1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \catcode`\@=11
4: \topmargin 0pt
5: \oddsidemargin 0pt
6: \headheight 0pt
7: \headsep 0pt
8: \textheight 9in
9: \textwidth 6.25in
10: \marginparwidth .875in
11: \def\numberbysection{\@addtoreset{equation}{section}
12: \def\theequation{\thesection.\arabic{equation}}}
13: \def\baselinestretch{1.1}
14: \numberbysection
15: \newcommand{\be}{\[}
16: \newcommand{\bea}{\begin{eqnarray*}}
17: \newcommand{\beq}{\begin{equation}}
18: \newcommand{\beqa}{\begin{eqnarray}}
19: \newcommand{\ee}{\]}
20: \newcommand{\eea}{\end{eqnarray*}}
21: \newcommand{\eeq}{\end{equation}}
22: \newcommand{\eeqa}{\end{eqnarray}}
23: \newcommand{\abs}[1]{\vert#1\vert}
24: \newcommand{\ap}{_{\rm ap}}
25: \newcommand{\chat}{{\widehat\chi}}
26: \newcommand{\cond}{{\rm cond}}
27: \newcommand{\cu}[1]{\langle\!\langle#1\rangle\!\rangle}
28: \renewcommand{\d}{{\rm d}}
29: \newcommand{\diag}{{\rm diag}}
30: \newcommand{\dif}{{\rm dif}}
31: \newcommand{\dom}{{\rm dom}}
32: \newcommand{\e}{{\rm e}}
33: \newcommand{\ev}{{\rm even}}
34: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}
35: \renewcommand{\i}{{\rm i}}
36: \newcommand{\lam}{\lambda}
37: \newcommand{\mean}[1]{\langle#1\rangle}
38: \newcommand{\od}{{\rm odd}}
39: \renewcommand{\r}{{\rm r}}
40: \newcommand{\ret}{{\rm ret}}
41: \newcommand{\s}{\sigma}
42: \newcommand{\spin}{{\rm spin}}
43: \newcommand{\st}{{^\star}}
44: \newcommand{\toinf}{\mathrel{\mathop{\longrightarrow}\limits_{N\to\infty}}}
45: \newcommand{\tot}{\leftrightarrow}
46: \newcommand{\var}{\mathop{\rm Var}\nolimits}
47: \newcommand{\C}{{\cal C}}
48: \renewcommand{\H}{{\cal H}}
49: \newcommand{\N}{{\cal N}}
50: \renewcommand{\P}{{\cal P}}
51: \renewcommand{\S}{\Sigma}
52: \newcommand{\Sp}{{\cal S}}
53: \newcommand{\T}{{\cal T}}
54: \newcommand{\1}{{\bf 1}}
55:
56: \begin{document}
57: \centerline{\Large\bf Metastable states of the Ising chain}
58: \vspace{.3cm}
59: \centerline{\Large\bf with Kawasaki dynamics}
60: \vspace{1.6cm}
61: \centerline{\large
62: G.~De Smedt$^{a,}$\footnote{desmedt@spht.saclay.cea.fr},
63: C.~Godr\`eche$^{b,}$\footnote{godreche@spec.saclay.cea.fr},
64: and J.M.~Luck$^{a,}$\footnote{luck@spht.saclay.cea.fr}}
65: \vspace{1cm}
66: \centerline{$^a$Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},
67: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}
68: \centerline{$^b$Service de Physique de l'\'Etat Condens\'e,
69: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}
70: \vspace{1cm}
71:
72: \begin{abstract}
73: We consider a ferromagnetic Ising chain
74: evolving under Kawasaki dynamics at zero temperature.
75: We investigate the statistics of the
76: metastable configurations in which the system gets blocked
77: (statistics of energy, spin correlations, distribution of domain sizes).
78: A systematic comparison is made with analytical predictions
79: for the ensemble of all blocked configurations
80: taken with equal a priori weights (Edwards approach).
81: \end{abstract}
82:
83: \vfill
84: \noindent To be submitted for publication to the European Physical Journal B
85:
86: \noindent P.A.C.S.: 05.70.Ln, 64.60.My, 75.40.Gb.
87:
88: \newpage
89: \setcounter{footnote}{0}
90: \section{Introduction}
91:
92: Glassy dynamics is often described as motion
93: in a complex energy landscape~\cite{gold},
94: with many valleys separated by barriers.
95: Valleys have been given various definitions~\cite{tap,ktw,sw,fv},
96: which are not equivalent from a dynamical point of view~\cite{bir}.
97: One common feature of all these approaches
98: is the exponential growth with the system size
99: of the number of valleys at fixed energy density $E$:
100: \beq
101: \N(N;E)\sim\exp(N S\ap(E)),
102: \label{ne}
103: \eeq
104: where $S\ap(E)$ is the `a priori' configurational entropy, or complexity.
105:
106: As valleys are so numerous, an important issue
107: concerns the possibility of giving an a priori statistical description of them.
108: The key question to address is then the following.
109: Assuming the initial configuration be random,
110: does the system sample all the possible valleys with equal statistical weights,
111: i.e., with a uniform or flat measure, or, to the contrary,
112: is the size of the basin of attraction of each valley relevant?
113: The assumption that an a priori statistical description
114: with a flat measure could hold
115: was first formulated by Edwards~\cite{edwards}
116: in the context of granular materials, and is therefore
117: referred to as the Edwards hypothesis.
118:
119: One-dimensional spin systems at zero temperature
120: provide an adequate setting for the investigation of the dynamical role
121: of valleys, and especially for testing the Edwards hypothesis.
122: Indeed, the blocked configurations reached by zero-temperature dynamics
123: are truly metastable states with infinite lifetimes.
124: In an earlier work~\cite{us} we have addressed this question for Ising chains
125: evolving under irreversible zero-temperature dynamics,
126: where each spin may flip at most once before a blocked state is reached.
127: These dynamical models can be exactly mapped
128: onto random sequential adsorption problems~\cite{rsa},
129: for which analytical tools are available.
130:
131: In the present work we pursue the investigation
132: of metastable states in Ising chains, by considering the richer
133: situation of zero-temperature Kawasaki (spin-exchange) dynamics~\cite{kawa}.
134: We allow the rate $W_0$ of diffusion processes
135: (constant-energy moves) to vary~\cite{cks}.
136: If the kinetic constraint $W_0=0$ is imposed~\cite{plk},
137: only irreversible processes are present~\cite{us}.
138: For the unconstrained dynamics $(W_0>0)$,
139: where diffusion processes are allowed, the dynamics is partly reversible.
140: Each spin only flips a finite number of times,
141: before the system globally reaches a blocked state after a finite time.
142: The statistics of this blocking time is studied in Section~\ref{s1}.
143: Our main goal is then again to test the Edwards hypothesis,
144: i.e., to evaluate various observables in the blocked configurations
145: reached by the dynamics,
146: starting from a random non-magnetized initial configuration.
147: We shall systematically compare the data of extensive numerical simulations,
148: presented in Section~\ref{s4},
149: with the predictions of the a priori approach, derived in Section~\ref{s3}.
150: Finally, some aspects of persistence,
151: in particular the distribution of the number of flips of a given spin,
152: are discussed in Section~\ref{s5}.
153:
154: To be more specific,
155: we consider a ferromagnetic chain of Ising spins $\s_n=\pm1$,
156: whose Hamiltonian reads
157: \beq
158: \H=-\sum_n\s_n\s_{n+1}.
159: \label{hamf}
160: \eeq
161:
162: In Kawasaki dynamics~\cite{kawa},
163: only pairs of opposite spins may be flipped $(+-\tot-+)$,
164: so that the magnetization is locally conserved.
