference @xmath74 in @xmath76 direction , the lowest landau level wave function for single electron in a magnetic field with landau gauge is : @xmath77 in which @xmath78 , @xmath79 are the equilibrium positions for each landau orbital . here
the magnetic length @xmath80 has been set to one .
the magnetic field is perpendicular to the cylinder surface and the number of orbits @xmath81 equals to the number of magnetic flux quantum penetrating from the surface . as each state or
each orbit occupies a constant area @xmath82 , the total area is @xmath83 for a finite size system .
the aspect ratio is defined as @xmath84 .
to study the quasiparticle tunneling on cylinder , we use a simple delta tunneling potential @xmath85 for a single particle similar to the disk geometry @xcite .
this potential allows a quasiparticle to tunnel along the @xmath86-direction while the y - direction does nt have translational symmetry .
the matrix element @xmath87 describes a particle tunneling from one single particle state @xmath88 to another state @xmath89 . if we set @xmath90 for convenience , it is easy to get the matrix element form as follows : @xmath91 where @xmath16 is the distance between two single states . in many - body case ,
the tunneling operator is just the summation of the single particle tunneling potential @xmath92 .
then the tunneling amplitude is obtained by calculating @xmath93 where @xmath94 and @xmath95 .
the matrix elements have non - zero components coming from the identical sets @xmath96 and @xmath97 except a single pair @xmath98 and @xmath99 which have constant value of momentum difference . taking moore - read state as an example ,
the @xmath20 case need the @xmath100 while the @xmath22 case need the @xmath101 where @xmath27 is the number of electrons .
thus , we easily get @xmath102 and @xmath103 , or the charge of the quasihole should appear in the gaussian factor .
in this section , we systematically study the scaling behavior of the tunneling amplitude in the case of @xmath30 , or moore - read state on a cylinder . in the language of the jack polynomial description , the moore - read state and its quasihole state
are labelled by root configurations @xmath104 , @xmath105 and @xmath106 respectively . here
the extra zero in the @xmath107 on the left means a flux quantum , or an abelian @xmath20 quasihole is created on the left edge of the cylinder .
the pattern in the @xmath108 means there are two @xmath22 quasiholes , one on each edge of the cylinder since the non - abelian majorana fermion modes are embedded in the @xmath22 quasihole excitation @xcite which must appear in pair . as was introduced in above section , the tunneling path of the quasihole in this case is the length of the finite cylinder in @xmath86-direction , namely @xmath109 . in this case
, we have a parameter @xmath0 that can smoothly tune the tunneling distance from zero to infinity . as a comparison , in the disk geometry
, the tunneling distance has a upper limit which is the radius of the system @xmath110 and @xmath111 for @xmath112 , thus the disk geometry is not suitable to large @xmath16 physics .
interestingly , in the limit of @xmath113 or @xmath114 , the two adjacent landau orbitals have practically zero overlap and thus the hamiltonian is dominated by the electrostatic repulsion .
this thin cylinder limit has ground state which is called a charge density wave state , or tao - thouless state @xcite on torus with occupation pattern @xmath115
. however previous studies in this limit @xcite shows that the topological properties of the fqh state does not change as varying @xmath0 , or there is no phase transition while varying the aspect ratio .
therefore , we feel more comfortable to say that the results calculated from eq.([tunnelingformula ] ) in the whole range of @xmath75 are really the tunneling amplitudes for fqh quasiholes . the bare ( a ) and rescaled tunneling amplitude @xmath116 with @xmath68 ( b ) for e/2 quasihole in the moore - read state as a function of the tunneling distance @xmath75 on cylinder .
the system size ranges from 8 to 18 electrons.,width=340,height=226 ] since the landau wavefunctions both on the disk and cylinder have the similar feature to gaussian wave , following the work on disk @xcite , we believe that the quasiparticle tunneling amplitude on cylinder has the same scaling behavior as that on disk in eq.([scalingconj ] ) . fig . [ halfcylinder](a ) shows that the tunneling amplitudes for @xmath20 quasihole as a function of the tunneling distance @xmath75 . by the help of jacks ,
the largest system size we have reached is @xmath117 electrons .
the data is the same as that on disk in fig .
[ diskhalf](a ) except that there are more data for large @xmath75 on cylinder . in the limit @xmath118 , or the cft limit , with the same argument as that on disk , the value of the tunneling amplitude are exactly the same as that in eq . ( [ abelianformula ] ) .
as @xmath75 increasing , the tunneling amplitudes dramatically drop to zero which dominated by the gaussian factor in eq.([scalingconj ] ) . in fig .
[ halfcylinder](b ) , we plot the rescaled data @xmath119 as a function of the tunneling distance . while @xmath120 , the scaling behavior is obviously better than that on disk as shown in fig .
[ diskhalf](b ) .
the data from @xmath121 electrons not only in the cft limit @xmath118 , but also in the large @xmath75 region collapse onto each other .
therefore we see that the scaling function still works , for charged @xmath20 abelian quasihole , in the region far away from the cft limit .
the next question is whether it is workable for the non - abelian quasihole tunneling amplitude , and whether the anomalous term in the scaling parameter still exist ? the unrescaled ( a ) and rescaled tunneling amplitude @xmath122 with @xmath68 ( b ) @xmath123 ( c ) for e/4 quasihole in the mr state as a function of the tunneling distance @xmath75 on cylinder .
, width=340,height=226 ] in fig .
[ quartercylinder](a ) , we plot the tunneling amplitude for @xmath22 quasihole as a function of @xmath75 for systems ranges from 8 to 18 electrons .
the values in the cft limit are still consistent to the eq.([quarterformula ] ) . comparing with the abelian case as shown in fig .
[ halfcylinder](a ) , the bare tunneling amplitudes for non - abelian @xmath22 quasihole do not decay monotonically as increasing tunneling distance .
there is a bump around @xmath124 which is consistent to the critial threshold value at which the interaction between two edges of fqh states can not be neglected @xcite .
the more interesting is that , as shown in fig .
[ halfcylinder](b ) , while the tunneling amplitudes are rescaled by eq.([scalingconj ] ) with a modified scaling parameter @xmath125 as expressed in eq.([modifiedalpha ] ) , the data approaching to the cft limit still scales very well .
this result has consistency with that in the disk geometry .
however , with this scaling parameter , the tunneling amplitudes for large @xmath75 do not collapse onto each other .
since the two edges of the cylinder in this case are getting more and more independent while increasing the edge - edge distance , it motivates us to look the scaling behavior without the correction term in eq.([modifiedalpha ] ) .
the results are shown in fig .
[ quartercylinder](c ) . here
we set the scaling parameter @xmath126 .
it is shown that the rescaled data for all systems scales very well in the range @xmath127 and obviously , the scaling behavior in the cft limit in this case is broken in this case . as a conclusion for this section
, we find that in the cylinder geometry , the tunneling amplitude for the abelian @xmath20 quasihole in the moore - read state obeys the scaling behavior of eq.([scalingconj ] ) with @xmath128 both in the region with short tunneling distance ( cft limit ) and in the region of long tunneling distance .
however , for the non - abelian @xmath22 quasihole , the above scaling parameter is only workable for @xmath127 where the @xmath129 is the critical distance , above which the two edges of the cylinder can be treated as two independent ones . as the previous study on disk , the data for the tunneling amplitudes in @xmath130 obey the similar scaling function with a non - trivial modification on the scaling parameter @xmath8 as expressed in eq .
( [ modifiedalpha ] ) .
to furtherly confirm that our conslusions for the case of @xmath41 are suitable for all the @xmath1 parafermion fqh states . in this section , we look at the scaling behavior and the related scaling parameter in the cases of @xmath31 and @xmath32 .
first of all , since we need to consider the relation between the scaling parameter @xmath8 and the scaling dimension of the quasiholes in fqh states , we list the quasihole charge , scaling dimension and its corresponding scaling parameters for abelian and non - abelian quasiholes in the following two tables .
.[tbl : abelianrr ] the scaling exponents for charge @xmath131 abelian quasiholes tunneling amplitudes scaling in the read - rezayi states .
@xmath132 is the scaling dimension for quasihole . [
cols="^,^,^,^",options="header " , ] the unrescaled tunneling amplitude @xmath133 ( a)(b ) and rescaled tunneling amplitudes @xmath134 ( c)(d ) for charge @xmath131 abelian quasihole in the @xmath1 states as a function of the tunneling distance @xmath75 on cylinder . ,
width=340,height=226 ] due to the limitation of storing the basis of the wavefunction , the system sizes for @xmath135 and @xmath136 parafermion states we considered are up to 21 and 28 electrons respectively . as shown in fig.[k34abelian ] , like in the @xmath30 case , the tunneling amplitudes for abelian quasiholes in the @xmath135 and @xmath136 states monotonically decay exponentially as increasing the tunneling distance . after being rescaled by eq .
( [ scalingconj ] ) , all the data locate on the same curve except for that of the smallest system size .
the scaling parameters @xmath137 we used in the plot are exact the expect values @xmath128 . here
we should notice that the finite size effect becomes more and more significant with increasing the @xmath4 index . the unrescaled tunneling amplitude @xmath133 ( a)(b ) and rescaled tunneling amplitudes @xmath134 ( c)-(f ) for charge @xmath138 non - abelian quasihole in the @xmath1 states as a function of the tunneling distance @xmath75 on cylinder.,width=340,height=226 ] again , for the smallest charged non - abelian quasihole in @xmath135 and @xmath136 states , as shown in fig.[k34nonabelian](a ) and
( b ) , there are still bump shapes in the tunneling amplitudes as varying @xmath75 . in the cft limit , although we do not have an analytical expression for the non - abelian quasihole tunneling amplitude as that for the moore - read state in eq .
( [ quarterformula ] ) , we believe that they still obey the similar scaling behavior .
therefore , we firstly replot the rescaled data with the scaling parameter modified by @xmath139 which is shown in fig.[k34nonabelian](c ) and ( d ) .
as expected , the data from different systems collapse together near the cft limit , or in the edge - edge interacting case .
this again verifies that the interaction of the two edges of the cylinder brings the anomalous correction to the scaling parameter . on the other hand , if this anomalous term is neglected , as shown in fig.[k34nonabelian](e ) and ( f ) , the data for the long distance tunneling scales better than that with the correction , although there are strong finite size effects .
in this work , we systematically study the scaling behavior of the quasihole tunneling amplitudes for the @xmath1 parafermion states on the cylinder geometry . comparing with the disk geometry we previously studied ,
the landau orbitals on cylinder do not have curvature difference and the tunneling distance can be smoothly tuned from zero to infinity via varying the aspect ratio . while the length of the finite cylinder @xmath75 decreasing from a thin cylinder limit to the cft limit , there is a critical length scale @xmath140 at which the two independent edges of the cylinder become interacting with each other .
therefore , the length scale of the tunneling distance is automatically separated into two regions .
our calculations reveals that the scaling behaviors for the overall region are in good agreement with the scaling conjecture in eq .
( [ scalingconj ] ) . for the abelian quasihole with charge @xmath47 in all @xmath1 states , by using the scaling parameter @xmath141 as expected from the analysis of the effective field theory , the data in two regions collapses onto each other very well . however , the things get more complicated in the non - abelian case . when @xmath127 , or with two independent edges , the scaling parameter @xmath8 is the same as that in the abelian case instead of substituting the quantum dimension @xmath142 by the one for non - abelian quasiholes .
and in the case of @xmath130 , or approaching to the cft limit , similar to the result on disk geometry , the scaling parameter need a modification which is shown in eq .
( [ modifiedalpha ] ) . based on the following reasons : ( 1 ) the tunneling amplitudes of the abelian quasiholes do not have this term in the cft limit ; ( 2 ) this anomalous term can be rewritten as @xmath24 where @xmath25 is the conformal dimension of the neutral component ; ( 3 ) our recent work elsewhere @xcite on the density difference between the bosonic and fermionic edge states shows that the width of the fermionic edge states in @xmath1 states is larger than that of bosonic edge states . or in other words , the neutral fermionic component in the edge state or quasihole state is more sensitive to the length scale of the edge - edge distance . we conclude that the anomalous correction term is contributed from the neutral component of the non - abelian quasiholes which can be treated as another feature of the non - abelian quasiholes .
this charactistics may be detected in the realistical systems , such as in the shot noise or the point contact interference experiments .
this work was supported by nsfc project no .
1127403 , 11547305 , fundamental research funds for the central universities no .
nj was also supported by chongqing graduate student research innovation project no .
xw was supported by the 973 program project no .
2012cb927404 , nsf - china grant no . 11174246 .
d. c. tsui , h. l stomer , and a. c. gossard , phys .
* 48 * , 1559 ( 1982 ) .
f. wilczek , phys .
lett . * 48 * , 1144 ( 1982 ) .
b. i. halperin , phys .
* 52 * , 1583 ( 1984 ) .
d. arovas , j. r. schrieffer , and f. wilczek , phys .
53 * , 722 ( 1984 ) .
m. h. freedman , proc .
natl , acad .
u. s. a. * 95 * , 98 ( 1998 ) .
j. preskill , in introduction to quantum computation , edited by h. -k .
lo , s. popescu , and t. p. spiller ( world scientific , singapore , 1998 ) .
a. kitaev , ann . phys .
( n. y. ) * 303 * , 2 ( 2003 ) . c. nayak , s. h. simon , a. stern , m. freedom , and s. das sarma , rev .
phys . * 80 * , 1803 ( 2008 ) .
r. l. willett , j. p. eisenstein , h. l. stormer , d. c. tsui , a. c. gossard , and j. h. english , phys .
lett . * 59 * , 1776 ( 1987 ) .
g. moore and n. read , nucl .
b * 360 * , 362 ( 1991 ) .
n. read and e. rezayi , phys .
b * 59 * , 8084 ( 1999 ) . v. a. fateev and a. b. zamalodchikov , sov . phys .
jetp * 62 * , 215 ( 1985 ) .
b. feigin , m. jimbo , t. miwa , and e. mukhin , int .
res . not .
* 2002 * , 1223 ( 2002 ) ; _ ibid _ , * 2003 * , 1015 ( 2003 ) .
b. a. bernevig and f. d. m. haldane , phys .
lett . * 100 * , 246802 ( 2008 ) .
b. a. bernevig and f. d. m. haldane , phys . rev
* 101 * , 246806 ( 2008 ) . b. a. bernevig and n. regnault , phys .
rev . lett . * 103 * , 206801 ( 2009 ) . c. de c. chamon , d. e. freed , s. a. kivelson , s. l. sondhi , and x. g. wen , phys .
b * 55 * , 2331 ( 1997 ) .
e. fradkin , c. nayak , a. tsvelik , and f. wilczek , nucl .
b * 516 * , 704 ( 1998 ) .
s. das sarma , m. freedman , and c. nayak , phys .
* 94 * , 166802 ( 2005 ) .
a. stern and b. i. halperin , phys .
lett . * 96 * , 016802 ( 2006 ) .
p. bonderson , a. kitaev , and k. shtengel , phys .
lett . * 96 * , 016803 ( 2006 ) .
b. rosenow , b. i. halperin , s. h. simon , and a. stern , phys . rev
100 * , 226803 ( 2008 ) .
w. bishara and c. nayak , phys .
b * 77 * , 165302 ( 2008 ) .
p. bonderson , k. shtengel , and j. k. slingerland , phys .
* 97 * , 016401 ( 2006 ) .
l. fidkowski , arxiv:0704.3291 ( unpublished ) .
p. bonderson , k. shtengel , and j. k. slingerland , ann .
( n.y ) . * 323 * , 2709 ( 2008 ) .
e. ardonne and e. -a .
kim , j. stat .
mech . : theory exp .
l04001 ( 2008 ) .
h. chen , z. -x .
hu , k. yang , e. h. rezayi , and x. wan , phys .
b * 80 * , 235305 ( 2009 ) . z. -x .
hu , k. h. lee , e. h. rezayi , x. wan , and k. yang , new j. phys . *
13 * , 035020 ( 2011 ) .
hu , k. h. lee , and x. wan , int . j. mod .
. ser . * 11 * , 70 ( 2012 ) .
q. li , n. jiang , z. zhu , and z .- x .
hu , new j. phys . *
17 * , 095006 ( 2015 ) .
r. tao and d. j. thouless , phys .
b * 28 * , 1142 ( 1983 ) .
d. j. thouless , surf .
sci . * 142 * , 147 ( 1984 ) .
a. seidel , h. fu , d - h .
lee , j. m. leinaas , and j. moore , phys .
lett . * 95 * , 266405 ( 2005 ) . e. j. bergholtz and a. karlhede , phys .
b * 77 * , 155308 ( 2008 )
. e. j. bergholtz , t. h. hansson , m. hermanns , a. karlhede , and s. viefers , phys .
b * 77 * , 165325 ( 2008 ) .
n. jiang and z. -x .
hu , unpublished . | quasiparticle tunneling between two counter - propagating edges through point contacts could provide information on the statistics of the quasiparticles .
previous study on a disk found a scaling behavior by varying the tunneling distance .
it was found that in the limit with zero tunneling distance , the abelian quasiparticles tunneling obey the scaling analysis while the non - abelian quasiparticles exhibit some non - trivial behaviors on the scaling exponents .
because of the limitation of disk geometry , we put the fractional quantum hall ( fqh ) state on the surface of a cylinder which has a larger tunable tunneling distance than that on disk by varying the aspect ratio @xmath0 .
we analyze the scaling behavior of the quasiholes , especially the non - abelian quasiholes in the read - rezayi @xmath1 parafermion states .
we aim to address the existance of the anomalous correction of the scaling parameter in the long tunneling distance . |
savannas ecosystems are characterized by the long - term coexistence between a continuous grass layer and scattered or clustered trees @xcite .
occurring in many regions of the world , in areas with very different climatic and ecological conditions , the spatial structure , persistence , and resilience of savannas have long intrigued ecologists @xcite . however , despite substantial research , the origin and nature of savannas have not yet been fully resolved and much remains to be learned .
savanna tree populations often exhibit pronounced , non - random spatial structures @xcite .
much research has therefore focused on explaining how spatial patterning in savannas arises @xcite . in most natural plant systems
both facilitative and competitive processes are simultaneously present @xcite and hard to disentangle @xcite .
some savanna studies have pointed toward the existence of short - distance facilitation @xcite , while others have demonstrated evidence of competition @xcite , with conflicting reports sometimes arriving from the same regions .
different classes of savannas , which can be characterized by how much rainfall they typically receive , should be affected by different sets of processes .
for example , in semiarid savannas water is extremely limited ( low mean annual precipitation ) and competition among trees is expected to be strong , but fire plays little role because there is typically not enough grass biomass to serve as fuel .
in contrast , humid savannas should be characterized by weaker competition among trees , but also by frequent and intense fires . in - between these extremes , in mesic savannas , trees likely have to contend with intermediate levels of both competition for water and fire @xcite .
competition among trees is mediated by roots that typically extend well beyond the crown @xcite .
additionally , fire can lead to local facilitation due to a protection effect , whereby vulnerable juvenile trees placed near adults are protected from fire by them @xcite .
we are particularly interested in how the interplay between these mechanisms governs the spatial arrangement of trees in mesic savannas , where both mechanisms may operate . on the other side ,
it has frequently been claimed that pattern formation in arid systems can be explained by a combination of long - distance competition and short - distance facilitation @xcite .
this combination of mechanisms is also known to produce spatial structures in many other natural systems @xcite .
although mesic savannas do not display the same range of highly regular spatial patterns that arise in arid systems ( e.g. , tigerbush ) , similar mechanisms might be at work .
specifically , the interaction between long - range competition and short - range facilitation might still play a role in pattern formation in savanna tree populations , but only for a limited range of parameter values and possibly modified by demographic stochasticity .
although the facilitation component has often been thought to be a key component in previous vegetation models @xcite , rietkerk and van de koppel @xcite , speculated , but did not show , that pattern formation could occur without short - range facilitation in the particular example of tidal freshwater marsh . in the case of savannas ,
as stated before , the presence of adult trees favor the establishment of new trees in the area , protecting the juveniles against fires . considering this effect , we take the facilitation component to its infinitesimally short spatial limit , and study its effect in the emergence of spatially periodic structures of trees . to our knowledge , this explanation , and the interrelation between long - range competition and local facilitation , has not been explored for a vegetation system
. one of our main results is that when considering the limit of local facilitation and nonlocal competition , clustering of trees appears .
here we develop a minimalistic model of savannas that considers two of the factors , as already mentioned , thought to be crucial to structure mesic savannas : tree - tree competition and fire , with a primary focus on spatially nonlocal competition .
employing standard tools used in the study of pattern formation phenomena in physics ( stability analysis and the structure function ) @xcite , we explore the conditions under which the model can produce non - homogeneous spatial distributions .
a key strength of our approach is that we are able to provide a complete and rigorous analysis of the patterns the model is capable of producing , and we identify which among these correspond to situations that are relevant for mesic savannas .
we further examine the role of demographic stochasticity in modifying both spatial patterns and the conditions under which trees persist in the system in the presence of fire , and discuss the implications of these results for the debate on whether the balance of processes affecting savanna trees is positive , negative , or is variable among systems .
this is the framework of our study : the role of long - range competition , facilitation and demographic fluctuations ( in the second part of the paper ) in the spatial structures of mesic savannas . to complete our work
we include an appendix where we study the effect of external fluctuations ( mimicking for e.g. rainfall ) on savanna dynamics .
our model is inspired by the one presented by calabrese et al . in @xcite .
it complements theirs by providing further analytical results that clearly demonstrate that this simple system , where we focus on the local limit of facilitation , can produce the full spectrum of spatial patterns reported from models employing both short - range facilitation and long - range inhibition ( competition ) .
in this section we derive the deterministic equation for the local density of trees , such that dynamics is of the logistic type and we only consider tree - tree competition and fire .
we study the formation of patterns via stability analysis and provide numerical simulations of our model , showing the emergence of spatial structures .
calabrese et al .
@xcite introduced a simple discrete - particle lattice savanna model that considers the birth - death dynamics of trees , and where tree - tree competition and fire are the principal ingredients .
these mechanisms act on the probability of establishment of a tree once a seed lands at a particular point on the lattice . in the discrete model ,
seeds land in the neighborhood of a parent tree with a rate @xmath0 , and establish as adult trees if they are able to survive both competition neighboring trees and fire .
as these two phenomena are independent , the probability of establishment is @xmath1 , where @xmath2 is the probability of surviving the competition , and @xmath3 is the probability of surviving a fire event . from this dynamics
, we write a deterministic differential equation describing the time evolution of the global density of trees ( mean field ) , @xmath4 , where the population has logistic growth at rate @xmath0 , and an exponential death term at rate @xmath5 .
it reads : @xmath6 generalizing eq .
( [ eq : mf ] ) , we propose an evolution equation for the space - dependent ( local ) density of trees , @xmath7 : @xmath8 we allow the probability of overcoming competition to depend on tree crowding in a local neighborhood , decaying exponentially with the density of surrounding trees as @xmath9 where @xmath10 is a parameter that modulates the strength of the competition , and @xmath11 is a positive kernel function that introduces a finite range of influence .
this model is related to earlier models of pattern formation in arid systems @xcite , and subsequent works @xcite , but it differs from standard kernel - based models in that the kernel function accounts for the interaction neighborhood , and not for the type of interaction with the distance . note also that the nonlocal term enters nonlinearly in the equation .
following @xcite , @xmath12 is assumed to be a saturating function of grass biomass , @xmath13 , similar to the implementation of fire of jeltsch _ et al . _ in @xcite @xmath14 where @xmath15 governs the resistance to fire , so @xmath16 means no resistance to fires .
notice how our model is close to the one in @xcite through the definitions of @xmath2 and @xmath3 , although we consider the probability of surviving a fire depending on the local density of trees , and in @xcite it depends on the global density .
the deterministic differential equation that considers tree - tree competition and fire for the spatial tree density is @xmath17 where @xmath18 thus , we have a logistic - type equation with an effective growth rate that depends nonlocally on the density itself , and which is a combination of long - range competition and local facilitation mechanisms ( fire ) .
the probability of surviving a fire is higher when the local density of trees is higher , as can be seen from the definition in equation ( [ probfire ] ) . in figure [ transition ]
we show numerical solutions for the mean field equation ( [ eq : mf ] ) ( lines ) and the spatially explicit model ( equation [ sav1 ] ) ( dots ) in the stationary state @xmath19 using different values of the competition . we have used a top - hat function as the competition kernel , @xmath11 ( see section [ lsa ] for more details on the kernel choice ) .
we observe a very good agreement of both descriptions which becomes worse when we get closer to the critical point @xmath20 , where the model presents a phase transition from a tree - grass coexistence to a grassland state .
this disagreement appears because while the mean field equation describes an infinite system , the eq .
( [ sav1 ] ) description forces us to choose a size for the system .
the model reproduces the long - term coexistence between grass and trees that is characteristic of savannas . to explore this coexistence , we study the long - time behavior of the system and analyze the homogeneous stationary solutions of eq .
( [ sav1 ] ) , which has two fixed points .
the first one is the absorbing state representing the absence of trees , @xmath21 , and the other can be obtained , in the general case , by numerically solving @xmath22 in the regime where @xmath23 is small ( near the critical point ) , if competition intensity , @xmath10 , is also small , it is possible to obtain an analytical expression for the critical value of the probability of surviving a fire , @xmath20 , @xmath24 outside of the limit where @xmath25 , we can solve eq .
( [ steq ] ) numerically in @xmath23 to show that the critical value of the fire resistance parameter , @xmath20 , does not depend on competition .
a steady state with trees is stable for higher fire survival probability ( fig .
[ transition ] ) .
the model , then , shows a transition from a state where grass is the only form of vegetation to another state where trees and grass coexist at @xmath20 . in what follows ,
we fix @xmath26 , so we choose our temporal scale in such a way that time is measured in units of @xmath5
. this choice does not qualitatively affect our results .
, as a function of the resistance to fires parameter , @xmath15 .
the lines come from the mean field solution , eq .
( [ steq ] ) , and the dots from the numerical integration of eq .
( [ sav1 ] ) over a square region of @xmath27 .
we have chosen @xmath26 , and @xmath28 . in the case of the spatial model ,
@xmath23 involves an average of the density of trees over the studied patch of savanna.,scaledwidth=40.0% ] the spatial patterns appearing in the nonlocal savanna model can be studied by performing a linear stability analysis @xcite of the stationary homogeneous solutions of equation ( [ sav1 ] ) , @xmath29 .
the stability analysis is performed by considering small harmonic perturbations around @xmath30 , @xmath31 , @xmath32 .
after some calculations ( [ appa ] ) , one arrives at the dispersion relation @xmath33}{(\sigma-\rho_{0}+1)}-1,&\end{aligned}\ ] ] where @xmath34 , @xmath35 , is the fourier transform of the kernel , @xmath36 the critical values of the parameters of the transition to pattern , @xmath37 and @xmath38 , and the fastest growing wavenumber @xmath39 are obtained from the simultaneous solution of @xmath40 note that @xmath39 represents the most unstable mode of the system , which means that it grows faster than the others and eventually dominates the state of the system .
therefore , it determines the length scale of the spatial pattern .
these two equations yield the values of the parameters @xmath10 and @xmath15 at which the maximum of the curve @xmath41 , right at @xmath39 , starts becoming positive .
this signals the formation of patterns in the solutions of eq .
( [ sav1 ] ) . as eq .
( [ eq : sec ] ) is explicitly written as @xmath42 the most unstable wavenumber @xmath39 can be obtained by evaluating the zeros of the derivative of the fourier transform of the kernel .
equation ( [ reldisper ] ) shows that competition , through the kernel function , fully determines the formation of patterns in the system .
the local facilitation appears in @xmath43 and it is not relevant in the formation of spatial structures .
if the fourier transform of @xmath44 never takes positive values , then @xmath45 is always negative and only the homogeneous solution is stable . however , when @xmath46 can take negative solutions then patterns may appear in the system .
what does this mean in biological terms ?
imagine that we have a family of kernels described by a parameter @xmath47 : @xmath48 ( @xmath49 gives the range of competition ) .
the kernels are more peaked around @xmath50 for @xmath51 and more box - like when @xmath52 .
it turns out that this family of functions has non - negative fourier transform for @xmath53 , so that no patterns appear in this case .
a lengthy discussion of this property in the context of competition of species can be found in @xcite .
thus , the shape of the competition kernel dictates whether or not patterns will appear in the system .
if pattern formation is possible , then the values of the fire and competition parameters govern the type of solution ( see below ) .
our central result for nonlocal competition is that , contrary to conventional wisdom , it can , in the limit of infinitesimally short ( purely local ) facilitation , promote the clustering of trees . whether or not this occurs depends entirely on the shape of the competition kernel . for large @xmath47
we have a box - like shape , and in these cases trees compete strongly with other trees , roughly within a distance @xmath49 from their position .
the mechanism behind this counterintuitive result is that trees farther than @xmath49 away from a resident tree area are not able to _ invade _ the zone defined by the radius r around the established tree ( their seeds do not establish there ) , so that an exclusion zone develops around it . for smaller @xmath47
there is less competition and the exclusion zones disappear . for a more detailed analysis
, one must choose an explicit form for the kernel function .
our choice is determined by the original @xmath54 taken in @xcite , so that it decays exponentially with the number of trees in a neighborhood of radius @xmath49 around a given tree .
thus , for @xmath44 we take the step function ( limit @xmath55 ) @xmath56 as noticed before , the idea behind the nonlocal competition is to capture the effect of the long roots of a tree .
the kernel function defines the area of influence of the roots , and it can be modeled at first order with the constant function of equation ( [ kerneldd ] ) .
thus the parameter r , which fixes the _ nonlocal _ interaction scale , must be of the order of the length of the roots @xcite .
since the roots are the responsible for the adsorption of resources ( water and soil nutrients ) , a strong long - range competition term implies strong resource depletion . for this kernel
the fourier transform is @xcite @xmath57 and its derivative is @xmath58 , where @xmath59 , and @xmath60 is the @xmath61-order bessel function . since @xmath34 can take positive and negative values , pattern solutions may arise in the system , that will in turn depend on the values of @xmath10 and @xmath15 .
the most unstable mode is numerically obtained as the first zero of @xmath62 , eq .
( [ derlam ] ) , which means the first zero of the bessel function @xmath63 .
this value only depends on @xmath49 , being independent of the resistance to fires and competition , and it is @xmath64 . because a pattern of @xmath65 cells is characterized by a wavenumber @xmath66 , where @xmath67 is the system size , the typical distance between clusters , @xmath68 , using the definition of the critical wavenumber is given by @xmath69 .
in other words , it is approximately the range of interaction @xmath49 .
this result is also independent of the other parameters of the system .
since we are interested in the effect of competition and fire on the distribution of savanna trees , we will try to fix all the parameters but @xmath15 and @xmath10 .
we will explore the effect of different values of these parameters on the results .
first , we have chosen ( as in @xcite ) the death rate @xmath26 , and solving eq .
( [ steq ] ) we will roughly estimate the birth rate , @xmath0 .
we will work in the limit of intermediate to high mean annual precipitation , so water is non - limiting and thus we can neglect the effects of competition ( @xmath70 ) . at this intermediate to high mean annual precipitation
the empirically observed upper limit of savanna tree cover is approximately @xmath71 @xcite . to reach this upper limit in the tree cover ,
disturbances must also be absent , implying no fire ( @xmath72 ) . in this limit ,
the mean field equation ( [ eq : mf ] ) is quantitatively accurate , as it is shown in figure [ transition ] , and the stationary mean field solution of the model depends only on the birth rate @xmath73 it can be solved for @xmath0 for a fixed @xmath71 , and it yields @xmath28 @xcite . in the following we just consider the dependence of our results on @xmath10 and @xmath15 .
in particular , @xmath74 .
the phase diagram of the model , computed numerically , is shown in fig
. [ phspace ] , where we plot the spatial character of the steady solution ( homogeneous or inhomogeneous ) as a function of @xmath10 and @xmath15 .
note that increasing competition enhances the inhomogeneous or pattern solution .
this is because , as we are now in the case of a kernel giving rise to clusters , increasing @xmath10 makes it more difficult to enter the exclusion zones in - between the clusters . for very strong competition ( high , unrealistic , @xmath10 ) , fire has no influence on the pattern . the critical line separating these two solutions ( pattern and homogeneous)can be analytically computed as a function of the parameters @xmath10 , @xmath15 , @xmath30 and @xmath75 ( see eq .
( [ criline ] ) in [ appb ] ) . in figure [ phspace ]
we have plotted ( with crosses ) this critical line separating homogeneous and pattern solutions for the step kernel .
note that the stationary density of trees , @xmath30 , must be computed numerically from eq .
( [ steq ] ) . ) for @xmath76 , @xmath77 , and a step kernel .
the absorbing - active transition is shown at @xmath20 with circles ( o ) . the homogeneous - pattern transition ( eq .
( [ criline ] ) ) is indicated with crosses ( x ) .
the diamond , the square , and the up - triangle show the value of the parameters @xmath15 and @xmath10 taken in figures [ patterns](a)-(c ) respectively .
the stars point out the transition to inhomogeneous solutions in the stochastic model as described in section [ stochastic ] , with @xmath78.,scaledwidth=40.0% ] with @xmath28 , in the absence of fire ( @xmath72 ) , and for weak competition , we can take the limits @xmath79 and @xmath72 of the dispersion relation eq .
( [ reldisper ] ) , leading to @xmath80 in fig .
[ transition ] , for large @xmath15 , it can be seen that typically @xmath81 , so eq .
( [ limitceroinfinity ] ) becomes negative .
this result means that in this limit , trees are uniformly distributed in the system as there is no competition , and space does not play a relevant role in the establishment of new trees .
such situation could be interpreted as favorable to forest leading to a fairly homogeneous density of trees .
this result agrees with the phase plane plotted in figure [ phspace ] . in biological terms , there are no exclusion zones in the system because there is no competition .
the previous analysis provides information , depending on the competition and fire parameters , about when the solution is spatially homogenous and when trees arrange in clusters .
however , the different shapes of the patterns have to be studied via numerical simulations @xcite of the whole equation of the model .
we have taken a finite square region of savanna with an area of @xmath82 ha . ,
allowed competition to occur in a circular area of radius @xmath83 , and employed periodic boundary conditions and a finite differences algorithm to obtain the numerical solution .
similarly to what has been observed in studies of semiarid water limited systems @xcite , different structures , including gaps , stripes , and tree spots , are obtained in the stationary state as we increase the strength of competition for a fixed value of the fire parameter or , on the other hand , as we decrease the resistance to fires for a given competition intensity . in both equivalent cases
, we observe this spectrum of patterns as far as we go to a more dry state of the system , where resources ( mainly water ) are more limited ( see figs . [ patterns](a)-[patterns](c ) ) and competition is consequently stronger .
this same sequence of appearance of patterns has been already observed in the presence of different short - range facilitation mechanisms @xcite .
it indicates that , when @xmath10 is increased ( i.e. the probability of surviving competition is decreased ) , new trees can not establish in the exclusion areas so clustering is enhanced . on the other hand , in the case of fire - prones savannas ,
previous works had only shown either tree spot @xcite or grass spots @xcite .
therefore , at some values of the parameter space ( see fig .
[ patterns]b ) , the patterns in our deterministic approach are not observed in mesic savannas , and should correspond to semiarid systems .
however , we will show in the following sections that under the parameter constraints of a mesic savanna , and considering the stochastic nature of the tree growth dynamics in the system ( i.e. demographic noise ) , our model shows realistic spatial structures . ) , ( b ) striped grass vs. tree ( @xmath84 ) , and ( c ) tree spots ( @xmath85 ) patterns in the deterministic model in a square patch of savanna of @xmath27 .
@xmath86 , @xmath87 , @xmath76 and @xmath77 in all the plots.,scaledwidth=48.0% ] a much more quantitative analysis of the periodicity in the patterns can be performed via the structure function .
this will be helpful to check the previous results and , especially , for the analysis of the data of the stochastic model of the next section , for which we will not present analytical results .
the structure function is defined as the modulus of the spatial fourier transform of the density of trees in the stationary state , @xmath88 where the average is a spherical average over the wavevectors with modulus @xmath89 .
the structure function is helpful to study spatial periodicities in the system , similar to the power spectrum of a temporal signal .
its maximum identifies dominant periodicities , which in our case are the distances between tree clusters .
note that the geometry of the different patterns can not be uncovered with the structure function , since it involves a spherical average . in fig .
[ maxstructure ] , we show the transition to patterns using the maximum of the structure function as a function of the competition parameter .
a peak appears when there are spatial structures in the system , so @xmath90\neq0 $ ] .
however , we do not have information about the values where the shapes of the patterns change . taking @xmath91 ,
the peak is always at @xmath92 for our deterministic savanna model , independently of the competition and fire resistance parameters , provided that they take values that ensure the emergence of patterns in the system ( see the line labeled by @xmath93 in fig .
[ strucboth ] ; for the definition of @xmath94 see next section ) .
this result is in good agreement with the theoretical result provided for the wavelength by the linear stability analysis @xmath95 , which is also independent of competition and resistance to fires . at long times .
the fire parameter is fixed at @xmath86 .
black circles refers to the deterministic model and red squares to the stochastic model , @xmath96.,scaledwidth=33.0% ] ) for different values of the demographic noise intensity .
@xmath97 , @xmath86 , @xmath91 , @xmath77 , @xmath76.,scaledwidth=33.0% ]
the perfectly periodic patterns emerging in fig . [ patterns ] from the deterministic model seem to be far from the disordered ones usually observed in aerial photographs of mesic savannas and shown by individual based models @xcite .
we have so far described a savanna system in terms of the density of trees with a deterministic dynamics .
the interpretation of the field @xmath98 is the density of tree ( active ) sites in a small volume , @xmath99 .
if we think of trees as reacting particles which are born and die probabilistically , then to provide a reasonable description of the underlying individual - based birth and death dynamics , we have to add a noise term to the standard deterministic equation .
it will take into account the _ intrinsic _ stochasticity present at the individual level in the system .
if we take a small volume , @xmath99 , the number of reactions taking place is proportional to the number of particles therein , @xmath100 , with small deviations .
if @xmath100 is large enough , the central limit theorem applies to the sum of @xmath100 independent random variables and predicts that the amplitude of the deviation is of the order of @xmath101 @xcite .
this stochasticity referred to as demographic noise .
the macroscopic equation is now stochastic , @xmath102- \nonumber \\ & -&\alpha\rho(\mathbf{x } , t)+\gamma\sqrt{\rho(\mathbf{x } , t)}\eta(\mathbf{x},t),\end{aligned}\ ] ] where @xmath103 ( but we take it as a constant , @xcite ) modulates the intensity of @xmath104 , a gaussian white noise term with zero mean and correlations given by dirac delta distributions @xmath105 the complete description of the dynamics in eq.([savsto ] ) should have the potential to describe more realistic patterns .
we first investigate the effect of demographic noise on the persistence of trees in the system .
we show in ( fig .
[ noisy - transition ] ) that the critical point , @xmath20 , depends on the value of the competition parameter @xmath10 .
this effect is rather small , so that when @xmath10 increases the transition to the grassland state appears only for a slightly larger @xmath15 ( i.e , less frequent fire ) .
the reason seems to be that fire frequency and intensity depend on grass biomass .
seasonally wet savannas support much more grass biomass that serves as fuel for fires during the dry season @xcite .
dry savannas have much lower grass biomass , so they do not burn as often or as intensely .
the shift of the critical value of @xmath15 when competition is stronger is consistent with the one showed in @xcite , as can be seen comparing figure 2 in @xcite with figure [ noisy - transition ] here . besides , the values obtained for @xmath20 are larger when we consider the demographic stochasticity @xcite neglected in the deterministic field approach . ) with @xmath78 and average the density of trees in the steady state.,scaledwidth=33.0% ] we explore numerically the stochastic savanna model using an algorithm developed in @xcite ( see [ sec : numerical ] ) .
note that the noise makes the transition to pattern smoother so the change from homogeneous to inhomogeneous spatial patterns is not as clear as it is in the limit where the demographic noise vanishes ( see fig . [ maxstructure ] ) .
the presence of demographic noise in the model , as shown in fig .
[ phspace ] ( red stars ) , also decreases the value of the competition strength at which patterns appear in the system , as has been observed in other systems .
mathematically , these new patterns appear since demographic noise maintains fourier modes of the solution which , due to the value of the parameters , would decay in a deterministic approach @xcite . biologically , exclusion zones are promoted by demographic noise , since it does not affect regions where there are not trees . on the other hand in vegetated areas fluctuations
may enhance tree density , leading to stronger competition .
the presence of demographic noise in the model allows the existence of patterns under more humid conditions .
this result is highly relevant for mesic savannas , as we expect competition to be of low to intermediate strength in such systems .
we show two examples of these irregular patterns in fig .
[ stopatterns](a ) and fig .
[ stopatterns](b ) .
unrealistic stripe - like patterns no longer appear in the stochastic model .
we have studied the dynamics of the system for some values of the fire and competition parameters .
demographic noise influences the spatial structures shown by the model .
the deterministic approach shows a full spectrum of patterns which are not visually realistic for mesic savannas ( but for arid systems ) .
the role of the noise is to transform this spectrum of regular , unrealistic patterns into more irregular ones ( figures [ stopatterns](a)-[stopatterns](d ) ) that remind the observed in aerial photographs of real mesic savannas . on the other hand ,
these patterns are statistically equivalent to the deterministic ones , as it is shown with the structure function in fig .
[ strucboth ] . the dominant scale in the solution
is given by the interaction radio , @xmath49 , and it is independent of the amplitude of the noise ( see the structure function in figure [ strucboth ] , peaked around @xmath106 independently of the noise ) . besides , over a certain treshold in the amplitude , demographic noise destroys the population of trees .
therefore , the model presents an active - absorbing transition with the noise strength , @xmath94 , being the control parameter . .
@xmath86 , @xmath87 , @xmath76 and @xmath77 in all the plots .
( a ) @xmath78 , @xmath107 .
( b ) @xmath78 , @xmath108 . ( c ) @xmath109 , @xmath110 .
( b ) @xmath78 , @xmath110.,scaledwidth=48.0% ]
understanding the mechanisms that produce spatial patterns in savanna tree populations has long been an area of interest among savanna ecologists @xcite .
a key step in such an analysis is defining the most parsimonious combination of mechanisms that will produce the pattern in question . in this paper
the combination of long - range competition for resources and the facilitation induced by fire are considered the responsible of the spatial structures , in the line of studies of vegetation pattern formation in arid systems , where also a combination of long - range inhibition and short - range facilitation is introduced @xcite .
the main difference is that the facilitation provided by the protection effect of adult trees against fires in our savanna model takes the short - range facilitation to its infinitesimally short limit ( i.e , local limit ) . under this assumption
we have studied the conditions under which our model could account for patterns .
we have shown that nonlocal competition combined with local facilitation induces the full range of observed spatial patterns , provided the competition term enters nonlinearly in the equation for the density of trees , and that competition is strong enough .
the key technical requirement for this effect to occur is that the competition kernel must be an almost constant function in a given competition region , and decay abruptly out of the region .
we verify this condition working with supergaussian kernel functions . in practice
, this means that competition kernels whose fourier transform takes negative values for some wavenumber values , will lead to competition driven clustering .
the other mechanism we have considered for a minimalistic but realistic savanna model , fire , has been shown to be relevant for the coexistence of trees and grass and for the shape of the patterns . however , competition is the main ingredient allowing pattern solutions to exist in the model .
if the shape of the kernel allows these types of solutions , then the specific values of fire and competition parameters determine the kind of spatial structure that develops .
it is also worth mentioning that one can observe the full spectrum of patterns in the limit where fires vanish ( @xmath72 ) , so there is no facilitation at all , provided competition is strong enough .
however , when there is no competition , @xmath70 , no patterns develop regardless of the value of the fire term .
therefore , we conclude that the nonlocal competition term is responsible for the emergence of clustered distributions of trees in the model , with the fire term playing a relevant role only to fix the value of the competition parameter at which patterns appear . in other words , for a given competition strength , patterns appear more readily when fire is combined with competition . a similar mechanism of competitive interactions between species has been shown to give rise to clusters of species in the context of classical ecological niche theory .
scheffer and van nes @xcite showed that species distribution in niche space was clustered , and pigolotti et al .
@xcite showed that this arises as an instability of the nonlocal nonlinear equation describing the competition of species .
long - distance competition for resources in combination with the local facilitation due to the protection effect of adult trees in the establishment of juvenile ones can explain the emergence of realistic structures of trees in mesic savannas .
in these environmental conditions , competition is limited , so we should restrict to small to intermediate values of the parameter @xmath10 , and the effect of fires is also worth to be taken into account . however , these two ingredients give a full range of patterns observed in vegetated systems , but not in the particular case of savannas .
it is necessary to consider the role of demographic noise , which is present in the system through the stochastic nature of the birth and death processes of individual trees . in this complete framework
our model shows irregular patterns of trees similar to the observed in real savannas .
the other important feature of savannas , the characteristic long - time coexistence of trees and grass is well captured with our model ( figures [ transition ] and [ noisy - transition ] ) .
besides , the presence of demographic noise , as it is shown in figure [ noisy - transition ] , makes our approach much more realistic , since the persistence of trees in the face of fires is related to the water in the system . on the other hand ,
demographic stochasticity causes tree extinction at lower fire frequencies ( larger @xmath15 ) than in the deterministic case .
this is because random fluctuations in tree density are of sufficient magnitude that this can hit zero even if the deterministic stationary tree density ( for a given fire frequency ) is greater than zero .
this effect vanishes if we increase the system size .
the demographic noise is proportional to the density of trees ( proportional to @xmath111 ) , so fluctuations are smaller if we study bigger patches of savannas . as usually happens in the study of critical phenomena in statistical mechanics , the extinction times due to demographic noise increase exponentially with the size of the system for those intensities of competition and fire that allow the presence of trees in the stationary state . over the critical line ,
this time will follow a power law scaling , and a logarithmic one when the stationary state of the deterministic model is already absorbing ( without trees ) @xcite .
we have shown the formation of patterns in a minimal savanna model , that considers the combination of long - range competition and local facilitation mechanisms as well as the transition from trees - grass coexistence to a grass only state .
the salient feature of the model is that it only considers nonlocal ( and nonlinear ) competition through a kernel function which defines the length of the interaction , while the facilitation is considered to have an infinitesimally short influence range .
our model thus differs from standard kernel - based savanna models that feature both short - range facilitation and long - range competition .
the same sequence of spatial patterns appears in both approaches , confirming rietkerk and van de koppel s @xcite suggestion that short - range facilitation does not induce spatial pattern formation by itself , and long - distance competition is also needed .
it also suggests that long - range competition could be not only a necessary , but also a sufficient condition to the appearance of spatial structures of trees . inspired by @xcite
, we have proposed a nonlocal deterministic macroscopic equation for the evolution of the local density of trees where fire and tree - tree competition are the dominant mechanisms .
if the kernel function falls off with distance very quickly ( the fourier transform is always positive ) the system only has homogenous solutions . in the opposite case
, patterns may appear depending on the value of the parameters ( @xmath10 and @xmath15 ) , and in a sequence similar to the spatial structures appearing in standard kernel - based models . under less favorable environmental conditions ,
trees tend to arrange in more robust structures to survive ( fig .
