BODY %%%%%% ------ Cut Here ---------------------------- %%%%% ----- LaTeX file ----- %%%%%%% \documentstyle[12pt]{article} \addtolength\oddsidemargin{-0.5in} \renewcommand\topmargin{-0.8in} \addtolength\textheight{1.6in} \renewcommand\textwidth{6.35in} \vfuzz=4pt \date{February 1993} \begin{document} \pretolerance=5000 \parskip=9pt plus 2pt minus 1pt \parindent=22pt \tolerance=10000 \hfuzz=5pt \def\Tr{\mathop{\rm trace}} \title{Microscopic Shock Profiles: Exact Solution of a Nonequilibrium System} \author{B. Derrida\\Service de Physique Th\'eorique, CE Saclay\\[-2pt]F91191 Gif-sur-Yvette, France\\ and\\ S. A. Janowsky\thanks{Permanent address: Department of Mathematics, Rutgers University, New Brunswick, NJ 08903.}, J. L. Lebowitz\footnotemark[1]\hspace{2.5pt} and E. R. Speer\footnotemark[1]\\ Institut Des Hautes Etudes Scientifiques\\[-2pt] 91440 Bures-sur-Yvette, France} \maketitle\vspace{-20pt} \renewcommand\baselinestretch{1.56}\large\normalsize \begin{abstract} The full microscopic structure of macroscopic shocks is obtained exactly in the one dimensional totally asymmetric simple exclusion process from the complete solution of the uniform stationary nonequilibrium state of a system containing two types of particles---``first'' and ``second'' class. The width of the shock as seen from a second class particle diverges as the inverse square of its strength when the latter goes to zero. In that limit the steady state exhibits long range correlations and weakly bound states of second class particles. \end{abstract} Short title: Microscopic Shock Profiles PACS: 05.40 05.60 05.70L 64.60 To be submitted to Europhysics Letters \subsection*{Introduction} There is much interest, physical and mathematical, in the behavior of model systems of particles on a lattice evolving microscopically under conservative stochastic dynamical rules and described macroscopically by a continuous density field satisfying equations of hydrodynamic type~\cite{Spohn,LPS}. Of particular interest are models for which the macroscopic equations produce shocks. In such cases it is clear that the hydrodynamic equations do not tell the full story; in fact, because of the discontinuities in the density and consequent infinities in the derivatives which enter the equations, shocks correspond to nonunique weak solutions of these equations, and to find the physical solution the equations have to be supplemented by additional conditions or be treated as the limit of equations with nonzero viscosity. The microscopic level has of course no room for such extra conditions and its study is therefore essential for a complete understanding of the behavior of shocks~\cite{Sardinia}. Unfortunately even molecular dynamics simulations of shocks in real particle systems can only be carried out very partially with the currently available supercomputers~\cite{Sardinia}, so model systems such as the one dimensional asymmetric simple exclusion process are important. In this model, particles on the lattice jump independently at random times with rate 1 to adjacent sites, choosing the site to their right with probability $p$, and to their left with probability $1-p$, provided the target site is unoccupied~\cite{Spohn,L}. The density profile of this system on macroscopic spatial and temporal scales $x$ and $t$ is described for $p\ne 1/2$ in the appropriate hydrodynamic scaling limit by the inviscid Burgers' equation~\cite{Spohn,LPS}, \begin{equation}\label{Burger} \frac{\partial u(x,t)}{\partial t} + (2p-1) \frac{\partial}{\partial x} [u(1-u)] = 0. \end{equation} This means that there exists a limiting scaling for which the actual random microscopic density exactly tracks the solution of the Burgers' equation. This equation can produce shocks; for example, there is a solution of (\ref{Burger}) such that $u(x,t)=u_+$ for $x>x_0(t)$ and $u(x,t)=u_-$ for $xu_-$ (when $p>1/2$) and the shock position $x_0(t)$ moves with the constant velocity $(2p-1)(1-u_+-u_-)$. We now ask about the corresponding microscopic behavior: can one see an abrupt change at that scale or is the jump in densities spread out over very large microscopic distances? To analyze the shock microscopically one must first locate it---that is, define precisely its position---on the microscopic level. This may be done by introducing into the system a special {\em second class particle}~\cite{ABL}, which evolves by a modified dynamical rule: it behaves like a regular (first class) particle when jumping to an empty site but must give way to any regular particle that tries to jump on it (by exchanging places with the latter). For $p>1/2$ this dynamics gives the second class particle a drift towards regions of positive density gradient and so keeps it near the shock. The position of this particle may then conveniently be taken as the definition of the (microscopic) shock location. With this definition it has been established that shocks are sharp even on the microscopic level~\cite{DKPS,FKS}; specifically, that the microscopic (ensemble averaged) particle density at site $j$, as viewed from the second class particle, has a time invariant distribution which approaches the densities $\rho_\pm=u_\pm$ at $\pm\infty$. In practice, this means that one can see the discontinuity on the scale of a few intermolecular distances, something consistent with computer simulations on real particle systems and with experiments~\cite{Sardinia}. Computer studies of this shock profile are described in~\cite{BCFN}, see also~\cite{JL}. In this note we present the exact microscopic structure of such a shock for the case $p=1$. We obtain our results by first finding the exact translation invariant stationary state of a related model system, introduced in \cite{FKS}, which considers a system with uniform densities $\rho_1$ and $\rho_2$ of first and second class particles ({\bf fcp} and {\bf scp}). It is in fact easy to show that when this uniform system is viewed from some chosen scp, all scp's to the right of the chosen one behave as fcp's while all scp's to the left of it behave as unoccupied sites or holes. With this identification the chosen scp in the uniform system sees exactly the same microscopic shock profile as a single scp in a system containing only fcp's in which the density approaches $\rho_+ =\rho_1+\rho_2$ at $+\infty$ and $\rho_- =\rho_1$ at $-\infty$~\cite{FKS}. This two species model is of independent interest as an example of a driven diffusive system~\cite{BS}. Such systems have particle conserving transition rates which do not satisfy detailed balance with respect to the stationary state and are expected to exhibit long range spatial correlations in two or more dimensions~\cite{BS,GLMS,Grinstein}. Our one dimensional model also exhibits such behavior when the density of the second class particles goes to zero. For $\rho_2$ nonzero, the exponential decay of correlations in the uniform system gives an exponential approach to the asymptotic densities for the one component shock, with a characteristic length $\xi$ which diverges as $(\rho_+ - \rho_-)^{-2}$ when $(\rho_+ - \rho_-) = \rho_2 \rightarrow 0$. In the limit $\rho_2\rightarrow0$ there still exists a microscopic ``shock'' consisting of an excess density of particles in front of any given scp and of holes in back of it, which decays to zero as $|j|^{-1/2}$; in contrast the mean field result found in \cite{BCFN} is $|j|^{-1}$. In this limit there are also long range correlations between the second class particles: a pair of scp's in a system with a uniform nonzero density of fcp's will form a bound state with the probability of separation $r$ between them decaying as $r^{-3/2}$. \subsection*{Exact Solution} The exact stationary state of our two species