165:
166: For simplicity, we limit ourselves to Monte-Carlo dynamics
167: with random sequential updating, and we assume that the flipping rate
168: only depends on the energy difference $\delta\H$ involved in the proposed move.
169: Any given pair of opposite spins thus has a probability
170: $W_{\delta\H}$ of flipping per unit time, with
171: \be
172: \delta\H=2(\s_{n-1}\s_n+\s_{n+1}\s_{n+2})\in\{-4,0,4\}.
173: \ee
174: The requirement of detailed balance
175: with respect to the Hamiltonian~(\ref{hamf}) at temperature $T=1/\beta$
176: yields a single condition,
177: \be
178: \frac{W_4}{W_{-4}}=\e^{-4\beta},
179: \ee
180: upon the rates $W_4$, $W_0$, $W_{-4}$.
181:
182: We furthermore restrict ourselves to zero-temperature dynamics,
183: hence $W_4=0$.
184: We choose time units such that $W_{-4}=1$, keeping $W_0$ as a free parameter.
185: The allowed moves and the corresponding rates are listed in Table~\ref{t1}.
186: The zero-temperature limits of the
187: Metropolis and heat-bath rules correspond respectively
188: to $W_0=1$ and $W_0=1/2$.
189: Hereafter we focus our attention on the range $0\le W_0\le1$.
190:
191: \begin{table}[htb]
192: \begin{center}
193: \begin{tabular}{|c|c|c|c|c|}
194: \hline
195: $\delta\H$&type&name&rate&moves\\
196: \hline
197: $-4$&irreversible&condensation&1&$\matrix{-+-\,+\to--+\,+\cr+-+\,-\to++-\,-}$\\
198: \hline
199: 0&reversible&diffusion&$W_0$&$\matrix{++-\,+\tot+-+\,+\cr-+-\,-\tot--+\,-}$\\
200: \hline
201: \end{tabular}
202: \caption{\small Allowed moves in zero-temperature Kawasaki dynamics.}
203: \label{t1}
204: \end{center}
205: \end{table}
206:
207: \section{Statistics of blocking time}
208: \label{s1}
209:
210: Let us first recall that for the kinetically constrained model $(W_0=0)$
211: a finite system consisting of~$N$ spins
212: reaches a blocked configuration after a finite blocking time $T_N\sim\ln N$,
213: which is the jamming time of the equivalent problem
214: of random sequential adsorption of hollow trimers~\cite{us}.
215: The blocked configurations are characterized by the property that
216: the spin patterns $+-+-\null$ and $-+-+\null$ are absent.
217: Equivalently, there are at most two consecutive unsatisfied bonds.
218:
219: In the present case $(W_0>0)$
220: the system still gets trapped in a blocked configuration.
221: However the diffusive motion of free $+$ spins in domains of $-$ spins,
222: and vice-versa, is allowed.
223: Each free spin will eventually be annihilated,
224: by meeting either another free spin or one of the boundaries of the domain.
225: Blocked configurations of Kawasaki dynamics
226: are therefore characterized by the property that
227: the patterns $+-+\null$ and $-+-\null$ are absent.
228: Equivalently, isolated spins are absent, or unsatisfied bonds are isolated.
229:
230: In order to understand the statistics of the blocking time $T_N$,
231: we consider first the regime $W_0\ll1$,
232: where the time scales of condensation and diffusion are well
233: separated~\cite{cks}.
234: The fast part of the dynamics, which occurs with unit rate,
235: is identical to the constrained dynamics considered in Refs.~\cite{us,plk}.
236: For intermediate times of order $1\ll t\ll1/W_0$,
237: the system is therefore approximately left in one of the final configurations
238: of the constrained dynamics.
239: The slow, diffusive part of the dynamics then takes place at rate $W_0\ll1$.
240: The late stages of the dynamics are governed by large domains,
241: on which a single free spin diffuses.
242: We assume that such large domains, of size $L\gg1$,
243: occur with an exponentially small probability
244: \be
245: f_\dif(L)\sim\exp(-L/\xi_\dif),
246: \ee
247: where $\xi_\dif$ is the relevant characteristic length.
248: The spin diffusion constant reads $D=W_0$ in our units.
249: For a random initial point,
250: the survival probability is known~\cite{trap} to decay as
251: $S(t;L)\approx(8L/\pi^2)\exp(-\pi^2W_0t/L^2)$.
252: The mean density of free spins at time~$t$ can therefore be estimated as
253: \beq
254: S(t)\approx\sum_L f_\dif(L)S(t;L)
255: \sim\int_0^\infty\exp\left(-\frac{L}{\xi_\dif}-\frac{\pi^2W_0t}{L^2}\right)
256: \label{sint}
257: L\,\d L.
258: \eeq
259: Evaluating the integral by the method of steepest descent,
260: we thus obtain stretched exponential decay for the density of free spins:
261: \beq
262: S(t)\sim t^{1/2}\,\exp\left(-\frac{(W_0t)^{1/3}}{A_\dif}\right),
263: \label{st}
264: \eeq
265: with
266: \beq
267: A_\dif=\left(\frac{4\xi_\dif^2}{27\pi^2}\right)^{1/3}.
268: \label{ad}
269: \eeq
270:
271: This behavior, already emphasized in Ref.~\cite{cks},
272: is a general characteristic feature of diffusion processes
273: in the presence of random traps~\cite{trap}.
274: The exponent $1/3$ is related to the one-dimensional geometry;
275: it would read $d/(d+2)$ for trapping problems in higher spatial dimension $d$.
276: The above result should hold for the late stages $(W_0t\gg1)$
277: of Kawasaki dynamics, for any finite value of $W_0$.
278: The characteristic length $\xi_\dif$ is expected to depend smoothly on $W_0$.
279:
280: For a finite system of~$N$ spins, the last free spin will be annihilated
281: at a time $T_N$ such that $NS(T_N)\sim1$,
282: hence $(W_0T_N)^{1/3}\sim A_\dif\ln N$.
283: More precisely, as the histories of spins diffusing
284: on different domains are statistically independent,
285: it can be argued along the lines of~\cite{us}
286: that $T_N$ is given according to extreme-value statistics~\cite{gumbel} as
287: \beq
288: (W_0T_N)^{1/3}\approx A_\dif\left(\,\ln\!\Bigl(N(\ln N)^{3/2}\Bigr)
289: +b+X_N\right).
290: \label{xdef}
291: \eeq
292: In this expression, the factor $(\ln N)^{3/2}$ takes account of the
293: $t^{1/2}$ prefactor in the survival probability~(\ref{st}),
294: while the effective constant~$b$ encompasses all subleading effects
295: that have been neglected,
296: and the random variable $X_N$ is distributed according to the Gumbel law
297: \beq
298: f(X)=\exp(-X-\e^{-X}).
299: \label{gum}
300: \eeq
301:
302: These predictions have been checked against extensive numerical simulations,
303: performed according to the
304: zero-temperature Kawasaki dynamics summarized in Table~\ref{t1},
305: with random sequential updates and periodic boundary conditions,
306: starting from a random initial configuration with zero magnetization.
307: The dynamics is run until the system gets trapped in a blocked configuration.
308: The blocking time $T_N$ is recorded for each sample.
309: This measurement has been performed for many samples
310: ($10^8$ spins in total for each value of $W_0$ and of $N$).
311: The mean $\mean{(W_0T_N)^{1/3}}$ is found to follow an almost perfect
312: linear law when plotted against $\ln(N(\ln N)^{3/2})$,
313: at least for~$N$ ranging from 50 to 3200,
314: whereas the same data plotted against $\ln N$ are bent in a significant way.
315: Using~(\ref{ad}), the slope of the latter plot yields
316: the value of the characteristic length $\xi_\dif$,
317: which is plotted in Figure~\ref{f1}, against $W_0$.