[ patterns](d ) ) .
biologically , trees are lumped in dense groups , separated by empty regions .
entrance of new trees in these _ exclusion zones _ is impossible due to the intense competition they experience there .
a great strength of our approach is that our deterministic analysis is formal , and we have shown the different spatial distributions of the trees that occur as competition becomes more intense , concluding that self organization of trees is a good mechanism to promote tree survival under adverse conditions @xcite .
trees tend to cluster in the high competition ( low resources ) limit ( fig .
[ patterns](d ) ) , due to the formation of exclusion zones caused by non - local competition , and not as a result of facilitation . however , because we are dealing with a deterministic model , the patterns are too regular and the transition between the grass - only and a tree - populated states is independent of tree competition .
we therefore considered stochasticity coming from the stochastic nature of individual birth and death events , to provide a more realistic description of savanna dynamics .
calabrese et al .
@xcite also noted that savanna - to - grassland transition was independent of competition intensity in the mean field approach , but not when demographic noise was included . in the present model , both the grassland to savanna transition and
the spatial structures that develop are influenced by demographic stochasticity . in the case of spatial structures ,
demographic noise is specially relevant , since it turns much of the unrealistic patterns of the deterministic model into more realistic ones , that remind the observed in real savannas .
it also allows the existence of periodic arrangements of trees in more humid systems , which means environmental conditions closer to mesic savannas .
we have quantified the characteristic spacing of spatial patterns through the structure function .
the irregular patterns produced by the stochastic model still have a dominant wavelength whose value is the same as in the deterministic model and depends only on the value of the range of the interaction , @xmath49 , in the kernel function .
the match between the typical spatial scale of the patterns and the characteristic distance over which nonlocal competition acts indicates that competition is responsible for the presence of clustered spatial structures .
r.m - g is supported by the jaepredoc program of csic .
r.m - g . and c.l .
acknowledge support from micinn ( spain ) and feder ( eu ) through grant no .
fis2007- 60327 fisicos .
we acknowledge federico vzquez and emilio hernndez - garca for their comments and discussion .
we also acknowledge the detailed reading and insightful comments of three anonymous referees which greatly helped to improve this manuscript .
this appendix shows the details of the linear stability analysis , in particular how it is obtained the dispersion relation in eq .
( [ reldisper ] ) .
we consider the stationary solution @xmath23 plus a small harmonic perturbation , @xmath112 where @xmath32 .
substituting eq .
( [ ansatz ] ) into the original equation ( [ sav1 ] ) , and retaining only linear terms in @xmath113,we arrive to the relation dispersion @xmath114 & \nonumber \\ & + bc\sigma\frac{1 - 2\rho_{0}}{\sigma+1-\rho_{0}}-1 , & \nonumber \\\end{aligned}\ ] ] where @xmath115 is the fourier transform of the kernel , @xmath116 , and @xmath117 , provided that we deal with normalized kernels . equation ( [ rd ] ) can be written as eq .
( [ reldisper ] ) using the definition of @xmath43 .
we show here the analytical expression for the critical line in the transition from homogeneous to inhomogeneous solutions . starting from eq .
( [ eq : first ] ) it is possible to write an expression for the value of the resistance to fires parameter , @xmath15 , at which the macroscopic equation ( [ sav1 ] ) starts showing pattern solutions , as a function of the competition parameter , @xmath10 , and the most unstable mode @xmath118 . considering the value of the parameters taken in our study , @xmath28 and @xmath26
, it is @xmath119}{10\left[1 - 2\rho_{0}+ \delta \hat{g}(k_{c})\rho_{0}(1+\rho_{0})-{\rm e}^{\delta\pi r^{2}\rho_{0}}/5\right ] } & \nonumber \\ & + \frac{(\rho_{0}-1)\sqrt{5\left[5(\rho_{0}-1)^{2 } ( \delta \hat{g}(k_{c})\rho_{0}-1)^{2}-4{\rm e}^{\delta\pi r^{2}\rho_{0}}\rho_{0}\right]}}{10\left[1 - 2\rho_{0}+ \delta \hat{g}(k_{c})\rho_{0}(1+\rho_{0})-{\rm e}^{\delta\pi r^{2}\rho_{0}}/5\right]}.&\end{aligned}\ ] ] this complicated expression must be evaluated numerically together with the solution of eq .
( [ steq ] ) for the stationary density of trees , which is also a function of the competition and fire parameters .
we show the results in figure [ phspace ] , where the curve , represented with the black crosses , fits perfectly with the numerical results from the linear stability analysis .
the integration of stochastic equations where the noise amplitude depends on the square root of the variable , @xmath120 , and there are absorbing states ( i.e , states where the system stays indefinitely ) , has awaken a great interest , specially in the study of critical phenomena ( i.e , properties of the system that appear when it is close to the critical point , often the absorbing state ) .
the amplitude of the fluctuations tends to zero there , and thus numerical instabilities may appear .
recently @xcite a very efficient method has been developed , but we have used in this work an older one , presented in @xcite , since its implementation is easier and it gives precise results working far from the transition point . it consists on discretizing the langevin equation , taking a step size @xmath121 in the variable . to apply the method to equation ( [ savsto ] ) , first of all we discretize the space .
particularly , we compute the integral in the exponential term approximating it by a sum of the field evaluated in the nodes of the discrete space @xmath122 then , we integrate the temporal dependence . the key of the algorithm is to prevent @xmath123 to take negative values . from a general equation @xmath124 where @xmath125 is a gaussian white noise with zero mean and delta correlated ,
it is @xmath126 where @xmath127 .
@xmath128 is a gaussian number with zero mean and unit variance . at this point ,
to prevent @xmath123 to take negative values , the author in @xcite proposes to dicretize the density setting @xmath129 and to truncate the gaussian distribution from where @xmath128 is obtained simetrically so that @xmath130 .
the negatives values are avoided requiring @xmath131 .
it can be done in many ways but following @xcite we use @xmath132 finally , rescaling the equation , we can achieve a discretized version in which positive and zero - mean noise are ensured at the cost of a `` quantized '' density .
one of the key ingredients for the long coexistence between grass and trees is the largely inhomogeneous temporal distribution of precipitations over time @xcite .
we have studied this environmental variability following the idea in @xcite , considering the switching between unstressed vegetation growth , given by the first term in ( [ sav1 ] ) , and drought - induced vegetation decay , represented with the second term in eq .
( [ sav1 ] ) . these processes
take place each time step with probability @xmath133 and @xmath134 , respectively . from now on ,
we call @xmath135&=&b_{eff}(\rho)\left[\rho(\mathbf{x},t)-\rho^{2}(\mathbf{x},t)\right ] , \nonumber \\ f_{d}[\rho(\mathbf{x},t)]&=&-\alpha\rho(\mathbf{x},t),\end{aligned}\ ] ] and @xmath136=\frac{1}{2}\left[f_{b}[\rho(\mathbf{x},t)]\pm f_{d}[\rho(\mathbf{x},t)]\right].\ ] ] the random dynamics of the system is written in terms of a stochastic partial differential equation , @xmath137+f_{-}[\rho(\mathbf{x},t)]\xi_{dn}(t),\ ] ] where @xmath138 is a dichotomous noise ( dmn ) , assuming values @xmath139 ( wet season ) and @xmath140 ( dry season ) with probability @xmath133 and @xmath134 , respectively .
if the rate of random switching , taken as the inverse of the integration time step , is relatively fast respect to the rate of convergence to equilibrium in each of the two states , we can replace the noise term in eq .
( [ dichsde ] ) with its average value , @xmath141 .
it is meaningful since the rainfall seasons are much shorter than the time needed to reach one of the equlibrium stationary states of death and birth processes , @xmath142 , respectively .
this substitution leads to a deterministic equation @xmath143+f_{-}[\rho(\mathbf{x},t)](1 - 2p),\ ] ] where we will be able to perform linear stability analysis as usual .
the new dispersion relation is easily obtained , @xmath144}{(\sigma-\rho_{0}+1)},&\end{aligned}\ ] ] which means that the main effect of the dichotomous noise is to renormalize the rates @xmath5 and @xmath0 .
the patterns observed now are the same as the ones in the deterministic case , though the regions where they emerge change in accordance with this renormalization .
thus , the effect of stochastic precipitation , as modeled with this random switching mechanism , is a change of the parameter values for the different transitions observed in the deterministic continuum model eq .
( [ sav1 ] ) . according to the value of @xmath133 , an absorbing - active phase transition
is observed , @xmath145 .
small values of p , meaning long dry season , lead to an absorbing state while increasing the probability of raining implies the appearence of trees in the system . in this latter case
, the solution can be either homogeneous or showing spatial patterns , depending on fire and competition .
this attempt to model rainfall has not been very succesful and does not give a lot of new information .
much effort of future research should be put on this point , trying to get much more realistic modelling of external environmental variability , according to empirical observations , with long runs of dry years and rare wet years .
calabrese , f. vzquez , c. lpez , m. san miguel , and v. grimm .
the individual and interactive effects of tree - tree establishment competition and fire on savanna structure and dynamics . the american naturalist , 175 , 3 , ( 2010 ) .
r. lefever , n. barbier , p. couteron , and o. lejeune .
deeply gapped vegetation patterns : on crown / root allometry , criticality and desertification . journal of theoretical biology , 2 , 261 , ( 2009 ) , 194 - 209 .
i. dornic , h. chat , and m.a .
muoz , integration of langevin equations with multiplicative noise and the viability of field theories for absorbing phase transitions .
physical review letters , 94 , ( 2005 ) , 100601 . | we propose a model equation for the dynamics of tree density in mesic savannas .
it considers long - range competition among trees and the effect of fire acting as a local facilitation mechanism . despite short - range facilitation
is taken to the local - range limit , the standard full spectrum of spatial structures obtained in general vegetation models is recovered .
long - range competition is thus the key ingredient for the development of patterns .
the long time coexistence between trees and grass , and how fires affect the survival of trees as well as the maintenance of the patterns is studied .
the influence of demographic noise is analyzed .
the stochastic system , under the parameter constraints typical of mesic savannas , shows irregular patterns characteristics of realistic situations .
the coexistence of trees and grass still remains at reasonable noise intensities . |
when a nonlinear system shows unexpectedly simple behavior , it may be a clue that some hidden structure awaits discovery .
for example , recall the classic detective story @xcite that began in the 1950s with the work of fermi , pasta , and ulam @xcite . in their numerical simulations of a chain of anharmonic oscillators ,
fermi et al .
were surprised to find the chain returning almost perfectly , again and again , to its initial state .
the struggle to understand these recurrences led zabusky and kruskal @xcite to the discovery of solitons in the korteweg
devries equation , which in turn sparked a series of results showing that this equation possessed many conserved quantities in fact , infinitely many @xcite .
then several other equations turned out to have the same properties . at the time these results seemed almost miraculous .
but by the mid-1970s the hidden structure responsible for all of them the complete integrability of certain infinite - dimensional hamiltonian systems @xcite had been made manifest by the inverse scattering transform @xcite and lax pairs @xcite .
something similar , though far less profound , has been happening again in nonlinear science .
the broad topic is still coupled oscillators , but unlike the conservative oscillators studied by fermi et al . , the oscillators in question now are dissipative and have stable limit cycles .
this latest story began around 1990 , when a few researchers noticed an enormous amount of neutral stability and seemingly low - dimensional behavior in their simulations of josephson junction arrays specifically , arrays of identical , overdamped junctions arranged in series and coupled through a common load @xcite .
then , just a year ago , antonsen et al .
@xcite uncovered similarly low - dimensional dynamics in the periodically forced version of the kuramoto model of biological oscillators @xcite .
this was particularly surprising because the oscillators in that model are non - identical .
as in the soliton story , these numerical observations then inspired a series of theoretical advances .
these included the discovery of constants of motion @xcite , and of a pair of transformations that established the low - dimensionality of the dynamics @xcite .
but what remained to be found was the final piece , the identification of the hidden structure . without it ,
it was unclear why the transformations and constants of motion should exist in the first place . in this paper
we show that the group of mbius transformations is the key to understanding this class of dynamical systems .
our analysis unifies the previous treatments of josephson arrays and the kuramoto model , and clarifies the geometric and algebraic structures responsible for their low - dimensional behavior .
one spin - off of our approach is a new set of constants of motion ; these generalize the constants found previously , and hold for a wider class of oscillator arrays .
the paper is organized as follows . to keep the treatment self - contained and to establish notation , section [ background ] reviews the relevant background about coupled oscillators and the mbius group . in section [ mobius group reduction ]
we show how to use mbius transformations to reduce the dynamics of oscillator arrays with global sinusoidal coupling , a class that includes the josephson and kuramoto models as special cases .
the reduced flow lives on a set of invariant three - dimensional manifolds , arising naturally as the so - called group orbits of the mbius group .
the results obtained in this way are then compared to previous findings ( section [ connections to previous results ] ) and used to generate new constants of motion via the classical cross ratio construction ( section [ characteristics of the motion ] ) . we explore the dynamics on the invariant manifolds in section [ chaos in josephson arrays ] , and show that the phase portraits for resistively coupled josephson arrays are filled with chaos and island chains , reminiscent of the pictures encountered in hamiltonian chaos and kam theory .
the theory developed here was originally motivated by simulations of the governing equations for a series array of @xmath0 identical , overdamped josephson junctions driven by a constant current and coupled through a resistive load . as shown in tsang et al .
@xcite , the dimensionless circuit equations for this system can be written as @xmath4 for @xmath5 .
the physical interpretation need not concern us here ; the important point for our purposes is that this set of @xmath0 ordinary differential equations ( odes ) displayed low - dimensional dynamics .
the same sort of low - dimensional behavior was later found in other kinds of oscillator arrays @xcite as well as in josephson arrays with other kinds of loads @xcite .
building on contributions from several teams of researchers @xcite , watanabe and strogatz @xcite showed that the system ( [ jj_resistive ] ) could be reduced from @xmath0 odes to three odes , in the following sense . consider a time - dependent transformation from a set of constant angles @xmath6 to a set of functions @xmath7 , defined via @xmath8 = \sqrt{\frac{1+\gamma(t)}{1-\gamma(t ) } } \tan\left[\frac{\theta_j-\theta(t)}{2}\right]\ ] ] for @xmath5 . by direct substitution
, one can check that the resulting functions @xmath7 simultaneously satisfy all @xmath0 equations in ( [ jj_resistive ] ) as long as the three variables @xmath9 and @xmath10 satisfy a certain closed set of odes @xcite .
watanabe and strogatz also noted that the same transformation can be used to reduce any system of the form @xmath11 for @xmath12 , where @xmath13 is any smooth , complex - valued , @xmath14-periodic function of the phases @xmath15 .
( here the overbar denotes complex conjugate .
also , note that @xmath16 has to be real - valued since @xmath17 is real . )
the functions @xmath13 and @xmath16 are allowed to depend on time and on any other auxiliary state variables in the system , for example , the charge on a load capacitor or the current through a load resistor for certain josephson junction arrays .
the key is that @xmath13 and @xmath16 must be the same for all oscillators , and thus do _ not _ depend on the index @xmath18 .
we call such systems _ sinusoidally coupled _ because the dependence on @xmath18 occurs solely through the first harmonics @xmath19 and @xmath20 . soon after the transformation ( [ ws_transformation ] )
was reported , goebel @xcite observed that it could be related to fractional linear transformations , and he used this fact to simplify some of the calculations in ref .
@xcite . at that point ,
research on the reducibility of josephson arrays paused for more than a decade .
the question of _ why _ this particular class of dynamical systems ( [ reducible ] ) should be reducible by fractional linear transformations was not pursued at that time , but will be addressed in section [ mobius group reduction ] .
ott and antonsen @xcite recently reopened the issue of low - dimensional dynamics , with their discovery of an ansatz that collapses the infinite - dimensional kuramoto model to a two - dimensional system of odes . to illustrate their ansatz in its simplest form ,
let us apply it to the class of identical oscillators governed by eq.([reducible ] ) , in the limit @xmath21 .
( note that this step involves two simplifying assumptions , namely , that @xmath0 is infinitely large and that the oscillators are identical .
the ott - antonsen ansatz applies more generally to systems of non - identical oscillators with frequencies chosen at random from a prescribed probability distribution indeed , this generalization was one of ott and antonsen s major advances but it is not needed for the issues that we wish to address . ) in the limit @xmath21 , the evolution of the system ( [ reducible ] ) is given by the continuity equation @xmath22 where the phase density @xmath23 is defined such that @xmath24 gives the fraction of phases that lie between @xmath25 and @xmath26 at time @xmath27 , and where the velocity field is the eulerian version of ( [ reducible ] ) : @xmath28 our earlier assumptions about the coefficient functions @xmath13 and @xmath16 now take the form that @xmath13 and @xmath16 may depend on @xmath27 but not on @xmath25 . the time - dependence of @xmath13 and @xmath16 can arise either explicitly ( through external forcing , say ) or implicitly ( through the time - dependence of the harmonics of @xmath29 or any auxiliary state variables in the system ) .
following ott and antonsen @xcite , suppose @xmath29 is of the form @xmath30 for some unknown function @xmath31 that is independent of @xmath25 .
( our definition of @xmath31 is , however , slightly different from that in ott and antonsen @xcite ; our @xmath31 is their @xmath32 . )
note that ( [ oa_ansatz ] ) is just an algebraic rearrangement of the usual form for the poisson kernel : @xmath33 where @xmath34 and @xmath35 are defined via @xmath36 in geometrical terms , the ansatz ( [ oa_ansatz ] ) defines a submanifold in the infinite - dimensional space of density functions @xmath29 . this _ poisson submanifold _ is two - dimensional and is parametrized by the complex number @xmath31 , or equivalently , by the polar coordinates @xmath34 and @xmath35 .
the intriguing fact discovered by ott and antonsen is that the poisson submanifold is invariant : if the density is initially a poisson kernel , it remains a poisson kernel for all time . to verify this , we substitute the velocity field ( [ velocity ] ) and the ansatz ( [ oa_ansatz ] ) into the continuity equation ( [ continuity ] ) , and find that the amplitude equations for each harmonic @xmath37 are simultaneously satisfied if and only if @xmath38 evolves according to @xmath39 this equation can be recast in a more physically meaningful form in terms of the complex order parameter , denoted by @xmath40 and defined as the centroid of the phases @xmath25 regarded as points @xmath41 on the unit circle : @xmath42 by substituting ( [ oa_ansatz ] ) into ( [ z ] ) we find that @xmath43 for all states on the poisson submanifold .
hence , @xmath40 satisfies the riccati equation @xmath44 when @xmath13 and @xmath16 are functions of @xmath40 alone , as in mean - field models , eq.([riccati ] ) constitutes a closed two - dimensional system for the flow on the poisson submanifold .
more generally , the system will be closed whenever @xmath13 and @xmath16 depend on @xmath29 only through its fourier coefficients .
we will show this explicitly in subsection [ fourier coefficients of the phase distribution ] , by finding formulas for all the higher fourier coefficients in terms of @xmath31 , and hence in terms of @xmath40 .
( however , as we will see , things become more complicated for states lying off the poisson submanifold .
then @xmath40 no longer coincides with @xmath31 and the closed system becomes three dimensional , involving @xmath45 as well as @xmath31 . )
the work of ott and antonsen @xcite raises several questions .
why should the set of poisson kernels be invariant ?
what is the relationship , if any , between the ansatz ( [ oa_ansatz ] ) and the transformation ( [ ws_transformation ] ) studied earlier ? why does ( [ ws_transformation ] ) reduce equations of the form ( [ reducible ] ) to a three - dimensional flow , whereas ( [ oa_ansatz ] ) reduces them to a two - dimensional flow ?
as we shall see , the answers have to do with the properties of the group of conformal mappings of the unit disc to itself . before showing how this group arises naturally in the dynamics of sinusoidally coupled oscillators ,
let us recall some of its relevant properties .
consider the set of all fractional linear transformations @xmath46 of the form @xmath47 where @xmath48 and @xmath49 are complex numbers , and the numerator is not a multiple of the denominator ( that is , @xmath50 ) .
this family of functions carries the structure of a group .
the group operation is composition of functions , the identity element is the identity map , and inverses are given by inverse functions . of most importance to us
is a subgroup @xmath51which we refer to as the _ mbius group_consisting of those fractional linear transformations that map the open unit disc @xmath52 onto itself in a one - to - one way .
these transformations and their inverses are analytic on @xmath53 and map its boundary ( the unit circle @xmath54 ) to itself .
all such automorphisms of the disc can be written @xcite in the form @xmath55 for some @xmath56 and @xmath57 .
the mbius group @xmath51 is in fact a three - dimensional lie group , with real parameters @xmath58 re@xmath59 , and im@xmath59 .
however , it turns out that a different parametrization of @xmath51 will be more notationally convenient in what follows , in the sense that it simplifies comparisons between our results and those in the prior literature .
specifically , we will view a typical element of @xmath51 as a mapping @xmath60 from the unit disc in the complex @xmath61-plane to the unit disc in the complex @xmath62-plane , with parametrization given by @xmath63 where @xmath57 and @xmath64 .
note that the inverse mapping @xmath65 has an appearance closer to that of the standard parametrization ( [ usual_automorphism_parametrization ] ) . a word about terminology : our definition of the mbius group is not the conventional one .
usually this term denotes the larger group of all fractional linear transformations ( or bilinear transformations , or linear fractional transformations ) , whereas we reserve the adjective mbius for the subgroup @xmath51 and its elements .
thus , from now on , when we say _ mbius transformation _ we specifically mean an element of the subgroup @xmath51 consisting of analytic automorphisms of the unit disc .
in this section we show that if the equations for the oscillator array are of the form ( [ reducible ] ) , then the oscillators phases @xmath7 evolve according to the action of the mbius group on the complex unit circle : @xmath66 for @xmath67 , where @xmath68 is a one - parameter family of mbius transformations and @xmath6 is a constant ( time - independent ) angle .
in other words , the time-@xmath27 flow map for the system is always a mbius map .
incidentally , this result is consistent with a basic topological fact : we know that different oscillators can not pass through each other on @xmath69 under the flow , so we expect the time-@xmath27 flow map to be an orientation - preserving homeomorphism of @xmath69 onto itself and indeed any mbius map is .
we begin the analysis with an algebraic method similar to that in goebel @xcite .
then , in sections [ geometric method of finding dotalpha ] and [ geometric method of finding dotpsi ] , we adopt a geometrical perspective and show that it answers several questions left open by the first method .
parametrize the one - parameter family of mbius transformations as @xmath70 where @xmath71 and @xmath72 , and let @xmath73 to verify that ( [ 2 ] ) gives an exact solution of ( [ reducible])subject to the constraint that the mbius parameters @xmath38 and @xmath74 obey appropriate odes , to be determined we compute the time - derivative of @xmath75 , keeping in mind that @xmath76 is constant : @xmath77 from ( [ inverse_parametrization ] ) , we get @xmath78 which when substituted into ( [ 3 ] ) yields @xmath79 where @xmath80 .
note that eq.([5 ] ) falls precisely into the algebraic form required by ( [ reducible ] ) .
thus , to derive the desired odes for @xmath38 and @xmath74 , we now subtract ( [ 5 ] ) from ( [ reducible ] ) to obtain @xmath0 equations of the form @xmath81 , for @xmath67 .
if the system contains at least three distinct oscillator phases , then @xmath82 , @xmath83 , and @xmath84 must generically be zero .
explicitly , @xmath85 the system ( [ 6 ] ) can be algebraically rearranged to give [ 7 ] @xmath86 equations ( [ 7a ] ) and ( [ 7b ] ) have been derived previously ; they appear as eqs.(10 ) and ( 11 ) , respectively , in pikovsky and rosenblum s work @xcite , where they were derived by applying the transformation ( [ ws_transformation ] ) .
both their approach and the one above are certainly quick and clean , but they require us to guess the transformation ahead of time , and reveal little about why this transformation works . incidentally , observe that under the change of variables @xmath87 , ( [ reducible ] ) becomes @xmath88 equation ( [ 1.1 ] )
is a riccati equation with the form of ( [ 7a])another coincidence that seems a bit surprising when approached this way . in the following subsection
, we will see how these ricatti equations emerge naturally from the infinitesimal generators of the mbius group .
now we change our view of mbius maps slightly . instead of thinking of @xmath60 as a map from the @xmath61-plane to the @xmath62-plane
, we view it as a map from the @xmath62-plane to itself .
this requires a small and temporary change in notation , but it makes things clearer , especially when we start to discuss differential equations on the complex plane .
we begin by recalling some basic facts and definitions .
suppose the coupled oscillator system contains just three distinct phases among its @xmath0 oscillators .
then by a property of mbius transformations , there exists a unique mbius transformation from any point @xmath90 to any other point @xmath91 in the state space @xmath92 .
if the system instead contains only one or two distinct phases , many mbius transformations will take @xmath93 to @xmath94 , so we can still reach every point of the phase space from every other point .
however , if the system contains more than three distinct phases , say @xmath0 , then there is not in general a mbius transformation that transforms @xmath95 to @xmath96 ; only some points are accessible from @xmath93 , while others are not . in the language of group theory
, we say that @xmath94 is in the _ group orbit _ of @xmath93 if there exists a mbius map @xmath60 such that @xmath97 .
then , as a direct consequence of the fact that mbius maps form a three - parameter group @xmath51 under composition , the group orbits of @xmath51 partition the phase space into three - dimensional manifolds ( when the phase space is at least three - dimensional ) .
to compute infinitesimal generators for @xmath51 , we compute the time derivatives of the three one - parameter families of curves corresponding to the three parameters of @xmath51 : @xmath45 , @xmath98 and @xmath99 .
each of the three families is obtained from the mbius transformation by setting two of the three parameters to zero , and leaving the remaining parameter free .
for example , if we set @xmath100 at @xmath101 , these three families are @xmath102 where @xmath103 is written in place of @xmath104 and likewise for @xmath105 and @xmath106 .
we continue using this shorthand in subsequent equations , writing @xmath107 in place of @xmath108 for any one - parameter function @xmath109 .
the time derivatives of the curves in ( [ 7.1 ] ) evaluated at @xmath100 then give a set of infinitesimal generators for @xmath51 : @xmath110 note that these three generators point out into the full @xmath0-dimensional complex space @xmath111 , as expected .
meanwhile , if we substitute @xmath112 ( where @xmath113 and @xmath114 are real functions ) into the original riccati dynamics ( [ 1.1 ] ) , we can rewrite this equation of motion in terms of the three infinitesimal generators : @xmath115 the implication of the rewritten form ( [ 7.3 ] ) is then given by a theorem from lie theory : if @xmath116 is a lie group acting on a submanifold with linearly independent infinitesimal generators @xmath117 , and @xmath118 is a vector field of the form @xmath119 where the coefficients @xmath120 depend only on time @xmath27 , then the trajectory of the dynamics @xmath121 from any initial condition @xmath122 can be expressed in the form @xmath123 for a unique family @xmath124 parameterized by @xmath27 . since the mbius group is a complex lie group , this result can be applied directly to conclude ( [ 7.3 ] ) has the solution @xmath125 where @xmath126 is a unique one - parameter family of mbius transformations .
although we have so far assumed that the components @xmath127 of @xmath128 lie on the complex unit circle , both ( [ 2 ] ) and ( [ 7.3 ] ) extend naturally to all of @xmath111 .
this implies that @xmath129 must evolve as @xmath130 for some family @xmath126 .
however , eq . ( [ 2 ] ) shows that @xmath131 for all @xmath132 .
so @xmath133 for all @xmath27 , meaning that @xmath38 satisfies ( [ 7.3 ] ) . since ( [ 7.3 ] )
is just a rewriting of ( [ 1.1 ] ) , the dynamics ( [ 7a ] ) for @xmath31 that we derived earlier are now placed in a geometrical context .
this approach reveals that @xmath38 is just the image of the origin under a one - parameter family of mbius maps , applied to any one complex plane of @xmath111 .
it is even more illuminating to compute the infinitesimal generators within the @xmath0-fold torus @xmath134 of phase values , i.e. , the quantities @xmath135 .
these turn out to be @xmath136 when expressed in terms of these infinitesimal generators , the equation of motion ( [ 7.3 ] ) becomes @xmath137 which is precisely what we earlier referred to as a sinusoidally coupled system ( [ reducible ] ) , and whose solution must therefore be of the form @xmath138 for some @xmath139 .
this calculation finally clarifies what is so special about sinusoidally coupled systems ( [ reducible ] ) : they are induced naturally by a flow on the mbius group .
this fact underlies their reducibility and all their other beautiful ( but non - generic ) properties .
we turn next to the dynamics of @xmath45 .
as we will show in the next section , the action of the mbius transformation involves a clockwise rotation of the oscillator phase density @xmath23 by @xmath141 and a counterclockwise rotation by @xmath142 .
hence , @xmath74 may be viewed as the overall counterclockwise rotation of the distribution at time @xmath27 relative to the initial distribution at @xmath100 . to support this interpretation ,
we show here that @xmath140 equals the average value of the vector field on the circle , given by @xmath143 observe the right side of the integrand ( [ 3 ] ) has two terms : @xmath144 by cauchy s formula , @xmath145 so @xmath146 simplifies to @xmath147 note that @xmath148 has a pole in the unit disc , so we make the change of variables @xmath149 to move this pole outside the circle .
evaluating the resulting integral yields @xmath150 which completes the demonstration that @xmath151 .
we can now go back and evaluate the average vector field in a different way to find the differential equation that governs @xmath74 .
differentiating @xmath152 with respect to time and substituting the result into @xmath153 , we obtain @xmath154 since @xmath68 obeys the ricatti equation , we can eliminate @xmath155 in the numerator above to get @xmath156 there are three integrals to evaluate here .
the third one involves a term @xmath157 which has a pole inside the unit circle , so we do the same change of variables as before , @xmath149 , to move the pole outside .
the corresponding integral then simplifies to @xmath158 where the final integration follows from cauchy s formula .
similiarly , we use cauchy s formula to integrate the first and second terms of the integrand in ( [ 7.9 ] ) , and thereby obtain the desired differential equation for @xmath45 , thus rederiving ( [ 7b ] ) found earlier .
it is natural to ask how the trigonometric transformation ( [ ws_transformation ] ) used in earlier studies @xcite relates to the mbius transformation ( [ 2 ] ) used above .
as we will see , ( [ ws_transformation ] ) may be viewed as a restriction of ( [ 2 ] ) to the complex unit circle .
first , by trigonometric identities , we have @xmath159 = i\frac{1- e^{i(\phi-\phi)}}{1 + e^{i(\phi-\phi)}}.\ ] ] to connect this to mbius transformations , consider what happens when we apply the map defined by ( [ 2 ] ) to a point @xmath160 on the unit circle . since the image is also a point on the unit circle , it can be written as @xmath161 for some angle @xmath25 .
next let @xmath162 and divide both sides of ( [ 2 ] ) by @xmath163 .
thus @xmath164 where @xmath165 .
substitution of ( [ trigtomobius ] ) into the right side of ( [ 9 ] ) gives @xmath166 = \frac{1-r}{1+r } \bigg(i\frac{1- e^{i(\theta-\theta)}}{1 + e^{i(\theta-\theta)}}\bigg).\ ] ] by the identity ( [ 9 ] ) , eq.([10 ] ) is equivalent to ( [ ws_transformation ] ) with @xmath167 .
we can now see how the mbius parameters @xmath31 and @xmath45 operate on the set of @xmath168 in @xmath169 . from the relationships between @xmath170 , @xmath171 , @xmath35 and the mbius parameters ,
the initial phase density is first rotated clockwise around @xmath69 by @xmath141 , then squeezed toward one side of the circle as a function of @xmath172 , and afterwards rotated counterclockwise by @xmath142 .
the squeeze , which takes uniform distributions to poisson kernels , can be thought of as a composition of inversions , dilations and translations in the complex plane .
in section [ ott - antonsen ansatz ] and in a previous paper @xcite , we used the ott - antonsen ansatz ( [ oa_ansatz ] ) to show that systems of identical oscillators with global sinusoidal coupling contain a degenerate two - dimensional manifold among the three - dimensional leaves of their phase space foliation .
this two - dimensional manifold , which we called the poisson submanifold , consists of phase densities @xmath23 that have the form of a poisson kernel .
we now rederive these results within the framework of mbius transformations .
let @xmath173 denote one instance of the transformation ( [ ws_transformation ] ) ; in other words , fix the parameters @xmath35 , @xmath171 and @xmath170 and let @xmath174 .
let @xmath175 denote the normalized uniform measure on @xmath69 ; thus @xmath176 the transformation @xmath173 maps @xmath175 to the measure @xmath177 , and , by the usual formula for transformation of single - variable measures , we have @xmath178 , where the prime denotes differentiation by @xmath25 . from this equation
it follows that @xmath179 has the form of the poisson kernel , because the inverse of the mbius transformation ( [ 2 ] ) is @xmath180 which implies @xmath181 then by differentiation and algebraic rearrangement , we obtain @xmath182 the integral of @xmath183 over @xmath184 is @xmath185 , so @xmath179 is indeed a normalized poisson kernel .
finally , if the phase distribution @xmath186 ever takes the form of a poisson kernel with parameters @xmath187 and @xmath188 , then we can set @xmath189 , @xmath190 and @xmath191 , and the above calculation shows that @xmath186 remains a poisson kernel for all future and past times .
hence , the set of normalized poisson kernels constitutes an invariant submanifold of the infinite - dimensional phase space .
the above demonstration also reveals that the poisson submanifold has dimension @xmath192 where @xmath193 is the number of state variables besides @xmath31 , @xmath45 and the oscillator phases .
more concretely , it implies that when the system lies on the poisson submanifold , we can write @xmath89 as depending only on @xmath31 ; it is not possible to require @xmath89 to depend on @xmath45 in any real coupling scheme . to see this , we first consider the case in which the system is closed and there are no additional state variables .
suppose @xmath89 does _ not _ depend only on @xmath31 .
then some of the state space trajectories cross when projected onto the unit disc of @xmath31 values . at the point of any crossing , the phase density @xmath23 has multiple @xmath89 values .
but by ( [ 13 ] ) , the phase density depends only on @xmath31 , so there is nothing in the state space that can distinguish between the different @xmath89 values at that point .
hence , @xmath89 must be expressible in terms of @xmath31 alone . by an analogous argument
, @xmath89 is also independent of @xmath45 on the poisson submanifold when there are @xmath193 other state variables besides the oscillator phases and mbius parameters . on the other hand , if the time - dependence of @xmath89 arises only via a dependence on @xmath31
, then @xmath34 and @xmath35 decouple from @xmath45 and the dynamics are two - dimensional regardless of whether the system is evolving on the poisson submanifold or not .
observe that we can always force @xmath45-independence for @xmath89 by throwing away enough information about the locations of the other phases .
for instance , in the extreme , we may simply make @xmath13 and @xmath16 constant .
finally , even when @xmath89 does not depend solely on @xmath31 , the dynamics still may be two dimensional .
for example , in the case of completely integrable systems @xcite , the variables @xmath34 and @xmath194 decouple from @xmath35 to foliate the phase space with two - dimensional tori .
the reduction of ( [ reducible ] ) by the three - parameter mbius group suggests that the corresponding system of coupled oscillators should have @xmath1 constants of motion .
as we will see , these conserved quantities are given by the cross ratios of the points @xmath195 on @xmath69 .
recall from complex analysis @xcite that the _ cross ratio _ of four distinct points @xmath196 is @xmath197 this quantity is conserved under mbius transformations : for all @xmath31 and @xmath45 , @xmath198 . hence , the @xmath199 cross ratios of the @xmath0 oscillator phases remain constant along the trajectories in phase space .
we denote the constant value of @xmath200 as @xmath201 .
of course , we could have defined the cross ratio for four - tuples of _ non - distinct _ points as well , but these quantities are trivially conserved regardless of the dynamics and hence do not reduce the dimension of the phase space . to show that exactly @xmath1 of the cross ratios are independent , consider the sequence : @xmath202 @xmath203 .
each cross ratio in the sequence includes a new point not in the cross ratios preceding it and therefore must be independent of them . hence , there are at least @xmath1 independent cross ratios . with a bit more work
( see the appendix ) , we can also confirm that the rest of the cross ratios are functionally dependent on these @xmath1 integrals .
since the state space of the phases is an @xmath0-fold torus of real variables , we expect that each of the constants of motion can be expressed in terms of real functions and variables .
indeed , if @xmath204 , @xmath205 , @xmath206 , @xmath207 lie on the unit circle , then the cross ratio @xmath200 lies on @xmath208 .
we see this explicitly by pulling out @xmath209 from the factor @xmath210 of @xmath211 , and likewise for the other three factors , and then canceling the factors @xmath212 in the numerator and denominator to find @xmath213 where @xmath214.\ ] ] this way of writing the cross ratio also suggests a relationship with the constants of motion reported by watanabe and strogatz @xcite for completely integrable systems ( those with @xmath215 and @xmath216 , where @xmath40 is the phase centroid ( [ z ] ) ) .
these constants of motion , which we will call _ ws integrals _ , take the form @xmath217 where any permutation of the indices generates another ws integral . as previously demonstrated @xcite , exactly @xmath218 of the @xmath219 index permutations of ( [ 17.2 ] ) are functionally independent . as we might anticipate , the ws integrals imply that the cross ratios are constants of motion : consider two distinct ws integrals @xmath220 and @xmath221 , where @xmath222 denotes the remaining product of factors .
assume @xmath222 is the same for both @xmath223 and @xmath224 .
then @xmath225 .
since @xmath226 , @xmath18 , @xmath193 , @xmath227 are arbitrary , we can generate all cross ratios via this procedure . additionally ,
if a single ws integral holds for a system in which the cross ratios are invariant , then _ all _ ws integrals hold , since we can arbitrarily permute the indices of the first ws integral by sequences of transpositions of the form @xmath228 in which @xmath227 and @xmath193 are interchanged .
when we introduced @xmath13 and @xmath16 in section [ background ] , we required that they depend on the phases _ only _ through the fourier coefficients of the phase density @xmath23 .
since the centroid ( [ z ] ) is the fourier coefficient corresponding to the first harmonic @xmath229 , this condition is met by standard kuramoto models , josephson junction series arrays , laser arrays and many other well - studied systems of globally coupled oscillators .
our goal now is to show that this condition implies the closure of ( [ 7 ] ) , in the sense that @xmath89 and @xmath140 depend only on @xmath31 and @xmath45 .
to do so , we will show that the fourier coefficient of all higher harmonics @xmath230 for any integer @xmath231 may be expressed in terms of @xmath31 and @xmath45 . for a fixed measure @xmath232 on @xmath184 and a transformation @xmath233 of this measure via the mbius map
@xmath60 , the fourier coefficient of @xmath230 is given by @xmath234 we use the notation @xmath235 as a reminder that @xmath40 is the phase centroid .
we assume that we can take a fourier expansion of @xmath232 , so @xmath236 where the constants @xmath237 are independent of @xmath238 .
since the phase distribution must be real and normalized , we know that @xmath239 and @xmath240 , so we can write @xmath241 where @xmath160 and @xmath242 .
the formula for @xmath235 then becomes : @xmath243 now , @xmath244 is analytic on the open disc @xmath53 and @xmath245 .
meanwhile , the remaining term of the integrand of ( [ 20 ] ) has the complex conjugate @xmath246 which features an order-1 pole at @xmath247 and an order-@xmath231 pole at @xmath248 .
the first residue evaluates to zero , while the second is given by @xmath249 \biggl|_{w = -e^{-i\psi}\alpha.}\ ] ] therefore , @xmath235 is equal to @xmath250 added to the complex conjugate of this second residue : @xmath251 for example , the centroid may be written in terms of @xmath31 and @xmath45 as @xmath252 this calculation reveals what is so special about the poisson submanifold .
recall from section [ invariant manifold of poisson kernels ] that poisson kernels arise when we take @xmath175 to be the uniform measure .
then @xmath253 for all @xmath254 and @xmath43 .
in this exceptional case , the centroid simply evolves according to the riccati equation ( [ riccati ] ) and the dynamics of @xmath31 and @xmath45 decouple in eqs .
( [ 7a ] ) , ( [ 7b ] ) .
( a similar observation about the crucial role of the uniform measure here was made by pikovsky and rosenblum @xcite . the centroid evolution equation ( [ 7a ] ) on the poisson submanifold was first written down by ott and antonsen ; see eq.(6 ) in ref .
@xcite . )
but for the generic case of states lying off the poisson submanifold , @xmath40 is no longer equal to @xmath31 and the reduced dynamics become fully three - dimensional , due to the coupling between @xmath31 and @xmath45 induced by the relation ( [ 25 ] ) and the dependence of @xmath13 and @xmath16 on @xmath40 and the higher fourier coefficients . in the next section
we will explore some of the possibilities for such three - dimensional flows .
although the leaves of the foliation imposed by the mbius group action are only three - dimensional , they often contain chaos for commonly studied @xmath13 and @xmath16 @xcite . in this section
, we showcase this phenomenon by specializing to the case of a resistively - loaded series array of overdamped josephson junctions . in several previous studies of sinusoidally
coupled oscillators in the continuum limit , it was found that under certain conditions , the fourier harmonics of the phase density @xmath23 evolved as if they were decoupled , at least near certain points in state space @xcite . in the spirit of these observations
, we can get a sense for how individual harmonics contribute to the chaos by starting the system ( [ 7 ] ) from sinusoidal phase densities with different wavenumbers @xmath255 . to be more precise , we choose an initial density @xmath256 at @xmath257 , we choose @xmath258 so that @xmath68 in eq .
( [ 2 ] ) is simply the identity map , and the time - dependent change of variables @xmath259 reduces to @xmath260 , initially .
thus , the corresponding density of @xmath238 is @xmath261 this density is independent of time , just as the angles @xmath6 were in the finite-@xmath0 case .
next we flow the density forward by @xmath259 , where the mbius parameters @xmath262 satisfy the reduced flow ( [ 7 ] ) .
then , by our earlier results , the resulting density @xmath23 automatically satisfies the governing equations ( [ continuity ] ) , ( [ velocity ] ) .
the three - dimensional plot of @xmath263 , @xmath264 and @xmath74 indicates how such a single - harmonic density evolves in time , revealing for example whether it exhibits chaos , follows a periodic orbit , or approaches a fixed point . to ease the notation , from now on we write @xmath31 in cartesian coordinates as @xmath265 then the reduced flow ( [ 7 ] ) becomes @xmath266 where @xmath267 we immediately see that for every fixed point of this system , @xmath268 and @xmath45 is arbitrary . if for some change of state variables @xmath269 , @xmath270 , and @xmath271 , the odes @xmath272 and @xmath273 constitute a closed two - dimensional system and @xmath274 receives all of its @xmath27-dependence through @xmath275 and @xmath276 , then there could be other fixed points for the physical system , namely where @xmath277 but @xmath278 .
examples of the second type of fixed point include the splay states found on the poisson submanifold @xcite .
as discussed in section [ reducible systems ] , series arrays of josephson junctions with a resistive load have dynamics given by eqs .
( [ jj_resistive ] ) , ( [ continuity ] ) , and ( [ velocity ] ) , with @xmath279 and @xmath280 , where @xmath281 and @xmath282 are dimensionless combinations of certain circuit parameters @xcite and @xmath283 is the complex order parameter ( [ z ] ) .
the dynamics of @xmath284 , @xmath285 and @xmath45 are given by substitution into ( [ 28 ] ) : @xmath286 with @xmath287 . from ( [
25 ] ) and ( [ 26 ] ) , @xmath288 $ ] .
we can now plot the phase portrait for ( [ 29 ] ) on the cylinder @xmath289 . in the simple case where @xmath31 decouples from @xmath45 ,
trajectories can be projected down onto the @xmath31-disc without intersecting themselves or each other . however , in the more typical case that @xmath31 and @xmath45 are interdependent , we use poincar sections at @xmath290 to sort out the structure . in these poincar sections ,
quasiperiodic trajectories ( ideally ) appear as closed curves or island chains , periodic trajectories appear as fixed points or period-@xmath291 points of integer period , and chaotic trajectories fill the remaining regions of the unit disc .
first , however , we must choose an appropriate @xmath281 and @xmath282 .
to do so , we consider their definitions in terms of the original circuit parameters : @xmath292 and @xmath293 , where @xmath0 is the number of junctions , @xmath294 the source current , @xmath295 the load resistance , @xmath296 the critical current of each josephson junction , and @xmath297 the intrinsic josephson junction resistance @xcite .
because the resistances must be positive in the physical system , we examine only @xmath298 in our simulations .
additionally , @xmath296 represents a positive current magnitude , while @xmath294 reflects both a source current magnitude and direction . since the circuit is symmetric with respect to reversal of the source circuit ( see fig . 1 of @xcite )
, the corresponding dynamical system is left unchanged by the reflection @xmath299 .
hence , we also restrict our study to positive values of @xmath282 .
if @xmath300 , ( [ 29 ] ) implies there are fixed points at @xmath301 , @xmath302 for arbitrary @xmath45 . in numerical experiments ,
the negative-@xmath303 line of fixed points appears to attract distributions , while the positive-@xmath303 line repels them .
along the bifurcation curve @xmath304 , the two rows of fixed points merge at @xmath305 , and we find computational evidence that a splay state ( for which @xmath306 ) emerges from their union and moves inside the unit disc along the @xmath284-axis toward the origin as @xmath281 is decreased or @xmath282 is increased .
we can see from ( [ 29 ] ) that any such state must lie on the @xmath284-axis for all parameter values , as it did in previous characterizations of the poisson submanifold @xcite .
for the submanifolds we examined , chaos only appeared in the portion of the first quadrant in the @xmath281-@xmath282 plane that did not contain the fixed points , and the chaos became more widespread as @xmath307 .
this is illustrated schematically in fig .