totally asymmetric simple exclusion process is based on earlier work~\cite{DEHP}, see also~\cite{DDM}, for a one component finite system with open boundaries; the solution can be extended to the partially asymmetric case~\cite{DJLS}. We work first on a ring of $N$ sites containing $N_1$ fcp's, $N_2$ scp's and $N_0\equiv N-N_1-N_2$ holes. A configuration is specified by an $N$-tuple $(\tau_1,\tau_2,\ldots,\tau_N)$, where $\tau_i=0$ if site $i$ is empty, $\tau_i=1$ if there is an fcp at site $i$, and $\tau_i=2$ if there is an scp at site $i$. During a time interval $dt$ each adjacent pair of sites in the system of the type $1\,0$, $1\,2$ or $2\,0$ has probability $dt$ of being exchanged. It is easy to see that any configuration can evolve into any other, so that the system is ergodic and there is a unique stationary probability measure. This measure which assigns a weight $w(\tau)$ to each configuration $ \tau = (\tau_1,\tau_2,\ldots,\tau_N)$ is determined by the stationary form of the master equation. It satisfies the following balance condition \begin{equation} \sum_{\{\,i\mid\tau_i\tau_{i+1}=10,\,12,\;{\rm or}\;20\,\}}w(\tau) \;\;=\;\;\sum_{\{\,i\mid\tau_i\tau_{i+1}=01,\,21,\;{\rm or}\;02\,\}} w(\tau^{i,i+1})\,,\label{bal} \end{equation} where $\tau^{ij}$ denotes the configuration obtained from $\tau=(\tau_1,\ldots,\tau_N)$ by interchanging $\tau_i$ and $\tau_j$. If either $N_0$, $N_1$ or $N_2$ equals zero then the system reduces to the one component asymmetric simple exclusion process whose stationary measure is well known: all configurations are equally likely. In the general case the stationary measure will be constructed with the aid of matrices or operators $D$ and $E$ which satisfy~\cite{DEHP} \begin{equation} DE = D+E \label{DE} \end{equation} and the matrix $A = DE-ED$ which, from (\ref{DE}), satisfies the relations \begin{equation}\label{DAAE} DA = AE = A. \end{equation} We will show that the probability in the stationary measure for fixed $N_1, N_2$ of the configuration $\tau$ is then given~\cite{DJLS}, up to an overall normalizing factor, by the weight $w(\tau)$ defined by \begin{equation} w(\tau_1,\ldots,\tau_N)=\Tr[X_1\cdots X_N],\label{fund} \end{equation} where $X_i=E$ if $\tau_i=0$, $X_i=D$ if $\tau_i=1$, and $X_i=A$ if $\tau_i=2$. To verify that (\ref{fund}) provides the weights of a stationary measure, we substitute this formula on both sides of (\ref{bal}); it is then convenient to write $X_1\cdots X_N$, possibly after a cyclic permutation which will not affect the trace, in terms of blocks of consecutive identical matrices: $X_1\cdots X_N=Y_1^{k_1}\cdots Y_m^{k_m}$, where each $Y_j$ is $D$, $A$, or $E$, and $Y_j\ne Y_{j+1}$, $Y_m\ne Y_1$. Now in the term on the left hand side of (\ref{bal}) indexed by $i$ we use the algebraic relations (\ref{DE}) and (\ref{DAAE}) to replace the product $X_iX_{i+1}$ by a sum of two operators or a single operator---that is, $DE$ is replaced by $D+E$ while $DA$ and $AE$ are replaced by $A$. The net effect of these replacements is that each $D$ block (respectively $E$ block) gives rise to a term in which the rightmost (respectively leftmost) factor in the block has disappeared; notice that in the case of a DE boundary we get a term for each block. Thus we find \begin{equation} \sum_{\{\,i\mid\tau_i\tau_{i+1}=10,\,12,\;{\rm or}\;20\,\}} \Tr[X_1\cdots X_iX_{i+1}\cdots X_N] = \sum_{\{\,j\,|\, Y_j=D,E\,\}} \Tr[Y_1^{k_1}\cdots Y_j^{k_j-1}\cdots Y_m^{k_m}]. \label{left} \end{equation} If we make the same replacements for the product $X_{i+1}X_i$ in each term on the right hand side of (\ref{bal}), then the net effect is that $D$ blocks lose their leftmost and the $E$ blocks their rightmost factors, and we find that (\ref{fund}) satisfies (\ref{bal}). To obtain the matrices $D$, $E$ and $A$, write $D=I+d$ and $E=I+e$, where $I$ is the identity matrix. Then (\ref{DE}) gives $de=I$ which shows that to obtain a nontrivial solution we must take infinite matrices: an explicit solution~\cite{DEHP} is that $d$ is the right and $e$ the left shift on semi-infinite vectors: \begin{equation} D_{ij} = \delta_{i,j} + \delta_{i,j-1}, \qquad E_{ij} = \delta_{i,j} + \delta_{i,j+1} \quad {\rm and} \quad A_{ij} = \delta_{i,1}\delta_{j,1},\qquad i,j\ge 1. \end{equation} $A$ is then the projector on the first component, $A=|1\rangle \langle 1|$ with $|1\rangle = [1,0,0,\ldots]$, and has trace 1. An important consequence of this matrix formulation and of the fact that the matrix $A$ is a one dimensional projector is the absence of correlations between regions separated by second class particles. In particular, in the infinite volume limit, there is a factorization about any site $i$ occupied by a second class particle, \begin{equation} \langle f(\tau) g(\tau) | \tau_i=2\rangle = \langle f(\tau) | \tau_i=2\rangle \langle g(\tau) | \tau_i=2\rangle, \end{equation} whenever $f$ (respectively $g$) depends only on the configuration to the left (respectively right) of $i$. \subsection*{Shock structure and other properties} Once the weight of each configuration in the steady state is known, one can in principle calculate all the correlation functions for arbitrary $N_1$, $N_2$ and $N$, and then consider various infinite volume limits $N\rightarrow\infty$. In practice this limit is actually taken in a ``grand canonical'' ensemble after introducing fugacity-like parameters which are later converted back to densities. We consider several situations: a finite density of both first class and second class particles ($N_1/N\rightarrow\rho_1$, $N_2/N\rightarrow\rho_2$), a single second class particle with a nonzero density of first class particles ($N_1/N\rightarrow\rho$, $N_2=1$), and two second class particles with a nonzero density of first class particles ($N_1/N\rightarrow\rho$, $N_2=2$). All these cases can be treated using similar techniques. Here we will summarize our results in the first two cases, leaving the detailed derivation for a future longer version of the present work~\cite{DJLS}, and will sketch the calculation in the third and easiest case, that of two second class particles. We first consider the case of finite densities of both types of particles, in which for each choice of $\rho_1$ and $\rho_2$ the infinite volume limit yields a translation invariant measure which is stationary for the two species asymmetric exclusion dynamics. One interesting feature of the measure that emerges from explicit computation~\cite{DJLS} is that, conditioned on the presence of an scp at some site $i$, the distribution of fcp's to the {\em right} of $i$ and of holes to the {\em left} of $i$ are just product measures with densities $\rho_1$ and $1-\rho_1-\rho_2$ respectively. To obtain from this measure the shock profile in the one component system with asymptotic densities $\rho_+=\rho_1+\rho_2$ and $\rho_-=\rho_1$, we pick any scp in the two component system and let its instantaneous position specify the origin of the lattice. We then define $\delta_j$ as the probability that site $j$ is occupied by an fcp for $j<0$ and that it is occupied by either an fcp or scp for $j>0$. An explicit computation gives~\cite{DJLS} \begin{equation}\label{eta1-1} \delta_{-1} = \rho_+\rho_-, \qquad \delta_{1} = \rho_+ + \rho_- - \rho_+\rho_-, \end{equation} and for all $j \geq 2$, \begin{equation} \delta_{-j} - \delta_{-1} = \delta_1 - \delta_j = \sum_{n=1}^{j-1} \sum_{p=0}^{n-1} { n \choose p} { n-1 \choose p} \frac{(\rho_+ \rho_-)^{n -p} [(1- \rho_+)(1- \rho_-)]^{p+1}}{p+1}. \label{deltaeta} \end{equation} That this is indeed the shock profile in the one component totally asymmetric simple exclusion process, seen in the randomly moving frame of a single scp, follows from the observation made earlier