318: The error on this estimate, containing a systematic and a statistical part,
319: is roughly comparable to the symbol size.
320: The length $\xi_\dif$ exhibits a rather weak dependence on $W_0$,
321: decreasing from $\xi_\dif=1.77$ in the $W_0\to0$ limit
322: to $\xi_\dif=1.61$ for $W_0=1$.
323:
324: \begin{figure}[htb]
325: \begin{center}
326: \includegraphics[angle=90,width=.7\linewidth]{f1.eps}
327: \caption{\small
328: Plot of the characteristic length $\xi_\dif$,
329: extracted from the size dependence of $\mean{(W_0T_N)^{1/3}}$,
330: against the rate $W_0$.
331: Symbols: numerical data.
332: Errors are comparable to the symbol size.
333: Line: third-degree polynomial fit.}
334: \label{f1}
335: \end{center}
336: \end{figure}
337:
338: Another confirmation of the above picture is provided by Figure~\ref{f2},
339: showing a histogram plot of the variable $X_N$
340: defined by~(\ref{xdef}), for 125,000 samples of size $N=800$, with $W_0=1$.
341: The parameter $A_\dif=0.339$, i.e., $\xi_\dif=1.61$,
342: is taken from the data of Figure~\ref{f1},
343: while the constant~$b$ is chosen by fitting the average $\mean{X_N}$.
344: A convincing agreement is found with the limit law~(\ref{gum}).
345:
346: \begin{figure}[htb]
347: \begin{center}
348: \includegraphics[angle=90,width=.7\linewidth]{f2.eps}
349: \caption{\small
350: Distribution of the variable $X_N$ defined by~(\ref{xdef}).
351: Histogram: numerical data for $N=800$ (see text).
352: Thick line: limit Gumbel law~(\ref{gum}).}
353: \label{f2}
354: \end{center}
355: \end{figure}
356:
357: \section{Blocked states: a priori ensemble}
358: \label{s3}
359:
360: As already stated, the blocked states of zero-temperature Kawasaki dynamics
361: are the configurations where isolated spins are absent.
362: Equivalently, unsatisfied bonds are isolated.
363:
364: This section is devoted to the statistical description of the blocked
365: configurations taken with equal a priori weights (Edwards approach).
366: We shall distinguish the {\it full} ensemble
367: of all the blocked configurations, irrespective of their energy,
368: the {\it restricted} ensemble of blocked configurations
369: with prescribed energy density $E$,
370: and the {\it canonical} ensemble of blocked configurations,
371: obtained by fixing the parameter $\beta$ conjugate to the energy density $E$
372: (see below).
373:
374: \subsection{Statistics of energy and configurational entropy}
375:
376: For a finite chain of~$N$ spins,
377: we first investigate the number $\N(N;E)$ of blocked configurations
378: with prescribed energy density $E$.
379: This number can be evaluated
380: by an elementary combinatorial reasoning~\cite{us}.
381: For the sake of generality,
382: we prefer to resort to the transfer-matrix formalism.
383: We introduce the partition function $Z_N(\beta)$,
384: defined as a sum over all the blocked configurations $\C=\{\s_n\}$
385: of the Boltzmann weight associated with the Hamiltonian~(\ref{hamf}).
386: We have
387: \beq
388: Z_N(\beta)=\sum_\C\e^{-\beta\H(\C)}\approx\int\N(N;E)\,\e^{-\beta NE}\,\d E.
389: \label{zn}
390: \eeq
391: In this framework, the parameter $\beta$ can be positive or negative,
392: and it is not related to physical temperature.
393: The transfer matrix is then a very useful tool.
394: Indeed the partition functions $Z_N^\alpha$,
395: labeled by the prescribed value $\alpha=(\s_{N-1},\s_N)$
396: of the last two spins, obey the recursion relation
397: \be
398: \pmatrix{Z_{N+1}^{++}\cr Z_{N+1}^{+-}\cr Z_{N+1}^{-+}\cr Z_{N+1}^{--}}
399: =\T\pmatrix{Z_N^{++}\cr Z_N^{+-}\cr Z_N^{-+}\cr Z_N^{--}},
400: \ee
401: where the $4\times4$ transfer matrix $\T$ reads
402: \be
403: \T=\pmatrix{\e^\beta&0&\e^\beta&0\cr\e^{-\beta}&0&0&0\cr
404: 0&0&0&\e^{-\beta}\cr 0&\e^\beta&0&\e^\beta}.
405: \ee
406:
407: The characteristic polynomial of $\T$ factors as
408: $\det(\lam\1-\T)=(\lam^2-\e^{\beta}\lam-1)(\lam^2-\e^{\beta}\lam+1)$.
409: The first (resp.~the second) factor yields eigenvalues $\lam_1$, $\lam_2$
410: (resp.~$\lam_3$, $\lam_4$), with
411: \be
412: \lam_{1,2}=\frad{1}{2}\left(\e^{\beta}\pm\sqrt{\e^{2\beta}+4}\right),\qquad
413: \lam_{3,4}=\frad{1}{2}\left(\e^{\beta}\pm\sqrt{\e^{2\beta}-4}\right).
414: \ee
415: The left and right eigenvectors $\langle L_a\vert$ and $\vert R_a\rangle$
416: are even (resp.~odd) for $a=1,2$ (resp.~$a=3,4$)
417: under the spin symmetry $+\tot-$.
418: We assume that they are normalized so that
419: $\langle L_a\vert R_b\rangle=\delta_{ab}$.
420: Their explicit expressions will not be needed hereafter.
421:
422: For large $N$, we have $Z_N(\beta)\sim\lam_1^N$,
423: as $\lam_1$ is the largest eigenvalue.
424: Using~(\ref{zn}), we obtain an exponential law of the form~(\ref{ne})
425: for $\N(N;E)$, where the a priori configurational entropy $S\ap(E)$
426: is related to $\ln\lam_1(\beta)$ by a Legendre transform:
427: \be
428: S\ap(E)-\ln\lam_1(\beta)=\beta E,
429: \qquad E=-\frac{\d\ln\lam_1}{\d\beta},
430: \qquad\beta=\frac{\d S\ap}{\d E}.
431: \ee
432: Explicitly, we have
433: \beq
434: E=-\frac{\e^\beta}{\sqrt{\e^{2\beta}+4}},\qquad
435: \e^\beta=\frac{-2E}{\sqrt{1-E^2}},
436: \label{eb}
437: \eeq
438: and the a priori entropy reads
439: \be
440: S\ap(E)=E\ln(-2E)+\frac{1-E}{2}\ln(1-E)-\frac{1+E}{2}\ln(1+E).
441: \ee
442: This entropy is non-zero for $-1<E<0$.
443: It takes its maximal value
444: \be
445: S\st=\ln\Phi=0.481212,
446: \ee
447: where $\Phi=(1+\sqrt{5})/2$ is the golden mean,
448: for
449: \beq
450: E\st=-\frac{1}{\sqrt{5}}=-0.447214,
451: \label{est}
452: \eeq
453: corresponding to $\beta=0$.
454: Equation~(\ref{est}) therefore yields the typical a priori energy density
455: of a blocked configuration.
456:
457: The result~(\ref{ne}) can be recast as follows.
458: Consider the full ensemble of all the blocked configurations,
459: irrespective of their energy.
460: The probability of observing, in that ensemble,
461: a blocked configuration with energy density~$E$ reads
462: \beq
463: P\ap(E)\sim\exp(-N\S\ap(E)),\qquad\S\ap(E)=S\st-S\ap(E).