[ pointplot ] ; the gradient of increasing darkness represents increasingly pervasive chaos . in submanifolds where the chaos was not widespread , the dynamics on the poincar sections were reminiscent of a kolmogorov - arnold - moser hamiltonian system with hierarchies of islands enclosing nested sets of closed orbits
nevertheless , we do not have an explicit hamiltonian for ( [ jj_resistive ] ) as we do for its averaged counterpart @xcite .
the increase in chaotic behavior is clearly visible in figs .
[ kamchaosa ] and fig .
[ kamchaosb ] , which show sequences of poincar sections corresponding to the points ( a ) and ( b ) in fig .
[ pointplot ] .
point ( a ) lies at @xmath308 , about 1/2 unit from the bifurcation curve @xmath304 , while point ( b ) lies at @xmath309 , about 1/3 unit from @xmath304 . as an example of the pattern of escalating chaos ,
observe that figs .
[ kamchaosb](a),(b),(c ) have larger , more dramatically overlapping chaotic regions than the corresponding plots ( a),(b),(d ) of fig .
[ kamchaosa ] .
although not shown , the chaotic trajectories that produced the scattered points in the poincar sections are phase coherent : they cycle smoothly and unidirectionally around the splay states throughout each period of @xmath45 . when the splay states are moved toward the edge of the unit disc by increasing @xmath281 or decreasing @xmath282
, these trajectories appear increasingly less prone to return to the same neighborhoods in the poincar sections , resulting in the observed amplification of chaotic behavior .
it is also possible to interpret the association between the parameter values and the intensity of the chaos in terms of the underlying physical parameters . in terms of these parameters
, the limit @xmath310 translates to @xmath311 or @xmath312 , which predicts that chaos should appear in real series arrays of josephson junctions if the source current is reduced to near the critical current of the junctions . even though the poincar sections in fig .
[ kamchaosa ] and fig .
[ kamchaosb ] show differing degrees of chaos , both series of plots depict a trend of decreasing chaotic behavior with increasing @xmath255 .
this stems from the dependence of @xmath16 on the phase centroid @xmath40 , which in turn arises because the oscillators are coupled only through their effect on the first harmonic of the phase density . for a coupling of this type , a sinusoidal phase density with a very short period and rapid oscillations ( high @xmath255 )
`` looks '' nearly identical ( in the riemann - lebesgue sense ) to a uniform density .
hence , in the limit of large @xmath255 , we see @xmath31 decoupling from @xmath45 , just as it does on the poisson submanifold ( recall that the poisson submanifold corresponds to a uniform density in @xmath238 , as shown in section [ invariant manifold of poisson kernels ] ) . from this perspective , then , chaos becomes increasingly dominant as we move `` away '' from the poisson submanifold , down toward small @xmath255 . finally , we point out a surprising feature in the poincar sections of ( a ) that was common in other simulations we performed . starting at @xmath313 , we see prominent sets of period-@xmath314 islands which appear for @xmath255 up to 8 in fig .
[ kamchaosa ] .
this ring of islands appears for higher @xmath255 as well and forms an increasingly larger and thinner band as @xmath255 is increased .
inside the dilating band , a set of nested orbits resembling the corresponding neutrally stable cycles of the poisson submanifold grows , filling the unit disc and approaching coincidence with the trajectories on the poisson submanifold .
we are currently unclear on why exactly @xmath315 islands emerge from the mbius group action on ( [ 26 ] ) , but pose this as an open question for future study . although it is tempting to try to extrapolate our numerical results to the case of non - identical oscillators , ott and antonsen @xcite have recently demonstrated that such systems contain a two - dimensional submanifold ( the generalization of the simpler poisson submanifold studied here ) that carries all the long - term dynamics of the phase centroid @xmath40 .
their results hold for the common case in which @xmath16 is a time - independent angular frequency with some distribution of values among the oscillators , and @xmath13 is a function of time , independent of oscillator variability . our numerical experiments , together with this new result , indicate that the widespread neutral stability in systems of identical , sinusoidally - coupled phase oscillators is a consequence of their special symmetries and underlying group - theoretic structure .
we show that the @xmath199 cross ratios of the oscillator phases are functionally dependent on the @xmath1 cross ratios @xmath316 .
to do so , we use the fact that the @xmath317 cross ratios corresponding to the @xmath317 permutations of @xmath318 can be written as elementary functions of @xmath319 : @xmath320 additionally , we can obtain new cross ratios from existing ones by multiplication : @xmath321 using these facts , we need to show that we can write @xmath322 for any distinct @xmath323 in terms of elements from @xmath316 . first , note that we can rewrite ( [ 15 ] ) as a function @xmath324 which takes two cross ratios @xmath319 and @xmath325 ( with indices in order ) , permutes the indices as necessary to eliminate @xmath326 , executes the multiplication and returns the product with its indices in order : @xmath327 observe , however , that @xmath324 is just short - hand for a composition of elementary functions from ( [ 14.2 ] ) : @xmath328 we can also define the analogous functions @xmath329 and @xmath330 : @xmath331 these functions have their own compositions like that of @xmath324 in ( [ 17 ] ) . let @xmath332 correspond to the permutation of @xmath322 in which the indices are in order
. we can write @xmath322 in terms of @xmath332 using one of the functions in ( [ 14.2 ] ) .
thus , the problem reduces to showing that we can obtain @xmath332 from the elements of @xmath316 by elimination of the indices between @xmath291 , @xmath333 , @xmath34 , @xmath334 using the operations @xmath324 , @xmath329 , @xmath330 .
if there are one or more indices between @xmath226 and @xmath18 , we say there is a _ gap _ between @xmath226 and @xmath18 . now observe that we can obtain the first gap between @xmath291 and @xmath333 using only @xmath319 with no gaps ; we grow this gap iteratively one index at a time by the operation : @xmath335 .
we can then grow the second gap between @xmath333 and @xmath34 to its full size using only @xmath319 that have no gaps between @xmath18 and @xmath193 or @xmath193 and @xmath227 ( each of which could be made from @xmath319 with no gaps ) using the operation : @xmath336 .
finally , we can create the third gap between @xmath34 and @xmath334 using only @xmath319 with no gaps between @xmath193 and @xmath227 ( which could be made from @xmath319 with fewer gaps ) using the operation : @xmath337 . | systems of @xmath0 identical phase oscillators with global sinusoidal coupling are known to display low - dimensional dynamics .
although this phenomenon was first observed about 20 years ago , its underlying cause has remained a puzzle .
here we expose the structure working behind the scenes of these systems , by proving that the governing equations are generated by the action of the mbius group , a three - parameter subgroup of fractional linear transformations that map the unit disc to itself . when there are no auxiliary state variables , the group action partitions the @xmath0-dimensional state space into three - dimensional invariant manifolds ( the group orbits ) .
the @xmath1 constants of motion associated with this foliation are the @xmath1 functionally independent cross ratios of the oscillator phases .
no further reduction is possible , in general ; numerical experiments on models of josephson junction arrays suggest that the invariant manifolds often contain three - dimensional regions of neutrally stable chaos .
* large arrays of coupled limit - cycle oscillators have been used to model diverse systems in physics , biology , chemistry , engineering and social science . the special case of phase oscillators coupled all - to - all through sinusoidal interactions has attracted mathematical interest because of its analytical tractability .
about 20 years ago , numerical experiments revealed that these systems display an exceptionally simple form of collective behavior : for all @xmath2 , where @xmath0 is the number of oscillators , all trajectories are confined to manifolds with @xmath3 fewer dimensions than the state space itself .
several insights have been obtained over the past two decades , but it has remained an open problem to pinpoint the symmetry or other structure that causes this non - generic behavior . here we show that group theory provides the explanation : the governing equations for these systems arise naturally from the action of the group of conformal mappings of the unit disc to itself .
this link unifies and explains the previous numerical and analytical results , and yields new constants of motion for this class of dynamical systems . * |
all biomembranes mainly consist of lipid bilayers , and functional proteins float on them .
such bilayers also appear in aqueous solutions of synthesized phospholipids .
therefore , bilayers in such solutions are intensively studied to understand the physical properties of biomembranes . moreover ,
these bilayers form vesicles that attract attention for studying model cells .
such vesicles can be effectively obtained by hydrating dry lipid films on substrates or test tubes .
( natural swelling method @xcite ) for example , dioleoylphosphatidylcholine ( dopc ) films , one of the typical phospholipids in biomembranes , provide micrometer - size vesicles .
these dopc vesicles normally grow with a multi - bilayer shell ( multi - lamellar vesicle ; mlv ) , although uni - lamellar vesicles ( ulvs ) are prefered for using as model cells .
therefore , effective methods to create ulvs have been proposed @xcite .
one of the authors has recently designed a novel method for preparing ulvs by hydrating dry lipid films mixed with sugar and salt @xcite . as some previous studies have pointed out
, an osmotic pressure would promote ulv formation @xcite .
the mechanism of ulv formation induced by osmotic pressure , however , has not been yet clarified . a key phenomenon to understand the mechanism of ulv formation is an `` unbinding transition '' ,
in which the inter - bilayer distance diverges infinitely and the bilayer stacking is unstabilized .
first , this transition was predicted theoretically by lipowsky @xcite .
recently , some experiments on this subject have been performed using small - angle x - ray scattering , small angle neutron scattering , and x - ray reflectivity @xcite .
the authors recently clarified the origin of the unbinding transition in terms of the interaction between bilayers ; a collaboration of the long- and short - range repulsive force , for example a steric repulsion due to membrane undulation @xcite and electrostatic repulsion , would be the origin of this transition @xcite .
therefore , the osmotic pressure would play an important role for the unbinding transition as these repulsions . in this study , we prepared two kinds of samples to investigate the effect of the osmotic pressure when dry lipid films include an additive before hydration .
one is an `` nai in film '' sample in which a dry dopc film mixed with nai was hydrated with pure water , that is , an `` nai in film '' sample is made by using the novel method to create ulvs .
the other is an `` nai in solution '' sample in which a dry dopc film was hydrated with an aqueous solution of nai for a control experiment . in this condition ,
dopc bilayers were in the liquid - crystalline phase at room temperature , and vesicles were effectively formed by hydration @xcite .
nai is suitable to prepare a homogeneous lipid film for `` nai in film '' , since it is a good solute for both water and methanol .
( the details are described in the experimental section . ) to investigate the dependence of the vesicle formation on the sample preparation , the structure of the multi - bilayer shell were observed using small - angle x - ray scattering ( saxs ) ; also , the shape of vesicles were observed by phase - contrast microscopy ( pcm ) .
dopc was purchased from sigma chemical inc . and nai from wako pure chemical industries ltd .
to prepare dry lipid films of `` nai in solution '' samples , dopc solutions of organic solvents , 1:2(v / v ) methanol / chloroform , were evaporated under n@xmath0 gas flow in a glass test tube , and kept in a vacuum at room temperature overnight .
the obtained lipid films were hydrated with an nai aqueous solution with various concentrations of nai . for `` nai in film '' samples , methanol solutions of various nai concentrations were prepared before mixing with chloroform .
then , dopc was dissolved in these organic solvents , the organic solvents were evaporated in a vacuum as the `` nai in solution '' samples , and the obtained dopc / nai films were hydrated in pure water .
the saxs experiments were performed at the bl40b2 beam port of spring8 at japan synchrotron radiation research institute ( jasri ) .
the incident x - rays were monochromatized by a double - crystal monochromator , and the wavelength was 1 ( @xmath1 ) .
the detector was an imaging - plate area detector placed at 1 m from the sample position .
the samples were prepared to be 1wt.% of dopc concentration , and to have the molar fractions of dopc : nai=10000:1 , 1000:1 , 100:1 , 10:1 , and 1:1 .
since all of the obtained two - dimensional data had no preferred orientation , they were azimuthally averaged to provide one - dimensional data .
all of the experiments were performed at room temperature .
the pcm experiments were performed by using nikon te-300 and recorded on s - vhs videotape at 30 frames / sec .
the dopc concentration was 1 mm ( about 0.08wt.% ) so as to avoid vesicle aggregation , and the molar ratio of dopc : nai was 1:1 .
the pcm experiments were also performed at room temperature as saxs experiments .
figure [ fig : saxs_all](a ) shows the saxs profiles obtained from the `` nai in solution '' samples .
all profiles have sharp bragg peaks due to the regular stacking of lipid bilayers , whose repeat distance , @xmath2 , is 63.5 .
this value is almost the same as that of an aqueous solution of dopc , @xmath3 @xcite .
this means that nai molecules in water had no effect on the lamellar structure .
figure [ fig : saxs_all](b ) shows the saxs profiles obtained from the `` nai in film '' samples . although the saxs profiles were essentially the same as the `` nai in solution '' samples at lower nai concentration , the bragg peaks split into two peaks at dopc : nai=10:1 , and completely disappeared at dopc : nai=1:1
this evidence imply that the osmotic pressure due to nai between bilayers destroys the multi - lamellar structure . to confirm that lipid bilayers still exist in the sample of dopc : nai=1:1 , we carried out a model fitting for the obtained saxs profile .
since the structure factor should be unity because of no correlation between the bilayers , the fitting was performed using @xmath4 where @xmath5 is the form factor of the bilayers , @xmath6 the scattering due to the concentration fluctuation of the lipids , and @xmath7 the constant background . in this model , @xmath6 was assumed to be lorentzian @xcite , @xmath8 where @xmath9 is the correlation length for the concentration fluctuation .
@xmath5 was assumed to be the fourier transform of three gaussians , which represents the bilayer electron density , as the following equation @xcite : @xmath10 \cos\left[\frac{d_{hh}q}{2}\right ] + \rho_r\sigma_c\exp\left[-\frac{(\sigma_cq)^2}{2}\right].\ ] ] here , @xmath11 , @xmath12 , @xmath13 , and @xmath14 are the parameters used to describe the electron density profiles of the lipid bilayers , as shown in fig .
[ fig : saxs_fq](b ) .
the fitting was performed with the least - square method for a @xmath15 range of @xmath16 [ @xmath17 .
the fitting model reproduced the experimental result well .
( b ) electron density profile obtained from the fitting .
@xmath11 is the standard deviation of the electron density of head groups , @xmath12 the standard deviation of the electron density of an end of hydrocarbon tails , @xmath13 the ratio of the peaks at the head groups to that at the end of hydrocarbon tails , and @xmath14 the head - to - head distance of a bilayer . ] as shown in fig .
[ fig : saxs_fq](a ) , a broad peak corresponding to the correlation between the head groups appeared around @xmath18 @xmath19 .
the fitting function reproduced the experimental result well , and the fitting parameters were evaluated to be @xmath11=2.38 , @xmath12=3.85 , @xmath13=-1.13 , and @xmath14=37.6 , which agrees with @xmath20 from the literature @xcite .
( the electron density profile calculated from these parameters is shown in fig .
[ fig : saxs_fq](b ) . )
since the electron density of the phosphate group is much greater than that of the other groups in the dopc bilayers , @xmath14 is the most characteristic parameter in the electron density profile .
therefore , the agreement of its value confirms that only the scattering due to the form factor was seen in the profile from the `` nai in film '' sample at dopc : nai=1:1 .
this suggests that the unbinding transition would occur in the `` nai in film '' sample , since the bragg peak due to the correlation between bilayers disappeared .
m and their shells were thick .
( b ) vesicles in the `` nai in film '' sample .
the size of the vesicles was over 10 @xmath21 m and their shell thicknesses were thin .
the scale bar is 10 @xmath21m.,scaledwidth=70.0% ] figure [ fig : pcm ] shows the vesicles observed by pcm .
the difference between the preparation procedures was clear : smaller vesicles with a thick shell formed in the `` nai in solution '' sample , whereas larger vesicles with a thin shell formed in the `` nai in film '' sample .
this is consistent with the disappearance of the bragg peak in the `` nai in film '' sample .
therefore , we concluded that an addition of nai to dopc films before hydration promotes ulv formation @xcite .
to discuss the relation between the unbinding transition and the ulv formation in the `` nai in film '' sample , we calculated the free - energy density profile of the outermost bilayer depending on the inter - bilayer distance . in this calculation , we assumed the following situation .
first , a lipid bilayer is known to stack on the surface of a substrate before hydration .
it is reasonable to assume that nai molecules in a dry lipid film locate between the bilayers , as shown in fig .
[ fig : calc_f](a ) @xcite . next
, only water can penetrate through the lipid bilayers on the hydration process , since the bilayer is known to be semipermeable .
furthermore , the free - energy density caused by osmotic pressure , @xmath22 , arises from the concentration difference of nai between inside and outside of the bilayers @xcite .
finally , the inter - bilayer distance increases up to a stable distance , where the free - energy profile has a local minimum . according to previous studies ,
three other considerable free - energy densities should be considered : the first one is due to the van der waals attractive force , @xmath23 , the second one is due to the hydration layers , @xmath24 , and the third one is from the steric repulsion due to the membrane undulation , @xmath25 @xcite .
therefore , the total free energy density @xmath26 can be described as follows : @xmath27,\\ f_{st}&=&0.42\frac{(k_bt)^2}{k_cd_w^2},\\ f_{osm}&=&-\frac{2xk_bt}{a}\ln[d_w],\label{eq : f_osm}\end{aligned}\ ] ] where , @xmath28 is the hamaker constant , @xmath29 the inter - bilayer distance , @xmath30 the bilayer thickness , @xmath31 the prefactor of @xmath24 , @xmath32 the decay length of hydration layers , @xmath33 the bending rigidity of bilayers , @xmath34 the molar ratio of nai to dopc , and @xmath35 the area per dopc molecule . .
the shift to upside with increasing @xmath34 is not due to an offset , but to the change of @xmath22.,title="fig:",height=188 ] .
the shift to upside with increasing @xmath34 is not due to an offset , but to the change of @xmath22.,title="fig:",height=188 ] .
.parameters determined by tristram - nagle @xcite [ cols="^,^,^,^,^,^,^",options="header " , ] [ tab : constant ] figure [ fig : calc_f](b ) shows the calculated free - energy densities as a function of @xmath29 by using these expressions with the parameters determined by tristram - nagle , as shown in table [ tab : constant ] .
a local minimum is seen at @xmath36 in the free - energy profile of the lowest nai content ( @xmath37 ) . the obtained @xmath2 value ( @xmath38 ) agrees with that obtained from the saxs profiles ( @xmath3963.5 ) .
therefore , this local minimum corresponds to the inter - bilayer distance of the bilayer stacking . with increasing the molar ratio of nai ,
this local minimum becomes shallow and disappears above @xmath40 due to an increase of the osmotic pressure .
this results in an increase of the inter - bilayer distance up to infinitely , _
i.e. _ , the unbinding transition takes place .
therefore , we concluded that the osmotic pressure due to the additive induced the unbinding transition on the hydration , which resulted in the ulv formation in the `` nai in film '' samples . it should be noted that the calculated critical concentrations to induce the unbinding transition were different from the saxs result ;
more than 10% of nai against dopc was required to induce the unbinding transition , while only about 0.04% of nai was required in the model calculation .
this would have originated from an over - estimation of the osmotic pressure , since the nai molecules should locate not only between bilayers , but also at the outside of bilayers in the multi - layer lipid film .
this problem is , however , not essential with respect to the mechanism of vesicle formation .
a difference of structure , depending on the process of mixing lipids with salt and hydration ( `` nai in solution '' and `` nai in film '' ) , was investigated with changing the molar ratio of nai to dopc using saxs and pcm .
the saxs experiment suggested that a large amount of nai in the `` nai in film '' sample induced the unbinding transition , and the pcm experiment showed that nai in the `` nai in film '' promoted the formation of ulvs .
the calculation of the free energy for the `` nai in film '' sample was also performed , and showed that the osmotic pressure was the origin of the unbinding transition . from these results
, we concluded that the osmotic pressure due to much amount of nai in the lipid film induces an unbinding transition which accelerates the formation of ulvs .
saxs experiments were carried at bl40b2 of spring-8 with the approval of the japan synchrotron radiation research institute ( jasri ) .
( proposal no .
2004b0520-nl2b - np ) we are grateful to dr .
k. inoue and dr .
s. sasaki ( jasri ) for experimental supports .
one of the authors ( h. s ) was supported by a grant - in - aid for scientific research ( no .
17540382 ) , by a grant - in - aid for the 21st century coe `` center for diversity and universality in physics '' from the ministry of education , culture , sports , science , and technology ( mext ) of japan , and by the yamada science foundation . .
. . in preparation
. in press .
the osmotic pressure @xmath41 due to the number of the molecule @xmath42 per volume @xmath43 is @xmath44 . in this case ,
@xmath42 and @xmath43 in a unit cell are @xmath45 and @xmath46 , respectively . on the other hand ,
the relation between @xmath22 and @xmath41 is described as @xmath47 from their definition .
therefore , we obtain eq .
( [ eq : f_osm ] ) under the assumption that any nai molecules are not ionized . | small - angle x - ray scattering and phase - contrast microscopy experiments were performed to investigate the effect of the osmotic pressure on vesicle formation in a dioleoylphosphatidylcholine ( dopc)/water / nai system .
multi - lamellar vesicles were formed when a pure lipid film was hydrated with an aqueous solution of nai . on the other hand , uni - lamellar vesicles ( ulvs )
were formed when a lipid film mixed with an enough amount of nai was hydrated . to confirm the effect of the osmotic pressure due to nai , a free - energy calculation was performed .
this result showed that the osmotic pressure induced an unbinding transition on the hydration process , which resulted in ulv formation . |
we consider finite graphs @xmath3 with vertex set @xmath4 and edge set @xmath5 .
the graphs might have multiple edges but no loops , and throughout this paper we assume that the graphs under consideration are connected ( otherwise we will mention this explicitely ) . terms and concepts that we do not define can be found in chapter 3 of @xcite . a perfect matching of a graph @xmath6 is a matching covering all vertices of @xmath6 , and @xmath7 denotes the number of odd components of @xmath6 .
tutte characterized the graphs with perfect matching .
[ tutte](tutte ) a graph @xmath6 has a perfect matching if and only if @xmath8 , for each @xmath9 . a class of graphs possessing a perfect matching is the class of @xmath0-graphs @xcite , which are @xmath0-regular graphs with @xmath10 for every odd @xmath9 , where @xmath11 is the set of edges of @xmath6 with precisely one end in @xmath12 .
the following conjectures are due to seymour @xcite .
[ seymourcoloring ] any @xmath0-graph is @xmath13-edge - colorable .
[ seymourperfects ] any @xmath0-graph contains @xmath2 perfect matchings such that each edge belongs to precisely two of them .
conjecture [ seymourperfects ] was first formulated by berge , fulkerson for @xmath14 ( berge fulkerson conjecture ) .
the berge fulkerson conjecture implies the following conjecture made by fan , raspaud @xcite .
[ fanraspaud ] any @xmath15-graph contains three perfect matchings @xmath16 such that @xmath17 .
conjecture [ seymourperfects ] implies that every @xmath0-graph has @xmath2 perfect matchings such that any three of them have an empty intersection . in the context of the conjectures of seymour ,
berge , fulkerson and of fan , raspaud , the following statement seems to be a natural generalization of conjecture [ fanraspaud ] .
[ fanraspaud_r ] for @xmath18 , any @xmath0-graph contains @xmath0 perfect matchings such that the intersection of any three of them is empty .
as the berge , fulkerson conjecture implies conjecture [ fanraspaud ] , conjecture [ seymourperfects ] implies conjecture [ fanraspaud_r ] . this paper studies the structure of minimum counter example to conjectures [ seymourcoloring ] and [ seymourperfects ] , and a variation of [ fanraspaud_r ] . a graph @xmath3 is matching covered if every edge belongs to a perfect matching .
it is factor - critical if @xmath19 has a perfect matching for every vertex @xmath20 , and it is bi - critical if @xmath21 has a perfect matching for every pair of vertices @xmath20 and @xmath22 .
a barrier in a matching covered graph is a set @xmath23 such that @xmath24 .
note that a single vertex in a matching covered graph is a barrier .
a non - bipartite , bi - critical and @xmath15-vertex - connected graph is a brick .
this paper proves that a minimal counter example to conjectures [ seymourcoloring ] and [ seymourperfects ] is a brick .
there are almost no results on these conjectures , and one might think about some variations of the original conjectures .
it is a natural question whether a perfect matching can be fixed in conjectures [ seymourperfects ] or [ fanraspaud ] , say : let @xmath6 be an @xmath0-graph ( @xmath25 ) and @xmath26 a perfect matching of @xmath6 , then @xmath6 has @xmath27 perfect matchings @xmath28 such that for any @xmath29 the intersection @xmath30 is empty .
we will show that this this is not true for any odd @xmath0 .
let @xmath3 be an @xmath0-graph .
an odd set @xmath23 with @xmath31 is non - trivial , if @xmath32 .
we start with the following theorem
. clearly , an @xmath0-graph is matching covered ( see exercise 3.4.4 in @xcite p. 113 ) , and let @xmath33 be any maximal barrier with @xmath34 .
then , @xmath35 has @xmath36 componets @xmath37 , which are factor - critical ( see lemma 4.6 .
in @xcite p. 198 ) .
since @xmath37 are odd , it follows that @xmath38 , for @xmath39 . on the other hand ,
the set @xmath33 can receive at most @xmath40 edges , thus @xmath33 is an independent set and @xmath41 , for @xmath39 .
case 1 : there is a maximal barrier @xmath33 with @xmath42 . if the components of @xmath35 are isolated vertices , then @xmath6 is bipartite and hence ( 1 ) .
if there is a component @xmath43 of @xmath35 , that is not an isolated vertex , then @xmath41 and @xmath6 satisfies the condition ( 2 ) of the theorem .
case 2 : for any maximal barrier @xmath33 , we have @xmath44 .
let @xmath45 .
@xmath46 is factor - critical , which means that for any vertex @xmath22 , the graph @xmath21 has a perfect matching , thus @xmath6 satisfies the condition ( 3 ) .
@xmath47 suppose that there is one and let @xmath49 be the vertices of the @xmath48-cut .
since @xmath6 is bi - critical , the graph @xmath21 has a perfect matching , thus all connected components @xmath37 of @xmath21 are even .
let @xmath50 denote the number of edges of @xmath6 that connect the vertices @xmath20 and @xmath22 .
for @xmath39 let @xmath51 and @xmath52 be the number of edges that connects @xmath20 and @xmath22 to @xmath43 , respectively .
clearly , @xmath53 let @xmath6 be a minimum counter - example to conjecture [ seymourcoloring ] .
since bipartite @xmath0-regular graphs are @xmath0-edge - colorable , @xmath6 is not bipartite .
next , we show that there is no non - trivial odd @xmath9 , such that @xmath31 .
suppose that there is one .
then consider the two graphs @xmath62 and @xmath63 that are obtained from @xmath6 by contracting @xmath12 and @xmath64 to a vertex , respectively .
clearly @xmath62 and @xmath63 are @xmath0-graphs ; moreover since they are smaller than @xmath6 , they are @xmath13-edge - colorable .
now , it is not hard to see that an @xmath13-edge - coloring of @xmath6 can be obtained from those of @xmath62 and @xmath63 , which would contradict the choice of @xmath6 .
let @xmath73 @xmath74 be an odd number and @xmath75 an unslicable @xmath0-graph .
let @xmath76 be a ( multi- ) cycle of length @xmath0 , where every edge has multiplicity @xmath36 , i.e. @xmath76 is @xmath69-regular .
replace every vertex @xmath22 of @xmath75 by a copy @xmath77 of @xmath76 to obtain a @xmath68-regular graph @xmath78 .
note that the _ old _ edges of @xmath75 in @xmath6 form a perfect matching @xmath26 of @xmath6 .
we first show that @xmath6 is an @xmath68-graph .
clearly @xmath79 is even .
let @xmath23 be an odd set and assume that @xmath80 .
if @xmath81 contains an edge of a cycle @xmath77 , then it contains at least @xmath69 of them , i.e. @xmath82 .
this implies that @xmath22 is a cut vertex in @xmath75 , contradicting the fact that @xmath75 is 2-vertex - connected .
thus @xmath81 contains at least one old edge of @xmath75 and hence @xmath83 , contradicting our assumption .
thus we may assume that @xmath81 contains only old edges of @xmath75 .
let @xmath84 be the subset of @xmath6 from which @xmath12 is obtained in the transfornmation from @xmath75 to @xmath6 .
@xmath12 is an odd set in @xmath6 and hence @xmath84 is an odd set in @xmath75 .
but then @xmath85 , contradicting the fact that @xmath75 is an @xmath0-graph .
thus @xmath6 is a @xmath68-graph .
now assume that @xmath6 has @xmath27 perfect matchings @xmath86 such that @xmath87 , for any @xmath29 .
consider @xmath88 .
since it is an odd cut , it follows that @xmath89 , for @xmath90 .
furthermore @xmath91 by the choice of @xmath26 .
if there is a perfect matching , say @xmath92 such that @xmath93 , then , since we assume that @xmath94 ( @xmath95 ) , the remaining @xmath96 perfect matching @xmath97 share @xmath98 edges .
thus there are @xmath99 , such that the intersection of @xmath26 , @xmath100 , and @xmath101 is not empty .
thus every perfect matching of @xmath102 contains precisely one edge of @xmath88 , and they are pairwise disjoint on @xmath88 . but
this implies that they induce @xmath27 pairwise disjoint perfect matchings @xmath103 on @xmath75 .
thus @xmath75 contains an @xmath27-graph , contradicting the fact that @xmath75 is unslicable . | an @xmath0-graph is an @xmath0-regular graph where every odd set of vertices is connected by at least @xmath0 edges to the rest of the graph .
seymour conjectured that any @xmath0-graph is @xmath1-edge - colorable , and also that any @xmath0-graph contains @xmath2 perfect matchings such that each edge belongs to two of them .
we show that the minimum counter - example to either of these conjectures is a brick .
furthermore we disprove a variant of a conjecture of fan , raspaud . |
suppose we are given a finite set @xmath1 , a finite cyclic group @xmath2 acting on @xmath1 , and a polynomial @xmath3 $ ] with integer coefficients .
following reiner , stanton , and white @xcite , we say that the triple @xmath4 exhibits the _
cyclic sieving phenomenon _ ( csp ) if for every integer @xmath5 , we have that @xmath6 where @xmath7 is a root of unity of multiplicitive order @xmath8 and @xmath9 is the fixed point set of the action of the power @xmath10 . the sizes of the fixed point sets determine the cycle structure of the canonical image of @xmath11 in the group of permutations of @xmath1 , @xmath12 .
therefore , to find the cycle structure of the image of any bijection @xmath13 , it is enough to determine the order of the action of @xmath11 on @xmath1 and find a polynomial @xmath14 such that @xmath15 exhibits the csp . the cyclic sieving phenomenon has been demonstrated in a variety of contexts .
the paper of reiner , stanton , and white @xcite itself includes examples involving noncrossing partitions , triangulations of polygons , and cosets of parabolic subgroups of coxeter groups .
an example of the csp with standard young tableaux is due to rhoades @xcite and will discussed further in section [ sec : rhoades ] .
now we turn to the csp of interest to this note .
let @xmath16 denote the longest element in the type @xmath17 coxeter group .
given generating set @xmath18 for @xmath17 , ( @xmath19 being the special " reflection ) , we will write a reduced expression for @xmath20 as a word in the subscripts . for example , @xmath21 can be written as @xmath22 we will abbreviate this product by @xmath23 .
it turns out that if we cyclically permute these letters , we always get another reduced expression for @xmath20 . said another way , @xmath24 for @xmath25 .
the same is not true for longest elements of other classical types . in type
a , we have @xmath26 , and for type d , @xmath27 let @xmath0 denote the set of reduced expressions for @xmath20 in type @xmath17 and let @xmath28 denote the action of placing the first letter of a word at the end .
then the orbit in @xmath29 of the word above is : @xmath30 @xmath31 as the length of @xmath20 is @xmath32 , we clearly have @xmath33 , and the size of any orbit divides @xmath32 . for an example of a smaller orbit , notice that the word @xmath34 has cyclic order 3 . for any word @xmath35 , ( e.g. , a reduced expression for @xmath20 ) ,
a _ descent _ of @xmath36 is defined to be a position @xmath37 in which @xmath38.the _ major index _ of @xmath36 , @xmath39 , is defined as the sum of the descent positions .
for example , the word @xmath40 has descents in positions @xmath41 , and @xmath42 , so its major index is @xmath43 .
let @xmath44 denote the generating function for this statistic on words in @xmath0 : @xmath45 the following is our main result .
[ theorem : main ] the triple @xmath46 exhibits the cyclic sieving phenomenon , where @xmath47 for example , let us consider the case @xmath48 .
we have @xmath49 let @xmath50 .
then we compute : @xmath51 thus , the @xmath52 reduced expressions for @xmath21 split into two orbits of size three ( the orbits of @xmath53 and @xmath54 ) and four orbits of size nine .
we prove theorem [ theorem : main ] by relating it to another instance of the csp , namely rhoades recent ( and deep ) result ( * ? ? ?
* thm 3.9 ) for the set @xmath55 of rectangular standard young tableaux with respect to the action of _ promotion _
( defined in section [ sec : prom ] ) . to make the connection , we rely on a pair of remarkable bijections due to haiman @xcite . the composition of haiman s bijections maps to @xmath0 from the set of square tableaux , @xmath56 . in this
note our main goal is to show that haiman s bijections carry the orbit structure of promotion on @xmath56 to the orbit structure of @xmath57 on @xmath0 .
we conclude this section by remarking that this approach was first outlined by rhoades ( * ? ? ?
* thm 8.1 ) .
one purpose of this article is to fill some nontrivial gaps in his argument .
a second is to justify the new observation that the polynomial @xmath14 can be expressed as the generating function for the major index on @xmath0 .
we thank brendon rhoades for encouraging us to write this note .
thanks also to kevin dilks , john stembridge , and alex yong for fruitful discussions on this and related topics , and to sergey fomin for comments on the manuscript .
for @xmath58 a partition , let @xmath59 denote the set of standard young tableaux of shape @xmath58 . if @xmath58 is a _ strict partition _ , i.e. , with no equal parts , then let @xmath60 denote the set of standard young tableaux of shifted shape @xmath58 .
we now describe the action of _ jeu de taquin promotion _ , first defined by schtzenberger @xcite
. we will consider promotion as a permutation of tableaux of a fixed shape ( resp . shifted shape ) , @xmath61 ( resp .
@xmath62 ) . given a tableau @xmath63 with @xmath64 , we form @xmath65 with the following algorithm .
( we denote the entry in row @xmath66 , column @xmath67 of a tableau @xmath63 , by @xmath68 . ) 1 .
remove the entry 1 in the upper left corner and decrease every other entry by 1 .
the empty box is initialized in position @xmath69 .
2 . perform jeu de taquin : 1 .
if there is no box to the right of the empty box and no box below the empty box , then go to 3 ) .
if there is a box to the right or below the empty box , then swap the empty box with the box containing the smaller entry , i.e. , @xmath70 . set @xmath71 , where @xmath72 are the coordinates of box swapped , and go to 2a ) .
3 . fill the empty box with @xmath73 .
here is an example : @xmath74 as a permutation , promotion naturally splits @xmath59 into disjoint orbits . for a general shape @xmath58 there seems to be no obvious pattern to the sizes of the orbits .
however , for certain shapes , notably haiman s generalized staircases " more can be said @xcite ( see also edelman and greene ( * ? ? ? * cor . 7.23 ) ) . in particular , rectangles fall into this category , with the following result .
[ thm : ordern ] if @xmath75 is a rectangle , then @xmath76 for all @xmath77 .
thus for @xmath78 square shapes @xmath58 , @xmath79 and the size of every orbit divides @xmath32 . with @xmath48 , here is an orbit of size 3 : @xmath80 there are @xmath52 standard young tableaux of shape @xmath81 , and there are @xmath52 reduced expressions in the set @xmath29 .
stanley first conjectured that @xmath0 and @xmath82 are equinumerous , and proctor suggested that rather than @xmath82 , a more direct correspondence might be given with @xmath83 , that is , with shifted standard tableaux of doubled staircase " shape .
( that the squares and doubled staircases are equinumerous follows easily from hook length formulas . )
haiman answers proctor s conjecture in such a way that the structure of promotion on doubled staircases corresponds precisely to cyclic permutation of words in @xmath0 ( * ? ? ?
* theorem 5.12 ) .
moreover , in ( * ? ? ?
* proposition 8.11 ) , he gives a bijection between standard young tableaux of square shape and those of doubled staircase shape that ( as we will show ) commutes with promotion .
as an example , his bijection carries the orbit in to this shifted orbit : @xmath84 both of these orbits of tableaux correspond to the orbit of the reduced word @xmath54 .
we first describe the bijection between reduced expressions and shifted standard tableaux of doubled staircase shape .
this bijection is described in section 5 of @xcite .
let @xmath63 in @xmath83 .
notice the largest entry in @xmath63 , ( i.e. , @xmath32 ) , occupies one of the outer corners .
let @xmath85 denote the row containing this largest entry , numbering the rows from the bottom up .
the _ promotion sequence _ of @xmath63 is defined to be @xmath86 , where @xmath87 . using the example above of @xmath88 we see @xmath89 , @xmath90 , and since @xmath91
, we have @xmath92 haiman s result is the following . [
theorem : phiandccommute ] the map @xmath93 is a bijection @xmath94
. by construction , then , we have @xmath95 i.e. , @xmath96 is an orbit - preserving bijection @xmath97 next , we will describe the bijection @xmath98 between squares and doubled staircases .
though not obvious from the definition below , we will demonstrate that @xmath99 commutes with promotion .
we assume the reader is familiar with the _ robinson - schensted - knuth insertion _
algorithm ( rsk ) .
( see ( * ? ? ?
* section 7.11 ) , for example . )
this is a map between words @xmath36 and pairs of tableaux @xmath100 .
we say @xmath101 is the _ insertion tableau _ and @xmath102 is the _ recording tableau_. there is a similar correspondence between words @xmath36 and pairs of shifted tableaux @xmath103 called _ shifted mixed insertion _ due to haiman @xcite .
( see also sagan @xcite and worley @xcite . )
serrano defined a semistandard generalization of shifted mixed insertion in @xcite . throughout this paper
we refer to semistandard shifted mixed insertion simply as _ mixed insertion_. details can be found in ( * ? ? ?
* section 1.1 ) .
[ theorem : bijection ] let @xmath36 be a word . if we view @xmath104 as a skew shifted standard young tableau and apply jeu de taquin to obtain a standard shifted young tableau , the result is @xmath105 ( independent of any choices in applying jeu de taquin ) .
for example , if @xmath106 , then @xmath107 @xmath108 performing jeu de taquin we see : @xmath109 haiman s bijection is precisely @xmath110 . that is , given a standard square tableau @xmath102 , we embed it in a shifted shape and apply jeu de taquin to create a standard shifted tableau .
that this is indeed a bijection follows from theorem [ theorem : bijection ] , but is originally found in ( * ? ? ?
* proposition 8.11 ) .
haiman s bijection @xmath99 applies more generally between rectangles and shifted trapezoids " , i.e. , for @xmath111 , we have @xmath112 .
all the results presented here extend to this generality , with similar proofs .
we restict to squares and doubled staircases for clarity of exposition .
we will now fix the tableaux @xmath101 and @xmath113 to ensure that the insertion word @xmath36 has particularly nice properties .
we will use the following lemma .
[ lem : insert ] fix a word @xmath36 .
let @xmath114 be the rsk insertion tableau and let @xmath115 be the mixed insertion tableau
. then the set of words that mixed insert into @xmath113 is contained in the set of words that rsk insert into @xmath101 .
now we apply lemma [ lem : insert ] to the word @xmath116 if we use rsk insertion , we find @xmath101 is an @xmath78 square tableau with all 1s in row first row , all 2s in the second row , and so on . with such a choice of @xmath101
it is not difficult to show that any other word @xmath117 inserting to @xmath101 has the property that in any initial subword @xmath118 , there are at least as many letters @xmath119 as letters @xmath120 . such words are sometimes called ( reverse ) _ lattice words _ or ( reverse ) _
yamanouchi words_. notice also that any such @xmath117 has @xmath73 copies of each letter @xmath37 , @xmath121 .
we call the words inserting to this choice of @xmath101 _ square words_. on the other hand , if we use mixed insertion on @xmath36 , we find @xmath113 as follows ( with @xmath122):@xmath123 in general , on the shifted half " of the tableau we see all 1s in the first row , all 2s in the second row , and so on . in the straight half "
we see only prime numbers , with @xmath124 on the first diagonal , @xmath125 on the second diagonal , and so on .
lemma [ lem : insert ] tells us that every @xmath117 that mixed inserts to @xmath113 is a square word .
but since the sets of recording tableaux for @xmath101 and for @xmath113 are equinumerous , we see that the set of words mixed inserting to @xmath113 is precisely the set of all square words .
yamanouchi words give a bijection with square standard young tableaux that circumvents insertion completely .
in reading the word from left to right , if @xmath126 , we put letter @xmath37 in the leftmost unoccupied position of row @xmath127 .
( see ( * ? ? ? *
proposition 7.10.3(d ) ) . )
we will soon characterize promotion in terms of operators on insertion words .
first , some lemmas . for a tableau @xmath63 ( shifted or not ) let @xmath128 denote the result of all but step ( 3 ) of promotion .
that is , we delete the smallest entry and perform jeu de taquin , but we do not fill in the empty box .
the following lemma says that , in both the shifted and unshifted cases , this can be expressed very simply in terms of our insertion word .
the first part of the lemma is a direct application of the theory of jeu de taquin ( see , e.g. , ( * ? ? ? *
a1.2 ) ) ; the second part is ( * ? ? ?
* lemma 3.9 ) .
[ lemma : recording ] for a word @xmath129 , let @xmath130 .
then we have @xmath131 and @xmath132 the operator @xmath133 acting on words @xmath134 is defined in the following way .
consider the subword of @xmath36 formed only by the letters @xmath120 and @xmath135 .
consider every @xmath135 as an opening bracket and every @xmath120 as a closing bracket , and pair them up accordingly .
the remaining word is of the form @xmath136 .
the operator @xmath133 leaves all of @xmath36 invariant , except for this subword , which it changes to @xmath137 .
( this operator is widely used in the theory of _
crystal graphs_. ) as an example , we calculate @xmath138 for the word @xmath139 . the subword formed from the letters @xmath140 and @xmath141 is@xmath142 which corresponds to the bracket sequence @xmath143 . removing paired brackets , one obtains @xmath144 , corresponding to the subword @xmath145 we change the last @xmath141 to a @xmath140 and keep the rest of the word unchanged , obtaining @xmath146 .
the following lemma shows that this operator leaves the recording tableau unchanged .
the unshifted case is found in work of lascoux , leclerc , and thibon ( * ? ? ?
* theorem 5.5.1 ) ; the shifted case follows from the unshifted case , and the fact that the mixed recording tableau of a word is uniquely determined by its rsk recording tableau ( theorem [ theorem : bijection ] ) .
[ lemma : operator ] recording tableaux are invariant under the operators @xmath147 .
that is , @xmath148 and @xmath149 let @xmath150 denote the composite operator given by applying first @xmath151 , then @xmath152 and so on .
it is clear that if @xmath153 is a square word , then @xmath154 is again a square word .
[ theorem : commute ] let @xmath153 be a square word .
then , @xmath155 and @xmath156 in other words , haiman s bijection commutes with promotion : @xmath157 by lemma [ lemma : recording ] , we see that @xmath158 is only one box away from @xmath159 .
further , repeated application of lemma [ lemma : operator ] shows that @xmath160 the same lemmas apply show @xmath161 is one box away from @xmath162 .
all that remains is to check that the box added by inserting 1 into @xmath163 ( resp .
@xmath164 ) is in the correct position . but
this follows from the observation that @xmath154 is a square word , and square words insert ( resp . mixed insert ) to squares ( resp . doubled staircases )
rhoades @xcite proved an instance of the csp related to the action of promotion on rectangular tableaux .
his result is quite deep , employing kahzdan - lusztig cellular representation theory in its proof .
recall that for any partition @xmath64 , we have that the standard tableaux of shape @xmath58 are enumerated by the frame - robinson - thrall _ hook length formula _ : @xmath165 where the product is over the boxes @xmath166 in @xmath58 and @xmath167 is the _ hook length _ at the box @xmath166 , i.e. , the number of boxes directly east or south of the box @xmath166 in @xmath58 , counting itself exactly once . to obtain the polynomial used for cyclic sieving , we replace the hook length formula with a natural @xmath168-analogue .
first , recall that for any @xmath169 , @xmath170_q : = 1 + q + \cdots + q^{n-1}$ ] and @xmath170_q ! : = [ n]_q [ n-1]_q \cdots [ 1]_q$ ] .