that each scp in the two component system behaves as if all the scp's at its left were holes and all scp's at its right were fcp's. Notice that these shocks have the symmetry property $\delta_j + \delta_{-j} = \rho_+ + \rho_-$. It is also easy to check from (\ref{eta1-1}) and (\ref{deltaeta}) that $\delta_j$ converges exponentially: $ |\delta_j - \rho_{\pm}| \sim \exp [-|j|/ \xi] $ as $j \rightarrow \pm \infty$ when $\rho_+ > \rho_-$, where the characteristic length $\xi$ measures the microscopic width of the shock as seen from the scp. This length $\xi$ diverges as $(\rho_+ - \rho_-)^{-2}$ when $\rho_+ \rightarrow \rho_-$. We can obtain the limiting shock shape when $\rho_-=\rho_+$ directly from (\ref{deltaeta}) or we can equivalently consider the case of a single scp in a system of fcp's at a uniform nonzero density $\rho$. Physically, one expects that the scp will be attracted by fluctuations with a positive density gradient of fcp's, so that the average density $ d_j$ of fcp's at position $j$ (in the frame where the scp is at the origin) will have a nontrivial profile. We have calculated this profile exactly \cite{DJLS} from the steady state (\ref{fund}) and found in the limit of an infinite system size ($N \rightarrow \infty$ with $N_1 / N \rightarrow \rho$ and $N_2=1$) that $d_1 = 2\rho -\rho^2$, $d_{-1} = \rho^2$, and for all $j \geq 2$, \begin{equation} d_{-j} - d_{-1} = d_1 - d_j = \sum_{n=1}^{j-1} \sum_{p=0}^{n-1} {n \choose p} {n-1 \choose p} \frac{(1 - \rho)^{2n - 2 p} \rho^{2 p + 2 }}{p+1}. \end{equation} From this exact expression, which is equivalent to that obtained from (\ref{deltaeta}) for $\delta_j$ when $\rho_2\searrow0$ and $\rho_1=\rho$, one can show that for large $j$ the profile is a power law: $ |d_j - \rho| \sim |j|^{- 1 / 2}$. The exponent $1/2$ disagrees with the exponent $1$ obtained from a mean field argument~\cite{BCFN}. The last case we consider is the case of a pair of scp's in an infinite system of fcp's at a nonzero density $\rho$. In that case, we find that the probability $P_\infty(r)$ that two scp's are a distance $r $ apart (in an infinite system) is given by \begin{equation}\label{P-separation} P_\infty(r) = \sum_{p=0}^{r-1} {r \choose p} {r-1 \choose p} \frac{\rho^{2n -2p-1} (1- \rho)^{2p+1}}{p+1}. \end{equation} This shows that the two second class particles form a bound state, of a rather unusual kind since $ P_\infty(r) \sim r^{-3/2}$ and the average distance is infinite. (We can say that this is an {\em algebraic} bound state in contrast with the normally encountered exponentially decaying bound states.) The derivation of (\ref{P-separation}) is rather easy when one uses the grand canonical ensemble. If we introduce a parameter $x$ which plays the role of a fugacity for the fcp, so that the weight of any configuration with $N_1$ first class particles is multiplied by $x^{N_1}$, then the probability (in the grand canonical ensemble) of finding the two second class particles a distance $r$ apart, on a ring of $N$ sites (for simplicity we choose here $N$ to be odd and $r$ to be the shorter distance between the two particles on the ring) is \begin{equation} P_N(r) = \langle 1 | C^{r-1} | 1\rangle \langle 1 | C^{N-r-1} | 1\rangle \left[\, \sum_{s=1}^{(N-1)/2} \langle 1 | C^{s-1} | 1\rangle \langle 1 | C^{N-s-1} | 1\rangle \right]^{-1}, \label{E1} \end{equation} where the matrix $C$ is defined by \begin{equation} C= x D + E, \end{equation} i.e., $C_{ij} = \delta_{i,j+1} +(1+x) \delta_{i,j} + x \delta_{i,j-1} $ for $i \geq 1 $ and $j \geq 1$. The expression (\ref{E1}) arises because when one sums over all the occupations of the $N-2$ sites one is adding at each site a matrix $D$ with weight $x$ and a matrix $E$ with weight $1$, giving at each site a matrix $C$. The limit $N \rightarrow \infty$ then follows easily: from the explicit expression of the matrix $C$, whose