464: \label{sap}
465: \eeq
466: The function $\S\ap(E)$ vanishes quadratically as
467: \be
468: \S\ap(E)\approx c\left(E-E\st\right)^2,\qquad c=\frac{5\sqrt{5}}{8}.
469: \ee
470: The bulk of the a priori distribution of $E$
471: is therefore a narrow Gaussian around $E\st$,
472: whose rescaled variance asymptotically reads
473: \be
474: N\var E=N(\mean{E^2}-E\st^2)\toinf\frac{1}{2c}=\frac{4\sqrt{5}}{25}=0.357771.
475: \ee
476:
477: The above results also allow to determine the higher cumulants
478: of the energy in the canonical ensemble (at fixed parameter $\beta$).
479: One has indeed
480: \be
481: \frac{1}{N}\ln\mean{\e^{sNE}}_\beta
482: =\frac{1}{N}\ln\frac{Z_N(\beta-s)}{Z_N(\beta)}
483: \toinf\ln\frac{\lam_1(\beta-s)}{\lam_1(\beta)}.
484: \ee
485: By expanding this result as a power series in $s$,
486: we obtain explicit expressions for the cumulants $\cu{E^k}$
487: as a function of $\beta$.
488:
489: As far as mean quantities are concerned,
490: the microcanonical or restricted ensemble (fixed energy density $E$)
491: and the canonical one (fixed conjugate parameter~$\beta$) are equivalent.
492: It is therefore justified to recast canonical results
493: in terms of the prescribed value~$E$ of the mean energy, using~(\ref{eb}).
494: Generalizing this equivalence prescription to the cumulants of the energy,
495: we obtain
496: \beqa
497: N\var{E}=N\cu{E^2}&\toinf&-E(1-E^2),\nonumber\\
498: N^2\cu{E^3}&\toinf&-E(1-E^2)(3E^2-1),\label{ecu}\\
499: N^3\cu{E^4}&\toinf&-E(1-E^2)(15E^4-12E^2+1).\nonumber
500: \eeqa
501: %The energy cumulants are definitely not mean quantities.
502: The results~(\ref{ecu}) can be given the following interpretation.
503: Fixing $\beta$ amounts to fixing the extensive part $NE$ of the energy,
504: while the equivalence prescription
505: is a natural Ansatz to describe the fluctuations of its non-extensive part.
506:
507: \subsection{Spin correlation function}
508:
509: The spin correlation function $C_n=\mean{\s_0\s_n}$ in the canonical
510: a priori ensemble can also be evaluated by the transfer-matrix method.
511: In the bulk of an infinitely long chain, and for $n\ge0$, we have
512: \beq
513: C_n=\frad{\langle L_1\vert\Sp\T^n\Sp\vert R_1\rangle}{\lam_1^n}
514: =\sum_a\langle L_1\vert\Sp\vert R_a\rangle\langle L_a\vert\Sp\vert R_1\rangle
515: \left(\frac{\lam_a}{\lam_1}\right)^n,
516: \label{ctm}
517: \eeq
518: where
519: \be
520: \Sp=\pmatrix{1&0&0&0\cr 0&-1&0&0\cr 0&0&1&0\cr 0&0&0&-1}
521: \ee
522: is the spin operator.
523:
524: Because of symmetry,
525: only the eigenvectors with $a=3,4$ contribute to the above sum.
526: The values $C_0=1$ and $C_1=-E$ allow
527: to determine the products of matrix elements entering~(\ref{ctm}),
528: without knowing the eigenvectors explicitly.
529: The following alternative reasoning can also be used.
530: Being a linear combination of $(\lam_3/\lam_1)^n$ and $(\lam_4/\lam_1)^n$,
531: $C_n$ can be shown to obey the three-term recursion relation
532: \be
533: (1-E)C_{n+2}+2EC_{n+1}+(1+E)C_n=0.
534: \ee
535: The initial values $C_0=1$ and $C_1=-E$ are therefore again sufficient
536: to determine the correlation function for all values of the distance $n$.
537:
538: For $-1/\sqrt{2}\le E\le0$,
539: which contains the range of final energies reached by the dynamics,
540: $\lam_{3,4}=\exp(\pm\i Q)$ are complex numbers with unit modulus, with
541: \beq
542: \tan Q=\frac{\sqrt{1-2E^2}}{-E}\qquad(0\le Q\le\pi/2).
543: \label{qap}
544: \eeq
545: We are thus led to the expression
546: \beq
547: C_n=\left(\frac{1+E}{1-E}\right)^{n/2}
548: \left(\cos nQ+\frac{E^2}{\sqrt{1-2E^2}}\,\sin nQ\right),
549: \label{cap}
550: \eeq
551: for $n\ge0$.
552: In the a priori ensemble,
553: the spin correlation function therefore exhibits an exponential decay,
554: of the form $\exp(-n/\xi_\spin)$, modulated by oscillations at wavevector~$Q$.
555: Both the correlation length
556: \beq
557: \xi_\spin=\frad{2}{\ln\frad{1-E}{1+E}}
558: \label{xiap}
559: \eeq
560: and the wavevector $Q$, given by~(\ref{qap}), depend continuously on energy.
561:
562: We finally quote for further reference the value of the reduced susceptibility
563: \beq
564: \chat=\sum_{n=-\infty}^\infty C_n.
565: \label{chat}
566: \eeq
567: After some algebra, we obtain the simple result
568: \beq
569: \chat=\frac{-E(1-E)}{1+E}.
570: \label{chatap}
571: \eeq
572:
573: \subsection{Distribution of domain sizes}
574:
575: Another characteristic feature of blocked states
576: is the distribution of domain sizes $f(\ell)$,
577: defined as the probability that a given domain
578: consists exactly of $\ell$ consecutive parallel spins.
579: Since isolated spins are absent, domains have at least size two ($\ell\ge2$).
580:
581: In the a priori ensemble, the distribution $f(\ell)$
582: can again be evaluated by the transfer-matrix method.
583: Indeed $\rho(\ell)$, the density (per unit length) of domains consisting of
584: exactly $\ell$ spins,
585: admits an expression similar to the middle side of~(\ref{ctm}), namely
586: \be
587: \rho(\ell)
588: =\frad{\langle L_1\vert\P\,\e^{(\ell-2)\beta}\vert R_1\rangle}{\lam_1^\ell}
589: =\langle L_1\vert\P\vert R_1\rangle
590: \;\frac{1+E}{1-E}\left(\frac{-2E}{1-E}\right)^{\ell-2},
591: \ee
592: where
593: \be
594: \P=\,(\mid\!+-\rangle\langle-+\!\mid)+(\mid\!-+\rangle\langle+-\!\mid)\,
595: =\pmatrix{0&0&0&0\cr 0&0&1&0\cr 0&1&0&0\cr 0&0&0&0}
596: \ee
597: is the appropriate domain boundary operator.
598: The explicit expression of the matrix element is not needed,
599: as it can be fixed by normalization.
600: The probability distribution $f(\ell)$ of domain sizes indeed
601: reads $f(\ell)=\rho(\ell)/\rho$, where
602: $\rho$ is the total density of domains (or equivalently, of domain walls).
603:
604: We thus obtain the geometric (i.e., discrete exponential) probability
605: distribution
606: \beq
607: f(\ell)=\frac{1+E}{1-E}\left(\frac{-2E}{1-E}\right)^{\ell-2}
608: \label{fellap}
609: \eeq
610: for $\ell\ge2$, and consistently
611: \beq
612: \rho=\sum_{\ell=2}^\infty\rho(\ell)=\frac{1}{\mean{\ell}}=\frac{1+E}{2}
613: =\langle L_1\vert\P\vert R_1\rangle.