[ thm : cs ] let @xmath171 be a rectangular shape and let @xmath172 .
let @xmath173 act on @xmath1 via promotion .
then , the triple @xmath4 exhibits the cyclic sieving phenomenon , where @xmath174_q!}{\pi_{(i , j ) \in \lambda } [ h_{ij}]_q}\ ] ] is the @xmath168-analogue of the hook length formula . now thanks to theorem [ theorem : commute ] we know that @xmath99 preserves orbits of promotion , and as a consequence we see the csp for doubled staircases .
[ cor : sieving1 ] let @xmath175 , and let @xmath176 act on @xmath1 via promotion .
then the triple @xmath4 exhibits the cyclic sieving phenomenon , where @xmath177_q!}{[n]_q^n\prod_{i=1}^{n-1 } ( [ i]_q\cdot [ 2n - i]_q)^i}\ ] ] is the @xmath168-analogue of the hook length formula for an @xmath178 square young diagram . because of theorem [ theorem : phiandccommute ] the set @xmath0 also exhibits the csp .
[ cor : sieving2 ] let @xmath179 and let @xmath14 as in corollary [ cor : sieving1 ] .
let @xmath176 act on @xmath1 by cyclic rotation of words .
then the triple @xmath4 exhibits the cyclic sieving phenomenon .
corollary [ cor : sieving2 ] is the csp for @xmath0 as stated by rhoades .
this is nearly our main result ( theorem [ theorem : main ] ) , but for the definition of @xmath14 . in spirit ,
if @xmath4 exhibits the csp , the polynomial @xmath14 should be some @xmath168-enumerator for the set @xmath1 .
that is , it should be expressible as @xmath180 where @xmath181 is an intrinsically defined statistic for the elements of @xmath1 .
indeed , nearly all known instances of the cyclic sieving phenomenon have this property .
for example , it is known ( ( * ? ? ?
* cor 7.21.5 ) ) that the @xmath168-analogue of the hook - length formula can be expressed as follows : @xmath182 where @xmath183 and for a tableau @xmath63 , @xmath184 is the sum of all @xmath37 such that @xmath37 appears in a row above @xmath185 .
thus @xmath14 in theorem [ thm : cs ] can be described in terms a statistic on young tableaux . with this point of view , corollaries [ cor : sieving1 ] and [ cor : sieving2 ] are aesthetically unsatisfying .
section [ sec : comb ] is given to showing that @xmath14 can be defined as the generating function for the major index on words in @xmath0 .
it would be interesting to find a combinatorial description for @xmath14 in terms of a statistic on @xmath186 as well , though we have no such description at present .
as stated in the introduction , we will show that @xmath187 if we specialize to square shapes , we see that @xmath188 and @xmath189 thus it suffices to exhibit a bijection between square tableaux and words in @xmath0 that preserves major index .
in fact , the composition @xmath190 has a stronger feature .
define the _
cyclic descent set _ of a word @xmath191 to be the set @xmath192 that is , we have descents in the usual way , but also a descent in position 0 if @xmath193
. then @xmath194 .
for example with @xmath195 , @xmath196 and @xmath197 . similarly , we follow @xcite in defining the cyclic descent set of a square ( in general , rectangular ) young tableau . for @xmath63 in @xmath56 , define @xmath198 to be the set of all @xmath37 such that @xmath37 appears in a row above @xmath185 , along with 0 if @xmath199 is above @xmath32 in @xmath65 .
major index is @xmath200 .
we will see that @xmath201 preserves cyclic descent sets , and hence , major index . using our earlier example of @xmath195 , one can check that @xmath202 has @xmath203 , and so @xmath204 .
first , we observe that both types of descent sets shift cyclically under their respective actions : @xmath208 and @xmath209 for words under cyclic rotation , this is obvious . for tableaux under promotion ,
this is a lemma of rhoades ( * ? ? ?
* lemma 3.3 ) .
because of this cyclic shifting , we see that @xmath210 if and only if @xmath211 .
thus , it suffices to show that @xmath212 if and only if @xmath213 .
( actually , it is easier to determine if @xmath199 is a descent . )
let @xmath214 be the shifted doubled staircase tableau corresponding to @xmath36 .
we have @xmath215 if and only if @xmath32 is in a higher row in @xmath216 than in @xmath217 .
but since @xmath32 occupies the same place in @xmath216 as @xmath199 occupies in @xmath217 , this is to say @xmath199 is above @xmath32 in @xmath217 . on the other hand , @xmath218 if and only if @xmath199 is above @xmath32 in @xmath63 .
it is straightforward to check that since @xmath217 is obtained from @xmath63 by jeu de taquin into the upper corner , the relative heights of @xmath32 and @xmath199 ( i.e. , whether @xmath32 is below or not ) are the same in @xmath217 as in @xmath63 .
this completes the proof .
[ thm : interp ] the @xmath168-analogue of the hook length formula for an @xmath178 square young diagram is , up to a shift , the major index generating function for reduced expressions of the longest element in the hyperoctahedral group : @xmath219_q!}{[n]_q^n\prod_{i=1}^{n-1 } ( [ i]_q\cdot [ 2n - i]_q)^i}.\ ] ] theorem [ thm : interp ] , along with corollary [ cor : sieving2 ] , completes the proof of our main result , theorem [ theorem : main ] . because this result can be stated purely in terms of the set @xmath0 and a natural statistic on this set , it would be interesting to obtain a self - contained proof , i.e. , one that does not appeal to haiman s or rhoades work .
why must a result about cyclic rotation of words rely on promotion of young tableaux ? | we show that the set @xmath0 of reduced expressions for the longest element in the hyperoctahedral group exhibits the cyclic sieving phenomenon .
more specifically , @xmath0 possesses a natural cyclic action given by moving the first letter of a word to the end , and we show that the orbit structure of this action is encoded by the generating function for the major index on @xmath0 . |
a general theory of games first introduced in @xcite has found several applications in the field of economics and engineering .
a solution concept or a notion of equilibrium was proposed by nash ( known as nash equilibrium ) in @xcite and was shown to exist in every finite normal - form game .
further generalizations of nash equilibrium such as correlated equilibrium and coarse correlated equilibrium were also introduced and studied .
it is well known that for every game the set correlated and coarse - correlated equilibria are convex subsets of the strategy space .
but in general the set of nash equilibria is not convex .
a number of methods have been proposed to compute a nash equilibium strategy .
lemke - howson s algorithm for bi - matrix games@xcite , global newton method@xcite , homotopy based methods@xcite are some of the few methods to compute a nash equilibrium strategy . for a general n - player game ,
the associated optimization problem is non - linear and non - convex and hence is difficult to solve .
it is known that the problem of computing nash equilibria in bi - matrix games is a linear complementarity problem and for the general n - player scenario it is a non - linear complementarity problem .
linear complimentarity problems ( the ones arising from games ) can be solved using lemke - howson s method , while non - linear complimentarity problems are in general hard to solve and require some sufficient conditions to be imposed on the problem to solve them which is not satisfied by every game . in this paper
we present optimization problems with biconvex objective function and linear constraints such that the set of global minima of the optimization problems is the same as the set of nash eqilibria of a n - player general - sum normal form game . global optimization algorithms
exist that can compute the global minima of such optimization problems@xcite .
the main idea in the formulation of these optimization problems is the fact that correlated or coarse - correlated equilibrium which are product of individual player s strategy is a nash equilibrium .
we further show that the objective function is an invex function i.e. the set of stationary points is the same as the set of global minima .
we also consider a projected gradient descent scheme and prove that is converges to a partial optimum of the objective function . the remainder of this paper is organised as follows : in section 2 , necessary definitions and notations are stated . in section 3 ,
functions with required properties are defined . in section 4 ,
properties of the functions defined in section 2 are proved . in section 5 ,
optimization problems are presented . in section 6 ,
the projected gradient descent algorithm is stated and convergence analysis is performed . in section 7 ,
simulation results of the projected gradient descent algorithm on certain test cases are presented . in section 8 , we summarize and present directions for future research .
in this section we shall state definitions , introduce variables and notations used later in this paper . a normal form game ( or simply a game ) ( @xmath0 )
is defined by tuple @xmath1 where , @xmath2 denotes the set of players ( @xmath3 ) , @xmath4 denotes the set of actions of player @xmath5 ( @xmath6 ) .
let @xmath7 and @xmath8 denotes the utility function of player @xmath5 . for every @xmath9
, @xmath10 denotes the set of probability distributions on @xmath11 .
@xmath10 is identified by the probability simplex @xmath12 .
@xmath13 denotes a generic element of @xmath10 .
let @xmath14 which is identified as a vector in @xmath15 where @xmath16 .
let @xmath17 .
let @xmath18 denote the set of probability distributions on @xmath19 .
@xmath18 is identified by the probability simplex @xmath20 where @xmath21 .
@xmath22 denotes a generic element in @xmath18 . for every @xmath9 , @xmath23 and @xmath24 denotes a generic element in @xmath25 .
similarly , this can be extended to more than one player .
@xmath26 . similarly define @xmath27 and @xmath28 denote a generic element in @xmath29 .
@xmath30 . for every @xmath9 , @xmath31 where @xmath32 . for every @xmath9 , @xmath33 where @xmath34 . for every @xmath9 , @xmath35 and @xmath36 .
similarly define @xmath37 .
@xmath38 is said to be a * nash equilibrium strategy of the game @xmath0 ( or just n.e . )
if @xmath39 .
let @xmath40 denote the set of nash equilibria strategies of game @xmath0 .
* @xmath41 is said to be a * correlated equilibrium strategy of the game @xmath0 ( or just c.e . ) if @xmath42 .
let @xmath43 denote the set of correlated equilibria of the game @xmath0 .
* @xmath41 is said to be a * coarse correlated equilibrium strategy of the game @xmath0 ( or just c.c.e . ) if @xmath44 .
let @xmath45 denote the set of coarse correlated equilibria of the game @xmath0 .
* define @xmath46 , s.t .
, @xmath47 where @xmath48 .
let the graph of the function @xmath49 be @xmath50 . in the following lemma
we summarize the relationship between the various equilibrium concepts defined.*lemma 2.1 : given a game @xmath0 . the following hold .
* * @xmath51 .
* @xmath52 . * @xmath53 .
the results in lemma follow directly from definitions .
@xmath54 is a * nash equilibrium profile of game @xmath0 if @xmath55 is a nash equilibrium strategy of game @xmath0 and @xmath56 .
* let @xmath57 and @xmath58 be two convex subsets of @xmath59 and @xmath60 respectively .
a function @xmath61 is said to be a * biconvex function if @xmath62 is a convex function and @xmath63 is a convex function .
@xmath64 is a * partial optimum of a biconvex function @xmath65 if @xmath66 and @xmath67 . for a detailed study of biconvex functions
see @xcite . *
* let @xmath68 be a subset of @xmath69 and @xmath70 .
@xmath71 is said to the global optimum of the optimization problem @xmath72 , if , @xmath73 .
in this section we shall define functions whose set of zeros is the same as the set of nash equilibria of the game @xmath0 .
the following theorem gives a necessary and sufficient condition for @xmath74 to be in @xmath75 .
* theorem 3.1 : given @xmath74
. then , @xmath76 iff @xmath77 . *
proof : [ @xmath78 assume @xmath76 .
fix @xmath79 . then@xmath80 and @xmath81 . therefore @xmath82 . since @xmath83 are arbitrary , @xmath84 , @xmath85 .
[ @xmath86 fix @xmath87 . from data ,
we know that @xmath88 . using the above
, we get , @xmath89 . from data , we also know that @xmath90 . therefore by substituting for the sum , we get , @xmath91 . similarly repeating the above procedure for actions of the third player we get , @xmath92 .
proceeding all the way upto player @xmath93 we get , @xmath94 . since @xmath41 , we know that @xmath95 . therefore , @xmath96 . since @xmath97 is arbitrary , @xmath98.@xmath99 * * using the above theorem we now define a non - negative function on @xmath100 such that the function takes the value zero on @xmath75 and is positive on @xmath101 .
let @xmath102 such that , @xmath103 .
* corollary 3.1 : given @xmath74
. then , @xmath104 iff @xmath76 . * from the definitions of coarse - correlated equilibrium and correlated equilibrium we now define the following non - negative functions on @xmath18 such that they take the value zero on the set of coarse - correlated equilibria ( @xmath45 ) and correlated equilibria ( @xmath43 ) respectively .
let @xmath105 , such that , @xmath106 and @xmath107 , such that , @xmath108 .
* lemma 3.1 : given @xmath109 . *
* @xmath110 iff @xmath111 . * @xmath112 iff @xmath113 .
* proof : follows directly from the definitions of correlated equilibrium and coarse correlated equilibrium in section 2.@xmath99 * let @xmath114 s.t .
the idea is that when @xmath116 and @xmath117 , then , @xmath118 is a best response to @xmath28 .
* lemma 3.2 : given @xmath116 .
@xmath117 iff @xmath55 is a nash equilibrium.*proof : [ @xmath78since @xmath117 , we have , @xmath119 . hence @xmath120 . since @xmath121 , @xmath122 and @xmath123 .
therefore , @xmath124 , which by definition of a nash equilibrium strategy in section 2 , implies @xmath55 is nash equilibrium . * * [ @xmath86 since @xmath55 is a nash equilibrium , we have , @xmath124 . since @xmath121 , @xmath122 and @xmath123 .
therefore , @xmath120 , which further implies , @xmath119 .
thus @xmath117.@xmath99
we now characterise the set of nash equilibria of a game ( @xmath0 ) using the functions @xmath125 and @xmath126 .
* theorem 3.2 : given @xmath54 .
* * @xmath127 is a nash equilibrium profile iff @xmath128 .
* @xmath127 is a nash equilibrium profile iff @xmath129 .
* @xmath127 is a nash equilibrium profile iff @xmath130 .
* proof : first we shall prove ( 1).[@xmath78 assume @xmath127 is a nash equilibrium .
then , by definition of nash equilibrium profile in section 2 , @xmath55 is a n.e . and
@xmath56 . by lemma 2.1 , since @xmath55 is a n.e .
@xmath131 and since @xmath56 , @xmath121 .
thus @xmath104 and @xmath110 by theorem 3.1 and lemma 3.1 respectively .
therefore @xmath128.assume @xmath128 .
since both @xmath132 and @xmath133 are non - negative , @xmath104 and @xmath110 . by theorem 3.1 , @xmath104 will imply @xmath121 and by lemma 3.1 @xmath110 will imply @xmath111 . since @xmath134 and @xmath56 ,
from lemma 2.1 , we have that @xmath55 is a n.e .
thus @xmath127 is a nash equilibrium . *
proof of ( 2 ) is similar to that of ( 1 ) and the proof of ( 3 ) follows from lemma 3.2 and corollary 3.1.@xmath99
in this section we shall prove certain properties of the functions constructed in section * no . first , we shall prove that @xmath132 is biconvex and that @xmath133 and @xmath126 are convex .
* lemma 4.1 : @xmath132 is a biconvex function i.e. @xmath135 is convex and @xmath136 is convex.*proof : @xmath137 where @xmath48 , @xmath138 is a linear function of @xmath41 and an affine function of @xmath139 . by proposition 1.1.4 in @xcite , @xmath140 is convex in @xmath41 and @xmath38 with the other fixed .
since sum of convex functions is convex , @xmath141 is convex in @xmath142 for every fixed @xmath38 and is convex in @xmath55 for every fixed @xmath143.@xmath99 * lemma 4.2 : @xmath144 and @xmath145 are convex functions of @xmath41.*proof : first we shall show @xmath133 is convex .
@xmath146 , @xmath147 is linear in @xmath41 .
since supremum of convex functions is convex , we have , @xmath146 , @xmath148 . since composition of nondecreasing function and convex function is convex , @xmath146 , @xmath149 , is convex .
therefore , @xmath150 is a convex function . * * * * * similarly we can show that @xmath145 is also a convex function.@xmath99 it is easy to show @xmath151 and @xmath145 are continuously differentiable on an open set containing their respective domains ( for a similar proof refer @xcite ) . let @xmath152^t$ ] , where @xmath153 and @xmath154 . for every @xmath155 , @xmath156\end{aligned}\ ] ] so as to compute @xmath157 , we shall write @xmath158 , where @xmath159 s.t . @xmath160 ( which is possible since @xmath138 is linear in @xmath142 ) .
therefore , @xmath161 the following lemma says that set of partial optima of @xmath132 , the set of stationary points of @xmath132 and the set of global minima of @xmath132 are all the same.*lemma 4.3 : given @xmath162 .
then the following are equivalent .
* * @xmath163 is a partial optimum of @xmath132 .
* @xmath163 is s.t . @xmath104 .
* @xmath163 is s.t . @xmath164 .
* proof : [ @xmath165 . since @xmath163 is a partial optimum of @xmath132 , @xmath166 .
hence , @xmath167 . therefore , @xmath168 . * [ @xmath169 . since @xmath104 , @xmath77 . substituting the above in the expression of @xmath170 and @xmath157 we get , @xmath164 .
[ @xmath171 . since @xmath132 is biconvex ( from lemma 4.1 ) ,
@xmath172 and @xmath173 are convex functions . from proposition 1.1.7 in @xcite
, we get , @xmath174 and @xmath175 . substituting @xmath176^t=0 $ ] ,
will give , @xmath177 and @xmath178 .
thus , @xmath163 is a partial optimum of @xmath132.@xmath99 so as to compute @xmath179 , we shall write @xmath180 where @xmath181 ( which is possible since @xmath147 is linear in @xmath142 ) . then @xmath182 .
the following lemma says that the set of global minima of @xmath133 and the set of stationary points of @xmath133 are the same .
* lemma 4.4 : given @xmath183 . @xmath184 iff @xmath185 . * proof : follows directly from the expression of the gradient and the convexity of @xmath133.@xmath99 * * a similar result can be derived for @xmath126 . in what follows in this paper results proved for @xmath133 can be extended to @xmath126 as well . in theorem 3.2
we showed that the set of zeros of @xmath186 is the same as the set of nash equilibrium profiles of the game @xmath0 . in the following lemma
we show that the set of zeros of @xmath186 is the same as the set of stationary points of the function @xmath186 .
* lemma 4.5 : given @xmath187 . @xmath188 iff @xmath189 . * proof : [ @xmath78 since @xmath188 and that @xmath132 and @xmath133 are non - negative , will imply that @xmath168 and @xmath184 .
thus , @xmath176^t=0 $ ] and @xmath185 by lemma 4.3 and 4.4 respectively
. therefore , @xmath190^t=0 $ ] . * * [ @xmath86since @xmath190^t=0 $ ] , we have @xmath191 . @xmath192 . by substituting the expressions for @xmath193 and @xmath194 we get , @xmath195 and @xmath196 .
therefore , @xmath197.@xmath99 lemma 4.5 shows that the function @xmath186 is invex .
similarly it can shown that @xmath198 is also invex . in following lemma
we show that @xmath199 is a biconvex function . as a consequence of this lemma , lemma 4.1 and lemma 3.3 in @xcite , we get , @xmath200 is a biconvex function .
* lemma 4.6 : @xmath199 is a biconvex function i.e. @xmath201 is a convex function and @xmath202 is a convex function . *
proof : proof is similar to that of lemma 4.1.@xmath99 * *
in this section we shall state the optimization problems obtained using the functions constructed in the previous sections such that the global minima of the optimization problem correspond to nash equilibria of the game @xmath0 .
first optimization problem ( @xmath203 ) is stated below : @xmath204 the constraints in the above optimization problem ensure that the feasible set is @xmath205 . the second optimization problem ( @xmath206 )
is stated below :
@xmath207 the following theorem says that the set of global minima of the optimization problem ( @xmath203 ) is the same as the set of nash equilibria profiles of the game @xmath0.*theorem 5.1 : for every game @xmath0 , there exists @xmath162 s.t .
@xmath188 . further given @xmath208 ,
@xmath188 iff @xmath163 is a nash equilibrium profile.*proof : since for every game there exists @xmath209 , s.t . , @xmath210 is a n.e .
( see @xcite ) .
thus by theorem 3.2 , @xmath163 with @xmath211 satisfies @xmath188 .
the other part follows directly from theorem 3.2.@xmath99 * * a similar claim can be proved for @xmath206 .
the above two optimization problems have a biconvex objective function with convex ( linear ) constraints .
global optimization algorithm exists that solves the above two optimization problems ( see @xcite ) .
in this section we shall consider a projected gradient descent algorithm to solve @xmath203 .
the algorithm is stated below : * input : * * @xmath212 : initial point for the algorithm , * @xmath0 : the underlying game , * @xmath213 : step size sequences chosen as follows : * * @xmath214 , * * @xmath215 , * * @xmath216 , * @xmath217 : projection operator ensuring that @xmath127 remains in @xmath205 .
* output : after sufficiently large number of iterations(@xmath218 ) the algorithm outputs the terminal strategy @xmath219.@xmath220 * in what follows we shall present the convergence analysis of the above projected gradient descent algorithm .
we shall analyse the behaviour of the above algorithm using the o.d.e .
method presented in @xcite . in order to use the results from @xcite , we need the gradient function to be lipschitz continuous on @xmath205 , which is proved in the following lemma.*lemma 6.1 : there exists @xmath221 , s.t . , @xmath222 , * @xmath223 * proof : it is easy to see that the function @xmath224 is twice continuously differentiable on an open set containing @xmath205 . thus @xmath225 is continuously diffrentiable on @xmath205 .
hence @xmath226 for some @xmath227 . by mean value theorem
, we have , @xmath225 is lipschitz continous with lipschitz constant @xmath228 .
let @xmath229 .
fix @xmath230 . clearly , @xmath231 .
therefore , we have , @xmath232 where @xmath233}|$ ] . since @xmath234
, we have , @xmath235 , where @xmath236 .
since sum of two lipschitz continuous functions is lipschitz continuous , we have , @xmath237 is lipschitz continous with lipschitz constant @xmath238.@xmath99 * in order to study the asymptotic behaviour of the recursion presented in the algorithm , by results in section 3.4 of @xcite , it is enough to study the asymptotic behaviour of the o.d.e .
, @xmath239 where @xmath240 i.e. the directional derivative of @xmath217 at @xmath241 along the direction @xmath242 . the above o.d.e .
is well posed i.e. has a unique solution for every initial point in @xmath205 ( for a proof see @xcite ) .
@xmath205 , is a cartesian product of simplices and hence the projection of @xmath243 on to @xmath205 is the same as projection of @xmath244 on to @xmath245 and @xmath246 on to @xmath18 i.e. @xmath247^t$ ] where @xmath248 denotes the projection operator which projects every vector in @xmath69 on to @xmath249 .
thus , in order to compute the directional derivative of @xmath217 , it is enough to consider the directional derivative of the projection operator on to individual simplices and then juxtaposing them would give us the directional derivative of @xmath217 .
the computation of the directional derivative of a projection operation on to a simplex can be found in @xcite which we shall state here .
let @xmath250 and @xmath251 .
then , @xmath252 where @xmath253 , s.t .
, @xmath254 .
let @xmath255 .
fix @xmath256 be a initial point of the o.d.e .
[ o.d.e . ] and
the corresponding unique solution be @xmath257 .
then , @xmath258 by substituing [ dd ] and the fact that @xmath259 in the above equation we get , @xmath260 where the last inequality follows from the application of cauchy schwartz and the fact that @xmath261 . therefore along every solution of the o.d.e .
] , the value of the potential function @xmath262 reduces and hence the above o.d.e . converges to an internally chain transitive invariant set contained in @xmath263 . in the following lemma
we shall prove that @xmath264 is an equilibrium point of o.d.e .
[ o.d.e.].*lemma 6.2 : if @xmath264 , then , @xmath265.*proof : if @xmath264 is such that @xmath266 , then @xmath267 .
assume @xmath268 .
since @xmath264 , @xmath269 . by cauchy schwartz inequality , @xmath270 and @xmath271 .
since their sum is zero , we get , @xmath272 and @xmath273 . hence , @xmath274 and @xmath275 . by , definition of @xmath276 in equation [ dd ] ,
we get , @xmath277 and @xmath278 . substituing for @xmath279 and @xmath280 in the expression for @xmath281 and @xmath282 and using the fact that @xmath283^t$ ] we get the desired result.@xmath99 * * in fact the converse is also true and the proof is similar to that of the previous lemma . therefore @xmath284 where @xmath285 denotes the set of equilibrium points of o.d.e.[o.d.e . ] .
the following lemma says that every point in the set @xmath286 is a partial optimum of the biconvex function @xmath186.*lemma 6.3 : @xmath287 , then , @xmath288 and @xmath289.*proof : if @xmath287 is such that @xmath266 , then by lemma 4.5 the result follows .
assume @xmath268 .
then by lemma 6.2 we have , @xmath277 and @xmath278 . * * by equation [ dd ] , @xmath290 and hence @xmath291 . therefore @xmath292 . by convexity of @xmath293 and
proposition 1.1.8 in @xcite , we get @xmath289 . by equation [ dd ] , @xmath294 and hence @xmath295 .
therefore @xmath296 .
since @xmath297 , we get , @xmath298 . thus by convexity of @xmath299 and by proposition 1.1.8 in @xcite , we have , @xmath300.@xmath99 even though the proof guarantees convergence to the set of partial optimum of the biconvex function in simulation on various test cases it was observed that the iterates converge to the set of nash equilibria of the game @xmath0 .
in the simulations carried out , in order to perform the projection operation in every iteration we use the procedure in @xcite .
we consider the following version of the standard rock - paper - scissor game .
@xmath301 in the above game , @xmath302 is the only nash equilibrium strategy .
having started the algorithm from a random initial point , variation of the objective function value and the strategies are shown in the plots below .
0.47 0.495 the plots in fig:[fig : rps_ap ] show that the action probabilities converge to the nash equilibrium of the game . as the action probabilities converge to nash equilibrium strategy the objective function value approaches zero as seen in fig:[fig : rps_obj ] .
the general form of jordan s game can be found in @xcite .
we consider the following version .
* player 3 action @xmath303 : @xmath304 * player 3 action @xmath305 : @xmath306 in the above game , @xmath307 is the only nash equilibrium strategy . having started the algorithm from a random initial point , variation of the objective function value and the strategies
are shown in the plots in fig:[fig : jg_ap ] and fig:[fig : jg_ap_obj ] .
0.435 0.485 0.435 0.45 simulations were also carried out on other versions of this game obtained from the general form in @xcite and convergence to nash equilibrium was observed .
the following game was introduced in @xcite in order to show non - convergence of certain class of algorithms .
the game is stated below .
@xmath308 in the above game , @xmath309 and @xmath310 are the two nash equilibrium strategies .
having started the algorithm from a random initial point , variation of the objective function value and the strategies are shown in the plots in fig:[fig : hm_ap ] and fig:[fig : hm_obj ] .
0.428 0.428 @xmath311 in the above game , @xmath312 is the set of nash equilbria .
having started the algorithm from a random initial point , variation of the objective function value and the strategies are shown in the plots in fig:[fig : ie_ap ] and fig:[fig : ie_obj ] .
0.46 0.4
we have presented optimization problems ( @xmath203 and @xmath206 ) such that the global minima of these optimization problems are nash equilibria of the game @xmath0 .
the objective functions were shown to be bi - convex and in case of @xmath203 the objective function was also shown to be an invex function .
we also considered a projected gradient descent scheme and proved that it converges to a partial optimum of the objective function . even though the proof gaurantees convergence to the set of partial optimum in various test cases considered we have seen convergence to a nash equilibrium strategy . in future
we wish to extend the above optimization problem formulation to discounted stochastic games and prove convergence to nash equilibrium or construct a counter example where the algorithm converges to a partial optimum which is not a nash equilibrium strategy .
99 von neumann j.and o. morgenstern .
theory of games and economic behaviour , princeton university press .
equilibrium points in n - person games .
proceedings of national academy of sciences , vol 44 , pp 48 - 49 , 1950 .
lemke c. e. and j. t. howson .
equilibrium points of bimatrix games .
siam journal on applied mathematics , vol 12 , pp 413 - 423 , 1964 .
s. govindan and r. wilson . a global newton method to compute nash equilibria .
journal of economic theory , vol 110,issue 1 , pp 65 - 86 , 2003 .
p c. a. floudas and v. vishweswaran .
a global optimization algorithm for certain classes of nonconvex nlps - i .
computers chem .
engng , vol .
1397 - 1417 , 1990 .
j. gorski , f. pfeuffer and k. klamroth .
biconvex sets and optimization with biconvex functions - a survey and extensions .
methods of operations res , vol 66 , issue 3 , pp 373 - 407 , 2007 . v. s. borkar .
stochastic approximations : a dynamical systems viewpoint .
dimitri p. bertsekas .
convex optimization theory .
r. d. mckelvey .
a liapunov function for nash equilibria . social science working paper , california institute of technology , 1998 .
yunmei chen and xiojing ye .
projection onto a simplex .
p. dupuis and a. nagurney .
dynamical systems and variational inequalities .
annals of operations research , vol 44 , pp 7 - 42 , 1993 .
sergiu hart and andreu mas - colell .
uncoupled dynamics do not lead to nash equilibrium .
, vol 93 , pp 1830 - 1836 , 2003 .
sergiu hart and andreu mas - colell .
stochastic uncoupled dynamics and nash equilibrium . games and economic behaviour , vol 57 ,
pp 286 - 303 , 2006 . | in this paper we present optimization problems with biconvex objective function and linear constraints such that the set of global minima of the optimization problems is the same as the set of nash eqilibria of a n - player general - sum normal form game .
we further show that the objective function is an invex function and consider a projected gradient descent algorithm .
we prove that the projected gradient descent scheme converges to a partial optimum of the objective function .
we also present simulation results on certain test cases showing convergence to a nash equilibrium strategy . |
the power spectrum of the cosmic microwave radiation ( cmbr ) carries much cosmological information about primordial density fluctuations in the early universe .
as photons leave the last scattering surface and travel across the universe , however , these brightness fluctuations are modified by intervening structures , causing secondary fluctuations , which we would expect to become more important on small angular scales .
the power spectrum of the cmbr alone can be used to determine cosmological parameters .
recently it has been shown , however , that a geometrical degeneracy effect prevents some combinations of cosmological parameters from being disentangled by the power spectrum alone ( @xcite ; @xcite ; @xcite ) .
the primordial density fluctuations and matter content determine the positions and magnitudes of the doppler peaks at the last scattering surface .
these fluctuations are transferred to apparent angular scales determined by their angular diameter distance . as a result ,
cold dark matter ( cdm ) models with the same primordial density fluctuations , matter content , and angular diameter distance can not be distinguished .
the models are `` effectively degenerate '' in the sense that their power spectrum is degenerate for parameter determination on intermediate and small scales .
although the observed power spectrum also depends on the time variation of the metric , via the integrated sachs - wolfe effect , this breaks the degeneracy only at large angular scales .
unfortunately observations of the power spectrum do not provide strong constraints on models at large scales , since this is where the statistics of the data are dominated by the cosmic variance due to the fact that we have only one realization of our cosmological model , the universe itself , and we encounter a sampling problem ) .
therefore we can determine , for example , only combinations such as @xmath2 and @xmath3 ( where @xmath4 and @xmath5 are the @xmath6 matter and baryon density parameters and @xmath7 is the dimensionless hubble parameter ) .
it has been noticed that gravitational lensing can break the geometric degeneracy at small angular scales , @xmath8 , in such a way that the cosmological parameters can be determined separately ( @xcite ; @xcite ) .
the effect of gravitational lensing on the cmbr was studied by several authors ( see for example @xcite ; @xcite ; @xcite ) .
static gravitational lenses do not change a smooth cmbr , but the fluctuations get distorted by lensing . as a result ,
power from the acoustic peaks is transferred to small angular scales , conserving the variance of the spectrum .
the amount of power transferred depends on the cosmological model , thus , in principle , we can determine separately @xmath4 , @xmath5 , and @xmath7 .
this method makes use of the small angular scale part of the power spectrum , where the amplitude of primordial fluctuations is declining and secondary fluctuations are becoming more important .
the question naturally arises : how do contributions to the power spectrum from secondary fluctuations influence parameter determination based on the small scale cmbr power spectrum ?
the most important secondary fluctuations are the thermal static and kinematic sunyaev - zeldovich ( ssz and ksz ) effects ( @xcite ) , the rees - sciama ( rs ) effect ( @xcite ) , the moving cluster of galaxies ( mcg ) effect ( @xcite ; @xcite ; @xcite ) , point sources ( @xcite ) , and , if the universe was re - ionized at some early stage , the ostriker - vishniac effect ( @xcite ; @xcite ) . in this paper
we concentrate on secondary effects caused by clusters of galaxies . since detailed reviews are available on the sz and the mcg effects ( @xcite ; @xcite ) , here we just summarize their major features .
the thermal sz effect is a change in the cmbr via inverse compton scattering by electrons in the hot atmosphere of an intervening cluster of galaxies .
we use the terms kinematic or static sz effect depending on whether or not the intracluster gas possesses bulk motion . to date
only the static thermal sz effect has been detected ( @xcite ) .
the mcg effect is a special type of rs effect , due to the time - varying gravitational field of a cluster of galaxies as it moves relative to the rest frame of the cmbr . unlike the original rs effect
, the mcg effect is not caused by intrinsic variation of the gravitational field , so that in the rest frame of the cluster , the photons fall into and climb out of the same gravitational field .
however , in the rest frame of the cluster the cmbr is not isotropic , but has a dipole pattern , being brighter in the direction of the cluster peculiar velocity vector .
photons passing the cluster are deflected towards its center .
thus in the direction of the cluster peculiar velocity vector ( ahead of the cluster ) one can see a cooler part of the dipole pattern . towards the tail of the cluster , one can see a brighter part of the dipole ( ahead of the cluster ) . when transferring back to the rest frame of the cmbr , we transfer the dipole out , but the fluctuations remain , showing a bipolar pattern of positive and negative peaks . at cluster center
there is no deflection , thus there is no effect .
the amplitude of the mcg effect is proportional to the product of the gravitational deflection angle and the peculiar velocity of the cluster .
the most important characteristics of the ssz , ksz , and mcg effects in the context of cosmology is that their amplitudes do not depend on the redshift of the clusters causing the effect .
using thermodynamic temperature units , their maximum amplitude are about 500 @xmath9k , 20 @xmath9k , and 10 @xmath9k , respectively .
the ssz and ksz effects have the same spatial dependence as the line of sight optical depth , the mcg effect has a unique bipolar pattern . assuming a king approximation for the total mass and an isothermal beta model for the intracluster gas , the full width at half maximum ( fwhm ) of the ssz and ksz effects @xmath10 , where @xmath11 is the core radius , depending on @xmath12 , which is typically between 2/3 and 1 .
the mcg effect has a much larger spatial extent , with fwhm for each part of the bipolar distribution @xmath13 .
the spectra of the effects are also important : the ssz effect has a unique spectrum which changes sign from negative to positive at about 218 ghz .
the ksz and mcg effects have the same frequency dependence as the primordial fluctuations .
the most important difference between the sz effect and the mcg effect is that the sz effect is caused by intracluster gas , the mcg effect is caused by gravitational lensing by the total mass regardless the physical nature of that mass .
therefore the sz effect only arises from clusters with intracluster gas .
clusters can produce significant mcg effects even if devoid of intracluster gas .
the effects of clusters of galaxies on the cmbr in a given cosmology have been a subject of intensive research since the late 1980s .
there are several different ways of extracting information from these effects .
source counts of the ssz effect were estimated by using the press - schechter mass function ( psmf ) and scaling relations ( @xcite ; @xcite ; 1994 ; @xcite ; @xcite , de luca , desert & puget 1995 , @xcite ; 1997 , @xcite ) .
the importance of the ssz effect was demonstrated and it was shown that thousands of detections are expected with the next generation of satellites .
contributions to the cmbr from the rs and the ksz effects were derived by tuluie , laguna & anninos ( 1996 ) and seljak ( 1996 ) for cdm models with zero cosmological constant .
tuluie et al .
used n - body simulations and a ray - tracing technique , seljak used n - body simulations and second order perturbation theory .
contributions from the ssz and ksz effects originating from large scale mass concentrations ( superclusters ) were studied by persi et al .
bersanelli et al .
( 1996 ) , in their extensive study of the cmbr for the planck mission , estimated the contribution to the power spectrum from the ssz and ksz effects .
aghanim et al .
( 1998 ) estimated the effects of the ksz and mcg effects on the cmbr including their contributions to the cmbr power spectrum .
aghanim et al .
simulated @xmath14 maps with pixel size of @xmath15 .
they used the psmf normalized to x - ray data ( assuming an x - ray luminosity - mass relation ) .
the total mass was assumed to have a navarro - frenk - white profile ( @xcite ) , and the intracluster gas was assumed to follow an isothermal beta model distribution .
the time evolution of the electron temperature and the core radius were assumed to follow models of bartlett & silk ( 1994 ) , which are based on self - similar models of kaiser ( 1986 ) . according to aghanim et al .
( 1998 ) s results , the ksz effect is many orders of magnitude stronger than the primordial cmbr on small angular scales , and therefore the effect would prevent the use of the power spectrum to break the geometric degeneracy .
atrio - barandela & mucket ( 1998 ) estimated the power spectra of the ssz effect in a standard dark matter dominated model with different lower mass cut - offs .
contributions to the power from the ostriker - vishniac effect in cdm models were estimated by jaffe & kamionkowski ( 1998 ) . in this paper
we estimate the contributions to the cmbr power spectrum from the ssz , ksz , and mcg effects on small angular scales adopting cold dark matter dominated models . in our models we assumed scale invariant primordial fluctuations with a processed spectrum having a power law form on cluster scales with a power law index of @xmath16 ( bahcall & fan 1998 ) .
this maybe used as a first approximation as long as contributions from very low and/or very high mass clusters are small ( cf .
our discussion about mass cut offs at section [ s : psmf ] ) .
we use three representative models in our study : _ model 1 _ , open cdm ( ocdm ) model : a low density open model with @xmath17 , @xmath18 , @xmath19 ; _ model 2 _ , flat lambda cdm ( cdm ) model : a low density flat model with @xmath17 , @xmath20 , @xmath21 ; _ model 3 _ , standard cdm ( scdm ) model : a flat model with @xmath22 , @xmath18 , and @xmath23 . in section [
s : method ] we outline our method of estimating the power spectra of secondary fluctuations caused by clusters of galaxies and discuss our normalization method for the psmf .
sections [ s : psmf ] and [ s : physpar ] describe how we used the psmf and the scaling relations to obtain masses and other physical parameters of clusters . in section [ s : powerspectr ]
we present the spherical harmonic expansion of the ssz , ksz and mcg effects , and our method of estimating their power spectra .
section [ s : simulation ] describes our simulations to evaluate the integrals over clusters .
sections [ s : results ] and [ s : discussion ] present our results and discuss the differences from previous work .
we used the press - schechter mass function ( psmf , @xcite ) as a distribution function for cluster masses .
we used @xmath24 as indicated by observations ( bahcall and fan 1998 , and references therein ) .
we used observationally determined cluster abundances as a constraint on the psmf . where necessary , we altered the model parameters resulting from the usual top hat spherical collapse model since that model is only an approximation . in our scdm model
we changed only the overall normalization of the psmf by multiplying it by 0.23 ( a similar normalization was used by @xcite ) .
this procedure is inconsistent with the interpretation of the psmf as a probability distribution ( it does not integrate to unity ) , but we use results for the scdm model only as a comparison to the other two models . in our lambda - cdm model
we multiplied the critical density threshold , @xmath25 , obtained from the spherical collapse model ( equation [ [ e : delta_c ] ] ) , by 1.23 ( which is equivalent to changing the @xmath26 normalization ) and made no other changes .
our ocdm model needed no adjustments . with these changes
all three models agree well with the present day ( @xmath27 ) observed mass spectrum ( figure [ f : psmf_obs ] ; bahcall & cen 1993 ) . for high masses ,
the first two models ( ocdm , and cdm ) also agree with the observed @xmath28 dependence of the high mass cumulative mass function ( figure [ f : psmf_z_obs ] ; bahcall , fan & cen 1997 ) .
the cdm and the scdm models agree with cmbr constraints , while the ocdm model is rejected by these constraints ( @xcite ) . as a cautionary note ,
it is useful to keep it in mind that taking _ all _ data into account none of these models are acceptable .
we assumed that the total mass distribution follows a truncated king profile ( @xcite ) .
for the intra - cluster gas we assumed an isothermal beta model ( cavaliere & fusco - femiano 1976 ) .
isothermal beta model fits to x - ray images of clusters give @xmath29 ( @xcite ) .
determinations of the @xmath12 parameter based on spectroscopy suggest @xmath30 ( @xcite ; @xcite ) .
numerical simulations imply a range for @xmath12 from 1 to about 1.3 ( @xmath31 : @xcite ) ; @xmath32 : @xcite ; @xcite suggest @xmath33 ) .
we follow the precepts of @xcite and adopt @xmath34 .
this choice provides a mass temperature function which is in a good agreement with the observed function ( @xcite ) .
the fitted x - ray spatial distribution is highly dependent on the x - ray structure of the core , and may be expected to be less reliable in the outer regions of clusters .
the sz effect is more sensitive to the outer regions ( the sz effect is proportional to the electron density as opposed to thermal bremsstrahlung , which is proportional to density squared ) . choosing the spectroscopically derived @xmath34 gives a smaller sz effect , and so our choice of @xmath12 should provide a conservative estimate of the contribution of the sz effect to the power spectrum .
a slightly larger @xmath12 would not change our results significantly . although the beta model describes the inner intra - cluster gas well , for more accurate sz work an improved cluster model , which fits the outer regions better , will be needed .
the other physical parameters were determined using the virial theorem , a spherical collapse model , and models of the intra - cluster gas by colafrancesco & vittorio ( 1994 ) .
we assumed a maxwellian distribution for the peculiar velocities , @xmath35 , and used results of n - body simulations by gramann et al .
( 1995 ) to normalize the distribution .
we took velocity bias into account , and assumed that the peculiar velocities are isotropically distributed .
we used analytical approximations to calculate the contributions of the ssz , ksz and mcg effects to the cmbr power spectrum .
these contributions are important only at small angular scales , where we can neglect the overlap between cluster images and ignore the weak cluster - cluster correlation , and therefore we can approximate the resulting power spectrum by summing the contributions from individual clusters ( similar methods were used by @xcite ; @xcite ; and @xcite ) .
we expanded the ssz , ksz , and mcg effects as laplace series ( i.e. , series of spherical harmonics ) , then determined the individual cluster contributions and summed over the clusters ( for a detailed description see @xcite ) .
our approximation breaks down at large angular scales , but at these scales primordial fluctuations dominate the cmbr , and the cluster contribution is only a minor perturbation , so that only a rough indication of the cluster effect is needed .
according to the press - schechter method , the co - moving number density of clusters of total mass @xmath36 at redshift @xmath28 ( the psmf ) is given by @xmath37 where @xmath4 is the matter density today in units of the critical density , @xmath38 is the current critical density of the universe ( we adopt a dimensionless hubble parameter @xmath39 in our work ) ; @xmath40 , where the present mass variance for a power law power spectrum , @xmath41 , is @xmath42 @xmath43 , @xmath44 is the mass within an @xmath45 mpc sphere , and @xmath26 is the normalization ( lacey & cole 1993 , press & schechter 1974 ) .
the over - density threshold linearly extrapolated to the present may be expressed as ( @xcite ; @xcite ) @xmath46 & $ \omega_0 < 1 $ , $ \lambda = 0 $ \cr 0.15 ( 12 \pi)^{2/3 } \omega_m^{0.0055 } d_\lambda(\omega_0 , 0)/d_\lambda(\omega_0 , z ) & $ \omega_0 < 1 $ , $ \lambda = 1 - \omega_0 $ \cr } , \ ] ] where the conformal time for open models is @xmath47 .\ ] ] for open models with cosmological constant @xmath18 the linear growth factor is ( peebles 1980 ) @xmath48 where @xmath49 , and the density parameter @xmath50 is @xmath51 for models with
a non - zero cosmological constant the integral can not be done analytically , thus we use an approximation ( lahav et al 1991 ; caroll , press & turner 1992 ) to obtain the equivalent expression to ( [ e : dz_lam0 ] ) , @xmath52 \biggl [ 1 + { \omega_\lambda(z ) \over 70 } \biggr ] \biggr\}^{-1 } .\ ] ] the density parameters , @xmath53 and @xmath54 , for spatially flat universes with @xmath55 , are @xmath56 the normalization of these growth functions is chosen so that at high redshifts they approximately match the time variation of density contrast in an einstein - de sitter ( @xmath22 ) universe , which is a good approximation to the early universe whatever its density parameter today .
the total number of clusters at redshift @xmath28 ( in a redshift interval of @xmath57 ) is @xmath58 where @xmath59 and @xmath60 are the lower and upper mass cut offs for clusters .
we used @xmath61 for the ssz and ksz effects , @xmath62 for the mcg effect , and @xmath63 for all effects .
the lower cut off , @xmath59 , for sz effects signifies the lowest cluster mass for which we expect a well - developed intracluster atmosphere . in the case of the mcg effect , @xmath59 is the mass limit from which we consider a mass concentration as a cluster ( `` formation '' ) .
we found that low mass clusters do not contribute substantially to the power spectrum , thus the lower cut off , @xmath59 , has little effect on our results .
the upper cut off , @xmath60 , has no effect on our results ( as long as it is large enough ) : the probability of getting such a large cluster is negligible , so that the contribution from more massive clusters is negligible .
we assumed a truncated king profile for the total mass distribution @xmath64 where @xmath65 is the core radius , @xmath66 is the cut off , and an isothermal @xmath12 model for the intra - cluster gas @xmath67 where @xmath68 and @xmath69 are the electron number density at radius @xmath70 and at the center of the cluster ( cavaliere & fusco - femiano 1976 ) .
analytical studies and numerical simulations show that the gas density profile scales with the total density , and that the gas central electron density may be expressed as @xmath71 where @xmath72 is the average hydrogen mass fraction , @xmath73 is the intra - cluster gas mass fraction .