elements can be interpreted in terms of random walks, one can show that for any $n$~\cite{DJLS}, \begin{equation} \langle 1 | C^n | 1 \rangle = \sum_{p=0}^{n} {n+1 \choose p} {n \choose p} \frac{x^p}{p+1}, \label{E2} \end{equation} which gives for large $n$ \begin{equation} \langle 1 | C^n | 1 \rangle \sim (\sqrt{x} +1)^{2n} \Big/ n^{3/2}. \label{E3} \end{equation} From this asymptotic behavior, it follows that the density $\rho$ is related in the large $N$ limit to the fugacity $x$ by $\rho = \sqrt{x}\big/ ( \sqrt{x} + 1 )$; then replacing $x$ by $\rho^2/(1- \rho)^2 $ in (\ref{E1}) and (\ref{E2}) one obtains in the large $N$ limit (\ref{P-separation}). \subsubsection*{Acknowledgments} This work was supported in part by AFOSR Grant 91-0010; SAJ was supported in part by NSF Mathematical Sciences Postdoctoral Research Fellowship DMS 90-07206. BD thanks the Mathematics Department of Rutgers University, where this work was initiated, for its hospitality. We also thank the ENS in Paris where it was completed. \renewcommand\baselinestretch{1.08}\large\normalsize \begin{thebibliography}{99} \bibitem{Spohn} H. Spohn, {\em Large-Scale Dynamics of Interacting Particles}, Texts and Monographs in Physics, Springer-Verlag, 1991; A. de Masi and E. Presutti, {\em Mathematical Methods for Hydrodynamic Limits}, Lecture Notes in Mathematics 1501, Springer-Verlag, 1991; and references therein. \bibitem{LPS} J. Lebowitz, E. Presutti and H. Spohn, ``Microscopic models of hydrodynamic behavior,'' {\em J. Stat. Phys.} {\bf 51}, 841--862 (1988). \bibitem{Sardinia} {\em Microscopic Simulations of Complex Hydrodynamic Phenomena}, Proceedings of the NATO Advanced Study Institute, Alghero, 1991, Plenum, 1992. \bibitem{L} T. M. Liggett, {\em Interacting Particle Systems,} Springer-Verlag, 1985. \bibitem{ABL} E. D. Andjel, M. Bramson, and T. M. Liggett, ``Shocks in the asymmetric exclusion process,'' {\em Prob. Th. Rel. Fields} {\bf 78}, 231-247 (1988). \bibitem{DKPS} A. DeMasi, C. Kipnis, E. Presutti, and E. Saada, ``Microscopic structure at the shock in the asymmetric simple exclusion,'' {\em Stoch. and Stoch. Rep.} {\bf 27}, 151--165 (1989). \bibitem{FKS} P. Ferrari, C. Kipnis, and E. Saada, ``Microscopic structure of traveling waves in the asymmetric simple exclusion,'' {\it Ann. Prob.} {\bf 19}, 226--244 (1991). \bibitem{BCFN} C. Boldrighini, G. Cosimi, S. Frigio and M. G. Nu\~nes, ``Computer simulation of shock waves in the completely asymmetric simple exclusion process,'' {\em J. Stat. Phys} {\bf 55}, 611--623 (1989). \bibitem{JL} S. A. Janowsky and J. L. Lebowitz, ``Finite Size Effects and Shock Fluctuations in the Asymmetric Simple Exclusion Process,'' {\em Phys. Rev. A} {\bf 45}, 618--625 (1992). \bibitem{BS} B. Schmittman: ``Critical behavior of the driven diffusive lattice gas,'' {\em Int. J. Mod. Phys.} {\bf B4}, 2269--2306 (1990). \bibitem{GLMS} P. Garrido, J. Lebowitz, C. Maes and H. Spohn, ``Long-range correlations for conservative dynamics,'' {\em Phys. Rev. A} {\bf 42}, 1954--1968 (1990). \bibitem{Grinstein} R. Bhagavatula, G. Grinstein, Y. He and C. Jayaprakash, ``Algebraic Correlations in Conserving Chaotic Systems,'' {\em Phys. Rev. Lett.} {\bf 69}, 3483--3486 (1992). \bibitem{DEHP} B. Derrida, M. R. Evans, V. Hakim and V. Pasquier, ``An Exact Solution of a 1D Asymmetric Exclusion Model Using a Matrix Formulation,'' {\em J. Phys. A}, to appear. \bibitem{DDM} B. Derrida, E. Domany and D. Mukamel, ``An exact solution of a one dimensional asymmetric exclusion model with open boundaries,'' {\em J. Stat.\ Phys.} {\bf 69}, 667--687 (1992). \bibitem{DJLS} B. Derrida, S. A. Janowsky, J. L. Lebowitz and E. R. Speer, ``Exact Solution of the Totally Asymmetric Simple Exclusion Process: Shock Profiles,'' in preparation. \end{thebibliography} \end{document} % -- CUT HERE -- -- CUT HERE -- -- CUT HERE -- -- CUT HERE -- %