614: \label{dens}
615: \eeq
616: The characteristic length of~(\ref{fellap}),
617: \beq
618: \xi_\dom=\frad{1}{\ln\frad{1-E}{-2E}},
619: \label{xiclap}
620: \eeq
621: is in general different from the spin correlation length $\xi_\spin$
622: of~(\ref{xiap}), except for the typical value of energy~(\ref{est}),
623: where $\xi_\spin=\xi_\dom=1/S\st$.
624:
625: The mean of the domain size distribution~(\ref{fellap})
626: agrees with~(\ref{dens}), while its variance reads
627: \be
628: \var\ell=\sum_{\ell=2}^\infty\ell^2 f(\ell)-\mean{\ell}^2
629: =\frac{-2E(1-E)}{(1+E)^2}.
630: \ee
631:
632: Consider now a large sample of~$N$ spins,
633: in the canonical a priori ensemble (fixed parameter $\beta$).
634: The number~$M$ of domains in the sample is such that $N=\ell_1+\cdots+\ell_M$
635: is the sum of~$M$ independent variables distributed according to $f(\ell)$,
636: neglecting boundary effects.
637: The expected number of domains and its variance are given by
638: \beq
639: \frac{\mean{M}}{N}\toinf\frac{1}{\mean{\ell}}=\frac{1+E}{2},\qquad
640: \frac{\var{M}}{N}\toinf\frac{\var{\ell}}{\mean{\ell}^3}=\frac{-E(1-E^2)}{4}.
641: \label{varm}
642: \eeq
643: One has therefore
644: \beq
645: N\var{E}=\frac{4\var{M}}{N}\toinf\frac{4\var{\ell}}{\mean{\ell}^3},
646: \label{evarap}
647: \eeq
648: so that the second result of~(\ref{varm})
649: agrees with expression~(\ref{ecu}) for $N\var{E}$.
650:
651: \section{Blocked states: dynamics}
652: \label{s4}
653:
654: In this section we compare the predictions of the a priori approach,
655: derived in Section~\ref{s3}, to the results of numerical simulations
656: concerning the blocked configurations reached by the dynamics.
657: We have used the rules of the zero-temperature Kawasaki dynamics summarized in
658: Table~\ref{t1}, with random sequential updates,
659: starting from a random non-magnetized initial configuration.
660:
661: \subsection{Mean energy}
662:
663: The first and the simplest quantity to be measured is the mean energy $E$
664: of the blocked configurations reached by the dynamics.
665:
666: Let us start with a reminder of the kinetically constrained model $(W_0=0)$.
667: For a random initial configuration, the two kinds of allowed defects, namely
668: isolated unsatisfied bonds (domain walls)
669: and domains of two unsatisfied bond (isolated spins)
670: occur respectively with the following densities in blocked
671: configurations~\cite{us,plk}:
672: \be
673: q_1=\frac{1}{2}\left(1-\e^{-5/4}
674: -\e^{-9/4}\int^{3/2}_1\e^{y^2}\,\d y\right)=0.219704,\qquad
675: q_2=\frac{\e^{-5/4}}{4}=0.071626,
676: \ee
677: so that the mean energy of blocked configurations reads
678: \beq
679: E_0=-1+2q_1+4q_2=-\e^{-9/4}\int^{3/2}_1\e^{y^2}\,\d y=-0.274087.
680: \label{e0}
681: \eeq
682:
683: In the present case $(W_0\ne0)$, the mean energy~$E$ of
684: blocked states is expected to be below this number.
685: Indeed the diffusive moves can only help relaxing more efficiently
686: the energy excess of the disordered initial state.
687:
688: Figure~\ref{f3} shows a plot of the mean final energy $E$
689: against the diffusive rate $W_0$.
690: Each data point corresponds to $10^8$ spins in total.
691: We have checked that no appreciable size dependence is to be observed.
692: The final energy is found to be well below~(\ref{e0}),
693: and below the typical a priori value~(\ref{est}).
694: It only exhibits a very weak dependence on the rate $W_0$,
695: increasing from the extrapolated minimum value
696: \beq
697: E(0)=-0.5279
698: \label{e00}
699: \eeq
700: in the $W_0\to0$ limit to the maximum value
701: \beq
702: E(1)=-0.51633
703: \label{e}
704: \eeq
705: for $W_0=1$.
706:
707: \begin{figure}[htb]
708: \begin{center}
709: \includegraphics[angle=90,width=.7\linewidth]{f3.eps}
710: \caption{\small
711: Plot of the mean final energy~$E$ against the rate $W_0$.
712: Symbols: numerical data.
713: Statistical errors are smaller than the symbol size.
714: Line: fifth-degree polynomial fit, yielding the extrapolated value~(\ref{e00})
715: in the $W_0\to0$ limit.}
716: \label{f3}
717: \end{center}
718: \end{figure}
719:
720: The regime $W_0\ll1$ again deserves some more attention.
721: The mean energy has a discontinuity at $W_0=0$.
722: Indeed the $W_0\to0$ limit of the energy, $E(0)$ given by~(\ref{e00}),
723: is different from that of the constrained dynamics, $E_0$ given by~(\ref{e0}).
724: This discontinuity can be analyzed as follows.
725: In the diffusive part of the dynamics,
726: free spins annihilate by meeting either each other or domain walls.
727: The collision of a single free spin with a domain wall
728: relaxes 4 units of energy,
729: while the coalescence between $n\ge2$ spins within a domain
730: relaxes $4(n-1)$ units of energy, i.e., $4(n-1)/n$ per spin.
731: If all the free spins were annihilated in meeting domain walls,
732: the final energy would assume the value $E=E_0-4p_2$.
733: The efficiency of the diffusive relaxation mechanism
734: can therefore be characterized by the ratio
735: \be
736: \eta=\frac{E_0-E(0)}{4p_2}=1-\sum_{n\ge2}\frac{\Pi_n}{n},
737: \ee
738: where $\Pi_n$ is the probability that a given free spin
739: gets annihilated in a coalescence of $n\ge2$ spins.
740: The extrapolated value~(\ref{e00}) of the final energy
741: yields a rather high efficiency: $\eta=0.886$.
742:
743: In view of the weak dependence of the final energy
744: on the rate $W_0$ (see Figure~\ref{f3}),
745: hereafter we restrict the numerical analysis to the case $W_0=1$
746: of zero-temperature Metropolis dynamics.
747:
748: \subsection{Higher cumulants of energy}
749:
750: We now turn to the statistics of the energy of the blocked configurations,
751: besides its mean value studied above.
752: In analogy with the a priori estimate~(\ref{sap}),
753: the final energy is expected to obey a large-deviation formula of the type
754: \be
755: P(E)\sim\exp(-N\S(E)),
756: \ee
757: with $\S(E)$ being the dynamical entropy.
758: This formula implies that the cumulants of the total energy scale
759: as $\cu{(NE)^k}\sim N$, just as in usual equilibrium situations.
760:
761: Instead of measuring the whole function $\S(E)$,
762: which would require a quite extensive numerical effort,
763: we have measured the first four cumulants of the energy
764: of the blocked configurations.
765: Table~\ref{t2} gives the measured values of the scaled
766: energy cumulants for $W_0=1$ and a random initial configuration.
767: For comparison we also list the predictions of the full
768: a priori ensemble~(\ref{ecu}) for $E=E\st$ of~(\ref{est}),
769: and of the restricted a priori ensemble,
770: obtained by inserting into~(\ref{ecu})
771: the observed value~(\ref{e}) of the mean energy.
772: The simulations have been performed on samples of various sizes
773: ranging from $N=50$ to 200, having $10^{10}$ spins in total.
774: No systematic size dependence is observed.
775: Statistical errors can be estimated to be of the order
776: of one unit of the least significant digit.
777: Both a priori schemes perform very unequally in predicting
778: the energy cumulants.