@xmath74 , the central mass density , is determined from the total mass by integrating equation ( [ e : king_ro ] ) .
little is known about the total mass and redshift dependence of the intra - cluster gas from observations . here , we adopt colafrancesco & vittorio ( 1994 ) s model which assumes that changes in the intra - cluster gas are driven by entropy variation and/or shock compression and heating . according to their model , the gas mass fraction may be expressed as @xmath75 where the normalization , @xmath76 , is based on local rich clusters , and we used @xmath77 and @xmath78 , which are consistent with available data .
using the virial radius to express the core radius , @xmath79 , and assuming spherical collapse , we obtain @xmath80^{1/3 } { 1 \over 1 + z } , \ ] ] where @xmath81 is the overdensity of the cluster relative to the background ( @xcite ) . for @xmath18 models ( our ocdm and scdm models )
the over density may be expressed as @xmath82 , \ ] ] where @xmath83 ( oukbir & blanchard 1997 ) . for spatially flat models with finite cosmological constant ( our cdm model ) we have @xmath84 , \ ] ] where we used the approximation of kitayama & suto ( 1996 ) .
numerical models of cluster formation show that cluster temperature scales with total mass . using the virial theorem and assuming spherical collapse with a recent - formation approximation in a standard cdm model , the electron temperature , @xmath85 , becomes @xmath86 where @xmath87 is the density contrast of a spherical top - hat perturbation relative to the background density just after virialization ( cf .
for example eke , cole & frenk 1996 ) .
the recent - formation approximation , however is valid only for @xmath22 . for our low matter density open model ( _ model 1 _ , ocdm )
, we use a model which takes into account accretion during the evolution of clusters , and leads to the following scaling : @xmath88 \ , { \rm kev } .\ ] ] this was derived for open models with zero cosmological constant ( @xcite ) , but since structure formation evolves similarly in a low density model with the same matter density and a zero cosmological constant , we use it for our _ model 2 _ ( cdm ) as an approximation .
we assumed a maxwellian for the cluster peculiar velocity distribution , @xmath35 , as expected from a gaussian initial density field : @xmath89 where @xmath90 is the maxwellian width of the peculiar velocity distribution .
the rms peculiar velocity from linear theory , smoothed with a top - hat window function of radius @xmath91 , @xmath92 , is given by @xmath93 where @xmath94 is the scale factor , and the moments , @xmath95 , are defined as @xmath96 where @xmath97 is the fourier transform of the power spectrum and equation ( [ e : pecvel1 ] ) uses the moment @xmath98 ( @xcite ) .
the velocity factor , @xmath99 , can be approximated as ( @xcite ; 1984 ) @xmath100 & $ \lambda = 1 - \omega_0 $ \cr } .\ ] ] the cluster peculiar velocity rms differs from this since we assume that clusters form at the peaks of the density distribution , and with this bias may be expressed as @xmath101\ ] ] ( bardeen et al 1986 ) .
colberg et al .
( 1998 ) calculated the velocity bias in a series of cdm models using a top - hat filter and processed cdm power spectra .
the correction factor has a weak dependence on @xmath4 : it is about 0.8 for low density and flat cdm models .
we obtain the maxwellian width in equation ( [ e : p_v_pec ] ) from the rms peculiar velocity from averaging a maxwellian : @xmath102 we expressed @xmath103 as @xmath104/ [ h(0 ) a(0 ) f(0 ) ] $ ] .
the normalization at @xmath6 was determined by using results on the peculiar velocity distribution from numerical simulations ( @xcite ) .
thus we obtain the following expression for the maxwellian width of the peculiar velocities , @xmath105 , with velocity bias for models with no cosmological constant ( ocdm , scdm , @xmath18 ) : @xmath106 for our cdm model ( @xmath107 ) we obtain @xmath108 , \\ & & \nonumber\end{aligned}\ ] ] this normalization is significantly larger than some recent measurements suggest ( bahcall & oh 1996 ) , but it is a good match to others ( gramann 1998 ) .
this uncertainty should be remembered when interpreting our final results .
ignoring the correlation between clusters , the power spectrum becomes @xmath109 where @xmath110 is the contribution from clusters with total mass @xmath36 at redshift @xmath28 , and @xmath111 denotes the ssz , ksz or mcg effects .
@xmath112 is the differential volume element ( assumed isotropy ) @xmath113^{-1/2 } , \ ] ] where the effective distance @xmath114 is @xmath115 & $ \lambda = 0$\cr { c \over h_0 } \int_0^z dx \bigl[\omega_0 ( 1 + x)^3 + 1 - \omega_0 \bigr]^{-1/2 } & $ \lambda = 1 - \omega_0$}\ ] ] ( peebles 1993 ) . in general , the coefficients @xmath116 may be determined by calculating the spherical harmonic expansion of the cluster image by averaging out the azimuthal parameter , @xmath117 , @xmath118 our task is to determine the @xmath119 coefficients .
the ssz and ksz effects are cylindrically symmetric for spherical clusters , therefore we may describe them using only one coordinate , the angular distance from the cluster center .
we separate the effects into amplitudes and geometrical form factors which carry their spatial dependence .
the ssz and ksz effects in thermodynamic temperature units may be expressed as @xmath120 where the central effects for the ssz and ksz effects are @xmath121 \ , \theta \
, \tau_0 , \ ] ] and @xmath122 \ , \beta^2 - \bigl[1 + \theta c_1(x ) + \theta^2 c_2(x ) \bigr]\ , \beta \ , p_1(\alpha ) \\ & + & \bigl [ d_0(x ) + \theta d_1(x ) \bigr ] \ , \beta^2 \ , p_2(\alpha ) \biggr\}\ , \tau_0 \nonumber .\end{aligned}\ ] ] in these expressions @xmath123 is the legendre polynomial of order of @xmath124 , @xmath125 is the dimensionless frequency , @xmath126 is the dimensionless temperature , @xmath127 is the angle between the cluster s peculiar velocity vector and its position vector , @xmath7 , @xmath128 , @xmath129 and @xmath130 are the planck constant , frequency , boltzmann constant , and temperature of the cmbr , @xmath131 k ( @xcite ) , and the lengthy expressions for the spectral functions @xmath132 , @xmath133 and @xmath134 may be found in nozawa , itoh & kohyama ( 1998 ) .
these functions arise from an expansion of the boltzmann equation and although they are inaccurate for high temperature clusters , their precision is sufficient for our purposes here ( for a discussion see @xcite and references therein ) . the optical depth through the cluster center for gas model ( [ e : n_el ] ) is @xmath135 and the geometrical form factor is @xmath136 where the function @xmath137 is defined as @xmath138 @xmath139 holds in the small angle approximation , and the integral , @xmath140 , is @xmath141 where @xmath142 is the gamma function , @xmath143 is gauss hyper - geometric function , and @xmath127 must be greater than 1/2 ( @xcite ) .
the geometrical form factor is normalized to one at the cluster center ( @xmath144 ) .
we may expand the ssz and ksz effects in legendre series as @xmath145 where @xmath146
are legendre coefficients of @xmath147 , and @xmath148 refers to the ssz or the ksz effect .
we determine the legendre coefficients using a small angle approximation , as @xmath149 \ , \theta \ , d \theta , \ ] ] where we used the approximation @xmath150 , \ ] ] where @xmath151 is a bessel function of the first kind and zero order ( @xcite ) .
we can convert legendre coefficients to laplace coefficients by expressing the laplace series of such a function as @xmath152 where @xmath153 , and @xmath154 and @xmath155 are the spherical harmonics and legendre polynomials .
therefore the conversion can be done as @xmath156 thus the laplace series of the ssz and ksz effects become @xmath157 using equation ( [ e : delta_sz ] ) , the contribution of one cluster to the power spectrum of the ssz and ksz effects becomes @xmath158 where @xmath159 and @xmath146 are given by equations ( [ e : delta_i_s_r ] ) , ( [ e : delta_i_k_r ] ) , and ( [ e : zeta_ell ] ) .
similarly , the mcg effect may be expressed as @xmath160 where @xmath161 is the geometrical form factor .
@xmath162 is the angle of the line of sight relative to the center of the cluster .
the azimuthal angle , @xmath163 , is measured in the plane of the sky from the direction of the tangential component of the peculiar velocity .
the maximum of the mcg effect is @xmath164 where @xmath165 is the maximum deflection angle , @xmath127 is the angle between the cluster s peculiar velocity vector , @xmath166 , and its position vector , and @xmath167 is the speed of light in vacuum .
for a spherically symmetric thin lens the deflection angle is given by @xmath168 where @xmath169 is the impact parameter at the source , and @xmath170 is the mass enclosed by a cylindrical volume with axis parallel to the line of sight and radius equal to the impact parameter @xmath169 ( cf .
for example schneider , ehlers & falco 1992 ) . using the king approximation for the density distribution ( equation [ [ e : king_ro ] ] ) ,
the total mass in cylindrical coordinates , @xmath171 , becomes @xmath172 where @xmath173 in the small angle approximation . a straightforward integration and equation ( [ e : delta_mcg0 ] ) lead to @xmath174 where the function @xmath175 is @xmath176 .\ ] ]
thus the geometrical form factor in our case becomes @xmath177 where @xmath178 is the maximum value of the function @xmath179 .
the mcg effect depends only on @xmath180 , therefore we need to determine only the @xmath181 terms in the spherical harmonic expansion . expressing the spherical harmonics by associated legendre polynomials , equation ( [ e : xi_g ] ) expands as @xmath182 , \ ] ] where we used the identity @xmath183 in order to obtain a real function , the imaginary terms must vanish , therefore we must have @xmath184 using orthogonality , expressing the spherical harmonics in terms of associated legendre polynomials and using equations ( [ e : xi_ell ] ) and ( [ e : xi_thetaphi ] ) , we obtain @xmath185 where @xmath186 the @xmath163 integral can be performed since @xmath175 and @xmath187 do not depend on @xmath163 giving @xmath188 from which we find @xmath189 \ , d \theta .\ ] ] here we used the small angle approximation for the associated legendre polynomials : @xmath190 , \ ] ] where @xmath191 is the bessel function of the first kind and order 1 ( for a derivation see appendix ) .
thus the laplace series of the mcg effect is @xmath192 where @xmath193 is given by equation ( [ e : xi_ell_1_j1 ] ) . for the power spectrum of the mcg effect , using equation ( [ e : delta_mcg_y ] ) , we obtain @xmath194 the observed effects are calculated by convolving the theoretical fluctuation pattern with the telescope s point spread function ( psf ) .
one advantage of using the spherical harmonic coefficients is that this convolution is just a multiplication in spherical harmonic space .
assuming an axially symmetric psf , @xmath91 , its legendre polynomial expansion may be expressed as @xmath195 where the unit vectors , @xmath196 and @xmath197 , point to an arbitrary direction ( where we want to evaluate the expansion ) and to the center of the psf . assuming a non - axially symmetric effect , its spherical harmonic expansion can be written as @xmath198 where @xmath199 runs from zero to infinity and @xmath117 runs from -@xmath199 to @xmath199 . using the addition theorem for spherical harmonics and their orthogonality ,
the convolution of these two functions , @xmath200 , becomes @xmath201
we used monte carlo simulations to generate an ensemble of clusters of galaxies with masses sampled from the psmf ( equation [ e : psmf ] ) with parameters those of our ocdm , cdm and scdm models .
we obtained the central electron number density and temperature , and the cluster core radius from scaling relations ( equations [ e : n_el_0 ] , [ e : t_eom1 ] , [ e : t_eopen ] , and [ e : r_cor ] ) .
we choose to sample the psmf using a rejection method .
the magnitude of the peculiar velocity may be sampled using an inversion method on the maxwellian ( [ e : p_v_pec ] ) , and yields @xmath202 where @xmath203 is the inverse of the incomplete gamma function and @xmath103 can be determined by using equations ( [ e : sigm_v_pec ] ) , ( [ e : sigma_p_l0 ] ) and ( [ e : sigma_p_lam ] ) .
@xmath204 is a uniformly distributed random number in ( 0,1 ) .
we assumed an isotropic distribution in space for the directions of the peculiar velocity vectors , and ignored correlations between cluster peculiar velocities .
the tangential and radial peculiar velocities are distributed as projections of equation ( [ e : v_p_rn ] ) accordingly . as an illustration , in figure
[ f : mcg1 ] we show results from one simulation using our scdm model projected on a grid .
the observational mass function ( figure [ f : psmf_obs ] ) is specified by @xmath205 , the mass contained within co - moving radius of 1.5 mpc . to convert @xmath205 to the virial mass , @xmath206 , which we use in the psmf , we assume that the mass profile near 1.5 mpc can be approximated with @xmath207 .
we obtain @xmath208 or , substituting the numerical values , @xmath209 we used q = 0.64 ( @xcite ) to obtain curves shown in figure [ f : psmf_z_obs ] .
the power spectrum for an ensemble of clusters may be determined by summing the individual contributions of the simulated clusters ( equation [ [ e : c_ell_int ] ] ) .
we binned clusters by their apparent core radii , @xmath210 , then we summed the amplitudes in each bin .
the numerical evaluation of integral ( [ e : c_ell_int ] ) , in this case , can be performed as @xmath211 and @xmath212 where the index @xmath213 runs over clusters whose core radii fall within the @xmath214 bin .
our results for the power spectra ( more exactly the dimensionless @xmath215 ) from our _ model 1 _ ( ocdm ) are shown on figure [ f : powerspocdm ] .
figures [ f : powersplcdm ] and [ f : powerspscdm ] show our results for _ model 2 _ , ( cdm ) and _ model 3 _ , ( scdm ) respectively . as a comparison , in each figure , we plot the corresponding primordial cmbr power spectrum ( solid line ) with @xmath216 normalization including the effects of gravitational lensing calculated by using a new version of cmbfast ( @xcite ; @xcite ) . on large angular scales (
up to about @xmath217 ) the cosmic variance dominates ( not shown ) . on
small angular scales the shape of the power spectra depends on the apparent angular sizes of the clusters and the amplitudes of the effects .
the apparent angular size depends on how the core radius and the angular diameter distance change with redshift , while the amplitude is sensitive to the gas content , gas temperature and total mass as a function of redshift .
figures [ f : powerspocdm]-[f : powerspscdm ] demonstrate that for small angular scales ( @xmath0 ) the contribution to the power spectrum from the ssz effect exceeds that of the primordial cmbr in all our models .
the contributions of the ksz and mcg effects become important only on small scales , but , at those scales , they may dominate over the lensed primordial fluctuations . due to the early structure formation
, there are more clusters at high redshift in our ocdm and cdm than in our scdm simulations .
therefore the contribution to the power spectrum from clusters in a low matter density model is substantially larger than in a scdm model . also , most clusters are closer to us in a scdm model , thus the contribution from clusters to the power spectrum peaks at higher angular scales ( lower @xmath199 ) than in low matter density models .
the ksz and mcg effects have their coherence length ( peak contributions ) at @xmath218 .
the coherence length of the ssz effect is about @xmath219 for our scdm model and at about @xmath220 for our ocdm and cdm models . in general , the contributions to the power spectrum from the ssz effect are about 2 and 3 orders of magnitude greater than those from the ksz and mcg effects . at very small angular scales , @xmath221 , the contribution to the power spectrum from the mcg effect might exceed that of the ssz or ksz effects , and even the primordial fluctuations in the cmbr , but this depends on the details of the evolution of cluster atmospheres .
our simulations give somewhat different results for the ksz and mcg effects than those of aghanim et al .
( 1998 ) . in our simulations
the amplitudes of the ksz and mcg effects for low matter density models are about an order of magnitude greater than those for our scdm model , and have a coherence length of about @xmath222 , while rising monotonically at smaller @xmath199 . according to aghanim et al.s simulations , with similar cut off to ours , @xmath223 , contributions from the ksz effect in all models constantly grow and show no signs of leveling off , and their amplitude has a very weak dependence on cosmological model .
contributions from the mcg effect on the other hand show a plateau in all aghanim et al.s models for @xmath224 , and for the scdm model , the mcg effect is larger than for the other two models .
quantitatively , our models show cluster - related effects that are weaker by a factor of 10 for the mcg effect in our scdm model and a factor of 100 for the ksz effect for all models .
we attribute these differences mostly to the different evolution models for the intracluster gas .
the ratio between the overall amplitudes of the ksz and mcg effects in our calculations is about the same as in aghanim et al .
( 1998 ) s results .
our results show that the power spectra of the ksz and mcg effects do not exceed the gravitationally lensed primordial power spectrum up to @xmath225 , while the power spectrum of the ssz effect becomes dominant at @xmath226 in all our models .
contributions to the power spectrum from the ssz and ksz effects based on aghanim et al.s model would exceed those from the cmbr at @xmath227 even if one takes gravitational lensing of the primordial cmbr into account .
our results are comparable to those obtained by tuluie et al .
( 1996 ) and seljak ( 1996 ) .
persi et al .
( 1995 ) s results for contributions to the power spectrum from the ssz ( ksz ) effect are about the same as ( an order of magnitude higher than ) our results suggest .
an observed power spectrum is made up from the sum of all astrophysical effects and noise .
we rely on the different frequency and/or power spectra of the secondary effects to separate these foregrounds from the primordial cmbr signal ( @xcite ) . of the effects discussed here , it should be easy to separate the ssz effect by using multi - frequency measurements of its unique spectrum .
the separation of primordial fluctuations in the cmbr and fluctuations caused by the ksz and mcg effects is more difficult since their frequency spectra are the same .
optimal filters have been designed to separate the ksz effect ( @xcite ; @xcite ) : here it helps to know the ssz effect for the same cluster , since that would give us a position and even an estimate for the expected amplitude of the effect .
aghanim et al .
( 1998 ) discussed methods to separate the mcg effect : this is facilitated by its unique dipole pattern with sharp peaks ( figure [ f : mcg1 ] ) .
primordial fluctuations are usually assumed to be gaussian , where the probability of getting such a strongly peaked bipolar pattern is small , and we would expect the strong small angular scale gradient near a known cluster of galaxies to be a definite indication of the presence of the mcg effect .
also , knowing the position of clusters helps to find the effect .
however , contributions from other effects , such as early ionization and discrete radio sources causes further confusion , and may be expected to make it difficult to determine the power from the sz or mcg effects .
we analyzed the contributions to the power spectrum from the ssz , ksz and mcg effects to check their impact on the determination of cosmological parameters , especially at large @xmath199 where gravitational lensing may break the geometric degeneracy . in figure
[ f : lens01 ] we show the small scale lensed primordial fluctuation power spectra of our three models ( ocdm , scdm , cdm ; solid lines ) with power spectra resulting from the sum of fluctuations due to the lensed primordial cmbr and the ssz effect ( long dashed lines ) , and from the sum of the lensed primordial cmbr , the ksz and the mcg effects ( short dashed lines ) . according to our models , if the fluctuations due to the ssz effect are fully separated , the ksz and mcg effects do not prevent the use of this part of the power spectrum to break the geometric degeneracy and distinguish between different cdm models . note that normalization at the first doppler peak , rather than the usual
_ cobe _ normalization , would lower the contributions to the power spectrum from primordial fluctuations in a cdm model relative to those from a scdm model , and thus secondary effects would become more important relative to the primordial cmbr fluctuations .
our simulations also show that the power spectrum of the ssz effect may itself be used to break the geometric degeneracy .
since the separation of the ssz effect from other secondary effects should be straightforward , we should be able to determine the power spectrum of the ssz effect alone .
as can be seen from figure [ f : powersp3 ] , this power spectrum depends on @xmath2 and @xmath3 , providing an additional constraint on these parameters .
we note , however , that the amplitude of the ssz effect is model dependent .
since the contributions to the power spectrum from the sz and mcg effects are model dependent , to evaluate fully their power spectra we need a better observationally - supported model for the intracluster gas .
sensitive , high - resolution all - sky , x - ray observations could map the emission from intracluster gas up to high redshift providing strong constraints on gas formation and evolution and thus a good basis for modeling the ssz and ksz effects ( @xcite ) .
number counts of clusters based on their ssz effect can also be used to constrain cosmological models ( @xcite ) .
there are many possibilities of using observations to break the geometric degeneracy .
for example measurements of the cmbr polarization , the hubble constant , or light curves of type ia supernovae have been discussed ( zaldarriaga et al . 1997 ; @xcite ; @xcite ) .
also , combination of measurements of the ssz effect and thermal bremsstrahlung ( x - ray ) emission from clusters can be used to determine the hubble constant for a large number of clusters , providing a statistical sample which might enable us to determine the hubble constant , and perhaps the acceleration parameter , to good accuracy ( @xcite ) .
secondary fluctuations introduce non - gaussianity into the primordial spectrum at small scales .
this non - gaussianity should be taken into account when estimating cmbr non - gaussianity at these scales .
winitzky ( 1998 ) estimated the effect of lensing and concluded that planck may observe non - gaussianity due to lensing near the angular scale of maximum effect , @xmath228 .
other processes , including the ssz , ksz , and especially the mcg effect , introduce a highly non - gaussian signal as is easily seen for the mcg effect on figure [ f : mcg1 ] .
a similar non - gaussian pattern arises from moving cosmic strings ( the kaiser - stebbins effect , compare our figure [ f : mcg1 ] to figure 6a of @xcite ) .
our results indicate that at @xmath229 the mcg effect might be comparable in strength to the primordial fluctuations .
evidence for non - gaussianity has been reported by ferreira , magueijo & gorski ( 1998 ) and gaztanaga , fosalba , & elizalde ( 1997 ) at angular scales @xmath230 and @xmath231 .
they do not exclude the possibility that this non - gaussianity has been introduced by foregrounds , but our results show that clusters can not introduce detectable non - gaussianity on such scales ( figure [ f : powersp3 ] ) .
we convolved our theoretical results ( figures [ f : powerspocdm ] - [ f : powerspscdm ] ) with the expected point spread functions ( psfs ) of instruments on the map and planck missions to estimate the level of the secondary fluctuations caused by clusters of galaxies on the observable power spectrum .
the observed @xmath232 values become in general , contributions from unresolved cluster static effects add to provide a cumulative contribution to the cmbr power spectrum .
contributions from the ksz and mcg effects from unresolved sources tend to cancel . in the case of the mcg effect
this is because each unresolved source contribution would be zero owing to the dipole spatial pattern of the effect . for small - scale ksz effects
there are several sources in the field of view of the telescope , and different sources have positive or negative contributions depending on the sign of their line of sight peculiar velocity , and therefore they tend to cancel each other .
the larger the beam size , the more effective is the cancellation of the mcg and ksz effects .
note that the spatial extension of the mcg effect is much larger than that of the ksz effect , so many clusters may be unresolved in their ksz and resolved in their mcg effect .
the mcg effect might be relatively more important at high redshifts , since it does not require a well - developed cluster atmosphere . in figures
[ f : map94 ] and [ f : planck353 ] we show the contributions to the power spectrum from primordial fluctuations , and the ssz , ksz and mcg effects , convolved with the psf of the planned @xmath238 ghz receiver on map , and the planned @xmath239 ghz bolometer on planck .
the amplitude of the fluctuations from the the ssz effect is negative at @xmath238 ghz and positive at @xmath239 ghz , but only the the absolute value of the effect contributes to the power spectrum .
the different maximum @xmath199 values for the map and planck systems ( @xmath240 1000 and 2000 , respectively ) can clearly be seen on figures [ f : map94 ] and [ f : planck353 ] .
because of these cutoffs , the observable power spectrum is dominated by primordial fluctuations at all @xmath199 for these missions . according to our results , the ssz effect may cause a 1% enhancement in the amplitude of the doppler peaks , which is at the limit of the sensitivity of the map and planck missions . from the analysis of the power spectrum ,
this would lead to an overestimation of the parameter @xmath2 by about 1% , .
the shift in the position of peaks as a function of @xmath199 caused by the ssz effect is less important since the spectrum of the ssz effect has only a weak dependence on @xmath199 . in table
[ t : rms_all ] we show the @xmath241 values of the contributions to the cmbr from the ssz , ksz , and mcg effects convolved with the the 94 ghz map and 353 ghz planck receivers for our three models ( ocdm ; cdm ; scdm ) . as a comparison
, we display the corresponding rms values of the primordial fluctuations .
the rms values of all these secondary effects are an order of magnitude smaller than rms values from primordial fluctuations .
the most important contribution is that of the ssz effect at these frequencies .
the ksz and mcg effects give similar contributions with the ksz effect being about a factor of two stronger .
aghanim et al .
( 1998 ) s results for the rms values of the mcg effect is a factor of 10 ( scdm ) or a factor of 3 ( ocdm and cdm ) larger than our results .
note , however , that rms values give only a crude estimate of the magnitude of the effects . at large angular scales the primordial fluctuations are about 100 ( for the ssz effect ) or @xmath242 ( ksz , mcg effects ) times stronger than the secondary fluctuations .
an ideal observation to measure the contribution to the power spectrum from the ssz effect would use high angular resolution ( @xmath226 ) and high frequency ( @xmath243 ghz ) .
suzie probes the power spectrum at angular scale @xmath244 at 140 ghz .
the 2@xmath245 upper limit on the power at this scale from suzie is @xmath246 ( @xcite ) .
unfortunately our models suggest that at this frequency the primordial contribution to the power spectrum is about 10 times stronger than that from the ssz effect .
a promising experiment is scuba , which probes the anisotropies at angular scale @xmath225 and frequency 348.4 ghz .
their preliminary 2@xmath245 upper limit on the power spectrum at this scale is @xmath247 .
much further work is planned , and should lower this limit by a factor of 3 - 10 ( @xcite ) .
smm is grateful to bristol university for a full scholarship , where most of this work was done .
this work was finished while smm held a national research council research associateship at nasa goddard space flight center .
we thank n. aghanim for comments on an earlier version of the manuscript , and our referee , dr bartlett , for his detailed comments and for helping us to clarify some aspects of our approximations .
we thank u. seljak and m. zaldarriaga for making the cmbfast code available .
we derive a small angle approximation for the associated legendre polynomial @xmath187 .
we express the associated legendre polynomial @xmath187 in terms of the legendre polynomial @xmath155 ( see for example @xcite ) as cccccccccc + model & & map & 94 ghz & & & & planck & 353 ghz & + + + & cmbr & ssz & pksz & mcg & & cmbr & ssz & pksz & mcg + & @xmath9k & @xmath9k & @xmath9k & @xmath9k & & @xmath9k & @xmath9k & @xmath9k & @xmath9k + + + 1 .
ocdm & 86 & 7.0 & 0.54 & 0.20 & & 100 & 14 & 0.81 & 0.33 + 2 .
lcdm & 119 & 5.9 & 0.49 & 0.20 & & 130 & 11 & 0.71 & 0.32 + 3 .
scdm & 93 & 3.2 & 0.36 & 0.12 & & 101 & 5.7 & 0.50 & 0.17 + | we estimate the contributions to the cosmic microwave background radiation ( cmbr ) power spectrum from the static and kinematic sunyaev - zeldovich ( sz ) effects , and from the moving cluster of galaxies ( mcg ) effect .
we conclude , in agreement with other studies , that at sufficiently small scales secondary fluctuations caused by clusters provide important contributions to the cmbr . at @xmath0 ,
these secondary fluctuations become important relative to lensed primordial fluctuations .
gravitational lensing at small angular scales has been proposed as a way to break the `` geometric degeneracy '' in determining fundamental cosmological parameters .
we show that this method requires the separation of the static sz effect , but the kinematic sz effect and the mcg effect are less important .
the power spectrum of secondary fluctuations caused by clusters of galaxies , if separated from the spectrum of lensed primordial fluctuations , might provide an independent constraint on several important cosmological parameters .
@xmath1 subject headings : cosmic microwave background galaxies : clusters : general methods : numerical |
bose - einstein condensates form one of the main topics of research at the moment .
one reason for this enormous interest is the fact that they combine concepts and techniques from different areas of physics , such as quantum optics , condensed matter physics , molecular physics and quantum chaos . on the experimental side ,
there has been a remarkable progress in confining and manipulating bose - einstein condensates @xcite , which has stimulated the theoretical research in the area .
there has been a large number of previous studies analyzing the dynamics of bose - einstein condensates in a double well potential using a mean - field approach , the gross - pitaevskii equation . in @xcite smerzi _
et al . _
discussed the occurrence of macroscopic self - trapping within one well .
the behavior of the system during an adiabatic change of parameters was studied in @xcite and generalizations of the linear two level crossing scenarios and the landau - zener formula were analyzed .
another line of investigation considers the mesoscopic regime in which the quantum and the classical , i.e. mean - field descriptions overlap and therefore semiclassical techniques can be used to study the system @xcite . in this paper
we present a semiclassical technique to analyze the spectral properties of a bose - einstein condensate .
this method has already been used to describe the vibrational spectra of molecules @xcite and provides an intuitive picture for the at first sight uninterpretable spectra . in this sense , the dynamics of a bose - einstein condensate in a triple well potential is analogous to the vibrations of a tri - atomic molecule . using the geometrical language of classical mechanics to describe the quantum system ,
we introduce a simple semiclassical method of visualization and classification of the quantum eigenstates which allows a characterization of the dynamics of the system .
furthermore , we investigate how far the correspondence between the mean - field system and the quantum many body system can be extended when the number of particles decreases . for our studies
we consider a bose - einstein condensate in a triple well potential , since the technique easily allows us to go beyond the standard double well potential analysis .
a triple well potential has a much richer structure @xcite and the power of the method can be shown without loss of clarity , still allowing a direct visualization of all relevant structures .
in the following analysis we consider a system consisting of bosonic particles in an external periodic potential @xmath0 with @xmath1 , @xmath2 and @xmath3 .
if a weak two - particle point - like interaction is assumed , then the hamiltonian in second quantization can be written as @xmath4{{\hat{{{\phi}}}}}(\vec{r } ) \nonumber \\ & + \frac{g}{2 } \int d^3r\ ; { { \hat{{{\phi}}}}}^\dagger(\vec{r})\,{{\hat{{{\phi}}}}}^\dagger(\vec{r})\,{{\hat{{{\phi}}}}}(\vec{r})\,{{\hat{{{\phi}}}}}(\vec{r})\,.\end{aligned}\ ] ] here , @xmath5 is the particle mass , @xmath6 is the coupling constant describing two - body interactions and @xmath7 is the s - wave scattering length . for a repulsive interaction ,
@xmath8 is positive while for an attractive interaction @xmath8 takes a negative value . for the rest of the paper we choose scaled units with @xmath9 .
the field operator @xmath10 can be expanded in terms of bosonic annihilation operators , @xmath11 where we assume that the basis functions @xmath12 of the one - particle hilbert space are exponentially localized in space and real , as is the case for the wannier functions @xcite .
the index @xmath13 describes basis functions in different wells and we will take into account only three different wells in order to model the three well potential .
the second index @xmath5 labels the excited states within a single well . assuming bose - einstein condensates
, we can restrict ourselves to the lowest energy state @xmath14 and neglect higher excited states ( see also @xcite for a careful discussion of this topic for a two well potential ) . experimentally such a system was realized in @xcite for a two well potential but the technique can in principle also be extended to three wells . expanding the hamiltonian in this basis and neglecting fourth order terms in the creation and annihilation operators from different basis functions ( modes ) yields the well - known bose - hubbard hamiltonian @xcite restricted to three wells .
so , the hamiltonian can be written in a symmetrized form as @xmath15 with @xmath16 here we neglect a constant energy shift . for convenience
, we will choose the nonlinear interaction strengths @xmath17 equal for each well in the following sections which is also in accordance with experimental realizations .
such hamiltonians have already been studied in great detail for the more restrictive two mode model ( e.g. in @xcite ) . the hamiltonian commutes with the particle number operator @xmath18 which expresses the conservation of the total number of particles .
the symmetrized form is more convenient when considering the semiclassical limit , as will become clear in the next paragraph .
hamiltonians of this kind have been used in molecular physics in order to describe and assign vibrational spectra @xcite . in the molecular case
they describe all kinds of vibrational degrees of freedom like stretches , bends , torsions etc . and include various resonant interactions corresponding to different simple rational ratios between the frequencies . the conserved particle number in our case of eq .
( [ eq - hqm ] ) corresponds to the polyad - type conserved quantities in the molecular systems . for the case of 30 particles considered in the following
, it is an easy numerical task to diagonalize the hamiltonian matrix and thus solve the problem .
however , one can not understand the underlying structure of this system from numerical values alone .
the aim of this paper is to present a method which allows an easy visual characterization of the eigenvectors of the hamiltonian , using the close correspondence with the classical system .
essential for our semiclassical classification and assignment of quantum states is a comparison between the quantum states and the corresponding classical dynamics . to this end , the first step is the construction of the classical hamiltonian function , which corresponds to the quantum hamiltonian given in eqs .
( [ eq - hqm])([eq - hqmb ] ) .
this is done by heisenberg s substitution rules @xcite @xmath19 there are two different lines of argumentation for this substitution .
first , it is exact for the harmonic oscillator where the well known classical hamiltonian @xmath20 is obtained by the replacement of the symmetrized product of an annihilation and a creation operator by the classical action .
this implies the correspondence @xmath21 between the classical action @xmath22 and the quantum number @xmath13 of the oscillator ( @xmath22 is here measured in units of @xmath23 ) .
this correspondence of eq .
( [ eq - inrel1 ] ) is also a result of the application of the semiclassical bohr - sommerfeld quantization rules to the harmonic oscillator . in more general cases we have to generalize the bohr - sommerfeld method to the ebk quantization .
then the argument holds for any bound system of any number of degrees of freedom as long as the system is close to integrable ( for general background information on semiclassics see @xcite ) . in general , the semiclassical methods give results correct in the lowest two orders in @xmath23 ( orders 0 and 1 ) and cause errors of order @xmath24 .
the application of the substitution rules of eq .
( [ eq - airel ] ) to the quantum hamiltonian of eqs .
( [ eq - hqm])([eq - hqmb ] ) gives @xmath25 a hamiltonian for the same system but expanded in another basis was analyzed in @xcite .
this function can be interpreted as the hamiltonian of a classical system of three coupled anharmonic oscillators described in action - angle variables @xmath26 and @xmath27 , where @xmath28 . as a method to construct the corresponding classical hamiltonian , the substitution rules of eq .
( [ eq - airel ] ) always give the correct result since in this direction ( quantum @xmath29 classical ) the correspondence is unique whenever it exist at all , in contrast to the other direction ( classical @xmath29 quantum ) with its notorious @xmath24 problems . at high excitation ( large quantum numbers )
there is a second argument for the semiclassical correspondence .
the application of a creation or annihilation operator to a number state @xmath30 has the effect @xmath31 in the limit of a large quantum number @xmath13 , the difference between @xmath13 and @xmath32 or @xmath33 is irrelevant in the square roots as well as in the states and the operators can simply be replaced by multiplication with the number @xmath34 .
this argument holds for condensates where a large number of particles goes into a superfluid state which is well described by a mean - field limit .
this is in line with the standard argument of semiclassical behavior in the limit of large quantum numbers .
interestingly , for systems of coupled anharmonic oscillators the semiclassical treatment is very good also for low excitation numbers . in this limit ,
we approach the integrable harmonic limit where the bohr - sommerfeld treatment gives the correct result .
the experience with molecular systems of the structure of eqs .
( [ eq - hqm])([eq - hqmb ] ) shows that a semiclassical treatment of such systems is globally quite good in most cases .
accordingly , we base our method of semiclassical assignment on this argument .
semiclassical arguments will be used later first to convert the eigenstates of the many - body hamiltonian ( [ eq - hqm ] ) into wave functions on the toroidal configuration space and second to compare these functions with important structures seen in the classical dynamics .
the integrable part @xmath35 of the hamiltonian , which does not contain interactions between the three oscillators , leaves all actions unchanged .
in contrast , @xmath36 changes the values of the actions ( particles in the wells ) because of its dependence on angles and introduces interactions between the three oscillators . in this sense
we call in the following @xmath36 the interaction part of the hamiltonian . in the picture of particles in the triple
well , @xmath36 describes tunneling terms between the various wells . the poisson bracket between @xmath37 and the observable @xmath38 the total action , is equal to zero , which corresponds to the quantum mechanically conserved number of particles .
note that the numerical value of @xmath39 differs by @xmath40 from the value of @xmath41 because of the zero point actions .
the symmetry @xmath42 can be used to reduce the number of degrees of freedom from three to two by a canonical transformation . using the generating function @xmath43 of the old angles @xmath44 and the new actions @xmath45 results in the transformations ( together with eq .
( [ eq - kcl ] ) ) @xmath46 where @xmath47 are the new angles conjugate to @xmath45 .
the hamiltonian in the new coordinates is given by @xmath48 with corresponding equations of motions @xmath49\cos\psi_1\nonumber\\ & \quad + \frac{k_{23}}{2}\sqrt{\frac{j_2}{k - j_1-j_2}}\cos\psi_2\,,\label{eq - diffhama}\\ \dot{\psi}_2 & = ( \omega_3 + 2x_3j_2)-(\omega_2 + 2x_2(k - j_1-j_2))\nonumber\\ & \quad -\frac{k_{23}}{2}\biggl[\,\sqrt{\frac{k - j_1-j_2}{j_2}}-\sqrt{\frac{j_2}{k - j_1-j_2 } } \;\biggr]\cos\psi_2\nonumber\\ & \quad + \frac{k_{12}}{2}\sqrt{\frac{j_1}{k - j_1-j_2}}\cos\psi_1\,,\label{eq - diffhamb}\\ \dot{j}_1 & = -k_{12}\,\sqrt{j_1(k - j_1-j_2)}\,\sin\psi_1\,,\label{eq - diffhamc}\\ \dot{j}_2 & = -k_{23}\,\sqrt{j_2(k - j_1-j_2)}\,\sin\psi_2\,.\label{eq - diffhamd}\end{aligned}\ ] ] the classical configuration space is a two dimensional torus spanned by the two angles @xmath50 and @xmath51 . in order to compare the classical and the quantum system we have to represent the states as wave functions on the classical configuration space . the way to do this
will be described in the following section .
the angle variables can be introduced in the quantum system by using the set of functions @xmath52 first introduced in molecular spectroscopy by sibert and mccoy @xcite .
these functions are similar to the bargmann states studied in @xcite in the context of a bose - einstein condensate .
this relation is well - known from the context of infinite lattices .
there , the sum is taken from @xmath53 to @xmath54 and corresponds to the representation of bloch functions in terms of wannier functions .
the angle variables @xmath55 , @xmath56 and @xmath57 span the brillouin zone .
however , in this example these functions are not orthogonal due to the fact that for fixed @xmath41 the sum is finite , @xmath58 = \!\!\!\!\!\!\!\sum_{n_1+n_2+n_3=n } \!\!\!\!\!\!\ ! e^{-i(n_1(\varphi_1-\varphi'_1)+n_2(\varphi_2-\varphi'_2)+n_3(\varphi_3-\varphi'_3))}\,.\end{gathered}\ ] ] for a large particle number @xmath41 the scalar product converges to a delta - comb .
there is a considerable deviation for the value of @xmath59 , which can play an important role when matrix elements are calculated .
but here we use these functions only for visualization and not for further algebraic manipulations .
the eigenfunctions of ( [ eq - hqm ] ) have the form @xmath60 the coefficients @xmath61 can be obtained by a numerical diagonalization in the number basis @xmath62 .
the eigenstates in the angle representation ( [ eq - phirep ] ) , i.e. the wave functions , are given by a fourier series : @xmath63 finally , we can reduce the number of degrees of freedom in this representation by using the same coordinate transformation as in the classical case of eq .
( [ eq - vartrans ] ) .
this leads to the expression @xmath64 the global phase factor @xmath65 can be ignored in the following considerations . it must be emphasized that the sum includes only a finite number of terms due to the finite number of combinations of numbers @xmath66 , @xmath67 and @xmath68 which sum to @xmath41 .
therefore the fourier expansion in eq .
( [ eq - eigfred ] ) has only a finite resolution . for a very small value of @xmath41
, this sum has just a few terms , so that only the coarse grain structure can be explored ; accordingly , the eigenfunctions @xmath69 show only diffuse structures . in our example , the configuration space of the reduced system is the two dimensional torus @xmath70 with total volume @xmath71 .
the total number of basis states for a given number of particles @xmath41 is @xmath72 .
accordingly , the eigenfunctions which are linear combinations of the @xmath73 basis functions can only show patterns with a resolution of the order @xmath74 in the area or a resolution of the order @xmath75 in each direction .
with @xmath59 we have @xmath76 eigenstates , giving a resolution of approximately @xmath77 in each direction . the reinterpretation of the expansion of an eigenstate into number states as a fourier series on the toroidal configuration space has the following semiclassical interpretation , where we write for the moment @xmath23 explicitly into the equations : if one naively quantizes the classical canonical variables @xmath78 using the schrdinger quantization @xmath79 = i\hbar\delta_{kl}$ ] , which imposes @xmath80 , then functions @xmath81 can be interpreted as wave functions in coordinate space . of course , the schrdinger quantization is correct only in cartesian coordinates and it does not commute with canonical transformations , in general yielding errors of the order of @xmath24 .
therefore the results have to be interpreted semiclassically .
note that due to our symmetric introduction of the quantum - classical correspondence in eq .
( [ eq - inrel1 ] ) , the errors to first order in @xmath23 cancel identically . because of these considerations we call the wave function from eq .
( [ eq - eigfred ] ) the _ semiclassical wave function_. in many semiclassical investigations , husimi functions are used to relate quantum wave functions of eigenstates to structures in the classical phase space .
this is the appropriate and natural procedure if the usual position and momentum coordinates are used .
it is less clear and in addition not necessary in our case where the whole dynamics is treated in action - angle variables .
let us explain this point in some detail : the description of the system by a hamiltonian of the functional structure of eqs .
( [ eq - hqm])([eq - hqmb ] ) in the quantum case or eq .
( [ eq - hcl ] ) in the classical case only makes sense for bound systems , it is not appropriate to describe scattering systems
. therefore we restrict the following discussion to bound states only . for any bound eigenstate in the standard position space
, there must be the same amount of wave running in one direction and in the opposite direction , otherwise it would not be a bound stationary state .
accordingly , the wave function can be chosen real .
the phases of the wave function do not play any important role and do not help for the classification of the states .
the canonically conjugate momenta have continuous values and wigner or husimi functions are defined without any problem on the classical phase space and indicate in many cases to which structure in the classical phase space some particular quantum state belongs .
the situation is very different in action angle variables . here
the configuration space is a torus with its very different global topology .
this causes great difficulties to define the usual wigner or husimi functions .
because of the periodicity of the configuration coordinates , the corresponding canonically conjugate variables ( here the actions ) only have discrete values in the quantum dynamics .
this makes it very tricky to convert the wave function into something defined on the continuous classical phase space . on the other hand
, we do not really need to do this , since we have the following simpler method to squeeze out of the wave functions information on the classical actions .
waves propagating in one direction on a torus always return to the starting point .
accordingly , wave functions for a bound state can have and in fact in most cases do have
strong running wave contributions and the phase of the function is essential and will be analyzed to help in the classification of the state . in a semiclassical spirit
the phase of a wave function can be interpreted as a classical action integral and accordingly the gradient of the phase function gives the value of the canonically conjugate momentum which in this case is the action .
if there is a sufficiently large patch of configuration space where the phase function comes close to a plane wave , then its gradient indicates the value of the actions which is represented by this part of the wave function .
this provides a kind of lift of the wave function from configuration space into phase space .
if there are closed loops on the torus along which the phase function is very regular ( and this usually happens along density crests which run along the classical organizing center as will be explained in detail in section [ sec - semiclass ] ) then we interpret this as representing a motion of almost constant action along this loop .
this idea is used to get longitudinal quantum numbers introduced in section [ sec - semiclass ] .
before we relate individual quantum states to guiding centers of the classical dynamics , we must get an overview of the classical dynamics and its skeleton . as an example , we discuss the classical dynamics for @xmath59 , i.e. for the value @xmath82 of the classically conserved total action @xmath39 . in the following ,
we choose parameter values @xmath83 , @xmath84 , @xmath85 and @xmath86 , which lead to a quantum mechanical energy interval of @xmath87 $ ] .
the classical reduced system exists in the energy interval @xmath88 $ ] .
furthermore , we measure all energies with respect to the quantum mechanical ground state of @xmath37 in eq .
( [ eq - hqm ] ) , i.e. we subtract the quantum mechanical zero point @xmath89 from the classical energies in order to facilitate the comparison between classical and quantum dynamics . to represent the classical dynamics graphically
, we show poincar sections in planes @xmath90 with positive orientation @xmath91 .
if an initial condition @xmath92 is chosen in the poincar section , then we first have to reconstruct the four corresponding coordinates in the phase space in order to start a trajectory of the flow through this point .
the two coordinates @xmath51 and @xmath93 coincide with the given coordinates in the domain of the poincar map .
the coordinate @xmath50 is obtained by the intersection condition and the remaining coordinate @xmath94 is calculated by an inversion of the hamiltonian function ( [ eq - hclred ] ) with respect to the coordinate @xmath94 for a fixed value of the energy and for the known values of the other three coordinates .
here some care is necessary since this inverse function is multivalued .
first we fix one orientation of the domain , i.e. we always search for solutions with @xmath95 . in principle
there can be several solutions with the same orientation and then it is necessary to ensure that all initial points used belong to the same branch .
poincar sections in planes @xmath96 look very similar to the ones in planes @xmath97 .
therefore it is sufficient to restrict ourselves to sections in @xmath90 only .
if the whole dynamics were governed by @xmath35 , then all actions would be constants of motion and all poincar sections would be foliated by invariant lines @xmath98 . including the interaction @xmath36 between the wells ( modes ) into the dynamics has the following effects . in regions of the phase space , where none of the resonances contained in @xmath36
has an important effect , the dynamics is in the kam regime ( see the extensive discussion of soft chaos in chapter 9 of @xcite ) and a large fraction of the phase space volume is still filled by invariant lines , which are continuous deformations of the invariant surfaces @xmath99 of the unperturbed @xmath35 dynamics .
we call such invariant surfaces primary tori .
this happens mainly in regions of phase space where the effective frequencies @xmath100 are far from simple rational ratios , for which there is a corresponding resonance coupling in @xmath36 , as explained in the next paragraph . for our particular choice of coupling terms in @xmath36 ,
only 1:1 resonances are relevant .
the effect of the coupling terms between the different modes can be described in the following way .
each term contains a cosine function whose argument is a difference between angles of the original degrees of freedom or one angle of the reduced system , see eqs .
( [ eq - hcl ] ) and ( [ eq - hclred ] ) .
because in our special case the arguments are differences of two angles with the same weight , we say that these terms describe 1:1 resonant interactions between the two degrees of freedom .
the right hand sides of the hamiltonian equations of motion ( [ eq - diffhama ] ) and ( [ eq - diffhamb ] ) for the angles @xmath101 ( @xmath102 ) of the reduced system , @xmath103 contain two contributions .
the first consists of the difference of two effective frequencies from eq .
( [ eq - omegaeff ] ) , and the second is the derivative of the coupling terms with respect to the action , which contains cosine functions . first , let us assume that we change some parameter , e.g. @xmath104 , to see how coupling sets in .
further we assume that the difference between the effective frequencies , i.e. the angle independent term on the right hand side , is different from zero .
let us say it has the value @xmath105 .
for a small value of @xmath104 the angle dependent terms are not able to cancel @xmath106 regardless of the value of the angles . the angle dependent terms have the maximal absolute value for angle values @xmath107 and @xmath108 because of the dependence on cosine functions .
when @xmath104 increases , then at one point it reaches a value , where the angle dependent terms are just able to cancel @xmath106 .
then the angle @xmath101 of the reduced system stops , @xmath109 , and we call this _ frequency locking_. this necessarily happens for angle values where the cosine functions have maximal absolute value , i.e. where the angles are 0 or @xmath108 . whether the appropriate angle values are 0 or @xmath108 depends on the signs of @xmath106 and of the terms in front of the cosine functions .
when the value of @xmath104 is further increased , then there is a whole interval of angle values where locking is possible .
the actual dynamics of the locked motion then performs small oscillations around the angle values 0 or @xmath108 .
this will be seen in the numerical results of the classical dynamics . in the quantum dynamics
the fluctuations around the coupling point of the angles are quantized and give rise to a discrete set of transversal quantum numbers , see section iv .
if only one of these resonant couplings is strong , then the dynamics is still close to integrable , and a large part of the phase space volume is filled by invariant tori , which show up as invariant lines in the poincar sections .
however , due to the rearrangement of phase space structures by the resonant coupling , the invariant surfaces in phase space are no longer primary tori , i.e. are no longer continuous deformations of invariant surfaces of the @xmath35 dynamics .
large bundles of secondary tori appear which are organized around periodic orbits ( in this case stable , elliptic ) representing the _ guiding centers _ for the new nonlinear modes
. there are also corresponding unstable periodic orbits , which in the integrable case are represented by separatrix crossings in poincar sections . in the nonintegrable cases , the separatrices break and turn into homoclinic tangles , which become the central structures of chaotic strips .
however , if only one resonant coupling has a strong effect and the others are not important , then the chaos strips are very thin and they still appear almost like separatrices .
if two or more linearly independent resonant couplings are strong , then chaos on large scales can appear .
these regions in phase space are resonance overlap zones @xcite .
however , also in strongly chaotic regions of phase space there are still simple short periodic orbits ( in this case unstable , normal hyperbolic or inverse hyperbolic ) which act as guiding centers of the flow .
then the dynamics is chaotic but nevertheless the flow follows some guiding center on the average .
this average flow is relevant for the comparison with quantum dynamics .
thus , also in the classically chaotic case we may find surprisingly simple and clean structures in a large part of the quantum wave functions .
in such cases it can be appropriate to imagine simple idealized classical guiding centers and interpret the quantum states as quantum excitations of these idealized structures .
let us give a short estimate of the size of structures which are relevant for our semiclassical considerations .
the range of action values is limited between 0 and @xmath39 due to eq .
( [ eq - kcl ] ) , the angle can vary over an interval of length @xmath110 . for each particular plot
only a part of this range is energetically accessible in reality . accordingly
the size of the poincar section is limited by @xmath111 . for semiclassical investigations structures of a size of @xmath23 or larger
are relevant .
we always use units in which @xmath23 has the numerical value 1 and also the values of all actions should be interpreted as being given in units of @xmath23 . therefore structures in our poincare plots are of interest in the following , if their size is at least in the order of one unit of action or has a relative size of @xmath112 compared to the size of the maximally possible domain of the map .
we perform almost all our calculations for the reduced system . on the other hand ,
the real object of interest is the original system of particles in three wells .
therefore we need a fast and easy method to transfer statements about the reduced system into the corresponding statements about the original system .
we have called this procedure the _ lift _ in the previous work on molecular systems @xcite .
let us assume a trajectory of the reduced system is given and we want to reconstruct the corresponding trajectory of the original system .
the first step of the procedure is the reconstruction of the cyclic angle .
it is done rigorously by using the hamiltonian equation of motion @xmath113 the right hand side of this equation does not depend on @xmath114 but only on the known values of the other coordinates as function of time .
accordingly we get @xmath115 by a simple integration with respect to time .
the experience with the molecular systems has shown that normally it is sufficient to approximate @xmath115 by @xmath116 times a constant effective frequency . in our case
@xmath114 is the only fast variable of the whole system and describes a fast oscillation superimposed on the motion of the whole system .
the initial value @xmath117 is rather irrelevant .
in contrast , the variables of the reduced system are slow variables describing the relative motion between the various degrees of freedom of the original system .
the next step of the lift procedure is to undo the canonical transformation and to go back to the coordinates of the original system . in this second step
the advantage of choosing the new actions equal to some of the old actions becomes evident .
the knowledge of the actions in the reduced system and of the constant value of @xmath39 gives immediately the values of the old actions , i.e. the values of the particle numbers in the three wells .
because of this simple connection between the actions of the reduced system and the actions of the original system we will switch very freely between the reduced and the original system in the following considerations .
in our case we have in the interaction part @xmath36 of the hamiltonian 1:1 couplings between the degrees of freedom 1 and 2 and between 2 and 3 respectively .
indirectly this also implies a 1:1 coupling between the degrees of freedom 1 and 3 .
accordingly , we have the following coupling schemes : the classical reduced system for energy @xmath118 . ( a ) poincar plot in the plane @xmath90 for variables @xmath51 and @xmath93 with @xmath94 fixed by energy conservation .
( b ) trajectory in a primary torus in the lower region of ( a ) for initial values @xmath119 .
( c ) trajectory in a primary torus in the upper region of ( a ) for initial values @xmath120 .
the points of the trajectories are given in equidistant time intervals @xmath121 in order to indicate the velocity by the distance between neighboring points.,width=321 ] _ type ( a ) _
: if the effective frequencies are not very close to each other , then no interaction term can cause frequency and phase coupling , and all three modes run independently with their own effective frequency .
this is the kam regime with many primary tori , where the motion is of quasiperiodic type with three independent frequencies .
the organization center of the reduced system is the complete configuration space @xmath70 . in poincar plots , we see many invariant lines which are continuous deformations of horizontal lines @xmath122
, i.e. of the invariant lines belonging to @xmath35 .
this type of motion appears mainly in the middle of the accessible energy interval for a given particle number . in fig .
[ fig - class55 ] we give some numerical results for the energy @xmath118 .
part ( a ) shows the poincar section and parts ( b ) and ( c ) show two segments of trajectories in the reduced configuration space .
the domain of the poincar map in ( a ) consists of two parts .
the range of @xmath93 values between approximately 10.2 and 20.5 is not accessible at this energy . at values of @xmath93
around 9 , we see many primary tori . a segment ( five revolutions in direction of @xmath50 ) of a typical trajectory belonging to one of them is shown in part ( b ) of the figure . in the long run
, the trajectory fills the whole configuration space quasiperiodically . in these primary tori the action @xmath93 is smaller than the action @xmath94 , so that the trajectories move faster in @xmath50 direction than in @xmath51 direction .
the opposite happens on the primary tori lying around @xmath93 values of 21 .
here the @xmath93 action is largest and therefore the quasiperiodic trajectories run with higher speed in the @xmath51 direction .
( for a numerical example , see a trajectory segment in fig .
[ fig - class55](c ) ) .
the other structures seen in fig . [ fig - class55](a ) belong to other types of motion , discussed below .
_ type ( b ) _ : if the effective frequencies of modes 2 and 3 are close but that of the first mode is not close , then we expect that modes 2 and 3 are locked but mode 1 is independent .
the motion is then quasiperiodic with two independent frequencies .
the organization center in the reduced system is a one dimensional curve with @xmath123 . in poincar plots in the plane @xmath124 , we see secondary islands .
this motion appears mainly for high energies .
figure [ fig - class80 ] gives some numerical results for @xmath125 .
part ( a ) shows a poincar section , again in the plane @xmath90 , and part ( b ) shows two periodic orbits in the configuration space .
the motion at the upper end of the accessible energy interval is close to integrable . at a very high energy , motion in @xmath50 direction
is preferred , since the linear frequency @xmath126 of original mode 1 is higher than the frequency @xmath127 of mode 3 . for decreasing energy ,
the kam island around the center at @xmath128 increases in size while the one around @xmath129 decreases . the central periodic orbit around @xmath130 remains stable for energies down to approximately @xmath131 , while the other one soon becomes unstable and its kam island disappears . in fig .
[ fig - class55](a ) we see clearly the large kam island belonging to the organization center @xmath130 , with center at @xmath132 .
( color online ) the classical reduced system for an energy @xmath125 .
( a ) poincar plot as in fig .
[ fig - class55](a ) .
( b ) periodic orbits crossing the poincar section in the centers of the kam islands ( black ) for initial values @xmath133 and at the border ( green ) for initial values @xmath134.,width=321 ] in contrast to the idealized organization center @xmath130 , the exact one is a periodic trajectory running in @xmath50 showing small wiggles in @xmath51 direction around the average value @xmath130 .
however for our considerations it is simpler and completely satisfactory to replace this true organization center , the true periodic orbit , by an idealized organization center , for which we just take the straight line @xmath130 .
the reader might remember the previous discussion of the onset of angle coupling and the values of the angles at which coupling sets in . in the spirit of this previous discussion
we define the idealized organization center as the subset of the configuration space defined by the angle restrictions exactly at the onset of the corresponding coupling scheme .
also the idealized semiclassical wave functions are given with respect to the corresponding idealized organization center .
_ type ( c ) _ :
if the effective frequencies of modes 1 and 2 are close , but that of the third mode is not close , then we expect that modes 1 and 2 are locked but mode 3 is independent .
then the motion is again quasiperiodic with two independent frequencies . in the reduced system ,
the organization center is a one dimensional curve which can be idealized by a line @xmath135 , where the constant usually is @xmath107 or @xmath108 according to the discussion in the beginning of this section .
the periodic orbit itself running in the @xmath51 direction is almost impossible to find in poincar maps with plane of intersection @xmath90 , since it violates the transversality of the map .
however , when it is stable , then there is a bundle of invariant tori around it . in poincar plots in the planes @xmath124 , these invariant tori appear as lines extending over all values of @xmath51 . in fig .
[ fig - class55](a ) they are the lines at the highest values of @xmath93 . in fig .
[ fig - class40 ] , we show some numerical results at energy @xmath136 . part ( a ) shows the poincar map and parts ( b ) and ( c ) show trajectories in configuration space . in fig .
[ fig - class40](a ) the lines at small values of @xmath93 belong to the tori around the organization center @xmath90 .
figure [ fig - class40](b ) shows a segment of a typical quasiperiodic orbit on one of these tori . while running monotonously in the negative @xmath51 direction , it oscillates in @xmath50 around the value 0 .
( color online ) the classical reduced system for an energy @xmath136 .
( a ) poincar plot as in fig .
[ fig - class55](a ) .
( b ) quasiperiodic orbit with initial values @xmath137 .
( c ) two periodic orbits : one oscillating along the diagonal with starting point @xmath138 ( black ) and the other rotating around the line @xmath139 ( green ) with starting point @xmath140.,width=321 ] _ type ( d ) _ : if the effective frequencies of modes 1 and 3 are very close , then also the weak indirect tunneling processes between modes 1 and 3 can cause coupling .
if the frequency of mode 2 is far from this common frequency , then mode 2 runs independently .
the corresponding organization center in the reduced system is the line @xmath141 , where again this constant is usually 0 or @xmath108 . in poincar plots in planes @xmath124 ,
we see secondary islands . in fig .
[ fig - class40](a ) , the two kam island of moderate size with centers at @xmath130 , @xmath142 and @xmath143 and @xmath144 respectively represent this type of motion .
the two periodic orbits belonging to the centers of these two kam islands are shown in fig .
[ fig - class40](c ) .
one oscillates along the diagonal and the other rotates around along the line @xmath145 . _
type ( e ) _ : if all three effective frequencies are close , then there are two possibilities : ( e1 ) : there is coupling between all three modes and the idealized organization center in the configuration space of the reduced system is a ( fixed)point . the actual trajectories oscillate around this coupling point and the relative angles @xmath101 do not rotate around the whole configuration torus .
this behavior , which dominates at very small energy , is shown in fig .
[ fig - class27 ] at energy @xmath146 .
only a limited range of @xmath51 values around the point zero is energetically accessible .
the same also holds for @xmath50 .
rotations around the configuration torus in either direction or the diagonal become possible only for a higher energy .
one of the organizing centers is represented in the poincar plot by a stable fixed point which lies at the center of the large kam island shown in fig .
[ fig - class27](a ) , and which is shown in configuration space in fig .
[ fig - class27](b ) as the figure - of - eight orbit mainly oscillating in the antidiagonal direction .
the other organizing center is an unstable periodic orbit belonging to the unstable fixed point near @xmath130 , @xmath147 in the poincar plot . in the configuration space plot of fig .
[ fig - class27](b ) , it is the orbit oscillating in the diagonal direction . at this energy ,
all trajectories in configuration space oscillate around the point @xmath148 , which acts as point organizing center .
the two periodic orbits of fig .
[ fig - class27](b ) then act as guiding structures for these fluctuations around the organizing center .
topologically speaking , all trajectories are contractible to a point on the configuration torus at very low energy . at the lower end of the accessible energy interval ,
the dynamics starts as almost integrable and for this case the invariant manifolds of the unstable fixed point mentioned above lie close to a figure - of - eight shape separatrix in the poincare section . for increasing energy
the system moves further away from integrable and the separatrix breaks and turns into a homoclinic tangle which is the central structure of a chaos strip . in fig .
[ fig - class27](a ) for energy @xmath146 this chaotic layer still has moderate size
. for higher energy it grows rapidly and turns into the large chaotic sea seen in fig .
[ fig - class40](a ) at energy @xmath136 .
( color online ) classical reduced system for energy @xmath146 .
( a ) poincar plot as in fig .
[ fig - class55](a ) .
( b ) trajectories through the stable fixed point ( black , double loop ) with initial values @xmath149 , and the unstable fixed point ( green , along the line @xmath150 ) with initial values @xmath151.,width=321 ] ( e2 ) : the couplings break and reestablish intermittently , and the dynamics shows large - scale chaos . the appearance of chaos in the case of two independent resonant interactions becoming active is a demonstration of chirikov s point of view of chaos being caused by resonance overlap @xcite . in poincar plots , we see large - scale chaos and eventually embedded in it remnants of islands and regular structures .
the beginning of chaos for small energies can be seen in fig .
[ fig - class27](a ) ; chaos on a large scale is evident in figs .
[ fig - class55](a ) and [ fig - class40](a ) .
in this section , we show examples of wave functions belonging to the various classes of motion described in the previous section . our method of classification has been developed in @xcite especially for hamiltonians given quantum mechanically in raising and lowering operators and classically in action - angle variables .
an analysis in a similar spirit of wave functions in the usual position space is rather common in molecular physics , two representative examples are @xcite .
in contrast to the procedure in action - angle space , the procedure in regular position space can be extended to scattering resonances , see @xcite .
our strategy of classification is as follows : first we expand the eigenstates @xmath152 in the representation @xmath153 , according to eq .
( [ eq - eigfred ] ) .
this representation of the eigenstates in the reduced configuration space is in complete analogy to the classical configuration space spanned by the angle variables @xmath50 and @xmath51 and therefore allows a direct comparison between the classical and quantum system .
we refer to these eigenfunctions @xmath69 as the _ semiclassical wave functions _ in order to indicate this resemblance .
we then check whether the density of the semiclassical wave function in the reduced configuration space resembles the structure of one of the organization centers described in the previous section ( types ( a ) ( e2 ) ) .
i.e. we check , whether the density is distributed over the whole configuration space without clear nodal structures ( type ( a ) ) , is concentrated along a few lines in the @xmath50 direction ( type ( b ) ) , in the @xmath51 direction ( type ( c ) ) or in the diagonal direction ( type ( d ) ) , is organized around the point center @xmath148 ( type ( e1 ) ) or shows random interferences between the pattern of different organization centers leading to irregular structures ( type ( e2 ) ) .
we call states , for which the density is located in a single crest along the organizing center , a transverse ground state to this organization center . in transversely excited states ,
the density is concentrated along various copies of the organization center , where these various copies are displaced relatively to each other and the wave function shows nodal structures between them .
in addition , we look for the phase advance in directions in or parallel to the organization structure
. the phase function must be continuous along curves which do not cross nodal lines . recall that the phase function can have singularities only in zero points of the density .
accordingly , the curve along a crest of high density must be a curve of continuous phase .
then the phase advance of such a curve must be some integer multiple of @xmath110 , say @xmath154 , and this number @xmath155 serves as one quantum number of the state .
these longitudinal quantum numbers , together with the transverse quantum numbers given by the nodal structures , provide a complete set of quantum numbers characterizing the state relative to its organization center .
we expect all states which can be related to an organization center to be close to a product of a plane wave in the longitudinal direction of this organization center and an oscillator function in transverse directions . in states belonging to classically chaotic motion , we do not see a simple and clear pattern in the density nor in the phase .
accordingly we are not able to give any assignment by quantum numbers to such states . in the following ,
the eigenstates @xmath156 , @xmath157 are sorted by increasing energy starting with the label 1 for the eigenstate with lowest energy .
we start our analysis of the semiclassical wave functions at the lower end of the accessible energy interval .
since the hamiltonian is dominated by quadratic anharmonicities , the smallest energy is realized by distributing the total excitation of 30 quanta ( particles ) evenly over the 3 basis modes ( potential wells ) . in the classical picture , this corresponds to the case where all three actions @xmath158 are close to each other .
thus the three effective frequencies ( [ eq - omegaeff ] ) are very similar and frequency and phase locking is easily established by the resonant coupling terms in the hamiltonian as explained in the beginning of section [ sec - class ] . in the classical configuration space ,
this mechanism restricts the trajectories to a small region of configuration space ( cf .
[ fig - class27 ] ) .
this behavior is confirmed in the quantum case . here , the wave functions are organized around a point , as can be seen in fig .
[ fig - punktzentrum ] , where the ground state and various excited states are plotted .
state @xmath159 is the ground state in this class . in this case
the ground state of an organization center coincides with the energetic groundstate @xmath159 of the whole system , but we will assign also a groundstate for the other types of guiding centers .
the state @xmath160 is the first transversal excitation in the antidiagonal direction , while the state @xmath161 represents the first transversal excitation in the diagonal direction .
the state @xmath162 represents the second transversal excitation in the antidiagonal direction , and the state @xmath163 is the combination of one transverse excitation in the diagonal and one in the antidiagonal direction .
state @xmath164 is the fourth excitation in the antidiagonal direction .
a point center does not have any longitudinal directions .
accordingly , there are no phase advances in longitudinal directions to be counted for the assignment and any state of this class is characterized by the two transverse excitation numbers @xmath165 , one in the diagonal direction and one in the antidiagonal direction .
thus we show only the density plots without the phases in fig .
[ fig - punktzentrum ] . in this scheme ,
the six states @xmath159 , @xmath160 , @xmath161 , @xmath162 , @xmath163 and @xmath164 have quantum numbers @xmath148 , @xmath166 , @xmath167 , @xmath168 , @xmath169 and @xmath170 , respectively .
gray scale plot of the squared modulus of the ( a ) ground state @xmath171 and the excited states ( b ) @xmath160 , ( c ) @xmath161 , ( d ) @xmath162 , ( e ) @xmath163 and ( f ) @xmath164 .
white color corresponds to low density and black to the highest density .
the range of the @xmath101 is @xmath172$].,width=302 ] note that the direction of excitation corresponds to the direction of oscillation of the classical periodic orbits shown in fig .
[ fig - class27](b ) .
the classical periods of these two orbits are @xmath173 for the antidiagonal one and @xmath174 for the diagonal one .
the quantum excitations in the corresponding direction increase the energy of the state by the classical frequency @xmath175 , where @xmath176 is the period of the orbit taken at an intermediate energy .
the quantum - classical correspondence can be described in the following way : all three original modes are frequency locked and the phases fluctuate around the coupling point .
the motion is similar to the one in a two dimensional anharmonic oscillator centered around the point @xmath148 .
this oscillator has its own normal modes and the states presented in fig .
[ fig - punktzentrum ] can be interpreted as some of the low lying excitation of this oscillator and described by the excitation numbers of these normal modes .
however , the reader should not confuse these modes of fluctuations around coupling points with the modes which are used to formulate the original hamiltonian in eqs .
( [ eq - hqm])([eq - hqmb ] ) .
compare also with the discussion of the onset of coupling given in section [ sec - class ] .
the wave functions in this class are therefore close to two dimensional oscillator functions and can be described approximately by @xmath177 where the functions @xmath178 are eigenfunctions of a one dimensional oscillator with harmonic and anharmonic contributions .
it is interesting to see what this means in the original coordinates @xmath179 , @xmath158 . using the transformation ( [ eq - vartrans ] ) , one obtains for the idealized eigenfunctions @xmath180 all three degrees of freedom are entangled for this type of guiding center .
the entanglement is the quantum analog of the phase locking in the classical picture .
altogether , we can assign 29 of the 496 eigenstates to this class of functions .
the highest energies for a given number of particles are achieved by putting almost all excitation into one mode , with the other two modes having very low excitation .
classically , these two modes have similar effective frequencies , ( see eq .
( [ eq - omegaeff ] ) ) , and therefore they are locked easily . in fig .
[ fig - psi10 ] we show as examples the densities and phases for the states @xmath182 and @xmath183 . in part
( a ) we see the density concentrated along the line @xmath124 ; thus , the transverse excitation number is @xmath184 . along this line
, the phase function is almost like a plane wave .
the total phase advance along one cycle around the organization center is @xmath185 .
accordingly , the longitudinal excitation number is @xmath186 .
note that the phase function has singular points far away from the places of high density . in part ( c ) of the figure
, we see the density concentrated along four lines in the @xmath51 direction .
the four density crests are separated by 3 nodal lines , which can be seen very clearly as lines of discontinuities in the phase plot in part ( d ) .
accordingly , the transverse excitation number of state @xmath183 is @xmath187 . along the density crests we count the total phase advance to obtain the longitudinal quantum number @xmath188 .
plot of the eigenfunctions @xmath182 and @xmath183 of the quantum system belonging to the @xmath90 guiding center . plot ( a ) shows @xmath189 , ( b ) shows @xmath190
, ( c ) shows @xmath191 , ( d ) @xmath192 . in the phase plots ,
the degree of darkness from white to black indicates the phase advance from @xmath107 to @xmath193.,width=302 ] the energy distance between two states which differ by one unit in @xmath155 and that have the same transverse quantum number is given by the frequency of the classical organizing center , the periodic orbit ( central fiber in the quasiperiodic motion shown in fig .
[ fig - class40](b ) ) , taken at an intermediate energy .
the classical motion behind this class of states is as follows : mode 3 runs with its own effective frequency independently of the other modes .
the quantum number @xmath155 is its action due to the semiclassical assignment of the phase function @xmath194 to the classical action integral , @xmath195 with @xmath196 and @xmath197 .
the phase @xmath194 is defined by @xmath198 and @xmath199 is the classical guiding center .
the path @xmath199 is simply the line @xmath90 for this class and the the number of particles in mode 3 can be directly assigned to the quantum number @xmath155 .
modes 1 and 2 run locked with total excitation , i. e. number of particles , @xmath200 , and the quantum number @xmath201 characterizes the fluctuations of the coupled motion around the coupling point .
the eigenfunctions in this class can therefore be written approximately as @xmath202 now we transform this expression back to the original coordinates , where we can interpret the actions @xmath158 directly as the number of particles in the potential well @xmath203 . using again transformation ( [ eq - vartrans ] ) ,
we can write the idealized wave functions of this class as @xmath204 this type of wave function shows entanglement between modes 1 and 2 , while mode 3 separates . the number of particles in mode 3 is given by @xmath155 while the transversal quantum number @xmath201 describes the transversal excitation of the organization center .
a total of 51 eigenstates can be assigned to this class of functions .
the states of this class look very similar to those in the previous subsection , only with the roles of the modes 1 and 3 interchanged and hence with @xmath50 and @xmath51 interchanged .
however , there is no perfect symmetry between classes c and b because there is no perfect equality between the modes 1 and 3 .
remember that @xmath206 .
this small perturbation of the symmetry is responsible that the states of class b loose their characteristics under smaller transverse excitations as the ones for class c. accordingly we can assign less states to class b , namely 42 only , than we have assigned to class c. if almost all the action @xmath39 is in mode 2 , then modes 1 and 3 have low actions and similar effective frequencies , whereas mode 2 has a quite different effective frequency . even though the hamiltonian does not contain a direct coupling between modes 1 and 3 , sometimes the small indirect coupling is sufficient to cause locking between modes 1 and 3 .
fig . [ fig - psi1psi2 ] shows the states @xmath208 and @xmath209 as two examples of semiclassical wave functions in this class .
the organization center is the diagonal @xmath210 .
state @xmath208 has the transverse quantum number @xmath211 relative to this center and state @xmath209 has @xmath212 .
the phase functions show that @xmath213 for state @xmath208 and @xmath214 for state @xmath208 .
the energy distance between two states , which differ by one unit in @xmath155 and that have the same transverse quantum number , is given by the frequency of the classical organizing center , namely the periodic orbit shown in fig .
[ fig - class40](c ) taken at an intermediate energy . the classical motion carrying these states is the following : the coupled motion of modes 1 and 3 has the number of particles @xmath155 while the rest of the total excitation @xmath200 is in mode 2 .
the transverse quantum number @xmath201 again characterizes the fluctuations around the coupling point . for the idealized wave functions of the reduced system , we obtain @xmath215 in the original coordinates , the wave function has the form @xmath216 in these coordinates , mode 2 separates from the other modes which are entangled .
the number of particles in mode 2 is given by @xmath200 , while the rest of the particles is in the entangled state of the other two modes , for which the quantum number @xmath201 is a measure of the fluctuations around the organization center .
we can assign 8 eigenstates to this class of functions .
plot of the eigenfunctions @xmath208 and @xmath209 of the quantum system belonging to the @xmath150 guiding center .
plot ( a ) shows @xmath217 , ( b ) @xmath218 , ( c ) shows @xmath219 , ( d ) @xmath220.,width=302 ] fig .
[ fig - torus ] shows the wave functions of the states @xmath222 and @xmath223 , which do not show any coupling .
these states belong to normal mode motion in the original modes .
this does not necessarily mean that they have a constant density , but the density is without any clear structure and the phase function is close to a plane wave globally .
as the two quantum numbers we count the phase advances around the two fundamental cycles of the toroidal configuration space . in part
( b ) of the figure we assign the quantum numbers @xmath224 , @xmath225 and from part ( d ) we read off @xmath226 and @xmath227 .
these states are described by the classical motion in the following way : the original mode 1 has the number of particles @xmath228 and original mode 3 has @xmath229 particles .
the rest of the excitation @xmath230 is in mode 2 .
all three modes run independently with their own effective frequency . thus phase functions of states of this class come close to a basis function ( i.e. they resemble a plane wave ) , even though the wave function can be a strong mixture of several basis functions .
the functional form of such states is therefore approximately given by @xmath231 or written in the original coordinates as @xmath232 these idealized functions factorize and the three degrees of freedom are completely disentangled .
there are 50 eigenstates in this class .
plot of the eigenfunctions @xmath222 and @xmath223 of the quantum system belonging to the @xmath70 guiding center .
plot ( a ) shows @xmath233 , ( b ) @xmath234 , ( c ) @xmath235 , ( d ) @xmath236.,width=302 ] finally , we give two examples of wave functions where we could not make any assignment to one of the organizing centers listed in the previous section . fig .
[ fig - chaos ] shows the densities and phases of states @xmath237 and @xmath238 .
neither in the density plots nor in the phase plots , can we discover any clean pattern related to one of the organizing centers .
the connection to the classical motion we interpret as follows : in classical chaos , any typical trajectory jumps around irregularly between the neighborhoods of various simple periodic orbits and therefore between various types of motion .
the corresponding quantum wave function should be random interference patterns of the structures belonging to the various organizing centers involved in the classical chaotic motion .
sometimes we can demix these interference patterns by forming appropriate linear combinations of several eigenfunctions of the hamiltonian .
plot of the eigenfunctions @xmath237 and @xmath238 of the quantum system belonging to class ( e2 ) .
plot ( a ) shows @xmath239 , ( b ) @xmath240 , ( c ) shows @xmath241 , ( d ) @xmath242.,width=302 ] concluding this section , we are able to characterize 180 of the 496 eigenstates within the scheme of guiding centers given by the classical motion ( excluding chaotic motion ) . our aim is not a complete assignment of all states , but rather to give an easy visual criterion in order to select states with different types of e.g. entanglement and localization properties as described in this section for each class . for these states , one can use the classical picture in order to understand the quantum mechanical structure , which allows a very intuitive treatment of the states .
the above graphical classification of the semiclassical wave functions is not strict and some functions allow ambiguous assignments .
such functions show characteristics of different classes and it is only a matter of degree in which class to put them .
for example , the phase functions in fig .
[ fig - psi1psi2 ] could be interpreted as continuous deformations of plane waves and therefore they could be assigned to type ( a ) as well .
finally , we wish to discuss the implications of our analysis for the time evolution in the classical description of the system . the classical system can be interpreted as an array of three bose - einstein condensates where the condensate in each well is described by the gross - pitaevskii equation and where the condensates interact weekly through josephson tunneling @xcite . in the previous section ,
we have used the classical system only to provide a tool for the classification of the quantum wave functions , and we have shown how close the quantum eigenfunctions resemble the classical guiding centers . in this section ,
we look in the other direction . starting from the classical system , i.e. the mean - field equations , we want to ask what information the structure of the quantum system can provide in order to solve the mean - field equations :
the analysis of a system of coupled nonlinear differential equations is very involved , while in the quantum system we only have to diagonalize the hamiltonian numerically and plot the eigenfunctions in configuration space .
since it is more convenient in this context to speak about complex occupation amplitudes , we introduce the new variables @xmath243 in these variables , the classical hamiltonian ( [ eq - hcl ] ) can be written as @xmath244 with canonically conjugate variables @xmath245 and corresponding equations of motion @xmath246 this system of three ordinary differential equations for the complex coefficients @xmath247 is equivalent to the six equations for the angles @xmath179 and the actions @xmath158 with @xmath28 .
the equations can also be derived from the gross - pitaevskii equation in coordinate space using an expansion of the condensate wave function in wannier functions @xcite .
one can also use the eigenstates of the one - particle hamiltonian , the so - called wannier - stark functions , resulting in the disappearance of the linear tunneling terms in the hamiltonian ( [ eq - hqm ] ) , while higher order coupling terms become important @xcite .
now we choose the initial conditions @xmath248 by using the semiclassical correspondence ( [ eq - inrel1 ] ) between the classical actions @xmath158 and the quantum numbers @xmath249 of a number state @xmath62 : @xmath250 in this way , we can construct initial conditions @xmath251 , where the action @xmath158 can be interpreted quantum mechanically via eq .
( [ eq - inrel ] ) as the number of particles in mode @xmath203 .
furthermore , we can use this correspondence in order to construct initial conditions resembling the properties of the eigenstates of the system .
before we explain this in more detail we first discuss the case of the basis vectors . here
we investigate to which extend we can attribute the same characteristics to the quantum mechanical number states @xmath62 and their classical analog defined by eq .
( [ eq - inrel ] ) .
accordingly , we define the initial conditions for the time evolution of eq .
( [ eq - gpe ] ) as @xmath252 in the following we will not explicitly write down the dependence of @xmath253 on the initial condition through the parameters @xmath254 and simply use @xmath255 .
we fix the three initial phases to zero , which corresponds to zero imaginary part of the @xmath256 . with this
initial conditions the time evolution can be calculated numerically , as shown in ( color online ) time evolution of eq .
( [ eq - gpe ] ) for an initial condition @xmath257 .
shown are squared modulus ( top ) and the phase of the first ( solid , black ) , second ( dashed , green ) and third ( dash - dotted , red ) mode . in the phase plot the first and second phase almost coincide and lie above the third phase which has a bigger phase velocity .
the time is measured with respect to @xmath258.,width=302 ] fig .
[ fig - gpevolution ] for initial values @xmath259 using eq .
( [ eq - gpe ] ) . in this example
the phases of @xmath260 and @xmath261 are locked , while @xmath262 evolves independently .
the difference in the amplitudes between mode 3 and the other two prohibits a coupling .
the amplitudes show a quite regular oscillation in all three modes .
this is motion of type ( c ) introduced in section [ sec - class ] .
physically interpreted , the wells 1 and 2 couple through josephson tunneling and the population between the two wells is exchanged periodically .
in contrast , the number of particles of well 3 stays approximately constant and much higher than the population of the other wells .
this behavior reflects the well - known macroscopic self - trapping found in the double well potential @xcite .
another type of this self - trapping effect in the type ( c ) dynamics can occur , when wells 1 and 2 have approximately the same population @xmath263 and well 3 is nearly empty .
one can also observe the other types of dynamics in the vectors @xmath264 , except type ( d ) , due to the very weak indirect coupling between modes 1 and 3 .
the different time evolutions @xmath264 can be easily assigned to the different guiding centers by looking at the phases : ( color online ) time evolution for an initial condition @xmath265 .
shown are squared modulus ( top ) and the phase of the first ( solid , black ) , second ( dashed , green ) and third ( dash - dotted , red ) mode . the time is measured with respect to @xmath258.,width=302 ] _ type ( a ) _ : all three phases behave independently and the amplitudes oscillate regularly .
the individual condensates in the different wells are completely decoupled and the population in each well stays approximately constant . _
type ( b ) _ : the dynamics shows the same behavior as for type ( c ) , but with phase locking between mode 2 and 3 .
_ type ( d ) _ : this type of motion is difficult to identify , because the indirect phase locking between modes 1 and 3 is very weak .
this leads to the effect that the phase velocities of these two phases are very close , but still distinguishable .
this is of course not a strict statement , and it depends on how long the time propagation is considered .
the problems with the classification of this type can also be seen in the quantum case in fig .
[ fig - psi1psi2 ] . in parts
( b ) resp .
( d ) , the phase singularities are not sharp but rather smooth , so these states could be assigned to type ( a ) as well . _ type ( e1 ) _ : in this case all three phases evolve with the same velocity and the amplitudes show similar regular oscillations as in types ( b ) and ( c ) for two locked phases . _
type ( e2 ) _ : this class is characterized by intermittencies as illustrated in fig .
[ fig - gpevolution2 ] .
the dynamics can be interpreted in such a way that the trajectories jump irregularly between different coupling schemes .
accordingly , frequency locking between different pairs of modes is only established temporarily during the time evolution . with this scheme
, we can classify the dynamics of all possible basis states @xmath264 , as shown in fig .
[ fig - basiscl ] .
the interesting point is that we can compare these results with the information that we extract from the semiclassical wave functions .
for this we compare for a given basis state @xmath62 all eigenfunctions ( [ eq - eigfred ] ) to which the basis state contributes significantly and assign a type ( a)(e2 ) to this basis state if possible .
the result is shown in fig .
[ fig - basisqm ] .
( color online ) characterization of the classical actions @xmath266 through direct numerical integration of eq .
( [ eq - gpe ] ) .
the action @xmath267 is given by @xmath268 .
plotted are time evolutions of type ( a ) ( @xmath269 , red ) , type ( b ) ( @xmath270 , green ) , type ( c ) ( @xmath271 , blue ) , type ( e1 ) ( @xmath272 , black ) and type ( e2 ) ( @xmath273 , cyan).,width=302 ] the points with no symbol indicate states which can not be assigned uniquely to a certain type . however , for the shown basis states one can see a close correspondence between the classical and the quantum system . only at the fringes
are there small deviations .
therefore the quantum mechanical analysis provides a grid of initial conditions for which we can predict the behavior of the solutions of the mean - field equations . finally
, we remark , that the classification of the basis states in fig .
[ fig - basiscl ] holds in principle also for an arbitrary choice of the initial phases in eq .
( [ eq - basisb ] ) . only at the fringes of the different zones
does the behavior of the time dynamics depend crucially on the initial conditions and there it can deviate from this classification .
( color online ) characterization of the classical actions @xmath266 through the semiclassical wave functions .
the action @xmath267 is given by @xmath268 . plotted
are actions whose quantum analog belongs to type ( a ) ( @xmath269 , red ) , type ( b ) ( @xmath270 , green ) , type ( c ) ( @xmath271 , blue ) , type ( d ) ( @xmath274 , magenta ) , type ( e1 ) ( * , black ) , and type ( e2 ) ( @xmath273 , cyan).,width=302 ] in the last section we discussed the close resemblance between the quantum and the classical picture by assigning the same characterization scheme with types ( a)(e2 ) to the basis functions and the solutions of the mean - field equations . in this section
we want to investigate , whether also the eigenstates of the quantum system can be reinterpreted classically , i. e. if they can be used to identify the different types of dynamical behavior in the system of the three bose - einstein condensates weakly coupled by josephson junctions . we construct the classical analog of eq .
( [ eq - eigf ] ) by defining the set of vectors @xmath275 which are related to the vectors @xmath276 by @xmath277 ( cf .
( [ eq - basisb ] ) ) .
however , note that the vectors @xmath278 , like the vectors @xmath253 of eq .
( [ eq - basisb ] ) , do not form a basis of @xmath279 . in analogy to eq .
( [ eq - eigexpansion ] ) one can write @xmath280 where the real - valued coefficients @xmath61 are taken from eq .
( [ eq - eigf ] ) .
( color online ) time evolution of the mean - field equations for an initial condition corresponding to the first quantum eigenstate .
shown are squared modulus ( top ) and the phase of the first ( @xmath281 , black ) , second ( @xmath282 , green ) and third ( @xmath283 , red ) mode . the time is measured with respect to @xmath258.,width=302 ] in this naive approach , the vector @xmath284 can be interpreted as the quantum expectation value of the action @xmath285 ( @xmath286 ) in the quantum state @xmath152 , @xmath287 where we have simply used the representation ( [ eq - eigexpansion ] ) of the eigenfunctions .
the initial phases are chosen equal zero like in the case of the basis vectors . in order to use this vector @xmath284 as initial conditions for the mean - field equations
, we must take the square root of each component , and to this end we define the new vector @xmath288 with components @xmath289 .
these vectors are normalized as @xmath290 where @xmath291 is the classically conserved total action of eq .
( [ eq - kcl ] ) . in the context of the gross - pitaevskii equation ,
the norm of the condensate wave function gives the number of particles in the condensate .
we get the additional term of @xmath292 for the number of particles compared to the many - particle hamiltonian ( [ eq - hqm ] ) , since we use the semiclassical correspondence of eq .
( [ eq - inrel1 ] ) . for bose - einstein condensates with a number of particles much larger than @xmath293
, one can ignore the term @xmath294 in eq .
( [ eq - inrel1 ] ) and obtain the standard correspondence between the particle numbers .
however , for @xmath59 , semiclassical studies like the present work show that the identification ( [ eq - inrel1 ] ) gives a much better agreement between classical and quantum mechanics . in order to obtain the normalization @xmath295 ,
one simply has to set @xmath296 and replace the nonlinearities @xmath297 by @xmath298 . in fig .
[ fig - gpeig1 ] , the time evolution for the initial condition @xmath299 is shown .
the time evolution shows approximately constant occupations @xmath300 ( upper panel ) , and the three phases are locked . in the reduced system
, this corresponds to a point in the neighborhood of a fixed point . for the parameter values chosen in this article
, there does not exist an exact fixed point of the hamiltonian flow of the reduced system , although this point serves as guiding center for the wave functions of type ( e1 ) . in that sense
the semiclassical wave functions behave very similarly in the neighborhood of a guiding center , while the solutions of the gross - pitaevskii equation are very sensitive to small deviations due to the nonlinearity of the time - evolution .
another example is shown in fig .
[ fig - gpeig444 ] for a type ( a ) motion .
( color online ) time evolution of eq .
( [ eq - gpe ] ) for the initial condition @xmath301 .
shown are squared modulus ( top ) and the phase of the first ( @xmath281 , black ) , second ( @xmath282 , green ) and third ( @xmath283 , red ) mode . the time is measured with respect to @xmath258.,width=302 ] the phases of the modes evolve independently and the amplitudes show tiny oscillations , due to the fact that the time evolution does not coincide with the corresponding idealized guiding center of type ( a ) . because the system is dominated by the anharmonicities the effective frequencies are almost linear in the actions according to eq .