779: The variance is rather accurately predicted by both schemes,
780: which perform equally poorly for the third cumulant,
781: while the restricted scheme is definitely better for the fourth cumulant.
782:
783: \begin{table}[htb]
784: \begin{center}
785: \begin{tabular}{|c|c|c|c|}
786: \hline
787: scaled cumulant&numerical result&full a priori&restricted a priori\\
788: \hline
789: $\cu{E}=E$&$-0.51633$&$-0.44721$&$-0.51633$\\
790: $N\cu{E^2}=N\var{E}$&{\hskip 1.2mm}$0.3446$&{\hskip 3.3mm}$0.35777$&{\hskip
791: 3.3mm}$0.37868$\\
792: $N^2\cu{E^3}$&{\hskip -0.9mm}$0.031$&$-0.14311$&$-0.07582$\\
793: $N^3\cu{E^4}$&{\hskip -6.3mm}$-0.46$&$-0.28622$&$-0.42906$\\
794: \hline
795: \end{tabular}
796: \caption{\small First four scaled cumulants of the energy
797: of the blocked configurations.
798: Comparison between numerical results
799: and predictions of the full and restricted a priori ensembles (see text).}
800: \label{t2}
801: \end{center}
802: \end{table}
803:
804: \subsection{Spin correlation function}
805:
806: We now turn to the spin correlation function
807: $C_n=\mean{\s_0\s_n}$ in the blocked configurations.
808: This quantity has two remarkable properties.
809: First, since isolated spins are not allowed in blocked configurations,
810: we have $C_1=1-2\rho=-E$ and $C_2=1-4\rho$,
811: where $\rho$ is the density of domain walls~(\ref{dens}), hence
812: \beq
813: C_2=2C_1-1.
814: \label{c1c2}
815: \eeq
816: Second, the total magnetization $M=\sum_n\s_n$ is exactly conserved
817: in Kawasaki dynamics.
818: Furthermore, for a homogeneous non-magnetized state,
819: we have $\mean{M^2}=N\chat$, with the definition~(\ref{chat}).
820: The quantity $\chat$ therefore has the same value in the initial
821: and final configurations, i.e., for a random initial state,
822: \beq
823: \chat=\sum_{n=-\infty}^\infty C_n=1.
824: \label{chat1}
825: \eeq
826:
827: Figure~\ref{f4}
828: shows a comparison between numerical data (circles and full line),
829: corresponding to $3\cdot10^9$ spins in total,
830: and the prediction~(\ref{cap}) of the restricted a priori ensemble
831: at the observed mean energy~(\ref{e}) (triangles and dashed line).
832: The a priori prediction and numerical data coincide
833: both for $n=1$ (by construction, as~$E$ has been imposed)
834: and for $n=2$ (as a consequence of~(\ref{c1c2})).
835: For larger distances, the spin correlations
836: oscillate in sign and fall off exponentially,
837: in qualitative agreement with the a priori prediction.
838: Both the period of oscillations and the decay length
839: are observed to be slightly larger than those of the a priori ensemble.
840: These observations remain however at a qualitative level.
841:
842: We also notice that the a priori ensemble
843: fails to reproduce the identity~(\ref{chat1}).
844: Indeed~$\chat$ of~(\ref{chatap}) equals unity for $E=1-\sqrt{2}=-0.414214$,
845: a value which neither agrees with the most probable
846: a priori energy $E\st$ of~(\ref{est}),
847: nor with the observed mean energy~(\ref{e}).
848: Conversely, the prediction for the correlation function
849: of the restricted a priori ensemble at energy~(\ref{e}),
850: plotted in Figure~\ref{f4},
851: has $\chat=1.6192$, which is significantly different from unity.
852:
853: \begin{figure}[htb]
854: \begin{center}
855: \includegraphics[angle=90,width=.7\linewidth]{f4.eps}
856: \caption{\small
857: Logarithmic plot of the absolute value of the
858: spin correlation function $\abs{C_n}$, against distance $n$.
859: Full (open) symbols show positive (negative) correlations.
860: Circles and full line: numerical data.
861: Statistical errors are smaller than the symbol size.
862: Triangles and dashed line: prediction~(\ref{cap}) of restricted
863: a priori ensemble at energy~(\ref{e}).}
864: \label{f4}
865: \end{center}
866: \end{figure}
867:
868: \subsection{Distribution of domain sizes}
869:
870: We have determined the distribution of domain sizes $f(\ell)$
871: by means of extensive numerical simulations, with a random initial state.
872: Figure~\ref{f5} shows our data corresponding to $3\cdot10^9$ spins in total.
873: The observed distribution is not exponential,
874: at variance with the prediction~(\ref{fellap}) of the a priori ensemble.
875: The full line suggests an exponential asymptotic fall-off of the
876: distribution, with a characteristic length $\xi_\dom=1.75$.
877: This length is close to the characteristic length
878: of domains with a single diffusive spin, $\xi_\dif=1.61$,
879: plotted in Figure~\ref{f1}.
880: As a matter of fact,
881: the equality $\xi_\dom=\xi_\dif$ is expected to hold in the $W_0\to0$ limit.
882: The prediction~(\ref{fellap}) of the restricted a priori ensemble
883: at the observed mean energy~(\ref{e})
884: is shown as a straight dashed line, whose inverse slope
885: is the prediction $\xi_\dom=2.603$~(\ref{xiclap}) of the a priori approach.
886:
887: \begin{figure}[htb]
888: \begin{center}
889: \includegraphics[angle=90,width=.7\linewidth]{f5.eps}
890: \caption{\small
891: Logarithmic plot of the probability distribution $f(\ell)$
892: against domain size $\ell$.
893: Circles: numerical data.
894: Statistical errors are much smaller than the symbol size.
895: Dashed straight line: prediction~(\ref{fellap}) of a priori ensemble
896: at energy~(\ref{e}).
897: Full straight line: guide to the eye with inverse slope $\xi_\dom=1.75$.}
898: \label{f5}
899: \end{center}
900: \end{figure}
901:
902: Another noticeable difference with the prediction of the a priori approach
903: is that the sizes of successive domains in the final states of the dynamics
904: are not independent.
905: Indeed, if they were so, the relation~(\ref{evarap}) would hold,
906: while the data yield $4\var{\ell}/\mean{\ell}^3=0.2790$,
907: a number significantly below the observed variance $N\var{E}=0.3446$,
908: listed in Table~\ref{t2}.
909: Hence domain sizes exhibit a weak but definite {\it positive} correlation.
910: Domain sizes which are neither statistically independent
911: nor exponentially distributed have also been observed recently~\cite{bfs}
912: in spin chains undergoing tapping dynamics.
913:
914: \section{Features of persistence}
915: \label{s5}
916:
917: We now turn to features related to the whole history of a single given spin,
918: say $\s_0$, the spin situated at the origin.
919: This kind of problems belongs to the realm of persistence.
920: In the present context,
921: the central quantity is $\nu$, the total number of times $\s_0$ flips
922: during the history of the sample, before a blocked configuration is reached.
923: The number $\nu$ of spin flips is finite with unit probability
924: in the limit of a large system.
925: Zero-temperature Kawasaki dynamics is therefore of type~${\cal F}$
926: in the classification of~\cite{ns}.
927: Furthermore $\nu$ is random, as it depends both on the initial configuration
928: and on the history of the system.
929: We are therefore interested in the distribution $p_\nu$ of the number of flips,
930: which is expected to have a well-behaved limit when
931: $\s_0$ is deep inside a large enough sample.
932:
933: Before presenting numerical data,
934: we first predict the main salient features of the distribution $p_\nu$,
935: following the lines of Section~\ref{s1}.
936: Consider again a large domain of $L\gg1$ spins,
937: containing the origin, on which a single free spin diffuses.
938: The spin $\s_0$ flips twice each time the free spin traverses the origin.
939: We are thus led to the following effective problem.
940:
941: Consider a random walker in the interval $-L_1<n<L_2$,
942: with absorbing boundaries at $n=-L_1$ and $n=L_2$.
943: The walker starts from the origin ($n=0$).
944: The probability that the walker returns to the origin before being
945: absorbed by either boundary reads
946: \beq
947: P_\ret(L_1,L_2)=1-\frac{1}{2}\left(\frac{1}{L_1}+\frac{1}{L_2}\right).
948: \label{pgamb}
949: \eeq
950: This formula relies on a well-known result in
951: the gambler's ruin problem~\cite{feller}.
952: Indeed, suppose that the walker's first jump is to the right
953: (resp.~to the left),
954: and consider $0\le n\le L_2$ (resp.~$0\le -n\le L_1$) as the gambler's wealth.
955: Then the ruin probability reads $P=1-1/L_2$ (resp.~$P=1-1/L_1$).
956: The expression~(\ref{pgamb}) is the arithmetical mean of both ruin
957: probabilities.
958:
959: The probability that $\s_0$ flips an even number $\nu=2k\gg1$ of times
960: therefore approximately reads
961: $p_{2k}(L_1,L_2)\sim(P_\ret(L_1,L_2))^k(1-P_\ret(L_1,L_2))$.
962: Hence the distribution of the number of flips can
963: be estimated, in analogy with~(\ref{sint}), as
964: \bea
965: p_{2k}&\approx&\sum_{L_1,L_2}(L_1+L_2-1)\,f_\dif(L_1+L_2-1)
966: \,P_\ret(L_1,L_2)^k(1-P_\ret(L_1,L_2))\\
967: &\sim&\left(\int_0^\infty\exp\left(-\frac{L}{\xi_\dif}-\frac{k}{2L}\right)
968: \,\d L\right)^2.
969: \eea
970: The integral entering this expression closely resembles
971: that entering~(\ref{sint}).
972: For $k\gg1$ it is legitimate to use the steepest-descent method.
973: The saddle point lies at $L_c=\sqrt{k\xi_\dif/2}$.
974: We thus obtain the stretched exponential law
975: \beq
976: p_{2k}\sim\sqrt{k}\,\exp\!\left(-2\sqrt{\frac{2k}{\xi_\dif}}\right)
977: \label{gp}
978: \eeq
979: for the distribution of the number of spin flips,
980: provided $\nu=2k$ is a large even number.
981:
982: The occurrence of odd numbers $\nu=2k+1$ of spin flips can also
983: be explained in the above framework,
984: if the spin $\s_0$ is either situated at an endpoint of a domain,
985: or involved in a coalescence event between two free spins.
986: Both effects are expected to scale as the inverse of the domain size $L$.
987: This leads us to predict that odd values of $\nu$ are suppressed
988: by a factor of order $1/L_c$, i.e.,
989: \beq
990: \frac{p_{2k+1}}{p_{2k}}\approx\frac{a}{\sqrt{k\xi_\dif}}.
991: \label{gi}
992: \eeq
993:
994: Figure~\ref{f6} shows a logarithmic plot of
995: the distribution $p_\nu$ of the number of flips.
996: The simulations again concern samples of various sizes
997: having $2\cdot 10^{10}$ spins in total, with $W_0=1$.
998: Even numbers of spin flips $\nu=2k$ (full symbols)
999: are clearly more frequent than odd numbers $\nu=2k+1$ (open symbols),
1000: especially for large values of $k$.
1001: From a quantitative viewpoint, the full lines on Figure~\ref{f6}
1002: show a common fit of the numerical data for $\nu>10$
1003: according to the asymptotic predictions~(\ref{gp}),~(\ref{gi}).
1004: We thus obtain $2/\sqrt{\xi_\dif}\approx1.55$,
1005: in agreement with the data of Figure~\ref{f1},
1006: $\xi_\dif\approx1.61$, i.e., $2/\sqrt{\xi_\dif}\approx1.57$.
1007: We also obtain $a\approx1.5$, albeit with a large uncertainty.
1008:
1009: \begin{figure}[htb]
1010: \begin{center}
1011: \includegraphics[angle=90,width=.7\linewidth]{f6.eps}
1012: \caption{\small
1013: Logarithmic plot of the distribution $p_\nu$ of number of flips.
1014: Full symbols: data for $\nu=2k$ even.
1015: Open symbols: data for $\nu=2k+1$ odd.
1016: Statistical errors are much smaller than the symbol size.
1017: Full lines: common fit described in the text.}
1018: \label{f6}
1019: \end{center}
1020: \end{figure}
1021:
1022: Besides the above analysis of the regime of large numbers of spin flips,
1023: due to long surviving free spins,
1024: our data yield yet other interesting informations.
1025: First, the persistence probability,
1026: i.e., the probability for a spin to never flip,
1027: has the following value:
1028: \be
1029: p_0=0.44739.
1030: \ee
1031: Then, the probability that a spin flips an even or an odd number of times reads
1032: \be
1033: P_\ev=\sum_{k\ge0}p_{2k}=0.72302,\qquad
1034: P_\od=\sum_{k\ge0}p_{2k+1}=1-P_\ev=0.27698.
1035: \ee
1036: These figures can be related to the overlap between
1037: a random initial configuration and the corresponding final one.
1038: We have indeed
1039: \be
1040: Q=\mean{\s_0(0)\s_0(\infty)}=\sum_{\nu\ge0}(-1)^\nu p_\nu=P_\ev-P_\od=0.44604.
1041: \ee
1042:
1043: The mean number of spin flips reads
1044: \be
1045: \mean{\nu}=\sum_{\nu\ge0}\nu\,p_\nu=2.06916.
1046: \ee
1047: This quantity can be used to determine the fractions
1048: $f_\cond$ of condensation moves
1049: and $f_\dif$ of diffusive moves in a typical history,
1050: starting from a random initial configuration.
1051: Indeed, on the one hand, a condensation move lowers the total energy
1052: by four units, while a diffusive moves leaves it unchanged
1053: (see Table~\ref{t1}).
1054: On the other hand, any move involves exactly two spin flips.
1055: We have therefore, using~(\ref{e}),
1056: \be
1057: f_\cond=1-f_\dif=-\frac{E}{2\mean{\nu}}=0.12477.
1058: \ee
1059:
1060: To close up, let us compare the above results
1061: to the case of constrained Kawasaki dynamics ($W_0=0$)~\cite{us,plk},
1062: where only condensation moves are allowed.
1063: In this situation, the above quantities
1064: can be simply related to the final energy $E_0$ of~(\ref{e0}).
1065: Indeed every spin flips at most once,
1066: so that only $p_0$ and $p_1=1-p_0$ are non-zero, and $f_\cond=1$.
1067: Starting again from a random configuration, we are left with
1068: \be
1069: 2(1-p_0)=1-Q=2\mean{\nu}=-E_0=0.274087.
1070: \ee
1071:
1072: \section{Discussion}
1073:
1074: First, we wish to emphasize the richness of the zero-temperature
1075: dynamics of the ferromagnetic Ising chain.
1076: There are indeed four different natural kinds of dynamics,
1077: summarized in Table~\ref{t3}.
1078: Only Glauber dynamics gives rise to a bona fide coarsening dynamics,
1079: obeying dynamical scaling with a typical domain size
1080: growing as $L(t)\sim\sqrt{t}$.
1081: With the three other dynamics,
1082: the system is left in a metastable configuration after a
1083: relatively short blocking time.