( [ eq - omegaeff ] ) .
therefore the slopes of the phase curves are proportional to the average values of the corresponding actions . to conclude , from the classical point of view the analysis of the corresponding quantum system offers a direct visual method for the understanding of the structure and can be used to identify the dynamical behavior of the system of the three weakly coupled bose - einstein condensates in the mean - field approximation simply by diagonalizing the quantum hamiltonian and plotting the eigenfunctions in the appropriate basis .
in our investigation of a bose - einstein condensate in a multi - well potential , we showed a close correspondence between the quantum mechanical description and a classical version where the bosonic creation and annihilation operators of the many particle system are replaced by c - numbers .
we truncated the many - particle hamiltonian to a few relevant modes and obtained a system of three coupled anharmonic oscillators . whether the truncation at a small number of modes is justified depends crucially on an appropriate choice of the expansion basis and on the external potential . in order to compare the quantum system with its classical counterpart
, we introduced the concept of the semiclassical wave functions defined on the same toroidal configuration space as in the classical system .
this choice of the quantum mechanical representation allowed us to compare the quantum system directly with the classical system . in both cases , for the classical and the quantum system , we used the conserved particle number resp . total action to reduce the degrees of freedom to two .
classically , we can identify various geometric structures in phase space that are connected to different types of motion in the configuration space .
these different types of motion belonging to the various guiding centers , are also found in the quantum mechanical wave functions .
so we used these guiding centers firstly to sort a large number of wave function into these different classes , and secondly to assign uniquely geometric quantum numbers to the wave functions within one class . in this geometric picture , the wave functions describe the quantum excitations of the underlying classical dynamics .
as an application , we can characterize the entanglement between the different modes and we can also determine the number of particles in each of the entangled modes using their associated quantum numbers .
in the last part of this article we analyzed the significance of the quantum mechanical classification of the wave functions for the classical dynamics . for this
we studied classical trajectories which have initial conditions corresponding to quantum mechanical number states , or which correspond to the eigenstates directly . in both cases ,
we could obtain the characteristics of the semiclassical classification also from the classical trajectories , although the classical dynamics is much more sensitive to deviations from the idealized guiding centers .
concluding , we showed that semiclassical wave functions provide an intuitive picture of the quantum mechanical many - particle eigenfunctions , and allow a direct classification of the dynamics .
we thank h. s. taylor for interesting discussions . support by dgapa under grant number in-118005
is gratefully acknowledged .
we thank the anonymous referee for an unusually detailed and careful referee report which has helped us a lot to improve the final version of the manuscript .
this work was supported by a fellowship within the postdoc - programme of the german academic exchange service ( daad ) . | we present a new approach for the analysis of bose - einstein condensates in a few mode approximation . this method has already been used to successfully analyze the vibrational modes in various molecular systems and offers a new perspective on the dynamics in many particle bosonic systems .
we discuss a system consisting of a bose - einstein condensate in a triple well potential .
such systems correspond to classical hamiltonian systems with three degrees of freedom .
the semiclassical approach allows a simple visualization of the eigenstates of the quantum system referring to the underlying classical dynamics . from this classification
we can read off the dynamical properties of the eigenstates such as particle exchange between the wells and entanglement without further calculations .
in addition , this approach offers new insights into the validity of the mean - field description of the many particle system by the gross - pitaevskii equation , since we make use of exactly this correspondence in our semiclassical analysis .
we choose a three mode system in order to visualize it easily and , moreover , to have a sufficiently interesting structure , although the method can also be extended to higher dimensional systems . |
the time variability of cosmic magnetic fields has always been an argument in favour of hydromagnetic dynamo action .
its understanding is crucial for insights in the interior dynamics of stars and planets .
the time dependence of convective dynamos is attributable to a non - stationary buoyancy flux as well as to a time dependent equilibration of the magnetic field .
the latter is subject of the study presented here .
how do dynamos saturate and in particular in which way is the saturation reflected in their time dependence ? in a general description ,
the infinite growth of a magnetic field due to an appropriate motion of a conducting fluid is inhibited owing to the backreaction of the lorentz force on the flow ; the resulting changes in the flow cause a reduction of dynamo action .
flows which are influenced by the lorentz force in this way are called saturated .
nevertheless , @xcite as well as @xcite demonstrate that saturated flows may lead to exponential growth of the magnetic field in a corresponding kinematic calculation . despite the fact that the magnetic field is saturated in the full non - linear system
, it can grow in a kinematic treatment , because both associated linearized stability problems are different .
the flows taken from a saturated dynamo simulation and then used in a kinematic calculation need only quench the growth of the particular magnetic field found in the nonlinear problem and can in principle allow others to grow .
as @xcite have pointed out there is at least one example , the benchmark dynamo case 1 @xcite , where the field taken from a saturated dynamo is also kinematically stable . in this study
, we show that there is in fact a whole class of saturated , chaotic , time - dependent dynamos for which the corresponding kinematic dynamo is stable . in order to assess kinematic stability in the sense explained above
we solve the mhd - equations for a boussinesq fluid in a rotating spherical shell . at the same time
we evolve a second passive tracer field using the induction equation .
while the tracer field experiences the self - consistent velocity field at each time step it does not contribute to the lorentz force .
this method has been used by @xcite for box simulations and a shell model and is closely related to the test - field method to determine mean - field coefficients @xcite . within a sample of 20 models ,
we identify two distinct dynamo regimes dependent on a modified rossby number @xcite in which the tracer field either grows exponentially or reaches a state aligned with the actual self - consistent magnetic field after an initial transition period .
moreover , differences in the kinematic stability of the dynamos are linked to differences in their time variability : exclusively kinematically unstable dynamos in the high rossby number regime show polarity reversals of the axial dipole field .
we attribute the complicated time behaviour of these models to an alternating growth of many competing dynamo modes . on the other hand
, the eigenvalue computation suggests that dynamos with low rossby number are dominated by only one fundamental mode which is repeatedly quenched and rebuilt .
all other modes in this case are clearly subcritical . in this sense ,
dynamo models in the low rossby number regime , i.e. fast rotators , exhibit a simple time dependence and their time - variability consists of fluctuations about their critical state .
we consider an electrically conducting boussinesq fluid in a rotating spherical shell and solve the mhd - equations as given by @xcite and described in detail by @xcite .
in addition , we compute the evolution of a passive tracer field with the help of a second induction equation @xmath0 while the tracer field , @xmath1 , experiences the self - consistent velocity field @xmath2 in each time step , it does not contribute to the lorentz force .
hence it does not act on the velocity field and is `` passive '' in this sense .
the initial conditions for the tracer field have been chosen arbitrarily with the help of a random number generator . moreover , for models 1015 , we added some random noise to the tracer field in each time step .
this enables us to perturb the tracer field permanently and prevents it from becoming aligned with the actual , self - consistent field . in these simulations
, we advance the tracer field for at least 10 magnetic diffusion times in order to test for kinematic stability . according to the scaling we used ,
the equations are governed by four parameters .
these are the ekman number @xmath3 , the ( modified ) rayleigh number @xmath4 , the prandtl number @xmath5 and the magnetic prandtl number @xmath6 . in these expressions
, @xmath7 denotes the kinematic viscosity , @xmath8 the rotation rate , @xmath9 the shell width , @xmath10 the thermal expansion coefficient , @xmath11 is the gravitational acceleration at the outer boundary , @xmath12 stands for the temperature difference between the inner and outer spherical boundaries , @xmath13 is the thermal and @xmath14 the magnetic diffusivity with the magnetic permeability @xmath15 and the electrical conductivity @xmath16 .
all four parameters have been varied to build up a sample of 20 dynamo models , see table 1 .
output parameters used here in order to interpret the results are the magnetic reynolds number , @xmath17 , the elsasser number , @xmath18 , and the rossby number , @xmath19 . in these expressions , @xmath20 and @xmath21 denote rms - values of the velocity and the magnetic field inside the shell , respectively , and @xmath22 is the density .
furthermore , we adopt the definition of a local rossby number proposed by @xcite , @xmath23 here , @xmath24 is the mean half wavelength of the flow and @xmath25 is the mean harmonic degree derived from the kinetic energy spectrum , @xmath26 the brackets in eq .
( [ eq:3 ] ) denote an average over time and radii , @xmath27 stands for the velocity component of harmonic degree @xmath28 .
versus local rossby number @xmath29 .
stars denote non - reversing dynamos which are kinematically stable , whereas triangles represent dynamos which do reverse and are kinematically unstable .
both regimes coincide with the dipolar and multipolar dynamo regimes identified by christensen and aubert ( 2006 ) .
there is one example ( diamonds ) , model 15 , which undergoes a transition between both regimes .
note that this example has a considerably lower relative dipole field strength in its second state . ] within our 20 examples ( see table [ tab1 ] ) we find 5 dynamos which are kinematically unstable and 14 which are kinematically stable .
one example ( model 15 ) belongs to both classes ; although in general unstable , the tracer field does not grow within certain periods of several magnetic diffusion times .
note that all dynamos considered here operate in the so called strong field regime , i.e. the elsasser number is of order unity or larger .
the equatorial symmetry is broken for most of the kinematically stable and all unstable models . except for model 1 , the quasi - steady benchmark dynamo @xcite ,
all models exhibit highly time - dependent or even chaotic behaviour .
this is demonstrated in fig .
[ figure7 ] for model 2 , the next simplest example to the benchmark dynamo .
this dynamo appears to be chaotic , and as an experiment we performed two simulations starting from almost identical initial conditions ( the difference between two initial conditions is a small deflection of the magnetic dipole axis in the second run ) .
the evolution from both initial conditions is shown in fig .
[ figure7 ] where the magnetic energy densities can be seen to diverge rapidly with time . , and the actual magnetic field , @xmath30 , versus time for model 8 ( dashed line ) and model 19 ( solid line ) . ] where @xmath31 is the outer shell radius .
note that the tracer field is completely aligned with the actual magnetic field .
both components are normalised due to their maxima and minima .
therefore the greyscale coding varies from -1 , white , to + 1 , black , and the contour lines correspond to @xmath32 . following contour plots
are presented in the same style . ] .
contour lines : see figure [ figure2 ] . ]
the regimes of kinematically stable and unstable dynamos can be clearly distinguished by the modified rossby number ( see table [ tab1 ] ) , @xmath29 .
models with low rossby number are kinematically stable whereas the tracer field grows exponentially for dynamos in the high rossby number regime .
the transition between both regimes occurs at @xmath33 .
there are two further properties related to both regimes which deserve mentioning .
all dynamos we found to be kinematically stable are dipolar and do not show any polarity reversals , while dynamos in the second regime are multipolar and do reverse .
this is also illustrated in fig .
[ figure1 ] . here
, the relative dipole field strength , @xmath34 , on the outer shell boundary is plotted versus the modified rossby number , @xmath29 ; @xmath34 is defined as the time - average ratio of the dipole field strength to the field strength in harmonic degrees 1 to 12 .
both regimes visible in fig .
[ figure1 ] coincide with those identified earlier by @xcite .
figure [ figure4 ] compares the magnetic energy densities of the tracer field for a kinematically stable ( model 8) and a kinematically unstable dynamo ( model 19 ) , varying with time .
while the tracer field grows rapidly after an initial transient phase in the latter case , it reaches a state aligned with the actual field if the dynamo is kinematically stable .
then , the energy density of the tracer field normalised with the energy density of the actual self - consistent field approaches a constant level which depends only on the initial conditions .
this is also confirmed by looking at the corresponding field configurations .
figure [ figure2 ] displays the radial component of the tracer field for model 8 , which differs from the actual field only by an overall scale factor .
therefore , only one contour plot is given .
on the other hand , although they have similar spatial scales , both field components are clearly not aligned but very different for model 19 ( see fig . [ figure3 ] ) .
[ cols="<,^,>,>,>,^,>,<,>,>,>",options="header " , ] model 15 is in general kinematically unstable but also exhibits periods of several magnetic diffusion times in which the tracer field stays stable .
according to its local rossby number , @xmath35 , it is located close to the boundry between both dynamo regimes and undergoes transitions from one to the other .
( dashed line which runs out of the figure at roughly 2.3 magnetic diffusion times ) .
as soon as the dynamo reverses it becomes kinematically unstable . ]
we could detect transitions from a kinematically stable to an unstable state ( see fig .
[ figure4a ] ) and vice versa .
as long as the tracer field remains stable , the tilt angle of the dipole axis fluctuates about the actual polarity state .
however , when the tracer field becomes unstable , also the polarity of the dipole field starts to reverse .
this coincidence is observed for transitions in both directions , i.e. the tilt angle of the dipole axis also stabilises when model 15 becomes intermittently stable . while the magnetic field is quite dipolar with @xmath36 for periods in which the polarity and the tracer field are stable the relative dipole field strength decreases drastically to @xmath37 otherwise .
the strong connection of field morphology , time dependence and saturation is not only present separately in several models but manifests itself in the time variation of a single dynamo model , too .
the existence of kinematically unstable dynamos was expected @xcite .
the finding of a class of kinematically stable but yet time - dependent or even chaotic dynamos , however , needs some further explanation .
the lack of growing modes for these models already suggests that almost all field configurations for the tracer field are decaying , except the one aligned with the actual , self - consistent field .
but this component of the tracer field is quenched by the saturated velocity field .
thus , the tracer field follows the actual field with time , apart from a different , arbitrary amplitude due to the linearity of the induction equation .
this interpretation is confirmed by looking at the spectrum of the time and azimuthally averaged dynamo operator @xmath9 , @xmath38 with eigenmodes @xmath39 and eigenvalues @xmath40 . in this , the operator @xmath9 is defined as @xmath41 note that @xmath9 , also known as mean - field dynamo operator @xcite , contains the mean velocity field @xmath42 as well as the so called mean - field coefficients @xmath43 and @xmath44 , which are tensors of second and third rank , respectively .
as noted by @xcite , these quantities appear inevitably as a consequence of averaging .
they depend on the velocity field and the magnetic diffusivity of the considered dynamo model only and have been determined with the help of the test field method @xcite . a detailed discussion on the applicability of mean - field concepts to direct numerical simulations of rotating magnetoconvection and a ( quasi-)stationary dynamo is provided by @xcite . a similar discussion for time dependent dynamos is not given here but will be subject of a forthcoming paper .
a recent review on the test - field method and its applications has been given by @xcite .
eigenvalues and eigenmodes of @xmath9 have been computed as reported by @xcite for model 2 . in fig .
[ figure5 ] the radial components of the first three ( dipolar ) eigenmodes , @xmath39 , are displayed .
all modes decay exponentially ; this had to be expected for kinematically stable dynamos ( see also the discussion in * ? ? ?
however , the decay rates are given here in units of @xmath45 , in which the molecular diffusivity @xmath46 is about 30 times smaller than the turbulent one inferred from components of @xmath44 .
thus , @xmath47 is much larger than one effective diffusion time and the first , fundamental , eigenmode is indeed close to its critical state . due to a noticable gap in decay rates after the fundamental mode , @xmath48
, this is not equally true for the subsequent eigenmodes .
they are much more diffusive , thus leaving the fundamental mode as the preferred eigenstate of the dynamo .
hence , the time dependence of model 2 may be understood in parts as a fluctuation of the fundamental mode about its critical state . of the time
averaged dynamo operator for model 2 .
the corresponding eigenvalues are @xmath49 and @xmath50 in units of @xmath45 .
note the huge drop in decay rates after @xmath51 . ]
the dominance of the first eigenstate is also revealed by a decomposition of the actual , time - dependent magnetic field of model 2 in a set of eigenmodes @xmath39 of @xmath9 , @xmath52 the time - dependent and in general complex mode coefficients @xmath53 have been computed as @xmath54 in which @xmath55 denotes the adjoint of the current @xmath56 , and @xmath57 is the vector potential of the actual , time - dependent field , @xmath58 .
the integration is carried out over the whole fluid domain @xmath59 . for a derivation of eq .
( [ eq:7 ] ) we refer to @xcite and @xcite
. already contributes 75% up to 85% to the total amount and its time variability reflects much of the time dependence of the axisymmetric magnetic field . ] in fig .
[ figure6 ] the energy contribution of the fundamental eigenmode @xmath60 is compared with the total axisymmetric magnetic energy density .
the fundamental eigenmode contributes at least 75% up to 85% to the total amount , revealing again its permanent dominance throughout the simulation .
the equilibration process for model 2 has been studied earlier by @xcite , too .
they found that in regions with high magnetic energy density , the lorentz force simply reduces locally the flow velocity without changing the overall pattern of convection .
they investigated possible changes in the velocity , if the magnetic field and thus the lorentz force is arbitrarily reduced at some instant in time and then recovers towards its equilibration value .
the kinematic effects relevant for dynamo action identified by them , an @xmath61-effect from helicity in the columnar convection and an anti-@xmath62 effect from the mean azimuthal flow , were present in the same proportions close and far from equilibrium conditions of the magnetic field .
their finding is supported by the study presented here .
saturation may reduce the amplitudes of @xmath61 and thus the growth rates of the eigenstates of the related dynamo operator , but does not change their relative order .
therefore , the preferred eigenstate stays the same throughout the simulation .
this is clear as there is a large gap between the growth rate of the fundamental eigenmode of the time - averaged dynamo operator @xmath9 and all other eigenmodes , as mentioned above .
so far we only analysed model 2 in detail . here
the velocity field is nearly symmetric with respect to the equatorial plane and the magnetic field belongs to the dipolar family .
contributions of quadrupolar type are not present . in a more complicated example with broken equatorial symmetry , we expect the fundamental quadrupolar mode to be excited . although its growth rate will be smaller than the one for the fundamental dipolar mode and probably subcritical , it is typically of the same order . in such a case a clear dominance of only one fundamental dipolar mode can no longer be deduced from the spectrum of the time - averaged dynamo operator @xmath9 , and a second , quadrupolar mode may become important . for the models in the high rossby number regime the findings of @xcite apply .
these models act kinematically as dynamos and the dynamo operator @xmath9 possesses in general growing eigenmodes .
a kinematic treatment of these dynamos does not reveal their actual time dependence .
however , the results presented here suggest that the regime of dipolar dynamos identified by @xcite is kinematically stable . for these models ,
the quenching of any magnetic field is fully captured in the velocity field and a kinematic treatment may indeed reproduce their actual time - dependence .
models of this dynamical regime are applicable to planetary dynamos and probably also to dynamos of fast rotating stars @xcite , thus covering a large range of magnetic reynolds numbers . hence , an attempt to explain the kinematic stability of these models due to a magnetic reynolds number which is close to its critical value fails .
we emphasise again that the transition between both regimes is governed by the local rossby number @xmath29 and not by @xmath63 , as can be already seen from table [ tab1 ] . in the low rossby number regime
, the rotational constraint leads to columnar structured flows , dipolar magnetic fields and finally to a rather simple time dependence , although these models operate in general far away from the dynamo threshold at @xmath64 .
dipolar dynamo models which show occasionally reversals are located close to the regime boundary in fig .
[ figure1 ] , with @xmath65 .
they resemble the geodynamo in many respects and are therefore of particular interest . explaining polarity reversals of an otherwise predominantly dipolar field , @xcite suggest that the geodynamo crosses the boundary towards the multipolar dynamo regime from time to time . with the help of scaling laws derived from numerical models ,
they indeed succeed in predicting a local rossby number of @xmath66 for the earth s core . adopting this viewpoint
we link the occurence of geomagnetic reversals to a change in the saturation process .
the quenching of a previously dipolar field may result in the preference of different , higher order modes if inertia gains importance in comparison to the coriolis force , and the dynamo undergoes an excursion into the kinematically unstable regime .
subsequently the dipole field is built up again , but it may have either polarity .
a computation of eigenmodes and a mode decomposition similar to ( [ eq:7a ] ) for model 15 seems to be a promising approach to confirm this picture .
note that from the viewpoint we take here , the existence of dipolar , stable periods for model 15 demands more explanation than the fact that it reverses .
however , whether inertia is indeed as important for the geodynamo as it is for present dynamo models is under debate ( e.g. * ? ? ?
in fact , the assumption of @xmath66 for the earth s core leads to a characteristic length scale of only a few hundred meters , on which the magnetic field would be highly diffusive @xcite .
fast rotating dynamos , characterised by a low rossby number , are kinematically stable . within this regime , a saturated velocity field taken from dynamo simulations does not lead to exponential growth of the magnetic field in a corresponding kinematic calculation .
hence , saturation may be understood as a quality of the velocity field , only . for these dynamos , saturation results in the unchanged preference of a fundamental eigenstate , whereas different eigenmodes may supersede each other if inertia gains importance .
this difference in the saturation process involves differences in the morphology of the magnetic field and its time dependence .
kinematically stable dynamos are dipolar and exhibit a rather simple time variability , which may be interpreted as the fluctuation of the fundamental mode about its critical state .
kinematically unstable dynamos are much more complicated .
the alternating growth of various modes leads to a multipolar field morphology and polarity reversals of the dipole field appear as a natural consequence .
we thank ulrich christensen and johannes wicht for many interesting discussions and support .
brandenburg , a. , 2009 .
advances in theory and simulations of large - scale dynamos , _ space sci .
_ , * 144 * , 87104 .
cattaneo , f. & tobias , s.m . , 2009 .
dynamo properties of the turbulent velocity field of a saturated dynamo ,
_ j. fluid mech .
_ , * 621 * , 205214 .
christensen , u.r . ,
aubert , j. , cardin , p. , dormy , e. , gibbons , s. , glatzmaier , g.a .
, grote , e. , honkura , y. , jones , c. , kono , m. , matsushima , m. , sakuraba , a. , takahashi , f. , tilgner , a. , wicht , j. & zhang , k. , 2001 . a numerical dynamo benchmark , _ phys .
_ , * 128 * , 2534 .
christensen , u.r .
& aubert , j. , 2006 .
scaling properties of convection - driven dynamos in rotating spherical shells and application to planetary magnetic fields , _ geophys .
j. int . _ ,
* 166 * , 97114 .
christensen , u.r .
& wicht , j. , 2007 .
numerical dynamo simulations , in _ treatise on geophysics , vol . 8 _ , pp . 245282 , ed .
schubert g. , elsevier , amsterdam .
christensen , u.r .
, holzwarth , v. & reiners , a. , 2009a .
energy flux determines magnetic field strength of planets and stars , _ nature _ , * 457 * , 167169 .
christensen , u.r . ,
schmitt , d. & rempel , m. , 2009b .
planetary dynamos from a solar perspective , _ space sci .
_ , * 144 * , 105126 .
hoyng , p. , 2009 .
statistical dynamo theory : mode excitation , _ phys .
e _ , * 79 * , 046320 , 113 .
krause , f. & rdler , k .- h . , 1980 .
mean - field magnetohydrodynamics and dynamo theory , pergamon press , oxford .
olson , p. & christensen , u.r .
dipole moment scaling for convection driven planetary dynamos , _ earth planet .
_ , * 250 * , 561571 .
olson , p. , christensen , u.r .
& glatzmaier , g.a . , 1999 .
numerical modeling of the geodynamo : mechanisms of field generation and equilibration , _ j. geophys .
_ , * 104 * , 1038310404 .
schrinner , m. , rdler , k .- h .
, schmitt , d. , rheinhardt , m. & christensen , u.r . , 2005 .
mean - field view on rotating magnetoconvection and a geodynamo model , _ astron .
_ , * 326 * , 245249 .
schrinner , m. , rdler , k .- h .
, schmitt , d. , rheinhardt , m. & christensen , u.r . , 2007 .
mean - field concept and direct numerical simulations of rotating magnetoconvection and the geodynamo , _ geophys .
fluid dynam .
_ , * 101 * , 81116 .
schrinner , m. , schmitt , d. , jiang , j. & hoyng , p. , 2009 . a new method for computing the eigenfunctions and their adjoints of the dynamo operator , _ mon .
r. astron .
_ , submitted .
sreenivasan , b. & jones , c.a . , 2006 .
the role of inertia in the evolution of spherical dynamos , _ geophys .
_ , * 164 * , 467476 .
tilgner , a. & brandenburg , a. , 2008 . a growing dynamo from a saturated roberts flow dynamo , _ mon . not .
r. astron .
_ , * 391 * , 14771481 . | in this study we address the question under which conditions a saturated velocity field stemming from geodynamo simulations leads to an exponential growth of the magnetic field in a corresponding kinematic calculation .
we perform global self - consistent geodynamo simulations and calculate the evolution of a kinematically advanced tracer field .
the self - consistent velocity field enters the induction equation in each time step , but the tracer field does not contribute to the lorentz force .
this experiment has been performed by @xcite and is closely related to the test field method by schrinner et al .
( 2005 , 2007 ) .
we find two dynamo regimes in which the tracer field either grows exponentially or approaches a state aligned with the actual self - consistent magnetic field after an initial transition period .
both regimes can be distinguished by the rossby number and coincide with the dipolar and multipolar dynamo regimes identified by @xcite .
dipolar dynamos with low rossby number are kinematically stable whereas the tracer field grows exponentially in the multipolar dynamo regime .
this difference in the saturation process for dynamos in both regimes comes along with differences in their time variability . within our sample of 20 models ,
solely kinematically unstable dynamos show dipole reversals and large excursions .
the complicated time behaviour of these dynamos presumably relates to the alternating growth of several competing dynamo modes .
on the other hand , dynamos in the low rossby number regime exhibit a rather simple time dependence and their saturation merely results in a fluctuation of the fundamental dynamo mode about its critical state .
[ firstpage ] dynamo : theories and simulations ; earth s core ; geomagnetic field ; magnetohydrodynamics . |
it is now generally agreed that the x - ray variability properties of @xmath3few solar mass black holes in x - ray binary systems ( xrbs ) and supermassive black holes in active galactic nuclei ( agn ) are similar . with a view to providing an independent method for determining physical parameters such as black hole mass and accretion rate
, there is considerable interest in determining how these parameters might scale with characteristic observable x - ray variability parameters .
x - ray variability can be quantified via the power spectral density ( psd ) and , to first order , the x - ray psds of agn are similar to those of xrbs @xcite , particularly to xrbs in the high - soft state ( e.g. * ? ? ?
* ; * ? ? ?
* ) . in this state the power , @xmath4 , at frequency , @xmath5 ,
is given by @xmath6 where @xmath7 at low frequencies bending , above a frequency @xmath8 , to a slope @xmath9 .
many authors have shown that the timescale @xmath10 , corresponding to the frequency @xmath8 , scales approximately linearly with mass ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) although the relationship shows considerable scatter .
however much of the scatter can be explained by an inverse scaling of with accretion rate ( where is the accretion rate in units of the eddington accretion rate , * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
the other variability parameter which has attracted attention is the normalisation of the high frequency ( hf ) psd .
this normalisation can be defined in a number of ways .
for example , by fitting a power law to the hf psds , after subtraction of the poisson noise contribution , @xcite measured the power at a fixed frequency ( @xmath11hz ) .
the normalised variability amplitude , nva , was then defined as the square root of the power , which was measured in units of ( counts s@xmath12)@xmath13 hz@xmath12 , divided by the average count rate .
the nvas for a sample of agn observed by exosat were therefore derived .
although black hole mass was assumed to be a driving parameter , in 1988 very few agn black hole masses were available so the nvas were plotted against luminosity as a proxy , showing a strong anticorrelation .
@xcite , using ginga observations , measured the timescale at which the psd crossed a particular power level ( @xmath14 , in rms@xmath13 hz@xmath12 units ) . scaling this timescale with mass from cyg x-1 , they estimated masses for 8 agn , although they noted that these masses were one or two orders of magnitude lower than masses derived by other methods .
@xcite defined an hf psd normalisation , @xmath15 , which is very similar to the nva of @xcite although , as most of their measurements were of xrbs , @xcite defined @xmath16 at the higher frequency of 1hz .
they also assumed @xmath17 whereas @xcite and @xcite measured @xmath18 . within the xrb sample of @xcite
there is no correlation of @xmath15 with mass ( their fig.7a ) , although the range of masses is small and mass uncertainties large . within their sample of agn ( their fig.7c )
there is an approximate inverse correlation of @xmath15 with mass , strengthened if ngc4395 is included ( fig.7b ) .
however the scatter is too large to define the relationship precisely within that sample .
however if it is assumed that agn and xrbs follow the same scaling relationship then @xcite find that @xmath19 .
@xcite fit a mixed ornstein - uhlenbeck model to cyg x-1 and to agn psds . in this model
the hf psd normalisation is defined by @xmath20 , the fractional amplitude of the driving noise field .
the psd slope is again fixed at @xmath17 with @xmath21 . for a sample of 10 agn
they find a strong correlation with @xmath22 ( 90 per cent confidence ) .
however such a relationship would not extrapolate to the xrb observations and so it is important to investigate the relationship between agn hf psd normalisation and mass using other datasets .
although less direct , the psd normalisation can be estimated from the normalised excess variance of the lightcurves , @xmath23 ( e.g. * ? ? ?
a number of authors ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
* ) have found an approximate inverse scaling of @xmath23 with mass .
if all of the frequencies sampled by the lightcurves lie above , and the hf psd slope is known , then gives a reasonable estimate of the hf psd normalisation , although @xcite show that is a noisy quantity when the psd is steep .
if is not known and lies within the frequencies sampled , or if the low frequency psd slope is unknown , then is a less good measure of hf psd normalisation .
the best way , therefore , to determine whether hf psd normalisation varies with mass is to derive hf psd normalisations directly from the psds . the hf psd is best determined from continuous long observations which suffer less from distortion by the window function of the sampling pattern than do observations which are split into many segments .
thus , which allows continuous observations of up to 130ksec , provides a better measurement of the hf psd ( e.g. * ? ? ?
* ; * ? ? ?
* and many others ) than do satellites in low earth orbits with maximum continuous observation lengths of @xmath24s , i.e. ginga ( e.g. * ? ? ?
* ) , asca ( e.g. * ? ? ?
* ) or rxte .
exosat was also useful as it allowed continous observations of up to 280ksec .
the exosat low energy ( le ) imaging telescope is not comparable in sensitivity at low energies ( 0.1 - 2 kev ) to . however the exosat medium energy proportional counter ( me ) is comparable at the higher energies ( approximately 1 - 9 kev ) where , also , the effects of absorption which may affect measurements of x - ray flux variations , are less .
in addition , the me has made long observations of some bright agn ( eg 3c273 , cena , ngc5506 ) for which long observations by do not yet exist . in this paper
the exosat me observations of agn are re - examined to determine whether they reveal any scaling of hf psd normalisation with mass .
@xcite provided initial nva measurements for a sample of agn observed by exosat , but a more precise analysis of all exosat me agn observations longer than 20ksec ( 32 agn in total ) was carried out by @xcite .
@xcite removed the small window artefacts from exosat psds using their own 1d version of the ` clean ' algorithm @xcite and hence derived nvas .
errors were determined from simulations .
@xcite confirmed the inverse scaling of nva with luminosity , suggesting that the lower nvas arose in larger mass objects from a larger emitting region , and also found a positive correlation of nva with photon energy index . where variability was not detected , ie the hypothesis that the lightcurve was constant
could not be rejected at the 95 per cent confidence level at least 90 per cent of the time , @xcite provide an nva upper limit , deduced from the photon counting noise level of the lowest noise observation .
the reader is referred to @xcite for full details .
.nvas from @xcite for agn with detected variability in exosat me observations .
black hole masses are from 1 @xcite , 2 @xcite , 3 @xcite , 4 @xcite , 5 @xcite , 6 @xcite , 7 @xcite . [ cols="<,>,^,^ " , ] in table [ tab : detect ] the nvas where variability was detected are listed together with the most accurate black hole masses available .
masses derived from stellar dynamical observations ( ngc3227 and ngc4151 ) , which are not subject to the uncertainty regarding the ` f - factor ' used to convert dynamical products from reverberation observations to black hole masses , are taken preferentially . as dynamical measurements are rare , reverberation masses are next taken ( 3c120 , ngc3783 , ngc4051 , 3c273 and ngc4593 ; see the caption to table 1 ) . for mcg-6 - 30 - 15
the mass estimated by @xcite is used .
this mass is the mean of the mass derived from the width of the stellar absorption lines and from the emission line width . for ngc5506 and ngc7314
the mass is derived from the stellar absorption line widths of 98 and 60 km s@xmath12 respectively @xcite using , as for mcg-6 - 30 - 15 , the @xmath25 relationships of @xcite and @xcite .
the latter , particularly , has been shown by @xcite to provide a good mass estimate for low mass systems .
the mean of the masses derived from these two relationships is taken and a 50 per cent uncertainty is assumed .
no reliable mass estimator could be found for mcg+8 - 11 - 11 and so this agn is not considered further . in table [ tab : undetect ] the nva upper limits for sources where variability was not detected are listed .
black hole masses are also given .
a simple maximum likelihood ( ml ) analysis , using the function within the qdp plotting package which takes only the errors on nva , was applied to the agn with detected variability ( except for mcg+8 - 11 - 11 ) .
this analysis shows that the relationship @xmath26 , with @xmath27 ( 90 per cent confidence ) is a good description of these data .
the qdp ml w - var statistic , which is similar to @xmath28 , is 6.1 for 8 d.o.f . , indicating that the errors are overestimated . to determine whether @xmath2 varies with the method of fitting ,
a number of other methods were examined , particularly fitexy @xcite and bayesian regression @xcite . in the fitexy method ,
errors in both directions are included .
the best value and uncertainty in @xmath2 were determined by measuring the minimum value of @xmath28 as a function of @xmath2 , with free normalisation ( fig .
[ fig : slopes ] ) .
the lowest overall @xmath28 of 4.24 corresponds to @xmath29 . for 8 d.o.f .
, this @xmath28 value confirms that the errors are over - estimates .
the 1@xmath30 ( ie @xmath31=1 ) uncertainty is 0.08 . the 90 per cent confidence uncertainty ( ie @xmath31=2.7 ) is 0.12 .
the bayesian regression analysis of @xcite , as implemented within idl gives an almost identical value of @xmath32 .
the standard deviation of 0.10 is slightly larger , probably because bayesian analysis takes account of the fact that the observed agn are just a random sample drawn from the parent distribution , a consideration addressed here in section [ sec : sample ] .
thus the results do not depend on the fitting method .
the best fit from the fitexy method is shown in fig .
[ fig : nva ] . unless stated otherwise , fitexy results will be quoted hereafter .
the nva upper limits are not included in this fit but are consistent with the fit . in principle the nva upper limits can be used to refine the value of @xmath2 .
however the result depends on how the ` upper limits ' are interpreted and what probability density functions ( pdfs ) are assigned to them , eg whether they are hard limits above which there is no probability of finding the datum , or limits with an associated measurement error .
eg the methodology of @xcite , as implemented within idl , treating the limits as hard , gives @xmath33 .
allowing some error in the limit allows slightly flatter values .
however as it is not clear how the limits of @xcite should be interpreted , it is concluded only that the upper limits are consistent with values of @xmath2 derived from detections , but not favouring values of @xmath2 much flatter than -0.5 . for almost all of the present sample
it is now known that is well below @xmath34hz ( e.g. see * ? ? ?
* ) and so the nvas reported by @xcite are a good measurement of the normalisation of the hf psd . in addition , where psd slopes could be reasonably measured
, @xmath18 is mostly close to 2 , as expected for the hf part of the psd .
however for ngc4051 @xmath35hz @xcite and the exosat observation length ( 207ks ) is long enough that a single power law fit would be noticeably flattened by the psd values below , where @xmath36 .
re - examination of the exosat psd shows that the nva could be underestimated by up to 50 per cent . with a corrected nva , @xmath37 . parameterising log(nva)=@xmath2log(m ) + @xmath38 , then here @xmath39 .
the only other source where a significant underestimation of the nva might be expected , based on its low mass , is ngc7314 .
however the exosat observation is only 22ks and so the hf psd is unlikely to be noticeably distorted . even allowing for an extreme 50 per cent underestimate similar to that for ngc4051 , and also allowing for the same underestimate in ngc4051 , @xmath2 increases only to @xmath40 .
measurement errors in the covariate , in this case mass , bias the slope towards zero ( e.g. * ? ? ?
although not discussed in detail here , that bias can be addressed via simulations where each data point is represented by a 2d pdf .
random selection of simulated data points from those pdfs gives values of @xmath2 which are typically flatter by 0.03 to 0.04 than those obtained by fitexy .
as the simulations include the measurement error twice , i.e. once in the value of the data point itself and once in the distribution of the pdf , but fitexy only includes the error once , the true values of @xmath2 may be @xmath41 steeper than those given above .
it is often questioned whether the x - ray emission mechanism in 3c273 is similar to that in seyfert galaxies , ie thermal compton scattering from a non - beamed corona , or whether it is synchrotron or synchrotron self - compton emission from a relativistic jet ( e.g. * ? ? ?
* ; * ? ? ?
* ; * ? ? ?
3c273 shows a bend in its long term psd which is consistent with arising from the same variability process as in seyfert galaxies @xcite but it is still unclear whether it should be included together with seyfert galaxies and non - blazar radio galaxies ( 3c120 ) , ie the remainder of table [ tab : detect ] , in any x - ray timing survey .
the fit was therefore repeated without 3c273 , providing a similar slope ( @xmath42 from fitexy or @xmath43 from bayesian regression ) whose extrapolation passes well within the error range for 3c273 .
these results suggest that the same process that drives x - ray variability in seyferts also drives variability in 3c273 although the x - ray emission location or emission mechanism need not be the same in both cases . with small samples such as that listed in table [ tab : detect ] , the results can depend on the choice of the values of the measured variables and/or of the sample content .
for example for mcg-6 - 30 - 15 one might take the mass derived only from the stellar absorption lines ( @xmath44 , * ? ? ?
* ) , which would not alter @xmath2 . or one might decide to take masses derived from reverberation for ngc3227 ( @xmath45 , * ? ? ?
* ) and ngc4151 .
the value of @xmath2 is again almost unchanged ( @xmath46 or @xmath47 if 3c273 is excluded ) although the fit , whilst remaining good , is very slightly worse .
one might even decide to exclude mcg-6 - 30 - 15 , ngc5506 and ngc7314 altogether , giving @xmath48 .
it is finally noted that @xcite propose a revision of black hole masses . using their ` classical ' ( ie all morphological type ) @xmath25 relationship ,
the masses of mcg-6 - 30 - 15 and ngc5506 would reduce by almost exactly a factor 2 , and that of ngc7314 by a factor 3 .
all masses based on reverberation mapping would reduce also by a factor 2 due to their downward revision of the ` f - factor '
. as these changes are almost entirely systematic , the various fits for @xmath2 are almost unchanged with typical values flattening by only @xmath49 ( eg from -0.54 to -0.52 ) with , if anything , very slightly reduced dispersion .
the overall conclusion is that @xmath2 is robust to minor changes in sample parameters with a value close to -0.54 .
it is shown here that nva , defined by @xcite and derived more precisely from exosat me observations by @xcite , correlates very well , within a sample of agn alone , with black hole mass . for a relationship @xmath50 then , even allowing for different choices of mass or sample content , @xmath2 remains close to -0.54 with a dispersion of @xmath51 .
the main contribution of the present work is thus to show that a close to inverse linear scaling of hf psd normalisation with mass , proposed by @xcite on the assumption that agn and xrbs follow the same scaling , occurs independently within a sample of agn alone .
the large scatter within the agn @xmath15 values of @xcite may have arisen because their values of @xmath15 were derived from earlier measurements of x - ray excess variance by satellites with large orbital gaps in their lightcurves , ie asca and rxte .
variations in accretion rate may perhaps add further scatter . within their sample of xrbs ,
@xcite show that @xmath15 is not constant for any given xrb , although they note that the range of variability with flux is not large .
they do , however , note that the value of @xmath15 in the soft state of cyg x-1 is larger than in the hard state which may indicate some variation of hf psd normalisation with accretion rate or state .
however the range of accretion rates within the sample listed in table [ tab : detect ] is not large .
thus , as there is no evidence for dispersion in the present value of @xmath2 over and above that expected from the errors in the mass and nva measurements , possible variation of nva with accretion rate are not considered further here although they can not be ruled out .
finally , the differences between the results of @xcite and those presented here are considered .
the equivalent value of the slope @xmath2 derived by @xcite from bayesian regression is @xmath52 ( 90 per cent confidence ) whereas the value presented here is @xmath53 ( 1@xmath30 , or @xmath54 at 90 per cent confidence ) , or -0.57 if ngc4051 is corrected .
possible reasons for the differences in slope include differences in analysis methods and in sample content .
regarding analysis methods , for the present sample all regression methods produce the same value of @xmath2 .
another analysis difference is that the nva values are derived from simple fits to the hf psds whereas the values of @xmath20 are derived from fitting a particular model , the mixed ornstein - uhlenbeck model , to the datasets .
although parameters in multiparameter fits usually interact , there is no obvious reason why this model should result in larger values of @xmath2 . similarly fixing the hf psd slope at @xmath17 ( @xcite ) compared with measuring @xmath17 ( @xcite ) and then correcting the nva
if @xmath18 is lower than 2 should have little effect .
large differences in sample content may , however , have a large effect on the value of @xmath2
. only 6 of the 10 agn in each sample are in common .
@xcite include the high accretion rate nls1s mkn766 and ark564 ( whose mass is uncertain ) .
however these two nls1s are not included here where the spread in accretion rate is more limited .
mkn766 and ark564 are both low mass agn and so have large leverage on the value of @xmath20 . if there is a slight increase of hf psd normalisation with accretion rate or ` state ' , as the observations of cyg x-1 by @xcite suggest , inclusion of these two agn would steepen @xmath20 . also some masses used here differ from those used by @xcite , taken from @xcite . with small samples ,
such differences in content can have a significant effect .
further studies to determine @xmath2 for other samples of agn are clearly merited .
the present results are also consistent with the work of @xcite who measure the excess variance , @xmath23 , for a sample of agn observed by .
they find that @xmath55 where @xmath56 , which is very close to the relationship derived here , for the agn with 80ksec minimum duration observations or @xmath57 for agn with a minimum of 40 ksec duration .
using nva measurements from @xcite it has been shown that , within a sample of agn observed by exosat , hf psd normalisation scales almost exactly inversely with black hole mass .
these observations support the proposal of @xcite that hf psd normalisation scales exactly inversely from xrbs to agn .
it is also noted that the quasar 3c273 fits well onto the scaling relationship derived for seyfert galaxies , suggesting that the same process which drives x - ray variability in seyfert galaxies also drives x - ray variability in 3c273 , even though the emission process or emission location may be different .
i thank dimitrios emmanoulopoulos and christian knigge for extensive discussions about statistics and regression and i thank dimitrios for tuition in the use of mathematica .
i thank liz bartlett for advice regarding idl and brandon kelly for discussions on regression and for advice as to how to run his bayesian regression code inside idl .
i thank the anonymous referee for a useful and informative report . | the old exosat medium energy measurements of high frequency ( hf ) agn power spectral normalisation are re - examined in the light of accurate black hole mass determinations which were not available when these data were first published @xcite .
it is found that the normalised variability amplitude ( nva ) , measured directly from the power spectrum , is proportional to @xmath0 where @xmath1 .
as nva is the square root of the power , these observations show that the normalisation of the hf power spectrum for this sample of agn varies very close to inversely with black hole mass .
almost the same value of @xmath2 is obtained whether the quasar 3c273 is included in the sample or not , suggesting that the same process that drives x - ray variability in seyfert galaxies applies also to 3c273 .
these observations support the work of @xcite who show that an almost exactly linear anticorrelation is required if the normalisations of the hf power spectra of agn and x - ray binary systems are to scale similarly .
these observations are also consistent with a number of studies showing that the short timescale variance of agn x - ray lightcurves varies approximately inversely with mass . |
many experimental groups are presently working on the formation of ultracold molecules and molecular quantum gases .
a very efficient scheme is the photoassociation reaction , where two colliding atoms absorb a photon to make a molecule @xcite ; moreover ultracold photoassociation spectroscopy @xcite is capable of providing deep insight into the properties of long range molecules , cold atomic collisions , or pair correlations in a condensate .
theoretical two - body scattering models have been developed to compute the photoassociation rates @xcite , in the framework of a perturbative treatment , adapted to situations where low - intensity cw lasers are employed and many - body effects are negligible . a linear variation of the photoassociation rate as a function of intensity is then predicted . in the work of bohn and julienne @xcite , which develops a semi - analytic theory yielding laser - induced energy shifts and l ine broadening ,
the departure from this linear behavior is described , the saturation limit at larger laser intensities being evaluated . in a condensate , the two - body model should fail and javanainen and mackie @xcite have developed a many - body theory which predicts a saturation limit occurring at lower intensities and attributed to rogue photodissociation .
the search for saturation effects has stimulated several experiments .
measurement , by mc kenzie _ et al _ @xcite , of the photoassociation rate in a sodium condensate as a function of laser intensity has shown a linear behavior in the intensity range considered , in agreement with the two - body bohn and julienne estimation , and in contrast with the prediction of javanainen and mackie . for a quantum degenerate lithium atomic gas , measurement @xcite of the intensity dependence of the photoassociation rate
has verified both the large value of this rate ( at a temperature @xmath0=600 nk ) and the two - body saturation effect at intensities @xmath1 30 w/@xmath2 .
the agreement with theory is considered as reasonable within the present experimental accuracy : the higher value of the measured saturation limit , and the oscillations in the experimental rate in the 30 - 80 w/@xmath2 intensity range should be confirmed by measurements with a smaller error bar . for a non - degenerate gas of cesium atoms , at a temperature of 40 @xmath3k , within an optical dipole trap , saturation limit is reached for intensities above 100 w/@xmath2@xcite , and the experimental values then lay around the theoretical curve : it is not clear whether such oscillations are due to uncertainties or should be attributed to a physical effect .
+ recently naidon and masnou - seeuws @xcite have developed a time - dependent many - body theory of photoassociation in an atomic bose - einstein condensate , showing that several physical effects depend on pair correlations .
they show that the two existing models for the description of such correlations @xcite , differ in their predictions of the photoassociation line shapes at large intensities .
precise measurements of the variation of the line shapes as a function of cw laser intensity should therefore give insight into the optimal description of many - body effects .
+ the aim of the present paper is to anticipate such measurements by revisiting the theoretical treatment of photoassociation line shapes , taking advantage of the efficiency of grid methods @xcite for the numerical description of cold collisions and long range molecules , particularly in cases where several channels are involved .
actually , the later methods have been widely used for the computation of photoassociation rates @xcite , since they can determine very precise energies and wave functions , both for bound and for continuum levels .
it has been demonstrated that due to their global character , mapped grid methods are ideally suited to precise computation of the perturbations occurring in molecular spectra , such as rb@xmath4@xcite or k@xmath4@xcite singlet - triplet mixing . in such cases , the perturbation of one level of a given series should be attributed to a large number of bound or continuum levels of the other series , and local perturbation treatments are poorly adapted .
however , up to know , attempts to compute the photoassociation line shapes via grid methods have not been very efficient @xcite and the interpretation of precise measurements is routinely performed with numerov type approaches @xcite .
the present paper is proposing an alternative way of implementing grid methods for the calculation of photoassociation line shapes , energy shifts , and rates .
+ the paper is organized as follows : the description of the present model is reported in section ii - iv .
the link with previous models , such as the bohn and julienne model is discussed in section v , where the various assumptions and approximations are underlined .
application to the cases of sodium photoassociation in the experimental conditions of ref .
@xcite is presented in section vi .
section vii will summaries and discuss the outlook of the method .
the physical process of photoassociation ( pa ) can be divided into three steps . _ initially _ ( step i ) , a thermal ensemble of cold atoms is in the ground electronic state ; the probability to find a colliding pair with relative motion at energy @xmath5 and angular momentum @xmath6 is given by : @xmath7 where @xmath0 is the temperature , and @xmath8 is the boltzmann constant .
the partition function @xmath9 was shown in @xcite to be well approximated by the classical expression @xmath10 for a gas of non - interacting pairs in a volume @xmath11 .
further discussion regarding @xmath9 could be found in @xcite .
+ the radial wave function which corresponds to the pair is represented with a continuum eigenfunction @xmath12 of the single - channel hamiltonian @xmath13 @xmath14 here , @xmath15 is the reduced mass of the pair , and @xmath16 is the born - oppenheimer potential curve for the relevant , here the ground , electronic state .
+ at the second step ( step ii ) , the intensity of a cw laser with frequency @xmath17 is ramped within a time scale that can be varied from hundreds of ns to several @xmath3s . along this _ preparation _ period ,
the continuum states are coupled to one or more bound vibrational levels @xmath18 that belong to one or more excited electronic states @xmath19 .
each of the levels in the excited state , denoted by @xmath20 , may decay spontaneously with a natural free width @xmath21 , which is typically at the order of ns .
the field that is applied on this stage varies with time , and so are the states of the total system .
+ finally , during the stable period of the experiment ( step iii ) , the laser reaches its peak constant intensity @xmath22 . for the trap - loss or multiphoton detection scheme which will be described here ,
the duration of this period is on the order of ms .
each of the energy levels of the system at this step can now be written as : @xmath23 where we include the possibility of populating bound levels @xmath24 of the ground state . in the present work
, we shall consider only this third step , which of course is influenced by the preparation step .
since we shall not treat explicitly the preparation step , and possible transitory effects , a rapid switching on of the photoassociation laser will be assumed , justifying sudden approximation .
the density of probability for a transition from initial state @xmath25 to a final state @xmath26 is given by : @xmath27 + at section iv it will be shown that a knowledge of @xmath28 is sufficient to extract the experimental parameters of the pa process .
an efficient method to compute the initial and the final states , @xmath29 and @xmath30 , is the subject of the next section .
the initial state is a thermal ensemble composed of atom pairs , with several angular momenta @xmath6 .
each pair is coupled to several @xmath31 partial waves components on the excited state . for simplicity
, the equations will be written hereby for one value @xmath6 and@xmath31 of the angular momentum . to obtain @xmath32
one has to compute the initial and final eigenstates of the hamiltonian without and with the light interaction . in the global grid method
used here , this is achieved by representation of the hamiltonian in a certain basis , and then a diagonalization .
the diagonalization step is the most expensive computationally , and it scales as @xmath33 where @xmath34 is the number of basis function .
a selection of an appropriate basis prior to the diagonalization is hence crucial for the efficiency of the method .
+ using numerical grid methods , a numerically complete set of of eigenvalues and eigenvectors of the time - independent schrdinger hamiltonian @xmath36 is obtained for the isolated molecule , for the ground and excited potentials .
this is achieved by using the mapped fourier grid method ( see for example the references @xcite ) . within the procedure
, a unitary transformation @xmath37 is performed on the hamiltonian eq .
( [ eq:01 ] ) , which is written in equally space grid @xmath35 .
the resulted hamiltonian @xmath38 , is represented on a grid with non - uniform distribution of points in the position space @xmath39 , to take advantage of the phase - space occupancy of the problem .
the potential energy operator is written simply as @xmath40 , and the kinetic energy operator reads : @xmath41 @xmath42 is the length of the grid , @xmath43 is the first derivative operator matrix , given by ( assuming even number of grid points ) : @xmath44}},i \ne j } \\ \end{array } } \right.\ ] ] and @xmath45 is the ( diagonal ) jacobian operator for the transformation . a sufficient representation for @xmath36 by an equally spaced grid demands typically @xmath46 basis functions for the example considered below in section [ sec : sodium ] .
as was shown in @xcite , the use of a mapped grid leads to a reduction of about an order of magnitude in the basis size , to several thousands . +
a diagonalization of @xmath47 gives the initial , interaction - free , wave functions and their energies for both electronic states @xmath48 and @xmath20 .
the eigenvalues with @xmath49 , correspond to bound , discrete states and the ones with @xmath50 refer to box - normalized quasi - continuum states .
the continuum under this picture is therefore tied to the box size from which is was deduced .
+ in order to take account of the spontaneous emission in the excited channel @xmath19 , an imaginary potential is introduced : @xcite @xmath51 where @xmath52 is the transition dipole moment , and @xmath53 .
a diagonalization of @xmath54 results a complex eigen - energies .
the imaginary part of the eigenvalues , @xmath55 is interpreted as the lifetime of the level @xmath56 .
a cw laser with constant intensity @xmath22 , detuned by @xmath57 relative to the atomic resonance line , introduces a radiative coupling between the ground state and the excited state . after making the rotating wave approximation ,
the full dressed hamiltonian under the interaction can be written as : @xmath58\ ] ] the coupling between the ground and the excited state is introduced by the rabi coupling operator through the transition dipole moment : @xmath59 , where @xmath60 is the spatially dependent transition dipole moment between the two electronic states , and @xmath61 is an angular factor depending upon the polarization of the laser and the orientation of the molecular axis . in the @xmath35 representation
@xmath62 is a local , i.e. diagonal , operator .
a diagonalization of @xmath63 gives the final states @xmath64 .
+ for the final states , one has to diagonalize the hamiltonian describing the interaction of the molecule with the radiation field routinely for many experimental parameters .
a direct diagonalization of @xmath65 even for the mapped grid representation , is inefficient , mainly due to mainly due to the redundancy in the representation of the basis for both potentials .
an improved representation for the coupled system is hence crucial
. this can be obtained by writing @xmath66 in the energy eigenstates of the single channels hamiltonians @xmath67 .
the unitary transformation @xmath68 is performed on @xmath65 to give @xmath69 the hamiltonian in the energy representation .
the eigenvectors that were obtained as the solutions for the single channel problem serve as the transformation matrices between the two representations , so that : @xmath70\ ] ] where @xmath71 are the matrices which contain the complete set of eigenvectors for the ground and excited states , @xmath48 and @xmath20 , respectively . + the coupled channel hamiltonian in the energy states picture
becomes : + * for the diagonal part in the excited channel subspace @xmath72 where @xmath73 is the detuning from the atomic resonance , and @xmath74 the natural width for the vibrational level @xmath75 . * for the diagonal matrix elements in the ground channel subspace : @xmath76 for clarity , explicit @xmath6 dependency is ommited .
we remark here again that @xmath77 include both the bound and the quasi - continuum parts of the spectrum .
@xmath78 is therefore discrete , for a given box normalization .
+ * for the non - diagonal part of @xmath79 : @xmath80 we can employ the r - centroid approximation for @xmath81 and write : @xmath82 where now @xmath83 denotes the frank - condon factor between vibrational levels from the two different electronic states .
the advantage of the energy representation is the possibility to pick the energy levels that participate in the process and thus reduce the required basis set .
we shall consider formally the case of an isolated resonance where the ground state continuum @xmath84 is coupled only to single bound level in the excited state @xmath85 .
the detuning from the atomic resonance @xmath57 could be replaced by @xmath86 , the detuning from the excited bound level .
a generalization to several bound excited states , or even a whole continuum , is straightforward . from this point on , we reduce all the indices for the excited state .
+ by moving to the energy representation we reduce the representation for the coupled channels problem by about one half , to the order of @xmath87 in the example discussed below in section [ sec : sodium ] .
note that so far , the representation of the continuum is done using a box unity - normalized wave functions .
this box - dependent basis for the continuum is arbitrary , and therefore it could be reduced .
+ in order to do so , one had to defined a box - independent frank - condon factor .
box - independent quantities for the continuum are obtained by moving from the box - normalized to the energy - normalized wave functions . the relation between the two
is given by : @xmath89 where the inverse of the level density , @xmath90 , is given simply , for a large enough box , by the energy difference between two adjacent quasi - continuum levels , given under the box - normalization constraint .
an energy normalized frank - condon profile : @xmath91 could now be obtained .
@xmath92 is a box - independent , numerically continuous function of the scattering energy .
in fact , @xmath92 is a physical measure for the energy dependence of the interaction and it thus the function that one would like to sample for obtaining the minimal energy representation .
the alternative basis has to capture the functional energy dependence of the frank - condon profile , using the smallest basis .
this could be done in principle with any sampling procedure @xcite .
the technical details regarding the numerical procedure we employed in this work is described at the appendix .
+ assuming such a minimal representation basis was found , we will denote it by @xmath93 for the ground state continuum , for @xmath94 , @xmath95 being the number of bound states .
the expression for the quasi - continuum diagonal part of the full hamiltonian ( eq.[eq : hdiag_g ] ) reads now : @xmath96 and all the other diagonal parts remained unchanged .
the nondiagonal part in the alternative basis is defined through the energy normalized frank - condon profile as : @xmath97 + a diagonalization of the hamiltonian ( eqs.[[eq : althdiag_g],[eq : alt_nond ] ] ) gives the set of eigenstates for the coupled hamiltonian , @xmath64 , which include components on both channels , with possible contributions from all the bound or free levels .
the energy eigenvalues which are obtained are again complex and @xmath98 is the width of the level @xmath64 .
the transition probability from any initially populated state @xmath93 to any final state @xmath64 , is simply : @xmath99 + it is important to notice here that because @xmath100 is an eigenstate of the coupled hamiltonian , a pair that reaches this state at the beginning of the final step of the experiment remains in the same state without any significant dynamics beside an exponential decay with the lifetime @xmath101 .
+ table 1 summarizes our procedure so far with the different representations and their typical necessary basis sizes .
we started from the hamiltonian on the equally spaced grid representation .
a minimal representation of the problem with such a grid demands a basis of about @xmath102 function .
we then used the phase space occupation of the problem to move to a mapped grid , and reduce the basis set in about an order on magnitude to @xmath103 , for each of the electronic states .
the transformation to the energy representation reduces the excited state manifold .
the reduced @xmath88 representation minimizes the basis for the ground state , so that finally we could solve the whole coupled channels problem with a basis at the order of @xmath104 , or even smaller . at this sizes of basis
the method become efficient , even compared to the semianalytic perturbative methods .
+ .summary of the different representations used in this paper . @xmath105 and @xmath106 are the typical basis set sizes for the ground and the excited electronic state , respectively , adapted to the example of section [ sec : sodium ] .
[ cols="^,^,^",options="header " , ] the thermal energy is commonly the smallest energy scale .
it ranges from about a hundred of nanokelvins for the coldest condensates , to a few hundreds microkelvins , for ` hot ' mots . for low temperatures ,
the narrow width of the populated energy band makes the coupling profile to be almost insignificant .
saturation could then be described quite well with the perturbative methods . as the temperature increases , the inner structure of the continuum is encrypted into deviations of the lineshape from a symmetric lorentzian , and an inverted region of @xmath107 as the intensity increases .
some of these deviations , especially the ones which are visible at the low intensity limit , for relatively high temperatures , could be well described by the single - path perturbative models . other features which result only at high intensities ,
could be calculated only with the help of the multi - path model .
+ the non perturbative model that we presented in this paper can serve as a useful tool to model easily other cases where non trivial lineshapes appear .
these include : * photoassociation close to threshold . *
accidentally degenerate or almost degenerate levels that originate from different electronic states . in both cases ,
interaction between several levels in the excited state need to be put in the framework of a higher order model .
the case of resonant coupling photoassociation @xcite is a good example . * active interaction between multiple continuum manifolds , where more than a single continuum is populated or depopulated during the process .
one possibility is when more then one partial wave participates in the scattering , e.g. , the @xmath108 resonance in @xmath109rb ( see ref .
@xcite ) . in cases of lower barriers , higher temperatures , or in the cases of processes that take place by explicitly populating the continuum ,
highly structured lineshapes might be resolved by our method .
finally , the dynamical approach which is the basis for the present study is an important building block for a future comprehensive time dependent description of the many body interaction in the description of pa .
this work was supported by the israel science foundation and by the european commission in the frame of the cold molecule research network under contract no .
hprn - ct-2002 - 00290 .
the fritz haber center is supporter by the minerva gesellschaft fur die forschung gmbh munchen , germany .
10 h. r. thorsheim , j. weiner , and p. s. julienne .
`` laser - induced photoassociation of ultracold sodium atoms . ''
lett . * 58 * , 2420 ( 1987 ) .
k. m. jones , e. tiesinga , p. d. lett , and p. s. julienne .
`` ultracold photoassociation spectroscopy : long range molecules and atomic scattering . ''
phys . * 78 * , 483 ( 2006 ) .
p. pillet , a. crubellier , a. bleton , o. dulieu , p. nosbaum , i. mourachko , and f. masnou - seeuws .
`` photoassociation in a gas of cold alkali atoms .
i : perturbative quantum approach . ''
b * 30 * , 2801 ( 1997 ) . r. napolitano . `` a two - state model for controlling scattering lengths and photoassociation spectral line shapes alkali - metal atoms by resonant light in the regime of finite ultracold temperature . ''
brazilian journal of physics * 27 * , 162 ( 1997 ) .
r. ct and a. dalgarno .
`` photoassociation intensities and radiative trap loss in lithium . ''
a * 58 * , 498 ( 1998 ) .
j. l. bohn and p. s. julienne .
`` semianalytic theory of laser - assisted resonant cold collisions . ''
a * 60 * , 414 ( 1999 ) .
e. taylor - juarros , r. ct , and k. kirby .
`` formation of ultracold polar molecules via raman excitation . ''
j. d * 31 * , 213 ( 2004 ) .
j. javanainen and m. mackie .
`` rate limit for photoassociation of a bose - einstein condensate . ''
lett . * 88 * , 090403 ( 2002 ) . c. mckenzie , j. h. denschlag , h. hffner , a. browaeys , l. e. de araujo , f. k. fatemi , k. m. jones , j. e. simsarian , d. cho , a. simoni , e. tiesinga , p. s. julienne , k. helmerson , p. d. lett , s. l. rolston , and w. d. phillips .
`` photoassociation of sodium in a bose - einstein condensate . ''
* 88 * , 120403 ( march 2002 ) .
i. d. prodan , m. pichler , m. junker , r. j. hulet , and j. l. bohn .
`` intensity dependence of photoassociation in a quantum degenerate atomic gas . '' phys .
* 91 * , 080402 ( august 2003 ) .
s. d. kraft , m. mudrich , m. u. staudt , j. lange , o. dulieu , r. wester , and m. weidemuller `` saturation of cs@xmath4 photoassociation in an optical dipole trap . ''
a * 71 * , 013417 ( 2005 ) .
p. naidon and f. masnou - seeuws .
`` photoassociation and optical feshbach resonances in an atomic bose - einstein condensate : treatment of correlation effects . ''
a * 73 * , 043611 ( april 2006 ) .
t. khler , t. gasenzer , and k. burnett .
`` microscopic theory of atom - molecule oscillations in a bose - einstein condensate . ''
a * 67 * , 013601 ( 2003 ) .
p. naidon and f. masnou - seeuws .
`` pair dynamics in the formation of molecules in a bose - einstein condensate . ''
a * 68 * , 033612 .
r. kosloff .
`` quantum molecular dynamics on grids . '' in r. e. wyatt and j. z. zhang , editors , dynamics of molecules and chemical reactions , pages 185 - 230 , marcel dekker , new york ( 1996 ) .
e. fattal , r. baer and r. kosloff , `` phase space approach for optimizing grid representation : the mapped fourier method . ''
e * 53 * , 1217 ( 1996 ) .
v. kokoouline , o. dulieu , r. kosloff , and f. masnou - seeuws .
`` mapped fourier methods for long range molecules : application to perturbations in the rb@xmath110 ) spectrum . ''
phys . * 110 * , 9865 ( 1999 ) .
k. willner , o. dulieu , and f. masnou - seeuws .
`` a mapped sine grid method for long range molecules and cold collisions . '' j. chem . phys . * 120 * , 548 ( 2004 ) .
c. m. dion , c. drag , o. dulieu , laburthe tolra , b. , f. masnou - seeuws , and p. pillet .
`` resonant coupling in the formation of ultracold ground state molecules via photoassociation . ''
lett . * 86 * , 2253 ( 2001 ) .
a. j. kerman , j. m. sage , s. sainis , t. bergeman , and d. demille .
`` photoassociation and state - selective detection of ultracold @xmath111 molecules . '' phys .
lett . * 92 * , 153001 ( 2004 ) .
s. azizi , m. aymar , and o. dulieu .
`` prospects for the formation of ultracold ground state polar molecules from mixed alkali atom pairs . ''
j. d * 31 * , 195 ( december 2004 ) .
j. m. sage , s. sainis , t. bergeman , and d. demille .
`` optical production of ultracold polar molecules . ''
* 94 * , 203001 ( 2005 ) .
v. kokoouline , o. dulieu , and f. masnou - seeuws .
`` theoretical treatment of channel mixing in excited rb@xmath4 and cs@xmath4 ultra - cold molecules .
perturbations in @xmath112 photoassociation and fluorescence spectra . ''
a * 62 * , 022504 ( 2000 ) .
v. kokoouline , o. dulieu , r. kosloff , and f. masnou - seeuws .
`` theoretical treatment of channel mixing in excited rb@xmath4 and cs@xmath4 ultra - cold molecules : determination of predissociation lifetimes with coordinate mapping . ''
a * 62 * , 032716 ( 2000 ) .
t. bergeman , j. qi , d. wang , y. huang , h. k. pechkis , e. e. eyler , p. l. gould , w. c. stwalley , r. a. cline , j. d. miller , and d. j. heinzen .
`` photoassociation of @xmath113rb atoms into 0@xmath114 states near the 5s + 5p atomic limit . ''
* 39 * , s813 ( 2006 ) . c. lisdat , o. dulieu , h.knockel , and e. tiemann .
`` inversion analysis of k@xmath4 coupled electronic states with the fourier grid method . ''
j. d * 17 * , 319 ( 2001 )
. p. pellegrini . .
thse de doctorat ( ph.d thesis ) , universit paris - sud , centre dorsay ( 2003 ) .
p. naidon . . thse de doctorat ( ph.d thesis ) , universit pierre et marie curie paris vi ( 2004 ) .
m. portier , s. moal , j. kim , m. leduc , c. cohen - tannoudji , and o. dulieu .
`` analysis of light - induced frequency shifts in the photoassociation of ultracold metastable helium atoms . ''
b : at . mol .
. phys . * 39 * s881 ( 2006 ) . c. p. koch , r. kosloff , e. luc - koenig , f. masnou - seeuws , and a. crubellier `` photoassociation with chirped laser pulses : calculation of the absolute number of molecules per pulse . ''
b : at . mol .
, * 39 * s1017 ( 2006 ) .
m. nest , h .-
. lett . * 352 * 486 ( 2002 ) .
f. laloe .
claude cohen - tannoudji , bernard diu . .
( hermann , paris , 1973 ) .
w.h . press ,
teukolsky , w.t .
vetterling , and b.p . flannery . .
( second edition , cambridge university press , 1995 ) .
u. fano .
`` effect of configuration interaction on intensities and phaseshifts . ''
phys . rev . *
124 * , 1866 ( 1961 ) .
m. l. du , a. dalgarno , and m. j. jamieson .
`` level shifts of discrete states embedded in a continuum . ''
* 5 * , 2980 ( 1991 ) .
j. l. bohn and p. s. julliene phys .
a * 60 * , 414 ( 1999 ) .
a. crubellier and e. luc - koenig .
`` threshold effects in the photoassociation of cold atoms : @xmath115 model in the milne formalism . ''
b * 39 * , 1417 ( 2006 )
. s. b. nagel , p. g. mickelson , a. d. saenz , y. n. martinez , y. c. chen , t. c. killian , p. pellegrini , and r. ct `` photoassociative spectroscopy at long range in ultracold strontium . ''
lett . * 94 * , 083004 ( 2005 ) .
two purposes are the objects of this appendix : the numerical details of the sampling of the continuum profile @xmath116 ( see eq . ( [ eq : fc_profile ] ) ) , and the demonstration of the energy scale that is needed for the calculation of the energy shift .
the sampling of the continuum profile @xmath116 could be done in principle by a global or local numerical sampling procedure ( see for example @xcite).the common procedures take advantage of an a - priori knowledge about the given function to be sampled . in this work we limited ourselves to a naive approach and determined the sampling of the profile a - posteriori according to the results that were obtained with a given basis .
the criterion we employed was the numerical convergence of the integral over a given moment of the interaction profile @xmath116 : @xmath117^n de } } \approx \sum_j \delta(e_j)[{\cal f}^e ( e_j)]^n\ ] ] where @xmath118 is the order of the moment , @xmath119 is the discrete sampling energy points , and @xmath120 is the energy difference between two adjacent points .
the upper bound for the profile @xmath121 is discussed below .
the procedure initiates with a small set of points , and a new point is added or rejected between any two old points according to its influence on the value of @xmath122 .
the quality of the procedure is determined by the tolerance parameter @xmath123 .
the number of basis functions for the sampling the @xmath124 continuum , as a function of the tolerance parameter is presented in figure a1 .
the various lines present different moments . + as can be understood , the convergence of the sampling becomes more efficient for higher moments , which emphasize more the features in the profile .
too high moments , however , smear completely the functional structure and so an optimum moment have to be found .
we have to comment here that in order to increase the resolution of the basis set with respect to the thermally populated energy near the threshold ( for @xmath125 , n is usually 10 ) , we inserted a small energy grid ( typically 50 - 100 basis functions ) with constant energy step at close - to - threshold energies .
+ in figure a2 the resulted energy shifts are plotted as a function of the tolerance parameter for the various moment orders .
the numerically exact value which was calculated using the perturbative method is @xmath126 . the values for moment order above the @xmath127 , convergence to this values within less
then 5% for tolerance value of @xmath128 . the basis set that was taken in this paper uses @xmath129 and @xmath130 .
larger moments than the chosen exhibit slower convergence and demand much smaller values of @xmath131 to produce an adequate basis set .
+ we must remark also , that the only physical parameter which demands such a care in the sampling is the energy shift .
the values for the lifetime are much less influenced by the size of the basis , and converged even at @xmath131 as small as @xmath132 . to demonstrate the importance of very large energy scale that are needed to calculate the energy shift correctly we present in figure a3 the calculated energy shift as a function of @xmath121 the upper bound of the energy grid .
convergence of the energy shift is achieved only when reaching hundreds of kelvins .
note , again , that only the energy shift exhibit a sensitivity for such high energies . for low laser
intensities the lifetime is found to converge @xmath121 at the order of the thermal energy grid .
convergence near the saturation limit demands extension of the grid up to the first minimum of the coupling profile , i.e.,@xmath133 , or tens of milikelvins .
a basis set for such energies contained less then 150 function and is adequate for calculating most of the lineshapes features , except for energy shifts . | a description of photoassociation by cw laser is formulated in the framework of grid methods .
the hamiltonian describing one or several bound states coupled to a multiple of continuum manifolds via a radiative field is written in the energy representation and diagonalized .
the generality of the treatment allows to compute accurately and efficiently physical properties such as non - linear high - intensity energy shifts , line shapes , and photoassociation rates both for isolated and non - isolated resonances .
application is given to sodium photoassociation in the experimental conditions of mc kenzie _ et al _ [ phys .
. lett . * 88 * , 090403 ( 2002 ) ] .
inverted region for the dependency of the rate vs. the intensity and non - symmetric lineshapes were predicted to occur above the saturation limit .
comparison with the model of bohn and julienne [ phys .
rev .
a * 60 * , 414 ( 1999 ) ] is discussed . |
much progress has been made in modelling nucleosynthesis in stars in general , but the origin of nitrogen still remains somewhat uncertain . in the classical picture , the apparent increase in nitrogen production with metallicity , as inferred from spectroscopy of galactic and extragalactic hii - regions , is a manifestation of nitrogen being either primary or secondary ( produced from primordial elements or with heavier elements as `` seeds '' ) .
however , reality often turns out to be more complex than our simplest models of it .
the origin of nitrogen is probably no exception . during the last few years
, reasonably accurate stellar nitrogen abundances have become available also for population ii stars ( see , e.g. , @xcite ) .
the rather large nitrogen abundances relative to iron seen in halo stars in the milky way suggest that a relatively large fraction of the nitrogen should come from type ii supernovae , especially at low metallicity .
however , recent stellar evolution models , including nucleosynthesis , all appear to predict significant nitrogen production in intermediate - mass ( i m ) stars ( @xcite )
. the typical production factor would be of the order @xmath0 of the initial stellar mass .
if im - stars of all metallicities produce that much nitrogen , it would require that metal - poor galaxies ( or regions within a galaxy ) with low n / o - ratios have very young stellar populations , in fact , they have to be so young that im - stars have not yet made any significant contribution to the interstellar nitrogen abundance of these objects , unless the production of oxygen is much greater at low metallicity . in this work
i have considered radial gradients of o / h and n / o in the milky way and five other nearby late - type galaxies in order to find additional clues to the origin of nitrogen .
the advantages of considering the abundance gradients within galaxies , rather than over - all trends obtained from the global properties of galaxies , are that the same stellar nucleosynthesis prescriptions ( yields ) must be able to simultaneously reproduce the radial abundance gradients in galaxies with quite different properties ( sizes , total masses , gas distributions , etc . ) and the very same yields must also be consistent with the fact that galaxies with similar o / h - gradients may have quite different n / o - gradients .
models of chemical evolution ( cems ) may thus be able to provide insight and constraints on stellar yields when several galaxies are modelled and analysed in parallel .
the oxygen and nitrogen abundances are taken from ( * ? ? ?
* ; * ? ? ?
* pilyugin et al . ( 2003 , 2004 ) ) . which provide a large compilation of data
that is consistent in the sense that all abundances are derived using the same strong - line calibration ( p - method , see @xcite ) .
data for hi and h@xmath1 ( co ) were collected from the literature ( see table [ params ] for references ) . to obtain a
consistent and , possibly , a more correct set of data for the gas distributions , the conversion from co - flux to h@xmath1 surface density has been redone using the oxygen - dependent calibration by ( * ? ? ?
* wilson ( 1995 ) ) . .[params ] model parameters and adopted distances . see sect . [ models ] for further details .
[ cols= " < , < , < , < , < , < , < , < , < " , ]
i present detailed , numerical cems for the milky way and five other nearby spirals , computed using the code described in detail in ( * ? ? ?
* mattsson ( 2008 ) ) .
since the halo contributes very little to the present - day abundance gradients , except for at very large galactic radii , where the disc density becomes comparable with the halo density ( see , e.g. , @xcite ) , the model is constructed as a single exponential disc , formed by exponentially decaying baryonic infall , i.e. , @xmath2\right\}^{-1}\exp\left[-{t\over\tau(r)}\right],\ ] ] where @xmath3 is the infall - time scale and @xmath4 denotes present time .
the star formation rate is prescribed by a modified `` silk - law '' ( @xcite ) , i.e. , @xmath5^{1+\varepsilon},\ ] ] where @xmath6 is the present - day ( @xmath7 ) angular frequency of the disc ( taken from @xcite ) and @xmath8 is a parameter regulating the star - formation efficiency relative to the milky way . to simulate the inside - out formation of a galactic disc , the infall time scale is assumed to have a linear radial dependence , @xmath9 , where @xmath10 are arbitrary constants and @xmath11 is the disc scale length . for @xmath12 , @xmath13 and @xmath14 kpc ,
the above relation is similar to that used by ( * ? ? ?
* romano et al .
( 2000 ) ) and ( * ? ? ?
* boissier & prantzos ( 1999 ) ) . for the gas - rich galaxies ( ngc 598 , ngc 2403 , ngc 6946 ) ,
the infall rate is assumed to be constant at all radii , indicated by @xmath15 in table [ params ] .
the nucleosynthesis prescriptions were taken from ( * ? ? ?
* chieffi & limongi ( 2004 ) ) for high mass stars and ( * ? ? ?
* van den hoek & groenewegen ( 1997 ) ) and ( * ? ? ?
* gaviln et al .
( 2005 ) ) for low and intermediate stars ( see fig . [ yields ] ) .
the latter set of yields has now been extended to low metallicities ( @xcite ) .
radial trends for o / h ( upper panels ) , n / o ( lower panels ) and the surface density of hydrogen .
black dots show the abundances from pilyugin et al .
( 2003 , 2004 ) , red lines show models with yields from gaviln et al .
( 2005 , 2007 ) and blue lines show models with yields from van den hoek & groenewegen ( 1997 ) .
the solar symbols show the solar values according to asplund et al .
( 2005).,title="fig:",width=166 ] radial trends for o / h ( upper panels ) , n / o ( lower panels ) and the surface density of hydrogen .
black dots show the abundances from pilyugin et al .
( 2003 , 2004 ) , red lines show models with yields from gaviln et al .
( 2005 , 2007 ) and blue lines show models with yields from van den hoek & groenewegen ( 1997 ) .
the solar symbols show the solar values according to asplund et al .
( 2005).,title="fig:",width=166 ] radial trends for o / h ( upper panels ) , n / o ( lower panels ) and the surface density of hydrogen .
black dots show the abundances from pilyugin et al .
( 2003 , 2004 ) , red lines show models with yields from gaviln et al .
( 2005 , 2007 ) and blue lines show models with yields from van den hoek & groenewegen ( 1997 ) .
the solar symbols show the solar values according to asplund et al .
( 2005).,title="fig:",width=166 ]
the modelling result using a @xcite imf is presented below in terms of o / h- and n / o - gradients and radial hydrogen distributions ( fig .
[ result1 ] and fig .
[ result2 ] ) . to avoid over - production of oxygen
the imf was truncated at @xmath16 for models with yields from van den ( * ? ? ?
* hoek & groenewegen ( 1997 ) ) and at @xmath17 for models with yields from ( * ? ? ?
* ; * ? ? ?
* gaviln et al .
( 2005 , 2007 ) ) .
same as fig .
[ result1 ] , but for ngc 2403 , ngc 3031 and ngc 6946.,title="fig:",width=166 ] same as fig .
[ result1 ] , but for ngc 2403 , ngc 3031 and ngc 6946.,title="fig:",width=166 ] same as fig .
[ result1 ] , but for ngc 2403 , ngc 3031 and ngc 6946.,title="fig:",width=166 ] i adjusted the remaining model parameters so that the o / h - gradients were reproduced and a reasonable agreement with the observed hydrogen distributions were obtained . for the milky way , the properties of the solar neighbourhood were also used as constraints .
the modelled n / o - gradients are then not in very good agreement with the observations - the only exceptions being ngc 598 and ngc 3031 , where the models reproduce the n / o - gradient reasonably well , although with different sets of yields . if the abundances can be trusted , which is very likely , this indicates a relatively strong correlation with metallicity of the integrated nitrogen yields . in contemporary stellar yields for im - stars ( see fig 1 . ) there is no strong dependence on metallicity , which perhaps indicates that the imf is varying .
note , also , that newer nitrogen yields , in general ( see @xcite ) , are larger than those by ( * ? ? ?
* van den hoek & groenewegen ( 1997 ) ) .
left panels : n- and o - yields from gaviln et al .
( 2005 , 2007 ) including the previously unpublished sets for low metallicity ( left ) .
right panels : the same from van den hoek & groenewegen ( 1997 ) . note that the n - yields by gaviln et al .
are roughly five times larger than those by van den hoek & groenewegen.,title="fig:",width=249 ] left panels : n- and o - yields from gaviln et al .
( 2005 , 2007 ) including the previously unpublished sets for low metallicity ( left ) .
right panels : the same from van den hoek & groenewegen ( 1997 ) . note that the n - yields by gaviln et al .
are roughly five times larger than those by van den hoek & groenewegen.,title="fig:",width=249 ] variations of the slope of the imf changes the n / o - gradient , but at the same time the over - all oxygen abundance will change ( see fig .
[ imf ] ) .
a ( * ? ? ?
* salpeter ( 1955 ) ) imf , @xmath18 , truncated at @xmath19 and @xmath20 gives a result similar to that with the ( * ? ? ?
* scalo ( 1986 ) ) imf for the o / h - gradient , but also a reasonable fit to the observed n / o - gradient for the milky way ( see fig . [ imf ] ) using the ( * ? ? ?
* ; * ? ? ?
* gaviln et al .
( 2005 , 2007 ) ) yields
. however , for the other five galaxies , this combination of imf and yields does not really improve the fit to the observed abundances .
hence , resolving the o / h - n / o - inconsistency by changing the imf requires that the imf varies over the galactic disc , or , effectively , from galaxy to galaxy .
a stellar - metallicity dependence in the nitrogen yields of i m stars is also a reasonable explanation .
since the n / o - gradients predicted by the models are too flat in three cases ( ngc 224 , ngc 2403 and ngc 6946 ) , the yields may need to increase with metallicity .
if the n - yields are as high as those by ( * ? ?
* gaviln et al .
( 2005 ) ) for im - stars around solar metallicities , but as low as those by ( * ? ? ?
* van den hoek & groenewegen ( 1997 ) ) at subsolar metallicities , the observed slopes might be reproduced . on the other hand ,
it is quite likely that such an _ ad - hoc _ modification of the yields may lead to gradients that are too steep for the milky way and ngc 3031 .
employing a standard imf ( @xcite ) with a high - mass cut - off at @xmath21 the o / h - gradients are well - reproduced by numerical cems .
however , the n / o - gradients can not , in general , be reproduced simultaneously .
this inconsistency can be explained in several ways , e.g. , 1 .
the production of nitrogen in intermediate - mass stars is strongly correlated with metallicity , 2 .
the imf varies over the galactic disc ( or , effectively , from galaxy to galaxy ) such that the n / o - ratio is affected , 3 . the infalling gas that forms the disc is not primordial , but significantly enriched with heavier elements ( from pop .
iii or pop .
ii stars ? ) , 4 . or the p - method produces systematic errors .
i believe that the explanations ( a ) and ( b ) are more likely then the other two . the main reasons are : ( 1 ) many ingredients of stellar evolution models that affects the nucleosynthetic yields ( e.g. , mass loss rates and the efficiency of hot - bottom burning ) are still highly uncertain , and
( 2 ) the turn - over in the imf at low stellar masses appears correlated with the jeans mass ( @xcite ) , ( 3 ) significant effects from chemically enriched infall is only expected in the outer parts of the discs , where the abundances are low , and ( 4 ) the p - method is in very good agreement with the standard electron - temperature method ( @xcite ) .
furthermore , the stellar yields and detailed properties of the imf are probably the greatest uncertainties in cems in general . variations of the imf . changing the slope of a simple power - law imf ( @xmath22 @xmath23 are left unchanged ) will shift the o / h - gradient up or down ( see left panel ) , but leave @xmath24 almost unaffected . a salpeter ( 1955 ) imf ( @xmath25 ) which provides an acceptable fit to the observed n / o - ratios ( right panel).,width=498 ] leonid pilyugin
is thanked for supplying the abundance data and explaining some issues regarding the p - method .
marta gaviln and collaborators are thanked for making their low - metallicity yields available .
asplund m. , grevesse n. & sauval a. j. , 2005 , aspc , 336 , 25a boissier s. & prantzos n. , 1999 , mnras , 307 , 857 boissier s. , prantzos n. , boselli a. & gavazzi g. , 2003 , mnras , 346 , 1215 chieffi & limongi , 2004 , apj , 608 , 405 cayrel r. , depagne e. , spite m. , et al .
2004 , a&a , 416 , 1117 chiappini c. , matteucci f. , & romano d. , 2001 , apj , 554 , 1044 corbelli e. , 2003 , mnras , 342 , 199 van dokkum p.g . , 2008 ,
apj , 674 , 29 ecuvillon a. , israelian g. , santos n.c . ,
et al . , 2004 , a&a , 418 , 703 gaviln m. , buell j.f . & moll m. , 2005 , a&a , 432 , 861 gaviln m. , buell j.f . & moll m. , 2007 , private communication . van den hoek l.b . & groenewegen m.a.t . , 1997 ,
a&as , 123 , 305 israelian g. , ecuvillon a. , rebolo r. , et al .
2004 , a&a , 421 , 649 izzard r.g . ,
, karakas a.i . & pols o.p . , 2004 , mnras , 350 , 407 karakas a. & lattanzio j.c . , 2007 , pasa , 24 , 103 koper e , israel f.p . , dame t.m . , & thaddeus p. , 1991 ,
apj , 383 , 11 marigo p. , 2001
, a&a , 370 , 194 mattsson l. , 2008 , in progress .
pilyugin l.s . , 2001a , a&a , 369 , 594 pilyugin l.s .
, 2001b , a&a , 373 , 56 pilyugin l.s . , ferrini f. & shkvarun r.v . , 2003 ,
a&a , 401 , 557 pilyugin l.s .
, vilchez j.m . &
contini t. , 2004 , a&a , 425 , 849 romano d. , matteucci f. , salucci p. & chiappini c. , 2000 , apj , 539 , 235 rots a.h .
, 1975 , a&a , 45 , 43 salpeter e.e . , 1955 , apj , 121 , 161 sanders d.b . , solomon p.m. & scoville n.z . , 1984 , apj , 276 , 182 scalo j. , 1986 , fund . of cosm
phys . , 11 , 1 sofue y. , tutui y. , honma m. , et al . , 1999 , apj , 523 , 136 spite m. , cayrel r. , plez b. , et al .
2005 , a&a , 430 , 655 tacconi l.s . & young j.s . , 1986 ,
apj , 308 , 600 thornley m.d . & wilson c.d . , 1995 ,
apj , 447 , 616 wevers b.m.h.r . , van der kruit p.c . &
allen r.j . , 1986 , a&as , 66 , 505 wilson c.d . , 1995 , apj , 448 , l97 wyse r.f.g .
& silk j. , 1989 , apj , 339 , 700 | models of galactic chemical evolution ( cems ) show that the shape of the stellar initial mass function ( imf ) and other assumptions regarding star formation affect the resultant abundance gradients in models of late - type galaxies . furthermore ,
intermediate mass ( i m ) stars undeniably play an important role in the buildup of nitrogen abundances in galaxies . here
i specifically discuss the nitrogen contribution from im / agb stars and how it affects the n / o - gradient . for this purpose
i have modelled the chemical evolution of a few nearby disc galaxies using different imfs and star formation prescriptions .
it is demonstrated that n / o - gradients may be used to constrain the nitrogen contribution from im / agb - stars . |
the topology or inherent shape and form of an object is important . in data analysis ,
the inherent form and structure of data clouds are important .
quite a few models of data form and structure are used in data analysis .
one of them is a hierarchically embedded set of clusters , a hierarchy .
it is traditional ( since at least the 1960s ) to impose such a form on data , and if useful to assess the goodness of fit . rather than fitting a hierarchical structure to data ( e.g. , @xcite ) ,
our recent work has taken a different orientation : we seek to find ( partial or global ) inherent hierarchical structure in data .
as we will describe in this article , there are interesting findings that result from this , and some very interesting perspectives are opened up for data analysis and , potentially , perspectives also on the physics ( or causal or generative mechanisms ) underlying the data . a formal definition of hierarchical structure is provided by ultrametric topology ( in turn , related closely to p - adic number theory ) .
we will return to this in section [ sect23 ] below .
first , though , we will summarize some of our findings .
ultrametricity is a pervasive property of observational data .
it arises as a limit case when data dimensionality or sparsity grows .
more strictly such a limit case is a regular lattice structure and ultrametricity is one possible representation for it .
notwithstanding alternative representations , ultrametricity offers computational efficiency ( related to tree depth / height being logarithmic in number of terminal nodes ) , linkage with dynamical or related functional properties ( phylogenetic interpretation ) , and processing tools based on well known p - adic or ultrametric theory ( examples : deriving a partition , or applying an ultrametric wavelet transform ) . in @xcite and other works , khrennikov has pointed to the importance of ultrametric topological analysis .
local ultrametricity is also of importance .
this can be used for forensic data exploration ( fingerprinting data sets ) : see @xcite and @xcite ; and to expedite search and discovery in information spaces : see @xcite as discussed by us in @xcite , @xcite , and @xcite . in section [ sect23 ]
we show how extent of ultrametricity is measured .
section [ sect3 ] presents our main results on the remarkable properties of very high dimensional , or very sparse , spaces . as dimensionality or sparsity
grow , so does the inherent hierarchical nature of the data in the space . in section [ hfda ]
we then discuss application to very high frequency time series modeling .
summarizing a full description in murtagh @xcite we explored two measures quantifying how ultrametric a data set is ,
lerman s and a new approach based on triangle invariance ( respectively , the second and third approaches described in this section ) .
the triangular inequality holds for a metric space : @xmath0 for any triplet of points @xmath1 .
in addition the properties of symmetry and positive definiteness are respected .
the `` strong triangular inequality '' or ultrametric inequality is : @xmath2 for any triplet @xmath1 .
an ultrametric space implies respect for a range of stringent properties .
for example , the triangle formed by any triplet is necessarily isosceles , with the two large sides equal ; or is equilateral . *
firstly , rammal et al .
@xcite used discrepancy between each pairwise distance and the corresponding subdominant ultrametric .
now , the subdominant ultrametric is also known as the ultrametric distance resulting from the single linkage agglomerative hierarchical clustering method .
closely related graph structures include the minimal spanning tree , and graph ( connected ) components . while the subdominant provides a good fit to the given distance ( or indeed dissimilarity ) , it suffers from the `` friends of friends '' or chaining effect . *
secondly , lerman @xcite developed a measure of ultrametricity , termed h - classifiability , using ranks of all pairwise given distances ( or dissimilarities ) .
the isosceles ( with small base ) or equilateral requirements of the ultrametric inequality impose constraints on the ranks .
the interval between median and maximum rank of every set of triplets must be empty for ultrametricity .
we have used extensively lerman s measure of degree of ultrametricity in a data set .
taking ranks provides scale invariance .
but the limitation of lerman s approach , we find , is that it is not reasonable to study ranks of real - valued ( values in non - negative reals ) distances defined on a large set of points . *
thirdly , our own measure of extent of ultrametricity @xcite can be described algorithmically .
we examine triplets of points ( exhaustively if possible , or otherwise through sampling ) , and determine the three angles formed by the associated triangle .
we select the smallest angle formed by the triplet points
. then we check if the other two remaining angles are approximately equal .
if they are equal then our triangle is isosceles with small base , or equilateral ( when all triangles are equal ) .
the approximation to equality is given by 2 degrees ( 0.0349 radians ) .
our motivation for the approximate ( `` fuzzy '' ) equality is that it makes our approach robust and independent of measurement precision .
a supposition for use of our measure of ultrametricity is that we can can define angles ( and hence triangle properties ) .
this in turn presupposes a scalar product .
thus we presuppose a complete normed vector space with a scalar product a hilbert space to provide our needed environment .
quite a general way to embed data , to be analyzed , in a euclidean space , is to use correspondence analysis @xcite .
this explains our interest in using correspondence analysis : it provides a convenient and versatile way to take input data in many varied formats ( e.g. , ranks or scores , presence / absence , frequency of occurrence , and many other forms of data ) and map them into a euclidean , factor space .
murtagh @xcite , and earlier work by rammal et al .
@xcite , has demonstrated the pervasiveness of ultrametricity , by focusing on the fact that sparse high - dimensional data tend to be ultrametric . in such work
it is shown how numbers of points in our clouds of data points are irrelevant ; but what counts is the ambient spatial dimensionality . among cases looked at
are statistically uniformly ( hence `` unclustered '' , or without structure in a certain sense ) distributed points , and statistically uniformly distributed hypercube vertices ( so the latter are random 0/1 valued vectors ) . using our ultrametricity measure , there is a clear tendency to ultrametricity as the spatial dimensionality ( hence spatial sparseness ) increases . as @xcite also show , gaussian data behave in the same way and a demonstration of this is seen in table [ tabunifgauss ] .
to provide an idea of consensus of these results , the 200,000-dimensional gaussian was repeated and yielded on successive runs values of the ultrametricity measure of : 0.96 , 0.98 , 0.96 .
.typical results , based on 300 sampled triangles from triplets of points . for uniform , the data are generated on [ 0 , 1 ] ; hypercube vertices are in @xmath3 , and for gaussian , the data are of mean 0 , and variance 1 .
dimen . is the ambient dimensionality .
isosc . is the number of isosceles triangles with small base , as a proportion of all triangles sampled .
equil . is the number of equilateral triangles as a proportion of triangles sampled .
um is the proportion of ultrametricity - respecting triangles (= 1 for all ultrametric ) . [ cols="<,<,<,<,<",options="header " , ] we find clearly distinguishable peaks in figure [ fig10 ] .
the lower and the higher peaks belong to the two arima components .
the central peak belongs to the inter - cluster distances .
we have shown that our methodology can be of use for time series segmentation and for model identifiability .
we will assess this further in future work .
given the use of a hilbert space as the essential springboard of all aspects of this work , it would appear that generalization of this work to multivariate time series analysis is straightforward .
what remains important , however , is the availability of very large embedding dimensionalities , i.e. very high frequency data streams .
what we have observed in all of this work is that in the limit of high dimensionality a hilbert space becomes ultrametric .
it has been our aim in this work to link observed data with an ultrametric topology for such data . the traditional approach in data analysis , of course ,
is to impose structure on the data .
this is done , for example , by using some agglomerative hierarchical clustering algorithm .
we can always do this ( modulo distance or other ties in the data ) .
then we can assess the degree of fit of such a ( tree or other ) structure to our data .
for our purposes , here , this is unsatisfactory .
* firstly , our aim was to show that ultrametricity can be naturally present in our data , globally or locally .
we did not want any `` measuring tool '' such as an agglomerative hierarchical clustering algorithm to overly influence this finding .
( unfortunately @xcite suffers from precisely this unhelpful influence of the `` measuring tool '' of the subdominant ultrametric .
in other respects , @xcite is a seminal paper . ) * secondly , let us assume that we did use hierarchical clustering , and then based our discussion around the goodness of fit .
this again is a traditional approach used in data analysis , and in statistical data modeling .
but such a discussion would have been unnecessary and futile . for , after all , if we have ultrametric properties in our data then many of the widely used hierarchical clustering algorithms will give precisely the same outcome , and furthermore the fit is by definition optimal .
we have described an application of this work to very high frequency signal processing .
the twin objectives are signal segmentation , and model identification .
we have noted that a considerable amount of this work is model - based : we require assumptions ( on clusters , and on model(s ) ) for identifiability .
motivation for this work includes the availability of very high frequency data streams in various fields ( physics , engineering , finance , meteorology , bio - engineering , and bio - medicine ) . by using a very large embedding dimensionality
, we are approaching the data analysis on a very gross scale , and hence furnishing a particular type of multiresolution analysis . that this is worthwhile has been shown in our case studies .
f. murtagh , from data to the physics using ultrametrics : new results in high dimensional data analysis , in a.yu .
khrennikov , z. raki and i.v .
volovich , eds . , p - adic mathematical physics , american institute of physics conf
. proc .
826 , 151161 , 2006 .
r. rammal , g. toulouse and m.a .
virasoro , ultrametricity for physicists , reviews of modern physics , 58 , 765788 , 1986 .
rohlf and d.r .
fisher , tests for hierarchical structure in random data sets , systematic zoology , 17 , 407412 , 1968 . | an ultrametric topology formalizes the notion of hierarchical structure .
an ultrametric embedding , referred to here as ultrametricity , is implied by a natural hierarchical embedding .
such hierarchical structure can be global in the data set , or local . by quantifying extent or degree of ultrametricity in a data set ,
we show that ultrametricity becomes pervasive as dimensionality and/or spatial sparsity increases .
this leads us to assert that very high dimensional data are of simple structure .
we exemplify this finding through a range of simulated data cases .
we discuss also application to very high frequency time series segmentation and modeling .
pacs : 02.50.-r , 05.45.tp , 89.65.gh , 89.20.-a |
Subsets and Splits
Data Mining Abstracts
This query searches for and retrieves 2000 documents containing the phrase 'data mining' in the abstract, providing limited but specific document samples.