1084: The constrained Glauber and Kawasaki dynamics
1085: have been analyzed in our previous work~\cite{us}.
1086: Each spin may flip at most once, before a global blocked state is reached.
1087: The blocking time scales with the number of spins as $\ln N$.
1088: These models can be exactly mapped onto
1089: the random sequential addition problem of dimers and hollow trimers,
1090: respectively, hence allowing an analytical treatment.
1091:
1092: \begin{table}[htb]
1093: \begin{center}
1094: \begin{tabular}{|l|l|l|l|}
1095: \hline
1096: $\matrix{\hbox{conserved}\hfill\cr\hbox{magnetization}}$
1097: &$\matrix{\hbox{diffusive}\cr\hbox{moves}\hfill}$
1098: &dynamics&behavior\\
1099: \hline
1100: no&yes&Glauber~\cite{glau}&coarsening~\cite{bray}\\
1101: no&no&constrained Glauber~\cite{cg}&metastability~\cite{us}\\
1102: yes&yes&Kawasaki~\cite{kawa}&metastability~[this work]\\
1103: yes&no&constrained Kawasaki~\cite{plk}&metastability~\cite{us}\\
1104: \hline
1105: \end{tabular}
1106: \caption{\small Four different zero-temperature dynamics of
1107: the ferromagnetic Ising chain, with appropriate references (brackets).}
1108: \label{t3}
1109: \end{center}
1110: \end{table}
1111:
1112: The last case not considered so far from the viewpoint of metastability,
1113: Kawasaki dynamics, has been the subject of this work.
1114: Because of the diffusive moves, the dynamics is only partly irreversible.
1115: This novel feature makes the Kawasaki problem both richer
1116: and hopefully closer to more realistic situations.
1117: The number of flips of a given spin, although finite with probability one,
1118: may be arbitrarily large.
1119: The blocking time grows as $(\ln N)^3$.
1120: On the other hand, analytical tools being no longer available,
1121: we needed to have recourse to numerical simulations.
1122:
1123: The present work
1124: demonstrates that there is no good a priori statistical description
1125: of the metastable states reached by Kawasaki dynamics.
1126: In other words, the Edwards hypothesis is invalidated,
1127: at least as an exact prescription.
1128: Systematic differences are indeed observed
1129: between numerical results and a priori predictions,
1130: especially in the pattern of spin correlations~(Figure~\ref{f4})
1131: and in the distribution of domain sizes~(Figure~\ref{f5}).
1132: The latter are neither statistically independent nor exponentially distributed.
1133:
1134: We also want to underline that the effective trapping description
1135: of the late stages of the zero-temperature Kawasaki dynamics,
1136: already emphasized in Ref.~\cite{cks},
1137: appears to yield quantitative predictions
1138: for several novel physical quantities,
1139: including the statistics of the blocking time
1140: ((\ref{xdef}), (\ref{gum}), Figure~\ref{f2}),
1141: and the distribution of the number of spin flips
1142: ((\ref{gp}), (\ref{gi}), Figure~\ref{f6}).
1143:
1144: \subsubsection*{Acknowledgements}
1145:
1146: Interesting discussions with Silvio Franz are gratefully acknowledged.
1147:
1148: \newpage
1149: \begin{thebibliography}{99}
1150:
1151: \bibitem{gold} M. Goldstein, J. Chem. Phys. {\bf 51}, 3728 (1969).
1152:
1153: \bibitem{tap} D.J. Thouless, P.W. Anderson, and R.G. Palmer, Phil. Mag.
1154: {\bf 35}, 593 (1977).
1155:
1156: \bibitem{ktw}
1157: T.R. Kirkpatrick and P.G. Wolynes, Phys. Rev. A {\bf 35}, 3072 (1987);
1158: T.R. Kirkpatrick and D. Thirumalai, Phys. Rev. B {\bf 36}, 5388 (1987);
1159: T.R. Kirkpatrick and P.G. Wolynes, Phys. Rev. B {\bf 36}, 8552 (1987);
1160: D. Thirumalai and T.R. Kirkpatrick, Phys. Rev. B {\bf 38}, 4881 (1988).
1161:
1162: \bibitem{sw} F.H. Stillinger and T.A. Weber, Phys. Rev. A {\bf 25}, 978 (1982);
1163: Science {\bf 225}, 983 (1984).
1164:
1165: \bibitem{fv} S. Franz and M.A. Virasoro, J. Phys. A {\bf 33}, 891 (2000).
1166:
1167: \bibitem{bir} G. Biroli and R. Monasson, Europhys. Lett. {\bf 50}, 155 (2000);
1168: G. Biroli and J. Kurchan, Phys. Rev. E {\bf 64}, 016101 (2001).
1169:
1170: \bibitem{edwards} S.F. Edwards, in {\it Granular Matter: An Interdisciplinary
1171: Approach}, ed. A. Mehta (Springer, New York, 1994).
1172:
1173: \bibitem{us} G. De Smedt, C. Godr\`eche, and J.M. Luck, Eur. Phys. J. B
1174: {\bf 27}, 363 (2002).
1175:
1176: \bibitem{rsa} For a review, see: J.W. Evans, Rev. Mod. Phys. {\bf 65}, 1281
1177: (1993).
1178:
1179: \bibitem{kawa} K. Kawasaki, Phys. Rev. {\bf 145}, 224 (1966).
1180:
1181: \bibitem{cks} S.J. Cornell, K. Kaski, and R.B. Stinchcombe, Phys. Rev. B
1182: {\bf 44}, 12263 (1991).
1183:
1184: \bibitem{plk} V. Privman, Phys. Rev. Lett. {\bf 69}, 3686 (1992);
1185: J.C. Lin and P.L. Taylor, Phys. Rev. E {\bf 48}, 4305 (1993);
1186: P.L. Krapivsky, J. Stat. Phys. {\bf 74}, 1211 (1994).
1187:
1188: \bibitem{trap}
1189: M.D. Donsker and S.R.S. Varadhan, Commun. Pure Appl. Math. {\bf 28},
1190: 525 (1975); {\bf 32}, 721 (1979);
1191: P. Grassberger and I. Procaccia, J. Chem. Phys. {\bf 77}, 6281 (1982);
1192: J.W. Haus and K.W. Kehr, Phys. Rep. {\bf 150}, 263 (1987);
1193: J.M. Luck, {\it Syst\`emes d\'esordonn\'es unidimensionnels}
1194: (Collection Al\'ea, Saclay, 1992).
1195:
1196: \bibitem{gumbel} E.J. Gumbel, {\it Statistics of Extremes} (Columbia
1197: University Press, 1958).
1198:
1199: \bibitem{bfs} J. Berg, S. Franz, and M. Sellitto, Eur. Phys. J. B {\bf 26},
1200: 349 (2002).
1201:
1202: \bibitem{ns} C.M. Newman and D.L. Stein, Physica A {\bf 279}, 159 (2000).
1203:
1204: \bibitem{feller} W. Feller, {\it An Introduction to Probability Theory and its
1205: Applications}, in 2 volumes (Wiley, New-York, 1966).
1206:
1207: \bibitem{glau} R.J. Glauber, J. Math. Phys. {\bf 4}, 294 (1963).
1208:
1209: \bibitem{bray} A.J. Bray, Adv. Phys. {\bf 43}, 357 (1994).
1210:
1211: \bibitem{cg}
1212: D.S. Dean and A. Lef\`evre, Phys. Rev. Lett. {\bf 86}, 5639 (2001);
1213: A. Lef\`evre and D.S. Dean, J. Phys. A {\bf 34}, L213 (2001);
1214: A. Prados and J.J. Brey, J. Phys. A {\bf 34}, L453 (2001).
1215:
1216: \end{thebibliography}
1217: \end{document}
1218: